[Rework] Update zstd to 1.4.5

58 files changed, 15354 insertions, 10112 deletions

diff --git a/contrib/zstd/CHANGELOG b/contrib/zstd/CHANGELOG
new file mode 100644
index 000000000..0ed939a5b
--- /dev/null
+++ b/contrib/zstd/CHANGELOG
@@ -0,0 +1,555 @@
+v1.4.5
+fix : Compression ratio regression on huge files (> 3 GB) using high levels (--ultra) and multithreading, by @terrelln
+perf: Improved decompression speed: x64 : +10% (clang) / +5% (gcc); ARM : from +15% to +50%, depending on SoC, by @terrelln
+perf: Automatically downsizes ZSTD_DCtx when too large for too long (#2069, by @bimbashreshta)
+perf: Improved fast compression speed on aarch64 (#2040, ~+3%, by @caoyzh)
+perf: Small level 1 compression speed gains (depending on compiler)
+cli : New --patch-from command, create and apply patches from files, by @bimbashreshta
+cli : New --filelist= : Provide a list of files to operate upon from a file
+cli : -b -d command can now benchmark decompression on multiple files
+cli : New --no-content-size command
+cli : New --show-default-cparams information command
+api : ZDICT_finalizeDictionary() is promoted to stable (#2111)
+api : new experimental parameter ZSTD_d_stableOutBuffer (#2094)
+build: Generate a single-file libzstd library (#2065, by @cwoffenden)
+build: Relative includes no longer require -I compiler flags for zstd lib subdirs (#2103, by @felixhandte)
+build: zstd now compiles cleanly under -pedantic (#2099)
+build: zstd now compiles with make-4.3
+build: Support mingw cross-compilation from Linux, by @Ericson2314
+build: Meson multi-thread build fix on windows
+build: Some misc icc fixes backed by new ci test on travis
+misc: bitflip analyzer tool, by @felixhandte
+misc: Extend largeNbDicts benchmark to compression
+misc: Edit-distance match finder in contrib/
+doc : Improved beginner CONTRIBUTING.md docs
+doc : New issue templates for zstd
+
+v1.4.4
+perf: Improved decompression speed, by > 10%, by @terrelln
+perf: Better compression speed when re-using a context, by @felixhandte
+perf: Fix compression ratio when compressing large files with small dictionary, by @senhuang42
+perf: zstd reference encoder can generate RLE blocks, by @bimbashrestha
+perf: minor generic speed optimization, by @davidbolvansky
+api: new ability to extract sequences from the parser for analysis, by @bimbashrestha
+api: fixed decoding of magic-less frames, by @terrelln
+api: fixed ZSTD_initCStream_advanced() performance with fast modes, reported by @QrczakMK
+cli: Named pipes support, by @bimbashrestha
+cli: short tar's extension support, by @stokito
+cli: command --output-dir-flat= , generates target files into requested directory, by @senhuang42
+cli: commands --stream-size=# and --size-hint=#, by @nmagerko
+cli: command --exclude-compressed, by @shashank0791
+cli: faster `-t` test mode
+cli: improved some error messages, by @vangyzen
+cli: fix command `-D dictionary` on Windows, reported by @artyompetrov
+cli: fix rare deadlock condition within dictionary builder, by @terrelln
+build: single-file decoder with emscripten compilation script, by @cwoffenden
+build: fixed zlibWrapper compilation on Visual Studio, reported by @bluenlive
+build: fixed deprecation warning for certain gcc version, reported by @jasonma163
+build: fix compilation on old gcc versions, by @cemeyer
+build: improved installation directories for cmake script, by Dmitri Shubin
+pack: modified pkgconfig, for better integration into openwrt, requested by @neheb
+misc: Improved documentation : ZSTD_CLEVEL, DYNAMIC_BMI2, ZSTD_CDict, function deprecation, zstd format
+misc: fixed educational decoder : accept larger literals section, and removed UNALIGNED() macro
+
+v1.4.3
+bug: Fix Dictionary Compression Ratio Regression by @cyan4973 (#1709)
+bug: Fix Buffer Overflow in legacy v0.3 decompression by @felixhandte (#1722)
+build: Add support for IAR C/C++ Compiler for Arm by @joseph0918 (#1705)
+
+v1.4.2
+bug: Fix bug in zstd-0.5 decoder by @terrelln (#1696)
+bug: Fix seekable decompression in-memory API by @iburinoc (#1695)
+misc: Validate blocks are smaller than size limit by @vivekmg (#1685)
+misc: Restructure source files by @ephiepark (#1679)
+
+v1.4.1
+bug: Fix data corruption in niche use cases by @terrelln (#1659)
+bug: Fuzz legacy modes, fix uncovered bugs by @terrelln (#1593, #1594, #1595)
+bug: Fix out of bounds read by @terrelln (#1590)
+perf: Improve decode speed by ~7% @mgrice (#1668)
+perf: Slightly improved compression ratio of level 3 and 4 (ZSTD_dfast) by @cyan4973 (#1681)
+perf: Slightly faster compression speed when re-using a context by @cyan4973 (#1658)
+perf: Improve compression ratio for small windowLog by @cyan4973 (#1624)
+perf: Faster compression speed in high compression mode for repetitive data by @terrelln (#1635)
+api: Add parameter to generate smaller dictionaries by @tyler-tran (#1656)
+cli: Recognize symlinks when built in C99 mode by @felixhandte (#1640)
+cli: Expose cpu load indicator for each file on -vv mode by @ephiepark (#1631)
+cli: Restrict read permissions on destination files by @chungy (#1644)
+cli: zstdgrep: handle -f flag by @felixhandte (#1618)
+cli: zstdcat: follow symlinks by @vejnar (#1604)
+doc: Remove extra size limit on compressed blocks by @felixhandte (#1689)
+doc: Fix typo by @yk-tanigawa (#1633)
+doc: Improve documentation on streaming buffer sizes by @cyan4973 (#1629)
+build: CMake: support building with LZ4 @leeyoung624 (#1626)
+build: CMake: install zstdless and zstdgrep by @leeyoung624 (#1647)
+build: CMake: respect existing uninstall target by @j301scott (#1619)
+build: Make: skip multithread tests when built without support by @michaelforney (#1620)
+build: Make: Fix examples/ test target by @sjnam (#1603)
+build: Meson: rename options out of deprecated namespace by @lzutao (#1665)
+build: Meson: fix build by @lzutao (#1602)
+build: Visual Studio: don't export symbols in static lib by @scharan (#1650)
+build: Visual Studio: fix linking by @absotively (#1639)
+build: Fix MinGW-W64 build by @myzhang1029 (#1600)
+misc: Expand decodecorpus coverage by @ephiepark (#1664)
+
+v1.4.0
+perf: Improve level 1 compression speed in most scenarios by 6% by @gbtucker and @terrelln
+api: Move the advanced API, including all functions in the staging section, to the stable section
+api: Make ZSTD_e_flush and ZSTD_e_end block for maximum forward progress
+api: Rename ZSTD_CCtxParam_getParameter to ZSTD_CCtxParams_getParameter
+api: Rename ZSTD_CCtxParam_setParameter to ZSTD_CCtxParams_setParameter
+api: Don't export ZSTDMT functions from the shared library by default
+api: Require ZSTD_MULTITHREAD to be defined to use ZSTDMT
+api: Add ZSTD_decompressBound() to provide an upper bound on decompressed size by @shakeelrao
+api: Fix ZSTD_decompressDCtx() corner cases with a dictionary
+api: Move ZSTD_getDictID_*() functions to the stable section
+api: Add ZSTD_c_literalCompressionMode flag to enable or disable literal compression by @terrelln
+api: Allow compression parameters to be set when a dictionary is used
+api: Allow setting parameters before or after ZSTD_CCtx_loadDictionary() is called
+api: Fix ZSTD_estimateCStreamSize_usingCCtxParams()
+api: Setting ZSTD_d_maxWindowLog to 0 means use the default
+cli: Ensure that a dictionary is not used to compress itself by @shakeelrao
+cli: Add --[no-]compress-literals flag to enable or disable literal compression
+doc: Update the examples to use the advanced API
+doc: Explain how to transition from old streaming functions to the advanced API in the header
+build: Improve the Windows release packages
+build: Improve CMake build by @hjmjohnson
+build: Build fixes for FreeBSD by @lwhsu
+build: Remove redundant warnings by @thatsafunnyname
+build: Fix tests on OpenBSD by @bket
+build: Extend fuzzer build system to work with the new clang engine
+build: CMake now creates the libzstd.so.1 symlink
+build: Improve Menson build by @lzutao
+misc: Fix symbolic link detection on FreeBSD
+misc: Use physical core count for -T0 on FreeBSD by @cemeyer
+misc: Fix zstd --list on truncated files by @kostmo
+misc: Improve logging in debug mode by @felixhandte
+misc: Add CirrusCI tests by @lwhsu
+misc: Optimize dictionary memory usage in corner cases
+misc: Improve the dictionary builder on small or homogeneous data
+misc: Fix spelling across the repo by @jsoref
+
+v1.3.8
+perf: better decompression speed on large files (+7%) and cold dictionaries (+15%)
+perf: slightly better compression ratio at high compression modes
+api : finalized advanced API, last stage before "stable" status
+api : new --rsyncable mode, by @terrelln
+api : support decompression of empty frames into NULL (used to be an error) (#1385)
+build: new set of macros to build a minimal size decoder, by @felixhandte
+build: fix compilation on MIPS32, reported by @clbr (#1441)
+build: fix compilation with multiple -arch flags, by @ryandesign
+build: highly upgraded meson build, by @lzutao
+build: improved buck support, by @obelisk
+build: fix cmake script : can create debug build, by @pitrou
+build: Makefile : grep works on both colored consoles and systems without color support
+build: fixed zstd-pgo, by @bmwiedemann
+cli : support ZSTD_CLEVEL environment variable, by @yijinfb (#1423)
+cli : --no-progress flag, preserving final summary (#1371), by @terrelln
+cli : ensure destination file is not source file (#1422)
+cli : clearer error messages, especially when input file not present
+doc : clarified zstd_compression_format.md, by @ulikunitz
+misc: fixed zstdgrep, returns 1 on failure, by @lzutao
+misc: NEWS renamed as CHANGELOG, in accordance with fboss
+
+v1.3.7
+perf: slightly better decompression speed on clang (depending on hardware target)
+fix : performance of dictionary compression for small input < 4 KB at levels 9 and 10
+build: no longer build backtrace by default in release mode; restrict further automatic mode
+build: control backtrace support through build macro BACKTRACE
+misc: added man pages for zstdless and zstdgrep, by @samrussell
+
+v1.3.6
+perf: much faster dictionary builder, by @jenniferliu
+perf: faster dictionary compression on small data when using multiple contexts, by @felixhandte
+perf: faster dictionary decompression when using a very large number of dictionaries simultaneously
+cli : fix : does no longer overwrite destination when source does not exist (#1082)
+cli : new command --adapt, for automatic compression level adaptation
+api : fix : block api can be streamed with > 4 GB, reported by @catid
+api : reduced ZSTD_DDict size by 2 KB
+api : minimum negative compression level is defined, and can be queried using ZSTD_minCLevel().
+build: support Haiku target, by @korli
+build: Read Legacy format is limited to v0.5+ by default. Can be changed at compile time with macro ZSTD_LEGACY_SUPPORT.
+doc : zstd_compression_format.md updated to match wording in IETF RFC 8478
+misc: tests/paramgrill, a parameter optimizer, by @GeorgeLu97
+
+v1.3.5
+perf: much faster dictionary compression, by @felixhandte
+perf: small quality improvement for dictionary generation, by @terrelln
+perf: slightly improved high compression levels (notably level 19)
+mem : automatic memory release for long duration contexts
+cli : fix : overlapLog can be manually set
+cli : fix : decoding invalid lz4 frames
+api : fix : performance degradation for dictionary compression when using advanced API, by @terrelln
+api : change : clarify ZSTD_CCtx_reset() vs ZSTD_CCtx_resetParameters(), by @terrelln
+build: select custom libzstd scope through control macros, by @GeorgeLu97
+build: OpenBSD patch, by @bket
+build: make and make all are compatible with -j
+doc : clarify zstd_compression_format.md, updated for IETF RFC process
+misc: pzstd compatible with reproducible compilation, by @lamby
+
+v1.3.4
+perf: faster speed (especially decoding speed) on recent cpus (haswell+)
+perf: much better performance associating --long with multi-threading, by @terrelln
+perf: better compression at levels 13-15
+cli : asynchronous compression by default, for faster experience (use --single-thread for former behavior)
+cli : smoother status report in multi-threading mode
+cli : added command --fast=#, for faster compression modes
+cli : fix crash when not overwriting existing files, by Pádraig Brady (@pixelb)
+api : `nbThreads` becomes `nbWorkers` : 1 triggers asynchronous mode
+api : compression levels can be negative, for even more speed
+api : ZSTD_getFrameProgression() : get precise progress status of ZSTDMT anytime
+api : ZSTDMT can accept new compression parameters during compression
+api : implemented all advanced dictionary decompression prototypes
+build: improved meson recipe, by Shawn Landden (@shawnl)
+build: VS2017 scripts, by @HaydnTrigg
+misc: all /contrib projects fixed
+misc: added /contrib/docker script by @gyscos
+
+v1.3.3
+perf: faster zstd_opt strategy (levels 16-19)
+fix : bug #944 : multithreading with shared ditionary and large data, reported by @gsliepen
+cli : fix : content size written in header by default
+cli : fix : improved LZ4 format support, by @felixhandte
+cli : new : hidden command `-S`, to benchmark multiple files while generating one result per file
+api : fix : support large skippable frames, by @terrelln
+api : fix : streaming interface was adding a useless 3-bytes null block to small frames
+api : change : when setting `pledgedSrcSize`, use `ZSTD_CONTENTSIZE_UNKNOWN` macro value to mean "unknown"
+build: fix : compilation under rhel6 and centos6, reported by @pixelb
+build: added `check` target
+
+v1.3.2
+new : long range mode, using --long command, by Stella Lau (@stellamplau)
+new : ability to generate and decode magicless frames (#591)
+changed : maximum nb of threads reduced to 200, to avoid address space exhaustion in 32-bits mode
+fix : multi-threading compression works with custom allocators
+fix : ZSTD_sizeof_CStream() was over-evaluating memory usage
+fix : a rare compression bug when compression generates very large distances and bunch of other conditions (only possible at --ultra -22)
+fix : 32-bits build can now decode large offsets (levels 21+)
+cli : added LZ4 frame support by default, by Felix Handte (@felixhandte)
+cli : improved --list output
+cli : new : can split input file for dictionary training, using command -B#
+cli : new : clean operation artefact on Ctrl-C interruption
+cli : fix : do not change /dev/null permissions when using command -t with root access, reported by @mike155 (#851)
+cli : fix : write file size in header in multiple-files mode
+api : added macro ZSTD_COMPRESSBOUND() for static allocation
+api : experimental : new advanced decompression API
+api : fix : sizeof_CCtx() used to over-estimate
+build: fix : no-multithread variant compiles without pool.c dependency, reported by Mitchell Blank Jr (@mitchblank) (#819)
+build: better compatibility with reproducible builds, by Bernhard M. Wiedemann (@bmwiedemann) (#818)
+example : added streaming_memory_usage
+license : changed /examples license to BSD + GPLv2
+license : fix a few header files to reflect new license (#825)
+
+v1.3.1
+New license : BSD + GPLv2
+perf: substantially decreased memory usage in Multi-threading mode, thanks to reports by Tino Reichardt (@mcmilk)
+perf: Multi-threading supports up to 256 threads. Cap at 256 when more are requested (#760)
+cli : improved and fixed --list command, by @ib (#772)
+cli : command -vV to list supported formats, by @ib (#771)
+build : fixed binary variants, reported by @svenha (#788)
+build : fix Visual compilation for non x86/x64 targets, reported by Greg Slazinski (@GregSlazinski) (#718)
+API exp : breaking change : ZSTD_getframeHeader() provides more information
+API exp : breaking change : pinned down values of error codes
+doc : fixed huffman example, by Ulrich Kunitz (@ulikunitz)
+new : contrib/adaptive-compression, I/O driven compression strength, by Paul Cruz (@paulcruz74)
+new : contrib/long_distance_matching, statistics by Stella Lau (@stellamplau)
+updated : contrib/linux-kernel, by Nick Terrell (@terrelln)
+
+v1.3.0
+cli : new : `--list` command, by Paul Cruz
+cli : changed : xz/lzma support enabled by default
+cli : changed : `-t *` continue processing list after a decompression error
+API : added : ZSTD_versionString()
+API : promoted to stable status : ZSTD_getFrameContentSize(), by Sean Purcell
+API exp : new advanced API : ZSTD_compress_generic(), ZSTD_CCtx_setParameter()
+API exp : new : API for static or external allocation : ZSTD_initStatic?Ctx()
+API exp : added : ZSTD_decompressBegin_usingDDict(), requested by Guy Riddle (#700)
+API exp : clarified memory estimation / measurement functions.
+API exp : changed : strongest strategy renamed ZSTD_btultra, fastest strategy ZSTD_fast set to 1
+tools : decodecorpus can generate random dictionary-compressed samples, by Paul Cruz
+new : contrib/seekable_format, demo and API, by Sean Purcell
+changed : contrib/linux-kernel, updated version and license, by Nick Terrell
+
+v1.2.0
+cli : changed : Multithreading enabled by default (use target zstd-nomt or HAVE_THREAD=0 to disable)
+cli : new : command -T0 means "detect and use nb of cores", by Sean Purcell
+cli : new : zstdmt symlink hardwired to `zstd -T0`
+cli : new : command --threads=# (#671)
+cli : changed : cover dictionary builder by default, for improved quality, by Nick Terrell
+cli : new : commands --train-cover and --train-legacy, to select dictionary algorithm and parameters
+cli : experimental targets `zstd4` and `xzstd4`, with support for lz4 format, by Sean Purcell
+cli : fix : does not output compressed data on console
+cli : fix : ignore symbolic links unless --force specified,
+API : breaking change : ZSTD_createCDict_advanced(), only use compressionParameters as argument
+API : added : prototypes ZSTD_*_usingCDict_advanced(), for direct control over frameParameters.
+API : improved: ZSTDMT_compressCCtx() reduced memory usage
+API : fix : ZSTDMT_compressCCtx() now provides srcSize in header (#634)
+API : fix : src size stored in frame header is controlled at end of frame
+API : fix : enforced consistent rules for pledgedSrcSize==0 (#641)
+API : fix : error code "GENERIC" replaced by "dstSizeTooSmall" when appropriate
+build: improved cmake script, by @Majlen
+build: enabled Multi-threading support for *BSD, by Baptiste Daroussin
+tools: updated Paramgrill. Command -O# provides best parameters for sample and speed target.
+new : contrib/linux-kernel version, by Nick Terrell
+
+v1.1.4
+cli : new : can compress in *.gz format, using --format=gzip command, by Przemyslaw Skibinski
+cli : new : advanced benchmark command --priority=rt
+cli : fix : write on sparse-enabled file systems in 32-bits mode, by @ds77
+cli : fix : --rm remains silent when input is stdin
+cli : experimental : xzstd, with support for xz/lzma decoding, by Przemyslaw Skibinski
+speed : improved decompression speed in streaming mode for single shot scenarios (+5%)
+memory: DDict (decompression dictionary) memory usage down from 150 KB to 20 KB
+arch: 32-bits variant able to generate and decode very long matches (>32 MB), by Sean Purcell
+API : new : ZSTD_findFrameCompressedSize(), ZSTD_getFrameContentSize(), ZSTD_findDecompressedSize()
+API : changed : dropped support of legacy versions <= v0.3 (can be changed by modifying ZSTD_LEGACY_SUPPORT value)
+build : new: meson build system in contrib/meson, by Dima Krasner
+build : improved cmake script, by @Majlen
+build : added -Wformat-security flag, as recommended by Padraig Brady
+doc : new : educational decoder, by Sean Purcell
+
+v1.1.3
+cli : zstd can decompress .gz files (can be disabled with `make zstd-nogz` or `make HAVE_ZLIB=0`)
+cli : new : experimental target `make zstdmt`, with multi-threading support
+cli : new : improved dictionary builder "cover" (experimental), by Nick Terrell, based on prior work by Giuseppe Ottaviano.
+cli : new : advanced commands for detailed parameters, by Przemyslaw Skibinski
+cli : fix zstdless on Mac OS-X, by Andrew Janke
+cli : fix #232 "compress non-files"
+dictBuilder : improved dictionary generation quality, thanks to Nick Terrell
+API : new : lib/compress/ZSTDMT_compress.h multithreading API (experimental)
+API : new : ZSTD_create?Dict_byReference(), requested by Bartosz Taudul
+API : new : ZDICT_finalizeDictionary()
+API : fix : ZSTD_initCStream_usingCDict() properly writes dictID into frame header, by Gregory Szorc (#511)
+API : fix : all symbols properly exposed in libzstd, by Nick Terrell
+build : support for Solaris target, by Przemyslaw Skibinski
+doc : clarified specification, by Sean Purcell
+
+v1.1.2
+API : streaming : decompression : changed : automatic implicit reset when chain-decoding new frames without init
+API : experimental : added : dictID retrieval functions, and ZSTD_initCStream_srcSize()
+API : zbuff : changed : prototypes now generate deprecation warnings
+lib : improved : faster decompression speed at ultra compression settings and 32-bits mode
+lib : changed : only public ZSTD_ symbols are now exposed
+lib : changed : reduced usage  of stack memory
+lib : fixed : several corner case bugs, by Nick Terrell
+cli : new : gzstd, experimental version able to decode .gz files, by Przemyslaw Skibinski
+cli : new : preserve file attributes
+cli : new : added zstdless and zstdgrep tools
+cli : fixed : status displays total amount decoded, even for file consisting of multiple frames (like pzstd)
+cli : fixed : zstdcat
+zlib_wrapper : added support for gz* functions, by Przemyslaw Skibinski
+install : better compatibility with FreeBSD, by Dimitry Andric
+source tree : changed : zbuff source files moved to lib/deprecated
+
+v1.1.1
+New : command -M#, --memory=, --memlimit=, --memlimit-decompress= to limit allowed memory consumption
+New : doc/zstd_manual.html, by Przemyslaw Skibinski
+Improved : slightly better compression ratio at --ultra levels (>= 20)
+Improved : better memory usage when using streaming compression API, thanks to @Rogier-5 report
+Added : API : ZSTD_initCStream_usingCDict(), ZSTD_initDStream_usingDDict() (experimental section)
+Added : example/multiple_streaming_compression.c
+Changed : zstd_errors.h is now installed within /include (and replaces errors_public.h)
+Updated man page
+Fixed : zstd-small, zstd-compress and zstd-decompress compilation targets
+
+v1.1.0
+New : contrib/pzstd, parallel version of zstd, by Nick Terrell
+added : NetBSD install target (#338)
+Improved : speed for batches of small files
+Improved : speed of zlib wrapper, by Przemyslaw Skibinski
+Changed : libzstd on Windows supports legacy formats, by Christophe Chevalier
+Fixed : CLI -d output to stdout by default when input is stdin (#322)
+Fixed : CLI correctly detects console on Mac OS-X
+Fixed : CLI supports recursive mode `-r` on Mac OS-X
+Fixed : Legacy decoders use unified error codes, reported by benrg (#341), fixed by Przemyslaw Skibinski
+Fixed : compatibility with OpenBSD, reported by Juan Francisco Cantero Hurtado (#319)
+Fixed : compatibility with Hurd, by Przemyslaw Skibinski (#365)
+Fixed : zstd-pgo, reported by octoploid (#329)
+
+v1.0.0
+Change Licensing, all project is now BSD, Copyright Facebook
+Small decompression speed improvement
+API : Streaming API supports legacy format
+API : ZDICT_getDictID(), ZSTD_sizeof_{CCtx, DCtx, CStream, DStream}(), ZSTD_setDStreamParameter()
+CLI supports legacy formats v0.4+
+Fixed : compression fails on certain huge files, reported by Jesse McGrew
+Enhanced documentation, by Przemyslaw Skibinski
+
+v0.8.1
+New streaming API
+Changed : --ultra now enables levels beyond 19
+Changed : -i# now selects benchmark time in second
+Fixed : ZSTD_compress* can now compress > 4 GB in a single pass, reported by Nick Terrell
+Fixed : speed regression on specific patterns (#272)
+Fixed : support for Z_SYNC_FLUSH, by Dmitry Krot (#291)
+Fixed : ICC compilation, by Przemyslaw Skibinski
+
+v0.8.0
+Improved : better speed on clang and gcc -O2, thanks to Eric Biggers
+New : Build on FreeBSD and DragonFly, thanks to JrMarino
+Changed : modified API : ZSTD_compressEnd()
+Fixed : legacy mode with ZSTD_HEAPMODE=0, by Christopher Bergqvist
+Fixed : premature end of frame when zero-sized raw block, reported by Eric Biggers
+Fixed : large dictionaries (> 384 KB), reported by Ilona Papava
+Fixed : checksum correctly checked in single-pass mode
+Fixed : combined --test amd --rm, reported by Andreas M. Nilsson
+Modified : minor compression level adaptations
+Updated : compression format specification to v0.2.0
+changed : zstd.h moved to /lib directory
+
+v0.7.5
+Transition version, supporting decoding of v0.8.x
+
+v0.7.4
+Added : homebrew for Mac, by Daniel Cade
+Added : more examples
+Fixed : segfault when using small dictionaries, reported by Felix Handte
+Modified : default compression level for CLI is now 3
+Updated : specification, to v0.1.1
+
+v0.7.3
+New : compression format specification
+New : `--` separator, stating that all following arguments are file names. Suggested by Chip Turner.
+New : `ZSTD_getDecompressedSize()`
+New : OpenBSD target, by Juan Francisco Cantero Hurtado
+New : `examples` directory
+fixed : dictBuilder using HC levels, reported by Bartosz Taudul
+fixed : legacy support from ZSTD_decompress_usingDDict(), reported by Felix Handte
+fixed : multi-blocks decoding with intermediate uncompressed blocks, reported by Greg Slazinski
+modified : removed "mem.h" and "error_public.h" dependencies from "zstd.h" (experimental section)
+modified : legacy functions no longer need magic number
+
+v0.7.2
+fixed : ZSTD_decompressBlock() using multiple consecutive blocks. Reported by Greg Slazinski.
+fixed : potential segfault on very large files (many gigabytes). Reported by Chip Turner.
+fixed : CLI displays system error message when destination file cannot be created (#231). Reported by Chip Turner.
+
+v0.7.1
+fixed : ZBUFF_compressEnd() called multiple times with too small `dst` buffer, reported by Christophe Chevalier
+fixed : dictBuilder fails if first sample is too small, reported by Руслан Ковалёв
+fixed : corruption issue, reported by cj
+modified : checksum enabled by default in command line mode
+
+v0.7.0
+New : Support for directory compression, using `-r`, thanks to Przemyslaw Skibinski
+New : Command `--rm`, to remove source file after successful de/compression
+New : Visual build scripts, by Christophe Chevalier
+New : Support for Sparse File-systems (do not use space for zero-filled sectors)
+New : Frame checksum support
+New : Support pass-through mode (when using `-df`)
+API : more efficient Dictionary API : `ZSTD_compress_usingCDict()`, `ZSTD_decompress_usingDDict()`
+API : create dictionary files from custom content, by Giuseppe Ottaviano
+API : support for custom malloc/free functions
+New : controllable Dictionary ID
+New : Support for skippable frames
+
+v0.6.1
+New : zlib wrapper API, thanks to Przemyslaw Skibinski
+New : Ability to compile compressor / decompressor separately
+Changed : new lib directory structure
+Fixed : Legacy codec v0.5 compatible with dictionary decompression
+Fixed : Decoder corruption error (#173)
+Fixed : null-string roundtrip (#176)
+New : benchmark mode can select directory as input
+Experimental : midipix support, VMS support
+
+v0.6.0
+Stronger high compression modes, thanks to Przemyslaw Skibinski
+API : ZSTD_getFrameParams() provides size of decompressed content
+New : highest compression modes require `--ultra` command to fully unleash their capacity
+Fixed : zstd cli return error code > 0 and removes dst file artifact when decompression fails, thanks to Chip Turner
+
+v0.5.1
+New : Optimal parsing => Very high compression modes, thanks to Przemyslaw Skibinski
+Changed : Dictionary builder integrated into libzstd and zstd cli
+Changed (!) : zstd cli now uses "multiple input files" as default mode. See `zstd -h`.
+Fix : high compression modes for big-endian platforms
+New : zstd cli : `-t` | `--test` command
+
+v0.5.0
+New : dictionary builder utility
+Changed : streaming & dictionary API
+Improved : better compression of small data
+
+v0.4.7
+Improved : small compression speed improvement in HC mode
+Changed : `zstd_decompress.c` has ZSTD_LEGACY_SUPPORT to 0 by default
+fix : bt search bug
+
+v0.4.6
+fix : fast compression mode on Windows
+New : cmake configuration file, thanks to Artyom Dymchenko
+Improved : high compression mode on repetitive data
+New : block-level API
+New : ZSTD_duplicateCCtx()
+
+v0.4.5
+new : -m/--multiple : compress/decompress multiple files
+
+v0.4.4
+Fixed : high compression modes for Windows 32 bits
+new : external dictionary API extended to buffered mode and accessible through command line
+new : windows DLL project, thanks to Christophe Chevalier
+
+v0.4.3 :
+new : external dictionary API
+new : zstd-frugal
+
+v0.4.2 :
+Generic minor improvements for small blocks
+Fixed : big-endian compatibility, by Peter Harris (#85)
+
+v0.4.1
+Fixed : ZSTD_LEGACY_SUPPORT=0 build mode (reported by Luben)
+removed `zstd.c`
+
+v0.4.0
+Command line utility compatible with high compression levels
+Removed zstdhc => merged into zstd
+Added : ZBUFF API (see zstd_buffered.h)
+Rolling buffer support
+
+v0.3.6
+small blocks params
+
+v0.3.5
+minor generic compression improvements
+
+v0.3.4
+Faster fast cLevels
+
+v0.3.3
+Small compression ratio improvement
+
+v0.3.2
+Fixed Visual Studio
+
+v0.3.1 :
+Small compression ratio improvement
+
+v0.3
+HC mode : compression levels 2-26
+
+v0.2.2
+Fix : Visual Studio 2013 & 2015 release compilation, by Christophe Chevalier
+
+v0.2.1
+Fix : Read errors, advanced fuzzer tests, by Hanno Böck
+
+v0.2.0
+**Breaking format change**
+Faster decompression speed
+Can still decode v0.1 format
+
+v0.1.3
+fix uninitialization warning, reported by Evan Nemerson
+
+v0.1.2
+frame concatenation support
+
+v0.1.1
+fix compression bug
+detects write-flush errors
+
+v0.1.0
+first release
diff --git a/contrib/zstd/CMakeLists.txt b/contrib/zstd/CMakeLists.txt
index 120e179ba..4f3f0e516 100644
--- a/contrib/zstd/CMakeLists.txt
+++ b/contrib/zstd/CMakeLists.txt
@@ -1,23 +1,27 @@
 SET(ZSTDSRC
-		cover.c
+		debug.c
 		divsufsort.c
 		entropy_common.c
 		error_private.c
 		fse_compress.c
 		fse_decompress.c
+		hist.c
 		huf_compress.c
 		huf_decompress.c
 		pool.c
 		threading.c
-		zdict.c
 		zstd_common.c
 		zstd_compress.c
+		zstd_compress_literals.c
+		zstd_compress_sequences.c
+		zstd_compress_superblock.c
+		zstd_ddict.c
 		zstd_decompress.c
+		zstd_decompress_block.c
 		zstd_double_fast.c
 		zstd_fast.c
 		zstd_lazy.c
 		zstd_ldm.c
-		zstdmt_compress.c
 		zstd_opt.c)
 
 ADD_LIBRARY(rspamd-zstd STATIC ${ZSTDSRC})
diff --git a/contrib/zstd/bitstream.h b/contrib/zstd/bitstream.h
index 2094823fe..37b99c01e 100644
--- a/contrib/zstd/bitstream.h
+++ b/contrib/zstd/bitstream.h
@@ -1,36 +1,15 @@
 /* ******************************************************************
-   bitstream
-   Part of FSE library
-   header file (to include)
-   Copyright (C) 2013-2017, Yann Collet.
-
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
-
-   Redistribution and use in source and binary forms, with or without
-   modification, are permitted provided that the following conditions are
-   met:
-
-       * Redistributions of source code must retain the above copyright
-   notice, this list of conditions and the following disclaimer.
-       * Redistributions in binary form must reproduce the above
-   copyright notice, this list of conditions and the following disclaimer
-   in the documentation and/or other materials provided with the
-   distribution.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-   You can contact the author at :
-   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ * bitstream
+ * Part of FSE library
+ * Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
+ *
+ * You can contact the author at :
+ * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
 ****************************************************************** */
 #ifndef BITSTREAM_H_MODULE
 #define BITSTREAM_H_MODULE
@@ -49,26 +28,18 @@ extern "C" {
 *  Dependencies
 ******************************************/
 #include "mem.h"            /* unaligned access routines */
+#include "compiler.h"       /* UNLIKELY() */
+#include "debug.h"          /* assert(), DEBUGLOG(), RAWLOG() */
 #include "error_private.h"  /* error codes and messages */
 
 
-/*-*************************************
-*  Debug
-***************************************/
-#if defined(BIT_DEBUG) && (BIT_DEBUG>=1)
-#  include <assert.h>
-#else
-#  ifndef assert
-#    define assert(condition) ((void)0)
-#  endif
-#endif
-
-
 /*=========================================
 *  Target specific
 =========================================*/
 #if defined(__BMI__) && defined(__GNUC__)
 #  include <immintrin.h>   /* support for bextr (experimental) */
+#elif defined(__ICCARM__)
+#  include <intrinsics.h>
 #endif
 
 #define STREAM_ACCUMULATOR_MIN_32  25
@@ -83,8 +54,7 @@ extern "C" {
  * A critical property of these streams is that they encode and decode in **reverse** direction.
  * So the first bit sequence you add will be the last to be read, like a LIFO stack.
  */
-typedef struct
-{
+typedef struct {
     size_t bitContainer;
     unsigned bitPos;
     char*  startPtr;
@@ -118,8 +88,7 @@ MEM_STATIC size_t BIT_closeCStream(BIT_CStream_t* bitC);
 /*-********************************************
 *  bitStream decoding API (read backward)
 **********************************************/
-typedef struct
-{
+typedef struct {
     size_t   bitContainer;
     unsigned bitsConsumed;
     const char* ptr;
@@ -167,16 +136,17 @@ MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits);
 /*-**************************************************************
 *  Internal functions
 ****************************************************************/
-MEM_STATIC unsigned BIT_highbit32 (register U32 val)
+MEM_STATIC unsigned BIT_highbit32 (U32 val)
 {
     assert(val != 0);
     {
 #   if defined(_MSC_VER)   /* Visual */
         unsigned long r=0;
-        _BitScanReverse ( &r, val );
-        return (unsigned) r;
+        return _BitScanReverse ( &r, val ) ? (unsigned)r : 0;
 #   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* Use GCC Intrinsic */
-        return 31 - __builtin_clz (val);
+        return __builtin_clz (val) ^ 31;
+#   elif defined(__ICCARM__)    /* IAR Intrinsic */
+        return 31 - __CLZ(val);
 #   else   /* Software version */
         static const unsigned DeBruijnClz[32] = { 0,  9,  1, 10, 13, 21,  2, 29,
                                                  11, 14, 16, 18, 22, 25,  3, 30,
@@ -236,7 +206,8 @@ MEM_STATIC void BIT_addBits(BIT_CStream_t* bitC,
 }
 
 /*! BIT_addBitsFast() :
- *  works only if `value` is _clean_, meaning all high bits above nbBits are 0 */
+ *  works only if `value` is _clean_,
+ *  meaning all high bits above nbBits are 0 */
 MEM_STATIC void BIT_addBitsFast(BIT_CStream_t* bitC,
                                 size_t value, unsigned nbBits)
 {
@@ -253,9 +224,9 @@ MEM_STATIC void BIT_flushBitsFast(BIT_CStream_t* bitC)
 {
     size_t const nbBytes = bitC->bitPos >> 3;
     assert(bitC->bitPos < sizeof(bitC->bitContainer) * 8);
+    assert(bitC->ptr <= bitC->endPtr);
     MEM_writeLEST(bitC->ptr, bitC->bitContainer);
     bitC->ptr += nbBytes;
-    assert(bitC->ptr <= bitC->endPtr);
     bitC->bitPos &= 7;
     bitC->bitContainer >>= nbBytes*8;
 }
@@ -269,6 +240,7 @@ MEM_STATIC void BIT_flushBits(BIT_CStream_t* bitC)
 {
     size_t const nbBytes = bitC->bitPos >> 3;
     assert(bitC->bitPos < sizeof(bitC->bitContainer) * 8);
+    assert(bitC->ptr <= bitC->endPtr);
     MEM_writeLEST(bitC->ptr, bitC->bitContainer);
     bitC->ptr += nbBytes;
     if (bitC->ptr > bitC->endPtr) bitC->ptr = bitC->endPtr;
@@ -352,17 +324,10 @@ MEM_STATIC size_t BIT_getUpperBits(size_t bitContainer, U32 const start)
 
 MEM_STATIC size_t BIT_getMiddleBits(size_t bitContainer, U32 const start, U32 const nbBits)
 {
-#if defined(__BMI__) && defined(__GNUC__) && __GNUC__*1000+__GNUC_MINOR__ >= 4008  /* experimental */
-#  if defined(__x86_64__)
-    if (sizeof(bitContainer)==8)
-        return _bextr_u64(bitContainer, start, nbBits);
-    else
-#  endif
-        return _bextr_u32(bitContainer, start, nbBits);
-#else
+    U32 const regMask = sizeof(bitContainer)*8 - 1;
+    /* if start > regMask, bitstream is corrupted, and result is undefined */
     assert(nbBits < BIT_MASK_SIZE);
-    return (bitContainer >> start) & BIT_mask[nbBits];
-#endif
+    return (bitContainer >> (start & regMask)) & BIT_mask[nbBits];
 }
 
 MEM_STATIC size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
@@ -379,9 +344,13 @@ MEM_STATIC size_t BIT_getLowerBits(size_t bitContainer, U32 const nbBits)
  * @return : value extracted */
 MEM_STATIC size_t BIT_lookBits(const BIT_DStream_t* bitD, U32 nbBits)
 {
-#if defined(__BMI__) && defined(__GNUC__)   /* experimental; fails if bitD->bitsConsumed + nbBits > sizeof(bitD->bitContainer)*8 */
+    /* arbitrate between double-shift and shift+mask */
+#if 1
+    /* if bitD->bitsConsumed + nbBits > sizeof(bitD->bitContainer)*8,
+     * bitstream is likely corrupted, and result is undefined */
     return BIT_getMiddleBits(bitD->bitContainer, (sizeof(bitD->bitContainer)*8) - bitD->bitsConsumed - nbBits, nbBits);
 #else
+    /* this code path is slower on my os-x laptop */
     U32 const regMask = sizeof(bitD->bitContainer)*8 - 1;
     return ((bitD->bitContainer << (bitD->bitsConsumed & regMask)) >> 1) >> ((regMask-nbBits) & regMask);
 #endif
@@ -405,7 +374,7 @@ MEM_STATIC void BIT_skipBits(BIT_DStream_t* bitD, U32 nbBits)
  *  Read (consume) next n bits from local register and update.
  *  Pay attention to not read more than nbBits contained into local register.
  * @return : extracted value. */
-MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
+MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, unsigned nbBits)
 {
     size_t const value = BIT_lookBits(bitD, nbBits);
     BIT_skipBits(bitD, nbBits);
@@ -414,7 +383,7 @@ MEM_STATIC size_t BIT_readBits(BIT_DStream_t* bitD, U32 nbBits)
 
 /*! BIT_readBitsFast() :
  *  unsafe version; only works only if nbBits >= 1 */
-MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
+MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, unsigned nbBits)
 {
     size_t const value = BIT_lookBitsFast(bitD, nbBits);
     assert(nbBits >= 1);
@@ -422,21 +391,35 @@ MEM_STATIC size_t BIT_readBitsFast(BIT_DStream_t* bitD, U32 nbBits)
     return value;
 }
 
+/*! BIT_reloadDStreamFast() :
+ *  Similar to BIT_reloadDStream(), but with two differences:
+ *  1. bitsConsumed <= sizeof(bitD->bitContainer)*8 must hold!
+ *  2. Returns BIT_DStream_overflow when bitD->ptr < bitD->limitPtr, at this
+ *     point you must use BIT_reloadDStream() to reload.
+ */
+MEM_STATIC BIT_DStream_status BIT_reloadDStreamFast(BIT_DStream_t* bitD)
+{
+    if (UNLIKELY(bitD->ptr < bitD->limitPtr))
+        return BIT_DStream_overflow;
+    assert(bitD->bitsConsumed <= sizeof(bitD->bitContainer)*8);
+    bitD->ptr -= bitD->bitsConsumed >> 3;
+    bitD->bitsConsumed &= 7;
+    bitD->bitContainer = MEM_readLEST(bitD->ptr);
+    return BIT_DStream_unfinished;
+}
+
 /*! BIT_reloadDStream() :
  *  Refill `bitD` from buffer previously set in BIT_initDStream() .
  *  This function is safe, it guarantees it will not read beyond src buffer.
  * @return : status of `BIT_DStream_t` internal register.
- *           when status == BIT_DStream_unfinished, internal register is filled with at least 25 or 57 bits */
+ *           when status == BIT_DStream_unfinished, internal register is filled with at least 25 or 57 bits */
 MEM_STATIC BIT_DStream_status BIT_reloadDStream(BIT_DStream_t* bitD)
 {
     if (bitD->bitsConsumed > (sizeof(bitD->bitContainer)*8))  /* overflow detected, like end of stream */
         return BIT_DStream_overflow;
 
     if (bitD->ptr >= bitD->limitPtr) {
-        bitD->ptr -= bitD->bitsConsumed >> 3;
-        bitD->bitsConsumed &= 7;
-        bitD->bitContainer = MEM_readLEST(bitD->ptr);
-        return BIT_DStream_unfinished;
+        return BIT_reloadDStreamFast(bitD);
     }
     if (bitD->ptr == bitD->start) {
         if (bitD->bitsConsumed < sizeof(bitD->bitContainer)*8) return BIT_DStream_endOfBuffer;
diff --git a/contrib/zstd/compiler.h b/contrib/zstd/compiler.h
index 3a7553c38..95e948352 100644
--- a/contrib/zstd/compiler.h
+++ b/contrib/zstd/compiler.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -15,13 +15,15 @@
 *  Compiler specifics
 *********************************************************/
 /* force inlining */
-#if defined (__GNUC__) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
+
+#if !defined(ZSTD_NO_INLINE)
+#if (defined(__GNUC__) && !defined(__STRICT_ANSI__)) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* C99 */
 #  define INLINE_KEYWORD inline
 #else
 #  define INLINE_KEYWORD
 #endif
 
-#if defined(__GNUC__)
+#if defined(__GNUC__) || defined(__ICCARM__)
 #  define FORCE_INLINE_ATTR __attribute__((always_inline))
 #elif defined(_MSC_VER)
 #  define FORCE_INLINE_ATTR __forceinline
@@ -29,9 +31,16 @@
 #  define FORCE_INLINE_ATTR
 #endif
 
+#else
+
+#define INLINE_KEYWORD
+#define FORCE_INLINE_ATTR
+
+#endif
+
 /**
  * FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant
- * parameters. They must be inlined for the compiler to elimininate the constant
+ * parameters. They must be inlined for the compiler to eliminate the constant
  * branches.
  */
 #define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR
@@ -52,25 +61,105 @@
 #  define HINT_INLINE static INLINE_KEYWORD FORCE_INLINE_ATTR
 #endif
 
+/* UNUSED_ATTR tells the compiler it is okay if the function is unused. */
+#if defined(__GNUC__)
+#  define UNUSED_ATTR __attribute__((unused))
+#else
+#  define UNUSED_ATTR
+#endif
+
 /* force no inlining */
 #ifdef _MSC_VER
 #  define FORCE_NOINLINE static __declspec(noinline)
 #else
-#  ifdef __GNUC__
+#  if defined(__GNUC__) || defined(__ICCARM__)
 #    define FORCE_NOINLINE static __attribute__((__noinline__))
 #  else
 #    define FORCE_NOINLINE static
 #  endif
 #endif
 
-/* prefetch */
-#if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86))  /* _mm_prefetch() is not defined outside of x86/x64 */
-#  include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
-#  define PREFETCH(ptr)   _mm_prefetch((const char*)ptr, _MM_HINT_T0)
-#elif defined(__GNUC__)
-#  define PREFETCH(ptr)   __builtin_prefetch(ptr, 0, 0)
+/* target attribute */
+#ifndef __has_attribute
+  #define __has_attribute(x) 0  /* Compatibility with non-clang compilers. */
+#endif
+#if defined(__GNUC__) || defined(__ICCARM__)
+#  define TARGET_ATTRIBUTE(target) __attribute__((__target__(target)))
+#else
+#  define TARGET_ATTRIBUTE(target)
+#endif
+
+/* Enable runtime BMI2 dispatch based on the CPU.
+ * Enabled for clang & gcc >=4.8 on x86 when BMI2 isn't enabled by default.
+ */
+#ifndef DYNAMIC_BMI2
+  #if ((defined(__clang__) && __has_attribute(__target__)) \
+      || (defined(__GNUC__) \
+          && (__GNUC__ >= 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)))) \
+      && (defined(__x86_64__) || defined(_M_X86)) \
+      && !defined(__BMI2__)
+  #  define DYNAMIC_BMI2 1
+  #else
+  #  define DYNAMIC_BMI2 0
+  #endif
+#endif
+
+/* prefetch
+ * can be disabled, by declaring NO_PREFETCH build macro */
+#if defined(NO_PREFETCH)
+#  define PREFETCH_L1(ptr)  (void)(ptr)  /* disabled */
+#  define PREFETCH_L2(ptr)  (void)(ptr)  /* disabled */
+#else
+#  if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86))  /* _mm_prefetch() is not defined outside of x86/x64 */
+#    include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
+#    define PREFETCH_L1(ptr)  _mm_prefetch((const char*)(ptr), _MM_HINT_T0)
+#    define PREFETCH_L2(ptr)  _mm_prefetch((const char*)(ptr), _MM_HINT_T1)
+#    elif defined(__aarch64__)
+#     define PREFETCH_L1(ptr)  __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr)))
+#     define PREFETCH_L2(ptr)  __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr)))
+#  elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
+#    define PREFETCH_L1(ptr)  __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
+#    define PREFETCH_L2(ptr)  __builtin_prefetch((ptr), 0 /* rw==read */, 2 /* locality */)
+#  else
+#    define PREFETCH_L1(ptr) (void)(ptr)  /* disabled */
+#    define PREFETCH_L2(ptr) (void)(ptr)  /* disabled */
+#  endif
+#endif  /* NO_PREFETCH */
+
+#define CACHELINE_SIZE 64
+
+#define PREFETCH_AREA(p, s)  {            \
+    const char* const _ptr = (const char*)(p);  \
+    size_t const _size = (size_t)(s);     \
+    size_t _pos;                          \
+    for (_pos=0; _pos<_size; _pos+=CACHELINE_SIZE) {  \
+        PREFETCH_L2(_ptr + _pos);         \
+    }                                     \
+}
+
+/* vectorization
+ * older GCC (pre gcc-4.3 picked as the cutoff) uses a different syntax */
+#if !defined(__INTEL_COMPILER) && !defined(__clang__) && defined(__GNUC__)
+#  if (__GNUC__ == 4 && __GNUC_MINOR__ > 3) || (__GNUC__ >= 5)
+#    define DONT_VECTORIZE __attribute__((optimize("no-tree-vectorize")))
+#  else
+#    define DONT_VECTORIZE _Pragma("GCC optimize(\"no-tree-vectorize\")")
+#  endif
+#else
+#  define DONT_VECTORIZE
+#endif
+
+/* Tell the compiler that a branch is likely or unlikely.
+ * Only use these macros if it causes the compiler to generate better code.
+ * If you can remove a LIKELY/UNLIKELY annotation without speed changes in gcc
+ * and clang, please do.
+ */
+#if defined(__GNUC__)
+#define LIKELY(x) (__builtin_expect((x), 1))
+#define UNLIKELY(x) (__builtin_expect((x), 0))
 #else
-#  define PREFETCH(ptr)   /* disabled */
+#define LIKELY(x) (x)
+#define UNLIKELY(x) (x)
 #endif
 
 /* disable warnings */
diff --git a/contrib/zstd/cover.c b/contrib/zstd/cover.c
deleted file mode 100644
index ec7271dba..000000000
--- a/contrib/zstd/cover.c
+++ /dev/null
@@ -1,1045 +0,0 @@
-/*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
- * All rights reserved.
- *
- * This source code is licensed under both the BSD-style license (found in the
- * LICENSE file in the root directory of this source tree) and the GPLv2 (found
- * in the COPYING file in the root directory of this source tree).
- * You may select, at your option, one of the above-listed licenses.
- */
-
-/* *****************************************************************************
- * Constructs a dictionary using a heuristic based on the following paper:
- *
- * Liao, Petri, Moffat, Wirth
- * Effective Construction of Relative Lempel-Ziv Dictionaries
- * Published in WWW 2016.
- *
- * Adapted from code originally written by @ot (Giuseppe Ottaviano).
- ******************************************************************************/
-
-/*-*************************************
-*  Dependencies
-***************************************/
-#include <stdio.h>  /* fprintf */
-#include <stdlib.h> /* malloc, free, qsort */
-#include <string.h> /* memset */
-#include <time.h>   /* clock */
-
-#include "mem.h" /* read */
-#include "pool.h"
-#include "threading.h"
-#include "zstd_internal.h" /* includes zstd.h */
-#ifndef ZDICT_STATIC_LINKING_ONLY
-#define ZDICT_STATIC_LINKING_ONLY
-#endif
-#include "zdict.h"
-
-/*-*************************************
-*  Constants
-***************************************/
-#define COVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((U32)-1) : ((U32)1 GB))
-
-/*-*************************************
-*  Console display
-***************************************/
-static int g_displayLevel = 2;
-#define DISPLAY(...)                                                           \
-  {                                                                            \
-    fprintf(stderr, __VA_ARGS__);                                              \
-    fflush(stderr);                                                            \
-  }
-#define LOCALDISPLAYLEVEL(displayLevel, l, ...)                                \
-  if (displayLevel >= l) {                                                     \
-    DISPLAY(__VA_ARGS__);                                                      \
-  } /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
-#define DISPLAYLEVEL(l, ...) LOCALDISPLAYLEVEL(g_displayLevel, l, __VA_ARGS__)
-
-#define LOCALDISPLAYUPDATE(displayLevel, l, ...)                               \
-  if (displayLevel >= l) {                                                     \
-    if ((clock() - g_time > refreshRate) || (displayLevel >= 4)) {             \
-      g_time = clock();                                                        \
-      DISPLAY(__VA_ARGS__);                                                    \
-    }                                                                          \
-  }
-#define DISPLAYUPDATE(l, ...) LOCALDISPLAYUPDATE(g_displayLevel, l, __VA_ARGS__)
-static const clock_t refreshRate = CLOCKS_PER_SEC * 15 / 100;
-static clock_t g_time = 0;
-
-/*-*************************************
-* Hash table
-***************************************
-* A small specialized hash map for storing activeDmers.
-* The map does not resize, so if it becomes full it will loop forever.
-* Thus, the map must be large enough to store every value.
-* The map implements linear probing and keeps its load less than 0.5.
-*/
-
-#define MAP_EMPTY_VALUE ((U32)-1)
-typedef struct COVER_map_pair_t_s {
-  U32 key;
-  U32 value;
-} COVER_map_pair_t;
-
-typedef struct COVER_map_s {
-  COVER_map_pair_t *data;
-  U32 sizeLog;
-  U32 size;
-  U32 sizeMask;
-} COVER_map_t;
-
-/**
- * Clear the map.
- */
-static void COVER_map_clear(COVER_map_t *map) {
-  memset(map->data, MAP_EMPTY_VALUE, map->size * sizeof(COVER_map_pair_t));
-}
-
-/**
- * Initializes a map of the given size.
- * Returns 1 on success and 0 on failure.
- * The map must be destroyed with COVER_map_destroy().
- * The map is only guaranteed to be large enough to hold size elements.
- */
-static int COVER_map_init(COVER_map_t *map, U32 size) {
-  map->sizeLog = ZSTD_highbit32(size) + 2;
-  map->size = (U32)1 << map->sizeLog;
-  map->sizeMask = map->size - 1;
-  map->data = (COVER_map_pair_t *)malloc(map->size * sizeof(COVER_map_pair_t));
-  if (!map->data) {
-    map->sizeLog = 0;
-    map->size = 0;
-    return 0;
-  }
-  COVER_map_clear(map);
-  return 1;
-}
-
-/**
- * Internal hash function
- */
-static const U32 prime4bytes = 2654435761U;
-static U32 COVER_map_hash(COVER_map_t *map, U32 key) {
-  return (key * prime4bytes) >> (32 - map->sizeLog);
-}
-
-/**
- * Helper function that returns the index that a key should be placed into.
- */
-static U32 COVER_map_index(COVER_map_t *map, U32 key) {
-  const U32 hash = COVER_map_hash(map, key);
-  U32 i;
-  for (i = hash;; i = (i + 1) & map->sizeMask) {
-    COVER_map_pair_t *pos = &map->data[i];
-    if (pos->value == MAP_EMPTY_VALUE) {
-      return i;
-    }
-    if (pos->key == key) {
-      return i;
-    }
-  }
-}
-
-/**
- * Returns the pointer to the value for key.
- * If key is not in the map, it is inserted and the value is set to 0.
- * The map must not be full.
- */
-static U32 *COVER_map_at(COVER_map_t *map, U32 key) {
-  COVER_map_pair_t *pos = &map->data[COVER_map_index(map, key)];
-  if (pos->value == MAP_EMPTY_VALUE) {
-    pos->key = key;
-    pos->value = 0;
-  }
-  return &pos->value;
-}
-
-/**
- * Deletes key from the map if present.
- */
-static void COVER_map_remove(COVER_map_t *map, U32 key) {
-  U32 i = COVER_map_index(map, key);
-  COVER_map_pair_t *del = &map->data[i];
-  U32 shift = 1;
-  if (del->value == MAP_EMPTY_VALUE) {
-    return;
-  }
-  for (i = (i + 1) & map->sizeMask;; i = (i + 1) & map->sizeMask) {
-    COVER_map_pair_t *const pos = &map->data[i];
-    /* If the position is empty we are done */
-    if (pos->value == MAP_EMPTY_VALUE) {
-      del->value = MAP_EMPTY_VALUE;
-      return;
-    }
-    /* If pos can be moved to del do so */
-    if (((i - COVER_map_hash(map, pos->key)) & map->sizeMask) >= shift) {
-      del->key = pos->key;
-      del->value = pos->value;
-      del = pos;
-      shift = 1;
-    } else {
-      ++shift;
-    }
-  }
-}
-
-/**
- * Destroyes a map that is inited with COVER_map_init().
- */
-static void COVER_map_destroy(COVER_map_t *map) {
-  if (map->data) {
-    free(map->data);
-  }
-  map->data = NULL;
-  map->size = 0;
-}
-
-/*-*************************************
-* Context
-***************************************/
-
-typedef struct {
-  const BYTE *samples;
-  size_t *offsets;
-  const size_t *samplesSizes;
-  size_t nbSamples;
-  U32 *suffix;
-  size_t suffixSize;
-  U32 *freqs;
-  U32 *dmerAt;
-  unsigned d;
-} COVER_ctx_t;
-
-/* We need a global context for qsort... */
-static COVER_ctx_t *g_ctx = NULL;
-
-/*-*************************************
-*  Helper functions
-***************************************/
-
-/**
- * Returns the sum of the sample sizes.
- */
-static size_t COVER_sum(const size_t *samplesSizes, unsigned nbSamples) {
-  size_t sum = 0;
-  size_t i;
-  for (i = 0; i < nbSamples; ++i) {
-    sum += samplesSizes[i];
-  }
-  return sum;
-}
-
-/**
- * Returns -1 if the dmer at lp is less than the dmer at rp.
- * Return 0 if the dmers at lp and rp are equal.
- * Returns 1 if the dmer at lp is greater than the dmer at rp.
- */
-static int COVER_cmp(COVER_ctx_t *ctx, const void *lp, const void *rp) {
-  U32 const lhs = *(U32 const *)lp;
-  U32 const rhs = *(U32 const *)rp;
-  return memcmp(ctx->samples + lhs, ctx->samples + rhs, ctx->d);
-}
-/**
- * Faster version for d <= 8.
- */
-static int COVER_cmp8(COVER_ctx_t *ctx, const void *lp, const void *rp) {
-  U64 const mask = (ctx->d == 8) ? (U64)-1 : (((U64)1 << (8 * ctx->d)) - 1);
-  U64 const lhs = MEM_readLE64(ctx->samples + *(U32 const *)lp) & mask;
-  U64 const rhs = MEM_readLE64(ctx->samples + *(U32 const *)rp) & mask;
-  if (lhs < rhs) {
-    return -1;
-  }
-  return (lhs > rhs);
-}
-
-/**
- * Same as COVER_cmp() except ties are broken by pointer value
- * NOTE: g_ctx must be set to call this function.  A global is required because
- * qsort doesn't take an opaque pointer.
- */
-static int COVER_strict_cmp(const void *lp, const void *rp) {
-  int result = COVER_cmp(g_ctx, lp, rp);
-  if (result == 0) {
-    result = lp < rp ? -1 : 1;
-  }
-  return result;
-}
-/**
- * Faster version for d <= 8.
- */
-static int COVER_strict_cmp8(const void *lp, const void *rp) {
-  int result = COVER_cmp8(g_ctx, lp, rp);
-  if (result == 0) {
-    result = lp < rp ? -1 : 1;
-  }
-  return result;
-}
-
-/**
- * Returns the first pointer in [first, last) whose element does not compare
- * less than value.  If no such element exists it returns last.
- */
-static const size_t *COVER_lower_bound(const size_t *first, const size_t *last,
-                                       size_t value) {
-  size_t count = last - first;
-  while (count != 0) {
-    size_t step = count / 2;
-    const size_t *ptr = first;
-    ptr += step;
-    if (*ptr < value) {
-      first = ++ptr;
-      count -= step + 1;
-    } else {
-      count = step;
-    }
-  }
-  return first;
-}
-
-/**
- * Generic groupBy function.
- * Groups an array sorted by cmp into groups with equivalent values.
- * Calls grp for each group.
- */
-static void
-COVER_groupBy(const void *data, size_t count, size_t size, COVER_ctx_t *ctx,
-              int (*cmp)(COVER_ctx_t *, const void *, const void *),
-              void (*grp)(COVER_ctx_t *, const void *, const void *)) {
-  const BYTE *ptr = (const BYTE *)data;
-  size_t num = 0;
-  while (num < count) {
-    const BYTE *grpEnd = ptr + size;
-    ++num;
-    while (num < count && cmp(ctx, ptr, grpEnd) == 0) {
-      grpEnd += size;
-      ++num;
-    }
-    grp(ctx, ptr, grpEnd);
-    ptr = grpEnd;
-  }
-}
-
-/*-*************************************
-*  Cover functions
-***************************************/
-
-/**
- * Called on each group of positions with the same dmer.
- * Counts the frequency of each dmer and saves it in the suffix array.
- * Fills `ctx->dmerAt`.
- */
-static void COVER_group(COVER_ctx_t *ctx, const void *group,
-                        const void *groupEnd) {
-  /* The group consists of all the positions with the same first d bytes. */
-  const U32 *grpPtr = (const U32 *)group;
-  const U32 *grpEnd = (const U32 *)groupEnd;
-  /* The dmerId is how we will reference this dmer.
-   * This allows us to map the whole dmer space to a much smaller space, the
-   * size of the suffix array.
-   */
-  const U32 dmerId = (U32)(grpPtr - ctx->suffix);
-  /* Count the number of samples this dmer shows up in */
-  U32 freq = 0;
-  /* Details */
-  const size_t *curOffsetPtr = ctx->offsets;
-  const size_t *offsetsEnd = ctx->offsets + ctx->nbSamples;
-  /* Once *grpPtr >= curSampleEnd this occurrence of the dmer is in a
-   * different sample than the last.
-   */
-  size_t curSampleEnd = ctx->offsets[0];
-  for (; grpPtr != grpEnd; ++grpPtr) {
-    /* Save the dmerId for this position so we can get back to it. */
-    ctx->dmerAt[*grpPtr] = dmerId;
-    /* Dictionaries only help for the first reference to the dmer.
-     * After that zstd can reference the match from the previous reference.
-     * So only count each dmer once for each sample it is in.
-     */
-    if (*grpPtr < curSampleEnd) {
-      continue;
-    }
-    freq += 1;
-    /* Binary search to find the end of the sample *grpPtr is in.
-     * In the common case that grpPtr + 1 == grpEnd we can skip the binary
-     * search because the loop is over.
-     */
-    if (grpPtr + 1 != grpEnd) {
-      const size_t *sampleEndPtr =
-          COVER_lower_bound(curOffsetPtr, offsetsEnd, *grpPtr);
-      curSampleEnd = *sampleEndPtr;
-      curOffsetPtr = sampleEndPtr + 1;
-    }
-  }
-  /* At this point we are never going to look at this segment of the suffix
-   * array again.  We take advantage of this fact to save memory.
-   * We store the frequency of the dmer in the first position of the group,
-   * which is dmerId.
-   */
-  ctx->suffix[dmerId] = freq;
-}
-
-/**
- * A segment is a range in the source as well as the score of the segment.
- */
-typedef struct {
-  U32 begin;
-  U32 end;
-  U32 score;
-} COVER_segment_t;
-
-/**
- * Selects the best segment in an epoch.
- * Segments of are scored according to the function:
- *
- * Let F(d) be the frequency of dmer d.
- * Let S_i be the dmer at position i of segment S which has length k.
- *
- *     Score(S) = F(S_1) + F(S_2) + ... + F(S_{k-d+1})
- *
- * Once the dmer d is in the dictionay we set F(d) = 0.
- */
-static COVER_segment_t COVER_selectSegment(const COVER_ctx_t *ctx, U32 *freqs,
-                                           COVER_map_t *activeDmers, U32 begin,
-                                           U32 end,
-                                           ZDICT_cover_params_t parameters) {
-  /* Constants */
-  const U32 k = parameters.k;
-  const U32 d = parameters.d;
-  const U32 dmersInK = k - d + 1;
-  /* Try each segment (activeSegment) and save the best (bestSegment) */
-  COVER_segment_t bestSegment = {0, 0, 0};
-  COVER_segment_t activeSegment;
-  /* Reset the activeDmers in the segment */
-  COVER_map_clear(activeDmers);
-  /* The activeSegment starts at the beginning of the epoch. */
-  activeSegment.begin = begin;
-  activeSegment.end = begin;
-  activeSegment.score = 0;
-  /* Slide the activeSegment through the whole epoch.
-   * Save the best segment in bestSegment.
-   */
-  while (activeSegment.end < end) {
-    /* The dmerId for the dmer at the next position */
-    U32 newDmer = ctx->dmerAt[activeSegment.end];
-    /* The entry in activeDmers for this dmerId */
-    U32 *newDmerOcc = COVER_map_at(activeDmers, newDmer);
-    /* If the dmer isn't already present in the segment add its score. */
-    if (*newDmerOcc == 0) {
-      /* The paper suggest using the L-0.5 norm, but experiments show that it
-       * doesn't help.
-       */
-      activeSegment.score += freqs[newDmer];
-    }
-    /* Add the dmer to the segment */
-    activeSegment.end += 1;
-    *newDmerOcc += 1;
-
-    /* If the window is now too large, drop the first position */
-    if (activeSegment.end - activeSegment.begin == dmersInK + 1) {
-      U32 delDmer = ctx->dmerAt[activeSegment.begin];
-      U32 *delDmerOcc = COVER_map_at(activeDmers, delDmer);
-      activeSegment.begin += 1;
-      *delDmerOcc -= 1;
-      /* If this is the last occurrence of the dmer, subtract its score */
-      if (*delDmerOcc == 0) {
-        COVER_map_remove(activeDmers, delDmer);
-        activeSegment.score -= freqs[delDmer];
-      }
-    }
-
-    /* If this segment is the best so far save it */
-    if (activeSegment.score > bestSegment.score) {
-      bestSegment = activeSegment;
-    }
-  }
-  {
-    /* Trim off the zero frequency head and tail from the segment. */
-    U32 newBegin = bestSegment.end;
-    U32 newEnd = bestSegment.begin;
-    U32 pos;
-    for (pos = bestSegment.begin; pos != bestSegment.end; ++pos) {
-      U32 freq = freqs[ctx->dmerAt[pos]];
-      if (freq != 0) {
-        newBegin = MIN(newBegin, pos);
-        newEnd = pos + 1;
-      }
-    }
-    bestSegment.begin = newBegin;
-    bestSegment.end = newEnd;
-  }
-  {
-    /* Zero out the frequency of each dmer covered by the chosen segment. */
-    U32 pos;
-    for (pos = bestSegment.begin; pos != bestSegment.end; ++pos) {
-      freqs[ctx->dmerAt[pos]] = 0;
-    }
-  }
-  return bestSegment;
-}
-
-/**
- * Check the validity of the parameters.
- * Returns non-zero if the parameters are valid and 0 otherwise.
- */
-static int COVER_checkParameters(ZDICT_cover_params_t parameters,
-                                 size_t maxDictSize) {
-  /* k and d are required parameters */
-  if (parameters.d == 0 || parameters.k == 0) {
-    return 0;
-  }
-  /* k <= maxDictSize */
-  if (parameters.k > maxDictSize) {
-    return 0;
-  }
-  /* d <= k */
-  if (parameters.d > parameters.k) {
-    return 0;
-  }
-  return 1;
-}
-
-/**
- * Clean up a context initialized with `COVER_ctx_init()`.
- */
-static void COVER_ctx_destroy(COVER_ctx_t *ctx) {
-  if (!ctx) {
-    return;
-  }
-  if (ctx->suffix) {
-    free(ctx->suffix);
-    ctx->suffix = NULL;
-  }
-  if (ctx->freqs) {
-    free(ctx->freqs);
-    ctx->freqs = NULL;
-  }
-  if (ctx->dmerAt) {
-    free(ctx->dmerAt);
-    ctx->dmerAt = NULL;
-  }
-  if (ctx->offsets) {
-    free(ctx->offsets);
-    ctx->offsets = NULL;
-  }
-}
-
-/**
- * Prepare a context for dictionary building.
- * The context is only dependent on the parameter `d` and can used multiple
- * times.
- * Returns 1 on success or zero on error.
- * The context must be destroyed with `COVER_ctx_destroy()`.
- */
-static int COVER_ctx_init(COVER_ctx_t *ctx, const void *samplesBuffer,
-                          const size_t *samplesSizes, unsigned nbSamples,
-                          unsigned d) {
-  const BYTE *const samples = (const BYTE *)samplesBuffer;
-  const size_t totalSamplesSize = COVER_sum(samplesSizes, nbSamples);
-  /* Checks */
-  if (totalSamplesSize < MAX(d, sizeof(U64)) ||
-      totalSamplesSize >= (size_t)COVER_MAX_SAMPLES_SIZE) {
-    DISPLAYLEVEL(1, "Total samples size is too large, maximum size is %u MB\n",
-                 (COVER_MAX_SAMPLES_SIZE >> 20));
-    return 0;
-  }
-  /* Zero the context */
-  memset(ctx, 0, sizeof(*ctx));
-  DISPLAYLEVEL(2, "Training on %u samples of total size %u\n", nbSamples,
-               (U32)totalSamplesSize);
-  ctx->samples = samples;
-  ctx->samplesSizes = samplesSizes;
-  ctx->nbSamples = nbSamples;
-  /* Partial suffix array */
-  ctx->suffixSize = totalSamplesSize - MAX(d, sizeof(U64)) + 1;
-  ctx->suffix = (U32 *)malloc(ctx->suffixSize * sizeof(U32));
-  /* Maps index to the dmerID */
-  ctx->dmerAt = (U32 *)malloc(ctx->suffixSize * sizeof(U32));
-  /* The offsets of each file */
-  ctx->offsets = (size_t *)malloc((nbSamples + 1) * sizeof(size_t));
-  if (!ctx->suffix || !ctx->dmerAt || !ctx->offsets) {
-    DISPLAYLEVEL(1, "Failed to allocate scratch buffers\n");
-    COVER_ctx_destroy(ctx);
-    return 0;
-  }
-  ctx->freqs = NULL;
-  ctx->d = d;
-
-  /* Fill offsets from the samlesSizes */
-  {
-    U32 i;
-    ctx->offsets[0] = 0;
-    for (i = 1; i <= nbSamples; ++i) {
-      ctx->offsets[i] = ctx->offsets[i - 1] + samplesSizes[i - 1];
-    }
-  }
-  DISPLAYLEVEL(2, "Constructing partial suffix array\n");
-  {
-    /* suffix is a partial suffix array.
-     * It only sorts suffixes by their first parameters.d bytes.
-     * The sort is stable, so each dmer group is sorted by position in input.
-     */
-    U32 i;
-    for (i = 0; i < ctx->suffixSize; ++i) {
-      ctx->suffix[i] = i;
-    }
-    /* qsort doesn't take an opaque pointer, so pass as a global */
-    g_ctx = ctx;
-    qsort(ctx->suffix, ctx->suffixSize, sizeof(U32),
-          (ctx->d <= 8 ? &COVER_strict_cmp8 : &COVER_strict_cmp));
-  }
-  DISPLAYLEVEL(2, "Computing frequencies\n");
-  /* For each dmer group (group of positions with the same first d bytes):
-   * 1. For each position we set dmerAt[position] = dmerID.  The dmerID is
-   *    (groupBeginPtr - suffix).  This allows us to go from position to
-   *    dmerID so we can look up values in freq.
-   * 2. We calculate how many samples the dmer occurs in and save it in
-   *    freqs[dmerId].
-   */
-  COVER_groupBy(ctx->suffix, ctx->suffixSize, sizeof(U32), ctx,
-                (ctx->d <= 8 ? &COVER_cmp8 : &COVER_cmp), &COVER_group);
-  ctx->freqs = ctx->suffix;
-  ctx->suffix = NULL;
-  return 1;
-}
-
-/**
- * Given the prepared context build the dictionary.
- */
-static size_t COVER_buildDictionary(const COVER_ctx_t *ctx, U32 *freqs,
-                                    COVER_map_t *activeDmers, void *dictBuffer,
-                                    size_t dictBufferCapacity,
-                                    ZDICT_cover_params_t parameters) {
-  BYTE *const dict = (BYTE *)dictBuffer;
-  size_t tail = dictBufferCapacity;
-  /* Divide the data up into epochs of equal size.
-   * We will select at least one segment from each epoch.
-   */
-  const U32 epochs = (U32)(dictBufferCapacity / parameters.k);
-  const U32 epochSize = (U32)(ctx->suffixSize / epochs);
-  size_t epoch;
-  DISPLAYLEVEL(2, "Breaking content into %u epochs of size %u\n", epochs,
-               epochSize);
-  /* Loop through the epochs until there are no more segments or the dictionary
-   * is full.
-   */
-  for (epoch = 0; tail > 0; epoch = (epoch + 1) % epochs) {
-    const U32 epochBegin = (U32)(epoch * epochSize);
-    const U32 epochEnd = epochBegin + epochSize;
-    size_t segmentSize;
-    /* Select a segment */
-    COVER_segment_t segment = COVER_selectSegment(
-        ctx, freqs, activeDmers, epochBegin, epochEnd, parameters);
-    /* If the segment covers no dmers, then we are out of content */
-    if (segment.score == 0) {
-      break;
-    }
-    /* Trim the segment if necessary and if it is too small then we are done */
-    segmentSize = MIN(segment.end - segment.begin + parameters.d - 1, tail);
-    if (segmentSize < parameters.d) {
-      break;
-    }
-    /* We fill the dictionary from the back to allow the best segments to be
-     * referenced with the smallest offsets.
-     */
-    tail -= segmentSize;
-    memcpy(dict + tail, ctx->samples + segment.begin, segmentSize);
-    DISPLAYUPDATE(
-        2, "\r%u%%       ",
-        (U32)(((dictBufferCapacity - tail) * 100) / dictBufferCapacity));
-  }
-  DISPLAYLEVEL(2, "\r%79s\r", "");
-  return tail;
-}
-
-ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
-    void *dictBuffer, size_t dictBufferCapacity, const void *samplesBuffer,
-    const size_t *samplesSizes, unsigned nbSamples,
-    ZDICT_cover_params_t parameters) {
-  BYTE *const dict = (BYTE *)dictBuffer;
-  COVER_ctx_t ctx;
-  COVER_map_t activeDmers;
-  /* Checks */
-  if (!COVER_checkParameters(parameters, dictBufferCapacity)) {
-    DISPLAYLEVEL(1, "Cover parameters incorrect\n");
-    return ERROR(GENERIC);
-  }
-  if (nbSamples == 0) {
-    DISPLAYLEVEL(1, "Cover must have at least one input file\n");
-    return ERROR(GENERIC);
-  }
-  if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
-    DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
-                 ZDICT_DICTSIZE_MIN);
-    return ERROR(dstSize_tooSmall);
-  }
-  /* Initialize global data */
-  g_displayLevel = parameters.zParams.notificationLevel;
-  /* Initialize context and activeDmers */
-  if (!COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
-                      parameters.d)) {
-    return ERROR(GENERIC);
-  }
-  if (!COVER_map_init(&activeDmers, parameters.k - parameters.d + 1)) {
-    DISPLAYLEVEL(1, "Failed to allocate dmer map: out of memory\n");
-    COVER_ctx_destroy(&ctx);
-    return ERROR(GENERIC);
-  }
-
-  DISPLAYLEVEL(2, "Building dictionary\n");
-  {
-    const size_t tail =
-        COVER_buildDictionary(&ctx, ctx.freqs, &activeDmers, dictBuffer,
-                              dictBufferCapacity, parameters);
-    const size_t dictionarySize = ZDICT_finalizeDictionary(
-        dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
-        samplesBuffer, samplesSizes, nbSamples, parameters.zParams);
-    if (!ZSTD_isError(dictionarySize)) {
-      DISPLAYLEVEL(2, "Constructed dictionary of size %u\n",
-                   (U32)dictionarySize);
-    }
-    COVER_ctx_destroy(&ctx);
-    COVER_map_destroy(&activeDmers);
-    return dictionarySize;
-  }
-}
-
-/**
- * COVER_best_t is used for two purposes:
- * 1. Synchronizing threads.
- * 2. Saving the best parameters and dictionary.
- *
- * All of the methods except COVER_best_init() are thread safe if zstd is
- * compiled with multithreaded support.
- */
-typedef struct COVER_best_s {
-  pthread_mutex_t mutex;
-  pthread_cond_t cond;
-  size_t liveJobs;
-  void *dict;
-  size_t dictSize;
-  ZDICT_cover_params_t parameters;
-  size_t compressedSize;
-} COVER_best_t;
-
-/**
- * Initialize the `COVER_best_t`.
- */
-static void COVER_best_init(COVER_best_t *best) {
-  if (best==NULL) return; /* compatible with init on NULL */
-  (void)pthread_mutex_init(&best->mutex, NULL);
-  (void)pthread_cond_init(&best->cond, NULL);
-  best->liveJobs = 0;
-  best->dict = NULL;
-  best->dictSize = 0;
-  best->compressedSize = (size_t)-1;
-  memset(&best->parameters, 0, sizeof(best->parameters));
-}
-
-/**
- * Wait until liveJobs == 0.
- */
-static void COVER_best_wait(COVER_best_t *best) {
-  if (!best) {
-    return;
-  }
-  pthread_mutex_lock(&best->mutex);
-  while (best->liveJobs != 0) {
-    pthread_cond_wait(&best->cond, &best->mutex);
-  }
-  pthread_mutex_unlock(&best->mutex);
-}
-
-/**
- * Call COVER_best_wait() and then destroy the COVER_best_t.
- */
-static void COVER_best_destroy(COVER_best_t *best) {
-  if (!best) {
-    return;
-  }
-  COVER_best_wait(best);
-  if (best->dict) {
-    free(best->dict);
-  }
-  pthread_mutex_destroy(&best->mutex);
-  pthread_cond_destroy(&best->cond);
-}
-
-/**
- * Called when a thread is about to be launched.
- * Increments liveJobs.
- */
-static void COVER_best_start(COVER_best_t *best) {
-  if (!best) {
-    return;
-  }
-  pthread_mutex_lock(&best->mutex);
-  ++best->liveJobs;
-  pthread_mutex_unlock(&best->mutex);
-}
-
-/**
- * Called when a thread finishes executing, both on error or success.
- * Decrements liveJobs and signals any waiting threads if liveJobs == 0.
- * If this dictionary is the best so far save it and its parameters.
- */
-static void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
-                              ZDICT_cover_params_t parameters, void *dict,
-                              size_t dictSize) {
-  if (!best) {
-    return;
-  }
-  {
-    size_t liveJobs;
-    pthread_mutex_lock(&best->mutex);
-    --best->liveJobs;
-    liveJobs = best->liveJobs;
-    /* If the new dictionary is better */
-    if (compressedSize < best->compressedSize) {
-      /* Allocate space if necessary */
-      if (!best->dict || best->dictSize < dictSize) {
-        if (best->dict) {
-          free(best->dict);
-        }
-        best->dict = malloc(dictSize);
-        if (!best->dict) {
-          best->compressedSize = ERROR(GENERIC);
-          best->dictSize = 0;
-          return;
-        }
-      }
-      /* Save the dictionary, parameters, and size */
-      memcpy(best->dict, dict, dictSize);
-      best->dictSize = dictSize;
-      best->parameters = parameters;
-      best->compressedSize = compressedSize;
-    }
-    pthread_mutex_unlock(&best->mutex);
-    if (liveJobs == 0) {
-      pthread_cond_broadcast(&best->cond);
-    }
-  }
-}
-
-/**
- * Parameters for COVER_tryParameters().
- */
-typedef struct COVER_tryParameters_data_s {
-  const COVER_ctx_t *ctx;
-  COVER_best_t *best;
-  size_t dictBufferCapacity;
-  ZDICT_cover_params_t parameters;
-} COVER_tryParameters_data_t;
-
-/**
- * Tries a set of parameters and upates the COVER_best_t with the results.
- * This function is thread safe if zstd is compiled with multithreaded support.
- * It takes its parameters as an *OWNING* opaque pointer to support threading.
- */
-static void COVER_tryParameters(void *opaque) {
-  /* Save parameters as local variables */
-  COVER_tryParameters_data_t *const data = (COVER_tryParameters_data_t *)opaque;
-  const COVER_ctx_t *const ctx = data->ctx;
-  const ZDICT_cover_params_t parameters = data->parameters;
-  size_t dictBufferCapacity = data->dictBufferCapacity;
-  size_t totalCompressedSize = ERROR(GENERIC);
-  /* Allocate space for hash table, dict, and freqs */
-  COVER_map_t activeDmers;
-  BYTE *const dict = (BYTE * const)malloc(dictBufferCapacity);
-  U32 *freqs = (U32 *)malloc(ctx->suffixSize * sizeof(U32));
-  if (!COVER_map_init(&activeDmers, parameters.k - parameters.d + 1)) {
-    DISPLAYLEVEL(1, "Failed to allocate dmer map: out of memory\n");
-    goto _cleanup;
-  }
-  if (!dict || !freqs) {
-    DISPLAYLEVEL(1, "Failed to allocate buffers: out of memory\n");
-    goto _cleanup;
-  }
-  /* Copy the frequencies because we need to modify them */
-  memcpy(freqs, ctx->freqs, ctx->suffixSize * sizeof(U32));
-  /* Build the dictionary */
-  {
-    const size_t tail = COVER_buildDictionary(ctx, freqs, &activeDmers, dict,
-                                              dictBufferCapacity, parameters);
-    dictBufferCapacity = ZDICT_finalizeDictionary(
-        dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
-        ctx->samples, ctx->samplesSizes, (unsigned)ctx->nbSamples,
-        parameters.zParams);
-    if (ZDICT_isError(dictBufferCapacity)) {
-      DISPLAYLEVEL(1, "Failed to finalize dictionary\n");
-      goto _cleanup;
-    }
-  }
-  /* Check total compressed size */
-  {
-    /* Pointers */
-    ZSTD_CCtx *cctx;
-    ZSTD_CDict *cdict;
-    void *dst;
-    /* Local variables */
-    size_t dstCapacity;
-    size_t i;
-    /* Allocate dst with enough space to compress the maximum sized sample */
-    {
-      size_t maxSampleSize = 0;
-      for (i = 0; i < ctx->nbSamples; ++i) {
-        maxSampleSize = MAX(ctx->samplesSizes[i], maxSampleSize);
-      }
-      dstCapacity = ZSTD_compressBound(maxSampleSize);
-      dst = malloc(dstCapacity);
-    }
-    /* Create the cctx and cdict */
-    cctx = ZSTD_createCCtx();
-    cdict = ZSTD_createCDict(dict, dictBufferCapacity,
-                             parameters.zParams.compressionLevel);
-    if (!dst || !cctx || !cdict) {
-      goto _compressCleanup;
-    }
-    /* Compress each sample and sum their sizes (or error) */
-    totalCompressedSize = dictBufferCapacity;
-    for (i = 0; i < ctx->nbSamples; ++i) {
-      const size_t size = ZSTD_compress_usingCDict(
-          cctx, dst, dstCapacity, ctx->samples + ctx->offsets[i],
-          ctx->samplesSizes[i], cdict);
-      if (ZSTD_isError(size)) {
-        totalCompressedSize = ERROR(GENERIC);
-        goto _compressCleanup;
-      }
-      totalCompressedSize += size;
-    }
-  _compressCleanup:
-    ZSTD_freeCCtx(cctx);
-    ZSTD_freeCDict(cdict);
-    if (dst) {
-      free(dst);
-    }
-  }
-
-_cleanup:
-  COVER_best_finish(data->best, totalCompressedSize, parameters, dict,
-                    dictBufferCapacity);
-  free(data);
-  COVER_map_destroy(&activeDmers);
-  if (dict) {
-    free(dict);
-  }
-  if (freqs) {
-    free(freqs);
-  }
-}
-
-ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
-    void *dictBuffer, size_t dictBufferCapacity, const void *samplesBuffer,
-    const size_t *samplesSizes, unsigned nbSamples,
-    ZDICT_cover_params_t *parameters) {
-  /* constants */
-  const unsigned nbThreads = parameters->nbThreads;
-  const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
-  const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
-  const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
-  const unsigned kMaxK = parameters->k == 0 ? 2000 : parameters->k;
-  const unsigned kSteps = parameters->steps == 0 ? 40 : parameters->steps;
-  const unsigned kStepSize = MAX((kMaxK - kMinK) / kSteps, 1);
-  const unsigned kIterations =
-      (1 + (kMaxD - kMinD) / 2) * (1 + (kMaxK - kMinK) / kStepSize);
-  /* Local variables */
-  const int displayLevel = parameters->zParams.notificationLevel;
-  unsigned iteration = 1;
-  unsigned d;
-  unsigned k;
-  COVER_best_t best;
-  POOL_ctx *pool = NULL;
-  /* Checks */
-  if (kMinK < kMaxD || kMaxK < kMinK) {
-    LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect parameters\n");
-    return ERROR(GENERIC);
-  }
-  if (nbSamples == 0) {
-    DISPLAYLEVEL(1, "Cover must have at least one input file\n");
-    return ERROR(GENERIC);
-  }
-  if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
-    DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
-                 ZDICT_DICTSIZE_MIN);
-    return ERROR(dstSize_tooSmall);
-  }
-  if (nbThreads > 1) {
-    pool = POOL_create(nbThreads, 1);
-    if (!pool) {
-      return ERROR(memory_allocation);
-    }
-  }
-  /* Initialization */
-  COVER_best_init(&best);
-  /* Turn down global display level to clean up display at level 2 and below */
-  g_displayLevel = displayLevel == 0 ? 0 : displayLevel - 1;
-  /* Loop through d first because each new value needs a new context */
-  LOCALDISPLAYLEVEL(displayLevel, 2, "Trying %u different sets of parameters\n",
-                    kIterations);
-  for (d = kMinD; d <= kMaxD; d += 2) {
-    /* Initialize the context for this value of d */
-    COVER_ctx_t ctx;
-    LOCALDISPLAYLEVEL(displayLevel, 3, "d=%u\n", d);
-    if (!COVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d)) {
-      LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to initialize context\n");
-      COVER_best_destroy(&best);
-      POOL_free(pool);
-      return ERROR(GENERIC);
-    }
-    /* Loop through k reusing the same context */
-    for (k = kMinK; k <= kMaxK; k += kStepSize) {
-      /* Prepare the arguments */
-      COVER_tryParameters_data_t *data = (COVER_tryParameters_data_t *)malloc(
-          sizeof(COVER_tryParameters_data_t));
-      LOCALDISPLAYLEVEL(displayLevel, 3, "k=%u\n", k);
-      if (!data) {
-        LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to allocate parameters\n");
-        COVER_best_destroy(&best);
-        COVER_ctx_destroy(&ctx);
-        POOL_free(pool);
-        return ERROR(GENERIC);
-      }
-      data->ctx = &ctx;
-      data->best = &best;
-      data->dictBufferCapacity = dictBufferCapacity;
-      data->parameters = *parameters;
-      data->parameters.k = k;
-      data->parameters.d = d;
-      data->parameters.steps = kSteps;
-      data->parameters.zParams.notificationLevel = g_displayLevel;
-      /* Check the parameters */
-      if (!COVER_checkParameters(data->parameters, dictBufferCapacity)) {
-        DISPLAYLEVEL(1, "Cover parameters incorrect\n");
-        free(data);
-        continue;
-      }
-      /* Call the function and pass ownership of data to it */
-      COVER_best_start(&best);
-      if (pool) {
-        POOL_add(pool, &COVER_tryParameters, data);
-      } else {
-        COVER_tryParameters(data);
-      }
-      /* Print status */
-      LOCALDISPLAYUPDATE(displayLevel, 2, "\r%u%%       ",
-                         (U32)((iteration * 100) / kIterations));
-      ++iteration;
-    }
-    COVER_best_wait(&best);
-    COVER_ctx_destroy(&ctx);
-  }
-  LOCALDISPLAYLEVEL(displayLevel, 2, "\r%79s\r", "");
-  /* Fill the output buffer and parameters with output of the best parameters */
-  {
-    const size_t dictSize = best.dictSize;
-    if (ZSTD_isError(best.compressedSize)) {
-      const size_t compressedSize = best.compressedSize;
-      COVER_best_destroy(&best);
-      POOL_free(pool);
-      return compressedSize;
-    }
-    *parameters = best.parameters;
-    memcpy(dictBuffer, best.dict, dictSize);
-    COVER_best_destroy(&best);
-    POOL_free(pool);
-    return dictSize;
-  }
-}
diff --git a/contrib/zstd/cpu.h b/contrib/zstd/cpu.h
new file mode 100644
index 000000000..6e8a974f6
--- /dev/null
+++ b/contrib/zstd/cpu.h
@@ -0,0 +1,215 @@
+/*
+ * Copyright (c) 2018-2020, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_COMMON_CPU_H
+#define ZSTD_COMMON_CPU_H
+
+/**
+ * Implementation taken from folly/CpuId.h
+ * https://github.com/facebook/folly/blob/master/folly/CpuId.h
+ */
+
+#include <string.h>
+
+#include "mem.h"
+
+#ifdef _MSC_VER
+#include <intrin.h>
+#endif
+
+typedef struct {
+    U32 f1c;
+    U32 f1d;
+    U32 f7b;
+    U32 f7c;
+} ZSTD_cpuid_t;
+
+MEM_STATIC ZSTD_cpuid_t ZSTD_cpuid(void) {
+    U32 f1c = 0;
+    U32 f1d = 0;
+    U32 f7b = 0;
+    U32 f7c = 0;
+#if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))
+    int reg[4];
+    __cpuid((int*)reg, 0);
+    {
+        int const n = reg[0];
+        if (n >= 1) {
+            __cpuid((int*)reg, 1);
+            f1c = (U32)reg[2];
+            f1d = (U32)reg[3];
+        }
+        if (n >= 7) {
+            __cpuidex((int*)reg, 7, 0);
+            f7b = (U32)reg[1];
+            f7c = (U32)reg[2];
+        }
+    }
+#elif defined(__i386__) && defined(__PIC__) && !defined(__clang__) && defined(__GNUC__)
+    /* The following block like the normal cpuid branch below, but gcc
+     * reserves ebx for use of its pic register so we must specially
+     * handle the save and restore to avoid clobbering the register
+     */
+    U32 n;
+    __asm__(
+        "pushl %%ebx\n\t"
+        "cpuid\n\t"
+        "popl %%ebx\n\t"
+        : "=a"(n)
+        : "a"(0)
+        : "ecx", "edx");
+    if (n >= 1) {
+      U32 f1a;
+      __asm__(
+          "pushl %%ebx\n\t"
+          "cpuid\n\t"
+          "popl %%ebx\n\t"
+          : "=a"(f1a), "=c"(f1c), "=d"(f1d)
+          : "a"(1));
+    }
+    if (n >= 7) {
+      __asm__(
+          "pushl %%ebx\n\t"
+          "cpuid\n\t"
+          "movl %%ebx, %%eax\n\t"
+          "popl %%ebx"
+          : "=a"(f7b), "=c"(f7c)
+          : "a"(7), "c"(0)
+          : "edx");
+    }
+#elif defined(__x86_64__) || defined(_M_X64) || defined(__i386__)
+    U32 n;
+    __asm__("cpuid" : "=a"(n) : "a"(0) : "ebx", "ecx", "edx");
+    if (n >= 1) {
+      U32 f1a;
+      __asm__("cpuid" : "=a"(f1a), "=c"(f1c), "=d"(f1d) : "a"(1) : "ebx");
+    }
+    if (n >= 7) {
+      U32 f7a;
+      __asm__("cpuid"
+              : "=a"(f7a), "=b"(f7b), "=c"(f7c)
+              : "a"(7), "c"(0)
+              : "edx");
+    }
+#endif
+    {
+        ZSTD_cpuid_t cpuid;
+        cpuid.f1c = f1c;
+        cpuid.f1d = f1d;
+        cpuid.f7b = f7b;
+        cpuid.f7c = f7c;
+        return cpuid;
+    }
+}
+
+#define X(name, r, bit)                                                        \
+  MEM_STATIC int ZSTD_cpuid_##name(ZSTD_cpuid_t const cpuid) {                 \
+    return ((cpuid.r) & (1U << bit)) != 0;                                     \
+  }
+
+/* cpuid(1): Processor Info and Feature Bits. */
+#define C(name, bit) X(name, f1c, bit)
+  C(sse3, 0)
+  C(pclmuldq, 1)
+  C(dtes64, 2)
+  C(monitor, 3)
+  C(dscpl, 4)
+  C(vmx, 5)
+  C(smx, 6)
+  C(eist, 7)
+  C(tm2, 8)
+  C(ssse3, 9)
+  C(cnxtid, 10)
+  C(fma, 12)
+  C(cx16, 13)
+  C(xtpr, 14)
+  C(pdcm, 15)
+  C(pcid, 17)
+  C(dca, 18)
+  C(sse41, 19)
+  C(sse42, 20)
+  C(x2apic, 21)
+  C(movbe, 22)
+  C(popcnt, 23)
+  C(tscdeadline, 24)
+  C(aes, 25)
+  C(xsave, 26)
+  C(osxsave, 27)
+  C(avx, 28)
+  C(f16c, 29)
+  C(rdrand, 30)
+#undef C
+#define D(name, bit) X(name, f1d, bit)
+  D(fpu, 0)
+  D(vme, 1)
+  D(de, 2)
+  D(pse, 3)
+  D(tsc, 4)
+  D(msr, 5)
+  D(pae, 6)
+  D(mce, 7)
+  D(cx8, 8)
+  D(apic, 9)
+  D(sep, 11)
+  D(mtrr, 12)
+  D(pge, 13)
+  D(mca, 14)
+  D(cmov, 15)
+  D(pat, 16)
+  D(pse36, 17)
+  D(psn, 18)
+  D(clfsh, 19)
+  D(ds, 21)
+  D(acpi, 22)
+  D(mmx, 23)
+  D(fxsr, 24)
+  D(sse, 25)
+  D(sse2, 26)
+  D(ss, 27)
+  D(htt, 28)
+  D(tm, 29)
+  D(pbe, 31)
+#undef D
+
+/* cpuid(7): Extended Features. */
+#define B(name, bit) X(name, f7b, bit)
+  B(bmi1, 3)
+  B(hle, 4)
+  B(avx2, 5)
+  B(smep, 7)
+  B(bmi2, 8)
+  B(erms, 9)
+  B(invpcid, 10)
+  B(rtm, 11)
+  B(mpx, 14)
+  B(avx512f, 16)
+  B(avx512dq, 17)
+  B(rdseed, 18)
+  B(adx, 19)
+  B(smap, 20)
+  B(avx512ifma, 21)
+  B(pcommit, 22)
+  B(clflushopt, 23)
+  B(clwb, 24)
+  B(avx512pf, 26)
+  B(avx512er, 27)
+  B(avx512cd, 28)
+  B(sha, 29)
+  B(avx512bw, 30)
+  B(avx512vl, 31)
+#undef B
+#define C(name, bit) X(name, f7c, bit)
+  C(prefetchwt1, 0)
+  C(avx512vbmi, 1)
+#undef C
+
+#undef X
+
+#endif /* ZSTD_COMMON_CPU_H */
diff --git a/contrib/zstd/debug.c b/contrib/zstd/debug.c
new file mode 100644
index 000000000..f303f4a2e
--- /dev/null
+++ b/contrib/zstd/debug.c
@@ -0,0 +1,24 @@
+/* ******************************************************************
+ * debug
+ * Part of FSE library
+ * Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
+ *
+ * You can contact the author at :
+ * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+
+/*
+ * This module only hosts one global variable
+ * which can be used to dynamically influence the verbosity of traces,
+ * such as DEBUGLOG and RAWLOG
+ */
+
+#include "debug.h"
+
+int g_debuglevel = DEBUGLEVEL;
diff --git a/contrib/zstd/debug.h b/contrib/zstd/debug.h
new file mode 100644
index 000000000..ac6224888
--- /dev/null
+++ b/contrib/zstd/debug.h
@@ -0,0 +1,114 @@
+/* ******************************************************************
+ * debug
+ * Part of FSE library
+ * Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
+ *
+ * You can contact the author at :
+ * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+
+/*
+ * The purpose of this header is to enable debug functions.
+ * They regroup assert(), DEBUGLOG() and RAWLOG() for run-time,
+ * and DEBUG_STATIC_ASSERT() for compile-time.
+ *
+ * By default, DEBUGLEVEL==0, which means run-time debug is disabled.
+ *
+ * Level 1 enables assert() only.
+ * Starting level 2, traces can be generated and pushed to stderr.
+ * The higher the level, the more verbose the traces.
+ *
+ * It's possible to dynamically adjust level using variable g_debug_level,
+ * which is only declared if DEBUGLEVEL>=2,
+ * and is a global variable, not multi-thread protected (use with care)
+ */
+
+#ifndef DEBUG_H_12987983217
+#define DEBUG_H_12987983217
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/* static assert is triggered at compile time, leaving no runtime artefact.
+ * static assert only works with compile-time constants.
+ * Also, this variant can only be used inside a function. */
+#define DEBUG_STATIC_ASSERT(c) (void)sizeof(char[(c) ? 1 : -1])
+
+
+/* DEBUGLEVEL is expected to be defined externally,
+ * typically through compiler command line.
+ * Value must be a number. */
+#ifndef DEBUGLEVEL
+#  define DEBUGLEVEL 0
+#endif
+
+
+/* DEBUGFILE can be defined externally,
+ * typically through compiler command line.
+ * note : currently useless.
+ * Value must be stderr or stdout */
+#ifndef DEBUGFILE
+#  define DEBUGFILE stderr
+#endif
+
+
+/* recommended values for DEBUGLEVEL :
+ * 0 : release mode, no debug, all run-time checks disabled
+ * 1 : enables assert() only, no display
+ * 2 : reserved, for currently active debug path
+ * 3 : events once per object lifetime (CCtx, CDict, etc.)
+ * 4 : events once per frame
+ * 5 : events once per block
+ * 6 : events once per sequence (verbose)
+ * 7+: events at every position (*very* verbose)
+ *
+ * It's generally inconvenient to output traces > 5.
+ * In which case, it's possible to selectively trigger high verbosity levels
+ * by modifying g_debug_level.
+ */
+
+#if (DEBUGLEVEL>=1)
+#  include <assert.h>
+#else
+#  ifndef assert   /* assert may be already defined, due to prior #include <assert.h> */
+#    define assert(condition) ((void)0)   /* disable assert (default) */
+#  endif
+#endif
+
+#if (DEBUGLEVEL>=2)
+#  include <stdio.h>
+extern int g_debuglevel; /* the variable is only declared,
+                            it actually lives in debug.c,
+                            and is shared by the whole process.
+                            It's not thread-safe.
+                            It's useful when enabling very verbose levels
+                            on selective conditions (such as position in src) */
+
+#  define RAWLOG(l, ...) {                                      \
+                if (l<=g_debuglevel) {                          \
+                    fprintf(stderr, __VA_ARGS__);               \
+            }   }
+#  define DEBUGLOG(l, ...) {                                    \
+                if (l<=g_debuglevel) {                          \
+                    fprintf(stderr, __FILE__ ": " __VA_ARGS__); \
+                    fprintf(stderr, " \n");                     \
+            }   }
+#else
+#  define RAWLOG(l, ...)      {}    /* disabled */
+#  define DEBUGLOG(l, ...)    {}    /* disabled */
+#endif
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* DEBUG_H_12987983217 */
diff --git a/contrib/zstd/entropy_common.c b/contrib/zstd/entropy_common.c
index b37a082fe..9d3e4e8e3 100644
--- a/contrib/zstd/entropy_common.c
+++ b/contrib/zstd/entropy_common.c
@@ -1,36 +1,16 @@
-/*
-   Common functions of New Generation Entropy library
-   Copyright (C) 2016, Yann Collet.
-
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
-
-   Redistribution and use in source and binary forms, with or without
-   modification, are permitted provided that the following conditions are
-   met:
-
-       * Redistributions of source code must retain the above copyright
-   notice, this list of conditions and the following disclaimer.
-       * Redistributions in binary form must reproduce the above
-   copyright notice, this list of conditions and the following disclaimer
-   in the documentation and/or other materials provided with the
-   distribution.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-    You can contact the author at :
-    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
-    - Public forum : https://groups.google.com/forum/#!forum/lz4c
-*************************************************************************** */
+/* ******************************************************************
+ * Common functions of New Generation Entropy library
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
+ *
+ *  You can contact the author at :
+ *  - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
 
 /* *************************************
 *  Dependencies
@@ -72,7 +52,21 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
     unsigned charnum = 0;
     int previous0 = 0;
 
-    if (hbSize < 4) return ERROR(srcSize_wrong);
+    if (hbSize < 4) {
+        /* This function only works when hbSize >= 4 */
+        char buffer[4];
+        memset(buffer, 0, sizeof(buffer));
+        memcpy(buffer, headerBuffer, hbSize);
+        {   size_t const countSize = FSE_readNCount(normalizedCounter, maxSVPtr, tableLogPtr,
+                                                    buffer, sizeof(buffer));
+            if (FSE_isError(countSize)) return countSize;
+            if (countSize > hbSize) return ERROR(corruption_detected);
+            return countSize;
+    }   }
+    assert(hbSize >= 4);
+
+    /* init */
+    memset(normalizedCounter, 0, (*maxSVPtr+1) * sizeof(normalizedCounter[0]));   /* all symbols not present in NCount have a frequency of 0 */
     bitStream = MEM_readLE32(ip);
     nbBits = (bitStream & 0xF) + FSE_MIN_TABLELOG;   /* extract tableLog */
     if (nbBits > FSE_TABLELOG_ABSOLUTE_MAX) return ERROR(tableLog_tooLarge);
@@ -105,6 +99,7 @@ size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSVPtr, unsigned* t
             if (n0 > *maxSVPtr) return ERROR(maxSymbolValue_tooSmall);
             while (charnum < n0) normalizedCounter[charnum++] = 0;
             if ((ip <= iend-7) || (ip + (bitCount>>3) <= iend-4)) {
+                assert((bitCount >> 3) <= 3); /* For first condition to work */
                 ip += bitCount>>3;
                 bitCount &= 7;
                 bitStream = MEM_readLE32(ip) >> bitCount;
diff --git a/contrib/zstd/error_private.c b/contrib/zstd/error_private.c
index 8045e445e..cd437529c 100644
--- a/contrib/zstd/error_private.c
+++ b/contrib/zstd/error_private.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -14,6 +14,10 @@
 
 const char* ERR_getErrorString(ERR_enum code)
 {
+#ifdef ZSTD_STRIP_ERROR_STRINGS
+    (void)code;
+    return "Error strings stripped";
+#else
     static const char* const notErrorCode = "Unspecified error code";
     switch( code )
     {
@@ -29,18 +33,23 @@ const char* ERR_getErrorString(ERR_enum code)
     case PREFIX(parameter_outOfBound): return "Parameter is out of bound";
     case PREFIX(init_missing): return "Context should be init first";
     case PREFIX(memory_allocation): return "Allocation error : not enough memory";
+    case PREFIX(workSpace_tooSmall): return "workSpace buffer is not large enough";
     case PREFIX(stage_wrong): return "Operation not authorized at current processing stage";
-    case PREFIX(dstSize_tooSmall): return "Destination buffer is too small";
-    case PREFIX(srcSize_wrong): return "Src size is incorrect";
     case PREFIX(tableLog_tooLarge): return "tableLog requires too much memory : unsupported";
     case PREFIX(maxSymbolValue_tooLarge): return "Unsupported max Symbol Value : too large";
     case PREFIX(maxSymbolValue_tooSmall): return "Specified maxSymbolValue is too small";
     case PREFIX(dictionary_corrupted): return "Dictionary is corrupted";
     case PREFIX(dictionary_wrong): return "Dictionary mismatch";
     case PREFIX(dictionaryCreation_failed): return "Cannot create Dictionary from provided samples";
+    case PREFIX(dstSize_tooSmall): return "Destination buffer is too small";
+    case PREFIX(srcSize_wrong): return "Src size is incorrect";
+    case PREFIX(dstBuffer_null): return "Operation on NULL destination buffer";
+        /* following error codes are not stable and may be removed or changed in a future version */
     case PREFIX(frameIndex_tooLarge): return "Frame index is too large";
     case PREFIX(seekableIO): return "An I/O error occurred when reading/seeking";
+    case PREFIX(dstBuffer_wrong): return "Destination buffer is wrong";
     case PREFIX(maxCode):
     default: return notErrorCode;
     }
+#endif
 }
diff --git a/contrib/zstd/error_private.h b/contrib/zstd/error_private.h
index 0d2fa7e34..982cf8e9f 100644
--- a/contrib/zstd/error_private.h
+++ b/contrib/zstd/error_private.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -49,7 +49,7 @@ typedef ZSTD_ErrorCode ERR_enum;
 /*-****************************************
 *  Error codes handling
 ******************************************/
-#undef ERROR   /* reported already defined on VS 2015 (Rich Geldreich) */
+#undef ERROR   /* already defined on Visual Studio */
 #define ERROR(name) ZSTD_ERROR(name)
 #define ZSTD_ERROR(name) ((size_t)-PREFIX(name))
 
@@ -57,6 +57,10 @@ ERR_STATIC unsigned ERR_isError(size_t code) { return (code > ERROR(maxCode)); }
 
 ERR_STATIC ERR_enum ERR_getErrorCode(size_t code) { if (!ERR_isError(code)) return (ERR_enum)0; return (ERR_enum) (0-code); }
 
+/* check and forward error code */
+#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return e
+#define CHECK_F(f)   { CHECK_V_F(_var_err__, f); }
+
 
 /*-****************************************
 *  Error Strings
diff --git a/contrib/zstd/fse.h b/contrib/zstd/fse.h
index 1c44f8375..ff54e70ea 100644
--- a/contrib/zstd/fse.h
+++ b/contrib/zstd/fse.h
@@ -1,35 +1,15 @@
 /* ******************************************************************
-   FSE : Finite State Entropy codec
-   Public Prototypes declaration
-   Copyright (C) 2013-2016, Yann Collet.
-
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
-
-   Redistribution and use in source and binary forms, with or without
-   modification, are permitted provided that the following conditions are
-   met:
-
-       * Redistributions of source code must retain the above copyright
-   notice, this list of conditions and the following disclaimer.
-       * Redistributions in binary form must reproduce the above
-   copyright notice, this list of conditions and the following disclaimer
-   in the documentation and/or other materials provided with the
-   distribution.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-   You can contact the author at :
-   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ * FSE : Finite State Entropy codec
+ * Public Prototypes declaration
+ * Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
+ *
+ * You can contact the author at :
+ * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
 ****************************************************************** */
 
 #if defined (__cplusplus)
@@ -72,6 +52,7 @@ extern "C" {
 #define FSE_VERSION_NUMBER  (FSE_VERSION_MAJOR *100*100 + FSE_VERSION_MINOR *100 + FSE_VERSION_RELEASE)
 FSE_PUBLIC_API unsigned FSE_versionNumber(void);   /**< library version number; to be used when checking dll version */
 
+
 /*-****************************************
 *  FSE simple functions
 ******************************************/
@@ -129,7 +110,7 @@ FSE_PUBLIC_API size_t FSE_compress2 (void* dst, size_t dstSize, const void* src,
 ******************************************/
 /*!
 FSE_compress() does the following:
-1. count symbol occurrence from source[] into table count[]
+1. count symbol occurrence from source[] into table count[] (see hist.h)
 2. normalize counters so that sum(count[]) == Power_of_2 (2^tableLog)
 3. save normalized counters to memory buffer using writeNCount()
 4. build encoding table 'CTable' from normalized counters
@@ -147,15 +128,6 @@ or to save and provide normalized distribution using external method.
 
 /* *** COMPRESSION *** */
 
-/*! FSE_count():
-    Provides the precise count of each byte within a table 'count'.
-    'count' is a table of unsigned int, of minimum size (*maxSymbolValuePtr+1).
-    *maxSymbolValuePtr will be updated if detected smaller than initial value.
-    @return : the count of the most frequent symbol (which is not identified).
-              if return == srcSize, there is only one symbol.
-              Can also return an error code, which can be tested with FSE_isError(). */
-FSE_PUBLIC_API size_t FSE_count(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
-
 /*! FSE_optimalTableLog():
     dynamically downsize 'tableLog' when conditions are met.
     It saves CPU time, by using smaller tables, while preserving or even improving compression ratio.
@@ -167,7 +139,8 @@ FSE_PUBLIC_API unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize
     'normalizedCounter' is a table of short, of minimum size (maxSymbolValue+1).
     @return : tableLog,
               or an errorCode, which can be tested using FSE_isError() */
-FSE_PUBLIC_API size_t FSE_normalizeCount(short* normalizedCounter, unsigned tableLog, const unsigned* count, size_t srcSize, unsigned maxSymbolValue);
+FSE_PUBLIC_API size_t FSE_normalizeCount(short* normalizedCounter, unsigned tableLog,
+                    const unsigned* count, size_t srcSize, unsigned maxSymbolValue);
 
 /*! FSE_NCountWriteBound():
     Provides the maximum possible size of an FSE normalized table, given 'maxSymbolValue' and 'tableLog'.
@@ -178,13 +151,14 @@ FSE_PUBLIC_API size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tab
     Compactly save 'normalizedCounter' into 'buffer'.
     @return : size of the compressed table,
               or an errorCode, which can be tested using FSE_isError(). */
-FSE_PUBLIC_API size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);
-
+FSE_PUBLIC_API size_t FSE_writeNCount (void* buffer, size_t bufferSize,
+                                 const short* normalizedCounter,
+                                 unsigned maxSymbolValue, unsigned tableLog);
 
 /*! Constructor and Destructor of FSE_CTable.
     Note that FSE_CTable size depends on 'tableLog' and 'maxSymbolValue' */
 typedef unsigned FSE_CTable;   /* don't allocate that. It's only meant to be more restrictive than void* */
-FSE_PUBLIC_API FSE_CTable* FSE_createCTable (unsigned tableLog, unsigned maxSymbolValue);
+FSE_PUBLIC_API FSE_CTable* FSE_createCTable (unsigned maxSymbolValue, unsigned tableLog);
 FSE_PUBLIC_API void        FSE_freeCTable (FSE_CTable* ct);
 
 /*! FSE_buildCTable():
@@ -250,7 +224,9 @@ If there is an error, the function will return an ErrorCode (which can be tested
     @return : size read from 'rBuffer',
               or an errorCode, which can be tested using FSE_isError().
               maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */
-FSE_PUBLIC_API size_t FSE_readNCount (short* normalizedCounter, unsigned* maxSymbolValuePtr, unsigned* tableLogPtr, const void* rBuffer, size_t rBuffSize);
+FSE_PUBLIC_API size_t FSE_readNCount (short* normalizedCounter,
+                           unsigned* maxSymbolValuePtr, unsigned* tableLogPtr,
+                           const void* rBuffer, size_t rBuffSize);
 
 /*! Constructor and Destructor of FSE_DTable.
     Note that its size depends on 'tableLog' */
@@ -312,7 +288,7 @@ If there is an error, the function will return an error code, which can be teste
 *******************************************/
 /* FSE buffer bounds */
 #define FSE_NCOUNTBOUND 512
-#define FSE_BLOCKBOUND(size) (size + (size>>7))
+#define FSE_BLOCKBOUND(size) (size + (size>>7) + 4 /* fse states */ + sizeof(size_t) /* bitContainer */)
 #define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size))   /* Macro version, useful for static allocation */
 
 /* It is possible to statically allocate FSE CTable/DTable as a table of FSE_CTable/FSE_DTable using below macros */
@@ -325,33 +301,8 @@ If there is an error, the function will return an error code, which can be teste
 
 
 /* *****************************************
-*  FSE advanced API
-*******************************************/
-/* FSE_count_wksp() :
- * Same as FSE_count(), but using an externally provided scratch buffer.
- * `workSpace` size must be table of >= `1024` unsigned
- */
-size_t FSE_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
-                 const void* source, size_t sourceSize, unsigned* workSpace);
-
-/** FSE_countFast() :
- *  same as FSE_count(), but blindly trusts that all byte values within src are <= *maxSymbolValuePtr
- */
-size_t FSE_countFast(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
-
-/* FSE_countFast_wksp() :
- * Same as FSE_countFast(), but using an externally provided scratch buffer.
- * `workSpace` must be a table of minimum `1024` unsigned
- */
-size_t FSE_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned* workSpace);
-
-/*! FSE_count_simple
- * Same as FSE_countFast(), but does not use any additional memory (not even on stack).
- * This function is unsafe, and will segfault if any value within `src` is `> *maxSymbolValuePtr` (presuming it's also the size of `count`).
-*/
-size_t FSE_count_simple(unsigned* count, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize);
-
-
+ *  FSE advanced API
+ ***************************************** */
 
 unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus);
 /**< same as FSE_optimalTableLog(), which used `minus==2` */
@@ -387,7 +338,7 @@ size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size
 typedef enum {
    FSE_repeat_none,  /**< Cannot use the previous table */
    FSE_repeat_check, /**< Can use the previous table but it must be checked */
-   FSE_repeat_valid  /**< Can use the previous table and it is asumed to be valid */
+   FSE_repeat_valid  /**< Can use the previous table and it is assumed to be valid */
  } FSE_repeat;
 
 /* *****************************************
@@ -541,7 +492,7 @@ MEM_STATIC void FSE_initCState(FSE_CState_t* statePtr, const FSE_CTable* ct)
     const U32 tableLog = MEM_read16(ptr);
     statePtr->value = (ptrdiff_t)1<<tableLog;
     statePtr->stateTable = u16ptr+2;
-    statePtr->symbolTT = ((const U32*)ct + 1 + (tableLog ? (1<<(tableLog-1)) : 1));
+    statePtr->symbolTT = ct + 1 + (tableLog ? (1<<(tableLog-1)) : 1);
     statePtr->stateLog = tableLog;
 }
 
@@ -560,7 +511,7 @@ MEM_STATIC void FSE_initCState2(FSE_CState_t* statePtr, const FSE_CTable* ct, U3
     }
 }
 
-MEM_STATIC void FSE_encodeSymbol(BIT_CStream_t* bitC, FSE_CState_t* statePtr, U32 symbol)
+MEM_STATIC void FSE_encodeSymbol(BIT_CStream_t* bitC, FSE_CState_t* statePtr, unsigned symbol)
 {
     FSE_symbolCompressionTransform const symbolTT = ((const FSE_symbolCompressionTransform*)(statePtr->symbolTT))[symbol];
     const U16* const stateTable = (const U16*)(statePtr->stateTable);
@@ -576,6 +527,39 @@ MEM_STATIC void FSE_flushCState(BIT_CStream_t* bitC, const FSE_CState_t* statePt
 }
 
 
+/* FSE_getMaxNbBits() :
+ * Approximate maximum cost of a symbol, in bits.
+ * Fractional get rounded up (i.e : a symbol with a normalized frequency of 3 gives the same result as a frequency of 2)
+ * note 1 : assume symbolValue is valid (<= maxSymbolValue)
+ * note 2 : if freq[symbolValue]==0, @return a fake cost of tableLog+1 bits */
+MEM_STATIC U32 FSE_getMaxNbBits(const void* symbolTTPtr, U32 symbolValue)
+{
+    const FSE_symbolCompressionTransform* symbolTT = (const FSE_symbolCompressionTransform*) symbolTTPtr;
+    return (symbolTT[symbolValue].deltaNbBits + ((1<<16)-1)) >> 16;
+}
+
+/* FSE_bitCost() :
+ * Approximate symbol cost, as fractional value, using fixed-point format (accuracyLog fractional bits)
+ * note 1 : assume symbolValue is valid (<= maxSymbolValue)
+ * note 2 : if freq[symbolValue]==0, @return a fake cost of tableLog+1 bits */
+MEM_STATIC U32 FSE_bitCost(const void* symbolTTPtr, U32 tableLog, U32 symbolValue, U32 accuracyLog)
+{
+    const FSE_symbolCompressionTransform* symbolTT = (const FSE_symbolCompressionTransform*) symbolTTPtr;
+    U32 const minNbBits = symbolTT[symbolValue].deltaNbBits >> 16;
+    U32 const threshold = (minNbBits+1) << 16;
+    assert(tableLog < 16);
+    assert(accuracyLog < 31-tableLog);  /* ensure enough room for renormalization double shift */
+    {   U32 const tableSize = 1 << tableLog;
+        U32 const deltaFromThreshold = threshold - (symbolTT[symbolValue].deltaNbBits + tableSize);
+        U32 const normalizedDeltaFromThreshold = (deltaFromThreshold << accuracyLog) >> tableLog;   /* linear interpolation (very approximate) */
+        U32 const bitMultiplier = 1 << accuracyLog;
+        assert(symbolTT[symbolValue].deltaNbBits + tableSize <= threshold);
+        assert(normalizedDeltaFromThreshold <= bitMultiplier);
+        return (minNbBits+1)*bitMultiplier - normalizedDeltaFromThreshold;
+    }
+}
+
+
 /* ======    Decompression    ====== */
 
 typedef struct {
diff --git a/contrib/zstd/fse_compress.c b/contrib/zstd/fse_compress.c
index 599280b90..e09b646df 100644
--- a/contrib/zstd/fse_compress.c
+++ b/contrib/zstd/fse_compress.c
@@ -1,35 +1,15 @@
 /* ******************************************************************
-   FSE : Finite State Entropy encoder
-   Copyright (C) 2013-2015, Yann Collet.
-
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
-
-   Redistribution and use in source and binary forms, with or without
-   modification, are permitted provided that the following conditions are
-   met:
-
-       * Redistributions of source code must retain the above copyright
-   notice, this list of conditions and the following disclaimer.
-       * Redistributions in binary form must reproduce the above
-   copyright notice, this list of conditions and the following disclaimer
-   in the documentation and/or other materials provided with the
-   distribution.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-    You can contact the author at :
-    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
-    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ * FSE : Finite State Entropy encoder
+ * Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
+ *
+ *  You can contact the author at :
+ *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
 ****************************************************************** */
 
 /* **************************************************************
@@ -37,9 +17,11 @@
 ****************************************************************/
 #include <stdlib.h>     /* malloc, free, qsort */
 #include <string.h>     /* memcpy, memset */
-#include <stdio.h>      /* printf (debug) */
-#include "bitstream.h"
 #include "compiler.h"
+#include "mem.h"        /* U32, U16, etc. */
+#include "debug.h"      /* assert, DEBUGLOG */
+#include "hist.h"       /* HIST_count_wksp */
+#include "bitstream.h"
 #define FSE_STATIC_LINKING_ONLY
 #include "fse.h"
 #include "error_private.h"
@@ -49,7 +31,6 @@
 *  Error Management
 ****************************************************************/
 #define FSE_isError ERR_isError
-#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
 
 
 /* **************************************************************
@@ -82,7 +63,9 @@
  * wkspSize should be sized to handle worst case situation, which is `1<<max_tableLog * sizeof(FSE_FUNCTION_TYPE)`
  * workSpace must also be properly aligned with FSE_FUNCTION_TYPE requirements
  */
-size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
+size_t FSE_buildCTable_wksp(FSE_CTable* ct,
+                      const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
+                            void* workSpace, size_t wkspSize)
 {
     U32 const tableSize = 1 << tableLog;
     U32 const tableMask = tableSize - 1;
@@ -100,14 +83,19 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsi
     if (((size_t)1 << tableLog) * sizeof(FSE_FUNCTION_TYPE) > wkspSize) return ERROR(tableLog_tooLarge);
     tableU16[-2] = (U16) tableLog;
     tableU16[-1] = (U16) maxSymbolValue;
+    assert(tableLog < 16);   /* required for threshold strategy to work */
 
     /* For explanations on how to distribute symbol values over the table :
-    *  http://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
+     * http://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
+
+     #ifdef __clang_analyzer__
+     memset(tableSymbol, 0, sizeof(*tableSymbol) * tableSize);   /* useless initialization, just to keep scan-build happy */
+     #endif
 
     /* symbol start positions */
     {   U32 u;
         cumul[0] = 0;
-        for (u=1; u<=maxSymbolValue+1; u++) {
+        for (u=1; u <= maxSymbolValue+1; u++) {
             if (normalizedCounter[u-1]==-1) {  /* Low proba symbol */
                 cumul[u] = cumul[u-1] + 1;
                 tableSymbol[highThreshold--] = (FSE_FUNCTION_TYPE)(u-1);
@@ -121,14 +109,16 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsi
     {   U32 position = 0;
         U32 symbol;
         for (symbol=0; symbol<=maxSymbolValue; symbol++) {
-            int nbOccurences;
-            for (nbOccurences=0; nbOccurences<normalizedCounter[symbol]; nbOccurences++) {
+            int nbOccurrences;
+            int const freq = normalizedCounter[symbol];
+            for (nbOccurrences=0; nbOccurrences<freq; nbOccurrences++) {
                 tableSymbol[position] = (FSE_FUNCTION_TYPE)symbol;
                 position = (position + step) & tableMask;
-                while (position > highThreshold) position = (position + step) & tableMask;   /* Low proba area */
+                while (position > highThreshold)
+                    position = (position + step) & tableMask;   /* Low proba area */
         }   }
 
-        if (position!=0) return ERROR(GENERIC);   /* Must have gone through all positions */
+        assert(position==0);  /* Must have initialized all positions */
     }
 
     /* Build table */
@@ -143,7 +133,10 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsi
         for (s=0; s<=maxSymbolValue; s++) {
             switch (normalizedCounter[s])
             {
-            case  0: break;
+            case  0:
+                /* filling nonetheless, for compatibility with FSE_getMaxNbBits() */
+                symbolTT[s].deltaNbBits = ((tableLog+1) << 16) - (1<<tableLog);
+                break;
 
             case -1:
             case  1:
@@ -160,6 +153,18 @@ size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsi
                     total +=  normalizedCounter[s];
     }   }   }   }
 
+#if 0  /* debug : symbol costs */
+    DEBUGLOG(5, "\n --- table statistics : ");
+    {   U32 symbol;
+        for (symbol=0; symbol<=maxSymbolValue; symbol++) {
+            DEBUGLOG(5, "%3u: w=%3i,   maxBits=%u, fracBits=%.2f",
+                symbol, normalizedCounter[symbol],
+                FSE_getMaxNbBits(symbolTT, symbol),
+                (double)FSE_bitCost(symbolTT, tableLog, symbol, 8) / 256);
+        }
+    }
+#endif
+
     return 0;
 }
 
@@ -174,8 +179,9 @@ size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned
 
 #ifndef FSE_COMMONDEFS_ONLY
 
+
 /*-**************************************************************
-*  FSE NCount encoding-decoding
+*  FSE NCount encoding
 ****************************************************************/
 size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
 {
@@ -183,9 +189,10 @@ size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
     return maxSymbolValue ? maxHeaderSize : FSE_NCOUNTBOUND;  /* maxSymbolValue==0 ? use default */
 }
 
-static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
-                                       const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
-                                       unsigned writeIsSafe)
+static size_t
+FSE_writeNCount_generic (void* header, size_t headerBufferSize,
+                   const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
+                         unsigned writeIsSafe)
 {
     BYTE* const ostart = (BYTE*) header;
     BYTE* out = ostart;
@@ -194,13 +201,12 @@ static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
     const int tableSize = 1 << tableLog;
     int remaining;
     int threshold;
-    U32 bitStream;
-    int bitCount;
-    unsigned charnum = 0;
-    int previous0 = 0;
+    U32 bitStream = 0;
+    int bitCount = 0;
+    unsigned symbol = 0;
+    unsigned const alphabetSize = maxSymbolValue + 1;
+    int previousIs0 = 0;
 
-    bitStream = 0;
-    bitCount  = 0;
     /* Table Size */
     bitStream += (tableLog-FSE_MIN_TABLELOG) << bitCount;
     bitCount  += 4;
@@ -210,48 +216,53 @@ static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
     threshold = tableSize;
     nbBits = tableLog+1;
 
-    while (remaining>1) {  /* stops at 1 */
-        if (previous0) {
-            unsigned start = charnum;
-            while (!normalizedCounter[charnum]) charnum++;
-            while (charnum >= start+24) {
+    while ((symbol < alphabetSize) && (remaining>1)) {  /* stops at 1 */
+        if (previousIs0) {
+            unsigned start = symbol;
+            while ((symbol < alphabetSize) && !normalizedCounter[symbol]) symbol++;
+            if (symbol == alphabetSize) break;   /* incorrect distribution */
+            while (symbol >= start+24) {
                 start+=24;
                 bitStream += 0xFFFFU << bitCount;
-                if ((!writeIsSafe) && (out > oend-2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+                if ((!writeIsSafe) && (out > oend-2))
+                    return ERROR(dstSize_tooSmall);   /* Buffer overflow */
                 out[0] = (BYTE) bitStream;
                 out[1] = (BYTE)(bitStream>>8);
                 out+=2;
                 bitStream>>=16;
             }
-            while (charnum >= start+3) {
+            while (symbol >= start+3) {
                 start+=3;
                 bitStream += 3 << bitCount;
                 bitCount += 2;
             }
-            bitStream += (charnum-start) << bitCount;
+            bitStream += (symbol-start) << bitCount;
             bitCount += 2;
             if (bitCount>16) {
-                if ((!writeIsSafe) && (out > oend - 2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+                if ((!writeIsSafe) && (out > oend - 2))
+                    return ERROR(dstSize_tooSmall);   /* Buffer overflow */
                 out[0] = (BYTE)bitStream;
                 out[1] = (BYTE)(bitStream>>8);
                 out += 2;
                 bitStream >>= 16;
                 bitCount -= 16;
         }   }
-        {   int count = normalizedCounter[charnum++];
-            int const max = (2*threshold-1)-remaining;
+        {   int count = normalizedCounter[symbol++];
+            int const max = (2*threshold-1) - remaining;
             remaining -= count < 0 ? -count : count;
             count++;   /* +1 for extra accuracy */
-            if (count>=threshold) count += max;   /* [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ */
+            if (count>=threshold)
+                count += max;   /* [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ */
             bitStream += count << bitCount;
             bitCount  += nbBits;
             bitCount  -= (count<max);
-            previous0  = (count==1);
+            previousIs0  = (count==1);
             if (remaining<1) return ERROR(GENERIC);
-            while (remaining<threshold) nbBits--, threshold>>=1;
+            while (remaining<threshold) { nbBits--; threshold>>=1; }
         }
         if (bitCount>16) {
-            if ((!writeIsSafe) && (out > oend - 2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+            if ((!writeIsSafe) && (out > oend - 2))
+                return ERROR(dstSize_tooSmall);   /* Buffer overflow */
             out[0] = (BYTE)bitStream;
             out[1] = (BYTE)(bitStream>>8);
             out += 2;
@@ -259,19 +270,23 @@ static size_t FSE_writeNCount_generic (void* header, size_t headerBufferSize,
             bitCount -= 16;
     }   }
 
+    if (remaining != 1)
+        return ERROR(GENERIC);  /* incorrect normalized distribution */
+    assert(symbol <= alphabetSize);
+
     /* flush remaining bitStream */
-    if ((!writeIsSafe) && (out > oend - 2)) return ERROR(dstSize_tooSmall);   /* Buffer overflow */
+    if ((!writeIsSafe) && (out > oend - 2))
+        return ERROR(dstSize_tooSmall);   /* Buffer overflow */
     out[0] = (BYTE)bitStream;
     out[1] = (BYTE)(bitStream>>8);
     out+= (bitCount+7) /8;
 
-    if (charnum > maxSymbolValue + 1) return ERROR(GENERIC);
-
     return (out-ostart);
 }
 
 
-size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
+size_t FSE_writeNCount (void* buffer, size_t bufferSize,
+                  const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
 {
     if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported */
     if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC);   /* Unsupported */
@@ -279,171 +294,13 @@ size_t FSE_writeNCount (void* buffer, size_t bufferSize, const short* normalized
     if (bufferSize < FSE_NCountWriteBound(maxSymbolValue, tableLog))
         return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 0);
 
-    return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 1);
+    return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 1 /* write in buffer is safe */);
 }
 
 
-
-/*-**************************************************************
-*  Counting histogram
-****************************************************************/
-/*! FSE_count_simple
-    This function counts byte values within `src`, and store the histogram into table `count`.
-    It doesn't use any additional memory.
-    But this function is unsafe : it doesn't check that all values within `src` can fit into `count`.
-    For this reason, prefer using a table `count` with 256 elements.
-    @return : count of most numerous element
-*/
-size_t FSE_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
-                        const void* src, size_t srcSize)
-{
-    const BYTE* ip = (const BYTE*)src;
-    const BYTE* const end = ip + srcSize;
-    unsigned maxSymbolValue = *maxSymbolValuePtr;
-    unsigned max=0;
-
-    memset(count, 0, (maxSymbolValue+1)*sizeof(*count));
-    if (srcSize==0) { *maxSymbolValuePtr = 0; return 0; }
-
-    while (ip<end) count[*ip++]++;
-
-    while (!count[maxSymbolValue]) maxSymbolValue--;
-    *maxSymbolValuePtr = maxSymbolValue;
-
-    { U32 s; for (s=0; s<=maxSymbolValue; s++) if (count[s] > max) max = count[s]; }
-
-    return (size_t)max;
-}
-
-
-/* FSE_count_parallel_wksp() :
- * Same as FSE_count_parallel(), but using an externally provided scratch buffer.
- * `workSpace` size must be a minimum of `1024 * sizeof(unsigned)`` */
-static size_t FSE_count_parallel_wksp(
-                                unsigned* count, unsigned* maxSymbolValuePtr,
-                                const void* source, size_t sourceSize,
-                                unsigned checkMax, unsigned* const workSpace)
-{
-    const BYTE* ip = (const BYTE*)source;
-    const BYTE* const iend = ip+sourceSize;
-    unsigned maxSymbolValue = *maxSymbolValuePtr;
-    unsigned max=0;
-    U32* const Counting1 = workSpace;
-    U32* const Counting2 = Counting1 + 256;
-    U32* const Counting3 = Counting2 + 256;
-    U32* const Counting4 = Counting3 + 256;
-
-    memset(Counting1, 0, 4*256*sizeof(unsigned));
-
-    /* safety checks */
-    if (!sourceSize) {
-        memset(count, 0, maxSymbolValue + 1);
-        *maxSymbolValuePtr = 0;
-        return 0;
-    }
-    if (!maxSymbolValue) maxSymbolValue = 255;            /* 0 == default */
-
-    /* by stripes of 16 bytes */
-    {   U32 cached = MEM_read32(ip); ip += 4;
-        while (ip < iend-15) {
-            U32 c = cached; cached = MEM_read32(ip); ip += 4;
-            Counting1[(BYTE) c     ]++;
-            Counting2[(BYTE)(c>>8) ]++;
-            Counting3[(BYTE)(c>>16)]++;
-            Counting4[       c>>24 ]++;
-            c = cached; cached = MEM_read32(ip); ip += 4;
-            Counting1[(BYTE) c     ]++;
-            Counting2[(BYTE)(c>>8) ]++;
-            Counting3[(BYTE)(c>>16)]++;
-            Counting4[       c>>24 ]++;
-            c = cached; cached = MEM_read32(ip); ip += 4;
-            Counting1[(BYTE) c     ]++;
-            Counting2[(BYTE)(c>>8) ]++;
-            Counting3[(BYTE)(c>>16)]++;
-            Counting4[       c>>24 ]++;
-            c = cached; cached = MEM_read32(ip); ip += 4;
-            Counting1[(BYTE) c     ]++;
-            Counting2[(BYTE)(c>>8) ]++;
-            Counting3[(BYTE)(c>>16)]++;
-            Counting4[       c>>24 ]++;
-        }
-        ip-=4;
-    }
-
-    /* finish last symbols */
-    while (ip<iend) Counting1[*ip++]++;
-
-    if (checkMax) {   /* verify stats will fit into destination table */
-        U32 s; for (s=255; s>maxSymbolValue; s--) {
-            Counting1[s] += Counting2[s] + Counting3[s] + Counting4[s];
-            if (Counting1[s]) return ERROR(maxSymbolValue_tooSmall);
-    }   }
-
-    {   U32 s; for (s=0; s<=maxSymbolValue; s++) {
-            count[s] = Counting1[s] + Counting2[s] + Counting3[s] + Counting4[s];
-            if (count[s] > max) max = count[s];
-    }   }
-
-    while (!count[maxSymbolValue]) maxSymbolValue--;
-    *maxSymbolValuePtr = maxSymbolValue;
-    return (size_t)max;
-}
-
-/* FSE_countFast_wksp() :
- * Same as FSE_countFast(), but using an externally provided scratch buffer.
- * `workSpace` size must be table of >= `1024` unsigned */
-size_t FSE_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
-                     const void* source, size_t sourceSize, unsigned* workSpace)
-{
-    if (sourceSize < 1500) return FSE_count_simple(count, maxSymbolValuePtr, source, sourceSize);
-    return FSE_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, 0, workSpace);
-}
-
-/* fast variant (unsafe : won't check if src contains values beyond count[] limit) */
-size_t FSE_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
-                     const void* source, size_t sourceSize)
-{
-    unsigned tmpCounters[1024];
-    return FSE_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, tmpCounters);
-}
-
-/* FSE_count_wksp() :
- * Same as FSE_count(), but using an externally provided scratch buffer.
- * `workSpace` size must be table of >= `1024` unsigned */
-size_t FSE_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
-                 const void* source, size_t sourceSize, unsigned* workSpace)
-{
-    if (*maxSymbolValuePtr < 255)
-        return FSE_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, 1, workSpace);
-    *maxSymbolValuePtr = 255;
-    return FSE_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, workSpace);
-}
-
-size_t FSE_count(unsigned* count, unsigned* maxSymbolValuePtr,
-                 const void* src, size_t srcSize)
-{
-    unsigned tmpCounters[1024];
-    return FSE_count_wksp(count, maxSymbolValuePtr, src, srcSize, tmpCounters);
-}
-
-
-
 /*-**************************************************************
 *  FSE Compression Code
 ****************************************************************/
-/*! FSE_sizeof_CTable() :
-    FSE_CTable is a variable size structure which contains :
-    `U16 tableLog;`
-    `U16 maxSymbolValue;`
-    `U16 nextStateNumber[1 << tableLog];`                         // This size is variable
-    `FSE_symbolCompressionTransform symbolTT[maxSymbolValue+1];`  // This size is variable
-Allocation is manual (C standard does not support variable-size structures).
-*/
-size_t FSE_sizeof_CTable (unsigned maxSymbolValue, unsigned tableLog)
-{
-    if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
-    return FSE_CTABLE_SIZE_U32 (tableLog, maxSymbolValue) * sizeof(U32);
-}
 
 FSE_CTable* FSE_createCTable (unsigned maxSymbolValue, unsigned tableLog)
 {
@@ -458,7 +315,7 @@ void FSE_freeCTable (FSE_CTable* ct) { free(ct); }
 /* provides the minimum logSize to safely represent a distribution */
 static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
 {
-    U32 minBitsSrc = BIT_highbit32((U32)(srcSize - 1)) + 1;
+    U32 minBitsSrc = BIT_highbit32((U32)(srcSize)) + 1;
     U32 minBitsSymbols = BIT_highbit32(maxSymbolValue) + 2;
     U32 minBits = minBitsSrc < minBitsSymbols ? minBitsSrc : minBitsSymbols;
     assert(srcSize > 1); /* Not supported, RLE should be used instead */
@@ -521,6 +378,9 @@ static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count,
     }
     ToDistribute = (1 << tableLog) - distributed;
 
+    if (ToDistribute == 0)
+        return 0;
+
     if ((total / ToDistribute) > lowOne) {
         /* risk of rounding to zero */
         lowOne = (U32)((total * 3) / (ToDistribute * 2));
@@ -540,7 +400,7 @@ static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count,
            find max, then give all remaining points to max */
         U32 maxV = 0, maxC = 0;
         for (s=0; s<=maxSymbolValue; s++)
-            if (count[s] > maxC) maxV=s, maxC=count[s];
+            if (count[s] > maxC) { maxV=s; maxC=count[s]; }
         norm[maxV] += (short)ToDistribute;
         return 0;
     }
@@ -548,7 +408,7 @@ static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count,
     if (total == 0) {
         /* all of the symbols were low enough for the lowOne or lowThreshold */
         for (s=0; ToDistribute > 0; s = (s+1)%(maxSymbolValue+1))
-            if (norm[s] > 0) ToDistribute--, norm[s]++;
+            if (norm[s] > 0) { ToDistribute--; norm[s]++; }
         return 0;
     }
 
@@ -582,7 +442,7 @@ size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
     if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported size */
     if (tableLog < FSE_minTableLog(total, maxSymbolValue)) return ERROR(GENERIC);   /* Too small tableLog, compression potentially impossible */
 
-    {   U32 const rtbTable[] = {     0, 473195, 504333, 520860, 550000, 700000, 750000, 830000 };
+    {   static U32 const rtbTable[] = {     0, 473195, 504333, 520860, 550000, 700000, 750000, 830000 };
         U64 const scale = 62 - tableLog;
         U64 const step = ((U64)1<<62) / total;   /* <== here, one division ! */
         U64 const vStep = 1ULL<<(scale-20);
@@ -604,7 +464,7 @@ size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
                     U64 restToBeat = vStep * rtbTable[proba];
                     proba += (count[s]*step) - ((U64)proba<<scale) > restToBeat;
                 }
-                if (proba > largestP) largestP=proba, largest=s;
+                if (proba > largestP) { largestP=proba; largest=s; }
                 normalizedCounter[s] = proba;
                 stillToDistribute -= proba;
         }   }
@@ -621,11 +481,11 @@ size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
         U32 s;
         U32 nTotal = 0;
         for (s=0; s<=maxSymbolValue; s++)
-            printf("%3i: %4i \n", s, normalizedCounter[s]);
+            RAWLOG(2, "%3i: %4i \n", s, normalizedCounter[s]);
         for (s=0; s<=maxSymbolValue; s++)
             nTotal += abs(normalizedCounter[s]);
         if (nTotal != (1U<<tableLog))
-            printf("Warning !!! Total == %u != %u !!!", nTotal, 1U<<tableLog);
+            RAWLOG(2, "Warning !!! Total == %u != %u !!!", nTotal, 1U<<tableLog);
         getchar();
     }
 #endif
@@ -765,9 +625,6 @@ size_t FSE_compress_usingCTable (void* dst, size_t dstSize,
 
 size_t FSE_compressBound(size_t size) { return FSE_COMPRESSBOUND(size); }
 
-#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return e
-#define CHECK_F(f)   { CHECK_V_F(_var_err__, f); }
-
 /* FSE_compress_wksp() :
  * Same as FSE_compress2(), but using an externally allocated scratch buffer (`workSpace`).
  * `wkspSize` size must be `(1<<tableLog)`.
@@ -778,7 +635,7 @@ size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t src
     BYTE* op = ostart;
     BYTE* const oend = ostart + dstSize;
 
-    U32   count[FSE_MAX_SYMBOL_VALUE+1];
+    unsigned count[FSE_MAX_SYMBOL_VALUE+1];
     S16   norm[FSE_MAX_SYMBOL_VALUE+1];
     FSE_CTable* CTable = (FSE_CTable*)workSpace;
     size_t const CTableSize = FSE_CTABLE_SIZE_U32(tableLog, maxSymbolValue);
@@ -792,7 +649,7 @@ size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t src
     if (!tableLog) tableLog = FSE_DEFAULT_TABLELOG;
 
     /* Scan input and build symbol stats */
-    {   CHECK_V_F(maxCount, FSE_count_wksp(count, &maxSymbolValue, src, srcSize, (unsigned*)scratchBuffer) );
+    {   CHECK_V_F(maxCount, HIST_count_wksp(count, &maxSymbolValue, src, srcSize, scratchBuffer, scratchBufferSize) );
         if (maxCount == srcSize) return 1;   /* only a single symbol in src : rle */
         if (maxCount == 1) return 0;         /* each symbol present maximum once => not compressible */
         if (maxCount < (srcSize >> 7)) return 0;   /* Heuristic : not compressible enough */
@@ -827,7 +684,7 @@ typedef struct {
 size_t FSE_compress2 (void* dst, size_t dstCapacity, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog)
 {
     fseWkspMax_t scratchBuffer;
-    FSE_STATIC_ASSERT(sizeof(scratchBuffer) >= FSE_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE));   /* compilation failures here means scratchBuffer is not large enough */
+    DEBUG_STATIC_ASSERT(sizeof(scratchBuffer) >= FSE_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE));   /* compilation failures here means scratchBuffer is not large enough */
     if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
     return FSE_compress_wksp(dst, dstCapacity, src, srcSize, maxSymbolValue, tableLog, &scratchBuffer, sizeof(scratchBuffer));
 }
diff --git a/contrib/zstd/fse_decompress.c b/contrib/zstd/fse_decompress.c
index 8e3f0035f..bcc2223cc 100644
--- a/contrib/zstd/fse_decompress.c
+++ b/contrib/zstd/fse_decompress.c
@@ -1,35 +1,15 @@
 /* ******************************************************************
-   FSE : Finite State Entropy decoder
-   Copyright (C) 2013-2015, Yann Collet.
-
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
-
-   Redistribution and use in source and binary forms, with or without
-   modification, are permitted provided that the following conditions are
-   met:
-
-       * Redistributions of source code must retain the above copyright
-   notice, this list of conditions and the following disclaimer.
-       * Redistributions in binary form must reproduce the above
-   copyright notice, this list of conditions and the following disclaimer
-   in the documentation and/or other materials provided with the
-   distribution.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-    You can contact the author at :
-    - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
-    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ * FSE : Finite State Entropy decoder
+ * Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
+ *
+ *  You can contact the author at :
+ *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
 ****************************************************************** */
 
 
@@ -49,10 +29,7 @@
 *  Error Management
 ****************************************************************/
 #define FSE_isError ERR_isError
-#define FSE_STATIC_ASSERT(c) { enum { FSE_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
-
-/* check and forward error code */
-#define CHECK_F(f) { size_t const e = f; if (FSE_isError(e)) return e; }
+#define FSE_STATIC_ASSERT(c) DEBUG_STATIC_ASSERT(c)   /* use only *after* variable declarations */
 
 
 /* **************************************************************
@@ -139,8 +116,8 @@ size_t FSE_buildDTable(FSE_DTable* dt, const short* normalizedCounter, unsigned
     {   U32 u;
         for (u=0; u<tableSize; u++) {
             FSE_FUNCTION_TYPE const symbol = (FSE_FUNCTION_TYPE)(tableDecode[u].symbol);
-            U16 nextState = symbolNext[symbol]++;
-            tableDecode[u].nbBits = (BYTE) (tableLog - BIT_highbit32 ((U32)nextState) );
+            U32 const nextState = symbolNext[symbol]++;
+            tableDecode[u].nbBits = (BYTE) (tableLog - BIT_highbit32(nextState) );
             tableDecode[u].newState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
     }   }
 
@@ -285,7 +262,7 @@ size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size
     /* normal FSE decoding mode */
     size_t const NCountLength = FSE_readNCount (counting, &maxSymbolValue, &tableLog, istart, cSrcSize);
     if (FSE_isError(NCountLength)) return NCountLength;
-    //if (NCountLength >= cSrcSize) return ERROR(srcSize_wrong);   /* too small input size; supposed to be already checked in NCountLength, only remaining case : NCountLength==cSrcSize */
+    /* if (NCountLength >= cSrcSize) return ERROR(srcSize_wrong); */  /* too small input size; supposed to be already checked in NCountLength, only remaining case : NCountLength==cSrcSize */
     if (tableLog > maxLog) return ERROR(tableLog_tooLarge);
     ip += NCountLength;
     cSrcSize -= NCountLength;
diff --git a/contrib/zstd/hist.c b/contrib/zstd/hist.c
new file mode 100644
index 000000000..c17b9725f
--- /dev/null
+++ b/contrib/zstd/hist.c
@@ -0,0 +1,183 @@
+/* ******************************************************************
+ * hist : Histogram functions
+ * part of Finite State Entropy project
+ * Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
+ *
+ *  You can contact the author at :
+ *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+/* --- dependencies --- */
+#include "mem.h"             /* U32, BYTE, etc. */
+#include "debug.h"           /* assert, DEBUGLOG */
+#include "error_private.h"   /* ERROR */
+#include "hist.h"
+
+
+/* --- Error management --- */
+unsigned HIST_isError(size_t code) { return ERR_isError(code); }
+
+/*-**************************************************************
+ *  Histogram functions
+ ****************************************************************/
+unsigned HIST_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
+                           const void* src, size_t srcSize)
+{
+    const BYTE* ip = (const BYTE*)src;
+    const BYTE* const end = ip + srcSize;
+    unsigned maxSymbolValue = *maxSymbolValuePtr;
+    unsigned largestCount=0;
+
+    memset(count, 0, (maxSymbolValue+1) * sizeof(*count));
+    if (srcSize==0) { *maxSymbolValuePtr = 0; return 0; }
+
+    while (ip<end) {
+        assert(*ip <= maxSymbolValue);
+        count[*ip++]++;
+    }
+
+    while (!count[maxSymbolValue]) maxSymbolValue--;
+    *maxSymbolValuePtr = maxSymbolValue;
+
+    {   U32 s;
+        for (s=0; s<=maxSymbolValue; s++)
+            if (count[s] > largestCount) largestCount = count[s];
+    }
+
+    return largestCount;
+}
+
+typedef enum { trustInput, checkMaxSymbolValue } HIST_checkInput_e;
+
+/* HIST_count_parallel_wksp() :
+ * store histogram into 4 intermediate tables, recombined at the end.
+ * this design makes better use of OoO cpus,
+ * and is noticeably faster when some values are heavily repeated.
+ * But it needs some additional workspace for intermediate tables.
+ * `workSpace` size must be a table of size >= HIST_WKSP_SIZE_U32.
+ * @return : largest histogram frequency,
+ *           or an error code (notably when histogram would be larger than *maxSymbolValuePtr). */
+static size_t HIST_count_parallel_wksp(
+                                unsigned* count, unsigned* maxSymbolValuePtr,
+                                const void* source, size_t sourceSize,
+                                HIST_checkInput_e check,
+                                U32* const workSpace)
+{
+    const BYTE* ip = (const BYTE*)source;
+    const BYTE* const iend = ip+sourceSize;
+    unsigned maxSymbolValue = *maxSymbolValuePtr;
+    unsigned max=0;
+    U32* const Counting1 = workSpace;
+    U32* const Counting2 = Counting1 + 256;
+    U32* const Counting3 = Counting2 + 256;
+    U32* const Counting4 = Counting3 + 256;
+
+    memset(workSpace, 0, 4*256*sizeof(unsigned));
+
+    /* safety checks */
+    if (!sourceSize) {
+        memset(count, 0, maxSymbolValue + 1);
+        *maxSymbolValuePtr = 0;
+        return 0;
+    }
+    if (!maxSymbolValue) maxSymbolValue = 255;            /* 0 == default */
+
+    /* by stripes of 16 bytes */
+    {   U32 cached = MEM_read32(ip); ip += 4;
+        while (ip < iend-15) {
+            U32 c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+            c = cached; cached = MEM_read32(ip); ip += 4;
+            Counting1[(BYTE) c     ]++;
+            Counting2[(BYTE)(c>>8) ]++;
+            Counting3[(BYTE)(c>>16)]++;
+            Counting4[       c>>24 ]++;
+        }
+        ip-=4;
+    }
+
+    /* finish last symbols */
+    while (ip<iend) Counting1[*ip++]++;
+
+    if (check) {   /* verify stats will fit into destination table */
+        U32 s; for (s=255; s>maxSymbolValue; s--) {
+            Counting1[s] += Counting2[s] + Counting3[s] + Counting4[s];
+            if (Counting1[s]) return ERROR(maxSymbolValue_tooSmall);
+    }   }
+
+    {   U32 s;
+        if (maxSymbolValue > 255) maxSymbolValue = 255;
+        for (s=0; s<=maxSymbolValue; s++) {
+            count[s] = Counting1[s] + Counting2[s] + Counting3[s] + Counting4[s];
+            if (count[s] > max) max = count[s];
+    }   }
+
+    while (!count[maxSymbolValue]) maxSymbolValue--;
+    *maxSymbolValuePtr = maxSymbolValue;
+    return (size_t)max;
+}
+
+/* HIST_countFast_wksp() :
+ * Same as HIST_countFast(), but using an externally provided scratch buffer.
+ * `workSpace` is a writable buffer which must be 4-bytes aligned,
+ * `workSpaceSize` must be >= HIST_WKSP_SIZE
+ */
+size_t HIST_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                          const void* source, size_t sourceSize,
+                          void* workSpace, size_t workSpaceSize)
+{
+    if (sourceSize < 1500) /* heuristic threshold */
+        return HIST_count_simple(count, maxSymbolValuePtr, source, sourceSize);
+    if ((size_t)workSpace & 3) return ERROR(GENERIC);  /* must be aligned on 4-bytes boundaries */
+    if (workSpaceSize < HIST_WKSP_SIZE) return ERROR(workSpace_tooSmall);
+    return HIST_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, trustInput, (U32*)workSpace);
+}
+
+/* fast variant (unsafe : won't check if src contains values beyond count[] limit) */
+size_t HIST_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
+                     const void* source, size_t sourceSize)
+{
+    unsigned tmpCounters[HIST_WKSP_SIZE_U32];
+    return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, tmpCounters, sizeof(tmpCounters));
+}
+
+/* HIST_count_wksp() :
+ * Same as HIST_count(), but using an externally provided scratch buffer.
+ * `workSpace` size must be table of >= HIST_WKSP_SIZE_U32 unsigned */
+size_t HIST_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                       const void* source, size_t sourceSize,
+                       void* workSpace, size_t workSpaceSize)
+{
+    if ((size_t)workSpace & 3) return ERROR(GENERIC);  /* must be aligned on 4-bytes boundaries */
+    if (workSpaceSize < HIST_WKSP_SIZE) return ERROR(workSpace_tooSmall);
+    if (*maxSymbolValuePtr < 255)
+        return HIST_count_parallel_wksp(count, maxSymbolValuePtr, source, sourceSize, checkMaxSymbolValue, (U32*)workSpace);
+    *maxSymbolValuePtr = 255;
+    return HIST_countFast_wksp(count, maxSymbolValuePtr, source, sourceSize, workSpace, workSpaceSize);
+}
+
+size_t HIST_count(unsigned* count, unsigned* maxSymbolValuePtr,
+                 const void* src, size_t srcSize)
+{
+    unsigned tmpCounters[HIST_WKSP_SIZE_U32];
+    return HIST_count_wksp(count, maxSymbolValuePtr, src, srcSize, tmpCounters, sizeof(tmpCounters));
+}
diff --git a/contrib/zstd/hist.h b/contrib/zstd/hist.h
new file mode 100644
index 000000000..77e3ec4fb
--- /dev/null
+++ b/contrib/zstd/hist.h
@@ -0,0 +1,75 @@
+/* ******************************************************************
+ * hist : Histogram functions
+ * part of Finite State Entropy project
+ * Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
+ *
+ *  You can contact the author at :
+ *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+****************************************************************** */
+
+/* --- dependencies --- */
+#include <stddef.h>   /* size_t */
+
+
+/* --- simple histogram functions --- */
+
+/*! HIST_count():
+ *  Provides the precise count of each byte within a table 'count'.
+ * 'count' is a table of unsigned int, of minimum size (*maxSymbolValuePtr+1).
+ *  Updates *maxSymbolValuePtr with actual largest symbol value detected.
+ * @return : count of the most frequent symbol (which isn't identified).
+ *           or an error code, which can be tested using HIST_isError().
+ *           note : if return == srcSize, there is only one symbol.
+ */
+size_t HIST_count(unsigned* count, unsigned* maxSymbolValuePtr,
+                  const void* src, size_t srcSize);
+
+unsigned HIST_isError(size_t code);  /**< tells if a return value is an error code */
+
+
+/* --- advanced histogram functions --- */
+
+#define HIST_WKSP_SIZE_U32 1024
+#define HIST_WKSP_SIZE    (HIST_WKSP_SIZE_U32 * sizeof(unsigned))
+/** HIST_count_wksp() :
+ *  Same as HIST_count(), but using an externally provided scratch buffer.
+ *  Benefit is this function will use very little stack space.
+ * `workSpace` is a writable buffer which must be 4-bytes aligned,
+ * `workSpaceSize` must be >= HIST_WKSP_SIZE
+ */
+size_t HIST_count_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                       const void* src, size_t srcSize,
+                       void* workSpace, size_t workSpaceSize);
+
+/** HIST_countFast() :
+ *  same as HIST_count(), but blindly trusts that all byte values within src are <= *maxSymbolValuePtr.
+ *  This function is unsafe, and will segfault if any value within `src` is `> *maxSymbolValuePtr`
+ */
+size_t HIST_countFast(unsigned* count, unsigned* maxSymbolValuePtr,
+                      const void* src, size_t srcSize);
+
+/** HIST_countFast_wksp() :
+ *  Same as HIST_countFast(), but using an externally provided scratch buffer.
+ * `workSpace` is a writable buffer which must be 4-bytes aligned,
+ * `workSpaceSize` must be >= HIST_WKSP_SIZE
+ */
+size_t HIST_countFast_wksp(unsigned* count, unsigned* maxSymbolValuePtr,
+                           const void* src, size_t srcSize,
+                           void* workSpace, size_t workSpaceSize);
+
+/*! HIST_count_simple() :
+ *  Same as HIST_countFast(), this function is unsafe,
+ *  and will segfault if any value within `src` is `> *maxSymbolValuePtr`.
+ *  It is also a bit slower for large inputs.
+ *  However, it does not need any additional memory (not even on stack).
+ * @return : count of the most frequent symbol.
+ *  Note this function doesn't produce any error (i.e. it must succeed).
+ */
+unsigned HIST_count_simple(unsigned* count, unsigned* maxSymbolValuePtr,
+                           const void* src, size_t srcSize);
diff --git a/contrib/zstd/huf.h b/contrib/zstd/huf.h
index 2b3015a84..ef432685d 100644
--- a/contrib/zstd/huf.h
+++ b/contrib/zstd/huf.h
@@ -1,35 +1,15 @@
 /* ******************************************************************
-   Huffman coder, part of New Generation Entropy library
-   header file
-   Copyright (C) 2013-2016, Yann Collet.
-
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
-
-   Redistribution and use in source and binary forms, with or without
-   modification, are permitted provided that the following conditions are
-   met:
-
-       * Redistributions of source code must retain the above copyright
-   notice, this list of conditions and the following disclaimer.
-       * Redistributions in binary form must reproduce the above
-   copyright notice, this list of conditions and the following disclaimer
-   in the documentation and/or other materials provided with the
-   distribution.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-   You can contact the author at :
-   - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ * huff0 huffman codec,
+ * part of Finite State Entropy library
+ * Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
+ *
+ * You can contact the author at :
+ * - Source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
 ****************************************************************** */
 
 #if defined (__cplusplus)
@@ -58,32 +38,32 @@ extern "C" {
 #endif
 
 
-/* *** simple functions *** */
-/**
-HUF_compress() :
-    Compress content from buffer 'src', of size 'srcSize', into buffer 'dst'.
-    'dst' buffer must be already allocated.
-    Compression runs faster if `dstCapacity` >= HUF_compressBound(srcSize).
-    `srcSize` must be <= `HUF_BLOCKSIZE_MAX` == 128 KB.
-    @return : size of compressed data (<= `dstCapacity`).
-    Special values : if return == 0, srcData is not compressible => Nothing is stored within dst !!!
-                     if return == 1, srcData is a single repeated byte symbol (RLE compression).
-                     if HUF_isError(return), compression failed (more details using HUF_getErrorName())
-*/
+/* ========================== */
+/* ***  simple functions  *** */
+/* ========================== */
+
+/** HUF_compress() :
+ *  Compress content from buffer 'src', of size 'srcSize', into buffer 'dst'.
+ * 'dst' buffer must be already allocated.
+ *  Compression runs faster if `dstCapacity` >= HUF_compressBound(srcSize).
+ * `srcSize` must be <= `HUF_BLOCKSIZE_MAX` == 128 KB.
+ * @return : size of compressed data (<= `dstCapacity`).
+ *  Special values : if return == 0, srcData is not compressible => Nothing is stored within dst !!!
+ *                   if HUF_isError(return), compression failed (more details using HUF_getErrorName())
+ */
 HUF_PUBLIC_API size_t HUF_compress(void* dst, size_t dstCapacity,
                              const void* src, size_t srcSize);
 
-/**
-HUF_decompress() :
-    Decompress HUF data from buffer 'cSrc', of size 'cSrcSize',
-    into already allocated buffer 'dst', of minimum size 'dstSize'.
-    `originalSize` : **must** be the ***exact*** size of original (uncompressed) data.
-    Note : in contrast with FSE, HUF_decompress can regenerate
-           RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
-           because it knows size to regenerate.
-    @return : size of regenerated data (== originalSize),
-              or an error code, which can be tested using HUF_isError()
-*/
+/** HUF_decompress() :
+ *  Decompress HUF data from buffer 'cSrc', of size 'cSrcSize',
+ *  into already allocated buffer 'dst', of minimum size 'dstSize'.
+ * `originalSize` : **must** be the ***exact*** size of original (uncompressed) data.
+ *  Note : in contrast with FSE, HUF_decompress can regenerate
+ *         RLE (cSrcSize==1) and uncompressed (cSrcSize==dstSize) data,
+ *         because it knows size to regenerate (originalSize).
+ * @return : size of regenerated data (== originalSize),
+ *           or an error code, which can be tested using HUF_isError()
+ */
 HUF_PUBLIC_API size_t HUF_decompress(void* dst,  size_t originalSize,
                                const void* cSrc, size_t cSrcSize);
 
@@ -100,39 +80,32 @@ HUF_PUBLIC_API const char* HUF_getErrorName(size_t code);  /**< provides error c
 /* ***   Advanced function   *** */
 
 /** HUF_compress2() :
- *  Same as HUF_compress(), but offers direct control over `maxSymbolValue` and `tableLog`.
- *  `tableLog` must be `<= HUF_TABLELOG_MAX` . */
-HUF_PUBLIC_API size_t HUF_compress2 (void* dst, size_t dstCapacity, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
+ *  Same as HUF_compress(), but offers control over `maxSymbolValue` and `tableLog`.
+ * `maxSymbolValue` must be <= HUF_SYMBOLVALUE_MAX .
+ * `tableLog` must be `<= HUF_TABLELOG_MAX` . */
+HUF_PUBLIC_API size_t HUF_compress2 (void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                               unsigned maxSymbolValue, unsigned tableLog);
 
 /** HUF_compress4X_wksp() :
  *  Same as HUF_compress2(), but uses externally allocated `workSpace`.
- *  `workspace` must have minimum alignment of 4, and be at least as large as following macro */
-#define HUF_WORKSPACE_SIZE (6 << 10)
+ * `workspace` must have minimum alignment of 4, and be at least as large as HUF_WORKSPACE_SIZE */
+#define HUF_WORKSPACE_SIZE ((6 << 10) + 256)
 #define HUF_WORKSPACE_SIZE_U32 (HUF_WORKSPACE_SIZE / sizeof(U32))
-HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);
-
-/**
- *  The minimum workspace size for the `workSpace` used in
- *  HUF_readDTableX2_wksp() and HUF_readDTableX4_wksp().
- *
- *  The space used depends on HUF_TABLELOG_MAX, ranging from ~1500 bytes when
- *  HUF_TABLE_LOG_MAX=12 to ~1850 bytes when HUF_TABLE_LOG_MAX=15.
- *  Buffer overflow errors may potentially occur if code modifications result in
- *  a required workspace size greater than that specified in the following
- *  macro.
- */
-#define HUF_DECOMPRESS_WORKSPACE_SIZE (2 << 10)
-#define HUF_DECOMPRESS_WORKSPACE_SIZE_U32 (HUF_DECOMPRESS_WORKSPACE_SIZE / sizeof(U32))
+HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity,
+                                     const void* src, size_t srcSize,
+                                     unsigned maxSymbolValue, unsigned tableLog,
+                                     void* workSpace, size_t wkspSize);
 
 #endif   /* HUF_H_298734234 */
 
 /* ******************************************************************
  *  WARNING !!
  *  The following section contains advanced and experimental definitions
- *  which shall never be used in the context of dll
+ *  which shall never be used in the context of a dynamic library,
  *  because they are not guaranteed to remain stable in the future.
  *  Only consider them in association with static linking.
- *******************************************************************/
+ * *****************************************************************/
 #if defined(HUF_STATIC_LINKING_ONLY) && !defined(HUF_H_HUF_STATIC_LINKING_ONLY)
 #define HUF_H_HUF_STATIC_LINKING_ONLY
 
@@ -141,11 +114,11 @@ HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity, const
 
 
 /* *** Constants *** */
-#define HUF_TABLELOG_MAX      12       /* max configured tableLog (for static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
-#define HUF_TABLELOG_DEFAULT  11       /* tableLog by default, when not specified */
+#define HUF_TABLELOG_MAX      12      /* max runtime value of tableLog (due to static allocation); can be modified up to HUF_ABSOLUTEMAX_TABLELOG */
+#define HUF_TABLELOG_DEFAULT  11      /* default tableLog value when none specified */
 #define HUF_SYMBOLVALUE_MAX  255
 
-#define HUF_TABLELOG_ABSOLUTEMAX  15   /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
+#define HUF_TABLELOG_ABSOLUTEMAX  15  /* absolute limit of HUF_MAX_TABLELOG. Beyond that value, code does not work */
 #if (HUF_TABLELOG_MAX > HUF_TABLELOG_ABSOLUTEMAX)
 #  error "HUF_TABLELOG_MAX is too large !"
 #endif
@@ -170,130 +143,195 @@ HUF_PUBLIC_API size_t HUF_compress4X_wksp (void* dst, size_t dstCapacity, const
 /* static allocation of HUF's DTable */
 typedef U32 HUF_DTable;
 #define HUF_DTABLE_SIZE(maxTableLog)   (1 + (1<<(maxTableLog)))
-#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
+#define HUF_CREATE_STATIC_DTABLEX1(DTable, maxTableLog) \
         HUF_DTable DTable[HUF_DTABLE_SIZE((maxTableLog)-1)] = { ((U32)((maxTableLog)-1) * 0x01000001) }
-#define HUF_CREATE_STATIC_DTABLEX4(DTable, maxTableLog) \
+#define HUF_CREATE_STATIC_DTABLEX2(DTable, maxTableLog) \
         HUF_DTable DTable[HUF_DTABLE_SIZE(maxTableLog)] = { ((U32)(maxTableLog) * 0x01000001) }
 
 
 /* ****************************************
 *  Advanced decompression functions
 ******************************************/
-size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
-size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+size_t HUF_decompress4X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+#ifndef HUF_FORCE_DECOMPRESS_X1
+size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+#endif
 
 size_t HUF_decompress4X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< decodes RLE and uncompressed */
 size_t HUF_decompress4X_hufOnly(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize); /**< considers RLE and uncompressed as errors */
 size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize); /**< considers RLE and uncompressed as errors */
-size_t HUF_decompress4X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
-size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< single-symbol decoder */
-size_t HUF_decompress4X4_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
-size_t HUF_decompress4X4_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< double-symbols decoder */
+size_t HUF_decompress4X1_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUF_decompress4X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< single-symbol decoder */
+#ifndef HUF_FORCE_DECOMPRESS_X1
+size_t HUF_decompress4X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< double-symbols decoder */
+#endif
 
 
 /* ****************************************
-*  HUF detailed API
-******************************************/
-/*!
-HUF_compress() does the following:
-1. count symbol occurrence from source[] into table count[] using FSE_count()
-2. (optional) refine tableLog using HUF_optimalTableLog()
-3. build Huffman table from count using HUF_buildCTable()
-4. save Huffman table to memory buffer using HUF_writeCTable()
-5. encode the data stream using HUF_compress4X_usingCTable()
-
-The following API allows targeting specific sub-functions for advanced tasks.
-For example, it's possible to compress several blocks using the same 'CTable',
-or to save and regenerate 'CTable' using external methods.
-*/
-/* FSE_count() : find it within "fse.h" */
+ *  HUF detailed API
+ * ****************************************/
+
+/*! HUF_compress() does the following:
+ *  1. count symbol occurrence from source[] into table count[] using FSE_count() (exposed within "fse.h")
+ *  2. (optional) refine tableLog using HUF_optimalTableLog()
+ *  3. build Huffman table from count using HUF_buildCTable()
+ *  4. save Huffman table to memory buffer using HUF_writeCTable()
+ *  5. encode the data stream using HUF_compress4X_usingCTable()
+ *
+ *  The following API allows targeting specific sub-functions for advanced tasks.
+ *  For example, it's possible to compress several blocks using the same 'CTable',
+ *  or to save and regenerate 'CTable' using external methods.
+ */
 unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue);
 typedef struct HUF_CElt_s HUF_CElt;   /* incomplete type */
-size_t HUF_buildCTable (HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue, unsigned maxNbBits);
+size_t HUF_buildCTable (HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue, unsigned maxNbBits);   /* @return : maxNbBits; CTable and count can overlap. In which case, CTable will overwrite count content */
 size_t HUF_writeCTable (void* dst, size_t maxDstSize, const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog);
 size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable);
+size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue);
+int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue);
 
 typedef enum {
    HUF_repeat_none,  /**< Cannot use the previous table */
    HUF_repeat_check, /**< Can use the previous table but it must be checked. Note : The previous table must have been constructed by HUF_compress{1, 4}X_repeat */
-   HUF_repeat_valid  /**< Can use the previous table and it is asumed to be valid */
+   HUF_repeat_valid  /**< Can use the previous table and it is assumed to be valid */
  } HUF_repeat;
 /** HUF_compress4X_repeat() :
-*   Same as HUF_compress4X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
-*   If it uses hufTable it does not modify hufTable or repeat.
-*   If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
-*   If preferRepeat then the old table will always be used if valid. */
-size_t HUF_compress4X_repeat(void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize, HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat);  /**< `workSpace` must be a table of at least HUF_WORKSPACE_SIZE_U32 unsigned */
+ *  Same as HUF_compress4X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
+ *  If it uses hufTable it does not modify hufTable or repeat.
+ *  If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
+ *  If preferRepeat then the old table will always be used if valid. */
+size_t HUF_compress4X_repeat(void* dst, size_t dstSize,
+                       const void* src, size_t srcSize,
+                       unsigned maxSymbolValue, unsigned tableLog,
+                       void* workSpace, size_t wkspSize,    /**< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
+                       HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2);
 
 /** HUF_buildCTable_wksp() :
  *  Same as HUF_buildCTable(), but using externally allocated scratch buffer.
- *  `workSpace` must be aligned on 4-bytes boundaries, and be at least as large as a table of 1024 unsigned.
+ * `workSpace` must be aligned on 4-bytes boundaries, and its size must be >= HUF_CTABLE_WORKSPACE_SIZE.
  */
-size_t HUF_buildCTable_wksp (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize);
+#define HUF_CTABLE_WORKSPACE_SIZE_U32 (2*HUF_SYMBOLVALUE_MAX +1 +1)
+#define HUF_CTABLE_WORKSPACE_SIZE (HUF_CTABLE_WORKSPACE_SIZE_U32 * sizeof(unsigned))
+size_t HUF_buildCTable_wksp (HUF_CElt* tree,
+                       const unsigned* count, U32 maxSymbolValue, U32 maxNbBits,
+                             void* workSpace, size_t wkspSize);
 
 /*! HUF_readStats() :
-    Read compact Huffman tree, saved by HUF_writeCTable().
-    `huffWeight` is destination buffer.
-    @return : size read from `src` , or an error Code .
-    Note : Needed by HUF_readCTable() and HUF_readDTableXn() . */
-size_t HUF_readStats(BYTE* huffWeight, size_t hwSize, U32* rankStats,
-                     U32* nbSymbolsPtr, U32* tableLogPtr,
+ *  Read compact Huffman tree, saved by HUF_writeCTable().
+ * `huffWeight` is destination buffer.
+ * @return : size read from `src` , or an error Code .
+ *  Note : Needed by HUF_readCTable() and HUF_readDTableXn() . */
+size_t HUF_readStats(BYTE* huffWeight, size_t hwSize,
+                     U32* rankStats, U32* nbSymbolsPtr, U32* tableLogPtr,
                      const void* src, size_t srcSize);
 
 /** HUF_readCTable() :
-*   Loading a CTable saved with HUF_writeCTable() */
-size_t HUF_readCTable (HUF_CElt* CTable, unsigned maxSymbolValue, const void* src, size_t srcSize);
+ *  Loading a CTable saved with HUF_writeCTable() */
+size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned *hasZeroWeights);
 
+/** HUF_getNbBits() :
+ *  Read nbBits from CTable symbolTable, for symbol `symbolValue` presumed <= HUF_SYMBOLVALUE_MAX
+ *  Note 1 : is not inlined, as HUF_CElt definition is private
+ *  Note 2 : const void* used, so that it can provide a statically allocated table as argument (which uses type U32) */
+U32 HUF_getNbBits(const void* symbolTable, U32 symbolValue);
 
 /*
-HUF_decompress() does the following:
-1. select the decompression algorithm (X2, X4) based on pre-computed heuristics
-2. build Huffman table from save, using HUF_readDTableXn()
-3. decode 1 or 4 segments in parallel using HUF_decompressSXn_usingDTable
-*/
+ * HUF_decompress() does the following:
+ * 1. select the decompression algorithm (X1, X2) based on pre-computed heuristics
+ * 2. build Huffman table from save, using HUF_readDTableX?()
+ * 3. decode 1 or 4 segments in parallel using HUF_decompress?X?_usingDTable()
+ */
 
 /** HUF_selectDecoder() :
-*   Tells which decoder is likely to decode faster,
-*   based on a set of pre-determined metrics.
-*   @return : 0==HUF_decompress4X2, 1==HUF_decompress4X4 .
-*   Assumption : 0 < cSrcSize < dstSize <= 128 KB */
+ *  Tells which decoder is likely to decode faster,
+ *  based on a set of pre-computed metrics.
+ * @return : 0==HUF_decompress4X1, 1==HUF_decompress4X2 .
+ *  Assumption : 0 < dstSize <= 128 KB */
 U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize);
 
+/**
+ *  The minimum workspace size for the `workSpace` used in
+ *  HUF_readDTableX1_wksp() and HUF_readDTableX2_wksp().
+ *
+ *  The space used depends on HUF_TABLELOG_MAX, ranging from ~1500 bytes when
+ *  HUF_TABLE_LOG_MAX=12 to ~1850 bytes when HUF_TABLE_LOG_MAX=15.
+ *  Buffer overflow errors may potentially occur if code modifications result in
+ *  a required workspace size greater than that specified in the following
+ *  macro.
+ */
+#define HUF_DECOMPRESS_WORKSPACE_SIZE (2 << 10)
+#define HUF_DECOMPRESS_WORKSPACE_SIZE_U32 (HUF_DECOMPRESS_WORKSPACE_SIZE / sizeof(U32))
+
+#ifndef HUF_FORCE_DECOMPRESS_X2
+size_t HUF_readDTableX1 (HUF_DTable* DTable, const void* src, size_t srcSize);
+size_t HUF_readDTableX1_wksp (HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize);
+#endif
+#ifndef HUF_FORCE_DECOMPRESS_X1
 size_t HUF_readDTableX2 (HUF_DTable* DTable, const void* src, size_t srcSize);
 size_t HUF_readDTableX2_wksp (HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize);
-size_t HUF_readDTableX4 (HUF_DTable* DTable, const void* src, size_t srcSize);
-size_t HUF_readDTableX4_wksp (HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize);
+#endif
 
 size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
+#ifndef HUF_FORCE_DECOMPRESS_X2
+size_t HUF_decompress4X1_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
+#endif
+#ifndef HUF_FORCE_DECOMPRESS_X1
 size_t HUF_decompress4X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
-size_t HUF_decompress4X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
+#endif
 
 
+/* ====================== */
 /* single stream variants */
+/* ====================== */
 
 size_t HUF_compress1X (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);
 size_t HUF_compress1X_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);  /**< `workSpace` must be a table of at least HUF_WORKSPACE_SIZE_U32 unsigned */
 size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable);
 /** HUF_compress1X_repeat() :
-*   Same as HUF_compress1X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
-*   If it uses hufTable it does not modify hufTable or repeat.
-*   If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
-*   If preferRepeat then the old table will always be used if valid. */
-size_t HUF_compress1X_repeat(void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize, HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat);  /**< `workSpace` must be a table of at least HUF_WORKSPACE_SIZE_U32 unsigned */
-
-size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
-size_t HUF_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
+ *  Same as HUF_compress1X_wksp(), but considers using hufTable if *repeat != HUF_repeat_none.
+ *  If it uses hufTable it does not modify hufTable or repeat.
+ *  If it doesn't, it sets *repeat = HUF_repeat_none, and it sets hufTable to the table used.
+ *  If preferRepeat then the old table will always be used if valid. */
+size_t HUF_compress1X_repeat(void* dst, size_t dstSize,
+                       const void* src, size_t srcSize,
+                       unsigned maxSymbolValue, unsigned tableLog,
+                       void* workSpace, size_t wkspSize,   /**< `workSpace` must be aligned on 4-bytes boundaries, `wkspSize` must be >= HUF_WORKSPACE_SIZE */
+                       HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2);
+
+size_t HUF_decompress1X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* single-symbol decoder */
+#ifndef HUF_FORCE_DECOMPRESS_X1
+size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /* double-symbol decoder */
+#endif
 
 size_t HUF_decompress1X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);
 size_t HUF_decompress1X_DCtx_wksp (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);
-size_t HUF_decompress1X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
-size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< single-symbol decoder */
-size_t HUF_decompress1X4_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
-size_t HUF_decompress1X4_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< double-symbols decoder */
+#ifndef HUF_FORCE_DECOMPRESS_X2
+size_t HUF_decompress1X1_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< single-symbol decoder */
+size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< single-symbol decoder */
+#endif
+#ifndef HUF_FORCE_DECOMPRESS_X1
+size_t HUF_decompress1X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize);   /**< double-symbols decoder */
+size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize);   /**< double-symbols decoder */
+#endif
 
 size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);   /**< automatic selection of sing or double symbol decoder, based on DTable */
+#ifndef HUF_FORCE_DECOMPRESS_X2
+size_t HUF_decompress1X1_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
+#endif
+#ifndef HUF_FORCE_DECOMPRESS_X1
 size_t HUF_decompress1X2_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
-size_t HUF_decompress1X4_usingDTable(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable);
+#endif
+
+/* BMI2 variants.
+ * If the CPU has BMI2 support, pass bmi2=1, otherwise pass bmi2=0.
+ */
+size_t HUF_decompress1X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2);
+#ifndef HUF_FORCE_DECOMPRESS_X2
+size_t HUF_decompress1X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2);
+#endif
+size_t HUF_decompress4X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2);
+size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2);
 
 #endif /* HUF_STATIC_LINKING_ONLY */
 
diff --git a/contrib/zstd/huf_compress.c b/contrib/zstd/huf_compress.c
index 2a47c1820..4efffbb54 100644
--- a/contrib/zstd/huf_compress.c
+++ b/contrib/zstd/huf_compress.c
@@ -1,35 +1,15 @@
 /* ******************************************************************
-   Huffman encoder, part of New Generation Entropy library
-   Copyright (C) 2013-2016, Yann Collet.
-
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
-
-   Redistribution and use in source and binary forms, with or without
-   modification, are permitted provided that the following conditions are
-   met:
-
-       * Redistributions of source code must retain the above copyright
-   notice, this list of conditions and the following disclaimer.
-       * Redistributions in binary form must reproduce the above
-   copyright notice, this list of conditions and the following disclaimer
-   in the documentation and/or other materials provided with the
-   distribution.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-    You can contact the author at :
-    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
-    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ * Huffman encoder, part of New Generation Entropy library
+ * Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
+ *
+ *  You can contact the author at :
+ *  - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
 ****************************************************************** */
 
 /* **************************************************************
@@ -45,7 +25,9 @@
 ****************************************************************/
 #include <string.h>     /* memcpy, memset */
 #include <stdio.h>      /* printf (debug) */
+#include "compiler.h"
 #include "bitstream.h"
+#include "hist.h"
 #define FSE_STATIC_LINKING_ONLY   /* FSE_optimalTableLog_internal */
 #include "fse.h"        /* header compression */
 #define HUF_STATIC_LINKING_ONLY
@@ -57,9 +39,7 @@
 *  Error Management
 ****************************************************************/
 #define HUF_isError ERR_isError
-#define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
-#define CHECK_V_F(e, f) size_t const e = f; if (ERR_isError(e)) return e
-#define CHECK_F(f)   { CHECK_V_F(_var_err__, f); }
+#define HUF_STATIC_ASSERT(c) DEBUG_STATIC_ASSERT(c)   /* use only *after* variable declarations */
 
 
 /* **************************************************************
@@ -80,46 +60,46 @@ unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxS
  * Note : all elements within weightTable are supposed to be <= HUF_TABLELOG_MAX.
  */
 #define MAX_FSE_TABLELOG_FOR_HUFF_HEADER 6
-size_t HUF_compressWeights (void* dst, size_t dstSize, const void* weightTable, size_t wtSize)
+static size_t HUF_compressWeights (void* dst, size_t dstSize, const void* weightTable, size_t wtSize)
 {
     BYTE* const ostart = (BYTE*) dst;
     BYTE* op = ostart;
     BYTE* const oend = ostart + dstSize;
 
-    U32 maxSymbolValue = HUF_TABLELOG_MAX;
+    unsigned maxSymbolValue = HUF_TABLELOG_MAX;
     U32 tableLog = MAX_FSE_TABLELOG_FOR_HUFF_HEADER;
 
     FSE_CTable CTable[FSE_CTABLE_SIZE_U32(MAX_FSE_TABLELOG_FOR_HUFF_HEADER, HUF_TABLELOG_MAX)];
     BYTE scratchBuffer[1<<MAX_FSE_TABLELOG_FOR_HUFF_HEADER];
 
-    U32 count[HUF_TABLELOG_MAX+1];
+    unsigned count[HUF_TABLELOG_MAX+1];
     S16 norm[HUF_TABLELOG_MAX+1];
 
     /* init conditions */
     if (wtSize <= 1) return 0;  /* Not compressible */
 
     /* Scan input and build symbol stats */
-    {   CHECK_V_F(maxCount, FSE_count_simple(count, &maxSymbolValue, weightTable, wtSize) );
+    {   unsigned const maxCount = HIST_count_simple(count, &maxSymbolValue, weightTable, wtSize);   /* never fails */
         if (maxCount == wtSize) return 1;   /* only a single symbol in src : rle */
-        if (maxCount == 1) return 0;         /* each symbol present maximum once => not compressible */
+        if (maxCount == 1) return 0;        /* each symbol present maximum once => not compressible */
     }
 
     tableLog = FSE_optimalTableLog(tableLog, wtSize, maxSymbolValue);
     CHECK_F( FSE_normalizeCount(norm, tableLog, count, wtSize, maxSymbolValue) );
 
     /* Write table description header */
-    {   CHECK_V_F(hSize, FSE_writeNCount(op, oend-op, norm, maxSymbolValue, tableLog) );
+    {   CHECK_V_F(hSize, FSE_writeNCount(op, (size_t)(oend-op), norm, maxSymbolValue, tableLog) );
         op += hSize;
     }
 
     /* Compress */
     CHECK_F( FSE_buildCTable_wksp(CTable, norm, maxSymbolValue, tableLog, scratchBuffer, sizeof(scratchBuffer)) );
-    {   CHECK_V_F(cSize, FSE_compress_usingCTable(op, oend - op, weightTable, wtSize, CTable) );
+    {   CHECK_V_F(cSize, FSE_compress_usingCTable(op, (size_t)(oend - op), weightTable, wtSize, CTable) );
         if (cSize == 0) return 0;   /* not enough space for compressed data */
         op += cSize;
     }
 
-    return op-ostart;
+    return (size_t)(op-ostart);
 }
 
 
@@ -132,7 +112,7 @@ struct HUF_CElt_s {
     `CTable` : Huffman tree to save, using huf representation.
     @return : size of saved CTable */
 size_t HUF_writeCTable (void* dst, size_t maxDstSize,
-                        const HUF_CElt* CTable, U32 maxSymbolValue, U32 huffLog)
+                        const HUF_CElt* CTable, unsigned maxSymbolValue, unsigned huffLog)
 {
     BYTE bitsToWeight[HUF_TABLELOG_MAX + 1];   /* precomputed conversion table */
     BYTE huffWeight[HUF_SYMBOLVALUE_MAX];
@@ -167,7 +147,7 @@ size_t HUF_writeCTable (void* dst, size_t maxDstSize,
 }
 
 
-size_t HUF_readCTable (HUF_CElt* CTable, U32 maxSymbolValue, const void* src, size_t srcSize)
+size_t HUF_readCTable (HUF_CElt* CTable, unsigned* maxSymbolValuePtr, const void* src, size_t srcSize, unsigned* hasZeroWeights)
 {
     BYTE huffWeight[HUF_SYMBOLVALUE_MAX + 1];   /* init not required, even though some static analyzer may complain */
     U32 rankVal[HUF_TABLELOG_ABSOLUTEMAX + 1];   /* large enough for values from 0 to 16 */
@@ -179,7 +159,7 @@ size_t HUF_readCTable (HUF_CElt* CTable, U32 maxSymbolValue, const void* src, si
 
     /* check result */
     if (tableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
-    if (nbSymbols > maxSymbolValue+1) return ERROR(maxSymbolValue_tooSmall);
+    if (nbSymbols > *maxSymbolValuePtr+1) return ERROR(maxSymbolValue_tooSmall);
 
     /* Prepare base value per rank */
     {   U32 n, nextRankStart = 0;
@@ -190,9 +170,11 @@ size_t HUF_readCTable (HUF_CElt* CTable, U32 maxSymbolValue, const void* src, si
     }   }
 
     /* fill nbBits */
+    *hasZeroWeights = 0;
     {   U32 n; for (n=0; n<nbSymbols; n++) {
             const U32 w = huffWeight[n];
-            CTable[n].nbBits = (BYTE)(tableLog + 1 - w);
+            *hasZeroWeights |= (w == 0);
+            CTable[n].nbBits = (BYTE)(tableLog + 1 - w) & -(w != 0);
     }   }
 
     /* fill val */
@@ -208,12 +190,20 @@ size_t HUF_readCTable (HUF_CElt* CTable, U32 maxSymbolValue, const void* src, si
                 min >>= 1;
         }   }
         /* assign value within rank, symbol order */
-        { U32 n; for (n=0; n<=maxSymbolValue; n++) CTable[n].val = valPerRank[CTable[n].nbBits]++; }
+        { U32 n; for (n=0; n<nbSymbols; n++) CTable[n].val = valPerRank[CTable[n].nbBits]++; }
     }
 
+    *maxSymbolValuePtr = nbSymbols - 1;
     return readSize;
 }
 
+U32 HUF_getNbBits(const void* symbolTable, U32 symbolValue)
+{
+    const HUF_CElt* table = (const HUF_CElt*)symbolTable;
+    assert(symbolValue <= HUF_SYMBOLVALUE_MAX);
+    return table[symbolValue].nbBits;
+}
+
 
 typedef struct nodeElt_s {
     U32 count;
@@ -230,7 +220,7 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
     /* there are several too large elements (at least >= 2) */
     {   int totalCost = 0;
         const U32 baseCost = 1 << (largestBits - maxNbBits);
-        U32 n = lastNonNull;
+        int n = (int)lastNonNull;
 
         while (huffNode[n].nbBits > maxNbBits) {
             totalCost += baseCost - (1 << (largestBits - huffNode[n].nbBits));
@@ -245,22 +235,22 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
         /* repay normalized cost */
         {   U32 const noSymbol = 0xF0F0F0F0;
             U32 rankLast[HUF_TABLELOG_MAX+2];
-            int pos;
 
             /* Get pos of last (smallest) symbol per rank */
             memset(rankLast, 0xF0, sizeof(rankLast));
             {   U32 currentNbBits = maxNbBits;
+                int pos;
                 for (pos=n ; pos >= 0; pos--) {
                     if (huffNode[pos].nbBits >= currentNbBits) continue;
                     currentNbBits = huffNode[pos].nbBits;   /* < maxNbBits */
-                    rankLast[maxNbBits-currentNbBits] = pos;
+                    rankLast[maxNbBits-currentNbBits] = (U32)pos;
             }   }
 
             while (totalCost > 0) {
-                U32 nBitsToDecrease = BIT_highbit32(totalCost) + 1;
+                U32 nBitsToDecrease = BIT_highbit32((U32)totalCost) + 1;
                 for ( ; nBitsToDecrease > 1; nBitsToDecrease--) {
-                    U32 highPos = rankLast[nBitsToDecrease];
-                    U32 lowPos = rankLast[nBitsToDecrease-1];
+                    U32 const highPos = rankLast[nBitsToDecrease];
+                    U32 const lowPos = rankLast[nBitsToDecrease-1];
                     if (highPos == noSymbol) continue;
                     if (lowPos == noSymbol) break;
                     {   U32 const highTotal = huffNode[highPos].count;
@@ -287,7 +277,8 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
                 if (rankLast[1] == noSymbol) {  /* special case : no rank 1 symbol (using maxNbBits-1); let's create one from largest rank 0 (using maxNbBits) */
                     while (huffNode[n].nbBits == maxNbBits) n--;
                     huffNode[n+1].nbBits--;
-                    rankLast[1] = n+1;
+                    assert(n >= 0);
+                    rankLast[1] = (U32)(n+1);
                     totalCost++;
                     continue;
                 }
@@ -299,29 +290,39 @@ static U32 HUF_setMaxHeight(nodeElt* huffNode, U32 lastNonNull, U32 maxNbBits)
     return maxNbBits;
 }
 
-
 typedef struct {
     U32 base;
     U32 current;
 } rankPos;
 
-static void HUF_sort(nodeElt* huffNode, const U32* count, U32 maxSymbolValue)
+typedef nodeElt huffNodeTable[HUF_CTABLE_WORKSPACE_SIZE_U32];
+
+#define RANK_POSITION_TABLE_SIZE 32
+
+typedef struct {
+  huffNodeTable huffNodeTbl;
+  rankPos rankPosition[RANK_POSITION_TABLE_SIZE];
+} HUF_buildCTable_wksp_tables;
+
+static void HUF_sort(nodeElt* huffNode, const unsigned* count, U32 maxSymbolValue, rankPos* rankPosition)
 {
-    rankPos rank[32];
     U32 n;
 
-    memset(rank, 0, sizeof(rank));
+    memset(rankPosition, 0, sizeof(*rankPosition) * RANK_POSITION_TABLE_SIZE);
     for (n=0; n<=maxSymbolValue; n++) {
         U32 r = BIT_highbit32(count[n] + 1);
-        rank[r].base ++;
+        rankPosition[r].base ++;
     }
-    for (n=30; n>0; n--) rank[n-1].base += rank[n].base;
-    for (n=0; n<32; n++) rank[n].current = rank[n].base;
+    for (n=30; n>0; n--) rankPosition[n-1].base += rankPosition[n].base;
+    for (n=0; n<32; n++) rankPosition[n].current = rankPosition[n].base;
     for (n=0; n<=maxSymbolValue; n++) {
         U32 const c = count[n];
         U32 const r = BIT_highbit32(c+1) + 1;
-        U32 pos = rank[r].current++;
-        while ((pos > rank[r].base) && (c > huffNode[pos-1].count)) huffNode[pos]=huffNode[pos-1], pos--;
+        U32 pos = rankPosition[r].current++;
+        while ((pos > rankPosition[r].base) && (c > huffNode[pos-1].count)) {
+            huffNode[pos] = huffNode[pos-1];
+            pos--;
+        }
         huffNode[pos].count = c;
         huffNode[pos].byte  = (BYTE)n;
     }
@@ -330,44 +331,48 @@ static void HUF_sort(nodeElt* huffNode, const U32* count, U32 maxSymbolValue)
 
 /** HUF_buildCTable_wksp() :
  *  Same as HUF_buildCTable(), but using externally allocated scratch buffer.
- *  `workSpace` must be aligned on 4-bytes boundaries, and be at least as large as a table of 1024 unsigned.
+ *  `workSpace` must be aligned on 4-bytes boundaries, and be at least as large as sizeof(HUF_buildCTable_wksp_tables).
  */
 #define STARTNODE (HUF_SYMBOLVALUE_MAX+1)
-typedef nodeElt huffNodeTable[2*HUF_SYMBOLVALUE_MAX+1 +1];
-size_t HUF_buildCTable_wksp (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize)
+
+size_t HUF_buildCTable_wksp (HUF_CElt* tree, const unsigned* count, U32 maxSymbolValue, U32 maxNbBits, void* workSpace, size_t wkspSize)
 {
-    nodeElt* const huffNode0 = (nodeElt*)workSpace;
+    HUF_buildCTable_wksp_tables* const wksp_tables = (HUF_buildCTable_wksp_tables*)workSpace;
+    nodeElt* const huffNode0 = wksp_tables->huffNodeTbl;
     nodeElt* const huffNode = huffNode0+1;
-    U32 n, nonNullRank;
+    int nonNullRank;
     int lowS, lowN;
-    U16 nodeNb = STARTNODE;
-    U32 nodeRoot;
+    int nodeNb = STARTNODE;
+    int n, nodeRoot;
 
     /* safety checks */
-    if (wkspSize < sizeof(huffNodeTable)) return ERROR(GENERIC);   /* workSpace is not large enough */
+    if (((size_t)workSpace & 3) != 0) return ERROR(GENERIC);  /* must be aligned on 4-bytes boundaries */
+    if (wkspSize < sizeof(HUF_buildCTable_wksp_tables))
+      return ERROR(workSpace_tooSmall);
     if (maxNbBits == 0) maxNbBits = HUF_TABLELOG_DEFAULT;
-    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(GENERIC);
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX)
+      return ERROR(maxSymbolValue_tooLarge);
     memset(huffNode0, 0, sizeof(huffNodeTable));
 
     /* sort, decreasing order */
-    HUF_sort(huffNode, count, maxSymbolValue);
+    HUF_sort(huffNode, count, maxSymbolValue, wksp_tables->rankPosition);
 
     /* init for parents */
-    nonNullRank = maxSymbolValue;
+    nonNullRank = (int)maxSymbolValue;
     while(huffNode[nonNullRank].count == 0) nonNullRank--;
     lowS = nonNullRank; nodeRoot = nodeNb + lowS - 1; lowN = nodeNb;
     huffNode[nodeNb].count = huffNode[lowS].count + huffNode[lowS-1].count;
-    huffNode[lowS].parent = huffNode[lowS-1].parent = nodeNb;
+    huffNode[lowS].parent = huffNode[lowS-1].parent = (U16)nodeNb;
     nodeNb++; lowS-=2;
     for (n=nodeNb; n<=nodeRoot; n++) huffNode[n].count = (U32)(1U<<30);
     huffNode0[0].count = (U32)(1U<<31);  /* fake entry, strong barrier */
 
     /* create parents */
     while (nodeNb <= nodeRoot) {
-        U32 n1 = (huffNode[lowS].count < huffNode[lowN].count) ? lowS-- : lowN++;
-        U32 n2 = (huffNode[lowS].count < huffNode[lowN].count) ? lowS-- : lowN++;
+        int const n1 = (huffNode[lowS].count < huffNode[lowN].count) ? lowS-- : lowN++;
+        int const n2 = (huffNode[lowS].count < huffNode[lowN].count) ? lowS-- : lowN++;
         huffNode[nodeNb].count = huffNode[n1].count + huffNode[n2].count;
-        huffNode[n1].parent = huffNode[n2].parent = nodeNb;
+        huffNode[n1].parent = huffNode[n2].parent = (U16)nodeNb;
         nodeNb++;
     }
 
@@ -379,24 +384,25 @@ size_t HUF_buildCTable_wksp (HUF_CElt* tree, const U32* count, U32 maxSymbolValu
         huffNode[n].nbBits = huffNode[ huffNode[n].parent ].nbBits + 1;
 
     /* enforce maxTableLog */
-    maxNbBits = HUF_setMaxHeight(huffNode, nonNullRank, maxNbBits);
+    maxNbBits = HUF_setMaxHeight(huffNode, (U32)nonNullRank, maxNbBits);
 
     /* fill result into tree (val, nbBits) */
     {   U16 nbPerRank[HUF_TABLELOG_MAX+1] = {0};
         U16 valPerRank[HUF_TABLELOG_MAX+1] = {0};
+        int const alphabetSize = (int)(maxSymbolValue + 1);
         if (maxNbBits > HUF_TABLELOG_MAX) return ERROR(GENERIC);   /* check fit into table */
         for (n=0; n<=nonNullRank; n++)
             nbPerRank[huffNode[n].nbBits]++;
         /* determine stating value per rank */
         {   U16 min = 0;
-            for (n=maxNbBits; n>0; n--) {
+            for (n=(int)maxNbBits; n>0; n--) {
                 valPerRank[n] = min;      /* get starting value within each rank */
                 min += nbPerRank[n];
                 min >>= 1;
         }   }
-        for (n=0; n<=maxSymbolValue; n++)
+        for (n=0; n<alphabetSize; n++)
             tree[huffNode[n].byte].nbBits = huffNode[n].nbBits;   /* push nbBits per symbol, symbol order */
-        for (n=0; n<=maxSymbolValue; n++)
+        for (n=0; n<alphabetSize; n++)
             tree[n].val = valPerRank[tree[n].nbBits]++;   /* assign value within rank, symbol order */
     }
 
@@ -404,15 +410,16 @@ size_t HUF_buildCTable_wksp (HUF_CElt* tree, const U32* count, U32 maxSymbolValu
 }
 
 /** HUF_buildCTable() :
+ * @return : maxNbBits
  *  Note : count is used before tree is written, so they can safely overlap
  */
-size_t HUF_buildCTable (HUF_CElt* tree, const U32* count, U32 maxSymbolValue, U32 maxNbBits)
+size_t HUF_buildCTable (HUF_CElt* tree, const unsigned* count, unsigned maxSymbolValue, unsigned maxNbBits)
 {
-    huffNodeTable nodeTable;
-    return HUF_buildCTable_wksp(tree, count, maxSymbolValue, maxNbBits, nodeTable, sizeof(nodeTable));
+    HUF_buildCTable_wksp_tables workspace;
+    return HUF_buildCTable_wksp(tree, count, maxSymbolValue, maxNbBits, &workspace, sizeof(workspace));
 }
 
-static size_t HUF_estimateCompressedSize(HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue)
+size_t HUF_estimateCompressedSize(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue)
 {
     size_t nbBits = 0;
     int s;
@@ -422,7 +429,7 @@ static size_t HUF_estimateCompressedSize(HUF_CElt* CTable, const unsigned* count
     return nbBits >> 3;
 }
 
-static int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue) {
+int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, unsigned maxSymbolValue) {
   int bad = 0;
   int s;
   for (s = 0; s <= (int)maxSymbolValue; ++s) {
@@ -431,13 +438,14 @@ static int HUF_validateCTable(const HUF_CElt* CTable, const unsigned* count, uns
   return !bad;
 }
 
-static void HUF_encodeSymbol(BIT_CStream_t* bitCPtr, U32 symbol, const HUF_CElt* CTable)
+size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
+
+FORCE_INLINE_TEMPLATE void
+HUF_encodeSymbol(BIT_CStream_t* bitCPtr, U32 symbol, const HUF_CElt* CTable)
 {
     BIT_addBitsFast(bitCPtr, CTable[symbol].val, CTable[symbol].nbBits);
 }
 
-size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
-
 #define HUF_FLUSHBITS(s)  BIT_flushBits(s)
 
 #define HUF_FLUSHBITS_1(stream) \
@@ -446,7 +454,10 @@ size_t HUF_compressBound(size_t size) { return HUF_COMPRESSBOUND(size); }
 #define HUF_FLUSHBITS_2(stream) \
     if (sizeof((stream)->bitContainer)*8 < HUF_TABLELOG_MAX*4+7) HUF_FLUSHBITS(stream)
 
-size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
+FORCE_INLINE_TEMPLATE size_t
+HUF_compress1X_usingCTable_internal_body(void* dst, size_t dstSize,
+                                   const void* src, size_t srcSize,
+                                   const HUF_CElt* CTable)
 {
     const BYTE* ip = (const BYTE*) src;
     BYTE* const ostart = (BYTE*)dst;
@@ -457,7 +468,7 @@ size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, si
 
     /* init */
     if (dstSize < 8) return 0;   /* not enough space to compress */
-    { size_t const initErr = BIT_initCStream(&bitC, op, oend-op);
+    { size_t const initErr = BIT_initCStream(&bitC, op, (size_t)(oend-op));
       if (HUF_isError(initErr)) return 0; }
 
     n = srcSize & ~3;  /* join to mod 4 */
@@ -490,8 +501,58 @@ size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, si
     return BIT_closeCStream(&bitC);
 }
 
+#if DYNAMIC_BMI2
 
-size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
+static TARGET_ATTRIBUTE("bmi2") size_t
+HUF_compress1X_usingCTable_internal_bmi2(void* dst, size_t dstSize,
+                                   const void* src, size_t srcSize,
+                                   const HUF_CElt* CTable)
+{
+    return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
+}
+
+static size_t
+HUF_compress1X_usingCTable_internal_default(void* dst, size_t dstSize,
+                                      const void* src, size_t srcSize,
+                                      const HUF_CElt* CTable)
+{
+    return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
+}
+
+static size_t
+HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
+                              const void* src, size_t srcSize,
+                              const HUF_CElt* CTable, const int bmi2)
+{
+    if (bmi2) {
+        return HUF_compress1X_usingCTable_internal_bmi2(dst, dstSize, src, srcSize, CTable);
+    }
+    return HUF_compress1X_usingCTable_internal_default(dst, dstSize, src, srcSize, CTable);
+}
+
+#else
+
+static size_t
+HUF_compress1X_usingCTable_internal(void* dst, size_t dstSize,
+                              const void* src, size_t srcSize,
+                              const HUF_CElt* CTable, const int bmi2)
+{
+    (void)bmi2;
+    return HUF_compress1X_usingCTable_internal_body(dst, dstSize, src, srcSize, CTable);
+}
+
+#endif
+
+size_t HUF_compress1X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
+{
+    return HUF_compress1X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, /* bmi2 */ 0);
+}
+
+
+static size_t
+HUF_compress4X_usingCTable_internal(void* dst, size_t dstSize,
+                              const void* src, size_t srcSize,
+                              const HUF_CElt* CTable, int bmi2)
 {
     size_t const segmentSize = (srcSize+3)/4;   /* first 3 segments */
     const BYTE* ip = (const BYTE*) src;
@@ -504,132 +565,162 @@ size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, si
     if (srcSize < 12) return 0;   /* no saving possible : too small input */
     op += 6;   /* jumpTable */
 
-    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable(op, oend-op, ip, segmentSize, CTable) );
+    assert(op <= oend);
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, bmi2) );
         if (cSize==0) return 0;
+        assert(cSize <= 65535);
         MEM_writeLE16(ostart, (U16)cSize);
         op += cSize;
     }
 
     ip += segmentSize;
-    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable(op, oend-op, ip, segmentSize, CTable) );
+    assert(op <= oend);
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, bmi2) );
         if (cSize==0) return 0;
+        assert(cSize <= 65535);
         MEM_writeLE16(ostart+2, (U16)cSize);
         op += cSize;
     }
 
     ip += segmentSize;
-    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable(op, oend-op, ip, segmentSize, CTable) );
+    assert(op <= oend);
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, segmentSize, CTable, bmi2) );
         if (cSize==0) return 0;
+        assert(cSize <= 65535);
         MEM_writeLE16(ostart+4, (U16)cSize);
         op += cSize;
     }
 
     ip += segmentSize;
-    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable(op, oend-op, ip, iend-ip, CTable) );
+    assert(op <= oend);
+    assert(ip <= iend);
+    {   CHECK_V_F(cSize, HUF_compress1X_usingCTable_internal(op, (size_t)(oend-op), ip, (size_t)(iend-ip), CTable, bmi2) );
         if (cSize==0) return 0;
         op += cSize;
     }
 
-    return op-ostart;
+    return (size_t)(op-ostart);
 }
 
+size_t HUF_compress4X_usingCTable(void* dst, size_t dstSize, const void* src, size_t srcSize, const HUF_CElt* CTable)
+{
+    return HUF_compress4X_usingCTable_internal(dst, dstSize, src, srcSize, CTable, /* bmi2 */ 0);
+}
+
+typedef enum { HUF_singleStream, HUF_fourStreams } HUF_nbStreams_e;
 
 static size_t HUF_compressCTable_internal(
                 BYTE* const ostart, BYTE* op, BYTE* const oend,
                 const void* src, size_t srcSize,
-                unsigned singleStream, const HUF_CElt* CTable)
+                HUF_nbStreams_e nbStreams, const HUF_CElt* CTable, const int bmi2)
 {
-    size_t const cSize = singleStream ?
-                         HUF_compress1X_usingCTable(op, oend - op, src, srcSize, CTable) :
-                         HUF_compress4X_usingCTable(op, oend - op, src, srcSize, CTable);
+    size_t const cSize = (nbStreams==HUF_singleStream) ?
+                         HUF_compress1X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, bmi2) :
+                         HUF_compress4X_usingCTable_internal(op, (size_t)(oend - op), src, srcSize, CTable, bmi2);
     if (HUF_isError(cSize)) { return cSize; }
     if (cSize==0) { return 0; }   /* uncompressible */
     op += cSize;
     /* check compressibility */
+    assert(op >= ostart);
     if ((size_t)(op-ostart) >= srcSize-1) { return 0; }
-    return op-ostart;
+    return (size_t)(op-ostart);
 }
 
-
-/* `workSpace` must a table of at least 1024 unsigned */
-static size_t HUF_compress_internal (
-                void* dst, size_t dstSize,
-                const void* src, size_t srcSize,
-                unsigned maxSymbolValue, unsigned huffLog,
-                unsigned singleStream,
-                void* workSpace, size_t wkspSize,
-                HUF_CElt* oldHufTable, HUF_repeat* repeat, int preferRepeat)
+typedef struct {
+    unsigned count[HUF_SYMBOLVALUE_MAX + 1];
+    HUF_CElt CTable[HUF_SYMBOLVALUE_MAX + 1];
+    HUF_buildCTable_wksp_tables buildCTable_wksp;
+} HUF_compress_tables_t;
+
+/* HUF_compress_internal() :
+ * `workSpace` must a table of at least HUF_WORKSPACE_SIZE_U32 unsigned */
+static size_t
+HUF_compress_internal (void* dst, size_t dstSize,
+                 const void* src, size_t srcSize,
+                       unsigned maxSymbolValue, unsigned huffLog,
+                       HUF_nbStreams_e nbStreams,
+                       void* workSpace, size_t wkspSize,
+                       HUF_CElt* oldHufTable, HUF_repeat* repeat, int preferRepeat,
+                 const int bmi2)
 {
+    HUF_compress_tables_t* const table = (HUF_compress_tables_t*)workSpace;
     BYTE* const ostart = (BYTE*)dst;
     BYTE* const oend = ostart + dstSize;
     BYTE* op = ostart;
 
-    U32* count;
-    size_t const countSize = sizeof(U32) * (HUF_SYMBOLVALUE_MAX + 1);
-    HUF_CElt* CTable;
-    size_t const CTableSize = sizeof(HUF_CElt) * (HUF_SYMBOLVALUE_MAX + 1);
+    HUF_STATIC_ASSERT(sizeof(*table) <= HUF_WORKSPACE_SIZE);
 
     /* checks & inits */
-    if (wkspSize < sizeof(huffNodeTable) + countSize + CTableSize) return ERROR(GENERIC);
-    if (!srcSize) return 0;  /* Uncompressed (note : 1 means rle, so first byte must be correct) */
-    if (!dstSize) return 0;  /* cannot fit within dst budget */
+    if (((size_t)workSpace & 3) != 0) return ERROR(GENERIC);  /* must be aligned on 4-bytes boundaries */
+    if (wkspSize < HUF_WORKSPACE_SIZE) return ERROR(workSpace_tooSmall);
+    if (!srcSize) return 0;  /* Uncompressed */
+    if (!dstSize) return 0;  /* cannot fit anything within dst budget */
     if (srcSize > HUF_BLOCKSIZE_MAX) return ERROR(srcSize_wrong);   /* current block size limit */
     if (huffLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
+    if (maxSymbolValue > HUF_SYMBOLVALUE_MAX) return ERROR(maxSymbolValue_tooLarge);
     if (!maxSymbolValue) maxSymbolValue = HUF_SYMBOLVALUE_MAX;
     if (!huffLog) huffLog = HUF_TABLELOG_DEFAULT;
 
-    count = (U32*)workSpace;
-    workSpace = (BYTE*)workSpace + countSize;
-    wkspSize -= countSize;
-    CTable = (HUF_CElt*)workSpace;
-    workSpace = (BYTE*)workSpace + CTableSize;
-    wkspSize -= CTableSize;
-
-    /* Heuristic : If we don't need to check the validity of the old table use the old table for small inputs */
+    /* Heuristic : If old table is valid, use it for small inputs */
     if (preferRepeat && repeat && *repeat == HUF_repeat_valid) {
-        return HUF_compressCTable_internal(ostart, op, oend, src, srcSize, singleStream, oldHufTable);
+        return HUF_compressCTable_internal(ostart, op, oend,
+                                           src, srcSize,
+                                           nbStreams, oldHufTable, bmi2);
     }
 
     /* Scan input and build symbol stats */
-    {   CHECK_V_F(largest, FSE_count_wksp (count, &maxSymbolValue, (const BYTE*)src, srcSize, (U32*)workSpace) );
+    {   CHECK_V_F(largest, HIST_count_wksp (table->count, &maxSymbolValue, (const BYTE*)src, srcSize, workSpace, wkspSize) );
         if (largest == srcSize) { *ostart = ((const BYTE*)src)[0]; return 1; }   /* single symbol, rle */
-        if (largest <= (srcSize >> 7)+1) return 0;   /* Fast heuristic : not compressible enough */
+        if (largest <= (srcSize >> 7)+4) return 0;   /* heuristic : probably not compressible enough */
     }
 
     /* Check validity of previous table */
-    if (repeat && *repeat == HUF_repeat_check && !HUF_validateCTable(oldHufTable, count, maxSymbolValue)) {
+    if ( repeat
+      && *repeat == HUF_repeat_check
+      && !HUF_validateCTable(oldHufTable, table->count, maxSymbolValue)) {
         *repeat = HUF_repeat_none;
     }
     /* Heuristic : use existing table for small inputs */
     if (preferRepeat && repeat && *repeat != HUF_repeat_none) {
-        return HUF_compressCTable_internal(ostart, op, oend, src, srcSize, singleStream, oldHufTable);
+        return HUF_compressCTable_internal(ostart, op, oend,
+                                           src, srcSize,
+                                           nbStreams, oldHufTable, bmi2);
     }
 
     /* Build Huffman Tree */
     huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue);
-    {   CHECK_V_F(maxBits, HUF_buildCTable_wksp (CTable, count, maxSymbolValue, huffLog, workSpace, wkspSize) );
+    {   size_t const maxBits = HUF_buildCTable_wksp(table->CTable, table->count,
+                                            maxSymbolValue, huffLog,
+                                            &table->buildCTable_wksp, sizeof(table->buildCTable_wksp));
+        CHECK_F(maxBits);
         huffLog = (U32)maxBits;
-        /* Zero the unused symbols so we can check it for validity */
-        memset(CTable + maxSymbolValue + 1, 0, CTableSize - (maxSymbolValue + 1) * sizeof(HUF_CElt));
+        /* Zero unused symbols in CTable, so we can check it for validity */
+        memset(table->CTable + (maxSymbolValue + 1), 0,
+               sizeof(table->CTable) - ((maxSymbolValue + 1) * sizeof(HUF_CElt)));
     }
 
     /* Write table description header */
-    {   CHECK_V_F(hSize, HUF_writeCTable (op, dstSize, CTable, maxSymbolValue, huffLog) );
-        /* Check if using the previous table will be beneficial */
+    {   CHECK_V_F(hSize, HUF_writeCTable (op, dstSize, table->CTable, maxSymbolValue, huffLog) );
+        /* Check if using previous huffman table is beneficial */
         if (repeat && *repeat != HUF_repeat_none) {
-            size_t const oldSize = HUF_estimateCompressedSize(oldHufTable, count, maxSymbolValue);
-            size_t const newSize = HUF_estimateCompressedSize(CTable, count, maxSymbolValue);
+            size_t const oldSize = HUF_estimateCompressedSize(oldHufTable, table->count, maxSymbolValue);
+            size_t const newSize = HUF_estimateCompressedSize(table->CTable, table->count, maxSymbolValue);
             if (oldSize <= hSize + newSize || hSize + 12 >= srcSize) {
-                return HUF_compressCTable_internal(ostart, op, oend, src, srcSize, singleStream, oldHufTable);
-            }
-        }
-        /* Use the new table */
+                return HUF_compressCTable_internal(ostart, op, oend,
+                                                   src, srcSize,
+                                                   nbStreams, oldHufTable, bmi2);
+        }   }
+
+        /* Use the new huffman table */
         if (hSize + 12ul >= srcSize) { return 0; }
         op += hSize;
         if (repeat) { *repeat = HUF_repeat_none; }
-        if (oldHufTable) { memcpy(oldHufTable, CTable, CTableSize); } /* Save the new table */
+        if (oldHufTable)
+            memcpy(oldHufTable, table->CTable, sizeof(table->CTable));  /* Save new table */
     }
-    return HUF_compressCTable_internal(ostart, op, oend, src, srcSize, singleStream, CTable);
+    return HUF_compressCTable_internal(ostart, op, oend,
+                                       src, srcSize,
+                                       nbStreams, table->CTable, bmi2);
 }
 
 
@@ -638,52 +729,70 @@ size_t HUF_compress1X_wksp (void* dst, size_t dstSize,
                       unsigned maxSymbolValue, unsigned huffLog,
                       void* workSpace, size_t wkspSize)
 {
-    return HUF_compress_internal(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, 1 /* single stream */, workSpace, wkspSize, NULL, NULL, 0);
+    return HUF_compress_internal(dst, dstSize, src, srcSize,
+                                 maxSymbolValue, huffLog, HUF_singleStream,
+                                 workSpace, wkspSize,
+                                 NULL, NULL, 0, 0 /*bmi2*/);
 }
 
 size_t HUF_compress1X_repeat (void* dst, size_t dstSize,
                       const void* src, size_t srcSize,
                       unsigned maxSymbolValue, unsigned huffLog,
                       void* workSpace, size_t wkspSize,
-                      HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat)
+                      HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2)
 {
-    return HUF_compress_internal(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, 1 /* single stream */, workSpace, wkspSize, hufTable, repeat, preferRepeat);
+    return HUF_compress_internal(dst, dstSize, src, srcSize,
+                                 maxSymbolValue, huffLog, HUF_singleStream,
+                                 workSpace, wkspSize, hufTable,
+                                 repeat, preferRepeat, bmi2);
 }
 
 size_t HUF_compress1X (void* dst, size_t dstSize,
                  const void* src, size_t srcSize,
                  unsigned maxSymbolValue, unsigned huffLog)
 {
-    unsigned workSpace[1024];
+    unsigned workSpace[HUF_WORKSPACE_SIZE_U32];
     return HUF_compress1X_wksp(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, workSpace, sizeof(workSpace));
 }
 
+/* HUF_compress4X_repeat():
+ * compress input using 4 streams.
+ * provide workspace to generate compression tables */
 size_t HUF_compress4X_wksp (void* dst, size_t dstSize,
                       const void* src, size_t srcSize,
                       unsigned maxSymbolValue, unsigned huffLog,
                       void* workSpace, size_t wkspSize)
 {
-    return HUF_compress_internal(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, 0 /* 4 streams */, workSpace, wkspSize, NULL, NULL, 0);
+    return HUF_compress_internal(dst, dstSize, src, srcSize,
+                                 maxSymbolValue, huffLog, HUF_fourStreams,
+                                 workSpace, wkspSize,
+                                 NULL, NULL, 0, 0 /*bmi2*/);
 }
 
+/* HUF_compress4X_repeat():
+ * compress input using 4 streams.
+ * re-use an existing huffman compression table */
 size_t HUF_compress4X_repeat (void* dst, size_t dstSize,
                       const void* src, size_t srcSize,
                       unsigned maxSymbolValue, unsigned huffLog,
                       void* workSpace, size_t wkspSize,
-                      HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat)
+                      HUF_CElt* hufTable, HUF_repeat* repeat, int preferRepeat, int bmi2)
 {
-    return HUF_compress_internal(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, 0 /* 4 streams */, workSpace, wkspSize, hufTable, repeat, preferRepeat);
+    return HUF_compress_internal(dst, dstSize, src, srcSize,
+                                 maxSymbolValue, huffLog, HUF_fourStreams,
+                                 workSpace, wkspSize,
+                                 hufTable, repeat, preferRepeat, bmi2);
 }
 
 size_t HUF_compress2 (void* dst, size_t dstSize,
                 const void* src, size_t srcSize,
                 unsigned maxSymbolValue, unsigned huffLog)
 {
-    unsigned workSpace[1024];
+    unsigned workSpace[HUF_WORKSPACE_SIZE_U32];
     return HUF_compress4X_wksp(dst, dstSize, src, srcSize, maxSymbolValue, huffLog, workSpace, sizeof(workSpace));
 }
 
 size_t HUF_compress (void* dst, size_t maxDstSize, const void* src, size_t srcSize)
 {
-    return HUF_compress2(dst, maxDstSize, src, (U32)srcSize, 255, HUF_TABLELOG_DEFAULT);
+    return HUF_compress2(dst, maxDstSize, src, srcSize, 255, HUF_TABLELOG_DEFAULT);
 }
diff --git a/contrib/zstd/huf_decompress.c b/contrib/zstd/huf_decompress.c
index 79ded96bf..ab5db481c 100644
--- a/contrib/zstd/huf_decompress.c
+++ b/contrib/zstd/huf_decompress.c
@@ -1,66 +1,103 @@
 /* ******************************************************************
-   Huffman decoder, part of New Generation Entropy library
-   Copyright (C) 2013-2016, Yann Collet.
-
-   BSD 2-Clause License (http://www.opensource.org/licenses/bsd-license.php)
-
-   Redistribution and use in source and binary forms, with or without
-   modification, are permitted provided that the following conditions are
-   met:
-
-       * Redistributions of source code must retain the above copyright
-   notice, this list of conditions and the following disclaimer.
-       * Redistributions in binary form must reproduce the above
-   copyright notice, this list of conditions and the following disclaimer
-   in the documentation and/or other materials provided with the
-   distribution.
-
-   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
-   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
-   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
-   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-
-    You can contact the author at :
-    - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
-    - Public forum : https://groups.google.com/forum/#!forum/lz4c
+ * huff0 huffman decoder,
+ * part of Finite State Entropy library
+ * Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
+ *
+ *  You can contact the author at :
+ *  - FSE+HUF source repository : https://github.com/Cyan4973/FiniteStateEntropy
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
 ****************************************************************** */
 
 /* **************************************************************
 *  Dependencies
 ****************************************************************/
 #include <string.h>     /* memcpy, memset */
-#include "bitstream.h"  /* BIT_* */
 #include "compiler.h"
-#include "fse.h"        /* header compression */
+#include "bitstream.h"  /* BIT_* */
+#include "fse.h"        /* to compress headers */
 #define HUF_STATIC_LINKING_ONLY
 #include "huf.h"
 #include "error_private.h"
 
+/* **************************************************************
+*  Macros
+****************************************************************/
+
+/* These two optional macros force the use one way or another of the two
+ * Huffman decompression implementations. You can't force in both directions
+ * at the same time.
+ */
+#if defined(HUF_FORCE_DECOMPRESS_X1) && \
+    defined(HUF_FORCE_DECOMPRESS_X2)
+#error "Cannot force the use of the X1 and X2 decoders at the same time!"
+#endif
+
 
 /* **************************************************************
 *  Error Management
 ****************************************************************/
 #define HUF_isError ERR_isError
-#define HUF_STATIC_ASSERT(c) { enum { HUF_static_assert = 1/(int)(!!(c)) }; }   /* use only *after* variable declarations */
 
 
 /* **************************************************************
 *  Byte alignment for workSpace management
 ****************************************************************/
-#define HUF_ALIGN(x, a) HUF_ALIGN_MASK((x), (a) - 1)
+#define HUF_ALIGN(x, a)         HUF_ALIGN_MASK((x), (a) - 1)
 #define HUF_ALIGN_MASK(x, mask) (((x) + (mask)) & ~(mask))
 
+
+/* **************************************************************
+*  BMI2 Variant Wrappers
+****************************************************************/
+#if DYNAMIC_BMI2
+
+#define HUF_DGEN(fn)                                                        \
+                                                                            \
+    static size_t fn##_default(                                             \
+                  void* dst,  size_t dstSize,                               \
+            const void* cSrc, size_t cSrcSize,                              \
+            const HUF_DTable* DTable)                                       \
+    {                                                                       \
+        return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
+    }                                                                       \
+                                                                            \
+    static TARGET_ATTRIBUTE("bmi2") size_t fn##_bmi2(                       \
+                  void* dst,  size_t dstSize,                               \
+            const void* cSrc, size_t cSrcSize,                              \
+            const HUF_DTable* DTable)                                       \
+    {                                                                       \
+        return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
+    }                                                                       \
+                                                                            \
+    static size_t fn(void* dst, size_t dstSize, void const* cSrc,           \
+                     size_t cSrcSize, HUF_DTable const* DTable, int bmi2)   \
+    {                                                                       \
+        if (bmi2) {                                                         \
+            return fn##_bmi2(dst, dstSize, cSrc, cSrcSize, DTable);         \
+        }                                                                   \
+        return fn##_default(dst, dstSize, cSrc, cSrcSize, DTable);          \
+    }
+
+#else
+
+#define HUF_DGEN(fn)                                                        \
+    static size_t fn(void* dst, size_t dstSize, void const* cSrc,           \
+                     size_t cSrcSize, HUF_DTable const* DTable, int bmi2)   \
+    {                                                                       \
+        (void)bmi2;                                                         \
+        return fn##_body(dst, dstSize, cSrc, cSrcSize, DTable);             \
+    }
+
+#endif
+
+
 /*-***************************/
 /*  generic DTableDesc       */
 /*-***************************/
-
 typedef struct { BYTE maxTableLog; BYTE tableType; BYTE tableLog; BYTE reserved; } DTableDesc;
 
 static DTableDesc HUF_getDTableDesc(const HUF_DTable* table)
@@ -71,19 +108,20 @@ static DTableDesc HUF_getDTableDesc(const HUF_DTable* table)
 }
 
 
+#ifndef HUF_FORCE_DECOMPRESS_X2
+
 /*-***************************/
 /*  single-symbol decoding   */
 /*-***************************/
+typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX1;   /* single-symbol decoding */
 
-typedef struct { BYTE byte; BYTE nbBits; } HUF_DEltX2;   /* single-symbol decoding */
-
-size_t HUF_readDTableX2_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize)
+size_t HUF_readDTableX1_wksp(HUF_DTable* DTable, const void* src, size_t srcSize, void* workSpace, size_t wkspSize)
 {
     U32 tableLog = 0;
     U32 nbSymbols = 0;
     size_t iSize;
     void* const dtPtr = DTable + 1;
-    HUF_DEltX2* const dt = (HUF_DEltX2*)dtPtr;
+    HUF_DEltX1* const dt = (HUF_DEltX1*)dtPtr;
 
     U32* rankVal;
     BYTE* huffWeight;
@@ -94,12 +132,9 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable, const void* src, size_t srcSize
     huffWeight = (BYTE *)((U32 *)workSpace + spaceUsed32);
     spaceUsed32 += HUF_ALIGN(HUF_SYMBOLVALUE_MAX + 1, sizeof(U32)) >> 2;
 
-    if ((spaceUsed32 << 2) > wkspSize)
-        return ERROR(tableLog_tooLarge);
-    workSpace = (U32 *)workSpace + spaceUsed32;
-    wkspSize -= (spaceUsed32 << 2);
+    if ((spaceUsed32 << 2) > wkspSize) return ERROR(tableLog_tooLarge);
 
-    HUF_STATIC_ASSERT(sizeof(DTableDesc) == sizeof(HUF_DTable));
+    DEBUG_STATIC_ASSERT(sizeof(DTableDesc) == sizeof(HUF_DTable));
     /* memset(huffWeight, 0, sizeof(huffWeight)); */   /* is not necessary, even though some analyzer complain ... */
 
     iSize = HUF_readStats(huffWeight, HUF_SYMBOLVALUE_MAX + 1, rankVal, &nbSymbols, &tableLog, src, srcSize);
@@ -123,29 +158,41 @@ size_t HUF_readDTableX2_wksp(HUF_DTable* DTable, const void* src, size_t srcSize
 
     /* fill DTable */
     {   U32 n;
-        for (n=0; n<nbSymbols; n++) {
-            U32 const w = huffWeight[n];
-            U32 const length = (1 << w) >> 1;
-            U32 u;
-            HUF_DEltX2 D;
-            D.byte = (BYTE)n; D.nbBits = (BYTE)(tableLog + 1 - w);
-            for (u = rankVal[w]; u < rankVal[w] + length; u++)
-                dt[u] = D;
-            rankVal[w] += length;
-    }   }
-
+        size_t const nEnd = nbSymbols;
+        for (n=0; n<nEnd; n++) {
+            size_t const w = huffWeight[n];
+            size_t const length = (1 << w) >> 1;
+            size_t const uStart = rankVal[w];
+            size_t const uEnd = uStart + length;
+            size_t u;
+            HUF_DEltX1 D;
+            D.byte = (BYTE)n;
+            D.nbBits = (BYTE)(tableLog + 1 - w);
+            rankVal[w] = (U32)uEnd;
+            if (length < 4) {
+                /* Use length in the loop bound so the compiler knows it is short. */
+                for (u = 0; u < length; ++u)
+                    dt[uStart + u] = D;
+            } else {
+                /* Unroll the loop 4 times, we know it is a power of 2. */
+                for (u = uStart; u < uEnd; u += 4) {
+                    dt[u + 0] = D;
+                    dt[u + 1] = D;
+                    dt[u + 2] = D;
+                    dt[u + 3] = D;
+    }   }   }   }
     return iSize;
 }
 
-size_t HUF_readDTableX2(HUF_DTable* DTable, const void* src, size_t srcSize)
+size_t HUF_readDTableX1(HUF_DTable* DTable, const void* src, size_t srcSize)
 {
     U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_readDTableX2_wksp(DTable, src, srcSize,
+    return HUF_readDTableX1_wksp(DTable, src, srcSize,
                                  workSpace, sizeof(workSpace));
 }
 
-
-static BYTE HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, const U32 dtLog)
+FORCE_INLINE_TEMPLATE BYTE
+HUF_decodeSymbolX1(BIT_DStream_t* Dstream, const HUF_DEltX1* dt, const U32 dtLog)
 {
     size_t const val = BIT_lookBitsFast(Dstream, dtLog); /* note : dtLog >= 1 */
     BYTE const c = dt[val].byte;
@@ -153,41 +200,44 @@ static BYTE HUF_decodeSymbolX2(BIT_DStream_t* Dstream, const HUF_DEltX2* dt, con
     return c;
 }
 
-#define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
-    *ptr++ = HUF_decodeSymbolX2(DStreamPtr, dt, dtLog)
+#define HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr) \
+    *ptr++ = HUF_decodeSymbolX1(DStreamPtr, dt, dtLog)
 
-#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
+#define HUF_DECODE_SYMBOLX1_1(ptr, DStreamPtr)  \
     if (MEM_64bits() || (HUF_TABLELOG_MAX<=12)) \
-        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+        HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr)
 
-#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
+#define HUF_DECODE_SYMBOLX1_2(ptr, DStreamPtr) \
     if (MEM_64bits()) \
-        HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr)
+        HUF_DECODE_SYMBOLX1_0(ptr, DStreamPtr)
 
-HINT_INLINE size_t HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX2* const dt, const U32 dtLog)
+HINT_INLINE size_t
+HUF_decodeStreamX1(BYTE* p, BIT_DStream_t* const bitDPtr, BYTE* const pEnd, const HUF_DEltX1* const dt, const U32 dtLog)
 {
     BYTE* const pStart = p;
 
     /* up to 4 symbols at a time */
-    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p <= pEnd-4)) {
-        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
-        HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
-        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
-        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-3)) {
+        HUF_DECODE_SYMBOLX1_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX1_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX1_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
     }
 
-    /* closer to the end */
-    while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) && (p < pEnd))
-        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+    /* [0-3] symbols remaining */
+    if (MEM_32bits())
+        while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd))
+            HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
 
-    /* no more data to retrieve from bitstream, hence no need to reload */
+    /* no more data to retrieve from bitstream, no need to reload */
     while (p < pEnd)
-        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
+        HUF_DECODE_SYMBOLX1_0(p, bitDPtr);
 
     return pEnd-pStart;
 }
 
-static size_t HUF_decompress1X2_usingDTable_internal(
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress1X1_usingDTable_internal_body(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
     const HUF_DTable* DTable)
@@ -195,63 +245,22 @@ static size_t HUF_decompress1X2_usingDTable_internal(
     BYTE* op = (BYTE*)dst;
     BYTE* const oend = op + dstSize;
     const void* dtPtr = DTable + 1;
-    const HUF_DEltX2* const dt = (const HUF_DEltX2*)dtPtr;
+    const HUF_DEltX1* const dt = (const HUF_DEltX1*)dtPtr;
     BIT_DStream_t bitD;
     DTableDesc const dtd = HUF_getDTableDesc(DTable);
     U32 const dtLog = dtd.tableLog;
 
-    { size_t const errorCode = BIT_initDStream(&bitD, cSrc, cSrcSize);
-      if (HUF_isError(errorCode)) return errorCode; }
+    CHECK_F( BIT_initDStream(&bitD, cSrc, cSrcSize) );
 
-    HUF_decodeStreamX2(op, &bitD, oend, dt, dtLog);
+    HUF_decodeStreamX1(op, &bitD, oend, dt, dtLog);
 
-    /* check */
     if (!BIT_endOfDStream(&bitD)) return ERROR(corruption_detected);
 
     return dstSize;
 }
 
-size_t HUF_decompress1X2_usingDTable(
-          void* dst,  size_t dstSize,
-    const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable)
-{
-    DTableDesc dtd = HUF_getDTableDesc(DTable);
-    if (dtd.tableType != 0) return ERROR(GENERIC);
-    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
-}
-
-size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
-                                   const void* cSrc, size_t cSrcSize,
-                                   void* workSpace, size_t wkspSize)
-{
-    const BYTE* ip = (const BYTE*) cSrc;
-
-    size_t const hSize = HUF_readDTableX2_wksp(DCtx, cSrc, cSrcSize, workSpace, wkspSize);
-    if (HUF_isError(hSize)) return hSize;
-    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
-    ip += hSize; cSrcSize -= hSize;
-
-    return HUF_decompress1X2_usingDTable_internal (dst, dstSize, ip, cSrcSize, DCtx);
-}
-
-
-size_t HUF_decompress1X2_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
-                              const void* cSrc, size_t cSrcSize)
-{
-    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_decompress1X2_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
-                                       workSpace, sizeof(workSpace));
-}
-
-size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
-    return HUF_decompress1X2_DCtx (DTable, dst, dstSize, cSrc, cSrcSize);
-}
-
-
-static size_t HUF_decompress4X2_usingDTable_internal(
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress4X1_usingDTable_internal_body(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
     const HUF_DTable* DTable)
@@ -262,8 +271,9 @@ static size_t HUF_decompress4X2_usingDTable_internal(
     {   const BYTE* const istart = (const BYTE*) cSrc;
         BYTE* const ostart = (BYTE*) dst;
         BYTE* const oend = ostart + dstSize;
+        BYTE* const olimit = oend - 3;
         const void* const dtPtr = DTable + 1;
-        const HUF_DEltX2* const dt = (const HUF_DEltX2*)dtPtr;
+        const HUF_DEltX1* const dt = (const HUF_DEltX1*)dtPtr;
 
         /* Init */
         BIT_DStream_t bitD1;
@@ -286,57 +296,57 @@ static size_t HUF_decompress4X2_usingDTable_internal(
         BYTE* op2 = opStart2;
         BYTE* op3 = opStart3;
         BYTE* op4 = opStart4;
-        U32 endSignal;
         DTableDesc const dtd = HUF_getDTableDesc(DTable);
         U32 const dtLog = dtd.tableLog;
+        U32 endSignal = 1;
 
         if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
-        { size_t const errorCode = BIT_initDStream(&bitD1, istart1, length1);
-          if (HUF_isError(errorCode)) return errorCode; }
-        { size_t const errorCode = BIT_initDStream(&bitD2, istart2, length2);
-          if (HUF_isError(errorCode)) return errorCode; }
-        { size_t const errorCode = BIT_initDStream(&bitD3, istart3, length3);
-          if (HUF_isError(errorCode)) return errorCode; }
-        { size_t const errorCode = BIT_initDStream(&bitD4, istart4, length4);
-          if (HUF_isError(errorCode)) return errorCode; }
-
-        /* 16-32 symbols per loop (4-8 symbols per stream) */
-        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
-        for ( ; (endSignal==BIT_DStream_unfinished) && (op4<(oend-7)) ; ) {
-            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
-            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
-            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
-            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
-            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
-            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
-            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
-            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
-            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
-            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
-            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
-            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
-            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
-            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
-            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
-            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
-            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
+        CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
+        CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
+        CHECK_F( BIT_initDStream(&bitD4, istart4, length4) );
+
+        /* up to 16 symbols per loop (4 symbols per stream) in 64-bit mode */
+        for ( ; (endSignal) & (op4 < olimit) ; ) {
+            HUF_DECODE_SYMBOLX1_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX1_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX1_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX1_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX1_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX1_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX1_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX1_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX1_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX1_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX1_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX1_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX1_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX1_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX1_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX1_0(op4, &bitD4);
+            endSignal &= BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished;
+            endSignal &= BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished;
+            endSignal &= BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished;
+            endSignal &= BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished;
         }
 
         /* check corruption */
+        /* note : should not be necessary : op# advance in lock step, and we control op4.
+         *        but curiously, binary generated by gcc 7.2 & 7.3 with -mbmi2 runs faster when >=1 test is present */
         if (op1 > opStart2) return ERROR(corruption_detected);
         if (op2 > opStart3) return ERROR(corruption_detected);
         if (op3 > opStart4) return ERROR(corruption_detected);
         /* note : op4 supposed already verified within main loop */
 
         /* finish bitStreams one by one */
-        HUF_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
-        HUF_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
-        HUF_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
-        HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
+        HUF_decodeStreamX1(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX1(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX1(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX1(op4, &bitD4, oend,     dt, dtLog);
 
         /* check */
-        endSignal = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
-        if (!endSignal) return ERROR(corruption_detected);
+        { U32 const endCheck = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
+          if (!endCheck) return ERROR(corruption_detected); }
 
         /* decoded size */
         return dstSize;
@@ -344,61 +354,123 @@ static size_t HUF_decompress4X2_usingDTable_internal(
 }
 
 
-size_t HUF_decompress4X2_usingDTable(
+typedef size_t (*HUF_decompress_usingDTable_t)(void *dst, size_t dstSize,
+                                               const void *cSrc,
+                                               size_t cSrcSize,
+                                               const HUF_DTable *DTable);
+
+HUF_DGEN(HUF_decompress1X1_usingDTable_internal)
+HUF_DGEN(HUF_decompress4X1_usingDTable_internal)
+
+
+
+size_t HUF_decompress1X1_usingDTable(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
     const HUF_DTable* DTable)
 {
     DTableDesc dtd = HUF_getDTableDesc(DTable);
     if (dtd.tableType != 0) return ERROR(GENERIC);
-    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+    return HUF_decompress1X1_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
 }
 
-
-size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+size_t HUF_decompress1X1_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
                                    const void* cSrc, size_t cSrcSize,
                                    void* workSpace, size_t wkspSize)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
-    size_t const hSize = HUF_readDTableX2_wksp (dctx, cSrc, cSrcSize,
+    size_t const hSize = HUF_readDTableX1_wksp(DCtx, cSrc, cSrcSize, workSpace, wkspSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress1X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, /* bmi2 */ 0);
+}
+
+
+size_t HUF_decompress1X1_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
+                              const void* cSrc, size_t cSrcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_decompress1X1_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
+                                       workSpace, sizeof(workSpace));
+}
+
+size_t HUF_decompress1X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX1(DTable, HUF_TABLELOG_MAX);
+    return HUF_decompress1X1_DCtx (DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+size_t HUF_decompress4X1_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    DTableDesc dtd = HUF_getDTableDesc(DTable);
+    if (dtd.tableType != 0) return ERROR(GENERIC);
+    return HUF_decompress4X1_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+}
+
+static size_t HUF_decompress4X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize, int bmi2)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUF_readDTableX1_wksp (dctx, cSrc, cSrcSize,
                                                 workSpace, wkspSize);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress4X2_usingDTable_internal (dst, dstSize, ip, cSrcSize, dctx);
+    return HUF_decompress4X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
+}
+
+size_t HUF_decompress4X1_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize)
+{
+    return HUF_decompress4X1_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, 0);
 }
 
 
-size_t HUF_decompress4X2_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+size_t HUF_decompress4X1_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
 {
     U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
+    return HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
                                        workSpace, sizeof(workSpace));
 }
-size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+size_t HUF_decompress4X1 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
 {
-    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
-    return HUF_decompress4X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+    HUF_CREATE_STATIC_DTABLEX1(DTable, HUF_TABLELOG_MAX);
+    return HUF_decompress4X1_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
 }
 
+#endif /* HUF_FORCE_DECOMPRESS_X2 */
+
+
+#ifndef HUF_FORCE_DECOMPRESS_X1
 
 /* *************************/
 /* double-symbols decoding */
 /* *************************/
-typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX4;  /* double-symbols decoding */
 
+typedef struct { U16 sequence; BYTE nbBits; BYTE length; } HUF_DEltX2;  /* double-symbols decoding */
 typedef struct { BYTE symbol; BYTE weight; } sortedSymbol_t;
+typedef U32 rankValCol_t[HUF_TABLELOG_MAX + 1];
+typedef rankValCol_t rankVal_t[HUF_TABLELOG_MAX];
 
-/* HUF_fillDTableX4Level2() :
+
+/* HUF_fillDTableX2Level2() :
  * `rankValOrigin` must be a table of at least (HUF_TABLELOG_MAX + 1) U32 */
-static void HUF_fillDTableX4Level2(HUF_DEltX4* DTable, U32 sizeLog, const U32 consumed,
+static void HUF_fillDTableX2Level2(HUF_DEltX2* DTable, U32 sizeLog, const U32 consumed,
                            const U32* rankValOrigin, const int minWeight,
                            const sortedSymbol_t* sortedSymbols, const U32 sortedListSize,
                            U32 nbBitsBaseline, U16 baseSeq)
 {
-    HUF_DEltX4 DElt;
+    HUF_DEltX2 DElt;
     U32 rankVal[HUF_TABLELOG_MAX + 1];
 
     /* get pre-calculated rankVal */
@@ -433,10 +505,8 @@ static void HUF_fillDTableX4Level2(HUF_DEltX4* DTable, U32 sizeLog, const U32 co
     }   }
 }
 
-typedef U32 rankValCol_t[HUF_TABLELOG_MAX + 1];
-typedef rankValCol_t rankVal_t[HUF_TABLELOG_MAX];
 
-static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
+static void HUF_fillDTableX2(HUF_DEltX2* DTable, const U32 targetLog,
                            const sortedSymbol_t* sortedList, const U32 sortedListSize,
                            const U32* rankStart, rankVal_t rankValOrigin, const U32 maxWeight,
                            const U32 nbBitsBaseline)
@@ -461,12 +531,12 @@ static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
             int minWeight = nbBits + scaleLog;
             if (minWeight < 1) minWeight = 1;
             sortedRank = rankStart[minWeight];
-            HUF_fillDTableX4Level2(DTable+start, targetLog-nbBits, nbBits,
+            HUF_fillDTableX2Level2(DTable+start, targetLog-nbBits, nbBits,
                            rankValOrigin[nbBits], minWeight,
                            sortedList+sortedRank, sortedListSize-sortedRank,
                            nbBitsBaseline, symbol);
         } else {
-            HUF_DEltX4 DElt;
+            HUF_DEltX2 DElt;
             MEM_writeLE16(&(DElt.sequence), symbol);
             DElt.nbBits = (BYTE)(nbBits);
             DElt.length = 1;
@@ -478,16 +548,16 @@ static void HUF_fillDTableX4(HUF_DEltX4* DTable, const U32 targetLog,
     }
 }
 
-size_t HUF_readDTableX4_wksp(HUF_DTable* DTable, const void* src,
-                             size_t srcSize, void* workSpace,
-                             size_t wkspSize)
+size_t HUF_readDTableX2_wksp(HUF_DTable* DTable,
+                       const void* src, size_t srcSize,
+                             void* workSpace, size_t wkspSize)
 {
     U32 tableLog, maxW, sizeOfSort, nbSymbols;
     DTableDesc dtd = HUF_getDTableDesc(DTable);
     U32 const maxTableLog = dtd.maxTableLog;
     size_t iSize;
     void* dtPtr = DTable+1;   /* force compiler to avoid strict-aliasing */
-    HUF_DEltX4* const dt = (HUF_DEltX4*)dtPtr;
+    HUF_DEltX2* const dt = (HUF_DEltX2*)dtPtr;
     U32 *rankStart;
 
     rankValCol_t* rankVal;
@@ -508,15 +578,12 @@ size_t HUF_readDTableX4_wksp(HUF_DTable* DTable, const void* src,
     weightList = (BYTE *)((U32 *)workSpace + spaceUsed32);
     spaceUsed32 += HUF_ALIGN(HUF_SYMBOLVALUE_MAX + 1, sizeof(U32)) >> 2;
 
-    if ((spaceUsed32 << 2) > wkspSize)
-        return ERROR(tableLog_tooLarge);
-    workSpace = (U32 *)workSpace + spaceUsed32;
-    wkspSize -= (spaceUsed32 << 2);
+    if ((spaceUsed32 << 2) > wkspSize) return ERROR(tableLog_tooLarge);
 
     rankStart = rankStart0 + 1;
     memset(rankStats, 0, sizeof(U32) * (2 * HUF_TABLELOG_MAX + 2 + 1));
 
-    HUF_STATIC_ASSERT(sizeof(HUF_DEltX4) == sizeof(HUF_DTable));   /* if compiler fails here, assertion is wrong */
+    DEBUG_STATIC_ASSERT(sizeof(HUF_DEltX2) == sizeof(HUF_DTable));   /* if compiler fails here, assertion is wrong */
     if (maxTableLog > HUF_TABLELOG_MAX) return ERROR(tableLog_tooLarge);
     /* memset(weightList, 0, sizeof(weightList)); */  /* is not necessary, even though some analyzer complain ... */
 
@@ -570,7 +637,7 @@ size_t HUF_readDTableX4_wksp(HUF_DTable* DTable, const void* src,
                     rankValPtr[w] = rankVal0[w] >> consumed;
     }   }   }   }
 
-    HUF_fillDTableX4(dt, maxTableLog,
+    HUF_fillDTableX2(dt, maxTableLog,
                    sortedSymbol, sizeOfSort,
                    rankStart0, rankVal, maxW,
                    tableLog+1);
@@ -581,14 +648,16 @@ size_t HUF_readDTableX4_wksp(HUF_DTable* DTable, const void* src,
     return iSize;
 }
 
-size_t HUF_readDTableX4(HUF_DTable* DTable, const void* src, size_t srcSize)
+size_t HUF_readDTableX2(HUF_DTable* DTable, const void* src, size_t srcSize)
 {
   U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-  return HUF_readDTableX4_wksp(DTable, src, srcSize,
+  return HUF_readDTableX2_wksp(DTable, src, srcSize,
                                workSpace, sizeof(workSpace));
 }
 
-static U32 HUF_decodeSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+
+FORCE_INLINE_TEMPLATE U32
+HUF_decodeSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, const U32 dtLog)
 {
     size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
     memcpy(op, dt+val, 2);
@@ -596,7 +665,8 @@ static U32 HUF_decodeSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4
     return dt[val].length;
 }
 
-static U32 HUF_decodeLastSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DEltX4* dt, const U32 dtLog)
+FORCE_INLINE_TEMPLATE U32
+HUF_decodeLastSymbolX2(void* op, BIT_DStream_t* DStream, const HUF_DEltX2* dt, const U32 dtLog)
 {
     size_t const val = BIT_lookBitsFast(DStream, dtLog);   /* note : dtLog >= 1 */
     memcpy(op, dt+val, 1);
@@ -611,45 +681,46 @@ static U32 HUF_decodeLastSymbolX4(void* op, BIT_DStream_t* DStream, const HUF_DE
     return 1;
 }
 
+#define HUF_DECODE_SYMBOLX2_0(ptr, DStreamPtr) \
+    ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog)
 
-#define HUF_DECODE_SYMBOLX4_0(ptr, DStreamPtr) \
-    ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
-
-#define HUF_DECODE_SYMBOLX4_1(ptr, DStreamPtr) \
+#define HUF_DECODE_SYMBOLX2_1(ptr, DStreamPtr) \
     if (MEM_64bits() || (HUF_TABLELOG_MAX<=12)) \
-        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+        ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog)
 
-#define HUF_DECODE_SYMBOLX4_2(ptr, DStreamPtr) \
+#define HUF_DECODE_SYMBOLX2_2(ptr, DStreamPtr) \
     if (MEM_64bits()) \
-        ptr += HUF_decodeSymbolX4(ptr, DStreamPtr, dt, dtLog)
+        ptr += HUF_decodeSymbolX2(ptr, DStreamPtr, dt, dtLog)
 
-HINT_INLINE size_t HUF_decodeStreamX4(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd, const HUF_DEltX4* const dt, const U32 dtLog)
+HINT_INLINE size_t
+HUF_decodeStreamX2(BYTE* p, BIT_DStream_t* bitDPtr, BYTE* const pEnd,
+                const HUF_DEltX2* const dt, const U32 dtLog)
 {
     BYTE* const pStart = p;
 
     /* up to 8 symbols at a time */
     while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p < pEnd-(sizeof(bitDPtr->bitContainer)-1))) {
-        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
-        HUF_DECODE_SYMBOLX4_1(p, bitDPtr);
-        HUF_DECODE_SYMBOLX4_2(p, bitDPtr);
-        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_1(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_2(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
     }
 
     /* closer to end : up to 2 symbols at a time */
     while ((BIT_reloadDStream(bitDPtr) == BIT_DStream_unfinished) & (p <= pEnd-2))
-        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);
 
     while (p <= pEnd-2)
-        HUF_DECODE_SYMBOLX4_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
+        HUF_DECODE_SYMBOLX2_0(p, bitDPtr);   /* no need to reload : reached the end of DStream */
 
     if (p < pEnd)
-        p += HUF_decodeLastSymbolX4(p, bitDPtr, dt, dtLog);
+        p += HUF_decodeLastSymbolX2(p, bitDPtr, dt, dtLog);
 
     return p-pStart;
 }
 
-
-static size_t HUF_decompress1X4_usingDTable_internal(
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress1X2_usingDTable_internal_body(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
     const HUF_DTable* DTable)
@@ -657,17 +728,15 @@ static size_t HUF_decompress1X4_usingDTable_internal(
     BIT_DStream_t bitD;
 
     /* Init */
-    {   size_t const errorCode = BIT_initDStream(&bitD, cSrc, cSrcSize);
-        if (HUF_isError(errorCode)) return errorCode;
-    }
+    CHECK_F( BIT_initDStream(&bitD, cSrc, cSrcSize) );
 
     /* decode */
     {   BYTE* const ostart = (BYTE*) dst;
         BYTE* const oend = ostart + dstSize;
         const void* const dtPtr = DTable+1;   /* force compiler to not use strict-aliasing */
-        const HUF_DEltX4* const dt = (const HUF_DEltX4*)dtPtr;
+        const HUF_DEltX2* const dt = (const HUF_DEltX2*)dtPtr;
         DTableDesc const dtd = HUF_getDTableDesc(DTable);
-        HUF_decodeStreamX4(ostart, &bitD, oend, dt, dtd.tableLog);
+        HUF_decodeStreamX2(ostart, &bitD, oend, dt, dtd.tableLog);
     }
 
     /* check */
@@ -677,47 +746,8 @@ static size_t HUF_decompress1X4_usingDTable_internal(
     return dstSize;
 }
 
-size_t HUF_decompress1X4_usingDTable(
-          void* dst,  size_t dstSize,
-    const void* cSrc, size_t cSrcSize,
-    const HUF_DTable* DTable)
-{
-    DTableDesc dtd = HUF_getDTableDesc(DTable);
-    if (dtd.tableType != 1) return ERROR(GENERIC);
-    return HUF_decompress1X4_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
-}
-
-size_t HUF_decompress1X4_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
-                                   const void* cSrc, size_t cSrcSize,
-                                   void* workSpace, size_t wkspSize)
-{
-    const BYTE* ip = (const BYTE*) cSrc;
-
-    size_t const hSize = HUF_readDTableX4_wksp(DCtx, cSrc, cSrcSize,
-                                               workSpace, wkspSize);
-    if (HUF_isError(hSize)) return hSize;
-    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
-    ip += hSize; cSrcSize -= hSize;
-
-    return HUF_decompress1X4_usingDTable_internal (dst, dstSize, ip, cSrcSize, DCtx);
-}
-
-
-size_t HUF_decompress1X4_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
-                              const void* cSrc, size_t cSrcSize)
-{
-    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_decompress1X4_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
-                                       workSpace, sizeof(workSpace));
-}
-
-size_t HUF_decompress1X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
-{
-    HUF_CREATE_STATIC_DTABLEX4(DTable, HUF_TABLELOG_MAX);
-    return HUF_decompress1X4_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
-}
-
-static size_t HUF_decompress4X4_usingDTable_internal(
+FORCE_INLINE_TEMPLATE size_t
+HUF_decompress4X2_usingDTable_internal_body(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
     const HUF_DTable* DTable)
@@ -727,8 +757,9 @@ static size_t HUF_decompress4X4_usingDTable_internal(
     {   const BYTE* const istart = (const BYTE*) cSrc;
         BYTE* const ostart = (BYTE*) dst;
         BYTE* const oend = ostart + dstSize;
+        BYTE* const olimit = oend - (sizeof(size_t)-1);
         const void* const dtPtr = DTable+1;
-        const HUF_DEltX4* const dt = (const HUF_DEltX4*)dtPtr;
+        const HUF_DEltX2* const dt = (const HUF_DEltX2*)dtPtr;
 
         /* Init */
         BIT_DStream_t bitD1;
@@ -751,41 +782,62 @@ static size_t HUF_decompress4X4_usingDTable_internal(
         BYTE* op2 = opStart2;
         BYTE* op3 = opStart3;
         BYTE* op4 = opStart4;
-        U32 endSignal;
+        U32 endSignal = 1;
         DTableDesc const dtd = HUF_getDTableDesc(DTable);
         U32 const dtLog = dtd.tableLog;
 
         if (length4 > cSrcSize) return ERROR(corruption_detected);   /* overflow */
-        { size_t const errorCode = BIT_initDStream(&bitD1, istart1, length1);
-          if (HUF_isError(errorCode)) return errorCode; }
-        { size_t const errorCode = BIT_initDStream(&bitD2, istart2, length2);
-          if (HUF_isError(errorCode)) return errorCode; }
-        { size_t const errorCode = BIT_initDStream(&bitD3, istart3, length3);
-          if (HUF_isError(errorCode)) return errorCode; }
-        { size_t const errorCode = BIT_initDStream(&bitD4, istart4, length4);
-          if (HUF_isError(errorCode)) return errorCode; }
+        CHECK_F( BIT_initDStream(&bitD1, istart1, length1) );
+        CHECK_F( BIT_initDStream(&bitD2, istart2, length2) );
+        CHECK_F( BIT_initDStream(&bitD3, istart3, length3) );
+        CHECK_F( BIT_initDStream(&bitD4, istart4, length4) );
 
         /* 16-32 symbols per loop (4-8 symbols per stream) */
-        endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
-        for ( ; (endSignal==BIT_DStream_unfinished) & (op4<(oend-(sizeof(bitD4.bitContainer)-1))) ; ) {
-            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
-            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
-            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
-            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
-            HUF_DECODE_SYMBOLX4_1(op1, &bitD1);
-            HUF_DECODE_SYMBOLX4_1(op2, &bitD2);
-            HUF_DECODE_SYMBOLX4_1(op3, &bitD3);
-            HUF_DECODE_SYMBOLX4_1(op4, &bitD4);
-            HUF_DECODE_SYMBOLX4_2(op1, &bitD1);
-            HUF_DECODE_SYMBOLX4_2(op2, &bitD2);
-            HUF_DECODE_SYMBOLX4_2(op3, &bitD3);
-            HUF_DECODE_SYMBOLX4_2(op4, &bitD4);
-            HUF_DECODE_SYMBOLX4_0(op1, &bitD1);
-            HUF_DECODE_SYMBOLX4_0(op2, &bitD2);
-            HUF_DECODE_SYMBOLX4_0(op3, &bitD3);
-            HUF_DECODE_SYMBOLX4_0(op4, &bitD4);
-
-            endSignal = BIT_reloadDStream(&bitD1) | BIT_reloadDStream(&bitD2) | BIT_reloadDStream(&bitD3) | BIT_reloadDStream(&bitD4);
+        for ( ; (endSignal) & (op4 < olimit); ) {
+#if defined(__clang__) && (defined(__x86_64__) || defined(__i386__))
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+            endSignal &= BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished;
+            endSignal &= BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished;
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+            endSignal &= BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished;
+            endSignal &= BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished;
+#else
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_1(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_1(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_1(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_1(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_2(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_2(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_2(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_2(op4, &bitD4);
+            HUF_DECODE_SYMBOLX2_0(op1, &bitD1);
+            HUF_DECODE_SYMBOLX2_0(op2, &bitD2);
+            HUF_DECODE_SYMBOLX2_0(op3, &bitD3);
+            HUF_DECODE_SYMBOLX2_0(op4, &bitD4);
+            endSignal = (U32)LIKELY(
+                        (BIT_reloadDStreamFast(&bitD1) == BIT_DStream_unfinished)
+                      & (BIT_reloadDStreamFast(&bitD2) == BIT_DStream_unfinished)
+                      & (BIT_reloadDStreamFast(&bitD3) == BIT_DStream_unfinished)
+                      & (BIT_reloadDStreamFast(&bitD4) == BIT_DStream_unfinished));
+#endif
         }
 
         /* check corruption */
@@ -795,10 +847,10 @@ static size_t HUF_decompress4X4_usingDTable_internal(
         /* note : op4 already verified within main loop */
 
         /* finish bitStreams one by one */
-        HUF_decodeStreamX4(op1, &bitD1, opStart2, dt, dtLog);
-        HUF_decodeStreamX4(op2, &bitD2, opStart3, dt, dtLog);
-        HUF_decodeStreamX4(op3, &bitD3, opStart4, dt, dtLog);
-        HUF_decodeStreamX4(op4, &bitD4, oend,     dt, dtLog);
+        HUF_decodeStreamX2(op1, &bitD1, opStart2, dt, dtLog);
+        HUF_decodeStreamX2(op2, &bitD2, opStart3, dt, dtLog);
+        HUF_decodeStreamX2(op3, &bitD3, opStart4, dt, dtLog);
+        HUF_decodeStreamX2(op4, &bitD4, oend,     dt, dtLog);
 
         /* check */
         { U32 const endCheck = BIT_endOfDStream(&bitD1) & BIT_endOfDStream(&bitD2) & BIT_endOfDStream(&bitD3) & BIT_endOfDStream(&bitD4);
@@ -809,60 +861,120 @@ static size_t HUF_decompress4X4_usingDTable_internal(
     }
 }
 
+HUF_DGEN(HUF_decompress1X2_usingDTable_internal)
+HUF_DGEN(HUF_decompress4X2_usingDTable_internal)
 
-size_t HUF_decompress4X4_usingDTable(
+size_t HUF_decompress1X2_usingDTable(
           void* dst,  size_t dstSize,
     const void* cSrc, size_t cSrcSize,
     const HUF_DTable* DTable)
 {
     DTableDesc dtd = HUF_getDTableDesc(DTable);
     if (dtd.tableType != 1) return ERROR(GENERIC);
-    return HUF_decompress4X4_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable);
+    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
 }
 
-
-size_t HUF_decompress4X4_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+size_t HUF_decompress1X2_DCtx_wksp(HUF_DTable* DCtx, void* dst, size_t dstSize,
                                    const void* cSrc, size_t cSrcSize,
                                    void* workSpace, size_t wkspSize)
 {
     const BYTE* ip = (const BYTE*) cSrc;
 
-    size_t hSize = HUF_readDTableX4_wksp(dctx, cSrc, cSrcSize,
+    size_t const hSize = HUF_readDTableX2_wksp(DCtx, cSrc, cSrcSize,
+                                               workSpace, wkspSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress1X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, DCtx, /* bmi2 */ 0);
+}
+
+
+size_t HUF_decompress1X2_DCtx(HUF_DTable* DCtx, void* dst, size_t dstSize,
+                              const void* cSrc, size_t cSrcSize)
+{
+    U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
+    return HUF_decompress1X2_DCtx_wksp(DCtx, dst, dstSize, cSrc, cSrcSize,
+                                       workSpace, sizeof(workSpace));
+}
+
+size_t HUF_decompress1X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+{
+    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
+    return HUF_decompress1X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+}
+
+size_t HUF_decompress4X2_usingDTable(
+          void* dst,  size_t dstSize,
+    const void* cSrc, size_t cSrcSize,
+    const HUF_DTable* DTable)
+{
+    DTableDesc dtd = HUF_getDTableDesc(DTable);
+    if (dtd.tableType != 1) return ERROR(GENERIC);
+    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+}
+
+static size_t HUF_decompress4X2_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize, int bmi2)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t hSize = HUF_readDTableX2_wksp(dctx, cSrc, cSrcSize,
                                          workSpace, wkspSize);
     if (HUF_isError(hSize)) return hSize;
     if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
     ip += hSize; cSrcSize -= hSize;
 
-    return HUF_decompress4X4_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx);
+    return HUF_decompress4X2_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
 }
 
+size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
+                                   const void* cSrc, size_t cSrcSize,
+                                   void* workSpace, size_t wkspSize)
+{
+    return HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, /* bmi2 */ 0);
+}
 
-size_t HUF_decompress4X4_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
+
+size_t HUF_decompress4X2_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
                               const void* cSrc, size_t cSrcSize)
 {
     U32 workSpace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    return HUF_decompress4X4_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
+    return HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
                                        workSpace, sizeof(workSpace));
 }
 
-size_t HUF_decompress4X4 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
+size_t HUF_decompress4X2 (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
 {
-    HUF_CREATE_STATIC_DTABLEX4(DTable, HUF_TABLELOG_MAX);
-    return HUF_decompress4X4_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
+    HUF_CREATE_STATIC_DTABLEX2(DTable, HUF_TABLELOG_MAX);
+    return HUF_decompress4X2_DCtx(DTable, dst, dstSize, cSrc, cSrcSize);
 }
 
+#endif /* HUF_FORCE_DECOMPRESS_X1 */
 
-/* ********************************/
-/* Generic decompression selector */
-/* ********************************/
+
+/* ***********************************/
+/* Universal decompression selectors */
+/* ***********************************/
 
 size_t HUF_decompress1X_usingDTable(void* dst, size_t maxDstSize,
                                     const void* cSrc, size_t cSrcSize,
                                     const HUF_DTable* DTable)
 {
     DTableDesc const dtd = HUF_getDTableDesc(DTable);
-    return dtd.tableType ? HUF_decompress1X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable) :
-                           HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+    (void)dtd;
+    assert(dtd.tableType == 0);
+    return HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+    (void)dtd;
+    assert(dtd.tableType == 1);
+    return HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+#else
+    return dtd.tableType ? HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0) :
+                           HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+#endif
 }
 
 size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize,
@@ -870,11 +982,22 @@ size_t HUF_decompress4X_usingDTable(void* dst, size_t maxDstSize,
                                     const HUF_DTable* DTable)
 {
     DTableDesc const dtd = HUF_getDTableDesc(DTable);
-    return dtd.tableType ? HUF_decompress4X4_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable) :
-                           HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+    (void)dtd;
+    assert(dtd.tableType == 0);
+    return HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+    (void)dtd;
+    assert(dtd.tableType == 1);
+    return HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+#else
+    return dtd.tableType ? HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0) :
+                           HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, /* bmi2 */ 0);
+#endif
 }
 
 
+#if !defined(HUF_FORCE_DECOMPRESS_X1) && !defined(HUF_FORCE_DECOMPRESS_X2)
 typedef struct { U32 tableTime; U32 decode256Time; } algo_time_t;
 static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, quad */] =
 {
@@ -896,22 +1019,35 @@ static const algo_time_t algoTime[16 /* Quantization */][3 /* single, double, qu
     {{1455,128}, {2422,124}, {4174,124}},   /* Q ==14 : 87-93% */
     {{ 722,128}, {1891,145}, {1936,146}},   /* Q ==15 : 93-99% */
 };
+#endif
 
 /** HUF_selectDecoder() :
-*   Tells which decoder is likely to decode faster,
-*   based on a set of pre-determined metrics.
-*   @return : 0==HUF_decompress4X2, 1==HUF_decompress4X4 .
-*   Assumption : 0 < cSrcSize, dstSize <= 128 KB */
+ *  Tells which decoder is likely to decode faster,
+ *  based on a set of pre-computed metrics.
+ * @return : 0==HUF_decompress4X1, 1==HUF_decompress4X2 .
+ *  Assumption : 0 < dstSize <= 128 KB */
 U32 HUF_selectDecoder (size_t dstSize, size_t cSrcSize)
 {
+    assert(dstSize > 0);
+    assert(dstSize <= 128*1024);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+    (void)dstSize;
+    (void)cSrcSize;
+    return 0;
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+    (void)dstSize;
+    (void)cSrcSize;
+    return 1;
+#else
     /* decoder timing evaluation */
-    U32 const Q = cSrcSize >= dstSize ? 15 : (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 */
-    U32 const D256 = (U32)(dstSize >> 8);
-    U32 const DTime0 = algoTime[Q][0].tableTime + (algoTime[Q][0].decode256Time * D256);
-    U32 DTime1 = algoTime[Q][1].tableTime + (algoTime[Q][1].decode256Time * D256);
-    DTime1 += DTime1 >> 3;  /* advantage to algorithm using less memory, for cache eviction */
-
-    return DTime1 < DTime0;
+    {   U32 const Q = (cSrcSize >= dstSize) ? 15 : (U32)(cSrcSize * 16 / dstSize);   /* Q < 16 */
+        U32 const D256 = (U32)(dstSize >> 8);
+        U32 const DTime0 = algoTime[Q][0].tableTime + (algoTime[Q][0].decode256Time * D256);
+        U32 DTime1 = algoTime[Q][1].tableTime + (algoTime[Q][1].decode256Time * D256);
+        DTime1 += DTime1 >> 3;  /* advantage to algorithm using less memory, to reduce cache eviction */
+        return DTime1 < DTime0;
+    }
+#endif
 }
 
 
@@ -919,7 +1055,9 @@ typedef size_t (*decompressionAlgo)(void* dst, size_t dstSize, const void* cSrc,
 
 size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize)
 {
-    static const decompressionAlgo decompress[2] = { HUF_decompress4X2, HUF_decompress4X4 };
+#if !defined(HUF_FORCE_DECOMPRESS_X1) && !defined(HUF_FORCE_DECOMPRESS_X2)
+    static const decompressionAlgo decompress[2] = { HUF_decompress4X1, HUF_decompress4X2 };
+#endif
 
     /* validation checks */
     if (dstSize == 0) return ERROR(dstSize_tooSmall);
@@ -928,7 +1066,17 @@ size_t HUF_decompress (void* dst, size_t dstSize, const void* cSrc, size_t cSrcS
     if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
 
     {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+        (void)algoNb;
+        assert(algoNb == 0);
+        return HUF_decompress4X1(dst, dstSize, cSrc, cSrcSize);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+        (void)algoNb;
+        assert(algoNb == 1);
+        return HUF_decompress4X2(dst, dstSize, cSrc, cSrcSize);
+#else
         return decompress[algoNb](dst, dstSize, cSrc, cSrcSize);
+#endif
     }
 }
 
@@ -941,8 +1089,18 @@ size_t HUF_decompress4X_DCtx (HUF_DTable* dctx, void* dst, size_t dstSize, const
     if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
 
     {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
-        return algoNb ? HUF_decompress4X4_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
-                        HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+        (void)algoNb;
+        assert(algoNb == 0);
+        return HUF_decompress4X1_DCtx(dctx, dst, dstSize, cSrc, cSrcSize);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+        (void)algoNb;
+        assert(algoNb == 1);
+        return HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize);
+#else
+        return algoNb ? HUF_decompress4X2_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) :
+                        HUF_decompress4X1_DCtx(dctx, dst, dstSize, cSrc, cSrcSize) ;
+#endif
     }
 }
 
@@ -964,8 +1122,19 @@ size_t HUF_decompress4X_hufOnly_wksp(HUF_DTable* dctx, void* dst,
     if (cSrcSize == 0) return ERROR(corruption_detected);
 
     {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
-        return algoNb ? HUF_decompress4X4_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize):
-                        HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+        (void)algoNb;
+        assert(algoNb == 0);
+        return HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+        (void)algoNb;
+        assert(algoNb == 1);
+        return HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize);
+#else
+        return algoNb ? HUF_decompress4X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                            cSrcSize, workSpace, wkspSize):
+                        HUF_decompress4X1_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize);
+#endif
     }
 }
 
@@ -980,10 +1149,22 @@ size_t HUF_decompress1X_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
     if (cSrcSize == 1) { memset(dst, *(const BYTE*)cSrc, dstSize); return dstSize; }   /* RLE */
 
     {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
-        return algoNb ? HUF_decompress1X4_DCtx_wksp(dctx, dst, dstSize, cSrc,
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+        (void)algoNb;
+        assert(algoNb == 0);
+        return HUF_decompress1X1_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                                cSrcSize, workSpace, wkspSize);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+        (void)algoNb;
+        assert(algoNb == 1);
+        return HUF_decompress1X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                                cSrcSize, workSpace, wkspSize);
+#else
+        return algoNb ? HUF_decompress1X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
                                 cSrcSize, workSpace, wkspSize):
-                        HUF_decompress1X2_DCtx_wksp(dctx, dst, dstSize, cSrc,
+                        HUF_decompress1X1_DCtx_wksp(dctx, dst, dstSize, cSrc,
                                 cSrcSize, workSpace, wkspSize);
+#endif
     }
 }
 
@@ -994,3 +1175,74 @@ size_t HUF_decompress1X_DCtx(HUF_DTable* dctx, void* dst, size_t dstSize,
     return HUF_decompress1X_DCtx_wksp(dctx, dst, dstSize, cSrc, cSrcSize,
                                       workSpace, sizeof(workSpace));
 }
+
+
+size_t HUF_decompress1X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2)
+{
+    DTableDesc const dtd = HUF_getDTableDesc(DTable);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+    (void)dtd;
+    assert(dtd.tableType == 0);
+    return HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+    (void)dtd;
+    assert(dtd.tableType == 1);
+    return HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+#else
+    return dtd.tableType ? HUF_decompress1X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2) :
+                           HUF_decompress1X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+#endif
+}
+
+#ifndef HUF_FORCE_DECOMPRESS_X2
+size_t HUF_decompress1X1_DCtx_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2)
+{
+    const BYTE* ip = (const BYTE*) cSrc;
+
+    size_t const hSize = HUF_readDTableX1_wksp(dctx, cSrc, cSrcSize, workSpace, wkspSize);
+    if (HUF_isError(hSize)) return hSize;
+    if (hSize >= cSrcSize) return ERROR(srcSize_wrong);
+    ip += hSize; cSrcSize -= hSize;
+
+    return HUF_decompress1X1_usingDTable_internal(dst, dstSize, ip, cSrcSize, dctx, bmi2);
+}
+#endif
+
+size_t HUF_decompress4X_usingDTable_bmi2(void* dst, size_t maxDstSize, const void* cSrc, size_t cSrcSize, const HUF_DTable* DTable, int bmi2)
+{
+    DTableDesc const dtd = HUF_getDTableDesc(DTable);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+    (void)dtd;
+    assert(dtd.tableType == 0);
+    return HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+    (void)dtd;
+    assert(dtd.tableType == 1);
+    return HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+#else
+    return dtd.tableType ? HUF_decompress4X2_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2) :
+                           HUF_decompress4X1_usingDTable_internal(dst, maxDstSize, cSrc, cSrcSize, DTable, bmi2);
+#endif
+}
+
+size_t HUF_decompress4X_hufOnly_wksp_bmi2(HUF_DTable* dctx, void* dst, size_t dstSize, const void* cSrc, size_t cSrcSize, void* workSpace, size_t wkspSize, int bmi2)
+{
+    /* validation checks */
+    if (dstSize == 0) return ERROR(dstSize_tooSmall);
+    if (cSrcSize == 0) return ERROR(corruption_detected);
+
+    {   U32 const algoNb = HUF_selectDecoder(dstSize, cSrcSize);
+#if defined(HUF_FORCE_DECOMPRESS_X1)
+        (void)algoNb;
+        assert(algoNb == 0);
+        return HUF_decompress4X1_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
+#elif defined(HUF_FORCE_DECOMPRESS_X2)
+        (void)algoNb;
+        assert(algoNb == 1);
+        return HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
+#else
+        return algoNb ? HUF_decompress4X2_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2) :
+                        HUF_decompress4X1_DCtx_wksp_bmi2(dctx, dst, dstSize, cSrc, cSrcSize, workSpace, wkspSize, bmi2);
+#endif
+    }
+}
diff --git a/contrib/zstd/mem.h b/contrib/zstd/mem.h
index 23335c314..89c8aea7d 100644
--- a/contrib/zstd/mem.h
+++ b/contrib/zstd/mem.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -39,10 +39,87 @@ extern "C" {
 #  define MEM_STATIC static  /* this version may generate warnings for unused static functions; disable the relevant warning */
 #endif
 
+#ifndef __has_builtin
+#  define __has_builtin(x) 0  /* compat. with non-clang compilers */
+#endif
+
 /* code only tested on 32 and 64 bits systems */
 #define MEM_STATIC_ASSERT(c)   { enum { MEM_static_assert = 1/(int)(!!(c)) }; }
 MEM_STATIC void MEM_check(void) { MEM_STATIC_ASSERT((sizeof(size_t)==4) || (sizeof(size_t)==8)); }
 
+/* detects whether we are being compiled under msan */
+#if defined (__has_feature)
+#  if __has_feature(memory_sanitizer)
+#    define MEMORY_SANITIZER 1
+#  endif
+#endif
+
+#if defined (MEMORY_SANITIZER)
+/* Not all platforms that support msan provide sanitizers/msan_interface.h.
+ * We therefore declare the functions we need ourselves, rather than trying to
+ * include the header file... */
+
+#include <stdint.h> /* intptr_t */
+
+/* Make memory region fully initialized (without changing its contents). */
+void __msan_unpoison(const volatile void *a, size_t size);
+
+/* Make memory region fully uninitialized (without changing its contents).
+   This is a legacy interface that does not update origin information. Use
+   __msan_allocated_memory() instead. */
+void __msan_poison(const volatile void *a, size_t size);
+
+/* Returns the offset of the first (at least partially) poisoned byte in the
+   memory range, or -1 if the whole range is good. */
+intptr_t __msan_test_shadow(const volatile void *x, size_t size);
+#endif
+
+/* detects whether we are being compiled under asan */
+#if defined (__has_feature)
+#  if __has_feature(address_sanitizer)
+#    define ADDRESS_SANITIZER 1
+#  endif
+#elif defined(__SANITIZE_ADDRESS__)
+#  define ADDRESS_SANITIZER 1
+#endif
+
+#if defined (ADDRESS_SANITIZER)
+/* Not all platforms that support asan provide sanitizers/asan_interface.h.
+ * We therefore declare the functions we need ourselves, rather than trying to
+ * include the header file... */
+
+/**
+ * Marks a memory region (<c>[addr, addr+size)</c>) as unaddressable.
+ *
+ * This memory must be previously allocated by your program. Instrumented
+ * code is forbidden from accessing addresses in this region until it is
+ * unpoisoned. This function is not guaranteed to poison the entire region -
+ * it could poison only a subregion of <c>[addr, addr+size)</c> due to ASan
+ * alignment restrictions.
+ *
+ * \note This function is not thread-safe because no two threads can poison or
+ * unpoison memory in the same memory region simultaneously.
+ *
+ * \param addr Start of memory region.
+ * \param size Size of memory region. */
+void __asan_poison_memory_region(void const volatile *addr, size_t size);
+
+/**
+ * Marks a memory region (<c>[addr, addr+size)</c>) as addressable.
+ *
+ * This memory must be previously allocated by your program. Accessing
+ * addresses in this region is allowed until this region is poisoned again.
+ * This function could unpoison a super-region of <c>[addr, addr+size)</c> due
+ * to ASan alignment restrictions.
+ *
+ * \note This function is not thread-safe because no two threads can
+ * poison or unpoison memory in the same memory region simultaneously.
+ *
+ * \param addr Start of memory region.
+ * \param size Size of memory region. */
+void __asan_unpoison_memory_region(void const volatile *addr, size_t size);
+#endif
+
 
 /*-**************************************************************
 *  Basic Types
@@ -56,18 +133,26 @@ MEM_STATIC void MEM_check(void) { MEM_STATIC_ASSERT((sizeof(size_t)==4) || (size
   typedef   int32_t S32;
   typedef  uint64_t U64;
   typedef   int64_t S64;
-  typedef  intptr_t iPtrDiff;
-  typedef uintptr_t uPtrDiff;
 #else
+# include <limits.h>
+#if CHAR_BIT != 8
+#  error "this implementation requires char to be exactly 8-bit type"
+#endif
   typedef unsigned char      BYTE;
+#if USHRT_MAX != 65535
+#  error "this implementation requires short to be exactly 16-bit type"
+#endif
   typedef unsigned short      U16;
   typedef   signed short      S16;
+#if UINT_MAX != 4294967295
+#  error "this implementation requires int to be exactly 32-bit type"
+#endif
   typedef unsigned int        U32;
   typedef   signed int        S32;
+/* note : there are no limits defined for long long type in C90.
+ * limits exist in C99, however, in such case, <stdint.h> is preferred */
   typedef unsigned long long  U64;
   typedef   signed long long  S64;
-  typedef ptrdiff_t      iPtrDiff;
-  typedef size_t         uPtrDiff;
 #endif
 
 
@@ -90,7 +175,7 @@ MEM_STATIC void MEM_check(void) { MEM_STATIC_ASSERT((sizeof(size_t)==4) || (size
 #ifndef MEM_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
 #  if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
 #    define MEM_FORCE_MEMORY_ACCESS 2
-#  elif defined(__INTEL_COMPILER) || defined(__GNUC__)
+#  elif defined(__INTEL_COMPILER) || defined(__GNUC__) || defined(__ICCARM__)
 #    define MEM_FORCE_MEMORY_ACCESS 1
 #  endif
 #endif
@@ -123,20 +208,26 @@ MEM_STATIC void MEM_write64(void* memPtr, U64 value) { *(U64*)memPtr = value; }
 /* currently only defined for gcc and icc */
 #if defined(_MSC_VER) || (defined(__INTEL_COMPILER) && defined(WIN32))
     __pragma( pack(push, 1) )
-    typedef union { U16 u16; U32 u32; U64 u64; size_t st; } unalign;
+    typedef struct { U16 v; } unalign16;
+    typedef struct { U32 v; } unalign32;
+    typedef struct { U64 v; } unalign64;
+    typedef struct { size_t v; } unalignArch;
     __pragma( pack(pop) )
 #else
-    typedef union { U16 u16; U32 u32; U64 u64; size_t st; } __attribute__((packed)) unalign;
+    typedef struct { U16 v; } __attribute__((packed)) unalign16;
+    typedef struct { U32 v; } __attribute__((packed)) unalign32;
+    typedef struct { U64 v; } __attribute__((packed)) unalign64;
+    typedef struct { size_t v; } __attribute__((packed)) unalignArch;
 #endif
 
-MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign*)ptr)->u16; }
-MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign*)ptr)->u32; }
-MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign*)ptr)->u64; }
-MEM_STATIC size_t MEM_readST(const void* ptr) { return ((const unalign*)ptr)->st; }
+MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign16*)ptr)->v; }
+MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign32*)ptr)->v; }
+MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign64*)ptr)->v; }
+MEM_STATIC size_t MEM_readST(const void* ptr) { return ((const unalignArch*)ptr)->v; }
 
-MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign*)memPtr)->u16 = value; }
-MEM_STATIC void MEM_write32(void* memPtr, U32 value) { ((unalign*)memPtr)->u32 = value; }
-MEM_STATIC void MEM_write64(void* memPtr, U64 value) { ((unalign*)memPtr)->u64 = value; }
+MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign16*)memPtr)->v = value; }
+MEM_STATIC void MEM_write32(void* memPtr, U32 value) { ((unalign32*)memPtr)->v = value; }
+MEM_STATIC void MEM_write64(void* memPtr, U64 value) { ((unalign64*)memPtr)->v = value; }
 
 #else
 
@@ -184,7 +275,8 @@ MEM_STATIC U32 MEM_swap32(U32 in)
 {
 #if defined(_MSC_VER)     /* Visual Studio */
     return _byteswap_ulong(in);
-#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+#elif (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \
+  || (defined(__clang__) && __has_builtin(__builtin_bswap32))
     return __builtin_bswap32(in);
 #else
     return  ((in << 24) & 0xff000000 ) |
@@ -198,7 +290,8 @@ MEM_STATIC U64 MEM_swap64(U64 in)
 {
 #if defined(_MSC_VER)     /* Visual Studio */
     return _byteswap_uint64(in);
-#elif defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)
+#elif (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \
+  || (defined(__clang__) && __has_builtin(__builtin_bswap64))
     return __builtin_bswap64(in);
 #else
     return  ((in << 56) & 0xff00000000000000ULL) |
diff --git a/contrib/zstd/pool.c b/contrib/zstd/pool.c
index 5f19f331b..aa4b4de0d 100644
--- a/contrib/zstd/pool.c
+++ b/contrib/zstd/pool.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -10,8 +10,9 @@
 
 
 /* ======   Dependencies   ======= */
-#include <stddef.h>  /* size_t */
-#include <stdlib.h>  /* malloc, calloc, free */
+#include <stddef.h>    /* size_t */
+#include "debug.h"     /* assert */
+#include "zstd_internal.h"  /* ZSTD_malloc, ZSTD_free */
 #include "pool.h"
 
 /* ======   Compiler specifics   ====== */
@@ -33,8 +34,9 @@ typedef struct POOL_job_s {
 struct POOL_ctx_s {
     ZSTD_customMem customMem;
     /* Keep track of the threads */
-    pthread_t *threads;
-    size_t numThreads;
+    ZSTD_pthread_t* threads;
+    size_t threadCapacity;
+    size_t threadLimit;
 
     /* The queue is a circular buffer */
     POOL_job *queue;
@@ -48,34 +50,37 @@ struct POOL_ctx_s {
     int queueEmpty;
 
     /* The mutex protects the queue */
-    pthread_mutex_t queueMutex;
+    ZSTD_pthread_mutex_t queueMutex;
     /* Condition variable for pushers to wait on when the queue is full */
-    pthread_cond_t queuePushCond;
+    ZSTD_pthread_cond_t queuePushCond;
     /* Condition variables for poppers to wait on when the queue is empty */
-    pthread_cond_t queuePopCond;
+    ZSTD_pthread_cond_t queuePopCond;
     /* Indicates if the queue is shutting down */
     int shutdown;
 };
 
 /* POOL_thread() :
-   Work thread for the thread pool.
-   Waits for jobs and executes them.
-   @returns : NULL on failure else non-null.
-*/
+ * Work thread for the thread pool.
+ * Waits for jobs and executes them.
+ * @returns : NULL on failure else non-null.
+ */
 static void* POOL_thread(void* opaque) {
     POOL_ctx* const ctx = (POOL_ctx*)opaque;
     if (!ctx) { return NULL; }
     for (;;) {
         /* Lock the mutex and wait for a non-empty queue or until shutdown */
-        pthread_mutex_lock(&ctx->queueMutex);
-
-        while (ctx->queueEmpty && !ctx->shutdown) {
-            pthread_cond_wait(&ctx->queuePopCond, &ctx->queueMutex);
-        }
-        /* empty => shutting down: so stop */
-        if (ctx->queueEmpty) {
-            pthread_mutex_unlock(&ctx->queueMutex);
-            return opaque;
+        ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+
+        while ( ctx->queueEmpty
+            || (ctx->numThreadsBusy >= ctx->threadLimit) ) {
+            if (ctx->shutdown) {
+                /* even if !queueEmpty, (possible if numThreadsBusy >= threadLimit),
+                 * a few threads will be shutdown while !queueEmpty,
+                 * but enough threads will remain active to finish the queue */
+                ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+                return opaque;
+            }
+            ZSTD_pthread_cond_wait(&ctx->queuePopCond, &ctx->queueMutex);
         }
         /* Pop a job off the queue */
         {   POOL_job const job = ctx->queue[ctx->queueHead];
@@ -83,62 +88,69 @@ static void* POOL_thread(void* opaque) {
             ctx->numThreadsBusy++;
             ctx->queueEmpty = ctx->queueHead == ctx->queueTail;
             /* Unlock the mutex, signal a pusher, and run the job */
-            pthread_mutex_unlock(&ctx->queueMutex);
-            pthread_cond_signal(&ctx->queuePushCond);
+            ZSTD_pthread_cond_signal(&ctx->queuePushCond);
+            ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
 
             job.function(job.opaque);
 
             /* If the intended queue size was 0, signal after finishing job */
+            ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+            ctx->numThreadsBusy--;
             if (ctx->queueSize == 1) {
-                pthread_mutex_lock(&ctx->queueMutex);
-                ctx->numThreadsBusy--;
-                pthread_mutex_unlock(&ctx->queueMutex);
-                pthread_cond_signal(&ctx->queuePushCond);
-        }   }
+                ZSTD_pthread_cond_signal(&ctx->queuePushCond);
+            }
+            ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+        }
     }  /* for (;;) */
-    /* Unreachable */
+    assert(0);  /* Unreachable */
 }
 
-POOL_ctx *POOL_create(size_t numThreads, size_t queueSize) {
+POOL_ctx* POOL_create(size_t numThreads, size_t queueSize) {
     return POOL_create_advanced(numThreads, queueSize, ZSTD_defaultCMem);
 }
 
-POOL_ctx *POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customMem customMem) {
-    POOL_ctx *ctx;
-    /* Check the parameters */
+POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize,
+                               ZSTD_customMem customMem) {
+    POOL_ctx* ctx;
+    /* Check parameters */
     if (!numThreads) { return NULL; }
     /* Allocate the context and zero initialize */
-    ctx = (POOL_ctx *)ZSTD_calloc(sizeof(POOL_ctx), customMem);
+    ctx = (POOL_ctx*)ZSTD_calloc(sizeof(POOL_ctx), customMem);
     if (!ctx) { return NULL; }
     /* Initialize the job queue.
-     * It needs one extra space since one space is wasted to differentiate empty
-     * and full queues.
+     * It needs one extra space since one space is wasted to differentiate
+     * empty and full queues.
      */
     ctx->queueSize = queueSize + 1;
-    ctx->queue = (POOL_job*) malloc(ctx->queueSize * sizeof(POOL_job));
+    ctx->queue = (POOL_job*)ZSTD_malloc(ctx->queueSize * sizeof(POOL_job), customMem);
     ctx->queueHead = 0;
     ctx->queueTail = 0;
     ctx->numThreadsBusy = 0;
     ctx->queueEmpty = 1;
-    (void)pthread_mutex_init(&ctx->queueMutex, NULL);
-    (void)pthread_cond_init(&ctx->queuePushCond, NULL);
-    (void)pthread_cond_init(&ctx->queuePopCond, NULL);
+    {
+        int error = 0;
+        error |= ZSTD_pthread_mutex_init(&ctx->queueMutex, NULL);
+        error |= ZSTD_pthread_cond_init(&ctx->queuePushCond, NULL);
+        error |= ZSTD_pthread_cond_init(&ctx->queuePopCond, NULL);
+        if (error) { POOL_free(ctx); return NULL; }
+    }
     ctx->shutdown = 0;
     /* Allocate space for the thread handles */
-    ctx->threads = (pthread_t*)ZSTD_malloc(numThreads * sizeof(pthread_t), customMem);
-    ctx->numThreads = 0;
+    ctx->threads = (ZSTD_pthread_t*)ZSTD_malloc(numThreads * sizeof(ZSTD_pthread_t), customMem);
+    ctx->threadCapacity = 0;
     ctx->customMem = customMem;
     /* Check for errors */
     if (!ctx->threads || !ctx->queue) { POOL_free(ctx); return NULL; }
     /* Initialize the threads */
     {   size_t i;
         for (i = 0; i < numThreads; ++i) {
-            if (pthread_create(&ctx->threads[i], NULL, &POOL_thread, ctx)) {
-                ctx->numThreads = i;
+            if (ZSTD_pthread_create(&ctx->threads[i], NULL, &POOL_thread, ctx)) {
+                ctx->threadCapacity = i;
                 POOL_free(ctx);
                 return NULL;
         }   }
-        ctx->numThreads = numThreads;
+        ctx->threadCapacity = numThreads;
+        ctx->threadLimit = numThreads;
     }
     return ctx;
 }
@@ -146,80 +158,146 @@ POOL_ctx *POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customM
 /*! POOL_join() :
     Shutdown the queue, wake any sleeping threads, and join all of the threads.
 */
-static void POOL_join(POOL_ctx *ctx) {
+static void POOL_join(POOL_ctx* ctx) {
     /* Shut down the queue */
-    pthread_mutex_lock(&ctx->queueMutex);
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
     ctx->shutdown = 1;
-    pthread_mutex_unlock(&ctx->queueMutex);
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
     /* Wake up sleeping threads */
-    pthread_cond_broadcast(&ctx->queuePushCond);
-    pthread_cond_broadcast(&ctx->queuePopCond);
+    ZSTD_pthread_cond_broadcast(&ctx->queuePushCond);
+    ZSTD_pthread_cond_broadcast(&ctx->queuePopCond);
     /* Join all of the threads */
     {   size_t i;
-        for (i = 0; i < ctx->numThreads; ++i) {
-            pthread_join(ctx->threads[i], NULL);
+        for (i = 0; i < ctx->threadCapacity; ++i) {
+            ZSTD_pthread_join(ctx->threads[i], NULL);  /* note : could fail */
     }   }
 }
 
 void POOL_free(POOL_ctx *ctx) {
     if (!ctx) { return; }
     POOL_join(ctx);
-    pthread_mutex_destroy(&ctx->queueMutex);
-    pthread_cond_destroy(&ctx->queuePushCond);
-    pthread_cond_destroy(&ctx->queuePopCond);
+    ZSTD_pthread_mutex_destroy(&ctx->queueMutex);
+    ZSTD_pthread_cond_destroy(&ctx->queuePushCond);
+    ZSTD_pthread_cond_destroy(&ctx->queuePopCond);
     ZSTD_free(ctx->queue, ctx->customMem);
     ZSTD_free(ctx->threads, ctx->customMem);
     ZSTD_free(ctx, ctx->customMem);
 }
 
+
+
 size_t POOL_sizeof(POOL_ctx *ctx) {
     if (ctx==NULL) return 0;  /* supports sizeof NULL */
     return sizeof(*ctx)
         + ctx->queueSize * sizeof(POOL_job)
-        + ctx->numThreads * sizeof(pthread_t);
+        + ctx->threadCapacity * sizeof(ZSTD_pthread_t);
+}
+
+
+/* @return : 0 on success, 1 on error */
+static int POOL_resize_internal(POOL_ctx* ctx, size_t numThreads)
+{
+    if (numThreads <= ctx->threadCapacity) {
+        if (!numThreads) return 1;
+        ctx->threadLimit = numThreads;
+        return 0;
+    }
+    /* numThreads > threadCapacity */
+    {   ZSTD_pthread_t* const threadPool = (ZSTD_pthread_t*)ZSTD_malloc(numThreads * sizeof(ZSTD_pthread_t), ctx->customMem);
+        if (!threadPool) return 1;
+        /* replace existing thread pool */
+        memcpy(threadPool, ctx->threads, ctx->threadCapacity * sizeof(*threadPool));
+        ZSTD_free(ctx->threads, ctx->customMem);
+        ctx->threads = threadPool;
+        /* Initialize additional threads */
+        {   size_t threadId;
+            for (threadId = ctx->threadCapacity; threadId < numThreads; ++threadId) {
+                if (ZSTD_pthread_create(&threadPool[threadId], NULL, &POOL_thread, ctx)) {
+                    ctx->threadCapacity = threadId;
+                    return 1;
+            }   }
+    }   }
+    /* successfully expanded */
+    ctx->threadCapacity = numThreads;
+    ctx->threadLimit = numThreads;
+    return 0;
+}
+
+/* @return : 0 on success, 1 on error */
+int POOL_resize(POOL_ctx* ctx, size_t numThreads)
+{
+    int result;
+    if (ctx==NULL) return 1;
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+    result = POOL_resize_internal(ctx, numThreads);
+    ZSTD_pthread_cond_broadcast(&ctx->queuePopCond);
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+    return result;
 }
 
 /**
  * Returns 1 if the queue is full and 0 otherwise.
  *
- * If the queueSize is 1 (the pool was created with an intended queueSize of 0),
- * then a queue is empty if there is a thread free and no job is waiting.
+ * When queueSize is 1 (pool was created with an intended queueSize of 0),
+ * then a queue is empty if there is a thread free _and_ no job is waiting.
  */
 static int isQueueFull(POOL_ctx const* ctx) {
     if (ctx->queueSize > 1) {
         return ctx->queueHead == ((ctx->queueTail + 1) % ctx->queueSize);
     } else {
-        return ctx->numThreadsBusy == ctx->numThreads ||
+        return (ctx->numThreadsBusy == ctx->threadLimit) ||
                !ctx->queueEmpty;
     }
 }
 
-void POOL_add(void* ctxVoid, POOL_function function, void *opaque) {
-    POOL_ctx* const ctx = (POOL_ctx*)ctxVoid;
-    if (!ctx) { return; }
 
-    pthread_mutex_lock(&ctx->queueMutex);
-    {   POOL_job const job = {function, opaque};
+static void POOL_add_internal(POOL_ctx* ctx, POOL_function function, void *opaque)
+{
+    POOL_job const job = {function, opaque};
+    assert(ctx != NULL);
+    if (ctx->shutdown) return;
 
-        /* Wait until there is space in the queue for the new job */
-        while (isQueueFull(ctx) && !ctx->shutdown) {
-          pthread_cond_wait(&ctx->queuePushCond, &ctx->queueMutex);
-        }
-        /* The queue is still going => there is space */
-        if (!ctx->shutdown) {
-            ctx->queueEmpty = 0;
-            ctx->queue[ctx->queueTail] = job;
-            ctx->queueTail = (ctx->queueTail + 1) % ctx->queueSize;
-        }
+    ctx->queueEmpty = 0;
+    ctx->queue[ctx->queueTail] = job;
+    ctx->queueTail = (ctx->queueTail + 1) % ctx->queueSize;
+    ZSTD_pthread_cond_signal(&ctx->queuePopCond);
+}
+
+void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque)
+{
+    assert(ctx != NULL);
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+    /* Wait until there is space in the queue for the new job */
+    while (isQueueFull(ctx) && (!ctx->shutdown)) {
+        ZSTD_pthread_cond_wait(&ctx->queuePushCond, &ctx->queueMutex);
     }
-    pthread_mutex_unlock(&ctx->queueMutex);
-    pthread_cond_signal(&ctx->queuePopCond);
+    POOL_add_internal(ctx, function, opaque);
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
 }
 
+
+int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque)
+{
+    assert(ctx != NULL);
+    ZSTD_pthread_mutex_lock(&ctx->queueMutex);
+    if (isQueueFull(ctx)) {
+        ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+        return 0;
+    }
+    POOL_add_internal(ctx, function, opaque);
+    ZSTD_pthread_mutex_unlock(&ctx->queueMutex);
+    return 1;
+}
+
+
 #else  /* ZSTD_MULTITHREAD  not defined */
+
+/* ========================== */
 /* No multi-threading support */
+/* ========================== */
+
 
-/* We don't need any data, but if it is empty malloc() might return NULL. */
+/* We don't need any data, but if it is empty, malloc() might return NULL. */
 struct POOL_ctx_s {
     int dummy;
 };
@@ -241,9 +319,20 @@ void POOL_free(POOL_ctx* ctx) {
     (void)ctx;
 }
 
-void POOL_add(void* ctx, POOL_function function, void* opaque) {
+int POOL_resize(POOL_ctx* ctx, size_t numThreads) {
+    (void)ctx; (void)numThreads;
+    return 0;
+}
+
+void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque) {
+    (void)ctx;
+    function(opaque);
+}
+
+int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque) {
     (void)ctx;
     function(opaque);
+    return 1;
 }
 
 size_t POOL_sizeof(POOL_ctx* ctx) {
diff --git a/contrib/zstd/pool.h b/contrib/zstd/pool.h
index 08c63715a..259bafc97 100644
--- a/contrib/zstd/pool.h
+++ b/contrib/zstd/pool.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -17,7 +17,8 @@ extern "C" {
 
 
 #include <stddef.h>   /* size_t */
-#include "zstd_internal.h"   /* ZSTD_customMem */
+#define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_customMem */
+#include "../zstd.h"
 
 typedef struct POOL_ctx_s POOL_ctx;
 
@@ -27,35 +28,53 @@ typedef struct POOL_ctx_s POOL_ctx;
  *  The maximum number of queued jobs before blocking is `queueSize`.
  * @return : POOL_ctx pointer on success, else NULL.
 */
-POOL_ctx *POOL_create(size_t numThreads, size_t queueSize);
+POOL_ctx* POOL_create(size_t numThreads, size_t queueSize);
 
-POOL_ctx *POOL_create_advanced(size_t numThreads, size_t queueSize, ZSTD_customMem customMem);
+POOL_ctx* POOL_create_advanced(size_t numThreads, size_t queueSize,
+                               ZSTD_customMem customMem);
 
 /*! POOL_free() :
-    Free a thread pool returned by POOL_create().
-*/
-void POOL_free(POOL_ctx *ctx);
+ *  Free a thread pool returned by POOL_create().
+ */
+void POOL_free(POOL_ctx* ctx);
+
+/*! POOL_resize() :
+ *  Expands or shrinks pool's number of threads.
+ *  This is more efficient than releasing + creating a new context,
+ *  since it tries to preserve and re-use existing threads.
+ * `numThreads` must be at least 1.
+ * @return : 0 when resize was successful,
+ *           !0 (typically 1) if there is an error.
+ *    note : only numThreads can be resized, queueSize remains unchanged.
+ */
+int POOL_resize(POOL_ctx* ctx, size_t numThreads);
 
 /*! POOL_sizeof() :
-    return memory usage of pool returned by POOL_create().
-*/
-size_t POOL_sizeof(POOL_ctx *ctx);
+ * @return threadpool memory usage
+ *  note : compatible with NULL (returns 0 in this case)
+ */
+size_t POOL_sizeof(POOL_ctx* ctx);
 
 /*! POOL_function :
-    The function type that can be added to a thread pool.
-*/
-typedef void (*POOL_function)(void *);
-/*! POOL_add_function :
-    The function type for a generic thread pool add function.
-*/
-typedef void (*POOL_add_function)(void *, POOL_function, void *);
+ *  The function type that can be added to a thread pool.
+ */
+typedef void (*POOL_function)(void*);
 
 /*! POOL_add() :
-    Add the job `function(opaque)` to the thread pool.
-    Possibly blocks until there is room in the queue.
-    Note : The function may be executed asynchronously, so `opaque` must live until the function has been completed.
-*/
-void POOL_add(void *ctx, POOL_function function, void *opaque);
+ *  Add the job `function(opaque)` to the thread pool. `ctx` must be valid.
+ *  Possibly blocks until there is room in the queue.
+ *  Note : The function may be executed asynchronously,
+ *         therefore, `opaque` must live until function has been completed.
+ */
+void POOL_add(POOL_ctx* ctx, POOL_function function, void* opaque);
+
+
+/*! POOL_tryAdd() :
+ *  Add the job `function(opaque)` to thread pool _if_ a worker is available.
+ *  Returns immediately even if not (does not block).
+ * @return : 1 if successful, 0 if not.
+ */
+int POOL_tryAdd(POOL_ctx* ctx, POOL_function function, void* opaque);
 
 
 #if defined (__cplusplus)
diff --git a/contrib/zstd/threading.c b/contrib/zstd/threading.c
index a82c975b2..e2edb313e 100644
--- a/contrib/zstd/threading.c
+++ b/contrib/zstd/threading.c
@@ -2,20 +2,23 @@
  * Copyright (c) 2016 Tino Reichardt
  * All rights reserved.
  *
+ * You can contact the author at:
+ * - zstdmt source repository: https://github.com/mcmilk/zstdmt
+ *
  * This source code is licensed under both the BSD-style license (found in the
  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
  * in the COPYING file in the root directory of this source tree).
- *
- * You can contact the author at:
- * - zstdmt source repository: https://github.com/mcmilk/zstdmt
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 /**
  * This file will hold wrapper for systems, which do not support pthreads
  */
 
-/* create fake symbol to avoid empty trnaslation unit warning */
-int g_ZSTD_threading_useles_symbol;
+#include "threading.h"
+
+/* create fake symbol to avoid empty translation unit warning */
+int g_ZSTD_threading_useless_symbol;
 
 #if defined(ZSTD_MULTITHREAD) && defined(_WIN32)
 
@@ -28,19 +31,18 @@ int g_ZSTD_threading_useles_symbol;
 /* ===  Dependencies  === */
 #include <process.h>
 #include <errno.h>
-#include "threading.h"
 
 
 /* ===  Implementation  === */
 
 static unsigned __stdcall worker(void *arg)
 {
-    pthread_t* const thread = (pthread_t*) arg;
+    ZSTD_pthread_t* const thread = (ZSTD_pthread_t*) arg;
     thread->arg = thread->start_routine(thread->arg);
     return 0;
 }
 
-int pthread_create(pthread_t* thread, const void* unused,
+int ZSTD_pthread_create(ZSTD_pthread_t* thread, const void* unused,
             void* (*start_routine) (void*), void* arg)
 {
     (void)unused;
@@ -54,16 +56,16 @@ int pthread_create(pthread_t* thread, const void* unused,
         return 0;
 }
 
-int _pthread_join(pthread_t * thread, void **value_ptr)
+int ZSTD_pthread_join(ZSTD_pthread_t thread, void **value_ptr)
 {
     DWORD result;
 
-    if (!thread->handle) return 0;
+    if (!thread.handle) return 0;
 
-    result = WaitForSingleObject(thread->handle, INFINITE);
+    result = WaitForSingleObject(thread.handle, INFINITE);
     switch (result) {
     case WAIT_OBJECT_0:
-        if (value_ptr) *value_ptr = thread->arg;
+        if (value_ptr) *value_ptr = thread.arg;
         return 0;
     case WAIT_ABANDONED:
         return EINVAL;
@@ -73,3 +75,47 @@ int _pthread_join(pthread_t * thread, void **value_ptr)
 }
 
 #endif   /* ZSTD_MULTITHREAD */
+
+#if defined(ZSTD_MULTITHREAD) && DEBUGLEVEL >= 1 && !defined(_WIN32)
+
+#include <stdlib.h>
+
+int ZSTD_pthread_mutex_init(ZSTD_pthread_mutex_t* mutex, pthread_mutexattr_t const* attr)
+{
+    *mutex = (pthread_mutex_t*)malloc(sizeof(pthread_mutex_t));
+    if (!*mutex)
+        return 1;
+    return pthread_mutex_init(*mutex, attr);
+}
+
+int ZSTD_pthread_mutex_destroy(ZSTD_pthread_mutex_t* mutex)
+{
+    if (!*mutex)
+        return 0;
+    {
+        int const ret = pthread_mutex_destroy(*mutex);
+        free(*mutex);
+        return ret;
+    }
+}
+
+int ZSTD_pthread_cond_init(ZSTD_pthread_cond_t* cond, pthread_condattr_t const* attr)
+{
+    *cond = (pthread_cond_t*)malloc(sizeof(pthread_cond_t));
+    if (!*cond)
+        return 1;
+    return pthread_cond_init(*cond, attr);
+}
+
+int ZSTD_pthread_cond_destroy(ZSTD_pthread_cond_t* cond)
+{
+    if (!*cond)
+        return 0;
+    {
+        int const ret = pthread_cond_destroy(*cond);
+        free(*cond);
+        return ret;
+    }
+}
+
+#endif
diff --git a/contrib/zstd/threading.h b/contrib/zstd/threading.h
index 8194bc6fa..fd0060d5a 100644
--- a/contrib/zstd/threading.h
+++ b/contrib/zstd/threading.h
@@ -2,17 +2,20 @@
  * Copyright (c) 2016 Tino Reichardt
  * All rights reserved.
  *
+ * You can contact the author at:
+ * - zstdmt source repository: https://github.com/mcmilk/zstdmt
+ *
  * This source code is licensed under both the BSD-style license (found in the
  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
  * in the COPYING file in the root directory of this source tree).
- *
- * You can contact the author at:
- * - zstdmt source repository: https://github.com/mcmilk/zstdmt
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 #ifndef THREADING_H_938743
 #define THREADING_H_938743
 
+#include "debug.h"
+
 #if defined (__cplusplus)
 extern "C" {
 #endif
@@ -44,59 +47,104 @@ extern "C" {
 
 
 /* mutex */
-#define pthread_mutex_t           CRITICAL_SECTION
-#define pthread_mutex_init(a,b)   (InitializeCriticalSection((a)), 0)
-#define pthread_mutex_destroy(a)  DeleteCriticalSection((a))
-#define pthread_mutex_lock(a)     EnterCriticalSection((a))
-#define pthread_mutex_unlock(a)   LeaveCriticalSection((a))
+#define ZSTD_pthread_mutex_t           CRITICAL_SECTION
+#define ZSTD_pthread_mutex_init(a, b)  ((void)(b), InitializeCriticalSection((a)), 0)
+#define ZSTD_pthread_mutex_destroy(a)  DeleteCriticalSection((a))
+#define ZSTD_pthread_mutex_lock(a)     EnterCriticalSection((a))
+#define ZSTD_pthread_mutex_unlock(a)   LeaveCriticalSection((a))
 
 /* condition variable */
-#define pthread_cond_t             CONDITION_VARIABLE
-#define pthread_cond_init(a, b)    (InitializeConditionVariable((a)), 0)
-#define pthread_cond_destroy(a)    /* No delete */
-#define pthread_cond_wait(a, b)    SleepConditionVariableCS((a), (b), INFINITE)
-#define pthread_cond_signal(a)     WakeConditionVariable((a))
-#define pthread_cond_broadcast(a)  WakeAllConditionVariable((a))
-
-/* pthread_create() and pthread_join() */
+#define ZSTD_pthread_cond_t             CONDITION_VARIABLE
+#define ZSTD_pthread_cond_init(a, b)    ((void)(b), InitializeConditionVariable((a)), 0)
+#define ZSTD_pthread_cond_destroy(a)    ((void)(a))
+#define ZSTD_pthread_cond_wait(a, b)    SleepConditionVariableCS((a), (b), INFINITE)
+#define ZSTD_pthread_cond_signal(a)     WakeConditionVariable((a))
+#define ZSTD_pthread_cond_broadcast(a)  WakeAllConditionVariable((a))
+
+/* ZSTD_pthread_create() and ZSTD_pthread_join() */
 typedef struct {
     HANDLE handle;
     void* (*start_routine)(void*);
     void* arg;
-} pthread_t;
+} ZSTD_pthread_t;
 
-int pthread_create(pthread_t* thread, const void* unused,
+int ZSTD_pthread_create(ZSTD_pthread_t* thread, const void* unused,
                    void* (*start_routine) (void*), void* arg);
 
-#define pthread_join(a, b) _pthread_join(&(a), (b))
-int _pthread_join(pthread_t* thread, void** value_ptr);
+int ZSTD_pthread_join(ZSTD_pthread_t thread, void** value_ptr);
 
 /**
  * add here more wrappers as required
  */
 
 
-#elif defined(ZSTD_MULTITHREAD)   /* posix assumed ; need a better detection method */
+#elif defined(ZSTD_MULTITHREAD)    /* posix assumed ; need a better detection method */
 /* ===   POSIX Systems   === */
 #  include <pthread.h>
 
+#if DEBUGLEVEL < 1
+
+#define ZSTD_pthread_mutex_t            pthread_mutex_t
+#define ZSTD_pthread_mutex_init(a, b)   pthread_mutex_init((a), (b))
+#define ZSTD_pthread_mutex_destroy(a)   pthread_mutex_destroy((a))
+#define ZSTD_pthread_mutex_lock(a)      pthread_mutex_lock((a))
+#define ZSTD_pthread_mutex_unlock(a)    pthread_mutex_unlock((a))
+
+#define ZSTD_pthread_cond_t             pthread_cond_t
+#define ZSTD_pthread_cond_init(a, b)    pthread_cond_init((a), (b))
+#define ZSTD_pthread_cond_destroy(a)    pthread_cond_destroy((a))
+#define ZSTD_pthread_cond_wait(a, b)    pthread_cond_wait((a), (b))
+#define ZSTD_pthread_cond_signal(a)     pthread_cond_signal((a))
+#define ZSTD_pthread_cond_broadcast(a)  pthread_cond_broadcast((a))
+
+#define ZSTD_pthread_t                  pthread_t
+#define ZSTD_pthread_create(a, b, c, d) pthread_create((a), (b), (c), (d))
+#define ZSTD_pthread_join(a, b)         pthread_join((a),(b))
+
+#else /* DEBUGLEVEL >= 1 */
+
+/* Debug implementation of threading.
+ * In this implementation we use pointers for mutexes and condition variables.
+ * This way, if we forget to init/destroy them the program will crash or ASAN
+ * will report leaks.
+ */
+
+#define ZSTD_pthread_mutex_t            pthread_mutex_t*
+int ZSTD_pthread_mutex_init(ZSTD_pthread_mutex_t* mutex, pthread_mutexattr_t const* attr);
+int ZSTD_pthread_mutex_destroy(ZSTD_pthread_mutex_t* mutex);
+#define ZSTD_pthread_mutex_lock(a)      pthread_mutex_lock(*(a))
+#define ZSTD_pthread_mutex_unlock(a)    pthread_mutex_unlock(*(a))
+
+#define ZSTD_pthread_cond_t             pthread_cond_t*
+int ZSTD_pthread_cond_init(ZSTD_pthread_cond_t* cond, pthread_condattr_t const* attr);
+int ZSTD_pthread_cond_destroy(ZSTD_pthread_cond_t* cond);
+#define ZSTD_pthread_cond_wait(a, b)    pthread_cond_wait(*(a), *(b))
+#define ZSTD_pthread_cond_signal(a)     pthread_cond_signal(*(a))
+#define ZSTD_pthread_cond_broadcast(a)  pthread_cond_broadcast(*(a))
+
+#define ZSTD_pthread_t                  pthread_t
+#define ZSTD_pthread_create(a, b, c, d) pthread_create((a), (b), (c), (d))
+#define ZSTD_pthread_join(a, b)         pthread_join((a),(b))
+
+#endif
+
 #else  /* ZSTD_MULTITHREAD not defined */
 /* No multithreading support */
 
-#define pthread_mutex_t int   /* #define rather than typedef, because sometimes pthread support is implicit, resulting in duplicated symbols */
-#define pthread_mutex_init(a,b)    ((void)a, 0)
-#define pthread_mutex_destroy(a)
-#define pthread_mutex_lock(a)
-#define pthread_mutex_unlock(a)
+typedef int ZSTD_pthread_mutex_t;
+#define ZSTD_pthread_mutex_init(a, b)   ((void)(a), (void)(b), 0)
+#define ZSTD_pthread_mutex_destroy(a)   ((void)(a))
+#define ZSTD_pthread_mutex_lock(a)      ((void)(a))
+#define ZSTD_pthread_mutex_unlock(a)    ((void)(a))
 
-#define pthread_cond_t int
-#define pthread_cond_init(a,b)     ((void)a, 0)
-#define pthread_cond_destroy(a)
-#define pthread_cond_wait(a,b)
-#define pthread_cond_signal(a)
-#define pthread_cond_broadcast(a)
+typedef int ZSTD_pthread_cond_t;
+#define ZSTD_pthread_cond_init(a, b)    ((void)(a), (void)(b), 0)
+#define ZSTD_pthread_cond_destroy(a)    ((void)(a))
+#define ZSTD_pthread_cond_wait(a, b)    ((void)(a), (void)(b))
+#define ZSTD_pthread_cond_signal(a)     ((void)(a))
+#define ZSTD_pthread_cond_broadcast(a)  ((void)(a))
 
-/* do not use pthread_t */
+/* do not use ZSTD_pthread_t */
 
 #endif /* ZSTD_MULTITHREAD */
 
diff --git a/contrib/zstd/zdict.c b/contrib/zstd/zdict.c
deleted file mode 100644
index 1bb8b0683..000000000
--- a/contrib/zstd/zdict.c
+++ /dev/null
@@ -1,1075 +0,0 @@
-/*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
- * All rights reserved.
- *
- * This source code is licensed under both the BSD-style license (found in the
- * LICENSE file in the root directory of this source tree) and the GPLv2 (found
- * in the COPYING file in the root directory of this source tree).
- * You may select, at your option, one of the above-listed licenses.
- */
-
-
-/*-**************************************
-*  Tuning parameters
-****************************************/
-#define MINRATIO 4   /* minimum nb of apparition to be selected in dictionary */
-#define ZDICT_MAX_SAMPLES_SIZE (2000U << 20)
-#define ZDICT_MIN_SAMPLES_SIZE (ZDICT_CONTENTSIZE_MIN * MINRATIO)
-
-
-/*-**************************************
-*  Compiler Options
-****************************************/
-/* Unix Large Files support (>4GB) */
-#define _FILE_OFFSET_BITS 64
-#if (defined(__sun__) && (!defined(__LP64__)))   /* Sun Solaris 32-bits requires specific definitions */
-#  define _LARGEFILE_SOURCE
-#elif ! defined(__LP64__)                        /* No point defining Large file for 64 bit */
-#  define _LARGEFILE64_SOURCE
-#endif
-
-
-/*-*************************************
-*  Dependencies
-***************************************/
-#include <stdlib.h>        /* malloc, free */
-#include <string.h>        /* memset */
-#include <stdio.h>         /* fprintf, fopen, ftello64 */
-#include <time.h>          /* clock */
-
-#include "mem.h"           /* read */
-#include "fse.h"           /* FSE_normalizeCount, FSE_writeNCount */
-#define HUF_STATIC_LINKING_ONLY
-#include "huf.h"           /* HUF_buildCTable, HUF_writeCTable */
-#include "zstd_internal.h" /* includes zstd.h */
-#include "xxhash.h"        /* XXH64 */
-#include "divsufsort.h"
-#ifndef ZDICT_STATIC_LINKING_ONLY
-#  define ZDICT_STATIC_LINKING_ONLY
-#endif
-#include "zdict.h"
-
-
-/*-*************************************
-*  Constants
-***************************************/
-#define KB *(1 <<10)
-#define MB *(1 <<20)
-#define GB *(1U<<30)
-
-#define DICTLISTSIZE_DEFAULT 10000
-
-#define NOISELENGTH 32
-
-static const int g_compressionLevel_default = 3;
-static const U32 g_selectivity_default = 9;
-
-
-/*-*************************************
-*  Console display
-***************************************/
-#define DISPLAY(...)         { fprintf(stderr, __VA_ARGS__); fflush( stderr ); }
-#define DISPLAYLEVEL(l, ...) if (notificationLevel>=l) { DISPLAY(__VA_ARGS__); }    /* 0 : no display;   1: errors;   2: default;  3: details;  4: debug */
-
-static clock_t ZDICT_clockSpan(clock_t nPrevious) { return clock() - nPrevious; }
-
-static void ZDICT_printHex(const void* ptr, size_t length)
-{
-    const BYTE* const b = (const BYTE*)ptr;
-    size_t u;
-    for (u=0; u<length; u++) {
-        BYTE c = b[u];
-        if (c<32 || c>126) c = '.';   /* non-printable char */
-        DISPLAY("%c", c);
-    }
-}
-
-
-/*-********************************************************
-*  Helper functions
-**********************************************************/
-unsigned ZDICT_isError(size_t errorCode) { return ERR_isError(errorCode); }
-
-const char* ZDICT_getErrorName(size_t errorCode) { return ERR_getErrorName(errorCode); }
-
-unsigned ZDICT_getDictID(const void* dictBuffer, size_t dictSize)
-{
-    if (dictSize < 8) return 0;
-    if (MEM_readLE32(dictBuffer) != ZSTD_MAGIC_DICTIONARY) return 0;
-    return MEM_readLE32((const char*)dictBuffer + 4);
-}
-
-
-/*-********************************************************
-*  Dictionary training functions
-**********************************************************/
-static unsigned ZDICT_NbCommonBytes (register size_t val)
-{
-    if (MEM_isLittleEndian()) {
-        if (MEM_64bits()) {
-#       if defined(_MSC_VER) && defined(_WIN64)
-            unsigned long r = 0;
-            _BitScanForward64( &r, (U64)val );
-            return (unsigned)(r>>3);
-#       elif defined(__GNUC__) && (__GNUC__ >= 3)
-            return (__builtin_ctzll((U64)val) >> 3);
-#       else
-            static const int DeBruijnBytePos[64] = { 0, 0, 0, 0, 0, 1, 1, 2, 0, 3, 1, 3, 1, 4, 2, 7, 0, 2, 3, 6, 1, 5, 3, 5, 1, 3, 4, 4, 2, 5, 6, 7, 7, 0, 1, 2, 3, 3, 4, 6, 2, 6, 5, 5, 3, 4, 5, 6, 7, 1, 2, 4, 6, 4, 4, 5, 7, 2, 6, 5, 7, 6, 7, 7 };
-            return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
-#       endif
-        } else { /* 32 bits */
-#       if defined(_MSC_VER)
-            unsigned long r=0;
-            _BitScanForward( &r, (U32)val );
-            return (unsigned)(r>>3);
-#       elif defined(__GNUC__) && (__GNUC__ >= 3)
-            return (__builtin_ctz((U32)val) >> 3);
-#       else
-            static const int DeBruijnBytePos[32] = { 0, 0, 3, 0, 3, 1, 3, 0, 3, 2, 2, 1, 3, 2, 0, 1, 3, 3, 1, 2, 2, 2, 2, 0, 3, 1, 2, 0, 1, 0, 1, 1 };
-            return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
-#       endif
-        }
-    } else {  /* Big Endian CPU */
-        if (MEM_64bits()) {
-#       if defined(_MSC_VER) && defined(_WIN64)
-            unsigned long r = 0;
-            _BitScanReverse64( &r, val );
-            return (unsigned)(r>>3);
-#       elif defined(__GNUC__) && (__GNUC__ >= 3)
-            return (__builtin_clzll(val) >> 3);
-#       else
-            unsigned r;
-            const unsigned n32 = sizeof(size_t)*4;   /* calculate this way due to compiler complaining in 32-bits mode */
-            if (!(val>>n32)) { r=4; } else { r=0; val>>=n32; }
-            if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
-            r += (!val);
-            return r;
-#       endif
-        } else { /* 32 bits */
-#       if defined(_MSC_VER)
-            unsigned long r = 0;
-            _BitScanReverse( &r, (unsigned long)val );
-            return (unsigned)(r>>3);
-#       elif defined(__GNUC__) && (__GNUC__ >= 3)
-            return (__builtin_clz((U32)val) >> 3);
-#       else
-            unsigned r;
-            if (!(val>>16)) { r=2; val>>=8; } else { r=0; val>>=24; }
-            r += (!val);
-            return r;
-#       endif
-    }   }
-}
-
-
-/*! ZDICT_count() :
-    Count the nb of common bytes between 2 pointers.
-    Note : this function presumes end of buffer followed by noisy guard band.
-*/
-static size_t ZDICT_count(const void* pIn, const void* pMatch)
-{
-    const char* const pStart = (const char*)pIn;
-    for (;;) {
-        size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
-        if (!diff) {
-            pIn = (const char*)pIn+sizeof(size_t);
-            pMatch = (const char*)pMatch+sizeof(size_t);
-            continue;
-        }
-        pIn = (const char*)pIn+ZDICT_NbCommonBytes(diff);
-        return (size_t)((const char*)pIn - pStart);
-    }
-}
-
-
-typedef struct {
-    U32 pos;
-    U32 length;
-    U32 savings;
-} dictItem;
-
-static void ZDICT_initDictItem(dictItem* d)
-{
-    d->pos = 1;
-    d->length = 0;
-    d->savings = (U32)(-1);
-}
-
-
-#define LLIMIT 64          /* heuristic determined experimentally */
-#define MINMATCHLENGTH 7   /* heuristic determined experimentally */
-static dictItem ZDICT_analyzePos(
-                       BYTE* doneMarks,
-                       const int* suffix, U32 start,
-                       const void* buffer, U32 minRatio, U32 notificationLevel)
-{
-    U32 lengthList[LLIMIT] = {0};
-    U32 cumulLength[LLIMIT] = {0};
-    U32 savings[LLIMIT] = {0};
-    const BYTE* b = (const BYTE*)buffer;
-    size_t length;
-    size_t maxLength = LLIMIT;
-    size_t pos = suffix[start];
-    U32 end = start;
-    dictItem solution;
-
-    /* init */
-    memset(&solution, 0, sizeof(solution));
-    doneMarks[pos] = 1;
-
-    /* trivial repetition cases */
-    if ( (MEM_read16(b+pos+0) == MEM_read16(b+pos+2))
-       ||(MEM_read16(b+pos+1) == MEM_read16(b+pos+3))
-       ||(MEM_read16(b+pos+2) == MEM_read16(b+pos+4)) ) {
-        /* skip and mark segment */
-        U16 u16 = MEM_read16(b+pos+4);
-        U32 u, e = 6;
-        while (MEM_read16(b+pos+e) == u16) e+=2 ;
-        if (b[pos+e] == b[pos+e-1]) e++;
-        for (u=1; u<e; u++)
-            doneMarks[pos+u] = 1;
-        return solution;
-    }
-
-    /* look forward */
-    do {
-        end++;
-        length = ZDICT_count(b + pos, b + suffix[end]);
-    } while (length >=MINMATCHLENGTH);
-
-    /* look backward */
-    do {
-        length = ZDICT_count(b + pos, b + *(suffix+start-1));
-        if (length >=MINMATCHLENGTH) start--;
-    } while(length >= MINMATCHLENGTH);
-
-    /* exit if not found a minimum nb of repetitions */
-    if (end-start < minRatio) {
-        U32 idx;
-        for(idx=start; idx<end; idx++)
-            doneMarks[suffix[idx]] = 1;
-        return solution;
-    }
-
-    {   int i;
-        U32 searchLength;
-        U32 refinedStart = start;
-        U32 refinedEnd = end;
-
-        DISPLAYLEVEL(4, "\n");
-        DISPLAYLEVEL(4, "found %3u matches of length >= %i at pos %7u  ", (U32)(end-start), MINMATCHLENGTH, (U32)pos);
-        DISPLAYLEVEL(4, "\n");
-
-        for (searchLength = MINMATCHLENGTH ; ; searchLength++) {
-            BYTE currentChar = 0;
-            U32 currentCount = 0;
-            U32 currentID = refinedStart;
-            U32 id;
-            U32 selectedCount = 0;
-            U32 selectedID = currentID;
-            for (id =refinedStart; id < refinedEnd; id++) {
-                if (b[ suffix[id] + searchLength] != currentChar) {
-                    if (currentCount > selectedCount) {
-                        selectedCount = currentCount;
-                        selectedID = currentID;
-                    }
-                    currentID = id;
-                    currentChar = b[ suffix[id] + searchLength];
-                    currentCount = 0;
-                }
-                currentCount ++;
-            }
-            if (currentCount > selectedCount) {  /* for last */
-                selectedCount = currentCount;
-                selectedID = currentID;
-            }
-
-            if (selectedCount < minRatio)
-                break;
-            refinedStart = selectedID;
-            refinedEnd = refinedStart + selectedCount;
-        }
-
-        /* evaluate gain based on new ref */
-        start = refinedStart;
-        pos = suffix[refinedStart];
-        end = start;
-        memset(lengthList, 0, sizeof(lengthList));
-
-        /* look forward */
-        do {
-            end++;
-            length = ZDICT_count(b + pos, b + suffix[end]);
-            if (length >= LLIMIT) length = LLIMIT-1;
-            lengthList[length]++;
-        } while (length >=MINMATCHLENGTH);
-
-        /* look backward */
-        length = MINMATCHLENGTH;
-        while ((length >= MINMATCHLENGTH) & (start > 0)) {
-            length = ZDICT_count(b + pos, b + suffix[start - 1]);
-            if (length >= LLIMIT) length = LLIMIT - 1;
-            lengthList[length]++;
-            if (length >= MINMATCHLENGTH) start--;
-        }
-
-        /* largest useful length */
-        memset(cumulLength, 0, sizeof(cumulLength));
-        cumulLength[maxLength-1] = lengthList[maxLength-1];
-        for (i=(int)(maxLength-2); i>=0; i--)
-            cumulLength[i] = cumulLength[i+1] + lengthList[i];
-
-        for (i=LLIMIT-1; i>=MINMATCHLENGTH; i--) if (cumulLength[i]>=minRatio) break;
-        maxLength = i;
-
-        /* reduce maxLength in case of final into repetitive data */
-        {   U32 l = (U32)maxLength;
-            BYTE const c = b[pos + maxLength-1];
-            while (b[pos+l-2]==c) l--;
-            maxLength = l;
-        }
-        if (maxLength < MINMATCHLENGTH) return solution;   /* skip : no long-enough solution */
-
-        /* calculate savings */
-        savings[5] = 0;
-        for (i=MINMATCHLENGTH; i<=(int)maxLength; i++)
-            savings[i] = savings[i-1] + (lengthList[i] * (i-3));
-
-        DISPLAYLEVEL(4, "Selected ref at position %u, of length %u : saves %u (ratio: %.2f)  \n",
-                     (U32)pos, (U32)maxLength, savings[maxLength], (double)savings[maxLength] / maxLength);
-
-        solution.pos = (U32)pos;
-        solution.length = (U32)maxLength;
-        solution.savings = savings[maxLength];
-
-        /* mark positions done */
-        {   U32 id;
-            for (id=start; id<end; id++) {
-                U32 p, pEnd;
-                U32 const testedPos = suffix[id];
-                if (testedPos == pos)
-                    length = solution.length;
-                else {
-                    length = ZDICT_count(b+pos, b+testedPos);
-                    if (length > solution.length) length = solution.length;
-                }
-                pEnd = (U32)(testedPos + length);
-                for (p=testedPos; p<pEnd; p++)
-                    doneMarks[p] = 1;
-    }   }   }
-
-    return solution;
-}
-
-
-static int isIncluded(const void* in, const void* container, size_t length)
-{
-    const char* const ip = (const char*) in;
-    const char* const into = (const char*) container;
-    size_t u;
-
-    for (u=0; u<length; u++) {  /* works because end of buffer is a noisy guard band */
-        if (ip[u] != into[u]) break;
-    }
-
-    return u==length;
-}
-
-/*! ZDICT_tryMerge() :
-    check if dictItem can be merged, do it if possible
-    @return : id of destination elt, 0 if not merged
-*/
-static U32 ZDICT_tryMerge(dictItem* table, dictItem elt, U32 eltNbToSkip, const void* buffer)
-{
-    const U32 tableSize = table->pos;
-    const U32 eltEnd = elt.pos + elt.length;
-    const char* const buf = (const char*) buffer;
-
-    /* tail overlap */
-    U32 u; for (u=1; u<tableSize; u++) {
-        if (u==eltNbToSkip) continue;
-        if ((table[u].pos > elt.pos) && (table[u].pos <= eltEnd)) {  /* overlap, existing > new */
-            /* append */
-            U32 const addedLength = table[u].pos - elt.pos;
-            table[u].length += addedLength;
-            table[u].pos = elt.pos;
-            table[u].savings += elt.savings * addedLength / elt.length;   /* rough approx */
-            table[u].savings += elt.length / 8;    /* rough approx bonus */
-            elt = table[u];
-            /* sort : improve rank */
-            while ((u>1) && (table[u-1].savings < elt.savings))
-            table[u] = table[u-1], u--;
-            table[u] = elt;
-            return u;
-    }   }
-
-    /* front overlap */
-    for (u=1; u<tableSize; u++) {
-        if (u==eltNbToSkip) continue;
-
-        if ((table[u].pos + table[u].length >= elt.pos) && (table[u].pos < elt.pos)) {  /* overlap, existing < new */
-            /* append */
-            int const addedLength = (int)eltEnd - (table[u].pos + table[u].length);
-            table[u].savings += elt.length / 8;    /* rough approx bonus */
-            if (addedLength > 0) {   /* otherwise, elt fully included into existing */
-                table[u].length += addedLength;
-                table[u].savings += elt.savings * addedLength / elt.length;   /* rough approx */
-            }
-            /* sort : improve rank */
-            elt = table[u];
-            while ((u>1) && (table[u-1].savings < elt.savings))
-                table[u] = table[u-1], u--;
-            table[u] = elt;
-            return u;
-        }
-
-        if (MEM_read64(buf + table[u].pos) == MEM_read64(buf + elt.pos + 1)) {
-            if (isIncluded(buf + table[u].pos, buf + elt.pos + 1, table[u].length)) {
-                size_t const addedLength = MAX( (int)elt.length - (int)table[u].length , 1 );
-                table[u].pos = elt.pos;
-                table[u].savings += (U32)(elt.savings * addedLength / elt.length);
-                table[u].length = MIN(elt.length, table[u].length + 1);
-                return u;
-            }
-        }
-    }
-
-    return 0;
-}
-
-
-static void ZDICT_removeDictItem(dictItem* table, U32 id)
-{
-    /* convention : table[0].pos stores nb of elts */
-    U32 const max = table[0].pos;
-    U32 u;
-    if (!id) return;   /* protection, should never happen */
-    for (u=id; u<max-1; u++)
-        table[u] = table[u+1];
-    table->pos--;
-}
-
-
-static void ZDICT_insertDictItem(dictItem* table, U32 maxSize, dictItem elt, const void* buffer)
-{
-    /* merge if possible */
-    U32 mergeId = ZDICT_tryMerge(table, elt, 0, buffer);
-    if (mergeId) {
-        U32 newMerge = 1;
-        while (newMerge) {
-            newMerge = ZDICT_tryMerge(table, table[mergeId], mergeId, buffer);
-            if (newMerge) ZDICT_removeDictItem(table, mergeId);
-            mergeId = newMerge;
-        }
-        return;
-    }
-
-    /* insert */
-    {   U32 current;
-        U32 nextElt = table->pos;
-        if (nextElt >= maxSize) nextElt = maxSize-1;
-        current = nextElt-1;
-        while (table[current].savings < elt.savings) {
-            table[current+1] = table[current];
-            current--;
-        }
-        table[current+1] = elt;
-        table->pos = nextElt+1;
-    }
-}
-
-
-static U32 ZDICT_dictSize(const dictItem* dictList)
-{
-    U32 u, dictSize = 0;
-    for (u=1; u<dictList[0].pos; u++)
-        dictSize += dictList[u].length;
-    return dictSize;
-}
-
-
-static size_t ZDICT_trainBuffer_legacy(dictItem* dictList, U32 dictListSize,
-                            const void* const buffer, size_t bufferSize,   /* buffer must end with noisy guard band */
-                            const size_t* fileSizes, unsigned nbFiles,
-                            U32 minRatio, U32 notificationLevel)
-{
-    int* const suffix0 = (int*)malloc((bufferSize+2)*sizeof(*suffix0));
-    int* const suffix = suffix0+1;
-    U32* reverseSuffix = (U32*)malloc((bufferSize)*sizeof(*reverseSuffix));
-    BYTE* doneMarks = (BYTE*)malloc((bufferSize+16)*sizeof(*doneMarks));   /* +16 for overflow security */
-    U32* filePos = (U32*)malloc(nbFiles * sizeof(*filePos));
-    size_t result = 0;
-    clock_t displayClock = 0;
-    clock_t const refreshRate = CLOCKS_PER_SEC * 3 / 10;
-
-#   define DISPLAYUPDATE(l, ...) if (notificationLevel>=l) { \
-            if (ZDICT_clockSpan(displayClock) > refreshRate)  \
-            { displayClock = clock(); DISPLAY(__VA_ARGS__); \
-            if (notificationLevel>=4) fflush(stderr); } }
-
-    /* init */
-    DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
-    if (!suffix0 || !reverseSuffix || !doneMarks || !filePos) {
-        result = ERROR(memory_allocation);
-        goto _cleanup;
-    }
-    if (minRatio < MINRATIO) minRatio = MINRATIO;
-    memset(doneMarks, 0, bufferSize+16);
-
-    /* limit sample set size (divsufsort limitation)*/
-    if (bufferSize > ZDICT_MAX_SAMPLES_SIZE) DISPLAYLEVEL(3, "sample set too large : reduced to %u MB ...\n", (U32)(ZDICT_MAX_SAMPLES_SIZE>>20));
-    while (bufferSize > ZDICT_MAX_SAMPLES_SIZE) bufferSize -= fileSizes[--nbFiles];
-
-    /* sort */
-    DISPLAYLEVEL(2, "sorting %u files of total size %u MB ...\n", nbFiles, (U32)(bufferSize>>20));
-    {   int const divSuftSortResult = divsufsort((const unsigned char*)buffer, suffix, (int)bufferSize, 0);
-        if (divSuftSortResult != 0) { result = ERROR(GENERIC); goto _cleanup; }
-    }
-    suffix[bufferSize] = (int)bufferSize;   /* leads into noise */
-    suffix0[0] = (int)bufferSize;           /* leads into noise */
-    /* build reverse suffix sort */
-    {   size_t pos;
-        for (pos=0; pos < bufferSize; pos++)
-            reverseSuffix[suffix[pos]] = (U32)pos;
-        /* note filePos tracks borders between samples.
-           It's not used at this stage, but planned to become useful in a later update */
-        filePos[0] = 0;
-        for (pos=1; pos<nbFiles; pos++)
-            filePos[pos] = (U32)(filePos[pos-1] + fileSizes[pos-1]);
-    }
-
-    DISPLAYLEVEL(2, "finding patterns ... \n");
-    DISPLAYLEVEL(3, "minimum ratio : %u \n", minRatio);
-
-    {   U32 cursor; for (cursor=0; cursor < bufferSize; ) {
-            dictItem solution;
-            if (doneMarks[cursor]) { cursor++; continue; }
-            solution = ZDICT_analyzePos(doneMarks, suffix, reverseSuffix[cursor], buffer, minRatio, notificationLevel);
-            if (solution.length==0) { cursor++; continue; }
-            ZDICT_insertDictItem(dictList, dictListSize, solution, buffer);
-            cursor += solution.length;
-            DISPLAYUPDATE(2, "\r%4.2f %% \r", (double)cursor / bufferSize * 100);
-    }   }
-
-_cleanup:
-    free(suffix0);
-    free(reverseSuffix);
-    free(doneMarks);
-    free(filePos);
-    return result;
-}
-
-
-static void ZDICT_fillNoise(void* buffer, size_t length)
-{
-    unsigned const prime1 = 2654435761U;
-    unsigned const prime2 = 2246822519U;
-    unsigned acc = prime1;
-    size_t p=0;;
-    for (p=0; p<length; p++) {
-        acc *= prime2;
-        ((unsigned char*)buffer)[p] = (unsigned char)(acc >> 21);
-    }
-}
-
-
-typedef struct
-{
-    ZSTD_CCtx* ref;
-    ZSTD_CCtx* zc;
-    void* workPlace;   /* must be ZSTD_BLOCKSIZE_MAX allocated */
-} EStats_ress_t;
-
-#define MAXREPOFFSET 1024
-
-static void ZDICT_countEStats(EStats_ress_t esr, ZSTD_parameters params,
-                            U32* countLit, U32* offsetcodeCount, U32* matchlengthCount, U32* litlengthCount, U32* repOffsets,
-                            const void* src, size_t srcSize, U32 notificationLevel)
-{
-    size_t const blockSizeMax = MIN (ZSTD_BLOCKSIZE_MAX, 1 << params.cParams.windowLog);
-    size_t cSize;
-
-    if (srcSize > blockSizeMax) srcSize = blockSizeMax;   /* protection vs large samples */
-    {  size_t const errorCode = ZSTD_copyCCtx(esr.zc, esr.ref, 0);
-            if (ZSTD_isError(errorCode)) { DISPLAYLEVEL(1, "warning : ZSTD_copyCCtx failed \n"); return; }
-    }
-    cSize = ZSTD_compressBlock(esr.zc, esr.workPlace, ZSTD_BLOCKSIZE_MAX, src, srcSize);
-    if (ZSTD_isError(cSize)) { DISPLAYLEVEL(3, "warning : could not compress sample size %u \n", (U32)srcSize); return; }
-
-    if (cSize) {  /* if == 0; block is not compressible */
-        const seqStore_t* seqStorePtr = ZSTD_getSeqStore(esr.zc);
-
-        /* literals stats */
-        {   const BYTE* bytePtr;
-            for(bytePtr = seqStorePtr->litStart; bytePtr < seqStorePtr->lit; bytePtr++)
-                countLit[*bytePtr]++;
-        }
-
-        /* seqStats */
-        {   U32 const nbSeq = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
-            ZSTD_seqToCodes(seqStorePtr);
-
-            {   const BYTE* codePtr = seqStorePtr->ofCode;
-                U32 u;
-                for (u=0; u<nbSeq; u++) offsetcodeCount[codePtr[u]]++;
-            }
-
-            {   const BYTE* codePtr = seqStorePtr->mlCode;
-                U32 u;
-                for (u=0; u<nbSeq; u++) matchlengthCount[codePtr[u]]++;
-            }
-
-            {   const BYTE* codePtr = seqStorePtr->llCode;
-                U32 u;
-                for (u=0; u<nbSeq; u++) litlengthCount[codePtr[u]]++;
-            }
-
-            if (nbSeq >= 2) { /* rep offsets */
-                const seqDef* const seq = seqStorePtr->sequencesStart;
-                U32 offset1 = seq[0].offset - 3;
-                U32 offset2 = seq[1].offset - 3;
-                if (offset1 >= MAXREPOFFSET) offset1 = 0;
-                if (offset2 >= MAXREPOFFSET) offset2 = 0;
-                repOffsets[offset1] += 3;
-                repOffsets[offset2] += 1;
-    }   }   }
-}
-
-static size_t ZDICT_totalSampleSize(const size_t* fileSizes, unsigned nbFiles)
-{
-    size_t total=0;
-    unsigned u;
-    for (u=0; u<nbFiles; u++) total += fileSizes[u];
-    return total;
-}
-
-typedef struct { U32 offset; U32 count; } offsetCount_t;
-
-static void ZDICT_insertSortCount(offsetCount_t table[ZSTD_REP_NUM+1], U32 val, U32 count)
-{
-    U32 u;
-    table[ZSTD_REP_NUM].offset = val;
-    table[ZSTD_REP_NUM].count = count;
-    for (u=ZSTD_REP_NUM; u>0; u--) {
-        offsetCount_t tmp;
-        if (table[u-1].count >= table[u].count) break;
-        tmp = table[u-1];
-        table[u-1] = table[u];
-        table[u] = tmp;
-    }
-}
-
-
-#define OFFCODE_MAX 30  /* only applicable to first block */
-static size_t ZDICT_analyzeEntropy(void*  dstBuffer, size_t maxDstSize,
-                                   unsigned compressionLevel,
-                             const void*  srcBuffer, const size_t* fileSizes, unsigned nbFiles,
-                             const void* dictBuffer, size_t  dictBufferSize,
-                                   unsigned notificationLevel)
-{
-    U32 countLit[256];
-    HUF_CREATE_STATIC_CTABLE(hufTable, 255);
-    U32 offcodeCount[OFFCODE_MAX+1];
-    short offcodeNCount[OFFCODE_MAX+1];
-    U32 offcodeMax = ZSTD_highbit32((U32)(dictBufferSize + 128 KB));
-    U32 matchLengthCount[MaxML+1];
-    short matchLengthNCount[MaxML+1];
-    U32 litLengthCount[MaxLL+1];
-    short litLengthNCount[MaxLL+1];
-    U32 repOffset[MAXREPOFFSET];
-    offsetCount_t bestRepOffset[ZSTD_REP_NUM+1];
-    EStats_ress_t esr;
-    ZSTD_parameters params;
-    U32 u, huffLog = 11, Offlog = OffFSELog, mlLog = MLFSELog, llLog = LLFSELog, total;
-    size_t pos = 0, errorCode;
-    size_t eSize = 0;
-    size_t const totalSrcSize = ZDICT_totalSampleSize(fileSizes, nbFiles);
-    size_t const averageSampleSize = totalSrcSize / (nbFiles + !nbFiles);
-    BYTE* dstPtr = (BYTE*)dstBuffer;
-
-    /* init */
-    esr.ref = ZSTD_createCCtx();
-    esr.zc = ZSTD_createCCtx();
-    esr.workPlace = malloc(ZSTD_BLOCKSIZE_MAX);
-    if (!esr.ref || !esr.zc || !esr.workPlace) {
-        eSize = ERROR(memory_allocation);
-        DISPLAYLEVEL(1, "Not enough memory \n");
-        goto _cleanup;
-    }
-    if (offcodeMax>OFFCODE_MAX) { eSize = ERROR(dictionaryCreation_failed); goto _cleanup; }   /* too large dictionary */
-    for (u=0; u<256; u++) countLit[u] = 1;   /* any character must be described */
-    for (u=0; u<=offcodeMax; u++) offcodeCount[u] = 1;
-    for (u=0; u<=MaxML; u++) matchLengthCount[u] = 1;
-    for (u=0; u<=MaxLL; u++) litLengthCount[u] = 1;
-    memset(repOffset, 0, sizeof(repOffset));
-    repOffset[1] = repOffset[4] = repOffset[8] = 1;
-    memset(bestRepOffset, 0, sizeof(bestRepOffset));
-    if (compressionLevel<=0) compressionLevel = g_compressionLevel_default;
-    params = ZSTD_getParams(compressionLevel, averageSampleSize, dictBufferSize);
-    {   size_t const beginResult = ZSTD_compressBegin_advanced(esr.ref, dictBuffer, dictBufferSize, params, 0);
-        if (ZSTD_isError(beginResult)) {
-            DISPLAYLEVEL(1, "error : ZSTD_compressBegin_advanced() failed : %s \n", ZSTD_getErrorName(beginResult));
-            eSize = ERROR(GENERIC);
-            goto _cleanup;
-    }   }
-
-    /* collect stats on all files */
-    for (u=0; u<nbFiles; u++) {
-        ZDICT_countEStats(esr, params,
-                          countLit, offcodeCount, matchLengthCount, litLengthCount, repOffset,
-                         (const char*)srcBuffer + pos, fileSizes[u],
-                          notificationLevel);
-        pos += fileSizes[u];
-    }
-
-    /* analyze */
-    errorCode = HUF_buildCTable (hufTable, countLit, 255, huffLog);
-    if (HUF_isError(errorCode)) {
-        eSize = ERROR(GENERIC);
-        DISPLAYLEVEL(1, "HUF_buildCTable error \n");
-        goto _cleanup;
-    }
-    huffLog = (U32)errorCode;
-
-    /* looking for most common first offsets */
-    {   U32 offset;
-        for (offset=1; offset<MAXREPOFFSET; offset++)
-            ZDICT_insertSortCount(bestRepOffset, offset, repOffset[offset]);
-    }
-    /* note : the result of this phase should be used to better appreciate the impact on statistics */
-
-    total=0; for (u=0; u<=offcodeMax; u++) total+=offcodeCount[u];
-    errorCode = FSE_normalizeCount(offcodeNCount, Offlog, offcodeCount, total, offcodeMax);
-    if (FSE_isError(errorCode)) {
-        eSize = ERROR(GENERIC);
-        DISPLAYLEVEL(1, "FSE_normalizeCount error with offcodeCount \n");
-        goto _cleanup;
-    }
-    Offlog = (U32)errorCode;
-
-    total=0; for (u=0; u<=MaxML; u++) total+=matchLengthCount[u];
-    errorCode = FSE_normalizeCount(matchLengthNCount, mlLog, matchLengthCount, total, MaxML);
-    if (FSE_isError(errorCode)) {
-        eSize = ERROR(GENERIC);
-        DISPLAYLEVEL(1, "FSE_normalizeCount error with matchLengthCount \n");
-        goto _cleanup;
-    }
-    mlLog = (U32)errorCode;
-
-    total=0; for (u=0; u<=MaxLL; u++) total+=litLengthCount[u];
-    errorCode = FSE_normalizeCount(litLengthNCount, llLog, litLengthCount, total, MaxLL);
-    if (FSE_isError(errorCode)) {
-        eSize = ERROR(GENERIC);
-        DISPLAYLEVEL(1, "FSE_normalizeCount error with litLengthCount \n");
-        goto _cleanup;
-    }
-    llLog = (U32)errorCode;
-
-    /* write result to buffer */
-    {   size_t const hhSize = HUF_writeCTable(dstPtr, maxDstSize, hufTable, 255, huffLog);
-        if (HUF_isError(hhSize)) {
-            eSize = ERROR(GENERIC);
-            DISPLAYLEVEL(1, "HUF_writeCTable error \n");
-            goto _cleanup;
-        }
-        dstPtr += hhSize;
-        maxDstSize -= hhSize;
-        eSize += hhSize;
-    }
-
-    {   size_t const ohSize = FSE_writeNCount(dstPtr, maxDstSize, offcodeNCount, OFFCODE_MAX, Offlog);
-        if (FSE_isError(ohSize)) {
-            eSize = ERROR(GENERIC);
-            DISPLAYLEVEL(1, "FSE_writeNCount error with offcodeNCount \n");
-            goto _cleanup;
-        }
-        dstPtr += ohSize;
-        maxDstSize -= ohSize;
-        eSize += ohSize;
-    }
-
-    {   size_t const mhSize = FSE_writeNCount(dstPtr, maxDstSize, matchLengthNCount, MaxML, mlLog);
-        if (FSE_isError(mhSize)) {
-            eSize = ERROR(GENERIC);
-            DISPLAYLEVEL(1, "FSE_writeNCount error with matchLengthNCount \n");
-            goto _cleanup;
-        }
-        dstPtr += mhSize;
-        maxDstSize -= mhSize;
-        eSize += mhSize;
-    }
-
-    {   size_t const lhSize = FSE_writeNCount(dstPtr, maxDstSize, litLengthNCount, MaxLL, llLog);
-        if (FSE_isError(lhSize)) {
-            eSize = ERROR(GENERIC);
-            DISPLAYLEVEL(1, "FSE_writeNCount error with litlengthNCount \n");
-            goto _cleanup;
-        }
-        dstPtr += lhSize;
-        maxDstSize -= lhSize;
-        eSize += lhSize;
-    }
-
-    if (maxDstSize<12) {
-        eSize = ERROR(GENERIC);
-        DISPLAYLEVEL(1, "not enough space to write RepOffsets \n");
-        goto _cleanup;
-    }
-# if 0
-    MEM_writeLE32(dstPtr+0, bestRepOffset[0].offset);
-    MEM_writeLE32(dstPtr+4, bestRepOffset[1].offset);
-    MEM_writeLE32(dstPtr+8, bestRepOffset[2].offset);
-#else
-    /* at this stage, we don't use the result of "most common first offset",
-       as the impact of statistics is not properly evaluated */
-    MEM_writeLE32(dstPtr+0, repStartValue[0]);
-    MEM_writeLE32(dstPtr+4, repStartValue[1]);
-    MEM_writeLE32(dstPtr+8, repStartValue[2]);
-#endif
-    eSize += 12;
-
-_cleanup:
-    ZSTD_freeCCtx(esr.ref);
-    ZSTD_freeCCtx(esr.zc);
-    free(esr.workPlace);
-
-    return eSize;
-}
-
-
-
-size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
-                          const void* customDictContent, size_t dictContentSize,
-                          const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
-                          ZDICT_params_t params)
-{
-    size_t hSize;
-#define HBUFFSIZE 256   /* should prove large enough for all entropy headers */
-    BYTE header[HBUFFSIZE];
-    int const compressionLevel = (params.compressionLevel <= 0) ? g_compressionLevel_default : params.compressionLevel;
-    U32 const notificationLevel = params.notificationLevel;
-
-    /* check conditions */
-    if (dictBufferCapacity < dictContentSize) return ERROR(dstSize_tooSmall);
-    if (dictContentSize < ZDICT_CONTENTSIZE_MIN) return ERROR(srcSize_wrong);
-    if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) return ERROR(dstSize_tooSmall);
-
-    /* dictionary header */
-    MEM_writeLE32(header, ZSTD_MAGIC_DICTIONARY);
-    {   U64 const randomID = XXH64(customDictContent, dictContentSize, 0);
-        U32 const compliantID = (randomID % ((1U<<31)-32768)) + 32768;
-        U32 const dictID = params.dictID ? params.dictID : compliantID;
-        MEM_writeLE32(header+4, dictID);
-    }
-    hSize = 8;
-
-    /* entropy tables */
-    DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
-    DISPLAYLEVEL(2, "statistics ... \n");
-    {   size_t const eSize = ZDICT_analyzeEntropy(header+hSize, HBUFFSIZE-hSize,
-                                  compressionLevel,
-                                  samplesBuffer, samplesSizes, nbSamples,
-                                  customDictContent, dictContentSize,
-                                  notificationLevel);
-        if (ZDICT_isError(eSize)) return eSize;
-        hSize += eSize;
-    }
-
-    /* copy elements in final buffer ; note : src and dst buffer can overlap */
-    if (hSize + dictContentSize > dictBufferCapacity) dictContentSize = dictBufferCapacity - hSize;
-    {   size_t const dictSize = hSize + dictContentSize;
-        char* dictEnd = (char*)dictBuffer + dictSize;
-        memmove(dictEnd - dictContentSize, customDictContent, dictContentSize);
-        memcpy(dictBuffer, header, hSize);
-        return dictSize;
-    }
-}
-
-
-size_t ZDICT_addEntropyTablesFromBuffer_advanced(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
-                                                 const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
-                                                 ZDICT_params_t params)
-{
-    int const compressionLevel = (params.compressionLevel <= 0) ? g_compressionLevel_default : params.compressionLevel;
-    U32 const notificationLevel = params.notificationLevel;
-    size_t hSize = 8;
-
-    /* calculate entropy tables */
-    DISPLAYLEVEL(2, "\r%70s\r", "");   /* clean display line */
-    DISPLAYLEVEL(2, "statistics ... \n");
-    {   size_t const eSize = ZDICT_analyzeEntropy((char*)dictBuffer+hSize, dictBufferCapacity-hSize,
-                                  compressionLevel,
-                                  samplesBuffer, samplesSizes, nbSamples,
-                                  (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize,
-                                  notificationLevel);
-        if (ZDICT_isError(eSize)) return eSize;
-        hSize += eSize;
-    }
-
-    /* add dictionary header (after entropy tables) */
-    MEM_writeLE32(dictBuffer, ZSTD_MAGIC_DICTIONARY);
-    {   U64 const randomID = XXH64((char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize, 0);
-        U32 const compliantID = (randomID % ((1U<<31)-32768)) + 32768;
-        U32 const dictID = params.dictID ? params.dictID : compliantID;
-        MEM_writeLE32((char*)dictBuffer+4, dictID);
-    }
-
-    if (hSize + dictContentSize < dictBufferCapacity)
-        memmove((char*)dictBuffer + hSize, (char*)dictBuffer + dictBufferCapacity - dictContentSize, dictContentSize);
-    return MIN(dictBufferCapacity, hSize+dictContentSize);
-}
-
-
-/*! ZDICT_trainFromBuffer_unsafe_legacy() :
-*   Warning : `samplesBuffer` must be followed by noisy guard band.
-*   @return : size of dictionary, or an error code which can be tested with ZDICT_isError()
-*/
-size_t ZDICT_trainFromBuffer_unsafe_legacy(
-                            void* dictBuffer, size_t maxDictSize,
-                            const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
-                            ZDICT_legacy_params_t params)
-{
-    U32 const dictListSize = MAX(MAX(DICTLISTSIZE_DEFAULT, nbSamples), (U32)(maxDictSize/16));
-    dictItem* const dictList = (dictItem*)malloc(dictListSize * sizeof(*dictList));
-    unsigned const selectivity = params.selectivityLevel == 0 ? g_selectivity_default : params.selectivityLevel;
-    unsigned const minRep = (selectivity > 30) ? MINRATIO : nbSamples >> selectivity;
-    size_t const targetDictSize = maxDictSize;
-    size_t const samplesBuffSize = ZDICT_totalSampleSize(samplesSizes, nbSamples);
-    size_t dictSize = 0;
-    U32 const notificationLevel = params.zParams.notificationLevel;
-
-    /* checks */
-    if (!dictList) return ERROR(memory_allocation);
-    if (maxDictSize < ZDICT_DICTSIZE_MIN) { free(dictList); return ERROR(dstSize_tooSmall); }   /* requested dictionary size is too small */
-    if (samplesBuffSize < ZDICT_MIN_SAMPLES_SIZE) { free(dictList); return ERROR(dictionaryCreation_failed); }   /* not enough source to create dictionary */
-
-    /* init */
-    ZDICT_initDictItem(dictList);
-
-    /* build dictionary */
-    ZDICT_trainBuffer_legacy(dictList, dictListSize,
-                       samplesBuffer, samplesBuffSize,
-                       samplesSizes, nbSamples,
-                       minRep, notificationLevel);
-
-    /* display best matches */
-    if (params.zParams.notificationLevel>= 3) {
-        U32 const nb = MIN(25, dictList[0].pos);
-        U32 const dictContentSize = ZDICT_dictSize(dictList);
-        U32 u;
-        DISPLAYLEVEL(3, "\n %u segments found, of total size %u \n", dictList[0].pos-1, dictContentSize);
-        DISPLAYLEVEL(3, "list %u best segments \n", nb-1);
-        for (u=1; u<nb; u++) {
-            U32 const pos = dictList[u].pos;
-            U32 const length = dictList[u].length;
-            U32 const printedLength = MIN(40, length);
-            if ((pos > samplesBuffSize) || ((pos + length) > samplesBuffSize))
-                return ERROR(GENERIC);   /* should never happen */
-            DISPLAYLEVEL(3, "%3u:%3u bytes at pos %8u, savings %7u bytes |",
-                         u, length, pos, dictList[u].savings);
-            ZDICT_printHex((const char*)samplesBuffer+pos, printedLength);
-            DISPLAYLEVEL(3, "| \n");
-    }   }
-
-
-    /* create dictionary */
-    {   U32 dictContentSize = ZDICT_dictSize(dictList);
-        if (dictContentSize < ZDICT_CONTENTSIZE_MIN) { free(dictList); return ERROR(dictionaryCreation_failed); }   /* dictionary content too small */
-        if (dictContentSize < targetDictSize/4) {
-            DISPLAYLEVEL(2, "!  warning : selected content significantly smaller than requested (%u < %u) \n", dictContentSize, (U32)maxDictSize);
-            if (samplesBuffSize < 10 * targetDictSize)
-                DISPLAYLEVEL(2, "!  consider increasing the number of samples (total size : %u MB)\n", (U32)(samplesBuffSize>>20));
-            if (minRep > MINRATIO) {
-                DISPLAYLEVEL(2, "!  consider increasing selectivity to produce larger dictionary (-s%u) \n", selectivity+1);
-                DISPLAYLEVEL(2, "!  note : larger dictionaries are not necessarily better, test its efficiency on samples \n");
-            }
-        }
-
-        if ((dictContentSize > targetDictSize*3) && (nbSamples > 2*MINRATIO) && (selectivity>1)) {
-            U32 proposedSelectivity = selectivity-1;
-            while ((nbSamples >> proposedSelectivity) <= MINRATIO) { proposedSelectivity--; }
-            DISPLAYLEVEL(2, "!  note : calculated dictionary significantly larger than requested (%u > %u) \n", dictContentSize, (U32)maxDictSize);
-            DISPLAYLEVEL(2, "!  consider increasing dictionary size, or produce denser dictionary (-s%u) \n", proposedSelectivity);
-            DISPLAYLEVEL(2, "!  always test dictionary efficiency on real samples \n");
-        }
-
-        /* limit dictionary size */
-        {   U32 const max = dictList->pos;   /* convention : nb of useful elts within dictList */
-            U32 currentSize = 0;
-            U32 n; for (n=1; n<max; n++) {
-                currentSize += dictList[n].length;
-                if (currentSize > targetDictSize) { currentSize -= dictList[n].length; break; }
-            }
-            dictList->pos = n;
-            dictContentSize = currentSize;
-        }
-
-        /* build dict content */
-        {   U32 u;
-            BYTE* ptr = (BYTE*)dictBuffer + maxDictSize;
-            for (u=1; u<dictList->pos; u++) {
-                U32 l = dictList[u].length;
-                ptr -= l;
-                if (ptr<(BYTE*)dictBuffer) { free(dictList); return ERROR(GENERIC); }   /* should not happen */
-                memcpy(ptr, (const char*)samplesBuffer+dictList[u].pos, l);
-        }   }
-
-        dictSize = ZDICT_addEntropyTablesFromBuffer_advanced(dictBuffer, dictContentSize, maxDictSize,
-                                                             samplesBuffer, samplesSizes, nbSamples,
-                                                             params.zParams);
-    }
-
-    /* clean up */
-    free(dictList);
-    return dictSize;
-}
-
-
-/* issue : samplesBuffer need to be followed by a noisy guard band.
-*  work around : duplicate the buffer, and add the noise */
-size_t ZDICT_trainFromBuffer_legacy(void* dictBuffer, size_t dictBufferCapacity,
-                              const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
-                              ZDICT_legacy_params_t params)
-{
-    size_t result;
-    void* newBuff;
-    size_t const sBuffSize = ZDICT_totalSampleSize(samplesSizes, nbSamples);
-    if (sBuffSize < ZDICT_MIN_SAMPLES_SIZE) return 0;   /* not enough content => no dictionary */
-
-    newBuff = malloc(sBuffSize + NOISELENGTH);
-    if (!newBuff) return ERROR(memory_allocation);
-
-    memcpy(newBuff, samplesBuffer, sBuffSize);
-    ZDICT_fillNoise((char*)newBuff + sBuffSize, NOISELENGTH);   /* guard band, for end of buffer condition */
-
-    result =
-        ZDICT_trainFromBuffer_unsafe_legacy(dictBuffer, dictBufferCapacity, newBuff,
-                                            samplesSizes, nbSamples, params);
-    free(newBuff);
-    return result;
-}
-
-
-size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
-                             const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
-{
-    ZDICT_cover_params_t params;
-    memset(&params, 0, sizeof(params));
-    params.d = 8;
-    params.steps = 4;
-    /* Default to level 6 since no compression level information is avaialble */
-    params.zParams.compressionLevel = 6;
-    return ZDICT_optimizeTrainFromBuffer_cover(dictBuffer, dictBufferCapacity,
-                                               samplesBuffer, samplesSizes,
-                                               nbSamples, &params);
-}
-
-size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
-                                        const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples)
-{
-    ZDICT_params_t params;
-    memset(&params, 0, sizeof(params));
-    return ZDICT_addEntropyTablesFromBuffer_advanced(dictBuffer, dictContentSize, dictBufferCapacity,
-                                                     samplesBuffer, samplesSizes, nbSamples,
-                                                     params);
-}
diff --git a/contrib/zstd/zdict.h b/contrib/zstd/zdict.h
deleted file mode 100644
index 5f0000b1c..000000000
--- a/contrib/zstd/zdict.h
+++ /dev/null
@@ -1,211 +0,0 @@
-/*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
- * All rights reserved.
- *
- * This source code is licensed under both the BSD-style license (found in the
- * LICENSE file in the root directory of this source tree) and the GPLv2 (found
- * in the COPYING file in the root directory of this source tree).
- * You may select, at your option, one of the above-listed licenses.
- */
-
-#ifndef DICTBUILDER_H_001
-#define DICTBUILDER_H_001
-
-#if defined (__cplusplus)
-extern "C" {
-#endif
-
-
-/*======  Dependencies  ======*/
-#include <stddef.h>  /* size_t */
-
-
-/* =====   ZDICTLIB_API : control library symbols visibility   ===== */
-#ifndef ZDICTLIB_VISIBILITY
-#  if defined(__GNUC__) && (__GNUC__ >= 4)
-#    define ZDICTLIB_VISIBILITY __attribute__ ((visibility ("default")))
-#  else
-#    define ZDICTLIB_VISIBILITY
-#  endif
-#endif
-#if defined(ZSTD_DLL_EXPORT) && (ZSTD_DLL_EXPORT==1)
-#  define ZDICTLIB_API __declspec(dllexport) ZDICTLIB_VISIBILITY
-#elif defined(ZSTD_DLL_IMPORT) && (ZSTD_DLL_IMPORT==1)
-#  define ZDICTLIB_API __declspec(dllimport) ZDICTLIB_VISIBILITY /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/
-#else
-#  define ZDICTLIB_API ZDICTLIB_VISIBILITY
-#endif
-
-
-/*! ZDICT_trainFromBuffer():
- * Train a dictionary from an array of samples.
- * Uses ZDICT_optimizeTrainFromBuffer_cover() single-threaded, with d=8 and steps=4.
- * Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
- * supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
- * The resulting dictionary will be saved into `dictBuffer`.
- * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
- *           or an error code, which can be tested with ZDICT_isError().
- * Note: ZDICT_trainFromBuffer() requires about 9 bytes of memory for each input byte.
- * Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
- *        It's obviously possible to target smaller or larger ones, just by specifying different `dictBufferCapacity`.
- *        In general, it's recommended to provide a few thousands samples, but this can vary a lot.
- *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
- */
-ZDICTLIB_API size_t ZDICT_trainFromBuffer(void* dictBuffer, size_t dictBufferCapacity,
-                       const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples);
-
-
-/*======   Helper functions   ======*/
-ZDICTLIB_API unsigned ZDICT_getDictID(const void* dictBuffer, size_t dictSize);  /**< extracts dictID; @return zero if error (not a valid dictionary) */
-ZDICTLIB_API unsigned ZDICT_isError(size_t errorCode);
-ZDICTLIB_API const char* ZDICT_getErrorName(size_t errorCode);
-
-
-
-#ifdef ZDICT_STATIC_LINKING_ONLY
-
-/* ====================================================================================
- * The definitions in this section are considered experimental.
- * They should never be used with a dynamic library, as they may change in the future.
- * They are provided for advanced usages.
- * Use them only in association with static linking.
- * ==================================================================================== */
-
-typedef struct {
-    int      compressionLevel;   /* 0 means default; target a specific zstd compression level */
-    unsigned notificationLevel;  /* Write to stderr; 0 = none (default); 1 = errors; 2 = progression; 3 = details; 4 = debug; */
-    unsigned dictID;             /* 0 means auto mode (32-bits random value); other : force dictID value */
-} ZDICT_params_t;
-
-/*! ZDICT_cover_params_t:
- *  For all values 0 means default.
- *  k and d are the only required parameters.
- */
-typedef struct {
-    unsigned k;                  /* Segment size : constraint: 0 < k : Reasonable range [16, 2048+] */
-    unsigned d;                  /* dmer size : constraint: 0 < d <= k : Reasonable range [6, 16] */
-    unsigned steps;              /* Number of steps : Only used for optimization : 0 means default (32) : Higher means more parameters checked */
-    unsigned nbThreads;          /* Number of threads : constraint: 0 < nbThreads : 1 means single-threaded : Only used for optimization : Ignored if ZSTD_MULTITHREAD is not defined */
-    ZDICT_params_t zParams;
-} ZDICT_cover_params_t;
-
-
-/*! ZDICT_trainFromBuffer_cover():
- * Train a dictionary from an array of samples using the COVER algorithm.
- * Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
- * supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
- * The resulting dictionary will be saved into `dictBuffer`.
- * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
- *           or an error code, which can be tested with ZDICT_isError().
- * Note: ZDICT_trainFromBuffer_cover() requires about 9 bytes of memory for each input byte.
- * Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
- *        It's obviously possible to target smaller or larger ones, just by specifying different `dictBufferCapacity`.
- *        In general, it's recommended to provide a few thousands samples, but this can vary a lot.
- *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
- */
-ZDICTLIB_API size_t ZDICT_trainFromBuffer_cover(
-    void *dictBuffer, size_t dictBufferCapacity, const void *samplesBuffer,
-    const size_t *samplesSizes, unsigned nbSamples,
-    ZDICT_cover_params_t parameters);
-
-/*! ZDICT_optimizeTrainFromBuffer_cover():
- * The same requirements as above hold for all the parameters except `parameters`.
- * This function tries many parameter combinations and picks the best parameters.
- * `*parameters` is filled with the best parameters found, and the dictionary
- * constructed with those parameters is stored in `dictBuffer`.
- *
- * All of the parameters d, k, steps are optional.
- * If d is non-zero then we don't check multiple values of d, otherwise we check d = {6, 8, 10, 12, 14, 16}.
- * if steps is zero it defaults to its default value.
- * If k is non-zero then we don't check multiple values of k, otherwise we check steps values in [16, 2048].
- *
- * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
- *           or an error code, which can be tested with ZDICT_isError().
- *           On success `*parameters` contains the parameters selected.
- * Note: ZDICT_optimizeTrainFromBuffer_cover() requires about 8 bytes of memory for each input byte and additionally another 5 bytes of memory for each byte of memory for each thread.
- */
-ZDICTLIB_API size_t ZDICT_optimizeTrainFromBuffer_cover(
-    void *dictBuffer, size_t dictBufferCapacity, const void *samplesBuffer,
-    const size_t *samplesSizes, unsigned nbSamples,
-    ZDICT_cover_params_t *parameters);
-
-/*! ZDICT_finalizeDictionary():
- * Given a custom content as a basis for dictionary, and a set of samples,
- * finalize dictionary by adding headers and statistics.
- *
- * Samples must be stored concatenated in a flat buffer `samplesBuffer`,
- * supplied with an array of sizes `samplesSizes`, providing the size of each sample in order.
- *
- * dictContentSize must be >= ZDICT_CONTENTSIZE_MIN bytes.
- * maxDictSize must be >= dictContentSize, and must be >= ZDICT_DICTSIZE_MIN bytes.
- *
- * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`),
- *           or an error code, which can be tested by ZDICT_isError().
- * Note: ZDICT_finalizeDictionary() will push notifications into stderr if instructed to, using notificationLevel>0.
- * Note 2: dictBuffer and dictContent can overlap
- */
-#define ZDICT_CONTENTSIZE_MIN 128
-#define ZDICT_DICTSIZE_MIN    256
-ZDICTLIB_API size_t ZDICT_finalizeDictionary(void* dictBuffer, size_t dictBufferCapacity,
-                                const void* dictContent, size_t dictContentSize,
-                                const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
-                                ZDICT_params_t parameters);
-
-typedef struct {
-    unsigned selectivityLevel;   /* 0 means default; larger => select more => larger dictionary */
-    ZDICT_params_t zParams;
-} ZDICT_legacy_params_t;
-
-/*! ZDICT_trainFromBuffer_legacy():
- * Train a dictionary from an array of samples.
- * Samples must be stored concatenated in a single flat buffer `samplesBuffer`,
- * supplied with an array of sizes `samplesSizes`, providing the size of each sample, in order.
- * The resulting dictionary will be saved into `dictBuffer`.
- * `parameters` is optional and can be provided with values set to 0 to mean "default".
- * @return: size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
- *           or an error code, which can be tested with ZDICT_isError().
- * Tips: In general, a reasonable dictionary has a size of ~ 100 KB.
- *        It's obviously possible to target smaller or larger ones, just by specifying different `dictBufferCapacity`.
- *        In general, it's recommended to provide a few thousands samples, but this can vary a lot.
- *        It's recommended that total size of all samples be about ~x100 times the target size of dictionary.
- * Note: ZDICT_trainFromBuffer_legacy() will send notifications into stderr if instructed to, using notificationLevel>0.
- */
-ZDICTLIB_API size_t ZDICT_trainFromBuffer_legacy(
-    void *dictBuffer, size_t dictBufferCapacity, const void *samplesBuffer,
-    const size_t *samplesSizes, unsigned nbSamples, ZDICT_legacy_params_t parameters);
-
-/* Deprecation warnings */
-/* It is generally possible to disable deprecation warnings from compiler,
-   for example with -Wno-deprecated-declarations for gcc
-   or _CRT_SECURE_NO_WARNINGS in Visual.
-   Otherwise, it's also possible to manually define ZDICT_DISABLE_DEPRECATE_WARNINGS */
-#ifdef ZDICT_DISABLE_DEPRECATE_WARNINGS
-#  define ZDICT_DEPRECATED(message) ZDICTLIB_API   /* disable deprecation warnings */
-#else
-#  define ZDICT_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
-#  if defined (__cplusplus) && (__cplusplus >= 201402) /* C++14 or greater */
-#    define ZDICT_DEPRECATED(message) [[deprecated(message)]] ZDICTLIB_API
-#  elif (ZDICT_GCC_VERSION >= 405) || defined(__clang__)
-#    define ZDICT_DEPRECATED(message) ZDICTLIB_API __attribute__((deprecated(message)))
-#  elif (ZDICT_GCC_VERSION >= 301)
-#    define ZDICT_DEPRECATED(message) ZDICTLIB_API __attribute__((deprecated))
-#  elif defined(_MSC_VER)
-#    define ZDICT_DEPRECATED(message) ZDICTLIB_API __declspec(deprecated(message))
-#  else
-#    pragma message("WARNING: You need to implement ZDICT_DEPRECATED for this compiler")
-#    define ZDICT_DEPRECATED(message) ZDICTLIB_API
-#  endif
-#endif /* ZDICT_DISABLE_DEPRECATE_WARNINGS */
-
-ZDICT_DEPRECATED("use ZDICT_finalizeDictionary() instead")
-size_t ZDICT_addEntropyTablesFromBuffer(void* dictBuffer, size_t dictContentSize, size_t dictBufferCapacity,
-                                  const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples);
-
-
-#endif   /* ZDICT_STATIC_LINKING_ONLY */
-
-#if defined (__cplusplus)
-}
-#endif
-
-#endif   /* DICTBUILDER_H_001 */
diff --git a/contrib/zstd/zstd_common.c b/contrib/zstd/zstd_common.c
index c2041053b..91fe3323a 100644
--- a/contrib/zstd/zstd_common.c
+++ b/contrib/zstd/zstd_common.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -30,23 +30,26 @@ const char* ZSTD_versionString(void) { return ZSTD_VERSION_STRING; }
 /*-****************************************
 *  ZSTD Error Management
 ******************************************/
+#undef ZSTD_isError   /* defined within zstd_internal.h */
 /*! ZSTD_isError() :
-*   tells if a return value is an error code */
+ *  tells if a return value is an error code
+ *  symbol is required for external callers */
 unsigned ZSTD_isError(size_t code) { return ERR_isError(code); }
 
 /*! ZSTD_getErrorName() :
-*   provides error code string from function result (useful for debugging) */
+ *  provides error code string from function result (useful for debugging) */
 const char* ZSTD_getErrorName(size_t code) { return ERR_getErrorName(code); }
 
 /*! ZSTD_getError() :
-*   convert a `size_t` function result into a proper ZSTD_errorCode enum */
+ *  convert a `size_t` function result into a proper ZSTD_errorCode enum */
 ZSTD_ErrorCode ZSTD_getErrorCode(size_t code) { return ERR_getErrorCode(code); }
 
 /*! ZSTD_getErrorString() :
-*   provides error code string from enum */
+ *  provides error code string from enum */
 const char* ZSTD_getErrorString(ZSTD_ErrorCode code) { return ERR_getErrorString(code); }
 
 
+
 /*=**************************************************************
 *  Custom allocator
 ****************************************************************/
diff --git a/contrib/zstd/zstd_compress.c b/contrib/zstd/zstd_compress.c
index b0e9195dd..526cb21f6 100644
--- a/contrib/zstd/zstd_compress.c
+++ b/contrib/zstd/zstd_compress.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -8,50 +8,41 @@
  * You may select, at your option, one of the above-listed licenses.
  */
 
-
-/*-*************************************
-*  Tuning parameters
-***************************************/
-#ifndef ZSTD_CLEVEL_DEFAULT
-#  define ZSTD_CLEVEL_DEFAULT 3
-#endif
-
-
 /*-*************************************
 *  Dependencies
 ***************************************/
+#include <limits.h>         /* INT_MAX */
 #include <string.h>         /* memset */
+#include "cpu.h"
 #include "mem.h"
+#include "hist.h"           /* HIST_countFast_wksp */
 #define FSE_STATIC_LINKING_ONLY   /* FSE_encodeSymbol */
 #include "fse.h"
 #define HUF_STATIC_LINKING_ONLY
 #include "huf.h"
-#include "zstd_compress.h"
+#include "zstd_compress_internal.h"
+#include "zstd_compress_sequences.h"
+#include "zstd_compress_literals.h"
 #include "zstd_fast.h"
 #include "zstd_double_fast.h"
 #include "zstd_lazy.h"
 #include "zstd_opt.h"
 #include "zstd_ldm.h"
+#include "zstd_compress_superblock.h"
 
 
 /*-*************************************
 *  Helper functions
 ***************************************/
+/* ZSTD_compressBound()
+ * Note that the result from this function is only compatible with the "normal"
+ * full-block strategy.
+ * When there are a lot of small blocks due to frequent flush in streaming mode
+ * the overhead of headers can make the compressed data to be larger than the
+ * return value of ZSTD_compressBound().
+ */
 size_t ZSTD_compressBound(size_t srcSize) {
-    size_t const lowLimit = 256 KB;
-    size_t const margin = (srcSize < lowLimit) ? (lowLimit-srcSize) >> 12 : 0;  /* from 64 to 0 */
-    return srcSize + (srcSize >> 8) + margin;
-}
-
-
-/*-*************************************
-*  Sequence storage
-***************************************/
-static void ZSTD_resetSeqStore(seqStore_t* ssPtr)
-{
-    ssPtr->lit = ssPtr->litStart;
-    ssPtr->sequences = ssPtr->sequencesStart;
-    ssPtr->longLengthID = 0;
+    return ZSTD_COMPRESSBOUND(srcSize);
 }
 
 
@@ -59,10 +50,15 @@ static void ZSTD_resetSeqStore(seqStore_t* ssPtr)
 *  Context memory management
 ***************************************/
 struct ZSTD_CDict_s {
-    void* dictBuffer;
     const void* dictContent;
     size_t dictContentSize;
-    ZSTD_CCtx* refContext;
+    U32* entropyWorkspace; /* entropy workspace of HUF_WORKSPACE_SIZE bytes */
+    ZSTD_cwksp workspace;
+    ZSTD_matchState_t matchState;
+    ZSTD_compressedBlockState_t cBlockState;
+    ZSTD_customMem customMem;
+    U32 dictID;
+    int compressionLevel; /* 0 indicates that advanced API was used to select CDict params */
 };  /* typedef'd to ZSTD_CDict within "zstd.h" */
 
 ZSTD_CCtx* ZSTD_createCCtx(void)
@@ -70,53 +66,97 @@ ZSTD_CCtx* ZSTD_createCCtx(void)
     return ZSTD_createCCtx_advanced(ZSTD_defaultCMem);
 }
 
-ZSTD_CCtx* ZSTD_createCCtx_advanced(ZSTD_customMem customMem)
+static void ZSTD_initCCtx(ZSTD_CCtx* cctx, ZSTD_customMem memManager)
 {
-    ZSTD_CCtx* cctx;
-
-    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+    assert(cctx != NULL);
+    memset(cctx, 0, sizeof(*cctx));
+    cctx->customMem = memManager;
+    cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
+    {   size_t const err = ZSTD_CCtx_reset(cctx, ZSTD_reset_parameters);
+        assert(!ZSTD_isError(err));
+        (void)err;
+    }
+}
 
-    cctx = (ZSTD_CCtx*) ZSTD_calloc(sizeof(ZSTD_CCtx), customMem);
-    if (!cctx) return NULL;
-    cctx->customMem = customMem;
-    cctx->requestedParams.compressionLevel = ZSTD_CLEVEL_DEFAULT;
+ZSTD_CCtx* ZSTD_createCCtx_advanced(ZSTD_customMem customMem)
+{
     ZSTD_STATIC_ASSERT(zcss_init==0);
     ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN==(0ULL - 1));
-    return cctx;
+    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+    {   ZSTD_CCtx* const cctx = (ZSTD_CCtx*)ZSTD_malloc(sizeof(ZSTD_CCtx), customMem);
+        if (!cctx) return NULL;
+        ZSTD_initCCtx(cctx, customMem);
+        return cctx;
+    }
 }
 
-ZSTD_CCtx* ZSTD_initStaticCCtx(void *workspace, size_t workspaceSize)
+ZSTD_CCtx* ZSTD_initStaticCCtx(void* workspace, size_t workspaceSize)
 {
-    ZSTD_CCtx* const cctx = (ZSTD_CCtx*) workspace;
+    ZSTD_cwksp ws;
+    ZSTD_CCtx* cctx;
     if (workspaceSize <= sizeof(ZSTD_CCtx)) return NULL;  /* minimum size */
     if ((size_t)workspace & 7) return NULL;  /* must be 8-aligned */
-    memset(workspace, 0, workspaceSize);   /* may be a bit generous, could memset be smaller ? */
-    cctx->staticSize = workspaceSize;
-    cctx->workSpace = (void*)(cctx+1);
-    cctx->workSpaceSize = workspaceSize - sizeof(ZSTD_CCtx);
+    ZSTD_cwksp_init(&ws, workspace, workspaceSize);
 
-    /* entropy space (never moves) */
-    if (cctx->workSpaceSize < sizeof(ZSTD_entropyCTables_t)) return NULL;
-    assert(((size_t)cctx->workSpace & (sizeof(void*)-1)) == 0);   /* ensure correct alignment */
-    cctx->entropy = (ZSTD_entropyCTables_t*)cctx->workSpace;
+    cctx = (ZSTD_CCtx*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CCtx));
+    if (cctx == NULL) return NULL;
 
+    memset(cctx, 0, sizeof(ZSTD_CCtx));
+    ZSTD_cwksp_move(&cctx->workspace, &ws);
+    cctx->staticSize = workspaceSize;
+
+    /* statically sized space. entropyWorkspace never moves (but prev/next block swap places) */
+    if (!ZSTD_cwksp_check_available(&cctx->workspace, HUF_WORKSPACE_SIZE + 2 * sizeof(ZSTD_compressedBlockState_t))) return NULL;
+    cctx->blockState.prevCBlock = (ZSTD_compressedBlockState_t*)ZSTD_cwksp_reserve_object(&cctx->workspace, sizeof(ZSTD_compressedBlockState_t));
+    cctx->blockState.nextCBlock = (ZSTD_compressedBlockState_t*)ZSTD_cwksp_reserve_object(&cctx->workspace, sizeof(ZSTD_compressedBlockState_t));
+    cctx->entropyWorkspace = (U32*)ZSTD_cwksp_reserve_object(&cctx->workspace, HUF_WORKSPACE_SIZE);
+    cctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
     return cctx;
 }
 
-size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx)
+/**
+ * Clears and frees all of the dictionaries in the CCtx.
+ */
+static void ZSTD_clearAllDicts(ZSTD_CCtx* cctx)
 {
-    if (cctx==NULL) return 0;   /* support free on NULL */
-    if (cctx->staticSize) return ERROR(memory_allocation);   /* not compatible with static CCtx */
-    ZSTD_free(cctx->workSpace, cctx->customMem);
-    cctx->workSpace = NULL;
-    ZSTD_freeCDict(cctx->cdictLocal);
-    cctx->cdictLocal = NULL;
+    ZSTD_free(cctx->localDict.dictBuffer, cctx->customMem);
+    ZSTD_freeCDict(cctx->localDict.cdict);
+    memset(&cctx->localDict, 0, sizeof(cctx->localDict));
+    memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));
+    cctx->cdict = NULL;
+}
+
+static size_t ZSTD_sizeof_localDict(ZSTD_localDict dict)
+{
+    size_t const bufferSize = dict.dictBuffer != NULL ? dict.dictSize : 0;
+    size_t const cdictSize = ZSTD_sizeof_CDict(dict.cdict);
+    return bufferSize + cdictSize;
+}
+
+static void ZSTD_freeCCtxContent(ZSTD_CCtx* cctx)
+{
+    assert(cctx != NULL);
+    assert(cctx->staticSize == 0);
+    ZSTD_clearAllDicts(cctx);
 #ifdef ZSTD_MULTITHREAD
-    ZSTDMT_freeCCtx(cctx->mtctx);
-    cctx->mtctx = NULL;
+    ZSTDMT_freeCCtx(cctx->mtctx); cctx->mtctx = NULL;
 #endif
-    ZSTD_free(cctx, cctx->customMem);
-    return 0;   /* reserved as a potential error code in the future */
+    ZSTD_cwksp_free(&cctx->workspace, cctx->customMem);
+}
+
+size_t ZSTD_freeCCtx(ZSTD_CCtx* cctx)
+{
+    if (cctx==NULL) return 0;   /* support free on NULL */
+    RETURN_ERROR_IF(cctx->staticSize, memory_allocation,
+                    "not compatible with static CCtx");
+    {
+        int cctxInWorkspace = ZSTD_cwksp_owns_buffer(&cctx->workspace, cctx);
+        ZSTD_freeCCtxContent(cctx);
+        if (!cctxInWorkspace) {
+            ZSTD_free(cctx, cctx->customMem);
+        }
+    }
+    return 0;
 }
 
 
@@ -125,7 +165,7 @@ static size_t ZSTD_sizeof_mtctx(const ZSTD_CCtx* cctx)
 #ifdef ZSTD_MULTITHREAD
     return ZSTDMT_sizeof_CCtx(cctx->mtctx);
 #else
-    (void) cctx;
+    (void)cctx;
     return 0;
 #endif
 }
@@ -134,12 +174,10 @@ static size_t ZSTD_sizeof_mtctx(const ZSTD_CCtx* cctx)
 size_t ZSTD_sizeof_CCtx(const ZSTD_CCtx* cctx)
 {
     if (cctx==NULL) return 0;   /* support sizeof on NULL */
-    DEBUGLOG(3, "sizeof(*cctx) : %u", (U32)sizeof(*cctx));
-    DEBUGLOG(3, "workSpaceSize (including streaming buffers): %u", (U32)cctx->workSpaceSize);
-    DEBUGLOG(3, "inner cdict : %u", (U32)ZSTD_sizeof_CDict(cctx->cdictLocal));
-    DEBUGLOG(3, "inner MTCTX : %u", (U32)ZSTD_sizeof_mtctx(cctx));
-    return sizeof(*cctx) + cctx->workSpaceSize
-           + ZSTD_sizeof_CDict(cctx->cdictLocal)
+    /* cctx may be in the workspace */
+    return (cctx->workspace.workspace == cctx ? 0 : sizeof(*cctx))
+           + ZSTD_cwksp_sizeof(&cctx->workspace)
+           + ZSTD_sizeof_localDict(cctx->localDict)
            + ZSTD_sizeof_mtctx(cctx);
 }
 
@@ -151,40 +189,15 @@ size_t ZSTD_sizeof_CStream(const ZSTD_CStream* zcs)
 /* private API call, for dictBuilder only */
 const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx) { return &(ctx->seqStore); }
 
-#define ZSTD_CLEVEL_CUSTOM 999
-
-static ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
-        ZSTD_CCtx_params params, U64 srcSizeHint, size_t dictSize)
-{
-    return (params.compressionLevel == ZSTD_CLEVEL_CUSTOM ?
-                    params.cParams :
-                    ZSTD_getCParams(params.compressionLevel, srcSizeHint, dictSize));
-}
-
-static void ZSTD_cLevelToCCtxParams_srcSize(ZSTD_CCtx_params* params, U64 srcSize)
-{
-    params->cParams = ZSTD_getCParamsFromCCtxParams(*params, srcSize, 0);
-    params->compressionLevel = ZSTD_CLEVEL_CUSTOM;
-}
-
-static void ZSTD_cLevelToCParams(ZSTD_CCtx* cctx)
-{
-    ZSTD_cLevelToCCtxParams_srcSize(
-            &cctx->requestedParams, cctx->pledgedSrcSizePlusOne-1);
-}
-
-static void ZSTD_cLevelToCCtxParams(ZSTD_CCtx_params* params)
-{
-    ZSTD_cLevelToCCtxParams_srcSize(params, 0);
-}
-
 static ZSTD_CCtx_params ZSTD_makeCCtxParamsFromCParams(
         ZSTD_compressionParameters cParams)
 {
     ZSTD_CCtx_params cctxParams;
     memset(&cctxParams, 0, sizeof(cctxParams));
     cctxParams.cParams = cParams;
-    cctxParams.compressionLevel = ZSTD_CLEVEL_CUSTOM;
+    cctxParams.compressionLevel = ZSTD_CLEVEL_DEFAULT;  /* should not matter, as all cParams are presumed properly defined */
+    assert(!ZSTD_checkCParams(cParams));
+    cctxParams.fParams.contentSizeFlag = 1;
     return cctxParams;
 }
 
@@ -198,6 +211,7 @@ static ZSTD_CCtx_params* ZSTD_createCCtxParams_advanced(
     if (!params) { return NULL; }
     params->customMem = customMem;
     params->compressionLevel = ZSTD_CLEVEL_DEFAULT;
+    params->fParams.contentSizeFlag = 1;
     return params;
 }
 
@@ -213,324 +227,690 @@ size_t ZSTD_freeCCtxParams(ZSTD_CCtx_params* params)
     return 0;
 }
 
-size_t ZSTD_resetCCtxParams(ZSTD_CCtx_params* params)
+size_t ZSTD_CCtxParams_reset(ZSTD_CCtx_params* params)
 {
-    return ZSTD_initCCtxParams(params, ZSTD_CLEVEL_DEFAULT);
+    return ZSTD_CCtxParams_init(params, ZSTD_CLEVEL_DEFAULT);
 }
 
-size_t ZSTD_initCCtxParams(ZSTD_CCtx_params* cctxParams, int compressionLevel) {
-    if (!cctxParams) { return ERROR(GENERIC); }
+size_t ZSTD_CCtxParams_init(ZSTD_CCtx_params* cctxParams, int compressionLevel) {
+    RETURN_ERROR_IF(!cctxParams, GENERIC, "NULL pointer!");
     memset(cctxParams, 0, sizeof(*cctxParams));
     cctxParams->compressionLevel = compressionLevel;
+    cctxParams->fParams.contentSizeFlag = 1;
     return 0;
 }
 
-size_t ZSTD_initCCtxParams_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params)
+size_t ZSTD_CCtxParams_init_advanced(ZSTD_CCtx_params* cctxParams, ZSTD_parameters params)
 {
-    if (!cctxParams) { return ERROR(GENERIC); }
-    CHECK_F( ZSTD_checkCParams(params.cParams) );
+    RETURN_ERROR_IF(!cctxParams, GENERIC, "NULL pointer!");
+    FORWARD_IF_ERROR( ZSTD_checkCParams(params.cParams) , "");
     memset(cctxParams, 0, sizeof(*cctxParams));
+    assert(!ZSTD_checkCParams(params.cParams));
     cctxParams->cParams = params.cParams;
     cctxParams->fParams = params.fParams;
-    cctxParams->compressionLevel = ZSTD_CLEVEL_CUSTOM;
+    cctxParams->compressionLevel = ZSTD_CLEVEL_DEFAULT;   /* should not matter, as all cParams are presumed properly defined */
     return 0;
 }
 
+/* ZSTD_assignParamsToCCtxParams() :
+ * params is presumed valid at this stage */
 static ZSTD_CCtx_params ZSTD_assignParamsToCCtxParams(
-        ZSTD_CCtx_params cctxParams, ZSTD_parameters params)
+        const ZSTD_CCtx_params* cctxParams, const ZSTD_parameters* params)
 {
-    ZSTD_CCtx_params ret = cctxParams;
-    ret.cParams = params.cParams;
-    ret.fParams = params.fParams;
+    ZSTD_CCtx_params ret = *cctxParams;
+    assert(!ZSTD_checkCParams(params->cParams));
+    ret.cParams = params->cParams;
+    ret.fParams = params->fParams;
+    ret.compressionLevel = ZSTD_CLEVEL_DEFAULT;   /* should not matter, as all cParams are presumed properly defined */
     return ret;
 }
 
-#define CLAMPCHECK(val,min,max) {            \
-    if (((val)<(min)) | ((val)>(max))) {     \
-        return ERROR(parameter_outOfBound);  \
-}   }
-
-size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, unsigned value)
+ZSTD_bounds ZSTD_cParam_getBounds(ZSTD_cParameter param)
 {
-    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
+    ZSTD_bounds bounds = { 0, 0, 0 };
 
     switch(param)
     {
-    case ZSTD_p_compressionLevel:
-        if (value == 0) return 0;  /* special value : 0 means "don't change anything" */
-        if (cctx->cdict) return ERROR(stage_wrong);
-        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
-
-    case ZSTD_p_windowLog:
-    case ZSTD_p_hashLog:
-    case ZSTD_p_chainLog:
-    case ZSTD_p_searchLog:
-    case ZSTD_p_minMatch:
-    case ZSTD_p_targetLength:
-    case ZSTD_p_compressionStrategy:
-        if (value == 0) return 0;  /* special value : 0 means "don't change anything" */
-        if (cctx->cdict) return ERROR(stage_wrong);
-        ZSTD_cLevelToCParams(cctx);  /* Can optimize if srcSize is known */
-        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
-
-    case ZSTD_p_contentSizeFlag:
-    case ZSTD_p_checksumFlag:
-    case ZSTD_p_dictIDFlag:
-        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
-
-    case ZSTD_p_forceMaxWindow :  /* Force back-references to remain < windowSize,
-                                   * even when referencing into Dictionary content
-                                   * default : 0 when using a CDict, 1 when using a Prefix */
-        cctx->loadedDictEnd = 0;
-        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
-
-    case ZSTD_p_nbThreads:
-        if (value==0) return 0;
-        DEBUGLOG(5, " setting nbThreads : %u", value);
-        if (value > 1 && cctx->staticSize) {
-            return ERROR(parameter_unsupported);  /* MT not compatible with static alloc */
-        }
-        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
-
-    case ZSTD_p_jobSize:
-        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+    case ZSTD_c_compressionLevel:
+        bounds.lowerBound = ZSTD_minCLevel();
+        bounds.upperBound = ZSTD_maxCLevel();
+        return bounds;
+
+    case ZSTD_c_windowLog:
+        bounds.lowerBound = ZSTD_WINDOWLOG_MIN;
+        bounds.upperBound = ZSTD_WINDOWLOG_MAX;
+        return bounds;
+
+    case ZSTD_c_hashLog:
+        bounds.lowerBound = ZSTD_HASHLOG_MIN;
+        bounds.upperBound = ZSTD_HASHLOG_MAX;
+        return bounds;
+
+    case ZSTD_c_chainLog:
+        bounds.lowerBound = ZSTD_CHAINLOG_MIN;
+        bounds.upperBound = ZSTD_CHAINLOG_MAX;
+        return bounds;
+
+    case ZSTD_c_searchLog:
+        bounds.lowerBound = ZSTD_SEARCHLOG_MIN;
+        bounds.upperBound = ZSTD_SEARCHLOG_MAX;
+        return bounds;
+
+    case ZSTD_c_minMatch:
+        bounds.lowerBound = ZSTD_MINMATCH_MIN;
+        bounds.upperBound = ZSTD_MINMATCH_MAX;
+        return bounds;
+
+    case ZSTD_c_targetLength:
+        bounds.lowerBound = ZSTD_TARGETLENGTH_MIN;
+        bounds.upperBound = ZSTD_TARGETLENGTH_MAX;
+        return bounds;
+
+    case ZSTD_c_strategy:
+        bounds.lowerBound = ZSTD_STRATEGY_MIN;
+        bounds.upperBound = ZSTD_STRATEGY_MAX;
+        return bounds;
+
+    case ZSTD_c_contentSizeFlag:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_checksumFlag:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_dictIDFlag:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_nbWorkers:
+        bounds.lowerBound = 0;
+#ifdef ZSTD_MULTITHREAD
+        bounds.upperBound = ZSTDMT_NBWORKERS_MAX;
+#else
+        bounds.upperBound = 0;
+#endif
+        return bounds;
 
-    case ZSTD_p_overlapSizeLog:
-        DEBUGLOG(5, " setting overlap with nbThreads == %u", cctx->requestedParams.nbThreads);
-        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+    case ZSTD_c_jobSize:
+        bounds.lowerBound = 0;
+#ifdef ZSTD_MULTITHREAD
+        bounds.upperBound = ZSTDMT_JOBSIZE_MAX;
+#else
+        bounds.upperBound = 0;
+#endif
+        return bounds;
 
-    case ZSTD_p_enableLongDistanceMatching:
-        if (cctx->cdict) return ERROR(stage_wrong);
-        if (value != 0) {
-            ZSTD_cLevelToCParams(cctx);
-        }
-        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+    case ZSTD_c_overlapLog:
+#ifdef ZSTD_MULTITHREAD
+        bounds.lowerBound = ZSTD_OVERLAPLOG_MIN;
+        bounds.upperBound = ZSTD_OVERLAPLOG_MAX;
+#else
+        bounds.lowerBound = 0;
+        bounds.upperBound = 0;
+#endif
+        return bounds;
+
+    case ZSTD_c_enableLongDistanceMatching:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_ldmHashLog:
+        bounds.lowerBound = ZSTD_LDM_HASHLOG_MIN;
+        bounds.upperBound = ZSTD_LDM_HASHLOG_MAX;
+        return bounds;
+
+    case ZSTD_c_ldmMinMatch:
+        bounds.lowerBound = ZSTD_LDM_MINMATCH_MIN;
+        bounds.upperBound = ZSTD_LDM_MINMATCH_MAX;
+        return bounds;
+
+    case ZSTD_c_ldmBucketSizeLog:
+        bounds.lowerBound = ZSTD_LDM_BUCKETSIZELOG_MIN;
+        bounds.upperBound = ZSTD_LDM_BUCKETSIZELOG_MAX;
+        return bounds;
+
+    case ZSTD_c_ldmHashRateLog:
+        bounds.lowerBound = ZSTD_LDM_HASHRATELOG_MIN;
+        bounds.upperBound = ZSTD_LDM_HASHRATELOG_MAX;
+        return bounds;
+
+    /* experimental parameters */
+    case ZSTD_c_rsyncable:
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_forceMaxWindow :
+        bounds.lowerBound = 0;
+        bounds.upperBound = 1;
+        return bounds;
+
+    case ZSTD_c_format:
+        ZSTD_STATIC_ASSERT(ZSTD_f_zstd1 < ZSTD_f_zstd1_magicless);
+        bounds.lowerBound = ZSTD_f_zstd1;
+        bounds.upperBound = ZSTD_f_zstd1_magicless;   /* note : how to ensure at compile time that this is the highest value enum ? */
+        return bounds;
+
+    case ZSTD_c_forceAttachDict:
+        ZSTD_STATIC_ASSERT(ZSTD_dictDefaultAttach < ZSTD_dictForceCopy);
+        bounds.lowerBound = ZSTD_dictDefaultAttach;
+        bounds.upperBound = ZSTD_dictForceLoad;       /* note : how to ensure at compile time that this is the highest value enum ? */
+        return bounds;
+
+    case ZSTD_c_literalCompressionMode:
+        ZSTD_STATIC_ASSERT(ZSTD_lcm_auto < ZSTD_lcm_huffman && ZSTD_lcm_huffman < ZSTD_lcm_uncompressed);
+        bounds.lowerBound = ZSTD_lcm_auto;
+        bounds.upperBound = ZSTD_lcm_uncompressed;
+        return bounds;
+
+    case ZSTD_c_targetCBlockSize:
+        bounds.lowerBound = ZSTD_TARGETCBLOCKSIZE_MIN;
+        bounds.upperBound = ZSTD_TARGETCBLOCKSIZE_MAX;
+        return bounds;
+
+    case ZSTD_c_srcSizeHint:
+        bounds.lowerBound = ZSTD_SRCSIZEHINT_MIN;
+        bounds.upperBound = ZSTD_SRCSIZEHINT_MAX;
+        return bounds;
 
-    case ZSTD_p_ldmHashLog:
-    case ZSTD_p_ldmMinMatch:
-        if (value == 0) return 0;  /* special value : 0 means "don't change anything" */
-        if (cctx->cdict) return ERROR(stage_wrong);
-        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+    default:
+        bounds.error = ERROR(parameter_unsupported);
+        return bounds;
+    }
+}
 
-    case ZSTD_p_ldmBucketSizeLog:
-    case ZSTD_p_ldmHashEveryLog:
-        if (cctx->cdict) return ERROR(stage_wrong);
-        return ZSTD_CCtxParam_setParameter(&cctx->requestedParams, param, value);
+/* ZSTD_cParam_clampBounds:
+ * Clamps the value into the bounded range.
+ */
+static size_t ZSTD_cParam_clampBounds(ZSTD_cParameter cParam, int* value)
+{
+    ZSTD_bounds const bounds = ZSTD_cParam_getBounds(cParam);
+    if (ZSTD_isError(bounds.error)) return bounds.error;
+    if (*value < bounds.lowerBound) *value = bounds.lowerBound;
+    if (*value > bounds.upperBound) *value = bounds.upperBound;
+    return 0;
+}
 
-    default: return ERROR(parameter_unsupported);
-    }
+#define BOUNDCHECK(cParam, val) { \
+    RETURN_ERROR_IF(!ZSTD_cParam_withinBounds(cParam,val), \
+                    parameter_outOfBound, "Param out of bounds"); \
 }
 
-size_t ZSTD_CCtxParam_setParameter(
-        ZSTD_CCtx_params* params, ZSTD_cParameter param, unsigned value)
+
+static int ZSTD_isUpdateAuthorized(ZSTD_cParameter param)
 {
     switch(param)
     {
-    case ZSTD_p_compressionLevel :
-        if ((int)value > ZSTD_maxCLevel()) value = ZSTD_maxCLevel();
-        if (value == 0) return 0;
-        params->compressionLevel = value;
-        return 0;
-
-    case ZSTD_p_windowLog :
-        if (value == 0) return 0;
-        CLAMPCHECK(value, ZSTD_WINDOWLOG_MIN, ZSTD_WINDOWLOG_MAX);
-        ZSTD_cLevelToCCtxParams(params);
-        params->cParams.windowLog = value;
-        return 0;
+    case ZSTD_c_compressionLevel:
+    case ZSTD_c_hashLog:
+    case ZSTD_c_chainLog:
+    case ZSTD_c_searchLog:
+    case ZSTD_c_minMatch:
+    case ZSTD_c_targetLength:
+    case ZSTD_c_strategy:
+        return 1;
 
-    case ZSTD_p_hashLog :
-        if (value == 0) return 0;
-        CLAMPCHECK(value, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
-        ZSTD_cLevelToCCtxParams(params);
-        params->cParams.hashLog = value;
+    case ZSTD_c_format:
+    case ZSTD_c_windowLog:
+    case ZSTD_c_contentSizeFlag:
+    case ZSTD_c_checksumFlag:
+    case ZSTD_c_dictIDFlag:
+    case ZSTD_c_forceMaxWindow :
+    case ZSTD_c_nbWorkers:
+    case ZSTD_c_jobSize:
+    case ZSTD_c_overlapLog:
+    case ZSTD_c_rsyncable:
+    case ZSTD_c_enableLongDistanceMatching:
+    case ZSTD_c_ldmHashLog:
+    case ZSTD_c_ldmMinMatch:
+    case ZSTD_c_ldmBucketSizeLog:
+    case ZSTD_c_ldmHashRateLog:
+    case ZSTD_c_forceAttachDict:
+    case ZSTD_c_literalCompressionMode:
+    case ZSTD_c_targetCBlockSize:
+    case ZSTD_c_srcSizeHint:
+    default:
         return 0;
+    }
+}
 
-    case ZSTD_p_chainLog :
-        if (value == 0) return 0;
-        CLAMPCHECK(value, ZSTD_CHAINLOG_MIN, ZSTD_CHAINLOG_MAX);
-        ZSTD_cLevelToCCtxParams(params);
-        params->cParams.chainLog = value;
-        return 0;
+size_t ZSTD_CCtx_setParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int value)
+{
+    DEBUGLOG(4, "ZSTD_CCtx_setParameter (%i, %i)", (int)param, value);
+    if (cctx->streamStage != zcss_init) {
+        if (ZSTD_isUpdateAuthorized(param)) {
+            cctx->cParamsChanged = 1;
+        } else {
+            RETURN_ERROR(stage_wrong, "can only set params in ctx init stage");
+    }   }
 
-    case ZSTD_p_searchLog :
-        if (value == 0) return 0;
-        CLAMPCHECK(value, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLOG_MAX);
-        ZSTD_cLevelToCCtxParams(params);
-        params->cParams.searchLog = value;
-        return 0;
+    switch(param)
+    {
+    case ZSTD_c_nbWorkers:
+        RETURN_ERROR_IF((value!=0) && cctx->staticSize, parameter_unsupported,
+                        "MT not compatible with static alloc");
+        break;
 
-    case ZSTD_p_minMatch :
-        if (value == 0) return 0;
-        CLAMPCHECK(value, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
-        ZSTD_cLevelToCCtxParams(params);
-        params->cParams.searchLength = value;
-        return 0;
+    case ZSTD_c_compressionLevel:
+    case ZSTD_c_windowLog:
+    case ZSTD_c_hashLog:
+    case ZSTD_c_chainLog:
+    case ZSTD_c_searchLog:
+    case ZSTD_c_minMatch:
+    case ZSTD_c_targetLength:
+    case ZSTD_c_strategy:
+    case ZSTD_c_ldmHashRateLog:
+    case ZSTD_c_format:
+    case ZSTD_c_contentSizeFlag:
+    case ZSTD_c_checksumFlag:
+    case ZSTD_c_dictIDFlag:
+    case ZSTD_c_forceMaxWindow:
+    case ZSTD_c_forceAttachDict:
+    case ZSTD_c_literalCompressionMode:
+    case ZSTD_c_jobSize:
+    case ZSTD_c_overlapLog:
+    case ZSTD_c_rsyncable:
+    case ZSTD_c_enableLongDistanceMatching:
+    case ZSTD_c_ldmHashLog:
+    case ZSTD_c_ldmMinMatch:
+    case ZSTD_c_ldmBucketSizeLog:
+    case ZSTD_c_targetCBlockSize:
+    case ZSTD_c_srcSizeHint:
+        break;
 
-    case ZSTD_p_targetLength :
-        if (value == 0) return 0;
-        CLAMPCHECK(value, ZSTD_TARGETLENGTH_MIN, ZSTD_TARGETLENGTH_MAX);
-        ZSTD_cLevelToCCtxParams(params);
-        params->cParams.targetLength = value;
-        return 0;
+    default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
+    }
+    return ZSTD_CCtxParams_setParameter(&cctx->requestedParams, param, value);
+}
 
-    case ZSTD_p_compressionStrategy :
-        if (value == 0) return 0;
-        CLAMPCHECK(value, (unsigned)ZSTD_fast, (unsigned)ZSTD_btultra);
-        ZSTD_cLevelToCCtxParams(params);
-        params->cParams.strategy = (ZSTD_strategy)value;
-        return 0;
+size_t ZSTD_CCtxParams_setParameter(ZSTD_CCtx_params* CCtxParams,
+                                    ZSTD_cParameter param, int value)
+{
+    DEBUGLOG(4, "ZSTD_CCtxParams_setParameter (%i, %i)", (int)param, value);
+    switch(param)
+    {
+    case ZSTD_c_format :
+        BOUNDCHECK(ZSTD_c_format, value);
+        CCtxParams->format = (ZSTD_format_e)value;
+        return (size_t)CCtxParams->format;
+
+    case ZSTD_c_compressionLevel : {
+        FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(param, &value), "");
+        if (value) {  /* 0 : does not change current level */
+            CCtxParams->compressionLevel = value;
+        }
+        if (CCtxParams->compressionLevel >= 0) return (size_t)CCtxParams->compressionLevel;
+        return 0;  /* return type (size_t) cannot represent negative values */
+    }
 
-    case ZSTD_p_contentSizeFlag :
+    case ZSTD_c_windowLog :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_windowLog, value);
+        CCtxParams->cParams.windowLog = (U32)value;
+        return CCtxParams->cParams.windowLog;
+
+    case ZSTD_c_hashLog :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_hashLog, value);
+        CCtxParams->cParams.hashLog = (U32)value;
+        return CCtxParams->cParams.hashLog;
+
+    case ZSTD_c_chainLog :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_chainLog, value);
+        CCtxParams->cParams.chainLog = (U32)value;
+        return CCtxParams->cParams.chainLog;
+
+    case ZSTD_c_searchLog :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_searchLog, value);
+        CCtxParams->cParams.searchLog = (U32)value;
+        return (size_t)value;
+
+    case ZSTD_c_minMatch :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_minMatch, value);
+        CCtxParams->cParams.minMatch = value;
+        return CCtxParams->cParams.minMatch;
+
+    case ZSTD_c_targetLength :
+        BOUNDCHECK(ZSTD_c_targetLength, value);
+        CCtxParams->cParams.targetLength = value;
+        return CCtxParams->cParams.targetLength;
+
+    case ZSTD_c_strategy :
+        if (value!=0)   /* 0 => use default */
+            BOUNDCHECK(ZSTD_c_strategy, value);
+        CCtxParams->cParams.strategy = (ZSTD_strategy)value;
+        return (size_t)CCtxParams->cParams.strategy;
+
+    case ZSTD_c_contentSizeFlag :
         /* Content size written in frame header _when known_ (default:1) */
-        DEBUGLOG(5, "set content size flag = %u", (value>0));
-        params->fParams.contentSizeFlag = value > 0;
-        return 0;
+        DEBUGLOG(4, "set content size flag = %u", (value!=0));
+        CCtxParams->fParams.contentSizeFlag = value != 0;
+        return CCtxParams->fParams.contentSizeFlag;
 
-    case ZSTD_p_checksumFlag :
+    case ZSTD_c_checksumFlag :
         /* A 32-bits content checksum will be calculated and written at end of frame (default:0) */
-        params->fParams.checksumFlag = value > 0;
-        return 0;
-
-    case ZSTD_p_dictIDFlag : /* When applicable, dictionary's dictID is provided in frame header (default:1) */
-        DEBUGLOG(5, "set dictIDFlag = %u", (value>0));
-        params->fParams.noDictIDFlag = (value == 0);
-        return 0;
+        CCtxParams->fParams.checksumFlag = value != 0;
+        return CCtxParams->fParams.checksumFlag;
+
+    case ZSTD_c_dictIDFlag : /* When applicable, dictionary's dictID is provided in frame header (default:1) */
+        DEBUGLOG(4, "set dictIDFlag = %u", (value!=0));
+        CCtxParams->fParams.noDictIDFlag = !value;
+        return !CCtxParams->fParams.noDictIDFlag;
+
+    case ZSTD_c_forceMaxWindow :
+        CCtxParams->forceWindow = (value != 0);
+        return CCtxParams->forceWindow;
+
+    case ZSTD_c_forceAttachDict : {
+        const ZSTD_dictAttachPref_e pref = (ZSTD_dictAttachPref_e)value;
+        BOUNDCHECK(ZSTD_c_forceAttachDict, pref);
+        CCtxParams->attachDictPref = pref;
+        return CCtxParams->attachDictPref;
+    }
 
-    case ZSTD_p_forceMaxWindow :
-        params->forceWindow = value > 0;
-        return 0;
+    case ZSTD_c_literalCompressionMode : {
+        const ZSTD_literalCompressionMode_e lcm = (ZSTD_literalCompressionMode_e)value;
+        BOUNDCHECK(ZSTD_c_literalCompressionMode, lcm);
+        CCtxParams->literalCompressionMode = lcm;
+        return CCtxParams->literalCompressionMode;
+    }
 
-    case ZSTD_p_nbThreads :
-        if (value == 0) return 0;
+    case ZSTD_c_nbWorkers :
 #ifndef ZSTD_MULTITHREAD
-        if (value > 1) return ERROR(parameter_unsupported);
+        RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
         return 0;
 #else
-        return ZSTDMT_initializeCCtxParameters(params, value);
+        FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(param, &value), "");
+        CCtxParams->nbWorkers = value;
+        return CCtxParams->nbWorkers;
 #endif
 
-    case ZSTD_p_jobSize :
+    case ZSTD_c_jobSize :
 #ifndef ZSTD_MULTITHREAD
-        return ERROR(parameter_unsupported);
+        RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
+        return 0;
 #else
-        if (params->nbThreads <= 1) return ERROR(parameter_unsupported);
-        return ZSTDMT_CCtxParam_setMTCtxParameter(params, ZSTDMT_p_sectionSize, value);
+        /* Adjust to the minimum non-default value. */
+        if (value != 0 && value < ZSTDMT_JOBSIZE_MIN)
+            value = ZSTDMT_JOBSIZE_MIN;
+        FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(param, &value), "");
+        assert(value >= 0);
+        CCtxParams->jobSize = value;
+        return CCtxParams->jobSize;
 #endif
 
-    case ZSTD_p_overlapSizeLog :
+    case ZSTD_c_overlapLog :
 #ifndef ZSTD_MULTITHREAD
-        return ERROR(parameter_unsupported);
+        RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
+        return 0;
 #else
-        if (params->nbThreads <= 1) return ERROR(parameter_unsupported);
-        return ZSTDMT_CCtxParam_setMTCtxParameter(params, ZSTDMT_p_overlapSectionLog, value);
+        FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(ZSTD_c_overlapLog, &value), "");
+        CCtxParams->overlapLog = value;
+        return CCtxParams->overlapLog;
 #endif
 
-    case ZSTD_p_enableLongDistanceMatching :
-        if (value != 0) {
-            ZSTD_cLevelToCCtxParams(params);
-            params->cParams.windowLog = ZSTD_LDM_WINDOW_LOG;
-        }
-        return ZSTD_ldm_initializeParameters(&params->ldmParams, value);
-
-    case ZSTD_p_ldmHashLog :
-        if (value == 0) return 0;
-        CLAMPCHECK(value, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
-        params->ldmParams.hashLog = value;
-        return 0;
-
-    case ZSTD_p_ldmMinMatch :
-        if (value == 0) return 0;
-        CLAMPCHECK(value, ZSTD_LDM_MINMATCH_MIN, ZSTD_LDM_MINMATCH_MAX);
-        params->ldmParams.minMatchLength = value;
+    case ZSTD_c_rsyncable :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR_IF(value!=0, parameter_unsupported, "not compiled with multithreading");
         return 0;
+#else
+        FORWARD_IF_ERROR(ZSTD_cParam_clampBounds(ZSTD_c_overlapLog, &value), "");
+        CCtxParams->rsyncable = value;
+        return CCtxParams->rsyncable;
+#endif
 
-    case ZSTD_p_ldmBucketSizeLog :
-        if (value > ZSTD_LDM_BUCKETSIZELOG_MAX) {
-            return ERROR(parameter_outOfBound);
-        }
-        params->ldmParams.bucketSizeLog = value;
-        return 0;
+    case ZSTD_c_enableLongDistanceMatching :
+        CCtxParams->ldmParams.enableLdm = (value!=0);
+        return CCtxParams->ldmParams.enableLdm;
+
+    case ZSTD_c_ldmHashLog :
+        if (value!=0)   /* 0 ==> auto */
+            BOUNDCHECK(ZSTD_c_ldmHashLog, value);
+        CCtxParams->ldmParams.hashLog = value;
+        return CCtxParams->ldmParams.hashLog;
+
+    case ZSTD_c_ldmMinMatch :
+        if (value!=0)   /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_ldmMinMatch, value);
+        CCtxParams->ldmParams.minMatchLength = value;
+        return CCtxParams->ldmParams.minMatchLength;
+
+    case ZSTD_c_ldmBucketSizeLog :
+        if (value!=0)   /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_ldmBucketSizeLog, value);
+        CCtxParams->ldmParams.bucketSizeLog = value;
+        return CCtxParams->ldmParams.bucketSizeLog;
+
+    case ZSTD_c_ldmHashRateLog :
+        RETURN_ERROR_IF(value > ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN,
+                        parameter_outOfBound, "Param out of bounds!");
+        CCtxParams->ldmParams.hashRateLog = value;
+        return CCtxParams->ldmParams.hashRateLog;
+
+    case ZSTD_c_targetCBlockSize :
+        if (value!=0)   /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_targetCBlockSize, value);
+        CCtxParams->targetCBlockSize = value;
+        return CCtxParams->targetCBlockSize;
+
+    case ZSTD_c_srcSizeHint :
+        if (value!=0)    /* 0 ==> default */
+            BOUNDCHECK(ZSTD_c_srcSizeHint, value);
+        CCtxParams->srcSizeHint = value;
+        return CCtxParams->srcSizeHint;
+
+    default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
+    }
+}
 
-    case ZSTD_p_ldmHashEveryLog :
-        if (value > ZSTD_WINDOWLOG_MAX - ZSTD_HASHLOG_MIN) {
-            return ERROR(parameter_outOfBound);
-        }
-        params->ldmParams.hashEveryLog = value;
-        return 0;
+size_t ZSTD_CCtx_getParameter(ZSTD_CCtx* cctx, ZSTD_cParameter param, int* value)
+{
+    return ZSTD_CCtxParams_getParameter(&cctx->requestedParams, param, value);
+}
 
-    default: return ERROR(parameter_unsupported);
+size_t ZSTD_CCtxParams_getParameter(
+        ZSTD_CCtx_params* CCtxParams, ZSTD_cParameter param, int* value)
+{
+    switch(param)
+    {
+    case ZSTD_c_format :
+        *value = CCtxParams->format;
+        break;
+    case ZSTD_c_compressionLevel :
+        *value = CCtxParams->compressionLevel;
+        break;
+    case ZSTD_c_windowLog :
+        *value = (int)CCtxParams->cParams.windowLog;
+        break;
+    case ZSTD_c_hashLog :
+        *value = (int)CCtxParams->cParams.hashLog;
+        break;
+    case ZSTD_c_chainLog :
+        *value = (int)CCtxParams->cParams.chainLog;
+        break;
+    case ZSTD_c_searchLog :
+        *value = CCtxParams->cParams.searchLog;
+        break;
+    case ZSTD_c_minMatch :
+        *value = CCtxParams->cParams.minMatch;
+        break;
+    case ZSTD_c_targetLength :
+        *value = CCtxParams->cParams.targetLength;
+        break;
+    case ZSTD_c_strategy :
+        *value = (unsigned)CCtxParams->cParams.strategy;
+        break;
+    case ZSTD_c_contentSizeFlag :
+        *value = CCtxParams->fParams.contentSizeFlag;
+        break;
+    case ZSTD_c_checksumFlag :
+        *value = CCtxParams->fParams.checksumFlag;
+        break;
+    case ZSTD_c_dictIDFlag :
+        *value = !CCtxParams->fParams.noDictIDFlag;
+        break;
+    case ZSTD_c_forceMaxWindow :
+        *value = CCtxParams->forceWindow;
+        break;
+    case ZSTD_c_forceAttachDict :
+        *value = CCtxParams->attachDictPref;
+        break;
+    case ZSTD_c_literalCompressionMode :
+        *value = CCtxParams->literalCompressionMode;
+        break;
+    case ZSTD_c_nbWorkers :
+#ifndef ZSTD_MULTITHREAD
+        assert(CCtxParams->nbWorkers == 0);
+#endif
+        *value = CCtxParams->nbWorkers;
+        break;
+    case ZSTD_c_jobSize :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR(parameter_unsupported, "not compiled with multithreading");
+#else
+        assert(CCtxParams->jobSize <= INT_MAX);
+        *value = (int)CCtxParams->jobSize;
+        break;
+#endif
+    case ZSTD_c_overlapLog :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR(parameter_unsupported, "not compiled with multithreading");
+#else
+        *value = CCtxParams->overlapLog;
+        break;
+#endif
+    case ZSTD_c_rsyncable :
+#ifndef ZSTD_MULTITHREAD
+        RETURN_ERROR(parameter_unsupported, "not compiled with multithreading");
+#else
+        *value = CCtxParams->rsyncable;
+        break;
+#endif
+    case ZSTD_c_enableLongDistanceMatching :
+        *value = CCtxParams->ldmParams.enableLdm;
+        break;
+    case ZSTD_c_ldmHashLog :
+        *value = CCtxParams->ldmParams.hashLog;
+        break;
+    case ZSTD_c_ldmMinMatch :
+        *value = CCtxParams->ldmParams.minMatchLength;
+        break;
+    case ZSTD_c_ldmBucketSizeLog :
+        *value = CCtxParams->ldmParams.bucketSizeLog;
+        break;
+    case ZSTD_c_ldmHashRateLog :
+        *value = CCtxParams->ldmParams.hashRateLog;
+        break;
+    case ZSTD_c_targetCBlockSize :
+        *value = (int)CCtxParams->targetCBlockSize;
+        break;
+    case ZSTD_c_srcSizeHint :
+        *value = (int)CCtxParams->srcSizeHint;
+        break;
+    default: RETURN_ERROR(parameter_unsupported, "unknown parameter");
     }
+    return 0;
 }
 
-/**
- * This function should be updated whenever ZSTD_CCtx_params is updated.
- * Parameters are copied manually before the dictionary is loaded.
- * The multithreading parameters jobSize and overlapSizeLog are set only if
- * nbThreads > 1.
- *
- * Pledged srcSize is treated as unknown.
+/** ZSTD_CCtx_setParametersUsingCCtxParams() :
+ *  just applies `params` into `cctx`
+ *  no action is performed, parameters are merely stored.
+ *  If ZSTDMT is enabled, parameters are pushed to cctx->mtctx.
+ *    This is possible even if a compression is ongoing.
+ *    In which case, new parameters will be applied on the fly, starting with next compression job.
  */
 size_t ZSTD_CCtx_setParametersUsingCCtxParams(
         ZSTD_CCtx* cctx, const ZSTD_CCtx_params* params)
 {
-    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
-    if (cctx->cdict) return ERROR(stage_wrong);
-
-    /* Assume the compression and frame parameters are validated */
-    cctx->requestedParams.cParams = params->cParams;
-    cctx->requestedParams.fParams = params->fParams;
-    cctx->requestedParams.compressionLevel = params->compressionLevel;
-
-    /* Set force window explicitly since it sets cctx->loadedDictEnd */
-    CHECK_F( ZSTD_CCtx_setParameter(
-                   cctx, ZSTD_p_forceMaxWindow, params->forceWindow) );
-
-    /* Set multithreading parameters explicitly */
-    CHECK_F( ZSTD_CCtx_setParameter(cctx, ZSTD_p_nbThreads, params->nbThreads) );
-    if (params->nbThreads > 1) {
-        CHECK_F( ZSTD_CCtx_setParameter(cctx, ZSTD_p_jobSize, params->jobSize) );
-        CHECK_F( ZSTD_CCtx_setParameter(
-                    cctx, ZSTD_p_overlapSizeLog, params->overlapSizeLog) );
-    }
-
-    /* Copy long distance matching parameters */
-    cctx->requestedParams.ldmParams = params->ldmParams;
-
-    /* customMem is used only for create/free params and can be ignored */
+    DEBUGLOG(4, "ZSTD_CCtx_setParametersUsingCCtxParams");
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "The context is in the wrong stage!");
+    RETURN_ERROR_IF(cctx->cdict, stage_wrong,
+                    "Can't override parameters with cdict attached (some must "
+                    "be inherited from the cdict).");
+
+    cctx->requestedParams = *params;
     return 0;
 }
 
 ZSTDLIB_API size_t ZSTD_CCtx_setPledgedSrcSize(ZSTD_CCtx* cctx, unsigned long long pledgedSrcSize)
 {
-    DEBUGLOG(5, " setting pledgedSrcSize to %u", (U32)pledgedSrcSize);
-    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
+    DEBUGLOG(4, "ZSTD_CCtx_setPledgedSrcSize to %u bytes", (U32)pledgedSrcSize);
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "Can't set pledgedSrcSize when not in init stage.");
     cctx->pledgedSrcSizePlusOne = pledgedSrcSize+1;
     return 0;
 }
 
+/**
+ * Initializes the local dict using the requested parameters.
+ * NOTE: This does not use the pledged src size, because it may be used for more
+ * than one compression.
+ */
+static size_t ZSTD_initLocalDict(ZSTD_CCtx* cctx)
+{
+    ZSTD_localDict* const dl = &cctx->localDict;
+    ZSTD_compressionParameters const cParams = ZSTD_getCParamsFromCCtxParams(
+            &cctx->requestedParams, ZSTD_CONTENTSIZE_UNKNOWN, dl->dictSize);
+    if (dl->dict == NULL) {
+        /* No local dictionary. */
+        assert(dl->dictBuffer == NULL);
+        assert(dl->cdict == NULL);
+        assert(dl->dictSize == 0);
+        return 0;
+    }
+    if (dl->cdict != NULL) {
+        assert(cctx->cdict == dl->cdict);
+        /* Local dictionary already initialized. */
+        return 0;
+    }
+    assert(dl->dictSize > 0);
+    assert(cctx->cdict == NULL);
+    assert(cctx->prefixDict.dict == NULL);
+
+    dl->cdict = ZSTD_createCDict_advanced(
+            dl->dict,
+            dl->dictSize,
+            ZSTD_dlm_byRef,
+            dl->dictContentType,
+            cParams,
+            cctx->customMem);
+    RETURN_ERROR_IF(!dl->cdict, memory_allocation, "ZSTD_createCDict_advanced failed");
+    cctx->cdict = dl->cdict;
+    return 0;
+}
+
 size_t ZSTD_CCtx_loadDictionary_advanced(
         ZSTD_CCtx* cctx, const void* dict, size_t dictSize,
-        ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictMode_e dictMode)
-{
-    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
-    if (cctx->staticSize) return ERROR(memory_allocation);  /* no malloc for static CCtx */
-    DEBUGLOG(5, "load dictionary of size %u", (U32)dictSize);
-    ZSTD_freeCDict(cctx->cdictLocal);  /* in case one already exists */
-    if (dict==NULL || dictSize==0) {   /* no dictionary mode */
-        cctx->cdictLocal = NULL;
-        cctx->cdict = NULL;
+        ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_dictContentType_e dictContentType)
+{
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "Can't load a dictionary when ctx is not in init stage.");
+    RETURN_ERROR_IF(cctx->staticSize, memory_allocation,
+                    "no malloc for static CCtx");
+    DEBUGLOG(4, "ZSTD_CCtx_loadDictionary_advanced (size: %u)", (U32)dictSize);
+    ZSTD_clearAllDicts(cctx);  /* in case one already exists */
+    if (dict == NULL || dictSize == 0)  /* no dictionary mode */
+        return 0;
+    if (dictLoadMethod == ZSTD_dlm_byRef) {
+        cctx->localDict.dict = dict;
     } else {
-        ZSTD_compressionParameters const cParams =
-                ZSTD_getCParamsFromCCtxParams(cctx->requestedParams, 0, dictSize);
-        cctx->cdictLocal = ZSTD_createCDict_advanced(
-                                dict, dictSize,
-                                dictLoadMethod, dictMode,
-                                cParams, cctx->customMem);
-        cctx->cdict = cctx->cdictLocal;
-        if (cctx->cdictLocal == NULL)
-            return ERROR(memory_allocation);
+        void* dictBuffer = ZSTD_malloc(dictSize, cctx->customMem);
+        RETURN_ERROR_IF(!dictBuffer, memory_allocation, "NULL pointer!");
+        memcpy(dictBuffer, dict, dictSize);
+        cctx->localDict.dictBuffer = dictBuffer;
+        cctx->localDict.dict = dictBuffer;
     }
+    cctx->localDict.dictSize = dictSize;
+    cctx->localDict.dictContentType = dictContentType;
     return 0;
 }
 
@@ -538,432 +918,669 @@ ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary_byReference(
       ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
 {
     return ZSTD_CCtx_loadDictionary_advanced(
-            cctx, dict, dictSize, ZSTD_dlm_byRef, ZSTD_dm_auto);
+            cctx, dict, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto);
 }
 
 ZSTDLIB_API size_t ZSTD_CCtx_loadDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
 {
     return ZSTD_CCtx_loadDictionary_advanced(
-            cctx, dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dm_auto);
+            cctx, dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dct_auto);
 }
 
 
 size_t ZSTD_CCtx_refCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
 {
-    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "Can't ref a dict when ctx not in init stage.");
+    /* Free the existing local cdict (if any) to save memory. */
+    ZSTD_clearAllDicts(cctx);
     cctx->cdict = cdict;
-    memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));  /* exclusive */
     return 0;
 }
 
 size_t ZSTD_CCtx_refPrefix(ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize)
 {
-    return ZSTD_CCtx_refPrefix_advanced(cctx, prefix, prefixSize, ZSTD_dm_rawContent);
+    return ZSTD_CCtx_refPrefix_advanced(cctx, prefix, prefixSize, ZSTD_dct_rawContent);
 }
 
 size_t ZSTD_CCtx_refPrefix_advanced(
-        ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictMode_e dictMode)
+        ZSTD_CCtx* cctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType)
 {
-    if (cctx->streamStage != zcss_init) return ERROR(stage_wrong);
-    cctx->cdict = NULL;   /* prefix discards any prior cdict */
-    cctx->prefixDict.dict = prefix;
-    cctx->prefixDict.dictSize = prefixSize;
-    cctx->prefixDict.dictMode = dictMode;
+    RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                    "Can't ref a prefix when ctx not in init stage.");
+    ZSTD_clearAllDicts(cctx);
+    if (prefix != NULL && prefixSize > 0) {
+        cctx->prefixDict.dict = prefix;
+        cctx->prefixDict.dictSize = prefixSize;
+        cctx->prefixDict.dictContentType = dictContentType;
+    }
     return 0;
 }
 
-static void ZSTD_startNewCompression(ZSTD_CCtx* cctx)
-{
-    cctx->streamStage = zcss_init;
-    cctx->pledgedSrcSizePlusOne = 0;
-}
-
 /*! ZSTD_CCtx_reset() :
  *  Also dumps dictionary */
-void ZSTD_CCtx_reset(ZSTD_CCtx* cctx)
+size_t ZSTD_CCtx_reset(ZSTD_CCtx* cctx, ZSTD_ResetDirective reset)
 {
-    ZSTD_startNewCompression(cctx);
-    cctx->cdict = NULL;
+    if ( (reset == ZSTD_reset_session_only)
+      || (reset == ZSTD_reset_session_and_parameters) ) {
+        cctx->streamStage = zcss_init;
+        cctx->pledgedSrcSizePlusOne = 0;
+    }
+    if ( (reset == ZSTD_reset_parameters)
+      || (reset == ZSTD_reset_session_and_parameters) ) {
+        RETURN_ERROR_IF(cctx->streamStage != zcss_init, stage_wrong,
+                        "Can't reset parameters only when not in init stage.");
+        ZSTD_clearAllDicts(cctx);
+        return ZSTD_CCtxParams_reset(&cctx->requestedParams);
+    }
+    return 0;
 }
 
+
 /** ZSTD_checkCParams() :
     control CParam values remain within authorized range.
     @return : 0, or an error code if one value is beyond authorized range */
 size_t ZSTD_checkCParams(ZSTD_compressionParameters cParams)
 {
-    CLAMPCHECK(cParams.windowLog, ZSTD_WINDOWLOG_MIN, ZSTD_WINDOWLOG_MAX);
-    CLAMPCHECK(cParams.chainLog, ZSTD_CHAINLOG_MIN, ZSTD_CHAINLOG_MAX);
-    CLAMPCHECK(cParams.hashLog, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
-    CLAMPCHECK(cParams.searchLog, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLOG_MAX);
-    CLAMPCHECK(cParams.searchLength, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
-    CLAMPCHECK(cParams.targetLength, ZSTD_TARGETLENGTH_MIN, ZSTD_TARGETLENGTH_MAX);
-    if ((U32)(cParams.strategy) > (U32)ZSTD_btultra)
-        return ERROR(parameter_unsupported);
+    BOUNDCHECK(ZSTD_c_windowLog, (int)cParams.windowLog);
+    BOUNDCHECK(ZSTD_c_chainLog,  (int)cParams.chainLog);
+    BOUNDCHECK(ZSTD_c_hashLog,   (int)cParams.hashLog);
+    BOUNDCHECK(ZSTD_c_searchLog, (int)cParams.searchLog);
+    BOUNDCHECK(ZSTD_c_minMatch,  (int)cParams.minMatch);
+    BOUNDCHECK(ZSTD_c_targetLength,(int)cParams.targetLength);
+    BOUNDCHECK(ZSTD_c_strategy,  cParams.strategy);
     return 0;
 }
 
 /** ZSTD_clampCParams() :
  *  make CParam values within valid range.
  *  @return : valid CParams */
-static ZSTD_compressionParameters ZSTD_clampCParams(ZSTD_compressionParameters cParams)
-{
-#   define CLAMP(val,min,max) {      \
-        if (val<min) val=min;        \
-        else if (val>max) val=max;   \
-    }
-    CLAMP(cParams.windowLog, ZSTD_WINDOWLOG_MIN, ZSTD_WINDOWLOG_MAX);
-    CLAMP(cParams.chainLog, ZSTD_CHAINLOG_MIN, ZSTD_CHAINLOG_MAX);
-    CLAMP(cParams.hashLog, ZSTD_HASHLOG_MIN, ZSTD_HASHLOG_MAX);
-    CLAMP(cParams.searchLog, ZSTD_SEARCHLOG_MIN, ZSTD_SEARCHLOG_MAX);
-    CLAMP(cParams.searchLength, ZSTD_SEARCHLENGTH_MIN, ZSTD_SEARCHLENGTH_MAX);
-    CLAMP(cParams.targetLength, ZSTD_TARGETLENGTH_MIN, ZSTD_TARGETLENGTH_MAX);
-    if ((U32)(cParams.strategy) > (U32)ZSTD_btultra) cParams.strategy = ZSTD_btultra;
+static ZSTD_compressionParameters
+ZSTD_clampCParams(ZSTD_compressionParameters cParams)
+{
+#   define CLAMP_TYPE(cParam, val, type) {                                \
+        ZSTD_bounds const bounds = ZSTD_cParam_getBounds(cParam);         \
+        if ((int)val<bounds.lowerBound) val=(type)bounds.lowerBound;      \
+        else if ((int)val>bounds.upperBound) val=(type)bounds.upperBound; \
+    }
+#   define CLAMP(cParam, val) CLAMP_TYPE(cParam, val, unsigned)
+    CLAMP(ZSTD_c_windowLog, cParams.windowLog);
+    CLAMP(ZSTD_c_chainLog,  cParams.chainLog);
+    CLAMP(ZSTD_c_hashLog,   cParams.hashLog);
+    CLAMP(ZSTD_c_searchLog, cParams.searchLog);
+    CLAMP(ZSTD_c_minMatch,  cParams.minMatch);
+    CLAMP(ZSTD_c_targetLength,cParams.targetLength);
+    CLAMP_TYPE(ZSTD_c_strategy,cParams.strategy, ZSTD_strategy);
     return cParams;
 }
 
 /** ZSTD_cycleLog() :
  *  condition for correct operation : hashLog > 1 */
-static U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat)
+U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat)
 {
     U32 const btScale = ((U32)strat >= (U32)ZSTD_btlazy2);
     return hashLog - btScale;
 }
 
 /** ZSTD_adjustCParams_internal() :
-    optimize `cPar` for a given input (`srcSize` and `dictSize`).
-    mostly downsizing to reduce memory consumption and initialization.
-    Both `srcSize` and `dictSize` are optional (use 0 if unknown),
-    but if both are 0, no optimization can be done.
-    Note : cPar is considered validated at this stage. Use ZSTD_checkParams() to ensure that. */
-ZSTD_compressionParameters ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize)
+ *  optimize `cPar` for a specified input (`srcSize` and `dictSize`).
+ *  mostly downsize to reduce memory consumption and initialization latency.
+ * `srcSize` can be ZSTD_CONTENTSIZE_UNKNOWN when not known.
+ *  note : `srcSize==0` means 0!
+ *  condition : cPar is presumed validated (can be checked using ZSTD_checkCParams()). */
+static ZSTD_compressionParameters
+ZSTD_adjustCParams_internal(ZSTD_compressionParameters cPar,
+                            unsigned long long srcSize,
+                            size_t dictSize)
 {
+    static const U64 minSrcSize = 513; /* (1<<9) + 1 */
+    static const U64 maxWindowResize = 1ULL << (ZSTD_WINDOWLOG_MAX-1);
     assert(ZSTD_checkCParams(cPar)==0);
-    if (srcSize+dictSize == 0) return cPar;   /* no size information available : no adjustment */
-
-    /* resize params, to use less memory when necessary */
-    {   U32 const minSrcSize = (srcSize==0) ? 500 : 0;
-        U64 const rSize = srcSize + dictSize + minSrcSize;
-        if (rSize < ((U64)1<<ZSTD_WINDOWLOG_MAX)) {
-            U32 const srcLog =
-                    MAX(ZSTD_HASHLOG_MIN, (rSize==1) ? 1 : ZSTD_highbit32((U32)(rSize)-1) + 1);
-            if (cPar.windowLog > srcLog) cPar.windowLog = srcLog;
-    }   }
-    if (cPar.hashLog > cPar.windowLog) cPar.hashLog = cPar.windowLog;
+
+    if (dictSize && srcSize == ZSTD_CONTENTSIZE_UNKNOWN)
+        srcSize = minSrcSize;
+
+    /* resize windowLog if input is small enough, to use less memory */
+    if ( (srcSize < maxWindowResize)
+      && (dictSize < maxWindowResize) )  {
+        U32 const tSize = (U32)(srcSize + dictSize);
+        static U32 const hashSizeMin = 1 << ZSTD_HASHLOG_MIN;
+        U32 const srcLog = (tSize < hashSizeMin) ? ZSTD_HASHLOG_MIN :
+                            ZSTD_highbit32(tSize-1) + 1;
+        if (cPar.windowLog > srcLog) cPar.windowLog = srcLog;
+    }
+    if (cPar.hashLog > cPar.windowLog+1) cPar.hashLog = cPar.windowLog+1;
     {   U32 const cycleLog = ZSTD_cycleLog(cPar.chainLog, cPar.strategy);
-        if (cycleLog > cPar.windowLog) cPar.chainLog -= (cycleLog - cPar.windowLog);
+        if (cycleLog > cPar.windowLog)
+            cPar.chainLog -= (cycleLog - cPar.windowLog);
     }
 
-    if (cPar.windowLog < ZSTD_WINDOWLOG_ABSOLUTEMIN) cPar.windowLog = ZSTD_WINDOWLOG_ABSOLUTEMIN;  /* required for frame header */
+    if (cPar.windowLog < ZSTD_WINDOWLOG_ABSOLUTEMIN)
+        cPar.windowLog = ZSTD_WINDOWLOG_ABSOLUTEMIN;  /* minimum wlog required for valid frame header */
 
     return cPar;
 }
 
-ZSTD_compressionParameters ZSTD_adjustCParams(ZSTD_compressionParameters cPar, unsigned long long srcSize, size_t dictSize)
+ZSTD_compressionParameters
+ZSTD_adjustCParams(ZSTD_compressionParameters cPar,
+                   unsigned long long srcSize,
+                   size_t dictSize)
 {
-    cPar = ZSTD_clampCParams(cPar);
+    cPar = ZSTD_clampCParams(cPar);   /* resulting cPar is necessarily valid (all parameters within range) */
+    if (srcSize == 0) srcSize = ZSTD_CONTENTSIZE_UNKNOWN;
     return ZSTD_adjustCParams_internal(cPar, srcSize, dictSize);
 }
 
-size_t ZSTD_estimateCCtxSize_advanced_usingCCtxParams(const ZSTD_CCtx_params* params)
+static ZSTD_compressionParameters ZSTD_getCParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize);
+static ZSTD_parameters ZSTD_getParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize);
+
+ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
+        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize)
+{
+    ZSTD_compressionParameters cParams;
+    if (srcSizeHint == ZSTD_CONTENTSIZE_UNKNOWN && CCtxParams->srcSizeHint > 0) {
+      srcSizeHint = CCtxParams->srcSizeHint;
+    }
+    cParams = ZSTD_getCParams_internal(CCtxParams->compressionLevel, srcSizeHint, dictSize);
+    if (CCtxParams->ldmParams.enableLdm) cParams.windowLog = ZSTD_LDM_DEFAULT_WINDOW_LOG;
+    if (CCtxParams->cParams.windowLog) cParams.windowLog = CCtxParams->cParams.windowLog;
+    if (CCtxParams->cParams.hashLog) cParams.hashLog = CCtxParams->cParams.hashLog;
+    if (CCtxParams->cParams.chainLog) cParams.chainLog = CCtxParams->cParams.chainLog;
+    if (CCtxParams->cParams.searchLog) cParams.searchLog = CCtxParams->cParams.searchLog;
+    if (CCtxParams->cParams.minMatch) cParams.minMatch = CCtxParams->cParams.minMatch;
+    if (CCtxParams->cParams.targetLength) cParams.targetLength = CCtxParams->cParams.targetLength;
+    if (CCtxParams->cParams.strategy) cParams.strategy = CCtxParams->cParams.strategy;
+    assert(!ZSTD_checkCParams(cParams));
+    /* srcSizeHint == 0 means 0 */
+    return ZSTD_adjustCParams_internal(cParams, srcSizeHint, dictSize);
+}
+
+static size_t
+ZSTD_sizeof_matchState(const ZSTD_compressionParameters* const cParams,
+                       const U32 forCCtx)
 {
-    /* Estimate CCtx size is supported for single-threaded compression only. */
-    if (params->nbThreads > 1) { return ERROR(GENERIC); }
+    size_t const chainSize = (cParams->strategy == ZSTD_fast) ? 0 : ((size_t)1 << cParams->chainLog);
+    size_t const hSize = ((size_t)1) << cParams->hashLog;
+    U32    const hashLog3 = (forCCtx && cParams->minMatch==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0;
+    size_t const h3Size = hashLog3 ? ((size_t)1) << hashLog3 : 0;
+    /* We don't use ZSTD_cwksp_alloc_size() here because the tables aren't
+     * surrounded by redzones in ASAN. */
+    size_t const tableSpace = chainSize * sizeof(U32)
+                            + hSize * sizeof(U32)
+                            + h3Size * sizeof(U32);
+    size_t const optPotentialSpace =
+        ZSTD_cwksp_alloc_size((MaxML+1) * sizeof(U32))
+      + ZSTD_cwksp_alloc_size((MaxLL+1) * sizeof(U32))
+      + ZSTD_cwksp_alloc_size((MaxOff+1) * sizeof(U32))
+      + ZSTD_cwksp_alloc_size((1<<Litbits) * sizeof(U32))
+      + ZSTD_cwksp_alloc_size((ZSTD_OPT_NUM+1) * sizeof(ZSTD_match_t))
+      + ZSTD_cwksp_alloc_size((ZSTD_OPT_NUM+1) * sizeof(ZSTD_optimal_t));
+    size_t const optSpace = (forCCtx && (cParams->strategy >= ZSTD_btopt))
+                                ? optPotentialSpace
+                                : 0;
+    DEBUGLOG(4, "chainSize: %u - hSize: %u - h3Size: %u",
+                (U32)chainSize, (U32)hSize, (U32)h3Size);
+    return tableSpace + optSpace;
+}
+
+size_t ZSTD_estimateCCtxSize_usingCCtxParams(const ZSTD_CCtx_params* params)
+{
+    RETURN_ERROR_IF(params->nbWorkers > 0, GENERIC, "Estimate CCtx size is supported for single-threaded compression only.");
     {   ZSTD_compressionParameters const cParams =
-                ZSTD_getCParamsFromCCtxParams(*params, 0, 0);
+                ZSTD_getCParamsFromCCtxParams(params, ZSTD_CONTENTSIZE_UNKNOWN, 0);
         size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << cParams.windowLog);
-        U32    const divider = (cParams.searchLength==3) ? 3 : 4;
+        U32    const divider = (cParams.minMatch==3) ? 3 : 4;
         size_t const maxNbSeq = blockSize / divider;
-        size_t const tokenSpace = blockSize + 11*maxNbSeq;
-        size_t const chainSize =
-                (cParams.strategy == ZSTD_fast) ? 0 : (1 << cParams.chainLog);
-        size_t const hSize = ((size_t)1) << cParams.hashLog;
-        U32    const hashLog3 = (cParams.searchLength>3) ?
-                                0 : MIN(ZSTD_HASHLOG3_MAX, cParams.windowLog);
-        size_t const h3Size = ((size_t)1) << hashLog3;
-        size_t const entropySpace = sizeof(ZSTD_entropyCTables_t);
-        size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
-
-        size_t const optBudget =
-                ((MaxML+1) + (MaxLL+1) + (MaxOff+1) + (1<<Litbits))*sizeof(U32)
-                + (ZSTD_OPT_NUM+1)*(sizeof(ZSTD_match_t) + sizeof(ZSTD_optimal_t));
-        size_t const optSpace = ((cParams.strategy == ZSTD_btopt) || (cParams.strategy == ZSTD_btultra)) ? optBudget : 0;
-
-        size_t const ldmSpace = params->ldmParams.enableLdm ?
-            ZSTD_ldm_getTableSize(params->ldmParams.hashLog,
-                                  params->ldmParams.bucketSizeLog) : 0;
-
-        size_t const neededSpace = entropySpace + tableSpace + tokenSpace +
-                                   optSpace + ldmSpace;
-
-        DEBUGLOG(5, "sizeof(ZSTD_CCtx) : %u", (U32)sizeof(ZSTD_CCtx));
-        DEBUGLOG(5, "estimate workSpace : %u", (U32)neededSpace);
-        return sizeof(ZSTD_CCtx) + neededSpace;
+        size_t const tokenSpace = ZSTD_cwksp_alloc_size(WILDCOPY_OVERLENGTH + blockSize)
+                                + ZSTD_cwksp_alloc_size(maxNbSeq * sizeof(seqDef))
+                                + 3 * ZSTD_cwksp_alloc_size(maxNbSeq * sizeof(BYTE));
+        size_t const entropySpace = ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE);
+        size_t const blockStateSpace = 2 * ZSTD_cwksp_alloc_size(sizeof(ZSTD_compressedBlockState_t));
+        size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 1);
+
+        size_t const ldmSpace = ZSTD_ldm_getTableSize(params->ldmParams);
+        size_t const ldmSeqSpace = ZSTD_cwksp_alloc_size(ZSTD_ldm_getMaxNbSeq(params->ldmParams, blockSize) * sizeof(rawSeq));
+
+        /* estimateCCtxSize is for one-shot compression. So no buffers should
+         * be needed. However, we still allocate two 0-sized buffers, which can
+         * take space under ASAN. */
+        size_t const bufferSpace = ZSTD_cwksp_alloc_size(0)
+                                 + ZSTD_cwksp_alloc_size(0);
+
+        size_t const cctxSpace = ZSTD_cwksp_alloc_size(sizeof(ZSTD_CCtx));
+
+        size_t const neededSpace =
+            cctxSpace +
+            entropySpace +
+            blockStateSpace +
+            ldmSpace +
+            ldmSeqSpace +
+            matchStateSize +
+            tokenSpace +
+            bufferSpace;
+
+        DEBUGLOG(5, "estimate workspace : %u", (U32)neededSpace);
+        return neededSpace;
     }
 }
 
-size_t ZSTD_estimateCCtxSize_advanced_usingCParams(ZSTD_compressionParameters cParams)
+size_t ZSTD_estimateCCtxSize_usingCParams(ZSTD_compressionParameters cParams)
 {
     ZSTD_CCtx_params const params = ZSTD_makeCCtxParamsFromCParams(cParams);
-    return ZSTD_estimateCCtxSize_advanced_usingCCtxParams(&params);
+    return ZSTD_estimateCCtxSize_usingCCtxParams(&params);
+}
+
+static size_t ZSTD_estimateCCtxSize_internal(int compressionLevel)
+{
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, 0);
+    return ZSTD_estimateCCtxSize_usingCParams(cParams);
 }
 
 size_t ZSTD_estimateCCtxSize(int compressionLevel)
 {
-    ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, 0, 0);
-    return ZSTD_estimateCCtxSize_advanced_usingCParams(cParams);
+    int level;
+    size_t memBudget = 0;
+    for (level=MIN(compressionLevel, 1); level<=compressionLevel; level++) {
+        size_t const newMB = ZSTD_estimateCCtxSize_internal(level);
+        if (newMB > memBudget) memBudget = newMB;
+    }
+    return memBudget;
 }
 
-size_t ZSTD_estimateCStreamSize_advanced_usingCCtxParams(const ZSTD_CCtx_params* params)
+size_t ZSTD_estimateCStreamSize_usingCCtxParams(const ZSTD_CCtx_params* params)
 {
-    if (params->nbThreads > 1) { return ERROR(GENERIC); }
-    {   size_t const CCtxSize = ZSTD_estimateCCtxSize_advanced_usingCCtxParams(params);
-        size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << params->cParams.windowLog);
-        size_t const inBuffSize = ((size_t)1 << params->cParams.windowLog) + blockSize;
+    RETURN_ERROR_IF(params->nbWorkers > 0, GENERIC, "Estimate CCtx size is supported for single-threaded compression only.");
+    {   ZSTD_compressionParameters const cParams =
+                ZSTD_getCParamsFromCCtxParams(params, ZSTD_CONTENTSIZE_UNKNOWN, 0);
+        size_t const CCtxSize = ZSTD_estimateCCtxSize_usingCCtxParams(params);
+        size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << cParams.windowLog);
+        size_t const inBuffSize = ((size_t)1 << cParams.windowLog) + blockSize;
         size_t const outBuffSize = ZSTD_compressBound(blockSize) + 1;
-        size_t const streamingSize = inBuffSize + outBuffSize;
+        size_t const streamingSize = ZSTD_cwksp_alloc_size(inBuffSize)
+                                   + ZSTD_cwksp_alloc_size(outBuffSize);
 
         return CCtxSize + streamingSize;
     }
 }
 
-size_t ZSTD_estimateCStreamSize_advanced_usingCParams(ZSTD_compressionParameters cParams)
+size_t ZSTD_estimateCStreamSize_usingCParams(ZSTD_compressionParameters cParams)
 {
     ZSTD_CCtx_params const params = ZSTD_makeCCtxParamsFromCParams(cParams);
-    return ZSTD_estimateCStreamSize_advanced_usingCCtxParams(&params);
+    return ZSTD_estimateCStreamSize_usingCCtxParams(&params);
 }
 
-size_t ZSTD_estimateCStreamSize(int compressionLevel) {
-    ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, 0, 0);
-    return ZSTD_estimateCStreamSize_advanced_usingCParams(cParams);
+static size_t ZSTD_estimateCStreamSize_internal(int compressionLevel)
+{
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, 0);
+    return ZSTD_estimateCStreamSize_usingCParams(cParams);
 }
 
-static U32 ZSTD_equivalentCParams(ZSTD_compressionParameters cParams1,
-                                  ZSTD_compressionParameters cParams2)
+size_t ZSTD_estimateCStreamSize(int compressionLevel)
 {
-    U32 bslog1 = MIN(cParams1.windowLog, ZSTD_BLOCKSIZELOG_MAX);
-    U32 bslog2 = MIN(cParams2.windowLog, ZSTD_BLOCKSIZELOG_MAX);
-    return (bslog1 == bslog2)   /* same block size */
-         & (cParams1.hashLog  == cParams2.hashLog)
-         & (cParams1.chainLog == cParams2.chainLog)
-         & (cParams1.strategy == cParams2.strategy)   /* opt parser space */
-         & ((cParams1.searchLength==3) == (cParams2.searchLength==3));  /* hashlog3 space */
+    int level;
+    size_t memBudget = 0;
+    for (level=MIN(compressionLevel, 1); level<=compressionLevel; level++) {
+        size_t const newMB = ZSTD_estimateCStreamSize_internal(level);
+        if (newMB > memBudget) memBudget = newMB;
+    }
+    return memBudget;
 }
 
-/** The parameters are equivalent if ldm is not enabled in both sets or
- *  all the parameters are equivalent. */
-static U32 ZSTD_equivalentLdmParams(ldmParams_t ldmParams1,
-                                    ldmParams_t ldmParams2)
+/* ZSTD_getFrameProgression():
+ * tells how much data has been consumed (input) and produced (output) for current frame.
+ * able to count progression inside worker threads (non-blocking mode).
+ */
+ZSTD_frameProgression ZSTD_getFrameProgression(const ZSTD_CCtx* cctx)
+{
+#ifdef ZSTD_MULTITHREAD
+    if (cctx->appliedParams.nbWorkers > 0) {
+        return ZSTDMT_getFrameProgression(cctx->mtctx);
+    }
+#endif
+    {   ZSTD_frameProgression fp;
+        size_t const buffered = (cctx->inBuff == NULL) ? 0 :
+                                cctx->inBuffPos - cctx->inToCompress;
+        if (buffered) assert(cctx->inBuffPos >= cctx->inToCompress);
+        assert(buffered <= ZSTD_BLOCKSIZE_MAX);
+        fp.ingested = cctx->consumedSrcSize + buffered;
+        fp.consumed = cctx->consumedSrcSize;
+        fp.produced = cctx->producedCSize;
+        fp.flushed  = cctx->producedCSize;   /* simplified; some data might still be left within streaming output buffer */
+        fp.currentJobID = 0;
+        fp.nbActiveWorkers = 0;
+        return fp;
+}   }
+
+/*! ZSTD_toFlushNow()
+ *  Only useful for multithreading scenarios currently (nbWorkers >= 1).
+ */
+size_t ZSTD_toFlushNow(ZSTD_CCtx* cctx)
 {
-    return (!ldmParams1.enableLdm && !ldmParams2.enableLdm) ||
-           (ldmParams1.enableLdm == ldmParams2.enableLdm &&
-            ldmParams1.hashLog == ldmParams2.hashLog &&
-            ldmParams1.bucketSizeLog == ldmParams2.bucketSizeLog &&
-            ldmParams1.minMatchLength == ldmParams2.minMatchLength &&
-            ldmParams1.hashEveryLog == ldmParams2.hashEveryLog);
+#ifdef ZSTD_MULTITHREAD
+    if (cctx->appliedParams.nbWorkers > 0) {
+        return ZSTDMT_toFlushNow(cctx->mtctx);
+    }
+#endif
+    (void)cctx;
+    return 0;   /* over-simplification; could also check if context is currently running in streaming mode, and in which case, report how many bytes are left to be flushed within output buffer */
 }
 
-/** Equivalence for resetCCtx purposes */
-static U32 ZSTD_equivalentParams(ZSTD_CCtx_params params1,
-                                 ZSTD_CCtx_params params2)
+static void ZSTD_assertEqualCParams(ZSTD_compressionParameters cParams1,
+                                    ZSTD_compressionParameters cParams2)
+{
+    (void)cParams1;
+    (void)cParams2;
+    assert(cParams1.windowLog    == cParams2.windowLog);
+    assert(cParams1.chainLog     == cParams2.chainLog);
+    assert(cParams1.hashLog      == cParams2.hashLog);
+    assert(cParams1.searchLog    == cParams2.searchLog);
+    assert(cParams1.minMatch     == cParams2.minMatch);
+    assert(cParams1.targetLength == cParams2.targetLength);
+    assert(cParams1.strategy     == cParams2.strategy);
+}
+
+void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs)
 {
-    return ZSTD_equivalentCParams(params1.cParams, params2.cParams) &&
-           ZSTD_equivalentLdmParams(params1.ldmParams, params2.ldmParams);
+    int i;
+    for (i = 0; i < ZSTD_REP_NUM; ++i)
+        bs->rep[i] = repStartValue[i];
+    bs->entropy.huf.repeatMode = HUF_repeat_none;
+    bs->entropy.fse.offcode_repeatMode = FSE_repeat_none;
+    bs->entropy.fse.matchlength_repeatMode = FSE_repeat_none;
+    bs->entropy.fse.litlength_repeatMode = FSE_repeat_none;
 }
 
-/*! ZSTD_continueCCtx() :
- *  reuse CCtx without reset (note : requires no dictionary) */
-static size_t ZSTD_continueCCtx(ZSTD_CCtx* cctx, ZSTD_CCtx_params params, U64 pledgedSrcSize)
+/*! ZSTD_invalidateMatchState()
+ *  Invalidate all the matches in the match finder tables.
+ *  Requires nextSrc and base to be set (can be NULL).
+ */
+static void ZSTD_invalidateMatchState(ZSTD_matchState_t* ms)
 {
-    U32 const end = (U32)(cctx->nextSrc - cctx->base);
-    DEBUGLOG(5, "continue mode");
-    cctx->appliedParams = params;
-    cctx->pledgedSrcSizePlusOne = pledgedSrcSize+1;
-    cctx->consumedSrcSize = 0;
-    if (pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN)
-        cctx->appliedParams.fParams.contentSizeFlag = 0;
-    DEBUGLOG(5, "pledged content size : %u ; flag : %u",
-        (U32)pledgedSrcSize, cctx->appliedParams.fParams.contentSizeFlag);
-    cctx->lowLimit = end;
-    cctx->dictLimit = end;
-    cctx->nextToUpdate = end+1;
-    cctx->stage = ZSTDcs_init;
-    cctx->dictID = 0;
-    cctx->loadedDictEnd = 0;
-    { int i; for (i=0; i<ZSTD_REP_NUM; i++) cctx->seqStore.rep[i] = repStartValue[i]; }
-    cctx->optState.litLengthSum = 0;  /* force reset of btopt stats */
-    XXH64_reset(&cctx->xxhState, 0);
+    ZSTD_window_clear(&ms->window);
+
+    ms->nextToUpdate = ms->window.dictLimit;
+    ms->loadedDictEnd = 0;
+    ms->opt.litLengthSum = 0;  /* force reset of btopt stats */
+    ms->dictMatchState = NULL;
+}
+
+/**
+ * Indicates whether this compression proceeds directly from user-provided
+ * source buffer to user-provided destination buffer (ZSTDb_not_buffered), or
+ * whether the context needs to buffer the input/output (ZSTDb_buffered).
+ */
+typedef enum {
+    ZSTDb_not_buffered,
+    ZSTDb_buffered
+} ZSTD_buffered_policy_e;
+
+/**
+ * Controls, for this matchState reset, whether the tables need to be cleared /
+ * prepared for the coming compression (ZSTDcrp_makeClean), or whether the
+ * tables can be left unclean (ZSTDcrp_leaveDirty), because we know that a
+ * subsequent operation will overwrite the table space anyways (e.g., copying
+ * the matchState contents in from a CDict).
+ */
+typedef enum {
+    ZSTDcrp_makeClean,
+    ZSTDcrp_leaveDirty
+} ZSTD_compResetPolicy_e;
+
+/**
+ * Controls, for this matchState reset, whether indexing can continue where it
+ * left off (ZSTDirp_continue), or whether it needs to be restarted from zero
+ * (ZSTDirp_reset).
+ */
+typedef enum {
+    ZSTDirp_continue,
+    ZSTDirp_reset
+} ZSTD_indexResetPolicy_e;
+
+typedef enum {
+    ZSTD_resetTarget_CDict,
+    ZSTD_resetTarget_CCtx
+} ZSTD_resetTarget_e;
+
+static size_t
+ZSTD_reset_matchState(ZSTD_matchState_t* ms,
+                      ZSTD_cwksp* ws,
+                const ZSTD_compressionParameters* cParams,
+                const ZSTD_compResetPolicy_e crp,
+                const ZSTD_indexResetPolicy_e forceResetIndex,
+                const ZSTD_resetTarget_e forWho)
+{
+    size_t const chainSize = (cParams->strategy == ZSTD_fast) ? 0 : ((size_t)1 << cParams->chainLog);
+    size_t const hSize = ((size_t)1) << cParams->hashLog;
+    U32    const hashLog3 = ((forWho == ZSTD_resetTarget_CCtx) && cParams->minMatch==3) ? MIN(ZSTD_HASHLOG3_MAX, cParams->windowLog) : 0;
+    size_t const h3Size = hashLog3 ? ((size_t)1) << hashLog3 : 0;
+
+    DEBUGLOG(4, "reset indices : %u", forceResetIndex == ZSTDirp_reset);
+    if (forceResetIndex == ZSTDirp_reset) {
+        ZSTD_window_init(&ms->window);
+        ZSTD_cwksp_mark_tables_dirty(ws);
+    }
+
+    ms->hashLog3 = hashLog3;
+
+    ZSTD_invalidateMatchState(ms);
+
+    assert(!ZSTD_cwksp_reserve_failed(ws)); /* check that allocation hasn't already failed */
+
+    ZSTD_cwksp_clear_tables(ws);
+
+    DEBUGLOG(5, "reserving table space");
+    /* table Space */
+    ms->hashTable = (U32*)ZSTD_cwksp_reserve_table(ws, hSize * sizeof(U32));
+    ms->chainTable = (U32*)ZSTD_cwksp_reserve_table(ws, chainSize * sizeof(U32));
+    ms->hashTable3 = (U32*)ZSTD_cwksp_reserve_table(ws, h3Size * sizeof(U32));
+    RETURN_ERROR_IF(ZSTD_cwksp_reserve_failed(ws), memory_allocation,
+                    "failed a workspace allocation in ZSTD_reset_matchState");
+
+    DEBUGLOG(4, "reset table : %u", crp!=ZSTDcrp_leaveDirty);
+    if (crp!=ZSTDcrp_leaveDirty) {
+        /* reset tables only */
+        ZSTD_cwksp_clean_tables(ws);
+    }
+
+    /* opt parser space */
+    if ((forWho == ZSTD_resetTarget_CCtx) && (cParams->strategy >= ZSTD_btopt)) {
+        DEBUGLOG(4, "reserving optimal parser space");
+        ms->opt.litFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (1<<Litbits) * sizeof(unsigned));
+        ms->opt.litLengthFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxLL+1) * sizeof(unsigned));
+        ms->opt.matchLengthFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxML+1) * sizeof(unsigned));
+        ms->opt.offCodeFreq = (unsigned*)ZSTD_cwksp_reserve_aligned(ws, (MaxOff+1) * sizeof(unsigned));
+        ms->opt.matchTable = (ZSTD_match_t*)ZSTD_cwksp_reserve_aligned(ws, (ZSTD_OPT_NUM+1) * sizeof(ZSTD_match_t));
+        ms->opt.priceTable = (ZSTD_optimal_t*)ZSTD_cwksp_reserve_aligned(ws, (ZSTD_OPT_NUM+1) * sizeof(ZSTD_optimal_t));
+    }
+
+    ms->cParams = *cParams;
+
+    RETURN_ERROR_IF(ZSTD_cwksp_reserve_failed(ws), memory_allocation,
+                    "failed a workspace allocation in ZSTD_reset_matchState");
+
     return 0;
 }
 
-typedef enum { ZSTDcrp_continue, ZSTDcrp_noMemset } ZSTD_compResetPolicy_e;
-typedef enum { ZSTDb_not_buffered, ZSTDb_buffered } ZSTD_buffered_policy_e;
+/* ZSTD_indexTooCloseToMax() :
+ * minor optimization : prefer memset() rather than reduceIndex()
+ * which is measurably slow in some circumstances (reported for Visual Studio).
+ * Works when re-using a context for a lot of smallish inputs :
+ * if all inputs are smaller than ZSTD_INDEXOVERFLOW_MARGIN,
+ * memset() will be triggered before reduceIndex().
+ */
+#define ZSTD_INDEXOVERFLOW_MARGIN (16 MB)
+static int ZSTD_indexTooCloseToMax(ZSTD_window_t w)
+{
+    return (size_t)(w.nextSrc - w.base) > (ZSTD_CURRENT_MAX - ZSTD_INDEXOVERFLOW_MARGIN);
+}
 
 /*! ZSTD_resetCCtx_internal() :
     note : `params` are assumed fully validated at this stage */
 static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
-                                      ZSTD_CCtx_params params, U64 pledgedSrcSize,
+                                      ZSTD_CCtx_params params,
+                                      U64 const pledgedSrcSize,
                                       ZSTD_compResetPolicy_e const crp,
                                       ZSTD_buffered_policy_e const zbuff)
 {
+    ZSTD_cwksp* const ws = &zc->workspace;
+    DEBUGLOG(4, "ZSTD_resetCCtx_internal: pledgedSrcSize=%u, wlog=%u",
+                (U32)pledgedSrcSize, params.cParams.windowLog);
     assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
 
-    if (crp == ZSTDcrp_continue) {
-        if (ZSTD_equivalentParams(params, zc->appliedParams)) {
-            DEBUGLOG(5, "ZSTD_equivalentParams()==1");
-            assert(!(params.ldmParams.enableLdm &&
-                     params.ldmParams.hashEveryLog == ZSTD_LDM_HASHEVERYLOG_NOTSET));
-            zc->entropy->hufCTable_repeatMode = HUF_repeat_none;
-            zc->entropy->offcode_repeatMode = FSE_repeat_none;
-            zc->entropy->matchlength_repeatMode = FSE_repeat_none;
-            zc->entropy->litlength_repeatMode = FSE_repeat_none;
-            return ZSTD_continueCCtx(zc, params, pledgedSrcSize);
-    }   }
+    zc->isFirstBlock = 1;
 
     if (params.ldmParams.enableLdm) {
         /* Adjust long distance matching parameters */
-        ZSTD_ldm_adjustParameters(&params.ldmParams, params.cParams.windowLog);
+        ZSTD_ldm_adjustParameters(&params.ldmParams, &params.cParams);
         assert(params.ldmParams.hashLog >= params.ldmParams.bucketSizeLog);
-        assert(params.ldmParams.hashEveryLog < 32);
-        zc->ldmState.hashPower =
-                ZSTD_ldm_getHashPower(params.ldmParams.minMatchLength);
+        assert(params.ldmParams.hashRateLog < 32);
+        zc->ldmState.hashPower = ZSTD_rollingHash_primePower(params.ldmParams.minMatchLength);
     }
 
-    {   size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, (size_t)1 << params.cParams.windowLog);
-        U32    const divider = (params.cParams.searchLength==3) ? 3 : 4;
+    {   size_t const windowSize = MAX(1, (size_t)MIN(((U64)1 << params.cParams.windowLog), pledgedSrcSize));
+        size_t const blockSize = MIN(ZSTD_BLOCKSIZE_MAX, windowSize);
+        U32    const divider = (params.cParams.minMatch==3) ? 3 : 4;
         size_t const maxNbSeq = blockSize / divider;
-        size_t const tokenSpace = blockSize + 11*maxNbSeq;
-        size_t const chainSize = (params.cParams.strategy == ZSTD_fast) ?
-                                0 : (1 << params.cParams.chainLog);
-        size_t const hSize = ((size_t)1) << params.cParams.hashLog;
-        U32    const hashLog3 = (params.cParams.searchLength>3) ?
-                                0 : MIN(ZSTD_HASHLOG3_MAX, params.cParams.windowLog);
-        size_t const h3Size = ((size_t)1) << hashLog3;
-        size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
+        size_t const tokenSpace = ZSTD_cwksp_alloc_size(WILDCOPY_OVERLENGTH + blockSize)
+                                + ZSTD_cwksp_alloc_size(maxNbSeq * sizeof(seqDef))
+                                + 3 * ZSTD_cwksp_alloc_size(maxNbSeq * sizeof(BYTE));
         size_t const buffOutSize = (zbuff==ZSTDb_buffered) ? ZSTD_compressBound(blockSize)+1 : 0;
-        size_t const buffInSize = (zbuff==ZSTDb_buffered) ? ((size_t)1 << params.cParams.windowLog) + blockSize : 0;
-        void* ptr;
-
-        /* Check if workSpace is large enough, alloc a new one if needed */
-        {   size_t const entropySpace = sizeof(ZSTD_entropyCTables_t);
-            size_t const optPotentialSpace = ((MaxML+1) + (MaxLL+1) + (MaxOff+1) + (1<<Litbits)) * sizeof(U32)
-                                  + (ZSTD_OPT_NUM+1) * (sizeof(ZSTD_match_t)+sizeof(ZSTD_optimal_t));
-            size_t const optSpace = ( (params.cParams.strategy == ZSTD_btopt)
-                                    || (params.cParams.strategy == ZSTD_btultra)) ?
-                                    optPotentialSpace : 0;
-            size_t const bufferSpace = buffInSize + buffOutSize;
-            size_t const ldmSpace = params.ldmParams.enableLdm
-                ? ZSTD_ldm_getTableSize(params.ldmParams.hashLog, params.ldmParams.bucketSizeLog)
-                : 0;
-            size_t const neededSpace = entropySpace + optSpace + ldmSpace +
-                                       tableSpace + tokenSpace + bufferSpace;
-
-            if (zc->workSpaceSize < neededSpace) {  /* too small : resize */
-                DEBUGLOG(5, "Need to update workSpaceSize from %uK to %uK \n",
-                            (unsigned)zc->workSpaceSize>>10,
-                            (unsigned)neededSpace>>10);
-                /* static cctx : no resize, error out */
-                if (zc->staticSize) return ERROR(memory_allocation);
-
-                zc->workSpaceSize = 0;
-                ZSTD_free(zc->workSpace, zc->customMem);
-                zc->workSpace = ZSTD_malloc(neededSpace, zc->customMem);
-                if (zc->workSpace == NULL) return ERROR(memory_allocation);
-                zc->workSpaceSize = neededSpace;
-                ptr = zc->workSpace;
-
-                /* entropy space */
-                assert(((size_t)zc->workSpace & 3) == 0);   /* ensure correct alignment */
-                assert(zc->workSpaceSize >= sizeof(ZSTD_entropyCTables_t));
-                zc->entropy = (ZSTD_entropyCTables_t*)zc->workSpace;
+        size_t const buffInSize = (zbuff==ZSTDb_buffered) ? windowSize + blockSize : 0;
+        size_t const matchStateSize = ZSTD_sizeof_matchState(&params.cParams, /* forCCtx */ 1);
+        size_t const maxNbLdmSeq = ZSTD_ldm_getMaxNbSeq(params.ldmParams, blockSize);
+
+        ZSTD_indexResetPolicy_e needsIndexReset = zc->initialized ? ZSTDirp_continue : ZSTDirp_reset;
+
+        if (ZSTD_indexTooCloseToMax(zc->blockState.matchState.window)) {
+            needsIndexReset = ZSTDirp_reset;
+        }
+
+        if (!zc->staticSize) ZSTD_cwksp_bump_oversized_duration(ws, 0);
+
+        /* Check if workspace is large enough, alloc a new one if needed */
+        {   size_t const cctxSpace = zc->staticSize ? ZSTD_cwksp_alloc_size(sizeof(ZSTD_CCtx)) : 0;
+            size_t const entropySpace = ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE);
+            size_t const blockStateSpace = 2 * ZSTD_cwksp_alloc_size(sizeof(ZSTD_compressedBlockState_t));
+            size_t const bufferSpace = ZSTD_cwksp_alloc_size(buffInSize) + ZSTD_cwksp_alloc_size(buffOutSize);
+            size_t const ldmSpace = ZSTD_ldm_getTableSize(params.ldmParams);
+            size_t const ldmSeqSpace = ZSTD_cwksp_alloc_size(maxNbLdmSeq * sizeof(rawSeq));
+
+            size_t const neededSpace =
+                cctxSpace +
+                entropySpace +
+                blockStateSpace +
+                ldmSpace +
+                ldmSeqSpace +
+                matchStateSize +
+                tokenSpace +
+                bufferSpace;
+
+            int const workspaceTooSmall = ZSTD_cwksp_sizeof(ws) < neededSpace;
+            int const workspaceWasteful = ZSTD_cwksp_check_wasteful(ws, neededSpace);
+
+            DEBUGLOG(4, "Need %zuKB workspace, including %zuKB for match state, and %zuKB for buffers",
+                        neededSpace>>10, matchStateSize>>10, bufferSpace>>10);
+            DEBUGLOG(4, "windowSize: %zu - blockSize: %zu", windowSize, blockSize);
+
+            if (workspaceTooSmall || workspaceWasteful) {
+                DEBUGLOG(4, "Resize workspaceSize from %zuKB to %zuKB",
+                            ZSTD_cwksp_sizeof(ws) >> 10,
+                            neededSpace >> 10);
+
+                RETURN_ERROR_IF(zc->staticSize, memory_allocation, "static cctx : no resize");
+
+                needsIndexReset = ZSTDirp_reset;
+
+                ZSTD_cwksp_free(ws, zc->customMem);
+                FORWARD_IF_ERROR(ZSTD_cwksp_create(ws, neededSpace, zc->customMem), "");
+
+                DEBUGLOG(5, "reserving object space");
+                /* Statically sized space.
+                 * entropyWorkspace never moves,
+                 * though prev/next block swap places */
+                assert(ZSTD_cwksp_check_available(ws, 2 * sizeof(ZSTD_compressedBlockState_t)));
+                zc->blockState.prevCBlock = (ZSTD_compressedBlockState_t*) ZSTD_cwksp_reserve_object(ws, sizeof(ZSTD_compressedBlockState_t));
+                RETURN_ERROR_IF(zc->blockState.prevCBlock == NULL, memory_allocation, "couldn't allocate prevCBlock");
+                zc->blockState.nextCBlock = (ZSTD_compressedBlockState_t*) ZSTD_cwksp_reserve_object(ws, sizeof(ZSTD_compressedBlockState_t));
+                RETURN_ERROR_IF(zc->blockState.nextCBlock == NULL, memory_allocation, "couldn't allocate nextCBlock");
+                zc->entropyWorkspace = (U32*) ZSTD_cwksp_reserve_object(ws, HUF_WORKSPACE_SIZE);
+                RETURN_ERROR_IF(zc->blockState.nextCBlock == NULL, memory_allocation, "couldn't allocate entropyWorkspace");
         }   }
 
+        ZSTD_cwksp_clear(ws);
+
         /* init params */
         zc->appliedParams = params;
+        zc->blockState.matchState.cParams = params.cParams;
         zc->pledgedSrcSizePlusOne = pledgedSrcSize+1;
         zc->consumedSrcSize = 0;
+        zc->producedCSize = 0;
         if (pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN)
             zc->appliedParams.fParams.contentSizeFlag = 0;
-        DEBUGLOG(5, "pledged content size : %u ; flag : %u",
-            (U32)pledgedSrcSize, zc->appliedParams.fParams.contentSizeFlag);
+        DEBUGLOG(4, "pledged content size : %u ; flag : %u",
+            (unsigned)pledgedSrcSize, zc->appliedParams.fParams.contentSizeFlag);
         zc->blockSize = blockSize;
 
         XXH64_reset(&zc->xxhState, 0);
         zc->stage = ZSTDcs_init;
         zc->dictID = 0;
-        zc->loadedDictEnd = 0;
-        zc->entropy->hufCTable_repeatMode = HUF_repeat_none;
-        zc->entropy->offcode_repeatMode = FSE_repeat_none;
-        zc->entropy->matchlength_repeatMode = FSE_repeat_none;
-        zc->entropy->litlength_repeatMode = FSE_repeat_none;
-        zc->nextToUpdate = 1;
-        zc->nextSrc = NULL;
-        zc->base = NULL;
-        zc->dictBase = NULL;
-        zc->dictLimit = 0;
-        zc->lowLimit = 0;
-        { int i; for (i=0; i<ZSTD_REP_NUM; i++) zc->seqStore.rep[i] = repStartValue[i]; }
-        zc->hashLog3 = hashLog3;
-        zc->optState.litLengthSum = 0;
-
-        ptr = zc->entropy + 1;
-
-        /* opt parser space */
-        if ((params.cParams.strategy == ZSTD_btopt) || (params.cParams.strategy == ZSTD_btultra)) {
-            DEBUGLOG(5, "reserving optimal parser space");
-            assert(((size_t)ptr & 3) == 0);  /* ensure ptr is properly aligned */
-            zc->optState.litFreq = (U32*)ptr;
-            zc->optState.litLengthFreq = zc->optState.litFreq + (1<<Litbits);
-            zc->optState.matchLengthFreq = zc->optState.litLengthFreq + (MaxLL+1);
-            zc->optState.offCodeFreq = zc->optState.matchLengthFreq + (MaxML+1);
-            ptr = zc->optState.offCodeFreq + (MaxOff+1);
-            zc->optState.matchTable = (ZSTD_match_t*)ptr;
-            ptr = zc->optState.matchTable + ZSTD_OPT_NUM+1;
-            zc->optState.priceTable = (ZSTD_optimal_t*)ptr;
-            ptr = zc->optState.priceTable + ZSTD_OPT_NUM+1;
-        }
 
-        /* ldm hash table */
-        /* initialize bucketOffsets table later for pointer alignment */
-        if (params.ldmParams.enableLdm) {
-            size_t const ldmHSize = ((size_t)1) << params.ldmParams.hashLog;
-            memset(ptr, 0, ldmHSize * sizeof(ldmEntry_t));
-            assert(((size_t)ptr & 3) == 0); /* ensure ptr is properly aligned */
-            zc->ldmState.hashTable = (ldmEntry_t*)ptr;
-            ptr = zc->ldmState.hashTable + ldmHSize;
-        }
+        ZSTD_reset_compressedBlockState(zc->blockState.prevCBlock);
 
-        /* table Space */
-        if (crp!=ZSTDcrp_noMemset) memset(ptr, 0, tableSpace);   /* reset tables only */
-        assert(((size_t)ptr & 3) == 0);  /* ensure ptr is properly aligned */
-        zc->hashTable = (U32*)(ptr);
-        zc->chainTable = zc->hashTable + hSize;
-        zc->hashTable3 = zc->chainTable + chainSize;
-        ptr = zc->hashTable3 + h3Size;
+        /* ZSTD_wildcopy() is used to copy into the literals buffer,
+         * so we have to oversize the buffer by WILDCOPY_OVERLENGTH bytes.
+         */
+        zc->seqStore.litStart = ZSTD_cwksp_reserve_buffer(ws, blockSize + WILDCOPY_OVERLENGTH);
+        zc->seqStore.maxNbLit = blockSize;
 
-        /* sequences storage */
-        zc->seqStore.sequencesStart = (seqDef*)ptr;
-        ptr = zc->seqStore.sequencesStart + maxNbSeq;
-        zc->seqStore.llCode = (BYTE*) ptr;
-        zc->seqStore.mlCode = zc->seqStore.llCode + maxNbSeq;
-        zc->seqStore.ofCode = zc->seqStore.mlCode + maxNbSeq;
-        zc->seqStore.litStart = zc->seqStore.ofCode + maxNbSeq;
-        ptr = zc->seqStore.litStart + blockSize;
+        /* buffers */
+        zc->inBuffSize = buffInSize;
+        zc->inBuff = (char*)ZSTD_cwksp_reserve_buffer(ws, buffInSize);
+        zc->outBuffSize = buffOutSize;
+        zc->outBuff = (char*)ZSTD_cwksp_reserve_buffer(ws, buffOutSize);
 
         /* ldm bucketOffsets table */
         if (params.ldmParams.enableLdm) {
+            /* TODO: avoid memset? */
             size_t const ldmBucketSize =
                   ((size_t)1) << (params.ldmParams.hashLog -
                                   params.ldmParams.bucketSizeLog);
-            memset(ptr, 0, ldmBucketSize);
-            zc->ldmState.bucketOffsets = (BYTE*)ptr;
-            ptr = zc->ldmState.bucketOffsets + ldmBucketSize;
+            zc->ldmState.bucketOffsets = ZSTD_cwksp_reserve_buffer(ws, ldmBucketSize);
+            memset(zc->ldmState.bucketOffsets, 0, ldmBucketSize);
         }
 
-        /* buffers */
-        zc->inBuffSize = buffInSize;
-        zc->inBuff = (char*)ptr;
-        zc->outBuffSize = buffOutSize;
-        zc->outBuff = zc->inBuff + buffInSize;
+        /* sequences storage */
+        ZSTD_referenceExternalSequences(zc, NULL, 0);
+        zc->seqStore.maxNbSeq = maxNbSeq;
+        zc->seqStore.llCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE));
+        zc->seqStore.mlCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE));
+        zc->seqStore.ofCode = ZSTD_cwksp_reserve_buffer(ws, maxNbSeq * sizeof(BYTE));
+        zc->seqStore.sequencesStart = (seqDef*)ZSTD_cwksp_reserve_aligned(ws, maxNbSeq * sizeof(seqDef));
+
+        FORWARD_IF_ERROR(ZSTD_reset_matchState(
+            &zc->blockState.matchState,
+            ws,
+            &params.cParams,
+            crp,
+            needsIndexReset,
+            ZSTD_resetTarget_CCtx), "");
+
+        /* ldm hash table */
+        if (params.ldmParams.enableLdm) {
+            /* TODO: avoid memset? */
+            size_t const ldmHSize = ((size_t)1) << params.ldmParams.hashLog;
+            zc->ldmState.hashTable = (ldmEntry_t*)ZSTD_cwksp_reserve_aligned(ws, ldmHSize * sizeof(ldmEntry_t));
+            memset(zc->ldmState.hashTable, 0, ldmHSize * sizeof(ldmEntry_t));
+            zc->ldmSequences = (rawSeq*)ZSTD_cwksp_reserve_aligned(ws, maxNbLdmSeq * sizeof(rawSeq));
+            zc->maxNbLdmSequences = maxNbLdmSeq;
+
+            ZSTD_window_init(&zc->ldmState.window);
+            ZSTD_window_clear(&zc->ldmState.window);
+            zc->ldmState.loadedDictEnd = 0;
+        }
+
+        DEBUGLOG(3, "wksp: finished allocating, %zd bytes remain available", ZSTD_cwksp_available_space(ws));
+        zc->initialized = 1;
 
         return 0;
     }
@@ -975,25 +1592,188 @@ static size_t ZSTD_resetCCtx_internal(ZSTD_CCtx* zc,
  *        do not use with extDict variant ! */
 void ZSTD_invalidateRepCodes(ZSTD_CCtx* cctx) {
     int i;
-    for (i=0; i<ZSTD_REP_NUM; i++) cctx->seqStore.rep[i] = 0;
+    for (i=0; i<ZSTD_REP_NUM; i++) cctx->blockState.prevCBlock->rep[i] = 0;
+    assert(!ZSTD_window_hasExtDict(cctx->blockState.matchState.window));
+}
+
+/* These are the approximate sizes for each strategy past which copying the
+ * dictionary tables into the working context is faster than using them
+ * in-place.
+ */
+static const size_t attachDictSizeCutoffs[ZSTD_STRATEGY_MAX+1] = {
+    8 KB,  /* unused */
+    8 KB,  /* ZSTD_fast */
+    16 KB, /* ZSTD_dfast */
+    32 KB, /* ZSTD_greedy */
+    32 KB, /* ZSTD_lazy */
+    32 KB, /* ZSTD_lazy2 */
+    32 KB, /* ZSTD_btlazy2 */
+    32 KB, /* ZSTD_btopt */
+    8 KB,  /* ZSTD_btultra */
+    8 KB   /* ZSTD_btultra2 */
+};
+
+static int ZSTD_shouldAttachDict(const ZSTD_CDict* cdict,
+                                 const ZSTD_CCtx_params* params,
+                                 U64 pledgedSrcSize)
+{
+    size_t cutoff = attachDictSizeCutoffs[cdict->matchState.cParams.strategy];
+    return ( pledgedSrcSize <= cutoff
+          || pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN
+          || params->attachDictPref == ZSTD_dictForceAttach )
+        && params->attachDictPref != ZSTD_dictForceCopy
+        && !params->forceWindow; /* dictMatchState isn't correctly
+                                 * handled in _enforceMaxDist */
+}
+
+static size_t
+ZSTD_resetCCtx_byAttachingCDict(ZSTD_CCtx* cctx,
+                        const ZSTD_CDict* cdict,
+                        ZSTD_CCtx_params params,
+                        U64 pledgedSrcSize,
+                        ZSTD_buffered_policy_e zbuff)
+{
+    {   const ZSTD_compressionParameters* const cdict_cParams = &cdict->matchState.cParams;
+        unsigned const windowLog = params.cParams.windowLog;
+        assert(windowLog != 0);
+        /* Resize working context table params for input only, since the dict
+         * has its own tables. */
+        /* pledgeSrcSize == 0 means 0! */
+        params.cParams = ZSTD_adjustCParams_internal(*cdict_cParams, pledgedSrcSize, 0);
+        params.cParams.windowLog = windowLog;
+        FORWARD_IF_ERROR(ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
+                                                 ZSTDcrp_makeClean, zbuff), "");
+        assert(cctx->appliedParams.cParams.strategy == cdict_cParams->strategy);
+    }
+
+    {   const U32 cdictEnd = (U32)( cdict->matchState.window.nextSrc
+                                  - cdict->matchState.window.base);
+        const U32 cdictLen = cdictEnd - cdict->matchState.window.dictLimit;
+        if (cdictLen == 0) {
+            /* don't even attach dictionaries with no contents */
+            DEBUGLOG(4, "skipping attaching empty dictionary");
+        } else {
+            DEBUGLOG(4, "attaching dictionary into context");
+            cctx->blockState.matchState.dictMatchState = &cdict->matchState;
+
+            /* prep working match state so dict matches never have negative indices
+             * when they are translated to the working context's index space. */
+            if (cctx->blockState.matchState.window.dictLimit < cdictEnd) {
+                cctx->blockState.matchState.window.nextSrc =
+                    cctx->blockState.matchState.window.base + cdictEnd;
+                ZSTD_window_clear(&cctx->blockState.matchState.window);
+            }
+            /* loadedDictEnd is expressed within the referential of the active context */
+            cctx->blockState.matchState.loadedDictEnd = cctx->blockState.matchState.window.dictLimit;
+    }   }
+
+    cctx->dictID = cdict->dictID;
+
+    /* copy block state */
+    memcpy(cctx->blockState.prevCBlock, &cdict->cBlockState, sizeof(cdict->cBlockState));
+
+    return 0;
+}
+
+static size_t ZSTD_resetCCtx_byCopyingCDict(ZSTD_CCtx* cctx,
+                            const ZSTD_CDict* cdict,
+                            ZSTD_CCtx_params params,
+                            U64 pledgedSrcSize,
+                            ZSTD_buffered_policy_e zbuff)
+{
+    const ZSTD_compressionParameters *cdict_cParams = &cdict->matchState.cParams;
+
+    DEBUGLOG(4, "copying dictionary into context");
+
+    {   unsigned const windowLog = params.cParams.windowLog;
+        assert(windowLog != 0);
+        /* Copy only compression parameters related to tables. */
+        params.cParams = *cdict_cParams;
+        params.cParams.windowLog = windowLog;
+        FORWARD_IF_ERROR(ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
+                                                 ZSTDcrp_leaveDirty, zbuff), "");
+        assert(cctx->appliedParams.cParams.strategy == cdict_cParams->strategy);
+        assert(cctx->appliedParams.cParams.hashLog == cdict_cParams->hashLog);
+        assert(cctx->appliedParams.cParams.chainLog == cdict_cParams->chainLog);
+    }
+
+    ZSTD_cwksp_mark_tables_dirty(&cctx->workspace);
+
+    /* copy tables */
+    {   size_t const chainSize = (cdict_cParams->strategy == ZSTD_fast) ? 0 : ((size_t)1 << cdict_cParams->chainLog);
+        size_t const hSize =  (size_t)1 << cdict_cParams->hashLog;
+
+        memcpy(cctx->blockState.matchState.hashTable,
+               cdict->matchState.hashTable,
+               hSize * sizeof(U32));
+        memcpy(cctx->blockState.matchState.chainTable,
+               cdict->matchState.chainTable,
+               chainSize * sizeof(U32));
+    }
+
+    /* Zero the hashTable3, since the cdict never fills it */
+    {   int const h3log = cctx->blockState.matchState.hashLog3;
+        size_t const h3Size = h3log ? ((size_t)1 << h3log) : 0;
+        assert(cdict->matchState.hashLog3 == 0);
+        memset(cctx->blockState.matchState.hashTable3, 0, h3Size * sizeof(U32));
+    }
+
+    ZSTD_cwksp_mark_tables_clean(&cctx->workspace);
+
+    /* copy dictionary offsets */
+    {   ZSTD_matchState_t const* srcMatchState = &cdict->matchState;
+        ZSTD_matchState_t* dstMatchState = &cctx->blockState.matchState;
+        dstMatchState->window       = srcMatchState->window;
+        dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
+        dstMatchState->loadedDictEnd= srcMatchState->loadedDictEnd;
+    }
+
+    cctx->dictID = cdict->dictID;
+
+    /* copy block state */
+    memcpy(cctx->blockState.prevCBlock, &cdict->cBlockState, sizeof(cdict->cBlockState));
+
+    return 0;
 }
 
+/* We have a choice between copying the dictionary context into the working
+ * context, or referencing the dictionary context from the working context
+ * in-place. We decide here which strategy to use. */
+static size_t ZSTD_resetCCtx_usingCDict(ZSTD_CCtx* cctx,
+                            const ZSTD_CDict* cdict,
+                            const ZSTD_CCtx_params* params,
+                            U64 pledgedSrcSize,
+                            ZSTD_buffered_policy_e zbuff)
+{
+
+    DEBUGLOG(4, "ZSTD_resetCCtx_usingCDict (pledgedSrcSize=%u)",
+                (unsigned)pledgedSrcSize);
+
+    if (ZSTD_shouldAttachDict(cdict, params, pledgedSrcSize)) {
+        return ZSTD_resetCCtx_byAttachingCDict(
+            cctx, cdict, *params, pledgedSrcSize, zbuff);
+    } else {
+        return ZSTD_resetCCtx_byCopyingCDict(
+            cctx, cdict, *params, pledgedSrcSize, zbuff);
+    }
+}
 
 /*! ZSTD_copyCCtx_internal() :
  *  Duplicate an existing context `srcCCtx` into another one `dstCCtx`.
- *  The "context", in this case, refers to the hash and chain tables, entropy
- *  tables, and dictionary offsets.
  *  Only works during stage ZSTDcs_init (i.e. after creation, but before first call to ZSTD_compressContinue()).
- *  pledgedSrcSize=0 means "empty" if fParams.contentSizeFlag=1
- *  @return : 0, or an error code */
+ *  The "context", in this case, refers to the hash and chain tables,
+ *  entropy tables, and dictionary references.
+ * `windowLog` value is enforced if != 0, otherwise value is copied from srcCCtx.
+ * @return : 0, or an error code */
 static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
                             const ZSTD_CCtx* srcCCtx,
                             ZSTD_frameParameters fParams,
-                            unsigned long long pledgedSrcSize,
+                            U64 pledgedSrcSize,
                             ZSTD_buffered_policy_e zbuff)
 {
     DEBUGLOG(5, "ZSTD_copyCCtx_internal");
-    if (srcCCtx->stage!=ZSTDcs_init) return ERROR(stage_wrong);
+    RETURN_ERROR_IF(srcCCtx->stage!=ZSTDcs_init, stage_wrong,
+                    "Can't copy a ctx that's not in init stage.");
 
     memcpy(&dstCCtx->customMem, &srcCCtx->customMem, sizeof(ZSTD_customMem));
     {   ZSTD_CCtx_params params = dstCCtx->requestedParams;
@@ -1001,32 +1781,47 @@ static size_t ZSTD_copyCCtx_internal(ZSTD_CCtx* dstCCtx,
         params.cParams = srcCCtx->appliedParams.cParams;
         params.fParams = fParams;
         ZSTD_resetCCtx_internal(dstCCtx, params, pledgedSrcSize,
-                                ZSTDcrp_noMemset, zbuff);
+                                ZSTDcrp_leaveDirty, zbuff);
+        assert(dstCCtx->appliedParams.cParams.windowLog == srcCCtx->appliedParams.cParams.windowLog);
+        assert(dstCCtx->appliedParams.cParams.strategy == srcCCtx->appliedParams.cParams.strategy);
+        assert(dstCCtx->appliedParams.cParams.hashLog == srcCCtx->appliedParams.cParams.hashLog);
+        assert(dstCCtx->appliedParams.cParams.chainLog == srcCCtx->appliedParams.cParams.chainLog);
+        assert(dstCCtx->blockState.matchState.hashLog3 == srcCCtx->blockState.matchState.hashLog3);
     }
 
+    ZSTD_cwksp_mark_tables_dirty(&dstCCtx->workspace);
+
     /* copy tables */
-    {   size_t const chainSize = (srcCCtx->appliedParams.cParams.strategy == ZSTD_fast) ? 0 : (1 << srcCCtx->appliedParams.cParams.chainLog);
+    {   size_t const chainSize = (srcCCtx->appliedParams.cParams.strategy == ZSTD_fast) ? 0 : ((size_t)1 << srcCCtx->appliedParams.cParams.chainLog);
         size_t const hSize =  (size_t)1 << srcCCtx->appliedParams.cParams.hashLog;
-        size_t const h3Size = (size_t)1 << srcCCtx->hashLog3;
-        size_t const tableSpace = (chainSize + hSize + h3Size) * sizeof(U32);
-        assert((U32*)dstCCtx->chainTable == (U32*)dstCCtx->hashTable + hSize);  /* chainTable must follow hashTable */
-        assert((U32*)dstCCtx->hashTable3 == (U32*)dstCCtx->chainTable + chainSize);
-        memcpy(dstCCtx->hashTable, srcCCtx->hashTable, tableSpace);   /* presumes all tables follow each other */
+        int const h3log = srcCCtx->blockState.matchState.hashLog3;
+        size_t const h3Size = h3log ? ((size_t)1 << h3log) : 0;
+
+        memcpy(dstCCtx->blockState.matchState.hashTable,
+               srcCCtx->blockState.matchState.hashTable,
+               hSize * sizeof(U32));
+        memcpy(dstCCtx->blockState.matchState.chainTable,
+               srcCCtx->blockState.matchState.chainTable,
+               chainSize * sizeof(U32));
+        memcpy(dstCCtx->blockState.matchState.hashTable3,
+               srcCCtx->blockState.matchState.hashTable3,
+               h3Size * sizeof(U32));
     }
 
+    ZSTD_cwksp_mark_tables_clean(&dstCCtx->workspace);
+
     /* copy dictionary offsets */
-    dstCCtx->nextToUpdate = srcCCtx->nextToUpdate;
-    dstCCtx->nextToUpdate3= srcCCtx->nextToUpdate3;
-    dstCCtx->nextSrc      = srcCCtx->nextSrc;
-    dstCCtx->base         = srcCCtx->base;
-    dstCCtx->dictBase     = srcCCtx->dictBase;
-    dstCCtx->dictLimit    = srcCCtx->dictLimit;
-    dstCCtx->lowLimit     = srcCCtx->lowLimit;
-    dstCCtx->loadedDictEnd= srcCCtx->loadedDictEnd;
-    dstCCtx->dictID       = srcCCtx->dictID;
-
-    /* copy entropy tables */
-    memcpy(dstCCtx->entropy, srcCCtx->entropy, sizeof(ZSTD_entropyCTables_t));
+    {
+        const ZSTD_matchState_t* srcMatchState = &srcCCtx->blockState.matchState;
+        ZSTD_matchState_t* dstMatchState = &dstCCtx->blockState.matchState;
+        dstMatchState->window       = srcMatchState->window;
+        dstMatchState->nextToUpdate = srcMatchState->nextToUpdate;
+        dstMatchState->loadedDictEnd= srcMatchState->loadedDictEnd;
+    }
+    dstCCtx->dictID = srcCCtx->dictID;
+
+    /* copy block state */
+    memcpy(dstCCtx->blockState.prevCBlock, srcCCtx->blockState.prevCBlock, sizeof(*srcCCtx->blockState.prevCBlock));
 
     return 0;
 }
@@ -1041,52 +1836,86 @@ size_t ZSTD_copyCCtx(ZSTD_CCtx* dstCCtx, const ZSTD_CCtx* srcCCtx, unsigned long
     ZSTD_frameParameters fParams = { 1 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
     ZSTD_buffered_policy_e const zbuff = (ZSTD_buffered_policy_e)(srcCCtx->inBuffSize>0);
     ZSTD_STATIC_ASSERT((U32)ZSTDb_buffered==1);
-    fParams.contentSizeFlag = pledgedSrcSize>0;
+    if (pledgedSrcSize==0) pledgedSrcSize = ZSTD_CONTENTSIZE_UNKNOWN;
+    fParams.contentSizeFlag = (pledgedSrcSize != ZSTD_CONTENTSIZE_UNKNOWN);
 
-    return ZSTD_copyCCtx_internal(dstCCtx, srcCCtx, fParams, pledgedSrcSize, zbuff);
+    return ZSTD_copyCCtx_internal(dstCCtx, srcCCtx,
+                                fParams, pledgedSrcSize,
+                                zbuff);
 }
 
 
+#define ZSTD_ROWSIZE 16
 /*! ZSTD_reduceTable() :
- *  reduce table indexes by `reducerValue` */
-static void ZSTD_reduceTable (U32* const table, U32 const size, U32 const reducerValue)
+ *  reduce table indexes by `reducerValue`, or squash to zero.
+ *  PreserveMark preserves "unsorted mark" for btlazy2 strategy.
+ *  It must be set to a clear 0/1 value, to remove branch during inlining.
+ *  Presume table size is a multiple of ZSTD_ROWSIZE
+ *  to help auto-vectorization */
+FORCE_INLINE_TEMPLATE void
+ZSTD_reduceTable_internal (U32* const table, U32 const size, U32 const reducerValue, int const preserveMark)
 {
-    U32 u;
-    for (u=0 ; u < size ; u++) {
-        if (table[u] < reducerValue) table[u] = 0;
-        else table[u] -= reducerValue;
-    }
+    int const nbRows = (int)size / ZSTD_ROWSIZE;
+    int cellNb = 0;
+    int rowNb;
+    assert((size & (ZSTD_ROWSIZE-1)) == 0);  /* multiple of ZSTD_ROWSIZE */
+    assert(size < (1U<<31));   /* can be casted to int */
+
+#if defined (MEMORY_SANITIZER) && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
+    /* To validate that the table re-use logic is sound, and that we don't
+     * access table space that we haven't cleaned, we re-"poison" the table
+     * space every time we mark it dirty.
+     *
+     * This function however is intended to operate on those dirty tables and
+     * re-clean them. So when this function is used correctly, we can unpoison
+     * the memory it operated on. This introduces a blind spot though, since
+     * if we now try to operate on __actually__ poisoned memory, we will not
+     * detect that. */
+    __msan_unpoison(table, size * sizeof(U32));
+#endif
+
+    for (rowNb=0 ; rowNb < nbRows ; rowNb++) {
+        int column;
+        for (column=0; column<ZSTD_ROWSIZE; column++) {
+            if (preserveMark) {
+                U32 const adder = (table[cellNb] == ZSTD_DUBT_UNSORTED_MARK) ? reducerValue : 0;
+                table[cellNb] += adder;
+            }
+            if (table[cellNb] < reducerValue) table[cellNb] = 0;
+            else table[cellNb] -= reducerValue;
+            cellNb++;
+    }   }
 }
 
-/*! ZSTD_ldm_reduceTable() :
- *  reduce table indexes by `reducerValue` */
-static void ZSTD_ldm_reduceTable(ldmEntry_t* const table, U32 const size,
-                                 U32 const reducerValue)
+static void ZSTD_reduceTable(U32* const table, U32 const size, U32 const reducerValue)
 {
-    U32 u;
-    for (u = 0; u < size; u++) {
-        if (table[u].offset < reducerValue) table[u].offset = 0;
-        else table[u].offset -= reducerValue;
-    }
+    ZSTD_reduceTable_internal(table, size, reducerValue, 0);
+}
+
+static void ZSTD_reduceTable_btlazy2(U32* const table, U32 const size, U32 const reducerValue)
+{
+    ZSTD_reduceTable_internal(table, size, reducerValue, 1);
 }
 
 /*! ZSTD_reduceIndex() :
 *   rescale all indexes to avoid future overflow (indexes are U32) */
-static void ZSTD_reduceIndex (ZSTD_CCtx* zc, const U32 reducerValue)
+static void ZSTD_reduceIndex (ZSTD_matchState_t* ms, ZSTD_CCtx_params const* params, const U32 reducerValue)
 {
-    { U32 const hSize = (U32)1 << zc->appliedParams.cParams.hashLog;
-      ZSTD_reduceTable(zc->hashTable, hSize, reducerValue); }
-
-    { U32 const chainSize = (zc->appliedParams.cParams.strategy == ZSTD_fast) ? 0 : ((U32)1 << zc->appliedParams.cParams.chainLog);
-      ZSTD_reduceTable(zc->chainTable, chainSize, reducerValue); }
+    {   U32 const hSize = (U32)1 << params->cParams.hashLog;
+        ZSTD_reduceTable(ms->hashTable, hSize, reducerValue);
+    }
 
-    { U32 const h3Size = (zc->hashLog3) ? (U32)1 << zc->hashLog3 : 0;
-      ZSTD_reduceTable(zc->hashTable3, h3Size, reducerValue); }
+    if (params->cParams.strategy != ZSTD_fast) {
+        U32 const chainSize = (U32)1 << params->cParams.chainLog;
+        if (params->cParams.strategy == ZSTD_btlazy2)
+            ZSTD_reduceTable_btlazy2(ms->chainTable, chainSize, reducerValue);
+        else
+            ZSTD_reduceTable(ms->chainTable, chainSize, reducerValue);
+    }
 
-    { if (zc->appliedParams.ldmParams.enableLdm) {
-          U32 const ldmHSize = (U32)1 << zc->appliedParams.ldmParams.hashLog;
-          ZSTD_ldm_reduceTable(zc->ldmState.hashTable, ldmHSize, reducerValue);
-      }
+    if (ms->hashLog3) {
+        U32 const h3Size = (U32)1 << ms->hashLog3;
+        ZSTD_reduceTable(ms->hashTable3, h3Size, reducerValue);
     }
 }
 
@@ -1097,152 +1926,21 @@ static void ZSTD_reduceIndex (ZSTD_CCtx* zc, const U32 reducerValue)
 
 /* See doc/zstd_compression_format.md for detailed format description */
 
-size_t ZSTD_noCompressBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
-{
-    if (srcSize + ZSTD_blockHeaderSize > dstCapacity) return ERROR(dstSize_tooSmall);
-    memcpy((BYTE*)dst + ZSTD_blockHeaderSize, src, srcSize);
-    MEM_writeLE24(dst, (U32)(srcSize << 2) + (U32)bt_raw);
-    return ZSTD_blockHeaderSize+srcSize;
-}
-
-
-static size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
-{
-    BYTE* const ostart = (BYTE* const)dst;
-    U32   const flSize = 1 + (srcSize>31) + (srcSize>4095);
-
-    if (srcSize + flSize > dstCapacity) return ERROR(dstSize_tooSmall);
-
-    switch(flSize)
-    {
-        case 1: /* 2 - 1 - 5 */
-            ostart[0] = (BYTE)((U32)set_basic + (srcSize<<3));
-            break;
-        case 2: /* 2 - 2 - 12 */
-            MEM_writeLE16(ostart, (U16)((U32)set_basic + (1<<2) + (srcSize<<4)));
-            break;
-        case 3: /* 2 - 2 - 20 */
-            MEM_writeLE32(ostart, (U32)((U32)set_basic + (3<<2) + (srcSize<<4)));
-            break;
-        default:   /* not necessary : flSize is {1,2,3} */
-            assert(0);
-    }
-
-    memcpy(ostart + flSize, src, srcSize);
-    return srcSize + flSize;
-}
-
-static size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
-{
-    BYTE* const ostart = (BYTE* const)dst;
-    U32   const flSize = 1 + (srcSize>31) + (srcSize>4095);
-
-    (void)dstCapacity;  /* dstCapacity already guaranteed to be >=4, hence large enough */
-
-    switch(flSize)
-    {
-        case 1: /* 2 - 1 - 5 */
-            ostart[0] = (BYTE)((U32)set_rle + (srcSize<<3));
-            break;
-        case 2: /* 2 - 2 - 12 */
-            MEM_writeLE16(ostart, (U16)((U32)set_rle + (1<<2) + (srcSize<<4)));
-            break;
-        case 3: /* 2 - 2 - 20 */
-            MEM_writeLE32(ostart, (U32)((U32)set_rle + (3<<2) + (srcSize<<4)));
-            break;
-        default:   /* not necessary : flSize is {1,2,3} */
-            assert(0);
-    }
-
-    ostart[flSize] = *(const BYTE*)src;
-    return flSize+1;
-}
-
-
-static size_t ZSTD_minGain(size_t srcSize) { return (srcSize >> 6) + 2; }
-
-static size_t ZSTD_compressLiterals (ZSTD_entropyCTables_t * entropy,
-                                     ZSTD_strategy strategy,
-                                     void* dst, size_t dstCapacity,
-                               const void* src, size_t srcSize)
-{
-    size_t const minGain = ZSTD_minGain(srcSize);
-    size_t const lhSize = 3 + (srcSize >= 1 KB) + (srcSize >= 16 KB);
-    BYTE*  const ostart = (BYTE*)dst;
-    U32 singleStream = srcSize < 256;
-    symbolEncodingType_e hType = set_compressed;
-    size_t cLitSize;
-
-
-    /* small ? don't even attempt compression (speed opt) */
-#   define LITERAL_NOENTROPY 63
-    {   size_t const minLitSize = entropy->hufCTable_repeatMode == HUF_repeat_valid ? 6 : LITERAL_NOENTROPY;
-        if (srcSize <= minLitSize) return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
-    }
-
-    if (dstCapacity < lhSize+1) return ERROR(dstSize_tooSmall);   /* not enough space for compression */
-    {   HUF_repeat repeat = entropy->hufCTable_repeatMode;
-        int const preferRepeat = strategy < ZSTD_lazy ? srcSize <= 1024 : 0;
-        if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1;
-        cLitSize = singleStream ? HUF_compress1X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
-                                      entropy->workspace, sizeof(entropy->workspace), (HUF_CElt*)entropy->hufCTable, &repeat, preferRepeat)
-                                : HUF_compress4X_repeat(ostart+lhSize, dstCapacity-lhSize, src, srcSize, 255, 11,
-                                      entropy->workspace, sizeof(entropy->workspace), (HUF_CElt*)entropy->hufCTable, &repeat, preferRepeat);
-        if (repeat != HUF_repeat_none) { hType = set_repeat; }    /* reused the existing table */
-        else { entropy->hufCTable_repeatMode = HUF_repeat_check; }       /* now have a table to reuse */
-    }
-
-    if ((cLitSize==0) | (cLitSize >= srcSize - minGain) | ERR_isError(cLitSize)) {
-        entropy->hufCTable_repeatMode = HUF_repeat_none;
-        return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
-    }
-    if (cLitSize==1) {
-        entropy->hufCTable_repeatMode = HUF_repeat_none;
-        return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
-    }
-
-    /* Build header */
-    switch(lhSize)
-    {
-    case 3: /* 2 - 2 - 10 - 10 */
-        {   U32 const lhc = hType + ((!singleStream) << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<14);
-            MEM_writeLE24(ostart, lhc);
-            break;
-        }
-    case 4: /* 2 - 2 - 14 - 14 */
-        {   U32 const lhc = hType + (2 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<18);
-            MEM_writeLE32(ostart, lhc);
-            break;
-        }
-    case 5: /* 2 - 2 - 18 - 18 */
-        {   U32 const lhc = hType + (3 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<22);
-            MEM_writeLE32(ostart, lhc);
-            ostart[4] = (BYTE)(cLitSize >> 10);
-            break;
-        }
-    default:   /* not possible : lhSize is {3,4,5} */
-        assert(0);
-    }
-    return lhSize+cLitSize;
-}
-
-
 void ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
 {
-    BYTE const LL_deltaCode = 19;
-    BYTE const ML_deltaCode = 36;
     const seqDef* const sequences = seqStorePtr->sequencesStart;
     BYTE* const llCodeTable = seqStorePtr->llCode;
     BYTE* const ofCodeTable = seqStorePtr->ofCode;
     BYTE* const mlCodeTable = seqStorePtr->mlCode;
     U32 const nbSeq = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
     U32 u;
+    assert(nbSeq <= seqStorePtr->maxNbSeq);
     for (u=0; u<nbSeq; u++) {
         U32 const llv = sequences[u].litLength;
         U32 const mlv = sequences[u].matchLength;
-        llCodeTable[u] = (llv> 63) ? (BYTE)ZSTD_highbit32(llv) + LL_deltaCode : LL_Code[llv];
+        llCodeTable[u] = (BYTE)ZSTD_LLcode(llv);
         ofCodeTable[u] = (BYTE)ZSTD_highbit32(sequences[u].offset);
-        mlCodeTable[u] = (mlv>127) ? (BYTE)ZSTD_highbit32(mlv) + ML_deltaCode : ML_Code[mlv];
+        mlCodeTable[u] = (BYTE)ZSTD_MLcode(mlv);
     }
     if (seqStorePtr->longLengthID==1)
         llCodeTable[seqStorePtr->longLengthPos] = MaxLL;
@@ -1250,154 +1948,33 @@ void ZSTD_seqToCodes(const seqStore_t* seqStorePtr)
         mlCodeTable[seqStorePtr->longLengthPos] = MaxML;
 }
 
-MEM_STATIC symbolEncodingType_e ZSTD_selectEncodingType(FSE_repeat* repeatMode,
-        size_t const mostFrequent, size_t nbSeq, U32 defaultNormLog)
-{
-#define MIN_SEQ_FOR_DYNAMIC_FSE   64
-#define MAX_SEQ_FOR_STATIC_FSE  1000
-
-    if ((mostFrequent == nbSeq) && (nbSeq > 2)) {
-        *repeatMode = FSE_repeat_check;
-        return set_rle;
-    }
-    if ((*repeatMode == FSE_repeat_valid) && (nbSeq < MAX_SEQ_FOR_STATIC_FSE)) {
-        return set_repeat;
-    }
-    if ((nbSeq < MIN_SEQ_FOR_DYNAMIC_FSE) || (mostFrequent < (nbSeq >> (defaultNormLog-1)))) {
-        *repeatMode = FSE_repeat_valid;
-        return set_basic;
-    }
-    *repeatMode = FSE_repeat_check;
-    return set_compressed;
-}
-
-MEM_STATIC size_t ZSTD_buildCTable(void* dst, size_t dstCapacity,
-        FSE_CTable* CTable, U32 FSELog, symbolEncodingType_e type,
-        U32* count, U32 max,
-        BYTE const* codeTable, size_t nbSeq,
-        S16 const* defaultNorm, U32 defaultNormLog, U32 defaultMax,
-        void* workspace, size_t workspaceSize)
+/* ZSTD_useTargetCBlockSize():
+ * Returns if target compressed block size param is being used.
+ * If used, compression will do best effort to make a compressed block size to be around targetCBlockSize.
+ * Returns 1 if true, 0 otherwise. */
+static int ZSTD_useTargetCBlockSize(const ZSTD_CCtx_params* cctxParams)
 {
-    BYTE* op = (BYTE*)dst;
-    BYTE const* const oend = op + dstCapacity;
-
-    switch (type) {
-    case set_rle:
-        *op = codeTable[0];
-        CHECK_F(FSE_buildCTable_rle(CTable, (BYTE)max));
-        return 1;
-    case set_repeat:
-        return 0;
-    case set_basic:
-        CHECK_F(FSE_buildCTable_wksp(CTable, defaultNorm, defaultMax, defaultNormLog, workspace, workspaceSize));
-        return 0;
-    case set_compressed: {
-        S16 norm[MaxSeq + 1];
-        size_t nbSeq_1 = nbSeq;
-        const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
-        if (count[codeTable[nbSeq-1]] > 1) {
-            count[codeTable[nbSeq-1]]--;
-            nbSeq_1--;
-        }
-        CHECK_F(FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max));
-        {   size_t const NCountSize = FSE_writeNCount(op, oend - op, norm, max, tableLog);   /* overflow protected */
-            if (FSE_isError(NCountSize)) return NCountSize;
-            CHECK_F(FSE_buildCTable_wksp(CTable, norm, max, tableLog, workspace, workspaceSize));
-            return NCountSize;
-        }
-    }
-    default: return assert(0), ERROR(GENERIC);
-    }
-}
-
-MEM_STATIC size_t ZSTD_encodeSequences(void* dst, size_t dstCapacity,
-    FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
-    FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
-    FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
-    seqDef const* sequences, size_t nbSeq, int longOffsets)
-{
-    BIT_CStream_t blockStream;
-    FSE_CState_t  stateMatchLength;
-    FSE_CState_t  stateOffsetBits;
-    FSE_CState_t  stateLitLength;
-
-    CHECK_E(BIT_initCStream(&blockStream, dst, dstCapacity), dstSize_tooSmall); /* not enough space remaining */
-
-    /* first symbols */
-    FSE_initCState2(&stateMatchLength, CTable_MatchLength, mlCodeTable[nbSeq-1]);
-    FSE_initCState2(&stateOffsetBits,  CTable_OffsetBits,  ofCodeTable[nbSeq-1]);
-    FSE_initCState2(&stateLitLength,   CTable_LitLength,   llCodeTable[nbSeq-1]);
-    BIT_addBits(&blockStream, sequences[nbSeq-1].litLength, LL_bits[llCodeTable[nbSeq-1]]);
-    if (MEM_32bits()) BIT_flushBits(&blockStream);
-    BIT_addBits(&blockStream, sequences[nbSeq-1].matchLength, ML_bits[mlCodeTable[nbSeq-1]]);
-    if (MEM_32bits()) BIT_flushBits(&blockStream);
-    if (longOffsets) {
-        U32 const ofBits = ofCodeTable[nbSeq-1];
-        int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
-        if (extraBits) {
-            BIT_addBits(&blockStream, sequences[nbSeq-1].offset, extraBits);
-            BIT_flushBits(&blockStream);
-        }
-        BIT_addBits(&blockStream, sequences[nbSeq-1].offset >> extraBits,
-                    ofBits - extraBits);
-    } else {
-        BIT_addBits(&blockStream, sequences[nbSeq-1].offset, ofCodeTable[nbSeq-1]);
-    }
-    BIT_flushBits(&blockStream);
-
-    {   size_t n;
-        for (n=nbSeq-2 ; n<nbSeq ; n--) {      /* intentional underflow */
-            BYTE const llCode = llCodeTable[n];
-            BYTE const ofCode = ofCodeTable[n];
-            BYTE const mlCode = mlCodeTable[n];
-            U32  const llBits = LL_bits[llCode];
-            U32  const ofBits = ofCode;                                     /* 32b*/  /* 64b*/
-            U32  const mlBits = ML_bits[mlCode];
-                                                                            /* (7)*/  /* (7)*/
-            FSE_encodeSymbol(&blockStream, &stateOffsetBits, ofCode);       /* 15 */  /* 15 */
-            FSE_encodeSymbol(&blockStream, &stateMatchLength, mlCode);      /* 24 */  /* 24 */
-            if (MEM_32bits()) BIT_flushBits(&blockStream);                  /* (7)*/
-            FSE_encodeSymbol(&blockStream, &stateLitLength, llCode);        /* 16 */  /* 33 */
-            if (MEM_32bits() || (ofBits+mlBits+llBits >= 64-7-(LLFSELog+MLFSELog+OffFSELog)))
-                BIT_flushBits(&blockStream);                                /* (7)*/
-            BIT_addBits(&blockStream, sequences[n].litLength, llBits);
-            if (MEM_32bits() && ((llBits+mlBits)>24)) BIT_flushBits(&blockStream);
-            BIT_addBits(&blockStream, sequences[n].matchLength, mlBits);
-            if (MEM_32bits() || (ofBits+mlBits+llBits > 56)) BIT_flushBits(&blockStream);
-            if (longOffsets) {
-                int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
-                if (extraBits) {
-                    BIT_addBits(&blockStream, sequences[n].offset, extraBits);
-                    BIT_flushBits(&blockStream);                            /* (7)*/
-                }
-                BIT_addBits(&blockStream, sequences[n].offset >> extraBits,
-                            ofBits - extraBits);                            /* 31 */
-            } else {
-                BIT_addBits(&blockStream, sequences[n].offset, ofBits);     /* 31 */
-            }
-            BIT_flushBits(&blockStream);                                    /* (7)*/
-    }   }
-
-    FSE_flushCState(&blockStream, &stateMatchLength);
-    FSE_flushCState(&blockStream, &stateOffsetBits);
-    FSE_flushCState(&blockStream, &stateLitLength);
-
-    {   size_t const streamSize = BIT_closeCStream(&blockStream);
-        if (streamSize==0) return ERROR(dstSize_tooSmall);   /* not enough space */
-        return streamSize;
-    }
+    DEBUGLOG(5, "ZSTD_useTargetCBlockSize (targetCBlockSize=%zu)", cctxParams->targetCBlockSize);
+    return (cctxParams->targetCBlockSize != 0);
 }
 
-MEM_STATIC size_t ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
-                              ZSTD_entropyCTables_t* entropy,
-                              ZSTD_compressionParameters const* cParams,
-                              void* dst, size_t dstCapacity)
+/* ZSTD_compressSequences_internal():
+ * actually compresses both literals and sequences */
+MEM_STATIC size_t
+ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
+                          const ZSTD_entropyCTables_t* prevEntropy,
+                                ZSTD_entropyCTables_t* nextEntropy,
+                          const ZSTD_CCtx_params* cctxParams,
+                                void* dst, size_t dstCapacity,
+                                void* entropyWorkspace, size_t entropyWkspSize,
+                          const int bmi2)
 {
-    const int longOffsets = cParams->windowLog > STREAM_ACCUMULATOR_MIN;
-    U32 count[MaxSeq+1];
-    FSE_CTable* CTable_LitLength = entropy->litlengthCTable;
-    FSE_CTable* CTable_OffsetBits = entropy->offcodeCTable;
-    FSE_CTable* CTable_MatchLength = entropy->matchlengthCTable;
+    const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
+    ZSTD_strategy const strategy = cctxParams->cParams.strategy;
+    unsigned count[MaxSeq+1];
+    FSE_CTable* CTable_LitLength = nextEntropy->fse.litlengthCTable;
+    FSE_CTable* CTable_OffsetBits = nextEntropy->fse.offcodeCTable;
+    FSE_CTable* CTable_MatchLength = nextEntropy->fse.matchlengthCTable;
     U32 LLtype, Offtype, MLtype;   /* compressed, raw or rle */
     const seqDef* const sequences = seqStorePtr->sequencesStart;
     const BYTE* const ofCodeTable = seqStorePtr->ofCode;
@@ -1406,132 +1983,244 @@ MEM_STATIC size_t ZSTD_compressSequences_internal(seqStore_t* seqStorePtr,
     BYTE* const ostart = (BYTE*)dst;
     BYTE* const oend = ostart + dstCapacity;
     BYTE* op = ostart;
-    size_t const nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
+    size_t const nbSeq = (size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
     BYTE* seqHead;
+    BYTE* lastNCount = NULL;
 
-    ZSTD_STATIC_ASSERT(sizeof(entropy->workspace) >= (1<<MAX(MLFSELog,LLFSELog)));
+    DEBUGLOG(5, "ZSTD_compressSequences_internal (nbSeq=%zu)", nbSeq);
+    ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
 
     /* Compress literals */
     {   const BYTE* const literals = seqStorePtr->litStart;
-        size_t const litSize = seqStorePtr->lit - literals;
+        size_t const litSize = (size_t)(seqStorePtr->lit - literals);
         size_t const cSize = ZSTD_compressLiterals(
-                entropy, cParams->strategy, op, dstCapacity, literals, litSize);
-        if (ZSTD_isError(cSize))
-          return cSize;
+                                    &prevEntropy->huf, &nextEntropy->huf,
+                                    cctxParams->cParams.strategy,
+                                    ZSTD_disableLiteralsCompression(cctxParams),
+                                    op, dstCapacity,
+                                    literals, litSize,
+                                    entropyWorkspace, entropyWkspSize,
+                                    bmi2);
+        FORWARD_IF_ERROR(cSize, "ZSTD_compressLiterals failed");
+        assert(cSize <= dstCapacity);
         op += cSize;
     }
 
     /* Sequences Header */
-    if ((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead */) return ERROR(dstSize_tooSmall);
-    if (nbSeq < 0x7F) *op++ = (BYTE)nbSeq;
-    else if (nbSeq < LONGNBSEQ) op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
-    else op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
-    if (nbSeq==0) return op - ostart;
+    RETURN_ERROR_IF((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead*/,
+                    dstSize_tooSmall, "Can't fit seq hdr in output buf!");
+    if (nbSeq < 128) {
+        *op++ = (BYTE)nbSeq;
+    } else if (nbSeq < LONGNBSEQ) {
+        op[0] = (BYTE)((nbSeq>>8) + 0x80);
+        op[1] = (BYTE)nbSeq;
+        op+=2;
+    } else {
+        op[0]=0xFF;
+        MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ));
+        op+=3;
+    }
+    assert(op <= oend);
+    if (nbSeq==0) {
+        /* Copy the old tables over as if we repeated them */
+        memcpy(&nextEntropy->fse, &prevEntropy->fse, sizeof(prevEntropy->fse));
+        return (size_t)(op - ostart);
+    }
 
     /* seqHead : flags for FSE encoding type */
     seqHead = op++;
+    assert(op <= oend);
 
     /* convert length/distances into codes */
     ZSTD_seqToCodes(seqStorePtr);
-    /* CTable for Literal Lengths */
-    {   U32 max = MaxLL;
-        size_t const mostFrequent = FSE_countFast_wksp(count, &max, llCodeTable, nbSeq, entropy->workspace);
-        LLtype = ZSTD_selectEncodingType(&entropy->litlength_repeatMode, mostFrequent, nbSeq, LL_defaultNormLog);
-        {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_LitLength, LLFSELog, (symbolEncodingType_e)LLtype,
-                    count, max, llCodeTable, nbSeq, LL_defaultNorm, LL_defaultNormLog, MaxLL,
-                    entropy->workspace, sizeof(entropy->workspace));
-            if (ZSTD_isError(countSize)) return countSize;
+    /* build CTable for Literal Lengths */
+    {   unsigned max = MaxLL;
+        size_t const mostFrequent = HIST_countFast_wksp(count, &max, llCodeTable, nbSeq, entropyWorkspace, entropyWkspSize);   /* can't fail */
+        DEBUGLOG(5, "Building LL table");
+        nextEntropy->fse.litlength_repeatMode = prevEntropy->fse.litlength_repeatMode;
+        LLtype = ZSTD_selectEncodingType(&nextEntropy->fse.litlength_repeatMode,
+                                        count, max, mostFrequent, nbSeq,
+                                        LLFSELog, prevEntropy->fse.litlengthCTable,
+                                        LL_defaultNorm, LL_defaultNormLog,
+                                        ZSTD_defaultAllowed, strategy);
+        assert(set_basic < set_compressed && set_rle < set_compressed);
+        assert(!(LLtype < set_compressed && nextEntropy->fse.litlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */
+        {   size_t const countSize = ZSTD_buildCTable(
+                op, (size_t)(oend - op),
+                CTable_LitLength, LLFSELog, (symbolEncodingType_e)LLtype,
+                count, max, llCodeTable, nbSeq,
+                LL_defaultNorm, LL_defaultNormLog, MaxLL,
+                prevEntropy->fse.litlengthCTable,
+                sizeof(prevEntropy->fse.litlengthCTable),
+                entropyWorkspace, entropyWkspSize);
+            FORWARD_IF_ERROR(countSize, "ZSTD_buildCTable for LitLens failed");
+            if (LLtype == set_compressed)
+                lastNCount = op;
             op += countSize;
+            assert(op <= oend);
     }   }
-    /* CTable for Offsets */
-    {   U32 max = MaxOff;
-        size_t const mostFrequent = FSE_countFast_wksp(count, &max, ofCodeTable, nbSeq, entropy->workspace);
-        Offtype = ZSTD_selectEncodingType(&entropy->offcode_repeatMode, mostFrequent, nbSeq, OF_defaultNormLog);
-        {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_OffsetBits, OffFSELog, (symbolEncodingType_e)Offtype,
-                    count, max, ofCodeTable, nbSeq, OF_defaultNorm, OF_defaultNormLog, MaxOff,
-                    entropy->workspace, sizeof(entropy->workspace));
-            if (ZSTD_isError(countSize)) return countSize;
+    /* build CTable for Offsets */
+    {   unsigned max = MaxOff;
+        size_t const mostFrequent = HIST_countFast_wksp(
+            count, &max, ofCodeTable, nbSeq, entropyWorkspace, entropyWkspSize);  /* can't fail */
+        /* We can only use the basic table if max <= DefaultMaxOff, otherwise the offsets are too large */
+        ZSTD_defaultPolicy_e const defaultPolicy = (max <= DefaultMaxOff) ? ZSTD_defaultAllowed : ZSTD_defaultDisallowed;
+        DEBUGLOG(5, "Building OF table");
+        nextEntropy->fse.offcode_repeatMode = prevEntropy->fse.offcode_repeatMode;
+        Offtype = ZSTD_selectEncodingType(&nextEntropy->fse.offcode_repeatMode,
+                                        count, max, mostFrequent, nbSeq,
+                                        OffFSELog, prevEntropy->fse.offcodeCTable,
+                                        OF_defaultNorm, OF_defaultNormLog,
+                                        defaultPolicy, strategy);
+        assert(!(Offtype < set_compressed && nextEntropy->fse.offcode_repeatMode != FSE_repeat_none)); /* We don't copy tables */
+        {   size_t const countSize = ZSTD_buildCTable(
+                op, (size_t)(oend - op),
+                CTable_OffsetBits, OffFSELog, (symbolEncodingType_e)Offtype,
+                count, max, ofCodeTable, nbSeq,
+                OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
+                prevEntropy->fse.offcodeCTable,
+                sizeof(prevEntropy->fse.offcodeCTable),
+                entropyWorkspace, entropyWkspSize);
+            FORWARD_IF_ERROR(countSize, "ZSTD_buildCTable for Offsets failed");
+            if (Offtype == set_compressed)
+                lastNCount = op;
             op += countSize;
+            assert(op <= oend);
     }   }
-    /* CTable for MatchLengths */
-    {   U32 max = MaxML;
-        size_t const mostFrequent = FSE_countFast_wksp(count, &max, mlCodeTable, nbSeq, entropy->workspace);
-        MLtype = ZSTD_selectEncodingType(&entropy->matchlength_repeatMode, mostFrequent, nbSeq, ML_defaultNormLog);
-        {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_MatchLength, MLFSELog, (symbolEncodingType_e)MLtype,
-                    count, max, mlCodeTable, nbSeq, ML_defaultNorm, ML_defaultNormLog, MaxML,
-                    entropy->workspace, sizeof(entropy->workspace));
-            if (ZSTD_isError(countSize)) return countSize;
+    /* build CTable for MatchLengths */
+    {   unsigned max = MaxML;
+        size_t const mostFrequent = HIST_countFast_wksp(
+            count, &max, mlCodeTable, nbSeq, entropyWorkspace, entropyWkspSize);   /* can't fail */
+        DEBUGLOG(5, "Building ML table (remaining space : %i)", (int)(oend-op));
+        nextEntropy->fse.matchlength_repeatMode = prevEntropy->fse.matchlength_repeatMode;
+        MLtype = ZSTD_selectEncodingType(&nextEntropy->fse.matchlength_repeatMode,
+                                        count, max, mostFrequent, nbSeq,
+                                        MLFSELog, prevEntropy->fse.matchlengthCTable,
+                                        ML_defaultNorm, ML_defaultNormLog,
+                                        ZSTD_defaultAllowed, strategy);
+        assert(!(MLtype < set_compressed && nextEntropy->fse.matchlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */
+        {   size_t const countSize = ZSTD_buildCTable(
+                op, (size_t)(oend - op),
+                CTable_MatchLength, MLFSELog, (symbolEncodingType_e)MLtype,
+                count, max, mlCodeTable, nbSeq,
+                ML_defaultNorm, ML_defaultNormLog, MaxML,
+                prevEntropy->fse.matchlengthCTable,
+                sizeof(prevEntropy->fse.matchlengthCTable),
+                entropyWorkspace, entropyWkspSize);
+            FORWARD_IF_ERROR(countSize, "ZSTD_buildCTable for MatchLengths failed");
+            if (MLtype == set_compressed)
+                lastNCount = op;
             op += countSize;
+            assert(op <= oend);
     }   }
 
     *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
 
-    {   size_t const streamSize = ZSTD_encodeSequences(op, oend - op,
-                CTable_MatchLength, mlCodeTable,
-                CTable_OffsetBits, ofCodeTable,
-                CTable_LitLength, llCodeTable,
-                sequences, nbSeq, longOffsets);
-        if (ZSTD_isError(streamSize)) return streamSize;
-        op += streamSize;
+    {   size_t const bitstreamSize = ZSTD_encodeSequences(
+                                        op, (size_t)(oend - op),
+                                        CTable_MatchLength, mlCodeTable,
+                                        CTable_OffsetBits, ofCodeTable,
+                                        CTable_LitLength, llCodeTable,
+                                        sequences, nbSeq,
+                                        longOffsets, bmi2);
+        FORWARD_IF_ERROR(bitstreamSize, "ZSTD_encodeSequences failed");
+        op += bitstreamSize;
+        assert(op <= oend);
+        /* zstd versions <= 1.3.4 mistakenly report corruption when
+         * FSE_readNCount() receives a buffer < 4 bytes.
+         * Fixed by https://github.com/facebook/zstd/pull/1146.
+         * This can happen when the last set_compressed table present is 2
+         * bytes and the bitstream is only one byte.
+         * In this exceedingly rare case, we will simply emit an uncompressed
+         * block, since it isn't worth optimizing.
+         */
+        if (lastNCount && (op - lastNCount) < 4) {
+            /* NCountSize >= 2 && bitstreamSize > 0 ==> lastCountSize == 3 */
+            assert(op - lastNCount == 3);
+            DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.3.4 by "
+                        "emitting an uncompressed block.");
+            return 0;
+        }
     }
 
-    return op - ostart;
+    DEBUGLOG(5, "compressed block size : %u", (unsigned)(op - ostart));
+    return (size_t)(op - ostart);
 }
 
-MEM_STATIC size_t ZSTD_compressSequences(seqStore_t* seqStorePtr,
-                              ZSTD_entropyCTables_t* entropy,
-                              ZSTD_compressionParameters const* cParams,
-                              void* dst, size_t dstCapacity,
-                              size_t srcSize)
-{
-    size_t const cSize = ZSTD_compressSequences_internal(seqStorePtr, entropy, cParams,
-                                                         dst, dstCapacity);
-    size_t const minGain = ZSTD_minGain(srcSize);
-    size_t const maxCSize = srcSize - minGain;
-    /* If the srcSize <= dstCapacity, then there is enough space to write a
-     * raw uncompressed block. Since we ran out of space, the block must not
-     * be compressible, so fall back to a raw uncompressed block.
+MEM_STATIC size_t
+ZSTD_compressSequences(seqStore_t* seqStorePtr,
+                       const ZSTD_entropyCTables_t* prevEntropy,
+                             ZSTD_entropyCTables_t* nextEntropy,
+                       const ZSTD_CCtx_params* cctxParams,
+                             void* dst, size_t dstCapacity,
+                             size_t srcSize,
+                             void* entropyWorkspace, size_t entropyWkspSize,
+                             int bmi2)
+{
+    size_t const cSize = ZSTD_compressSequences_internal(
+                            seqStorePtr, prevEntropy, nextEntropy, cctxParams,
+                            dst, dstCapacity,
+                            entropyWorkspace, entropyWkspSize, bmi2);
+    if (cSize == 0) return 0;
+    /* When srcSize <= dstCapacity, there is enough space to write a raw uncompressed block.
+     * Since we ran out of space, block must be not compressible, so fall back to raw uncompressed block.
      */
-    int const uncompressibleError = cSize == ERROR(dstSize_tooSmall) && srcSize <= dstCapacity;
+    if ((cSize == ERROR(dstSize_tooSmall)) & (srcSize <= dstCapacity))
+        return 0;  /* block not compressed */
+    FORWARD_IF_ERROR(cSize, "ZSTD_compressSequences_internal failed");
 
-    if (ZSTD_isError(cSize) && !uncompressibleError)
-        return cSize;
     /* Check compressibility */
-    if (cSize >= maxCSize || uncompressibleError) {
-        entropy->hufCTable_repeatMode = HUF_repeat_none;
-        entropy->offcode_repeatMode = FSE_repeat_none;
-        entropy->matchlength_repeatMode = FSE_repeat_none;
-        entropy->litlength_repeatMode = FSE_repeat_none;
-        return 0;
+    {   size_t const maxCSize = srcSize - ZSTD_minGain(srcSize, cctxParams->cParams.strategy);
+        if (cSize >= maxCSize) return 0;  /* block not compressed */
     }
-    assert(!ZSTD_isError(cSize));
 
-    /* confirm repcodes */
-    { int i; for (i=0; i<ZSTD_REP_NUM; i++) seqStorePtr->rep[i] = seqStorePtr->repToConfirm[i]; }
     return cSize;
 }
 
 /* ZSTD_selectBlockCompressor() :
  * Not static, but internal use only (used by long distance matcher)
  * assumption : strat is a valid strategy */
-typedef size_t (*ZSTD_blockCompressor) (ZSTD_CCtx* ctx, const void* src, size_t srcSize);
-ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, int extDict)
+ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_dictMode_e dictMode)
 {
-    static const ZSTD_blockCompressor blockCompressor[2][(unsigned)ZSTD_btultra+1] = {
+    static const ZSTD_blockCompressor blockCompressor[3][ZSTD_STRATEGY_MAX+1] = {
         { ZSTD_compressBlock_fast  /* default for 0 */,
-          ZSTD_compressBlock_fast, ZSTD_compressBlock_doubleFast, ZSTD_compressBlock_greedy,
-          ZSTD_compressBlock_lazy, ZSTD_compressBlock_lazy2, ZSTD_compressBlock_btlazy2,
-          ZSTD_compressBlock_btopt, ZSTD_compressBlock_btultra },
+          ZSTD_compressBlock_fast,
+          ZSTD_compressBlock_doubleFast,
+          ZSTD_compressBlock_greedy,
+          ZSTD_compressBlock_lazy,
+          ZSTD_compressBlock_lazy2,
+          ZSTD_compressBlock_btlazy2,
+          ZSTD_compressBlock_btopt,
+          ZSTD_compressBlock_btultra,
+          ZSTD_compressBlock_btultra2 },
         { ZSTD_compressBlock_fast_extDict  /* default for 0 */,
-          ZSTD_compressBlock_fast_extDict, ZSTD_compressBlock_doubleFast_extDict, ZSTD_compressBlock_greedy_extDict,
-          ZSTD_compressBlock_lazy_extDict,ZSTD_compressBlock_lazy2_extDict, ZSTD_compressBlock_btlazy2_extDict,
-          ZSTD_compressBlock_btopt_extDict, ZSTD_compressBlock_btultra_extDict }
+          ZSTD_compressBlock_fast_extDict,
+          ZSTD_compressBlock_doubleFast_extDict,
+          ZSTD_compressBlock_greedy_extDict,
+          ZSTD_compressBlock_lazy_extDict,
+          ZSTD_compressBlock_lazy2_extDict,
+          ZSTD_compressBlock_btlazy2_extDict,
+          ZSTD_compressBlock_btopt_extDict,
+          ZSTD_compressBlock_btultra_extDict,
+          ZSTD_compressBlock_btultra_extDict },
+        { ZSTD_compressBlock_fast_dictMatchState  /* default for 0 */,
+          ZSTD_compressBlock_fast_dictMatchState,
+          ZSTD_compressBlock_doubleFast_dictMatchState,
+          ZSTD_compressBlock_greedy_dictMatchState,
+          ZSTD_compressBlock_lazy_dictMatchState,
+          ZSTD_compressBlock_lazy2_dictMatchState,
+          ZSTD_compressBlock_btlazy2_dictMatchState,
+          ZSTD_compressBlock_btopt_dictMatchState,
+          ZSTD_compressBlock_btultra_dictMatchState,
+          ZSTD_compressBlock_btultra_dictMatchState }
     };
+    ZSTD_blockCompressor selectedCompressor;
     ZSTD_STATIC_ASSERT((unsigned)ZSTD_fast == 1);
-    assert((U32)strat >= (U32)ZSTD_fast);
-    assert((U32)strat <= (U32)ZSTD_btultra);
 
-    return blockCompressor[extDict!=0][(U32)strat];
+    assert(ZSTD_cParam_withinBounds(ZSTD_c_strategy, strat));
+    selectedCompressor = blockCompressor[(int)dictMode][(int)strat];
+    assert(selectedCompressor != NULL);
+    return selectedCompressor;
 }
 
 static void ZSTD_storeLastLiterals(seqStore_t* seqStorePtr,
@@ -1541,33 +2230,347 @@ static void ZSTD_storeLastLiterals(seqStore_t* seqStorePtr,
     seqStorePtr->lit += lastLLSize;
 }
 
-static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+void ZSTD_resetSeqStore(seqStore_t* ssPtr)
 {
-    const BYTE* const base = zc->base;
-    const BYTE* const istart = (const BYTE*)src;
-    const U32 current = (U32)(istart-base);
-    size_t lastLLSize;
-    const BYTE* anchor;
-    U32 const extDict = zc->lowLimit < zc->dictLimit;
-    const ZSTD_blockCompressor blockCompressor =
-        zc->appliedParams.ldmParams.enableLdm
-            ? (extDict ? ZSTD_compressBlock_ldm_extDict : ZSTD_compressBlock_ldm)
-            : ZSTD_selectBlockCompressor(zc->appliedParams.cParams.strategy, extDict);
+    ssPtr->lit = ssPtr->litStart;
+    ssPtr->sequences = ssPtr->sequencesStart;
+    ssPtr->longLengthID = 0;
+}
+
+typedef enum { ZSTDbss_compress, ZSTDbss_noCompress } ZSTD_buildSeqStore_e;
 
-    if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1) return 0;   /* don't even attempt compression below a certain srcSize */
+static size_t ZSTD_buildSeqStore(ZSTD_CCtx* zc, const void* src, size_t srcSize)
+{
+    ZSTD_matchState_t* const ms = &zc->blockState.matchState;
+    DEBUGLOG(5, "ZSTD_buildSeqStore (srcSize=%zu)", srcSize);
+    assert(srcSize <= ZSTD_BLOCKSIZE_MAX);
+    /* Assert that we have correctly flushed the ctx params into the ms's copy */
+    ZSTD_assertEqualCParams(zc->appliedParams.cParams, ms->cParams);
+    if (srcSize < MIN_CBLOCK_SIZE+ZSTD_blockHeaderSize+1) {
+        ZSTD_ldm_skipSequences(&zc->externSeqStore, srcSize, zc->appliedParams.cParams.minMatch);
+        return ZSTDbss_noCompress; /* don't even attempt compression below a certain srcSize */
+    }
     ZSTD_resetSeqStore(&(zc->seqStore));
-    if (current > zc->nextToUpdate + 384)
-        zc->nextToUpdate = current - MIN(192, (U32)(current - zc->nextToUpdate - 384));   /* limited update after finding a very long match */
+    /* required for optimal parser to read stats from dictionary */
+    ms->opt.symbolCosts = &zc->blockState.prevCBlock->entropy;
+    /* tell the optimal parser how we expect to compress literals */
+    ms->opt.literalCompressionMode = zc->appliedParams.literalCompressionMode;
+    /* a gap between an attached dict and the current window is not safe,
+     * they must remain adjacent,
+     * and when that stops being the case, the dict must be unset */
+    assert(ms->dictMatchState == NULL || ms->loadedDictEnd == ms->window.dictLimit);
+
+    /* limited update after a very long match */
+    {   const BYTE* const base = ms->window.base;
+        const BYTE* const istart = (const BYTE*)src;
+        const U32 current = (U32)(istart-base);
+        if (sizeof(ptrdiff_t)==8) assert(istart - base < (ptrdiff_t)(U32)(-1));   /* ensure no overflow */
+        if (current > ms->nextToUpdate + 384)
+            ms->nextToUpdate = current - MIN(192, (U32)(current - ms->nextToUpdate - 384));
+    }
+
+    /* select and store sequences */
+    {   ZSTD_dictMode_e const dictMode = ZSTD_matchState_dictMode(ms);
+        size_t lastLLSize;
+        {   int i;
+            for (i = 0; i < ZSTD_REP_NUM; ++i)
+                zc->blockState.nextCBlock->rep[i] = zc->blockState.prevCBlock->rep[i];
+        }
+        if (zc->externSeqStore.pos < zc->externSeqStore.size) {
+            assert(!zc->appliedParams.ldmParams.enableLdm);
+            /* Updates ldmSeqStore.pos */
+            lastLLSize =
+                ZSTD_ldm_blockCompress(&zc->externSeqStore,
+                                       ms, &zc->seqStore,
+                                       zc->blockState.nextCBlock->rep,
+                                       src, srcSize);
+            assert(zc->externSeqStore.pos <= zc->externSeqStore.size);
+        } else if (zc->appliedParams.ldmParams.enableLdm) {
+            rawSeqStore_t ldmSeqStore = {NULL, 0, 0, 0};
+
+            ldmSeqStore.seq = zc->ldmSequences;
+            ldmSeqStore.capacity = zc->maxNbLdmSequences;
+            /* Updates ldmSeqStore.size */
+            FORWARD_IF_ERROR(ZSTD_ldm_generateSequences(&zc->ldmState, &ldmSeqStore,
+                                               &zc->appliedParams.ldmParams,
+                                               src, srcSize), "");
+            /* Updates ldmSeqStore.pos */
+            lastLLSize =
+                ZSTD_ldm_blockCompress(&ldmSeqStore,
+                                       ms, &zc->seqStore,
+                                       zc->blockState.nextCBlock->rep,
+                                       src, srcSize);
+            assert(ldmSeqStore.pos == ldmSeqStore.size);
+        } else {   /* not long range mode */
+            ZSTD_blockCompressor const blockCompressor = ZSTD_selectBlockCompressor(zc->appliedParams.cParams.strategy, dictMode);
+            lastLLSize = blockCompressor(ms, &zc->seqStore, zc->blockState.nextCBlock->rep, src, srcSize);
+        }
+        {   const BYTE* const lastLiterals = (const BYTE*)src + srcSize - lastLLSize;
+            ZSTD_storeLastLiterals(&zc->seqStore, lastLiterals, lastLLSize);
+    }   }
+    return ZSTDbss_compress;
+}
+
+static void ZSTD_copyBlockSequences(ZSTD_CCtx* zc)
+{
+    const seqStore_t* seqStore = ZSTD_getSeqStore(zc);
+    const seqDef* seqs = seqStore->sequencesStart;
+    size_t seqsSize = seqStore->sequences - seqs;
+
+    ZSTD_Sequence* outSeqs = &zc->seqCollector.seqStart[zc->seqCollector.seqIndex];
+    size_t i; size_t position; int repIdx;
+
+    assert(zc->seqCollector.seqIndex + 1 < zc->seqCollector.maxSequences);
+    for (i = 0, position = 0; i < seqsSize; ++i) {
+        outSeqs[i].offset = seqs[i].offset;
+        outSeqs[i].litLength = seqs[i].litLength;
+        outSeqs[i].matchLength = seqs[i].matchLength + MINMATCH;
+
+        if (i == seqStore->longLengthPos) {
+            if (seqStore->longLengthID == 1) {
+                outSeqs[i].litLength += 0x10000;
+            } else if (seqStore->longLengthID == 2) {
+                outSeqs[i].matchLength += 0x10000;
+            }
+        }
+
+        if (outSeqs[i].offset <= ZSTD_REP_NUM) {
+            outSeqs[i].rep = outSeqs[i].offset;
+            repIdx = (unsigned int)i - outSeqs[i].offset;
+
+            if (outSeqs[i].litLength == 0) {
+                if (outSeqs[i].offset < 3) {
+                    --repIdx;
+                } else {
+                    repIdx = (unsigned int)i - 1;
+                }
+                ++outSeqs[i].rep;
+            }
+            assert(repIdx >= -3);
+            outSeqs[i].offset = repIdx >= 0 ? outSeqs[repIdx].offset : repStartValue[-repIdx - 1];
+            if (outSeqs[i].rep == 4) {
+                --outSeqs[i].offset;
+            }
+        } else {
+            outSeqs[i].offset -= ZSTD_REP_NUM;
+        }
+
+        position += outSeqs[i].litLength;
+        outSeqs[i].matchPos = (unsigned int)position;
+        position += outSeqs[i].matchLength;
+    }
+    zc->seqCollector.seqIndex += seqsSize;
+}
+
+size_t ZSTD_getSequences(ZSTD_CCtx* zc, ZSTD_Sequence* outSeqs,
+    size_t outSeqsSize, const void* src, size_t srcSize)
+{
+    const size_t dstCapacity = ZSTD_compressBound(srcSize);
+    void* dst = ZSTD_malloc(dstCapacity, ZSTD_defaultCMem);
+    SeqCollector seqCollector;
+
+    RETURN_ERROR_IF(dst == NULL, memory_allocation, "NULL pointer!");
+
+    seqCollector.collectSequences = 1;
+    seqCollector.seqStart = outSeqs;
+    seqCollector.seqIndex = 0;
+    seqCollector.maxSequences = outSeqsSize;
+    zc->seqCollector = seqCollector;
+
+    ZSTD_compress2(zc, dst, dstCapacity, src, srcSize);
+    ZSTD_free(dst, ZSTD_defaultCMem);
+    return zc->seqCollector.seqIndex;
+}
+
+/* Returns true if the given block is a RLE block */
+static int ZSTD_isRLE(const BYTE *ip, size_t length) {
+    size_t i;
+    if (length < 2) return 1;
+    for (i = 1; i < length; ++i) {
+        if (ip[0] != ip[i]) return 0;
+    }
+    return 1;
+}
+
+/* Returns true if the given block may be RLE.
+ * This is just a heuristic based on the compressibility.
+ * It may return both false positives and false negatives.
+ */
+static int ZSTD_maybeRLE(seqStore_t const* seqStore)
+{
+    size_t const nbSeqs = (size_t)(seqStore->sequences - seqStore->sequencesStart);
+    size_t const nbLits = (size_t)(seqStore->lit - seqStore->litStart);
+
+    return nbSeqs < 4 && nbLits < 10;
+}
 
-    lastLLSize = blockCompressor(zc, src, srcSize);
+static void ZSTD_confirmRepcodesAndEntropyTables(ZSTD_CCtx* zc)
+{
+    ZSTD_compressedBlockState_t* const tmp = zc->blockState.prevCBlock;
+    zc->blockState.prevCBlock = zc->blockState.nextCBlock;
+    zc->blockState.nextCBlock = tmp;
+}
+
+static size_t ZSTD_compressBlock_internal(ZSTD_CCtx* zc,
+                                        void* dst, size_t dstCapacity,
+                                        const void* src, size_t srcSize, U32 frame)
+{
+    /* This the upper bound for the length of an rle block.
+     * This isn't the actual upper bound. Finding the real threshold
+     * needs further investigation.
+     */
+    const U32 rleMaxLength = 25;
+    size_t cSize;
+    const BYTE* ip = (const BYTE*)src;
+    BYTE* op = (BYTE*)dst;
+    DEBUGLOG(5, "ZSTD_compressBlock_internal (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u)",
+                (unsigned)dstCapacity, (unsigned)zc->blockState.matchState.window.dictLimit,
+                (unsigned)zc->blockState.matchState.nextToUpdate);
+
+    {   const size_t bss = ZSTD_buildSeqStore(zc, src, srcSize);
+        FORWARD_IF_ERROR(bss, "ZSTD_buildSeqStore failed");
+        if (bss == ZSTDbss_noCompress) { cSize = 0; goto out; }
+    }
+
+    if (zc->seqCollector.collectSequences) {
+        ZSTD_copyBlockSequences(zc);
+        return 0;
+    }
+
+    /* encode sequences and literals */
+    cSize = ZSTD_compressSequences(&zc->seqStore,
+            &zc->blockState.prevCBlock->entropy, &zc->blockState.nextCBlock->entropy,
+            &zc->appliedParams,
+            dst, dstCapacity,
+            srcSize,
+            zc->entropyWorkspace, HUF_WORKSPACE_SIZE /* statically allocated in resetCCtx */,
+            zc->bmi2);
+
+    if (frame &&
+        /* We don't want to emit our first block as a RLE even if it qualifies because
+         * doing so will cause the decoder (cli only) to throw a "should consume all input error."
+         * This is only an issue for zstd <= v1.4.3
+         */
+        !zc->isFirstBlock &&
+        cSize < rleMaxLength &&
+        ZSTD_isRLE(ip, srcSize))
+    {
+        cSize = 1;
+        op[0] = ip[0];
+    }
+
+out:
+    if (!ZSTD_isError(cSize) && cSize > 1) {
+        ZSTD_confirmRepcodesAndEntropyTables(zc);
+    }
+    /* We check that dictionaries have offset codes available for the first
+     * block. After the first block, the offcode table might not have large
+     * enough codes to represent the offsets in the data.
+     */
+    if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+        zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+
+    return cSize;
+}
+
+static size_t ZSTD_compressBlock_targetCBlockSize_body(ZSTD_CCtx* zc,
+                               void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                               const size_t bss, U32 lastBlock)
+{
+    DEBUGLOG(6, "Attempting ZSTD_compressSuperBlock()");
+    if (bss == ZSTDbss_compress) {
+        if (/* We don't want to emit our first block as a RLE even if it qualifies because
+            * doing so will cause the decoder (cli only) to throw a "should consume all input error."
+            * This is only an issue for zstd <= v1.4.3
+            */
+            !zc->isFirstBlock &&
+            ZSTD_maybeRLE(&zc->seqStore) &&
+            ZSTD_isRLE((BYTE const*)src, srcSize))
+        {
+            return ZSTD_rleCompressBlock(dst, dstCapacity, *(BYTE const*)src, srcSize, lastBlock);
+        }
+        /* Attempt superblock compression.
+         *
+         * Note that compressed size of ZSTD_compressSuperBlock() is not bound by the
+         * standard ZSTD_compressBound(). This is a problem, because even if we have
+         * space now, taking an extra byte now could cause us to run out of space later
+         * and violate ZSTD_compressBound().
+         *
+         * Define blockBound(blockSize) = blockSize + ZSTD_blockHeaderSize.
+         *
+         * In order to respect ZSTD_compressBound() we must attempt to emit a raw
+         * uncompressed block in these cases:
+         *   * cSize == 0: Return code for an uncompressed block.
+         *   * cSize == dstSize_tooSmall: We may have expanded beyond blockBound(srcSize).
+         *     ZSTD_noCompressBlock() will return dstSize_tooSmall if we are really out of
+         *     output space.
+         *   * cSize >= blockBound(srcSize): We have expanded the block too much so
+         *     emit an uncompressed block.
+         */
+        {
+            size_t const cSize = ZSTD_compressSuperBlock(zc, dst, dstCapacity, src, srcSize, lastBlock);
+            if (cSize != ERROR(dstSize_tooSmall)) {
+                size_t const maxCSize = srcSize - ZSTD_minGain(srcSize, zc->appliedParams.cParams.strategy);
+                FORWARD_IF_ERROR(cSize, "ZSTD_compressSuperBlock failed");
+                if (cSize != 0 && cSize < maxCSize + ZSTD_blockHeaderSize) {
+                    ZSTD_confirmRepcodesAndEntropyTables(zc);
+                    return cSize;
+                }
+            }
+        }
+    }
+
+    DEBUGLOG(6, "Resorting to ZSTD_noCompressBlock()");
+    /* Superblock compression failed, attempt to emit a single no compress block.
+     * The decoder will be able to stream this block since it is uncompressed.
+     */
+    return ZSTD_noCompressBlock(dst, dstCapacity, src, srcSize, lastBlock);
+}
+
+static size_t ZSTD_compressBlock_targetCBlockSize(ZSTD_CCtx* zc,
+                               void* dst, size_t dstCapacity,
+                               const void* src, size_t srcSize,
+                               U32 lastBlock)
+{
+    size_t cSize = 0;
+    const size_t bss = ZSTD_buildSeqStore(zc, src, srcSize);
+    DEBUGLOG(5, "ZSTD_compressBlock_targetCBlockSize (dstCapacity=%u, dictLimit=%u, nextToUpdate=%u, srcSize=%zu)",
+                (unsigned)dstCapacity, (unsigned)zc->blockState.matchState.window.dictLimit, (unsigned)zc->blockState.matchState.nextToUpdate, srcSize);
+    FORWARD_IF_ERROR(bss, "ZSTD_buildSeqStore failed");
 
-    /* Last literals */
-    anchor = (const BYTE*)src + srcSize - lastLLSize;
-    ZSTD_storeLastLiterals(&zc->seqStore, anchor, lastLLSize);
+    cSize = ZSTD_compressBlock_targetCBlockSize_body(zc, dst, dstCapacity, src, srcSize, bss, lastBlock);
+    FORWARD_IF_ERROR(cSize, "ZSTD_compressBlock_targetCBlockSize_body failed");
 
-    return ZSTD_compressSequences(&zc->seqStore, zc->entropy, &zc->appliedParams.cParams, dst, dstCapacity, srcSize);
+    if (zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode == FSE_repeat_valid)
+        zc->blockState.prevCBlock->entropy.fse.offcode_repeatMode = FSE_repeat_check;
+
+    return cSize;
 }
 
+static void ZSTD_overflowCorrectIfNeeded(ZSTD_matchState_t* ms,
+                                         ZSTD_cwksp* ws,
+                                         ZSTD_CCtx_params const* params,
+                                         void const* ip,
+                                         void const* iend)
+{
+    if (ZSTD_window_needOverflowCorrection(ms->window, iend)) {
+        U32 const maxDist = (U32)1 << params->cParams.windowLog;
+        U32 const cycleLog = ZSTD_cycleLog(params->cParams.chainLog, params->cParams.strategy);
+        U32 const correction = ZSTD_window_correctOverflow(&ms->window, cycleLog, maxDist, ip);
+        ZSTD_STATIC_ASSERT(ZSTD_CHAINLOG_MAX <= 30);
+        ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX_32 <= 30);
+        ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX <= 31);
+        ZSTD_cwksp_mark_tables_dirty(ws);
+        ZSTD_reduceIndex(ms, params, correction);
+        ZSTD_cwksp_mark_tables_clean(ws);
+        if (ms->nextToUpdate < correction) ms->nextToUpdate = 0;
+        else ms->nextToUpdate -= correction;
+        /* invalidate dictionaries on overflow correction */
+        ms->loadedDictEnd = 0;
+        ms->dictMatchState = NULL;
+    }
+}
 
 /*! ZSTD_compress_frameChunk() :
 *   Compress a chunk of data into one or multiple blocks.
@@ -1586,88 +2589,96 @@ static size_t ZSTD_compress_frameChunk (ZSTD_CCtx* cctx,
     const BYTE* ip = (const BYTE*)src;
     BYTE* const ostart = (BYTE*)dst;
     BYTE* op = ostart;
-    U32 const maxDist = 1 << cctx->appliedParams.cParams.windowLog;
+    U32 const maxDist = (U32)1 << cctx->appliedParams.cParams.windowLog;
 
+    assert(cctx->appliedParams.cParams.windowLog <= ZSTD_WINDOWLOG_MAX);
+
+    DEBUGLOG(5, "ZSTD_compress_frameChunk (blockSize=%u)", (unsigned)blockSize);
     if (cctx->appliedParams.fParams.checksumFlag && srcSize)
         XXH64_update(&cctx->xxhState, src, srcSize);
 
     while (remaining) {
+        ZSTD_matchState_t* const ms = &cctx->blockState.matchState;
         U32 const lastBlock = lastFrameChunk & (blockSize >= remaining);
-        size_t cSize;
 
-        if (dstCapacity < ZSTD_blockHeaderSize + MIN_CBLOCK_SIZE)
-            return ERROR(dstSize_tooSmall);   /* not enough space to store compressed block */
+        RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize + MIN_CBLOCK_SIZE,
+                        dstSize_tooSmall,
+                        "not enough space to store compressed block");
         if (remaining < blockSize) blockSize = remaining;
 
-        /* preemptive overflow correction */
-        if (cctx->lowLimit > (3U<<29)) {
-            U32 const cycleMask = (1 << ZSTD_cycleLog(cctx->appliedParams.cParams.hashLog, cctx->appliedParams.cParams.strategy)) - 1;
-            U32 const current = (U32)(ip - cctx->base);
-            U32 const newCurrent = (current & cycleMask) + (1 << cctx->appliedParams.cParams.windowLog);
-            U32 const correction = current - newCurrent;
-            ZSTD_STATIC_ASSERT(ZSTD_WINDOWLOG_MAX_64 <= 30);
-            ZSTD_reduceIndex(cctx, correction);
-            cctx->base += correction;
-            cctx->dictBase += correction;
-            cctx->lowLimit -= correction;
-            cctx->dictLimit -= correction;
-            if (cctx->nextToUpdate < correction) cctx->nextToUpdate = 0;
-            else cctx->nextToUpdate -= correction;
-        }
+        ZSTD_overflowCorrectIfNeeded(
+            ms, &cctx->workspace, &cctx->appliedParams, ip, ip + blockSize);
+        ZSTD_checkDictValidity(&ms->window, ip + blockSize, maxDist, &ms->loadedDictEnd, &ms->dictMatchState);
 
-        if ((U32)(ip+blockSize - cctx->base) > cctx->loadedDictEnd + maxDist) {
-            /* enforce maxDist */
-            U32 const newLowLimit = (U32)(ip+blockSize - cctx->base) - maxDist;
-            if (cctx->lowLimit < newLowLimit) cctx->lowLimit = newLowLimit;
-            if (cctx->dictLimit < cctx->lowLimit) cctx->dictLimit = cctx->lowLimit;
-        }
+        /* Ensure hash/chain table insertion resumes no sooner than lowlimit */
+        if (ms->nextToUpdate < ms->window.lowLimit) ms->nextToUpdate = ms->window.lowLimit;
 
-        cSize = ZSTD_compressBlock_internal(cctx, op+ZSTD_blockHeaderSize, dstCapacity-ZSTD_blockHeaderSize, ip, blockSize);
-        if (ZSTD_isError(cSize)) return cSize;
+        {   size_t cSize;
+            if (ZSTD_useTargetCBlockSize(&cctx->appliedParams)) {
+                cSize = ZSTD_compressBlock_targetCBlockSize(cctx, op, dstCapacity, ip, blockSize, lastBlock);
+                FORWARD_IF_ERROR(cSize, "ZSTD_compressBlock_targetCBlockSize failed");
+                assert(cSize > 0);
+                assert(cSize <= blockSize + ZSTD_blockHeaderSize);
+            } else {
+                cSize = ZSTD_compressBlock_internal(cctx,
+                                        op+ZSTD_blockHeaderSize, dstCapacity-ZSTD_blockHeaderSize,
+                                        ip, blockSize, 1 /* frame */);
+                FORWARD_IF_ERROR(cSize, "ZSTD_compressBlock_internal failed");
+
+                if (cSize == 0) {  /* block is not compressible */
+                    cSize = ZSTD_noCompressBlock(op, dstCapacity, ip, blockSize, lastBlock);
+                    FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
+                } else {
+                    U32 const cBlockHeader = cSize == 1 ?
+                        lastBlock + (((U32)bt_rle)<<1) + (U32)(blockSize << 3) :
+                        lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
+                    MEM_writeLE24(op, cBlockHeader);
+                    cSize += ZSTD_blockHeaderSize;
+                }
+            }
 
-        if (cSize == 0) {  /* block is not compressible */
-            U32 const cBlockHeader24 = lastBlock + (((U32)bt_raw)<<1) + (U32)(blockSize << 3);
-            if (blockSize + ZSTD_blockHeaderSize > dstCapacity) return ERROR(dstSize_tooSmall);
-            MEM_writeLE32(op, cBlockHeader24);   /* no pb, 4th byte will be overwritten */
-            memcpy(op + ZSTD_blockHeaderSize, ip, blockSize);
-            cSize = ZSTD_blockHeaderSize+blockSize;
-        } else {
-            U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
-            MEM_writeLE24(op, cBlockHeader24);
-            cSize += ZSTD_blockHeaderSize;
-        }
 
-        remaining -= blockSize;
-        dstCapacity -= cSize;
-        ip += blockSize;
-        op += cSize;
-    }
+            ip += blockSize;
+            assert(remaining >= blockSize);
+            remaining -= blockSize;
+            op += cSize;
+            assert(dstCapacity >= cSize);
+            dstCapacity -= cSize;
+            cctx->isFirstBlock = 0;
+            DEBUGLOG(5, "ZSTD_compress_frameChunk: adding a block of size %u",
+                        (unsigned)cSize);
+    }   }
 
     if (lastFrameChunk && (op>ostart)) cctx->stage = ZSTDcs_ending;
-    return op-ostart;
+    return (size_t)(op-ostart);
 }
 
 
 static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
-                                    ZSTD_CCtx_params params, U64 pledgedSrcSize, U32 dictID)
+                                    const ZSTD_CCtx_params* params, U64 pledgedSrcSize, U32 dictID)
 {   BYTE* const op = (BYTE*)dst;
     U32   const dictIDSizeCodeLength = (dictID>0) + (dictID>=256) + (dictID>=65536);   /* 0-3 */
-    U32   const dictIDSizeCode = params.fParams.noDictIDFlag ? 0 : dictIDSizeCodeLength;   /* 0-3 */
-    U32   const checksumFlag = params.fParams.checksumFlag>0;
-    U32   const windowSize = 1U << params.cParams.windowLog;
-    U32   const singleSegment = params.fParams.contentSizeFlag && (windowSize >= pledgedSrcSize);
-    BYTE  const windowLogByte = (BYTE)((params.cParams.windowLog - ZSTD_WINDOWLOG_ABSOLUTEMIN) << 3);
-    U32   const fcsCode = params.fParams.contentSizeFlag ?
+    U32   const dictIDSizeCode = params->fParams.noDictIDFlag ? 0 : dictIDSizeCodeLength;   /* 0-3 */
+    U32   const checksumFlag = params->fParams.checksumFlag>0;
+    U32   const windowSize = (U32)1 << params->cParams.windowLog;
+    U32   const singleSegment = params->fParams.contentSizeFlag && (windowSize >= pledgedSrcSize);
+    BYTE  const windowLogByte = (BYTE)((params->cParams.windowLog - ZSTD_WINDOWLOG_ABSOLUTEMIN) << 3);
+    U32   const fcsCode = params->fParams.contentSizeFlag ?
                      (pledgedSrcSize>=256) + (pledgedSrcSize>=65536+256) + (pledgedSrcSize>=0xFFFFFFFFU) : 0;  /* 0-3 */
-    BYTE  const frameHeaderDecriptionByte = (BYTE)(dictIDSizeCode + (checksumFlag<<2) + (singleSegment<<5) + (fcsCode<<6) );
-    size_t pos;
-
-    if (dstCapacity < ZSTD_frameHeaderSize_max) return ERROR(dstSize_tooSmall);
-    DEBUGLOG(5, "ZSTD_writeFrameHeader : dictIDFlag : %u ; dictID : %u ; dictIDSizeCode : %u",
-                !params.fParams.noDictIDFlag, dictID,  dictIDSizeCode);
-
-    MEM_writeLE32(dst, ZSTD_MAGICNUMBER);
-    op[4] = frameHeaderDecriptionByte; pos=5;
+    BYTE  const frameHeaderDescriptionByte = (BYTE)(dictIDSizeCode + (checksumFlag<<2) + (singleSegment<<5) + (fcsCode<<6) );
+    size_t pos=0;
+
+    assert(!(params->fParams.contentSizeFlag && pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN));
+    RETURN_ERROR_IF(dstCapacity < ZSTD_FRAMEHEADERSIZE_MAX, dstSize_tooSmall,
+                    "dst buf is too small to fit worst-case frame header size.");
+    DEBUGLOG(4, "ZSTD_writeFrameHeader : dictIDFlag : %u ; dictID : %u ; dictIDSizeCode : %u",
+                !params->fParams.noDictIDFlag, (unsigned)dictID, (unsigned)dictIDSizeCode);
+
+    if (params->format == ZSTD_f_zstd1) {
+        MEM_writeLE32(dst, ZSTD_MAGICNUMBER);
+        pos = 4;
+    }
+    op[pos++] = frameHeaderDescriptionByte;
     if (!singleSegment) op[pos++] = windowLogByte;
     switch(dictIDSizeCode)
     {
@@ -1688,129 +2699,189 @@ static size_t ZSTD_writeFrameHeader(void* dst, size_t dstCapacity,
     return pos;
 }
 
+/* ZSTD_writeLastEmptyBlock() :
+ * output an empty Block with end-of-frame mark to complete a frame
+ * @return : size of data written into `dst` (== ZSTD_blockHeaderSize (defined in zstd_internal.h))
+ *           or an error code if `dstCapacity` is too small (<ZSTD_blockHeaderSize)
+ */
+size_t ZSTD_writeLastEmptyBlock(void* dst, size_t dstCapacity)
+{
+    RETURN_ERROR_IF(dstCapacity < ZSTD_blockHeaderSize, dstSize_tooSmall,
+                    "dst buf is too small to write frame trailer empty block.");
+    {   U32 const cBlockHeader24 = 1 /*lastBlock*/ + (((U32)bt_raw)<<1);  /* 0 size */
+        MEM_writeLE24(dst, cBlockHeader24);
+        return ZSTD_blockHeaderSize;
+    }
+}
+
+size_t ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSeq)
+{
+    RETURN_ERROR_IF(cctx->stage != ZSTDcs_init, stage_wrong,
+                    "wrong cctx stage");
+    RETURN_ERROR_IF(cctx->appliedParams.ldmParams.enableLdm,
+                    parameter_unsupported,
+                    "incompatible with ldm");
+    cctx->externSeqStore.seq = seq;
+    cctx->externSeqStore.size = nbSeq;
+    cctx->externSeqStore.capacity = nbSeq;
+    cctx->externSeqStore.pos = 0;
+    return 0;
+}
+
 
 static size_t ZSTD_compressContinue_internal (ZSTD_CCtx* cctx,
                               void* dst, size_t dstCapacity,
                         const void* src, size_t srcSize,
                                U32 frame, U32 lastFrameChunk)
 {
-    const BYTE* const ip = (const BYTE*) src;
+    ZSTD_matchState_t* const ms = &cctx->blockState.matchState;
     size_t fhSize = 0;
 
-    DEBUGLOG(5, "ZSTD_compressContinue_internal");
-    DEBUGLOG(5, "stage: %u", cctx->stage);
-    if (cctx->stage==ZSTDcs_created) return ERROR(stage_wrong);   /* missing init (ZSTD_compressBegin) */
+    DEBUGLOG(5, "ZSTD_compressContinue_internal, stage: %u, srcSize: %u",
+                cctx->stage, (unsigned)srcSize);
+    RETURN_ERROR_IF(cctx->stage==ZSTDcs_created, stage_wrong,
+                    "missing init (ZSTD_compressBegin)");
 
     if (frame && (cctx->stage==ZSTDcs_init)) {
-        fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, cctx->appliedParams,
+        fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, &cctx->appliedParams,
                                        cctx->pledgedSrcSizePlusOne-1, cctx->dictID);
-        if (ZSTD_isError(fhSize)) return fhSize;
+        FORWARD_IF_ERROR(fhSize, "ZSTD_writeFrameHeader failed");
+        assert(fhSize <= dstCapacity);
         dstCapacity -= fhSize;
         dst = (char*)dst + fhSize;
         cctx->stage = ZSTDcs_ongoing;
     }
 
-    /* Check if blocks follow each other */
-    if (src != cctx->nextSrc) {
-        /* not contiguous */
-        ptrdiff_t const delta = cctx->nextSrc - ip;
-        cctx->lowLimit = cctx->dictLimit;
-        cctx->dictLimit = (U32)(cctx->nextSrc - cctx->base);
-        cctx->dictBase = cctx->base;
-        cctx->base -= delta;
-        cctx->nextToUpdate = cctx->dictLimit;
-        if (cctx->dictLimit - cctx->lowLimit < HASH_READ_SIZE) cctx->lowLimit = cctx->dictLimit;   /* too small extDict */
-    }
+    if (!srcSize) return fhSize;  /* do not generate an empty block if no input */
 
-    /* if input and dictionary overlap : reduce dictionary (area presumed modified by input) */
-    if ((ip+srcSize > cctx->dictBase + cctx->lowLimit) & (ip < cctx->dictBase + cctx->dictLimit)) {
-        ptrdiff_t const highInputIdx = (ip + srcSize) - cctx->dictBase;
-        U32 const lowLimitMax = (highInputIdx > (ptrdiff_t)cctx->dictLimit) ? cctx->dictLimit : (U32)highInputIdx;
-        cctx->lowLimit = lowLimitMax;
+    if (!ZSTD_window_update(&ms->window, src, srcSize)) {
+        ms->nextToUpdate = ms->window.dictLimit;
+    }
+    if (cctx->appliedParams.ldmParams.enableLdm) {
+        ZSTD_window_update(&cctx->ldmState.window, src, srcSize);
     }
 
-    cctx->nextSrc = ip + srcSize;
+    if (!frame) {
+        /* overflow check and correction for block mode */
+        ZSTD_overflowCorrectIfNeeded(
+            ms, &cctx->workspace, &cctx->appliedParams,
+            src, (BYTE const*)src + srcSize);
+    }
 
-    if (srcSize) {
-        size_t const cSize = frame ?
+    DEBUGLOG(5, "ZSTD_compressContinue_internal (blockSize=%u)", (unsigned)cctx->blockSize);
+    {   size_t const cSize = frame ?
                              ZSTD_compress_frameChunk (cctx, dst, dstCapacity, src, srcSize, lastFrameChunk) :
-                             ZSTD_compressBlock_internal (cctx, dst, dstCapacity, src, srcSize);
-        if (ZSTD_isError(cSize)) return cSize;
+                             ZSTD_compressBlock_internal (cctx, dst, dstCapacity, src, srcSize, 0 /* frame */);
+        FORWARD_IF_ERROR(cSize, "%s", frame ? "ZSTD_compress_frameChunk failed" : "ZSTD_compressBlock_internal failed");
         cctx->consumedSrcSize += srcSize;
+        cctx->producedCSize += (cSize + fhSize);
+        assert(!(cctx->appliedParams.fParams.contentSizeFlag && cctx->pledgedSrcSizePlusOne == 0));
+        if (cctx->pledgedSrcSizePlusOne != 0) {  /* control src size */
+            ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN == (unsigned long long)-1);
+            RETURN_ERROR_IF(
+                cctx->consumedSrcSize+1 > cctx->pledgedSrcSizePlusOne,
+                srcSize_wrong,
+                "error : pledgedSrcSize = %u, while realSrcSize >= %u",
+                (unsigned)cctx->pledgedSrcSizePlusOne-1,
+                (unsigned)cctx->consumedSrcSize);
+        }
         return cSize + fhSize;
-    } else
-        return fhSize;
+    }
 }
 
 size_t ZSTD_compressContinue (ZSTD_CCtx* cctx,
                               void* dst, size_t dstCapacity,
                         const void* src, size_t srcSize)
 {
+    DEBUGLOG(5, "ZSTD_compressContinue (srcSize=%u)", (unsigned)srcSize);
     return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 1 /* frame mode */, 0 /* last chunk */);
 }
 
 
 size_t ZSTD_getBlockSize(const ZSTD_CCtx* cctx)
 {
-    ZSTD_compressionParameters const cParams =
-            ZSTD_getCParamsFromCCtxParams(cctx->appliedParams, 0, 0);
-    return MIN (ZSTD_BLOCKSIZE_MAX, 1 << cParams.windowLog);
+    ZSTD_compressionParameters const cParams = cctx->appliedParams.cParams;
+    assert(!ZSTD_checkCParams(cParams));
+    return MIN (ZSTD_BLOCKSIZE_MAX, (U32)1 << cParams.windowLog);
 }
 
 size_t ZSTD_compressBlock(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
 {
-    size_t const blockSizeMax = ZSTD_getBlockSize(cctx);
-    if (srcSize > blockSizeMax) return ERROR(srcSize_wrong);
+    DEBUGLOG(5, "ZSTD_compressBlock: srcSize = %u", (unsigned)srcSize);
+    { size_t const blockSizeMax = ZSTD_getBlockSize(cctx);
+      RETURN_ERROR_IF(srcSize > blockSizeMax, srcSize_wrong, "input is larger than a block"); }
+
     return ZSTD_compressContinue_internal(cctx, dst, dstCapacity, src, srcSize, 0 /* frame mode */, 0 /* last chunk */);
 }
 
 /*! ZSTD_loadDictionaryContent() :
  *  @return : 0, or an error code
  */
-static size_t ZSTD_loadDictionaryContent(ZSTD_CCtx* zc, const void* src, size_t srcSize)
+static size_t ZSTD_loadDictionaryContent(ZSTD_matchState_t* ms,
+                                         ldmState_t* ls,
+                                         ZSTD_cwksp* ws,
+                                         ZSTD_CCtx_params const* params,
+                                         const void* src, size_t srcSize,
+                                         ZSTD_dictTableLoadMethod_e dtlm)
 {
-    const BYTE* const ip = (const BYTE*) src;
+    const BYTE* ip = (const BYTE*) src;
     const BYTE* const iend = ip + srcSize;
 
-    /* input becomes current prefix */
-    zc->lowLimit = zc->dictLimit;
-    zc->dictLimit = (U32)(zc->nextSrc - zc->base);
-    zc->dictBase = zc->base;
-    zc->base += ip - zc->nextSrc;
-    zc->nextToUpdate = zc->dictLimit;
-    zc->loadedDictEnd = zc->appliedParams.forceWindow ? 0 : (U32)(iend - zc->base);
+    ZSTD_window_update(&ms->window, src, srcSize);
+    ms->loadedDictEnd = params->forceWindow ? 0 : (U32)(iend - ms->window.base);
+
+    if (params->ldmParams.enableLdm && ls != NULL) {
+        ZSTD_window_update(&ls->window, src, srcSize);
+        ls->loadedDictEnd = params->forceWindow ? 0 : (U32)(iend - ls->window.base);
+    }
+
+    /* Assert that we the ms params match the params we're being given */
+    ZSTD_assertEqualCParams(params->cParams, ms->cParams);
 
-    zc->nextSrc = iend;
     if (srcSize <= HASH_READ_SIZE) return 0;
 
-    switch(zc->appliedParams.cParams.strategy)
-    {
-    case ZSTD_fast:
-        ZSTD_fillHashTable (zc, iend, zc->appliedParams.cParams.searchLength);
-        break;
-    case ZSTD_dfast:
-        ZSTD_fillDoubleHashTable (zc, iend, zc->appliedParams.cParams.searchLength);
-        break;
+    while (iend - ip > HASH_READ_SIZE) {
+        size_t const remaining = (size_t)(iend - ip);
+        size_t const chunk = MIN(remaining, ZSTD_CHUNKSIZE_MAX);
+        const BYTE* const ichunk = ip + chunk;
 
-    case ZSTD_greedy:
-    case ZSTD_lazy:
-    case ZSTD_lazy2:
-        if (srcSize >= HASH_READ_SIZE)
-            ZSTD_insertAndFindFirstIndex(zc, iend-HASH_READ_SIZE, zc->appliedParams.cParams.searchLength);
-        break;
+        ZSTD_overflowCorrectIfNeeded(ms, ws, params, ip, ichunk);
 
-    case ZSTD_btlazy2:
-    case ZSTD_btopt:
-    case ZSTD_btultra:
-        if (srcSize >= HASH_READ_SIZE)
-            ZSTD_updateTree(zc, iend-HASH_READ_SIZE, iend, 1 << zc->appliedParams.cParams.searchLog, zc->appliedParams.cParams.searchLength);
-        break;
+        if (params->ldmParams.enableLdm && ls != NULL)
+            ZSTD_ldm_fillHashTable(ls, (const BYTE*)src, (const BYTE*)src + srcSize, &params->ldmParams);
 
-    default:
-        assert(0);  /* not possible : not a valid strategy id */
+        switch(params->cParams.strategy)
+        {
+        case ZSTD_fast:
+            ZSTD_fillHashTable(ms, ichunk, dtlm);
+            break;
+        case ZSTD_dfast:
+            ZSTD_fillDoubleHashTable(ms, ichunk, dtlm);
+            break;
+
+        case ZSTD_greedy:
+        case ZSTD_lazy:
+        case ZSTD_lazy2:
+            if (chunk >= HASH_READ_SIZE)
+                ZSTD_insertAndFindFirstIndex(ms, ichunk-HASH_READ_SIZE);
+            break;
+
+        case ZSTD_btlazy2:   /* we want the dictionary table fully sorted */
+        case ZSTD_btopt:
+        case ZSTD_btultra:
+        case ZSTD_btultra2:
+            if (chunk >= HASH_READ_SIZE)
+                ZSTD_updateTree(ms, ichunk-HASH_READ_SIZE, ichunk);
+            break;
+
+        default:
+            assert(0);  /* not possible : not a valid strategy id */
+        }
+
+        ip = ichunk;
     }
 
-    zc->nextToUpdate = (U32)(iend - zc->base);
+    ms->nextToUpdate = (U32)(iend - ms->window.base);
     return 0;
 }
 
@@ -1821,81 +2892,123 @@ static size_t ZSTD_loadDictionaryContent(ZSTD_CCtx* zc, const void* src, size_t
    NOTE: This behavior is not standard and could be improved in the future. */
 static size_t ZSTD_checkDictNCount(short* normalizedCounter, unsigned dictMaxSymbolValue, unsigned maxSymbolValue) {
     U32 s;
-    if (dictMaxSymbolValue < maxSymbolValue) return ERROR(dictionary_corrupted);
+    RETURN_ERROR_IF(dictMaxSymbolValue < maxSymbolValue, dictionary_corrupted, "dict fse tables don't have all symbols");
     for (s = 0; s <= maxSymbolValue; ++s) {
-        if (normalizedCounter[s] == 0) return ERROR(dictionary_corrupted);
+        RETURN_ERROR_IF(normalizedCounter[s] == 0, dictionary_corrupted, "dict fse tables don't have all symbols");
     }
     return 0;
 }
 
-
-/* Dictionary format :
- * See :
- * https://github.com/facebook/zstd/blob/master/doc/zstd_compression_format.md#dictionary-format
- */
-/*! ZSTD_loadZstdDictionary() :
- * @return : 0, or an error code
- *  assumptions : magic number supposed already checked
- *                dictSize supposed > 8
- */
-static size_t ZSTD_loadZstdDictionary(ZSTD_CCtx* cctx, const void* dict, size_t dictSize)
+size_t ZSTD_loadCEntropy(ZSTD_compressedBlockState_t* bs, void* workspace,
+                         short* offcodeNCount, unsigned* offcodeMaxValue,
+                         const void* const dict, size_t dictSize)
 {
-    const BYTE* dictPtr = (const BYTE*)dict;
+    const BYTE* dictPtr = (const BYTE*)dict;    /* skip magic num and dict ID */
     const BYTE* const dictEnd = dictPtr + dictSize;
-    short offcodeNCount[MaxOff+1];
-    unsigned offcodeMaxValue = MaxOff;
+    dictPtr += 8;
+    bs->entropy.huf.repeatMode = HUF_repeat_check;
 
-    ZSTD_STATIC_ASSERT(sizeof(cctx->entropy->workspace) >= (1<<MAX(MLFSELog,LLFSELog)));
+    {   unsigned maxSymbolValue = 255;
+        unsigned hasZeroWeights = 1;
+        size_t const hufHeaderSize = HUF_readCTable((HUF_CElt*)bs->entropy.huf.CTable, &maxSymbolValue, dictPtr,
+            dictEnd-dictPtr, &hasZeroWeights);
 
-    dictPtr += 4;   /* skip magic number */
-    cctx->dictID = cctx->appliedParams.fParams.noDictIDFlag ? 0 :  MEM_readLE32(dictPtr);
-    dictPtr += 4;
+        /* We only set the loaded table as valid if it contains all non-zero
+         * weights. Otherwise, we set it to check */
+        if (!hasZeroWeights)
+            bs->entropy.huf.repeatMode = HUF_repeat_valid;
 
-    {   size_t const hufHeaderSize = HUF_readCTable((HUF_CElt*)cctx->entropy->hufCTable, 255, dictPtr, dictEnd-dictPtr);
-        if (HUF_isError(hufHeaderSize)) return ERROR(dictionary_corrupted);
+        RETURN_ERROR_IF(HUF_isError(hufHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(maxSymbolValue < 255, dictionary_corrupted, "");
         dictPtr += hufHeaderSize;
     }
 
     {   unsigned offcodeLog;
-        size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
-        if (FSE_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
-        if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
+        size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
+        RETURN_ERROR_IF(FSE_isError(offcodeHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(offcodeLog > OffFSELog, dictionary_corrupted, "");
         /* Defer checking offcodeMaxValue because we need to know the size of the dictionary content */
-        CHECK_E( FSE_buildCTable_wksp(cctx->entropy->offcodeCTable, offcodeNCount, offcodeMaxValue, offcodeLog, cctx->entropy->workspace, sizeof(cctx->entropy->workspace)),
-                 dictionary_corrupted);
+        /* fill all offset symbols to avoid garbage at end of table */
+        RETURN_ERROR_IF(FSE_isError(FSE_buildCTable_wksp(
+                bs->entropy.fse.offcodeCTable,
+                offcodeNCount, MaxOff, offcodeLog,
+                workspace, HUF_WORKSPACE_SIZE)),
+            dictionary_corrupted, "");
         dictPtr += offcodeHeaderSize;
     }
 
     {   short matchlengthNCount[MaxML+1];
         unsigned matchlengthMaxValue = MaxML, matchlengthLog;
         size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
-        if (FSE_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
-        if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
+        RETURN_ERROR_IF(FSE_isError(matchlengthHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(matchlengthLog > MLFSELog, dictionary_corrupted, "");
         /* Every match length code must have non-zero probability */
-        CHECK_F( ZSTD_checkDictNCount(matchlengthNCount, matchlengthMaxValue, MaxML));
-        CHECK_E( FSE_buildCTable_wksp(cctx->entropy->matchlengthCTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog, cctx->entropy->workspace, sizeof(cctx->entropy->workspace)),
-                 dictionary_corrupted);
+        FORWARD_IF_ERROR( ZSTD_checkDictNCount(matchlengthNCount, matchlengthMaxValue, MaxML), "");
+        RETURN_ERROR_IF(FSE_isError(FSE_buildCTable_wksp(
+                bs->entropy.fse.matchlengthCTable,
+                matchlengthNCount, matchlengthMaxValue, matchlengthLog,
+                workspace, HUF_WORKSPACE_SIZE)),
+            dictionary_corrupted, "");
         dictPtr += matchlengthHeaderSize;
     }
 
     {   short litlengthNCount[MaxLL+1];
         unsigned litlengthMaxValue = MaxLL, litlengthLog;
         size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
-        if (FSE_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
-        if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
+        RETURN_ERROR_IF(FSE_isError(litlengthHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(litlengthLog > LLFSELog, dictionary_corrupted, "");
         /* Every literal length code must have non-zero probability */
-        CHECK_F( ZSTD_checkDictNCount(litlengthNCount, litlengthMaxValue, MaxLL));
-        CHECK_E( FSE_buildCTable_wksp(cctx->entropy->litlengthCTable, litlengthNCount, litlengthMaxValue, litlengthLog, cctx->entropy->workspace, sizeof(cctx->entropy->workspace)),
-                 dictionary_corrupted);
+        FORWARD_IF_ERROR( ZSTD_checkDictNCount(litlengthNCount, litlengthMaxValue, MaxLL), "");
+        RETURN_ERROR_IF(FSE_isError(FSE_buildCTable_wksp(
+                bs->entropy.fse.litlengthCTable,
+                litlengthNCount, litlengthMaxValue, litlengthLog,
+                workspace, HUF_WORKSPACE_SIZE)),
+            dictionary_corrupted, "");
         dictPtr += litlengthHeaderSize;
     }
 
-    if (dictPtr+12 > dictEnd) return ERROR(dictionary_corrupted);
-    cctx->seqStore.rep[0] = MEM_readLE32(dictPtr+0);
-    cctx->seqStore.rep[1] = MEM_readLE32(dictPtr+4);
-    cctx->seqStore.rep[2] = MEM_readLE32(dictPtr+8);
+    RETURN_ERROR_IF(dictPtr+12 > dictEnd, dictionary_corrupted, "");
+    bs->rep[0] = MEM_readLE32(dictPtr+0);
+    bs->rep[1] = MEM_readLE32(dictPtr+4);
+    bs->rep[2] = MEM_readLE32(dictPtr+8);
     dictPtr += 12;
 
+    return dictPtr - (const BYTE*)dict;
+}
+
+/* Dictionary format :
+ * See :
+ * https://github.com/facebook/zstd/blob/master/doc/zstd_compression_format.md#dictionary-format
+ */
+/*! ZSTD_loadZstdDictionary() :
+ * @return : dictID, or an error code
+ *  assumptions : magic number supposed already checked
+ *                dictSize supposed >= 8
+ */
+static size_t ZSTD_loadZstdDictionary(ZSTD_compressedBlockState_t* bs,
+                                      ZSTD_matchState_t* ms,
+                                      ZSTD_cwksp* ws,
+                                      ZSTD_CCtx_params const* params,
+                                      const void* dict, size_t dictSize,
+                                      ZSTD_dictTableLoadMethod_e dtlm,
+                                      void* workspace)
+{
+    const BYTE* dictPtr = (const BYTE*)dict;
+    const BYTE* const dictEnd = dictPtr + dictSize;
+    short offcodeNCount[MaxOff+1];
+    unsigned offcodeMaxValue = MaxOff;
+    size_t dictID;
+    size_t eSize;
+
+    ZSTD_STATIC_ASSERT(HUF_WORKSPACE_SIZE >= (1<<MAX(MLFSELog,LLFSELog)));
+    assert(dictSize >= 8);
+    assert(MEM_readLE32(dictPtr) == ZSTD_MAGIC_DICTIONARY);
+
+    dictID = params->fParams.noDictIDFlag ? 0 :  MEM_readLE32(dictPtr + 4 /* skip magic number */ );
+    eSize = ZSTD_loadCEntropy(bs, workspace, offcodeNCount, &offcodeMaxValue, dict, dictSize);
+    FORWARD_IF_ERROR(eSize, "ZSTD_loadCEntropy failed");
+    dictPtr += eSize;
+
     {   size_t const dictContentSize = (size_t)(dictEnd - dictPtr);
         U32 offcodeMax = MaxOff;
         if (dictContentSize <= ((U32)-1) - 128 KB) {
@@ -1903,86 +3016,123 @@ static size_t ZSTD_loadZstdDictionary(ZSTD_CCtx* cctx, const void* dict, size_t
             offcodeMax = ZSTD_highbit32(maxOffset); /* Calculate minimum offset code required to represent maxOffset */
         }
         /* All offset values <= dictContentSize + 128 KB must be representable */
-        CHECK_F (ZSTD_checkDictNCount(offcodeNCount, offcodeMaxValue, MIN(offcodeMax, MaxOff)));
+        FORWARD_IF_ERROR(ZSTD_checkDictNCount(offcodeNCount, offcodeMaxValue, MIN(offcodeMax, MaxOff)), "");
         /* All repCodes must be <= dictContentSize and != 0*/
         {   U32 u;
             for (u=0; u<3; u++) {
-                if (cctx->seqStore.rep[u] == 0) return ERROR(dictionary_corrupted);
-                if (cctx->seqStore.rep[u] > dictContentSize) return ERROR(dictionary_corrupted);
+                RETURN_ERROR_IF(bs->rep[u] == 0, dictionary_corrupted, "");
+                RETURN_ERROR_IF(bs->rep[u] > dictContentSize, dictionary_corrupted, "");
         }   }
 
-        cctx->entropy->hufCTable_repeatMode = HUF_repeat_valid;
-        cctx->entropy->offcode_repeatMode = FSE_repeat_valid;
-        cctx->entropy->matchlength_repeatMode = FSE_repeat_valid;
-        cctx->entropy->litlength_repeatMode = FSE_repeat_valid;
-        return ZSTD_loadDictionaryContent(cctx, dictPtr, dictContentSize);
+        bs->entropy.fse.offcode_repeatMode = FSE_repeat_valid;
+        bs->entropy.fse.matchlength_repeatMode = FSE_repeat_valid;
+        bs->entropy.fse.litlength_repeatMode = FSE_repeat_valid;
+        FORWARD_IF_ERROR(ZSTD_loadDictionaryContent(
+            ms, NULL, ws, params, dictPtr, dictContentSize, dtlm), "");
+        return dictID;
     }
 }
 
 /** ZSTD_compress_insertDictionary() :
-*   @return : 0, or an error code */
-static size_t ZSTD_compress_insertDictionary(ZSTD_CCtx* cctx,
-                                       const void* dict, size_t dictSize,
-                                             ZSTD_dictMode_e dictMode)
+*   @return : dictID, or an error code */
+static size_t
+ZSTD_compress_insertDictionary(ZSTD_compressedBlockState_t* bs,
+                               ZSTD_matchState_t* ms,
+                               ldmState_t* ls,
+                               ZSTD_cwksp* ws,
+                         const ZSTD_CCtx_params* params,
+                         const void* dict, size_t dictSize,
+                               ZSTD_dictContentType_e dictContentType,
+                               ZSTD_dictTableLoadMethod_e dtlm,
+                               void* workspace)
 {
-    DEBUGLOG(5, "ZSTD_compress_insertDictionary");
-    if ((dict==NULL) || (dictSize<=8)) return 0;
+    DEBUGLOG(4, "ZSTD_compress_insertDictionary (dictSize=%u)", (U32)dictSize);
+    if ((dict==NULL) || (dictSize<8)) {
+        RETURN_ERROR_IF(dictContentType == ZSTD_dct_fullDict, dictionary_wrong, "");
+        return 0;
+    }
+
+    ZSTD_reset_compressedBlockState(bs);
 
     /* dict restricted modes */
-    if (dictMode==ZSTD_dm_rawContent)
-        return ZSTD_loadDictionaryContent(cctx, dict, dictSize);
+    if (dictContentType == ZSTD_dct_rawContent)
+        return ZSTD_loadDictionaryContent(ms, ls, ws, params, dict, dictSize, dtlm);
 
     if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) {
-        if (dictMode == ZSTD_dm_auto) {
-            DEBUGLOG(5, "raw content dictionary detected");
-            return ZSTD_loadDictionaryContent(cctx, dict, dictSize);
+        if (dictContentType == ZSTD_dct_auto) {
+            DEBUGLOG(4, "raw content dictionary detected");
+            return ZSTD_loadDictionaryContent(
+                ms, ls, ws, params, dict, dictSize, dtlm);
         }
-        if (dictMode == ZSTD_dm_fullDict)
-            return ERROR(dictionary_wrong);
+        RETURN_ERROR_IF(dictContentType == ZSTD_dct_fullDict, dictionary_wrong, "");
         assert(0);   /* impossible */
     }
 
     /* dict as full zstd dictionary */
-    return ZSTD_loadZstdDictionary(cctx, dict, dictSize);
+    return ZSTD_loadZstdDictionary(
+        bs, ms, ws, params, dict, dictSize, dtlm, workspace);
 }
 
+#define ZSTD_USE_CDICT_PARAMS_SRCSIZE_CUTOFF (128 KB)
+#define ZSTD_USE_CDICT_PARAMS_DICTSIZE_MULTIPLIER (6)
+
 /*! ZSTD_compressBegin_internal() :
  * @return : 0, or an error code */
 static size_t ZSTD_compressBegin_internal(ZSTD_CCtx* cctx,
-                             const void* dict, size_t dictSize,
-                             ZSTD_dictMode_e dictMode,
-                             const ZSTD_CDict* cdict,
-                                   ZSTD_CCtx_params params, U64 pledgedSrcSize,
-                                   ZSTD_buffered_policy_e zbuff)
-{
-    DEBUGLOG(4, "ZSTD_compressBegin_internal");
-    DEBUGLOG(4, "dict ? %s", dict ? "dict" : (cdict ? "cdict" : "none"));
-    DEBUGLOG(4, "dictMode : %u", (U32)dictMode);
+                                    const void* dict, size_t dictSize,
+                                    ZSTD_dictContentType_e dictContentType,
+                                    ZSTD_dictTableLoadMethod_e dtlm,
+                                    const ZSTD_CDict* cdict,
+                                    const ZSTD_CCtx_params* params, U64 pledgedSrcSize,
+                                    ZSTD_buffered_policy_e zbuff)
+{
+    DEBUGLOG(4, "ZSTD_compressBegin_internal: wlog=%u", params->cParams.windowLog);
     /* params are supposed to be fully validated at this point */
-    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+    assert(!ZSTD_isError(ZSTD_checkCParams(params->cParams)));
     assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
-
-    if (cdict && cdict->dictContentSize>0) {
-        return ZSTD_copyCCtx_internal(cctx, cdict->refContext,
-                                      params.fParams, pledgedSrcSize,
-                                      zbuff);
+    if ( (cdict)
+      && (cdict->dictContentSize > 0)
+      && ( pledgedSrcSize < ZSTD_USE_CDICT_PARAMS_SRCSIZE_CUTOFF
+        || pledgedSrcSize < cdict->dictContentSize * ZSTD_USE_CDICT_PARAMS_DICTSIZE_MULTIPLIER
+        || pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN
+        || cdict->compressionLevel == 0)
+      && (params->attachDictPref != ZSTD_dictForceLoad) ) {
+        return ZSTD_resetCCtx_usingCDict(cctx, cdict, params, pledgedSrcSize, zbuff);
     }
 
-    CHECK_F( ZSTD_resetCCtx_internal(cctx, params, pledgedSrcSize,
-                                     ZSTDcrp_continue, zbuff) );
-    return ZSTD_compress_insertDictionary(cctx, dict, dictSize, dictMode);
+    FORWARD_IF_ERROR( ZSTD_resetCCtx_internal(cctx, *params, pledgedSrcSize,
+                                     ZSTDcrp_makeClean, zbuff) , "");
+    {   size_t const dictID = cdict ?
+                ZSTD_compress_insertDictionary(
+                        cctx->blockState.prevCBlock, &cctx->blockState.matchState,
+                        &cctx->ldmState, &cctx->workspace, &cctx->appliedParams, cdict->dictContent,
+                        cdict->dictContentSize, dictContentType, dtlm,
+                        cctx->entropyWorkspace)
+              : ZSTD_compress_insertDictionary(
+                        cctx->blockState.prevCBlock, &cctx->blockState.matchState,
+                        &cctx->ldmState, &cctx->workspace, &cctx->appliedParams, dict, dictSize,
+                        dictContentType, dtlm, cctx->entropyWorkspace);
+        FORWARD_IF_ERROR(dictID, "ZSTD_compress_insertDictionary failed");
+        assert(dictID <= UINT_MAX);
+        cctx->dictID = (U32)dictID;
+    }
+    return 0;
 }
 
-size_t ZSTD_compressBegin_advanced_internal(
-                                    ZSTD_CCtx* cctx,
+size_t ZSTD_compressBegin_advanced_internal(ZSTD_CCtx* cctx,
                                     const void* dict, size_t dictSize,
-                                    ZSTD_dictMode_e dictMode,
-                                    ZSTD_CCtx_params params,
+                                    ZSTD_dictContentType_e dictContentType,
+                                    ZSTD_dictTableLoadMethod_e dtlm,
+                                    const ZSTD_CDict* cdict,
+                                    const ZSTD_CCtx_params* params,
                                     unsigned long long pledgedSrcSize)
 {
+    DEBUGLOG(4, "ZSTD_compressBegin_advanced_internal: wlog=%u", params->cParams.windowLog);
     /* compression parameters verification and optimization */
-    CHECK_F( ZSTD_checkCParams(params.cParams) );
-    return ZSTD_compressBegin_internal(cctx, dict, dictSize, dictMode, NULL,
+    FORWARD_IF_ERROR( ZSTD_checkCParams(params->cParams) , "");
+    return ZSTD_compressBegin_internal(cctx,
+                                       dict, dictSize, dictContentType, dtlm,
+                                       cdict,
                                        params, pledgedSrcSize,
                                        ZSTDb_not_buffered);
 }
@@ -1994,19 +3144,21 @@ size_t ZSTD_compressBegin_advanced(ZSTD_CCtx* cctx,
                                    ZSTD_parameters params, unsigned long long pledgedSrcSize)
 {
     ZSTD_CCtx_params const cctxParams =
-            ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
-    return ZSTD_compressBegin_advanced_internal(cctx, dict, dictSize, ZSTD_dm_auto,
-                                                cctxParams,
-                                                pledgedSrcSize);
+            ZSTD_assignParamsToCCtxParams(&cctx->requestedParams, &params);
+    return ZSTD_compressBegin_advanced_internal(cctx,
+                                            dict, dictSize, ZSTD_dct_auto, ZSTD_dtlm_fast,
+                                            NULL /*cdict*/,
+                                            &cctxParams, pledgedSrcSize);
 }
 
 size_t ZSTD_compressBegin_usingDict(ZSTD_CCtx* cctx, const void* dict, size_t dictSize, int compressionLevel)
 {
-    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, 0, dictSize);
+    ZSTD_parameters const params = ZSTD_getParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize);
     ZSTD_CCtx_params const cctxParams =
-            ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
-    return ZSTD_compressBegin_internal(cctx, dict, dictSize, ZSTD_dm_auto, NULL,
-                                       cctxParams, 0, ZSTDb_not_buffered);
+            ZSTD_assignParamsToCCtxParams(&cctx->requestedParams, &params);
+    DEBUGLOG(4, "ZSTD_compressBegin_usingDict (dictSize=%u)", (unsigned)dictSize);
+    return ZSTD_compressBegin_internal(cctx, dict, dictSize, ZSTD_dct_auto, ZSTD_dtlm_fast, NULL,
+                                       &cctxParams, ZSTD_CONTENTSIZE_UNKNOWN, ZSTDb_not_buffered);
 }
 
 size_t ZSTD_compressBegin(ZSTD_CCtx* cctx, int compressionLevel)
@@ -2024,13 +3176,13 @@ static size_t ZSTD_writeEpilogue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity)
     BYTE* op = ostart;
     size_t fhSize = 0;
 
-    DEBUGLOG(5, "ZSTD_writeEpilogue");
-    if (cctx->stage == ZSTDcs_created) return ERROR(stage_wrong);  /* init missing */
+    DEBUGLOG(4, "ZSTD_writeEpilogue");
+    RETURN_ERROR_IF(cctx->stage == ZSTDcs_created, stage_wrong, "init missing");
 
     /* special case : empty frame */
     if (cctx->stage == ZSTDcs_init) {
-        fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, cctx->appliedParams, 0, 0);
-        if (ZSTD_isError(fhSize)) return fhSize;
+        fhSize = ZSTD_writeFrameHeader(dst, dstCapacity, &cctx->appliedParams, 0, 0);
+        FORWARD_IF_ERROR(fhSize, "ZSTD_writeFrameHeader failed");
         dstCapacity -= fhSize;
         op += fhSize;
         cctx->stage = ZSTDcs_ongoing;
@@ -2039,7 +3191,7 @@ static size_t ZSTD_writeEpilogue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity)
     if (cctx->stage != ZSTDcs_ending) {
         /* write one last empty block, make it the "last" block */
         U32 const cBlockHeader24 = 1 /* last block */ + (((U32)bt_raw)<<1) + 0;
-        if (dstCapacity<4) return ERROR(dstSize_tooSmall);
+        RETURN_ERROR_IF(dstCapacity<4, dstSize_tooSmall, "no room for epilogue");
         MEM_writeLE32(op, cBlockHeader24);
         op += ZSTD_blockHeaderSize;
         dstCapacity -= ZSTD_blockHeaderSize;
@@ -2047,7 +3199,8 @@ static size_t ZSTD_writeEpilogue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity)
 
     if (cctx->appliedParams.fParams.checksumFlag) {
         U32 const checksum = (U32) XXH64_digest(&cctx->xxhState);
-        if (dstCapacity<4) return ERROR(dstSize_tooSmall);
+        RETURN_ERROR_IF(dstCapacity<4, dstSize_tooSmall, "no room for checksum");
+        DEBUGLOG(4, "ZSTD_writeEpilogue: write checksum : %08X", (unsigned)checksum);
         MEM_writeLE32(op, checksum);
         op += 4;
     }
@@ -2056,7 +3209,6 @@ static size_t ZSTD_writeEpilogue(ZSTD_CCtx* cctx, void* dst, size_t dstCapacity)
     return op-ostart;
 }
 
-
 size_t ZSTD_compressEnd (ZSTD_CCtx* cctx,
                          void* dst, size_t dstCapacity,
                    const void* src, size_t srcSize)
@@ -2065,43 +3217,53 @@ size_t ZSTD_compressEnd (ZSTD_CCtx* cctx,
     size_t const cSize = ZSTD_compressContinue_internal(cctx,
                                 dst, dstCapacity, src, srcSize,
                                 1 /* frame mode */, 1 /* last chunk */);
-    if (ZSTD_isError(cSize)) return cSize;
+    FORWARD_IF_ERROR(cSize, "ZSTD_compressContinue_internal failed");
     endResult = ZSTD_writeEpilogue(cctx, (char*)dst + cSize, dstCapacity-cSize);
-    if (ZSTD_isError(endResult)) return endResult;
-    if (cctx->appliedParams.fParams.contentSizeFlag) {  /* control src size */
-        DEBUGLOG(5, "end of frame : controlling src size");
-        if (cctx->pledgedSrcSizePlusOne != cctx->consumedSrcSize+1) {
-            DEBUGLOG(5, "error : pledgedSrcSize = %u, while realSrcSize = %u",
-                (U32)cctx->pledgedSrcSizePlusOne-1, (U32)cctx->consumedSrcSize);
-            return ERROR(srcSize_wrong);
-    }   }
+    FORWARD_IF_ERROR(endResult, "ZSTD_writeEpilogue failed");
+    assert(!(cctx->appliedParams.fParams.contentSizeFlag && cctx->pledgedSrcSizePlusOne == 0));
+    if (cctx->pledgedSrcSizePlusOne != 0) {  /* control src size */
+        ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_UNKNOWN == (unsigned long long)-1);
+        DEBUGLOG(4, "end of frame : controlling src size");
+        RETURN_ERROR_IF(
+            cctx->pledgedSrcSizePlusOne != cctx->consumedSrcSize+1,
+            srcSize_wrong,
+             "error : pledgedSrcSize = %u, while realSrcSize = %u",
+            (unsigned)cctx->pledgedSrcSizePlusOne-1,
+            (unsigned)cctx->consumedSrcSize);
+    }
     return cSize + endResult;
 }
 
 
 static size_t ZSTD_compress_internal (ZSTD_CCtx* cctx,
-                               void* dst, size_t dstCapacity,
-                         const void* src, size_t srcSize,
-                         const void* dict,size_t dictSize,
-                               ZSTD_parameters params)
+                                      void* dst, size_t dstCapacity,
+                                const void* src, size_t srcSize,
+                                const void* dict,size_t dictSize,
+                                const ZSTD_parameters* params)
 {
     ZSTD_CCtx_params const cctxParams =
-            ZSTD_assignParamsToCCtxParams(cctx->requestedParams, params);
+            ZSTD_assignParamsToCCtxParams(&cctx->requestedParams, params);
+    DEBUGLOG(4, "ZSTD_compress_internal");
     return ZSTD_compress_advanced_internal(cctx,
-                                          dst, dstCapacity,
-                                          src, srcSize,
-                                          dict, dictSize,
-                                          cctxParams);
+                                           dst, dstCapacity,
+                                           src, srcSize,
+                                           dict, dictSize,
+                                           &cctxParams);
 }
 
-size_t ZSTD_compress_advanced (ZSTD_CCtx* ctx,
+size_t ZSTD_compress_advanced (ZSTD_CCtx* cctx,
                                void* dst, size_t dstCapacity,
                          const void* src, size_t srcSize,
                          const void* dict,size_t dictSize,
                                ZSTD_parameters params)
 {
-    CHECK_F(ZSTD_checkCParams(params.cParams));
-    return ZSTD_compress_internal(ctx, dst, dstCapacity, src, srcSize, dict, dictSize, params);
+    DEBUGLOG(4, "ZSTD_compress_advanced");
+    FORWARD_IF_ERROR(ZSTD_checkCParams(params.cParams), "");
+    return ZSTD_compress_internal(cctx,
+                                  dst, dstCapacity,
+                                  src, srcSize,
+                                  dict, dictSize,
+                                  &params);
 }
 
 /* Internal */
@@ -2110,34 +3272,47 @@ size_t ZSTD_compress_advanced_internal(
         void* dst, size_t dstCapacity,
         const void* src, size_t srcSize,
         const void* dict,size_t dictSize,
-        ZSTD_CCtx_params params)
+        const ZSTD_CCtx_params* params)
 {
-    CHECK_F( ZSTD_compressBegin_internal(cctx, dict, dictSize, ZSTD_dm_auto, NULL,
-                                         params, srcSize, ZSTDb_not_buffered) );
+    DEBUGLOG(4, "ZSTD_compress_advanced_internal (srcSize:%u)", (unsigned)srcSize);
+    FORWARD_IF_ERROR( ZSTD_compressBegin_internal(cctx,
+                         dict, dictSize, ZSTD_dct_auto, ZSTD_dtlm_fast, NULL,
+                         params, srcSize, ZSTDb_not_buffered) , "");
     return ZSTD_compressEnd(cctx, dst, dstCapacity, src, srcSize);
 }
 
-size_t ZSTD_compress_usingDict(ZSTD_CCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize,
-                               const void* dict, size_t dictSize, int compressionLevel)
+size_t ZSTD_compress_usingDict(ZSTD_CCtx* cctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize,
+                         const void* dict, size_t dictSize,
+                               int compressionLevel)
 {
-    ZSTD_parameters params = ZSTD_getParams(compressionLevel, srcSize, dict ? dictSize : 0);
-    params.fParams.contentSizeFlag = 1;
-    return ZSTD_compress_internal(ctx, dst, dstCapacity, src, srcSize, dict, dictSize, params);
+    ZSTD_parameters const params = ZSTD_getParams_internal(compressionLevel, srcSize, dict ? dictSize : 0);
+    ZSTD_CCtx_params cctxParams = ZSTD_assignParamsToCCtxParams(&cctx->requestedParams, &params);
+    DEBUGLOG(4, "ZSTD_compress_usingDict (srcSize=%u)", (unsigned)srcSize);
+    assert(params.fParams.contentSizeFlag == 1);
+    return ZSTD_compress_advanced_internal(cctx, dst, dstCapacity, src, srcSize, dict, dictSize, &cctxParams);
 }
 
-size_t ZSTD_compressCCtx (ZSTD_CCtx* ctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel)
+size_t ZSTD_compressCCtx(ZSTD_CCtx* cctx,
+                         void* dst, size_t dstCapacity,
+                   const void* src, size_t srcSize,
+                         int compressionLevel)
 {
-    return ZSTD_compress_usingDict(ctx, dst, dstCapacity, src, srcSize, NULL, 0, compressionLevel);
+    DEBUGLOG(4, "ZSTD_compressCCtx (srcSize=%u)", (unsigned)srcSize);
+    assert(cctx != NULL);
+    return ZSTD_compress_usingDict(cctx, dst, dstCapacity, src, srcSize, NULL, 0, compressionLevel);
 }
 
-size_t ZSTD_compress(void* dst, size_t dstCapacity, const void* src, size_t srcSize, int compressionLevel)
+size_t ZSTD_compress(void* dst, size_t dstCapacity,
+               const void* src, size_t srcSize,
+                     int compressionLevel)
 {
     size_t result;
     ZSTD_CCtx ctxBody;
-    memset(&ctxBody, 0, sizeof(ctxBody));
-    ctxBody.customMem = ZSTD_defaultCMem;
+    ZSTD_initCCtx(&ctxBody, ZSTD_defaultCMem);
     result = ZSTD_compressCCtx(&ctxBody, dst, dstCapacity, src, srcSize, compressionLevel);
-    ZSTD_free(ctxBody.workSpace, ZSTD_defaultCMem);  /* can't free ctxBody itself, as it's on stack; free only heap content */
+    ZSTD_freeCCtxContent(&ctxBody);   /* can't free ctxBody itself, as it's on stack; free only heap content */
     return result;
 }
 
@@ -2150,54 +3325,77 @@ size_t ZSTD_estimateCDictSize_advanced(
         size_t dictSize, ZSTD_compressionParameters cParams,
         ZSTD_dictLoadMethod_e dictLoadMethod)
 {
-    DEBUGLOG(5, "sizeof(ZSTD_CDict) : %u", (U32)sizeof(ZSTD_CDict));
-    DEBUGLOG(5, "CCtx estimate : %u",
-             (U32)ZSTD_estimateCCtxSize_advanced_usingCParams(cParams));
-    return sizeof(ZSTD_CDict) + ZSTD_estimateCCtxSize_advanced_usingCParams(cParams)
-           + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
+    DEBUGLOG(5, "sizeof(ZSTD_CDict) : %u", (unsigned)sizeof(ZSTD_CDict));
+    return ZSTD_cwksp_alloc_size(sizeof(ZSTD_CDict))
+         + ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE)
+         + ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0)
+         + (dictLoadMethod == ZSTD_dlm_byRef ? 0
+            : ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(dictSize, sizeof(void *))));
 }
 
 size_t ZSTD_estimateCDictSize(size_t dictSize, int compressionLevel)
 {
-    ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, 0, dictSize);
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize);
     return ZSTD_estimateCDictSize_advanced(dictSize, cParams, ZSTD_dlm_byCopy);
 }
 
 size_t ZSTD_sizeof_CDict(const ZSTD_CDict* cdict)
 {
     if (cdict==NULL) return 0;   /* support sizeof on NULL */
-    DEBUGLOG(5, "sizeof(*cdict) : %u", (U32)sizeof(*cdict));
-    DEBUGLOG(5, "ZSTD_sizeof_CCtx : %u", (U32)ZSTD_sizeof_CCtx(cdict->refContext));
-    return ZSTD_sizeof_CCtx(cdict->refContext) + (cdict->dictBuffer ? cdict->dictContentSize : 0) + sizeof(*cdict);
+    DEBUGLOG(5, "sizeof(*cdict) : %u", (unsigned)sizeof(*cdict));
+    /* cdict may be in the workspace */
+    return (cdict->workspace.workspace == cdict ? 0 : sizeof(*cdict))
+        + ZSTD_cwksp_sizeof(&cdict->workspace);
 }
 
 static size_t ZSTD_initCDict_internal(
                     ZSTD_CDict* cdict,
               const void* dictBuffer, size_t dictSize,
                     ZSTD_dictLoadMethod_e dictLoadMethod,
-                    ZSTD_dictMode_e dictMode,
+                    ZSTD_dictContentType_e dictContentType,
                     ZSTD_compressionParameters cParams)
 {
-    DEBUGLOG(5, "ZSTD_initCDict_internal, mode %u", (U32)dictMode);
+    DEBUGLOG(3, "ZSTD_initCDict_internal (dictContentType:%u)", (unsigned)dictContentType);
+    assert(!ZSTD_checkCParams(cParams));
+    cdict->matchState.cParams = cParams;
     if ((dictLoadMethod == ZSTD_dlm_byRef) || (!dictBuffer) || (!dictSize)) {
-        cdict->dictBuffer = NULL;
         cdict->dictContent = dictBuffer;
     } else {
-        void* const internalBuffer = ZSTD_malloc(dictSize, cdict->refContext->customMem);
-        cdict->dictBuffer = internalBuffer;
+         void *internalBuffer = ZSTD_cwksp_reserve_object(&cdict->workspace, ZSTD_cwksp_align(dictSize, sizeof(void*)));
+        RETURN_ERROR_IF(!internalBuffer, memory_allocation, "NULL pointer!");
         cdict->dictContent = internalBuffer;
-        if (!internalBuffer) return ERROR(memory_allocation);
         memcpy(internalBuffer, dictBuffer, dictSize);
     }
     cdict->dictContentSize = dictSize;
 
-    {   ZSTD_CCtx_params cctxParams = cdict->refContext->requestedParams;
-        cctxParams.cParams = cParams;
-        CHECK_F( ZSTD_compressBegin_internal(cdict->refContext,
-                                        cdict->dictContent, dictSize, dictMode,
-                                        NULL,
-                                        cctxParams, ZSTD_CONTENTSIZE_UNKNOWN,
-                                        ZSTDb_not_buffered) );
+    cdict->entropyWorkspace = (U32*)ZSTD_cwksp_reserve_object(&cdict->workspace, HUF_WORKSPACE_SIZE);
+
+
+    /* Reset the state to no dictionary */
+    ZSTD_reset_compressedBlockState(&cdict->cBlockState);
+    FORWARD_IF_ERROR(ZSTD_reset_matchState(
+        &cdict->matchState,
+        &cdict->workspace,
+        &cParams,
+        ZSTDcrp_makeClean,
+        ZSTDirp_reset,
+        ZSTD_resetTarget_CDict), "");
+    /* (Maybe) load the dictionary
+     * Skips loading the dictionary if it is < 8 bytes.
+     */
+    {   ZSTD_CCtx_params params;
+        memset(&params, 0, sizeof(params));
+        params.compressionLevel = ZSTD_CLEVEL_DEFAULT;
+        params.fParams.contentSizeFlag = 1;
+        params.cParams = cParams;
+        {   size_t const dictID = ZSTD_compress_insertDictionary(
+                    &cdict->cBlockState, &cdict->matchState, NULL, &cdict->workspace,
+                    &params, cdict->dictContent, cdict->dictContentSize,
+                    dictContentType, ZSTD_dtlm_full, cdict->entropyWorkspace);
+            FORWARD_IF_ERROR(dictID, "ZSTD_compress_insertDictionary failed");
+            assert(dictID <= (size_t)(U32)-1);
+            cdict->dictID = (U32)dictID;
+        }
     }
 
     return 0;
@@ -2205,24 +3403,38 @@ static size_t ZSTD_initCDict_internal(
 
 ZSTD_CDict* ZSTD_createCDict_advanced(const void* dictBuffer, size_t dictSize,
                                       ZSTD_dictLoadMethod_e dictLoadMethod,
-                                      ZSTD_dictMode_e dictMode,
+                                      ZSTD_dictContentType_e dictContentType,
                                       ZSTD_compressionParameters cParams, ZSTD_customMem customMem)
 {
-    DEBUGLOG(5, "ZSTD_createCDict_advanced, mode %u", (U32)dictMode);
+    DEBUGLOG(3, "ZSTD_createCDict_advanced, mode %u", (unsigned)dictContentType);
     if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
 
-    {   ZSTD_CDict* const cdict = (ZSTD_CDict*)ZSTD_malloc(sizeof(ZSTD_CDict), customMem);
-        ZSTD_CCtx* const cctx = ZSTD_createCCtx_advanced(customMem);
-
-        if (!cdict || !cctx) {
-            ZSTD_free(cdict, customMem);
-            ZSTD_freeCCtx(cctx);
+    {   size_t const workspaceSize =
+            ZSTD_cwksp_alloc_size(sizeof(ZSTD_CDict)) +
+            ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE) +
+            ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0) +
+            (dictLoadMethod == ZSTD_dlm_byRef ? 0
+             : ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(dictSize, sizeof(void*))));
+        void* const workspace = ZSTD_malloc(workspaceSize, customMem);
+        ZSTD_cwksp ws;
+        ZSTD_CDict* cdict;
+
+        if (!workspace) {
+            ZSTD_free(workspace, customMem);
             return NULL;
         }
-        cdict->refContext = cctx;
+
+        ZSTD_cwksp_init(&ws, workspace, workspaceSize);
+
+        cdict = (ZSTD_CDict*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CDict));
+        assert(cdict != NULL);
+        ZSTD_cwksp_move(&cdict->workspace, &ws);
+        cdict->customMem = customMem;
+        cdict->compressionLevel = 0; /* signals advanced API usage */
+
         if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
                                         dictBuffer, dictSize,
-                                        dictLoadMethod, dictMode,
+                                        dictLoadMethod, dictContentType,
                                         cParams) )) {
             ZSTD_freeCDict(cdict);
             return NULL;
@@ -2234,27 +3446,32 @@ ZSTD_CDict* ZSTD_createCDict_advanced(const void* dictBuffer, size_t dictSize,
 
 ZSTD_CDict* ZSTD_createCDict(const void* dict, size_t dictSize, int compressionLevel)
 {
-    ZSTD_compressionParameters cParams = ZSTD_getCParams(compressionLevel, 0, dictSize);
-    return ZSTD_createCDict_advanced(dict, dictSize,
-                                     ZSTD_dlm_byCopy, ZSTD_dm_auto,
-                                     cParams, ZSTD_defaultCMem);
+    ZSTD_compressionParameters cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize);
+    ZSTD_CDict* cdict = ZSTD_createCDict_advanced(dict, dictSize,
+                                                  ZSTD_dlm_byCopy, ZSTD_dct_auto,
+                                                  cParams, ZSTD_defaultCMem);
+    if (cdict)
+        cdict->compressionLevel = compressionLevel == 0 ? ZSTD_CLEVEL_DEFAULT : compressionLevel;
+    return cdict;
 }
 
 ZSTD_CDict* ZSTD_createCDict_byReference(const void* dict, size_t dictSize, int compressionLevel)
 {
-    ZSTD_compressionParameters cParams = ZSTD_getCParams(compressionLevel, 0, dictSize);
+    ZSTD_compressionParameters cParams = ZSTD_getCParams_internal(compressionLevel, ZSTD_CONTENTSIZE_UNKNOWN, dictSize);
     return ZSTD_createCDict_advanced(dict, dictSize,
-                                     ZSTD_dlm_byRef, ZSTD_dm_auto,
+                                     ZSTD_dlm_byRef, ZSTD_dct_auto,
                                      cParams, ZSTD_defaultCMem);
 }
 
 size_t ZSTD_freeCDict(ZSTD_CDict* cdict)
 {
     if (cdict==NULL) return 0;   /* support free on NULL */
-    {   ZSTD_customMem const cMem = cdict->refContext->customMem;
-        ZSTD_freeCCtx(cdict->refContext);
-        ZSTD_free(cdict->dictBuffer, cMem);
-        ZSTD_free(cdict, cMem);
+    {   ZSTD_customMem const cMem = cdict->customMem;
+        int cdictInWorkspace = ZSTD_cwksp_owns_buffer(&cdict->workspace, cdict);
+        ZSTD_cwksp_free(&cdict->workspace, cMem);
+        if (!cdictInWorkspace) {
+            ZSTD_free(cdict, cMem);
+        }
         return 0;
     }
 }
@@ -2272,43 +3489,48 @@ size_t ZSTD_freeCDict(ZSTD_CDict* cdict)
  *  Note : there is no corresponding "free" function.
  *         Since workspace was allocated externally, it must be freed externally.
  */
-ZSTD_CDict* ZSTD_initStaticCDict(void* workspace, size_t workspaceSize,
+const ZSTD_CDict* ZSTD_initStaticCDict(
+                                 void* workspace, size_t workspaceSize,
                            const void* dict, size_t dictSize,
                                  ZSTD_dictLoadMethod_e dictLoadMethod,
-                                 ZSTD_dictMode_e dictMode,
+                                 ZSTD_dictContentType_e dictContentType,
                                  ZSTD_compressionParameters cParams)
 {
-    size_t const cctxSize = ZSTD_estimateCCtxSize_advanced_usingCParams(cParams);
-    size_t const neededSize = sizeof(ZSTD_CDict) + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize)
-                            + cctxSize;
-    ZSTD_CDict* const cdict = (ZSTD_CDict*) workspace;
-    void* ptr;
-    DEBUGLOG(5, "(size_t)workspace & 7 : %u", (U32)(size_t)workspace & 7);
+    size_t const matchStateSize = ZSTD_sizeof_matchState(&cParams, /* forCCtx */ 0);
+    size_t const neededSize = ZSTD_cwksp_alloc_size(sizeof(ZSTD_CDict))
+                            + (dictLoadMethod == ZSTD_dlm_byRef ? 0
+                               : ZSTD_cwksp_alloc_size(ZSTD_cwksp_align(dictSize, sizeof(void*))))
+                            + ZSTD_cwksp_alloc_size(HUF_WORKSPACE_SIZE)
+                            + matchStateSize;
+    ZSTD_CDict* cdict;
+
     if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
-    DEBUGLOG(5, "(workspaceSize < neededSize) : (%u < %u) => %u",
-        (U32)workspaceSize, (U32)neededSize, (U32)(workspaceSize < neededSize));
-    if (workspaceSize < neededSize) return NULL;
 
-    if (dictLoadMethod == ZSTD_dlm_byCopy) {
-        memcpy(cdict+1, dict, dictSize);
-        dict = cdict+1;
-        ptr = (char*)workspace + sizeof(ZSTD_CDict) + dictSize;
-    } else {
-        ptr = cdict+1;
+    {
+        ZSTD_cwksp ws;
+        ZSTD_cwksp_init(&ws, workspace, workspaceSize);
+        cdict = (ZSTD_CDict*)ZSTD_cwksp_reserve_object(&ws, sizeof(ZSTD_CDict));
+        if (cdict == NULL) return NULL;
+        ZSTD_cwksp_move(&cdict->workspace, &ws);
     }
-    cdict->refContext = ZSTD_initStaticCCtx(ptr, cctxSize);
+
+    DEBUGLOG(4, "(workspaceSize < neededSize) : (%u < %u) => %u",
+        (unsigned)workspaceSize, (unsigned)neededSize, (unsigned)(workspaceSize < neededSize));
+    if (workspaceSize < neededSize) return NULL;
 
     if (ZSTD_isError( ZSTD_initCDict_internal(cdict,
                                               dict, dictSize,
-                                              ZSTD_dlm_byRef, dictMode,
+                                              dictLoadMethod, dictContentType,
                                               cParams) ))
         return NULL;
 
     return cdict;
 }
 
-ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict) {
-    return cdict->refContext->appliedParams.cParams;
+ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict)
+{
+    assert(cdict != NULL);
+    return cdict->matchState.cParams;
 }
 
 /* ZSTD_compressBegin_usingCDict_advanced() :
@@ -2317,15 +3539,32 @@ size_t ZSTD_compressBegin_usingCDict_advanced(
     ZSTD_CCtx* const cctx, const ZSTD_CDict* const cdict,
     ZSTD_frameParameters const fParams, unsigned long long const pledgedSrcSize)
 {
-    if (cdict==NULL) return ERROR(dictionary_wrong);
+    DEBUGLOG(4, "ZSTD_compressBegin_usingCDict_advanced");
+    RETURN_ERROR_IF(cdict==NULL, dictionary_wrong, "NULL pointer!");
     {   ZSTD_CCtx_params params = cctx->requestedParams;
-        params.cParams = ZSTD_getCParamsFromCDict(cdict);
+        params.cParams = ( pledgedSrcSize < ZSTD_USE_CDICT_PARAMS_SRCSIZE_CUTOFF
+                        || pledgedSrcSize < cdict->dictContentSize * ZSTD_USE_CDICT_PARAMS_DICTSIZE_MULTIPLIER
+                        || pledgedSrcSize == ZSTD_CONTENTSIZE_UNKNOWN
+                        || cdict->compressionLevel == 0 )
+                      && (params.attachDictPref != ZSTD_dictForceLoad) ?
+                ZSTD_getCParamsFromCDict(cdict)
+              : ZSTD_getCParams(cdict->compressionLevel,
+                                pledgedSrcSize,
+                                cdict->dictContentSize);
+        /* Increase window log to fit the entire dictionary and source if the
+         * source size is known. Limit the increase to 19, which is the
+         * window log for compression level 1 with the largest source size.
+         */
+        if (pledgedSrcSize != ZSTD_CONTENTSIZE_UNKNOWN) {
+            U32 const limitedSrcSize = (U32)MIN(pledgedSrcSize, 1U << 19);
+            U32 const limitedSrcLog = limitedSrcSize > 1 ? ZSTD_highbit32(limitedSrcSize - 1) + 1 : 1;
+            params.cParams.windowLog = MAX(params.cParams.windowLog, limitedSrcLog);
+        }
         params.fParams = fParams;
-        DEBUGLOG(5, "ZSTD_compressBegin_usingCDict_advanced");
         return ZSTD_compressBegin_internal(cctx,
-                                           NULL, 0, ZSTD_dm_auto,
+                                           NULL, 0, ZSTD_dct_auto, ZSTD_dtlm_fast,
                                            cdict,
-                                           params, pledgedSrcSize,
+                                           &params, pledgedSrcSize,
                                            ZSTDb_not_buffered);
     }
 }
@@ -2336,8 +3575,8 @@ size_t ZSTD_compressBegin_usingCDict_advanced(
 size_t ZSTD_compressBegin_usingCDict(ZSTD_CCtx* cctx, const ZSTD_CDict* cdict)
 {
     ZSTD_frameParameters const fParams = { 0 /*content*/, 0 /*checksum*/, 0 /*noDictID*/ };
-    DEBUGLOG(5, "ZSTD_compressBegin_usingCDict : dictIDFlag == %u", !fParams.noDictIDFlag);
-    return ZSTD_compressBegin_usingCDict_advanced(cctx, cdict, fParams, 0);
+    DEBUGLOG(4, "ZSTD_compressBegin_usingCDict : dictIDFlag == %u", !fParams.noDictIDFlag);
+    return ZSTD_compressBegin_usingCDict_advanced(cctx, cdict, fParams, ZSTD_CONTENTSIZE_UNKNOWN);
 }
 
 size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
@@ -2345,7 +3584,7 @@ size_t ZSTD_compress_usingCDict_advanced(ZSTD_CCtx* cctx,
                                 const void* src, size_t srcSize,
                                 const ZSTD_CDict* cdict, ZSTD_frameParameters fParams)
 {
-    CHECK_F (ZSTD_compressBegin_usingCDict_advanced(cctx, cdict, fParams, srcSize));   /* will check if cdict != NULL */
+    FORWARD_IF_ERROR(ZSTD_compressBegin_usingCDict_advanced(cctx, cdict, fParams, srcSize), "");   /* will check if cdict != NULL */
     return ZSTD_compressEnd(cctx, dst, dstCapacity, src, srcSize);
 }
 
@@ -2371,6 +3610,7 @@ size_t ZSTD_compress_usingCDict(ZSTD_CCtx* cctx,
 
 ZSTD_CStream* ZSTD_createCStream(void)
 {
+    DEBUGLOG(3, "ZSTD_createCStream");
     return ZSTD_createCStream_advanced(ZSTD_defaultCMem);
 }
 
@@ -2400,76 +3640,71 @@ size_t ZSTD_CStreamOutSize(void)
     return ZSTD_compressBound(ZSTD_BLOCKSIZE_MAX) + ZSTD_blockHeaderSize + 4 /* 32-bits hash */ ;
 }
 
-static size_t ZSTD_resetCStream_internal(ZSTD_CStream* zcs,
-                    const void* dict, size_t dictSize, ZSTD_dictMode_e dictMode,
-                    const ZSTD_CDict* cdict,
-                    const ZSTD_CCtx_params params, unsigned long long pledgedSrcSize)
+static size_t ZSTD_resetCStream_internal(ZSTD_CStream* cctx,
+                    const void* const dict, size_t const dictSize, ZSTD_dictContentType_e const dictContentType,
+                    const ZSTD_CDict* const cdict,
+                    ZSTD_CCtx_params params, unsigned long long const pledgedSrcSize)
 {
     DEBUGLOG(4, "ZSTD_resetCStream_internal");
+    /* Finalize the compression parameters */
+    params.cParams = ZSTD_getCParamsFromCCtxParams(&params, pledgedSrcSize, dictSize);
     /* params are supposed to be fully validated at this point */
     assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
     assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
 
-    CHECK_F( ZSTD_compressBegin_internal(zcs,
-                                        dict, dictSize, dictMode,
-                                        cdict,
-                                        params, pledgedSrcSize,
-                                        ZSTDb_buffered) );
-
-    zcs->inToCompress = 0;
-    zcs->inBuffPos = 0;
-    zcs->inBuffTarget = zcs->blockSize;
-    zcs->outBuffContentSize = zcs->outBuffFlushedSize = 0;
-    zcs->streamStage = zcss_load;
-    zcs->frameEnded = 0;
+    FORWARD_IF_ERROR( ZSTD_compressBegin_internal(cctx,
+                                         dict, dictSize, dictContentType, ZSTD_dtlm_fast,
+                                         cdict,
+                                         &params, pledgedSrcSize,
+                                         ZSTDb_buffered) , "");
+
+    cctx->inToCompress = 0;
+    cctx->inBuffPos = 0;
+    cctx->inBuffTarget = cctx->blockSize
+                      + (cctx->blockSize == pledgedSrcSize);   /* for small input: avoid automatic flush on reaching end of block, since it would require to add a 3-bytes null block to end frame */
+    cctx->outBuffContentSize = cctx->outBuffFlushedSize = 0;
+    cctx->streamStage = zcss_load;
+    cctx->frameEnded = 0;
     return 0;   /* ready to go */
 }
 
-size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pledgedSrcSize)
+/* ZSTD_resetCStream():
+ * pledgedSrcSize == 0 means "unknown" */
+size_t ZSTD_resetCStream(ZSTD_CStream* zcs, unsigned long long pss)
 {
-    ZSTD_CCtx_params params = zcs->requestedParams;
-    params.fParams.contentSizeFlag = (pledgedSrcSize > 0);
-    params.cParams = ZSTD_getCParamsFromCCtxParams(params, pledgedSrcSize, 0);
-    DEBUGLOG(5, "ZSTD_resetCStream");
-    return ZSTD_resetCStream_internal(zcs, NULL, 0, ZSTD_dm_auto, zcs->cdict, params, pledgedSrcSize);
+    /* temporary : 0 interpreted as "unknown" during transition period.
+     * Users willing to specify "unknown" **must** use ZSTD_CONTENTSIZE_UNKNOWN.
+     * 0 will be interpreted as "empty" in the future.
+     */
+    U64 const pledgedSrcSize = (pss==0) ? ZSTD_CONTENTSIZE_UNKNOWN : pss;
+    DEBUGLOG(4, "ZSTD_resetCStream: pledgedSrcSize = %u", (unsigned)pledgedSrcSize);
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) , "");
+    return 0;
 }
 
 /*! ZSTD_initCStream_internal() :
- *  Note : not static, but hidden (not exposed). Used by zstdmt_compress.c
+ *  Note : for lib/compress only. Used by zstdmt_compress.c.
  *  Assumption 1 : params are valid
  *  Assumption 2 : either dict, or cdict, is defined, not both */
 size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs,
                     const void* dict, size_t dictSize, const ZSTD_CDict* cdict,
-                    ZSTD_CCtx_params params, unsigned long long pledgedSrcSize)
+                    const ZSTD_CCtx_params* params,
+                    unsigned long long pledgedSrcSize)
 {
-    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
+    DEBUGLOG(4, "ZSTD_initCStream_internal");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) , "");
+    assert(!ZSTD_isError(ZSTD_checkCParams(params->cParams)));
+    zcs->requestedParams = *params;
     assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
-
-    if (dict && dictSize >= 8) {
-        DEBUGLOG(5, "loading dictionary of size %u", (U32)dictSize);
-        if (zcs->staticSize) {   /* static CCtx : never uses malloc */
-            /* incompatible with internal cdict creation */
-            return ERROR(memory_allocation);
-        }
-        ZSTD_freeCDict(zcs->cdictLocal);
-        zcs->cdictLocal = ZSTD_createCDict_advanced(dict, dictSize,
-                                            ZSTD_dlm_byCopy, ZSTD_dm_auto,
-                                            params.cParams, zcs->customMem);
-        zcs->cdict = zcs->cdictLocal;
-        if (zcs->cdictLocal == NULL) return ERROR(memory_allocation);
+    if (dict) {
+        FORWARD_IF_ERROR( ZSTD_CCtx_loadDictionary(zcs, dict, dictSize) , "");
     } else {
-        if (cdict) {
-            params.cParams = ZSTD_getCParamsFromCDict(cdict);  /* cParams are enforced from cdict */
-        }
-        ZSTD_freeCDict(zcs->cdictLocal);
-        zcs->cdictLocal = NULL;
-        zcs->cdict = cdict;
+        /* Dictionary is cleared if !cdict */
+        FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, cdict) , "");
     }
-
-    params.compressionLevel = ZSTD_CLEVEL_CUSTOM;
-    zcs->requestedParams = params;
-
-    return ZSTD_resetCStream_internal(zcs, NULL, 0, ZSTD_dm_auto, zcs->cdict, params, pledgedSrcSize);
+    return 0;
 }
 
 /* ZSTD_initCStream_usingCDict_advanced() :
@@ -2478,88 +3713,112 @@ size_t ZSTD_initCStream_usingCDict_advanced(ZSTD_CStream* zcs,
                                             const ZSTD_CDict* cdict,
                                             ZSTD_frameParameters fParams,
                                             unsigned long long pledgedSrcSize)
-{   /* cannot handle NULL cdict (does not know what to do) */
-    if (!cdict) return ERROR(dictionary_wrong);
-    {   ZSTD_CCtx_params params = zcs->requestedParams;
-        params.cParams = ZSTD_getCParamsFromCDict(cdict);
-        params.fParams = fParams;
-        return ZSTD_initCStream_internal(zcs,
-                                NULL, 0, cdict,
-                                params, pledgedSrcSize);
-    }
+{
+    DEBUGLOG(4, "ZSTD_initCStream_usingCDict_advanced");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) , "");
+    zcs->requestedParams.fParams = fParams;
+    FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, cdict) , "");
+    return 0;
 }
 
 /* note : cdict must outlive compression session */
 size_t ZSTD_initCStream_usingCDict(ZSTD_CStream* zcs, const ZSTD_CDict* cdict)
 {
-    ZSTD_frameParameters const fParams = { 0 /* contentSize */, 0 /* checksum */, 0 /* hideDictID */ };
-    return ZSTD_initCStream_usingCDict_advanced(zcs, cdict, fParams, 0);  /* note : will check that cdict != NULL */
+    DEBUGLOG(4, "ZSTD_initCStream_usingCDict");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, cdict) , "");
+    return 0;
 }
 
+
+/* ZSTD_initCStream_advanced() :
+ * pledgedSrcSize must be exact.
+ * if srcSize is not known at init time, use value ZSTD_CONTENTSIZE_UNKNOWN.
+ * dict is loaded with default parameters ZSTD_dct_auto and ZSTD_dlm_byCopy. */
 size_t ZSTD_initCStream_advanced(ZSTD_CStream* zcs,
                                  const void* dict, size_t dictSize,
-                                 ZSTD_parameters params, unsigned long long pledgedSrcSize)
+                                 ZSTD_parameters params, unsigned long long pss)
 {
-    ZSTD_CCtx_params const cctxParams =
-            ZSTD_assignParamsToCCtxParams(zcs->requestedParams, params);
-    CHECK_F( ZSTD_checkCParams(params.cParams) );
-    return ZSTD_initCStream_internal(zcs, dict, dictSize, NULL, cctxParams, pledgedSrcSize);
+    /* for compatibility with older programs relying on this behavior.
+     * Users should now specify ZSTD_CONTENTSIZE_UNKNOWN.
+     * This line will be removed in the future.
+     */
+    U64 const pledgedSrcSize = (pss==0 && params.fParams.contentSizeFlag==0) ? ZSTD_CONTENTSIZE_UNKNOWN : pss;
+    DEBUGLOG(4, "ZSTD_initCStream_advanced");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) , "");
+    FORWARD_IF_ERROR( ZSTD_checkCParams(params.cParams) , "");
+    zcs->requestedParams = ZSTD_assignParamsToCCtxParams(&zcs->requestedParams, &params);
+    FORWARD_IF_ERROR( ZSTD_CCtx_loadDictionary(zcs, dict, dictSize) , "");
+    return 0;
 }
 
 size_t ZSTD_initCStream_usingDict(ZSTD_CStream* zcs, const void* dict, size_t dictSize, int compressionLevel)
 {
-    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, 0, dictSize);
-    ZSTD_CCtx_params const cctxParams =
-            ZSTD_assignParamsToCCtxParams(zcs->requestedParams, params);
-    return ZSTD_initCStream_internal(zcs, dict, dictSize, NULL, cctxParams, 0);
+    DEBUGLOG(4, "ZSTD_initCStream_usingDict");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_loadDictionary(zcs, dict, dictSize) , "");
+    return 0;
 }
 
-size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs, int compressionLevel, unsigned long long pledgedSrcSize)
+size_t ZSTD_initCStream_srcSize(ZSTD_CStream* zcs, int compressionLevel, unsigned long long pss)
 {
-    ZSTD_CCtx_params cctxParams;
-    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, pledgedSrcSize, 0);
-    cctxParams = ZSTD_assignParamsToCCtxParams(zcs->requestedParams, params);
-    cctxParams.fParams.contentSizeFlag = (pledgedSrcSize>0);
-    return ZSTD_initCStream_internal(zcs, NULL, 0, NULL, cctxParams, pledgedSrcSize);
+    /* temporary : 0 interpreted as "unknown" during transition period.
+     * Users willing to specify "unknown" **must** use ZSTD_CONTENTSIZE_UNKNOWN.
+     * 0 will be interpreted as "empty" in the future.
+     */
+    U64 const pledgedSrcSize = (pss==0) ? ZSTD_CONTENTSIZE_UNKNOWN : pss;
+    DEBUGLOG(4, "ZSTD_initCStream_srcSize");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, NULL) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setPledgedSrcSize(zcs, pledgedSrcSize) , "");
+    return 0;
 }
 
 size_t ZSTD_initCStream(ZSTD_CStream* zcs, int compressionLevel)
 {
-    return ZSTD_initCStream_srcSize(zcs, compressionLevel, 0);
+    DEBUGLOG(4, "ZSTD_initCStream");
+    FORWARD_IF_ERROR( ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_refCDict(zcs, NULL) , "");
+    FORWARD_IF_ERROR( ZSTD_CCtx_setParameter(zcs, ZSTD_c_compressionLevel, compressionLevel) , "");
+    return 0;
 }
 
 /*======   Compression   ======*/
 
-MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity,
-                           const void* src, size_t srcSize)
+static size_t ZSTD_nextInputSizeHint(const ZSTD_CCtx* cctx)
 {
-    size_t const length = MIN(dstCapacity, srcSize);
-    if (length) memcpy(dst, src, length);
-    return length;
+    size_t hintInSize = cctx->inBuffTarget - cctx->inBuffPos;
+    if (hintInSize==0) hintInSize = cctx->blockSize;
+    return hintInSize;
 }
 
 /** ZSTD_compressStream_generic():
- *  internal function for all *compressStream*() variants and *compress_generic()
+ *  internal function for all *compressStream*() variants
+ *  non-static, because can be called from zstdmt_compress.c
  * @return : hint size for next input */
-size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
-                                   ZSTD_outBuffer* output,
-                                   ZSTD_inBuffer* input,
-                                   ZSTD_EndDirective const flushMode)
+static size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
+                                          ZSTD_outBuffer* output,
+                                          ZSTD_inBuffer* input,
+                                          ZSTD_EndDirective const flushMode)
 {
     const char* const istart = (const char*)input->src;
-    const char* const iend = istart + input->size;
-    const char* ip = istart + input->pos;
+    const char* const iend = input->size != 0 ? istart + input->size : istart;
+    const char* ip = input->pos != 0 ? istart + input->pos : istart;
     char* const ostart = (char*)output->dst;
-    char* const oend = ostart + output->size;
-    char* op = ostart + output->pos;
+    char* const oend = output->size != 0 ? ostart + output->size : ostart;
+    char* op = output->pos != 0 ? ostart + output->pos : ostart;
     U32 someMoreWork = 1;
 
     /* check expectations */
-    DEBUGLOG(5, "ZSTD_compressStream_generic, flush=%u", (U32)flushMode);
+    DEBUGLOG(5, "ZSTD_compressStream_generic, flush=%u", (unsigned)flushMode);
     assert(zcs->inBuff != NULL);
-    assert(zcs->inBuffSize>0);
-    assert(zcs->outBuff!= NULL);
-    assert(zcs->outBuffSize>0);
+    assert(zcs->inBuffSize > 0);
+    assert(zcs->outBuff !=  NULL);
+    assert(zcs->outBuffSize > 0);
     assert(output->pos <= output->size);
     assert(input->pos <= input->size);
 
@@ -2567,8 +3826,7 @@ size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
         switch(zcs->streamStage)
         {
         case zcss_init:
-            /* call ZSTD_initCStream() first ! */
-            return ERROR(init_missing);
+            RETURN_ERROR(init_missing, "call ZSTD_initCStream() first!");
 
         case zcss_load:
             if ( (flushMode == ZSTD_e_end)
@@ -2577,12 +3835,12 @@ size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                 /* shortcut to compression pass directly into output buffer */
                 size_t const cSize = ZSTD_compressEnd(zcs,
                                                 op, oend-op, ip, iend-ip);
-                DEBUGLOG(4, "ZSTD_compressEnd : %u", (U32)cSize);
-                if (ZSTD_isError(cSize)) return cSize;
+                DEBUGLOG(4, "ZSTD_compressEnd : cSize=%u", (unsigned)cSize);
+                FORWARD_IF_ERROR(cSize, "ZSTD_compressEnd failed");
                 ip = iend;
                 op += cSize;
                 zcs->frameEnded = 1;
-                ZSTD_startNewCompression(zcs);
+                ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
                 someMoreWork = 0; break;
             }
             /* complete loading into inBuffer */
@@ -2591,7 +3849,8 @@ size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                                         zcs->inBuff + zcs->inBuffPos, toLoad,
                                         ip, iend-ip);
                 zcs->inBuffPos += loaded;
-                ip += loaded;
+                if (loaded != 0)
+                    ip += loaded;
                 if ( (flushMode == ZSTD_e_continue)
                   && (zcs->inBuffPos < zcs->inBuffTarget) ) {
                     /* not enough input to fill full block : stop here */
@@ -2619,14 +3878,14 @@ size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                                     zcs->inBuff + zcs->inToCompress, iSize) :
                         ZSTD_compressContinue(zcs, cDst, oSize,
                                     zcs->inBuff + zcs->inToCompress, iSize);
-                if (ZSTD_isError(cSize)) return cSize;
+                FORWARD_IF_ERROR(cSize, "%s", lastBlock ? "ZSTD_compressEnd failed" : "ZSTD_compressContinue failed");
                 zcs->frameEnded = lastBlock;
                 /* prepare next block */
                 zcs->inBuffTarget = zcs->inBuffPos + zcs->blockSize;
                 if (zcs->inBuffTarget > zcs->inBuffSize)
                     zcs->inBuffPos = 0, zcs->inBuffTarget = zcs->blockSize;
                 DEBUGLOG(5, "inBuffTarget:%u / inBuffSize:%u",
-                         (U32)zcs->inBuffTarget, (U32)zcs->inBuffSize);
+                         (unsigned)zcs->inBuffTarget, (unsigned)zcs->inBuffSize);
                 if (!lastBlock)
                     assert(zcs->inBuffTarget <= zcs->inBuffSize);
                 zcs->inToCompress = zcs->inBuffPos;
@@ -2635,7 +3894,7 @@ size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                     if (zcs->frameEnded) {
                         DEBUGLOG(5, "Frame completed directly in outBuffer");
                         someMoreWork = 0;
-                        ZSTD_startNewCompression(zcs);
+                        ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
                     }
                     break;
                 }
@@ -2647,11 +3906,12 @@ size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
         case zcss_flush:
             DEBUGLOG(5, "flush stage");
             {   size_t const toFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize;
-                size_t const flushed = ZSTD_limitCopy(op, oend-op,
+                size_t const flushed = ZSTD_limitCopy(op, (size_t)(oend-op),
                             zcs->outBuff + zcs->outBuffFlushedSize, toFlush);
                 DEBUGLOG(5, "toFlush: %u into %u ==> flushed: %u",
-                            (U32)toFlush, (U32)(oend-op), (U32)flushed);
-                op += flushed;
+                            (unsigned)toFlush, (unsigned)(oend-op), (unsigned)flushed);
+                if (flushed)
+                    op += flushed;
                 zcs->outBuffFlushedSize += flushed;
                 if (toFlush!=flushed) {
                     /* flush not fully completed, presumably because dst is too small */
@@ -2663,7 +3923,7 @@ size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
                 if (zcs->frameEnded) {
                     DEBUGLOG(5, "Frame completed on flush");
                     someMoreWork = 0;
-                    ZSTD_startNewCompression(zcs);
+                    ZSTD_CCtx_reset(zcs, ZSTD_reset_session_only);
                     break;
                 }
                 zcs->streamStage = zcss_load;
@@ -2678,82 +3938,115 @@ size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
     input->pos = ip - istart;
     output->pos = op - ostart;
     if (zcs->frameEnded) return 0;
-    {   size_t hintInSize = zcs->inBuffTarget - zcs->inBuffPos;
-        if (hintInSize==0) hintInSize = zcs->blockSize;
-        return hintInSize;
+    return ZSTD_nextInputSizeHint(zcs);
+}
+
+static size_t ZSTD_nextInputSizeHint_MTorST(const ZSTD_CCtx* cctx)
+{
+#ifdef ZSTD_MULTITHREAD
+    if (cctx->appliedParams.nbWorkers >= 1) {
+        assert(cctx->mtctx != NULL);
+        return ZSTDMT_nextInputSizeHint(cctx->mtctx);
     }
+#endif
+    return ZSTD_nextInputSizeHint(cctx);
+
 }
 
 size_t ZSTD_compressStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
 {
-    /* check conditions */
-    if (output->pos > output->size) return ERROR(GENERIC);
-    if (input->pos  > input->size)  return ERROR(GENERIC);
-
-    return ZSTD_compressStream_generic(zcs, output, input, ZSTD_e_continue);
+    FORWARD_IF_ERROR( ZSTD_compressStream2(zcs, output, input, ZSTD_e_continue) , "");
+    return ZSTD_nextInputSizeHint_MTorST(zcs);
 }
 
 
-size_t ZSTD_compress_generic (ZSTD_CCtx* cctx,
-                              ZSTD_outBuffer* output,
-                              ZSTD_inBuffer* input,
-                              ZSTD_EndDirective endOp)
+size_t ZSTD_compressStream2( ZSTD_CCtx* cctx,
+                             ZSTD_outBuffer* output,
+                             ZSTD_inBuffer* input,
+                             ZSTD_EndDirective endOp)
 {
+    DEBUGLOG(5, "ZSTD_compressStream2, endOp=%u ", (unsigned)endOp);
     /* check conditions */
-    if (output->pos > output->size) return ERROR(GENERIC);
-    if (input->pos  > input->size)  return ERROR(GENERIC);
+    RETURN_ERROR_IF(output->pos > output->size, GENERIC, "invalid buffer");
+    RETURN_ERROR_IF(input->pos  > input->size, GENERIC, "invalid buffer");
     assert(cctx!=NULL);
 
     /* transparent initialization stage */
     if (cctx->streamStage == zcss_init) {
-        ZSTD_prefixDict const prefixDict = cctx->prefixDict;
         ZSTD_CCtx_params params = cctx->requestedParams;
+        ZSTD_prefixDict const prefixDict = cctx->prefixDict;
+        FORWARD_IF_ERROR( ZSTD_initLocalDict(cctx) , ""); /* Init the local dict if present. */
+        memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));   /* single usage */
+        assert(prefixDict.dict==NULL || cctx->cdict==NULL);    /* only one can be set */
+        DEBUGLOG(4, "ZSTD_compressStream2 : transparent init stage");
+        if (endOp == ZSTD_e_end) cctx->pledgedSrcSizePlusOne = input->size + 1;  /* auto-fix pledgedSrcSize */
         params.cParams = ZSTD_getCParamsFromCCtxParams(
-                cctx->requestedParams, cctx->pledgedSrcSizePlusOne-1, 0 /*dictSize*/);
-        memset(&cctx->prefixDict, 0, sizeof(cctx->prefixDict));  /* single usage */
-        assert(prefixDict.dict==NULL || cctx->cdict==NULL);   /* only one can be set */
+                &cctx->requestedParams, cctx->pledgedSrcSizePlusOne-1, 0 /*dictSize*/);
+
 
 #ifdef ZSTD_MULTITHREAD
-        if (params.nbThreads > 1) {
-            if (cctx->mtctx == NULL || cctx->appliedParams.nbThreads != params.nbThreads) {
-                ZSTDMT_freeCCtx(cctx->mtctx);
-                cctx->mtctx = ZSTDMT_createCCtx_advanced(params.nbThreads, cctx->customMem);
-                if (cctx->mtctx == NULL) return ERROR(memory_allocation);
+        if ((cctx->pledgedSrcSizePlusOne-1) <= ZSTDMT_JOBSIZE_MIN) {
+            params.nbWorkers = 0; /* do not invoke multi-threading when src size is too small */
+        }
+        if (params.nbWorkers > 0) {
+            /* mt context creation */
+            if (cctx->mtctx == NULL) {
+                DEBUGLOG(4, "ZSTD_compressStream2: creating new mtctx for nbWorkers=%u",
+                            params.nbWorkers);
+                cctx->mtctx = ZSTDMT_createCCtx_advanced((U32)params.nbWorkers, cctx->customMem);
+                RETURN_ERROR_IF(cctx->mtctx == NULL, memory_allocation, "NULL pointer!");
             }
-            DEBUGLOG(4, "call ZSTDMT_initCStream_internal as nbThreads=%u", params.nbThreads);
-            CHECK_F( ZSTDMT_initCStream_internal(
-                             cctx->mtctx,
-                             prefixDict.dict, prefixDict.dictSize, ZSTD_dm_rawContent,
-                             cctx->cdict, params, cctx->pledgedSrcSizePlusOne-1) );
+            /* mt compression */
+            DEBUGLOG(4, "call ZSTDMT_initCStream_internal as nbWorkers=%u", params.nbWorkers);
+            FORWARD_IF_ERROR( ZSTDMT_initCStream_internal(
+                        cctx->mtctx,
+                        prefixDict.dict, prefixDict.dictSize, prefixDict.dictContentType,
+                        cctx->cdict, params, cctx->pledgedSrcSizePlusOne-1) , "");
             cctx->streamStage = zcss_load;
-            cctx->appliedParams.nbThreads = params.nbThreads;
+            cctx->appliedParams.nbWorkers = params.nbWorkers;
         } else
 #endif
-        {
-            CHECK_F( ZSTD_resetCStream_internal(
-                             cctx, prefixDict.dict, prefixDict.dictSize,
-                             prefixDict.dictMode, cctx->cdict, params,
-                             cctx->pledgedSrcSizePlusOne-1) );
+        {   FORWARD_IF_ERROR( ZSTD_resetCStream_internal(cctx,
+                            prefixDict.dict, prefixDict.dictSize, prefixDict.dictContentType,
+                            cctx->cdict,
+                            params, cctx->pledgedSrcSizePlusOne-1) , "");
+            assert(cctx->streamStage == zcss_load);
+            assert(cctx->appliedParams.nbWorkers == 0);
     }   }
+    /* end of transparent initialization stage */
 
     /* compression stage */
 #ifdef ZSTD_MULTITHREAD
-    if (cctx->appliedParams.nbThreads > 1) {
-        size_t const flushMin = ZSTDMT_compressStream_generic(cctx->mtctx, output, input, endOp);
-        DEBUGLOG(5, "ZSTDMT_compressStream_generic : %u", (U32)flushMin);
-        if ( ZSTD_isError(flushMin)
-          || (endOp == ZSTD_e_end && flushMin == 0) ) { /* compression completed */
-            ZSTD_startNewCompression(cctx);
+    if (cctx->appliedParams.nbWorkers > 0) {
+        int const forceMaxProgress = (endOp == ZSTD_e_flush || endOp == ZSTD_e_end);
+        size_t flushMin;
+        assert(forceMaxProgress || endOp == ZSTD_e_continue /* Protection for a new flush type */);
+        if (cctx->cParamsChanged) {
+            ZSTDMT_updateCParams_whileCompressing(cctx->mtctx, &cctx->requestedParams);
+            cctx->cParamsChanged = 0;
         }
+        do {
+            flushMin = ZSTDMT_compressStream_generic(cctx->mtctx, output, input, endOp);
+            if ( ZSTD_isError(flushMin)
+              || (endOp == ZSTD_e_end && flushMin == 0) ) { /* compression completed */
+                ZSTD_CCtx_reset(cctx, ZSTD_reset_session_only);
+            }
+            FORWARD_IF_ERROR(flushMin, "ZSTDMT_compressStream_generic failed");
+        } while (forceMaxProgress && flushMin != 0 && output->pos < output->size);
+        DEBUGLOG(5, "completed ZSTD_compressStream2 delegating to ZSTDMT_compressStream_generic");
+        /* Either we don't require maximum forward progress, we've finished the
+         * flush, or we are out of output space.
+         */
+        assert(!forceMaxProgress || flushMin == 0 || output->pos == output->size);
         return flushMin;
     }
 #endif
-    CHECK_F( ZSTD_compressStream_generic(cctx, output, input, endOp) );
-    DEBUGLOG(5, "completed ZSTD_compress_generic");
+    FORWARD_IF_ERROR( ZSTD_compressStream_generic(cctx, output, input, endOp) , "");
+    DEBUGLOG(5, "completed ZSTD_compressStream2");
     return cctx->outBuffContentSize - cctx->outBuffFlushedSize; /* remaining to flush */
 }
 
-size_t ZSTD_compress_generic_simpleArgs (
+size_t ZSTD_compressStream2_simpleArgs (
                             ZSTD_CCtx* cctx,
                             void* dst, size_t dstCapacity, size_t* dstPos,
                       const void* src, size_t srcSize, size_t* srcPos,
@@ -2761,37 +4054,57 @@ size_t ZSTD_compress_generic_simpleArgs (
 {
     ZSTD_outBuffer output = { dst, dstCapacity, *dstPos };
     ZSTD_inBuffer  input  = { src, srcSize, *srcPos };
-    /* ZSTD_compress_generic() will check validity of dstPos and srcPos */
-    size_t const cErr = ZSTD_compress_generic(cctx, &output, &input, endOp);
+    /* ZSTD_compressStream2() will check validity of dstPos and srcPos */
+    size_t const cErr = ZSTD_compressStream2(cctx, &output, &input, endOp);
     *dstPos = output.pos;
     *srcPos = input.pos;
     return cErr;
 }
 
+size_t ZSTD_compress2(ZSTD_CCtx* cctx,
+                      void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    DEBUGLOG(4, "ZSTD_compress2 (srcSize=%u)", (unsigned)srcSize);
+    ZSTD_CCtx_reset(cctx, ZSTD_reset_session_only);
+    {   size_t oPos = 0;
+        size_t iPos = 0;
+        size_t const result = ZSTD_compressStream2_simpleArgs(cctx,
+                                        dst, dstCapacity, &oPos,
+                                        src, srcSize, &iPos,
+                                        ZSTD_e_end);
+        FORWARD_IF_ERROR(result, "ZSTD_compressStream2_simpleArgs failed");
+        if (result != 0) {  /* compression not completed, due to lack of output space */
+            assert(oPos == dstCapacity);
+            RETURN_ERROR(dstSize_tooSmall, "");
+        }
+        assert(iPos == srcSize);   /* all input is expected consumed */
+        return oPos;
+    }
+}
 
 /*======   Finalize   ======*/
 
 /*! ZSTD_flushStream() :
-*   @return : amount of data remaining to flush */
+ * @return : amount of data remaining to flush */
 size_t ZSTD_flushStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
 {
     ZSTD_inBuffer input = { NULL, 0, 0 };
-    if (output->pos > output->size) return ERROR(GENERIC);
-    CHECK_F( ZSTD_compressStream_generic(zcs, output, &input, ZSTD_e_flush) );
-    return zcs->outBuffContentSize - zcs->outBuffFlushedSize;  /* remaining to flush */
+    return ZSTD_compressStream2(zcs, output, &input, ZSTD_e_flush);
 }
 
 
 size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
 {
     ZSTD_inBuffer input = { NULL, 0, 0 };
-    if (output->pos > output->size) return ERROR(GENERIC);
-    CHECK_F( ZSTD_compressStream_generic(zcs, output, &input, ZSTD_e_end) );
+    size_t const remainingToFlush = ZSTD_compressStream2(zcs, output, &input, ZSTD_e_end);
+    FORWARD_IF_ERROR( remainingToFlush , "ZSTD_compressStream2 failed");
+    if (zcs->appliedParams.nbWorkers > 0) return remainingToFlush;   /* minimal estimation */
+    /* single thread mode : attempt to calculate remaining to flush more precisely */
     {   size_t const lastBlockSize = zcs->frameEnded ? 0 : ZSTD_BLOCKHEADERSIZE;
-        size_t const checksumSize = zcs->frameEnded ? 0 : zcs->appliedParams.fParams.checksumFlag * 4;
-        size_t const toFlush = zcs->outBuffContentSize - zcs->outBuffFlushedSize + lastBlockSize + checksumSize;
-        DEBUGLOG(5, "ZSTD_endStream : remaining to flush : %u",
-                (unsigned)toFlush);
+        size_t const checksumSize = (size_t)(zcs->frameEnded ? 0 : zcs->appliedParams.fParams.checksumFlag * 4);
+        size_t const toFlush = remainingToFlush + lastBlockSize + checksumSize;
+        DEBUGLOG(4, "ZSTD_endStream : remaining to flush : %u", (unsigned)toFlush);
         return toFlush;
     }
 }
@@ -2801,163 +4114,165 @@ size_t ZSTD_endStream(ZSTD_CStream* zcs, ZSTD_outBuffer* output)
 
 #define ZSTD_MAX_CLEVEL     22
 int ZSTD_maxCLevel(void) { return ZSTD_MAX_CLEVEL; }
+int ZSTD_minCLevel(void) { return (int)-ZSTD_TARGETLENGTH_MAX; }
 
 static const ZSTD_compressionParameters ZSTD_defaultCParameters[4][ZSTD_MAX_CLEVEL+1] = {
-{   /* "default" - guarantees a monotonically increasing memory budget */
+{   /* "default" - for any srcSize > 256 KB */
     /* W,  C,  H,  S,  L, TL, strat */
-    { 18, 12, 12,  1,  7, 16, ZSTD_fast    },  /* level  0 - never used */
-    { 19, 13, 14,  1,  7, 16, ZSTD_fast    },  /* level  1 */
-    { 19, 15, 16,  1,  6, 16, ZSTD_fast    },  /* level  2 */
-    { 20, 16, 17,  1,  5, 16, ZSTD_dfast   },  /* level  3 */
-    { 20, 17, 18,  1,  5, 16, ZSTD_dfast   },  /* level  4 */
-    { 20, 17, 18,  2,  5, 16, ZSTD_greedy  },  /* level  5 */
-    { 21, 17, 19,  2,  5, 16, ZSTD_lazy    },  /* level  6 */
-    { 21, 18, 19,  3,  5, 16, ZSTD_lazy    },  /* level  7 */
-    { 21, 18, 20,  3,  5, 16, ZSTD_lazy2   },  /* level  8 */
-    { 21, 19, 20,  3,  5, 16, ZSTD_lazy2   },  /* level  9 */
-    { 21, 19, 21,  4,  5, 16, ZSTD_lazy2   },  /* level 10 */
-    { 22, 20, 22,  4,  5, 16, ZSTD_lazy2   },  /* level 11 */
-    { 22, 20, 22,  5,  5, 16, ZSTD_lazy2   },  /* level 12 */
-    { 22, 21, 22,  5,  5, 16, ZSTD_lazy2   },  /* level 13 */
-    { 22, 21, 22,  6,  5, 16, ZSTD_lazy2   },  /* level 14 */
-    { 22, 21, 22,  5,  5, 16, ZSTD_btlazy2 },  /* level 15 */
-    { 23, 22, 22,  5,  5, 16, ZSTD_btlazy2 },  /* level 16 */
-    { 23, 22, 22,  4,  5, 24, ZSTD_btopt   },  /* level 17 */
-    { 23, 22, 22,  5,  4, 32, ZSTD_btopt   },  /* level 18 */
-    { 23, 23, 22,  6,  3, 48, ZSTD_btopt   },  /* level 19 */
-    { 25, 25, 23,  7,  3, 64, ZSTD_btultra },  /* level 20 */
-    { 26, 26, 24,  7,  3,256, ZSTD_btultra },  /* level 21 */
-    { 27, 27, 25,  9,  3,512, ZSTD_btultra },  /* level 22 */
+    { 19, 12, 13,  1,  6,  1, ZSTD_fast    },  /* base for negative levels */
+    { 19, 13, 14,  1,  7,  0, ZSTD_fast    },  /* level  1 */
+    { 20, 15, 16,  1,  6,  0, ZSTD_fast    },  /* level  2 */
+    { 21, 16, 17,  1,  5,  0, ZSTD_dfast   },  /* level  3 */
+    { 21, 18, 18,  1,  5,  0, ZSTD_dfast   },  /* level  4 */
+    { 21, 18, 19,  2,  5,  2, ZSTD_greedy  },  /* level  5 */
+    { 21, 19, 19,  3,  5,  4, ZSTD_greedy  },  /* level  6 */
+    { 21, 19, 19,  3,  5,  8, ZSTD_lazy    },  /* level  7 */
+    { 21, 19, 19,  3,  5, 16, ZSTD_lazy2   },  /* level  8 */
+    { 21, 19, 20,  4,  5, 16, ZSTD_lazy2   },  /* level  9 */
+    { 22, 20, 21,  4,  5, 16, ZSTD_lazy2   },  /* level 10 */
+    { 22, 21, 22,  4,  5, 16, ZSTD_lazy2   },  /* level 11 */
+    { 22, 21, 22,  5,  5, 16, ZSTD_lazy2   },  /* level 12 */
+    { 22, 21, 22,  5,  5, 32, ZSTD_btlazy2 },  /* level 13 */
+    { 22, 22, 23,  5,  5, 32, ZSTD_btlazy2 },  /* level 14 */
+    { 22, 23, 23,  6,  5, 32, ZSTD_btlazy2 },  /* level 15 */
+    { 22, 22, 22,  5,  5, 48, ZSTD_btopt   },  /* level 16 */
+    { 23, 23, 22,  5,  4, 64, ZSTD_btopt   },  /* level 17 */
+    { 23, 23, 22,  6,  3, 64, ZSTD_btultra },  /* level 18 */
+    { 23, 24, 22,  7,  3,256, ZSTD_btultra2},  /* level 19 */
+    { 25, 25, 23,  7,  3,256, ZSTD_btultra2},  /* level 20 */
+    { 26, 26, 24,  7,  3,512, ZSTD_btultra2},  /* level 21 */
+    { 27, 27, 25,  9,  3,999, ZSTD_btultra2},  /* level 22 */
 },
 {   /* for srcSize <= 256 KB */
     /* W,  C,  H,  S,  L,  T, strat */
-    {  0,  0,  0,  0,  0,  0, ZSTD_fast    },  /* level  0 - not used */
-    { 18, 13, 14,  1,  6,  8, ZSTD_fast    },  /* level  1 */
-    { 18, 14, 13,  1,  5,  8, ZSTD_dfast   },  /* level  2 */
-    { 18, 16, 15,  1,  5,  8, ZSTD_dfast   },  /* level  3 */
-    { 18, 15, 17,  1,  5,  8, ZSTD_greedy  },  /* level  4.*/
-    { 18, 16, 17,  4,  5,  8, ZSTD_greedy  },  /* level  5.*/
-    { 18, 16, 17,  3,  5,  8, ZSTD_lazy    },  /* level  6.*/
-    { 18, 17, 17,  4,  4,  8, ZSTD_lazy    },  /* level  7 */
-    { 18, 17, 17,  4,  4,  8, ZSTD_lazy2   },  /* level  8 */
-    { 18, 17, 17,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
-    { 18, 17, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
-    { 18, 18, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 11.*/
-    { 18, 18, 17,  7,  4,  8, ZSTD_lazy2   },  /* level 12.*/
-    { 18, 19, 17,  6,  4,  8, ZSTD_btlazy2 },  /* level 13 */
-    { 18, 18, 18,  4,  4, 16, ZSTD_btopt   },  /* level 14.*/
-    { 18, 18, 18,  4,  3, 16, ZSTD_btopt   },  /* level 15.*/
-    { 18, 19, 18,  6,  3, 32, ZSTD_btopt   },  /* level 16.*/
-    { 18, 19, 18,  8,  3, 64, ZSTD_btopt   },  /* level 17.*/
-    { 18, 19, 18,  9,  3,128, ZSTD_btopt   },  /* level 18.*/
-    { 18, 19, 18, 10,  3,256, ZSTD_btopt   },  /* level 19.*/
-    { 18, 19, 18, 11,  3,512, ZSTD_btultra },  /* level 20.*/
-    { 18, 19, 18, 12,  3,512, ZSTD_btultra },  /* level 21.*/
-    { 18, 19, 18, 13,  3,512, ZSTD_btultra },  /* level 22.*/
+    { 18, 12, 13,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
+    { 18, 13, 14,  1,  6,  0, ZSTD_fast    },  /* level  1 */
+    { 18, 14, 14,  1,  5,  0, ZSTD_dfast   },  /* level  2 */
+    { 18, 16, 16,  1,  4,  0, ZSTD_dfast   },  /* level  3 */
+    { 18, 16, 17,  2,  5,  2, ZSTD_greedy  },  /* level  4.*/
+    { 18, 18, 18,  3,  5,  2, ZSTD_greedy  },  /* level  5.*/
+    { 18, 18, 19,  3,  5,  4, ZSTD_lazy    },  /* level  6.*/
+    { 18, 18, 19,  4,  4,  4, ZSTD_lazy    },  /* level  7 */
+    { 18, 18, 19,  4,  4,  8, ZSTD_lazy2   },  /* level  8 */
+    { 18, 18, 19,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
+    { 18, 18, 19,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
+    { 18, 18, 19,  5,  4, 12, ZSTD_btlazy2 },  /* level 11.*/
+    { 18, 19, 19,  7,  4, 12, ZSTD_btlazy2 },  /* level 12.*/
+    { 18, 18, 19,  4,  4, 16, ZSTD_btopt   },  /* level 13 */
+    { 18, 18, 19,  4,  3, 32, ZSTD_btopt   },  /* level 14.*/
+    { 18, 18, 19,  6,  3,128, ZSTD_btopt   },  /* level 15.*/
+    { 18, 19, 19,  6,  3,128, ZSTD_btultra },  /* level 16.*/
+    { 18, 19, 19,  8,  3,256, ZSTD_btultra },  /* level 17.*/
+    { 18, 19, 19,  6,  3,128, ZSTD_btultra2},  /* level 18.*/
+    { 18, 19, 19,  8,  3,256, ZSTD_btultra2},  /* level 19.*/
+    { 18, 19, 19, 10,  3,512, ZSTD_btultra2},  /* level 20.*/
+    { 18, 19, 19, 12,  3,512, ZSTD_btultra2},  /* level 21.*/
+    { 18, 19, 19, 13,  3,999, ZSTD_btultra2},  /* level 22.*/
 },
 {   /* for srcSize <= 128 KB */
     /* W,  C,  H,  S,  L,  T, strat */
-    { 17, 12, 12,  1,  7,  8, ZSTD_fast    },  /* level  0 - not used */
-    { 17, 12, 13,  1,  6,  8, ZSTD_fast    },  /* level  1 */
-    { 17, 13, 16,  1,  5,  8, ZSTD_fast    },  /* level  2 */
-    { 17, 16, 16,  2,  5,  8, ZSTD_dfast   },  /* level  3 */
-    { 17, 13, 15,  3,  4,  8, ZSTD_greedy  },  /* level  4 */
-    { 17, 15, 17,  4,  4,  8, ZSTD_greedy  },  /* level  5 */
-    { 17, 16, 17,  3,  4,  8, ZSTD_lazy    },  /* level  6 */
-    { 17, 15, 17,  4,  4,  8, ZSTD_lazy2   },  /* level  7 */
+    { 17, 12, 12,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
+    { 17, 12, 13,  1,  6,  0, ZSTD_fast    },  /* level  1 */
+    { 17, 13, 15,  1,  5,  0, ZSTD_fast    },  /* level  2 */
+    { 17, 15, 16,  2,  5,  0, ZSTD_dfast   },  /* level  3 */
+    { 17, 17, 17,  2,  4,  0, ZSTD_dfast   },  /* level  4 */
+    { 17, 16, 17,  3,  4,  2, ZSTD_greedy  },  /* level  5 */
+    { 17, 17, 17,  3,  4,  4, ZSTD_lazy    },  /* level  6 */
+    { 17, 17, 17,  3,  4,  8, ZSTD_lazy2   },  /* level  7 */
     { 17, 17, 17,  4,  4,  8, ZSTD_lazy2   },  /* level  8 */
     { 17, 17, 17,  5,  4,  8, ZSTD_lazy2   },  /* level  9 */
     { 17, 17, 17,  6,  4,  8, ZSTD_lazy2   },  /* level 10 */
-    { 17, 17, 17,  7,  4,  8, ZSTD_lazy2   },  /* level 11 */
-    { 17, 17, 17,  8,  4,  8, ZSTD_lazy2   },  /* level 12 */
-    { 17, 18, 17,  6,  4,  8, ZSTD_btlazy2 },  /* level 13.*/
-    { 17, 17, 17,  7,  3,  8, ZSTD_btopt   },  /* level 14.*/
-    { 17, 17, 17,  7,  3, 16, ZSTD_btopt   },  /* level 15.*/
-    { 17, 18, 17,  7,  3, 32, ZSTD_btopt   },  /* level 16.*/
-    { 17, 18, 17,  7,  3, 64, ZSTD_btopt   },  /* level 17.*/
-    { 17, 18, 17,  7,  3,256, ZSTD_btopt   },  /* level 18.*/
-    { 17, 18, 17,  8,  3,256, ZSTD_btopt   },  /* level 19.*/
-    { 17, 18, 17,  9,  3,256, ZSTD_btultra },  /* level 20.*/
-    { 17, 18, 17, 10,  3,256, ZSTD_btultra },  /* level 21.*/
-    { 17, 18, 17, 11,  3,512, ZSTD_btultra },  /* level 22.*/
+    { 17, 17, 17,  5,  4,  8, ZSTD_btlazy2 },  /* level 11 */
+    { 17, 18, 17,  7,  4, 12, ZSTD_btlazy2 },  /* level 12 */
+    { 17, 18, 17,  3,  4, 12, ZSTD_btopt   },  /* level 13.*/
+    { 17, 18, 17,  4,  3, 32, ZSTD_btopt   },  /* level 14.*/
+    { 17, 18, 17,  6,  3,256, ZSTD_btopt   },  /* level 15.*/
+    { 17, 18, 17,  6,  3,128, ZSTD_btultra },  /* level 16.*/
+    { 17, 18, 17,  8,  3,256, ZSTD_btultra },  /* level 17.*/
+    { 17, 18, 17, 10,  3,512, ZSTD_btultra },  /* level 18.*/
+    { 17, 18, 17,  5,  3,256, ZSTD_btultra2},  /* level 19.*/
+    { 17, 18, 17,  7,  3,512, ZSTD_btultra2},  /* level 20.*/
+    { 17, 18, 17,  9,  3,512, ZSTD_btultra2},  /* level 21.*/
+    { 17, 18, 17, 11,  3,999, ZSTD_btultra2},  /* level 22.*/
 },
 {   /* for srcSize <= 16 KB */
     /* W,  C,  H,  S,  L,  T, strat */
-    { 14, 12, 12,  1,  7,  6, ZSTD_fast    },  /* level  0 - not used */
-    { 14, 14, 14,  1,  6,  6, ZSTD_fast    },  /* level  1 */
-    { 14, 14, 14,  1,  4,  6, ZSTD_fast    },  /* level  2 */
-    { 14, 14, 14,  1,  4,  6, ZSTD_dfast   },  /* level  3.*/
-    { 14, 14, 14,  4,  4,  6, ZSTD_greedy  },  /* level  4.*/
-    { 14, 14, 14,  3,  4,  6, ZSTD_lazy    },  /* level  5.*/
-    { 14, 14, 14,  4,  4,  6, ZSTD_lazy2   },  /* level  6 */
-    { 14, 14, 14,  5,  4,  6, ZSTD_lazy2   },  /* level  7 */
-    { 14, 14, 14,  6,  4,  6, ZSTD_lazy2   },  /* level  8.*/
-    { 14, 15, 14,  6,  4,  6, ZSTD_btlazy2 },  /* level  9.*/
-    { 14, 15, 14,  3,  3,  6, ZSTD_btopt   },  /* level 10.*/
-    { 14, 15, 14,  6,  3,  8, ZSTD_btopt   },  /* level 11.*/
-    { 14, 15, 14,  6,  3, 16, ZSTD_btopt   },  /* level 12.*/
-    { 14, 15, 14,  6,  3, 24, ZSTD_btopt   },  /* level 13.*/
-    { 14, 15, 15,  6,  3, 48, ZSTD_btopt   },  /* level 14.*/
-    { 14, 15, 15,  6,  3, 64, ZSTD_btopt   },  /* level 15.*/
-    { 14, 15, 15,  6,  3, 96, ZSTD_btopt   },  /* level 16.*/
-    { 14, 15, 15,  6,  3,128, ZSTD_btopt   },  /* level 17.*/
-    { 14, 15, 15,  6,  3,256, ZSTD_btopt   },  /* level 18.*/
-    { 14, 15, 15,  7,  3,256, ZSTD_btopt   },  /* level 19.*/
-    { 14, 15, 15,  8,  3,256, ZSTD_btultra },  /* level 20.*/
-    { 14, 15, 15,  9,  3,256, ZSTD_btultra },  /* level 21.*/
-    { 14, 15, 15, 10,  3,256, ZSTD_btultra },  /* level 22.*/
+    { 14, 12, 13,  1,  5,  1, ZSTD_fast    },  /* base for negative levels */
+    { 14, 14, 15,  1,  5,  0, ZSTD_fast    },  /* level  1 */
+    { 14, 14, 15,  1,  4,  0, ZSTD_fast    },  /* level  2 */
+    { 14, 14, 15,  2,  4,  0, ZSTD_dfast   },  /* level  3 */
+    { 14, 14, 14,  4,  4,  2, ZSTD_greedy  },  /* level  4 */
+    { 14, 14, 14,  3,  4,  4, ZSTD_lazy    },  /* level  5.*/
+    { 14, 14, 14,  4,  4,  8, ZSTD_lazy2   },  /* level  6 */
+    { 14, 14, 14,  6,  4,  8, ZSTD_lazy2   },  /* level  7 */
+    { 14, 14, 14,  8,  4,  8, ZSTD_lazy2   },  /* level  8.*/
+    { 14, 15, 14,  5,  4,  8, ZSTD_btlazy2 },  /* level  9.*/
+    { 14, 15, 14,  9,  4,  8, ZSTD_btlazy2 },  /* level 10.*/
+    { 14, 15, 14,  3,  4, 12, ZSTD_btopt   },  /* level 11.*/
+    { 14, 15, 14,  4,  3, 24, ZSTD_btopt   },  /* level 12.*/
+    { 14, 15, 14,  5,  3, 32, ZSTD_btultra },  /* level 13.*/
+    { 14, 15, 15,  6,  3, 64, ZSTD_btultra },  /* level 14.*/
+    { 14, 15, 15,  7,  3,256, ZSTD_btultra },  /* level 15.*/
+    { 14, 15, 15,  5,  3, 48, ZSTD_btultra2},  /* level 16.*/
+    { 14, 15, 15,  6,  3,128, ZSTD_btultra2},  /* level 17.*/
+    { 14, 15, 15,  7,  3,256, ZSTD_btultra2},  /* level 18.*/
+    { 14, 15, 15,  8,  3,256, ZSTD_btultra2},  /* level 19.*/
+    { 14, 15, 15,  8,  3,512, ZSTD_btultra2},  /* level 20.*/
+    { 14, 15, 15,  9,  3,512, ZSTD_btultra2},  /* level 21.*/
+    { 14, 15, 15, 10,  3,999, ZSTD_btultra2},  /* level 22.*/
 },
 };
 
-#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=1)
-/* This function just controls
- * the monotonic memory budget increase of ZSTD_defaultCParameters[0].
- * Run once, on first ZSTD_getCParams() usage, if ZSTD_DEBUG is enabled
- */
-MEM_STATIC void ZSTD_check_compressionLevel_monotonicIncrease_memoryBudget(void)
+/*! ZSTD_getCParams_internal() :
+ * @return ZSTD_compressionParameters structure for a selected compression level, srcSize and dictSize.
+ *  Note: srcSizeHint 0 means 0, use ZSTD_CONTENTSIZE_UNKNOWN for unknown.
+ *        Use dictSize == 0 for unknown or unused. */
+static ZSTD_compressionParameters ZSTD_getCParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize)
 {
-    int level;
-    for (level=1; level<ZSTD_maxCLevel(); level++) {
-        ZSTD_compressionParameters const c1 = ZSTD_defaultCParameters[0][level];
-        ZSTD_compressionParameters const c2 = ZSTD_defaultCParameters[0][level+1];
-        assert(c1.windowLog <= c2.windowLog);
-#       define ZSTD_TABLECOST(h,c) ((1<<(h)) + (1<<(c)))
-        assert(ZSTD_TABLECOST(c1.hashLog, c1.chainLog) <= ZSTD_TABLECOST(c2.hashLog, c2.chainLog));
+    int const unknown = srcSizeHint == ZSTD_CONTENTSIZE_UNKNOWN;
+    size_t const addedSize = unknown && dictSize > 0 ? 500 : 0;
+    U64 const rSize = unknown && dictSize == 0 ? ZSTD_CONTENTSIZE_UNKNOWN : srcSizeHint+dictSize+addedSize;
+    U32 const tableID = (rSize <= 256 KB) + (rSize <= 128 KB) + (rSize <= 16 KB);
+    int row = compressionLevel;
+    DEBUGLOG(5, "ZSTD_getCParams_internal (cLevel=%i)", compressionLevel);
+    if (compressionLevel == 0) row = ZSTD_CLEVEL_DEFAULT;   /* 0 == default */
+    if (compressionLevel < 0) row = 0;   /* entry 0 is baseline for fast mode */
+    if (compressionLevel > ZSTD_MAX_CLEVEL) row = ZSTD_MAX_CLEVEL;
+    {   ZSTD_compressionParameters cp = ZSTD_defaultCParameters[tableID][row];
+        if (compressionLevel < 0) cp.targetLength = (unsigned)(-compressionLevel);   /* acceleration factor */
+        /* refine parameters based on srcSize & dictSize */
+        return ZSTD_adjustCParams_internal(cp, srcSizeHint, dictSize);
     }
 }
-#endif
 
 /*! ZSTD_getCParams() :
-*   @return ZSTD_compressionParameters structure for a selected compression level, `srcSize` and `dictSize`.
-*   Size values are optional, provide 0 if not known or unused */
+ * @return ZSTD_compressionParameters structure for a selected compression level, srcSize and dictSize.
+ *  Size values are optional, provide 0 if not known or unused */
 ZSTD_compressionParameters ZSTD_getCParams(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize)
 {
-    size_t const addedSize = srcSizeHint ? 0 : 500;
-    U64 const rSize = srcSizeHint+dictSize ? srcSizeHint+dictSize+addedSize : (U64)-1;
-    U32 const tableID = (rSize <= 256 KB) + (rSize <= 128 KB) + (rSize <= 16 KB);   /* intentional underflow for srcSizeHint == 0 */
-
-#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=1)
-    static int g_monotonicTest = 1;
-    if (g_monotonicTest) {
-        ZSTD_check_compressionLevel_monotonicIncrease_memoryBudget();
-        g_monotonicTest=0;
-    }
-#endif
-
-    if (compressionLevel <= 0) compressionLevel = ZSTD_CLEVEL_DEFAULT;   /* 0 == default; no negative compressionLevel yet */
-    if (compressionLevel > ZSTD_MAX_CLEVEL) compressionLevel = ZSTD_MAX_CLEVEL;
-    { ZSTD_compressionParameters const cp = ZSTD_defaultCParameters[tableID][compressionLevel];
-      return ZSTD_adjustCParams_internal(cp, srcSizeHint, dictSize); }
-
+    if (srcSizeHint == 0) srcSizeHint = ZSTD_CONTENTSIZE_UNKNOWN;
+    return ZSTD_getCParams_internal(compressionLevel, srcSizeHint, dictSize);
 }
 
 /*! ZSTD_getParams() :
-*   same as ZSTD_getCParams(), but @return a `ZSTD_parameters` object (instead of `ZSTD_compressionParameters`).
-*   All fields of `ZSTD_frameParameters` are set to default (0) */
-ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize) {
+ *  same idea as ZSTD_getCParams()
+ * @return a `ZSTD_parameters` structure (instead of `ZSTD_compressionParameters`).
+ *  Fields of `ZSTD_frameParameters` are set to default values */
+static ZSTD_parameters ZSTD_getParams_internal(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize) {
     ZSTD_parameters params;
-    ZSTD_compressionParameters const cParams = ZSTD_getCParams(compressionLevel, srcSizeHint, dictSize);
+    ZSTD_compressionParameters const cParams = ZSTD_getCParams_internal(compressionLevel, srcSizeHint, dictSize);
+    DEBUGLOG(5, "ZSTD_getParams (cLevel=%i)", compressionLevel);
     memset(&params, 0, sizeof(params));
     params.cParams = cParams;
+    params.fParams.contentSizeFlag = 1;
     return params;
 }
+
+/*! ZSTD_getParams() :
+ *  same idea as ZSTD_getCParams()
+ * @return a `ZSTD_parameters` structure (instead of `ZSTD_compressionParameters`).
+ *  Fields of `ZSTD_frameParameters` are set to default values */
+ZSTD_parameters ZSTD_getParams(int compressionLevel, unsigned long long srcSizeHint, size_t dictSize) {
+    if (srcSizeHint == 0) srcSizeHint = ZSTD_CONTENTSIZE_UNKNOWN;
+    return ZSTD_getParams_internal(compressionLevel, srcSizeHint, dictSize);
+}
diff --git a/contrib/zstd/zstd_compress_internal.h b/contrib/zstd/zstd_compress_internal.h
new file mode 100644
index 000000000..c500687ce
--- /dev/null
+++ b/contrib/zstd/zstd_compress_internal.h
@@ -0,0 +1,1125 @@
+/*
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* This header contains definitions
+ * that shall **only** be used by modules within lib/compress.
+ */
+
+#ifndef ZSTD_COMPRESS_H
+#define ZSTD_COMPRESS_H
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include "zstd_internal.h"
+#include "zstd_cwksp.h"
+#ifdef ZSTD_MULTITHREAD
+#  include "zstdmt_compress.h"
+#endif
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+
+/*-*************************************
+*  Constants
+***************************************/
+#define kSearchStrength      8
+#define HASH_READ_SIZE       8
+#define ZSTD_DUBT_UNSORTED_MARK 1   /* For btlazy2 strategy, index ZSTD_DUBT_UNSORTED_MARK==1 means "unsorted".
+                                       It could be confused for a real successor at index "1", if sorted as larger than its predecessor.
+                                       It's not a big deal though : candidate will just be sorted again.
+                                       Additionally, candidate position 1 will be lost.
+                                       But candidate 1 cannot hide a large tree of candidates, so it's a minimal loss.
+                                       The benefit is that ZSTD_DUBT_UNSORTED_MARK cannot be mishandled after table re-use with a different strategy.
+                                       This constant is required by ZSTD_compressBlock_btlazy2() and ZSTD_reduceTable_internal() */
+
+
+/*-*************************************
+*  Context memory management
+***************************************/
+typedef enum { ZSTDcs_created=0, ZSTDcs_init, ZSTDcs_ongoing, ZSTDcs_ending } ZSTD_compressionStage_e;
+typedef enum { zcss_init=0, zcss_load, zcss_flush } ZSTD_cStreamStage;
+
+typedef struct ZSTD_prefixDict_s {
+    const void* dict;
+    size_t dictSize;
+    ZSTD_dictContentType_e dictContentType;
+} ZSTD_prefixDict;
+
+typedef struct {
+    void* dictBuffer;
+    void const* dict;
+    size_t dictSize;
+    ZSTD_dictContentType_e dictContentType;
+    ZSTD_CDict* cdict;
+} ZSTD_localDict;
+
+typedef struct {
+    U32 CTable[HUF_CTABLE_SIZE_U32(255)];
+    HUF_repeat repeatMode;
+} ZSTD_hufCTables_t;
+
+typedef struct {
+    FSE_CTable offcodeCTable[FSE_CTABLE_SIZE_U32(OffFSELog, MaxOff)];
+    FSE_CTable matchlengthCTable[FSE_CTABLE_SIZE_U32(MLFSELog, MaxML)];
+    FSE_CTable litlengthCTable[FSE_CTABLE_SIZE_U32(LLFSELog, MaxLL)];
+    FSE_repeat offcode_repeatMode;
+    FSE_repeat matchlength_repeatMode;
+    FSE_repeat litlength_repeatMode;
+} ZSTD_fseCTables_t;
+
+typedef struct {
+    ZSTD_hufCTables_t huf;
+    ZSTD_fseCTables_t fse;
+} ZSTD_entropyCTables_t;
+
+typedef struct {
+    U32 off;
+    U32 len;
+} ZSTD_match_t;
+
+typedef struct {
+    int price;
+    U32 off;
+    U32 mlen;
+    U32 litlen;
+    U32 rep[ZSTD_REP_NUM];
+} ZSTD_optimal_t;
+
+typedef enum { zop_dynamic=0, zop_predef } ZSTD_OptPrice_e;
+
+typedef struct {
+    /* All tables are allocated inside cctx->workspace by ZSTD_resetCCtx_internal() */
+    unsigned* litFreq;           /* table of literals statistics, of size 256 */
+    unsigned* litLengthFreq;     /* table of litLength statistics, of size (MaxLL+1) */
+    unsigned* matchLengthFreq;   /* table of matchLength statistics, of size (MaxML+1) */
+    unsigned* offCodeFreq;       /* table of offCode statistics, of size (MaxOff+1) */
+    ZSTD_match_t* matchTable;    /* list of found matches, of size ZSTD_OPT_NUM+1 */
+    ZSTD_optimal_t* priceTable;  /* All positions tracked by optimal parser, of size ZSTD_OPT_NUM+1 */
+
+    U32  litSum;                 /* nb of literals */
+    U32  litLengthSum;           /* nb of litLength codes */
+    U32  matchLengthSum;         /* nb of matchLength codes */
+    U32  offCodeSum;             /* nb of offset codes */
+    U32  litSumBasePrice;        /* to compare to log2(litfreq) */
+    U32  litLengthSumBasePrice;  /* to compare to log2(llfreq)  */
+    U32  matchLengthSumBasePrice;/* to compare to log2(mlfreq)  */
+    U32  offCodeSumBasePrice;    /* to compare to log2(offreq)  */
+    ZSTD_OptPrice_e priceType;   /* prices can be determined dynamically, or follow a pre-defined cost structure */
+    const ZSTD_entropyCTables_t* symbolCosts;  /* pre-calculated dictionary statistics */
+    ZSTD_literalCompressionMode_e literalCompressionMode;
+} optState_t;
+
+typedef struct {
+  ZSTD_entropyCTables_t entropy;
+  U32 rep[ZSTD_REP_NUM];
+} ZSTD_compressedBlockState_t;
+
+typedef struct {
+    BYTE const* nextSrc;    /* next block here to continue on current prefix */
+    BYTE const* base;       /* All regular indexes relative to this position */
+    BYTE const* dictBase;   /* extDict indexes relative to this position */
+    U32 dictLimit;          /* below that point, need extDict */
+    U32 lowLimit;           /* below that point, no more valid data */
+} ZSTD_window_t;
+
+typedef struct ZSTD_matchState_t ZSTD_matchState_t;
+struct ZSTD_matchState_t {
+    ZSTD_window_t window;   /* State for window round buffer management */
+    U32 loadedDictEnd;      /* index of end of dictionary, within context's referential.
+                             * When loadedDictEnd != 0, a dictionary is in use, and still valid.
+                             * This relies on a mechanism to set loadedDictEnd=0 when dictionary is no longer within distance.
+                             * Such mechanism is provided within ZSTD_window_enforceMaxDist() and ZSTD_checkDictValidity().
+                             * When dict referential is copied into active context (i.e. not attached),
+                             * loadedDictEnd == dictSize, since referential starts from zero.
+                             */
+    U32 nextToUpdate;       /* index from which to continue table update */
+    U32 hashLog3;           /* dispatch table for matches of len==3 : larger == faster, more memory */
+    U32* hashTable;
+    U32* hashTable3;
+    U32* chainTable;
+    optState_t opt;         /* optimal parser state */
+    const ZSTD_matchState_t* dictMatchState;
+    ZSTD_compressionParameters cParams;
+};
+
+typedef struct {
+    ZSTD_compressedBlockState_t* prevCBlock;
+    ZSTD_compressedBlockState_t* nextCBlock;
+    ZSTD_matchState_t matchState;
+} ZSTD_blockState_t;
+
+typedef struct {
+    U32 offset;
+    U32 checksum;
+} ldmEntry_t;
+
+typedef struct {
+    ZSTD_window_t window;   /* State for the window round buffer management */
+    ldmEntry_t* hashTable;
+    U32 loadedDictEnd;
+    BYTE* bucketOffsets;    /* Next position in bucket to insert entry */
+    U64 hashPower;          /* Used to compute the rolling hash.
+                             * Depends on ldmParams.minMatchLength */
+} ldmState_t;
+
+typedef struct {
+    U32 enableLdm;          /* 1 if enable long distance matching */
+    U32 hashLog;            /* Log size of hashTable */
+    U32 bucketSizeLog;      /* Log bucket size for collision resolution, at most 8 */
+    U32 minMatchLength;     /* Minimum match length */
+    U32 hashRateLog;       /* Log number of entries to skip */
+    U32 windowLog;          /* Window log for the LDM */
+} ldmParams_t;
+
+typedef struct {
+    U32 offset;
+    U32 litLength;
+    U32 matchLength;
+} rawSeq;
+
+typedef struct {
+  rawSeq* seq;     /* The start of the sequences */
+  size_t pos;      /* The position where reading stopped. <= size. */
+  size_t size;     /* The number of sequences. <= capacity. */
+  size_t capacity; /* The capacity starting from `seq` pointer */
+} rawSeqStore_t;
+
+typedef struct {
+    int collectSequences;
+    ZSTD_Sequence* seqStart;
+    size_t seqIndex;
+    size_t maxSequences;
+} SeqCollector;
+
+struct ZSTD_CCtx_params_s {
+    ZSTD_format_e format;
+    ZSTD_compressionParameters cParams;
+    ZSTD_frameParameters fParams;
+
+    int compressionLevel;
+    int forceWindow;           /* force back-references to respect limit of
+                                * 1<<wLog, even for dictionary */
+    size_t targetCBlockSize;   /* Tries to fit compressed block size to be around targetCBlockSize.
+                                * No target when targetCBlockSize == 0.
+                                * There is no guarantee on compressed block size */
+    int srcSizeHint;           /* User's best guess of source size.
+                                * Hint is not valid when srcSizeHint == 0.
+                                * There is no guarantee that hint is close to actual source size */
+
+    ZSTD_dictAttachPref_e attachDictPref;
+    ZSTD_literalCompressionMode_e literalCompressionMode;
+
+    /* Multithreading: used to pass parameters to mtctx */
+    int nbWorkers;
+    size_t jobSize;
+    int overlapLog;
+    int rsyncable;
+
+    /* Long distance matching parameters */
+    ldmParams_t ldmParams;
+
+    /* Internal use, for createCCtxParams() and freeCCtxParams() only */
+    ZSTD_customMem customMem;
+};  /* typedef'd to ZSTD_CCtx_params within "zstd.h" */
+
+struct ZSTD_CCtx_s {
+    ZSTD_compressionStage_e stage;
+    int cParamsChanged;                  /* == 1 if cParams(except wlog) or compression level are changed in requestedParams. Triggers transmission of new params to ZSTDMT (if available) then reset to 0. */
+    int bmi2;                            /* == 1 if the CPU supports BMI2 and 0 otherwise. CPU support is determined dynamically once per context lifetime. */
+    ZSTD_CCtx_params requestedParams;
+    ZSTD_CCtx_params appliedParams;
+    U32   dictID;
+
+    ZSTD_cwksp workspace; /* manages buffer for dynamic allocations */
+    size_t blockSize;
+    unsigned long long pledgedSrcSizePlusOne;  /* this way, 0 (default) == unknown */
+    unsigned long long consumedSrcSize;
+    unsigned long long producedCSize;
+    XXH64_state_t xxhState;
+    ZSTD_customMem customMem;
+    size_t staticSize;
+    SeqCollector seqCollector;
+    int isFirstBlock;
+    int initialized;
+
+    seqStore_t seqStore;      /* sequences storage ptrs */
+    ldmState_t ldmState;      /* long distance matching state */
+    rawSeq* ldmSequences;     /* Storage for the ldm output sequences */
+    size_t maxNbLdmSequences;
+    rawSeqStore_t externSeqStore; /* Mutable reference to external sequences */
+    ZSTD_blockState_t blockState;
+    U32* entropyWorkspace;  /* entropy workspace of HUF_WORKSPACE_SIZE bytes */
+
+    /* streaming */
+    char*  inBuff;
+    size_t inBuffSize;
+    size_t inToCompress;
+    size_t inBuffPos;
+    size_t inBuffTarget;
+    char*  outBuff;
+    size_t outBuffSize;
+    size_t outBuffContentSize;
+    size_t outBuffFlushedSize;
+    ZSTD_cStreamStage streamStage;
+    U32    frameEnded;
+
+    /* Dictionary */
+    ZSTD_localDict localDict;
+    const ZSTD_CDict* cdict;
+    ZSTD_prefixDict prefixDict;   /* single-usage dictionary */
+
+    /* Multi-threading */
+#ifdef ZSTD_MULTITHREAD
+    ZSTDMT_CCtx* mtctx;
+#endif
+};
+
+typedef enum { ZSTD_dtlm_fast, ZSTD_dtlm_full } ZSTD_dictTableLoadMethod_e;
+
+typedef enum { ZSTD_noDict = 0, ZSTD_extDict = 1, ZSTD_dictMatchState = 2 } ZSTD_dictMode_e;
+
+
+typedef size_t (*ZSTD_blockCompressor) (
+        ZSTD_matchState_t* bs, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, ZSTD_dictMode_e dictMode);
+
+
+MEM_STATIC U32 ZSTD_LLcode(U32 litLength)
+{
+    static const BYTE LL_Code[64] = {  0,  1,  2,  3,  4,  5,  6,  7,
+                                       8,  9, 10, 11, 12, 13, 14, 15,
+                                      16, 16, 17, 17, 18, 18, 19, 19,
+                                      20, 20, 20, 20, 21, 21, 21, 21,
+                                      22, 22, 22, 22, 22, 22, 22, 22,
+                                      23, 23, 23, 23, 23, 23, 23, 23,
+                                      24, 24, 24, 24, 24, 24, 24, 24,
+                                      24, 24, 24, 24, 24, 24, 24, 24 };
+    static const U32 LL_deltaCode = 19;
+    return (litLength > 63) ? ZSTD_highbit32(litLength) + LL_deltaCode : LL_Code[litLength];
+}
+
+/* ZSTD_MLcode() :
+ * note : mlBase = matchLength - MINMATCH;
+ *        because it's the format it's stored in seqStore->sequences */
+MEM_STATIC U32 ZSTD_MLcode(U32 mlBase)
+{
+    static const BYTE ML_Code[128] = { 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15,
+                                      16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31,
+                                      32, 32, 33, 33, 34, 34, 35, 35, 36, 36, 36, 36, 37, 37, 37, 37,
+                                      38, 38, 38, 38, 38, 38, 38, 38, 39, 39, 39, 39, 39, 39, 39, 39,
+                                      40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40, 40,
+                                      41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41, 41,
+                                      42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42,
+                                      42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42 };
+    static const U32 ML_deltaCode = 36;
+    return (mlBase > 127) ? ZSTD_highbit32(mlBase) + ML_deltaCode : ML_Code[mlBase];
+}
+
+typedef struct repcodes_s {
+    U32 rep[3];
+} repcodes_t;
+
+MEM_STATIC repcodes_t ZSTD_updateRep(U32 const rep[3], U32 const offset, U32 const ll0)
+{
+    repcodes_t newReps;
+    if (offset >= ZSTD_REP_NUM) {  /* full offset */
+        newReps.rep[2] = rep[1];
+        newReps.rep[1] = rep[0];
+        newReps.rep[0] = offset - ZSTD_REP_MOVE;
+    } else {   /* repcode */
+        U32 const repCode = offset + ll0;
+        if (repCode > 0) {  /* note : if repCode==0, no change */
+            U32 const currentOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
+            newReps.rep[2] = (repCode >= 2) ? rep[1] : rep[2];
+            newReps.rep[1] = rep[0];
+            newReps.rep[0] = currentOffset;
+        } else {   /* repCode == 0 */
+            memcpy(&newReps, rep, sizeof(newReps));
+        }
+    }
+    return newReps;
+}
+
+/* ZSTD_cParam_withinBounds:
+ * @return 1 if value is within cParam bounds,
+ * 0 otherwise */
+MEM_STATIC int ZSTD_cParam_withinBounds(ZSTD_cParameter cParam, int value)
+{
+    ZSTD_bounds const bounds = ZSTD_cParam_getBounds(cParam);
+    if (ZSTD_isError(bounds.error)) return 0;
+    if (value < bounds.lowerBound) return 0;
+    if (value > bounds.upperBound) return 0;
+    return 1;
+}
+
+/* ZSTD_noCompressBlock() :
+ * Writes uncompressed block to dst buffer from given src.
+ * Returns the size of the block */
+MEM_STATIC size_t ZSTD_noCompressBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize, U32 lastBlock)
+{
+    U32 const cBlockHeader24 = lastBlock + (((U32)bt_raw)<<1) + (U32)(srcSize << 3);
+    RETURN_ERROR_IF(srcSize + ZSTD_blockHeaderSize > dstCapacity,
+                    dstSize_tooSmall, "dst buf too small for uncompressed block");
+    MEM_writeLE24(dst, cBlockHeader24);
+    memcpy((BYTE*)dst + ZSTD_blockHeaderSize, src, srcSize);
+    return ZSTD_blockHeaderSize + srcSize;
+}
+
+MEM_STATIC size_t ZSTD_rleCompressBlock (void* dst, size_t dstCapacity, BYTE src, size_t srcSize, U32 lastBlock)
+{
+    BYTE* const op = (BYTE*)dst;
+    U32 const cBlockHeader = lastBlock + (((U32)bt_rle)<<1) + (U32)(srcSize << 3);
+    RETURN_ERROR_IF(dstCapacity < 4, dstSize_tooSmall, "");
+    MEM_writeLE24(op, cBlockHeader);
+    op[3] = src;
+    return 4;
+}
+
+
+/* ZSTD_minGain() :
+ * minimum compression required
+ * to generate a compress block or a compressed literals section.
+ * note : use same formula for both situations */
+MEM_STATIC size_t ZSTD_minGain(size_t srcSize, ZSTD_strategy strat)
+{
+    U32 const minlog = (strat>=ZSTD_btultra) ? (U32)(strat) - 1 : 6;
+    ZSTD_STATIC_ASSERT(ZSTD_btultra == 8);
+    assert(ZSTD_cParam_withinBounds(ZSTD_c_strategy, strat));
+    return (srcSize >> minlog) + 2;
+}
+
+MEM_STATIC int ZSTD_disableLiteralsCompression(const ZSTD_CCtx_params* cctxParams)
+{
+    switch (cctxParams->literalCompressionMode) {
+    case ZSTD_lcm_huffman:
+        return 0;
+    case ZSTD_lcm_uncompressed:
+        return 1;
+    default:
+        assert(0 /* impossible: pre-validated */);
+        /* fall-through */
+    case ZSTD_lcm_auto:
+        return (cctxParams->cParams.strategy == ZSTD_fast) && (cctxParams->cParams.targetLength > 0);
+    }
+}
+
+/*! ZSTD_safecopyLiterals() :
+ *  memcpy() function that won't read beyond more than WILDCOPY_OVERLENGTH bytes past ilimit_w.
+ *  Only called when the sequence ends past ilimit_w, so it only needs to be optimized for single
+ *  large copies.
+ */
+static void ZSTD_safecopyLiterals(BYTE* op, BYTE const* ip, BYTE const* const iend, BYTE const* ilimit_w) {
+    assert(iend > ilimit_w);
+    if (ip <= ilimit_w) {
+        ZSTD_wildcopy(op, ip, ilimit_w - ip, ZSTD_no_overlap);
+        op += ilimit_w - ip;
+        ip = ilimit_w;
+    }
+    while (ip < iend) *op++ = *ip++;
+}
+
+/*! ZSTD_storeSeq() :
+ *  Store a sequence (litlen, litPtr, offCode and mlBase) into seqStore_t.
+ *  `offCode` : distance to match + ZSTD_REP_MOVE (values <= ZSTD_REP_MOVE are repCodes).
+ *  `mlBase` : matchLength - MINMATCH
+ *  Allowed to overread literals up to litLimit.
+*/
+HINT_INLINE UNUSED_ATTR
+void ZSTD_storeSeq(seqStore_t* seqStorePtr, size_t litLength, const BYTE* literals, const BYTE* litLimit, U32 offCode, size_t mlBase)
+{
+    BYTE const* const litLimit_w = litLimit - WILDCOPY_OVERLENGTH;
+    BYTE const* const litEnd = literals + litLength;
+#if defined(DEBUGLEVEL) && (DEBUGLEVEL >= 6)
+    static const BYTE* g_start = NULL;
+    if (g_start==NULL) g_start = (const BYTE*)literals;  /* note : index only works for compression within a single segment */
+    {   U32 const pos = (U32)((const BYTE*)literals - g_start);
+        DEBUGLOG(6, "Cpos%7u :%3u literals, match%4u bytes at offCode%7u",
+               pos, (U32)litLength, (U32)mlBase+MINMATCH, (U32)offCode);
+    }
+#endif
+    assert((size_t)(seqStorePtr->sequences - seqStorePtr->sequencesStart) < seqStorePtr->maxNbSeq);
+    /* copy Literals */
+    assert(seqStorePtr->maxNbLit <= 128 KB);
+    assert(seqStorePtr->lit + litLength <= seqStorePtr->litStart + seqStorePtr->maxNbLit);
+    assert(literals + litLength <= litLimit);
+    if (litEnd <= litLimit_w) {
+        /* Common case we can use wildcopy.
+	 * First copy 16 bytes, because literals are likely short.
+	 */
+        assert(WILDCOPY_OVERLENGTH >= 16);
+        ZSTD_copy16(seqStorePtr->lit, literals);
+        if (litLength > 16) {
+            ZSTD_wildcopy(seqStorePtr->lit+16, literals+16, (ptrdiff_t)litLength-16, ZSTD_no_overlap);
+        }
+    } else {
+        ZSTD_safecopyLiterals(seqStorePtr->lit, literals, litEnd, litLimit_w);
+    }
+    seqStorePtr->lit += litLength;
+
+    /* literal Length */
+    if (litLength>0xFFFF) {
+        assert(seqStorePtr->longLengthID == 0); /* there can only be a single long length */
+        seqStorePtr->longLengthID = 1;
+        seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    }
+    seqStorePtr->sequences[0].litLength = (U16)litLength;
+
+    /* match offset */
+    seqStorePtr->sequences[0].offset = offCode + 1;
+
+    /* match Length */
+    if (mlBase>0xFFFF) {
+        assert(seqStorePtr->longLengthID == 0); /* there can only be a single long length */
+        seqStorePtr->longLengthID = 2;
+        seqStorePtr->longLengthPos = (U32)(seqStorePtr->sequences - seqStorePtr->sequencesStart);
+    }
+    seqStorePtr->sequences[0].matchLength = (U16)mlBase;
+
+    seqStorePtr->sequences++;
+}
+
+
+/*-*************************************
+*  Match length counter
+***************************************/
+static unsigned ZSTD_NbCommonBytes (size_t val)
+{
+    if (MEM_isLittleEndian()) {
+        if (MEM_64bits()) {
+#       if defined(_MSC_VER) && defined(_WIN64)
+            unsigned long r = 0;
+            return _BitScanForward64( &r, (U64)val ) ? (unsigned)(r >> 3) : 0;
+#       elif defined(__GNUC__) && (__GNUC__ >= 4)
+            return (__builtin_ctzll((U64)val) >> 3);
+#       else
+            static const int DeBruijnBytePos[64] = { 0, 0, 0, 0, 0, 1, 1, 2,
+                                                     0, 3, 1, 3, 1, 4, 2, 7,
+                                                     0, 2, 3, 6, 1, 5, 3, 5,
+                                                     1, 3, 4, 4, 2, 5, 6, 7,
+                                                     7, 0, 1, 2, 3, 3, 4, 6,
+                                                     2, 6, 5, 5, 3, 4, 5, 6,
+                                                     7, 1, 2, 4, 6, 4, 4, 5,
+                                                     7, 2, 6, 5, 7, 6, 7, 7 };
+            return DeBruijnBytePos[((U64)((val & -(long long)val) * 0x0218A392CDABBD3FULL)) >> 58];
+#       endif
+        } else { /* 32 bits */
+#       if defined(_MSC_VER)
+            unsigned long r=0;
+            return _BitScanForward( &r, (U32)val ) ? (unsigned)(r >> 3) : 0;
+#       elif defined(__GNUC__) && (__GNUC__ >= 3)
+            return (__builtin_ctz((U32)val) >> 3);
+#       else
+            static const int DeBruijnBytePos[32] = { 0, 0, 3, 0, 3, 1, 3, 0,
+                                                     3, 2, 2, 1, 3, 2, 0, 1,
+                                                     3, 3, 1, 2, 2, 2, 2, 0,
+                                                     3, 1, 2, 0, 1, 0, 1, 1 };
+            return DeBruijnBytePos[((U32)((val & -(S32)val) * 0x077CB531U)) >> 27];
+#       endif
+        }
+    } else {  /* Big Endian CPU */
+        if (MEM_64bits()) {
+#       if defined(_MSC_VER) && defined(_WIN64)
+            unsigned long r = 0;
+            return _BitScanReverse64( &r, val ) ? (unsigned)(r >> 3) : 0;
+#       elif defined(__GNUC__) && (__GNUC__ >= 4)
+            return (__builtin_clzll(val) >> 3);
+#       else
+            unsigned r;
+            const unsigned n32 = sizeof(size_t)*4;   /* calculate this way due to compiler complaining in 32-bits mode */
+            if (!(val>>n32)) { r=4; } else { r=0; val>>=n32; }
+            if (!(val>>16)) { r+=2; val>>=8; } else { val>>=24; }
+            r += (!val);
+            return r;
+#       endif
+        } else { /* 32 bits */
+#       if defined(_MSC_VER)
+            unsigned long r = 0;
+            return _BitScanReverse( &r, (unsigned long)val ) ? (unsigned)(r >> 3) : 0;
+#       elif defined(__GNUC__) && (__GNUC__ >= 3)
+            return (__builtin_clz((U32)val) >> 3);
+#       else
+            unsigned r;
+            if (!(val>>16)) { r=2; val>>=8; } else { r=0; val>>=24; }
+            r += (!val);
+            return r;
+#       endif
+    }   }
+}
+
+
+MEM_STATIC size_t ZSTD_count(const BYTE* pIn, const BYTE* pMatch, const BYTE* const pInLimit)
+{
+    const BYTE* const pStart = pIn;
+    const BYTE* const pInLoopLimit = pInLimit - (sizeof(size_t)-1);
+
+    if (pIn < pInLoopLimit) {
+        { size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
+          if (diff) return ZSTD_NbCommonBytes(diff); }
+        pIn+=sizeof(size_t); pMatch+=sizeof(size_t);
+        while (pIn < pInLoopLimit) {
+            size_t const diff = MEM_readST(pMatch) ^ MEM_readST(pIn);
+            if (!diff) { pIn+=sizeof(size_t); pMatch+=sizeof(size_t); continue; }
+            pIn += ZSTD_NbCommonBytes(diff);
+            return (size_t)(pIn - pStart);
+    }   }
+    if (MEM_64bits() && (pIn<(pInLimit-3)) && (MEM_read32(pMatch) == MEM_read32(pIn))) { pIn+=4; pMatch+=4; }
+    if ((pIn<(pInLimit-1)) && (MEM_read16(pMatch) == MEM_read16(pIn))) { pIn+=2; pMatch+=2; }
+    if ((pIn<pInLimit) && (*pMatch == *pIn)) pIn++;
+    return (size_t)(pIn - pStart);
+}
+
+/** ZSTD_count_2segments() :
+ *  can count match length with `ip` & `match` in 2 different segments.
+ *  convention : on reaching mEnd, match count continue starting from iStart
+ */
+MEM_STATIC size_t
+ZSTD_count_2segments(const BYTE* ip, const BYTE* match,
+                     const BYTE* iEnd, const BYTE* mEnd, const BYTE* iStart)
+{
+    const BYTE* const vEnd = MIN( ip + (mEnd - match), iEnd);
+    size_t const matchLength = ZSTD_count(ip, match, vEnd);
+    if (match + matchLength != mEnd) return matchLength;
+    DEBUGLOG(7, "ZSTD_count_2segments: found a 2-parts match (current length==%zu)", matchLength);
+    DEBUGLOG(7, "distance from match beginning to end dictionary = %zi", mEnd - match);
+    DEBUGLOG(7, "distance from current pos to end buffer = %zi", iEnd - ip);
+    DEBUGLOG(7, "next byte : ip==%02X, istart==%02X", ip[matchLength], *iStart);
+    DEBUGLOG(7, "final match length = %zu", matchLength + ZSTD_count(ip+matchLength, iStart, iEnd));
+    return matchLength + ZSTD_count(ip+matchLength, iStart, iEnd);
+}
+
+
+/*-*************************************
+ *  Hashes
+ ***************************************/
+static const U32 prime3bytes = 506832829U;
+static U32    ZSTD_hash3(U32 u, U32 h) { return ((u << (32-24)) * prime3bytes)  >> (32-h) ; }
+MEM_STATIC size_t ZSTD_hash3Ptr(const void* ptr, U32 h) { return ZSTD_hash3(MEM_readLE32(ptr), h); } /* only in zstd_opt.h */
+
+static const U32 prime4bytes = 2654435761U;
+static U32    ZSTD_hash4(U32 u, U32 h) { return (u * prime4bytes) >> (32-h) ; }
+static size_t ZSTD_hash4Ptr(const void* ptr, U32 h) { return ZSTD_hash4(MEM_read32(ptr), h); }
+
+static const U64 prime5bytes = 889523592379ULL;
+static size_t ZSTD_hash5(U64 u, U32 h) { return (size_t)(((u  << (64-40)) * prime5bytes) >> (64-h)) ; }
+static size_t ZSTD_hash5Ptr(const void* p, U32 h) { return ZSTD_hash5(MEM_readLE64(p), h); }
+
+static const U64 prime6bytes = 227718039650203ULL;
+static size_t ZSTD_hash6(U64 u, U32 h) { return (size_t)(((u  << (64-48)) * prime6bytes) >> (64-h)) ; }
+static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h); }
+
+static const U64 prime7bytes = 58295818150454627ULL;
+static size_t ZSTD_hash7(U64 u, U32 h) { return (size_t)(((u  << (64-56)) * prime7bytes) >> (64-h)) ; }
+static size_t ZSTD_hash7Ptr(const void* p, U32 h) { return ZSTD_hash7(MEM_readLE64(p), h); }
+
+static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
+static size_t ZSTD_hash8(U64 u, U32 h) { return (size_t)(((u) * prime8bytes) >> (64-h)) ; }
+static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h); }
+
+MEM_STATIC size_t ZSTD_hashPtr(const void* p, U32 hBits, U32 mls)
+{
+    switch(mls)
+    {
+    default:
+    case 4: return ZSTD_hash4Ptr(p, hBits);
+    case 5: return ZSTD_hash5Ptr(p, hBits);
+    case 6: return ZSTD_hash6Ptr(p, hBits);
+    case 7: return ZSTD_hash7Ptr(p, hBits);
+    case 8: return ZSTD_hash8Ptr(p, hBits);
+    }
+}
+
+/** ZSTD_ipow() :
+ * Return base^exponent.
+ */
+static U64 ZSTD_ipow(U64 base, U64 exponent)
+{
+    U64 power = 1;
+    while (exponent) {
+      if (exponent & 1) power *= base;
+      exponent >>= 1;
+      base *= base;
+    }
+    return power;
+}
+
+#define ZSTD_ROLL_HASH_CHAR_OFFSET 10
+
+/** ZSTD_rollingHash_append() :
+ * Add the buffer to the hash value.
+ */
+static U64 ZSTD_rollingHash_append(U64 hash, void const* buf, size_t size)
+{
+    BYTE const* istart = (BYTE const*)buf;
+    size_t pos;
+    for (pos = 0; pos < size; ++pos) {
+        hash *= prime8bytes;
+        hash += istart[pos] + ZSTD_ROLL_HASH_CHAR_OFFSET;
+    }
+    return hash;
+}
+
+/** ZSTD_rollingHash_compute() :
+ * Compute the rolling hash value of the buffer.
+ */
+MEM_STATIC U64 ZSTD_rollingHash_compute(void const* buf, size_t size)
+{
+    return ZSTD_rollingHash_append(0, buf, size);
+}
+
+/** ZSTD_rollingHash_primePower() :
+ * Compute the primePower to be passed to ZSTD_rollingHash_rotate() for a hash
+ * over a window of length bytes.
+ */
+MEM_STATIC U64 ZSTD_rollingHash_primePower(U32 length)
+{
+    return ZSTD_ipow(prime8bytes, length - 1);
+}
+
+/** ZSTD_rollingHash_rotate() :
+ * Rotate the rolling hash by one byte.
+ */
+MEM_STATIC U64 ZSTD_rollingHash_rotate(U64 hash, BYTE toRemove, BYTE toAdd, U64 primePower)
+{
+    hash -= (toRemove + ZSTD_ROLL_HASH_CHAR_OFFSET) * primePower;
+    hash *= prime8bytes;
+    hash += toAdd + ZSTD_ROLL_HASH_CHAR_OFFSET;
+    return hash;
+}
+
+/*-*************************************
+*  Round buffer management
+***************************************/
+#if (ZSTD_WINDOWLOG_MAX_64 > 31)
+# error "ZSTD_WINDOWLOG_MAX is too large : would overflow ZSTD_CURRENT_MAX"
+#endif
+/* Max current allowed */
+#define ZSTD_CURRENT_MAX ((3U << 29) + (1U << ZSTD_WINDOWLOG_MAX))
+/* Maximum chunk size before overflow correction needs to be called again */
+#define ZSTD_CHUNKSIZE_MAX                                                     \
+    ( ((U32)-1)                  /* Maximum ending current index */            \
+    - ZSTD_CURRENT_MAX)          /* Maximum beginning lowLimit */
+
+/**
+ * ZSTD_window_clear():
+ * Clears the window containing the history by simply setting it to empty.
+ */
+MEM_STATIC void ZSTD_window_clear(ZSTD_window_t* window)
+{
+    size_t const endT = (size_t)(window->nextSrc - window->base);
+    U32 const end = (U32)endT;
+
+    window->lowLimit = end;
+    window->dictLimit = end;
+}
+
+/**
+ * ZSTD_window_hasExtDict():
+ * Returns non-zero if the window has a non-empty extDict.
+ */
+MEM_STATIC U32 ZSTD_window_hasExtDict(ZSTD_window_t const window)
+{
+    return window.lowLimit < window.dictLimit;
+}
+
+/**
+ * ZSTD_matchState_dictMode():
+ * Inspects the provided matchState and figures out what dictMode should be
+ * passed to the compressor.
+ */
+MEM_STATIC ZSTD_dictMode_e ZSTD_matchState_dictMode(const ZSTD_matchState_t *ms)
+{
+    return ZSTD_window_hasExtDict(ms->window) ?
+        ZSTD_extDict :
+        ms->dictMatchState != NULL ?
+            ZSTD_dictMatchState :
+            ZSTD_noDict;
+}
+
+/**
+ * ZSTD_window_needOverflowCorrection():
+ * Returns non-zero if the indices are getting too large and need overflow
+ * protection.
+ */
+MEM_STATIC U32 ZSTD_window_needOverflowCorrection(ZSTD_window_t const window,
+                                                  void const* srcEnd)
+{
+    U32 const current = (U32)((BYTE const*)srcEnd - window.base);
+    return current > ZSTD_CURRENT_MAX;
+}
+
+/**
+ * ZSTD_window_correctOverflow():
+ * Reduces the indices to protect from index overflow.
+ * Returns the correction made to the indices, which must be applied to every
+ * stored index.
+ *
+ * The least significant cycleLog bits of the indices must remain the same,
+ * which may be 0. Every index up to maxDist in the past must be valid.
+ * NOTE: (maxDist & cycleMask) must be zero.
+ */
+MEM_STATIC U32 ZSTD_window_correctOverflow(ZSTD_window_t* window, U32 cycleLog,
+                                           U32 maxDist, void const* src)
+{
+    /* preemptive overflow correction:
+     * 1. correction is large enough:
+     *    lowLimit > (3<<29) ==> current > 3<<29 + 1<<windowLog
+     *    1<<windowLog <= newCurrent < 1<<chainLog + 1<<windowLog
+     *
+     *    current - newCurrent
+     *    > (3<<29 + 1<<windowLog) - (1<<windowLog + 1<<chainLog)
+     *    > (3<<29) - (1<<chainLog)
+     *    > (3<<29) - (1<<30)             (NOTE: chainLog <= 30)
+     *    > 1<<29
+     *
+     * 2. (ip+ZSTD_CHUNKSIZE_MAX - cctx->base) doesn't overflow:
+     *    After correction, current is less than (1<<chainLog + 1<<windowLog).
+     *    In 64-bit mode we are safe, because we have 64-bit ptrdiff_t.
+     *    In 32-bit mode we are safe, because (chainLog <= 29), so
+     *    ip+ZSTD_CHUNKSIZE_MAX - cctx->base < 1<<32.
+     * 3. (cctx->lowLimit + 1<<windowLog) < 1<<32:
+     *    windowLog <= 31 ==> 3<<29 + 1<<windowLog < 7<<29 < 1<<32.
+     */
+    U32 const cycleMask = (1U << cycleLog) - 1;
+    U32 const current = (U32)((BYTE const*)src - window->base);
+    U32 const currentCycle0 = current & cycleMask;
+    /* Exclude zero so that newCurrent - maxDist >= 1. */
+    U32 const currentCycle1 = currentCycle0 == 0 ? (1U << cycleLog) : currentCycle0;
+    U32 const newCurrent = currentCycle1 + maxDist;
+    U32 const correction = current - newCurrent;
+    assert((maxDist & cycleMask) == 0);
+    assert(current > newCurrent);
+    /* Loose bound, should be around 1<<29 (see above) */
+    assert(correction > 1<<28);
+
+    window->base += correction;
+    window->dictBase += correction;
+    if (window->lowLimit <= correction) window->lowLimit = 1;
+    else window->lowLimit -= correction;
+    if (window->dictLimit <= correction) window->dictLimit = 1;
+    else window->dictLimit -= correction;
+
+    /* Ensure we can still reference the full window. */
+    assert(newCurrent >= maxDist);
+    assert(newCurrent - maxDist >= 1);
+    /* Ensure that lowLimit and dictLimit didn't underflow. */
+    assert(window->lowLimit <= newCurrent);
+    assert(window->dictLimit <= newCurrent);
+
+    DEBUGLOG(4, "Correction of 0x%x bytes to lowLimit=0x%x", correction,
+             window->lowLimit);
+    return correction;
+}
+
+/**
+ * ZSTD_window_enforceMaxDist():
+ * Updates lowLimit so that:
+ *    (srcEnd - base) - lowLimit == maxDist + loadedDictEnd
+ *
+ * It ensures index is valid as long as index >= lowLimit.
+ * This must be called before a block compression call.
+ *
+ * loadedDictEnd is only defined if a dictionary is in use for current compression.
+ * As the name implies, loadedDictEnd represents the index at end of dictionary.
+ * The value lies within context's referential, it can be directly compared to blockEndIdx.
+ *
+ * If loadedDictEndPtr is NULL, no dictionary is in use, and we use loadedDictEnd == 0.
+ * If loadedDictEndPtr is not NULL, we set it to zero after updating lowLimit.
+ * This is because dictionaries are allowed to be referenced fully
+ * as long as the last byte of the dictionary is in the window.
+ * Once input has progressed beyond window size, dictionary cannot be referenced anymore.
+ *
+ * In normal dict mode, the dictionary lies between lowLimit and dictLimit.
+ * In dictMatchState mode, lowLimit and dictLimit are the same,
+ * and the dictionary is below them.
+ * forceWindow and dictMatchState are therefore incompatible.
+ */
+MEM_STATIC void
+ZSTD_window_enforceMaxDist(ZSTD_window_t* window,
+                     const void* blockEnd,
+                           U32   maxDist,
+                           U32*  loadedDictEndPtr,
+                     const ZSTD_matchState_t** dictMatchStatePtr)
+{
+    U32 const blockEndIdx = (U32)((BYTE const*)blockEnd - window->base);
+    U32 const loadedDictEnd = (loadedDictEndPtr != NULL) ? *loadedDictEndPtr : 0;
+    DEBUGLOG(5, "ZSTD_window_enforceMaxDist: blockEndIdx=%u, maxDist=%u, loadedDictEnd=%u",
+                (unsigned)blockEndIdx, (unsigned)maxDist, (unsigned)loadedDictEnd);
+
+    /* - When there is no dictionary : loadedDictEnd == 0.
+         In which case, the test (blockEndIdx > maxDist) is merely to avoid
+         overflowing next operation `newLowLimit = blockEndIdx - maxDist`.
+       - When there is a standard dictionary :
+         Index referential is copied from the dictionary,
+         which means it starts from 0.
+         In which case, loadedDictEnd == dictSize,
+         and it makes sense to compare `blockEndIdx > maxDist + dictSize`
+         since `blockEndIdx` also starts from zero.
+       - When there is an attached dictionary :
+         loadedDictEnd is expressed within the referential of the context,
+         so it can be directly compared against blockEndIdx.
+    */
+    if (blockEndIdx > maxDist + loadedDictEnd) {
+        U32 const newLowLimit = blockEndIdx - maxDist;
+        if (window->lowLimit < newLowLimit) window->lowLimit = newLowLimit;
+        if (window->dictLimit < window->lowLimit) {
+            DEBUGLOG(5, "Update dictLimit to match lowLimit, from %u to %u",
+                        (unsigned)window->dictLimit, (unsigned)window->lowLimit);
+            window->dictLimit = window->lowLimit;
+        }
+        /* On reaching window size, dictionaries are invalidated */
+        if (loadedDictEndPtr) *loadedDictEndPtr = 0;
+        if (dictMatchStatePtr) *dictMatchStatePtr = NULL;
+    }
+}
+
+/* Similar to ZSTD_window_enforceMaxDist(),
+ * but only invalidates dictionary
+ * when input progresses beyond window size.
+ * assumption : loadedDictEndPtr and dictMatchStatePtr are valid (non NULL)
+ *              loadedDictEnd uses same referential as window->base
+ *              maxDist is the window size */
+MEM_STATIC void
+ZSTD_checkDictValidity(const ZSTD_window_t* window,
+                       const void* blockEnd,
+                             U32   maxDist,
+                             U32*  loadedDictEndPtr,
+                       const ZSTD_matchState_t** dictMatchStatePtr)
+{
+    assert(loadedDictEndPtr != NULL);
+    assert(dictMatchStatePtr != NULL);
+    {   U32 const blockEndIdx = (U32)((BYTE const*)blockEnd - window->base);
+        U32 const loadedDictEnd = *loadedDictEndPtr;
+        DEBUGLOG(5, "ZSTD_checkDictValidity: blockEndIdx=%u, maxDist=%u, loadedDictEnd=%u",
+                    (unsigned)blockEndIdx, (unsigned)maxDist, (unsigned)loadedDictEnd);
+        assert(blockEndIdx >= loadedDictEnd);
+
+        if (blockEndIdx > loadedDictEnd + maxDist) {
+            /* On reaching window size, dictionaries are invalidated.
+             * For simplification, if window size is reached anywhere within next block,
+             * the dictionary is invalidated for the full block.
+             */
+            DEBUGLOG(6, "invalidating dictionary for current block (distance > windowSize)");
+            *loadedDictEndPtr = 0;
+            *dictMatchStatePtr = NULL;
+        } else {
+            if (*loadedDictEndPtr != 0) {
+                DEBUGLOG(6, "dictionary considered valid for current block");
+    }   }   }
+}
+
+MEM_STATIC void ZSTD_window_init(ZSTD_window_t* window) {
+    memset(window, 0, sizeof(*window));
+    window->base = (BYTE const*)"";
+    window->dictBase = (BYTE const*)"";
+    window->dictLimit = 1;    /* start from 1, so that 1st position is valid */
+    window->lowLimit = 1;     /* it ensures first and later CCtx usages compress the same */
+    window->nextSrc = window->base + 1;   /* see issue #1241 */
+}
+
+/**
+ * ZSTD_window_update():
+ * Updates the window by appending [src, src + srcSize) to the window.
+ * If it is not contiguous, the current prefix becomes the extDict, and we
+ * forget about the extDict. Handles overlap of the prefix and extDict.
+ * Returns non-zero if the segment is contiguous.
+ */
+MEM_STATIC U32 ZSTD_window_update(ZSTD_window_t* window,
+                                  void const* src, size_t srcSize)
+{
+    BYTE const* const ip = (BYTE const*)src;
+    U32 contiguous = 1;
+    DEBUGLOG(5, "ZSTD_window_update");
+    if (srcSize == 0)
+        return contiguous;
+    assert(window->base != NULL);
+    assert(window->dictBase != NULL);
+    /* Check if blocks follow each other */
+    if (src != window->nextSrc) {
+        /* not contiguous */
+        size_t const distanceFromBase = (size_t)(window->nextSrc - window->base);
+        DEBUGLOG(5, "Non contiguous blocks, new segment starts at %u", window->dictLimit);
+        window->lowLimit = window->dictLimit;
+        assert(distanceFromBase == (size_t)(U32)distanceFromBase);  /* should never overflow */
+        window->dictLimit = (U32)distanceFromBase;
+        window->dictBase = window->base;
+        window->base = ip - distanceFromBase;
+        /* ms->nextToUpdate = window->dictLimit; */
+        if (window->dictLimit - window->lowLimit < HASH_READ_SIZE) window->lowLimit = window->dictLimit;   /* too small extDict */
+        contiguous = 0;
+    }
+    window->nextSrc = ip + srcSize;
+    /* if input and dictionary overlap : reduce dictionary (area presumed modified by input) */
+    if ( (ip+srcSize > window->dictBase + window->lowLimit)
+       & (ip < window->dictBase + window->dictLimit)) {
+        ptrdiff_t const highInputIdx = (ip + srcSize) - window->dictBase;
+        U32 const lowLimitMax = (highInputIdx > (ptrdiff_t)window->dictLimit) ? window->dictLimit : (U32)highInputIdx;
+        window->lowLimit = lowLimitMax;
+        DEBUGLOG(5, "Overlapping extDict and input : new lowLimit = %u", window->lowLimit);
+    }
+    return contiguous;
+}
+
+/**
+ * Returns the lowest allowed match index. It may either be in the ext-dict or the prefix.
+ */
+MEM_STATIC U32 ZSTD_getLowestMatchIndex(const ZSTD_matchState_t* ms, U32 current, unsigned windowLog)
+{
+    U32    const maxDistance = 1U << windowLog;
+    U32    const lowestValid = ms->window.lowLimit;
+    U32    const withinWindow = (current - lowestValid > maxDistance) ? current - maxDistance : lowestValid;
+    U32    const isDictionary = (ms->loadedDictEnd != 0);
+    U32    const matchLowest = isDictionary ? lowestValid : withinWindow;
+    return matchLowest;
+}
+
+/**
+ * Returns the lowest allowed match index in the prefix.
+ */
+MEM_STATIC U32 ZSTD_getLowestPrefixIndex(const ZSTD_matchState_t* ms, U32 current, unsigned windowLog)
+{
+    U32    const maxDistance = 1U << windowLog;
+    U32    const lowestValid = ms->window.dictLimit;
+    U32    const withinWindow = (current - lowestValid > maxDistance) ? current - maxDistance : lowestValid;
+    U32    const isDictionary = (ms->loadedDictEnd != 0);
+    U32    const matchLowest = isDictionary ? lowestValid : withinWindow;
+    return matchLowest;
+}
+
+
+
+/* debug functions */
+#if (DEBUGLEVEL>=2)
+
+MEM_STATIC double ZSTD_fWeight(U32 rawStat)
+{
+    U32 const fp_accuracy = 8;
+    U32 const fp_multiplier = (1 << fp_accuracy);
+    U32 const newStat = rawStat + 1;
+    U32 const hb = ZSTD_highbit32(newStat);
+    U32 const BWeight = hb * fp_multiplier;
+    U32 const FWeight = (newStat << fp_accuracy) >> hb;
+    U32 const weight = BWeight + FWeight;
+    assert(hb + fp_accuracy < 31);
+    return (double)weight / fp_multiplier;
+}
+
+/* display a table content,
+ * listing each element, its frequency, and its predicted bit cost */
+MEM_STATIC void ZSTD_debugTable(const U32* table, U32 max)
+{
+    unsigned u, sum;
+    for (u=0, sum=0; u<=max; u++) sum += table[u];
+    DEBUGLOG(2, "total nb elts: %u", sum);
+    for (u=0; u<=max; u++) {
+        DEBUGLOG(2, "%2u: %5u  (%.2f)",
+                u, table[u], ZSTD_fWeight(sum) - ZSTD_fWeight(table[u]) );
+    }
+}
+
+#endif
+
+
+#if defined (__cplusplus)
+}
+#endif
+
+/* ===============================================================
+ * Shared internal declarations
+ * These prototypes may be called from sources not in lib/compress
+ * =============================================================== */
+
+/* ZSTD_loadCEntropy() :
+ * dict : must point at beginning of a valid zstd dictionary.
+ * return : size of dictionary header (size of magic number + dict ID + entropy tables)
+ * assumptions : magic number supposed already checked
+ *               and dictSize >= 8 */
+size_t ZSTD_loadCEntropy(ZSTD_compressedBlockState_t* bs, void* workspace,
+                         short* offcodeNCount, unsigned* offcodeMaxValue,
+                         const void* const dict, size_t dictSize);
+
+void ZSTD_reset_compressedBlockState(ZSTD_compressedBlockState_t* bs);
+
+/* ==============================================================
+ * Private declarations
+ * These prototypes shall only be called from within lib/compress
+ * ============================================================== */
+
+/* ZSTD_getCParamsFromCCtxParams() :
+ * cParams are built depending on compressionLevel, src size hints,
+ * LDM and manually set compression parameters.
+ * Note: srcSizeHint == 0 means 0!
+ */
+ZSTD_compressionParameters ZSTD_getCParamsFromCCtxParams(
+        const ZSTD_CCtx_params* CCtxParams, U64 srcSizeHint, size_t dictSize);
+
+/*! ZSTD_initCStream_internal() :
+ *  Private use only. Init streaming operation.
+ *  expects params to be valid.
+ *  must receive dict, or cdict, or none, but not both.
+ *  @return : 0, or an error code */
+size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs,
+                     const void* dict, size_t dictSize,
+                     const ZSTD_CDict* cdict,
+                     const ZSTD_CCtx_params* params, unsigned long long pledgedSrcSize);
+
+void ZSTD_resetSeqStore(seqStore_t* ssPtr);
+
+/*! ZSTD_getCParamsFromCDict() :
+ *  as the name implies */
+ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict);
+
+/* ZSTD_compressBegin_advanced_internal() :
+ * Private use only. To be called from zstdmt_compress.c. */
+size_t ZSTD_compressBegin_advanced_internal(ZSTD_CCtx* cctx,
+                                    const void* dict, size_t dictSize,
+                                    ZSTD_dictContentType_e dictContentType,
+                                    ZSTD_dictTableLoadMethod_e dtlm,
+                                    const ZSTD_CDict* cdict,
+                                    const ZSTD_CCtx_params* params,
+                                    unsigned long long pledgedSrcSize);
+
+/* ZSTD_compress_advanced_internal() :
+ * Private use only. To be called from zstdmt_compress.c. */
+size_t ZSTD_compress_advanced_internal(ZSTD_CCtx* cctx,
+                                       void* dst, size_t dstCapacity,
+                                 const void* src, size_t srcSize,
+                                 const void* dict,size_t dictSize,
+                                 const ZSTD_CCtx_params* params);
+
+
+/* ZSTD_writeLastEmptyBlock() :
+ * output an empty Block with end-of-frame mark to complete a frame
+ * @return : size of data written into `dst` (== ZSTD_blockHeaderSize (defined in zstd_internal.h))
+ *           or an error code if `dstCapacity` is too small (<ZSTD_blockHeaderSize)
+ */
+size_t ZSTD_writeLastEmptyBlock(void* dst, size_t dstCapacity);
+
+
+/* ZSTD_referenceExternalSequences() :
+ * Must be called before starting a compression operation.
+ * seqs must parse a prefix of the source.
+ * This cannot be used when long range matching is enabled.
+ * Zstd will use these sequences, and pass the literals to a secondary block
+ * compressor.
+ * @return : An error code on failure.
+ * NOTE: seqs are not verified! Invalid sequences can cause out-of-bounds memory
+ * access and data corruption.
+ */
+size_t ZSTD_referenceExternalSequences(ZSTD_CCtx* cctx, rawSeq* seq, size_t nbSeq);
+
+/** ZSTD_cycleLog() :
+ *  condition for correct operation : hashLog > 1 */
+U32 ZSTD_cycleLog(U32 hashLog, ZSTD_strategy strat);
+
+#endif /* ZSTD_COMPRESS_H */
diff --git a/contrib/zstd/zstd_compress_literals.c b/contrib/zstd/zstd_compress_literals.c
new file mode 100644
index 000000000..17e7168d8
--- /dev/null
+++ b/contrib/zstd/zstd_compress_literals.c
@@ -0,0 +1,158 @@
+/*
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+ /*-*************************************
+ *  Dependencies
+ ***************************************/
+#include "zstd_compress_literals.h"
+
+size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    BYTE* const ostart = (BYTE* const)dst;
+    U32   const flSize = 1 + (srcSize>31) + (srcSize>4095);
+
+    RETURN_ERROR_IF(srcSize + flSize > dstCapacity, dstSize_tooSmall, "");
+
+    switch(flSize)
+    {
+        case 1: /* 2 - 1 - 5 */
+            ostart[0] = (BYTE)((U32)set_basic + (srcSize<<3));
+            break;
+        case 2: /* 2 - 2 - 12 */
+            MEM_writeLE16(ostart, (U16)((U32)set_basic + (1<<2) + (srcSize<<4)));
+            break;
+        case 3: /* 2 - 2 - 20 */
+            MEM_writeLE32(ostart, (U32)((U32)set_basic + (3<<2) + (srcSize<<4)));
+            break;
+        default:   /* not necessary : flSize is {1,2,3} */
+            assert(0);
+    }
+
+    memcpy(ostart + flSize, src, srcSize);
+    DEBUGLOG(5, "Raw literals: %u -> %u", (U32)srcSize, (U32)(srcSize + flSize));
+    return srcSize + flSize;
+}
+
+size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+{
+    BYTE* const ostart = (BYTE* const)dst;
+    U32   const flSize = 1 + (srcSize>31) + (srcSize>4095);
+
+    (void)dstCapacity;  /* dstCapacity already guaranteed to be >=4, hence large enough */
+
+    switch(flSize)
+    {
+        case 1: /* 2 - 1 - 5 */
+            ostart[0] = (BYTE)((U32)set_rle + (srcSize<<3));
+            break;
+        case 2: /* 2 - 2 - 12 */
+            MEM_writeLE16(ostart, (U16)((U32)set_rle + (1<<2) + (srcSize<<4)));
+            break;
+        case 3: /* 2 - 2 - 20 */
+            MEM_writeLE32(ostart, (U32)((U32)set_rle + (3<<2) + (srcSize<<4)));
+            break;
+        default:   /* not necessary : flSize is {1,2,3} */
+            assert(0);
+    }
+
+    ostart[flSize] = *(const BYTE*)src;
+    DEBUGLOG(5, "RLE literals: %u -> %u", (U32)srcSize, (U32)flSize + 1);
+    return flSize+1;
+}
+
+size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
+                              ZSTD_hufCTables_t* nextHuf,
+                              ZSTD_strategy strategy, int disableLiteralCompression,
+                              void* dst, size_t dstCapacity,
+                        const void* src, size_t srcSize,
+                              void* entropyWorkspace, size_t entropyWorkspaceSize,
+                        const int bmi2)
+{
+    size_t const minGain = ZSTD_minGain(srcSize, strategy);
+    size_t const lhSize = 3 + (srcSize >= 1 KB) + (srcSize >= 16 KB);
+    BYTE*  const ostart = (BYTE*)dst;
+    U32 singleStream = srcSize < 256;
+    symbolEncodingType_e hType = set_compressed;
+    size_t cLitSize;
+
+    DEBUGLOG(5,"ZSTD_compressLiterals (disableLiteralCompression=%i srcSize=%u)",
+                disableLiteralCompression, (U32)srcSize);
+
+    /* Prepare nextEntropy assuming reusing the existing table */
+    memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+
+    if (disableLiteralCompression)
+        return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
+
+    /* small ? don't even attempt compression (speed opt) */
+#   define COMPRESS_LITERALS_SIZE_MIN 63
+    {   size_t const minLitSize = (prevHuf->repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN;
+        if (srcSize <= minLitSize) return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
+    }
+
+    RETURN_ERROR_IF(dstCapacity < lhSize+1, dstSize_tooSmall, "not enough space for compression");
+    {   HUF_repeat repeat = prevHuf->repeatMode;
+        int const preferRepeat = strategy < ZSTD_lazy ? srcSize <= 1024 : 0;
+        if (repeat == HUF_repeat_valid && lhSize == 3) singleStream = 1;
+        cLitSize = singleStream ?
+            HUF_compress1X_repeat(
+                ostart+lhSize, dstCapacity-lhSize, src, srcSize,
+                HUF_SYMBOLVALUE_MAX, HUF_TABLELOG_DEFAULT, entropyWorkspace, entropyWorkspaceSize,
+                (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2) :
+            HUF_compress4X_repeat(
+                ostart+lhSize, dstCapacity-lhSize, src, srcSize,
+                HUF_SYMBOLVALUE_MAX, HUF_TABLELOG_DEFAULT, entropyWorkspace, entropyWorkspaceSize,
+                (HUF_CElt*)nextHuf->CTable, &repeat, preferRepeat, bmi2);
+        if (repeat != HUF_repeat_none) {
+            /* reused the existing table */
+            DEBUGLOG(5, "Reusing previous huffman table");
+            hType = set_repeat;
+        }
+    }
+
+    if ((cLitSize==0) | (cLitSize >= srcSize - minGain) | ERR_isError(cLitSize)) {
+        memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+        return ZSTD_noCompressLiterals(dst, dstCapacity, src, srcSize);
+    }
+    if (cLitSize==1) {
+        memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+        return ZSTD_compressRleLiteralsBlock(dst, dstCapacity, src, srcSize);
+    }
+
+    if (hType == set_compressed) {
+        /* using a newly constructed table */
+        nextHuf->repeatMode = HUF_repeat_check;
+    }
+
+    /* Build header */
+    switch(lhSize)
+    {
+    case 3: /* 2 - 2 - 10 - 10 */
+        {   U32 const lhc = hType + ((!singleStream) << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<14);
+            MEM_writeLE24(ostart, lhc);
+            break;
+        }
+    case 4: /* 2 - 2 - 14 - 14 */
+        {   U32 const lhc = hType + (2 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<18);
+            MEM_writeLE32(ostart, lhc);
+            break;
+        }
+    case 5: /* 2 - 2 - 18 - 18 */
+        {   U32 const lhc = hType + (3 << 2) + ((U32)srcSize<<4) + ((U32)cLitSize<<22);
+            MEM_writeLE32(ostart, lhc);
+            ostart[4] = (BYTE)(cLitSize >> 10);
+            break;
+        }
+    default:  /* not possible : lhSize is {3,4,5} */
+        assert(0);
+    }
+    DEBUGLOG(5, "Compressed literals: %u -> %u", (U32)srcSize, (U32)(lhSize+cLitSize));
+    return lhSize+cLitSize;
+}
diff --git a/contrib/zstd/zstd_compress_literals.h b/contrib/zstd/zstd_compress_literals.h
new file mode 100644
index 000000000..8b0870574
--- /dev/null
+++ b/contrib/zstd/zstd_compress_literals.h
@@ -0,0 +1,29 @@
+/*
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_COMPRESS_LITERALS_H
+#define ZSTD_COMPRESS_LITERALS_H
+
+#include "zstd_compress_internal.h" /* ZSTD_hufCTables_t, ZSTD_minGain() */
+
+
+size_t ZSTD_noCompressLiterals (void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+size_t ZSTD_compressRleLiteralsBlock (void* dst, size_t dstCapacity, const void* src, size_t srcSize);
+
+size_t ZSTD_compressLiterals (ZSTD_hufCTables_t const* prevHuf,
+                              ZSTD_hufCTables_t* nextHuf,
+                              ZSTD_strategy strategy, int disableLiteralCompression,
+                              void* dst, size_t dstCapacity,
+                        const void* src, size_t srcSize,
+                              void* entropyWorkspace, size_t entropyWorkspaceSize,
+                        const int bmi2);
+
+#endif /* ZSTD_COMPRESS_LITERALS_H */
diff --git a/contrib/zstd/zstd_compress_sequences.c b/contrib/zstd/zstd_compress_sequences.c
new file mode 100644
index 000000000..f9f8097c8
--- /dev/null
+++ b/contrib/zstd/zstd_compress_sequences.c
@@ -0,0 +1,419 @@
+/*
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+ /*-*************************************
+ *  Dependencies
+ ***************************************/
+#include "zstd_compress_sequences.h"
+
+/**
+ * -log2(x / 256) lookup table for x in [0, 256).
+ * If x == 0: Return 0
+ * Else: Return floor(-log2(x / 256) * 256)
+ */
+static unsigned const kInverseProbabilityLog256[256] = {
+    0,    2048, 1792, 1642, 1536, 1453, 1386, 1329, 1280, 1236, 1197, 1162,
+    1130, 1100, 1073, 1047, 1024, 1001, 980,  960,  941,  923,  906,  889,
+    874,  859,  844,  830,  817,  804,  791,  779,  768,  756,  745,  734,
+    724,  714,  704,  694,  685,  676,  667,  658,  650,  642,  633,  626,
+    618,  610,  603,  595,  588,  581,  574,  567,  561,  554,  548,  542,
+    535,  529,  523,  517,  512,  506,  500,  495,  489,  484,  478,  473,
+    468,  463,  458,  453,  448,  443,  438,  434,  429,  424,  420,  415,
+    411,  407,  402,  398,  394,  390,  386,  382,  377,  373,  370,  366,
+    362,  358,  354,  350,  347,  343,  339,  336,  332,  329,  325,  322,
+    318,  315,  311,  308,  305,  302,  298,  295,  292,  289,  286,  282,
+    279,  276,  273,  270,  267,  264,  261,  258,  256,  253,  250,  247,
+    244,  241,  239,  236,  233,  230,  228,  225,  222,  220,  217,  215,
+    212,  209,  207,  204,  202,  199,  197,  194,  192,  190,  187,  185,
+    182,  180,  178,  175,  173,  171,  168,  166,  164,  162,  159,  157,
+    155,  153,  151,  149,  146,  144,  142,  140,  138,  136,  134,  132,
+    130,  128,  126,  123,  121,  119,  117,  115,  114,  112,  110,  108,
+    106,  104,  102,  100,  98,   96,   94,   93,   91,   89,   87,   85,
+    83,   82,   80,   78,   76,   74,   73,   71,   69,   67,   66,   64,
+    62,   61,   59,   57,   55,   54,   52,   50,   49,   47,   46,   44,
+    42,   41,   39,   37,   36,   34,   33,   31,   30,   28,   26,   25,
+    23,   22,   20,   19,   17,   16,   14,   13,   11,   10,   8,    7,
+    5,    4,    2,    1,
+};
+
+static unsigned ZSTD_getFSEMaxSymbolValue(FSE_CTable const* ctable) {
+  void const* ptr = ctable;
+  U16 const* u16ptr = (U16 const*)ptr;
+  U32 const maxSymbolValue = MEM_read16(u16ptr + 1);
+  return maxSymbolValue;
+}
+
+/**
+ * Returns the cost in bytes of encoding the normalized count header.
+ * Returns an error if any of the helper functions return an error.
+ */
+static size_t ZSTD_NCountCost(unsigned const* count, unsigned const max,
+                              size_t const nbSeq, unsigned const FSELog)
+{
+    BYTE wksp[FSE_NCOUNTBOUND];
+    S16 norm[MaxSeq + 1];
+    const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
+    FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq, max), "");
+    return FSE_writeNCount(wksp, sizeof(wksp), norm, max, tableLog);
+}
+
+/**
+ * Returns the cost in bits of encoding the distribution described by count
+ * using the entropy bound.
+ */
+static size_t ZSTD_entropyCost(unsigned const* count, unsigned const max, size_t const total)
+{
+    unsigned cost = 0;
+    unsigned s;
+    for (s = 0; s <= max; ++s) {
+        unsigned norm = (unsigned)((256 * count[s]) / total);
+        if (count[s] != 0 && norm == 0)
+            norm = 1;
+        assert(count[s] < total);
+        cost += count[s] * kInverseProbabilityLog256[norm];
+    }
+    return cost >> 8;
+}
+
+/**
+ * Returns the cost in bits of encoding the distribution in count using ctable.
+ * Returns an error if ctable cannot represent all the symbols in count.
+ */
+size_t ZSTD_fseBitCost(
+    FSE_CTable const* ctable,
+    unsigned const* count,
+    unsigned const max)
+{
+    unsigned const kAccuracyLog = 8;
+    size_t cost = 0;
+    unsigned s;
+    FSE_CState_t cstate;
+    FSE_initCState(&cstate, ctable);
+    if (ZSTD_getFSEMaxSymbolValue(ctable) < max) {
+        DEBUGLOG(5, "Repeat FSE_CTable has maxSymbolValue %u < %u",
+                    ZSTD_getFSEMaxSymbolValue(ctable), max);
+        return ERROR(GENERIC);
+    }
+    for (s = 0; s <= max; ++s) {
+        unsigned const tableLog = cstate.stateLog;
+        unsigned const badCost = (tableLog + 1) << kAccuracyLog;
+        unsigned const bitCost = FSE_bitCost(cstate.symbolTT, tableLog, s, kAccuracyLog);
+        if (count[s] == 0)
+            continue;
+        if (bitCost >= badCost) {
+            DEBUGLOG(5, "Repeat FSE_CTable has Prob[%u] == 0", s);
+            return ERROR(GENERIC);
+        }
+        cost += (size_t)count[s] * bitCost;
+    }
+    return cost >> kAccuracyLog;
+}
+
+/**
+ * Returns the cost in bits of encoding the distribution in count using the
+ * table described by norm. The max symbol support by norm is assumed >= max.
+ * norm must be valid for every symbol with non-zero probability in count.
+ */
+size_t ZSTD_crossEntropyCost(short const* norm, unsigned accuracyLog,
+                             unsigned const* count, unsigned const max)
+{
+    unsigned const shift = 8 - accuracyLog;
+    size_t cost = 0;
+    unsigned s;
+    assert(accuracyLog <= 8);
+    for (s = 0; s <= max; ++s) {
+        unsigned const normAcc = (norm[s] != -1) ? (unsigned)norm[s] : 1;
+        unsigned const norm256 = normAcc << shift;
+        assert(norm256 > 0);
+        assert(norm256 < 256);
+        cost += count[s] * kInverseProbabilityLog256[norm256];
+    }
+    return cost >> 8;
+}
+
+symbolEncodingType_e
+ZSTD_selectEncodingType(
+        FSE_repeat* repeatMode, unsigned const* count, unsigned const max,
+        size_t const mostFrequent, size_t nbSeq, unsigned const FSELog,
+        FSE_CTable const* prevCTable,
+        short const* defaultNorm, U32 defaultNormLog,
+        ZSTD_defaultPolicy_e const isDefaultAllowed,
+        ZSTD_strategy const strategy)
+{
+    ZSTD_STATIC_ASSERT(ZSTD_defaultDisallowed == 0 && ZSTD_defaultAllowed != 0);
+    if (mostFrequent == nbSeq) {
+        *repeatMode = FSE_repeat_none;
+        if (isDefaultAllowed && nbSeq <= 2) {
+            /* Prefer set_basic over set_rle when there are 2 or less symbols,
+             * since RLE uses 1 byte, but set_basic uses 5-6 bits per symbol.
+             * If basic encoding isn't possible, always choose RLE.
+             */
+            DEBUGLOG(5, "Selected set_basic");
+            return set_basic;
+        }
+        DEBUGLOG(5, "Selected set_rle");
+        return set_rle;
+    }
+    if (strategy < ZSTD_lazy) {
+        if (isDefaultAllowed) {
+            size_t const staticFse_nbSeq_max = 1000;
+            size_t const mult = 10 - strategy;
+            size_t const baseLog = 3;
+            size_t const dynamicFse_nbSeq_min = (((size_t)1 << defaultNormLog) * mult) >> baseLog;  /* 28-36 for offset, 56-72 for lengths */
+            assert(defaultNormLog >= 5 && defaultNormLog <= 6);  /* xx_DEFAULTNORMLOG */
+            assert(mult <= 9 && mult >= 7);
+            if ( (*repeatMode == FSE_repeat_valid)
+              && (nbSeq < staticFse_nbSeq_max) ) {
+                DEBUGLOG(5, "Selected set_repeat");
+                return set_repeat;
+            }
+            if ( (nbSeq < dynamicFse_nbSeq_min)
+              || (mostFrequent < (nbSeq >> (defaultNormLog-1))) ) {
+                DEBUGLOG(5, "Selected set_basic");
+                /* The format allows default tables to be repeated, but it isn't useful.
+                 * When using simple heuristics to select encoding type, we don't want
+                 * to confuse these tables with dictionaries. When running more careful
+                 * analysis, we don't need to waste time checking both repeating tables
+                 * and default tables.
+                 */
+                *repeatMode = FSE_repeat_none;
+                return set_basic;
+            }
+        }
+    } else {
+        size_t const basicCost = isDefaultAllowed ? ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, count, max) : ERROR(GENERIC);
+        size_t const repeatCost = *repeatMode != FSE_repeat_none ? ZSTD_fseBitCost(prevCTable, count, max) : ERROR(GENERIC);
+        size_t const NCountCost = ZSTD_NCountCost(count, max, nbSeq, FSELog);
+        size_t const compressedCost = (NCountCost << 3) + ZSTD_entropyCost(count, max, nbSeq);
+
+        if (isDefaultAllowed) {
+            assert(!ZSTD_isError(basicCost));
+            assert(!(*repeatMode == FSE_repeat_valid && ZSTD_isError(repeatCost)));
+        }
+        assert(!ZSTD_isError(NCountCost));
+        assert(compressedCost < ERROR(maxCode));
+        DEBUGLOG(5, "Estimated bit costs: basic=%u\trepeat=%u\tcompressed=%u",
+                    (unsigned)basicCost, (unsigned)repeatCost, (unsigned)compressedCost);
+        if (basicCost <= repeatCost && basicCost <= compressedCost) {
+            DEBUGLOG(5, "Selected set_basic");
+            assert(isDefaultAllowed);
+            *repeatMode = FSE_repeat_none;
+            return set_basic;
+        }
+        if (repeatCost <= compressedCost) {
+            DEBUGLOG(5, "Selected set_repeat");
+            assert(!ZSTD_isError(repeatCost));
+            return set_repeat;
+        }
+        assert(compressedCost < basicCost && compressedCost < repeatCost);
+    }
+    DEBUGLOG(5, "Selected set_compressed");
+    *repeatMode = FSE_repeat_check;
+    return set_compressed;
+}
+
+size_t
+ZSTD_buildCTable(void* dst, size_t dstCapacity,
+                FSE_CTable* nextCTable, U32 FSELog, symbolEncodingType_e type,
+                unsigned* count, U32 max,
+                const BYTE* codeTable, size_t nbSeq,
+                const S16* defaultNorm, U32 defaultNormLog, U32 defaultMax,
+                const FSE_CTable* prevCTable, size_t prevCTableSize,
+                void* entropyWorkspace, size_t entropyWorkspaceSize)
+{
+    BYTE* op = (BYTE*)dst;
+    const BYTE* const oend = op + dstCapacity;
+    DEBUGLOG(6, "ZSTD_buildCTable (dstCapacity=%u)", (unsigned)dstCapacity);
+
+    switch (type) {
+    case set_rle:
+        FORWARD_IF_ERROR(FSE_buildCTable_rle(nextCTable, (BYTE)max), "");
+        RETURN_ERROR_IF(dstCapacity==0, dstSize_tooSmall, "not enough space");
+        *op = codeTable[0];
+        return 1;
+    case set_repeat:
+        memcpy(nextCTable, prevCTable, prevCTableSize);
+        return 0;
+    case set_basic:
+        FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, defaultNorm, defaultMax, defaultNormLog, entropyWorkspace, entropyWorkspaceSize), "");  /* note : could be pre-calculated */
+        return 0;
+    case set_compressed: {
+        S16 norm[MaxSeq + 1];
+        size_t nbSeq_1 = nbSeq;
+        const U32 tableLog = FSE_optimalTableLog(FSELog, nbSeq, max);
+        if (count[codeTable[nbSeq-1]] > 1) {
+            count[codeTable[nbSeq-1]]--;
+            nbSeq_1--;
+        }
+        assert(nbSeq_1 > 1);
+        FORWARD_IF_ERROR(FSE_normalizeCount(norm, tableLog, count, nbSeq_1, max), "");
+        {   size_t const NCountSize = FSE_writeNCount(op, oend - op, norm, max, tableLog);   /* overflow protected */
+            FORWARD_IF_ERROR(NCountSize, "FSE_writeNCount failed");
+            FORWARD_IF_ERROR(FSE_buildCTable_wksp(nextCTable, norm, max, tableLog, entropyWorkspace, entropyWorkspaceSize), "");
+            return NCountSize;
+        }
+    }
+    default: assert(0); RETURN_ERROR(GENERIC, "impossible to reach");
+    }
+}
+
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_encodeSequences_body(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets)
+{
+    BIT_CStream_t blockStream;
+    FSE_CState_t  stateMatchLength;
+    FSE_CState_t  stateOffsetBits;
+    FSE_CState_t  stateLitLength;
+
+    RETURN_ERROR_IF(
+        ERR_isError(BIT_initCStream(&blockStream, dst, dstCapacity)),
+        dstSize_tooSmall, "not enough space remaining");
+    DEBUGLOG(6, "available space for bitstream : %i  (dstCapacity=%u)",
+                (int)(blockStream.endPtr - blockStream.startPtr),
+                (unsigned)dstCapacity);
+
+    /* first symbols */
+    FSE_initCState2(&stateMatchLength, CTable_MatchLength, mlCodeTable[nbSeq-1]);
+    FSE_initCState2(&stateOffsetBits,  CTable_OffsetBits,  ofCodeTable[nbSeq-1]);
+    FSE_initCState2(&stateLitLength,   CTable_LitLength,   llCodeTable[nbSeq-1]);
+    BIT_addBits(&blockStream, sequences[nbSeq-1].litLength, LL_bits[llCodeTable[nbSeq-1]]);
+    if (MEM_32bits()) BIT_flushBits(&blockStream);
+    BIT_addBits(&blockStream, sequences[nbSeq-1].matchLength, ML_bits[mlCodeTable[nbSeq-1]]);
+    if (MEM_32bits()) BIT_flushBits(&blockStream);
+    if (longOffsets) {
+        U32 const ofBits = ofCodeTable[nbSeq-1];
+        unsigned const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
+        if (extraBits) {
+            BIT_addBits(&blockStream, sequences[nbSeq-1].offset, extraBits);
+            BIT_flushBits(&blockStream);
+        }
+        BIT_addBits(&blockStream, sequences[nbSeq-1].offset >> extraBits,
+                    ofBits - extraBits);
+    } else {
+        BIT_addBits(&blockStream, sequences[nbSeq-1].offset, ofCodeTable[nbSeq-1]);
+    }
+    BIT_flushBits(&blockStream);
+
+    {   size_t n;
+        for (n=nbSeq-2 ; n<nbSeq ; n--) {      /* intentional underflow */
+            BYTE const llCode = llCodeTable[n];
+            BYTE const ofCode = ofCodeTable[n];
+            BYTE const mlCode = mlCodeTable[n];
+            U32  const llBits = LL_bits[llCode];
+            U32  const ofBits = ofCode;
+            U32  const mlBits = ML_bits[mlCode];
+            DEBUGLOG(6, "encoding: litlen:%2u - matchlen:%2u - offCode:%7u",
+                        (unsigned)sequences[n].litLength,
+                        (unsigned)sequences[n].matchLength + MINMATCH,
+                        (unsigned)sequences[n].offset);
+                                                                            /* 32b*/  /* 64b*/
+                                                                            /* (7)*/  /* (7)*/
+            FSE_encodeSymbol(&blockStream, &stateOffsetBits, ofCode);       /* 15 */  /* 15 */
+            FSE_encodeSymbol(&blockStream, &stateMatchLength, mlCode);      /* 24 */  /* 24 */
+            if (MEM_32bits()) BIT_flushBits(&blockStream);                  /* (7)*/
+            FSE_encodeSymbol(&blockStream, &stateLitLength, llCode);        /* 16 */  /* 33 */
+            if (MEM_32bits() || (ofBits+mlBits+llBits >= 64-7-(LLFSELog+MLFSELog+OffFSELog)))
+                BIT_flushBits(&blockStream);                                /* (7)*/
+            BIT_addBits(&blockStream, sequences[n].litLength, llBits);
+            if (MEM_32bits() && ((llBits+mlBits)>24)) BIT_flushBits(&blockStream);
+            BIT_addBits(&blockStream, sequences[n].matchLength, mlBits);
+            if (MEM_32bits() || (ofBits+mlBits+llBits > 56)) BIT_flushBits(&blockStream);
+            if (longOffsets) {
+                unsigned const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN-1);
+                if (extraBits) {
+                    BIT_addBits(&blockStream, sequences[n].offset, extraBits);
+                    BIT_flushBits(&blockStream);                            /* (7)*/
+                }
+                BIT_addBits(&blockStream, sequences[n].offset >> extraBits,
+                            ofBits - extraBits);                            /* 31 */
+            } else {
+                BIT_addBits(&blockStream, sequences[n].offset, ofBits);     /* 31 */
+            }
+            BIT_flushBits(&blockStream);                                    /* (7)*/
+            DEBUGLOG(7, "remaining space : %i", (int)(blockStream.endPtr - blockStream.ptr));
+    }   }
+
+    DEBUGLOG(6, "ZSTD_encodeSequences: flushing ML state with %u bits", stateMatchLength.stateLog);
+    FSE_flushCState(&blockStream, &stateMatchLength);
+    DEBUGLOG(6, "ZSTD_encodeSequences: flushing Off state with %u bits", stateOffsetBits.stateLog);
+    FSE_flushCState(&blockStream, &stateOffsetBits);
+    DEBUGLOG(6, "ZSTD_encodeSequences: flushing LL state with %u bits", stateLitLength.stateLog);
+    FSE_flushCState(&blockStream, &stateLitLength);
+
+    {   size_t const streamSize = BIT_closeCStream(&blockStream);
+        RETURN_ERROR_IF(streamSize==0, dstSize_tooSmall, "not enough space");
+        return streamSize;
+    }
+}
+
+static size_t
+ZSTD_encodeSequences_default(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets)
+{
+    return ZSTD_encodeSequences_body(dst, dstCapacity,
+                                    CTable_MatchLength, mlCodeTable,
+                                    CTable_OffsetBits, ofCodeTable,
+                                    CTable_LitLength, llCodeTable,
+                                    sequences, nbSeq, longOffsets);
+}
+
+
+#if DYNAMIC_BMI2
+
+static TARGET_ATTRIBUTE("bmi2") size_t
+ZSTD_encodeSequences_bmi2(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets)
+{
+    return ZSTD_encodeSequences_body(dst, dstCapacity,
+                                    CTable_MatchLength, mlCodeTable,
+                                    CTable_OffsetBits, ofCodeTable,
+                                    CTable_LitLength, llCodeTable,
+                                    sequences, nbSeq, longOffsets);
+}
+
+#endif
+
+size_t ZSTD_encodeSequences(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets, int bmi2)
+{
+    DEBUGLOG(5, "ZSTD_encodeSequences: dstCapacity = %u", (unsigned)dstCapacity);
+#if DYNAMIC_BMI2
+    if (bmi2) {
+        return ZSTD_encodeSequences_bmi2(dst, dstCapacity,
+                                         CTable_MatchLength, mlCodeTable,
+                                         CTable_OffsetBits, ofCodeTable,
+                                         CTable_LitLength, llCodeTable,
+                                         sequences, nbSeq, longOffsets);
+    }
+#endif
+    (void)bmi2;
+    return ZSTD_encodeSequences_default(dst, dstCapacity,
+                                        CTable_MatchLength, mlCodeTable,
+                                        CTable_OffsetBits, ofCodeTable,
+                                        CTable_LitLength, llCodeTable,
+                                        sequences, nbSeq, longOffsets);
+}
diff --git a/contrib/zstd/zstd_compress_sequences.h b/contrib/zstd/zstd_compress_sequences.h
new file mode 100644
index 000000000..b884e45b4
--- /dev/null
+++ b/contrib/zstd/zstd_compress_sequences.h
@@ -0,0 +1,54 @@
+/*
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_COMPRESS_SEQUENCES_H
+#define ZSTD_COMPRESS_SEQUENCES_H
+
+#include "fse.h" /* FSE_repeat, FSE_CTable */
+#include "zstd_internal.h" /* symbolEncodingType_e, ZSTD_strategy */
+
+typedef enum {
+    ZSTD_defaultDisallowed = 0,
+    ZSTD_defaultAllowed = 1
+} ZSTD_defaultPolicy_e;
+
+symbolEncodingType_e
+ZSTD_selectEncodingType(
+        FSE_repeat* repeatMode, unsigned const* count, unsigned const max,
+        size_t const mostFrequent, size_t nbSeq, unsigned const FSELog,
+        FSE_CTable const* prevCTable,
+        short const* defaultNorm, U32 defaultNormLog,
+        ZSTD_defaultPolicy_e const isDefaultAllowed,
+        ZSTD_strategy const strategy);
+
+size_t
+ZSTD_buildCTable(void* dst, size_t dstCapacity,
+                FSE_CTable* nextCTable, U32 FSELog, symbolEncodingType_e type,
+                unsigned* count, U32 max,
+                const BYTE* codeTable, size_t nbSeq,
+                const S16* defaultNorm, U32 defaultNormLog, U32 defaultMax,
+                const FSE_CTable* prevCTable, size_t prevCTableSize,
+                void* entropyWorkspace, size_t entropyWorkspaceSize);
+
+size_t ZSTD_encodeSequences(
+            void* dst, size_t dstCapacity,
+            FSE_CTable const* CTable_MatchLength, BYTE const* mlCodeTable,
+            FSE_CTable const* CTable_OffsetBits, BYTE const* ofCodeTable,
+            FSE_CTable const* CTable_LitLength, BYTE const* llCodeTable,
+            seqDef const* sequences, size_t nbSeq, int longOffsets, int bmi2);
+
+size_t ZSTD_fseBitCost(
+    FSE_CTable const* ctable,
+    unsigned const* count,
+    unsigned const max);
+
+size_t ZSTD_crossEntropyCost(short const* norm, unsigned accuracyLog,
+                             unsigned const* count, unsigned const max);
+#endif /* ZSTD_COMPRESS_SEQUENCES_H */
diff --git a/contrib/zstd/zstd_compress_superblock.c b/contrib/zstd/zstd_compress_superblock.c
new file mode 100644
index 000000000..d72a8000a
--- /dev/null
+++ b/contrib/zstd/zstd_compress_superblock.c
@@ -0,0 +1,845 @@
+/*
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+ /*-*************************************
+ *  Dependencies
+ ***************************************/
+#include "zstd_compress_superblock.h"
+
+#include "zstd_internal.h"  /* ZSTD_getSequenceLength */
+#include "hist.h"                     /* HIST_countFast_wksp */
+#include "zstd_compress_internal.h"
+#include "zstd_compress_sequences.h"
+#include "zstd_compress_literals.h"
+
+/*-*************************************
+*  Superblock entropy buffer structs
+***************************************/
+/** ZSTD_hufCTablesMetadata_t :
+ *  Stores Literals Block Type for a super-block in hType, and
+ *  huffman tree description in hufDesBuffer.
+ *  hufDesSize refers to the size of huffman tree description in bytes.
+ *  This metadata is populated in ZSTD_buildSuperBlockEntropy_literal() */
+typedef struct {
+    symbolEncodingType_e hType;
+    BYTE hufDesBuffer[500]; /* TODO give name to this value */
+    size_t hufDesSize;
+} ZSTD_hufCTablesMetadata_t;
+
+/** ZSTD_fseCTablesMetadata_t :
+ *  Stores symbol compression modes for a super-block in {ll, ol, ml}Type, and
+ *  fse tables in fseTablesBuffer.
+ *  fseTablesSize refers to the size of fse tables in bytes.
+ *  This metadata is populated in ZSTD_buildSuperBlockEntropy_sequences() */
+typedef struct {
+    symbolEncodingType_e llType;
+    symbolEncodingType_e ofType;
+    symbolEncodingType_e mlType;
+    BYTE fseTablesBuffer[500]; /* TODO give name to this value */
+    size_t fseTablesSize;
+    size_t lastCountSize; /* This is to account for bug in 1.3.4. More detail in ZSTD_compressSubBlock_sequences() */
+} ZSTD_fseCTablesMetadata_t;
+
+typedef struct {
+    ZSTD_hufCTablesMetadata_t hufMetadata;
+    ZSTD_fseCTablesMetadata_t fseMetadata;
+} ZSTD_entropyCTablesMetadata_t;
+
+
+/** ZSTD_buildSuperBlockEntropy_literal() :
+ *  Builds entropy for the super-block literals.
+ *  Stores literals block type (raw, rle, compressed, repeat) and
+ *  huffman description table to hufMetadata.
+ *  @return : size of huffman description table or error code */
+static size_t ZSTD_buildSuperBlockEntropy_literal(void* const src, size_t srcSize,
+                                            const ZSTD_hufCTables_t* prevHuf,
+                                                  ZSTD_hufCTables_t* nextHuf,
+                                                  ZSTD_hufCTablesMetadata_t* hufMetadata,
+                                                  const int disableLiteralsCompression,
+                                                  void* workspace, size_t wkspSize)
+{
+    BYTE* const wkspStart = (BYTE*)workspace;
+    BYTE* const wkspEnd = wkspStart + wkspSize;
+    BYTE* const countWkspStart = wkspStart;
+    unsigned* const countWksp = (unsigned*)workspace;
+    const size_t countWkspSize = (HUF_SYMBOLVALUE_MAX + 1) * sizeof(unsigned);
+    BYTE* const nodeWksp = countWkspStart + countWkspSize;
+    const size_t nodeWkspSize = wkspEnd-nodeWksp;
+    unsigned maxSymbolValue = 255;
+    unsigned huffLog = HUF_TABLELOG_DEFAULT;
+    HUF_repeat repeat = prevHuf->repeatMode;
+
+    DEBUGLOG(5, "ZSTD_buildSuperBlockEntropy_literal (srcSize=%zu)", srcSize);
+
+    /* Prepare nextEntropy assuming reusing the existing table */
+    memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+
+    if (disableLiteralsCompression) {
+        DEBUGLOG(5, "set_basic - disabled");
+        hufMetadata->hType = set_basic;
+        return 0;
+    }
+
+    /* small ? don't even attempt compression (speed opt) */
+#   define COMPRESS_LITERALS_SIZE_MIN 63
+    {   size_t const minLitSize = (prevHuf->repeatMode == HUF_repeat_valid) ? 6 : COMPRESS_LITERALS_SIZE_MIN;
+        if (srcSize <= minLitSize) {
+            DEBUGLOG(5, "set_basic - too small");
+            hufMetadata->hType = set_basic;
+            return 0;
+        }
+    }
+
+    /* Scan input and build symbol stats */
+    {   size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)src, srcSize, workspace, wkspSize);
+        FORWARD_IF_ERROR(largest, "HIST_count_wksp failed");
+        if (largest == srcSize) {
+            DEBUGLOG(5, "set_rle");
+            hufMetadata->hType = set_rle;
+            return 0;
+        }
+        if (largest <= (srcSize >> 7)+4) {
+            DEBUGLOG(5, "set_basic - no gain");
+            hufMetadata->hType = set_basic;
+            return 0;
+        }
+    }
+
+    /* Validate the previous Huffman table */
+    if (repeat == HUF_repeat_check && !HUF_validateCTable((HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue)) {
+        repeat = HUF_repeat_none;
+    }
+
+    /* Build Huffman Tree */
+    memset(nextHuf->CTable, 0, sizeof(nextHuf->CTable));
+    huffLog = HUF_optimalTableLog(huffLog, srcSize, maxSymbolValue);
+    {   size_t const maxBits = HUF_buildCTable_wksp((HUF_CElt*)nextHuf->CTable, countWksp,
+                                                    maxSymbolValue, huffLog,
+                                                    nodeWksp, nodeWkspSize);
+        FORWARD_IF_ERROR(maxBits, "HUF_buildCTable_wksp");
+        huffLog = (U32)maxBits;
+        {   /* Build and write the CTable */
+            size_t const newCSize = HUF_estimateCompressedSize(
+                    (HUF_CElt*)nextHuf->CTable, countWksp, maxSymbolValue);
+            size_t const hSize = HUF_writeCTable(
+                    hufMetadata->hufDesBuffer, sizeof(hufMetadata->hufDesBuffer),
+                    (HUF_CElt*)nextHuf->CTable, maxSymbolValue, huffLog);
+            /* Check against repeating the previous CTable */
+            if (repeat != HUF_repeat_none) {
+                size_t const oldCSize = HUF_estimateCompressedSize(
+                        (HUF_CElt const*)prevHuf->CTable, countWksp, maxSymbolValue);
+                if (oldCSize < srcSize && (oldCSize <= hSize + newCSize || hSize + 12 >= srcSize)) {
+                    DEBUGLOG(5, "set_repeat - smaller");
+                    memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+                    hufMetadata->hType = set_repeat;
+                    return 0;
+                }
+            }
+            if (newCSize + hSize >= srcSize) {
+                DEBUGLOG(5, "set_basic - no gains");
+                memcpy(nextHuf, prevHuf, sizeof(*prevHuf));
+                hufMetadata->hType = set_basic;
+                return 0;
+            }
+            DEBUGLOG(5, "set_compressed (hSize=%u)", (U32)hSize);
+            hufMetadata->hType = set_compressed;
+            nextHuf->repeatMode = HUF_repeat_check;
+            return hSize;
+        }
+    }
+}
+
+/** ZSTD_buildSuperBlockEntropy_sequences() :
+ *  Builds entropy for the super-block sequences.
+ *  Stores symbol compression modes and fse table to fseMetadata.
+ *  @return : size of fse tables or error code */
+static size_t ZSTD_buildSuperBlockEntropy_sequences(seqStore_t* seqStorePtr,
+                                              const ZSTD_fseCTables_t* prevEntropy,
+                                                    ZSTD_fseCTables_t* nextEntropy,
+                                              const ZSTD_CCtx_params* cctxParams,
+                                                    ZSTD_fseCTablesMetadata_t* fseMetadata,
+                                                    void* workspace, size_t wkspSize)
+{
+    BYTE* const wkspStart = (BYTE*)workspace;
+    BYTE* const wkspEnd = wkspStart + wkspSize;
+    BYTE* const countWkspStart = wkspStart;
+    unsigned* const countWksp = (unsigned*)workspace;
+    const size_t countWkspSize = (MaxSeq + 1) * sizeof(unsigned);
+    BYTE* const cTableWksp = countWkspStart + countWkspSize;
+    const size_t cTableWkspSize = wkspEnd-cTableWksp;
+    ZSTD_strategy const strategy = cctxParams->cParams.strategy;
+    FSE_CTable* CTable_LitLength = nextEntropy->litlengthCTable;
+    FSE_CTable* CTable_OffsetBits = nextEntropy->offcodeCTable;
+    FSE_CTable* CTable_MatchLength = nextEntropy->matchlengthCTable;
+    const BYTE* const ofCodeTable = seqStorePtr->ofCode;
+    const BYTE* const llCodeTable = seqStorePtr->llCode;
+    const BYTE* const mlCodeTable = seqStorePtr->mlCode;
+    size_t const nbSeq = seqStorePtr->sequences - seqStorePtr->sequencesStart;
+    BYTE* const ostart = fseMetadata->fseTablesBuffer;
+    BYTE* const oend = ostart + sizeof(fseMetadata->fseTablesBuffer);
+    BYTE* op = ostart;
+
+    assert(cTableWkspSize >= (1 << MaxFSELog) * sizeof(FSE_FUNCTION_TYPE));
+    DEBUGLOG(5, "ZSTD_buildSuperBlockEntropy_sequences (nbSeq=%zu)", nbSeq);
+    memset(workspace, 0, wkspSize);
+
+    fseMetadata->lastCountSize = 0;
+    /* convert length/distances into codes */
+    ZSTD_seqToCodes(seqStorePtr);
+    /* build CTable for Literal Lengths */
+    {   U32 LLtype;
+        unsigned max = MaxLL;
+        size_t const mostFrequent = HIST_countFast_wksp(countWksp, &max, llCodeTable, nbSeq, workspace, wkspSize);  /* can't fail */
+        DEBUGLOG(5, "Building LL table");
+        nextEntropy->litlength_repeatMode = prevEntropy->litlength_repeatMode;
+        LLtype = ZSTD_selectEncodingType(&nextEntropy->litlength_repeatMode,
+                                        countWksp, max, mostFrequent, nbSeq,
+                                        LLFSELog, prevEntropy->litlengthCTable,
+                                        LL_defaultNorm, LL_defaultNormLog,
+                                        ZSTD_defaultAllowed, strategy);
+        assert(set_basic < set_compressed && set_rle < set_compressed);
+        assert(!(LLtype < set_compressed && nextEntropy->litlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */
+        {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_LitLength, LLFSELog, (symbolEncodingType_e)LLtype,
+                                                    countWksp, max, llCodeTable, nbSeq, LL_defaultNorm, LL_defaultNormLog, MaxLL,
+                                                    prevEntropy->litlengthCTable, sizeof(prevEntropy->litlengthCTable),
+                                                    cTableWksp, cTableWkspSize);
+            FORWARD_IF_ERROR(countSize, "ZSTD_buildCTable for LitLens failed");
+            if (LLtype == set_compressed)
+                fseMetadata->lastCountSize = countSize;
+            op += countSize;
+            fseMetadata->llType = (symbolEncodingType_e) LLtype;
+    }   }
+    /* build CTable for Offsets */
+    {   U32 Offtype;
+        unsigned max = MaxOff;
+        size_t const mostFrequent = HIST_countFast_wksp(countWksp, &max, ofCodeTable, nbSeq, workspace, wkspSize);  /* can't fail */
+        /* We can only use the basic table if max <= DefaultMaxOff, otherwise the offsets are too large */
+        ZSTD_defaultPolicy_e const defaultPolicy = (max <= DefaultMaxOff) ? ZSTD_defaultAllowed : ZSTD_defaultDisallowed;
+        DEBUGLOG(5, "Building OF table");
+        nextEntropy->offcode_repeatMode = prevEntropy->offcode_repeatMode;
+        Offtype = ZSTD_selectEncodingType(&nextEntropy->offcode_repeatMode,
+                                        countWksp, max, mostFrequent, nbSeq,
+                                        OffFSELog, prevEntropy->offcodeCTable,
+                                        OF_defaultNorm, OF_defaultNormLog,
+                                        defaultPolicy, strategy);
+        assert(!(Offtype < set_compressed && nextEntropy->offcode_repeatMode != FSE_repeat_none)); /* We don't copy tables */
+        {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_OffsetBits, OffFSELog, (symbolEncodingType_e)Offtype,
+                                                    countWksp, max, ofCodeTable, nbSeq, OF_defaultNorm, OF_defaultNormLog, DefaultMaxOff,
+                                                    prevEntropy->offcodeCTable, sizeof(prevEntropy->offcodeCTable),
+                                                    cTableWksp, cTableWkspSize);
+            FORWARD_IF_ERROR(countSize, "ZSTD_buildCTable for Offsets failed");
+            if (Offtype == set_compressed)
+                fseMetadata->lastCountSize = countSize;
+            op += countSize;
+            fseMetadata->ofType = (symbolEncodingType_e) Offtype;
+    }   }
+    /* build CTable for MatchLengths */
+    {   U32 MLtype;
+        unsigned max = MaxML;
+        size_t const mostFrequent = HIST_countFast_wksp(countWksp, &max, mlCodeTable, nbSeq, workspace, wkspSize);   /* can't fail */
+        DEBUGLOG(5, "Building ML table (remaining space : %i)", (int)(oend-op));
+        nextEntropy->matchlength_repeatMode = prevEntropy->matchlength_repeatMode;
+        MLtype = ZSTD_selectEncodingType(&nextEntropy->matchlength_repeatMode,
+                                        countWksp, max, mostFrequent, nbSeq,
+                                        MLFSELog, prevEntropy->matchlengthCTable,
+                                        ML_defaultNorm, ML_defaultNormLog,
+                                        ZSTD_defaultAllowed, strategy);
+        assert(!(MLtype < set_compressed && nextEntropy->matchlength_repeatMode != FSE_repeat_none)); /* We don't copy tables */
+        {   size_t const countSize = ZSTD_buildCTable(op, oend - op, CTable_MatchLength, MLFSELog, (symbolEncodingType_e)MLtype,
+                                                    countWksp, max, mlCodeTable, nbSeq, ML_defaultNorm, ML_defaultNormLog, MaxML,
+                                                    prevEntropy->matchlengthCTable, sizeof(prevEntropy->matchlengthCTable),
+                                                    cTableWksp, cTableWkspSize);
+            FORWARD_IF_ERROR(countSize, "ZSTD_buildCTable for MatchLengths failed");
+            if (MLtype == set_compressed)
+                fseMetadata->lastCountSize = countSize;
+            op += countSize;
+            fseMetadata->mlType = (symbolEncodingType_e) MLtype;
+    }   }
+    assert((size_t) (op-ostart) <= sizeof(fseMetadata->fseTablesBuffer));
+    return op-ostart;
+}
+
+
+/** ZSTD_buildSuperBlockEntropy() :
+ *  Builds entropy for the super-block.
+ *  @return : 0 on success or error code */
+static size_t
+ZSTD_buildSuperBlockEntropy(seqStore_t* seqStorePtr,
+                      const ZSTD_entropyCTables_t* prevEntropy,
+                            ZSTD_entropyCTables_t* nextEntropy,
+                      const ZSTD_CCtx_params* cctxParams,
+                            ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                            void* workspace, size_t wkspSize)
+{
+    size_t const litSize = seqStorePtr->lit - seqStorePtr->litStart;
+    DEBUGLOG(5, "ZSTD_buildSuperBlockEntropy");
+    entropyMetadata->hufMetadata.hufDesSize =
+        ZSTD_buildSuperBlockEntropy_literal(seqStorePtr->litStart, litSize,
+                                            &prevEntropy->huf, &nextEntropy->huf,
+                                            &entropyMetadata->hufMetadata,
+                                            ZSTD_disableLiteralsCompression(cctxParams),
+                                            workspace, wkspSize);
+    FORWARD_IF_ERROR(entropyMetadata->hufMetadata.hufDesSize, "ZSTD_buildSuperBlockEntropy_literal failed");
+    entropyMetadata->fseMetadata.fseTablesSize =
+        ZSTD_buildSuperBlockEntropy_sequences(seqStorePtr,
+                                              &prevEntropy->fse, &nextEntropy->fse,
+                                              cctxParams,
+                                              &entropyMetadata->fseMetadata,
+                                              workspace, wkspSize);
+    FORWARD_IF_ERROR(entropyMetadata->fseMetadata.fseTablesSize, "ZSTD_buildSuperBlockEntropy_sequences failed");
+    return 0;
+}
+
+/** ZSTD_compressSubBlock_literal() :
+ *  Compresses literals section for a sub-block.
+ *  When we have to write the Huffman table we will sometimes choose a header
+ *  size larger than necessary. This is because we have to pick the header size
+ *  before we know the table size + compressed size, so we have a bound on the
+ *  table size. If we guessed incorrectly, we fall back to uncompressed literals.
+ *
+ *  We write the header when writeEntropy=1 and set entropyWrriten=1 when we succeeded
+ *  in writing the header, otherwise it is set to 0.
+ *
+ *  hufMetadata->hType has literals block type info.
+ *      If it is set_basic, all sub-blocks literals section will be Raw_Literals_Block.
+ *      If it is set_rle, all sub-blocks literals section will be RLE_Literals_Block.
+ *      If it is set_compressed, first sub-block's literals section will be Compressed_Literals_Block
+ *      If it is set_compressed, first sub-block's literals section will be Treeless_Literals_Block
+ *      and the following sub-blocks' literals sections will be Treeless_Literals_Block.
+ *  @return : compressed size of literals section of a sub-block
+ *            Or 0 if it unable to compress.
+ *            Or error code */
+static size_t ZSTD_compressSubBlock_literal(const HUF_CElt* hufTable,
+                                    const ZSTD_hufCTablesMetadata_t* hufMetadata,
+                                    const BYTE* literals, size_t litSize,
+                                    void* dst, size_t dstSize,
+                                    const int bmi2, int writeEntropy, int* entropyWritten)
+{
+    size_t const header = writeEntropy ? 200 : 0;
+    size_t const lhSize = 3 + (litSize >= (1 KB - header)) + (litSize >= (16 KB - header));
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstSize;
+    BYTE* op = ostart + lhSize;
+    U32 const singleStream = lhSize == 3;
+    symbolEncodingType_e hType = writeEntropy ? hufMetadata->hType : set_repeat;
+    size_t cLitSize = 0;
+
+    (void)bmi2; /* TODO bmi2... */
+
+    DEBUGLOG(5, "ZSTD_compressSubBlock_literal (litSize=%zu, lhSize=%zu, writeEntropy=%d)", litSize, lhSize, writeEntropy);
+
+    *entropyWritten = 0;
+    if (litSize == 0 || hufMetadata->hType == set_basic) {
+      DEBUGLOG(5, "ZSTD_compressSubBlock_literal using raw literal");
+      return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
+    } else if (hufMetadata->hType == set_rle) {
+      DEBUGLOG(5, "ZSTD_compressSubBlock_literal using rle literal");
+      return ZSTD_compressRleLiteralsBlock(dst, dstSize, literals, litSize);
+    }
+
+    assert(litSize > 0);
+    assert(hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat);
+
+    if (writeEntropy && hufMetadata->hType == set_compressed) {
+        memcpy(op, hufMetadata->hufDesBuffer, hufMetadata->hufDesSize);
+        op += hufMetadata->hufDesSize;
+        cLitSize += hufMetadata->hufDesSize;
+        DEBUGLOG(5, "ZSTD_compressSubBlock_literal (hSize=%zu)", hufMetadata->hufDesSize);
+    }
+
+    /* TODO bmi2 */
+    {   const size_t cSize = singleStream ? HUF_compress1X_usingCTable(op, oend-op, literals, litSize, hufTable)
+                                          : HUF_compress4X_usingCTable(op, oend-op, literals, litSize, hufTable);
+        op += cSize;
+        cLitSize += cSize;
+        if (cSize == 0 || ERR_isError(cSize)) {
+            DEBUGLOG(5, "Failed to write entropy tables %s", ZSTD_getErrorName(cSize));
+            return 0;
+        }
+        /* If we expand and we aren't writing a header then emit uncompressed */
+        if (!writeEntropy && cLitSize >= litSize) {
+            DEBUGLOG(5, "ZSTD_compressSubBlock_literal using raw literal because uncompressible");
+            return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
+        }
+        /* If we are writing headers then allow expansion that doesn't change our header size. */
+        if (lhSize < (size_t)(3 + (cLitSize >= 1 KB) + (cLitSize >= 16 KB))) {
+            assert(cLitSize > litSize);
+            DEBUGLOG(5, "Literals expanded beyond allowed header size");
+            return ZSTD_noCompressLiterals(dst, dstSize, literals, litSize);
+        }
+        DEBUGLOG(5, "ZSTD_compressSubBlock_literal (cSize=%zu)", cSize);
+    }
+
+    /* Build header */
+    switch(lhSize)
+    {
+    case 3: /* 2 - 2 - 10 - 10 */
+        {   U32 const lhc = hType + ((!singleStream) << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<14);
+            MEM_writeLE24(ostart, lhc);
+            break;
+        }
+    case 4: /* 2 - 2 - 14 - 14 */
+        {   U32 const lhc = hType + (2 << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<18);
+            MEM_writeLE32(ostart, lhc);
+            break;
+        }
+    case 5: /* 2 - 2 - 18 - 18 */
+        {   U32 const lhc = hType + (3 << 2) + ((U32)litSize<<4) + ((U32)cLitSize<<22);
+            MEM_writeLE32(ostart, lhc);
+            ostart[4] = (BYTE)(cLitSize >> 10);
+            break;
+        }
+    default:  /* not possible : lhSize is {3,4,5} */
+        assert(0);
+    }
+    *entropyWritten = 1;
+    DEBUGLOG(5, "Compressed literals: %u -> %u", (U32)litSize, (U32)(op-ostart));
+    return op-ostart;
+}
+
+static size_t ZSTD_seqDecompressedSize(seqStore_t const* seqStore, const seqDef* sequences, size_t nbSeq, size_t litSize, int lastSequence) {
+    const seqDef* const sstart = sequences;
+    const seqDef* const send = sequences + nbSeq;
+    const seqDef* sp = sstart;
+    size_t matchLengthSum = 0;
+    size_t litLengthSum = 0;
+    while (send-sp > 0) {
+        ZSTD_sequenceLength const seqLen = ZSTD_getSequenceLength(seqStore, sp);
+        litLengthSum += seqLen.litLength;
+        matchLengthSum += seqLen.matchLength;
+        sp++;
+    }
+    assert(litLengthSum <= litSize);
+    if (!lastSequence) {
+        assert(litLengthSum == litSize);
+    }
+    return matchLengthSum + litSize;
+}
+
+/** ZSTD_compressSubBlock_sequences() :
+ *  Compresses sequences section for a sub-block.
+ *  fseMetadata->llType, fseMetadata->ofType, and fseMetadata->mlType have
+ *  symbol compression modes for the super-block.
+ *  The first successfully compressed block will have these in its header.
+ *  We set entropyWritten=1 when we succeed in compressing the sequences.
+ *  The following sub-blocks will always have repeat mode.
+ *  @return : compressed size of sequences section of a sub-block
+ *            Or 0 if it is unable to compress
+ *            Or error code. */
+static size_t ZSTD_compressSubBlock_sequences(const ZSTD_fseCTables_t* fseTables,
+                                              const ZSTD_fseCTablesMetadata_t* fseMetadata,
+                                              const seqDef* sequences, size_t nbSeq,
+                                              const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
+                                              const ZSTD_CCtx_params* cctxParams,
+                                              void* dst, size_t dstCapacity,
+                                              const int bmi2, int writeEntropy, int* entropyWritten)
+{
+    const int longOffsets = cctxParams->cParams.windowLog > STREAM_ACCUMULATOR_MIN;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstCapacity;
+    BYTE* op = ostart;
+    BYTE* seqHead;
+
+    DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (nbSeq=%zu, writeEntropy=%d, longOffsets=%d)", nbSeq, writeEntropy, longOffsets);
+
+    *entropyWritten = 0;
+    /* Sequences Header */
+    RETURN_ERROR_IF((oend-op) < 3 /*max nbSeq Size*/ + 1 /*seqHead*/,
+                    dstSize_tooSmall, "");
+    if (nbSeq < 0x7F)
+        *op++ = (BYTE)nbSeq;
+    else if (nbSeq < LONGNBSEQ)
+        op[0] = (BYTE)((nbSeq>>8) + 0x80), op[1] = (BYTE)nbSeq, op+=2;
+    else
+        op[0]=0xFF, MEM_writeLE16(op+1, (U16)(nbSeq - LONGNBSEQ)), op+=3;
+    if (nbSeq==0) {
+        return op - ostart;
+    }
+
+    /* seqHead : flags for FSE encoding type */
+    seqHead = op++;
+
+    DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (seqHeadSize=%u)", (unsigned)(op-ostart));
+
+    if (writeEntropy) {
+        const U32 LLtype = fseMetadata->llType;
+        const U32 Offtype = fseMetadata->ofType;
+        const U32 MLtype = fseMetadata->mlType;
+        DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (fseTablesSize=%zu)", fseMetadata->fseTablesSize);
+        *seqHead = (BYTE)((LLtype<<6) + (Offtype<<4) + (MLtype<<2));
+        memcpy(op, fseMetadata->fseTablesBuffer, fseMetadata->fseTablesSize);
+        op += fseMetadata->fseTablesSize;
+    } else {
+        const U32 repeat = set_repeat;
+        *seqHead = (BYTE)((repeat<<6) + (repeat<<4) + (repeat<<2));
+    }
+
+    {   size_t const bitstreamSize = ZSTD_encodeSequences(
+                                        op, oend - op,
+                                        fseTables->matchlengthCTable, mlCode,
+                                        fseTables->offcodeCTable, ofCode,
+                                        fseTables->litlengthCTable, llCode,
+                                        sequences, nbSeq,
+                                        longOffsets, bmi2);
+        FORWARD_IF_ERROR(bitstreamSize, "ZSTD_encodeSequences failed");
+        op += bitstreamSize;
+        /* zstd versions <= 1.3.4 mistakenly report corruption when
+         * FSE_readNCount() receives a buffer < 4 bytes.
+         * Fixed by https://github.com/facebook/zstd/pull/1146.
+         * This can happen when the last set_compressed table present is 2
+         * bytes and the bitstream is only one byte.
+         * In this exceedingly rare case, we will simply emit an uncompressed
+         * block, since it isn't worth optimizing.
+         */
+#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+        if (writeEntropy && fseMetadata->lastCountSize && fseMetadata->lastCountSize + bitstreamSize < 4) {
+            /* NCountSize >= 2 && bitstreamSize > 0 ==> lastCountSize == 3 */
+            assert(fseMetadata->lastCountSize + bitstreamSize == 3);
+            DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.3.4 by "
+                        "emitting an uncompressed block.");
+            return 0;
+        }
+#endif
+        DEBUGLOG(5, "ZSTD_compressSubBlock_sequences (bitstreamSize=%zu)", bitstreamSize);
+    }
+
+    /* zstd versions <= 1.4.0 mistakenly report error when
+     * sequences section body size is less than 3 bytes.
+     * Fixed by https://github.com/facebook/zstd/pull/1664.
+     * This can happen when the previous sequences section block is compressed
+     * with rle mode and the current block's sequences section is compressed
+     * with repeat mode where sequences section body size can be 1 byte.
+     */
+#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    if (op-seqHead < 4) {
+        DEBUGLOG(5, "Avoiding bug in zstd decoder in versions <= 1.4.0 by emitting "
+                    "an uncompressed block when sequences are < 4 bytes");
+        return 0;
+    }
+#endif
+
+    *entropyWritten = 1;
+    return op - ostart;
+}
+
+/** ZSTD_compressSubBlock() :
+ *  Compresses a single sub-block.
+ *  @return : compressed size of the sub-block
+ *            Or 0 if it failed to compress. */
+static size_t ZSTD_compressSubBlock(const ZSTD_entropyCTables_t* entropy,
+                                    const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                                    const seqDef* sequences, size_t nbSeq,
+                                    const BYTE* literals, size_t litSize,
+                                    const BYTE* llCode, const BYTE* mlCode, const BYTE* ofCode,
+                                    const ZSTD_CCtx_params* cctxParams,
+                                    void* dst, size_t dstCapacity,
+                                    const int bmi2,
+                                    int writeLitEntropy, int writeSeqEntropy,
+                                    int* litEntropyWritten, int* seqEntropyWritten,
+                                    U32 lastBlock)
+{
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstCapacity;
+    BYTE* op = ostart + ZSTD_blockHeaderSize;
+    DEBUGLOG(5, "ZSTD_compressSubBlock (litSize=%zu, nbSeq=%zu, writeLitEntropy=%d, writeSeqEntropy=%d, lastBlock=%d)",
+                litSize, nbSeq, writeLitEntropy, writeSeqEntropy, lastBlock);
+    {   size_t cLitSize = ZSTD_compressSubBlock_literal((const HUF_CElt*)entropy->huf.CTable,
+                                                        &entropyMetadata->hufMetadata, literals, litSize,
+                                                        op, oend-op, bmi2, writeLitEntropy, litEntropyWritten);
+        FORWARD_IF_ERROR(cLitSize, "ZSTD_compressSubBlock_literal failed");
+        if (cLitSize == 0) return 0;
+        op += cLitSize;
+    }
+    {   size_t cSeqSize = ZSTD_compressSubBlock_sequences(&entropy->fse,
+                                                  &entropyMetadata->fseMetadata,
+                                                  sequences, nbSeq,
+                                                  llCode, mlCode, ofCode,
+                                                  cctxParams,
+                                                  op, oend-op,
+                                                  bmi2, writeSeqEntropy, seqEntropyWritten);
+        FORWARD_IF_ERROR(cSeqSize, "ZSTD_compressSubBlock_sequences failed");
+        if (cSeqSize == 0) return 0;
+        op += cSeqSize;
+    }
+    /* Write block header */
+    {   size_t cSize = (op-ostart)-ZSTD_blockHeaderSize;
+        U32 const cBlockHeader24 = lastBlock + (((U32)bt_compressed)<<1) + (U32)(cSize << 3);
+        MEM_writeLE24(ostart, cBlockHeader24);
+    }
+    return op-ostart;
+}
+
+static size_t ZSTD_estimateSubBlockSize_literal(const BYTE* literals, size_t litSize,
+                                                const ZSTD_hufCTables_t* huf,
+                                                const ZSTD_hufCTablesMetadata_t* hufMetadata,
+                                                void* workspace, size_t wkspSize,
+                                                int writeEntropy)
+{
+    unsigned* const countWksp = (unsigned*)workspace;
+    unsigned maxSymbolValue = 255;
+    size_t literalSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */
+
+    if (hufMetadata->hType == set_basic) return litSize;
+    else if (hufMetadata->hType == set_rle) return 1;
+    else if (hufMetadata->hType == set_compressed || hufMetadata->hType == set_repeat) {
+        size_t const largest = HIST_count_wksp (countWksp, &maxSymbolValue, (const BYTE*)literals, litSize, workspace, wkspSize);
+        if (ZSTD_isError(largest)) return litSize;
+        {   size_t cLitSizeEstimate = HUF_estimateCompressedSize((const HUF_CElt*)huf->CTable, countWksp, maxSymbolValue);
+            if (writeEntropy) cLitSizeEstimate += hufMetadata->hufDesSize;
+            return cLitSizeEstimate + literalSectionHeaderSize;
+    }   }
+    assert(0); /* impossible */
+    return 0;
+}
+
+static size_t ZSTD_estimateSubBlockSize_symbolType(symbolEncodingType_e type,
+                        const BYTE* codeTable, unsigned maxCode,
+                        size_t nbSeq, const FSE_CTable* fseCTable,
+                        const U32* additionalBits,
+                        short const* defaultNorm, U32 defaultNormLog,
+                        void* workspace, size_t wkspSize)
+{
+    unsigned* const countWksp = (unsigned*)workspace;
+    const BYTE* ctp = codeTable;
+    const BYTE* const ctStart = ctp;
+    const BYTE* const ctEnd = ctStart + nbSeq;
+    size_t cSymbolTypeSizeEstimateInBits = 0;
+    unsigned max = maxCode;
+
+    HIST_countFast_wksp(countWksp, &max, codeTable, nbSeq, workspace, wkspSize);  /* can't fail */
+    if (type == set_basic) {
+        cSymbolTypeSizeEstimateInBits = ZSTD_crossEntropyCost(defaultNorm, defaultNormLog, countWksp, max);
+    } else if (type == set_rle) {
+        cSymbolTypeSizeEstimateInBits = 0;
+    } else if (type == set_compressed || type == set_repeat) {
+        cSymbolTypeSizeEstimateInBits = ZSTD_fseBitCost(fseCTable, countWksp, max);
+    }
+    if (ZSTD_isError(cSymbolTypeSizeEstimateInBits)) return nbSeq * 10;
+    while (ctp < ctEnd) {
+        if (additionalBits) cSymbolTypeSizeEstimateInBits += additionalBits[*ctp];
+        else cSymbolTypeSizeEstimateInBits += *ctp; /* for offset, offset code is also the number of additional bits */
+        ctp++;
+    }
+    return cSymbolTypeSizeEstimateInBits / 8;
+}
+
+static size_t ZSTD_estimateSubBlockSize_sequences(const BYTE* ofCodeTable,
+                                                  const BYTE* llCodeTable,
+                                                  const BYTE* mlCodeTable,
+                                                  size_t nbSeq,
+                                                  const ZSTD_fseCTables_t* fseTables,
+                                                  const ZSTD_fseCTablesMetadata_t* fseMetadata,
+                                                  void* workspace, size_t wkspSize,
+                                                  int writeEntropy)
+{
+    size_t sequencesSectionHeaderSize = 3; /* Use hard coded size of 3 bytes */
+    size_t cSeqSizeEstimate = 0;
+    cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->ofType, ofCodeTable, MaxOff,
+                                         nbSeq, fseTables->offcodeCTable, NULL,
+                                         OF_defaultNorm, OF_defaultNormLog,
+                                         workspace, wkspSize);
+    cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->llType, llCodeTable, MaxLL,
+                                         nbSeq, fseTables->litlengthCTable, LL_bits,
+                                         LL_defaultNorm, LL_defaultNormLog,
+                                         workspace, wkspSize);
+    cSeqSizeEstimate += ZSTD_estimateSubBlockSize_symbolType(fseMetadata->mlType, mlCodeTable, MaxML,
+                                         nbSeq, fseTables->matchlengthCTable, ML_bits,
+                                         ML_defaultNorm, ML_defaultNormLog,
+                                         workspace, wkspSize);
+    if (writeEntropy) cSeqSizeEstimate += fseMetadata->fseTablesSize;
+    return cSeqSizeEstimate + sequencesSectionHeaderSize;
+}
+
+static size_t ZSTD_estimateSubBlockSize(const BYTE* literals, size_t litSize,
+                                        const BYTE* ofCodeTable,
+                                        const BYTE* llCodeTable,
+                                        const BYTE* mlCodeTable,
+                                        size_t nbSeq,
+                                        const ZSTD_entropyCTables_t* entropy,
+                                        const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                                        void* workspace, size_t wkspSize,
+                                        int writeLitEntropy, int writeSeqEntropy) {
+    size_t cSizeEstimate = 0;
+    cSizeEstimate += ZSTD_estimateSubBlockSize_literal(literals, litSize,
+                                                         &entropy->huf, &entropyMetadata->hufMetadata,
+                                                         workspace, wkspSize, writeLitEntropy);
+    cSizeEstimate += ZSTD_estimateSubBlockSize_sequences(ofCodeTable, llCodeTable, mlCodeTable,
+                                                         nbSeq, &entropy->fse, &entropyMetadata->fseMetadata,
+                                                         workspace, wkspSize, writeSeqEntropy);
+    return cSizeEstimate + ZSTD_blockHeaderSize;
+}
+
+static int ZSTD_needSequenceEntropyTables(ZSTD_fseCTablesMetadata_t const* fseMetadata)
+{
+    if (fseMetadata->llType == set_compressed || fseMetadata->llType == set_rle)
+        return 1;
+    if (fseMetadata->mlType == set_compressed || fseMetadata->mlType == set_rle)
+        return 1;
+    if (fseMetadata->ofType == set_compressed || fseMetadata->ofType == set_rle)
+        return 1;
+    return 0;
+}
+
+/** ZSTD_compressSubBlock_multi() :
+ *  Breaks super-block into multiple sub-blocks and compresses them.
+ *  Entropy will be written to the first block.
+ *  The following blocks will use repeat mode to compress.
+ *  All sub-blocks are compressed blocks (no raw or rle blocks).
+ *  @return : compressed size of the super block (which is multiple ZSTD blocks)
+ *            Or 0 if it failed to compress. */
+static size_t ZSTD_compressSubBlock_multi(const seqStore_t* seqStorePtr,
+                            const ZSTD_compressedBlockState_t* prevCBlock,
+                            ZSTD_compressedBlockState_t* nextCBlock,
+                            const ZSTD_entropyCTablesMetadata_t* entropyMetadata,
+                            const ZSTD_CCtx_params* cctxParams,
+                                  void* dst, size_t dstCapacity,
+                            const void* src, size_t srcSize,
+                            const int bmi2, U32 lastBlock,
+                            void* workspace, size_t wkspSize)
+{
+    const seqDef* const sstart = seqStorePtr->sequencesStart;
+    const seqDef* const send = seqStorePtr->sequences;
+    const seqDef* sp = sstart;
+    const BYTE* const lstart = seqStorePtr->litStart;
+    const BYTE* const lend = seqStorePtr->lit;
+    const BYTE* lp = lstart;
+    BYTE const* ip = (BYTE const*)src;
+    BYTE const* const iend = ip + srcSize;
+    BYTE* const ostart = (BYTE*)dst;
+    BYTE* const oend = ostart + dstCapacity;
+    BYTE* op = ostart;
+    const BYTE* llCodePtr = seqStorePtr->llCode;
+    const BYTE* mlCodePtr = seqStorePtr->mlCode;
+    const BYTE* ofCodePtr = seqStorePtr->ofCode;
+    size_t targetCBlockSize = cctxParams->targetCBlockSize;
+    size_t litSize, seqCount;
+    int writeLitEntropy = entropyMetadata->hufMetadata.hType == set_compressed;
+    int writeSeqEntropy = 1;
+    int lastSequence = 0;
+
+    DEBUGLOG(5, "ZSTD_compressSubBlock_multi (litSize=%u, nbSeq=%u)",
+                (unsigned)(lend-lp), (unsigned)(send-sstart));
+
+    litSize = 0;
+    seqCount = 0;
+    do {
+        size_t cBlockSizeEstimate = 0;
+        if (sstart == send) {
+            lastSequence = 1;
+        } else {
+            const seqDef* const sequence = sp + seqCount;
+            lastSequence = sequence == send - 1;
+            litSize += ZSTD_getSequenceLength(seqStorePtr, sequence).litLength;
+            seqCount++;
+        }
+        if (lastSequence) {
+            assert(lp <= lend);
+            assert(litSize <= (size_t)(lend - lp));
+            litSize = (size_t)(lend - lp);
+        }
+        /* I think there is an optimization opportunity here.
+         * Calling ZSTD_estimateSubBlockSize for every sequence can be wasteful
+         * since it recalculates estimate from scratch.
+         * For example, it would recount literal distribution and symbol codes everytime.
+         */
+        cBlockSizeEstimate = ZSTD_estimateSubBlockSize(lp, litSize, ofCodePtr, llCodePtr, mlCodePtr, seqCount,
+                                                       &nextCBlock->entropy, entropyMetadata,
+                                                       workspace, wkspSize, writeLitEntropy, writeSeqEntropy);
+        if (cBlockSizeEstimate > targetCBlockSize || lastSequence) {
+            int litEntropyWritten = 0;
+            int seqEntropyWritten = 0;
+            const size_t decompressedSize = ZSTD_seqDecompressedSize(seqStorePtr, sp, seqCount, litSize, lastSequence);
+            const size_t cSize = ZSTD_compressSubBlock(&nextCBlock->entropy, entropyMetadata,
+                                                       sp, seqCount,
+                                                       lp, litSize,
+                                                       llCodePtr, mlCodePtr, ofCodePtr,
+                                                       cctxParams,
+                                                       op, oend-op,
+                                                       bmi2, writeLitEntropy, writeSeqEntropy,
+                                                       &litEntropyWritten, &seqEntropyWritten,
+                                                       lastBlock && lastSequence);
+            FORWARD_IF_ERROR(cSize, "ZSTD_compressSubBlock failed");
+            if (cSize > 0 && cSize < decompressedSize) {
+                DEBUGLOG(5, "Committed the sub-block");
+                assert(ip + decompressedSize <= iend);
+                ip += decompressedSize;
+                sp += seqCount;
+                lp += litSize;
+                op += cSize;
+                llCodePtr += seqCount;
+                mlCodePtr += seqCount;
+                ofCodePtr += seqCount;
+                litSize = 0;
+                seqCount = 0;
+                /* Entropy only needs to be written once */
+                if (litEntropyWritten) {
+                    writeLitEntropy = 0;
+                }
+                if (seqEntropyWritten) {
+                    writeSeqEntropy = 0;
+                }
+            }
+        }
+    } while (!lastSequence);
+    if (writeLitEntropy) {
+        DEBUGLOG(5, "ZSTD_compressSubBlock_multi has literal entropy tables unwritten");
+        memcpy(&nextCBlock->entropy.huf, &prevCBlock->entropy.huf, sizeof(prevCBlock->entropy.huf));
+    }
+    if (writeSeqEntropy && ZSTD_needSequenceEntropyTables(&entropyMetadata->fseMetadata)) {
+        /* If we haven't written our entropy tables, then we've violated our contract and
+         * must emit an uncompressed block.
+         */
+        DEBUGLOG(5, "ZSTD_compressSubBlock_multi has sequence entropy tables unwritten");
+        return 0;
+    }
+    if (ip < iend) {
+        size_t const cSize = ZSTD_noCompressBlock(op, oend - op, ip, iend - ip, lastBlock);
+        DEBUGLOG(5, "ZSTD_compressSubBlock_multi last sub-block uncompressed, %zu bytes", (size_t)(iend - ip));
+        FORWARD_IF_ERROR(cSize, "ZSTD_noCompressBlock failed");
+        assert(cSize != 0);
+        op += cSize;
+        /* We have to regenerate the repcodes because we've skipped some sequences */
+        if (sp < send) {
+            seqDef const* seq;
+            repcodes_t rep;
+            memcpy(&rep, prevCBlock->rep, sizeof(rep));
+            for (seq = sstart; seq < sp; ++seq) {
+                rep = ZSTD_updateRep(rep.rep, seq->offset - 1, ZSTD_getSequenceLength(seqStorePtr, seq).litLength == 0);
+            }
+            memcpy(nextCBlock->rep, &rep, sizeof(rep));
+        }
+    }
+    DEBUGLOG(5, "ZSTD_compressSubBlock_multi compressed");
+    return op-ostart;
+}
+
+size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
+                               void* dst, size_t dstCapacity,
+                               void const* src, size_t srcSize,
+                               unsigned lastBlock) {
+    ZSTD_entropyCTablesMetadata_t entropyMetadata;
+
+    FORWARD_IF_ERROR(ZSTD_buildSuperBlockEntropy(&zc->seqStore,
+          &zc->blockState.prevCBlock->entropy,
+          &zc->blockState.nextCBlock->entropy,
+          &zc->appliedParams,
+          &entropyMetadata,
+          zc->entropyWorkspace, HUF_WORKSPACE_SIZE /* statically allocated in resetCCtx */), "");
+
+    return ZSTD_compressSubBlock_multi(&zc->seqStore,
+            zc->blockState.prevCBlock,
+            zc->blockState.nextCBlock,
+            &entropyMetadata,
+            &zc->appliedParams,
+            dst, dstCapacity,
+            src, srcSize,
+            zc->bmi2, lastBlock,
+            zc->entropyWorkspace, HUF_WORKSPACE_SIZE /* statically allocated in resetCCtx */);
+}
diff --git a/contrib/zstd/zstd_compress_superblock.h b/contrib/zstd/zstd_compress_superblock.h
new file mode 100644
index 000000000..07f4cb1dc
--- /dev/null
+++ b/contrib/zstd/zstd_compress_superblock.h
@@ -0,0 +1,32 @@
+/*
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_COMPRESS_ADVANCED_H
+#define ZSTD_COMPRESS_ADVANCED_H
+
+/*-*************************************
+*  Dependencies
+***************************************/
+
+#include "../zstd.h" /* ZSTD_CCtx */
+
+/*-*************************************
+*  Target Compressed Block Size
+***************************************/
+
+/* ZSTD_compressSuperBlock() :
+ * Used to compress a super block when targetCBlockSize is being used.
+ * The given block will be compressed into multiple sub blocks that are around targetCBlockSize. */
+size_t ZSTD_compressSuperBlock(ZSTD_CCtx* zc,
+                               void* dst, size_t dstCapacity,
+                               void const* src, size_t srcSize,
+                               unsigned lastBlock);
+
+#endif /* ZSTD_COMPRESS_ADVANCED_H */
diff --git a/contrib/zstd/zstd_cwksp.h b/contrib/zstd/zstd_cwksp.h
new file mode 100644
index 000000000..91f812fa3
--- /dev/null
+++ b/contrib/zstd/zstd_cwksp.h
@@ -0,0 +1,525 @@
+/*
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+#ifndef ZSTD_CWKSP_H
+#define ZSTD_CWKSP_H
+
+/*-*************************************
+*  Dependencies
+***************************************/
+#include "zstd_internal.h"
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/*-*************************************
+*  Constants
+***************************************/
+
+/* Since the workspace is effectively its own little malloc implementation /
+ * arena, when we run under ASAN, we should similarly insert redzones between
+ * each internal element of the workspace, so ASAN will catch overruns that
+ * reach outside an object but that stay inside the workspace.
+ *
+ * This defines the size of that redzone.
+ */
+#ifndef ZSTD_CWKSP_ASAN_REDZONE_SIZE
+#define ZSTD_CWKSP_ASAN_REDZONE_SIZE 128
+#endif
+
+/*-*************************************
+*  Structures
+***************************************/
+typedef enum {
+    ZSTD_cwksp_alloc_objects,
+    ZSTD_cwksp_alloc_buffers,
+    ZSTD_cwksp_alloc_aligned
+} ZSTD_cwksp_alloc_phase_e;
+
+/**
+ * Zstd fits all its internal datastructures into a single continuous buffer,
+ * so that it only needs to perform a single OS allocation (or so that a buffer
+ * can be provided to it and it can perform no allocations at all). This buffer
+ * is called the workspace.
+ *
+ * Several optimizations complicate that process of allocating memory ranges
+ * from this workspace for each internal datastructure:
+ *
+ * - These different internal datastructures have different setup requirements:
+ *
+ *   - The static objects need to be cleared once and can then be trivially
+ *     reused for each compression.
+ *
+ *   - Various buffers don't need to be initialized at all--they are always
+ *     written into before they're read.
+ *
+ *   - The matchstate tables have a unique requirement that they don't need
+ *     their memory to be totally cleared, but they do need the memory to have
+ *     some bound, i.e., a guarantee that all values in the memory they've been
+ *     allocated is less than some maximum value (which is the starting value
+ *     for the indices that they will then use for compression). When this
+ *     guarantee is provided to them, they can use the memory without any setup
+ *     work. When it can't, they have to clear the area.
+ *
+ * - These buffers also have different alignment requirements.
+ *
+ * - We would like to reuse the objects in the workspace for multiple
+ *   compressions without having to perform any expensive reallocation or
+ *   reinitialization work.
+ *
+ * - We would like to be able to efficiently reuse the workspace across
+ *   multiple compressions **even when the compression parameters change** and
+ *   we need to resize some of the objects (where possible).
+ *
+ * To attempt to manage this buffer, given these constraints, the ZSTD_cwksp
+ * abstraction was created. It works as follows:
+ *
+ * Workspace Layout:
+ *
+ * [                        ... workspace ...                         ]
+ * [objects][tables ... ->] free space [<- ... aligned][<- ... buffers]
+ *
+ * The various objects that live in the workspace are divided into the
+ * following categories, and are allocated separately:
+ *
+ * - Static objects: this is optionally the enclosing ZSTD_CCtx or ZSTD_CDict,
+ *   so that literally everything fits in a single buffer. Note: if present,
+ *   this must be the first object in the workspace, since ZSTD_free{CCtx,
+ *   CDict}() rely on a pointer comparison to see whether one or two frees are
+ *   required.
+ *
+ * - Fixed size objects: these are fixed-size, fixed-count objects that are
+ *   nonetheless "dynamically" allocated in the workspace so that we can
+ *   control how they're initialized separately from the broader ZSTD_CCtx.
+ *   Examples:
+ *   - Entropy Workspace
+ *   - 2 x ZSTD_compressedBlockState_t
+ *   - CDict dictionary contents
+ *
+ * - Tables: these are any of several different datastructures (hash tables,
+ *   chain tables, binary trees) that all respect a common format: they are
+ *   uint32_t arrays, all of whose values are between 0 and (nextSrc - base).
+ *   Their sizes depend on the cparams.
+ *
+ * - Aligned: these buffers are used for various purposes that require 4 byte
+ *   alignment, but don't require any initialization before they're used.
+ *
+ * - Buffers: these buffers are used for various purposes that don't require
+ *   any alignment or initialization before they're used. This means they can
+ *   be moved around at no cost for a new compression.
+ *
+ * Allocating Memory:
+ *
+ * The various types of objects must be allocated in order, so they can be
+ * correctly packed into the workspace buffer. That order is:
+ *
+ * 1. Objects
+ * 2. Buffers
+ * 3. Aligned
+ * 4. Tables
+ *
+ * Attempts to reserve objects of different types out of order will fail.
+ */
+typedef struct {
+    void* workspace;
+    void* workspaceEnd;
+
+    void* objectEnd;
+    void* tableEnd;
+    void* tableValidEnd;
+    void* allocStart;
+
+    int allocFailed;
+    int workspaceOversizedDuration;
+    ZSTD_cwksp_alloc_phase_e phase;
+} ZSTD_cwksp;
+
+/*-*************************************
+*  Functions
+***************************************/
+
+MEM_STATIC size_t ZSTD_cwksp_available_space(ZSTD_cwksp* ws);
+
+MEM_STATIC void ZSTD_cwksp_assert_internal_consistency(ZSTD_cwksp* ws) {
+    (void)ws;
+    assert(ws->workspace <= ws->objectEnd);
+    assert(ws->objectEnd <= ws->tableEnd);
+    assert(ws->objectEnd <= ws->tableValidEnd);
+    assert(ws->tableEnd <= ws->allocStart);
+    assert(ws->tableValidEnd <= ws->allocStart);
+    assert(ws->allocStart <= ws->workspaceEnd);
+}
+
+/**
+ * Align must be a power of 2.
+ */
+MEM_STATIC size_t ZSTD_cwksp_align(size_t size, size_t const align) {
+    size_t const mask = align - 1;
+    assert((align & mask) == 0);
+    return (size + mask) & ~mask;
+}
+
+/**
+ * Use this to determine how much space in the workspace we will consume to
+ * allocate this object. (Normally it should be exactly the size of the object,
+ * but under special conditions, like ASAN, where we pad each object, it might
+ * be larger.)
+ *
+ * Since tables aren't currently redzoned, you don't need to call through this
+ * to figure out how much space you need for the matchState tables. Everything
+ * else is though.
+ */
+MEM_STATIC size_t ZSTD_cwksp_alloc_size(size_t size) {
+#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    return size + 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE;
+#else
+    return size;
+#endif
+}
+
+MEM_STATIC void ZSTD_cwksp_internal_advance_phase(
+        ZSTD_cwksp* ws, ZSTD_cwksp_alloc_phase_e phase) {
+    assert(phase >= ws->phase);
+    if (phase > ws->phase) {
+        if (ws->phase < ZSTD_cwksp_alloc_buffers &&
+                phase >= ZSTD_cwksp_alloc_buffers) {
+            ws->tableValidEnd = ws->objectEnd;
+        }
+        if (ws->phase < ZSTD_cwksp_alloc_aligned &&
+                phase >= ZSTD_cwksp_alloc_aligned) {
+            /* If unaligned allocations down from a too-large top have left us
+             * unaligned, we need to realign our alloc ptr. Technically, this
+             * can consume space that is unaccounted for in the neededSpace
+             * calculation. However, I believe this can only happen when the
+             * workspace is too large, and specifically when it is too large
+             * by a larger margin than the space that will be consumed. */
+            /* TODO: cleaner, compiler warning friendly way to do this??? */
+            ws->allocStart = (BYTE*)ws->allocStart - ((size_t)ws->allocStart & (sizeof(U32)-1));
+            if (ws->allocStart < ws->tableValidEnd) {
+                ws->tableValidEnd = ws->allocStart;
+            }
+        }
+        ws->phase = phase;
+    }
+}
+
+/**
+ * Returns whether this object/buffer/etc was allocated in this workspace.
+ */
+MEM_STATIC int ZSTD_cwksp_owns_buffer(const ZSTD_cwksp* ws, const void* ptr) {
+    return (ptr != NULL) && (ws->workspace <= ptr) && (ptr <= ws->workspaceEnd);
+}
+
+/**
+ * Internal function. Do not use directly.
+ */
+MEM_STATIC void* ZSTD_cwksp_reserve_internal(
+        ZSTD_cwksp* ws, size_t bytes, ZSTD_cwksp_alloc_phase_e phase) {
+    void* alloc;
+    void* bottom = ws->tableEnd;
+    ZSTD_cwksp_internal_advance_phase(ws, phase);
+    alloc = (BYTE *)ws->allocStart - bytes;
+
+#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    /* over-reserve space */
+    alloc = (BYTE *)alloc - 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE;
+#endif
+
+    DEBUGLOG(5, "cwksp: reserving %p %zd bytes, %zd bytes remaining",
+        alloc, bytes, ZSTD_cwksp_available_space(ws) - bytes);
+    ZSTD_cwksp_assert_internal_consistency(ws);
+    assert(alloc >= bottom);
+    if (alloc < bottom) {
+        DEBUGLOG(4, "cwksp: alloc failed!");
+        ws->allocFailed = 1;
+        return NULL;
+    }
+    if (alloc < ws->tableValidEnd) {
+        ws->tableValidEnd = alloc;
+    }
+    ws->allocStart = alloc;
+
+#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    /* Move alloc so there's ZSTD_CWKSP_ASAN_REDZONE_SIZE unused space on
+     * either size. */
+    alloc = (BYTE *)alloc + ZSTD_CWKSP_ASAN_REDZONE_SIZE;
+    __asan_unpoison_memory_region(alloc, bytes);
+#endif
+
+    return alloc;
+}
+
+/**
+ * Reserves and returns unaligned memory.
+ */
+MEM_STATIC BYTE* ZSTD_cwksp_reserve_buffer(ZSTD_cwksp* ws, size_t bytes) {
+    return (BYTE*)ZSTD_cwksp_reserve_internal(ws, bytes, ZSTD_cwksp_alloc_buffers);
+}
+
+/**
+ * Reserves and returns memory sized on and aligned on sizeof(unsigned).
+ */
+MEM_STATIC void* ZSTD_cwksp_reserve_aligned(ZSTD_cwksp* ws, size_t bytes) {
+    assert((bytes & (sizeof(U32)-1)) == 0);
+    return ZSTD_cwksp_reserve_internal(ws, ZSTD_cwksp_align(bytes, sizeof(U32)), ZSTD_cwksp_alloc_aligned);
+}
+
+/**
+ * Aligned on sizeof(unsigned). These buffers have the special property that
+ * their values remain constrained, allowing us to re-use them without
+ * memset()-ing them.
+ */
+MEM_STATIC void* ZSTD_cwksp_reserve_table(ZSTD_cwksp* ws, size_t bytes) {
+    const ZSTD_cwksp_alloc_phase_e phase = ZSTD_cwksp_alloc_aligned;
+    void* alloc = ws->tableEnd;
+    void* end = (BYTE *)alloc + bytes;
+    void* top = ws->allocStart;
+
+    DEBUGLOG(5, "cwksp: reserving %p table %zd bytes, %zd bytes remaining",
+        alloc, bytes, ZSTD_cwksp_available_space(ws) - bytes);
+    assert((bytes & (sizeof(U32)-1)) == 0);
+    ZSTD_cwksp_internal_advance_phase(ws, phase);
+    ZSTD_cwksp_assert_internal_consistency(ws);
+    assert(end <= top);
+    if (end > top) {
+        DEBUGLOG(4, "cwksp: table alloc failed!");
+        ws->allocFailed = 1;
+        return NULL;
+    }
+    ws->tableEnd = end;
+
+#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    __asan_unpoison_memory_region(alloc, bytes);
+#endif
+
+    return alloc;
+}
+
+/**
+ * Aligned on sizeof(void*).
+ */
+MEM_STATIC void* ZSTD_cwksp_reserve_object(ZSTD_cwksp* ws, size_t bytes) {
+    size_t roundedBytes = ZSTD_cwksp_align(bytes, sizeof(void*));
+    void* alloc = ws->objectEnd;
+    void* end = (BYTE*)alloc + roundedBytes;
+
+#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    /* over-reserve space */
+    end = (BYTE *)end + 2 * ZSTD_CWKSP_ASAN_REDZONE_SIZE;
+#endif
+
+    DEBUGLOG(5,
+        "cwksp: reserving %p object %zd bytes (rounded to %zd), %zd bytes remaining",
+        alloc, bytes, roundedBytes, ZSTD_cwksp_available_space(ws) - roundedBytes);
+    assert(((size_t)alloc & (sizeof(void*)-1)) == 0);
+    assert((bytes & (sizeof(void*)-1)) == 0);
+    ZSTD_cwksp_assert_internal_consistency(ws);
+    /* we must be in the first phase, no advance is possible */
+    if (ws->phase != ZSTD_cwksp_alloc_objects || end > ws->workspaceEnd) {
+        DEBUGLOG(4, "cwksp: object alloc failed!");
+        ws->allocFailed = 1;
+        return NULL;
+    }
+    ws->objectEnd = end;
+    ws->tableEnd = end;
+    ws->tableValidEnd = end;
+
+#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    /* Move alloc so there's ZSTD_CWKSP_ASAN_REDZONE_SIZE unused space on
+     * either size. */
+    alloc = (BYTE *)alloc + ZSTD_CWKSP_ASAN_REDZONE_SIZE;
+    __asan_unpoison_memory_region(alloc, bytes);
+#endif
+
+    return alloc;
+}
+
+MEM_STATIC void ZSTD_cwksp_mark_tables_dirty(ZSTD_cwksp* ws) {
+    DEBUGLOG(4, "cwksp: ZSTD_cwksp_mark_tables_dirty");
+
+#if defined (MEMORY_SANITIZER) && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
+    /* To validate that the table re-use logic is sound, and that we don't
+     * access table space that we haven't cleaned, we re-"poison" the table
+     * space every time we mark it dirty. */
+    {
+        size_t size = (BYTE*)ws->tableValidEnd - (BYTE*)ws->objectEnd;
+        assert(__msan_test_shadow(ws->objectEnd, size) == -1);
+        __msan_poison(ws->objectEnd, size);
+    }
+#endif
+
+    assert(ws->tableValidEnd >= ws->objectEnd);
+    assert(ws->tableValidEnd <= ws->allocStart);
+    ws->tableValidEnd = ws->objectEnd;
+    ZSTD_cwksp_assert_internal_consistency(ws);
+}
+
+MEM_STATIC void ZSTD_cwksp_mark_tables_clean(ZSTD_cwksp* ws) {
+    DEBUGLOG(4, "cwksp: ZSTD_cwksp_mark_tables_clean");
+    assert(ws->tableValidEnd >= ws->objectEnd);
+    assert(ws->tableValidEnd <= ws->allocStart);
+    if (ws->tableValidEnd < ws->tableEnd) {
+        ws->tableValidEnd = ws->tableEnd;
+    }
+    ZSTD_cwksp_assert_internal_consistency(ws);
+}
+
+/**
+ * Zero the part of the allocated tables not already marked clean.
+ */
+MEM_STATIC void ZSTD_cwksp_clean_tables(ZSTD_cwksp* ws) {
+    DEBUGLOG(4, "cwksp: ZSTD_cwksp_clean_tables");
+    assert(ws->tableValidEnd >= ws->objectEnd);
+    assert(ws->tableValidEnd <= ws->allocStart);
+    if (ws->tableValidEnd < ws->tableEnd) {
+        memset(ws->tableValidEnd, 0, (BYTE*)ws->tableEnd - (BYTE*)ws->tableValidEnd);
+    }
+    ZSTD_cwksp_mark_tables_clean(ws);
+}
+
+/**
+ * Invalidates table allocations.
+ * All other allocations remain valid.
+ */
+MEM_STATIC void ZSTD_cwksp_clear_tables(ZSTD_cwksp* ws) {
+    DEBUGLOG(4, "cwksp: clearing tables!");
+
+#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    {
+        size_t size = (BYTE*)ws->tableValidEnd - (BYTE*)ws->objectEnd;
+        __asan_poison_memory_region(ws->objectEnd, size);
+    }
+#endif
+
+    ws->tableEnd = ws->objectEnd;
+    ZSTD_cwksp_assert_internal_consistency(ws);
+}
+
+/**
+ * Invalidates all buffer, aligned, and table allocations.
+ * Object allocations remain valid.
+ */
+MEM_STATIC void ZSTD_cwksp_clear(ZSTD_cwksp* ws) {
+    DEBUGLOG(4, "cwksp: clearing!");
+
+#if defined (MEMORY_SANITIZER) && !defined (ZSTD_MSAN_DONT_POISON_WORKSPACE)
+    /* To validate that the context re-use logic is sound, and that we don't
+     * access stuff that this compression hasn't initialized, we re-"poison"
+     * the workspace (or at least the non-static, non-table parts of it)
+     * every time we start a new compression. */
+    {
+        size_t size = (BYTE*)ws->workspaceEnd - (BYTE*)ws->tableValidEnd;
+        __msan_poison(ws->tableValidEnd, size);
+    }
+#endif
+
+#if defined (ADDRESS_SANITIZER) && !defined (ZSTD_ASAN_DONT_POISON_WORKSPACE)
+    {
+        size_t size = (BYTE*)ws->workspaceEnd - (BYTE*)ws->objectEnd;
+        __asan_poison_memory_region(ws->objectEnd, size);
+    }
+#endif
+
+    ws->tableEnd = ws->objectEnd;
+    ws->allocStart = ws->workspaceEnd;
+    ws->allocFailed = 0;
+    if (ws->phase > ZSTD_cwksp_alloc_buffers) {
+        ws->phase = ZSTD_cwksp_alloc_buffers;
+    }
+    ZSTD_cwksp_assert_internal_consistency(ws);
+}
+
+/**
+ * The provided workspace takes ownership of the buffer [start, start+size).
+ * Any existing values in the workspace are ignored (the previously managed
+ * buffer, if present, must be separately freed).
+ */
+MEM_STATIC void ZSTD_cwksp_init(ZSTD_cwksp* ws, void* start, size_t size) {
+    DEBUGLOG(4, "cwksp: init'ing workspace with %zd bytes", size);
+    assert(((size_t)start & (sizeof(void*)-1)) == 0); /* ensure correct alignment */
+    ws->workspace = start;
+    ws->workspaceEnd = (BYTE*)start + size;
+    ws->objectEnd = ws->workspace;
+    ws->tableValidEnd = ws->objectEnd;
+    ws->phase = ZSTD_cwksp_alloc_objects;
+    ZSTD_cwksp_clear(ws);
+    ws->workspaceOversizedDuration = 0;
+    ZSTD_cwksp_assert_internal_consistency(ws);
+}
+
+MEM_STATIC size_t ZSTD_cwksp_create(ZSTD_cwksp* ws, size_t size, ZSTD_customMem customMem) {
+    void* workspace = ZSTD_malloc(size, customMem);
+    DEBUGLOG(4, "cwksp: creating new workspace with %zd bytes", size);
+    RETURN_ERROR_IF(workspace == NULL, memory_allocation, "NULL pointer!");
+    ZSTD_cwksp_init(ws, workspace, size);
+    return 0;
+}
+
+MEM_STATIC void ZSTD_cwksp_free(ZSTD_cwksp* ws, ZSTD_customMem customMem) {
+    void *ptr = ws->workspace;
+    DEBUGLOG(4, "cwksp: freeing workspace");
+    memset(ws, 0, sizeof(ZSTD_cwksp));
+    ZSTD_free(ptr, customMem);
+}
+
+/**
+ * Moves the management of a workspace from one cwksp to another. The src cwksp
+ * is left in an invalid state (src must be re-init()'ed before its used again).
+ */
+MEM_STATIC void ZSTD_cwksp_move(ZSTD_cwksp* dst, ZSTD_cwksp* src) {
+    *dst = *src;
+    memset(src, 0, sizeof(ZSTD_cwksp));
+}
+
+MEM_STATIC size_t ZSTD_cwksp_sizeof(const ZSTD_cwksp* ws) {
+    return (size_t)((BYTE*)ws->workspaceEnd - (BYTE*)ws->workspace);
+}
+
+MEM_STATIC int ZSTD_cwksp_reserve_failed(const ZSTD_cwksp* ws) {
+    return ws->allocFailed;
+}
+
+/*-*************************************
+*  Functions Checking Free Space
+***************************************/
+
+MEM_STATIC size_t ZSTD_cwksp_available_space(ZSTD_cwksp* ws) {
+    return (size_t)((BYTE*)ws->allocStart - (BYTE*)ws->tableEnd);
+}
+
+MEM_STATIC int ZSTD_cwksp_check_available(ZSTD_cwksp* ws, size_t additionalNeededSpace) {
+    return ZSTD_cwksp_available_space(ws) >= additionalNeededSpace;
+}
+
+MEM_STATIC int ZSTD_cwksp_check_too_large(ZSTD_cwksp* ws, size_t additionalNeededSpace) {
+    return ZSTD_cwksp_check_available(
+        ws, additionalNeededSpace * ZSTD_WORKSPACETOOLARGE_FACTOR);
+}
+
+MEM_STATIC int ZSTD_cwksp_check_wasteful(ZSTD_cwksp* ws, size_t additionalNeededSpace) {
+    return ZSTD_cwksp_check_too_large(ws, additionalNeededSpace)
+        && ws->workspaceOversizedDuration > ZSTD_WORKSPACETOOLARGE_MAXDURATION;
+}
+
+MEM_STATIC void ZSTD_cwksp_bump_oversized_duration(
+        ZSTD_cwksp* ws, size_t additionalNeededSpace) {
+    if (ZSTD_cwksp_check_too_large(ws, additionalNeededSpace)) {
+        ws->workspaceOversizedDuration++;
+    } else {
+        ws->workspaceOversizedDuration = 0;
+    }
+}
+
+#if defined (__cplusplus)
+}
+#endif
+
+#endif /* ZSTD_CWKSP_H */
diff --git a/contrib/zstd/zstd_ddict.c b/contrib/zstd/zstd_ddict.c
new file mode 100644
index 000000000..47e985fca
--- /dev/null
+++ b/contrib/zstd/zstd_ddict.c
@@ -0,0 +1,244 @@
+/*
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* zstd_ddict.c :
+ * concentrates all logic that needs to know the internals of ZSTD_DDict object */
+
+/*-*******************************************************
+*  Dependencies
+*********************************************************/
+#include <string.h>      /* memcpy, memmove, memset */
+#include "cpu.h"         /* bmi2 */
+#include "mem.h"         /* low level memory routines */
+#define FSE_STATIC_LINKING_ONLY
+#include "fse.h"
+#define HUF_STATIC_LINKING_ONLY
+#include "huf.h"
+#include "zstd_decompress_internal.h"
+#include "zstd_ddict.h"
+
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+#  include "../legacy/zstd_legacy.h"
+#endif
+
+
+
+/*-*******************************************************
+*  Types
+*********************************************************/
+struct ZSTD_DDict_s {
+    void* dictBuffer;
+    const void* dictContent;
+    size_t dictSize;
+    ZSTD_entropyDTables_t entropy;
+    U32 dictID;
+    U32 entropyPresent;
+    ZSTD_customMem cMem;
+};  /* typedef'd to ZSTD_DDict within "zstd.h" */
+
+const void* ZSTD_DDict_dictContent(const ZSTD_DDict* ddict)
+{
+    assert(ddict != NULL);
+    return ddict->dictContent;
+}
+
+size_t ZSTD_DDict_dictSize(const ZSTD_DDict* ddict)
+{
+    assert(ddict != NULL);
+    return ddict->dictSize;
+}
+
+void ZSTD_copyDDictParameters(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
+{
+    DEBUGLOG(4, "ZSTD_copyDDictParameters");
+    assert(dctx != NULL);
+    assert(ddict != NULL);
+    dctx->dictID = ddict->dictID;
+    dctx->prefixStart = ddict->dictContent;
+    dctx->virtualStart = ddict->dictContent;
+    dctx->dictEnd = (const BYTE*)ddict->dictContent + ddict->dictSize;
+    dctx->previousDstEnd = dctx->dictEnd;
+#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    dctx->dictContentBeginForFuzzing = dctx->prefixStart;
+    dctx->dictContentEndForFuzzing = dctx->previousDstEnd;
+#endif
+    if (ddict->entropyPresent) {
+        dctx->litEntropy = 1;
+        dctx->fseEntropy = 1;
+        dctx->LLTptr = ddict->entropy.LLTable;
+        dctx->MLTptr = ddict->entropy.MLTable;
+        dctx->OFTptr = ddict->entropy.OFTable;
+        dctx->HUFptr = ddict->entropy.hufTable;
+        dctx->entropy.rep[0] = ddict->entropy.rep[0];
+        dctx->entropy.rep[1] = ddict->entropy.rep[1];
+        dctx->entropy.rep[2] = ddict->entropy.rep[2];
+    } else {
+        dctx->litEntropy = 0;
+        dctx->fseEntropy = 0;
+    }
+}
+
+
+static size_t
+ZSTD_loadEntropy_intoDDict(ZSTD_DDict* ddict,
+                           ZSTD_dictContentType_e dictContentType)
+{
+    ddict->dictID = 0;
+    ddict->entropyPresent = 0;
+    if (dictContentType == ZSTD_dct_rawContent) return 0;
+
+    if (ddict->dictSize < 8) {
+        if (dictContentType == ZSTD_dct_fullDict)
+            return ERROR(dictionary_corrupted);   /* only accept specified dictionaries */
+        return 0;   /* pure content mode */
+    }
+    {   U32 const magic = MEM_readLE32(ddict->dictContent);
+        if (magic != ZSTD_MAGIC_DICTIONARY) {
+            if (dictContentType == ZSTD_dct_fullDict)
+                return ERROR(dictionary_corrupted);   /* only accept specified dictionaries */
+            return 0;   /* pure content mode */
+        }
+    }
+    ddict->dictID = MEM_readLE32((const char*)ddict->dictContent + ZSTD_FRAMEIDSIZE);
+
+    /* load entropy tables */
+    RETURN_ERROR_IF(ZSTD_isError(ZSTD_loadDEntropy(
+            &ddict->entropy, ddict->dictContent, ddict->dictSize)),
+        dictionary_corrupted, "");
+    ddict->entropyPresent = 1;
+    return 0;
+}
+
+
+static size_t ZSTD_initDDict_internal(ZSTD_DDict* ddict,
+                                      const void* dict, size_t dictSize,
+                                      ZSTD_dictLoadMethod_e dictLoadMethod,
+                                      ZSTD_dictContentType_e dictContentType)
+{
+    if ((dictLoadMethod == ZSTD_dlm_byRef) || (!dict) || (!dictSize)) {
+        ddict->dictBuffer = NULL;
+        ddict->dictContent = dict;
+        if (!dict) dictSize = 0;
+    } else {
+        void* const internalBuffer = ZSTD_malloc(dictSize, ddict->cMem);
+        ddict->dictBuffer = internalBuffer;
+        ddict->dictContent = internalBuffer;
+        if (!internalBuffer) return ERROR(memory_allocation);
+        memcpy(internalBuffer, dict, dictSize);
+    }
+    ddict->dictSize = dictSize;
+    ddict->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
+
+    /* parse dictionary content */
+    FORWARD_IF_ERROR( ZSTD_loadEntropy_intoDDict(ddict, dictContentType) , "");
+
+    return 0;
+}
+
+ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize,
+                                      ZSTD_dictLoadMethod_e dictLoadMethod,
+                                      ZSTD_dictContentType_e dictContentType,
+                                      ZSTD_customMem customMem)
+{
+    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
+
+    {   ZSTD_DDict* const ddict = (ZSTD_DDict*) ZSTD_malloc(sizeof(ZSTD_DDict), customMem);
+        if (ddict == NULL) return NULL;
+        ddict->cMem = customMem;
+        {   size_t const initResult = ZSTD_initDDict_internal(ddict,
+                                            dict, dictSize,
+                                            dictLoadMethod, dictContentType);
+            if (ZSTD_isError(initResult)) {
+                ZSTD_freeDDict(ddict);
+                return NULL;
+        }   }
+        return ddict;
+    }
+}
+
+/*! ZSTD_createDDict() :
+*   Create a digested dictionary, to start decompression without startup delay.
+*   `dict` content is copied inside DDict.
+*   Consequently, `dict` can be released after `ZSTD_DDict` creation */
+ZSTD_DDict* ZSTD_createDDict(const void* dict, size_t dictSize)
+{
+    ZSTD_customMem const allocator = { NULL, NULL, NULL };
+    return ZSTD_createDDict_advanced(dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dct_auto, allocator);
+}
+
+/*! ZSTD_createDDict_byReference() :
+ *  Create a digested dictionary, to start decompression without startup delay.
+ *  Dictionary content is simply referenced, it will be accessed during decompression.
+ *  Warning : dictBuffer must outlive DDict (DDict must be freed before dictBuffer) */
+ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize)
+{
+    ZSTD_customMem const allocator = { NULL, NULL, NULL };
+    return ZSTD_createDDict_advanced(dictBuffer, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto, allocator);
+}
+
+
+const ZSTD_DDict* ZSTD_initStaticDDict(
+                                void* sBuffer, size_t sBufferSize,
+                                const void* dict, size_t dictSize,
+                                ZSTD_dictLoadMethod_e dictLoadMethod,
+                                ZSTD_dictContentType_e dictContentType)
+{
+    size_t const neededSpace = sizeof(ZSTD_DDict)
+                             + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
+    ZSTD_DDict* const ddict = (ZSTD_DDict*)sBuffer;
+    assert(sBuffer != NULL);
+    assert(dict != NULL);
+    if ((size_t)sBuffer & 7) return NULL;   /* 8-aligned */
+    if (sBufferSize < neededSpace) return NULL;
+    if (dictLoadMethod == ZSTD_dlm_byCopy) {
+        memcpy(ddict+1, dict, dictSize);  /* local copy */
+        dict = ddict+1;
+    }
+    if (ZSTD_isError( ZSTD_initDDict_internal(ddict,
+                                              dict, dictSize,
+                                              ZSTD_dlm_byRef, dictContentType) ))
+        return NULL;
+    return ddict;
+}
+
+
+size_t ZSTD_freeDDict(ZSTD_DDict* ddict)
+{
+    if (ddict==NULL) return 0;   /* support free on NULL */
+    {   ZSTD_customMem const cMem = ddict->cMem;
+        ZSTD_free(ddict->dictBuffer, cMem);
+        ZSTD_free(ddict, cMem);
+        return 0;
+    }
+}
+
+/*! ZSTD_estimateDDictSize() :
+ *  Estimate amount of memory that will be needed to create a dictionary for decompression.
+ *  Note : dictionary created by reference using ZSTD_dlm_byRef are smaller */
+size_t ZSTD_estimateDDictSize(size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod)
+{
+    return sizeof(ZSTD_DDict) + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
+}
+
+size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict)
+{
+    if (ddict==NULL) return 0;   /* support sizeof on NULL */
+    return sizeof(*ddict) + (ddict->dictBuffer ? ddict->dictSize : 0) ;
+}
+
+/*! ZSTD_getDictID_fromDDict() :
+ *  Provides the dictID of the dictionary loaded into `ddict`.
+ *  If @return == 0, the dictionary is not conformant to Zstandard specification, or empty.
+ *  Non-conformant dictionaries can still be loaded, but as content-only dictionaries. */
+unsigned ZSTD_getDictID_fromDDict(const ZSTD_DDict* ddict)
+{
+    if (ddict==NULL) return 0;
+    return ZSTD_getDictID_fromDict(ddict->dictContent, ddict->dictSize);
+}
diff --git a/contrib/zstd/zstd_ddict.h b/contrib/zstd/zstd_ddict.h
new file mode 100644
index 000000000..af307efd3
--- /dev/null
+++ b/contrib/zstd/zstd_ddict.h
@@ -0,0 +1,44 @@
+/*
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+#ifndef ZSTD_DDICT_H
+#define ZSTD_DDICT_H
+
+/*-*******************************************************
+ *  Dependencies
+ *********************************************************/
+#include <stddef.h>   /* size_t */
+#include "../zstd.h"     /* ZSTD_DDict, and several public functions */
+
+
+/*-*******************************************************
+ *  Interface
+ *********************************************************/
+
+/* note: several prototypes are already published in `zstd.h` :
+ * ZSTD_createDDict()
+ * ZSTD_createDDict_byReference()
+ * ZSTD_createDDict_advanced()
+ * ZSTD_freeDDict()
+ * ZSTD_initStaticDDict()
+ * ZSTD_sizeof_DDict()
+ * ZSTD_estimateDDictSize()
+ * ZSTD_getDictID_fromDict()
+ */
+
+const void* ZSTD_DDict_dictContent(const ZSTD_DDict* ddict);
+size_t ZSTD_DDict_dictSize(const ZSTD_DDict* ddict);
+
+void ZSTD_copyDDictParameters(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict);
+
+
+
+#endif /* ZSTD_DDICT_H */
diff --git a/contrib/zstd/zstd_decompress.c b/contrib/zstd/zstd_decompress.c
index 91518990e..e466647f0 100644
--- a/contrib/zstd/zstd_decompress.c
+++ b/contrib/zstd/zstd_decompress.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -14,8 +14,9 @@
 *****************************************************************/
 /*!
  * HEAPMODE :
- * Select how default decompression function ZSTD_decompress() will allocate memory,
- * in memory stack (0), or in memory heap (1, requires malloc())
+ * Select how default decompression function ZSTD_decompress() allocates its context,
+ * on stack (0), or into heap (1, default; requires malloc()).
+ * Note that functions with explicit context such as ZSTD_decompressDCtx() are unaffected.
  */
 #ifndef ZSTD_HEAPMODE
 #  define ZSTD_HEAPMODE 1
@@ -23,19 +24,31 @@
 
 /*!
 *  LEGACY_SUPPORT :
-*  if set to 1, ZSTD_decompress() can decode older formats (v0.1+)
+*  if set to 1+, ZSTD_decompress() can decode older formats (v0.1+)
 */
 #ifndef ZSTD_LEGACY_SUPPORT
 #  define ZSTD_LEGACY_SUPPORT 0
 #endif
 
 /*!
-*  MAXWINDOWSIZE_DEFAULT :
-*  maximum window size accepted by DStream, by default.
-*  Frames requiring more memory will be rejected.
-*/
+ *  MAXWINDOWSIZE_DEFAULT :
+ *  maximum window size accepted by DStream __by default__.
+ *  Frames requiring more memory will be rejected.
+ *  It's possible to set a different limit using ZSTD_DCtx_setMaxWindowSize().
+ */
 #ifndef ZSTD_MAXWINDOWSIZE_DEFAULT
-#  define ZSTD_MAXWINDOWSIZE_DEFAULT ((1 << ZSTD_WINDOWLOG_MAX) + 1)   /* defined within zstd.h */
+#  define ZSTD_MAXWINDOWSIZE_DEFAULT (((U32)1 << ZSTD_WINDOWLOG_LIMIT_DEFAULT) + 1)
+#endif
+
+/*!
+ *  NO_FORWARD_PROGRESS_MAX :
+ *  maximum allowed nb of calls to ZSTD_decompressStream()
+ *  without any forward progress
+ *  (defined as: no byte read from input, and no byte flushed to output)
+ *  before triggering an error.
+ */
+#ifndef ZSTD_NO_FORWARD_PROGRESS_MAX
+#  define ZSTD_NO_FORWARD_PROGRESS_MAX 16
 #endif
 
 
@@ -43,102 +56,25 @@
 *  Dependencies
 *********************************************************/
 #include <string.h>      /* memcpy, memmove, memset */
+#include "cpu.h"         /* bmi2 */
 #include "mem.h"         /* low level memory routines */
 #define FSE_STATIC_LINKING_ONLY
 #include "fse.h"
 #define HUF_STATIC_LINKING_ONLY
 #include "huf.h"
-#include "zstd_internal.h"
+#include "zstd_internal.h"  /* blockProperties_t */
+#include "zstd_decompress_internal.h"   /* ZSTD_DCtx */
+#include "zstd_ddict.h"  /* ZSTD_DDictDictContent */
+#include "zstd_decompress_block.h"   /* ZSTD_decompressBlock_internal */
 
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
-#  include "zstd_legacy.h"
+#  include "../legacy/zstd_legacy.h"
 #endif
 
 
-/*-*************************************
-*  Errors
-***************************************/
-#define ZSTD_isError ERR_isError   /* for inlining */
-#define FSE_isError  ERR_isError
-#define HUF_isError  ERR_isError
-
-
-/*_*******************************************************
-*  Memory operations
-**********************************************************/
-static void ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
-
-
 /*-*************************************************************
 *   Context management
 ***************************************************************/
-typedef enum { ZSTDds_getFrameHeaderSize, ZSTDds_decodeFrameHeader,
-               ZSTDds_decodeBlockHeader, ZSTDds_decompressBlock,
-               ZSTDds_decompressLastBlock, ZSTDds_checkChecksum,
-               ZSTDds_decodeSkippableHeader, ZSTDds_skipFrame } ZSTD_dStage;
-
-typedef enum { zdss_init=0, zdss_loadHeader,
-               zdss_read, zdss_load, zdss_flush } ZSTD_dStreamStage;
-
-typedef struct {
-    FSE_DTable LLTable[FSE_DTABLE_SIZE_U32(LLFSELog)];
-    FSE_DTable OFTable[FSE_DTABLE_SIZE_U32(OffFSELog)];
-    FSE_DTable MLTable[FSE_DTABLE_SIZE_U32(MLFSELog)];
-    HUF_DTable hufTable[HUF_DTABLE_SIZE(HufLog)];  /* can accommodate HUF_decompress4X */
-    U32 workspace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];
-    U32 rep[ZSTD_REP_NUM];
-} ZSTD_entropyDTables_t;
-
-struct ZSTD_DCtx_s
-{
-    const FSE_DTable* LLTptr;
-    const FSE_DTable* MLTptr;
-    const FSE_DTable* OFTptr;
-    const HUF_DTable* HUFptr;
-    ZSTD_entropyDTables_t entropy;
-    const void* previousDstEnd;   /* detect continuity */
-    const void* base;             /* start of current segment */
-    const void* vBase;            /* virtual start of previous segment if it was just before current one */
-    const void* dictEnd;          /* end of previous segment */
-    size_t expected;
-    ZSTD_frameHeader fParams;
-    U64 decodedSize;
-    blockType_e bType;            /* used in ZSTD_decompressContinue(), store blockType between block header decoding and block decompression stages */
-    ZSTD_dStage stage;
-    U32 litEntropy;
-    U32 fseEntropy;
-    XXH64_state_t xxhState;
-    size_t headerSize;
-    U32 dictID;
-    const BYTE* litPtr;
-    ZSTD_customMem customMem;
-    size_t litSize;
-    size_t rleSize;
-    size_t staticSize;
-
-    /* streaming */
-    ZSTD_DDict* ddictLocal;
-    const ZSTD_DDict* ddict;
-    ZSTD_dStreamStage streamStage;
-    char*  inBuff;
-    size_t inBuffSize;
-    size_t inPos;
-    size_t maxWindowSize;
-    char*  outBuff;
-    size_t outBuffSize;
-    size_t outStart;
-    size_t outEnd;
-    size_t lhSize;
-    void* legacyContext;
-    U32 previousLegacyVersion;
-    U32 legacyVersion;
-    U32 hostageByte;
-
-    /* workspace */
-    BYTE litBuffer[ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH];
-    BYTE headerBuffer[ZSTD_FRAMEHEADERSIZE_MAX];
-};  /* typedef'd to ZSTD_DCtx within "zstd.h" */
-
 size_t ZSTD_sizeof_DCtx (const ZSTD_DCtx* dctx)
 {
     if (dctx==NULL) return 0;   /* support sizeof NULL */
@@ -149,38 +85,51 @@ size_t ZSTD_sizeof_DCtx (const ZSTD_DCtx* dctx)
 
 size_t ZSTD_estimateDCtxSize(void) { return sizeof(ZSTD_DCtx); }
 
-size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
+
+static size_t ZSTD_startingInputLength(ZSTD_format_e format)
 {
-    dctx->expected = ZSTD_frameHeaderSize_prefix;
-    dctx->stage = ZSTDds_getFrameHeaderSize;
-    dctx->decodedSize = 0;
-    dctx->previousDstEnd = NULL;
-    dctx->base = NULL;
-    dctx->vBase = NULL;
-    dctx->dictEnd = NULL;
-    dctx->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
-    dctx->litEntropy = dctx->fseEntropy = 0;
-    dctx->dictID = 0;
-    MEM_STATIC_ASSERT(sizeof(dctx->entropy.rep) == sizeof(repStartValue));
-    memcpy(dctx->entropy.rep, repStartValue, sizeof(repStartValue));  /* initial repcodes */
-    dctx->LLTptr = dctx->entropy.LLTable;
-    dctx->MLTptr = dctx->entropy.MLTable;
-    dctx->OFTptr = dctx->entropy.OFTable;
-    dctx->HUFptr = dctx->entropy.hufTable;
-    return 0;
+    size_t const startingInputLength = ZSTD_FRAMEHEADERSIZE_PREFIX(format);
+    /* only supports formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless */
+    assert( (format == ZSTD_f_zstd1) || (format == ZSTD_f_zstd1_magicless) );
+    return startingInputLength;
 }
 
 static void ZSTD_initDCtx_internal(ZSTD_DCtx* dctx)
 {
-    ZSTD_decompressBegin(dctx);   /* cannot fail */
+    dctx->format = ZSTD_f_zstd1;  /* ZSTD_decompressBegin() invokes ZSTD_startingInputLength() with argument dctx->format */
     dctx->staticSize  = 0;
     dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
     dctx->ddict       = NULL;
     dctx->ddictLocal  = NULL;
+    dctx->dictEnd     = NULL;
+    dctx->ddictIsCold = 0;
+    dctx->dictUses = ZSTD_dont_use;
     dctx->inBuff      = NULL;
     dctx->inBuffSize  = 0;
     dctx->outBuffSize = 0;
     dctx->streamStage = zdss_init;
+    dctx->legacyContext = NULL;
+    dctx->previousLegacyVersion = 0;
+    dctx->noForwardProgress = 0;
+    dctx->oversizedDuration = 0;
+    dctx->bmi2 = ZSTD_cpuid_bmi2(ZSTD_cpuid());
+    dctx->outBufferMode = ZSTD_obm_buffered;
+#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    dctx->dictContentEndForFuzzing = NULL;
+#endif
+}
+
+ZSTD_DCtx* ZSTD_initStaticDCtx(void *workspace, size_t workspaceSize)
+{
+    ZSTD_DCtx* const dctx = (ZSTD_DCtx*) workspace;
+
+    if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
+    if (workspaceSize < sizeof(ZSTD_DCtx)) return NULL;  /* minimum size */
+
+    ZSTD_initDCtx_internal(dctx);
+    dctx->staticSize = workspaceSize;
+    dctx->inBuff = (char*)(dctx+1);
+    return dctx;
 }
 
 ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
@@ -190,38 +139,31 @@ ZSTD_DCtx* ZSTD_createDCtx_advanced(ZSTD_customMem customMem)
     {   ZSTD_DCtx* const dctx = (ZSTD_DCtx*)ZSTD_malloc(sizeof(*dctx), customMem);
         if (!dctx) return NULL;
         dctx->customMem = customMem;
-        dctx->legacyContext = NULL;
-        dctx->previousLegacyVersion = 0;
         ZSTD_initDCtx_internal(dctx);
         return dctx;
     }
 }
 
-ZSTD_DCtx* ZSTD_initStaticDCtx(void *workspace, size_t workspaceSize)
+ZSTD_DCtx* ZSTD_createDCtx(void)
 {
-    ZSTD_DCtx* dctx = (ZSTD_DCtx*) workspace;
-
-    if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
-    if (workspaceSize < sizeof(ZSTD_DCtx)) return NULL;  /* minimum size */
-
-    ZSTD_initDCtx_internal(dctx);
-    dctx->staticSize = workspaceSize;
-    dctx->inBuff = (char*)(dctx+1);
-    return dctx;
+    DEBUGLOG(3, "ZSTD_createDCtx");
+    return ZSTD_createDCtx_advanced(ZSTD_defaultCMem);
 }
 
-ZSTD_DCtx* ZSTD_createDCtx(void)
+static void ZSTD_clearDict(ZSTD_DCtx* dctx)
 {
-    return ZSTD_createDCtx_advanced(ZSTD_defaultCMem);
+    ZSTD_freeDDict(dctx->ddictLocal);
+    dctx->ddictLocal = NULL;
+    dctx->ddict = NULL;
+    dctx->dictUses = ZSTD_dont_use;
 }
 
 size_t ZSTD_freeDCtx(ZSTD_DCtx* dctx)
 {
     if (dctx==NULL) return 0;   /* support free on NULL */
-    if (dctx->staticSize) return ERROR(memory_allocation);   /* not compatible with static DCtx */
+    RETURN_ERROR_IF(dctx->staticSize, memory_allocation, "not compatible with static DCtx");
     {   ZSTD_customMem const cMem = dctx->customMem;
-        ZSTD_freeDDict(dctx->ddictLocal);
-        dctx->ddictLocal = NULL;
+        ZSTD_clearDict(dctx);
         ZSTD_free(dctx->inBuff, cMem);
         dctx->inBuff = NULL;
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
@@ -242,8 +184,8 @@ void ZSTD_copyDCtx(ZSTD_DCtx* dstDCtx, const ZSTD_DCtx* srcDCtx)
 
 
 /*-*************************************************************
-*   Decompression section
-***************************************************************/
+ *   Frame header decoding
+ ***************************************************************/
 
 /*! ZSTD_isFrame() :
  *  Tells if the content of `buffer` starts with a valid Frame Identifier.
@@ -252,10 +194,10 @@ void ZSTD_copyDCtx(ZSTD_DCtx* dstDCtx, const ZSTD_DCtx* srcDCtx)
  *  Note 3 : Skippable Frame Identifiers are considered valid. */
 unsigned ZSTD_isFrame(const void* buffer, size_t size)
 {
-    if (size < 4) return 0;
+    if (size < ZSTD_FRAMEIDSIZE) return 0;
     {   U32 const magic = MEM_readLE32(buffer);
         if (magic == ZSTD_MAGICNUMBER) return 1;
-        if ((magic & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) return 1;
+        if ((magic & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) return 1;
     }
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
     if (ZSTD_isLegacy(buffer, size)) return 1;
@@ -263,54 +205,73 @@ unsigned ZSTD_isFrame(const void* buffer, size_t size)
     return 0;
 }
 
-
-/** ZSTD_frameHeaderSize() :
-*   srcSize must be >= ZSTD_frameHeaderSize_prefix.
-*   @return : size of the Frame Header */
-size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize)
+/** ZSTD_frameHeaderSize_internal() :
+ *  srcSize must be large enough to reach header size fields.
+ *  note : only works for formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless.
+ * @return : size of the Frame Header
+ *           or an error code, which can be tested with ZSTD_isError() */
+static size_t ZSTD_frameHeaderSize_internal(const void* src, size_t srcSize, ZSTD_format_e format)
 {
-    if (srcSize < ZSTD_frameHeaderSize_prefix) return ERROR(srcSize_wrong);
-    {   BYTE const fhd = ((const BYTE*)src)[4];
+    size_t const minInputSize = ZSTD_startingInputLength(format);
+    RETURN_ERROR_IF(srcSize < minInputSize, srcSize_wrong, "");
+
+    {   BYTE const fhd = ((const BYTE*)src)[minInputSize-1];
         U32 const dictID= fhd & 3;
         U32 const singleSegment = (fhd >> 5) & 1;
         U32 const fcsId = fhd >> 6;
-        return ZSTD_frameHeaderSize_prefix + !singleSegment + ZSTD_did_fieldSize[dictID] + ZSTD_fcs_fieldSize[fcsId]
-                + (singleSegment && !fcsId);
+        return minInputSize + !singleSegment
+             + ZSTD_did_fieldSize[dictID] + ZSTD_fcs_fieldSize[fcsId]
+             + (singleSegment && !fcsId);
     }
 }
 
+/** ZSTD_frameHeaderSize() :
+ *  srcSize must be >= ZSTD_frameHeaderSize_prefix.
+ * @return : size of the Frame Header,
+ *           or an error code (if srcSize is too small) */
+size_t ZSTD_frameHeaderSize(const void* src, size_t srcSize)
+{
+    return ZSTD_frameHeaderSize_internal(src, srcSize, ZSTD_f_zstd1);
+}
 
-/** ZSTD_getFrameHeader() :
-*   decode Frame Header, or require larger `srcSize`.
-*   @return : 0, `zfhPtr` is correctly filled,
-*            >0, `srcSize` is too small, result is expected `srcSize`,
-*             or an error code, which can be tested using ZSTD_isError() */
-size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize)
+
+/** ZSTD_getFrameHeader_advanced() :
+ *  decode Frame Header, or require larger `srcSize`.
+ *  note : only works for formats ZSTD_f_zstd1 and ZSTD_f_zstd1_magicless
+ * @return : 0, `zfhPtr` is correctly filled,
+ *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
+ *           or an error code, which can be tested using ZSTD_isError() */
+size_t ZSTD_getFrameHeader_advanced(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize, ZSTD_format_e format)
 {
     const BYTE* ip = (const BYTE*)src;
-    if (srcSize < ZSTD_frameHeaderSize_prefix) return ZSTD_frameHeaderSize_prefix;
+    size_t const minInputSize = ZSTD_startingInputLength(format);
+
+    memset(zfhPtr, 0, sizeof(*zfhPtr));   /* not strictly necessary, but static analyzer do not understand that zfhPtr is only going to be read only if return value is zero, since they are 2 different signals */
+    if (srcSize < minInputSize) return minInputSize;
+    RETURN_ERROR_IF(src==NULL, GENERIC, "invalid parameter");
 
-    if (MEM_readLE32(src) != ZSTD_MAGICNUMBER) {
-        if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {
+    if ( (format != ZSTD_f_zstd1_magicless)
+      && (MEM_readLE32(src) != ZSTD_MAGICNUMBER) ) {
+        if ((MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
             /* skippable frame */
-            if (srcSize < ZSTD_skippableHeaderSize)
-                return ZSTD_skippableHeaderSize; /* magic number + frame length */
+            if (srcSize < ZSTD_SKIPPABLEHEADERSIZE)
+                return ZSTD_SKIPPABLEHEADERSIZE; /* magic number + frame length */
             memset(zfhPtr, 0, sizeof(*zfhPtr));
-            zfhPtr->frameContentSize = MEM_readLE32((const char *)src + 4);
+            zfhPtr->frameContentSize = MEM_readLE32((const char *)src + ZSTD_FRAMEIDSIZE);
             zfhPtr->frameType = ZSTD_skippableFrame;
             return 0;
         }
-        return ERROR(prefix_unknown);
+        RETURN_ERROR(prefix_unknown, "");
     }
 
     /* ensure there is enough `srcSize` to fully read/decode frame header */
-    {   size_t const fhsize = ZSTD_frameHeaderSize(src, srcSize);
+    {   size_t const fhsize = ZSTD_frameHeaderSize_internal(src, srcSize, format);
         if (srcSize < fhsize) return fhsize;
         zfhPtr->headerSize = (U32)fhsize;
     }
 
-    {   BYTE const fhdByte = ip[4];
-        size_t pos = 5;
+    {   BYTE const fhdByte = ip[minInputSize-1];
+        size_t pos = minInputSize;
         U32 const dictIDSizeCode = fhdByte&3;
         U32 const checksumFlag = (fhdByte>>2)&1;
         U32 const singleSegment = (fhdByte>>5)&1;
@@ -318,14 +279,13 @@ size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t src
         U64 windowSize = 0;
         U32 dictID = 0;
         U64 frameContentSize = ZSTD_CONTENTSIZE_UNKNOWN;
-        if ((fhdByte & 0x08) != 0)
-            return ERROR(frameParameter_unsupported); /* reserved bits, must be zero */
+        RETURN_ERROR_IF((fhdByte & 0x08) != 0, frameParameter_unsupported,
+                        "reserved bits, must be zero");
 
         if (!singleSegment) {
             BYTE const wlByte = ip[pos++];
             U32 const windowLog = (wlByte >> 3) + ZSTD_WINDOWLOG_ABSOLUTEMIN;
-            if (windowLog > ZSTD_WINDOWLOG_MAX)
-                return ERROR(frameParameter_windowTooLarge);
+            RETURN_ERROR_IF(windowLog > ZSTD_WINDOWLOG_MAX, frameParameter_windowTooLarge, "");
             windowSize = (1ULL << windowLog);
             windowSize += (windowSize >> 3) * (wlByte&7);
         }
@@ -357,6 +317,18 @@ size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t src
     return 0;
 }
 
+/** ZSTD_getFrameHeader() :
+ *  decode Frame Header, or require larger `srcSize`.
+ *  note : this function does not consume input, it only reads it.
+ * @return : 0, `zfhPtr` is correctly filled,
+ *          >0, `srcSize` is too small, value is wanted `srcSize` amount,
+ *           or an error code, which can be tested using ZSTD_isError() */
+size_t ZSTD_getFrameHeader(ZSTD_frameHeader* zfhPtr, const void* src, size_t srcSize)
+{
+    return ZSTD_getFrameHeader_advanced(zfhPtr, src, srcSize, ZSTD_f_zstd1);
+}
+
+
 /** ZSTD_getFrameContentSize() :
  *  compatible with legacy mode
  * @return : decompressed size of the single frame pointed to be `src` if known, otherwise
@@ -380,6 +352,23 @@ unsigned long long ZSTD_getFrameContentSize(const void *src, size_t srcSize)
     }   }
 }
 
+static size_t readSkippableFrameSize(void const* src, size_t srcSize)
+{
+    size_t const skippableHeaderSize = ZSTD_SKIPPABLEHEADERSIZE;
+    U32 sizeU32;
+
+    RETURN_ERROR_IF(srcSize < ZSTD_SKIPPABLEHEADERSIZE, srcSize_wrong, "");
+
+    sizeU32 = MEM_readLE32((BYTE const*)src + ZSTD_FRAMEIDSIZE);
+    RETURN_ERROR_IF((U32)(sizeU32 + ZSTD_SKIPPABLEHEADERSIZE) < sizeU32,
+                    frameParameter_unsupported, "");
+    {
+        size_t const skippableSize = skippableHeaderSize + sizeU32;
+        RETURN_ERROR_IF(skippableSize > srcSize, srcSize_wrong, "");
+        return skippableSize;
+    }
+}
+
 /** ZSTD_findDecompressedSize() :
  *  compatible with legacy mode
  *  `srcSize` must be the exact length of some number of ZSTD compressed and/or
@@ -389,18 +378,15 @@ unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize)
 {
     unsigned long long totalDstSize = 0;
 
-    while (srcSize >= ZSTD_frameHeaderSize_prefix) {
-        const U32 magicNumber = MEM_readLE32(src);
+    while (srcSize >= ZSTD_startingInputLength(ZSTD_f_zstd1)) {
+        U32 const magicNumber = MEM_readLE32(src);
 
-        if ((magicNumber & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {
-            size_t skippableSize;
-            if (srcSize < ZSTD_skippableHeaderSize)
-                return ERROR(srcSize_wrong);
-            skippableSize = MEM_readLE32((const BYTE *)src + 4) +
-                            ZSTD_skippableHeaderSize;
-            if (srcSize < skippableSize) {
+        if ((magicNumber & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
+            size_t const skippableSize = readSkippableFrameSize(src, srcSize);
+            if (ZSTD_isError(skippableSize)) {
                 return ZSTD_CONTENTSIZE_ERROR;
             }
+            assert(skippableSize <= srcSize);
 
             src = (const BYTE *)src + skippableSize;
             srcSize -= skippableSize;
@@ -422,19 +408,17 @@ unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize)
             src = (const BYTE *)src + frameSrcSize;
             srcSize -= frameSrcSize;
         }
-    }
+    }  /* while (srcSize >= ZSTD_frameHeaderSize_prefix) */
 
-    if (srcSize) {
-        return ZSTD_CONTENTSIZE_ERROR;
-    }
+    if (srcSize) return ZSTD_CONTENTSIZE_ERROR;
 
     return totalDstSize;
 }
 
 /** ZSTD_getDecompressedSize() :
-*   compatible with legacy mode
-*   @return : decompressed size if known, 0 otherwise
-              note : 0 can mean any of the following :
+ *  compatible with legacy mode
+ * @return : decompressed size if known, 0 otherwise
+             note : 0 can mean any of the following :
                    - frame content is empty
                    - decompressed size field is not present in frame header
                    - frame header unknown / not supported
@@ -442,1084 +426,223 @@ unsigned long long ZSTD_findDecompressedSize(const void* src, size_t srcSize)
 unsigned long long ZSTD_getDecompressedSize(const void* src, size_t srcSize)
 {
     unsigned long long const ret = ZSTD_getFrameContentSize(src, srcSize);
-    return ret >= ZSTD_CONTENTSIZE_ERROR ? 0 : ret;
+    ZSTD_STATIC_ASSERT(ZSTD_CONTENTSIZE_ERROR < ZSTD_CONTENTSIZE_UNKNOWN);
+    return (ret >= ZSTD_CONTENTSIZE_ERROR) ? 0 : ret;
 }
 
 
 /** ZSTD_decodeFrameHeader() :
-*   `headerSize` must be the size provided by ZSTD_frameHeaderSize().
-*   @return : 0 if success, or an error code, which can be tested using ZSTD_isError() */
+ * `headerSize` must be the size provided by ZSTD_frameHeaderSize().
+ * @return : 0 if success, or an error code, which can be tested using ZSTD_isError() */
 static size_t ZSTD_decodeFrameHeader(ZSTD_DCtx* dctx, const void* src, size_t headerSize)
 {
-    size_t const result = ZSTD_getFrameHeader(&(dctx->fParams), src, headerSize);
-    if (ZSTD_isError(result)) return result;  /* invalid header */
-    if (result>0) return ERROR(srcSize_wrong);   /* headerSize too small */
-    if (dctx->fParams.dictID && (dctx->dictID != dctx->fParams.dictID))
-        return ERROR(dictionary_wrong);
+    size_t const result = ZSTD_getFrameHeader_advanced(&(dctx->fParams), src, headerSize, dctx->format);
+    if (ZSTD_isError(result)) return result;    /* invalid header */
+    RETURN_ERROR_IF(result>0, srcSize_wrong, "headerSize too small");
+#ifndef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    /* Skip the dictID check in fuzzing mode, because it makes the search
+     * harder.
+     */
+    RETURN_ERROR_IF(dctx->fParams.dictID && (dctx->dictID != dctx->fParams.dictID),
+                    dictionary_wrong, "");
+#endif
     if (dctx->fParams.checksumFlag) XXH64_reset(&dctx->xxhState, 0);
     return 0;
 }
 
-
-/*! ZSTD_getcBlockSize() :
-*   Provides the size of compressed block from block header `src` */
-size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
-                          blockProperties_t* bpPtr)
-{
-    if (srcSize < ZSTD_blockHeaderSize) return ERROR(srcSize_wrong);
-    {   U32 const cBlockHeader = MEM_readLE24(src);
-        U32 const cSize = cBlockHeader >> 3;
-        bpPtr->lastBlock = cBlockHeader & 1;
-        bpPtr->blockType = (blockType_e)((cBlockHeader >> 1) & 3);
-        bpPtr->origSize = cSize;   /* only useful for RLE */
-        if (bpPtr->blockType == bt_rle) return 1;
-        if (bpPtr->blockType == bt_reserved) return ERROR(corruption_detected);
-        return cSize;
-    }
-}
-
-
-static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity,
-                          const void* src, size_t srcSize)
-{
-    if (srcSize > dstCapacity) return ERROR(dstSize_tooSmall);
-    memcpy(dst, src, srcSize);
-    return srcSize;
-}
-
-
-static size_t ZSTD_setRleBlock(void* dst, size_t dstCapacity,
-                         const void* src, size_t srcSize,
-                               size_t regenSize)
-{
-    if (srcSize != 1) return ERROR(srcSize_wrong);
-    if (regenSize > dstCapacity) return ERROR(dstSize_tooSmall);
-    memset(dst, *(const BYTE*)src, regenSize);
-    return regenSize;
-}
-
-/*! ZSTD_decodeLiteralsBlock() :
-    @return : nb of bytes read from src (< srcSize ) */
-size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
-                          const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+static ZSTD_frameSizeInfo ZSTD_errorFrameSizeInfo(size_t ret)
 {
-    if (srcSize < MIN_CBLOCK_SIZE) return ERROR(corruption_detected);
-
-    {   const BYTE* const istart = (const BYTE*) src;
-        symbolEncodingType_e const litEncType = (symbolEncodingType_e)(istart[0] & 3);
-
-        switch(litEncType)
-        {
-        case set_repeat:
-            if (dctx->litEntropy==0) return ERROR(dictionary_corrupted);
-            /* fall-through */
-        case set_compressed:
-            if (srcSize < 5) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for case 3 */
-            {   size_t lhSize, litSize, litCSize;
-                U32 singleStream=0;
-                U32 const lhlCode = (istart[0] >> 2) & 3;
-                U32 const lhc = MEM_readLE32(istart);
-                switch(lhlCode)
-                {
-                case 0: case 1: default:   /* note : default is impossible, since lhlCode into [0..3] */
-                    /* 2 - 2 - 10 - 10 */
-                    singleStream = !lhlCode;
-                    lhSize = 3;
-                    litSize  = (lhc >> 4) & 0x3FF;
-                    litCSize = (lhc >> 14) & 0x3FF;
-                    break;
-                case 2:
-                    /* 2 - 2 - 14 - 14 */
-                    lhSize = 4;
-                    litSize  = (lhc >> 4) & 0x3FFF;
-                    litCSize = lhc >> 18;
-                    break;
-                case 3:
-                    /* 2 - 2 - 18 - 18 */
-                    lhSize = 5;
-                    litSize  = (lhc >> 4) & 0x3FFFF;
-                    litCSize = (lhc >> 22) + (istart[4] << 10);
-                    break;
-                }
-                if (litSize > ZSTD_BLOCKSIZE_MAX) return ERROR(corruption_detected);
-                if (litCSize + lhSize > srcSize) return ERROR(corruption_detected);
-
-                if (HUF_isError((litEncType==set_repeat) ?
-                                    ( singleStream ?
-                                        HUF_decompress1X_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->HUFptr) :
-                                        HUF_decompress4X_usingDTable(dctx->litBuffer, litSize, istart+lhSize, litCSize, dctx->HUFptr) ) :
-                                    ( singleStream ?
-                                        HUF_decompress1X2_DCtx_wksp(dctx->entropy.hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize,
-                                                                    dctx->entropy.workspace, sizeof(dctx->entropy.workspace)) :
-                                        HUF_decompress4X_hufOnly_wksp(dctx->entropy.hufTable, dctx->litBuffer, litSize, istart+lhSize, litCSize,
-                                                                      dctx->entropy.workspace, sizeof(dctx->entropy.workspace)))))
-                    return ERROR(corruption_detected);
-
-                dctx->litPtr = dctx->litBuffer;
-                dctx->litSize = litSize;
-                dctx->litEntropy = 1;
-                if (litEncType==set_compressed) dctx->HUFptr = dctx->entropy.hufTable;
-                memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
-                return litCSize + lhSize;
-            }
-
-        case set_basic:
-            {   size_t litSize, lhSize;
-                U32 const lhlCode = ((istart[0]) >> 2) & 3;
-                switch(lhlCode)
-                {
-                case 0: case 2: default:   /* note : default is impossible, since lhlCode into [0..3] */
-                    lhSize = 1;
-                    litSize = istart[0] >> 3;
-                    break;
-                case 1:
-                    lhSize = 2;
-                    litSize = MEM_readLE16(istart) >> 4;
-                    break;
-                case 3:
-                    lhSize = 3;
-                    litSize = MEM_readLE24(istart) >> 4;
-                    break;
-                }
-
-                if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
-                    if (litSize+lhSize > srcSize) return ERROR(corruption_detected);
-                    memcpy(dctx->litBuffer, istart+lhSize, litSize);
-                    dctx->litPtr = dctx->litBuffer;
-                    dctx->litSize = litSize;
-                    memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
-                    return lhSize+litSize;
-                }
-                /* direct reference into compressed stream */
-                dctx->litPtr = istart+lhSize;
-                dctx->litSize = litSize;
-                return lhSize+litSize;
-            }
-
-        case set_rle:
-            {   U32 const lhlCode = ((istart[0]) >> 2) & 3;
-                size_t litSize, lhSize;
-                switch(lhlCode)
-                {
-                case 0: case 2: default:   /* note : default is impossible, since lhlCode into [0..3] */
-                    lhSize = 1;
-                    litSize = istart[0] >> 3;
-                    break;
-                case 1:
-                    lhSize = 2;
-                    litSize = MEM_readLE16(istart) >> 4;
-                    break;
-                case 3:
-                    lhSize = 3;
-                    litSize = MEM_readLE24(istart) >> 4;
-                    if (srcSize<4) return ERROR(corruption_detected);   /* srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4 */
-                    break;
-                }
-                if (litSize > ZSTD_BLOCKSIZE_MAX) return ERROR(corruption_detected);
-                memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
-                dctx->litPtr = dctx->litBuffer;
-                dctx->litSize = litSize;
-                return lhSize+1;
-            }
-        default:
-            return ERROR(corruption_detected);   /* impossible */
-        }
-    }
+    ZSTD_frameSizeInfo frameSizeInfo;
+    frameSizeInfo.compressedSize = ret;
+    frameSizeInfo.decompressedBound = ZSTD_CONTENTSIZE_ERROR;
+    return frameSizeInfo;
 }
 
-
-typedef union {
-    FSE_decode_t realData;
-    U32 alignedBy4;
-} FSE_decode_t4;
-
-/* Default FSE distribution table for Literal Lengths */
-static const FSE_decode_t4 LL_defaultDTable[(1<<LL_DEFAULTNORMLOG)+1] = {
-    { { LL_DEFAULTNORMLOG, 1, 1 } }, /* header : tableLog, fastMode, fastMode */
-     /* base, symbol, bits */
-    { {  0,  0,  4 } }, { { 16,  0,  4 } }, { { 32,  1,  5 } }, { {  0,  3,  5 } },
-    { {  0,  4,  5 } }, { {  0,  6,  5 } }, { {  0,  7,  5 } }, { {  0,  9,  5 } },
-    { {  0, 10,  5 } }, { {  0, 12,  5 } }, { {  0, 14,  6 } }, { {  0, 16,  5 } },
-    { {  0, 18,  5 } }, { {  0, 19,  5 } }, { {  0, 21,  5 } }, { {  0, 22,  5 } },
-    { {  0, 24,  5 } }, { { 32, 25,  5 } }, { {  0, 26,  5 } }, { {  0, 27,  6 } },
-    { {  0, 29,  6 } }, { {  0, 31,  6 } }, { { 32,  0,  4 } }, { {  0,  1,  4 } },
-    { {  0,  2,  5 } }, { { 32,  4,  5 } }, { {  0,  5,  5 } }, { { 32,  7,  5 } },
-    { {  0,  8,  5 } }, { { 32, 10,  5 } }, { {  0, 11,  5 } }, { {  0, 13,  6 } },
-    { { 32, 16,  5 } }, { {  0, 17,  5 } }, { { 32, 19,  5 } }, { {  0, 20,  5 } },
-    { { 32, 22,  5 } }, { {  0, 23,  5 } }, { {  0, 25,  4 } }, { { 16, 25,  4 } },
-    { { 32, 26,  5 } }, { {  0, 28,  6 } }, { {  0, 30,  6 } }, { { 48,  0,  4 } },
-    { { 16,  1,  4 } }, { { 32,  2,  5 } }, { { 32,  3,  5 } }, { { 32,  5,  5 } },
-    { { 32,  6,  5 } }, { { 32,  8,  5 } }, { { 32,  9,  5 } }, { { 32, 11,  5 } },
-    { { 32, 12,  5 } }, { {  0, 15,  6 } }, { { 32, 17,  5 } }, { { 32, 18,  5 } },
-    { { 32, 20,  5 } }, { { 32, 21,  5 } }, { { 32, 23,  5 } }, { { 32, 24,  5 } },
-    { {  0, 35,  6 } }, { {  0, 34,  6 } }, { {  0, 33,  6 } }, { {  0, 32,  6 } },
-};   /* LL_defaultDTable */
-
-/* Default FSE distribution table for Match Lengths */
-static const FSE_decode_t4 ML_defaultDTable[(1<<ML_DEFAULTNORMLOG)+1] = {
-    { { ML_DEFAULTNORMLOG, 1, 1 } }, /* header : tableLog, fastMode, fastMode */
-    /* base, symbol, bits */
-    { {  0,  0,  6 } }, { {  0,  1,  4 } }, { { 32,  2,  5 } }, { {  0,  3,  5 } },
-    { {  0,  5,  5 } }, { {  0,  6,  5 } }, { {  0,  8,  5 } }, { {  0, 10,  6 } },
-    { {  0, 13,  6 } }, { {  0, 16,  6 } }, { {  0, 19,  6 } }, { {  0, 22,  6 } },
-    { {  0, 25,  6 } }, { {  0, 28,  6 } }, { {  0, 31,  6 } }, { {  0, 33,  6 } },
-    { {  0, 35,  6 } }, { {  0, 37,  6 } }, { {  0, 39,  6 } }, { {  0, 41,  6 } },
-    { {  0, 43,  6 } }, { {  0, 45,  6 } }, { { 16,  1,  4 } }, { {  0,  2,  4 } },
-    { { 32,  3,  5 } }, { {  0,  4,  5 } }, { { 32,  6,  5 } }, { {  0,  7,  5 } },
-    { {  0,  9,  6 } }, { {  0, 12,  6 } }, { {  0, 15,  6 } }, { {  0, 18,  6 } },
-    { {  0, 21,  6 } }, { {  0, 24,  6 } }, { {  0, 27,  6 } }, { {  0, 30,  6 } },
-    { {  0, 32,  6 } }, { {  0, 34,  6 } }, { {  0, 36,  6 } }, { {  0, 38,  6 } },
-    { {  0, 40,  6 } }, { {  0, 42,  6 } }, { {  0, 44,  6 } }, { { 32,  1,  4 } },
-    { { 48,  1,  4 } }, { { 16,  2,  4 } }, { { 32,  4,  5 } }, { { 32,  5,  5 } },
-    { { 32,  7,  5 } }, { { 32,  8,  5 } }, { {  0, 11,  6 } }, { {  0, 14,  6 } },
-    { {  0, 17,  6 } }, { {  0, 20,  6 } }, { {  0, 23,  6 } }, { {  0, 26,  6 } },
-    { {  0, 29,  6 } }, { {  0, 52,  6 } }, { {  0, 51,  6 } }, { {  0, 50,  6 } },
-    { {  0, 49,  6 } }, { {  0, 48,  6 } }, { {  0, 47,  6 } }, { {  0, 46,  6 } },
-};   /* ML_defaultDTable */
-
-/* Default FSE distribution table for Offset Codes */
-static const FSE_decode_t4 OF_defaultDTable[(1<<OF_DEFAULTNORMLOG)+1] = {
-    { { OF_DEFAULTNORMLOG, 1, 1 } }, /* header : tableLog, fastMode, fastMode */
-    /* base, symbol, bits */
-    { {  0,  0,  5 } }, { {  0,  6,  4 } },
-    { {  0,  9,  5 } }, { {  0, 15,  5 } },
-    { {  0, 21,  5 } }, { {  0,  3,  5 } },
-    { {  0,  7,  4 } }, { {  0, 12,  5 } },
-    { {  0, 18,  5 } }, { {  0, 23,  5 } },
-    { {  0,  5,  5 } }, { {  0,  8,  4 } },
-    { {  0, 14,  5 } }, { {  0, 20,  5 } },
-    { {  0,  2,  5 } }, { { 16,  7,  4 } },
-    { {  0, 11,  5 } }, { {  0, 17,  5 } },
-    { {  0, 22,  5 } }, { {  0,  4,  5 } },
-    { { 16,  8,  4 } }, { {  0, 13,  5 } },
-    { {  0, 19,  5 } }, { {  0,  1,  5 } },
-    { { 16,  6,  4 } }, { {  0, 10,  5 } },
-    { {  0, 16,  5 } }, { {  0, 28,  5 } },
-    { {  0, 27,  5 } }, { {  0, 26,  5 } },
-    { {  0, 25,  5 } }, { {  0, 24,  5 } },
-};   /* OF_defaultDTable */
-
-/*! ZSTD_buildSeqTable() :
-    @return : nb bytes read from src,
-              or an error code if it fails, testable with ZSTD_isError()
-*/
-static size_t ZSTD_buildSeqTable(FSE_DTable* DTableSpace, const FSE_DTable** DTablePtr,
-                                 symbolEncodingType_e type, U32 max, U32 maxLog,
-                                 const void* src, size_t srcSize,
-                                 const FSE_decode_t4* defaultTable, U32 flagRepeatTable)
+static ZSTD_frameSizeInfo ZSTD_findFrameSizeInfo(const void* src, size_t srcSize)
 {
-    const void* const tmpPtr = defaultTable;   /* bypass strict aliasing */
-    switch(type)
-    {
-    case set_rle :
-        if (!srcSize) return ERROR(srcSize_wrong);
-        if ( (*(const BYTE*)src) > max) return ERROR(corruption_detected);
-        FSE_buildDTable_rle(DTableSpace, *(const BYTE*)src);
-        *DTablePtr = DTableSpace;
-        return 1;
-    case set_basic :
-        *DTablePtr = (const FSE_DTable*)tmpPtr;
-        return 0;
-    case set_repeat:
-        if (!flagRepeatTable) return ERROR(corruption_detected);
-        return 0;
-    default :   /* impossible */
-    case set_compressed :
-        {   U32 tableLog;
-            S16 norm[MaxSeq+1];
-            size_t const headerSize = FSE_readNCount(norm, &max, &tableLog, src, srcSize);
-            if (FSE_isError(headerSize)) return ERROR(corruption_detected);
-            if (tableLog > maxLog) return ERROR(corruption_detected);
-            FSE_buildDTable(DTableSpace, norm, max, tableLog);
-            *DTablePtr = DTableSpace;
-            return headerSize;
-    }   }
-}
+    ZSTD_frameSizeInfo frameSizeInfo;
+    memset(&frameSizeInfo, 0, sizeof(ZSTD_frameSizeInfo));
 
-size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
-                             const void* src, size_t srcSize)
-{
-    const BYTE* const istart = (const BYTE* const)src;
-    const BYTE* const iend = istart + srcSize;
-    const BYTE* ip = istart;
-    DEBUGLOG(5, "ZSTD_decodeSeqHeaders");
-
-    /* check */
-    if (srcSize < MIN_SEQUENCES_SIZE) return ERROR(srcSize_wrong);
-
-    /* SeqHead */
-    {   int nbSeq = *ip++;
-        if (!nbSeq) { *nbSeqPtr=0; return 1; }
-        if (nbSeq > 0x7F) {
-            if (nbSeq == 0xFF) {
-                if (ip+2 > iend) return ERROR(srcSize_wrong);
-                nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2;
-            } else {
-                if (ip >= iend) return ERROR(srcSize_wrong);
-                nbSeq = ((nbSeq-0x80)<<8) + *ip++;
-            }
-        }
-        *nbSeqPtr = nbSeq;
-    }
-
-    /* FSE table descriptors */
-    if (ip+4 > iend) return ERROR(srcSize_wrong); /* minimum possible size */
-    {   symbolEncodingType_e const LLtype = (symbolEncodingType_e)(*ip >> 6);
-        symbolEncodingType_e const OFtype = (symbolEncodingType_e)((*ip >> 4) & 3);
-        symbolEncodingType_e const MLtype = (symbolEncodingType_e)((*ip >> 2) & 3);
-        ip++;
-
-        /* Build DTables */
-        {   size_t const llhSize = ZSTD_buildSeqTable(dctx->entropy.LLTable, &dctx->LLTptr,
-                                                      LLtype, MaxLL, LLFSELog,
-                                                      ip, iend-ip, LL_defaultDTable, dctx->fseEntropy);
-            if (ZSTD_isError(llhSize)) return ERROR(corruption_detected);
-            ip += llhSize;
-        }
-        {   size_t const ofhSize = ZSTD_buildSeqTable(dctx->entropy.OFTable, &dctx->OFTptr,
-                                                      OFtype, MaxOff, OffFSELog,
-                                                      ip, iend-ip, OF_defaultDTable, dctx->fseEntropy);
-            if (ZSTD_isError(ofhSize)) return ERROR(corruption_detected);
-            ip += ofhSize;
-        }
-        {   size_t const mlhSize = ZSTD_buildSeqTable(dctx->entropy.MLTable, &dctx->MLTptr,
-                                                      MLtype, MaxML, MLFSELog,
-                                                      ip, iend-ip, ML_defaultDTable, dctx->fseEntropy);
-            if (ZSTD_isError(mlhSize)) return ERROR(corruption_detected);
-            ip += mlhSize;
-        }
-    }
-
-    return ip-istart;
-}
-
-
-typedef struct {
-    size_t litLength;
-    size_t matchLength;
-    size_t offset;
-    const BYTE* match;
-} seq_t;
-
-typedef struct {
-    BIT_DStream_t DStream;
-    FSE_DState_t stateLL;
-    FSE_DState_t stateOffb;
-    FSE_DState_t stateML;
-    size_t prevOffset[ZSTD_REP_NUM];
-    const BYTE* base;
-    size_t pos;
-    uPtrDiff gotoDict;
-} seqState_t;
-
-
-FORCE_NOINLINE
-size_t ZSTD_execSequenceLast7(BYTE* op,
-                              BYTE* const oend, seq_t sequence,
-                              const BYTE** litPtr, const BYTE* const litLimit,
-                              const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
-{
-    BYTE* const oLitEnd = op + sequence.litLength;
-    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
-    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
-    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;
-    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
-    const BYTE* match = oLitEnd - sequence.offset;
-
-    /* check */
-    if (oMatchEnd>oend) return ERROR(dstSize_tooSmall); /* last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend */
-    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
-    if (oLitEnd <= oend_w) return ERROR(GENERIC);   /* Precondition */
-
-    /* copy literals */
-    if (op < oend_w) {
-        ZSTD_wildcopy(op, *litPtr, oend_w - op);
-        *litPtr += oend_w - op;
-        op = oend_w;
-    }
-    while (op < oLitEnd) *op++ = *(*litPtr)++;
-
-    /* copy Match */
-    if (sequence.offset > (size_t)(oLitEnd - base)) {
-        /* offset beyond prefix */
-        if (sequence.offset > (size_t)(oLitEnd - vBase)) return ERROR(corruption_detected);
-        match = dictEnd - (base-match);
-        if (match + sequence.matchLength <= dictEnd) {
-            memmove(oLitEnd, match, sequence.matchLength);
-            return sequenceLength;
-        }
-        /* span extDict & currentPrefixSegment */
-        {   size_t const length1 = dictEnd - match;
-            memmove(oLitEnd, match, length1);
-            op = oLitEnd + length1;
-            sequence.matchLength -= length1;
-            match = base;
-    }   }
-    while (op < oMatchEnd) *op++ = *match++;
-    return sequenceLength;
-}
-
-
-typedef enum { ZSTD_lo_isRegularOffset, ZSTD_lo_isLongOffset=1 } ZSTD_longOffset_e;
-
-
-static seq_t ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets)
-{
-    seq_t seq;
-
-    U32 const llCode = FSE_peekSymbol(&seqState->stateLL);
-    U32 const mlCode = FSE_peekSymbol(&seqState->stateML);
-    U32 const ofCode = FSE_peekSymbol(&seqState->stateOffb);   /* <= maxOff, by table construction */
-
-    U32 const llBits = LL_bits[llCode];
-    U32 const mlBits = ML_bits[mlCode];
-    U32 const ofBits = ofCode;
-    U32 const totalBits = llBits+mlBits+ofBits;
-
-    static const U32 LL_base[MaxLL+1] = {
-                             0,    1,    2,     3,     4,     5,     6,      7,
-                             8,    9,   10,    11,    12,    13,    14,     15,
-                            16,   18,   20,    22,    24,    28,    32,     40,
-                            48,   64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
-                            0x2000, 0x4000, 0x8000, 0x10000 };
-
-    static const U32 ML_base[MaxML+1] = {
-                             3,  4,  5,    6,     7,     8,     9,    10,
-                            11, 12, 13,   14,    15,    16,    17,    18,
-                            19, 20, 21,   22,    23,    24,    25,    26,
-                            27, 28, 29,   30,    31,    32,    33,    34,
-                            35, 37, 39,   41,    43,    47,    51,    59,
-                            67, 83, 99, 0x83, 0x103, 0x203, 0x403, 0x803,
-                            0x1003, 0x2003, 0x4003, 0x8003, 0x10003 };
-
-    static const U32 OF_base[MaxOff+1] = {
-                     0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
-                     0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
-                     0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
-                     0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD };
-
-    /* sequence */
-    {   size_t offset;
-        if (!ofCode)
-            offset = 0;
-        else {
-            ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1);
-            if (longOffsets) {
-                int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN);
-                offset = OF_base[ofCode] + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
-                if (MEM_32bits() || extraBits) BIT_reloadDStream(&seqState->DStream);
-                if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits);
-            } else {
-                offset = OF_base[ofCode] + BIT_readBitsFast(&seqState->DStream, ofBits);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
-                if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
-            }
-        }
-
-        if (ofCode <= 1) {
-            offset += (llCode==0);
-            if (offset) {
-                size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
-                temp += !temp;   /* 0 is not valid; input is corrupted; force offset to 1 */
-                if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
-                seqState->prevOffset[1] = seqState->prevOffset[0];
-                seqState->prevOffset[0] = offset = temp;
-            } else {
-                offset = seqState->prevOffset[0];
-            }
-        } else {
-            seqState->prevOffset[2] = seqState->prevOffset[1];
-            seqState->prevOffset[1] = seqState->prevOffset[0];
-            seqState->prevOffset[0] = offset;
-        }
-        seq.offset = offset;
-    }
-
-    seq.matchLength = ML_base[mlCode]
-                    + ((mlCode>31) ? BIT_readBitsFast(&seqState->DStream, mlBits) : 0);  /* <=  16 bits */
-    if (MEM_32bits() && (mlBits+llBits>24)) BIT_reloadDStream(&seqState->DStream);
-
-    seq.litLength = LL_base[llCode]
-                  + ((llCode>15) ? BIT_readBitsFast(&seqState->DStream, llBits) : 0);    /* <=  16 bits */
-    if (  MEM_32bits()
-      || (totalBits > 64 - 7 - (LLFSELog+MLFSELog+OffFSELog)) )
-       BIT_reloadDStream(&seqState->DStream);
-
-    DEBUGLOG(6, "seq: litL=%u, matchL=%u, offset=%u",
-                (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
-
-    /* ANS state update */
-    FSE_updateState(&seqState->stateLL, &seqState->DStream);    /* <=  9 bits */
-    FSE_updateState(&seqState->stateML, &seqState->DStream);    /* <=  9 bits */
-    if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
-    FSE_updateState(&seqState->stateOffb, &seqState->DStream);  /* <=  8 bits */
-
-    return seq;
-}
-
-
-HINT_INLINE
-size_t ZSTD_execSequence(BYTE* op,
-                         BYTE* const oend, seq_t sequence,
-                         const BYTE** litPtr, const BYTE* const litLimit,
-                         const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
-{
-    BYTE* const oLitEnd = op + sequence.litLength;
-    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
-    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
-    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;
-    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
-    const BYTE* match = oLitEnd - sequence.offset;
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
+    if (ZSTD_isLegacy(src, srcSize))
+        return ZSTD_findFrameSizeInfoLegacy(src, srcSize);
+#endif
 
-    /* check */
-    if (oMatchEnd>oend) return ERROR(dstSize_tooSmall); /* last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend */
-    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
-    if (oLitEnd>oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, base, vBase, dictEnd);
-
-    /* copy Literals */
-    ZSTD_copy8(op, *litPtr);
-    if (sequence.litLength > 8)
-        ZSTD_wildcopy(op+8, (*litPtr)+8, sequence.litLength - 8);   /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */
-    op = oLitEnd;
-    *litPtr = iLitEnd;   /* update for next sequence */
-
-    /* copy Match */
-    if (sequence.offset > (size_t)(oLitEnd - base)) {
-        /* offset beyond prefix -> go into extDict */
-        if (sequence.offset > (size_t)(oLitEnd - vBase))
-            return ERROR(corruption_detected);
-        match = dictEnd + (match - base);
-        if (match + sequence.matchLength <= dictEnd) {
-            memmove(oLitEnd, match, sequence.matchLength);
-            return sequenceLength;
-        }
-        /* span extDict & currentPrefixSegment */
-        {   size_t const length1 = dictEnd - match;
-            memmove(oLitEnd, match, length1);
-            op = oLitEnd + length1;
-            sequence.matchLength -= length1;
-            match = base;
-            if (op > oend_w || sequence.matchLength < MINMATCH) {
-              U32 i;
-              for (i = 0; i < sequence.matchLength; ++i) op[i] = match[i];
-              return sequenceLength;
-            }
-    }   }
-    /* Requirement: op <= oend_w && sequence.matchLength >= MINMATCH */
-
-    /* match within prefix */
-    if (sequence.offset < 8) {
-        /* close range match, overlap */
-        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
-        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
-        int const sub2 = dec64table[sequence.offset];
-        op[0] = match[0];
-        op[1] = match[1];
-        op[2] = match[2];
-        op[3] = match[3];
-        match += dec32table[sequence.offset];
-        ZSTD_copy4(op+4, match);
-        match -= sub2;
+    if ((srcSize >= ZSTD_SKIPPABLEHEADERSIZE)
+        && (MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
+        frameSizeInfo.compressedSize = readSkippableFrameSize(src, srcSize);
+        assert(ZSTD_isError(frameSizeInfo.compressedSize) ||
+               frameSizeInfo.compressedSize <= srcSize);
+        return frameSizeInfo;
     } else {
-        ZSTD_copy8(op, match);
-    }
-    op += 8; match += 8;
+        const BYTE* ip = (const BYTE*)src;
+        const BYTE* const ipstart = ip;
+        size_t remainingSize = srcSize;
+        size_t nbBlocks = 0;
+        ZSTD_frameHeader zfh;
 
-    if (oMatchEnd > oend-(16-MINMATCH)) {
-        if (op < oend_w) {
-            ZSTD_wildcopy(op, match, oend_w - op);
-            match += oend_w - op;
-            op = oend_w;
+        /* Extract Frame Header */
+        {   size_t const ret = ZSTD_getFrameHeader(&zfh, src, srcSize);
+            if (ZSTD_isError(ret))
+                return ZSTD_errorFrameSizeInfo(ret);
+            if (ret > 0)
+                return ZSTD_errorFrameSizeInfo(ERROR(srcSize_wrong));
         }
-        while (op < oMatchEnd) *op++ = *match++;
-    } else {
-        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
-    }
-    return sequenceLength;
-}
-
-
-static size_t ZSTD_decompressSequences(
-                               ZSTD_DCtx* dctx,
-                               void* dst, size_t maxDstSize,
-                         const void* seqStart, size_t seqSize,
-                         const ZSTD_longOffset_e isLongOffset)
-{
-    const BYTE* ip = (const BYTE*)seqStart;
-    const BYTE* const iend = ip + seqSize;
-    BYTE* const ostart = (BYTE* const)dst;
-    BYTE* const oend = ostart + maxDstSize;
-    BYTE* op = ostart;
-    const BYTE* litPtr = dctx->litPtr;
-    const BYTE* const litEnd = litPtr + dctx->litSize;
-    const BYTE* const base = (const BYTE*) (dctx->base);
-    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
-    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
-    int nbSeq;
-    DEBUGLOG(5, "ZSTD_decompressSequences");
-
-    /* Build Decoding Tables */
-    {   size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, seqSize);
-        DEBUGLOG(5, "ZSTD_decodeSeqHeaders: size=%u, nbSeq=%i",
-                    (U32)seqHSize, nbSeq);
-        if (ZSTD_isError(seqHSize)) return seqHSize;
-        ip += seqHSize;
-    }
-
-    /* Regen sequences */
-    if (nbSeq) {
-        seqState_t seqState;
-        dctx->fseEntropy = 1;
-        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
-        CHECK_E(BIT_initDStream(&seqState.DStream, ip, iend-ip), corruption_detected);
-        FSE_initDState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
-        FSE_initDState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
-        FSE_initDState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
-
-        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && nbSeq ; ) {
-            nbSeq--;
-            {   seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset);
-                size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, base, vBase, dictEnd);
-                DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
-                if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
-                op += oneSeqSize;
-        }   }
 
-        /* check if reached exact end */
-        DEBUGLOG(5, "after decode loop, remaining nbSeq : %i", nbSeq);
-        if (nbSeq) return ERROR(corruption_detected);
-        /* save reps for next block */
-        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
-    }
-
-    /* last literal segment */
-    {   size_t const lastLLSize = litEnd - litPtr;
-        if (lastLLSize > (size_t)(oend-op)) return ERROR(dstSize_tooSmall);
-        memcpy(op, litPtr, lastLLSize);
-        op += lastLLSize;
-    }
-
-    return op-ostart;
-}
+        ip += zfh.headerSize;
+        remainingSize -= zfh.headerSize;
 
+        /* Iterate over each block */
+        while (1) {
+            blockProperties_t blockProperties;
+            size_t const cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+            if (ZSTD_isError(cBlockSize))
+                return ZSTD_errorFrameSizeInfo(cBlockSize);
 
+            if (ZSTD_blockHeaderSize + cBlockSize > remainingSize)
+                return ZSTD_errorFrameSizeInfo(ERROR(srcSize_wrong));
 
-HINT_INLINE
-seq_t ZSTD_decodeSequenceLong(seqState_t* seqState, ZSTD_longOffset_e const longOffsets)
-{
-    seq_t seq;
-
-    U32 const llCode = FSE_peekSymbol(&seqState->stateLL);
-    U32 const mlCode = FSE_peekSymbol(&seqState->stateML);
-    U32 const ofCode = FSE_peekSymbol(&seqState->stateOffb);   /* <= maxOff, by table construction */
-
-    U32 const llBits = LL_bits[llCode];
-    U32 const mlBits = ML_bits[mlCode];
-    U32 const ofBits = ofCode;
-    U32 const totalBits = llBits+mlBits+ofBits;
-
-    static const U32 LL_base[MaxLL+1] = {
-                             0,  1,    2,     3,     4,     5,     6,      7,
-                             8,  9,   10,    11,    12,    13,    14,     15,
-                            16, 18,   20,    22,    24,    28,    32,     40,
-                            48, 64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
-                            0x2000, 0x4000, 0x8000, 0x10000 };
-
-    static const U32 ML_base[MaxML+1] = {
-                             3,  4,  5,    6,     7,     8,     9,    10,
-                            11, 12, 13,   14,    15,    16,    17,    18,
-                            19, 20, 21,   22,    23,    24,    25,    26,
-                            27, 28, 29,   30,    31,    32,    33,    34,
-                            35, 37, 39,   41,    43,    47,    51,    59,
-                            67, 83, 99, 0x83, 0x103, 0x203, 0x403, 0x803,
-                            0x1003, 0x2003, 0x4003, 0x8003, 0x10003 };
-
-    static const U32 OF_base[MaxOff+1] = {
-                     0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
-                     0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
-                     0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
-                     0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD };
-
-    /* sequence */
-    {   size_t offset;
-        if (!ofCode)
-            offset = 0;
-        else {
-            ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1);
-            if (longOffsets) {
-                int const extraBits = ofBits - MIN(ofBits, STREAM_ACCUMULATOR_MIN);
-                offset = OF_base[ofCode] + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
-                if (MEM_32bits() || extraBits) BIT_reloadDStream(&seqState->DStream);
-                if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits);
-            } else {
-                offset = OF_base[ofCode] + BIT_readBitsFast(&seqState->DStream, ofBits);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
-                if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
-            }
-        }
+            ip += ZSTD_blockHeaderSize + cBlockSize;
+            remainingSize -= ZSTD_blockHeaderSize + cBlockSize;
+            nbBlocks++;
 
-        if (ofCode <= 1) {
-            offset += (llCode==0);
-            if (offset) {
-                size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
-                temp += !temp;   /* 0 is not valid; input is corrupted; force offset to 1 */
-                if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
-                seqState->prevOffset[1] = seqState->prevOffset[0];
-                seqState->prevOffset[0] = offset = temp;
-            } else {
-                offset = seqState->prevOffset[0];
-            }
-        } else {
-            seqState->prevOffset[2] = seqState->prevOffset[1];
-            seqState->prevOffset[1] = seqState->prevOffset[0];
-            seqState->prevOffset[0] = offset;
+            if (blockProperties.lastBlock) break;
         }
-        seq.offset = offset;
-    }
-
-    seq.matchLength = ML_base[mlCode] + ((mlCode>31) ? BIT_readBitsFast(&seqState->DStream, mlBits) : 0);  /* <=  16 bits */
-    if (MEM_32bits() && (mlBits+llBits>24)) BIT_reloadDStream(&seqState->DStream);
-
-    seq.litLength = LL_base[llCode] + ((llCode>15) ? BIT_readBitsFast(&seqState->DStream, llBits) : 0);    /* <=  16 bits */
-    if (MEM_32bits() ||
-       (totalBits > 64 - 7 - (LLFSELog+MLFSELog+OffFSELog)) ) BIT_reloadDStream(&seqState->DStream);
-
-    {   size_t const pos = seqState->pos + seq.litLength;
-        seq.match = seqState->base + pos - seq.offset;    /* single memory segment */
-        if (seq.offset > pos) seq.match += seqState->gotoDict;   /* separate memory segment */
-        seqState->pos = pos + seq.matchLength;
-    }
 
-    /* ANS state update */
-    FSE_updateState(&seqState->stateLL, &seqState->DStream);    /* <=  9 bits */
-    FSE_updateState(&seqState->stateML, &seqState->DStream);    /* <=  9 bits */
-    if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
-    FSE_updateState(&seqState->stateOffb, &seqState->DStream);  /* <=  8 bits */
-
-    return seq;
-}
-
-
-HINT_INLINE
-size_t ZSTD_execSequenceLong(BYTE* op,
-                             BYTE* const oend, seq_t sequence,
-                             const BYTE** litPtr, const BYTE* const litLimit,
-                             const BYTE* const base, const BYTE* const vBase, const BYTE* const dictEnd)
-{
-    BYTE* const oLitEnd = op + sequence.litLength;
-    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
-    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
-    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;
-    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
-    const BYTE* match = sequence.match;
-
-    /* check */
-    if (oMatchEnd>oend) return ERROR(dstSize_tooSmall); /* last match must start at a minimum distance of WILDCOPY_OVERLENGTH from oend */
-    if (iLitEnd > litLimit) return ERROR(corruption_detected);   /* over-read beyond lit buffer */
-    if (oLitEnd>oend_w) return ZSTD_execSequenceLast7(op, oend, sequence, litPtr, litLimit, base, vBase, dictEnd);
-
-    /* copy Literals */
-    ZSTD_copy8(op, *litPtr);
-    if (sequence.litLength > 8)
-        ZSTD_wildcopy(op+8, (*litPtr)+8, sequence.litLength - 8);   /* note : since oLitEnd <= oend-WILDCOPY_OVERLENGTH, no risk of overwrite beyond oend */
-    op = oLitEnd;
-    *litPtr = iLitEnd;   /* update for next sequence */
-
-    /* copy Match */
-    if (sequence.offset > (size_t)(oLitEnd - base)) {
-        /* offset beyond prefix */
-        if (sequence.offset > (size_t)(oLitEnd - vBase)) return ERROR(corruption_detected);
-        if (match + sequence.matchLength <= dictEnd) {
-            memmove(oLitEnd, match, sequence.matchLength);
-            return sequenceLength;
+        /* Final frame content checksum */
+        if (zfh.checksumFlag) {
+            if (remainingSize < 4)
+                return ZSTD_errorFrameSizeInfo(ERROR(srcSize_wrong));
+            ip += 4;
         }
-        /* span extDict & currentPrefixSegment */
-        {   size_t const length1 = dictEnd - match;
-            memmove(oLitEnd, match, length1);
-            op = oLitEnd + length1;
-            sequence.matchLength -= length1;
-            match = base;
-            if (op > oend_w || sequence.matchLength < MINMATCH) {
-              U32 i;
-              for (i = 0; i < sequence.matchLength; ++i) op[i] = match[i];
-              return sequenceLength;
-            }
-    }   }
-    assert(op <= oend_w);
-    assert(sequence.matchLength >= MINMATCH);
-
-    /* match within prefix */
-    if (sequence.offset < 8) {
-        /* close range match, overlap */
-        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
-        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
-        int const sub2 = dec64table[sequence.offset];
-        op[0] = match[0];
-        op[1] = match[1];
-        op[2] = match[2];
-        op[3] = match[3];
-        match += dec32table[sequence.offset];
-        ZSTD_copy4(op+4, match);
-        match -= sub2;
-    } else {
-        ZSTD_copy8(op, match);
-    }
-    op += 8; match += 8;
 
-    if (oMatchEnd > oend-(16-MINMATCH)) {
-        if (op < oend_w) {
-            ZSTD_wildcopy(op, match, oend_w - op);
-            match += oend_w - op;
-            op = oend_w;
-        }
-        while (op < oMatchEnd) *op++ = *match++;
-    } else {
-        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8);   /* works even if matchLength < 8 */
+        frameSizeInfo.compressedSize = ip - ipstart;
+        frameSizeInfo.decompressedBound = (zfh.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN)
+                                        ? zfh.frameContentSize
+                                        : nbBlocks * zfh.blockSizeMax;
+        return frameSizeInfo;
     }
-    return sequenceLength;
 }
 
-static size_t ZSTD_decompressSequencesLong(
-                               ZSTD_DCtx* dctx,
-                               void* dst, size_t maxDstSize,
-                         const void* seqStart, size_t seqSize,
-                         const ZSTD_longOffset_e isLongOffset)
+/** ZSTD_findFrameCompressedSize() :
+ *  compatible with legacy mode
+ *  `src` must point to the start of a ZSTD frame, ZSTD legacy frame, or skippable frame
+ *  `srcSize` must be at least as large as the frame contained
+ *  @return : the compressed size of the frame starting at `src` */
+size_t ZSTD_findFrameCompressedSize(const void *src, size_t srcSize)
 {
-    const BYTE* ip = (const BYTE*)seqStart;
-    const BYTE* const iend = ip + seqSize;
-    BYTE* const ostart = (BYTE* const)dst;
-    BYTE* const oend = ostart + maxDstSize;
-    BYTE* op = ostart;
-    const BYTE* litPtr = dctx->litPtr;
-    const BYTE* const litEnd = litPtr + dctx->litSize;
-    const BYTE* const base = (const BYTE*) (dctx->base);
-    const BYTE* const vBase = (const BYTE*) (dctx->vBase);
-    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
-    int nbSeq;
-
-    /* Build Decoding Tables */
-    {   size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, seqSize);
-        if (ZSTD_isError(seqHSize)) return seqHSize;
-        ip += seqHSize;
-    }
-
-    /* Regen sequences */
-    if (nbSeq) {
-#define STORED_SEQS 4
-#define STOSEQ_MASK (STORED_SEQS-1)
-#define ADVANCED_SEQS 4
-        seq_t sequences[STORED_SEQS];
-        int const seqAdvance = MIN(nbSeq, ADVANCED_SEQS);
-        seqState_t seqState;
-        int seqNb;
-        dctx->fseEntropy = 1;
-        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
-        seqState.base = base;
-        seqState.pos = (size_t)(op-base);
-        seqState.gotoDict = (uPtrDiff)dictEnd - (uPtrDiff)base; /* cast to avoid undefined behaviour */
-        CHECK_E(BIT_initDStream(&seqState.DStream, ip, iend-ip), corruption_detected);
-        FSE_initDState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
-        FSE_initDState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
-        FSE_initDState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
-
-        /* prepare in advance */
-        for (seqNb=0; (BIT_reloadDStream(&seqState.DStream) <= BIT_DStream_completed) && seqNb<seqAdvance; seqNb++) {
-            sequences[seqNb] = ZSTD_decodeSequenceLong(&seqState, isLongOffset);
-        }
-        if (seqNb<seqAdvance) return ERROR(corruption_detected);
-
-        /* decode and decompress */
-        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && seqNb<nbSeq ; seqNb++) {
-            seq_t const sequence = ZSTD_decodeSequenceLong(&seqState, isLongOffset);
-            size_t const oneSeqSize = ZSTD_execSequenceLong(op, oend, sequences[(seqNb-ADVANCED_SEQS) & STOSEQ_MASK], &litPtr, litEnd, base, vBase, dictEnd);
-            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
-            PREFETCH(sequence.match);
-            sequences[seqNb&STOSEQ_MASK] = sequence;
-            op += oneSeqSize;
-        }
-        if (seqNb<nbSeq) return ERROR(corruption_detected);
-
-        /* finish queue */
-        seqNb -= seqAdvance;
-        for ( ; seqNb<nbSeq ; seqNb++) {
-            size_t const oneSeqSize = ZSTD_execSequenceLong(op, oend, sequences[seqNb&STOSEQ_MASK], &litPtr, litEnd, base, vBase, dictEnd);
-            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
-            op += oneSeqSize;
-        }
-
-        /* save reps for next block */
-        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
-    }
-
-    /* last literal segment */
-    {   size_t const lastLLSize = litEnd - litPtr;
-        if (lastLLSize > (size_t)(oend-op)) return ERROR(dstSize_tooSmall);
-        memcpy(op, litPtr, lastLLSize);
-        op += lastLLSize;
-    }
-
-    return op-ostart;
-}
-
-
-static size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
-                            void* dst, size_t dstCapacity,
-                      const void* src, size_t srcSize, const int frame)
-{   /* blockType == blockCompressed */
-    const BYTE* ip = (const BYTE*)src;
-    /* isLongOffset must be true if there are long offsets.
-     * Offsets are long if they are larger than 2^STREAM_ACCUMULATOR_MIN.
-     * We don't expect that to be the case in 64-bit mode.
-     * If we are in block mode we don't know the window size, so we have to be
-     * conservative.
-     */
-    ZSTD_longOffset_e const isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (!frame || dctx->fParams.windowSize > (1ULL << STREAM_ACCUMULATOR_MIN)));
-    /* windowSize could be any value at this point, since it is only validated
-     * in the streaming API.
-     */
-    DEBUGLOG(5, "ZSTD_decompressBlock_internal");
-
-    if (srcSize >= ZSTD_BLOCKSIZE_MAX) return ERROR(srcSize_wrong);
-
-    /* Decode literals section */
-    {   size_t const litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize);
-        DEBUGLOG(5, "ZSTD_decodeLiteralsBlock : %u", (U32)litCSize);
-        if (ZSTD_isError(litCSize)) return litCSize;
-        ip += litCSize;
-        srcSize -= litCSize;
-    }
-    if (frame && dctx->fParams.windowSize > (1<<23))
-        return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize, isLongOffset);
-    return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, isLongOffset);
+    ZSTD_frameSizeInfo const frameSizeInfo = ZSTD_findFrameSizeInfo(src, srcSize);
+    return frameSizeInfo.compressedSize;
 }
 
-
-static void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst)
-{
-    if (dst != dctx->previousDstEnd) {   /* not contiguous */
-        dctx->dictEnd = dctx->previousDstEnd;
-        dctx->vBase = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
-        dctx->base = dst;
-        dctx->previousDstEnd = dst;
+/** ZSTD_decompressBound() :
+ *  compatible with legacy mode
+ *  `src` must point to the start of a ZSTD frame or a skippeable frame
+ *  `srcSize` must be at least as large as the frame contained
+ *  @return : the maximum decompressed size of the compressed source
+ */
+unsigned long long ZSTD_decompressBound(const void* src, size_t srcSize)
+{
+    unsigned long long bound = 0;
+    /* Iterate over each frame */
+    while (srcSize > 0) {
+        ZSTD_frameSizeInfo const frameSizeInfo = ZSTD_findFrameSizeInfo(src, srcSize);
+        size_t const compressedSize = frameSizeInfo.compressedSize;
+        unsigned long long const decompressedBound = frameSizeInfo.decompressedBound;
+        if (ZSTD_isError(compressedSize) || decompressedBound == ZSTD_CONTENTSIZE_ERROR)
+            return ZSTD_CONTENTSIZE_ERROR;
+        assert(srcSize >= compressedSize);
+        src = (const BYTE*)src + compressedSize;
+        srcSize -= compressedSize;
+        bound += decompressedBound;
     }
+    return bound;
 }
 
-size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx,
-                            void* dst, size_t dstCapacity,
-                      const void* src, size_t srcSize)
-{
-    size_t dSize;
-    ZSTD_checkContinuity(dctx, dst);
-    dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 0);
-    dctx->previousDstEnd = (char*)dst + dSize;
-    return dSize;
-}
 
+/*-*************************************************************
+ *   Frame decoding
+ ***************************************************************/
 
 /** ZSTD_insertBlock() :
-    insert `src` block into `dctx` history. Useful to track uncompressed blocks. */
-ZSTDLIB_API size_t ZSTD_insertBlock(ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize)
+ *  insert `src` block into `dctx` history. Useful to track uncompressed blocks. */
+size_t ZSTD_insertBlock(ZSTD_DCtx* dctx, const void* blockStart, size_t blockSize)
 {
+    DEBUGLOG(5, "ZSTD_insertBlock: %u bytes", (unsigned)blockSize);
     ZSTD_checkContinuity(dctx, blockStart);
     dctx->previousDstEnd = (const char*)blockStart + blockSize;
     return blockSize;
 }
 
 
-size_t ZSTD_generateNxBytes(void* dst, size_t dstCapacity, BYTE byte, size_t length)
+static size_t ZSTD_copyRawBlock(void* dst, size_t dstCapacity,
+                          const void* src, size_t srcSize)
 {
-    if (length > dstCapacity) return ERROR(dstSize_tooSmall);
-    memset(dst, byte, length);
-    return length;
+    DEBUGLOG(5, "ZSTD_copyRawBlock");
+    if (dst == NULL) {
+        if (srcSize == 0) return 0;
+        RETURN_ERROR(dstBuffer_null, "");
+    }
+    RETURN_ERROR_IF(srcSize > dstCapacity, dstSize_tooSmall, "");
+    memcpy(dst, src, srcSize);
+    return srcSize;
 }
 
-/** ZSTD_findFrameCompressedSize() :
- *  compatible with legacy mode
- *  `src` must point to the start of a ZSTD frame, ZSTD legacy frame, or skippable frame
- *  `srcSize` must be at least as large as the frame contained
- *  @return : the compressed size of the frame starting at `src` */
-size_t ZSTD_findFrameCompressedSize(const void *src, size_t srcSize)
+static size_t ZSTD_setRleBlock(void* dst, size_t dstCapacity,
+                               BYTE b,
+                               size_t regenSize)
 {
-#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
-    if (ZSTD_isLegacy(src, srcSize))
-        return ZSTD_findFrameCompressedSizeLegacy(src, srcSize);
-#endif
-    if ( (srcSize >= ZSTD_skippableHeaderSize)
-      && (MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START ) {
-        return ZSTD_skippableHeaderSize + MEM_readLE32((const BYTE*)src + 4);
-    } else {
-        const BYTE* ip = (const BYTE*)src;
-        const BYTE* const ipstart = ip;
-        size_t remainingSize = srcSize;
-        ZSTD_frameHeader zfh;
-
-        /* Extract Frame Header */
-        {   size_t const ret = ZSTD_getFrameHeader(&zfh, src, srcSize);
-            if (ZSTD_isError(ret)) return ret;
-            if (ret > 0) return ERROR(srcSize_wrong);
-        }
-
-        ip += zfh.headerSize;
-        remainingSize -= zfh.headerSize;
-
-        /* Loop on each block */
-        while (1) {
-            blockProperties_t blockProperties;
-            size_t const cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
-            if (ZSTD_isError(cBlockSize)) return cBlockSize;
-
-            if (ZSTD_blockHeaderSize + cBlockSize > remainingSize)
-                return ERROR(srcSize_wrong);
-
-            ip += ZSTD_blockHeaderSize + cBlockSize;
-            remainingSize -= ZSTD_blockHeaderSize + cBlockSize;
-
-            if (blockProperties.lastBlock) break;
-        }
-
-        if (zfh.checksumFlag) {   /* Final frame content checksum */
-            if (remainingSize < 4) return ERROR(srcSize_wrong);
-            ip += 4;
-            remainingSize -= 4;
-        }
-
-        return ip - ipstart;
+    if (dst == NULL) {
+        if (regenSize == 0) return 0;
+        RETURN_ERROR(dstBuffer_null, "");
     }
+    RETURN_ERROR_IF(regenSize > dstCapacity, dstSize_tooSmall, "");
+    memset(dst, b, regenSize);
+    return regenSize;
 }
 
+
 /*! ZSTD_decompressFrame() :
-*   @dctx must be properly initialized */
+ * @dctx must be properly initialized
+ *  will update *srcPtr and *srcSizePtr,
+ *  to make *srcPtr progress by one frame. */
 static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
                                    void* dst, size_t dstCapacity,
                              const void** srcPtr, size_t *srcSizePtr)
 {
     const BYTE* ip = (const BYTE*)(*srcPtr);
     BYTE* const ostart = (BYTE* const)dst;
-    BYTE* const oend = ostart + dstCapacity;
+    BYTE* const oend = dstCapacity != 0 ? ostart + dstCapacity : ostart;
     BYTE* op = ostart;
-    size_t remainingSize = *srcSizePtr;
+    size_t remainingSrcSize = *srcSizePtr;
+
+    DEBUGLOG(4, "ZSTD_decompressFrame (srcSize:%i)", (int)*srcSizePtr);
 
     /* check */
-    if (remainingSize < ZSTD_frameHeaderSize_min+ZSTD_blockHeaderSize)
-        return ERROR(srcSize_wrong);
+    RETURN_ERROR_IF(
+        remainingSrcSize < ZSTD_FRAMEHEADERSIZE_MIN(dctx->format)+ZSTD_blockHeaderSize,
+        srcSize_wrong, "");
 
     /* Frame Header */
-    {   size_t const frameHeaderSize = ZSTD_frameHeaderSize(ip, ZSTD_frameHeaderSize_prefix);
+    {   size_t const frameHeaderSize = ZSTD_frameHeaderSize_internal(
+                ip, ZSTD_FRAMEHEADERSIZE_PREFIX(dctx->format), dctx->format);
         if (ZSTD_isError(frameHeaderSize)) return frameHeaderSize;
-        if (remainingSize < frameHeaderSize+ZSTD_blockHeaderSize)
-            return ERROR(srcSize_wrong);
-        CHECK_F( ZSTD_decodeFrameHeader(dctx, ip, frameHeaderSize) );
-        ip += frameHeaderSize; remainingSize -= frameHeaderSize;
+        RETURN_ERROR_IF(remainingSrcSize < frameHeaderSize+ZSTD_blockHeaderSize,
+                        srcSize_wrong, "");
+        FORWARD_IF_ERROR( ZSTD_decodeFrameHeader(dctx, ip, frameHeaderSize) , "");
+        ip += frameHeaderSize; remainingSrcSize -= frameHeaderSize;
     }
 
     /* Loop on each block */
     while (1) {
         size_t decodedSize;
         blockProperties_t blockProperties;
-        size_t const cBlockSize = ZSTD_getcBlockSize(ip, remainingSize, &blockProperties);
+        size_t const cBlockSize = ZSTD_getcBlockSize(ip, remainingSrcSize, &blockProperties);
         if (ZSTD_isError(cBlockSize)) return cBlockSize;
 
         ip += ZSTD_blockHeaderSize;
-        remainingSize -= ZSTD_blockHeaderSize;
-        if (cBlockSize > remainingSize) return ERROR(srcSize_wrong);
+        remainingSrcSize -= ZSTD_blockHeaderSize;
+        RETURN_ERROR_IF(cBlockSize > remainingSrcSize, srcSize_wrong, "");
 
         switch(blockProperties.blockType)
         {
@@ -1530,41 +653,44 @@ static size_t ZSTD_decompressFrame(ZSTD_DCtx* dctx,
             decodedSize = ZSTD_copyRawBlock(op, oend-op, ip, cBlockSize);
             break;
         case bt_rle :
-            decodedSize = ZSTD_generateNxBytes(op, oend-op, *ip, blockProperties.origSize);
+            decodedSize = ZSTD_setRleBlock(op, oend-op, *ip, blockProperties.origSize);
             break;
         case bt_reserved :
         default:
-            return ERROR(corruption_detected);
+            RETURN_ERROR(corruption_detected, "invalid block type");
         }
 
         if (ZSTD_isError(decodedSize)) return decodedSize;
         if (dctx->fParams.checksumFlag)
             XXH64_update(&dctx->xxhState, op, decodedSize);
-        op += decodedSize;
+        if (decodedSize != 0)
+            op += decodedSize;
+        assert(ip != NULL);
         ip += cBlockSize;
-        remainingSize -= cBlockSize;
+        remainingSrcSize -= cBlockSize;
         if (blockProperties.lastBlock) break;
     }
 
+    if (dctx->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN) {
+        RETURN_ERROR_IF((U64)(op-ostart) != dctx->fParams.frameContentSize,
+                        corruption_detected, "");
+    }
     if (dctx->fParams.checksumFlag) { /* Frame content checksum verification */
         U32 const checkCalc = (U32)XXH64_digest(&dctx->xxhState);
         U32 checkRead;
-        if (remainingSize<4) return ERROR(checksum_wrong);
+        RETURN_ERROR_IF(remainingSrcSize<4, checksum_wrong, "");
         checkRead = MEM_readLE32(ip);
-        if (checkRead != checkCalc) return ERROR(checksum_wrong);
+        RETURN_ERROR_IF(checkRead != checkCalc, checksum_wrong, "");
         ip += 4;
-        remainingSize -= 4;
+        remainingSrcSize -= 4;
     }
 
     /* Allow caller to get size read */
     *srcPtr = ip;
-    *srcSizePtr = remainingSize;
+    *srcSizePtr = remainingSrcSize;
     return op-ostart;
 }
 
-static const void* ZSTD_DDictDictContent(const ZSTD_DDict* ddict);
-static size_t ZSTD_DDictDictSize(const ZSTD_DDict* ddict);
-
 static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
                                         void* dst, size_t dstCapacity,
                                   const void* src, size_t srcSize,
@@ -1572,26 +698,30 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
                                   const ZSTD_DDict* ddict)
 {
     void* const dststart = dst;
+    int moreThan1Frame = 0;
+
+    DEBUGLOG(5, "ZSTD_decompressMultiFrame");
     assert(dict==NULL || ddict==NULL);  /* either dict or ddict set, not both */
 
     if (ddict) {
-        dict = ZSTD_DDictDictContent(ddict);
-        dictSize = ZSTD_DDictDictSize(ddict);
+        dict = ZSTD_DDict_dictContent(ddict);
+        dictSize = ZSTD_DDict_dictSize(ddict);
     }
 
-    while (srcSize >= ZSTD_frameHeaderSize_prefix) {
-        U32 magicNumber;
+    while (srcSize >= ZSTD_startingInputLength(dctx->format)) {
 
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT >= 1)
         if (ZSTD_isLegacy(src, srcSize)) {
             size_t decodedSize;
             size_t const frameSize = ZSTD_findFrameCompressedSizeLegacy(src, srcSize);
             if (ZSTD_isError(frameSize)) return frameSize;
-            /* legacy support is not compatible with static dctx */
-            if (dctx->staticSize) return ERROR(memory_allocation);
+            RETURN_ERROR_IF(dctx->staticSize, memory_allocation,
+                "legacy support is not compatible with static dctx");
 
             decodedSize = ZSTD_decompressLegacy(dst, dstCapacity, src, frameSize, dict, dictSize);
+            if (ZSTD_isError(decodedSize)) return decodedSize;
 
+            assert(decodedSize <=- dstCapacity);
             dst = (BYTE*)dst + decodedSize;
             dstCapacity -= decodedSize;
 
@@ -1602,43 +732,54 @@ static size_t ZSTD_decompressMultiFrame(ZSTD_DCtx* dctx,
         }
 #endif
 
-        magicNumber = MEM_readLE32(src);
-        if (magicNumber != ZSTD_MAGICNUMBER) {
-            if ((magicNumber & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {
-                size_t skippableSize;
-                if (srcSize < ZSTD_skippableHeaderSize)
-                    return ERROR(srcSize_wrong);
-                skippableSize = MEM_readLE32((const BYTE *)src + 4) +
-                                ZSTD_skippableHeaderSize;
-                if (srcSize < skippableSize) return ERROR(srcSize_wrong);
+        {   U32 const magicNumber = MEM_readLE32(src);
+            DEBUGLOG(4, "reading magic number %08X (expecting %08X)",
+                        (unsigned)magicNumber, ZSTD_MAGICNUMBER);
+            if ((magicNumber & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {
+                size_t const skippableSize = readSkippableFrameSize(src, srcSize);
+                FORWARD_IF_ERROR(skippableSize, "readSkippableFrameSize failed");
+                assert(skippableSize <= srcSize);
 
                 src = (const BYTE *)src + skippableSize;
                 srcSize -= skippableSize;
                 continue;
-            }
-            return ERROR(prefix_unknown);
-        }
+        }   }
 
         if (ddict) {
             /* we were called from ZSTD_decompress_usingDDict */
-            CHECK_F(ZSTD_decompressBegin_usingDDict(dctx, ddict));
+            FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDDict(dctx, ddict), "");
         } else {
             /* this will initialize correctly with no dict if dict == NULL, so
              * use this in all cases but ddict */
-            CHECK_F(ZSTD_decompressBegin_usingDict(dctx, dict, dictSize));
+            FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDict(dctx, dict, dictSize), "");
         }
         ZSTD_checkContinuity(dctx, dst);
 
         {   const size_t res = ZSTD_decompressFrame(dctx, dst, dstCapacity,
                                                     &src, &srcSize);
+            RETURN_ERROR_IF(
+                (ZSTD_getErrorCode(res) == ZSTD_error_prefix_unknown)
+             && (moreThan1Frame==1),
+                srcSize_wrong,
+                "at least one frame successfully completed, but following "
+                "bytes are garbage: it's more likely to be a srcSize error, "
+                "specifying more bytes than compressed size of frame(s). This "
+                "error message replaces ERROR(prefix_unknown), which would be "
+                "confusing, as the first header is actually correct. Note that "
+                "one could be unlucky, it might be a corruption error instead, "
+                "happening right at the place where we expect zstd magic "
+                "bytes. But this is _much_ less likely than a srcSize field "
+                "error.");
             if (ZSTD_isError(res)) return res;
-            /* no need to bound check, ZSTD_decompressFrame already has */
-            dst = (BYTE*)dst + res;
+            assert(res <= dstCapacity);
+            if (res != 0)
+                dst = (BYTE*)dst + res;
             dstCapacity -= res;
         }
+        moreThan1Frame = 1;
     }  /* while (srcSize >= ZSTD_frameHeaderSize_prefix) */
 
-    if (srcSize) return ERROR(srcSize_wrong); /* input not entirely consumed */
+    RETURN_ERROR_IF(srcSize, srcSize_wrong, "input not entirely consumed");
 
     return (BYTE*)dst - (BYTE*)dststart;
 }
@@ -1652,9 +793,26 @@ size_t ZSTD_decompress_usingDict(ZSTD_DCtx* dctx,
 }
 
 
+static ZSTD_DDict const* ZSTD_getDDict(ZSTD_DCtx* dctx)
+{
+    switch (dctx->dictUses) {
+    default:
+        assert(0 /* Impossible */);
+        /* fall-through */
+    case ZSTD_dont_use:
+        ZSTD_clearDict(dctx);
+        return NULL;
+    case ZSTD_use_indefinitely:
+        return dctx->ddict;
+    case ZSTD_use_once:
+        dctx->dictUses = ZSTD_dont_use;
+        return dctx->ddict;
+    }
+}
+
 size_t ZSTD_decompressDCtx(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
 {
-    return ZSTD_decompress_usingDict(dctx, dst, dstCapacity, src, srcSize, NULL, 0);
+    return ZSTD_decompress_usingDDict(dctx, dst, dstCapacity, src, srcSize, ZSTD_getDDict(dctx));
 }
 
 
@@ -1663,12 +821,13 @@ size_t ZSTD_decompress(void* dst, size_t dstCapacity, const void* src, size_t sr
 #if defined(ZSTD_HEAPMODE) && (ZSTD_HEAPMODE>=1)
     size_t regenSize;
     ZSTD_DCtx* const dctx = ZSTD_createDCtx();
-    if (dctx==NULL) return ERROR(memory_allocation);
+    RETURN_ERROR_IF(dctx==NULL, memory_allocation, "NULL pointer!");
     regenSize = ZSTD_decompressDCtx(dctx, dst, dstCapacity, src, srcSize);
     ZSTD_freeDCtx(dctx);
     return regenSize;
 #else   /* stack mode */
     ZSTD_DCtx dctx;
+    ZSTD_initDCtx_internal(&dctx);
     return ZSTD_decompressDCtx(&dctx, dst, dstCapacity, src, srcSize);
 #endif
 }
@@ -1680,6 +839,24 @@ size_t ZSTD_decompress(void* dst, size_t dstCapacity, const void* src, size_t sr
 ****************************************/
 size_t ZSTD_nextSrcSizeToDecompress(ZSTD_DCtx* dctx) { return dctx->expected; }
 
+/**
+ * Similar to ZSTD_nextSrcSizeToDecompress(), but when when a block input can be streamed,
+ * we allow taking a partial block as the input. Currently only raw uncompressed blocks can
+ * be streamed.
+ *
+ * For blocks that can be streamed, this allows us to reduce the latency until we produce
+ * output, and avoid copying the input.
+ *
+ * @param inputSize - The total amount of input that the caller currently has.
+ */
+static size_t ZSTD_nextSrcSizeToDecompressWithInputSize(ZSTD_DCtx* dctx, size_t inputSize) {
+    if (!(dctx->stage == ZSTDds_decompressBlock || dctx->stage == ZSTDds_decompressLastBlock))
+        return dctx->expected;
+    if (dctx->bType != bt_raw)
+        return dctx->expected;
+    return MIN(MAX(inputSize, 1), dctx->expected);
+}
+
 ZSTD_nextInputType_e ZSTD_nextInputType(ZSTD_DCtx* dctx) {
     switch(dctx->stage)
     {
@@ -1710,36 +887,34 @@ static int ZSTD_isSkipFrame(ZSTD_DCtx* dctx) { return dctx->stage == ZSTDds_skip
  *            or an error code, which can be tested using ZSTD_isError() */
 size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, const void* src, size_t srcSize)
 {
-    DEBUGLOG(5, "ZSTD_decompressContinue");
+    DEBUGLOG(5, "ZSTD_decompressContinue (srcSize:%u)", (unsigned)srcSize);
     /* Sanity check */
-    if (srcSize != dctx->expected) return ERROR(srcSize_wrong);   /* unauthorized */
+    RETURN_ERROR_IF(srcSize != ZSTD_nextSrcSizeToDecompressWithInputSize(dctx, srcSize), srcSize_wrong, "not allowed");
     if (dstCapacity) ZSTD_checkContinuity(dctx, dst);
 
     switch (dctx->stage)
     {
     case ZSTDds_getFrameHeaderSize :
-        if (srcSize != ZSTD_frameHeaderSize_prefix) return ERROR(srcSize_wrong);      /* unauthorized */
         assert(src != NULL);
-        if ((MEM_readLE32(src) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {        /* skippable frame */
-            memcpy(dctx->headerBuffer, src, ZSTD_frameHeaderSize_prefix);
-            dctx->expected = ZSTD_skippableHeaderSize - ZSTD_frameHeaderSize_prefix;  /* magic number + skippable frame length */
-            dctx->stage = ZSTDds_decodeSkippableHeader;
-            return 0;
-        }
-        dctx->headerSize = ZSTD_frameHeaderSize(src, ZSTD_frameHeaderSize_prefix);
+        if (dctx->format == ZSTD_f_zstd1) {  /* allows header */
+            assert(srcSize >= ZSTD_FRAMEIDSIZE);  /* to read skippable magic number */
+            if ((MEM_readLE32(src) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {        /* skippable frame */
+                memcpy(dctx->headerBuffer, src, srcSize);
+                dctx->expected = ZSTD_SKIPPABLEHEADERSIZE - srcSize;  /* remaining to load to get full skippable frame header */
+                dctx->stage = ZSTDds_decodeSkippableHeader;
+                return 0;
+        }   }
+        dctx->headerSize = ZSTD_frameHeaderSize_internal(src, srcSize, dctx->format);
         if (ZSTD_isError(dctx->headerSize)) return dctx->headerSize;
-        memcpy(dctx->headerBuffer, src, ZSTD_frameHeaderSize_prefix);
-        if (dctx->headerSize > ZSTD_frameHeaderSize_prefix) {
-            dctx->expected = dctx->headerSize - ZSTD_frameHeaderSize_prefix;
-            dctx->stage = ZSTDds_decodeFrameHeader;
-            return 0;
-        }
-        dctx->expected = 0;   /* not necessary to copy more */
-        /* fall-through */
+        memcpy(dctx->headerBuffer, src, srcSize);
+        dctx->expected = dctx->headerSize - srcSize;
+        dctx->stage = ZSTDds_decodeFrameHeader;
+        return 0;
+
     case ZSTDds_decodeFrameHeader:
         assert(src != NULL);
-        memcpy(dctx->headerBuffer + ZSTD_frameHeaderSize_prefix, src, dctx->expected);
-        CHECK_F(ZSTD_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize));
+        memcpy(dctx->headerBuffer + (dctx->headerSize - srcSize), src, srcSize);
+        FORWARD_IF_ERROR(ZSTD_decodeFrameHeader(dctx, dctx->headerBuffer, dctx->headerSize), "");
         dctx->expected = ZSTD_blockHeaderSize;
         dctx->stage = ZSTDds_decodeBlockHeader;
         return 0;
@@ -1748,6 +923,7 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
         {   blockProperties_t bp;
             size_t const cBlockSize = ZSTD_getcBlockSize(src, ZSTD_blockHeaderSize, &bp);
             if (ZSTD_isError(cBlockSize)) return cBlockSize;
+            RETURN_ERROR_IF(cBlockSize > dctx->fParams.blockSizeMax, corruption_detected, "Block Size Exceeds Maximum");
             dctx->expected = cBlockSize;
             dctx->bType = bp.blockType;
             dctx->rleSize = bp.origSize;
@@ -1770,37 +946,51 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
             }
             return 0;
         }
+
     case ZSTDds_decompressLastBlock:
     case ZSTDds_decompressBlock:
-        DEBUGLOG(5, "case ZSTDds_decompressBlock");
+        DEBUGLOG(5, "ZSTD_decompressContinue: case ZSTDds_decompressBlock");
         {   size_t rSize;
             switch(dctx->bType)
             {
             case bt_compressed:
-                DEBUGLOG(5, "case bt_compressed");
+                DEBUGLOG(5, "ZSTD_decompressContinue: case bt_compressed");
                 rSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 1);
+                dctx->expected = 0;  /* Streaming not supported */
                 break;
             case bt_raw :
+                assert(srcSize <= dctx->expected);
                 rSize = ZSTD_copyRawBlock(dst, dstCapacity, src, srcSize);
+                FORWARD_IF_ERROR(rSize, "ZSTD_copyRawBlock failed");
+                assert(rSize == srcSize);
+                dctx->expected -= rSize;
                 break;
             case bt_rle :
-                rSize = ZSTD_setRleBlock(dst, dstCapacity, src, srcSize, dctx->rleSize);
+                rSize = ZSTD_setRleBlock(dst, dstCapacity, *(const BYTE*)src, dctx->rleSize);
+                dctx->expected = 0;  /* Streaming not supported */
                 break;
             case bt_reserved :   /* should never happen */
             default:
-                return ERROR(corruption_detected);
+                RETURN_ERROR(corruption_detected, "invalid block type");
             }
-            if (ZSTD_isError(rSize)) return rSize;
-            DEBUGLOG(5, "decoded size from block : %u", (U32)rSize);
+            FORWARD_IF_ERROR(rSize, "");
+            RETURN_ERROR_IF(rSize > dctx->fParams.blockSizeMax, corruption_detected, "Decompressed Block Size Exceeds Maximum");
+            DEBUGLOG(5, "ZSTD_decompressContinue: decoded size from block : %u", (unsigned)rSize);
             dctx->decodedSize += rSize;
             if (dctx->fParams.checksumFlag) XXH64_update(&dctx->xxhState, dst, rSize);
+            dctx->previousDstEnd = (char*)dst + rSize;
+
+            /* Stay on the same stage until we are finished streaming the block. */
+            if (dctx->expected > 0) {
+                return rSize;
+            }
 
             if (dctx->stage == ZSTDds_decompressLastBlock) {   /* end of frame */
-                DEBUGLOG(4, "decoded size from frame : %u", (U32)dctx->decodedSize);
-                if (dctx->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN) {
-                    if (dctx->decodedSize != dctx->fParams.frameContentSize) {
-                        return ERROR(corruption_detected);
-                }   }
+                DEBUGLOG(4, "ZSTD_decompressContinue: decoded size from frame : %u", (unsigned)dctx->decodedSize);
+                RETURN_ERROR_IF(
+                    dctx->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
+                 && dctx->decodedSize != dctx->fParams.frameContentSize,
+                    corruption_detected, "");
                 if (dctx->fParams.checksumFlag) {  /* another round for frame checksum */
                     dctx->expected = 4;
                     dctx->stage = ZSTDds_checkChecksum;
@@ -1811,35 +1001,37 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
             } else {
                 dctx->stage = ZSTDds_decodeBlockHeader;
                 dctx->expected = ZSTD_blockHeaderSize;
-                dctx->previousDstEnd = (char*)dst + rSize;
             }
             return rSize;
         }
+
     case ZSTDds_checkChecksum:
-        DEBUGLOG(4, "case ZSTDds_checkChecksum");
         assert(srcSize == 4);  /* guaranteed by dctx->expected */
         {   U32 const h32 = (U32)XXH64_digest(&dctx->xxhState);
             U32 const check32 = MEM_readLE32(src);
-            DEBUGLOG(4, "calculated %08X :: %08X read", h32, check32);
-            if (check32 != h32) return ERROR(checksum_wrong);
+            DEBUGLOG(4, "ZSTD_decompressContinue: checksum : calculated %08X :: %08X read", (unsigned)h32, (unsigned)check32);
+            RETURN_ERROR_IF(check32 != h32, checksum_wrong, "");
             dctx->expected = 0;
             dctx->stage = ZSTDds_getFrameHeaderSize;
             return 0;
         }
+
     case ZSTDds_decodeSkippableHeader:
-        {   assert(src != NULL);
-            memcpy(dctx->headerBuffer + ZSTD_frameHeaderSize_prefix, src, dctx->expected);
-            dctx->expected = MEM_readLE32(dctx->headerBuffer + 4);
-            dctx->stage = ZSTDds_skipFrame;
-            return 0;
-        }
+        assert(src != NULL);
+        assert(srcSize <= ZSTD_SKIPPABLEHEADERSIZE);
+        memcpy(dctx->headerBuffer + (ZSTD_SKIPPABLEHEADERSIZE - srcSize), src, srcSize);   /* complete skippable header */
+        dctx->expected = MEM_readLE32(dctx->headerBuffer + ZSTD_FRAMEIDSIZE);   /* note : dctx->expected can grow seriously large, beyond local buffer size */
+        dctx->stage = ZSTDds_skipFrame;
+        return 0;
+
     case ZSTDds_skipFrame:
-        {   dctx->expected = 0;
-            dctx->stage = ZSTDds_getFrameHeaderSize;
-            return 0;
-        }
+        dctx->expected = 0;
+        dctx->stage = ZSTDds_getFrameHeaderSize;
+        return 0;
+
     default:
-        return ERROR(GENERIC);   /* impossible */
+        assert(0);   /* impossible */
+        RETURN_ERROR(GENERIC, "impossible to reach");   /* some compiler require default to do something */
     }
 }
 
@@ -1847,64 +1039,95 @@ size_t ZSTD_decompressContinue(ZSTD_DCtx* dctx, void* dst, size_t dstCapacity, c
 static size_t ZSTD_refDictContent(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
 {
     dctx->dictEnd = dctx->previousDstEnd;
-    dctx->vBase = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->base));
-    dctx->base = dict;
+    dctx->virtualStart = (const char*)dict - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->prefixStart));
+    dctx->prefixStart = dict;
     dctx->previousDstEnd = (const char*)dict + dictSize;
+#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    dctx->dictContentBeginForFuzzing = dctx->prefixStart;
+    dctx->dictContentEndForFuzzing = dctx->previousDstEnd;
+#endif
     return 0;
 }
 
-/* ZSTD_loadEntropy() :
- * dict : must point at beginning of a valid zstd dictionary
+/*! ZSTD_loadDEntropy() :
+ *  dict : must point at beginning of a valid zstd dictionary.
  * @return : size of entropy tables read */
-static size_t ZSTD_loadEntropy(ZSTD_entropyDTables_t* entropy, const void* const dict, size_t const dictSize)
+size_t
+ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
+                  const void* const dict, size_t const dictSize)
 {
     const BYTE* dictPtr = (const BYTE*)dict;
     const BYTE* const dictEnd = dictPtr + dictSize;
 
-    if (dictSize <= 8) return ERROR(dictionary_corrupted);
+    RETURN_ERROR_IF(dictSize <= 8, dictionary_corrupted, "dict is too small");
+    assert(MEM_readLE32(dict) == ZSTD_MAGIC_DICTIONARY);   /* dict must be valid */
     dictPtr += 8;   /* skip header = magic + dictID */
 
-
-    {   size_t const hSize = HUF_readDTableX4_wksp(
-            entropy->hufTable, dictPtr, dictEnd - dictPtr,
-            entropy->workspace, sizeof(entropy->workspace));
-        if (HUF_isError(hSize)) return ERROR(dictionary_corrupted);
+    ZSTD_STATIC_ASSERT(offsetof(ZSTD_entropyDTables_t, OFTable) == offsetof(ZSTD_entropyDTables_t, LLTable) + sizeof(entropy->LLTable));
+    ZSTD_STATIC_ASSERT(offsetof(ZSTD_entropyDTables_t, MLTable) == offsetof(ZSTD_entropyDTables_t, OFTable) + sizeof(entropy->OFTable));
+    ZSTD_STATIC_ASSERT(sizeof(entropy->LLTable) + sizeof(entropy->OFTable) + sizeof(entropy->MLTable) >= HUF_DECOMPRESS_WORKSPACE_SIZE);
+    {   void* const workspace = &entropy->LLTable;   /* use fse tables as temporary workspace; implies fse tables are grouped together */
+        size_t const workspaceSize = sizeof(entropy->LLTable) + sizeof(entropy->OFTable) + sizeof(entropy->MLTable);
+#ifdef HUF_FORCE_DECOMPRESS_X1
+        /* in minimal huffman, we always use X1 variants */
+        size_t const hSize = HUF_readDTableX1_wksp(entropy->hufTable,
+                                                dictPtr, dictEnd - dictPtr,
+                                                workspace, workspaceSize);
+#else
+        size_t const hSize = HUF_readDTableX2_wksp(entropy->hufTable,
+                                                dictPtr, dictEnd - dictPtr,
+                                                workspace, workspaceSize);
+#endif
+        RETURN_ERROR_IF(HUF_isError(hSize), dictionary_corrupted, "");
         dictPtr += hSize;
     }
 
     {   short offcodeNCount[MaxOff+1];
-        U32 offcodeMaxValue = MaxOff, offcodeLog;
+        unsigned offcodeMaxValue = MaxOff, offcodeLog;
         size_t const offcodeHeaderSize = FSE_readNCount(offcodeNCount, &offcodeMaxValue, &offcodeLog, dictPtr, dictEnd-dictPtr);
-        if (FSE_isError(offcodeHeaderSize)) return ERROR(dictionary_corrupted);
-        if (offcodeLog > OffFSELog) return ERROR(dictionary_corrupted);
-        CHECK_E(FSE_buildDTable(entropy->OFTable, offcodeNCount, offcodeMaxValue, offcodeLog), dictionary_corrupted);
+        RETURN_ERROR_IF(FSE_isError(offcodeHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(offcodeMaxValue > MaxOff, dictionary_corrupted, "");
+        RETURN_ERROR_IF(offcodeLog > OffFSELog, dictionary_corrupted, "");
+        ZSTD_buildFSETable( entropy->OFTable,
+                            offcodeNCount, offcodeMaxValue,
+                            OF_base, OF_bits,
+                            offcodeLog);
         dictPtr += offcodeHeaderSize;
     }
 
     {   short matchlengthNCount[MaxML+1];
         unsigned matchlengthMaxValue = MaxML, matchlengthLog;
         size_t const matchlengthHeaderSize = FSE_readNCount(matchlengthNCount, &matchlengthMaxValue, &matchlengthLog, dictPtr, dictEnd-dictPtr);
-        if (FSE_isError(matchlengthHeaderSize)) return ERROR(dictionary_corrupted);
-        if (matchlengthLog > MLFSELog) return ERROR(dictionary_corrupted);
-        CHECK_E(FSE_buildDTable(entropy->MLTable, matchlengthNCount, matchlengthMaxValue, matchlengthLog), dictionary_corrupted);
+        RETURN_ERROR_IF(FSE_isError(matchlengthHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(matchlengthMaxValue > MaxML, dictionary_corrupted, "");
+        RETURN_ERROR_IF(matchlengthLog > MLFSELog, dictionary_corrupted, "");
+        ZSTD_buildFSETable( entropy->MLTable,
+                            matchlengthNCount, matchlengthMaxValue,
+                            ML_base, ML_bits,
+                            matchlengthLog);
         dictPtr += matchlengthHeaderSize;
     }
 
     {   short litlengthNCount[MaxLL+1];
         unsigned litlengthMaxValue = MaxLL, litlengthLog;
         size_t const litlengthHeaderSize = FSE_readNCount(litlengthNCount, &litlengthMaxValue, &litlengthLog, dictPtr, dictEnd-dictPtr);
-        if (FSE_isError(litlengthHeaderSize)) return ERROR(dictionary_corrupted);
-        if (litlengthLog > LLFSELog) return ERROR(dictionary_corrupted);
-        CHECK_E(FSE_buildDTable(entropy->LLTable, litlengthNCount, litlengthMaxValue, litlengthLog), dictionary_corrupted);
+        RETURN_ERROR_IF(FSE_isError(litlengthHeaderSize), dictionary_corrupted, "");
+        RETURN_ERROR_IF(litlengthMaxValue > MaxLL, dictionary_corrupted, "");
+        RETURN_ERROR_IF(litlengthLog > LLFSELog, dictionary_corrupted, "");
+        ZSTD_buildFSETable( entropy->LLTable,
+                            litlengthNCount, litlengthMaxValue,
+                            LL_base, LL_bits,
+                            litlengthLog);
         dictPtr += litlengthHeaderSize;
     }
 
-    if (dictPtr+12 > dictEnd) return ERROR(dictionary_corrupted);
+    RETURN_ERROR_IF(dictPtr+12 > dictEnd, dictionary_corrupted, "");
     {   int i;
         size_t const dictContentSize = (size_t)(dictEnd - (dictPtr+12));
         for (i=0; i<3; i++) {
             U32 const rep = MEM_readLE32(dictPtr); dictPtr += 4;
-            if (rep==0 || rep >= dictContentSize) return ERROR(dictionary_corrupted);
+            RETURN_ERROR_IF(rep==0 || rep > dictContentSize,
+                            dictionary_corrupted, "");
             entropy->rep[i] = rep;
     }   }
 
@@ -1918,11 +1141,11 @@ static size_t ZSTD_decompress_insertDictionary(ZSTD_DCtx* dctx, const void* dict
         if (magic != ZSTD_MAGIC_DICTIONARY) {
             return ZSTD_refDictContent(dctx, dict, dictSize);   /* pure content mode */
     }   }
-    dctx->dictID = MEM_readLE32((const char*)dict + 4);
+    dctx->dictID = MEM_readLE32((const char*)dict + ZSTD_FRAMEIDSIZE);
 
     /* load entropy tables */
-    {   size_t const eSize = ZSTD_loadEntropy(&dctx->entropy, dict, dictSize);
-        if (ZSTD_isError(eSize)) return ERROR(dictionary_corrupted);
+    {   size_t const eSize = ZSTD_loadDEntropy(&dctx->entropy, dict, dictSize);
+        RETURN_ERROR_IF(ZSTD_isError(eSize), dictionary_corrupted, "");
         dict = (const char*)dict + eSize;
         dictSize -= eSize;
     }
@@ -1932,183 +1155,59 @@ static size_t ZSTD_decompress_insertDictionary(ZSTD_DCtx* dctx, const void* dict
     return ZSTD_refDictContent(dctx, dict, dictSize);
 }
 
-size_t ZSTD_decompressBegin_usingDict(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
-{
-    CHECK_F( ZSTD_decompressBegin(dctx) );
-    if (dict && dictSize)
-        CHECK_E(ZSTD_decompress_insertDictionary(dctx, dict, dictSize), dictionary_corrupted);
-    return 0;
-}
-
-
-/* ======   ZSTD_DDict   ====== */
-
-struct ZSTD_DDict_s {
-    void* dictBuffer;
-    const void* dictContent;
-    size_t dictSize;
-    ZSTD_entropyDTables_t entropy;
-    U32 dictID;
-    U32 entropyPresent;
-    ZSTD_customMem cMem;
-};  /* typedef'd to ZSTD_DDict within "zstd.h" */
-
-static const void* ZSTD_DDictDictContent(const ZSTD_DDict* ddict)
-{
-    return ddict->dictContent;
-}
-
-static size_t ZSTD_DDictDictSize(const ZSTD_DDict* ddict)
-{
-    return ddict->dictSize;
-}
-
-size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dstDCtx, const ZSTD_DDict* ddict)
-{
-    CHECK_F( ZSTD_decompressBegin(dstDCtx) );
-    if (ddict) {   /* support begin on NULL */
-        dstDCtx->dictID = ddict->dictID;
-        dstDCtx->base = ddict->dictContent;
-        dstDCtx->vBase = ddict->dictContent;
-        dstDCtx->dictEnd = (const BYTE*)ddict->dictContent + ddict->dictSize;
-        dstDCtx->previousDstEnd = dstDCtx->dictEnd;
-        if (ddict->entropyPresent) {
-            dstDCtx->litEntropy = 1;
-            dstDCtx->fseEntropy = 1;
-            dstDCtx->LLTptr = ddict->entropy.LLTable;
-            dstDCtx->MLTptr = ddict->entropy.MLTable;
-            dstDCtx->OFTptr = ddict->entropy.OFTable;
-            dstDCtx->HUFptr = ddict->entropy.hufTable;
-            dstDCtx->entropy.rep[0] = ddict->entropy.rep[0];
-            dstDCtx->entropy.rep[1] = ddict->entropy.rep[1];
-            dstDCtx->entropy.rep[2] = ddict->entropy.rep[2];
-        } else {
-            dstDCtx->litEntropy = 0;
-            dstDCtx->fseEntropy = 0;
-        }
-    }
-    return 0;
-}
-
-static size_t ZSTD_loadEntropy_inDDict(ZSTD_DDict* ddict)
+size_t ZSTD_decompressBegin(ZSTD_DCtx* dctx)
 {
-    ddict->dictID = 0;
-    ddict->entropyPresent = 0;
-    if (ddict->dictSize < 8) return 0;
-    {   U32 const magic = MEM_readLE32(ddict->dictContent);
-        if (magic != ZSTD_MAGIC_DICTIONARY) return 0;   /* pure content mode */
-    }
-    ddict->dictID = MEM_readLE32((const char*)ddict->dictContent + 4);
-
-    /* load entropy tables */
-    CHECK_E( ZSTD_loadEntropy(&ddict->entropy, ddict->dictContent, ddict->dictSize), dictionary_corrupted );
-    ddict->entropyPresent = 1;
+    assert(dctx != NULL);
+    dctx->expected = ZSTD_startingInputLength(dctx->format);  /* dctx->format must be properly set */
+    dctx->stage = ZSTDds_getFrameHeaderSize;
+    dctx->decodedSize = 0;
+    dctx->previousDstEnd = NULL;
+    dctx->prefixStart = NULL;
+    dctx->virtualStart = NULL;
+    dctx->dictEnd = NULL;
+    dctx->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
+    dctx->litEntropy = dctx->fseEntropy = 0;
+    dctx->dictID = 0;
+    dctx->bType = bt_reserved;
+    ZSTD_STATIC_ASSERT(sizeof(dctx->entropy.rep) == sizeof(repStartValue));
+    memcpy(dctx->entropy.rep, repStartValue, sizeof(repStartValue));  /* initial repcodes */
+    dctx->LLTptr = dctx->entropy.LLTable;
+    dctx->MLTptr = dctx->entropy.MLTable;
+    dctx->OFTptr = dctx->entropy.OFTable;
+    dctx->HUFptr = dctx->entropy.hufTable;
     return 0;
 }
 
-
-static size_t ZSTD_initDDict_internal(ZSTD_DDict* ddict, const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod)
+size_t ZSTD_decompressBegin_usingDict(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
 {
-    if ((dictLoadMethod == ZSTD_dlm_byRef) || (!dict) || (!dictSize)) {
-        ddict->dictBuffer = NULL;
-        ddict->dictContent = dict;
-    } else {
-        void* const internalBuffer = ZSTD_malloc(dictSize, ddict->cMem);
-        ddict->dictBuffer = internalBuffer;
-        ddict->dictContent = internalBuffer;
-        if (!internalBuffer) return ERROR(memory_allocation);
-        memcpy(internalBuffer, dict, dictSize);
-    }
-    ddict->dictSize = dictSize;
-    ddict->entropy.hufTable[0] = (HUF_DTable)((HufLog)*0x1000001);  /* cover both little and big endian */
-
-    /* parse dictionary content */
-    CHECK_F( ZSTD_loadEntropy_inDDict(ddict) );
-
+    FORWARD_IF_ERROR( ZSTD_decompressBegin(dctx) , "");
+    if (dict && dictSize)
+        RETURN_ERROR_IF(
+            ZSTD_isError(ZSTD_decompress_insertDictionary(dctx, dict, dictSize)),
+            dictionary_corrupted, "");
     return 0;
 }
 
-ZSTD_DDict* ZSTD_createDDict_advanced(const void* dict, size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod, ZSTD_customMem customMem)
-{
-    if (!customMem.customAlloc ^ !customMem.customFree) return NULL;
-
-    {   ZSTD_DDict* const ddict = (ZSTD_DDict*) ZSTD_malloc(sizeof(ZSTD_DDict), customMem);
-        if (!ddict) return NULL;
-        ddict->cMem = customMem;
-
-        if (ZSTD_isError( ZSTD_initDDict_internal(ddict, dict, dictSize, dictLoadMethod) )) {
-            ZSTD_freeDDict(ddict);
-            return NULL;
-        }
-
-        return ddict;
-    }
-}
-
-/*! ZSTD_createDDict() :
-*   Create a digested dictionary, to start decompression without startup delay.
-*   `dict` content is copied inside DDict.
-*   Consequently, `dict` can be released after `ZSTD_DDict` creation */
-ZSTD_DDict* ZSTD_createDDict(const void* dict, size_t dictSize)
-{
-    ZSTD_customMem const allocator = { NULL, NULL, NULL };
-    return ZSTD_createDDict_advanced(dict, dictSize, ZSTD_dlm_byCopy, allocator);
-}
-
-/*! ZSTD_createDDict_byReference() :
- *  Create a digested dictionary, to start decompression without startup delay.
- *  Dictionary content is simply referenced, it will be accessed during decompression.
- *  Warning : dictBuffer must outlive DDict (DDict must be freed before dictBuffer) */
-ZSTD_DDict* ZSTD_createDDict_byReference(const void* dictBuffer, size_t dictSize)
-{
-    ZSTD_customMem const allocator = { NULL, NULL, NULL };
-    return ZSTD_createDDict_advanced(dictBuffer, dictSize, ZSTD_dlm_byRef, allocator);
-}
 
+/* ======   ZSTD_DDict   ====== */
 
-ZSTD_DDict* ZSTD_initStaticDDict(void* workspace, size_t workspaceSize,
-                                 const void* dict, size_t dictSize,
-                                 ZSTD_dictLoadMethod_e dictLoadMethod)
+size_t ZSTD_decompressBegin_usingDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
 {
-    size_t const neededSpace =
-            sizeof(ZSTD_DDict) + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
-    ZSTD_DDict* const ddict = (ZSTD_DDict*)workspace;
-    assert(workspace != NULL);
-    assert(dict != NULL);
-    if ((size_t)workspace & 7) return NULL;  /* 8-aligned */
-    if (workspaceSize < neededSpace) return NULL;
-    if (dictLoadMethod == ZSTD_dlm_byCopy) {
-        memcpy(ddict+1, dict, dictSize);  /* local copy */
-        dict = ddict+1;
+    DEBUGLOG(4, "ZSTD_decompressBegin_usingDDict");
+    assert(dctx != NULL);
+    if (ddict) {
+        const char* const dictStart = (const char*)ZSTD_DDict_dictContent(ddict);
+        size_t const dictSize = ZSTD_DDict_dictSize(ddict);
+        const void* const dictEnd = dictStart + dictSize;
+        dctx->ddictIsCold = (dctx->dictEnd != dictEnd);
+        DEBUGLOG(4, "DDict is %s",
+                    dctx->ddictIsCold ? "~cold~" : "hot!");
     }
-    if (ZSTD_isError( ZSTD_initDDict_internal(ddict, dict, dictSize, ZSTD_dlm_byRef) ))
-        return NULL;
-    return ddict;
-}
-
-
-size_t ZSTD_freeDDict(ZSTD_DDict* ddict)
-{
-    if (ddict==NULL) return 0;   /* support free on NULL */
-    {   ZSTD_customMem const cMem = ddict->cMem;
-        ZSTD_free(ddict->dictBuffer, cMem);
-        ZSTD_free(ddict, cMem);
-        return 0;
+    FORWARD_IF_ERROR( ZSTD_decompressBegin(dctx) , "");
+    if (ddict) {   /* NULL ddict is equivalent to no dictionary */
+        ZSTD_copyDDictParameters(dctx, ddict);
     }
-}
-
-/*! ZSTD_estimateDDictSize() :
- *  Estimate amount of memory that will be needed to create a dictionary for decompression.
- *  Note : dictionary created by reference using ZSTD_dlm_byRef are smaller */
-size_t ZSTD_estimateDDictSize(size_t dictSize, ZSTD_dictLoadMethod_e dictLoadMethod)
-{
-    return sizeof(ZSTD_DDict) + (dictLoadMethod == ZSTD_dlm_byRef ? 0 : dictSize);
-}
-
-size_t ZSTD_sizeof_DDict(const ZSTD_DDict* ddict)
-{
-    if (ddict==NULL) return 0;   /* support sizeof on NULL */
-    return sizeof(*ddict) + (ddict->dictBuffer ? ddict->dictSize : 0) ;
+    return 0;
 }
 
 /*! ZSTD_getDictID_fromDict() :
@@ -2119,21 +1218,11 @@ unsigned ZSTD_getDictID_fromDict(const void* dict, size_t dictSize)
 {
     if (dictSize < 8) return 0;
     if (MEM_readLE32(dict) != ZSTD_MAGIC_DICTIONARY) return 0;
-    return MEM_readLE32((const char*)dict + 4);
-}
-
-/*! ZSTD_getDictID_fromDDict() :
- *  Provides the dictID of the dictionary loaded into `ddict`.
- *  If @return == 0, the dictionary is not conformant to Zstandard specification, or empty.
- *  Non-conformant dictionaries can still be loaded, but as content-only dictionaries. */
-unsigned ZSTD_getDictID_fromDDict(const ZSTD_DDict* ddict)
-{
-    if (ddict==NULL) return 0;
-    return ZSTD_getDictID_fromDict(ddict->dictContent, ddict->dictSize);
+    return MEM_readLE32((const char*)dict + ZSTD_FRAMEIDSIZE);
 }
 
 /*! ZSTD_getDictID_fromFrame() :
- *  Provides the dictID required to decompresse frame stored within `src`.
+ *  Provides the dictID required to decompress frame stored within `src`.
  *  If @return == 0, the dictID could not be decoded.
  *  This could for one of the following reasons :
  *  - The frame does not require a dictionary (most common case).
@@ -2175,6 +1264,7 @@ size_t ZSTD_decompress_usingDDict(ZSTD_DCtx* dctx,
 
 ZSTD_DStream* ZSTD_createDStream(void)
 {
+    DEBUGLOG(3, "ZSTD_createDStream");
     return ZSTD_createDStream_advanced(ZSTD_defaultCMem);
 }
 
@@ -2194,64 +1284,201 @@ size_t ZSTD_freeDStream(ZSTD_DStream* zds)
 }
 
 
-/* *** Initialization *** */
+/* ***  Initialization  *** */
 
 size_t ZSTD_DStreamInSize(void)  { return ZSTD_BLOCKSIZE_MAX + ZSTD_blockHeaderSize; }
 size_t ZSTD_DStreamOutSize(void) { return ZSTD_BLOCKSIZE_MAX; }
 
+size_t ZSTD_DCtx_loadDictionary_advanced(ZSTD_DCtx* dctx,
+                                   const void* dict, size_t dictSize,
+                                         ZSTD_dictLoadMethod_e dictLoadMethod,
+                                         ZSTD_dictContentType_e dictContentType)
+{
+    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
+    ZSTD_clearDict(dctx);
+    if (dict && dictSize != 0) {
+        dctx->ddictLocal = ZSTD_createDDict_advanced(dict, dictSize, dictLoadMethod, dictContentType, dctx->customMem);
+        RETURN_ERROR_IF(dctx->ddictLocal == NULL, memory_allocation, "NULL pointer!");
+        dctx->ddict = dctx->ddictLocal;
+        dctx->dictUses = ZSTD_use_indefinitely;
+    }
+    return 0;
+}
+
+size_t ZSTD_DCtx_loadDictionary_byReference(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_DCtx_loadDictionary_advanced(dctx, dict, dictSize, ZSTD_dlm_byRef, ZSTD_dct_auto);
+}
+
+size_t ZSTD_DCtx_loadDictionary(ZSTD_DCtx* dctx, const void* dict, size_t dictSize)
+{
+    return ZSTD_DCtx_loadDictionary_advanced(dctx, dict, dictSize, ZSTD_dlm_byCopy, ZSTD_dct_auto);
+}
+
+size_t ZSTD_DCtx_refPrefix_advanced(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize, ZSTD_dictContentType_e dictContentType)
+{
+    FORWARD_IF_ERROR(ZSTD_DCtx_loadDictionary_advanced(dctx, prefix, prefixSize, ZSTD_dlm_byRef, dictContentType), "");
+    dctx->dictUses = ZSTD_use_once;
+    return 0;
+}
+
+size_t ZSTD_DCtx_refPrefix(ZSTD_DCtx* dctx, const void* prefix, size_t prefixSize)
+{
+    return ZSTD_DCtx_refPrefix_advanced(dctx, prefix, prefixSize, ZSTD_dct_rawContent);
+}
+
+
+/* ZSTD_initDStream_usingDict() :
+ * return : expected size, aka ZSTD_startingInputLength().
+ * this function cannot fail */
 size_t ZSTD_initDStream_usingDict(ZSTD_DStream* zds, const void* dict, size_t dictSize)
 {
-    zds->streamStage = zdss_loadHeader;
-    zds->lhSize = zds->inPos = zds->outStart = zds->outEnd = 0;
-    ZSTD_freeDDict(zds->ddictLocal);
-    if (dict && dictSize >= 8) {
-        zds->ddictLocal = ZSTD_createDDict(dict, dictSize);
-        if (zds->ddictLocal == NULL) return ERROR(memory_allocation);
-    } else zds->ddictLocal = NULL;
-    zds->ddict = zds->ddictLocal;
-    zds->legacyVersion = 0;
-    zds->hostageByte = 0;
-    return ZSTD_frameHeaderSize_prefix;
+    DEBUGLOG(4, "ZSTD_initDStream_usingDict");
+    FORWARD_IF_ERROR( ZSTD_DCtx_reset(zds, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_DCtx_loadDictionary(zds, dict, dictSize) , "");
+    return ZSTD_startingInputLength(zds->format);
 }
 
+/* note : this variant can't fail */
 size_t ZSTD_initDStream(ZSTD_DStream* zds)
 {
-    return ZSTD_initDStream_usingDict(zds, NULL, 0);
+    DEBUGLOG(4, "ZSTD_initDStream");
+    return ZSTD_initDStream_usingDDict(zds, NULL);
 }
 
 /* ZSTD_initDStream_usingDDict() :
- * ddict will just be referenced, and must outlive decompression session */
-size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* zds, const ZSTD_DDict* ddict)
+ * ddict will just be referenced, and must outlive decompression session
+ * this function cannot fail */
+size_t ZSTD_initDStream_usingDDict(ZSTD_DStream* dctx, const ZSTD_DDict* ddict)
 {
-    size_t const initResult = ZSTD_initDStream(zds);
-    zds->ddict = ddict;
-    return initResult;
+    FORWARD_IF_ERROR( ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only) , "");
+    FORWARD_IF_ERROR( ZSTD_DCtx_refDDict(dctx, ddict) , "");
+    return ZSTD_startingInputLength(dctx->format);
 }
 
-size_t ZSTD_resetDStream(ZSTD_DStream* zds)
+/* ZSTD_resetDStream() :
+ * return : expected size, aka ZSTD_startingInputLength().
+ * this function cannot fail */
+size_t ZSTD_resetDStream(ZSTD_DStream* dctx)
 {
-    zds->streamStage = zdss_loadHeader;
-    zds->lhSize = zds->inPos = zds->outStart = zds->outEnd = 0;
-    zds->legacyVersion = 0;
-    zds->hostageByte = 0;
-    return ZSTD_frameHeaderSize_prefix;
+    FORWARD_IF_ERROR(ZSTD_DCtx_reset(dctx, ZSTD_reset_session_only), "");
+    return ZSTD_startingInputLength(dctx->format);
 }
 
-size_t ZSTD_setDStreamParameter(ZSTD_DStream* zds,
-                                ZSTD_DStreamParameter_e paramType, unsigned paramValue)
+
+size_t ZSTD_DCtx_refDDict(ZSTD_DCtx* dctx, const ZSTD_DDict* ddict)
 {
-    switch(paramType)
-    {
-        default : return ERROR(parameter_unsupported);
-        case DStream_p_maxWindowSize : zds->maxWindowSize = paramValue ? paramValue : (U32)(-1); break;
+    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
+    ZSTD_clearDict(dctx);
+    if (ddict) {
+        dctx->ddict = ddict;
+        dctx->dictUses = ZSTD_use_indefinitely;
+    }
+    return 0;
+}
+
+/* ZSTD_DCtx_setMaxWindowSize() :
+ * note : no direct equivalence in ZSTD_DCtx_setParameter,
+ * since this version sets windowSize, and the other sets windowLog */
+size_t ZSTD_DCtx_setMaxWindowSize(ZSTD_DCtx* dctx, size_t maxWindowSize)
+{
+    ZSTD_bounds const bounds = ZSTD_dParam_getBounds(ZSTD_d_windowLogMax);
+    size_t const min = (size_t)1 << bounds.lowerBound;
+    size_t const max = (size_t)1 << bounds.upperBound;
+    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
+    RETURN_ERROR_IF(maxWindowSize < min, parameter_outOfBound, "");
+    RETURN_ERROR_IF(maxWindowSize > max, parameter_outOfBound, "");
+    dctx->maxWindowSize = maxWindowSize;
+    return 0;
+}
+
+size_t ZSTD_DCtx_setFormat(ZSTD_DCtx* dctx, ZSTD_format_e format)
+{
+    return ZSTD_DCtx_setParameter(dctx, ZSTD_d_format, format);
+}
+
+ZSTD_bounds ZSTD_dParam_getBounds(ZSTD_dParameter dParam)
+{
+    ZSTD_bounds bounds = { 0, 0, 0 };
+    switch(dParam) {
+        case ZSTD_d_windowLogMax:
+            bounds.lowerBound = ZSTD_WINDOWLOG_ABSOLUTEMIN;
+            bounds.upperBound = ZSTD_WINDOWLOG_MAX;
+            return bounds;
+        case ZSTD_d_format:
+            bounds.lowerBound = (int)ZSTD_f_zstd1;
+            bounds.upperBound = (int)ZSTD_f_zstd1_magicless;
+            ZSTD_STATIC_ASSERT(ZSTD_f_zstd1 < ZSTD_f_zstd1_magicless);
+            return bounds;
+        case ZSTD_d_stableOutBuffer:
+            bounds.lowerBound = (int)ZSTD_obm_buffered;
+            bounds.upperBound = (int)ZSTD_obm_stable;
+            return bounds;
+        default:;
+    }
+    bounds.error = ERROR(parameter_unsupported);
+    return bounds;
+}
+
+/* ZSTD_dParam_withinBounds:
+ * @return 1 if value is within dParam bounds,
+ * 0 otherwise */
+static int ZSTD_dParam_withinBounds(ZSTD_dParameter dParam, int value)
+{
+    ZSTD_bounds const bounds = ZSTD_dParam_getBounds(dParam);
+    if (ZSTD_isError(bounds.error)) return 0;
+    if (value < bounds.lowerBound) return 0;
+    if (value > bounds.upperBound) return 0;
+    return 1;
+}
+
+#define CHECK_DBOUNDS(p,v) {                \
+    RETURN_ERROR_IF(!ZSTD_dParam_withinBounds(p, v), parameter_outOfBound, ""); \
+}
+
+size_t ZSTD_DCtx_setParameter(ZSTD_DCtx* dctx, ZSTD_dParameter dParam, int value)
+{
+    RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
+    switch(dParam) {
+        case ZSTD_d_windowLogMax:
+            if (value == 0) value = ZSTD_WINDOWLOG_LIMIT_DEFAULT;
+            CHECK_DBOUNDS(ZSTD_d_windowLogMax, value);
+            dctx->maxWindowSize = ((size_t)1) << value;
+            return 0;
+        case ZSTD_d_format:
+            CHECK_DBOUNDS(ZSTD_d_format, value);
+            dctx->format = (ZSTD_format_e)value;
+            return 0;
+        case ZSTD_d_stableOutBuffer:
+            CHECK_DBOUNDS(ZSTD_d_stableOutBuffer, value);
+            dctx->outBufferMode = (ZSTD_outBufferMode_e)value;
+            return 0;
+        default:;
+    }
+    RETURN_ERROR(parameter_unsupported, "");
+}
+
+size_t ZSTD_DCtx_reset(ZSTD_DCtx* dctx, ZSTD_ResetDirective reset)
+{
+    if ( (reset == ZSTD_reset_session_only)
+      || (reset == ZSTD_reset_session_and_parameters) ) {
+        dctx->streamStage = zdss_init;
+        dctx->noForwardProgress = 0;
+    }
+    if ( (reset == ZSTD_reset_parameters)
+      || (reset == ZSTD_reset_session_and_parameters) ) {
+        RETURN_ERROR_IF(dctx->streamStage != zdss_init, stage_wrong, "");
+        ZSTD_clearDict(dctx);
+        dctx->format = ZSTD_f_zstd1;
+        dctx->maxWindowSize = ZSTD_MAXWINDOWSIZE_DEFAULT;
     }
     return 0;
 }
 
 
-size_t ZSTD_sizeof_DStream(const ZSTD_DStream* zds)
+size_t ZSTD_sizeof_DStream(const ZSTD_DStream* dctx)
 {
-    return ZSTD_sizeof_DCtx(zds);
+    return ZSTD_sizeof_DCtx(dctx);
 }
 
 size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long long frameContentSize)
@@ -2260,7 +1487,8 @@ size_t ZSTD_decodingBufferSize_min(unsigned long long windowSize, unsigned long
     unsigned long long const neededRBSize = windowSize + blockSize + (WILDCOPY_OVERLENGTH * 2);
     unsigned long long const neededSize = MIN(frameContentSize, neededRBSize);
     size_t const minRBSize = (size_t) neededSize;
-    if ((unsigned long long)minRBSize != neededSize) return ERROR(frameParameter_windowTooLarge);
+    RETURN_ERROR_IF((unsigned long long)minRBSize != neededSize,
+                    frameParameter_windowTooLarge, "");
     return minRBSize;
 }
 
@@ -2272,84 +1500,174 @@ size_t ZSTD_estimateDStreamSize(size_t windowSize)
     return ZSTD_estimateDCtxSize() + inBuffSize + outBuffSize;
 }
 
-ZSTDLIB_API size_t ZSTD_estimateDStreamSize_fromFrame(const void* src, size_t srcSize)
+size_t ZSTD_estimateDStreamSize_fromFrame(const void* src, size_t srcSize)
 {
-    U32 const windowSizeMax = 1U << ZSTD_WINDOWLOG_MAX;   /* note : should be user-selectable */
+    U32 const windowSizeMax = 1U << ZSTD_WINDOWLOG_MAX;   /* note : should be user-selectable, but requires an additional parameter (or a dctx) */
     ZSTD_frameHeader zfh;
     size_t const err = ZSTD_getFrameHeader(&zfh, src, srcSize);
     if (ZSTD_isError(err)) return err;
-    if (err>0) return ERROR(srcSize_wrong);
-    if (zfh.windowSize > windowSizeMax)
-        return ERROR(frameParameter_windowTooLarge);
+    RETURN_ERROR_IF(err>0, srcSize_wrong, "");
+    RETURN_ERROR_IF(zfh.windowSize > windowSizeMax,
+                    frameParameter_windowTooLarge, "");
     return ZSTD_estimateDStreamSize((size_t)zfh.windowSize);
 }
 
 
 /* *****   Decompression   ***** */
 
-MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
+static int ZSTD_DCtx_isOverflow(ZSTD_DStream* zds, size_t const neededInBuffSize, size_t const neededOutBuffSize)
 {
-    size_t const length = MIN(dstCapacity, srcSize);
-    memcpy(dst, src, length);
-    return length;
+    return (zds->inBuffSize + zds->outBuffSize) >= (neededInBuffSize + neededOutBuffSize) * ZSTD_WORKSPACETOOLARGE_FACTOR;
 }
 
+static void ZSTD_DCtx_updateOversizedDuration(ZSTD_DStream* zds, size_t const neededInBuffSize, size_t const neededOutBuffSize)
+{
+    if (ZSTD_DCtx_isOverflow(zds, neededInBuffSize, neededOutBuffSize))
+        zds->oversizedDuration++;
+    else
+        zds->oversizedDuration = 0;
+}
+
+static int ZSTD_DCtx_isOversizedTooLong(ZSTD_DStream* zds)
+{
+    return zds->oversizedDuration >= ZSTD_WORKSPACETOOLARGE_MAXDURATION;
+}
+
+/* Checks that the output buffer hasn't changed if ZSTD_obm_stable is used. */
+static size_t ZSTD_checkOutBuffer(ZSTD_DStream const* zds, ZSTD_outBuffer const* output)
+{
+    ZSTD_outBuffer const expect = zds->expectedOutBuffer;
+    /* No requirement when ZSTD_obm_stable is not enabled. */
+    if (zds->outBufferMode != ZSTD_obm_stable)
+        return 0;
+    /* Any buffer is allowed in zdss_init, this must be the same for every other call until
+     * the context is reset.
+     */
+    if (zds->streamStage == zdss_init)
+        return 0;
+    /* The buffer must match our expectation exactly. */
+    if (expect.dst == output->dst && expect.pos == output->pos && expect.size == output->size)
+        return 0;
+    RETURN_ERROR(dstBuffer_wrong, "ZSTD_obm_stable enabled but output differs!");
+}
+
+/* Calls ZSTD_decompressContinue() with the right parameters for ZSTD_decompressStream()
+ * and updates the stage and the output buffer state. This call is extracted so it can be
+ * used both when reading directly from the ZSTD_inBuffer, and in buffered input mode.
+ * NOTE: You must break after calling this function since the streamStage is modified.
+ */
+static size_t ZSTD_decompressContinueStream(
+            ZSTD_DStream* zds, char** op, char* oend,
+            void const* src, size_t srcSize) {
+    int const isSkipFrame = ZSTD_isSkipFrame(zds);
+    if (zds->outBufferMode == ZSTD_obm_buffered) {
+        size_t const dstSize = isSkipFrame ? 0 : zds->outBuffSize - zds->outStart;
+        size_t const decodedSize = ZSTD_decompressContinue(zds,
+                zds->outBuff + zds->outStart, dstSize, src, srcSize);
+        FORWARD_IF_ERROR(decodedSize, "");
+        if (!decodedSize && !isSkipFrame) {
+            zds->streamStage = zdss_read;
+        } else {
+            zds->outEnd = zds->outStart + decodedSize;
+            zds->streamStage = zdss_flush;
+        }
+    } else {
+        /* Write directly into the output buffer */
+        size_t const dstSize = isSkipFrame ? 0 : oend - *op;
+        size_t const decodedSize = ZSTD_decompressContinue(zds, *op, dstSize, src, srcSize);
+        FORWARD_IF_ERROR(decodedSize, "");
+        *op += decodedSize;
+        /* Flushing is not needed. */
+        zds->streamStage = zdss_read;
+        assert(*op <= oend);
+        assert(zds->outBufferMode == ZSTD_obm_stable);
+    }
+    return 0;
+}
 
 size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
 {
-    const char* const istart = (const char*)(input->src) + input->pos;
-    const char* const iend = (const char*)(input->src) + input->size;
+    const char* const src = (const char*)input->src;
+    const char* const istart = input->pos != 0 ? src + input->pos : src;
+    const char* const iend = input->size != 0 ? src + input->size : src;
     const char* ip = istart;
-    char* const ostart = (char*)(output->dst) + output->pos;
-    char* const oend = (char*)(output->dst) + output->size;
+    char* const dst = (char*)output->dst;
+    char* const ostart = output->pos != 0 ? dst + output->pos : dst;
+    char* const oend = output->size != 0 ? dst + output->size : dst;
     char* op = ostart;
     U32 someMoreWork = 1;
 
     DEBUGLOG(5, "ZSTD_decompressStream");
+    RETURN_ERROR_IF(
+        input->pos > input->size,
+        srcSize_wrong,
+        "forbidden. in: pos: %u   vs size: %u",
+        (U32)input->pos, (U32)input->size);
+    RETURN_ERROR_IF(
+        output->pos > output->size,
+        dstSize_tooSmall,
+        "forbidden. out: pos: %u   vs size: %u",
+        (U32)output->pos, (U32)output->size);
     DEBUGLOG(5, "input size : %u", (U32)(input->size - input->pos));
-#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
-    if (zds->legacyVersion) {
-        /* legacy support is incompatible with static dctx */
-        if (zds->staticSize) return ERROR(memory_allocation);
-        return ZSTD_decompressLegacyStream(zds->legacyContext, zds->legacyVersion, output, input);
-    }
-#endif
+    FORWARD_IF_ERROR(ZSTD_checkOutBuffer(zds, output), "");
 
     while (someMoreWork) {
         switch(zds->streamStage)
         {
         case zdss_init :
-            ZSTD_resetDStream(zds);   /* transparent reset on starting decoding a new frame */
+            DEBUGLOG(5, "stage zdss_init => transparent reset ");
+            zds->streamStage = zdss_loadHeader;
+            zds->lhSize = zds->inPos = zds->outStart = zds->outEnd = 0;
+            zds->legacyVersion = 0;
+            zds->hostageByte = 0;
+            zds->expectedOutBuffer = *output;
             /* fall-through */
 
         case zdss_loadHeader :
-            {   size_t const hSize = ZSTD_getFrameHeader(&zds->fParams, zds->headerBuffer, zds->lhSize);
+            DEBUGLOG(5, "stage zdss_loadHeader (srcSize : %u)", (U32)(iend - ip));
+#if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
+            if (zds->legacyVersion) {
+                RETURN_ERROR_IF(zds->staticSize, memory_allocation,
+                    "legacy support is incompatible with static dctx");
+                {   size_t const hint = ZSTD_decompressLegacyStream(zds->legacyContext, zds->legacyVersion, output, input);
+                    if (hint==0) zds->streamStage = zdss_init;
+                    return hint;
+            }   }
+#endif
+            {   size_t const hSize = ZSTD_getFrameHeader_advanced(&zds->fParams, zds->headerBuffer, zds->lhSize, zds->format);
+                DEBUGLOG(5, "header size : %u", (U32)hSize);
                 if (ZSTD_isError(hSize)) {
 #if defined(ZSTD_LEGACY_SUPPORT) && (ZSTD_LEGACY_SUPPORT>=1)
                     U32 const legacyVersion = ZSTD_isLegacy(istart, iend-istart);
                     if (legacyVersion) {
-                        const void* const dict = zds->ddict ? zds->ddict->dictContent : NULL;
-                        size_t const dictSize = zds->ddict ? zds->ddict->dictSize : 0;
-                        /* legacy support is incompatible with static dctx */
-                        if (zds->staticSize) return ERROR(memory_allocation);
-                        CHECK_F(ZSTD_initLegacyStream(&zds->legacyContext,
+                        ZSTD_DDict const* const ddict = ZSTD_getDDict(zds);
+                        const void* const dict = ddict ? ZSTD_DDict_dictContent(ddict) : NULL;
+                        size_t const dictSize = ddict ? ZSTD_DDict_dictSize(ddict) : 0;
+                        DEBUGLOG(5, "ZSTD_decompressStream: detected legacy version v0.%u", legacyVersion);
+                        RETURN_ERROR_IF(zds->staticSize, memory_allocation,
+                            "legacy support is incompatible with static dctx");
+                        FORWARD_IF_ERROR(ZSTD_initLegacyStream(&zds->legacyContext,
                                     zds->previousLegacyVersion, legacyVersion,
-                                    dict, dictSize));
+                                    dict, dictSize), "");
                         zds->legacyVersion = zds->previousLegacyVersion = legacyVersion;
-                        return ZSTD_decompressLegacyStream(zds->legacyContext, legacyVersion, output, input);
-                    }
+                        {   size_t const hint = ZSTD_decompressLegacyStream(zds->legacyContext, legacyVersion, output, input);
+                            if (hint==0) zds->streamStage = zdss_init;   /* or stay in stage zdss_loadHeader */
+                            return hint;
+                    }   }
 #endif
-                    return hSize; /* error */
+                    return hSize;   /* error */
                 }
                 if (hSize != 0) {   /* need more input */
                     size_t const toLoad = hSize - zds->lhSize;   /* if hSize!=0, hSize > zds->lhSize */
-                    if (toLoad > (size_t)(iend-ip)) {   /* not enough input to load full header */
-                        if (iend-ip > 0) {
-                            memcpy(zds->headerBuffer + zds->lhSize, ip, iend-ip);
-                            zds->lhSize += iend-ip;
+                    size_t const remainingInput = (size_t)(iend-ip);
+                    assert(iend >= ip);
+                    if (toLoad > remainingInput) {   /* not enough input to load full header */
+                        if (remainingInput > 0) {
+                            memcpy(zds->headerBuffer + zds->lhSize, ip, remainingInput);
+                            zds->lhSize += remainingInput;
                         }
                         input->pos = input->size;
-                        return (MAX(ZSTD_frameHeaderSize_min, hSize) - zds->lhSize) + ZSTD_blockHeaderSize;   /* remaining header bytes + next block header */
+                        return (MAX((size_t)ZSTD_FRAMEHEADERSIZE_MIN(zds->format), hSize) - zds->lhSize) + ZSTD_blockHeaderSize;   /* remaining header bytes + next block header */
                     }
                     assert(ip != NULL);
                     memcpy(zds->headerBuffer + zds->lhSize, ip, toLoad); zds->lhSize = hSize; ip += toLoad;
@@ -2357,12 +1675,15 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
             }   }
 
             /* check for single-pass mode opportunity */
-            if (zds->fParams.frameContentSize && zds->fParams.windowSize /* skippable frame if == 0 */
+            if (zds->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
+                && zds->fParams.frameType != ZSTD_skippableFrame
                 && (U64)(size_t)(oend-op) >= zds->fParams.frameContentSize) {
                 size_t const cSize = ZSTD_findFrameCompressedSize(istart, iend-istart);
                 if (cSize <= (size_t)(iend-istart)) {
-                    size_t const decompressedSize = ZSTD_decompress_usingDDict(zds, op, oend-op, istart, cSize, zds->ddict);
+                    /* shortcut : using single-pass mode */
+                    size_t const decompressedSize = ZSTD_decompress_usingDDict(zds, op, oend-op, istart, cSize, ZSTD_getDDict(zds));
                     if (ZSTD_isError(decompressedSize)) return decompressedSize;
+                    DEBUGLOG(4, "shortcut to single-pass ZSTD_decompress_usingDDict()")
                     ip = istart + cSize;
                     op += decompressedSize;
                     zds->expected = 0;
@@ -2371,55 +1692,75 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                     break;
             }   }
 
+            /* Check output buffer is large enough for ZSTD_odm_stable. */
+            if (zds->outBufferMode == ZSTD_obm_stable
+                && zds->fParams.frameType != ZSTD_skippableFrame
+                && zds->fParams.frameContentSize != ZSTD_CONTENTSIZE_UNKNOWN
+                && (U64)(size_t)(oend-op) < zds->fParams.frameContentSize) {
+                RETURN_ERROR(dstSize_tooSmall, "ZSTD_obm_stable passed but ZSTD_outBuffer is too small");
+            }
+
             /* Consume header (see ZSTDds_decodeFrameHeader) */
             DEBUGLOG(4, "Consume header");
-            CHECK_F(ZSTD_decompressBegin_usingDDict(zds, zds->ddict));
+            FORWARD_IF_ERROR(ZSTD_decompressBegin_usingDDict(zds, ZSTD_getDDict(zds)), "");
 
-            if ((MEM_readLE32(zds->headerBuffer) & 0xFFFFFFF0U) == ZSTD_MAGIC_SKIPPABLE_START) {  /* skippable frame */
-                zds->expected = MEM_readLE32(zds->headerBuffer + 4);
+            if ((MEM_readLE32(zds->headerBuffer) & ZSTD_MAGIC_SKIPPABLE_MASK) == ZSTD_MAGIC_SKIPPABLE_START) {  /* skippable frame */
+                zds->expected = MEM_readLE32(zds->headerBuffer + ZSTD_FRAMEIDSIZE);
                 zds->stage = ZSTDds_skipFrame;
             } else {
-                CHECK_F(ZSTD_decodeFrameHeader(zds, zds->headerBuffer, zds->lhSize));
+                FORWARD_IF_ERROR(ZSTD_decodeFrameHeader(zds, zds->headerBuffer, zds->lhSize), "");
                 zds->expected = ZSTD_blockHeaderSize;
                 zds->stage = ZSTDds_decodeBlockHeader;
             }
 
             /* control buffer memory usage */
-            DEBUGLOG(4, "Control max buffer memory usage");
+            DEBUGLOG(4, "Control max memory usage (%u KB <= max %u KB)",
+                        (U32)(zds->fParams.windowSize >>10),
+                        (U32)(zds->maxWindowSize >> 10) );
             zds->fParams.windowSize = MAX(zds->fParams.windowSize, 1U << ZSTD_WINDOWLOG_ABSOLUTEMIN);
-            if (zds->fParams.windowSize > zds->maxWindowSize) return ERROR(frameParameter_windowTooLarge);
+            RETURN_ERROR_IF(zds->fParams.windowSize > zds->maxWindowSize,
+                            frameParameter_windowTooLarge, "");
 
             /* Adapt buffer sizes to frame header instructions */
             {   size_t const neededInBuffSize = MAX(zds->fParams.blockSizeMax, 4 /* frame checksum */);
-                size_t const neededOutBuffSize = ZSTD_decodingBufferSize_min(zds->fParams.windowSize, zds->fParams.frameContentSize);
-                if ((zds->inBuffSize < neededInBuffSize) || (zds->outBuffSize < neededOutBuffSize)) {
-                    size_t const bufferSize = neededInBuffSize + neededOutBuffSize;
-                    DEBUGLOG(4, "inBuff  : from %u to %u",
-                                (U32)zds->inBuffSize, (U32)neededInBuffSize);
-                    DEBUGLOG(4, "outBuff : from %u to %u",
-                                (U32)zds->outBuffSize, (U32)neededOutBuffSize);
-                    if (zds->staticSize) {  /* static DCtx */
-                        DEBUGLOG(4, "staticSize : %u", (U32)zds->staticSize);
-                        assert(zds->staticSize >= sizeof(ZSTD_DCtx));  /* controlled at init */
-                        if (bufferSize > zds->staticSize - sizeof(ZSTD_DCtx))
-                            return ERROR(memory_allocation);
-                    } else {
-                        ZSTD_free(zds->inBuff, zds->customMem);
-                        zds->inBuffSize = 0;
-                        zds->outBuffSize = 0;
-                        zds->inBuff = (char*)ZSTD_malloc(bufferSize, zds->customMem);
-                        if (zds->inBuff == NULL) return ERROR(memory_allocation);
-                    }
-                    zds->inBuffSize = neededInBuffSize;
-                    zds->outBuff = zds->inBuff + zds->inBuffSize;
-                    zds->outBuffSize = neededOutBuffSize;
-            }   }
+                size_t const neededOutBuffSize = zds->outBufferMode == ZSTD_obm_buffered
+                        ? ZSTD_decodingBufferSize_min(zds->fParams.windowSize, zds->fParams.frameContentSize)
+                        : 0;
+
+                ZSTD_DCtx_updateOversizedDuration(zds, neededInBuffSize, neededOutBuffSize);
+
+                {   int const tooSmall = (zds->inBuffSize < neededInBuffSize) || (zds->outBuffSize < neededOutBuffSize);
+                    int const tooLarge = ZSTD_DCtx_isOversizedTooLong(zds);
+
+                    if (tooSmall || tooLarge) {
+                        size_t const bufferSize = neededInBuffSize + neededOutBuffSize;
+                        DEBUGLOG(4, "inBuff  : from %u to %u",
+                                    (U32)zds->inBuffSize, (U32)neededInBuffSize);
+                        DEBUGLOG(4, "outBuff : from %u to %u",
+                                    (U32)zds->outBuffSize, (U32)neededOutBuffSize);
+                        if (zds->staticSize) {  /* static DCtx */
+                            DEBUGLOG(4, "staticSize : %u", (U32)zds->staticSize);
+                            assert(zds->staticSize >= sizeof(ZSTD_DCtx));  /* controlled at init */
+                            RETURN_ERROR_IF(
+                                bufferSize > zds->staticSize - sizeof(ZSTD_DCtx),
+                                memory_allocation, "");
+                        } else {
+                            ZSTD_free(zds->inBuff, zds->customMem);
+                            zds->inBuffSize = 0;
+                            zds->outBuffSize = 0;
+                            zds->inBuff = (char*)ZSTD_malloc(bufferSize, zds->customMem);
+                            RETURN_ERROR_IF(zds->inBuff == NULL, memory_allocation, "");
+                        }
+                        zds->inBuffSize = neededInBuffSize;
+                        zds->outBuff = zds->inBuff + zds->inBuffSize;
+                        zds->outBuffSize = neededOutBuffSize;
+            }   }   }
             zds->streamStage = zdss_read;
             /* fall-through */
 
         case zdss_read:
             DEBUGLOG(5, "stage zdss_read");
-            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds);
+            {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompressWithInputSize(zds, iend - ip);
                 DEBUGLOG(5, "neededInSize = %u", (U32)neededInSize);
                 if (neededInSize==0) {  /* end of frame */
                     zds->streamStage = zdss_init;
@@ -2427,42 +1768,40 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
                     break;
                 }
                 if ((size_t)(iend-ip) >= neededInSize) {  /* decode directly from src */
-                    int const isSkipFrame = ZSTD_isSkipFrame(zds);
-                    size_t const decodedSize = ZSTD_decompressContinue(zds,
-                        zds->outBuff + zds->outStart, (isSkipFrame ? 0 : zds->outBuffSize - zds->outStart),
-                        ip, neededInSize);
-                    if (ZSTD_isError(decodedSize)) return decodedSize;
+                    FORWARD_IF_ERROR(ZSTD_decompressContinueStream(zds, &op, oend, ip, neededInSize), "");
                     ip += neededInSize;
-                    if (!decodedSize && !isSkipFrame) break;   /* this was just a header */
-                    zds->outEnd = zds->outStart + decodedSize;
-                    zds->streamStage = zdss_flush;
+                    /* Function modifies the stage so we must break */
                     break;
             }   }
             if (ip==iend) { someMoreWork = 0; break; }   /* no more input */
             zds->streamStage = zdss_load;
             /* fall-through */
+
         case zdss_load:
             {   size_t const neededInSize = ZSTD_nextSrcSizeToDecompress(zds);
-                size_t const toLoad = neededInSize - zds->inPos;   /* should always be <= remaining space within inBuff */
+                size_t const toLoad = neededInSize - zds->inPos;
+                int const isSkipFrame = ZSTD_isSkipFrame(zds);
                 size_t loadedSize;
-                if (toLoad > zds->inBuffSize - zds->inPos) return ERROR(corruption_detected);   /* should never happen */
-                loadedSize = ZSTD_limitCopy(zds->inBuff + zds->inPos, toLoad, ip, iend-ip);
+                /* At this point we shouldn't be decompressing a block that we can stream. */
+                assert(neededInSize == ZSTD_nextSrcSizeToDecompressWithInputSize(zds, iend - ip));
+                if (isSkipFrame) {
+                    loadedSize = MIN(toLoad, (size_t)(iend-ip));
+                } else {
+                    RETURN_ERROR_IF(toLoad > zds->inBuffSize - zds->inPos,
+                                    corruption_detected,
+                                    "should never happen");
+                    loadedSize = ZSTD_limitCopy(zds->inBuff + zds->inPos, toLoad, ip, iend-ip);
+                }
                 ip += loadedSize;
                 zds->inPos += loadedSize;
                 if (loadedSize < toLoad) { someMoreWork = 0; break; }   /* not enough input, wait for more */
 
                 /* decode loaded input */
-                {  const int isSkipFrame = ZSTD_isSkipFrame(zds);
-                   size_t const decodedSize = ZSTD_decompressContinue(zds,
-                        zds->outBuff + zds->outStart, zds->outBuffSize - zds->outStart,
-                        zds->inBuff, neededInSize);
-                    if (ZSTD_isError(decodedSize)) return decodedSize;
-                    zds->inPos = 0;   /* input is consumed */
-                    if (!decodedSize && !isSkipFrame) { zds->streamStage = zdss_read; break; }   /* this was just a header */
-                    zds->outEnd = zds->outStart +  decodedSize;
-            }   }
-            zds->streamStage = zdss_flush;
-            /* fall-through */
+                zds->inPos = 0;   /* input is consumed */
+                FORWARD_IF_ERROR(ZSTD_decompressContinueStream(zds, &op, oend, zds->inBuff, neededInSize), "");
+                /* Function modifies the stage so we must break */
+                break;
+            }
         case zdss_flush:
             {   size_t const toFlushSize = zds->outEnd - zds->outStart;
                 size_t const flushedSize = ZSTD_limitCopy(op, oend-op, zds->outBuff + zds->outStart, toFlushSize);
@@ -2483,12 +1822,28 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
             someMoreWork = 0;
             break;
 
-        default: return ERROR(GENERIC);   /* impossible */
+        default:
+            assert(0);    /* impossible */
+            RETURN_ERROR(GENERIC, "impossible to reach");   /* some compiler require default to do something */
     }   }
 
     /* result */
-    input->pos += (size_t)(ip-istart);
-    output->pos += (size_t)(op-ostart);
+    input->pos = (size_t)(ip - (const char*)(input->src));
+    output->pos = (size_t)(op - (char*)(output->dst));
+
+    /* Update the expected output buffer for ZSTD_obm_stable. */
+    zds->expectedOutBuffer = *output;
+
+    if ((ip==istart) && (op==ostart)) {  /* no forward progress */
+        zds->noForwardProgress ++;
+        if (zds->noForwardProgress >= ZSTD_NO_FORWARD_PROGRESS_MAX) {
+            RETURN_ERROR_IF(op==oend, dstSize_tooSmall, "");
+            RETURN_ERROR_IF(ip==iend, srcSize_wrong, "");
+            assert(0);
+        }
+    } else {
+        zds->noForwardProgress = 0;
+    }
     {   size_t nextSrcSizeHint = ZSTD_nextSrcSizeToDecompress(zds);
         if (!nextSrcSizeHint) {   /* frame fully decoded */
             if (zds->outEnd == zds->outStart) {  /* output fully flushed */
@@ -2509,8 +1864,22 @@ size_t ZSTD_decompressStream(ZSTD_DStream* zds, ZSTD_outBuffer* output, ZSTD_inB
             return 1;
         }  /* nextSrcSizeHint==0 */
         nextSrcSizeHint += ZSTD_blockHeaderSize * (ZSTD_nextInputType(zds) == ZSTDnit_block);   /* preload header of next block */
-        if (zds->inPos > nextSrcSizeHint) return ERROR(GENERIC);   /* should never happen */
-        nextSrcSizeHint -= zds->inPos;   /* already loaded*/
+        assert(zds->inPos <= nextSrcSizeHint);
+        nextSrcSizeHint -= zds->inPos;   /* part already loaded*/
         return nextSrcSizeHint;
     }
 }
+
+size_t ZSTD_decompressStream_simpleArgs (
+                            ZSTD_DCtx* dctx,
+                            void* dst, size_t dstCapacity, size_t* dstPos,
+                      const void* src, size_t srcSize, size_t* srcPos)
+{
+    ZSTD_outBuffer output = { dst, dstCapacity, *dstPos };
+    ZSTD_inBuffer  input  = { src, srcSize, *srcPos };
+    /* ZSTD_compress_generic() will check validity of dstPos and srcPos */
+    size_t const cErr = ZSTD_decompressStream(dctx, &output, &input);
+    *dstPos = output.pos;
+    *srcPos = input.pos;
+    return cErr;
+}
diff --git a/contrib/zstd/zstd_decompress_block.c b/contrib/zstd/zstd_decompress_block.c
new file mode 100644
index 000000000..fcb381b99
--- /dev/null
+++ b/contrib/zstd/zstd_decompress_block.c
@@ -0,0 +1,1432 @@
+/*
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+/* zstd_decompress_block :
+ * this module takes care of decompressing _compressed_ block */
+
+/*-*******************************************************
+*  Dependencies
+*********************************************************/
+#include <string.h>      /* memcpy, memmove, memset */
+#include "compiler.h"    /* prefetch */
+#include "cpu.h"         /* bmi2 */
+#include "mem.h"         /* low level memory routines */
+#define FSE_STATIC_LINKING_ONLY
+#include "fse.h"
+#define HUF_STATIC_LINKING_ONLY
+#include "huf.h"
+#include "zstd_internal.h"
+#include "zstd_decompress_internal.h"   /* ZSTD_DCtx */
+#include "zstd_ddict.h"  /* ZSTD_DDictDictContent */
+#include "zstd_decompress_block.h"
+
+/*_*******************************************************
+*  Macros
+**********************************************************/
+
+/* These two optional macros force the use one way or another of the two
+ * ZSTD_decompressSequences implementations. You can't force in both directions
+ * at the same time.
+ */
+#if defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
+    defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
+#error "Cannot force the use of the short and the long ZSTD_decompressSequences variants!"
+#endif
+
+
+/*_*******************************************************
+*  Memory operations
+**********************************************************/
+static void ZSTD_copy4(void* dst, const void* src) { memcpy(dst, src, 4); }
+
+
+/*-*************************************************************
+ *   Block decoding
+ ***************************************************************/
+
+/*! ZSTD_getcBlockSize() :
+ *  Provides the size of compressed block from block header `src` */
+size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
+                          blockProperties_t* bpPtr)
+{
+    RETURN_ERROR_IF(srcSize < ZSTD_blockHeaderSize, srcSize_wrong, "");
+
+    {   U32 const cBlockHeader = MEM_readLE24(src);
+        U32 const cSize = cBlockHeader >> 3;
+        bpPtr->lastBlock = cBlockHeader & 1;
+        bpPtr->blockType = (blockType_e)((cBlockHeader >> 1) & 3);
+        bpPtr->origSize = cSize;   /* only useful for RLE */
+        if (bpPtr->blockType == bt_rle) return 1;
+        RETURN_ERROR_IF(bpPtr->blockType == bt_reserved, corruption_detected, "");
+        return cSize;
+    }
+}
+
+
+/* Hidden declaration for fullbench */
+size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
+                          const void* src, size_t srcSize);
+/*! ZSTD_decodeLiteralsBlock() :
+ * @return : nb of bytes read from src (< srcSize )
+ *  note : symbol not declared but exposed for fullbench */
+size_t ZSTD_decodeLiteralsBlock(ZSTD_DCtx* dctx,
+                          const void* src, size_t srcSize)   /* note : srcSize < BLOCKSIZE */
+{
+    DEBUGLOG(5, "ZSTD_decodeLiteralsBlock");
+    RETURN_ERROR_IF(srcSize < MIN_CBLOCK_SIZE, corruption_detected, "");
+
+    {   const BYTE* const istart = (const BYTE*) src;
+        symbolEncodingType_e const litEncType = (symbolEncodingType_e)(istart[0] & 3);
+
+        switch(litEncType)
+        {
+        case set_repeat:
+            DEBUGLOG(5, "set_repeat flag : re-using stats from previous compressed literals block");
+            RETURN_ERROR_IF(dctx->litEntropy==0, dictionary_corrupted, "");
+            /* fall-through */
+
+        case set_compressed:
+            RETURN_ERROR_IF(srcSize < 5, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 3; here we need up to 5 for case 3");
+            {   size_t lhSize, litSize, litCSize;
+                U32 singleStream=0;
+                U32 const lhlCode = (istart[0] >> 2) & 3;
+                U32 const lhc = MEM_readLE32(istart);
+                size_t hufSuccess;
+                switch(lhlCode)
+                {
+                case 0: case 1: default:   /* note : default is impossible, since lhlCode into [0..3] */
+                    /* 2 - 2 - 10 - 10 */
+                    singleStream = !lhlCode;
+                    lhSize = 3;
+                    litSize  = (lhc >> 4) & 0x3FF;
+                    litCSize = (lhc >> 14) & 0x3FF;
+                    break;
+                case 2:
+                    /* 2 - 2 - 14 - 14 */
+                    lhSize = 4;
+                    litSize  = (lhc >> 4) & 0x3FFF;
+                    litCSize = lhc >> 18;
+                    break;
+                case 3:
+                    /* 2 - 2 - 18 - 18 */
+                    lhSize = 5;
+                    litSize  = (lhc >> 4) & 0x3FFFF;
+                    litCSize = (lhc >> 22) + ((size_t)istart[4] << 10);
+                    break;
+                }
+                RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected, "");
+                RETURN_ERROR_IF(litCSize + lhSize > srcSize, corruption_detected, "");
+
+                /* prefetch huffman table if cold */
+                if (dctx->ddictIsCold && (litSize > 768 /* heuristic */)) {
+                    PREFETCH_AREA(dctx->HUFptr, sizeof(dctx->entropy.hufTable));
+                }
+
+                if (litEncType==set_repeat) {
+                    if (singleStream) {
+                        hufSuccess = HUF_decompress1X_usingDTable_bmi2(
+                            dctx->litBuffer, litSize, istart+lhSize, litCSize,
+                            dctx->HUFptr, dctx->bmi2);
+                    } else {
+                        hufSuccess = HUF_decompress4X_usingDTable_bmi2(
+                            dctx->litBuffer, litSize, istart+lhSize, litCSize,
+                            dctx->HUFptr, dctx->bmi2);
+                    }
+                } else {
+                    if (singleStream) {
+#if defined(HUF_FORCE_DECOMPRESS_X2)
+                        hufSuccess = HUF_decompress1X_DCtx_wksp(
+                            dctx->entropy.hufTable, dctx->litBuffer, litSize,
+                            istart+lhSize, litCSize, dctx->workspace,
+                            sizeof(dctx->workspace));
+#else
+                        hufSuccess = HUF_decompress1X1_DCtx_wksp_bmi2(
+                            dctx->entropy.hufTable, dctx->litBuffer, litSize,
+                            istart+lhSize, litCSize, dctx->workspace,
+                            sizeof(dctx->workspace), dctx->bmi2);
+#endif
+                    } else {
+                        hufSuccess = HUF_decompress4X_hufOnly_wksp_bmi2(
+                            dctx->entropy.hufTable, dctx->litBuffer, litSize,
+                            istart+lhSize, litCSize, dctx->workspace,
+                            sizeof(dctx->workspace), dctx->bmi2);
+                    }
+                }
+
+                RETURN_ERROR_IF(HUF_isError(hufSuccess), corruption_detected, "");
+
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                dctx->litEntropy = 1;
+                if (litEncType==set_compressed) dctx->HUFptr = dctx->entropy.hufTable;
+                memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+                return litCSize + lhSize;
+            }
+
+        case set_basic:
+            {   size_t litSize, lhSize;
+                U32 const lhlCode = ((istart[0]) >> 2) & 3;
+                switch(lhlCode)
+                {
+                case 0: case 2: default:   /* note : default is impossible, since lhlCode into [0..3] */
+                    lhSize = 1;
+                    litSize = istart[0] >> 3;
+                    break;
+                case 1:
+                    lhSize = 2;
+                    litSize = MEM_readLE16(istart) >> 4;
+                    break;
+                case 3:
+                    lhSize = 3;
+                    litSize = MEM_readLE24(istart) >> 4;
+                    break;
+                }
+
+                if (lhSize+litSize+WILDCOPY_OVERLENGTH > srcSize) {  /* risk reading beyond src buffer with wildcopy */
+                    RETURN_ERROR_IF(litSize+lhSize > srcSize, corruption_detected, "");
+                    memcpy(dctx->litBuffer, istart+lhSize, litSize);
+                    dctx->litPtr = dctx->litBuffer;
+                    dctx->litSize = litSize;
+                    memset(dctx->litBuffer + dctx->litSize, 0, WILDCOPY_OVERLENGTH);
+                    return lhSize+litSize;
+                }
+                /* direct reference into compressed stream */
+                dctx->litPtr = istart+lhSize;
+                dctx->litSize = litSize;
+                return lhSize+litSize;
+            }
+
+        case set_rle:
+            {   U32 const lhlCode = ((istart[0]) >> 2) & 3;
+                size_t litSize, lhSize;
+                switch(lhlCode)
+                {
+                case 0: case 2: default:   /* note : default is impossible, since lhlCode into [0..3] */
+                    lhSize = 1;
+                    litSize = istart[0] >> 3;
+                    break;
+                case 1:
+                    lhSize = 2;
+                    litSize = MEM_readLE16(istart) >> 4;
+                    break;
+                case 3:
+                    lhSize = 3;
+                    litSize = MEM_readLE24(istart) >> 4;
+                    RETURN_ERROR_IF(srcSize<4, corruption_detected, "srcSize >= MIN_CBLOCK_SIZE == 3; here we need lhSize+1 = 4");
+                    break;
+                }
+                RETURN_ERROR_IF(litSize > ZSTD_BLOCKSIZE_MAX, corruption_detected, "");
+                memset(dctx->litBuffer, istart[lhSize], litSize + WILDCOPY_OVERLENGTH);
+                dctx->litPtr = dctx->litBuffer;
+                dctx->litSize = litSize;
+                return lhSize+1;
+            }
+        default:
+            RETURN_ERROR(corruption_detected, "impossible");
+        }
+    }
+}
+
+/* Default FSE distribution tables.
+ * These are pre-calculated FSE decoding tables using default distributions as defined in specification :
+ * https://github.com/facebook/zstd/blob/master/doc/zstd_compression_format.md#default-distributions
+ * They were generated programmatically with following method :
+ * - start from default distributions, present in /lib/common/zstd_internal.h
+ * - generate tables normally, using ZSTD_buildFSETable()
+ * - printout the content of tables
+ * - pretify output, report below, test with fuzzer to ensure it's correct */
+
+/* Default FSE distribution table for Literal Lengths */
+static const ZSTD_seqSymbol LL_defaultDTable[(1<<LL_DEFAULTNORMLOG)+1] = {
+     {  1,  1,  1, LL_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
+     /* nextState, nbAddBits, nbBits, baseVal */
+     {  0,  0,  4,    0},  { 16,  0,  4,    0},
+     { 32,  0,  5,    1},  {  0,  0,  5,    3},
+     {  0,  0,  5,    4},  {  0,  0,  5,    6},
+     {  0,  0,  5,    7},  {  0,  0,  5,    9},
+     {  0,  0,  5,   10},  {  0,  0,  5,   12},
+     {  0,  0,  6,   14},  {  0,  1,  5,   16},
+     {  0,  1,  5,   20},  {  0,  1,  5,   22},
+     {  0,  2,  5,   28},  {  0,  3,  5,   32},
+     {  0,  4,  5,   48},  { 32,  6,  5,   64},
+     {  0,  7,  5,  128},  {  0,  8,  6,  256},
+     {  0, 10,  6, 1024},  {  0, 12,  6, 4096},
+     { 32,  0,  4,    0},  {  0,  0,  4,    1},
+     {  0,  0,  5,    2},  { 32,  0,  5,    4},
+     {  0,  0,  5,    5},  { 32,  0,  5,    7},
+     {  0,  0,  5,    8},  { 32,  0,  5,   10},
+     {  0,  0,  5,   11},  {  0,  0,  6,   13},
+     { 32,  1,  5,   16},  {  0,  1,  5,   18},
+     { 32,  1,  5,   22},  {  0,  2,  5,   24},
+     { 32,  3,  5,   32},  {  0,  3,  5,   40},
+     {  0,  6,  4,   64},  { 16,  6,  4,   64},
+     { 32,  7,  5,  128},  {  0,  9,  6,  512},
+     {  0, 11,  6, 2048},  { 48,  0,  4,    0},
+     { 16,  0,  4,    1},  { 32,  0,  5,    2},
+     { 32,  0,  5,    3},  { 32,  0,  5,    5},
+     { 32,  0,  5,    6},  { 32,  0,  5,    8},
+     { 32,  0,  5,    9},  { 32,  0,  5,   11},
+     { 32,  0,  5,   12},  {  0,  0,  6,   15},
+     { 32,  1,  5,   18},  { 32,  1,  5,   20},
+     { 32,  2,  5,   24},  { 32,  2,  5,   28},
+     { 32,  3,  5,   40},  { 32,  4,  5,   48},
+     {  0, 16,  6,65536},  {  0, 15,  6,32768},
+     {  0, 14,  6,16384},  {  0, 13,  6, 8192},
+};   /* LL_defaultDTable */
+
+/* Default FSE distribution table for Offset Codes */
+static const ZSTD_seqSymbol OF_defaultDTable[(1<<OF_DEFAULTNORMLOG)+1] = {
+    {  1,  1,  1, OF_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
+    /* nextState, nbAddBits, nbBits, baseVal */
+    {  0,  0,  5,    0},     {  0,  6,  4,   61},
+    {  0,  9,  5,  509},     {  0, 15,  5,32765},
+    {  0, 21,  5,2097149},   {  0,  3,  5,    5},
+    {  0,  7,  4,  125},     {  0, 12,  5, 4093},
+    {  0, 18,  5,262141},    {  0, 23,  5,8388605},
+    {  0,  5,  5,   29},     {  0,  8,  4,  253},
+    {  0, 14,  5,16381},     {  0, 20,  5,1048573},
+    {  0,  2,  5,    1},     { 16,  7,  4,  125},
+    {  0, 11,  5, 2045},     {  0, 17,  5,131069},
+    {  0, 22,  5,4194301},   {  0,  4,  5,   13},
+    { 16,  8,  4,  253},     {  0, 13,  5, 8189},
+    {  0, 19,  5,524285},    {  0,  1,  5,    1},
+    { 16,  6,  4,   61},     {  0, 10,  5, 1021},
+    {  0, 16,  5,65533},     {  0, 28,  5,268435453},
+    {  0, 27,  5,134217725}, {  0, 26,  5,67108861},
+    {  0, 25,  5,33554429},  {  0, 24,  5,16777213},
+};   /* OF_defaultDTable */
+
+
+/* Default FSE distribution table for Match Lengths */
+static const ZSTD_seqSymbol ML_defaultDTable[(1<<ML_DEFAULTNORMLOG)+1] = {
+    {  1,  1,  1, ML_DEFAULTNORMLOG},  /* header : fastMode, tableLog */
+    /* nextState, nbAddBits, nbBits, baseVal */
+    {  0,  0,  6,    3},  {  0,  0,  4,    4},
+    { 32,  0,  5,    5},  {  0,  0,  5,    6},
+    {  0,  0,  5,    8},  {  0,  0,  5,    9},
+    {  0,  0,  5,   11},  {  0,  0,  6,   13},
+    {  0,  0,  6,   16},  {  0,  0,  6,   19},
+    {  0,  0,  6,   22},  {  0,  0,  6,   25},
+    {  0,  0,  6,   28},  {  0,  0,  6,   31},
+    {  0,  0,  6,   34},  {  0,  1,  6,   37},
+    {  0,  1,  6,   41},  {  0,  2,  6,   47},
+    {  0,  3,  6,   59},  {  0,  4,  6,   83},
+    {  0,  7,  6,  131},  {  0,  9,  6,  515},
+    { 16,  0,  4,    4},  {  0,  0,  4,    5},
+    { 32,  0,  5,    6},  {  0,  0,  5,    7},
+    { 32,  0,  5,    9},  {  0,  0,  5,   10},
+    {  0,  0,  6,   12},  {  0,  0,  6,   15},
+    {  0,  0,  6,   18},  {  0,  0,  6,   21},
+    {  0,  0,  6,   24},  {  0,  0,  6,   27},
+    {  0,  0,  6,   30},  {  0,  0,  6,   33},
+    {  0,  1,  6,   35},  {  0,  1,  6,   39},
+    {  0,  2,  6,   43},  {  0,  3,  6,   51},
+    {  0,  4,  6,   67},  {  0,  5,  6,   99},
+    {  0,  8,  6,  259},  { 32,  0,  4,    4},
+    { 48,  0,  4,    4},  { 16,  0,  4,    5},
+    { 32,  0,  5,    7},  { 32,  0,  5,    8},
+    { 32,  0,  5,   10},  { 32,  0,  5,   11},
+    {  0,  0,  6,   14},  {  0,  0,  6,   17},
+    {  0,  0,  6,   20},  {  0,  0,  6,   23},
+    {  0,  0,  6,   26},  {  0,  0,  6,   29},
+    {  0,  0,  6,   32},  {  0, 16,  6,65539},
+    {  0, 15,  6,32771},  {  0, 14,  6,16387},
+    {  0, 13,  6, 8195},  {  0, 12,  6, 4099},
+    {  0, 11,  6, 2051},  {  0, 10,  6, 1027},
+};   /* ML_defaultDTable */
+
+
+static void ZSTD_buildSeqTable_rle(ZSTD_seqSymbol* dt, U32 baseValue, U32 nbAddBits)
+{
+    void* ptr = dt;
+    ZSTD_seqSymbol_header* const DTableH = (ZSTD_seqSymbol_header*)ptr;
+    ZSTD_seqSymbol* const cell = dt + 1;
+
+    DTableH->tableLog = 0;
+    DTableH->fastMode = 0;
+
+    cell->nbBits = 0;
+    cell->nextState = 0;
+    assert(nbAddBits < 255);
+    cell->nbAdditionalBits = (BYTE)nbAddBits;
+    cell->baseValue = baseValue;
+}
+
+
+/* ZSTD_buildFSETable() :
+ * generate FSE decoding table for one symbol (ll, ml or off)
+ * cannot fail if input is valid =>
+ * all inputs are presumed validated at this stage */
+void
+ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
+            const short* normalizedCounter, unsigned maxSymbolValue,
+            const U32* baseValue, const U32* nbAdditionalBits,
+            unsigned tableLog)
+{
+    ZSTD_seqSymbol* const tableDecode = dt+1;
+    U16 symbolNext[MaxSeq+1];
+
+    U32 const maxSV1 = maxSymbolValue + 1;
+    U32 const tableSize = 1 << tableLog;
+    U32 highThreshold = tableSize-1;
+
+    /* Sanity Checks */
+    assert(maxSymbolValue <= MaxSeq);
+    assert(tableLog <= MaxFSELog);
+
+    /* Init, lay down lowprob symbols */
+    {   ZSTD_seqSymbol_header DTableH;
+        DTableH.tableLog = tableLog;
+        DTableH.fastMode = 1;
+        {   S16 const largeLimit= (S16)(1 << (tableLog-1));
+            U32 s;
+            for (s=0; s<maxSV1; s++) {
+                if (normalizedCounter[s]==-1) {
+                    tableDecode[highThreshold--].baseValue = s;
+                    symbolNext[s] = 1;
+                } else {
+                    if (normalizedCounter[s] >= largeLimit) DTableH.fastMode=0;
+                    assert(normalizedCounter[s]>=0);
+                    symbolNext[s] = (U16)normalizedCounter[s];
+        }   }   }
+        memcpy(dt, &DTableH, sizeof(DTableH));
+    }
+
+    /* Spread symbols */
+    {   U32 const tableMask = tableSize-1;
+        U32 const step = FSE_TABLESTEP(tableSize);
+        U32 s, position = 0;
+        for (s=0; s<maxSV1; s++) {
+            int i;
+            for (i=0; i<normalizedCounter[s]; i++) {
+                tableDecode[position].baseValue = s;
+                position = (position + step) & tableMask;
+                while (position > highThreshold) position = (position + step) & tableMask;   /* lowprob area */
+        }   }
+        assert(position == 0); /* position must reach all cells once, otherwise normalizedCounter is incorrect */
+    }
+
+    /* Build Decoding table */
+    {   U32 u;
+        for (u=0; u<tableSize; u++) {
+            U32 const symbol = tableDecode[u].baseValue;
+            U32 const nextState = symbolNext[symbol]++;
+            tableDecode[u].nbBits = (BYTE) (tableLog - BIT_highbit32(nextState) );
+            tableDecode[u].nextState = (U16) ( (nextState << tableDecode[u].nbBits) - tableSize);
+            assert(nbAdditionalBits[symbol] < 255);
+            tableDecode[u].nbAdditionalBits = (BYTE)nbAdditionalBits[symbol];
+            tableDecode[u].baseValue = baseValue[symbol];
+    }   }
+}
+
+
+/*! ZSTD_buildSeqTable() :
+ * @return : nb bytes read from src,
+ *           or an error code if it fails */
+static size_t ZSTD_buildSeqTable(ZSTD_seqSymbol* DTableSpace, const ZSTD_seqSymbol** DTablePtr,
+                                 symbolEncodingType_e type, unsigned max, U32 maxLog,
+                                 const void* src, size_t srcSize,
+                                 const U32* baseValue, const U32* nbAdditionalBits,
+                                 const ZSTD_seqSymbol* defaultTable, U32 flagRepeatTable,
+                                 int ddictIsCold, int nbSeq)
+{
+    switch(type)
+    {
+    case set_rle :
+        RETURN_ERROR_IF(!srcSize, srcSize_wrong, "");
+        RETURN_ERROR_IF((*(const BYTE*)src) > max, corruption_detected, "");
+        {   U32 const symbol = *(const BYTE*)src;
+            U32 const baseline = baseValue[symbol];
+            U32 const nbBits = nbAdditionalBits[symbol];
+            ZSTD_buildSeqTable_rle(DTableSpace, baseline, nbBits);
+        }
+        *DTablePtr = DTableSpace;
+        return 1;
+    case set_basic :
+        *DTablePtr = defaultTable;
+        return 0;
+    case set_repeat:
+        RETURN_ERROR_IF(!flagRepeatTable, corruption_detected, "");
+        /* prefetch FSE table if used */
+        if (ddictIsCold && (nbSeq > 24 /* heuristic */)) {
+            const void* const pStart = *DTablePtr;
+            size_t const pSize = sizeof(ZSTD_seqSymbol) * (SEQSYMBOL_TABLE_SIZE(maxLog));
+            PREFETCH_AREA(pStart, pSize);
+        }
+        return 0;
+    case set_compressed :
+        {   unsigned tableLog;
+            S16 norm[MaxSeq+1];
+            size_t const headerSize = FSE_readNCount(norm, &max, &tableLog, src, srcSize);
+            RETURN_ERROR_IF(FSE_isError(headerSize), corruption_detected, "");
+            RETURN_ERROR_IF(tableLog > maxLog, corruption_detected, "");
+            ZSTD_buildFSETable(DTableSpace, norm, max, baseValue, nbAdditionalBits, tableLog);
+            *DTablePtr = DTableSpace;
+            return headerSize;
+        }
+    default :
+        assert(0);
+        RETURN_ERROR(GENERIC, "impossible");
+    }
+}
+
+size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
+                             const void* src, size_t srcSize)
+{
+    const BYTE* const istart = (const BYTE* const)src;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* ip = istart;
+    int nbSeq;
+    DEBUGLOG(5, "ZSTD_decodeSeqHeaders");
+
+    /* check */
+    RETURN_ERROR_IF(srcSize < MIN_SEQUENCES_SIZE, srcSize_wrong, "");
+
+    /* SeqHead */
+    nbSeq = *ip++;
+    if (!nbSeq) {
+        *nbSeqPtr=0;
+        RETURN_ERROR_IF(srcSize != 1, srcSize_wrong, "");
+        return 1;
+    }
+    if (nbSeq > 0x7F) {
+        if (nbSeq == 0xFF) {
+            RETURN_ERROR_IF(ip+2 > iend, srcSize_wrong, "");
+            nbSeq = MEM_readLE16(ip) + LONGNBSEQ, ip+=2;
+        } else {
+            RETURN_ERROR_IF(ip >= iend, srcSize_wrong, "");
+            nbSeq = ((nbSeq-0x80)<<8) + *ip++;
+        }
+    }
+    *nbSeqPtr = nbSeq;
+
+    /* FSE table descriptors */
+    RETURN_ERROR_IF(ip+1 > iend, srcSize_wrong, ""); /* minimum possible size: 1 byte for symbol encoding types */
+    {   symbolEncodingType_e const LLtype = (symbolEncodingType_e)(*ip >> 6);
+        symbolEncodingType_e const OFtype = (symbolEncodingType_e)((*ip >> 4) & 3);
+        symbolEncodingType_e const MLtype = (symbolEncodingType_e)((*ip >> 2) & 3);
+        ip++;
+
+        /* Build DTables */
+        {   size_t const llhSize = ZSTD_buildSeqTable(dctx->entropy.LLTable, &dctx->LLTptr,
+                                                      LLtype, MaxLL, LLFSELog,
+                                                      ip, iend-ip,
+                                                      LL_base, LL_bits,
+                                                      LL_defaultDTable, dctx->fseEntropy,
+                                                      dctx->ddictIsCold, nbSeq);
+            RETURN_ERROR_IF(ZSTD_isError(llhSize), corruption_detected, "ZSTD_buildSeqTable failed");
+            ip += llhSize;
+        }
+
+        {   size_t const ofhSize = ZSTD_buildSeqTable(dctx->entropy.OFTable, &dctx->OFTptr,
+                                                      OFtype, MaxOff, OffFSELog,
+                                                      ip, iend-ip,
+                                                      OF_base, OF_bits,
+                                                      OF_defaultDTable, dctx->fseEntropy,
+                                                      dctx->ddictIsCold, nbSeq);
+            RETURN_ERROR_IF(ZSTD_isError(ofhSize), corruption_detected, "ZSTD_buildSeqTable failed");
+            ip += ofhSize;
+        }
+
+        {   size_t const mlhSize = ZSTD_buildSeqTable(dctx->entropy.MLTable, &dctx->MLTptr,
+                                                      MLtype, MaxML, MLFSELog,
+                                                      ip, iend-ip,
+                                                      ML_base, ML_bits,
+                                                      ML_defaultDTable, dctx->fseEntropy,
+                                                      dctx->ddictIsCold, nbSeq);
+            RETURN_ERROR_IF(ZSTD_isError(mlhSize), corruption_detected, "ZSTD_buildSeqTable failed");
+            ip += mlhSize;
+        }
+    }
+
+    return ip-istart;
+}
+
+
+typedef struct {
+    size_t litLength;
+    size_t matchLength;
+    size_t offset;
+    const BYTE* match;
+} seq_t;
+
+typedef struct {
+    size_t state;
+    const ZSTD_seqSymbol* table;
+} ZSTD_fseState;
+
+typedef struct {
+    BIT_DStream_t DStream;
+    ZSTD_fseState stateLL;
+    ZSTD_fseState stateOffb;
+    ZSTD_fseState stateML;
+    size_t prevOffset[ZSTD_REP_NUM];
+    const BYTE* prefixStart;
+    const BYTE* dictEnd;
+    size_t pos;
+} seqState_t;
+
+/*! ZSTD_overlapCopy8() :
+ *  Copies 8 bytes from ip to op and updates op and ip where ip <= op.
+ *  If the offset is < 8 then the offset is spread to at least 8 bytes.
+ *
+ *  Precondition: *ip <= *op
+ *  Postcondition: *op - *op >= 8
+ */
+HINT_INLINE void ZSTD_overlapCopy8(BYTE** op, BYTE const** ip, size_t offset) {
+    assert(*ip <= *op);
+    if (offset < 8) {
+        /* close range match, overlap */
+        static const U32 dec32table[] = { 0, 1, 2, 1, 4, 4, 4, 4 };   /* added */
+        static const int dec64table[] = { 8, 8, 8, 7, 8, 9,10,11 };   /* subtracted */
+        int const sub2 = dec64table[offset];
+        (*op)[0] = (*ip)[0];
+        (*op)[1] = (*ip)[1];
+        (*op)[2] = (*ip)[2];
+        (*op)[3] = (*ip)[3];
+        *ip += dec32table[offset];
+        ZSTD_copy4(*op+4, *ip);
+        *ip -= sub2;
+    } else {
+        ZSTD_copy8(*op, *ip);
+    }
+    *ip += 8;
+    *op += 8;
+    assert(*op - *ip >= 8);
+}
+
+/*! ZSTD_safecopy() :
+ *  Specialized version of memcpy() that is allowed to READ up to WILDCOPY_OVERLENGTH past the input buffer
+ *  and write up to 16 bytes past oend_w (op >= oend_w is allowed).
+ *  This function is only called in the uncommon case where the sequence is near the end of the block. It
+ *  should be fast for a single long sequence, but can be slow for several short sequences.
+ *
+ *  @param ovtype controls the overlap detection
+ *         - ZSTD_no_overlap: The source and destination are guaranteed to be at least WILDCOPY_VECLEN bytes apart.
+ *         - ZSTD_overlap_src_before_dst: The src and dst may overlap and may be any distance apart.
+ *           The src buffer must be before the dst buffer.
+ */
+static void ZSTD_safecopy(BYTE* op, BYTE* const oend_w, BYTE const* ip, ptrdiff_t length, ZSTD_overlap_e ovtype) {
+    ptrdiff_t const diff = op - ip;
+    BYTE* const oend = op + length;
+
+    assert((ovtype == ZSTD_no_overlap && (diff <= -8 || diff >= 8 || op >= oend_w)) ||
+           (ovtype == ZSTD_overlap_src_before_dst && diff >= 0));
+
+    if (length < 8) {
+        /* Handle short lengths. */
+        while (op < oend) *op++ = *ip++;
+        return;
+    }
+    if (ovtype == ZSTD_overlap_src_before_dst) {
+        /* Copy 8 bytes and ensure the offset >= 8 when there can be overlap. */
+        assert(length >= 8);
+        ZSTD_overlapCopy8(&op, &ip, diff);
+        assert(op - ip >= 8);
+        assert(op <= oend);
+    }
+
+    if (oend <= oend_w) {
+        /* No risk of overwrite. */
+        ZSTD_wildcopy(op, ip, length, ovtype);
+        return;
+    }
+    if (op <= oend_w) {
+        /* Wildcopy until we get close to the end. */
+        assert(oend > oend_w);
+        ZSTD_wildcopy(op, ip, oend_w - op, ovtype);
+        ip += oend_w - op;
+        op = oend_w;
+    }
+    /* Handle the leftovers. */
+    while (op < oend) *op++ = *ip++;
+}
+
+/* ZSTD_execSequenceEnd():
+ * This version handles cases that are near the end of the output buffer. It requires
+ * more careful checks to make sure there is no overflow. By separating out these hard
+ * and unlikely cases, we can speed up the common cases.
+ *
+ * NOTE: This function needs to be fast for a single long sequence, but doesn't need
+ * to be optimized for many small sequences, since those fall into ZSTD_execSequence().
+ */
+FORCE_NOINLINE
+size_t ZSTD_execSequenceEnd(BYTE* op,
+                            BYTE* const oend, seq_t sequence,
+                            const BYTE** litPtr, const BYTE* const litLimit,
+                            const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;
+
+    /* bounds checks : careful of address space overflow in 32-bit mode */
+    RETURN_ERROR_IF(sequenceLength > (size_t)(oend - op), dstSize_tooSmall, "last match must fit within dstBuffer");
+    RETURN_ERROR_IF(sequence.litLength > (size_t)(litLimit - *litPtr), corruption_detected, "try to read beyond literal buffer");
+    assert(op < op + sequenceLength);
+    assert(oLitEnd < op + sequenceLength);
+
+    /* copy literals */
+    ZSTD_safecopy(op, oend_w, *litPtr, sequence.litLength, ZSTD_no_overlap);
+    op = oLitEnd;
+    *litPtr = iLitEnd;
+
+    /* copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
+        /* offset beyond prefix */
+        RETURN_ERROR_IF(sequence.offset > (size_t)(oLitEnd - virtualStart), corruption_detected, "");
+        match = dictEnd - (prefixStart-match);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = prefixStart;
+    }   }
+    ZSTD_safecopy(op, oend_w, match, sequence.matchLength, ZSTD_overlap_src_before_dst);
+    return sequenceLength;
+}
+
+HINT_INLINE
+size_t ZSTD_execSequence(BYTE* op,
+                         BYTE* const oend, seq_t sequence,
+                         const BYTE** litPtr, const BYTE* const litLimit,
+                         const BYTE* const prefixStart, const BYTE* const virtualStart, const BYTE* const dictEnd)
+{
+    BYTE* const oLitEnd = op + sequence.litLength;
+    size_t const sequenceLength = sequence.litLength + sequence.matchLength;
+    BYTE* const oMatchEnd = op + sequenceLength;   /* risk : address space overflow (32-bits) */
+    BYTE* const oend_w = oend - WILDCOPY_OVERLENGTH;   /* risk : address space underflow on oend=NULL */
+    const BYTE* const iLitEnd = *litPtr + sequence.litLength;
+    const BYTE* match = oLitEnd - sequence.offset;
+
+    assert(op != NULL /* Precondition */);
+    assert(oend_w < oend /* No underflow */);
+    /* Handle edge cases in a slow path:
+     *   - Read beyond end of literals
+     *   - Match end is within WILDCOPY_OVERLIMIT of oend
+     *   - 32-bit mode and the match length overflows
+     */
+    if (UNLIKELY(
+            iLitEnd > litLimit ||
+            oMatchEnd > oend_w ||
+            (MEM_32bits() && (size_t)(oend - op) < sequenceLength + WILDCOPY_OVERLENGTH)))
+        return ZSTD_execSequenceEnd(op, oend, sequence, litPtr, litLimit, prefixStart, virtualStart, dictEnd);
+
+    /* Assumptions (everything else goes into ZSTD_execSequenceEnd()) */
+    assert(op <= oLitEnd /* No overflow */);
+    assert(oLitEnd < oMatchEnd /* Non-zero match & no overflow */);
+    assert(oMatchEnd <= oend /* No underflow */);
+    assert(iLitEnd <= litLimit /* Literal length is in bounds */);
+    assert(oLitEnd <= oend_w /* Can wildcopy literals */);
+    assert(oMatchEnd <= oend_w /* Can wildcopy matches */);
+
+    /* Copy Literals:
+     * Split out litLength <= 16 since it is nearly always true. +1.6% on gcc-9.
+     * We likely don't need the full 32-byte wildcopy.
+     */
+    assert(WILDCOPY_OVERLENGTH >= 16);
+    ZSTD_copy16(op, (*litPtr));
+    if (UNLIKELY(sequence.litLength > 16)) {
+        ZSTD_wildcopy(op+16, (*litPtr)+16, sequence.litLength-16, ZSTD_no_overlap);
+    }
+    op = oLitEnd;
+    *litPtr = iLitEnd;   /* update for next sequence */
+
+    /* Copy Match */
+    if (sequence.offset > (size_t)(oLitEnd - prefixStart)) {
+        /* offset beyond prefix -> go into extDict */
+        RETURN_ERROR_IF(UNLIKELY(sequence.offset > (size_t)(oLitEnd - virtualStart)), corruption_detected, "");
+        match = dictEnd + (match - prefixStart);
+        if (match + sequence.matchLength <= dictEnd) {
+            memmove(oLitEnd, match, sequence.matchLength);
+            return sequenceLength;
+        }
+        /* span extDict & currentPrefixSegment */
+        {   size_t const length1 = dictEnd - match;
+            memmove(oLitEnd, match, length1);
+            op = oLitEnd + length1;
+            sequence.matchLength -= length1;
+            match = prefixStart;
+    }   }
+    /* Match within prefix of 1 or more bytes */
+    assert(op <= oMatchEnd);
+    assert(oMatchEnd <= oend_w);
+    assert(match >= prefixStart);
+    assert(sequence.matchLength >= 1);
+
+    /* Nearly all offsets are >= WILDCOPY_VECLEN bytes, which means we can use wildcopy
+     * without overlap checking.
+     */
+    if (LIKELY(sequence.offset >= WILDCOPY_VECLEN)) {
+        /* We bet on a full wildcopy for matches, since we expect matches to be
+         * longer than literals (in general). In silesia, ~10% of matches are longer
+         * than 16 bytes.
+         */
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength, ZSTD_no_overlap);
+        return sequenceLength;
+    }
+    assert(sequence.offset < WILDCOPY_VECLEN);
+
+    /* Copy 8 bytes and spread the offset to be >= 8. */
+    ZSTD_overlapCopy8(&op, &match, sequence.offset);
+
+    /* If the match length is > 8 bytes, then continue with the wildcopy. */
+    if (sequence.matchLength > 8) {
+        assert(op < oMatchEnd);
+        ZSTD_wildcopy(op, match, (ptrdiff_t)sequence.matchLength-8, ZSTD_overlap_src_before_dst);
+    }
+    return sequenceLength;
+}
+
+static void
+ZSTD_initFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, const ZSTD_seqSymbol* dt)
+{
+    const void* ptr = dt;
+    const ZSTD_seqSymbol_header* const DTableH = (const ZSTD_seqSymbol_header*)ptr;
+    DStatePtr->state = BIT_readBits(bitD, DTableH->tableLog);
+    DEBUGLOG(6, "ZSTD_initFseState : val=%u using %u bits",
+                (U32)DStatePtr->state, DTableH->tableLog);
+    BIT_reloadDStream(bitD);
+    DStatePtr->table = dt + 1;
+}
+
+FORCE_INLINE_TEMPLATE void
+ZSTD_updateFseState(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD)
+{
+    ZSTD_seqSymbol const DInfo = DStatePtr->table[DStatePtr->state];
+    U32 const nbBits = DInfo.nbBits;
+    size_t const lowBits = BIT_readBits(bitD, nbBits);
+    DStatePtr->state = DInfo.nextState + lowBits;
+}
+
+FORCE_INLINE_TEMPLATE void
+ZSTD_updateFseStateWithDInfo(ZSTD_fseState* DStatePtr, BIT_DStream_t* bitD, ZSTD_seqSymbol const DInfo)
+{
+    U32 const nbBits = DInfo.nbBits;
+    size_t const lowBits = BIT_readBits(bitD, nbBits);
+    DStatePtr->state = DInfo.nextState + lowBits;
+}
+
+/* We need to add at most (ZSTD_WINDOWLOG_MAX_32 - 1) bits to read the maximum
+ * offset bits. But we can only read at most (STREAM_ACCUMULATOR_MIN_32 - 1)
+ * bits before reloading. This value is the maximum number of bytes we read
+ * after reloading when we are decoding long offsets.
+ */
+#define LONG_OFFSETS_MAX_EXTRA_BITS_32                       \
+    (ZSTD_WINDOWLOG_MAX_32 > STREAM_ACCUMULATOR_MIN_32       \
+        ? ZSTD_WINDOWLOG_MAX_32 - STREAM_ACCUMULATOR_MIN_32  \
+        : 0)
+
+typedef enum { ZSTD_lo_isRegularOffset, ZSTD_lo_isLongOffset=1 } ZSTD_longOffset_e;
+typedef enum { ZSTD_p_noPrefetch=0, ZSTD_p_prefetch=1 } ZSTD_prefetch_e;
+
+FORCE_INLINE_TEMPLATE seq_t
+ZSTD_decodeSequence(seqState_t* seqState, const ZSTD_longOffset_e longOffsets, const ZSTD_prefetch_e prefetch)
+{
+    seq_t seq;
+    ZSTD_seqSymbol const llDInfo = seqState->stateLL.table[seqState->stateLL.state];
+    ZSTD_seqSymbol const mlDInfo = seqState->stateML.table[seqState->stateML.state];
+    ZSTD_seqSymbol const ofDInfo = seqState->stateOffb.table[seqState->stateOffb.state];
+    U32 const llBase = llDInfo.baseValue;
+    U32 const mlBase = mlDInfo.baseValue;
+    U32 const ofBase = ofDInfo.baseValue;
+    BYTE const llBits = llDInfo.nbAdditionalBits;
+    BYTE const mlBits = mlDInfo.nbAdditionalBits;
+    BYTE const ofBits = ofDInfo.nbAdditionalBits;
+    BYTE const totalBits = llBits+mlBits+ofBits;
+
+    /* sequence */
+    {   size_t offset;
+        if (ofBits > 1) {
+            ZSTD_STATIC_ASSERT(ZSTD_lo_isLongOffset == 1);
+            ZSTD_STATIC_ASSERT(LONG_OFFSETS_MAX_EXTRA_BITS_32 == 5);
+            assert(ofBits <= MaxOff);
+            if (MEM_32bits() && longOffsets && (ofBits >= STREAM_ACCUMULATOR_MIN_32)) {
+                U32 const extraBits = ofBits - MIN(ofBits, 32 - seqState->DStream.bitsConsumed);
+                offset = ofBase + (BIT_readBitsFast(&seqState->DStream, ofBits - extraBits) << extraBits);
+                BIT_reloadDStream(&seqState->DStream);
+                if (extraBits) offset += BIT_readBitsFast(&seqState->DStream, extraBits);
+                assert(extraBits <= LONG_OFFSETS_MAX_EXTRA_BITS_32);   /* to avoid another reload */
+            } else {
+                offset = ofBase + BIT_readBitsFast(&seqState->DStream, ofBits/*>0*/);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
+                if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
+            }
+            seqState->prevOffset[2] = seqState->prevOffset[1];
+            seqState->prevOffset[1] = seqState->prevOffset[0];
+            seqState->prevOffset[0] = offset;
+        } else {
+            U32 const ll0 = (llBase == 0);
+            if (LIKELY((ofBits == 0))) {
+                if (LIKELY(!ll0))
+                    offset = seqState->prevOffset[0];
+                else {
+                    offset = seqState->prevOffset[1];
+                    seqState->prevOffset[1] = seqState->prevOffset[0];
+                    seqState->prevOffset[0] = offset;
+                }
+            } else {
+                offset = ofBase + ll0 + BIT_readBitsFast(&seqState->DStream, 1);
+                {   size_t temp = (offset==3) ? seqState->prevOffset[0] - 1 : seqState->prevOffset[offset];
+                    temp += !temp;   /* 0 is not valid; input is corrupted; force offset to 1 */
+                    if (offset != 1) seqState->prevOffset[2] = seqState->prevOffset[1];
+                    seqState->prevOffset[1] = seqState->prevOffset[0];
+                    seqState->prevOffset[0] = offset = temp;
+        }   }   }
+        seq.offset = offset;
+    }
+
+    seq.matchLength = mlBase;
+    if (mlBits > 0)
+        seq.matchLength += BIT_readBitsFast(&seqState->DStream, mlBits/*>0*/);
+
+    if (MEM_32bits() && (mlBits+llBits >= STREAM_ACCUMULATOR_MIN_32-LONG_OFFSETS_MAX_EXTRA_BITS_32))
+        BIT_reloadDStream(&seqState->DStream);
+    if (MEM_64bits() && UNLIKELY(totalBits >= STREAM_ACCUMULATOR_MIN_64-(LLFSELog+MLFSELog+OffFSELog)))
+        BIT_reloadDStream(&seqState->DStream);
+    /* Ensure there are enough bits to read the rest of data in 64-bit mode. */
+    ZSTD_STATIC_ASSERT(16+LLFSELog+MLFSELog+OffFSELog < STREAM_ACCUMULATOR_MIN_64);
+
+    seq.litLength = llBase;
+    if (llBits > 0)
+        seq.litLength += BIT_readBitsFast(&seqState->DStream, llBits/*>0*/);
+
+    if (MEM_32bits())
+        BIT_reloadDStream(&seqState->DStream);
+
+    DEBUGLOG(6, "seq: litL=%u, matchL=%u, offset=%u",
+                (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
+
+    if (prefetch == ZSTD_p_prefetch) {
+        size_t const pos = seqState->pos + seq.litLength;
+        const BYTE* const matchBase = (seq.offset > pos) ? seqState->dictEnd : seqState->prefixStart;
+        seq.match = matchBase + pos - seq.offset;  /* note : this operation can overflow when seq.offset is really too large, which can only happen when input is corrupted.
+                                                    * No consequence though : no memory access will occur, offset is only used for prefetching */
+        seqState->pos = pos + seq.matchLength;
+    }
+
+    /* ANS state update
+     * gcc-9.0.0 does 2.5% worse with ZSTD_updateFseStateWithDInfo().
+     * clang-9.2.0 does 7% worse with ZSTD_updateFseState().
+     * Naturally it seems like ZSTD_updateFseStateWithDInfo() should be the
+     * better option, so it is the default for other compilers. But, if you
+     * measure that it is worse, please put up a pull request.
+     */
+    {
+#if defined(__GNUC__) && !defined(__clang__)
+        const int kUseUpdateFseState = 1;
+#else
+        const int kUseUpdateFseState = 0;
+#endif
+        if (kUseUpdateFseState) {
+            ZSTD_updateFseState(&seqState->stateLL, &seqState->DStream);    /* <=  9 bits */
+            ZSTD_updateFseState(&seqState->stateML, &seqState->DStream);    /* <=  9 bits */
+            if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
+            ZSTD_updateFseState(&seqState->stateOffb, &seqState->DStream);  /* <=  8 bits */
+        } else {
+            ZSTD_updateFseStateWithDInfo(&seqState->stateLL, &seqState->DStream, llDInfo);    /* <=  9 bits */
+            ZSTD_updateFseStateWithDInfo(&seqState->stateML, &seqState->DStream, mlDInfo);    /* <=  9 bits */
+            if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);    /* <= 18 bits */
+            ZSTD_updateFseStateWithDInfo(&seqState->stateOffb, &seqState->DStream, ofDInfo);  /* <=  8 bits */
+        }
+    }
+
+    return seq;
+}
+
+#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+static int ZSTD_dictionaryIsActive(ZSTD_DCtx const* dctx, BYTE const* prefixStart, BYTE const* oLitEnd)
+{
+    size_t const windowSize = dctx->fParams.windowSize;
+    /* No dictionary used. */
+    if (dctx->dictContentEndForFuzzing == NULL) return 0;
+    /* Dictionary is our prefix. */
+    if (prefixStart == dctx->dictContentBeginForFuzzing) return 1;
+    /* Dictionary is not our ext-dict. */
+    if (dctx->dictEnd != dctx->dictContentEndForFuzzing) return 0;
+    /* Dictionary is not within our window size. */
+    if ((size_t)(oLitEnd - prefixStart) >= windowSize) return 0;
+    /* Dictionary is active. */
+    return 1;
+}
+
+MEM_STATIC void ZSTD_assertValidSequence(
+        ZSTD_DCtx const* dctx,
+        BYTE const* op, BYTE const* oend,
+        seq_t const seq,
+        BYTE const* prefixStart, BYTE const* virtualStart)
+{
+    size_t const windowSize = dctx->fParams.windowSize;
+    size_t const sequenceSize = seq.litLength + seq.matchLength;
+    BYTE const* const oLitEnd = op + seq.litLength;
+    DEBUGLOG(6, "Checking sequence: litL=%u matchL=%u offset=%u",
+            (U32)seq.litLength, (U32)seq.matchLength, (U32)seq.offset);
+    assert(op <= oend);
+    assert((size_t)(oend - op) >= sequenceSize);
+    assert(sequenceSize <= ZSTD_BLOCKSIZE_MAX);
+    if (ZSTD_dictionaryIsActive(dctx, prefixStart, oLitEnd)) {
+        size_t const dictSize = (size_t)((char const*)dctx->dictContentEndForFuzzing - (char const*)dctx->dictContentBeginForFuzzing);
+        /* Offset must be within the dictionary. */
+        assert(seq.offset <= (size_t)(oLitEnd - virtualStart));
+        assert(seq.offset <= windowSize + dictSize);
+    } else {
+        /* Offset must be within our window. */
+        assert(seq.offset <= windowSize);
+    }
+}
+#endif
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
+FORCE_INLINE_TEMPLATE size_t
+DONT_VECTORIZE
+ZSTD_decompressSequences_body( ZSTD_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize, int nbSeq,
+                         const ZSTD_longOffset_e isLongOffset,
+                         const int frame)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* const oend = ostart + maxDstSize;
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
+    const BYTE* const vBase = (const BYTE*) (dctx->virtualStart);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+    DEBUGLOG(5, "ZSTD_decompressSequences_body");
+    (void)frame;
+
+    /* Regen sequences */
+    if (nbSeq) {
+        seqState_t seqState;
+        size_t error = 0;
+        dctx->fseEntropy = 1;
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
+        RETURN_ERROR_IF(
+            ERR_isError(BIT_initDStream(&seqState.DStream, ip, iend-ip)),
+            corruption_detected, "");
+        ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
+        ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
+        ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
+        assert(dst != NULL);
+
+        ZSTD_STATIC_ASSERT(
+                BIT_DStream_unfinished < BIT_DStream_completed &&
+                BIT_DStream_endOfBuffer < BIT_DStream_completed &&
+                BIT_DStream_completed < BIT_DStream_overflow);
+
+#if defined(__GNUC__) && defined(__x86_64__)
+        /* Align the decompression loop to 32 + 16 bytes.
+         *
+         * zstd compiled with gcc-9 on an Intel i9-9900k shows 10% decompression
+         * speed swings based on the alignment of the decompression loop. This
+         * performance swing is caused by parts of the decompression loop falling
+         * out of the DSB. The entire decompression loop should fit in the DSB,
+         * when it can't we get much worse performance. You can measure if you've
+         * hit the good case or the bad case with this perf command for some
+         * compressed file test.zst:
+         *
+         *   perf stat -e cycles -e instructions -e idq.all_dsb_cycles_any_uops \
+         *             -e idq.all_mite_cycles_any_uops -- ./zstd -tq test.zst
+         *
+         * If you see most cycles served out of the MITE you've hit the bad case.
+         * If you see most cycles served out of the DSB you've hit the good case.
+         * If it is pretty even then you may be in an okay case.
+         *
+         * I've been able to reproduce this issue on the following CPUs:
+         *   - Kabylake: Macbook Pro (15-inch, 2019) 2.4 GHz Intel Core i9
+         *               Use Instruments->Counters to get DSB/MITE cycles.
+         *               I never got performance swings, but I was able to
+         *               go from the good case of mostly DSB to half of the
+         *               cycles served from MITE.
+         *   - Coffeelake: Intel i9-9900k
+         *
+         * I haven't been able to reproduce the instability or DSB misses on any
+         * of the following CPUS:
+         *   - Haswell
+         *   - Broadwell: Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GH
+         *   - Skylake
+         *
+         * If you are seeing performance stability this script can help test.
+         * It tests on 4 commits in zstd where I saw performance change.
+         *
+         *   https://gist.github.com/terrelln/9889fc06a423fd5ca6e99351564473f4
+         */
+        __asm__(".p2align 5");
+        __asm__("nop");
+        __asm__(".p2align 4");
+#endif
+        for ( ; ; ) {
+            seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset, ZSTD_p_noPrefetch);
+            size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequence, &litPtr, litEnd, prefixStart, vBase, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+            assert(!ZSTD_isError(oneSeqSize));
+            if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequence, prefixStart, vBase);
+#endif
+            DEBUGLOG(6, "regenerated sequence size : %u", (U32)oneSeqSize);
+            BIT_reloadDStream(&(seqState.DStream));
+            /* gcc and clang both don't like early returns in this loop.
+             * gcc doesn't like early breaks either.
+             * Instead save an error and report it at the end.
+             * When there is an error, don't increment op, so we don't
+             * overwrite.
+             */
+            if (UNLIKELY(ZSTD_isError(oneSeqSize))) error = oneSeqSize;
+            else op += oneSeqSize;
+            if (UNLIKELY(!--nbSeq)) break;
+        }
+
+        /* check if reached exact end */
+        DEBUGLOG(5, "ZSTD_decompressSequences_body: after decode loop, remaining nbSeq : %i", nbSeq);
+        if (ZSTD_isError(error)) return error;
+        RETURN_ERROR_IF(nbSeq, corruption_detected, "");
+        RETURN_ERROR_IF(BIT_reloadDStream(&seqState.DStream) < BIT_DStream_completed, corruption_detected, "");
+        /* save reps for next block */
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
+    }
+
+    /* last literal segment */
+    {   size_t const lastLLSize = litEnd - litPtr;
+        RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
+        if (op != NULL) {
+            memcpy(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return op-ostart;
+}
+
+static size_t
+ZSTD_decompressSequences_default(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset,
+                           const int frame)
+{
+    return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_decompressSequencesLong_body(
+                               ZSTD_DCtx* dctx,
+                               void* dst, size_t maxDstSize,
+                         const void* seqStart, size_t seqSize, int nbSeq,
+                         const ZSTD_longOffset_e isLongOffset,
+                         const int frame)
+{
+    const BYTE* ip = (const BYTE*)seqStart;
+    const BYTE* const iend = ip + seqSize;
+    BYTE* const ostart = (BYTE* const)dst;
+    BYTE* const oend = ostart + maxDstSize;
+    BYTE* op = ostart;
+    const BYTE* litPtr = dctx->litPtr;
+    const BYTE* const litEnd = litPtr + dctx->litSize;
+    const BYTE* const prefixStart = (const BYTE*) (dctx->prefixStart);
+    const BYTE* const dictStart = (const BYTE*) (dctx->virtualStart);
+    const BYTE* const dictEnd = (const BYTE*) (dctx->dictEnd);
+    (void)frame;
+
+    /* Regen sequences */
+    if (nbSeq) {
+#define STORED_SEQS 4
+#define STORED_SEQS_MASK (STORED_SEQS-1)
+#define ADVANCED_SEQS 4
+        seq_t sequences[STORED_SEQS];
+        int const seqAdvance = MIN(nbSeq, ADVANCED_SEQS);
+        seqState_t seqState;
+        int seqNb;
+        dctx->fseEntropy = 1;
+        { int i; for (i=0; i<ZSTD_REP_NUM; i++) seqState.prevOffset[i] = dctx->entropy.rep[i]; }
+        seqState.prefixStart = prefixStart;
+        seqState.pos = (size_t)(op-prefixStart);
+        seqState.dictEnd = dictEnd;
+        assert(dst != NULL);
+        assert(iend >= ip);
+        RETURN_ERROR_IF(
+            ERR_isError(BIT_initDStream(&seqState.DStream, ip, iend-ip)),
+            corruption_detected, "");
+        ZSTD_initFseState(&seqState.stateLL, &seqState.DStream, dctx->LLTptr);
+        ZSTD_initFseState(&seqState.stateOffb, &seqState.DStream, dctx->OFTptr);
+        ZSTD_initFseState(&seqState.stateML, &seqState.DStream, dctx->MLTptr);
+
+        /* prepare in advance */
+        for (seqNb=0; (BIT_reloadDStream(&seqState.DStream) <= BIT_DStream_completed) && (seqNb<seqAdvance); seqNb++) {
+            sequences[seqNb] = ZSTD_decodeSequence(&seqState, isLongOffset, ZSTD_p_prefetch);
+            PREFETCH_L1(sequences[seqNb].match); PREFETCH_L1(sequences[seqNb].match + sequences[seqNb].matchLength - 1); /* note : it's safe to invoke PREFETCH() on any memory address, including invalid ones */
+        }
+        RETURN_ERROR_IF(seqNb<seqAdvance, corruption_detected, "");
+
+        /* decode and decompress */
+        for ( ; (BIT_reloadDStream(&(seqState.DStream)) <= BIT_DStream_completed) && (seqNb<nbSeq) ; seqNb++) {
+            seq_t const sequence = ZSTD_decodeSequence(&seqState, isLongOffset, ZSTD_p_prefetch);
+            size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequences[(seqNb-ADVANCED_SEQS) & STORED_SEQS_MASK], &litPtr, litEnd, prefixStart, dictStart, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+            assert(!ZSTD_isError(oneSeqSize));
+            if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[(seqNb-ADVANCED_SEQS) & STORED_SEQS_MASK], prefixStart, dictStart);
+#endif
+            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+            PREFETCH_L1(sequence.match); PREFETCH_L1(sequence.match + sequence.matchLength - 1); /* note : it's safe to invoke PREFETCH() on any memory address, including invalid ones */
+            sequences[seqNb & STORED_SEQS_MASK] = sequence;
+            op += oneSeqSize;
+        }
+        RETURN_ERROR_IF(seqNb<nbSeq, corruption_detected, "");
+
+        /* finish queue */
+        seqNb -= seqAdvance;
+        for ( ; seqNb<nbSeq ; seqNb++) {
+            size_t const oneSeqSize = ZSTD_execSequence(op, oend, sequences[seqNb&STORED_SEQS_MASK], &litPtr, litEnd, prefixStart, dictStart, dictEnd);
+#if defined(FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION) && defined(FUZZING_ASSERT_VALID_SEQUENCE)
+            assert(!ZSTD_isError(oneSeqSize));
+            if (frame) ZSTD_assertValidSequence(dctx, op, oend, sequences[seqNb&STORED_SEQS_MASK], prefixStart, dictStart);
+#endif
+            if (ZSTD_isError(oneSeqSize)) return oneSeqSize;
+            op += oneSeqSize;
+        }
+
+        /* save reps for next block */
+        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) dctx->entropy.rep[i] = (U32)(seqState.prevOffset[i]); }
+    }
+
+    /* last literal segment */
+    {   size_t const lastLLSize = litEnd - litPtr;
+        RETURN_ERROR_IF(lastLLSize > (size_t)(oend-op), dstSize_tooSmall, "");
+        if (op != NULL) {
+            memcpy(op, litPtr, lastLLSize);
+            op += lastLLSize;
+        }
+    }
+
+    return op-ostart;
+}
+
+static size_t
+ZSTD_decompressSequencesLong_default(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset,
+                           const int frame)
+{
+    return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
+
+
+
+#if DYNAMIC_BMI2
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
+static TARGET_ATTRIBUTE("bmi2") size_t
+DONT_VECTORIZE
+ZSTD_decompressSequences_bmi2(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset,
+                           const int frame)
+{
+    return ZSTD_decompressSequences_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
+static TARGET_ATTRIBUTE("bmi2") size_t
+ZSTD_decompressSequencesLong_bmi2(ZSTD_DCtx* dctx,
+                                 void* dst, size_t maxDstSize,
+                           const void* seqStart, size_t seqSize, int nbSeq,
+                           const ZSTD_longOffset_e isLongOffset,
+                           const int frame)
+{
+    return ZSTD_decompressSequencesLong_body(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
+
+#endif /* DYNAMIC_BMI2 */
+
+typedef size_t (*ZSTD_decompressSequences_t)(
+                            ZSTD_DCtx* dctx,
+                            void* dst, size_t maxDstSize,
+                            const void* seqStart, size_t seqSize, int nbSeq,
+                            const ZSTD_longOffset_e isLongOffset,
+                            const int frame);
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
+static size_t
+ZSTD_decompressSequences(ZSTD_DCtx* dctx, void* dst, size_t maxDstSize,
+                   const void* seqStart, size_t seqSize, int nbSeq,
+                   const ZSTD_longOffset_e isLongOffset,
+                   const int frame)
+{
+    DEBUGLOG(5, "ZSTD_decompressSequences");
+#if DYNAMIC_BMI2
+    if (dctx->bmi2) {
+        return ZSTD_decompressSequences_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+    }
+#endif
+  return ZSTD_decompressSequences_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG */
+
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
+/* ZSTD_decompressSequencesLong() :
+ * decompression function triggered when a minimum share of offsets is considered "long",
+ * aka out of cache.
+ * note : "long" definition seems overloaded here, sometimes meaning "wider than bitstream register", and sometimes meaning "farther than memory cache distance".
+ * This function will try to mitigate main memory latency through the use of prefetching */
+static size_t
+ZSTD_decompressSequencesLong(ZSTD_DCtx* dctx,
+                             void* dst, size_t maxDstSize,
+                             const void* seqStart, size_t seqSize, int nbSeq,
+                             const ZSTD_longOffset_e isLongOffset,
+                             const int frame)
+{
+    DEBUGLOG(5, "ZSTD_decompressSequencesLong");
+#if DYNAMIC_BMI2
+    if (dctx->bmi2) {
+        return ZSTD_decompressSequencesLong_bmi2(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+    }
+#endif
+  return ZSTD_decompressSequencesLong_default(dctx, dst, maxDstSize, seqStart, seqSize, nbSeq, isLongOffset, frame);
+}
+#endif /* ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT */
+
+
+
+#if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
+    !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
+/* ZSTD_getLongOffsetsShare() :
+ * condition : offTable must be valid
+ * @return : "share" of long offsets (arbitrarily defined as > (1<<23))
+ *           compared to maximum possible of (1<<OffFSELog) */
+static unsigned
+ZSTD_getLongOffsetsShare(const ZSTD_seqSymbol* offTable)
+{
+    const void* ptr = offTable;
+    U32 const tableLog = ((const ZSTD_seqSymbol_header*)ptr)[0].tableLog;
+    const ZSTD_seqSymbol* table = offTable + 1;
+    U32 const max = 1 << tableLog;
+    U32 u, total = 0;
+    DEBUGLOG(5, "ZSTD_getLongOffsetsShare: (tableLog=%u)", tableLog);
+
+    assert(max <= (1 << OffFSELog));  /* max not too large */
+    for (u=0; u<max; u++) {
+        if (table[u].nbAdditionalBits > 22) total += 1;
+    }
+
+    assert(tableLog <= OffFSELog);
+    total <<= (OffFSELog - tableLog);  /* scale to OffFSELog */
+
+    return total;
+}
+#endif
+
+size_t
+ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
+                              void* dst, size_t dstCapacity,
+                        const void* src, size_t srcSize, const int frame)
+{   /* blockType == blockCompressed */
+    const BYTE* ip = (const BYTE*)src;
+    /* isLongOffset must be true if there are long offsets.
+     * Offsets are long if they are larger than 2^STREAM_ACCUMULATOR_MIN.
+     * We don't expect that to be the case in 64-bit mode.
+     * In block mode, window size is not known, so we have to be conservative.
+     * (note: but it could be evaluated from current-lowLimit)
+     */
+    ZSTD_longOffset_e const isLongOffset = (ZSTD_longOffset_e)(MEM_32bits() && (!frame || (dctx->fParams.windowSize > (1ULL << STREAM_ACCUMULATOR_MIN))));
+    DEBUGLOG(5, "ZSTD_decompressBlock_internal (size : %u)", (U32)srcSize);
+
+    RETURN_ERROR_IF(srcSize >= ZSTD_BLOCKSIZE_MAX, srcSize_wrong, "");
+
+    /* Decode literals section */
+    {   size_t const litCSize = ZSTD_decodeLiteralsBlock(dctx, src, srcSize);
+        DEBUGLOG(5, "ZSTD_decodeLiteralsBlock : %u", (U32)litCSize);
+        if (ZSTD_isError(litCSize)) return litCSize;
+        ip += litCSize;
+        srcSize -= litCSize;
+    }
+
+    /* Build Decoding Tables */
+    {
+        /* These macros control at build-time which decompressor implementation
+         * we use. If neither is defined, we do some inspection and dispatch at
+         * runtime.
+         */
+#if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
+    !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
+        int usePrefetchDecoder = dctx->ddictIsCold;
+#endif
+        int nbSeq;
+        size_t const seqHSize = ZSTD_decodeSeqHeaders(dctx, &nbSeq, ip, srcSize);
+        if (ZSTD_isError(seqHSize)) return seqHSize;
+        ip += seqHSize;
+        srcSize -= seqHSize;
+
+        RETURN_ERROR_IF(dst == NULL && nbSeq > 0, dstSize_tooSmall, "NULL not handled");
+
+#if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
+    !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
+        if ( !usePrefetchDecoder
+          && (!frame || (dctx->fParams.windowSize > (1<<24)))
+          && (nbSeq>ADVANCED_SEQS) ) {  /* could probably use a larger nbSeq limit */
+            U32 const shareLongOffsets = ZSTD_getLongOffsetsShare(dctx->OFTptr);
+            U32 const minShare = MEM_64bits() ? 7 : 20; /* heuristic values, correspond to 2.73% and 7.81% */
+            usePrefetchDecoder = (shareLongOffsets >= minShare);
+        }
+#endif
+
+        dctx->ddictIsCold = 0;
+
+#if !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT) && \
+    !defined(ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG)
+        if (usePrefetchDecoder)
+#endif
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_SHORT
+            return ZSTD_decompressSequencesLong(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
+#endif
+
+#ifndef ZSTD_FORCE_DECOMPRESS_SEQUENCES_LONG
+        /* else */
+        return ZSTD_decompressSequences(dctx, dst, dstCapacity, ip, srcSize, nbSeq, isLongOffset, frame);
+#endif
+    }
+}
+
+
+void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst)
+{
+    if (dst != dctx->previousDstEnd) {   /* not contiguous */
+        dctx->dictEnd = dctx->previousDstEnd;
+        dctx->virtualStart = (const char*)dst - ((const char*)(dctx->previousDstEnd) - (const char*)(dctx->prefixStart));
+        dctx->prefixStart = dst;
+        dctx->previousDstEnd = dst;
+    }
+}
+
+
+size_t ZSTD_decompressBlock(ZSTD_DCtx* dctx,
+                            void* dst, size_t dstCapacity,
+                      const void* src, size_t srcSize)
+{
+    size_t dSize;
+    ZSTD_checkContinuity(dctx, dst);
+    dSize = ZSTD_decompressBlock_internal(dctx, dst, dstCapacity, src, srcSize, /* frame */ 0);
+    dctx->previousDstEnd = (char*)dst + dSize;
+    return dSize;
+}
diff --git a/contrib/zstd/zstd_decompress_block.h b/contrib/zstd/zstd_decompress_block.h
new file mode 100644
index 000000000..a065f8d93
--- /dev/null
+++ b/contrib/zstd/zstd_decompress_block.h
@@ -0,0 +1,59 @@
+/*
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+#ifndef ZSTD_DEC_BLOCK_H
+#define ZSTD_DEC_BLOCK_H
+
+/*-*******************************************************
+ *  Dependencies
+ *********************************************************/
+#include <stddef.h>   /* size_t */
+#include "zstd.h"    /* DCtx, and some public functions */
+#include "zstd_internal.h"  /* blockProperties_t, and some public functions */
+#include "zstd_decompress_internal.h"  /* ZSTD_seqSymbol */
+
+
+/* ===   Prototypes   === */
+
+/* note: prototypes already published within `zstd.h` :
+ * ZSTD_decompressBlock()
+ */
+
+/* note: prototypes already published within `zstd_internal.h` :
+ * ZSTD_getcBlockSize()
+ * ZSTD_decodeSeqHeaders()
+ */
+
+
+/* ZSTD_decompressBlock_internal() :
+ * decompress block, starting at `src`,
+ * into destination buffer `dst`.
+ * @return : decompressed block size,
+ *           or an error code (which can be tested using ZSTD_isError())
+ */
+size_t ZSTD_decompressBlock_internal(ZSTD_DCtx* dctx,
+                               void* dst, size_t dstCapacity,
+                         const void* src, size_t srcSize, const int frame);
+
+/* ZSTD_buildFSETable() :
+ * generate FSE decoding table for one symbol (ll, ml or off)
+ * this function must be called with valid parameters only
+ * (dt is large enough, normalizedCounter distribution total is a power of 2, max is within range, etc.)
+ * in which case it cannot fail.
+ * Internal use only.
+ */
+void ZSTD_buildFSETable(ZSTD_seqSymbol* dt,
+             const short* normalizedCounter, unsigned maxSymbolValue,
+             const U32* baseValue, const U32* nbAdditionalBits,
+                   unsigned tableLog);
+
+
+#endif /* ZSTD_DEC_BLOCK_H */
diff --git a/contrib/zstd/zstd_decompress_internal.h b/contrib/zstd/zstd_decompress_internal.h
new file mode 100644
index 000000000..a5cafc04e
--- /dev/null
+++ b/contrib/zstd/zstd_decompress_internal.h
@@ -0,0 +1,189 @@
+/*
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
+ * All rights reserved.
+ *
+ * This source code is licensed under both the BSD-style license (found in the
+ * LICENSE file in the root directory of this source tree) and the GPLv2 (found
+ * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
+ */
+
+
+/* zstd_decompress_internal:
+ * objects and definitions shared within lib/decompress modules */
+
+ #ifndef ZSTD_DECOMPRESS_INTERNAL_H
+ #define ZSTD_DECOMPRESS_INTERNAL_H
+
+
+/*-*******************************************************
+ *  Dependencies
+ *********************************************************/
+#include "mem.h"             /* BYTE, U16, U32 */
+#include "zstd_internal.h"   /* ZSTD_seqSymbol */
+
+
+
+/*-*******************************************************
+ *  Constants
+ *********************************************************/
+static const U32 LL_base[MaxLL+1] = {
+                 0,    1,    2,     3,     4,     5,     6,      7,
+                 8,    9,   10,    11,    12,    13,    14,     15,
+                16,   18,   20,    22,    24,    28,    32,     40,
+                48,   64, 0x80, 0x100, 0x200, 0x400, 0x800, 0x1000,
+                0x2000, 0x4000, 0x8000, 0x10000 };
+
+static const U32 OF_base[MaxOff+1] = {
+                 0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
+                 0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
+                 0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
+                 0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD, 0x1FFFFFFD, 0x3FFFFFFD, 0x7FFFFFFD };
+
+static const U32 OF_bits[MaxOff+1] = {
+                     0,  1,  2,  3,  4,  5,  6,  7,
+                     8,  9, 10, 11, 12, 13, 14, 15,
+                    16, 17, 18, 19, 20, 21, 22, 23,
+                    24, 25, 26, 27, 28, 29, 30, 31 };
+
+static const U32 ML_base[MaxML+1] = {
+                     3,  4,  5,    6,     7,     8,     9,    10,
+                    11, 12, 13,   14,    15,    16,    17,    18,
+                    19, 20, 21,   22,    23,    24,    25,    26,
+                    27, 28, 29,   30,    31,    32,    33,    34,
+                    35, 37, 39,   41,    43,    47,    51,    59,
+                    67, 83, 99, 0x83, 0x103, 0x203, 0x403, 0x803,
+                    0x1003, 0x2003, 0x4003, 0x8003, 0x10003 };
+
+
+/*-*******************************************************
+ *  Decompression types
+ *********************************************************/
+ typedef struct {
+     U32 fastMode;
+     U32 tableLog;
+ } ZSTD_seqSymbol_header;
+
+ typedef struct {
+     U16  nextState;
+     BYTE nbAdditionalBits;
+     BYTE nbBits;
+     U32  baseValue;
+ } ZSTD_seqSymbol;
+
+ #define SEQSYMBOL_TABLE_SIZE(log)   (1 + (1 << (log)))
+
+typedef struct {
+    ZSTD_seqSymbol LLTable[SEQSYMBOL_TABLE_SIZE(LLFSELog)];    /* Note : Space reserved for FSE Tables */
+    ZSTD_seqSymbol OFTable[SEQSYMBOL_TABLE_SIZE(OffFSELog)];   /* is also used as temporary workspace while building hufTable during DDict creation */
+    ZSTD_seqSymbol MLTable[SEQSYMBOL_TABLE_SIZE(MLFSELog)];    /* and therefore must be at least HUF_DECOMPRESS_WORKSPACE_SIZE large */
+    HUF_DTable hufTable[HUF_DTABLE_SIZE(HufLog)];  /* can accommodate HUF_decompress4X */
+    U32 rep[ZSTD_REP_NUM];
+} ZSTD_entropyDTables_t;
+
+typedef enum { ZSTDds_getFrameHeaderSize, ZSTDds_decodeFrameHeader,
+               ZSTDds_decodeBlockHeader, ZSTDds_decompressBlock,
+               ZSTDds_decompressLastBlock, ZSTDds_checkChecksum,
+               ZSTDds_decodeSkippableHeader, ZSTDds_skipFrame } ZSTD_dStage;
+
+typedef enum { zdss_init=0, zdss_loadHeader,
+               zdss_read, zdss_load, zdss_flush } ZSTD_dStreamStage;
+
+typedef enum {
+    ZSTD_use_indefinitely = -1,  /* Use the dictionary indefinitely */
+    ZSTD_dont_use = 0,           /* Do not use the dictionary (if one exists free it) */
+    ZSTD_use_once = 1            /* Use the dictionary once and set to ZSTD_dont_use */
+} ZSTD_dictUses_e;
+
+typedef enum {
+    ZSTD_obm_buffered = 0,  /* Buffer the output */
+    ZSTD_obm_stable = 1     /* ZSTD_outBuffer is stable */
+} ZSTD_outBufferMode_e;
+
+struct ZSTD_DCtx_s
+{
+    const ZSTD_seqSymbol* LLTptr;
+    const ZSTD_seqSymbol* MLTptr;
+    const ZSTD_seqSymbol* OFTptr;
+    const HUF_DTable* HUFptr;
+    ZSTD_entropyDTables_t entropy;
+    U32 workspace[HUF_DECOMPRESS_WORKSPACE_SIZE_U32];   /* space needed when building huffman tables */
+    const void* previousDstEnd;   /* detect continuity */
+    const void* prefixStart;      /* start of current segment */
+    const void* virtualStart;     /* virtual start of previous segment if it was just before current one */
+    const void* dictEnd;          /* end of previous segment */
+    size_t expected;
+    ZSTD_frameHeader fParams;
+    U64 decodedSize;
+    blockType_e bType;            /* used in ZSTD_decompressContinue(), store blockType between block header decoding and block decompression stages */
+    ZSTD_dStage stage;
+    U32 litEntropy;
+    U32 fseEntropy;
+    XXH64_state_t xxhState;
+    size_t headerSize;
+    ZSTD_format_e format;
+    const BYTE* litPtr;
+    ZSTD_customMem customMem;
+    size_t litSize;
+    size_t rleSize;
+    size_t staticSize;
+    int bmi2;                     /* == 1 if the CPU supports BMI2 and 0 otherwise. CPU support is determined dynamically once per context lifetime. */
+
+    /* dictionary */
+    ZSTD_DDict* ddictLocal;
+    const ZSTD_DDict* ddict;     /* set by ZSTD_initDStream_usingDDict(), or ZSTD_DCtx_refDDict() */
+    U32 dictID;
+    int ddictIsCold;             /* if == 1 : dictionary is "new" for working context, and presumed "cold" (not in cpu cache) */
+    ZSTD_dictUses_e dictUses;
+
+    /* streaming */
+    ZSTD_dStreamStage streamStage;
+    char*  inBuff;
+    size_t inBuffSize;
+    size_t inPos;
+    size_t maxWindowSize;
+    char*  outBuff;
+    size_t outBuffSize;
+    size_t outStart;
+    size_t outEnd;
+    size_t lhSize;
+    void* legacyContext;
+    U32 previousLegacyVersion;
+    U32 legacyVersion;
+    U32 hostageByte;
+    int noForwardProgress;
+    ZSTD_outBufferMode_e outBufferMode;
+    ZSTD_outBuffer expectedOutBuffer;
+
+    /* workspace */
+    BYTE litBuffer[ZSTD_BLOCKSIZE_MAX + WILDCOPY_OVERLENGTH];
+    BYTE headerBuffer[ZSTD_FRAMEHEADERSIZE_MAX];
+
+    size_t oversizedDuration;
+
+#ifdef FUZZING_BUILD_MODE_UNSAFE_FOR_PRODUCTION
+    void const* dictContentBeginForFuzzing;
+    void const* dictContentEndForFuzzing;
+#endif
+};  /* typedef'd to ZSTD_DCtx within "zstd.h" */
+
+
+/*-*******************************************************
+ *  Shared internal functions
+ *********************************************************/
+
+/*! ZSTD_loadDEntropy() :
+ *  dict : must point at beginning of a valid zstd dictionary.
+ * @return : size of dictionary header (size of magic number + dict ID + entropy tables) */
+size_t ZSTD_loadDEntropy(ZSTD_entropyDTables_t* entropy,
+                   const void* const dict, size_t const dictSize);
+
+/*! ZSTD_checkContinuity() :
+ *  check if next `dst` follows previous position, where decompression ended.
+ *  If yes, do nothing (continue on current segment).
+ *  If not, classify previous segment as "external dictionary", and start a new segment.
+ *  This function cannot fail. */
+void ZSTD_checkContinuity(ZSTD_DCtx* dctx, const void* dst);
+
+
+#endif /* ZSTD_DECOMPRESS_INTERNAL_H */
diff --git a/contrib/zstd/zstd_double_fast.c b/contrib/zstd/zstd_double_fast.c
index 876a36042..27eed66cf 100644
--- a/contrib/zstd/zstd_double_fast.c
+++ b/contrib/zstd/zstd_double_fast.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -8,270 +8,481 @@
  * You may select, at your option, one of the above-listed licenses.
  */
 
+#include "zstd_compress_internal.h"
 #include "zstd_double_fast.h"
 
 
-void ZSTD_fillDoubleHashTable(ZSTD_CCtx* cctx, const void* end, const U32 mls)
+void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
+                              void const* end, ZSTD_dictTableLoadMethod_e dtlm)
 {
-    U32* const hashLarge = cctx->hashTable;
-    U32  const hBitsL = cctx->appliedParams.cParams.hashLog;
-    U32* const hashSmall = cctx->chainTable;
-    U32  const hBitsS = cctx->appliedParams.cParams.chainLog;
-    const BYTE* const base = cctx->base;
-    const BYTE* ip = base + cctx->nextToUpdate;
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashLarge = ms->hashTable;
+    U32  const hBitsL = cParams->hashLog;
+    U32  const mls = cParams->minMatch;
+    U32* const hashSmall = ms->chainTable;
+    U32  const hBitsS = cParams->chainLog;
+    const BYTE* const base = ms->window.base;
+    const BYTE* ip = base + ms->nextToUpdate;
     const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
-    const size_t fastHashFillStep = 3;
-
-    while(ip <= iend) {
-        hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip - base);
-        hashLarge[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip - base);
-        ip += fastHashFillStep;
-    }
+    const U32 fastHashFillStep = 3;
+
+    /* Always insert every fastHashFillStep position into the hash tables.
+     * Insert the other positions into the large hash table if their entry
+     * is empty.
+     */
+    for (; ip + fastHashFillStep - 1 <= iend; ip += fastHashFillStep) {
+        U32 const current = (U32)(ip - base);
+        U32 i;
+        for (i = 0; i < fastHashFillStep; ++i) {
+            size_t const smHash = ZSTD_hashPtr(ip + i, hBitsS, mls);
+            size_t const lgHash = ZSTD_hashPtr(ip + i, hBitsL, 8);
+            if (i == 0)
+                hashSmall[smHash] = current + i;
+            if (i == 0 || hashLarge[lgHash] == 0)
+                hashLarge[lgHash] = current + i;
+            /* Only load extra positions for ZSTD_dtlm_full */
+            if (dtlm == ZSTD_dtlm_fast)
+                break;
+    }   }
 }
 
 
 FORCE_INLINE_TEMPLATE
-size_t ZSTD_compressBlock_doubleFast_generic(ZSTD_CCtx* cctx,
-                                 const void* src, size_t srcSize,
-                                 const U32 mls)
+size_t ZSTD_compressBlock_doubleFast_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize,
+        U32 const mls /* template */, ZSTD_dictMode_e const dictMode)
 {
-    U32* const hashLong = cctx->hashTable;
-    const U32 hBitsL = cctx->appliedParams.cParams.hashLog;
-    U32* const hashSmall = cctx->chainTable;
-    const U32 hBitsS = cctx->appliedParams.cParams.chainLog;
-    seqStore_t* seqStorePtr = &(cctx->seqStore);
-    const BYTE* const base = cctx->base;
+    ZSTD_compressionParameters const* cParams = &ms->cParams;
+    U32* const hashLong = ms->hashTable;
+    const U32 hBitsL = cParams->hashLog;
+    U32* const hashSmall = ms->chainTable;
+    const U32 hBitsS = cParams->chainLog;
+    const BYTE* const base = ms->window.base;
     const BYTE* const istart = (const BYTE*)src;
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
-    const U32 lowestIndex = cctx->dictLimit;
-    const BYTE* const lowest = base + lowestIndex;
+    const U32 endIndex = (U32)((size_t)(istart - base) + srcSize);
+    /* presumes that, if there is a dictionary, it must be using Attach mode */
+    const U32 prefixLowestIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
+    const BYTE* const prefixLowest = base + prefixLowestIndex;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - HASH_READ_SIZE;
-    U32 offset_1=seqStorePtr->rep[0], offset_2=seqStorePtr->rep[1];
+    U32 offset_1=rep[0], offset_2=rep[1];
     U32 offsetSaved = 0;
 
+    const ZSTD_matchState_t* const dms = ms->dictMatchState;
+    const ZSTD_compressionParameters* const dictCParams =
+                                     dictMode == ZSTD_dictMatchState ?
+                                     &dms->cParams : NULL;
+    const U32* const dictHashLong  = dictMode == ZSTD_dictMatchState ?
+                                     dms->hashTable : NULL;
+    const U32* const dictHashSmall = dictMode == ZSTD_dictMatchState ?
+                                     dms->chainTable : NULL;
+    const U32 dictStartIndex       = dictMode == ZSTD_dictMatchState ?
+                                     dms->window.dictLimit : 0;
+    const BYTE* const dictBase     = dictMode == ZSTD_dictMatchState ?
+                                     dms->window.base : NULL;
+    const BYTE* const dictStart    = dictMode == ZSTD_dictMatchState ?
+                                     dictBase + dictStartIndex : NULL;
+    const BYTE* const dictEnd      = dictMode == ZSTD_dictMatchState ?
+                                     dms->window.nextSrc : NULL;
+    const U32 dictIndexDelta       = dictMode == ZSTD_dictMatchState ?
+                                     prefixLowestIndex - (U32)(dictEnd - dictBase) :
+                                     0;
+    const U32 dictHBitsL           = dictMode == ZSTD_dictMatchState ?
+                                     dictCParams->hashLog : hBitsL;
+    const U32 dictHBitsS           = dictMode == ZSTD_dictMatchState ?
+                                     dictCParams->chainLog : hBitsS;
+    const U32 dictAndPrefixLength  = (U32)((ip - prefixLowest) + (dictEnd - dictStart));
+
+    DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_generic");
+
+    assert(dictMode == ZSTD_noDict || dictMode == ZSTD_dictMatchState);
+
+    /* if a dictionary is attached, it must be within window range */
+    if (dictMode == ZSTD_dictMatchState) {
+        assert(ms->window.dictLimit + (1U << cParams->windowLog) >= endIndex);
+    }
+
     /* init */
-    ip += (ip==lowest);
-    {   U32 const maxRep = (U32)(ip-lowest);
+    ip += (dictAndPrefixLength == 0);
+    if (dictMode == ZSTD_noDict) {
+        U32 const current = (U32)(ip - base);
+        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
+        U32 const maxRep = current - windowLow;
         if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
         if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
     }
+    if (dictMode == ZSTD_dictMatchState) {
+        /* dictMatchState repCode checks don't currently handle repCode == 0
+         * disabling. */
+        assert(offset_1 <= dictAndPrefixLength);
+        assert(offset_2 <= dictAndPrefixLength);
+    }
 
     /* Main Search Loop */
     while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
         size_t mLength;
+        U32 offset;
         size_t const h2 = ZSTD_hashPtr(ip, hBitsL, 8);
         size_t const h = ZSTD_hashPtr(ip, hBitsS, mls);
+        size_t const dictHL = ZSTD_hashPtr(ip, dictHBitsL, 8);
+        size_t const dictHS = ZSTD_hashPtr(ip, dictHBitsS, mls);
         U32 const current = (U32)(ip-base);
         U32 const matchIndexL = hashLong[h2];
-        U32 const matchIndexS = hashSmall[h];
+        U32 matchIndexS = hashSmall[h];
         const BYTE* matchLong = base + matchIndexL;
         const BYTE* match = base + matchIndexS;
+        const U32 repIndex = current + 1 - offset_1;
+        const BYTE* repMatch = (dictMode == ZSTD_dictMatchState
+                            && repIndex < prefixLowestIndex) ?
+                               dictBase + (repIndex - dictIndexDelta) :
+                               base + repIndex;
         hashLong[h2] = hashSmall[h] = current;   /* update hash tables */
 
-        assert(offset_1 <= current);   /* supposed guaranteed by construction */
-        if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
-            /* favor repcode */
+        /* check dictMatchState repcode */
+        if (dictMode == ZSTD_dictMatchState
+            && ((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
+            && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+            const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
+            ip++;
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, mLength-MINMATCH);
+            goto _match_stored;
+        }
+
+        /* check noDict repcode */
+        if ( dictMode == ZSTD_noDict
+          && ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1)))) {
             mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
             ip++;
-            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mLength-MINMATCH);
-        } else {
-            U32 offset;
-            if ( (matchIndexL > lowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip)) ) {
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, mLength-MINMATCH);
+            goto _match_stored;
+        }
+
+        if (matchIndexL > prefixLowestIndex) {
+            /* check prefix long match */
+            if (MEM_read64(matchLong) == MEM_read64(ip)) {
                 mLength = ZSTD_count(ip+8, matchLong+8, iend) + 8;
                 offset = (U32)(ip-matchLong);
-                while (((ip>anchor) & (matchLong>lowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
-            } else if ( (matchIndexS > lowestIndex) && (MEM_read32(match) == MEM_read32(ip)) ) {
-                size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
-                U32 const matchIndexL3 = hashLong[hl3];
-                const BYTE* matchL3 = base + matchIndexL3;
-                hashLong[hl3] = current + 1;
-                if ( (matchIndexL3 > lowestIndex) && (MEM_read64(matchL3) == MEM_read64(ip+1)) ) {
+                while (((ip>anchor) & (matchLong>prefixLowest)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; } /* catch up */
+                goto _match_found;
+            }
+        } else if (dictMode == ZSTD_dictMatchState) {
+            /* check dictMatchState long match */
+            U32 const dictMatchIndexL = dictHashLong[dictHL];
+            const BYTE* dictMatchL = dictBase + dictMatchIndexL;
+            assert(dictMatchL < dictEnd);
+
+            if (dictMatchL > dictStart && MEM_read64(dictMatchL) == MEM_read64(ip)) {
+                mLength = ZSTD_count_2segments(ip+8, dictMatchL+8, iend, dictEnd, prefixLowest) + 8;
+                offset = (U32)(current - dictMatchIndexL - dictIndexDelta);
+                while (((ip>anchor) & (dictMatchL>dictStart)) && (ip[-1] == dictMatchL[-1])) { ip--; dictMatchL--; mLength++; } /* catch up */
+                goto _match_found;
+        }   }
+
+        if (matchIndexS > prefixLowestIndex) {
+            /* check prefix short match */
+            if (MEM_read32(match) == MEM_read32(ip)) {
+                goto _search_next_long;
+            }
+        } else if (dictMode == ZSTD_dictMatchState) {
+            /* check dictMatchState short match */
+            U32 const dictMatchIndexS = dictHashSmall[dictHS];
+            match = dictBase + dictMatchIndexS;
+            matchIndexS = dictMatchIndexS + dictIndexDelta;
+
+            if (match > dictStart && MEM_read32(match) == MEM_read32(ip)) {
+                goto _search_next_long;
+        }   }
+
+        ip += ((ip-anchor) >> kSearchStrength) + 1;
+#if defined(__aarch64__)
+        PREFETCH_L1(ip+256);
+#endif
+        continue;
+
+_search_next_long:
+
+        {   size_t const hl3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
+            size_t const dictHLNext = ZSTD_hashPtr(ip+1, dictHBitsL, 8);
+            U32 const matchIndexL3 = hashLong[hl3];
+            const BYTE* matchL3 = base + matchIndexL3;
+            hashLong[hl3] = current + 1;
+
+            /* check prefix long +1 match */
+            if (matchIndexL3 > prefixLowestIndex) {
+                if (MEM_read64(matchL3) == MEM_read64(ip+1)) {
                     mLength = ZSTD_count(ip+9, matchL3+8, iend) + 8;
                     ip++;
                     offset = (U32)(ip-matchL3);
-                    while (((ip>anchor) & (matchL3>lowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
-                } else {
-                    mLength = ZSTD_count(ip+4, match+4, iend) + 4;
-                    offset = (U32)(ip-match);
-                    while (((ip>anchor) & (match>lowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+                    while (((ip>anchor) & (matchL3>prefixLowest)) && (ip[-1] == matchL3[-1])) { ip--; matchL3--; mLength++; } /* catch up */
+                    goto _match_found;
                 }
-            } else {
-                ip += ((ip-anchor) >> g_searchStrength) + 1;
-                continue;
-            }
+            } else if (dictMode == ZSTD_dictMatchState) {
+                /* check dict long +1 match */
+                U32 const dictMatchIndexL3 = dictHashLong[dictHLNext];
+                const BYTE* dictMatchL3 = dictBase + dictMatchIndexL3;
+                assert(dictMatchL3 < dictEnd);
+                if (dictMatchL3 > dictStart && MEM_read64(dictMatchL3) == MEM_read64(ip+1)) {
+                    mLength = ZSTD_count_2segments(ip+1+8, dictMatchL3+8, iend, dictEnd, prefixLowest) + 8;
+                    ip++;
+                    offset = (U32)(current + 1 - dictMatchIndexL3 - dictIndexDelta);
+                    while (((ip>anchor) & (dictMatchL3>dictStart)) && (ip[-1] == dictMatchL3[-1])) { ip--; dictMatchL3--; mLength++; } /* catch up */
+                    goto _match_found;
+        }   }   }
+
+        /* if no long +1 match, explore the short match we found */
+        if (dictMode == ZSTD_dictMatchState && matchIndexS < prefixLowestIndex) {
+            mLength = ZSTD_count_2segments(ip+4, match+4, iend, dictEnd, prefixLowest) + 4;
+            offset = (U32)(current - matchIndexS);
+            while (((ip>anchor) & (match>dictStart)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+        } else {
+            mLength = ZSTD_count(ip+4, match+4, iend) + 4;
+            offset = (U32)(ip - match);
+            while (((ip>anchor) & (match>prefixLowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+        }
 
-            offset_2 = offset_1;
-            offset_1 = offset;
+        /* fall-through */
 
-            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
-        }
+_match_found:
+        offset_2 = offset_1;
+        offset_1 = offset;
+
+        ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
 
+_match_stored:
         /* match found */
         ip += mLength;
         anchor = ip;
 
         if (ip <= ilimit) {
-            /* Fill Table */
-            hashLong[ZSTD_hashPtr(base+current+2, hBitsL, 8)] =
-                hashSmall[ZSTD_hashPtr(base+current+2, hBitsS, mls)] = current+2;  /* here because current+2 could be > iend-8 */
-            hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] =
-                hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base);
+            /* Complementary insertion */
+            /* done after iLimit test, as candidates could be > iend-8 */
+            {   U32 const indexToInsert = current+2;
+                hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
+                hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
+                hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
+                hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
+            }
 
             /* check immediate repcode */
-            while ( (ip <= ilimit)
-                 && ( (offset_2>0)
-                 & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
-                /* store sequence */
-                size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
-                { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; } /* swap offset_2 <=> offset_1 */
-                hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
-                hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
-                ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, rLength-MINMATCH);
-                ip += rLength;
-                anchor = ip;
-                continue;   /* faster when present ... (?) */
-    }   }   }
+            if (dictMode == ZSTD_dictMatchState) {
+                while (ip <= ilimit) {
+                    U32 const current2 = (U32)(ip-base);
+                    U32 const repIndex2 = current2 - offset_2;
+                    const BYTE* repMatch2 = dictMode == ZSTD_dictMatchState
+                        && repIndex2 < prefixLowestIndex ?
+                            dictBase + repIndex2 - dictIndexDelta :
+                            base + repIndex2;
+                    if ( ((U32)((prefixLowestIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
+                       && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                        const BYTE* const repEnd2 = repIndex2 < prefixLowestIndex ? dictEnd : iend;
+                        size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixLowest) + 4;
+                        U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                        ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, repLength2-MINMATCH);
+                        hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
+                        hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
+                        ip += repLength2;
+                        anchor = ip;
+                        continue;
+                    }
+                    break;
+            }   }
+
+            if (dictMode == ZSTD_noDict) {
+                while ( (ip <= ilimit)
+                     && ( (offset_2>0)
+                        & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
+                    /* store sequence */
+                    size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+                    U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff;  /* swap offset_2 <=> offset_1 */
+                    hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = (U32)(ip-base);
+                    hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = (U32)(ip-base);
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, rLength-MINMATCH);
+                    ip += rLength;
+                    anchor = ip;
+                    continue;   /* faster when present ... (?) */
+        }   }   }
+    }   /* while (ip < ilimit) */
 
     /* save reps for next block */
-    seqStorePtr->repToConfirm[0] = offset_1 ? offset_1 : offsetSaved;
-    seqStorePtr->repToConfirm[1] = offset_2 ? offset_2 : offsetSaved;
+    rep[0] = offset_1 ? offset_1 : offsetSaved;
+    rep[1] = offset_2 ? offset_2 : offsetSaved;
 
     /* Return the last literals size */
-    return iend - anchor;
+    return (size_t)(iend - anchor);
 }
 
 
-size_t ZSTD_compressBlock_doubleFast(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+size_t ZSTD_compressBlock_doubleFast(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
 {
-    const U32 mls = ctx->appliedParams.cParams.searchLength;
+    const U32 mls = ms->cParams.minMatch;
     switch(mls)
     {
     default: /* includes case 3 */
     case 4 :
-        return ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 4);
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 4, ZSTD_noDict);
     case 5 :
-        return ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 5);
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 5, ZSTD_noDict);
     case 6 :
-        return ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 6);
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 6, ZSTD_noDict);
     case 7 :
-        return ZSTD_compressBlock_doubleFast_generic(ctx, src, srcSize, 7);
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 7, ZSTD_noDict);
     }
 }
 
 
-static size_t ZSTD_compressBlock_doubleFast_extDict_generic(ZSTD_CCtx* ctx,
-                                 const void* src, size_t srcSize,
-                                 const U32 mls)
+size_t ZSTD_compressBlock_doubleFast_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
 {
-    U32* const hashLong = ctx->hashTable;
-    U32  const hBitsL = ctx->appliedParams.cParams.hashLog;
-    U32* const hashSmall = ctx->chainTable;
-    U32  const hBitsS = ctx->appliedParams.cParams.chainLog;
-    seqStore_t* seqStorePtr = &(ctx->seqStore);
-    const BYTE* const base = ctx->base;
-    const BYTE* const dictBase = ctx->dictBase;
+    const U32 mls = ms->cParams.minMatch;
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 4, ZSTD_dictMatchState);
+    case 5 :
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 5, ZSTD_dictMatchState);
+    case 6 :
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 6, ZSTD_dictMatchState);
+    case 7 :
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, 7, ZSTD_dictMatchState);
+    }
+}
+
+
+static size_t ZSTD_compressBlock_doubleFast_extDict_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize,
+        U32 const mls /* template */)
+{
+    ZSTD_compressionParameters const* cParams = &ms->cParams;
+    U32* const hashLong = ms->hashTable;
+    U32  const hBitsL = cParams->hashLog;
+    U32* const hashSmall = ms->chainTable;
+    U32  const hBitsS = cParams->chainLog;
     const BYTE* const istart = (const BYTE*)src;
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
-    const U32   lowestIndex = ctx->lowLimit;
-    const BYTE* const dictStart = dictBase + lowestIndex;
-    const U32   dictLimit = ctx->dictLimit;
-    const BYTE* const lowPrefixPtr = base + dictLimit;
-    const BYTE* const dictEnd = dictBase + dictLimit;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - 8;
-    U32 offset_1=seqStorePtr->rep[0], offset_2=seqStorePtr->rep[1];
+    const BYTE* const base = ms->window.base;
+    const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
+    const U32   lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);
+    const U32   dictStartIndex = lowLimit;
+    const U32   dictLimit = ms->window.dictLimit;
+    const U32   prefixStartIndex = (dictLimit > lowLimit) ? dictLimit : lowLimit;
+    const BYTE* const prefixStart = base + prefixStartIndex;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const BYTE* const dictStart = dictBase + dictStartIndex;
+    const BYTE* const dictEnd = dictBase + prefixStartIndex;
+    U32 offset_1=rep[0], offset_2=rep[1];
+
+    DEBUGLOG(5, "ZSTD_compressBlock_doubleFast_extDict_generic (srcSize=%zu)", srcSize);
+
+    /* if extDict is invalidated due to maxDistance, switch to "regular" variant */
+    if (prefixStartIndex == dictStartIndex)
+        return ZSTD_compressBlock_doubleFast_generic(ms, seqStore, rep, src, srcSize, mls, ZSTD_noDict);
 
     /* Search Loop */
     while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
         const size_t hSmall = ZSTD_hashPtr(ip, hBitsS, mls);
         const U32 matchIndex = hashSmall[hSmall];
-        const BYTE* matchBase = matchIndex < dictLimit ? dictBase : base;
+        const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base;
         const BYTE* match = matchBase + matchIndex;
 
         const size_t hLong = ZSTD_hashPtr(ip, hBitsL, 8);
         const U32 matchLongIndex = hashLong[hLong];
-        const BYTE* matchLongBase = matchLongIndex < dictLimit ? dictBase : base;
+        const BYTE* const matchLongBase = matchLongIndex < prefixStartIndex ? dictBase : base;
         const BYTE* matchLong = matchLongBase + matchLongIndex;
 
         const U32 current = (U32)(ip-base);
         const U32 repIndex = current + 1 - offset_1;   /* offset_1 expected <= current +1 */
-        const BYTE* repBase = repIndex < dictLimit ? dictBase : base;
-        const BYTE* repMatch = repBase + repIndex;
+        const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
+        const BYTE* const repMatch = repBase + repIndex;
         size_t mLength;
         hashSmall[hSmall] = hashLong[hLong] = current;   /* update hash table */
 
-        if ( (((U32)((dictLimit-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > lowestIndex))
-           && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
-            const BYTE* repMatchEnd = repIndex < dictLimit ? dictEnd : iend;
-            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, lowPrefixPtr) + 4;
+        if ((((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex doesn't overlap dict + prefix */
+            & (repIndex > dictStartIndex))
+          && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+            const BYTE* repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
             ip++;
-            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mLength-MINMATCH);
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, mLength-MINMATCH);
         } else {
-            if ((matchLongIndex > lowestIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
-                const BYTE* matchEnd = matchLongIndex < dictLimit ? dictEnd : iend;
-                const BYTE* lowMatchPtr = matchLongIndex < dictLimit ? dictStart : lowPrefixPtr;
+            if ((matchLongIndex > dictStartIndex) && (MEM_read64(matchLong) == MEM_read64(ip))) {
+                const BYTE* const matchEnd = matchLongIndex < prefixStartIndex ? dictEnd : iend;
+                const BYTE* const lowMatchPtr = matchLongIndex < prefixStartIndex ? dictStart : prefixStart;
                 U32 offset;
-                mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, lowPrefixPtr) + 8;
+                mLength = ZSTD_count_2segments(ip+8, matchLong+8, iend, matchEnd, prefixStart) + 8;
                 offset = current - matchLongIndex;
                 while (((ip>anchor) & (matchLong>lowMatchPtr)) && (ip[-1] == matchLong[-1])) { ip--; matchLong--; mLength++; }   /* catch up */
                 offset_2 = offset_1;
                 offset_1 = offset;
-                ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
 
-            } else if ((matchIndex > lowestIndex) && (MEM_read32(match) == MEM_read32(ip))) {
+            } else if ((matchIndex > dictStartIndex) && (MEM_read32(match) == MEM_read32(ip))) {
                 size_t const h3 = ZSTD_hashPtr(ip+1, hBitsL, 8);
                 U32 const matchIndex3 = hashLong[h3];
-                const BYTE* const match3Base = matchIndex3 < dictLimit ? dictBase : base;
+                const BYTE* const match3Base = matchIndex3 < prefixStartIndex ? dictBase : base;
                 const BYTE* match3 = match3Base + matchIndex3;
                 U32 offset;
                 hashLong[h3] = current + 1;
-                if ( (matchIndex3 > lowestIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
-                    const BYTE* matchEnd = matchIndex3 < dictLimit ? dictEnd : iend;
-                    const BYTE* lowMatchPtr = matchIndex3 < dictLimit ? dictStart : lowPrefixPtr;
-                    mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, lowPrefixPtr) + 8;
+                if ( (matchIndex3 > dictStartIndex) && (MEM_read64(match3) == MEM_read64(ip+1)) ) {
+                    const BYTE* const matchEnd = matchIndex3 < prefixStartIndex ? dictEnd : iend;
+                    const BYTE* const lowMatchPtr = matchIndex3 < prefixStartIndex ? dictStart : prefixStart;
+                    mLength = ZSTD_count_2segments(ip+9, match3+8, iend, matchEnd, prefixStart) + 8;
                     ip++;
                     offset = current+1 - matchIndex3;
                     while (((ip>anchor) & (match3>lowMatchPtr)) && (ip[-1] == match3[-1])) { ip--; match3--; mLength++; } /* catch up */
                 } else {
-                    const BYTE* matchEnd = matchIndex < dictLimit ? dictEnd : iend;
-                    const BYTE* lowMatchPtr = matchIndex < dictLimit ? dictStart : lowPrefixPtr;
-                    mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, lowPrefixPtr) + 4;
+                    const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
+                    const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
+                    mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
                     offset = current - matchIndex;
                     while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
                 }
                 offset_2 = offset_1;
                 offset_1 = offset;
-                ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
 
             } else {
-                ip += ((ip-anchor) >> g_searchStrength) + 1;
+                ip += ((ip-anchor) >> kSearchStrength) + 1;
                 continue;
         }   }
 
-        /* found a match : store it */
+        /* move to next sequence start */
         ip += mLength;
         anchor = ip;
 
         if (ip <= ilimit) {
-            /* Fill Table */
-            hashSmall[ZSTD_hashPtr(base+current+2, hBitsS, mls)] = current+2;
-            hashLong[ZSTD_hashPtr(base+current+2, hBitsL, 8)] = current+2;
-            hashSmall[ZSTD_hashPtr(ip-2, hBitsS, mls)] = (U32)(ip-2-base);
-            hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
+            /* Complementary insertion */
+            /* done after iLimit test, as candidates could be > iend-8 */
+            {   U32 const indexToInsert = current+2;
+                hashLong[ZSTD_hashPtr(base+indexToInsert, hBitsL, 8)] = indexToInsert;
+                hashLong[ZSTD_hashPtr(ip-2, hBitsL, 8)] = (U32)(ip-2-base);
+                hashSmall[ZSTD_hashPtr(base+indexToInsert, hBitsS, mls)] = indexToInsert;
+                hashSmall[ZSTD_hashPtr(ip-1, hBitsS, mls)] = (U32)(ip-1-base);
+            }
+
             /* check immediate repcode */
             while (ip <= ilimit) {
                 U32 const current2 = (U32)(ip-base);
                 U32 const repIndex2 = current2 - offset_2;
-                const BYTE* repMatch2 = repIndex2 < dictLimit ? dictBase + repIndex2 : base + repIndex2;
-                if ( (((U32)((dictLimit-1) - repIndex2) >= 3) & (repIndex2 > lowestIndex))  /* intentional overflow */
-                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
-                    const BYTE* const repEnd2 = repIndex2 < dictLimit ? dictEnd : iend;
-                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, lowPrefixPtr) + 4;
-                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
-                    ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, repLength2-MINMATCH);
+                const BYTE* repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
+                if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3)   /* intentional overflow : ensure repIndex2 doesn't overlap dict + prefix */
+                    & (repIndex2 > dictStartIndex))
+                  && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
+                    U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, repLength2-MINMATCH);
                     hashSmall[ZSTD_hashPtr(ip, hBitsS, mls)] = current2;
                     hashLong[ZSTD_hashPtr(ip, hBitsL, 8)] = current2;
                     ip += repLength2;
@@ -282,27 +493,29 @@ static size_t ZSTD_compressBlock_doubleFast_extDict_generic(ZSTD_CCtx* ctx,
     }   }   }
 
     /* save reps for next block */
-    seqStorePtr->repToConfirm[0] = offset_1; seqStorePtr->repToConfirm[1] = offset_2;
+    rep[0] = offset_1;
+    rep[1] = offset_2;
 
     /* Return the last literals size */
-    return iend - anchor;
+    return (size_t)(iend - anchor);
 }
 
 
-size_t ZSTD_compressBlock_doubleFast_extDict(ZSTD_CCtx* ctx,
-                         const void* src, size_t srcSize)
+size_t ZSTD_compressBlock_doubleFast_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
 {
-    U32 const mls = ctx->appliedParams.cParams.searchLength;
+    U32 const mls = ms->cParams.minMatch;
     switch(mls)
     {
     default: /* includes case 3 */
     case 4 :
-        return ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 4);
+        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 4);
     case 5 :
-        return ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 5);
+        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 5);
     case 6 :
-        return ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 6);
+        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 6);
     case 7 :
-        return ZSTD_compressBlock_doubleFast_extDict_generic(ctx, src, srcSize, 7);
+        return ZSTD_compressBlock_doubleFast_extDict_generic(ms, seqStore, rep, src, srcSize, 7);
     }
 }
diff --git a/contrib/zstd/zstd_double_fast.h b/contrib/zstd/zstd_double_fast.h
index 3dba6c710..73a2002f3 100644
--- a/contrib/zstd/zstd_double_fast.h
+++ b/contrib/zstd/zstd_double_fast.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,15 +11,25 @@
 #ifndef ZSTD_DOUBLE_FAST_H
 #define ZSTD_DOUBLE_FAST_H
 
-#include "zstd_compress.h"
-
 #if defined (__cplusplus)
 extern "C" {
 #endif
 
-void ZSTD_fillDoubleHashTable(ZSTD_CCtx* cctx, const void* end, const U32 mls);
-size_t ZSTD_compressBlock_doubleFast(ZSTD_CCtx* ctx, const void* src, size_t srcSize);
-size_t ZSTD_compressBlock_doubleFast_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize);
+#include "mem.h"      /* U32 */
+#include "zstd_compress_internal.h"     /* ZSTD_CCtx, size_t */
+
+void ZSTD_fillDoubleHashTable(ZSTD_matchState_t* ms,
+                              void const* end, ZSTD_dictTableLoadMethod_e dtlm);
+size_t ZSTD_compressBlock_doubleFast(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_doubleFast_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_doubleFast_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+
 
 #if defined (__cplusplus)
 }
diff --git a/contrib/zstd/zstd_errors.h b/contrib/zstd/zstd_errors.h
index bde4304c9..998398e7e 100644
--- a/contrib/zstd/zstd_errors.h
+++ b/contrib/zstd/zstd_errors.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -35,12 +35,20 @@ extern "C" {
 #  define ZSTDERRORLIB_API ZSTDERRORLIB_VISIBILITY
 #endif
 
-/*-****************************************
- *  error codes list
- *  note : this API is still considered unstable
- *         and shall not be used with a dynamic library.
- *         only static linking is allowed
- ******************************************/
+/*-*********************************************
+ *  Error codes list
+ *-*********************************************
+ *  Error codes _values_ are pinned down since v1.3.1 only.
+ *  Therefore, don't rely on values if you may link to any version < v1.3.1.
+ *
+ *  Only values < 100 are considered stable.
+ *
+ *  note 1 : this API shall be used with static linking only.
+ *           dynamic linking is not yet officially supported.
+ *  note 2 : Prefer relying on the enum than on its value whenever possible
+ *           This is the only supported way to use the error list < v1.3.1
+ *  note 3 : ZSTD_isError() is always correct, whatever the library version.
+ **********************************************/
 typedef enum {
   ZSTD_error_no_error = 0,
   ZSTD_error_GENERIC  = 1,
@@ -61,11 +69,15 @@ typedef enum {
   ZSTD_error_stage_wrong       = 60,
   ZSTD_error_init_missing      = 62,
   ZSTD_error_memory_allocation = 64,
+  ZSTD_error_workSpace_tooSmall= 66,
   ZSTD_error_dstSize_tooSmall = 70,
   ZSTD_error_srcSize_wrong    = 72,
+  ZSTD_error_dstBuffer_null   = 74,
+  /* following error codes are __NOT STABLE__, they can be removed or changed in future versions */
   ZSTD_error_frameIndex_tooLarge = 100,
   ZSTD_error_seekableIO          = 102,
-  ZSTD_error_maxCode = 120  /* never EVER use this value directly, it may change in future versions! Use ZSTD_isError() instead */
+  ZSTD_error_dstBuffer_wrong     = 104,
+  ZSTD_error_maxCode = 120  /* never EVER use this value directly, it can change in future versions! Use ZSTD_isError() instead */
 } ZSTD_ErrorCode;
 
 /*! ZSTD_getErrorCode() :
diff --git a/contrib/zstd/zstd_fast.c b/contrib/zstd/zstd_fast.c
index 2e057017b..85a3a7a91 100644
--- a/contrib/zstd/zstd_fast.c
+++ b/contrib/zstd/zstd_fast.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -8,77 +8,306 @@
  * You may select, at your option, one of the above-listed licenses.
  */
 
+#include "zstd_compress_internal.h"  /* ZSTD_hashPtr, ZSTD_count, ZSTD_storeSeq */
 #include "zstd_fast.h"
 
 
-void ZSTD_fillHashTable (ZSTD_CCtx* zc, const void* end, const U32 mls)
+void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
+                        const void* const end,
+                        ZSTD_dictTableLoadMethod_e dtlm)
 {
-    U32* const hashTable = zc->hashTable;
-    U32  const hBits = zc->appliedParams.cParams.hashLog;
-    const BYTE* const base = zc->base;
-    const BYTE* ip = base + zc->nextToUpdate;
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashTable = ms->hashTable;
+    U32  const hBits = cParams->hashLog;
+    U32  const mls = cParams->minMatch;
+    const BYTE* const base = ms->window.base;
+    const BYTE* ip = base + ms->nextToUpdate;
     const BYTE* const iend = ((const BYTE*)end) - HASH_READ_SIZE;
-    const size_t fastHashFillStep = 3;
+    const U32 fastHashFillStep = 3;
 
-    while(ip <= iend) {
-        hashTable[ZSTD_hashPtr(ip, hBits, mls)] = (U32)(ip - base);
-        ip += fastHashFillStep;
-    }
+    /* Always insert every fastHashFillStep position into the hash table.
+     * Insert the other positions if their hash entry is empty.
+     */
+    for ( ; ip + fastHashFillStep < iend + 2; ip += fastHashFillStep) {
+        U32 const current = (U32)(ip - base);
+        size_t const hash0 = ZSTD_hashPtr(ip, hBits, mls);
+        hashTable[hash0] = current;
+        if (dtlm == ZSTD_dtlm_fast) continue;
+        /* Only load extra positions for ZSTD_dtlm_full */
+        {   U32 p;
+            for (p = 1; p < fastHashFillStep; ++p) {
+                size_t const hash = ZSTD_hashPtr(ip + p, hBits, mls);
+                if (hashTable[hash] == 0) {  /* not yet filled */
+                    hashTable[hash] = current + p;
+    }   }   }   }
 }
 
 
-FORCE_INLINE_TEMPLATE
-size_t ZSTD_compressBlock_fast_generic(ZSTD_CCtx* cctx,
-                                 const void* src, size_t srcSize,
-                                 const U32 mls)
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_compressBlock_fast_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize,
+        U32 const mls)
 {
-    U32* const hashTable = cctx->hashTable;
-    U32  const hBits = cctx->appliedParams.cParams.hashLog;
-    seqStore_t* seqStorePtr = &(cctx->seqStore);
-    const BYTE* const base = cctx->base;
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashTable = ms->hashTable;
+    U32 const hlog = cParams->hashLog;
+    /* support stepSize of 0 */
+    size_t const stepSize = cParams->targetLength + !(cParams->targetLength) + 1;
+    const BYTE* const base = ms->window.base;
     const BYTE* const istart = (const BYTE*)src;
-    const BYTE* ip = istart;
+    /* We check ip0 (ip + 0) and ip1 (ip + 1) each loop */
+    const BYTE* ip0 = istart;
+    const BYTE* ip1;
     const BYTE* anchor = istart;
-    const U32   lowestIndex = cctx->dictLimit;
-    const BYTE* const lowest = base + lowestIndex;
+    const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
+    const U32   prefixStartIndex = ZSTD_getLowestPrefixIndex(ms, endIndex, cParams->windowLog);
+    const BYTE* const prefixStart = base + prefixStartIndex;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - HASH_READ_SIZE;
-    U32 offset_1=seqStorePtr->rep[0], offset_2=seqStorePtr->rep[1];
+    U32 offset_1=rep[0], offset_2=rep[1];
     U32 offsetSaved = 0;
 
     /* init */
-    ip += (ip==lowest);
-    {   U32 const maxRep = (U32)(ip-lowest);
+    DEBUGLOG(5, "ZSTD_compressBlock_fast_generic");
+    ip0 += (ip0 == prefixStart);
+    ip1 = ip0 + 1;
+    {   U32 const current = (U32)(ip0 - base);
+        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, cParams->windowLog);
+        U32 const maxRep = current - windowLow;
         if (offset_2 > maxRep) offsetSaved = offset_2, offset_2 = 0;
         if (offset_1 > maxRep) offsetSaved = offset_1, offset_1 = 0;
     }
 
     /* Main Search Loop */
+#ifdef __INTEL_COMPILER
+    /* From intel 'The vector pragma indicates that the loop should be
+     * vectorized if it is legal to do so'. Can be used together with
+     * #pragma ivdep (but have opted to exclude that because intel
+     * warns against using it).*/
+    #pragma vector always
+#endif
+    while (ip1 < ilimit) {   /* < instead of <=, because check at ip0+2 */
+        size_t mLength;
+        BYTE const* ip2 = ip0 + 2;
+        size_t const h0 = ZSTD_hashPtr(ip0, hlog, mls);
+        U32 const val0 = MEM_read32(ip0);
+        size_t const h1 = ZSTD_hashPtr(ip1, hlog, mls);
+        U32 const val1 = MEM_read32(ip1);
+        U32 const current0 = (U32)(ip0-base);
+        U32 const current1 = (U32)(ip1-base);
+        U32 const matchIndex0 = hashTable[h0];
+        U32 const matchIndex1 = hashTable[h1];
+        BYTE const* repMatch = ip2 - offset_1;
+        const BYTE* match0 = base + matchIndex0;
+        const BYTE* match1 = base + matchIndex1;
+        U32 offcode;
+
+#if defined(__aarch64__)
+        PREFETCH_L1(ip0+256);
+#endif
+
+        hashTable[h0] = current0;   /* update hash table */
+        hashTable[h1] = current1;   /* update hash table */
+
+        assert(ip0 + 1 == ip1);
+
+        if ((offset_1 > 0) & (MEM_read32(repMatch) == MEM_read32(ip2))) {
+            mLength = (ip2[-1] == repMatch[-1]) ? 1 : 0;
+            ip0 = ip2 - mLength;
+            match0 = repMatch - mLength;
+            mLength += 4;
+            offcode = 0;
+            goto _match;
+        }
+        if ((matchIndex0 > prefixStartIndex) && MEM_read32(match0) == val0) {
+            /* found a regular match */
+            goto _offset;
+        }
+        if ((matchIndex1 > prefixStartIndex) && MEM_read32(match1) == val1) {
+            /* found a regular match after one literal */
+            ip0 = ip1;
+            match0 = match1;
+            goto _offset;
+        }
+        {   size_t const step = ((size_t)(ip0-anchor) >> (kSearchStrength - 1)) + stepSize;
+            assert(step >= 2);
+            ip0 += step;
+            ip1 += step;
+            continue;
+        }
+_offset: /* Requires: ip0, match0 */
+        /* Compute the offset code */
+        offset_2 = offset_1;
+        offset_1 = (U32)(ip0-match0);
+        offcode = offset_1 + ZSTD_REP_MOVE;
+        mLength = 4;
+        /* Count the backwards match length */
+        while (((ip0>anchor) & (match0>prefixStart))
+             && (ip0[-1] == match0[-1])) { ip0--; match0--; mLength++; } /* catch up */
+
+_match: /* Requires: ip0, match0, offcode */
+        /* Count the forward length */
+        mLength += ZSTD_count(ip0+mLength, match0+mLength, iend);
+        ZSTD_storeSeq(seqStore, (size_t)(ip0-anchor), anchor, iend, offcode, mLength-MINMATCH);
+        /* match found */
+        ip0 += mLength;
+        anchor = ip0;
+
+        if (ip0 <= ilimit) {
+            /* Fill Table */
+            assert(base+current0+2 > istart);  /* check base overflow */
+            hashTable[ZSTD_hashPtr(base+current0+2, hlog, mls)] = current0+2;  /* here because current+2 could be > iend-8 */
+            hashTable[ZSTD_hashPtr(ip0-2, hlog, mls)] = (U32)(ip0-2-base);
+
+            if (offset_2 > 0) { /* offset_2==0 means offset_2 is invalidated */
+                while ( (ip0 <= ilimit) && (MEM_read32(ip0) == MEM_read32(ip0 - offset_2)) ) {
+                    /* store sequence */
+                    size_t const rLength = ZSTD_count(ip0+4, ip0+4-offset_2, iend) + 4;
+                    { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; } /* swap offset_2 <=> offset_1 */
+                    hashTable[ZSTD_hashPtr(ip0, hlog, mls)] = (U32)(ip0-base);
+                    ip0 += rLength;
+                    ZSTD_storeSeq(seqStore, 0 /*litLen*/, anchor, iend, 0 /*offCode*/, rLength-MINMATCH);
+                    anchor = ip0;
+                    continue;   /* faster when present (confirmed on gcc-8) ... (?) */
+        }   }   }
+        ip1 = ip0 + 1;
+    }
+
+    /* save reps for next block */
+    rep[0] = offset_1 ? offset_1 : offsetSaved;
+    rep[1] = offset_2 ? offset_2 : offsetSaved;
+
+    /* Return the last literals size */
+    return (size_t)(iend - anchor);
+}
+
+
+size_t ZSTD_compressBlock_fast(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    U32 const mls = ms->cParams.minMatch;
+    assert(ms->dictMatchState == NULL);
+    switch(mls)
+    {
+    default: /* includes case 3 */
+    case 4 :
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 4);
+    case 5 :
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 5);
+    case 6 :
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 6);
+    case 7 :
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, 7);
+    }
+}
+
+FORCE_INLINE_TEMPLATE
+size_t ZSTD_compressBlock_fast_dictMatchState_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize, U32 const mls)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashTable = ms->hashTable;
+    U32 const hlog = cParams->hashLog;
+    /* support stepSize of 0 */
+    U32 const stepSize = cParams->targetLength + !(cParams->targetLength);
+    const BYTE* const base = ms->window.base;
+    const BYTE* const istart = (const BYTE*)src;
+    const BYTE* ip = istart;
+    const BYTE* anchor = istart;
+    const U32   prefixStartIndex = ms->window.dictLimit;
+    const BYTE* const prefixStart = base + prefixStartIndex;
+    const BYTE* const iend = istart + srcSize;
+    const BYTE* const ilimit = iend - HASH_READ_SIZE;
+    U32 offset_1=rep[0], offset_2=rep[1];
+    U32 offsetSaved = 0;
+
+    const ZSTD_matchState_t* const dms = ms->dictMatchState;
+    const ZSTD_compressionParameters* const dictCParams = &dms->cParams ;
+    const U32* const dictHashTable = dms->hashTable;
+    const U32 dictStartIndex       = dms->window.dictLimit;
+    const BYTE* const dictBase     = dms->window.base;
+    const BYTE* const dictStart    = dictBase + dictStartIndex;
+    const BYTE* const dictEnd      = dms->window.nextSrc;
+    const U32 dictIndexDelta       = prefixStartIndex - (U32)(dictEnd - dictBase);
+    const U32 dictAndPrefixLength  = (U32)(ip - prefixStart + dictEnd - dictStart);
+    const U32 dictHLog             = dictCParams->hashLog;
+
+    /* if a dictionary is still attached, it necessarily means that
+     * it is within window size. So we just check it. */
+    const U32 maxDistance = 1U << cParams->windowLog;
+    const U32 endIndex = (U32)((size_t)(ip - base) + srcSize);
+    assert(endIndex - prefixStartIndex <= maxDistance);
+    (void)maxDistance; (void)endIndex;   /* these variables are not used when assert() is disabled */
+
+    /* ensure there will be no no underflow
+     * when translating a dict index into a local index */
+    assert(prefixStartIndex >= (U32)(dictEnd - dictBase));
+
+    /* init */
+    DEBUGLOG(5, "ZSTD_compressBlock_fast_dictMatchState_generic");
+    ip += (dictAndPrefixLength == 0);
+    /* dictMatchState repCode checks don't currently handle repCode == 0
+     * disabling. */
+    assert(offset_1 <= dictAndPrefixLength);
+    assert(offset_2 <= dictAndPrefixLength);
+
+    /* Main Search Loop */
     while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
         size_t mLength;
-        size_t const h = ZSTD_hashPtr(ip, hBits, mls);
+        size_t const h = ZSTD_hashPtr(ip, hlog, mls);
         U32 const current = (U32)(ip-base);
         U32 const matchIndex = hashTable[h];
         const BYTE* match = base + matchIndex;
+        const U32 repIndex = current + 1 - offset_1;
+        const BYTE* repMatch = (repIndex < prefixStartIndex) ?
+                               dictBase + (repIndex - dictIndexDelta) :
+                               base + repIndex;
         hashTable[h] = current;   /* update hash table */
 
-        if ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1))) {
-            mLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
+        if ( ((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow : ensure repIndex isn't overlapping dict + prefix */
+          && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+            const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
+            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixStart) + 4;
             ip++;
-            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mLength-MINMATCH);
-        } else {
-            U32 offset;
-            if ( (matchIndex <= lowestIndex) || (MEM_read32(match) != MEM_read32(ip)) ) {
-                ip += ((ip-anchor) >> g_searchStrength) + 1;
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, mLength-MINMATCH);
+        } else if ( (matchIndex <= prefixStartIndex) ) {
+            size_t const dictHash = ZSTD_hashPtr(ip, dictHLog, mls);
+            U32 const dictMatchIndex = dictHashTable[dictHash];
+            const BYTE* dictMatch = dictBase + dictMatchIndex;
+            if (dictMatchIndex <= dictStartIndex ||
+                MEM_read32(dictMatch) != MEM_read32(ip)) {
+                assert(stepSize >= 1);
+                ip += ((ip-anchor) >> kSearchStrength) + stepSize;
                 continue;
+            } else {
+                /* found a dict match */
+                U32 const offset = (U32)(current-dictMatchIndex-dictIndexDelta);
+                mLength = ZSTD_count_2segments(ip+4, dictMatch+4, iend, dictEnd, prefixStart) + 4;
+                while (((ip>anchor) & (dictMatch>dictStart))
+                     && (ip[-1] == dictMatch[-1])) {
+                    ip--; dictMatch--; mLength++;
+                } /* catch up */
+                offset_2 = offset_1;
+                offset_1 = offset;
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
             }
+        } else if (MEM_read32(match) != MEM_read32(ip)) {
+            /* it's not a match, and we're not going to check the dictionary */
+            assert(stepSize >= 1);
+            ip += ((ip-anchor) >> kSearchStrength) + stepSize;
+            continue;
+        } else {
+            /* found a regular match */
+            U32 const offset = (U32)(ip-match);
             mLength = ZSTD_count(ip+4, match+4, iend) + 4;
-            offset = (U32)(ip-match);
-            while (((ip>anchor) & (match>lowest)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
+            while (((ip>anchor) & (match>prefixStart))
+                 && (ip[-1] == match[-1])) { ip--; match--; mLength++; } /* catch up */
             offset_2 = offset_1;
             offset_1 = offset;
-
-            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
         }
 
         /* match found */
@@ -87,127 +316,150 @@ size_t ZSTD_compressBlock_fast_generic(ZSTD_CCtx* cctx,
 
         if (ip <= ilimit) {
             /* Fill Table */
-            hashTable[ZSTD_hashPtr(base+current+2, hBits, mls)] = current+2;  /* here because current+2 could be > iend-8 */
-            hashTable[ZSTD_hashPtr(ip-2, hBits, mls)] = (U32)(ip-2-base);
+            assert(base+current+2 > istart);  /* check base overflow */
+            hashTable[ZSTD_hashPtr(base+current+2, hlog, mls)] = current+2;  /* here because current+2 could be > iend-8 */
+            hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
+
             /* check immediate repcode */
-            while ( (ip <= ilimit)
-                 && ( (offset_2>0)
-                 & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
-                /* store sequence */
-                size_t const rLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
-                { U32 const tmpOff = offset_2; offset_2 = offset_1; offset_1 = tmpOff; }  /* swap offset_2 <=> offset_1 */
-                hashTable[ZSTD_hashPtr(ip, hBits, mls)] = (U32)(ip-base);
-                ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, rLength-MINMATCH);
-                ip += rLength;
-                anchor = ip;
-                continue;   /* faster when present ... (?) */
-    }   }   }
+            while (ip <= ilimit) {
+                U32 const current2 = (U32)(ip-base);
+                U32 const repIndex2 = current2 - offset_2;
+                const BYTE* repMatch2 = repIndex2 < prefixStartIndex ?
+                        dictBase - dictIndexDelta + repIndex2 :
+                        base + repIndex2;
+                if ( ((U32)((prefixStartIndex-1) - (U32)repIndex2) >= 3 /* intentional overflow */)
+                   && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
+                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, repLength2-MINMATCH);
+                    hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2;
+                    ip += repLength2;
+                    anchor = ip;
+                    continue;
+                }
+                break;
+            }
+        }
+    }
 
     /* save reps for next block */
-    seqStorePtr->repToConfirm[0] = offset_1 ? offset_1 : offsetSaved;
-    seqStorePtr->repToConfirm[1] = offset_2 ? offset_2 : offsetSaved;
+    rep[0] = offset_1 ? offset_1 : offsetSaved;
+    rep[1] = offset_2 ? offset_2 : offsetSaved;
 
     /* Return the last literals size */
-    return iend - anchor;
+    return (size_t)(iend - anchor);
 }
 
-
-size_t ZSTD_compressBlock_fast(ZSTD_CCtx* ctx,
-                       const void* src, size_t srcSize)
+size_t ZSTD_compressBlock_fast_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
 {
-    const U32 mls = ctx->appliedParams.cParams.searchLength;
+    U32 const mls = ms->cParams.minMatch;
+    assert(ms->dictMatchState != NULL);
     switch(mls)
     {
     default: /* includes case 3 */
     case 4 :
-        return ZSTD_compressBlock_fast_generic(ctx, src, srcSize, 4);
+        return ZSTD_compressBlock_fast_dictMatchState_generic(ms, seqStore, rep, src, srcSize, 4);
     case 5 :
-        return ZSTD_compressBlock_fast_generic(ctx, src, srcSize, 5);
+        return ZSTD_compressBlock_fast_dictMatchState_generic(ms, seqStore, rep, src, srcSize, 5);
     case 6 :
-        return ZSTD_compressBlock_fast_generic(ctx, src, srcSize, 6);
+        return ZSTD_compressBlock_fast_dictMatchState_generic(ms, seqStore, rep, src, srcSize, 6);
     case 7 :
-        return ZSTD_compressBlock_fast_generic(ctx, src, srcSize, 7);
+        return ZSTD_compressBlock_fast_dictMatchState_generic(ms, seqStore, rep, src, srcSize, 7);
     }
 }
 
 
-static size_t ZSTD_compressBlock_fast_extDict_generic(ZSTD_CCtx* ctx,
-                                 const void* src, size_t srcSize,
-                                 const U32 mls)
+static size_t ZSTD_compressBlock_fast_extDict_generic(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize, U32 const mls)
 {
-    U32* hashTable = ctx->hashTable;
-    const U32 hBits = ctx->appliedParams.cParams.hashLog;
-    seqStore_t* seqStorePtr = &(ctx->seqStore);
-    const BYTE* const base = ctx->base;
-    const BYTE* const dictBase = ctx->dictBase;
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashTable = ms->hashTable;
+    U32 const hlog = cParams->hashLog;
+    /* support stepSize of 0 */
+    U32 const stepSize = cParams->targetLength + !(cParams->targetLength);
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
     const BYTE* const istart = (const BYTE*)src;
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
-    const U32   lowestIndex = ctx->lowLimit;
-    const BYTE* const dictStart = dictBase + lowestIndex;
-    const U32   dictLimit = ctx->dictLimit;
-    const BYTE* const lowPrefixPtr = base + dictLimit;
-    const BYTE* const dictEnd = dictBase + dictLimit;
+    const U32   endIndex = (U32)((size_t)(istart - base) + srcSize);
+    const U32   lowLimit = ZSTD_getLowestMatchIndex(ms, endIndex, cParams->windowLog);
+    const U32   dictStartIndex = lowLimit;
+    const BYTE* const dictStart = dictBase + dictStartIndex;
+    const U32   dictLimit = ms->window.dictLimit;
+    const U32   prefixStartIndex = dictLimit < lowLimit ? lowLimit : dictLimit;
+    const BYTE* const prefixStart = base + prefixStartIndex;
+    const BYTE* const dictEnd = dictBase + prefixStartIndex;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - 8;
-    U32 offset_1=seqStorePtr->rep[0], offset_2=seqStorePtr->rep[1];
+    U32 offset_1=rep[0], offset_2=rep[1];
+
+    DEBUGLOG(5, "ZSTD_compressBlock_fast_extDict_generic (offset_1=%u)", offset_1);
+
+    /* switch to "regular" variant if extDict is invalidated due to maxDistance */
+    if (prefixStartIndex == dictStartIndex)
+        return ZSTD_compressBlock_fast_generic(ms, seqStore, rep, src, srcSize, mls);
 
     /* Search Loop */
     while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
-        const size_t h = ZSTD_hashPtr(ip, hBits, mls);
-        const U32 matchIndex = hashTable[h];
-        const BYTE* matchBase = matchIndex < dictLimit ? dictBase : base;
-        const BYTE* match = matchBase + matchIndex;
-        const U32 current = (U32)(ip-base);
-        const U32 repIndex = current + 1 - offset_1;   /* offset_1 expected <= current +1 */
-        const BYTE* repBase = repIndex < dictLimit ? dictBase : base;
-        const BYTE* repMatch = repBase + repIndex;
-        size_t mLength;
+        const size_t h = ZSTD_hashPtr(ip, hlog, mls);
+        const U32    matchIndex = hashTable[h];
+        const BYTE* const matchBase = matchIndex < prefixStartIndex ? dictBase : base;
+        const BYTE*  match = matchBase + matchIndex;
+        const U32    current = (U32)(ip-base);
+        const U32    repIndex = current + 1 - offset_1;
+        const BYTE* const repBase = repIndex < prefixStartIndex ? dictBase : base;
+        const BYTE* const repMatch = repBase + repIndex;
         hashTable[h] = current;   /* update hash table */
+        DEBUGLOG(7, "offset_1 = %u , current = %u", offset_1, current);
+        assert(offset_1 <= current +1);   /* check repIndex */
 
-        if ( (((U32)((dictLimit-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > lowestIndex))
+        if ( (((U32)((prefixStartIndex-1) - repIndex) >= 3) /* intentional underflow */ & (repIndex > dictStartIndex))
            && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
-            const BYTE* repMatchEnd = repIndex < dictLimit ? dictEnd : iend;
-            mLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, lowPrefixPtr) + 4;
+            const BYTE* const repMatchEnd = repIndex < prefixStartIndex ? dictEnd : iend;
+            size_t const rLength = ZSTD_count_2segments(ip+1 +4, repMatch +4, iend, repMatchEnd, prefixStart) + 4;
             ip++;
-            ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, 0, mLength-MINMATCH);
+            ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, 0, rLength-MINMATCH);
+            ip += rLength;
+            anchor = ip;
         } else {
-            if ( (matchIndex < lowestIndex) ||
+            if ( (matchIndex < dictStartIndex) ||
                  (MEM_read32(match) != MEM_read32(ip)) ) {
-                ip += ((ip-anchor) >> g_searchStrength) + 1;
+                assert(stepSize >= 1);
+                ip += ((ip-anchor) >> kSearchStrength) + stepSize;
                 continue;
             }
-            {   const BYTE* matchEnd = matchIndex < dictLimit ? dictEnd : iend;
-                const BYTE* lowMatchPtr = matchIndex < dictLimit ? dictStart : lowPrefixPtr;
-                U32 offset;
-                mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, lowPrefixPtr) + 4;
+            {   const BYTE* const matchEnd = matchIndex < prefixStartIndex ? dictEnd : iend;
+                const BYTE* const lowMatchPtr = matchIndex < prefixStartIndex ? dictStart : prefixStart;
+                U32 const offset = current - matchIndex;
+                size_t mLength = ZSTD_count_2segments(ip+4, match+4, iend, matchEnd, prefixStart) + 4;
                 while (((ip>anchor) & (match>lowMatchPtr)) && (ip[-1] == match[-1])) { ip--; match--; mLength++; }   /* catch up */
-                offset = current - matchIndex;
-                offset_2 = offset_1;
-                offset_1 = offset;
-                ZSTD_storeSeq(seqStorePtr, ip-anchor, anchor, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+                offset_2 = offset_1; offset_1 = offset;  /* update offset history */
+                ZSTD_storeSeq(seqStore, (size_t)(ip-anchor), anchor, iend, offset + ZSTD_REP_MOVE, mLength-MINMATCH);
+                ip += mLength;
+                anchor = ip;
         }   }
 
-        /* found a match : store it */
-        ip += mLength;
-        anchor = ip;
-
         if (ip <= ilimit) {
             /* Fill Table */
-            hashTable[ZSTD_hashPtr(base+current+2, hBits, mls)] = current+2;
-            hashTable[ZSTD_hashPtr(ip-2, hBits, mls)] = (U32)(ip-2-base);
+            hashTable[ZSTD_hashPtr(base+current+2, hlog, mls)] = current+2;
+            hashTable[ZSTD_hashPtr(ip-2, hlog, mls)] = (U32)(ip-2-base);
             /* check immediate repcode */
             while (ip <= ilimit) {
                 U32 const current2 = (U32)(ip-base);
                 U32 const repIndex2 = current2 - offset_2;
-                const BYTE* repMatch2 = repIndex2 < dictLimit ? dictBase + repIndex2 : base + repIndex2;
-                if ( (((U32)((dictLimit-1) - repIndex2) >= 3) & (repIndex2 > lowestIndex))  /* intentional overflow */
+                const BYTE* const repMatch2 = repIndex2 < prefixStartIndex ? dictBase + repIndex2 : base + repIndex2;
+                if ( (((U32)((prefixStartIndex-1) - repIndex2) >= 3) & (repIndex2 > dictStartIndex))  /* intentional overflow */
                    && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
-                    const BYTE* const repEnd2 = repIndex2 < dictLimit ? dictEnd : iend;
-                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, lowPrefixPtr) + 4;
-                    U32 tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset;   /* swap offset_2 <=> offset_1 */
-                    ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, repLength2-MINMATCH);
-                    hashTable[ZSTD_hashPtr(ip, hBits, mls)] = current2;
+                    const BYTE* const repEnd2 = repIndex2 < prefixStartIndex ? dictEnd : iend;
+                    size_t const repLength2 = ZSTD_count_2segments(ip+4, repMatch2+4, iend, repEnd2, prefixStart) + 4;
+                    { U32 const tmpOffset = offset_2; offset_2 = offset_1; offset_1 = tmpOffset; }  /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0 /*litlen*/, anchor, iend, 0 /*offcode*/, repLength2-MINMATCH);
+                    hashTable[ZSTD_hashPtr(ip, hlog, mls)] = current2;
                     ip += repLength2;
                     anchor = ip;
                     continue;
@@ -216,27 +468,29 @@ static size_t ZSTD_compressBlock_fast_extDict_generic(ZSTD_CCtx* ctx,
     }   }   }
 
     /* save reps for next block */
-    seqStorePtr->repToConfirm[0] = offset_1; seqStorePtr->repToConfirm[1] = offset_2;
+    rep[0] = offset_1;
+    rep[1] = offset_2;
 
     /* Return the last literals size */
-    return iend - anchor;
+    return (size_t)(iend - anchor);
 }
 
 
-size_t ZSTD_compressBlock_fast_extDict(ZSTD_CCtx* ctx,
-                         const void* src, size_t srcSize)
+size_t ZSTD_compressBlock_fast_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
 {
-    U32 const mls = ctx->appliedParams.cParams.searchLength;
+    U32 const mls = ms->cParams.minMatch;
     switch(mls)
     {
     default: /* includes case 3 */
     case 4 :
-        return ZSTD_compressBlock_fast_extDict_generic(ctx, src, srcSize, 4);
+        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 4);
     case 5 :
-        return ZSTD_compressBlock_fast_extDict_generic(ctx, src, srcSize, 5);
+        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 5);
     case 6 :
-        return ZSTD_compressBlock_fast_extDict_generic(ctx, src, srcSize, 6);
+        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 6);
     case 7 :
-        return ZSTD_compressBlock_fast_extDict_generic(ctx, src, srcSize, 7);
+        return ZSTD_compressBlock_fast_extDict_generic(ms, seqStore, rep, src, srcSize, 7);
     }
 }
diff --git a/contrib/zstd/zstd_fast.h b/contrib/zstd/zstd_fast.h
index 4205141a9..3e15cc85c 100644
--- a/contrib/zstd/zstd_fast.h
+++ b/contrib/zstd/zstd_fast.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,17 +11,24 @@
 #ifndef ZSTD_FAST_H
 #define ZSTD_FAST_H
 
-#include "zstd_compress.h"
-
 #if defined (__cplusplus)
 extern "C" {
 #endif
 
-void ZSTD_fillHashTable(ZSTD_CCtx* zc, const void* end, const U32 mls);
-size_t ZSTD_compressBlock_fast(ZSTD_CCtx* ctx,
-                         const void* src, size_t srcSize);
-size_t ZSTD_compressBlock_fast_extDict(ZSTD_CCtx* ctx,
-                         const void* src, size_t srcSize);
+#include "mem.h"      /* U32 */
+#include "zstd_compress_internal.h"
+
+void ZSTD_fillHashTable(ZSTD_matchState_t* ms,
+                        void const* end, ZSTD_dictTableLoadMethod_e dtlm);
+size_t ZSTD_compressBlock_fast(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_fast_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_fast_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
 
 #if defined (__cplusplus)
 }
diff --git a/contrib/zstd/zstd_internal.h b/contrib/zstd/zstd_internal.h
index e7d02b5a0..ea3d61837 100644
--- a/contrib/zstd/zstd_internal.h
+++ b/contrib/zstd/zstd_internal.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,17 +11,25 @@
 #ifndef ZSTD_CCOMMON_H_MODULE
 #define ZSTD_CCOMMON_H_MODULE
 
+/* this module contains definitions which must be identical
+ * across compression, decompression and dictBuilder.
+ * It also contains a few functions useful to at least 2 of them
+ * and which benefit from being inlined */
 
 /*-*************************************
 *  Dependencies
 ***************************************/
+#ifdef __aarch64__
+#include <arm_neon.h>
+#endif
 #include "compiler.h"
 #include "mem.h"
+#include "debug.h"                 /* assert, DEBUGLOG, RAWLOG, g_debuglevel */
 #include "error_private.h"
 #ifndef ZSTD_STATIC_LINKING_ONLY
 #define ZSTD_STATIC_LINKING_ONLY
 #endif
-#include "zstd.h"
+#include "../zstd.h"
 #define FSE_STATIC_LINKING_ONLY
 #include "fse.h"
 #define HUF_STATIC_LINKING_ONLY
@@ -31,43 +39,15 @@
 #endif
 #include "xxhash.h"                /* XXH_reset, update, digest */
 
-
 #if defined (__cplusplus)
 extern "C" {
 #endif
 
-
-/*-*************************************
-*  Debug
-***************************************/
-#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=1)
-#  include <assert.h>
-#else
-#  ifndef assert
-#    define assert(condition) ((void)0)
-#  endif
-#endif
-
-#define ZSTD_STATIC_ASSERT(c) { enum { ZSTD_static_assert = 1/(int)(!!(c)) }; }
-
-#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=2)
-#  include <stdio.h>
-/* recommended values for ZSTD_DEBUG display levels :
- * 1 : no display, enables assert() only
- * 2 : reserved for currently active debugging path
- * 3 : events once per object lifetime (CCtx, CDict)
- * 4 : events once per frame
- * 5 : events once per block
- * 6 : events once per sequence (*very* verbose) */
-#  define DEBUGLOG(l, ...) {                         \
-                if (l<=ZSTD_DEBUG) {                 \
-                    fprintf(stderr, __FILE__ ": ");  \
-                    fprintf(stderr, __VA_ARGS__);    \
-                    fprintf(stderr, " \n");          \
-            }   }
-#else
-#  define DEBUGLOG(l, ...)      {}    /* disabled */
-#endif
+/* ---- static assert (debug) --- */
+#define ZSTD_STATIC_ASSERT(c) DEBUG_STATIC_ASSERT(c)
+#define ZSTD_isError ERR_isError   /* for inlining */
+#define FSE_isError  ERR_isError
+#define HUF_isError  ERR_isError
 
 
 /*-*************************************
@@ -77,8 +57,81 @@ extern "C" {
 #undef MAX
 #define MIN(a,b) ((a)<(b) ? (a) : (b))
 #define MAX(a,b) ((a)>(b) ? (a) : (b))
-#define CHECK_F(f) { size_t const errcod = f; if (ERR_isError(errcod)) return errcod; }  /* check and Forward error code */
-#define CHECK_E(f, e) { size_t const errcod = f; if (ERR_isError(errcod)) return ERROR(e); }  /* check and send Error code */
+
+/**
+ * Ignore: this is an internal helper.
+ *
+ * This is a helper function to help force C99-correctness during compilation.
+ * Under strict compilation modes, variadic macro arguments can't be empty.
+ * However, variadic function arguments can be. Using a function therefore lets
+ * us statically check that at least one (string) argument was passed,
+ * independent of the compilation flags.
+ */
+static INLINE_KEYWORD UNUSED_ATTR
+void _force_has_format_string(const char *format, ...) {
+  (void)format;
+}
+
+/**
+ * Ignore: this is an internal helper.
+ *
+ * We want to force this function invocation to be syntactically correct, but
+ * we don't want to force runtime evaluation of its arguments.
+ */
+#define _FORCE_HAS_FORMAT_STRING(...) \
+  if (0) { \
+    _force_has_format_string(__VA_ARGS__); \
+  }
+
+/**
+ * Return the specified error if the condition evaluates to true.
+ *
+ * In debug modes, prints additional information.
+ * In order to do that (particularly, printing the conditional that failed),
+ * this can't just wrap RETURN_ERROR().
+ */
+#define RETURN_ERROR_IF(cond, err, ...) \
+  if (cond) { \
+    RAWLOG(3, "%s:%d: ERROR!: check %s failed, returning %s", \
+           __FILE__, __LINE__, ZSTD_QUOTE(cond), ZSTD_QUOTE(ERROR(err))); \
+    _FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
+    RAWLOG(3, ": " __VA_ARGS__); \
+    RAWLOG(3, "\n"); \
+    return ERROR(err); \
+  }
+
+/**
+ * Unconditionally return the specified error.
+ *
+ * In debug modes, prints additional information.
+ */
+#define RETURN_ERROR(err, ...) \
+  do { \
+    RAWLOG(3, "%s:%d: ERROR!: unconditional check failed, returning %s", \
+           __FILE__, __LINE__, ZSTD_QUOTE(ERROR(err))); \
+    _FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
+    RAWLOG(3, ": " __VA_ARGS__); \
+    RAWLOG(3, "\n"); \
+    return ERROR(err); \
+  } while(0);
+
+/**
+ * If the provided expression evaluates to an error code, returns that error code.
+ *
+ * In debug modes, prints additional information.
+ */
+#define FORWARD_IF_ERROR(err, ...) \
+  do { \
+    size_t const err_code = (err); \
+    if (ERR_isError(err_code)) { \
+      RAWLOG(3, "%s:%d: ERROR!: forwarding error in %s: %s", \
+             __FILE__, __LINE__, ZSTD_QUOTE(err), ERR_getErrorName(err_code)); \
+      _FORCE_HAS_FORMAT_STRING(__VA_ARGS__); \
+      RAWLOG(3, ": " __VA_ARGS__); \
+      RAWLOG(3, "\n"); \
+      return err_code; \
+    } \
+  } while(0);
 
 
 /*-*************************************
@@ -87,9 +140,7 @@ extern "C" {
 #define ZSTD_OPT_NUM    (1<<12)
 
 #define ZSTD_REP_NUM      3                 /* number of repcodes */
-#define ZSTD_REP_CHECK    (ZSTD_REP_NUM)    /* number of repcodes to check by the optimal parser */
 #define ZSTD_REP_MOVE     (ZSTD_REP_NUM-1)
-#define ZSTD_REP_MOVE_OPT (ZSTD_REP_NUM)
 static const U32 repStartValue[ZSTD_REP_NUM] = { 1, 4, 8 };
 
 #define KB *(1 <<10)
@@ -107,10 +158,14 @@ static const U32 repStartValue[ZSTD_REP_NUM] = { 1, 4, 8 };
 static const size_t ZSTD_fcs_fieldSize[4] = { 0, 2, 4, 8 };
 static const size_t ZSTD_did_fieldSize[4] = { 0, 1, 2, 4 };
 
+#define ZSTD_FRAMEIDSIZE 4   /* magic number size */
+
 #define ZSTD_BLOCKHEADERSIZE 3   /* C standard doesn't allow `static const` variable to be init using another `static const` variable */
 static const size_t ZSTD_blockHeaderSize = ZSTD_BLOCKHEADERSIZE;
 typedef enum { bt_raw, bt_rle, bt_compressed, bt_reserved } blockType_e;
 
+#define ZSTD_FRAMECHECKSUMSIZE 4
+
 #define MIN_SEQUENCES_SIZE 1 /* nbSeq==0 */
 #define MIN_CBLOCK_SIZE (1 /*litCSize*/ + 1 /* RLE or RAW */ + MIN_SEQUENCES_SIZE /* nbSeq==0 */)   /* for a non-null block */
 
@@ -123,36 +178,50 @@ typedef enum { set_basic, set_rle, set_compressed, set_repeat } symbolEncodingTy
 
 #define Litbits  8
 #define MaxLit ((1<<Litbits) - 1)
-#define MaxML  52
-#define MaxLL  35
-#define MaxOff 28
+#define MaxML   52
+#define MaxLL   35
+#define DefaultMaxOff 28
+#define MaxOff  31
 #define MaxSeq MAX(MaxLL, MaxML)   /* Assumption : MaxOff < MaxLL,MaxML */
 #define MLFSELog    9
 #define LLFSELog    9
 #define OffFSELog   8
+#define MaxFSELog  MAX(MAX(MLFSELog, LLFSELog), OffFSELog)
 
-static const U32 LL_bits[MaxLL+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-                                      1, 1, 1, 1, 2, 2, 3, 3, 4, 6, 7, 8, 9,10,11,12,
+static const U32 LL_bits[MaxLL+1] = { 0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3,
+                                      4, 6, 7, 8, 9,10,11,12,
                                      13,14,15,16 };
-static const S16 LL_defaultNorm[MaxLL+1] = { 4, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 1, 1, 1,
-                                             2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 2, 1, 1, 1, 1, 1,
+static const S16 LL_defaultNorm[MaxLL+1] = { 4, 3, 2, 2, 2, 2, 2, 2,
+                                             2, 2, 2, 2, 2, 1, 1, 1,
+                                             2, 2, 2, 2, 2, 2, 2, 2,
+                                             2, 3, 2, 1, 1, 1, 1, 1,
                                             -1,-1,-1,-1 };
 #define LL_DEFAULTNORMLOG 6  /* for static allocation */
 static const U32 LL_defaultNormLog = LL_DEFAULTNORMLOG;
 
-static const U32 ML_bits[MaxML+1] = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-                                      0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-                                      1, 1, 1, 1, 2, 2, 3, 3, 4, 4, 5, 7, 8, 9,10,11,
+static const U32 ML_bits[MaxML+1] = { 0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0,
+                                      0, 0, 0, 0, 0, 0, 0, 0,
+                                      1, 1, 1, 1, 2, 2, 3, 3,
+                                      4, 4, 5, 7, 8, 9,10,11,
                                      12,13,14,15,16 };
-static const S16 ML_defaultNorm[MaxML+1] = { 1, 4, 3, 2, 2, 2, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
-                                             1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
-                                             1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,-1,-1,
+static const S16 ML_defaultNorm[MaxML+1] = { 1, 4, 3, 2, 2, 2, 2, 2,
+                                             2, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1, 1, 1,
+                                             1, 1, 1, 1, 1, 1,-1,-1,
                                             -1,-1,-1,-1,-1 };
 #define ML_DEFAULTNORMLOG 6  /* for static allocation */
 static const U32 ML_defaultNormLog = ML_DEFAULTNORMLOG;
 
-static const S16 OF_defaultNorm[MaxOff+1] = { 1, 1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1, 1, 1, 1, 1,
-                                              1, 1, 1, 1, 1, 1, 1, 1,-1,-1,-1,-1,-1 };
+static const S16 OF_defaultNorm[DefaultMaxOff+1] = { 1, 1, 1, 1, 1, 1, 2, 2,
+                                                     2, 1, 1, 1, 1, 1, 1, 1,
+                                                     1, 1, 1, 1, 1, 1, 1, 1,
+                                                    -1,-1,-1,-1,-1 };
 #define OF_DEFAULTNORMLOG 5  /* for static allocation */
 static const U32 OF_defaultNormLog = OF_DEFAULTNORMLOG;
 
@@ -160,45 +229,109 @@ static const U32 OF_defaultNormLog = OF_DEFAULTNORMLOG;
 /*-*******************************************
 *  Shared functions to include for inlining
 *********************************************/
-static void ZSTD_copy8(void* dst, const void* src) { memcpy(dst, src, 8); }
+static void ZSTD_copy8(void* dst, const void* src) {
+#ifdef __aarch64__
+    vst1_u8((uint8_t*)dst, vld1_u8((const uint8_t*)src));
+#else
+    memcpy(dst, src, 8);
+#endif
+}
+
 #define COPY8(d,s) { ZSTD_copy8(d,s); d+=8; s+=8; }
+static void ZSTD_copy16(void* dst, const void* src) {
+#ifdef __aarch64__
+    vst1q_u8((uint8_t*)dst, vld1q_u8((const uint8_t*)src));
+#else
+    memcpy(dst, src, 16);
+#endif
+}
+#define COPY16(d,s) { ZSTD_copy16(d,s); d+=16; s+=16; }
+
+#define WILDCOPY_OVERLENGTH 32
+#define WILDCOPY_VECLEN 16
+
+typedef enum {
+    ZSTD_no_overlap,
+    ZSTD_overlap_src_before_dst
+    /*  ZSTD_overlap_dst_before_src, */
+} ZSTD_overlap_e;
 
 /*! ZSTD_wildcopy() :
-*   custom version of memcpy(), can copy up to 7 bytes too many (8 bytes if length==0) */
-#define WILDCOPY_OVERLENGTH 8
-MEM_STATIC void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length)
+ *  Custom version of memcpy(), can over read/write up to WILDCOPY_OVERLENGTH bytes (if length==0)
+ *  @param ovtype controls the overlap detection
+ *         - ZSTD_no_overlap: The source and destination are guaranteed to be at least WILDCOPY_VECLEN bytes apart.
+ *         - ZSTD_overlap_src_before_dst: The src and dst may overlap, but they MUST be at least 8 bytes apart.
+ *           The src buffer must be before the dst buffer.
+ */
+MEM_STATIC FORCE_INLINE_ATTR
+void ZSTD_wildcopy(void* dst, const void* src, ptrdiff_t length, ZSTD_overlap_e const ovtype)
 {
+    ptrdiff_t diff = (BYTE*)dst - (const BYTE*)src;
     const BYTE* ip = (const BYTE*)src;
     BYTE* op = (BYTE*)dst;
     BYTE* const oend = op + length;
-    do
-        COPY8(op, ip)
-    while (op < oend);
+
+    assert(diff >= 8 || (ovtype == ZSTD_no_overlap && diff <= -WILDCOPY_VECLEN));
+
+    if (ovtype == ZSTD_overlap_src_before_dst && diff < WILDCOPY_VECLEN) {
+        /* Handle short offset copies. */
+        do {
+            COPY8(op, ip)
+        } while (op < oend);
+    } else {
+        assert(diff >= WILDCOPY_VECLEN || diff <= -WILDCOPY_VECLEN);
+        /* Separate out the first COPY16() call because the copy length is
+         * almost certain to be short, so the branches have different
+         * probabilities. Since it is almost certain to be short, only do
+         * one COPY16() in the first call. Then, do two calls per loop since
+         * at that point it is more likely to have a high trip count.
+         */
+#ifndef __aarch64__
+        do {
+            COPY16(op, ip);
+        }
+        while (op < oend);
+#else
+        COPY16(op, ip);
+        if (op >= oend) return;
+        do {
+            COPY16(op, ip);
+            COPY16(op, ip);
+        }
+        while (op < oend);
+#endif
+    }
 }
 
-MEM_STATIC void ZSTD_wildcopy_e(void* dst, const void* src, void* dstEnd)   /* should be faster for decoding, but strangely, not verified on all platform */
+MEM_STATIC size_t ZSTD_limitCopy(void* dst, size_t dstCapacity, const void* src, size_t srcSize)
 {
-    const BYTE* ip = (const BYTE*)src;
-    BYTE* op = (BYTE*)dst;
-    BYTE* const oend = (BYTE*)dstEnd;
-    do
-        COPY8(op, ip)
-    while (op < oend);
+    size_t const length = MIN(dstCapacity, srcSize);
+    if (length > 0) {
+        memcpy(dst, src, length);
+    }
+    return length;
 }
 
+/* define "workspace is too large" as this number of times larger than needed */
+#define ZSTD_WORKSPACETOOLARGE_FACTOR 3
+
+/* when workspace is continuously too large
+ * during at least this number of times,
+ * context's memory usage is considered wasteful,
+ * because it's sized to handle a worst case scenario which rarely happens.
+ * In which case, resize it down to free some memory */
+#define ZSTD_WORKSPACETOOLARGE_MAXDURATION 128
+
 
 /*-*******************************************
-*  Private interfaces
+*  Private declarations
 *********************************************/
-typedef struct ZSTD_stats_s ZSTD_stats_t;
-
 typedef struct seqDef_s {
     U32 offset;
     U16 litLength;
     U16 matchLength;
 } seqDef;
 
-
 typedef struct {
     seqDef* sequencesStart;
     seqDef* sequences;
@@ -207,105 +340,50 @@ typedef struct {
     BYTE* llCode;
     BYTE* mlCode;
     BYTE* ofCode;
+    size_t maxNbSeq;
+    size_t maxNbLit;
     U32   longLengthID;   /* 0 == no longLength; 1 == Lit.longLength; 2 == Match.longLength; */
     U32   longLengthPos;
-    U32   rep[ZSTD_REP_NUM];
-    U32   repToConfirm[ZSTD_REP_NUM];
 } seqStore_t;
 
 typedef struct {
-    U32 off;
-    U32 len;
-} ZSTD_match_t;
-
-typedef struct {
-    U32 price;
-    U32 off;
-    U32 mlen;
-    U32 litlen;
-    U32 rep[ZSTD_REP_NUM];
-} ZSTD_optimal_t;
-
-typedef struct {
-    U32* litFreq;
-    U32* litLengthFreq;
-    U32* matchLengthFreq;
-    U32* offCodeFreq;
-    ZSTD_match_t* matchTable;
-    ZSTD_optimal_t* priceTable;
-
-    U32  matchLengthSum;
-    U32  matchSum;
-    U32  litLengthSum;
-    U32  litSum;
-    U32  offCodeSum;
-    U32  log2matchLengthSum;
-    U32  log2matchSum;
-    U32  log2litLengthSum;
-    U32  log2litSum;
-    U32  log2offCodeSum;
-    U32  factor;
-    U32  staticPrices;
-    U32  cachedPrice;
-    U32  cachedLitLength;
-    const BYTE* cachedLiterals;
-} optState_t;
-
-typedef struct {
-    U32 offset;
-    U32 checksum;
-} ldmEntry_t;
-
-typedef struct {
-    ldmEntry_t* hashTable;
-    BYTE* bucketOffsets;    /* Next position in bucket to insert entry */
-    U64 hashPower;          /* Used to compute the rolling hash.
-                             * Depends on ldmParams.minMatchLength */
-} ldmState_t;
+    U32 litLength;
+    U32 matchLength;
+} ZSTD_sequenceLength;
 
-typedef struct {
-    U32 enableLdm;          /* 1 if enable long distance matching */
-    U32 hashLog;            /* Log size of hashTable */
-    U32 bucketSizeLog;      /* Log bucket size for collision resolution, at most 8 */
-    U32 minMatchLength;     /* Minimum match length */
-    U32 hashEveryLog;       /* Log number of entries to skip */
-} ldmParams_t;
+/**
+ * Returns the ZSTD_sequenceLength for the given sequences. It handles the decoding of long sequences
+ * indicated by longLengthPos and longLengthID, and adds MINMATCH back to matchLength.
+ */
+MEM_STATIC ZSTD_sequenceLength ZSTD_getSequenceLength(seqStore_t const* seqStore, seqDef const* seq)
+{
+    ZSTD_sequenceLength seqLen;
+    seqLen.litLength = seq->litLength;
+    seqLen.matchLength = seq->matchLength + MINMATCH;
+    if (seqStore->longLengthPos == (U32)(seq - seqStore->sequencesStart)) {
+        if (seqStore->longLengthID == 1) {
+            seqLen.litLength += 0xFFFF;
+        }
+        if (seqStore->longLengthID == 2) {
+            seqLen.matchLength += 0xFFFF;
+        }
+    }
+    return seqLen;
+}
 
+/**
+ * Contains the compressed frame size and an upper-bound for the decompressed frame size.
+ * Note: before using `compressedSize`, check for errors using ZSTD_isError().
+ *       similarly, before using `decompressedBound`, check for errors using:
+ *          `decompressedBound != ZSTD_CONTENTSIZE_ERROR`
+ */
 typedef struct {
-    U32 hufCTable[HUF_CTABLE_SIZE_U32(255)];
-    FSE_CTable offcodeCTable[FSE_CTABLE_SIZE_U32(OffFSELog, MaxOff)];
-    FSE_CTable matchlengthCTable[FSE_CTABLE_SIZE_U32(MLFSELog, MaxML)];
-    FSE_CTable litlengthCTable[FSE_CTABLE_SIZE_U32(LLFSELog, MaxLL)];
-    U32 workspace[HUF_WORKSPACE_SIZE_U32];
-    HUF_repeat hufCTable_repeatMode;
-    FSE_repeat offcode_repeatMode;
-    FSE_repeat matchlength_repeatMode;
-    FSE_repeat litlength_repeatMode;
-} ZSTD_entropyCTables_t;
-
-struct ZSTD_CCtx_params_s {
-    ZSTD_compressionParameters cParams;
-    ZSTD_frameParameters fParams;
-
-    int compressionLevel;
-    U32 forceWindow;           /* force back-references to respect limit of
-                                * 1<<wLog, even for dictionary */
+    size_t compressedSize;
+    unsigned long long decompressedBound;
+} ZSTD_frameSizeInfo;   /* decompress & legacy */
 
-    /* Multithreading: used to pass parameters to mtctx */
-    U32 nbThreads;
-    unsigned jobSize;
-    unsigned overlapSizeLog;
-
-    /* Long distance matching parameters */
-    ldmParams_t ldmParams;
-
-    /* For use with createCCtxParams() and freeCCtxParams() only */
-    ZSTD_customMem customMem;
-
-};  /* typedef'd to ZSTD_CCtx_params within "zstd.h" */
-
-const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx);
-void ZSTD_seqToCodes(const seqStore_t* seqStorePtr);
+const seqStore_t* ZSTD_getSeqStore(const ZSTD_CCtx* ctx);   /* compress & dictBuilder */
+void ZSTD_seqToCodes(const seqStore_t* seqStorePtr);   /* compress, dictBuilder, decodeCorpus (shouldn't get its definition from here) */
 
 /* custom memory allocation functions */
 void* ZSTD_malloc(size_t size, ZSTD_customMem customMem);
@@ -313,91 +391,57 @@ void* ZSTD_calloc(size_t size, ZSTD_customMem customMem);
 void ZSTD_free(void* ptr, ZSTD_customMem customMem);
 
 
-/*======  common function  ======*/
-
-MEM_STATIC U32 ZSTD_highbit32(U32 val)
+MEM_STATIC U32 ZSTD_highbit32(U32 val)   /* compress, dictBuilder, decodeCorpus */
 {
     assert(val != 0);
     {
 #   if defined(_MSC_VER)   /* Visual */
         unsigned long r=0;
-        _BitScanReverse(&r, val);
-        return (unsigned)r;
+        return _BitScanReverse(&r, val) ? (unsigned)r : 0;
 #   elif defined(__GNUC__) && (__GNUC__ >= 3)   /* GCC Intrinsic */
-        return 31 - __builtin_clz(val);
+        return __builtin_clz (val) ^ 31;
+#   elif defined(__ICCARM__)    /* IAR Intrinsic */
+        return 31 - __CLZ(val);
 #   else   /* Software version */
-        static const int DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
+        static const U32 DeBruijnClz[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 };
         U32 v = val;
-        int r;
         v |= v >> 1;
         v |= v >> 2;
         v |= v >> 4;
         v |= v >> 8;
         v |= v >> 16;
-        r = DeBruijnClz[(U32)(v * 0x07C4ACDDU) >> 27];
-        return r;
+        return DeBruijnClz[(v * 0x07C4ACDDU) >> 27];
 #   endif
     }
 }
 
 
-/* hidden functions */
-
 /* ZSTD_invalidateRepCodes() :
  * ensures next compression will not use repcodes from previous block.
  * Note : only works with regular variant;
  *        do not use with extDict variant ! */
-void ZSTD_invalidateRepCodes(ZSTD_CCtx* cctx);
-
-
-/*! ZSTD_initCStream_internal() :
- *  Private use only. Init streaming operation.
- *  expects params to be valid.
- *  must receive dict, or cdict, or none, but not both.
- *  @return : 0, or an error code */
-size_t ZSTD_initCStream_internal(ZSTD_CStream* zcs,
-                     const void* dict, size_t dictSize,
-                     const ZSTD_CDict* cdict,
-                     ZSTD_CCtx_params  params, unsigned long long pledgedSrcSize);
-
-/*! ZSTD_compressStream_generic() :
- *  Private use only. To be called from zstdmt_compress.c in single-thread mode. */
-size_t ZSTD_compressStream_generic(ZSTD_CStream* zcs,
-                                   ZSTD_outBuffer* output,
-                                   ZSTD_inBuffer* input,
-                                   ZSTD_EndDirective const flushMode);
-
-/*! ZSTD_getCParamsFromCDict() :
- *  as the name implies */
-ZSTD_compressionParameters ZSTD_getCParamsFromCDict(const ZSTD_CDict* cdict);
-
-/* ZSTD_compressBegin_advanced_internal() :
- * Private use only. To be called from zstdmt_compress.c. */
-size_t ZSTD_compressBegin_advanced_internal(ZSTD_CCtx* cctx,
-                                    const void* dict, size_t dictSize,
-                                    ZSTD_dictMode_e dictMode,
-                                    ZSTD_CCtx_params params,
-                                    unsigned long long pledgedSrcSize);
-
-/* ZSTD_compress_advanced_internal() :
- * Private use only. To be called from zstdmt_compress.c. */
-size_t ZSTD_compress_advanced_internal(ZSTD_CCtx* cctx,
-                                       void* dst, size_t dstCapacity,
-                                 const void* src, size_t srcSize,
-                                 const void* dict,size_t dictSize,
-                                 ZSTD_CCtx_params params);
+void ZSTD_invalidateRepCodes(ZSTD_CCtx* cctx);   /* zstdmt, adaptive_compression (shouldn't get this definition from here) */
+
 
 typedef struct {
     blockType_e blockType;
     U32 lastBlock;
     U32 origSize;
-} blockProperties_t;
+} blockProperties_t;   /* declared here for decompress and fullbench */
 
 /*! ZSTD_getcBlockSize() :
-*   Provides the size of compressed block from block header `src` */
+ *  Provides the size of compressed block from block header `src` */
+/* Used by: decompress, fullbench (does not get its definition from here) */
 size_t ZSTD_getcBlockSize(const void* src, size_t srcSize,
                           blockProperties_t* bpPtr);
 
+/*! ZSTD_decodeSeqHeaders() :
+ *  decode sequence header from src */
+/* Used by: decompress, fullbench (does not get its definition from here) */
+size_t ZSTD_decodeSeqHeaders(ZSTD_DCtx* dctx, int* nbSeqPtr,
+                       const void* src, size_t srcSize);
+
+
 #if defined (__cplusplus)
 }
 #endif
diff --git a/contrib/zstd/zstd_lazy.c b/contrib/zstd/zstd_lazy.c
index 2a7f6a0fe..4cf5c88b5 100644
--- a/contrib/zstd/zstd_lazy.c
+++ b/contrib/zstd/zstd_lazy.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -8,273 +8,432 @@
  * You may select, at your option, one of the above-listed licenses.
  */
 
+#include "zstd_compress_internal.h"
 #include "zstd_lazy.h"
 
 
 /*-*************************************
 *  Binary Tree search
 ***************************************/
-/** ZSTD_insertBt1() : add one or multiple positions to tree.
-*   ip : assumed <= iend-8 .
-*   @return : nb of positions added */
-static U32 ZSTD_insertBt1(ZSTD_CCtx* zc, const BYTE* const ip, const U32 mls, const BYTE* const iend, U32 nbCompares,
-                          U32 extDict)
+
+static void
+ZSTD_updateDUBT(ZSTD_matchState_t* ms,
+                const BYTE* ip, const BYTE* iend,
+                U32 mls)
 {
-    U32*   const hashTable = zc->hashTable;
-    U32    const hashLog = zc->appliedParams.cParams.hashLog;
-    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
-    U32*   const bt = zc->chainTable;
-    U32    const btLog  = zc->appliedParams.cParams.chainLog - 1;
-    U32    const btMask = (1 << btLog) - 1;
-    U32 matchIndex = hashTable[h];
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const hashTable = ms->hashTable;
+    U32  const hashLog = cParams->hashLog;
+
+    U32* const bt = ms->chainTable;
+    U32  const btLog  = cParams->chainLog - 1;
+    U32  const btMask = (1 << btLog) - 1;
+
+    const BYTE* const base = ms->window.base;
+    U32 const target = (U32)(ip - base);
+    U32 idx = ms->nextToUpdate;
+
+    if (idx != target)
+        DEBUGLOG(7, "ZSTD_updateDUBT, from %u to %u (dictLimit:%u)",
+                    idx, target, ms->window.dictLimit);
+    assert(ip + 8 <= iend);   /* condition for ZSTD_hashPtr */
+    (void)iend;
+
+    assert(idx >= ms->window.dictLimit);   /* condition for valid base+idx */
+    for ( ; idx < target ; idx++) {
+        size_t const h  = ZSTD_hashPtr(base + idx, hashLog, mls);   /* assumption : ip + 8 <= iend */
+        U32    const matchIndex = hashTable[h];
+
+        U32*   const nextCandidatePtr = bt + 2*(idx&btMask);
+        U32*   const sortMarkPtr  = nextCandidatePtr + 1;
+
+        DEBUGLOG(8, "ZSTD_updateDUBT: insert %u", idx);
+        hashTable[h] = idx;   /* Update Hash Table */
+        *nextCandidatePtr = matchIndex;   /* update BT like a chain */
+        *sortMarkPtr = ZSTD_DUBT_UNSORTED_MARK;
+    }
+    ms->nextToUpdate = target;
+}
+
+
+/** ZSTD_insertDUBT1() :
+ *  sort one already inserted but unsorted position
+ *  assumption : current >= btlow == (current - btmask)
+ *  doesn't fail */
+static void
+ZSTD_insertDUBT1(ZSTD_matchState_t* ms,
+                 U32 current, const BYTE* inputEnd,
+                 U32 nbCompares, U32 btLow,
+                 const ZSTD_dictMode_e dictMode)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const bt = ms->chainTable;
+    U32  const btLog  = cParams->chainLog - 1;
+    U32  const btMask = (1 << btLog) - 1;
     size_t commonLengthSmaller=0, commonLengthLarger=0;
-    const BYTE* const base = zc->base;
-    const BYTE* const dictBase = zc->dictBase;
-    const U32 dictLimit = zc->dictLimit;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const ip = (current>=dictLimit) ? base + current : dictBase + current;
+    const BYTE* const iend = (current>=dictLimit) ? inputEnd : dictBase + dictLimit;
     const BYTE* const dictEnd = dictBase + dictLimit;
     const BYTE* const prefixStart = base + dictLimit;
     const BYTE* match;
-    const U32 current = (U32)(ip-base);
-    const U32 btLow = btMask >= current ? 0 : current - btMask;
     U32* smallerPtr = bt + 2*(current&btMask);
     U32* largerPtr  = smallerPtr + 1;
+    U32 matchIndex = *smallerPtr;   /* this candidate is unsorted : next sorted candidate is reached through *smallerPtr, while *largerPtr contains previous unsorted candidate (which is already saved and can be overwritten) */
     U32 dummy32;   /* to be nullified at the end */
-    U32 const windowLow = zc->lowLimit;
-    U32 matchEndIdx = current+8;
-    size_t bestLength = 8;
-#ifdef ZSTD_C_PREDICT
-    U32 predictedSmall = *(bt + 2*((current-1)&btMask) + 0);
-    U32 predictedLarge = *(bt + 2*((current-1)&btMask) + 1);
-    predictedSmall += (predictedSmall>0);
-    predictedLarge += (predictedLarge>0);
-#endif /* ZSTD_C_PREDICT */
+    U32 const windowValid = ms->window.lowLimit;
+    U32 const maxDistance = 1U << cParams->windowLog;
+    U32 const windowLow = (current - windowValid > maxDistance) ? current - maxDistance : windowValid;
 
-    assert(ip <= iend-8);   /* required for h calculation */
-    hashTable[h] = current;   /* Update Hash Table */
+
+    DEBUGLOG(8, "ZSTD_insertDUBT1(%u) (dictLimit=%u, lowLimit=%u)",
+                current, dictLimit, windowLow);
+    assert(current >= btLow);
+    assert(ip < iend);   /* condition for ZSTD_count */
 
     while (nbCompares-- && (matchIndex > windowLow)) {
         U32* const nextPtr = bt + 2*(matchIndex & btMask);
         size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
-
-#ifdef ZSTD_C_PREDICT   /* note : can create issues when hlog small <= 11 */
-        const U32* predictPtr = bt + 2*((matchIndex-1) & btMask);   /* written this way, as bt is a roll buffer */
-        if (matchIndex == predictedSmall) {
-            /* no need to check length, result known */
-            *smallerPtr = matchIndex;
-            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
-            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
-            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
-            predictedSmall = predictPtr[1] + (predictPtr[1]>0);
-            continue;
-        }
-        if (matchIndex == predictedLarge) {
-            *largerPtr = matchIndex;
-            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
-            largerPtr = nextPtr;
-            matchIndex = nextPtr[0];
-            predictedLarge = predictPtr[0] + (predictPtr[0]>0);
-            continue;
-        }
-#endif
-        if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
-            match = base + matchIndex;
-            if (match[matchLength] == ip[matchLength])
-                matchLength += ZSTD_count(ip+matchLength+1, match+matchLength+1, iend) +1;
+        assert(matchIndex < current);
+        /* note : all candidates are now supposed sorted,
+         * but it's still possible to have nextPtr[1] == ZSTD_DUBT_UNSORTED_MARK
+         * when a real index has the same value as ZSTD_DUBT_UNSORTED_MARK */
+
+        if ( (dictMode != ZSTD_extDict)
+          || (matchIndex+matchLength >= dictLimit)  /* both in current segment*/
+          || (current < dictLimit) /* both in extDict */) {
+            const BYTE* const mBase = ( (dictMode != ZSTD_extDict)
+                                     || (matchIndex+matchLength >= dictLimit)) ?
+                                        base : dictBase;
+            assert( (matchIndex+matchLength >= dictLimit)   /* might be wrong if extDict is incorrectly set to 0 */
+                 || (current < dictLimit) );
+            match = mBase + matchIndex;
+            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
         } else {
             match = dictBase + matchIndex;
             matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
             if (matchIndex+matchLength >= dictLimit)
-                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+                match = base + matchIndex;   /* preparation for next read of match[matchLength] */
         }
 
-        if (matchLength > bestLength) {
-            bestLength = matchLength;
-            if (matchLength > matchEndIdx - matchIndex)
-                matchEndIdx = matchIndex + (U32)matchLength;
-        }
+        DEBUGLOG(8, "ZSTD_insertDUBT1: comparing %u with %u : found %u common bytes ",
+                    current, matchIndex, (U32)matchLength);
 
-        if (ip+matchLength == iend)   /* equal : no way to know if inf or sup */
+        if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
             break;   /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt tree */
+        }
 
         if (match[matchLength] < ip[matchLength]) {  /* necessarily within buffer */
-            /* match+1 is smaller than current */
+            /* match is smaller than current */
             *smallerPtr = matchIndex;             /* update smaller idx */
             commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
             if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
-            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
-            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            DEBUGLOG(8, "ZSTD_insertDUBT1: %u (>btLow=%u) is smaller : next => %u",
+                        matchIndex, btLow, nextPtr[1]);
+            smallerPtr = nextPtr+1;               /* new "candidate" => larger than match, which was smaller than target */
+            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous and closer to current */
         } else {
             /* match is larger than current */
             *largerPtr = matchIndex;
             commonLengthLarger = matchLength;
             if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            DEBUGLOG(8, "ZSTD_insertDUBT1: %u (>btLow=%u) is larger => %u",
+                        matchIndex, btLow, nextPtr[0]);
             largerPtr = nextPtr;
             matchIndex = nextPtr[0];
     }   }
 
     *smallerPtr = *largerPtr = 0;
-    if (bestLength > 384) return MIN(192, (U32)(bestLength - 384));   /* speed optimization */
-    if (matchEndIdx > current + 8) return matchEndIdx - (current + 8);
-    return 1;
 }
 
 
-static size_t ZSTD_insertBtAndFindBestMatch (
-                        ZSTD_CCtx* zc,
-                        const BYTE* const ip, const BYTE* const iend,
-                        size_t* offsetPtr,
-                        U32 nbCompares, const U32 mls,
-                        U32 extDict)
+static size_t
+ZSTD_DUBT_findBetterDictMatch (
+        ZSTD_matchState_t* ms,
+        const BYTE* const ip, const BYTE* const iend,
+        size_t* offsetPtr,
+        size_t bestLength,
+        U32 nbCompares,
+        U32 const mls,
+        const ZSTD_dictMode_e dictMode)
 {
-    U32*   const hashTable = zc->hashTable;
-    U32    const hashLog = zc->appliedParams.cParams.hashLog;
-    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
-    U32*   const bt = zc->chainTable;
-    U32    const btLog  = zc->appliedParams.cParams.chainLog - 1;
-    U32    const btMask = (1 << btLog) - 1;
-    U32 matchIndex  = hashTable[h];
+    const ZSTD_matchState_t * const dms = ms->dictMatchState;
+    const ZSTD_compressionParameters* const dmsCParams = &dms->cParams;
+    const U32 * const dictHashTable = dms->hashTable;
+    U32         const hashLog = dmsCParams->hashLog;
+    size_t      const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32               dictMatchIndex = dictHashTable[h];
+
+    const BYTE* const base = ms->window.base;
+    const BYTE* const prefixStart = base + ms->window.dictLimit;
+    U32         const current = (U32)(ip-base);
+    const BYTE* const dictBase = dms->window.base;
+    const BYTE* const dictEnd = dms->window.nextSrc;
+    U32         const dictHighLimit = (U32)(dms->window.nextSrc - dms->window.base);
+    U32         const dictLowLimit = dms->window.lowLimit;
+    U32         const dictIndexDelta = ms->window.lowLimit - dictHighLimit;
+
+    U32*        const dictBt = dms->chainTable;
+    U32         const btLog  = dmsCParams->chainLog - 1;
+    U32         const btMask = (1 << btLog) - 1;
+    U32         const btLow = (btMask >= dictHighLimit - dictLowLimit) ? dictLowLimit : dictHighLimit - btMask;
+
     size_t commonLengthSmaller=0, commonLengthLarger=0;
-    const BYTE* const base = zc->base;
-    const BYTE* const dictBase = zc->dictBase;
-    const U32 dictLimit = zc->dictLimit;
-    const BYTE* const dictEnd = dictBase + dictLimit;
-    const BYTE* const prefixStart = base + dictLimit;
-    const U32 current = (U32)(ip-base);
-    const U32 btLow = btMask >= current ? 0 : current - btMask;
-    const U32 windowLow = zc->lowLimit;
-    U32* smallerPtr = bt + 2*(current&btMask);
-    U32* largerPtr  = bt + 2*(current&btMask) + 1;
-    U32 matchEndIdx = current+8;
-    U32 dummy32;   /* to be nullified at the end */
-    size_t bestLength = 0;
 
-    assert(ip <= iend-8);   /* required for h calculation */
-    hashTable[h] = current;   /* Update Hash Table */
+    (void)dictMode;
+    assert(dictMode == ZSTD_dictMatchState);
 
-    while (nbCompares-- && (matchIndex > windowLow)) {
-        U32* const nextPtr = bt + 2*(matchIndex & btMask);
+    while (nbCompares-- && (dictMatchIndex > dictLowLimit)) {
+        U32* const nextPtr = dictBt + 2*(dictMatchIndex & btMask);
         size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
-        const BYTE* match;
-
-        if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
-            match = base + matchIndex;
-            if (match[matchLength] == ip[matchLength])
-                matchLength += ZSTD_count(ip+matchLength+1, match+matchLength+1, iend) +1;
-        } else {
-            match = dictBase + matchIndex;
-            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
-            if (matchIndex+matchLength >= dictLimit)
-                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
-        }
+        const BYTE* match = dictBase + dictMatchIndex;
+        matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+        if (dictMatchIndex+matchLength >= dictHighLimit)
+            match = base + dictMatchIndex + dictIndexDelta;   /* to prepare for next usage of match[matchLength] */
 
         if (matchLength > bestLength) {
-            if (matchLength > matchEndIdx - matchIndex)
-                matchEndIdx = matchIndex + (U32)matchLength;
-            if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(current-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) )
+            U32 matchIndex = dictMatchIndex + dictIndexDelta;
+            if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(current-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) ) {
+                DEBUGLOG(9, "ZSTD_DUBT_findBetterDictMatch(%u) : found better match length %u -> %u and offsetCode %u -> %u (dictMatchIndex %u, matchIndex %u)",
+                    current, (U32)bestLength, (U32)matchLength, (U32)*offsetPtr, ZSTD_REP_MOVE + current - matchIndex, dictMatchIndex, matchIndex);
                 bestLength = matchLength, *offsetPtr = ZSTD_REP_MOVE + current - matchIndex;
-            if (ip+matchLength == iend)   /* equal : no way to know if inf or sup */
+            }
+            if (ip+matchLength == iend) {   /* reached end of input : ip[matchLength] is not valid, no way to know if it's larger or smaller than match */
                 break;   /* drop, to guarantee consistency (miss a little bit of compression) */
+            }
         }
 
         if (match[matchLength] < ip[matchLength]) {
-            /* match is smaller than current */
-            *smallerPtr = matchIndex;             /* update smaller idx */
+            if (dictMatchIndex <= btLow) { break; }   /* beyond tree size, stop the search */
             commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
-            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
-            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
-            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            dictMatchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
         } else {
             /* match is larger than current */
-            *largerPtr = matchIndex;
+            if (dictMatchIndex <= btLow) { break; }   /* beyond tree size, stop the search */
             commonLengthLarger = matchLength;
-            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
-            largerPtr = nextPtr;
-            matchIndex = nextPtr[0];
-    }   }
-
-    *smallerPtr = *largerPtr = 0;
+            dictMatchIndex = nextPtr[0];
+        }
+    }
 
-    zc->nextToUpdate = (matchEndIdx > current + 8) ? matchEndIdx - 8 : current+1;
+    if (bestLength >= MINMATCH) {
+        U32 const mIndex = current - ((U32)*offsetPtr - ZSTD_REP_MOVE); (void)mIndex;
+        DEBUGLOG(8, "ZSTD_DUBT_findBetterDictMatch(%u) : found match of length %u and offsetCode %u (pos %u)",
+                    current, (U32)bestLength, (U32)*offsetPtr, mIndex);
+    }
     return bestLength;
+
 }
 
 
-void ZSTD_updateTree(ZSTD_CCtx* zc, const BYTE* const ip, const BYTE* const iend, const U32 nbCompares, const U32 mls)
+static size_t
+ZSTD_DUBT_findBestMatch(ZSTD_matchState_t* ms,
+                        const BYTE* const ip, const BYTE* const iend,
+                        size_t* offsetPtr,
+                        U32 const mls,
+                        const ZSTD_dictMode_e dictMode)
 {
-    const BYTE* const base = zc->base;
-    const U32 target = (U32)(ip - base);
-    U32 idx = zc->nextToUpdate;
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32*   const hashTable = ms->hashTable;
+    U32    const hashLog = cParams->hashLog;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32          matchIndex  = hashTable[h];
+
+    const BYTE* const base = ms->window.base;
+    U32    const current = (U32)(ip-base);
+    U32    const windowLow = ZSTD_getLowestMatchIndex(ms, current, cParams->windowLog);
+
+    U32*   const bt = ms->chainTable;
+    U32    const btLog  = cParams->chainLog - 1;
+    U32    const btMask = (1 << btLog) - 1;
+    U32    const btLow = (btMask >= current) ? 0 : current - btMask;
+    U32    const unsortLimit = MAX(btLow, windowLow);
+
+    U32*         nextCandidate = bt + 2*(matchIndex&btMask);
+    U32*         unsortedMark = bt + 2*(matchIndex&btMask) + 1;
+    U32          nbCompares = 1U << cParams->searchLog;
+    U32          nbCandidates = nbCompares;
+    U32          previousCandidate = 0;
+
+    DEBUGLOG(7, "ZSTD_DUBT_findBestMatch (%u) ", current);
+    assert(ip <= iend-8);   /* required for h calculation */
+
+    /* reach end of unsorted candidates list */
+    while ( (matchIndex > unsortLimit)
+         && (*unsortedMark == ZSTD_DUBT_UNSORTED_MARK)
+         && (nbCandidates > 1) ) {
+        DEBUGLOG(8, "ZSTD_DUBT_findBestMatch: candidate %u is unsorted",
+                    matchIndex);
+        *unsortedMark = previousCandidate;  /* the unsortedMark becomes a reversed chain, to move up back to original position */
+        previousCandidate = matchIndex;
+        matchIndex = *nextCandidate;
+        nextCandidate = bt + 2*(matchIndex&btMask);
+        unsortedMark = bt + 2*(matchIndex&btMask) + 1;
+        nbCandidates --;
+    }
+
+    /* nullify last candidate if it's still unsorted
+     * simplification, detrimental to compression ratio, beneficial for speed */
+    if ( (matchIndex > unsortLimit)
+      && (*unsortedMark==ZSTD_DUBT_UNSORTED_MARK) ) {
+        DEBUGLOG(7, "ZSTD_DUBT_findBestMatch: nullify last unsorted candidate %u",
+                    matchIndex);
+        *nextCandidate = *unsortedMark = 0;
+    }
+
+    /* batch sort stacked candidates */
+    matchIndex = previousCandidate;
+    while (matchIndex) {  /* will end on matchIndex == 0 */
+        U32* const nextCandidateIdxPtr = bt + 2*(matchIndex&btMask) + 1;
+        U32 const nextCandidateIdx = *nextCandidateIdxPtr;
+        ZSTD_insertDUBT1(ms, matchIndex, iend,
+                         nbCandidates, unsortLimit, dictMode);
+        matchIndex = nextCandidateIdx;
+        nbCandidates++;
+    }
+
+    /* find longest match */
+    {   size_t commonLengthSmaller = 0, commonLengthLarger = 0;
+        const BYTE* const dictBase = ms->window.dictBase;
+        const U32 dictLimit = ms->window.dictLimit;
+        const BYTE* const dictEnd = dictBase + dictLimit;
+        const BYTE* const prefixStart = base + dictLimit;
+        U32* smallerPtr = bt + 2*(current&btMask);
+        U32* largerPtr  = bt + 2*(current&btMask) + 1;
+        U32 matchEndIdx = current + 8 + 1;
+        U32 dummy32;   /* to be nullified at the end */
+        size_t bestLength = 0;
+
+        matchIndex  = hashTable[h];
+        hashTable[h] = current;   /* Update Hash Table */
+
+        while (nbCompares-- && (matchIndex > windowLow)) {
+            U32* const nextPtr = bt + 2*(matchIndex & btMask);
+            size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+            const BYTE* match;
+
+            if ((dictMode != ZSTD_extDict) || (matchIndex+matchLength >= dictLimit)) {
+                match = base + matchIndex;
+                matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
+            } else {
+                match = dictBase + matchIndex;
+                matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+                if (matchIndex+matchLength >= dictLimit)
+                    match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+            }
+
+            if (matchLength > bestLength) {
+                if (matchLength > matchEndIdx - matchIndex)
+                    matchEndIdx = matchIndex + (U32)matchLength;
+                if ( (4*(int)(matchLength-bestLength)) > (int)(ZSTD_highbit32(current-matchIndex+1) - ZSTD_highbit32((U32)offsetPtr[0]+1)) )
+                    bestLength = matchLength, *offsetPtr = ZSTD_REP_MOVE + current - matchIndex;
+                if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
+                    if (dictMode == ZSTD_dictMatchState) {
+                        nbCompares = 0; /* in addition to avoiding checking any
+                                         * further in this loop, make sure we
+                                         * skip checking in the dictionary. */
+                    }
+                    break;   /* drop, to guarantee consistency (miss a little bit of compression) */
+                }
+            }
+
+            if (match[matchLength] < ip[matchLength]) {
+                /* match is smaller than current */
+                *smallerPtr = matchIndex;             /* update smaller idx */
+                commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+                if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+                smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
+                matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            } else {
+                /* match is larger than current */
+                *largerPtr = matchIndex;
+                commonLengthLarger = matchLength;
+                if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+                largerPtr = nextPtr;
+                matchIndex = nextPtr[0];
+        }   }
+
+        *smallerPtr = *largerPtr = 0;
+
+        if (dictMode == ZSTD_dictMatchState && nbCompares) {
+            bestLength = ZSTD_DUBT_findBetterDictMatch(
+                    ms, ip, iend,
+                    offsetPtr, bestLength, nbCompares,
+                    mls, dictMode);
+        }
 
-    while(idx < target)
-        idx += ZSTD_insertBt1(zc, base+idx, mls, iend, nbCompares, 0);
+        assert(matchEndIdx > current+8); /* ensure nextToUpdate is increased */
+        ms->nextToUpdate = matchEndIdx - 8;   /* skip repetitive patterns */
+        if (bestLength >= MINMATCH) {
+            U32 const mIndex = current - ((U32)*offsetPtr - ZSTD_REP_MOVE); (void)mIndex;
+            DEBUGLOG(8, "ZSTD_DUBT_findBestMatch(%u) : found match of length %u and offsetCode %u (pos %u)",
+                        current, (U32)bestLength, (U32)*offsetPtr, mIndex);
+        }
+        return bestLength;
+    }
 }
 
+
 /** ZSTD_BtFindBestMatch() : Tree updater, providing best match */
-static size_t ZSTD_BtFindBestMatch (
-                        ZSTD_CCtx* zc,
-                        const BYTE* const ip, const BYTE* const iLimit,
-                        size_t* offsetPtr,
-                        const U32 maxNbAttempts, const U32 mls)
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_BtFindBestMatch( ZSTD_matchState_t* ms,
+                const BYTE* const ip, const BYTE* const iLimit,
+                      size_t* offsetPtr,
+                const U32 mls /* template */,
+                const ZSTD_dictMode_e dictMode)
 {
-    if (ip < zc->base + zc->nextToUpdate) return 0;   /* skipped area */
-    ZSTD_updateTree(zc, ip, iLimit, maxNbAttempts, mls);
-    return ZSTD_insertBtAndFindBestMatch(zc, ip, iLimit, offsetPtr, maxNbAttempts, mls, 0);
+    DEBUGLOG(7, "ZSTD_BtFindBestMatch");
+    if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
+    ZSTD_updateDUBT(ms, ip, iLimit, mls);
+    return ZSTD_DUBT_findBestMatch(ms, ip, iLimit, offsetPtr, mls, dictMode);
 }
 
 
-static size_t ZSTD_BtFindBestMatch_selectMLS (
-                        ZSTD_CCtx* zc,   /* Index table will be updated */
-                        const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr,
-                        const U32 maxNbAttempts, const U32 matchLengthSearch)
+static size_t
+ZSTD_BtFindBestMatch_selectMLS (  ZSTD_matchState_t* ms,
+                            const BYTE* ip, const BYTE* const iLimit,
+                                  size_t* offsetPtr)
 {
-    switch(matchLengthSearch)
+    switch(ms->cParams.minMatch)
     {
     default : /* includes case 3 */
-    case 4 : return ZSTD_BtFindBestMatch(zc, ip, iLimit, offsetPtr, maxNbAttempts, 4);
-    case 5 : return ZSTD_BtFindBestMatch(zc, ip, iLimit, offsetPtr, maxNbAttempts, 5);
+    case 4 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 4, ZSTD_noDict);
+    case 5 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 5, ZSTD_noDict);
     case 7 :
-    case 6 : return ZSTD_BtFindBestMatch(zc, ip, iLimit, offsetPtr, maxNbAttempts, 6);
+    case 6 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 6, ZSTD_noDict);
     }
 }
 
 
-void ZSTD_updateTree_extDict(ZSTD_CCtx* zc, const BYTE* const ip, const BYTE* const iend, const U32 nbCompares, const U32 mls)
-{
-    const BYTE* const base = zc->base;
-    const U32 target = (U32)(ip - base);
-    U32 idx = zc->nextToUpdate;
-
-    while (idx < target) idx += ZSTD_insertBt1(zc, base+idx, mls, iend, nbCompares, 1);
-}
-
-
-/** Tree updater, providing best match */
-static size_t ZSTD_BtFindBestMatch_extDict (
-                        ZSTD_CCtx* zc,
-                        const BYTE* const ip, const BYTE* const iLimit,
-                        size_t* offsetPtr,
-                        const U32 maxNbAttempts, const U32 mls)
+static size_t ZSTD_BtFindBestMatch_dictMatchState_selectMLS (
+                        ZSTD_matchState_t* ms,
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* offsetPtr)
 {
-    if (ip < zc->base + zc->nextToUpdate) return 0;   /* skipped area */
-    ZSTD_updateTree_extDict(zc, ip, iLimit, maxNbAttempts, mls);
-    return ZSTD_insertBtAndFindBestMatch(zc, ip, iLimit, offsetPtr, maxNbAttempts, mls, 1);
+    switch(ms->cParams.minMatch)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 4, ZSTD_dictMatchState);
+    case 5 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 5, ZSTD_dictMatchState);
+    case 7 :
+    case 6 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 6, ZSTD_dictMatchState);
+    }
 }
 
 
-static size_t ZSTD_BtFindBestMatch_selectMLS_extDict (
-                        ZSTD_CCtx* zc,   /* Index table will be updated */
+static size_t ZSTD_BtFindBestMatch_extDict_selectMLS (
+                        ZSTD_matchState_t* ms,
                         const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr,
-                        const U32 maxNbAttempts, const U32 matchLengthSearch)
+                        size_t* offsetPtr)
 {
-    switch(matchLengthSearch)
+    switch(ms->cParams.minMatch)
     {
     default : /* includes case 3 */
-    case 4 : return ZSTD_BtFindBestMatch_extDict(zc, ip, iLimit, offsetPtr, maxNbAttempts, 4);
-    case 5 : return ZSTD_BtFindBestMatch_extDict(zc, ip, iLimit, offsetPtr, maxNbAttempts, 5);
+    case 4 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 4, ZSTD_extDict);
+    case 5 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 5, ZSTD_extDict);
     case 7 :
-    case 6 : return ZSTD_BtFindBestMatch_extDict(zc, ip, iLimit, offsetPtr, maxNbAttempts, 6);
+    case 6 : return ZSTD_BtFindBestMatch(ms, ip, iLimit, offsetPtr, 6, ZSTD_extDict);
     }
 }
 
@@ -283,19 +442,22 @@ static size_t ZSTD_BtFindBestMatch_selectMLS_extDict (
 /* *********************************
 *  Hash Chain
 ***********************************/
-#define NEXT_IN_CHAIN(d, mask)   chainTable[(d) & mask]
+#define NEXT_IN_CHAIN(d, mask)   chainTable[(d) & (mask)]
 
 /* Update chains up to ip (excluded)
    Assumption : always within prefix (i.e. not within extDict) */
-U32 ZSTD_insertAndFindFirstIndex (ZSTD_CCtx* zc, const BYTE* ip, U32 mls)
+static U32 ZSTD_insertAndFindFirstIndex_internal(
+                        ZSTD_matchState_t* ms,
+                        const ZSTD_compressionParameters* const cParams,
+                        const BYTE* ip, U32 const mls)
 {
-    U32* const hashTable  = zc->hashTable;
-    const U32 hashLog = zc->appliedParams.cParams.hashLog;
-    U32* const chainTable = zc->chainTable;
-    const U32 chainMask = (1 << zc->appliedParams.cParams.chainLog) - 1;
-    const BYTE* const base = zc->base;
+    U32* const hashTable  = ms->hashTable;
+    const U32 hashLog = cParams->hashLog;
+    U32* const chainTable = ms->chainTable;
+    const U32 chainMask = (1 << cParams->chainLog) - 1;
+    const BYTE* const base = ms->window.base;
     const U32 target = (U32)(ip - base);
-    U32 idx = zc->nextToUpdate;
+    U32 idx = ms->nextToUpdate;
 
     while(idx < target) { /* catch up */
         size_t const h = ZSTD_hashPtr(base+idx, hashLog, mls);
@@ -304,45 +466,56 @@ U32 ZSTD_insertAndFindFirstIndex (ZSTD_CCtx* zc, const BYTE* ip, U32 mls)
         idx++;
     }
 
-    zc->nextToUpdate = target;
+    ms->nextToUpdate = target;
     return hashTable[ZSTD_hashPtr(ip, hashLog, mls)];
 }
 
+U32 ZSTD_insertAndFindFirstIndex(ZSTD_matchState_t* ms, const BYTE* ip) {
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    return ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, ms->cParams.minMatch);
+}
+
 
 /* inlining is important to hardwire a hot branch (template emulation) */
 FORCE_INLINE_TEMPLATE
 size_t ZSTD_HcFindBestMatch_generic (
-                        ZSTD_CCtx* zc,   /* Index table will be updated */
+                        ZSTD_matchState_t* ms,
                         const BYTE* const ip, const BYTE* const iLimit,
                         size_t* offsetPtr,
-                        const U32 maxNbAttempts, const U32 mls, const U32 extDict)
+                        const U32 mls, const ZSTD_dictMode_e dictMode)
 {
-    U32* const chainTable = zc->chainTable;
-    const U32 chainSize = (1 << zc->appliedParams.cParams.chainLog);
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32* const chainTable = ms->chainTable;
+    const U32 chainSize = (1 << cParams->chainLog);
     const U32 chainMask = chainSize-1;
-    const BYTE* const base = zc->base;
-    const BYTE* const dictBase = zc->dictBase;
-    const U32 dictLimit = zc->dictLimit;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
     const BYTE* const prefixStart = base + dictLimit;
     const BYTE* const dictEnd = dictBase + dictLimit;
-    const U32 lowLimit = zc->lowLimit;
     const U32 current = (U32)(ip-base);
+    const U32 maxDistance = 1U << cParams->windowLog;
+    const U32 lowestValid = ms->window.lowLimit;
+    const U32 withinMaxDistance = (current - lowestValid > maxDistance) ? current - maxDistance : lowestValid;
+    const U32 isDictionary = (ms->loadedDictEnd != 0);
+    const U32 lowLimit = isDictionary ? lowestValid : withinMaxDistance;
     const U32 minChain = current > chainSize ? current - chainSize : 0;
-    int nbAttempts=maxNbAttempts;
+    U32 nbAttempts = 1U << cParams->searchLog;
     size_t ml=4-1;
 
     /* HC4 match finder */
-    U32 matchIndex = ZSTD_insertAndFindFirstIndex (zc, ip, mls);
+    U32 matchIndex = ZSTD_insertAndFindFirstIndex_internal(ms, cParams, ip, mls);
 
     for ( ; (matchIndex>lowLimit) & (nbAttempts>0) ; nbAttempts--) {
-        const BYTE* match;
         size_t currentMl=0;
-        if ((!extDict) || matchIndex >= dictLimit) {
-            match = base + matchIndex;
+        if ((dictMode != ZSTD_extDict) || matchIndex >= dictLimit) {
+            const BYTE* const match = base + matchIndex;
+            assert(matchIndex >= dictLimit);   /* ensures this is true if dictMode != ZSTD_extDict */
             if (match[ml] == ip[ml])   /* potentially better */
                 currentMl = ZSTD_count(ip, match, iLimit);
         } else {
-            match = dictBase + matchIndex;
+            const BYTE* const match = dictBase + matchIndex;
+            assert(match+4 <= dictEnd);
             if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
                 currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dictEnd, prefixStart) + 4;
         }
@@ -358,40 +531,87 @@ size_t ZSTD_HcFindBestMatch_generic (
         matchIndex = NEXT_IN_CHAIN(matchIndex, chainMask);
     }
 
+    if (dictMode == ZSTD_dictMatchState) {
+        const ZSTD_matchState_t* const dms = ms->dictMatchState;
+        const U32* const dmsChainTable = dms->chainTable;
+        const U32 dmsChainSize         = (1 << dms->cParams.chainLog);
+        const U32 dmsChainMask         = dmsChainSize - 1;
+        const U32 dmsLowestIndex       = dms->window.dictLimit;
+        const BYTE* const dmsBase      = dms->window.base;
+        const BYTE* const dmsEnd       = dms->window.nextSrc;
+        const U32 dmsSize              = (U32)(dmsEnd - dmsBase);
+        const U32 dmsIndexDelta        = dictLimit - dmsSize;
+        const U32 dmsMinChain = dmsSize > dmsChainSize ? dmsSize - dmsChainSize : 0;
+
+        matchIndex = dms->hashTable[ZSTD_hashPtr(ip, dms->cParams.hashLog, mls)];
+
+        for ( ; (matchIndex>dmsLowestIndex) & (nbAttempts>0) ; nbAttempts--) {
+            size_t currentMl=0;
+            const BYTE* const match = dmsBase + matchIndex;
+            assert(match+4 <= dmsEnd);
+            if (MEM_read32(match) == MEM_read32(ip))   /* assumption : matchIndex <= dictLimit-4 (by table construction) */
+                currentMl = ZSTD_count_2segments(ip+4, match+4, iLimit, dmsEnd, prefixStart) + 4;
+
+            /* save best solution */
+            if (currentMl > ml) {
+                ml = currentMl;
+                *offsetPtr = current - (matchIndex + dmsIndexDelta) + ZSTD_REP_MOVE;
+                if (ip+currentMl == iLimit) break; /* best possible, avoids read overflow on next attempt */
+            }
+
+            if (matchIndex <= dmsMinChain) break;
+            matchIndex = dmsChainTable[matchIndex & dmsChainMask];
+        }
+    }
+
     return ml;
 }
 
 
 FORCE_INLINE_TEMPLATE size_t ZSTD_HcFindBestMatch_selectMLS (
-                        ZSTD_CCtx* zc,
+                        ZSTD_matchState_t* ms,
                         const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr,
-                        const U32 maxNbAttempts, const U32 matchLengthSearch)
+                        size_t* offsetPtr)
 {
-    switch(matchLengthSearch)
+    switch(ms->cParams.minMatch)
     {
     default : /* includes case 3 */
-    case 4 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 4, 0);
-    case 5 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 5, 0);
+    case 4 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, ZSTD_noDict);
+    case 5 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, ZSTD_noDict);
     case 7 :
-    case 6 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 6, 0);
+    case 6 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, ZSTD_noDict);
+    }
+}
+
+
+static size_t ZSTD_HcFindBestMatch_dictMatchState_selectMLS (
+                        ZSTD_matchState_t* ms,
+                        const BYTE* ip, const BYTE* const iLimit,
+                        size_t* offsetPtr)
+{
+    switch(ms->cParams.minMatch)
+    {
+    default : /* includes case 3 */
+    case 4 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, ZSTD_dictMatchState);
+    case 5 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, ZSTD_dictMatchState);
+    case 7 :
+    case 6 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, ZSTD_dictMatchState);
     }
 }
 
 
 FORCE_INLINE_TEMPLATE size_t ZSTD_HcFindBestMatch_extDict_selectMLS (
-                        ZSTD_CCtx* zc,
+                        ZSTD_matchState_t* ms,
                         const BYTE* ip, const BYTE* const iLimit,
-                        size_t* offsetPtr,
-                        const U32 maxNbAttempts, const U32 matchLengthSearch)
+                        size_t* offsetPtr)
 {
-    switch(matchLengthSearch)
+    switch(ms->cParams.minMatch)
     {
     default : /* includes case 3 */
-    case 4 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 4, 1);
-    case 5 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 5, 1);
+    case 4 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 4, ZSTD_extDict);
+    case 5 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 5, ZSTD_extDict);
     case 7 :
-    case 6 : return ZSTD_HcFindBestMatch_generic(zc, ip, iLimit, offsetPtr, maxNbAttempts, 6, 1);
+    case 6 : return ZSTD_HcFindBestMatch_generic(ms, ip, iLimit, offsetPtr, 6, ZSTD_extDict);
     }
 }
 
@@ -399,58 +619,108 @@ FORCE_INLINE_TEMPLATE size_t ZSTD_HcFindBestMatch_extDict_selectMLS (
 /* *******************************
 *  Common parser - lazy strategy
 *********************************/
-FORCE_INLINE_TEMPLATE
-size_t ZSTD_compressBlock_lazy_generic(ZSTD_CCtx* ctx,
-                                       const void* src, size_t srcSize,
-                                       const U32 searchMethod, const U32 depth)
+typedef enum { search_hashChain, search_binaryTree } searchMethod_e;
+
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_compressBlock_lazy_generic(
+                        ZSTD_matchState_t* ms, seqStore_t* seqStore,
+                        U32 rep[ZSTD_REP_NUM],
+                        const void* src, size_t srcSize,
+                        const searchMethod_e searchMethod, const U32 depth,
+                        ZSTD_dictMode_e const dictMode)
 {
-    seqStore_t* seqStorePtr = &(ctx->seqStore);
     const BYTE* const istart = (const BYTE*)src;
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - 8;
-    const BYTE* const base = ctx->base + ctx->dictLimit;
-
-    U32 const maxSearches = 1 << ctx->appliedParams.cParams.searchLog;
-    U32 const mls = ctx->appliedParams.cParams.searchLength;
-
-    typedef size_t (*searchMax_f)(ZSTD_CCtx* zc, const BYTE* ip, const BYTE* iLimit,
-                        size_t* offsetPtr,
-                        U32 maxNbAttempts, U32 matchLengthSearch);
-    searchMax_f const searchMax = searchMethod ? ZSTD_BtFindBestMatch_selectMLS : ZSTD_HcFindBestMatch_selectMLS;
-    U32 offset_1 = seqStorePtr->rep[0], offset_2 = seqStorePtr->rep[1], savedOffset=0;
+    const BYTE* const base = ms->window.base;
+    const U32 prefixLowestIndex = ms->window.dictLimit;
+    const BYTE* const prefixLowest = base + prefixLowestIndex;
+
+    typedef size_t (*searchMax_f)(
+                        ZSTD_matchState_t* ms,
+                        const BYTE* ip, const BYTE* iLimit, size_t* offsetPtr);
+    searchMax_f const searchMax = dictMode == ZSTD_dictMatchState ?
+        (searchMethod==search_binaryTree ? ZSTD_BtFindBestMatch_dictMatchState_selectMLS
+                                         : ZSTD_HcFindBestMatch_dictMatchState_selectMLS) :
+        (searchMethod==search_binaryTree ? ZSTD_BtFindBestMatch_selectMLS
+                                         : ZSTD_HcFindBestMatch_selectMLS);
+    U32 offset_1 = rep[0], offset_2 = rep[1], savedOffset=0;
+
+    const ZSTD_matchState_t* const dms = ms->dictMatchState;
+    const U32 dictLowestIndex      = dictMode == ZSTD_dictMatchState ?
+                                     dms->window.dictLimit : 0;
+    const BYTE* const dictBase     = dictMode == ZSTD_dictMatchState ?
+                                     dms->window.base : NULL;
+    const BYTE* const dictLowest   = dictMode == ZSTD_dictMatchState ?
+                                     dictBase + dictLowestIndex : NULL;
+    const BYTE* const dictEnd      = dictMode == ZSTD_dictMatchState ?
+                                     dms->window.nextSrc : NULL;
+    const U32 dictIndexDelta       = dictMode == ZSTD_dictMatchState ?
+                                     prefixLowestIndex - (U32)(dictEnd - dictBase) :
+                                     0;
+    const U32 dictAndPrefixLength = (U32)((ip - prefixLowest) + (dictEnd - dictLowest));
+
+    DEBUGLOG(5, "ZSTD_compressBlock_lazy_generic (dictMode=%u)", (U32)dictMode);
 
     /* init */
-    ip += (ip==base);
-    ctx->nextToUpdate3 = ctx->nextToUpdate;
-    {   U32 const maxRep = (U32)(ip-base);
+    ip += (dictAndPrefixLength == 0);
+    if (dictMode == ZSTD_noDict) {
+        U32 const current = (U32)(ip - base);
+        U32 const windowLow = ZSTD_getLowestPrefixIndex(ms, current, ms->cParams.windowLog);
+        U32 const maxRep = current - windowLow;
         if (offset_2 > maxRep) savedOffset = offset_2, offset_2 = 0;
         if (offset_1 > maxRep) savedOffset = offset_1, offset_1 = 0;
     }
+    if (dictMode == ZSTD_dictMatchState) {
+        /* dictMatchState repCode checks don't currently handle repCode == 0
+         * disabling. */
+        assert(offset_1 <= dictAndPrefixLength);
+        assert(offset_2 <= dictAndPrefixLength);
+    }
 
     /* Match Loop */
+#if defined(__GNUC__) && defined(__x86_64__)
+    /* I've measured random a 5% speed loss on levels 5 & 6 (greedy) when the
+     * code alignment is perturbed. To fix the instability align the loop on 32-bytes.
+     */
+    __asm__(".p2align 5");
+#endif
     while (ip < ilimit) {
         size_t matchLength=0;
         size_t offset=0;
         const BYTE* start=ip+1;
 
         /* check repCode */
-        if ((offset_1>0) & (MEM_read32(ip+1) == MEM_read32(ip+1 - offset_1))) {
-            /* repcode : we take it */
+        if (dictMode == ZSTD_dictMatchState) {
+            const U32 repIndex = (U32)(ip - base) + 1 - offset_1;
+            const BYTE* repMatch = (dictMode == ZSTD_dictMatchState
+                                && repIndex < prefixLowestIndex) ?
+                                   dictBase + (repIndex - dictIndexDelta) :
+                                   base + repIndex;
+            if (((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
+                && (MEM_read32(repMatch) == MEM_read32(ip+1)) ) {
+                const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
+                matchLength = ZSTD_count_2segments(ip+1+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
+                if (depth==0) goto _storeSequence;
+            }
+        }
+        if ( dictMode == ZSTD_noDict
+          && ((offset_1 > 0) & (MEM_read32(ip+1-offset_1) == MEM_read32(ip+1)))) {
             matchLength = ZSTD_count(ip+1+4, ip+1+4-offset_1, iend) + 4;
             if (depth==0) goto _storeSequence;
         }
 
         /* first search (depth 0) */
-        {   size_t offsetFound = 99999999;
-            size_t const ml2 = searchMax(ctx, ip, iend, &offsetFound, maxSearches, mls);
+        {   size_t offsetFound = 999999999;
+            size_t const ml2 = searchMax(ms, ip, iend, &offsetFound);
             if (ml2 > matchLength)
                 matchLength = ml2, start = ip, offset=offsetFound;
         }
 
         if (matchLength < 4) {
-            ip += ((ip-anchor) >> g_searchStrength) + 1;   /* jump faster over incompressible sections */
+            ip += ((ip-anchor) >> kSearchStrength) + 1;   /* jump faster over incompressible sections */
             continue;
         }
 
@@ -458,15 +728,31 @@ size_t ZSTD_compressBlock_lazy_generic(ZSTD_CCtx* ctx,
         if (depth>=1)
         while (ip<ilimit) {
             ip ++;
-            if ((offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+            if ( (dictMode == ZSTD_noDict)
+              && (offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
                 size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
                 int const gain2 = (int)(mlRep * 3);
                 int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offset+1) + 1);
                 if ((mlRep >= 4) && (gain2 > gain1))
                     matchLength = mlRep, offset = 0, start = ip;
             }
-            {   size_t offset2=99999999;
-                size_t const ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls);
+            if (dictMode == ZSTD_dictMatchState) {
+                const U32 repIndex = (U32)(ip - base) - offset_1;
+                const BYTE* repMatch = repIndex < prefixLowestIndex ?
+                               dictBase + (repIndex - dictIndexDelta) :
+                               base + repIndex;
+                if (((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
+                    && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
+                    const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
+                    size_t const mlRep = ZSTD_count_2segments(ip+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
+                    int const gain2 = (int)(mlRep * 3);
+                    int const gain1 = (int)(matchLength*3 - ZSTD_highbit32((U32)offset+1) + 1);
+                    if ((mlRep >= 4) && (gain2 > gain1))
+                        matchLength = mlRep, offset = 0, start = ip;
+                }
+            }
+            {   size_t offset2=999999999;
+                size_t const ml2 = searchMax(ms, ip, iend, &offset2);
                 int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
                 int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 4);
                 if ((ml2 >= 4) && (gain2 > gain1)) {
@@ -477,15 +763,31 @@ size_t ZSTD_compressBlock_lazy_generic(ZSTD_CCtx* ctx,
             /* let's find an even better one */
             if ((depth==2) && (ip<ilimit)) {
                 ip ++;
-                if ((offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
-                    size_t const ml2 = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
-                    int const gain2 = (int)(ml2 * 4);
+                if ( (dictMode == ZSTD_noDict)
+                  && (offset) && ((offset_1>0) & (MEM_read32(ip) == MEM_read32(ip - offset_1)))) {
+                    size_t const mlRep = ZSTD_count(ip+4, ip+4-offset_1, iend) + 4;
+                    int const gain2 = (int)(mlRep * 4);
                     int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 1);
-                    if ((ml2 >= 4) && (gain2 > gain1))
-                        matchLength = ml2, offset = 0, start = ip;
+                    if ((mlRep >= 4) && (gain2 > gain1))
+                        matchLength = mlRep, offset = 0, start = ip;
+                }
+                if (dictMode == ZSTD_dictMatchState) {
+                    const U32 repIndex = (U32)(ip - base) - offset_1;
+                    const BYTE* repMatch = repIndex < prefixLowestIndex ?
+                                   dictBase + (repIndex - dictIndexDelta) :
+                                   base + repIndex;
+                    if (((U32)((prefixLowestIndex-1) - repIndex) >= 3 /* intentional underflow */)
+                        && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
+                        const BYTE* repMatchEnd = repIndex < prefixLowestIndex ? dictEnd : iend;
+                        size_t const mlRep = ZSTD_count_2segments(ip+4, repMatch+4, iend, repMatchEnd, prefixLowest) + 4;
+                        int const gain2 = (int)(mlRep * 4);
+                        int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 1);
+                        if ((mlRep >= 4) && (gain2 > gain1))
+                            matchLength = mlRep, offset = 0, start = ip;
+                    }
                 }
-                {   size_t offset2=99999999;
-                    size_t const ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls);
+                {   size_t offset2=999999999;
+                    size_t const ml2 = searchMax(ms, ip, iend, &offset2);
                     int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
                     int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 7);
                     if ((ml2 >= 4) && (gain2 > gain1)) {
@@ -502,96 +804,166 @@ size_t ZSTD_compressBlock_lazy_generic(ZSTD_CCtx* ctx,
          */
         /* catch up */
         if (offset) {
-            while ( (start > anchor)
-                 && (start > base+offset-ZSTD_REP_MOVE)
-                 && (start[-1] == (start-offset+ZSTD_REP_MOVE)[-1]) )  /* only search for offset within prefix */
-                { start--; matchLength++; }
+            if (dictMode == ZSTD_noDict) {
+                while ( ((start > anchor) & (start - (offset-ZSTD_REP_MOVE) > prefixLowest))
+                     && (start[-1] == (start-(offset-ZSTD_REP_MOVE))[-1]) )  /* only search for offset within prefix */
+                    { start--; matchLength++; }
+            }
+            if (dictMode == ZSTD_dictMatchState) {
+                U32 const matchIndex = (U32)((start-base) - (offset - ZSTD_REP_MOVE));
+                const BYTE* match = (matchIndex < prefixLowestIndex) ? dictBase + matchIndex - dictIndexDelta : base + matchIndex;
+                const BYTE* const mStart = (matchIndex < prefixLowestIndex) ? dictLowest : prefixLowest;
+                while ((start>anchor) && (match>mStart) && (start[-1] == match[-1])) { start--; match--; matchLength++; }  /* catch up */
+            }
             offset_2 = offset_1; offset_1 = (U32)(offset - ZSTD_REP_MOVE);
         }
         /* store sequence */
 _storeSequence:
         {   size_t const litLength = start - anchor;
-            ZSTD_storeSeq(seqStorePtr, litLength, anchor, (U32)offset, matchLength-MINMATCH);
+            ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offset, matchLength-MINMATCH);
             anchor = ip = start + matchLength;
         }
 
         /* check immediate repcode */
-        while ( (ip <= ilimit)
-             && ((offset_2>0)
-             & (MEM_read32(ip) == MEM_read32(ip - offset_2)) )) {
-            /* store sequence */
-            matchLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
-            offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset; /* swap repcodes */
-            ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, matchLength-MINMATCH);
-            ip += matchLength;
-            anchor = ip;
-            continue;   /* faster when present ... (?) */
-    }   }
+        if (dictMode == ZSTD_dictMatchState) {
+            while (ip <= ilimit) {
+                U32 const current2 = (U32)(ip-base);
+                U32 const repIndex = current2 - offset_2;
+                const BYTE* repMatch = dictMode == ZSTD_dictMatchState
+                    && repIndex < prefixLowestIndex ?
+                        dictBase - dictIndexDelta + repIndex :
+                        base + repIndex;
+                if ( ((U32)((prefixLowestIndex-1) - (U32)repIndex) >= 3 /* intentional overflow */)
+                   && (MEM_read32(repMatch) == MEM_read32(ip)) ) {
+                    const BYTE* const repEnd2 = repIndex < prefixLowestIndex ? dictEnd : iend;
+                    matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd2, prefixLowest) + 4;
+                    offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset;   /* swap offset_2 <=> offset_1 */
+                    ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, matchLength-MINMATCH);
+                    ip += matchLength;
+                    anchor = ip;
+                    continue;
+                }
+                break;
+            }
+        }
+
+        if (dictMode == ZSTD_noDict) {
+            while ( ((ip <= ilimit) & (offset_2>0))
+                 && (MEM_read32(ip) == MEM_read32(ip - offset_2)) ) {
+                /* store sequence */
+                matchLength = ZSTD_count(ip+4, ip+4-offset_2, iend) + 4;
+                offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset; /* swap repcodes */
+                ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, matchLength-MINMATCH);
+                ip += matchLength;
+                anchor = ip;
+                continue;   /* faster when present ... (?) */
+    }   }   }
 
     /* Save reps for next block */
-    seqStorePtr->repToConfirm[0] = offset_1 ? offset_1 : savedOffset;
-    seqStorePtr->repToConfirm[1] = offset_2 ? offset_2 : savedOffset;
+    rep[0] = offset_1 ? offset_1 : savedOffset;
+    rep[1] = offset_2 ? offset_2 : savedOffset;
 
     /* Return the last literals size */
-    return iend - anchor;
+    return (size_t)(iend - anchor);
+}
+
+
+size_t ZSTD_compressBlock_btlazy2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_lazy2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_noDict);
 }
 
+size_t ZSTD_compressBlock_lazy(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_noDict);
+}
+
+size_t ZSTD_compressBlock_greedy(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_noDict);
+}
 
-size_t ZSTD_compressBlock_btlazy2(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+size_t ZSTD_compressBlock_btlazy2_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ctx, src, srcSize, 1, 2);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2, ZSTD_dictMatchState);
 }
 
-size_t ZSTD_compressBlock_lazy2(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+size_t ZSTD_compressBlock_lazy2_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ctx, src, srcSize, 0, 2);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2, ZSTD_dictMatchState);
 }
 
-size_t ZSTD_compressBlock_lazy(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+size_t ZSTD_compressBlock_lazy_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ctx, src, srcSize, 0, 1);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1, ZSTD_dictMatchState);
 }
 
-size_t ZSTD_compressBlock_greedy(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+size_t ZSTD_compressBlock_greedy_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_generic(ctx, src, srcSize, 0, 0);
+    return ZSTD_compressBlock_lazy_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0, ZSTD_dictMatchState);
 }
 
 
 FORCE_INLINE_TEMPLATE
-size_t ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
-                                     const void* src, size_t srcSize,
-                                     const U32 searchMethod, const U32 depth)
+size_t ZSTD_compressBlock_lazy_extDict_generic(
+                        ZSTD_matchState_t* ms, seqStore_t* seqStore,
+                        U32 rep[ZSTD_REP_NUM],
+                        const void* src, size_t srcSize,
+                        const searchMethod_e searchMethod, const U32 depth)
 {
-    seqStore_t* seqStorePtr = &(ctx->seqStore);
     const BYTE* const istart = (const BYTE*)src;
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - 8;
-    const BYTE* const base = ctx->base;
-    const U32 dictLimit = ctx->dictLimit;
-    const U32 lowestIndex = ctx->lowLimit;
+    const BYTE* const base = ms->window.base;
+    const U32 dictLimit = ms->window.dictLimit;
     const BYTE* const prefixStart = base + dictLimit;
-    const BYTE* const dictBase = ctx->dictBase;
+    const BYTE* const dictBase = ms->window.dictBase;
     const BYTE* const dictEnd  = dictBase + dictLimit;
-    const BYTE* const dictStart  = dictBase + ctx->lowLimit;
+    const BYTE* const dictStart  = dictBase + ms->window.lowLimit;
+    const U32 windowLog = ms->cParams.windowLog;
 
-    const U32 maxSearches = 1 << ctx->appliedParams.cParams.searchLog;
-    const U32 mls = ctx->appliedParams.cParams.searchLength;
+    typedef size_t (*searchMax_f)(
+                        ZSTD_matchState_t* ms,
+                        const BYTE* ip, const BYTE* iLimit, size_t* offsetPtr);
+    searchMax_f searchMax = searchMethod==search_binaryTree ? ZSTD_BtFindBestMatch_extDict_selectMLS : ZSTD_HcFindBestMatch_extDict_selectMLS;
 
-    typedef size_t (*searchMax_f)(ZSTD_CCtx* zc, const BYTE* ip, const BYTE* iLimit,
-                        size_t* offsetPtr,
-                        U32 maxNbAttempts, U32 matchLengthSearch);
-    searchMax_f searchMax = searchMethod ? ZSTD_BtFindBestMatch_selectMLS_extDict : ZSTD_HcFindBestMatch_extDict_selectMLS;
+    U32 offset_1 = rep[0], offset_2 = rep[1];
 
-    U32 offset_1 = seqStorePtr->rep[0], offset_2 = seqStorePtr->rep[1];
+    DEBUGLOG(5, "ZSTD_compressBlock_lazy_extDict_generic");
 
     /* init */
-    ctx->nextToUpdate3 = ctx->nextToUpdate;
     ip += (ip == prefixStart);
 
     /* Match Loop */
+#if defined(__GNUC__) && defined(__x86_64__)
+    /* I've measured random a 5% speed loss on levels 5 & 6 (greedy) when the
+     * code alignment is perturbed. To fix the instability align the loop on 32-bytes.
+     */
+    __asm__(".p2align 5");
+#endif
     while (ip < ilimit) {
         size_t matchLength=0;
         size_t offset=0;
@@ -599,10 +971,11 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
         U32 current = (U32)(ip-base);
 
         /* check repCode */
-        {   const U32 repIndex = (U32)(current+1 - offset_1);
+        {   const U32 windowLow = ZSTD_getLowestMatchIndex(ms, current+1, windowLog);
+            const U32 repIndex = (U32)(current+1 - offset_1);
             const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
             const BYTE* const repMatch = repBase + repIndex;
-            if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))   /* intentional overflow */
+            if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > windowLow))   /* intentional overflow */
             if (MEM_read32(ip+1) == MEM_read32(repMatch)) {
                 /* repcode detected we should take it */
                 const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
@@ -611,14 +984,14 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
         }   }
 
         /* first search (depth 0) */
-        {   size_t offsetFound = 99999999;
-            size_t const ml2 = searchMax(ctx, ip, iend, &offsetFound, maxSearches, mls);
+        {   size_t offsetFound = 999999999;
+            size_t const ml2 = searchMax(ms, ip, iend, &offsetFound);
             if (ml2 > matchLength)
                 matchLength = ml2, start = ip, offset=offsetFound;
         }
 
          if (matchLength < 4) {
-            ip += ((ip-anchor) >> g_searchStrength) + 1;   /* jump faster over incompressible sections */
+            ip += ((ip-anchor) >> kSearchStrength) + 1;   /* jump faster over incompressible sections */
             continue;
         }
 
@@ -629,10 +1002,11 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
             current++;
             /* check repCode */
             if (offset) {
+                const U32 windowLow = ZSTD_getLowestMatchIndex(ms, current, windowLog);
                 const U32 repIndex = (U32)(current - offset_1);
                 const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
                 const BYTE* const repMatch = repBase + repIndex;
-                if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
+                if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > windowLow))  /* intentional overflow */
                 if (MEM_read32(ip) == MEM_read32(repMatch)) {
                     /* repcode detected */
                     const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
@@ -644,8 +1018,8 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
             }   }
 
             /* search match, depth 1 */
-            {   size_t offset2=99999999;
-                size_t const ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls);
+            {   size_t offset2=999999999;
+                size_t const ml2 = searchMax(ms, ip, iend, &offset2);
                 int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
                 int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 4);
                 if ((ml2 >= 4) && (gain2 > gain1)) {
@@ -659,10 +1033,11 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
                 current++;
                 /* check repCode */
                 if (offset) {
+                    const U32 windowLow = ZSTD_getLowestMatchIndex(ms, current, windowLog);
                     const U32 repIndex = (U32)(current - offset_1);
                     const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
                     const BYTE* const repMatch = repBase + repIndex;
-                    if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
+                    if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > windowLow))  /* intentional overflow */
                     if (MEM_read32(ip) == MEM_read32(repMatch)) {
                         /* repcode detected */
                         const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
@@ -674,8 +1049,8 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
                 }   }
 
                 /* search match, depth 2 */
-                {   size_t offset2=99999999;
-                    size_t const ml2 = searchMax(ctx, ip, iend, &offset2, maxSearches, mls);
+                {   size_t offset2=999999999;
+                    size_t const ml2 = searchMax(ms, ip, iend, &offset2);
                     int const gain2 = (int)(ml2*4 - ZSTD_highbit32((U32)offset2+1));   /* raw approx */
                     int const gain1 = (int)(matchLength*4 - ZSTD_highbit32((U32)offset+1) + 7);
                     if ((ml2 >= 4) && (gain2 > gain1)) {
@@ -697,22 +1072,24 @@ size_t ZSTD_compressBlock_lazy_extDict_generic(ZSTD_CCtx* ctx,
         /* store sequence */
 _storeSequence:
         {   size_t const litLength = start - anchor;
-            ZSTD_storeSeq(seqStorePtr, litLength, anchor, (U32)offset, matchLength-MINMATCH);
+            ZSTD_storeSeq(seqStore, litLength, anchor, iend, (U32)offset, matchLength-MINMATCH);
             anchor = ip = start + matchLength;
         }
 
         /* check immediate repcode */
         while (ip <= ilimit) {
-            const U32 repIndex = (U32)((ip-base) - offset_2);
+            const U32 repCurrent = (U32)(ip-base);
+            const U32 windowLow = ZSTD_getLowestMatchIndex(ms, repCurrent, windowLog);
+            const U32 repIndex = repCurrent - offset_2;
             const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
             const BYTE* const repMatch = repBase + repIndex;
-            if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > lowestIndex))  /* intentional overflow */
+            if (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex > windowLow))  /* intentional overflow */
             if (MEM_read32(ip) == MEM_read32(repMatch)) {
                 /* repcode detected we should take it */
                 const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
                 matchLength = ZSTD_count_2segments(ip+4, repMatch+4, iend, repEnd, prefixStart) + 4;
                 offset = offset_2; offset_2 = offset_1; offset_1 = (U32)offset;   /* swap offset history */
-                ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, matchLength-MINMATCH);
+                ZSTD_storeSeq(seqStore, 0, anchor, iend, 0, matchLength-MINMATCH);
                 ip += matchLength;
                 anchor = ip;
                 continue;   /* faster when present ... (?) */
@@ -721,29 +1098,41 @@ _storeSequence:
     }   }
 
     /* Save reps for next block */
-    seqStorePtr->repToConfirm[0] = offset_1; seqStorePtr->repToConfirm[1] = offset_2;
+    rep[0] = offset_1;
+    rep[1] = offset_2;
 
     /* Return the last literals size */
-    return iend - anchor;
+    return (size_t)(iend - anchor);
 }
 
 
-size_t ZSTD_compressBlock_greedy_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+size_t ZSTD_compressBlock_greedy_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_lazy_extDict_generic(ctx, src, srcSize, 0, 0);
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 0);
 }
 
-size_t ZSTD_compressBlock_lazy_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+size_t ZSTD_compressBlock_lazy_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+
 {
-    return ZSTD_compressBlock_lazy_extDict_generic(ctx, src, srcSize, 0, 1);
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 1);
 }
 
-size_t ZSTD_compressBlock_lazy2_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+size_t ZSTD_compressBlock_lazy2_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+
 {
-    return ZSTD_compressBlock_lazy_extDict_generic(ctx, src, srcSize, 0, 2);
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_hashChain, 2);
 }
 
-size_t ZSTD_compressBlock_btlazy2_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+size_t ZSTD_compressBlock_btlazy2_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize)
+
 {
-    return ZSTD_compressBlock_lazy_extDict_generic(ctx, src, srcSize, 1, 2);
+    return ZSTD_compressBlock_lazy_extDict_generic(ms, seqStore, rep, src, srcSize, search_binaryTree, 2);
 }
diff --git a/contrib/zstd/zstd_lazy.h b/contrib/zstd/zstd_lazy.h
index a9c4daed2..581936f03 100644
--- a/contrib/zstd/zstd_lazy.h
+++ b/contrib/zstd/zstd_lazy.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,25 +11,54 @@
 #ifndef ZSTD_LAZY_H
 #define ZSTD_LAZY_H
 
-#include "zstd_compress.h"
-
 #if defined (__cplusplus)
 extern "C" {
 #endif
 
-U32 ZSTD_insertAndFindFirstIndex (ZSTD_CCtx* zc, const BYTE* ip, U32 mls);
-void ZSTD_updateTree(ZSTD_CCtx* zc, const BYTE* const ip, const BYTE* const iend, const U32 nbCompares, const U32 mls);
-void ZSTD_updateTree_extDict(ZSTD_CCtx* zc, const BYTE* const ip, const BYTE* const iend, const U32 nbCompares, const U32 mls);
+#include "zstd_compress_internal.h"
+
+U32 ZSTD_insertAndFindFirstIndex(ZSTD_matchState_t* ms, const BYTE* ip);
+
+void ZSTD_preserveUnsortedMark (U32* const table, U32 const size, U32 const reducerValue);  /*! used in ZSTD_reduceIndex(). preemptively increase value of ZSTD_DUBT_UNSORTED_MARK */
+
+size_t ZSTD_compressBlock_btlazy2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_greedy(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
 
-size_t ZSTD_compressBlock_btlazy2(ZSTD_CCtx* ctx, const void* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy2(ZSTD_CCtx* ctx, const void* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy(ZSTD_CCtx* ctx, const void* src, size_t srcSize);
-size_t ZSTD_compressBlock_greedy(ZSTD_CCtx* ctx, const void* src, size_t srcSize);
+size_t ZSTD_compressBlock_btlazy2_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy2_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_greedy_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
 
-size_t ZSTD_compressBlock_greedy_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize);
-size_t ZSTD_compressBlock_lazy2_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize);
-size_t ZSTD_compressBlock_btlazy2_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize);
+size_t ZSTD_compressBlock_greedy_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_lazy2_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btlazy2_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
 
 #if defined (__cplusplus)
 }
diff --git a/contrib/zstd/zstd_ldm.c b/contrib/zstd/zstd_ldm.c
index e40007c19..3916da494 100644
--- a/contrib/zstd/zstd_ldm.c
+++ b/contrib/zstd/zstd_ldm.c
@@ -1,48 +1,64 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
  * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 #include "zstd_ldm.h"
 
+#include "debug.h"
 #include "zstd_fast.h"          /* ZSTD_fillHashTable() */
 #include "zstd_double_fast.h"   /* ZSTD_fillDoubleHashTable() */
 
 #define LDM_BUCKET_SIZE_LOG 3
 #define LDM_MIN_MATCH_LENGTH 64
-#define LDM_HASH_LOG 20
+#define LDM_HASH_RLOG 7
 #define LDM_HASH_CHAR_OFFSET 10
 
-size_t ZSTD_ldm_initializeParameters(ldmParams_t* params, U32 enableLdm)
+void ZSTD_ldm_adjustParameters(ldmParams_t* params,
+                               ZSTD_compressionParameters const* cParams)
 {
+    params->windowLog = cParams->windowLog;
     ZSTD_STATIC_ASSERT(LDM_BUCKET_SIZE_LOG <= ZSTD_LDM_BUCKETSIZELOG_MAX);
-    params->enableLdm = enableLdm>0;
-    params->hashLog = LDM_HASH_LOG;
-    params->bucketSizeLog = LDM_BUCKET_SIZE_LOG;
-    params->minMatchLength = LDM_MIN_MATCH_LENGTH;
-    params->hashEveryLog = ZSTD_LDM_HASHEVERYLOG_NOTSET;
-    return 0;
+    DEBUGLOG(4, "ZSTD_ldm_adjustParameters");
+    if (!params->bucketSizeLog) params->bucketSizeLog = LDM_BUCKET_SIZE_LOG;
+    if (!params->minMatchLength) params->minMatchLength = LDM_MIN_MATCH_LENGTH;
+    if (cParams->strategy >= ZSTD_btopt) {
+      /* Get out of the way of the optimal parser */
+      U32 const minMatch = MAX(cParams->targetLength, params->minMatchLength);
+      assert(minMatch >= ZSTD_LDM_MINMATCH_MIN);
+      assert(minMatch <= ZSTD_LDM_MINMATCH_MAX);
+      params->minMatchLength = minMatch;
+    }
+    if (params->hashLog == 0) {
+        params->hashLog = MAX(ZSTD_HASHLOG_MIN, params->windowLog - LDM_HASH_RLOG);
+        assert(params->hashLog <= ZSTD_HASHLOG_MAX);
+    }
+    if (params->hashRateLog == 0) {
+        params->hashRateLog = params->windowLog < params->hashLog
+                                   ? 0
+                                   : params->windowLog - params->hashLog;
+    }
+    params->bucketSizeLog = MIN(params->bucketSizeLog, params->hashLog);
 }
 
-void ZSTD_ldm_adjustParameters(ldmParams_t* params, U32 windowLog)
+size_t ZSTD_ldm_getTableSize(ldmParams_t params)
 {
-    if (params->hashEveryLog == ZSTD_LDM_HASHEVERYLOG_NOTSET) {
-        params->hashEveryLog =
-                windowLog < params->hashLog ? 0 : windowLog - params->hashLog;
-    }
-    params->bucketSizeLog = MIN(params->bucketSizeLog, params->hashLog);
+    size_t const ldmHSize = ((size_t)1) << params.hashLog;
+    size_t const ldmBucketSizeLog = MIN(params.bucketSizeLog, params.hashLog);
+    size_t const ldmBucketSize = ((size_t)1) << (params.hashLog - ldmBucketSizeLog);
+    size_t const totalSize = ZSTD_cwksp_alloc_size(ldmBucketSize)
+                           + ZSTD_cwksp_alloc_size(ldmHSize * sizeof(ldmEntry_t));
+    return params.enableLdm ? totalSize : 0;
 }
 
-size_t ZSTD_ldm_getTableSize(U32 hashLog, U32 bucketSizeLog) {
-    size_t const ldmHSize = ((size_t)1) << hashLog;
-    size_t const ldmBucketSizeLog = MIN(bucketSizeLog, hashLog);
-    size_t const ldmBucketSize =
-        ((size_t)1) << (hashLog - ldmBucketSizeLog);
-    return ldmBucketSize + (ldmHSize * (sizeof(ldmEntry_t)));
+size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize)
+{
+    return params.enableLdm ? (maxChunkSize / params.minMatchLength) : 0;
 }
 
 /** ZSTD_ldm_getSmallHash() :
@@ -104,20 +120,20 @@ static void ZSTD_ldm_insertEntry(ldmState_t* ldmState,
  *
  *  Gets the small hash, checksum, and tag from the rollingHash.
  *
- *  If the tag matches (1 << ldmParams.hashEveryLog)-1, then
+ *  If the tag matches (1 << ldmParams.hashRateLog)-1, then
  *  creates an ldmEntry from the offset, and inserts it into the hash table.
  *
  *  hBits is the length of the small hash, which is the most significant hBits
  *  of rollingHash. The checksum is the next 32 most significant bits, followed
- *  by ldmParams.hashEveryLog bits that make up the tag. */
+ *  by ldmParams.hashRateLog bits that make up the tag. */
 static void ZSTD_ldm_makeEntryAndInsertByTag(ldmState_t* ldmState,
                                              U64 const rollingHash,
                                              U32 const hBits,
                                              U32 const offset,
                                              ldmParams_t const ldmParams)
 {
-    U32 const tag = ZSTD_ldm_getTag(rollingHash, hBits, ldmParams.hashEveryLog);
-    U32 const tagMask = ((U32)1 << ldmParams.hashEveryLog) - 1;
+    U32 const tag = ZSTD_ldm_getTag(rollingHash, hBits, ldmParams.hashRateLog);
+    U32 const tagMask = ((U32)1 << ldmParams.hashRateLog) - 1;
     if (tag == tagMask) {
         U32 const hash = ZSTD_ldm_getSmallHash(rollingHash, hBits);
         U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
@@ -128,55 +144,6 @@ static void ZSTD_ldm_makeEntryAndInsertByTag(ldmState_t* ldmState,
     }
 }
 
-/** ZSTD_ldm_getRollingHash() :
- *  Get a 64-bit hash using the first len bytes from buf.
- *
- *  Giving bytes s = s_1, s_2, ... s_k, the hash is defined to be
- *  H(s) = s_1*(a^(k-1)) + s_2*(a^(k-2)) + ... + s_k*(a^0)
- *
- *  where the constant a is defined to be prime8bytes.
- *
- *  The implementation adds an offset to each byte, so
- *  H(s) = (s_1 + HASH_CHAR_OFFSET)*(a^(k-1)) + ... */
-static U64 ZSTD_ldm_getRollingHash(const BYTE* buf, U32 len)
-{
-    U64 ret = 0;
-    U32 i;
-    for (i = 0; i < len; i++) {
-        ret *= prime8bytes;
-        ret += buf[i] + LDM_HASH_CHAR_OFFSET;
-    }
-    return ret;
-}
-
-/** ZSTD_ldm_ipow() :
- *  Return base^exp. */
-static U64 ZSTD_ldm_ipow(U64 base, U64 exp)
-{
-    U64 ret = 1;
-    while (exp) {
-        if (exp & 1) { ret *= base; }
-        exp >>= 1;
-        base *= base;
-    }
-    return ret;
-}
-
-U64 ZSTD_ldm_getHashPower(U32 minMatchLength) {
-    assert(minMatchLength >= ZSTD_LDM_MINMATCH_MIN);
-    return ZSTD_ldm_ipow(prime8bytes, minMatchLength - 1);
-}
-
-/** ZSTD_ldm_updateHash() :
- *  Updates hash by removing toRemove and adding toAdd. */
-static U64 ZSTD_ldm_updateHash(U64 hash, BYTE toRemove, BYTE toAdd, U64 hashPower)
-{
-    hash -= ((toRemove + LDM_HASH_CHAR_OFFSET) * hashPower);
-    hash *= prime8bytes;
-    hash += toAdd + LDM_HASH_CHAR_OFFSET;
-    return hash;
-}
-
 /** ZSTD_ldm_countBackwardsMatch() :
  *  Returns the number of bytes that match backwards before pIn and pMatch.
  *
@@ -201,21 +168,19 @@ static size_t ZSTD_ldm_countBackwardsMatch(
  *
  *  The tables for the other strategies are filled within their
  *  block compressors. */
-static size_t ZSTD_ldm_fillFastTables(ZSTD_CCtx* zc, const void* end)
+static size_t ZSTD_ldm_fillFastTables(ZSTD_matchState_t* ms,
+                                      void const* end)
 {
     const BYTE* const iend = (const BYTE*)end;
-    const U32 mls = zc->appliedParams.cParams.searchLength;
 
-    switch(zc->appliedParams.cParams.strategy)
+    switch(ms->cParams.strategy)
     {
     case ZSTD_fast:
-        ZSTD_fillHashTable(zc, iend, mls);
-        zc->nextToUpdate = (U32)(iend - zc->base);
+        ZSTD_fillHashTable(ms, iend, ZSTD_dtlm_fast);
         break;
 
     case ZSTD_dfast:
-        ZSTD_fillDoubleHashTable(zc, iend, mls);
-        zc->nextToUpdate = (U32)(iend - zc->base);
+        ZSTD_fillDoubleHashTable(ms, iend, ZSTD_dtlm_fast);
         break;
 
     case ZSTD_greedy:
@@ -224,6 +189,7 @@ static size_t ZSTD_ldm_fillFastTables(ZSTD_CCtx* zc, const void* end)
     case ZSTD_btlazy2:
     case ZSTD_btopt:
     case ZSTD_btultra:
+    case ZSTD_btultra2:
         break;
     default:
         assert(0);  /* not possible : not a valid strategy id */
@@ -247,9 +213,9 @@ static U64 ZSTD_ldm_fillLdmHashTable(ldmState_t* state,
     const BYTE* cur = lastHashed + 1;
 
     while (cur < iend) {
-        rollingHash = ZSTD_ldm_updateHash(rollingHash, cur[-1],
-                                          cur[ldmParams.minMatchLength-1],
-                                          state->hashPower);
+        rollingHash = ZSTD_rollingHash_rotate(rollingHash, cur[-1],
+                                              cur[ldmParams.minMatchLength-1],
+                                              state->hashPower);
         ZSTD_ldm_makeEntryAndInsertByTag(state,
                                          rollingHash, hBits,
                                          (U32)(cur - base), ldmParams);
@@ -258,75 +224,82 @@ static U64 ZSTD_ldm_fillLdmHashTable(ldmState_t* state,
     return rollingHash;
 }
 
+void ZSTD_ldm_fillHashTable(
+            ldmState_t* state, const BYTE* ip,
+            const BYTE* iend, ldmParams_t const* params)
+{
+    DEBUGLOG(5, "ZSTD_ldm_fillHashTable");
+    if ((size_t)(iend - ip) >= params->minMatchLength) {
+        U64 startingHash = ZSTD_rollingHash_compute(ip, params->minMatchLength);
+        ZSTD_ldm_fillLdmHashTable(
+            state, startingHash, ip, iend - params->minMatchLength, state->window.base,
+            params->hashLog - params->bucketSizeLog,
+            *params);
+    }
+}
+
 
 /** ZSTD_ldm_limitTableUpdate() :
  *
  *  Sets cctx->nextToUpdate to a position corresponding closer to anchor
  *  if it is far way
  *  (after a long match, only update tables a limited amount). */
-static void ZSTD_ldm_limitTableUpdate(ZSTD_CCtx* cctx, const BYTE* anchor)
+static void ZSTD_ldm_limitTableUpdate(ZSTD_matchState_t* ms, const BYTE* anchor)
 {
-    U32 const current = (U32)(anchor - cctx->base);
-    if (current > cctx->nextToUpdate + 1024) {
-        cctx->nextToUpdate =
-            current - MIN(512, current - cctx->nextToUpdate - 1024);
+    U32 const current = (U32)(anchor - ms->window.base);
+    if (current > ms->nextToUpdate + 1024) {
+        ms->nextToUpdate =
+            current - MIN(512, current - ms->nextToUpdate - 1024);
     }
 }
 
-typedef size_t (*ZSTD_blockCompressor) (ZSTD_CCtx* ctx, const void* src, size_t srcSize);
-/* defined in zstd_compress.c */
-ZSTD_blockCompressor ZSTD_selectBlockCompressor(ZSTD_strategy strat, int extDict);
-
-FORCE_INLINE_TEMPLATE
-size_t ZSTD_compressBlock_ldm_generic(ZSTD_CCtx* cctx,
-                                      const void* src, size_t srcSize)
+static size_t ZSTD_ldm_generateSequences_internal(
+        ldmState_t* ldmState, rawSeqStore_t* rawSeqStore,
+        ldmParams_t const* params, void const* src, size_t srcSize)
 {
-    ldmState_t* const ldmState = &(cctx->ldmState);
-    const ldmParams_t ldmParams = cctx->appliedParams.ldmParams;
-    const U64 hashPower = ldmState->hashPower;
-    const U32 hBits = ldmParams.hashLog - ldmParams.bucketSizeLog;
-    const U32 ldmBucketSize = ((U32)1 << ldmParams.bucketSizeLog);
-    const U32 ldmTagMask = ((U32)1 << ldmParams.hashEveryLog) - 1;
-    seqStore_t* const seqStorePtr = &(cctx->seqStore);
-    const BYTE* const base = cctx->base;
-    const BYTE* const istart = (const BYTE*)src;
-    const BYTE* ip = istart;
-    const BYTE* anchor = istart;
-    const U32   lowestIndex = cctx->dictLimit;
-    const BYTE* const lowest = base + lowestIndex;
-    const BYTE* const iend = istart + srcSize;
-    const BYTE* const ilimit = iend - MAX(ldmParams.minMatchLength, HASH_READ_SIZE);
-
-    const ZSTD_blockCompressor blockCompressor =
-        ZSTD_selectBlockCompressor(cctx->appliedParams.cParams.strategy, 0);
-    U32* const repToConfirm = seqStorePtr->repToConfirm;
-    U32 savedRep[ZSTD_REP_NUM];
+    /* LDM parameters */
+    int const extDict = ZSTD_window_hasExtDict(ldmState->window);
+    U32 const minMatchLength = params->minMatchLength;
+    U64 const hashPower = ldmState->hashPower;
+    U32 const hBits = params->hashLog - params->bucketSizeLog;
+    U32 const ldmBucketSize = 1U << params->bucketSizeLog;
+    U32 const hashRateLog = params->hashRateLog;
+    U32 const ldmTagMask = (1U << params->hashRateLog) - 1;
+    /* Prefix and extDict parameters */
+    U32 const dictLimit = ldmState->window.dictLimit;
+    U32 const lowestIndex = extDict ? ldmState->window.lowLimit : dictLimit;
+    BYTE const* const base = ldmState->window.base;
+    BYTE const* const dictBase = extDict ? ldmState->window.dictBase : NULL;
+    BYTE const* const dictStart = extDict ? dictBase + lowestIndex : NULL;
+    BYTE const* const dictEnd = extDict ? dictBase + dictLimit : NULL;
+    BYTE const* const lowPrefixPtr = base + dictLimit;
+    /* Input bounds */
+    BYTE const* const istart = (BYTE const*)src;
+    BYTE const* const iend = istart + srcSize;
+    BYTE const* const ilimit = iend - MAX(minMatchLength, HASH_READ_SIZE);
+    /* Input positions */
+    BYTE const* anchor = istart;
+    BYTE const* ip = istart;
+    /* Rolling hash */
+    BYTE const* lastHashed = NULL;
     U64 rollingHash = 0;
-    const BYTE* lastHashed = NULL;
-    size_t i, lastLiterals;
-
-    /* Save seqStorePtr->rep and copy repToConfirm */
-    for (i = 0; i < ZSTD_REP_NUM; i++)
-        savedRep[i] = repToConfirm[i] = seqStorePtr->rep[i];
 
-    /* Main Search Loop */
-    while (ip < ilimit) {   /* < instead of <=, because repcode check at (ip+1) */
+    while (ip <= ilimit) {
         size_t mLength;
         U32 const current = (U32)(ip - base);
         size_t forwardMatchLength = 0, backwardMatchLength = 0;
         ldmEntry_t* bestEntry = NULL;
         if (ip != istart) {
-            rollingHash = ZSTD_ldm_updateHash(rollingHash, lastHashed[0],
-                                              lastHashed[ldmParams.minMatchLength],
-                                              hashPower);
+            rollingHash = ZSTD_rollingHash_rotate(rollingHash, lastHashed[0],
+                                                  lastHashed[minMatchLength],
+                                                  hashPower);
         } else {
-            rollingHash = ZSTD_ldm_getRollingHash(ip, ldmParams.minMatchLength);
+            rollingHash = ZSTD_rollingHash_compute(ip, minMatchLength);
         }
         lastHashed = ip;
 
         /* Do not insert and do not look for a match */
-        if (ZSTD_ldm_getTag(rollingHash, hBits, ldmParams.hashEveryLog) !=
-                ldmTagMask) {
+        if (ZSTD_ldm_getTag(rollingHash, hBits, hashRateLog) != ldmTagMask) {
            ip++;
            continue;
         }
@@ -336,27 +309,49 @@ size_t ZSTD_compressBlock_ldm_generic(ZSTD_CCtx* cctx,
             ldmEntry_t* const bucket =
                 ZSTD_ldm_getBucket(ldmState,
                                    ZSTD_ldm_getSmallHash(rollingHash, hBits),
-                                   ldmParams);
+                                   *params);
             ldmEntry_t* cur;
             size_t bestMatchLength = 0;
             U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
 
             for (cur = bucket; cur < bucket + ldmBucketSize; ++cur) {
-                const BYTE* const pMatch = cur->offset + base;
                 size_t curForwardMatchLength, curBackwardMatchLength,
                        curTotalMatchLength;
                 if (cur->checksum != checksum || cur->offset <= lowestIndex) {
                     continue;
                 }
-
-                curForwardMatchLength = ZSTD_count(ip, pMatch, iend);
-                if (curForwardMatchLength < ldmParams.minMatchLength) {
-                    continue;
+                if (extDict) {
+                    BYTE const* const curMatchBase =
+                        cur->offset < dictLimit ? dictBase : base;
+                    BYTE const* const pMatch = curMatchBase + cur->offset;
+                    BYTE const* const matchEnd =
+                        cur->offset < dictLimit ? dictEnd : iend;
+                    BYTE const* const lowMatchPtr =
+                        cur->offset < dictLimit ? dictStart : lowPrefixPtr;
+
+                    curForwardMatchLength = ZSTD_count_2segments(
+                                                ip, pMatch, iend,
+                                                matchEnd, lowPrefixPtr);
+                    if (curForwardMatchLength < minMatchLength) {
+                        continue;
+                    }
+                    curBackwardMatchLength =
+                        ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
+                                                     lowMatchPtr);
+                    curTotalMatchLength = curForwardMatchLength +
+                                          curBackwardMatchLength;
+                } else { /* !extDict */
+                    BYTE const* const pMatch = base + cur->offset;
+                    curForwardMatchLength = ZSTD_count(ip, pMatch, iend);
+                    if (curForwardMatchLength < minMatchLength) {
+                        continue;
+                    }
+                    curBackwardMatchLength =
+                        ZSTD_ldm_countBackwardsMatch(ip, anchor, pMatch,
+                                                     lowPrefixPtr);
+                    curTotalMatchLength = curForwardMatchLength +
+                                          curBackwardMatchLength;
                 }
-                curBackwardMatchLength = ZSTD_ldm_countBackwardsMatch(
-                                             ip, anchor, pMatch, lowest);
-                curTotalMatchLength = curForwardMatchLength +
-                                      curBackwardMatchLength;
 
                 if (curTotalMatchLength > bestMatchLength) {
                     bestMatchLength = curTotalMatchLength;
@@ -371,7 +366,7 @@ size_t ZSTD_compressBlock_ldm_generic(ZSTD_CCtx* cctx,
         if (bestEntry == NULL) {
             ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash,
                                              hBits, current,
-                                             ldmParams);
+                                             *params);
             ip++;
             continue;
         }
@@ -380,324 +375,245 @@ size_t ZSTD_compressBlock_ldm_generic(ZSTD_CCtx* cctx,
         mLength = forwardMatchLength + backwardMatchLength;
         ip -= backwardMatchLength;
 
-        /* Call the block compressor on the remaining literals */
         {
+            /* Store the sequence:
+             * ip = current - backwardMatchLength
+             * The match is at (bestEntry->offset - backwardMatchLength)
+             */
             U32 const matchIndex = bestEntry->offset;
-            const BYTE* const match = base + matchIndex - backwardMatchLength;
-            U32 const offset = (U32)(ip - match);
-
-            /* Overwrite rep codes */
-            for (i = 0; i < ZSTD_REP_NUM; i++)
-                seqStorePtr->rep[i] = repToConfirm[i];
-
-            /* Fill tables for block compressor */
-            ZSTD_ldm_limitTableUpdate(cctx, anchor);
-            ZSTD_ldm_fillFastTables(cctx, anchor);
-
-            /* Call block compressor and get remaining literals */
-            lastLiterals = blockCompressor(cctx, anchor, ip - anchor);
-            cctx->nextToUpdate = (U32)(ip - base);
-
-            /* Update repToConfirm with the new offset */
-            for (i = ZSTD_REP_NUM - 1; i > 0; i--)
-                repToConfirm[i] = repToConfirm[i-1];
-            repToConfirm[0] = offset;
-
-            /* Store the sequence with the leftover literals */
-            ZSTD_storeSeq(seqStorePtr, lastLiterals, ip - lastLiterals,
-                          offset + ZSTD_REP_MOVE, mLength - MINMATCH);
+            U32 const offset = current - matchIndex;
+            rawSeq* const seq = rawSeqStore->seq + rawSeqStore->size;
+
+            /* Out of sequence storage */
+            if (rawSeqStore->size == rawSeqStore->capacity)
+                return ERROR(dstSize_tooSmall);
+            seq->litLength = (U32)(ip - anchor);
+            seq->matchLength = (U32)mLength;
+            seq->offset = offset;
+            rawSeqStore->size++;
         }
 
         /* Insert the current entry into the hash table */
         ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash, hBits,
                                          (U32)(lastHashed - base),
-                                         ldmParams);
+                                         *params);
 
         assert(ip + backwardMatchLength == lastHashed);
 
         /* Fill the hash table from lastHashed+1 to ip+mLength*/
         /* Heuristic: don't need to fill the entire table at end of block */
-        if (ip + mLength < ilimit) {
+        if (ip + mLength <= ilimit) {
             rollingHash = ZSTD_ldm_fillLdmHashTable(
                               ldmState, rollingHash, lastHashed,
-                              ip + mLength, base, hBits, ldmParams);
+                              ip + mLength, base, hBits, *params);
             lastHashed = ip + mLength - 1;
         }
         ip += mLength;
         anchor = ip;
-        /* Check immediate repcode */
-        while ( (ip < ilimit)
-             && ( (repToConfirm[1] > 0) && (repToConfirm[1] <= (U32)(ip-lowest))
-             && (MEM_read32(ip) == MEM_read32(ip - repToConfirm[1])) )) {
-
-            size_t const rLength = ZSTD_count(ip+4, ip+4-repToConfirm[1],
-                                              iend) + 4;
-            /* Swap repToConfirm[1] <=> repToConfirm[0] */
-            {
-                U32 const tmpOff = repToConfirm[1];
-                repToConfirm[1] = repToConfirm[0];
-                repToConfirm[0] = tmpOff;
-            }
-
-            ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, rLength-MINMATCH);
-
-            /* Fill the  hash table from lastHashed+1 to ip+rLength*/
-            if (ip + rLength < ilimit) {
-                rollingHash = ZSTD_ldm_fillLdmHashTable(
-                                ldmState, rollingHash, lastHashed,
-                                ip + rLength, base, hBits, ldmParams);
-                lastHashed = ip + rLength - 1;
-            }
-            ip += rLength;
-            anchor = ip;
-        }
     }
-
-    /* Overwrite rep */
-    for (i = 0; i < ZSTD_REP_NUM; i++)
-        seqStorePtr->rep[i] = repToConfirm[i];
-
-    ZSTD_ldm_limitTableUpdate(cctx, anchor);
-    ZSTD_ldm_fillFastTables(cctx, anchor);
-
-    lastLiterals = blockCompressor(cctx, anchor, iend - anchor);
-    cctx->nextToUpdate = (U32)(iend - base);
-
-    /* Restore seqStorePtr->rep */
-    for (i = 0; i < ZSTD_REP_NUM; i++)
-        seqStorePtr->rep[i] = savedRep[i];
-
-    /* Return the last literals size */
-    return lastLiterals;
+    return iend - anchor;
 }
 
-size_t ZSTD_compressBlock_ldm(ZSTD_CCtx* ctx,
-                              const void* src, size_t srcSize)
+/*! ZSTD_ldm_reduceTable() :
+ *  reduce table indexes by `reducerValue` */
+static void ZSTD_ldm_reduceTable(ldmEntry_t* const table, U32 const size,
+                                 U32 const reducerValue)
 {
-    return ZSTD_compressBlock_ldm_generic(ctx, src, srcSize);
+    U32 u;
+    for (u = 0; u < size; u++) {
+        if (table[u].offset < reducerValue) table[u].offset = 0;
+        else table[u].offset -= reducerValue;
+    }
 }
 
-static size_t ZSTD_compressBlock_ldm_extDict_generic(
-                                 ZSTD_CCtx* ctx,
-                                 const void* src, size_t srcSize)
+size_t ZSTD_ldm_generateSequences(
+        ldmState_t* ldmState, rawSeqStore_t* sequences,
+        ldmParams_t const* params, void const* src, size_t srcSize)
 {
-    ldmState_t* const ldmState = &(ctx->ldmState);
-    const ldmParams_t ldmParams = ctx->appliedParams.ldmParams;
-    const U64 hashPower = ldmState->hashPower;
-    const U32 hBits = ldmParams.hashLog - ldmParams.bucketSizeLog;
-    const U32 ldmBucketSize = ((U32)1 << ldmParams.bucketSizeLog);
-    const U32 ldmTagMask = ((U32)1 << ldmParams.hashEveryLog) - 1;
-    seqStore_t* const seqStorePtr = &(ctx->seqStore);
-    const BYTE* const base = ctx->base;
-    const BYTE* const dictBase = ctx->dictBase;
-    const BYTE* const istart = (const BYTE*)src;
-    const BYTE* ip = istart;
-    const BYTE* anchor = istart;
-    const U32   lowestIndex = ctx->lowLimit;
-    const BYTE* const dictStart = dictBase + lowestIndex;
-    const U32   dictLimit = ctx->dictLimit;
-    const BYTE* const lowPrefixPtr = base + dictLimit;
-    const BYTE* const dictEnd = dictBase + dictLimit;
-    const BYTE* const iend = istart + srcSize;
-    const BYTE* const ilimit = iend - MAX(ldmParams.minMatchLength, HASH_READ_SIZE);
-
-    const ZSTD_blockCompressor blockCompressor =
-        ZSTD_selectBlockCompressor(ctx->appliedParams.cParams.strategy, 1);
-    U32* const repToConfirm = seqStorePtr->repToConfirm;
-    U32 savedRep[ZSTD_REP_NUM];
-    U64 rollingHash = 0;
-    const BYTE* lastHashed = NULL;
-    size_t i, lastLiterals;
-
-    /* Save seqStorePtr->rep and copy repToConfirm */
-    for (i = 0; i < ZSTD_REP_NUM; i++) {
-        savedRep[i] = repToConfirm[i] = seqStorePtr->rep[i];
-    }
-
-    /* Search Loop */
-    while (ip < ilimit) {  /* < instead of <=, because (ip+1) */
-        size_t mLength;
-        const U32 current = (U32)(ip-base);
-        size_t forwardMatchLength = 0, backwardMatchLength = 0;
-        ldmEntry_t* bestEntry = NULL;
-        if (ip != istart) {
-          rollingHash = ZSTD_ldm_updateHash(rollingHash, lastHashed[0],
-                                       lastHashed[ldmParams.minMatchLength],
-                                       hashPower);
+    U32 const maxDist = 1U << params->windowLog;
+    BYTE const* const istart = (BYTE const*)src;
+    BYTE const* const iend = istart + srcSize;
+    size_t const kMaxChunkSize = 1 << 20;
+    size_t const nbChunks = (srcSize / kMaxChunkSize) + ((srcSize % kMaxChunkSize) != 0);
+    size_t chunk;
+    size_t leftoverSize = 0;
+
+    assert(ZSTD_CHUNKSIZE_MAX >= kMaxChunkSize);
+    /* Check that ZSTD_window_update() has been called for this chunk prior
+     * to passing it to this function.
+     */
+    assert(ldmState->window.nextSrc >= (BYTE const*)src + srcSize);
+    /* The input could be very large (in zstdmt), so it must be broken up into
+     * chunks to enforce the maximum distance and handle overflow correction.
+     */
+    assert(sequences->pos <= sequences->size);
+    assert(sequences->size <= sequences->capacity);
+    for (chunk = 0; chunk < nbChunks && sequences->size < sequences->capacity; ++chunk) {
+        BYTE const* const chunkStart = istart + chunk * kMaxChunkSize;
+        size_t const remaining = (size_t)(iend - chunkStart);
+        BYTE const *const chunkEnd =
+            (remaining < kMaxChunkSize) ? iend : chunkStart + kMaxChunkSize;
+        size_t const chunkSize = chunkEnd - chunkStart;
+        size_t newLeftoverSize;
+        size_t const prevSize = sequences->size;
+
+        assert(chunkStart < iend);
+        /* 1. Perform overflow correction if necessary. */
+        if (ZSTD_window_needOverflowCorrection(ldmState->window, chunkEnd)) {
+            U32 const ldmHSize = 1U << params->hashLog;
+            U32 const correction = ZSTD_window_correctOverflow(
+                &ldmState->window, /* cycleLog */ 0, maxDist, chunkStart);
+            ZSTD_ldm_reduceTable(ldmState->hashTable, ldmHSize, correction);
+            /* invalidate dictionaries on overflow correction */
+            ldmState->loadedDictEnd = 0;
+        }
+        /* 2. We enforce the maximum offset allowed.
+         *
+         * kMaxChunkSize should be small enough that we don't lose too much of
+         * the window through early invalidation.
+         * TODO: * Test the chunk size.
+         *       * Try invalidation after the sequence generation and test the
+         *         the offset against maxDist directly.
+         *
+         * NOTE: Because of dictionaries + sequence splitting we MUST make sure
+         * that any offset used is valid at the END of the sequence, since it may
+         * be split into two sequences. This condition holds when using
+         * ZSTD_window_enforceMaxDist(), but if we move to checking offsets
+         * against maxDist directly, we'll have to carefully handle that case.
+         */
+        ZSTD_window_enforceMaxDist(&ldmState->window, chunkEnd, maxDist, &ldmState->loadedDictEnd, NULL);
+        /* 3. Generate the sequences for the chunk, and get newLeftoverSize. */
+        newLeftoverSize = ZSTD_ldm_generateSequences_internal(
+            ldmState, sequences, params, chunkStart, chunkSize);
+        if (ZSTD_isError(newLeftoverSize))
+            return newLeftoverSize;
+        /* 4. We add the leftover literals from previous iterations to the first
+         *    newly generated sequence, or add the `newLeftoverSize` if none are
+         *    generated.
+         */
+        /* Prepend the leftover literals from the last call */
+        if (prevSize < sequences->size) {
+            sequences->seq[prevSize].litLength += (U32)leftoverSize;
+            leftoverSize = newLeftoverSize;
         } else {
-            rollingHash = ZSTD_ldm_getRollingHash(ip, ldmParams.minMatchLength);
+            assert(newLeftoverSize == chunkSize);
+            leftoverSize += chunkSize;
         }
-        lastHashed = ip;
+    }
+    return 0;
+}
 
-        if (ZSTD_ldm_getTag(rollingHash, hBits, ldmParams.hashEveryLog) !=
-                ldmTagMask) {
-            /* Don't insert and don't look for a match */
-           ip++;
-           continue;
+void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize, U32 const minMatch) {
+    while (srcSize > 0 && rawSeqStore->pos < rawSeqStore->size) {
+        rawSeq* seq = rawSeqStore->seq + rawSeqStore->pos;
+        if (srcSize <= seq->litLength) {
+            /* Skip past srcSize literals */
+            seq->litLength -= (U32)srcSize;
+            return;
         }
-
-        /* Get the best entry and compute the match lengths */
-        {
-            ldmEntry_t* const bucket =
-                ZSTD_ldm_getBucket(ldmState,
-                                   ZSTD_ldm_getSmallHash(rollingHash, hBits),
-                                   ldmParams);
-            ldmEntry_t* cur;
-            size_t bestMatchLength = 0;
-            U32 const checksum = ZSTD_ldm_getChecksum(rollingHash, hBits);
-
-            for (cur = bucket; cur < bucket + ldmBucketSize; ++cur) {
-                const BYTE* const curMatchBase =
-                    cur->offset < dictLimit ? dictBase : base;
-                const BYTE* const pMatch = curMatchBase + cur->offset;
-                const BYTE* const matchEnd =
-                    cur->offset < dictLimit ? dictEnd : iend;
-                const BYTE* const lowMatchPtr =
-                    cur->offset < dictLimit ? dictStart : lowPrefixPtr;
-                size_t curForwardMatchLength, curBackwardMatchLength,
-                       curTotalMatchLength;
-
-                if (cur->checksum != checksum || cur->offset <= lowestIndex) {
-                    continue;
-                }
-
-                curForwardMatchLength = ZSTD_count_2segments(
-                                            ip, pMatch, iend,
-                                            matchEnd, lowPrefixPtr);
-                if (curForwardMatchLength < ldmParams.minMatchLength) {
-                    continue;
-                }
-                curBackwardMatchLength = ZSTD_ldm_countBackwardsMatch(
-                                             ip, anchor, pMatch, lowMatchPtr);
-                curTotalMatchLength = curForwardMatchLength +
-                                      curBackwardMatchLength;
-
-                if (curTotalMatchLength > bestMatchLength) {
-                    bestMatchLength = curTotalMatchLength;
-                    forwardMatchLength = curForwardMatchLength;
-                    backwardMatchLength = curBackwardMatchLength;
-                    bestEntry = cur;
+        srcSize -= seq->litLength;
+        seq->litLength = 0;
+        if (srcSize < seq->matchLength) {
+            /* Skip past the first srcSize of the match */
+            seq->matchLength -= (U32)srcSize;
+            if (seq->matchLength < minMatch) {
+                /* The match is too short, omit it */
+                if (rawSeqStore->pos + 1 < rawSeqStore->size) {
+                    seq[1].litLength += seq[0].matchLength;
                 }
+                rawSeqStore->pos++;
             }
+            return;
         }
+        srcSize -= seq->matchLength;
+        seq->matchLength = 0;
+        rawSeqStore->pos++;
+    }
+}
 
-        /* No match found -- continue searching */
-        if (bestEntry == NULL) {
-            ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash, hBits,
-                                             (U32)(lastHashed - base),
-                                             ldmParams);
-            ip++;
-            continue;
+/**
+ * If the sequence length is longer than remaining then the sequence is split
+ * between this block and the next.
+ *
+ * Returns the current sequence to handle, or if the rest of the block should
+ * be literals, it returns a sequence with offset == 0.
+ */
+static rawSeq maybeSplitSequence(rawSeqStore_t* rawSeqStore,
+                                 U32 const remaining, U32 const minMatch)
+{
+    rawSeq sequence = rawSeqStore->seq[rawSeqStore->pos];
+    assert(sequence.offset > 0);
+    /* Likely: No partial sequence */
+    if (remaining >= sequence.litLength + sequence.matchLength) {
+        rawSeqStore->pos++;
+        return sequence;
+    }
+    /* Cut the sequence short (offset == 0 ==> rest is literals). */
+    if (remaining <= sequence.litLength) {
+        sequence.offset = 0;
+    } else if (remaining < sequence.litLength + sequence.matchLength) {
+        sequence.matchLength = remaining - sequence.litLength;
+        if (sequence.matchLength < minMatch) {
+            sequence.offset = 0;
         }
+    }
+    /* Skip past `remaining` bytes for the future sequences. */
+    ZSTD_ldm_skipSequences(rawSeqStore, remaining, minMatch);
+    return sequence;
+}
 
-        /* Match found */
-        mLength = forwardMatchLength + backwardMatchLength;
-        ip -= backwardMatchLength;
-
-        /* Call the block compressor on the remaining literals */
-        {
-            /* ip = current - backwardMatchLength
-             * The match is at (bestEntry->offset - backwardMatchLength) */
-            U32 const matchIndex = bestEntry->offset;
-            U32 const offset = current - matchIndex;
-
-            /* Overwrite rep codes */
-            for (i = 0; i < ZSTD_REP_NUM; i++)
-                seqStorePtr->rep[i] = repToConfirm[i];
-
-            /* Fill the hash table for the block compressor */
-            ZSTD_ldm_limitTableUpdate(ctx, anchor);
-            ZSTD_ldm_fillFastTables(ctx, anchor);
+size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
+    ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+    void const* src, size_t srcSize)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    unsigned const minMatch = cParams->minMatch;
+    ZSTD_blockCompressor const blockCompressor =
+        ZSTD_selectBlockCompressor(cParams->strategy, ZSTD_matchState_dictMode(ms));
+    /* Input bounds */
+    BYTE const* const istart = (BYTE const*)src;
+    BYTE const* const iend = istart + srcSize;
+    /* Input positions */
+    BYTE const* ip = istart;
+
+    DEBUGLOG(5, "ZSTD_ldm_blockCompress: srcSize=%zu", srcSize);
+    assert(rawSeqStore->pos <= rawSeqStore->size);
+    assert(rawSeqStore->size <= rawSeqStore->capacity);
+    /* Loop through each sequence and apply the block compressor to the lits */
+    while (rawSeqStore->pos < rawSeqStore->size && ip < iend) {
+        /* maybeSplitSequence updates rawSeqStore->pos */
+        rawSeq const sequence = maybeSplitSequence(rawSeqStore,
+                                                   (U32)(iend - ip), minMatch);
+        int i;
+        /* End signal */
+        if (sequence.offset == 0)
+            break;
 
-            /* Call block compressor and get remaining literals  */
-            lastLiterals = blockCompressor(ctx, anchor, ip - anchor);
-            ctx->nextToUpdate = (U32)(ip - base);
+        assert(ip + sequence.litLength + sequence.matchLength <= iend);
 
-            /* Update repToConfirm with the new offset */
+        /* Fill tables for block compressor */
+        ZSTD_ldm_limitTableUpdate(ms, ip);
+        ZSTD_ldm_fillFastTables(ms, ip);
+        /* Run the block compressor */
+        DEBUGLOG(5, "pos %u : calling block compressor on segment of size %u", (unsigned)(ip-istart), sequence.litLength);
+        {
+            size_t const newLitLength =
+                blockCompressor(ms, seqStore, rep, ip, sequence.litLength);
+            ip += sequence.litLength;
+            /* Update the repcodes */
             for (i = ZSTD_REP_NUM - 1; i > 0; i--)
-                repToConfirm[i] = repToConfirm[i-1];
-            repToConfirm[0] = offset;
-
-            /* Store the sequence with the leftover literals */
-            ZSTD_storeSeq(seqStorePtr, lastLiterals, ip - lastLiterals,
-                          offset + ZSTD_REP_MOVE, mLength - MINMATCH);
-        }
-
-        /* Insert the current entry into the hash table */
-        ZSTD_ldm_makeEntryAndInsertByTag(ldmState, rollingHash, hBits,
-                                         (U32)(lastHashed - base),
-                                         ldmParams);
-
-        /* Fill the hash table from lastHashed+1 to ip+mLength */
-        assert(ip + backwardMatchLength == lastHashed);
-        if (ip + mLength < ilimit) {
-            rollingHash = ZSTD_ldm_fillLdmHashTable(
-                              ldmState, rollingHash, lastHashed,
-                              ip + mLength, base, hBits,
-                              ldmParams);
-            lastHashed = ip + mLength - 1;
-        }
-        ip += mLength;
-        anchor = ip;
-
-        /* check immediate repcode */
-        while (ip < ilimit) {
-            U32 const current2 = (U32)(ip-base);
-            U32 const repIndex2 = current2 - repToConfirm[1];
-            const BYTE* repMatch2 = repIndex2 < dictLimit ?
-                                    dictBase + repIndex2 : base + repIndex2;
-            if ( (((U32)((dictLimit-1) - repIndex2) >= 3) &
-                        (repIndex2 > lowestIndex))  /* intentional overflow */
-               && (MEM_read32(repMatch2) == MEM_read32(ip)) ) {
-                const BYTE* const repEnd2 = repIndex2 < dictLimit ?
-                                            dictEnd : iend;
-                size_t const repLength2 =
-                        ZSTD_count_2segments(ip+4, repMatch2+4, iend,
-                                             repEnd2, lowPrefixPtr) + 4;
-
-                U32 tmpOffset = repToConfirm[1];
-                repToConfirm[1] = repToConfirm[0];
-                repToConfirm[0] = tmpOffset;
-
-                ZSTD_storeSeq(seqStorePtr, 0, anchor, 0, repLength2-MINMATCH);
-
-                /* Fill the  hash table from lastHashed+1 to ip+repLength2*/
-                if (ip + repLength2 < ilimit) {
-                    rollingHash = ZSTD_ldm_fillLdmHashTable(
-                                      ldmState, rollingHash, lastHashed,
-                                      ip + repLength2, base, hBits,
-                                      ldmParams);
-                    lastHashed = ip + repLength2 - 1;
-                }
-                ip += repLength2;
-                anchor = ip;
-                continue;
-            }
-            break;
+                rep[i] = rep[i-1];
+            rep[0] = sequence.offset;
+            /* Store the sequence */
+            ZSTD_storeSeq(seqStore, newLitLength, ip - newLitLength, iend,
+                          sequence.offset + ZSTD_REP_MOVE,
+                          sequence.matchLength - MINMATCH);
+            ip += sequence.matchLength;
         }
     }
-
-    /* Overwrite rep */
-    for (i = 0; i < ZSTD_REP_NUM; i++)
-        seqStorePtr->rep[i] = repToConfirm[i];
-
-    ZSTD_ldm_limitTableUpdate(ctx, anchor);
-    ZSTD_ldm_fillFastTables(ctx, anchor);
-
-    /* Call the block compressor one last time on the last literals */
-    lastLiterals = blockCompressor(ctx, anchor, iend - anchor);
-    ctx->nextToUpdate = (U32)(iend - base);
-
-    /* Restore seqStorePtr->rep */
-    for (i = 0; i < ZSTD_REP_NUM; i++)
-        seqStorePtr->rep[i] = savedRep[i];
-
-    /* Return the last literals size */
-    return lastLiterals;
-}
-
-size_t ZSTD_compressBlock_ldm_extDict(ZSTD_CCtx* ctx,
-                                      const void* src, size_t srcSize)
-{
-    return ZSTD_compressBlock_ldm_extDict_generic(ctx, src, srcSize);
+    /* Fill the tables for the block compressor */
+    ZSTD_ldm_limitTableUpdate(ms, ip);
+    ZSTD_ldm_fillFastTables(ms, ip);
+    /* Compress the last literals */
+    return blockCompressor(ms, seqStore, rep, ip, iend - ip);
 }
diff --git a/contrib/zstd/zstd_ldm.h b/contrib/zstd/zstd_ldm.h
index 7a6248399..229ea05a9 100644
--- a/contrib/zstd/zstd_ldm.h
+++ b/contrib/zstd/zstd_ldm.h
@@ -1,64 +1,107 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
  * in the COPYING file in the root directory of this source tree).
+ * You may select, at your option, one of the above-listed licenses.
  */
 
 #ifndef ZSTD_LDM_H
 #define ZSTD_LDM_H
 
-#include "zstd_compress.h"
-
 #if defined (__cplusplus)
 extern "C" {
 #endif
 
+#include "zstd_compress_internal.h"   /* ldmParams_t, U32 */
+#include "../zstd.h"   /* ZSTD_CCtx, size_t */
+
 /*-*************************************
 *  Long distance matching
 ***************************************/
 
-#define ZSTD_LDM_WINDOW_LOG 27
-#define ZSTD_LDM_HASHEVERYLOG_NOTSET 9999
+#define ZSTD_LDM_DEFAULT_WINDOW_LOG ZSTD_WINDOWLOG_LIMIT_DEFAULT
 
-/** ZSTD_compressBlock_ldm_generic() :
+void ZSTD_ldm_fillHashTable(
+            ldmState_t* state, const BYTE* ip,
+            const BYTE* iend, ldmParams_t const* params);
+
+/**
+ * ZSTD_ldm_generateSequences():
+ *
+ * Generates the sequences using the long distance match finder.
+ * Generates long range matching sequences in `sequences`, which parse a prefix
+ * of the source. `sequences` must be large enough to store every sequence,
+ * which can be checked with `ZSTD_ldm_getMaxNbSeq()`.
+ * @returns 0 or an error code.
  *
- *  This is a block compressor intended for long distance matching.
+ * NOTE: The user must have called ZSTD_window_update() for all of the input
+ * they have, even if they pass it to ZSTD_ldm_generateSequences() in chunks.
+ * NOTE: This function returns an error if it runs out of space to store
+ *       sequences.
+ */
+size_t ZSTD_ldm_generateSequences(
+            ldmState_t* ldms, rawSeqStore_t* sequences,
+            ldmParams_t const* params, void const* src, size_t srcSize);
+
+/**
+ * ZSTD_ldm_blockCompress():
  *
- *  The function searches for matches of length at least
- *  ldmParams.minMatchLength using a hash table in cctx->ldmState.
- *  Matches can be at a distance of up to cParams.windowLog.
+ * Compresses a block using the predefined sequences, along with a secondary
+ * block compressor. The literals section of every sequence is passed to the
+ * secondary block compressor, and those sequences are interspersed with the
+ * predefined sequences. Returns the length of the last literals.
+ * Updates `rawSeqStore.pos` to indicate how many sequences have been consumed.
+ * `rawSeqStore.seq` may also be updated to split the last sequence between two
+ * blocks.
+ * @return The length of the last literals.
  *
- *  Upon finding a match, the unmatched literals are compressed using a
- *  ZSTD_blockCompressor (depending on the strategy in the compression
- *  parameters), which stores the matched sequences. The "long distance"
- *  match is then stored with the remaining literals from the
- *  ZSTD_blockCompressor. */
-size_t ZSTD_compressBlock_ldm(ZSTD_CCtx* cctx, const void* src, size_t srcSize);
-size_t ZSTD_compressBlock_ldm_extDict(ZSTD_CCtx* ctx,
-                                      const void* src, size_t srcSize);
+ * NOTE: The source must be at most the maximum block size, but the predefined
+ * sequences can be any size, and may be longer than the block. In the case that
+ * they are longer than the block, the last sequences may need to be split into
+ * two. We handle that case correctly, and update `rawSeqStore` appropriately.
+ * NOTE: This function does not return any errors.
+ */
+size_t ZSTD_ldm_blockCompress(rawSeqStore_t* rawSeqStore,
+            ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+            void const* src, size_t srcSize);
 
-/** ZSTD_ldm_initializeParameters() :
- *  Initialize the long distance matching parameters to their default values. */
-size_t ZSTD_ldm_initializeParameters(ldmParams_t* params, U32 enableLdm);
+/**
+ * ZSTD_ldm_skipSequences():
+ *
+ * Skip past `srcSize` bytes worth of sequences in `rawSeqStore`.
+ * Avoids emitting matches less than `minMatch` bytes.
+ * Must be called for data with is not passed to ZSTD_ldm_blockCompress().
+ */
+void ZSTD_ldm_skipSequences(rawSeqStore_t* rawSeqStore, size_t srcSize,
+    U32 const minMatch);
 
-/** ZSTD_ldm_getTableSize() :
- *  Estimate the space needed for long distance matching tables. */
-size_t ZSTD_ldm_getTableSize(U32 hashLog, U32 bucketSizeLog);
 
 /** ZSTD_ldm_getTableSize() :
- *  Return prime8bytes^(minMatchLength-1) */
-U64 ZSTD_ldm_getHashPower(U32 minMatchLength);
+ *  Estimate the space needed for long distance matching tables or 0 if LDM is
+ *  disabled.
+ */
+size_t ZSTD_ldm_getTableSize(ldmParams_t params);
+
+/** ZSTD_ldm_getSeqSpace() :
+ *  Return an upper bound on the number of sequences that can be produced by
+ *  the long distance matcher, or 0 if LDM is disabled.
+ */
+size_t ZSTD_ldm_getMaxNbSeq(ldmParams_t params, size_t maxChunkSize);
 
 /** ZSTD_ldm_adjustParameters() :
- *  If the params->hashEveryLog is not set, set it to its default value based on
+ *  If the params->hashRateLog is not set, set it to its default value based on
  *  windowLog and params->hashLog.
  *
  *  Ensures that params->bucketSizeLog is <= params->hashLog (setting it to
- *  params->hashLog if it is not). */
-void ZSTD_ldm_adjustParameters(ldmParams_t* params, U32 windowLog);
+ *  params->hashLog if it is not).
+ *
+ *  Ensures that the minMatchLength >= targetLength during optimal parsing.
+ */
+void ZSTD_ldm_adjustParameters(ldmParams_t* params,
+                               ZSTD_compressionParameters const* cParams);
 
 #if defined (__cplusplus)
 }
diff --git a/contrib/zstd/zstd_opt.c b/contrib/zstd/zstd_opt.c
index fd102da2d..36fff050c 100644
--- a/contrib/zstd/zstd_opt.c
+++ b/contrib/zstd/zstd_opt.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Przemyslaw Skibinski, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -8,199 +8,321 @@
  * You may select, at your option, one of the above-listed licenses.
  */
 
+#include "zstd_compress_internal.h"
+#include "hist.h"
 #include "zstd_opt.h"
-#include "zstd_lazy.h"
 
 
-#define ZSTD_LITFREQ_ADD    2
-#define ZSTD_FREQ_DIV       4
-#define ZSTD_MAX_PRICE      (1<<30)
+#define ZSTD_LITFREQ_ADD    2   /* scaling factor for litFreq, so that frequencies adapt faster to new stats */
+#define ZSTD_FREQ_DIV       4   /* log factor when using previous stats to init next stats */
+#define ZSTD_MAX_PRICE     (1<<30)
+
+#define ZSTD_PREDEF_THRESHOLD 1024   /* if srcSize < ZSTD_PREDEF_THRESHOLD, symbols' cost is assumed static, directly determined by pre-defined distributions */
+
 
 /*-*************************************
 *  Price functions for optimal parser
 ***************************************/
-static void ZSTD_setLog2Prices(optState_t* optPtr)
+
+#if 0    /* approximation at bit level */
+#  define BITCOST_ACCURACY 0
+#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
+#  define WEIGHT(stat)  ((void)opt, ZSTD_bitWeight(stat))
+#elif 0  /* fractional bit accuracy */
+#  define BITCOST_ACCURACY 8
+#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
+#  define WEIGHT(stat,opt) ((void)opt, ZSTD_fracWeight(stat))
+#else    /* opt==approx, ultra==accurate */
+#  define BITCOST_ACCURACY 8
+#  define BITCOST_MULTIPLIER (1 << BITCOST_ACCURACY)
+#  define WEIGHT(stat,opt) (opt ? ZSTD_fracWeight(stat) : ZSTD_bitWeight(stat))
+#endif
+
+MEM_STATIC U32 ZSTD_bitWeight(U32 stat)
 {
-    optPtr->log2matchLengthSum = ZSTD_highbit32(optPtr->matchLengthSum+1);
-    optPtr->log2litLengthSum = ZSTD_highbit32(optPtr->litLengthSum+1);
-    optPtr->log2litSum = ZSTD_highbit32(optPtr->litSum+1);
-    optPtr->log2offCodeSum = ZSTD_highbit32(optPtr->offCodeSum+1);
-    optPtr->factor = 1 + ((optPtr->litSum>>5) / optPtr->litLengthSum) + ((optPtr->litSum<<1) / (optPtr->litSum + optPtr->matchSum));
+    return (ZSTD_highbit32(stat+1) * BITCOST_MULTIPLIER);
 }
 
+MEM_STATIC U32 ZSTD_fracWeight(U32 rawStat)
+{
+    U32 const stat = rawStat + 1;
+    U32 const hb = ZSTD_highbit32(stat);
+    U32 const BWeight = hb * BITCOST_MULTIPLIER;
+    U32 const FWeight = (stat << BITCOST_ACCURACY) >> hb;
+    U32 const weight = BWeight + FWeight;
+    assert(hb + BITCOST_ACCURACY < 31);
+    return weight;
+}
 
-static void ZSTD_rescaleFreqs(optState_t* optPtr, const BYTE* src, size_t srcSize)
+#if (DEBUGLEVEL>=2)
+/* debugging function,
+ * @return price in bytes as fractional value
+ * for debug messages only */
+MEM_STATIC double ZSTD_fCost(U32 price)
 {
-    unsigned u;
-
-    optPtr->cachedLiterals = NULL;
-    optPtr->cachedPrice = optPtr->cachedLitLength = 0;
-    optPtr->staticPrices = 0;
-
-    if (optPtr->litLengthSum == 0) {
-        if (srcSize <= 1024) optPtr->staticPrices = 1;
-
-        assert(optPtr->litFreq!=NULL);
-        for (u=0; u<=MaxLit; u++)
-            optPtr->litFreq[u] = 0;
-        for (u=0; u<srcSize; u++)
-            optPtr->litFreq[src[u]]++;
-
-        optPtr->litSum = 0;
-        optPtr->litLengthSum = MaxLL+1;
-        optPtr->matchLengthSum = MaxML+1;
-        optPtr->offCodeSum = (MaxOff+1);
-        optPtr->matchSum = (ZSTD_LITFREQ_ADD<<Litbits);
-
-        for (u=0; u<=MaxLit; u++) {
-            optPtr->litFreq[u] = 1 + (optPtr->litFreq[u]>>ZSTD_FREQ_DIV);
-            optPtr->litSum += optPtr->litFreq[u];
-        }
-        for (u=0; u<=MaxLL; u++)
-            optPtr->litLengthFreq[u] = 1;
-        for (u=0; u<=MaxML; u++)
-            optPtr->matchLengthFreq[u] = 1;
-        for (u=0; u<=MaxOff; u++)
-            optPtr->offCodeFreq[u] = 1;
-    } else {
-        optPtr->matchLengthSum = 0;
-        optPtr->litLengthSum = 0;
-        optPtr->offCodeSum = 0;
-        optPtr->matchSum = 0;
-        optPtr->litSum = 0;
-
-        for (u=0; u<=MaxLit; u++) {
-            optPtr->litFreq[u] = 1 + (optPtr->litFreq[u]>>(ZSTD_FREQ_DIV+1));
-            optPtr->litSum += optPtr->litFreq[u];
-        }
-        for (u=0; u<=MaxLL; u++) {
-            optPtr->litLengthFreq[u] = 1 + (optPtr->litLengthFreq[u]>>(ZSTD_FREQ_DIV+1));
-            optPtr->litLengthSum += optPtr->litLengthFreq[u];
-        }
-        for (u=0; u<=MaxML; u++) {
-            optPtr->matchLengthFreq[u] = 1 + (optPtr->matchLengthFreq[u]>>ZSTD_FREQ_DIV);
-            optPtr->matchLengthSum += optPtr->matchLengthFreq[u];
-            optPtr->matchSum += optPtr->matchLengthFreq[u] * (u + 3);
-        }
-        optPtr->matchSum *= ZSTD_LITFREQ_ADD;
-        for (u=0; u<=MaxOff; u++) {
-            optPtr->offCodeFreq[u] = 1 + (optPtr->offCodeFreq[u]>>ZSTD_FREQ_DIV);
-            optPtr->offCodeSum += optPtr->offCodeFreq[u];
-        }
-    }
+    return (double)price / (BITCOST_MULTIPLIER*8);
+}
+#endif
 
-    ZSTD_setLog2Prices(optPtr);
+static int ZSTD_compressedLiterals(optState_t const* const optPtr)
+{
+    return optPtr->literalCompressionMode != ZSTD_lcm_uncompressed;
+}
+
+static void ZSTD_setBasePrices(optState_t* optPtr, int optLevel)
+{
+    if (ZSTD_compressedLiterals(optPtr))
+        optPtr->litSumBasePrice = WEIGHT(optPtr->litSum, optLevel);
+    optPtr->litLengthSumBasePrice = WEIGHT(optPtr->litLengthSum, optLevel);
+    optPtr->matchLengthSumBasePrice = WEIGHT(optPtr->matchLengthSum, optLevel);
+    optPtr->offCodeSumBasePrice = WEIGHT(optPtr->offCodeSum, optLevel);
 }
 
 
-static U32 ZSTD_getLiteralPrice(optState_t* optPtr, U32 litLength, const BYTE* literals)
+/* ZSTD_downscaleStat() :
+ * reduce all elements in table by a factor 2^(ZSTD_FREQ_DIV+malus)
+ * return the resulting sum of elements */
+static U32 ZSTD_downscaleStat(unsigned* table, U32 lastEltIndex, int malus)
 {
-    U32 price, u;
+    U32 s, sum=0;
+    DEBUGLOG(5, "ZSTD_downscaleStat (nbElts=%u)", (unsigned)lastEltIndex+1);
+    assert(ZSTD_FREQ_DIV+malus > 0 && ZSTD_FREQ_DIV+malus < 31);
+    for (s=0; s<lastEltIndex+1; s++) {
+        table[s] = 1 + (table[s] >> (ZSTD_FREQ_DIV+malus));
+        sum += table[s];
+    }
+    return sum;
+}
 
-    if (optPtr->staticPrices)
-        return ZSTD_highbit32((U32)litLength+1) + (litLength*6);
+/* ZSTD_rescaleFreqs() :
+ * if first block (detected by optPtr->litLengthSum == 0) : init statistics
+ *    take hints from dictionary if there is one
+ *    or init from zero, using src for literals stats, or flat 1 for match symbols
+ * otherwise downscale existing stats, to be used as seed for next block.
+ */
+static void
+ZSTD_rescaleFreqs(optState_t* const optPtr,
+            const BYTE* const src, size_t const srcSize,
+                  int const optLevel)
+{
+    int const compressedLiterals = ZSTD_compressedLiterals(optPtr);
+    DEBUGLOG(5, "ZSTD_rescaleFreqs (srcSize=%u)", (unsigned)srcSize);
+    optPtr->priceType = zop_dynamic;
+
+    if (optPtr->litLengthSum == 0) {  /* first block : init */
+        if (srcSize <= ZSTD_PREDEF_THRESHOLD) {  /* heuristic */
+            DEBUGLOG(5, "(srcSize <= ZSTD_PREDEF_THRESHOLD) => zop_predef");
+            optPtr->priceType = zop_predef;
+        }
 
-    if (litLength == 0)
-        return optPtr->log2litLengthSum - ZSTD_highbit32(optPtr->litLengthFreq[0]+1);
+        assert(optPtr->symbolCosts != NULL);
+        if (optPtr->symbolCosts->huf.repeatMode == HUF_repeat_valid) {
+            /* huffman table presumed generated by dictionary */
+            optPtr->priceType = zop_dynamic;
+
+            if (compressedLiterals) {
+                unsigned lit;
+                assert(optPtr->litFreq != NULL);
+                optPtr->litSum = 0;
+                for (lit=0; lit<=MaxLit; lit++) {
+                    U32 const scaleLog = 11;   /* scale to 2K */
+                    U32 const bitCost = HUF_getNbBits(optPtr->symbolCosts->huf.CTable, lit);
+                    assert(bitCost <= scaleLog);
+                    optPtr->litFreq[lit] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
+                    optPtr->litSum += optPtr->litFreq[lit];
+            }   }
+
+            {   unsigned ll;
+                FSE_CState_t llstate;
+                FSE_initCState(&llstate, optPtr->symbolCosts->fse.litlengthCTable);
+                optPtr->litLengthSum = 0;
+                for (ll=0; ll<=MaxLL; ll++) {
+                    U32 const scaleLog = 10;   /* scale to 1K */
+                    U32 const bitCost = FSE_getMaxNbBits(llstate.symbolTT, ll);
+                    assert(bitCost < scaleLog);
+                    optPtr->litLengthFreq[ll] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
+                    optPtr->litLengthSum += optPtr->litLengthFreq[ll];
+            }   }
+
+            {   unsigned ml;
+                FSE_CState_t mlstate;
+                FSE_initCState(&mlstate, optPtr->symbolCosts->fse.matchlengthCTable);
+                optPtr->matchLengthSum = 0;
+                for (ml=0; ml<=MaxML; ml++) {
+                    U32 const scaleLog = 10;
+                    U32 const bitCost = FSE_getMaxNbBits(mlstate.symbolTT, ml);
+                    assert(bitCost < scaleLog);
+                    optPtr->matchLengthFreq[ml] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
+                    optPtr->matchLengthSum += optPtr->matchLengthFreq[ml];
+            }   }
+
+            {   unsigned of;
+                FSE_CState_t ofstate;
+                FSE_initCState(&ofstate, optPtr->symbolCosts->fse.offcodeCTable);
+                optPtr->offCodeSum = 0;
+                for (of=0; of<=MaxOff; of++) {
+                    U32 const scaleLog = 10;
+                    U32 const bitCost = FSE_getMaxNbBits(ofstate.symbolTT, of);
+                    assert(bitCost < scaleLog);
+                    optPtr->offCodeFreq[of] = bitCost ? 1 << (scaleLog-bitCost) : 1 /*minimum to calculate cost*/;
+                    optPtr->offCodeSum += optPtr->offCodeFreq[of];
+            }   }
+
+        } else {  /* not a dictionary */
+
+            assert(optPtr->litFreq != NULL);
+            if (compressedLiterals) {
+                unsigned lit = MaxLit;
+                HIST_count_simple(optPtr->litFreq, &lit, src, srcSize);   /* use raw first block to init statistics */
+                optPtr->litSum = ZSTD_downscaleStat(optPtr->litFreq, MaxLit, 1);
+            }
 
-    /* literals */
-    if (optPtr->cachedLiterals == literals) {
-        U32 const additional = litLength - optPtr->cachedLitLength;
-        const BYTE* literals2 = optPtr->cachedLiterals + optPtr->cachedLitLength;
-        price = optPtr->cachedPrice + additional * optPtr->log2litSum;
-        for (u=0; u < additional; u++)
-            price -= ZSTD_highbit32(optPtr->litFreq[literals2[u]]+1);
-        optPtr->cachedPrice = price;
-        optPtr->cachedLitLength = litLength;
-    } else {
-        price = litLength * optPtr->log2litSum;
-        for (u=0; u < litLength; u++)
-            price -= ZSTD_highbit32(optPtr->litFreq[literals[u]]+1);
+            {   unsigned ll;
+                for (ll=0; ll<=MaxLL; ll++)
+                    optPtr->litLengthFreq[ll] = 1;
+            }
+            optPtr->litLengthSum = MaxLL+1;
+
+            {   unsigned ml;
+                for (ml=0; ml<=MaxML; ml++)
+                    optPtr->matchLengthFreq[ml] = 1;
+            }
+            optPtr->matchLengthSum = MaxML+1;
+
+            {   unsigned of;
+                for (of=0; of<=MaxOff; of++)
+                    optPtr->offCodeFreq[of] = 1;
+            }
+            optPtr->offCodeSum = MaxOff+1;
 
-        if (litLength >= 12) {
-            optPtr->cachedLiterals = literals;
-            optPtr->cachedPrice = price;
-            optPtr->cachedLitLength = litLength;
         }
-    }
 
-    /* literal Length */
-    {   const BYTE LL_deltaCode = 19;
-        const BYTE llCode = (litLength>63) ? (BYTE)ZSTD_highbit32(litLength) + LL_deltaCode : LL_Code[litLength];
-        price += LL_bits[llCode] + optPtr->log2litLengthSum - ZSTD_highbit32(optPtr->litLengthFreq[llCode]+1);
+    } else {   /* new block : re-use previous statistics, scaled down */
+
+        if (compressedLiterals)
+            optPtr->litSum = ZSTD_downscaleStat(optPtr->litFreq, MaxLit, 1);
+        optPtr->litLengthSum = ZSTD_downscaleStat(optPtr->litLengthFreq, MaxLL, 0);
+        optPtr->matchLengthSum = ZSTD_downscaleStat(optPtr->matchLengthFreq, MaxML, 0);
+        optPtr->offCodeSum = ZSTD_downscaleStat(optPtr->offCodeFreq, MaxOff, 0);
     }
 
-    return price;
+    ZSTD_setBasePrices(optPtr, optLevel);
+}
+
+/* ZSTD_rawLiteralsCost() :
+ * price of literals (only) in specified segment (which length can be 0).
+ * does not include price of literalLength symbol */
+static U32 ZSTD_rawLiteralsCost(const BYTE* const literals, U32 const litLength,
+                                const optState_t* const optPtr,
+                                int optLevel)
+{
+    if (litLength == 0) return 0;
+
+    if (!ZSTD_compressedLiterals(optPtr))
+        return (litLength << 3) * BITCOST_MULTIPLIER;  /* Uncompressed - 8 bytes per literal. */
+
+    if (optPtr->priceType == zop_predef)
+        return (litLength*6) * BITCOST_MULTIPLIER;  /* 6 bit per literal - no statistic used */
+
+    /* dynamic statistics */
+    {   U32 price = litLength * optPtr->litSumBasePrice;
+        U32 u;
+        for (u=0; u < litLength; u++) {
+            assert(WEIGHT(optPtr->litFreq[literals[u]], optLevel) <= optPtr->litSumBasePrice);   /* literal cost should never be negative */
+            price -= WEIGHT(optPtr->litFreq[literals[u]], optLevel);
+        }
+        return price;
+    }
 }
 
+/* ZSTD_litLengthPrice() :
+ * cost of literalLength symbol */
+static U32 ZSTD_litLengthPrice(U32 const litLength, const optState_t* const optPtr, int optLevel)
+{
+    if (optPtr->priceType == zop_predef) return WEIGHT(litLength, optLevel);
+
+    /* dynamic statistics */
+    {   U32 const llCode = ZSTD_LLcode(litLength);
+        return (LL_bits[llCode] * BITCOST_MULTIPLIER)
+             + optPtr->litLengthSumBasePrice
+             - WEIGHT(optPtr->litLengthFreq[llCode], optLevel);
+    }
+}
 
-FORCE_INLINE_TEMPLATE U32 ZSTD_getPrice(optState_t* optPtr, U32 litLength, const BYTE* literals, U32 offset, U32 matchLength, const int ultra)
+/* ZSTD_getMatchPrice() :
+ * Provides the cost of the match part (offset + matchLength) of a sequence
+ * Must be combined with ZSTD_fullLiteralsCost() to get the full cost of a sequence.
+ * optLevel: when <2, favors small offset for decompression speed (improved cache efficiency) */
+FORCE_INLINE_TEMPLATE U32
+ZSTD_getMatchPrice(U32 const offset,
+                   U32 const matchLength,
+             const optState_t* const optPtr,
+                   int const optLevel)
 {
-    /* offset */
     U32 price;
-    BYTE const offCode = (BYTE)ZSTD_highbit32(offset+1);
+    U32 const offCode = ZSTD_highbit32(offset+1);
+    U32 const mlBase = matchLength - MINMATCH;
+    assert(matchLength >= MINMATCH);
 
-    if (optPtr->staticPrices)
-        return ZSTD_getLiteralPrice(optPtr, litLength, literals) + ZSTD_highbit32((U32)matchLength+1) + 16 + offCode;
+    if (optPtr->priceType == zop_predef)  /* fixed scheme, do not use statistics */
+        return WEIGHT(mlBase, optLevel) + ((16 + offCode) * BITCOST_MULTIPLIER);
 
-    price = offCode + optPtr->log2offCodeSum - ZSTD_highbit32(optPtr->offCodeFreq[offCode]+1);
-    if (!ultra && offCode >= 20) price += (offCode-19)*2;
+    /* dynamic statistics */
+    price = (offCode * BITCOST_MULTIPLIER) + (optPtr->offCodeSumBasePrice - WEIGHT(optPtr->offCodeFreq[offCode], optLevel));
+    if ((optLevel<2) /*static*/ && offCode >= 20)
+        price += (offCode-19)*2 * BITCOST_MULTIPLIER; /* handicap for long distance offsets, favor decompression speed */
 
     /* match Length */
-    {   const BYTE ML_deltaCode = 36;
-        const BYTE mlCode = (matchLength>127) ? (BYTE)ZSTD_highbit32(matchLength) + ML_deltaCode : ML_Code[matchLength];
-        price += ML_bits[mlCode] + optPtr->log2matchLengthSum - ZSTD_highbit32(optPtr->matchLengthFreq[mlCode]+1);
+    {   U32 const mlCode = ZSTD_MLcode(mlBase);
+        price += (ML_bits[mlCode] * BITCOST_MULTIPLIER) + (optPtr->matchLengthSumBasePrice - WEIGHT(optPtr->matchLengthFreq[mlCode], optLevel));
     }
 
-    return price + ZSTD_getLiteralPrice(optPtr, litLength, literals) + optPtr->factor;
-}
+    price += BITCOST_MULTIPLIER / 5;   /* heuristic : make matches a bit more costly to favor less sequences -> faster decompression speed */
 
+    DEBUGLOG(8, "ZSTD_getMatchPrice(ml:%u) = %u", matchLength, price);
+    return price;
+}
 
-static void ZSTD_updatePrice(optState_t* optPtr, U32 litLength, const BYTE* literals, U32 offset, U32 matchLength)
+/* ZSTD_updateStats() :
+ * assumption : literals + litLengtn <= iend */
+static void ZSTD_updateStats(optState_t* const optPtr,
+                             U32 litLength, const BYTE* literals,
+                             U32 offsetCode, U32 matchLength)
 {
-    U32 u;
-
     /* literals */
-    optPtr->litSum += litLength*ZSTD_LITFREQ_ADD;
-    for (u=0; u < litLength; u++)
-        optPtr->litFreq[literals[u]] += ZSTD_LITFREQ_ADD;
+    if (ZSTD_compressedLiterals(optPtr)) {
+        U32 u;
+        for (u=0; u < litLength; u++)
+            optPtr->litFreq[literals[u]] += ZSTD_LITFREQ_ADD;
+        optPtr->litSum += litLength*ZSTD_LITFREQ_ADD;
+    }
 
     /* literal Length */
-    {   const BYTE LL_deltaCode = 19;
-        const BYTE llCode = (litLength>63) ? (BYTE)ZSTD_highbit32(litLength) + LL_deltaCode : LL_Code[litLength];
+    {   U32 const llCode = ZSTD_LLcode(litLength);
         optPtr->litLengthFreq[llCode]++;
         optPtr->litLengthSum++;
     }
 
-    /* match offset */
-    {   BYTE const offCode = (BYTE)ZSTD_highbit32(offset+1);
-        optPtr->offCodeSum++;
+    /* match offset code (0-2=>repCode; 3+=>offset+2) */
+    {   U32 const offCode = ZSTD_highbit32(offsetCode+1);
+        assert(offCode <= MaxOff);
         optPtr->offCodeFreq[offCode]++;
+        optPtr->offCodeSum++;
     }
 
     /* match Length */
-    {   const BYTE ML_deltaCode = 36;
-        const BYTE mlCode = (matchLength>127) ? (BYTE)ZSTD_highbit32(matchLength) + ML_deltaCode : ML_Code[matchLength];
+    {   U32 const mlBase = matchLength - MINMATCH;
+        U32 const mlCode = ZSTD_MLcode(mlBase);
         optPtr->matchLengthFreq[mlCode]++;
         optPtr->matchLengthSum++;
     }
-
-    ZSTD_setLog2Prices(optPtr);
 }
 
 
-#define SET_PRICE(pos, mlen_, offset_, litlen_, price_)   \
-    {                                                 \
-        while (last_pos < pos)  { opt[last_pos+1].price = ZSTD_MAX_PRICE; last_pos++; } \
-        opt[pos].mlen = mlen_;                         \
-        opt[pos].off = offset_;                        \
-        opt[pos].litlen = litlen_;                     \
-        opt[pos].price = price_;                       \
-    }
-
-
-/* function safe only for comparisons */
-static U32 ZSTD_readMINMATCH(const void* memPtr, U32 length)
+/* ZSTD_readMINMATCH() :
+ * function safe only for comparisons
+ * assumption : memPtr must be at least 4 bytes before end of buffer */
+MEM_STATIC U32 ZSTD_readMINMATCH(const void* memPtr, U32 length)
 {
     switch (length)
     {
@@ -216,21 +338,24 @@ static U32 ZSTD_readMINMATCH(const void* memPtr, U32 length)
 
 /* Update hashTable3 up to ip (excluded)
    Assumption : always within prefix (i.e. not within extDict) */
-static
-U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_CCtx* zc, const BYTE* ip)
+static U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_matchState_t* ms,
+                                              U32* nextToUpdate3,
+                                              const BYTE* const ip)
 {
-    U32* const hashTable3  = zc->hashTable3;
-    U32 const hashLog3  = zc->hashLog3;
-    const BYTE* const base = zc->base;
-    U32 idx = zc->nextToUpdate3;
-    const U32 target = zc->nextToUpdate3 = (U32)(ip - base);
-    const size_t hash3 = ZSTD_hash3Ptr(ip, hashLog3);
+    U32* const hashTable3 = ms->hashTable3;
+    U32 const hashLog3 = ms->hashLog3;
+    const BYTE* const base = ms->window.base;
+    U32 idx = *nextToUpdate3;
+    U32 const target = (U32)(ip - base);
+    size_t const hash3 = ZSTD_hash3Ptr(ip, hashLog3);
+    assert(hashLog3 > 0);
 
     while(idx < target) {
         hashTable3[ZSTD_hash3Ptr(base+idx, hashLog3)] = idx;
         idx++;
     }
 
+    *nextToUpdate3 = target;
     return hashTable3[hash3];
 }
 
@@ -238,102 +363,326 @@ U32 ZSTD_insertAndFindFirstIndexHash3 (ZSTD_CCtx* zc, const BYTE* ip)
 /*-*************************************
 *  Binary Tree search
 ***************************************/
-static U32 ZSTD_insertBtAndGetAllMatches (
-                        ZSTD_CCtx* zc,
-                        const BYTE* const ip, const BYTE* const iLimit,
-                        U32 nbCompares, const U32 mls,
-                        U32 extDict, ZSTD_match_t* matches, const U32 minMatchLen)
+/** ZSTD_insertBt1() : add one or multiple positions to tree.
+ *  ip : assumed <= iend-8 .
+ * @return : nb of positions added */
+static U32 ZSTD_insertBt1(
+                ZSTD_matchState_t* ms,
+                const BYTE* const ip, const BYTE* const iend,
+                U32 const mls, const int extDict)
 {
-    const BYTE* const base = zc->base;
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32*   const hashTable = ms->hashTable;
+    U32    const hashLog = cParams->hashLog;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
+    U32*   const bt = ms->chainTable;
+    U32    const btLog  = cParams->chainLog - 1;
+    U32    const btMask = (1 << btLog) - 1;
+    U32 matchIndex = hashTable[h];
+    size_t commonLengthSmaller=0, commonLengthLarger=0;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const dictBase = ms->window.dictBase;
+    const U32 dictLimit = ms->window.dictLimit;
+    const BYTE* const dictEnd = dictBase + dictLimit;
+    const BYTE* const prefixStart = base + dictLimit;
+    const BYTE* match;
     const U32 current = (U32)(ip-base);
-    const U32 hashLog = zc->appliedParams.cParams.hashLog;
-    const size_t h  = ZSTD_hashPtr(ip, hashLog, mls);
-    U32* const hashTable = zc->hashTable;
+    const U32 btLow = btMask >= current ? 0 : current - btMask;
+    U32* smallerPtr = bt + 2*(current&btMask);
+    U32* largerPtr  = smallerPtr + 1;
+    U32 dummy32;   /* to be nullified at the end */
+    U32 const windowLow = ms->window.lowLimit;
+    U32 matchEndIdx = current+8+1;
+    size_t bestLength = 8;
+    U32 nbCompares = 1U << cParams->searchLog;
+#ifdef ZSTD_C_PREDICT
+    U32 predictedSmall = *(bt + 2*((current-1)&btMask) + 0);
+    U32 predictedLarge = *(bt + 2*((current-1)&btMask) + 1);
+    predictedSmall += (predictedSmall>0);
+    predictedLarge += (predictedLarge>0);
+#endif /* ZSTD_C_PREDICT */
+
+    DEBUGLOG(8, "ZSTD_insertBt1 (%u)", current);
+
+    assert(ip <= iend-8);   /* required for h calculation */
+    hashTable[h] = current;   /* Update Hash Table */
+
+    assert(windowLow > 0);
+    while (nbCompares-- && (matchIndex >= windowLow)) {
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        assert(matchIndex < current);
+
+#ifdef ZSTD_C_PREDICT   /* note : can create issues when hlog small <= 11 */
+        const U32* predictPtr = bt + 2*((matchIndex-1) & btMask);   /* written this way, as bt is a roll buffer */
+        if (matchIndex == predictedSmall) {
+            /* no need to check length, result known */
+            *smallerPtr = matchIndex;
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
+            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            predictedSmall = predictPtr[1] + (predictPtr[1]>0);
+            continue;
+        }
+        if (matchIndex == predictedLarge) {
+            *largerPtr = matchIndex;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+            predictedLarge = predictPtr[0] + (predictPtr[0]>0);
+            continue;
+        }
+#endif
+
+        if (!extDict || (matchIndex+matchLength >= dictLimit)) {
+            assert(matchIndex+matchLength >= dictLimit);   /* might be wrong if actually extDict */
+            match = base + matchIndex;
+            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iend);
+        } else {
+            match = dictBase + matchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iend, dictEnd, prefixStart);
+            if (matchIndex+matchLength >= dictLimit)
+                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+        }
+
+        if (matchLength > bestLength) {
+            bestLength = matchLength;
+            if (matchLength > matchEndIdx - matchIndex)
+                matchEndIdx = matchIndex + (U32)matchLength;
+        }
+
+        if (ip+matchLength == iend) {   /* equal : no way to know if inf or sup */
+            break;   /* drop , to guarantee consistency ; miss a bit of compression, but other solutions can corrupt tree */
+        }
+
+        if (match[matchLength] < ip[matchLength]) {  /* necessarily within buffer */
+            /* match is smaller than current */
+            *smallerPtr = matchIndex;             /* update smaller idx */
+            commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+            if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            smallerPtr = nextPtr+1;               /* new "candidate" => larger than match, which was smaller than target */
+            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous and closer to current */
+        } else {
+            /* match is larger than current */
+            *largerPtr = matchIndex;
+            commonLengthLarger = matchLength;
+            if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop searching */
+            largerPtr = nextPtr;
+            matchIndex = nextPtr[0];
+    }   }
+
+    *smallerPtr = *largerPtr = 0;
+    {   U32 positions = 0;
+        if (bestLength > 384) positions = MIN(192, (U32)(bestLength - 384));   /* speed optimization */
+        assert(matchEndIdx > current + 8);
+        return MAX(positions, matchEndIdx - (current + 8));
+    }
+}
+
+FORCE_INLINE_TEMPLATE
+void ZSTD_updateTree_internal(
+                ZSTD_matchState_t* ms,
+                const BYTE* const ip, const BYTE* const iend,
+                const U32 mls, const ZSTD_dictMode_e dictMode)
+{
+    const BYTE* const base = ms->window.base;
+    U32 const target = (U32)(ip - base);
+    U32 idx = ms->nextToUpdate;
+    DEBUGLOG(6, "ZSTD_updateTree_internal, from %u to %u  (dictMode:%u)",
+                idx, target, dictMode);
+
+    while(idx < target) {
+        U32 const forward = ZSTD_insertBt1(ms, base+idx, iend, mls, dictMode == ZSTD_extDict);
+        assert(idx < (U32)(idx + forward));
+        idx += forward;
+    }
+    assert((size_t)(ip - base) <= (size_t)(U32)(-1));
+    assert((size_t)(iend - base) <= (size_t)(U32)(-1));
+    ms->nextToUpdate = target;
+}
+
+void ZSTD_updateTree(ZSTD_matchState_t* ms, const BYTE* ip, const BYTE* iend) {
+    ZSTD_updateTree_internal(ms, ip, iend, ms->cParams.minMatch, ZSTD_noDict);
+}
+
+FORCE_INLINE_TEMPLATE
+U32 ZSTD_insertBtAndGetAllMatches (
+                    ZSTD_match_t* matches,   /* store result (found matches) in this table (presumed large enough) */
+                    ZSTD_matchState_t* ms,
+                    U32* nextToUpdate3,
+                    const BYTE* const ip, const BYTE* const iLimit, const ZSTD_dictMode_e dictMode,
+                    const U32 rep[ZSTD_REP_NUM],
+                    U32 const ll0,   /* tells if associated literal length is 0 or not. This value must be 0 or 1 */
+                    const U32 lengthToBeat,
+                    U32 const mls /* template */)
+{
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
+    const BYTE* const base = ms->window.base;
+    U32 const current = (U32)(ip-base);
+    U32 const hashLog = cParams->hashLog;
+    U32 const minMatch = (mls==3) ? 3 : 4;
+    U32* const hashTable = ms->hashTable;
+    size_t const h  = ZSTD_hashPtr(ip, hashLog, mls);
     U32 matchIndex  = hashTable[h];
-    U32* const bt   = zc->chainTable;
-    const U32 btLog = zc->appliedParams.cParams.chainLog - 1;
-    const U32 btMask= (1U << btLog) - 1;
+    U32* const bt   = ms->chainTable;
+    U32 const btLog = cParams->chainLog - 1;
+    U32 const btMask= (1U << btLog) - 1;
     size_t commonLengthSmaller=0, commonLengthLarger=0;
-    const BYTE* const dictBase = zc->dictBase;
-    const U32 dictLimit = zc->dictLimit;
+    const BYTE* const dictBase = ms->window.dictBase;
+    U32 const dictLimit = ms->window.dictLimit;
     const BYTE* const dictEnd = dictBase + dictLimit;
     const BYTE* const prefixStart = base + dictLimit;
-    const U32 btLow = btMask >= current ? 0 : current - btMask;
-    const U32 windowLow = zc->lowLimit;
+    U32 const btLow = (btMask >= current) ? 0 : current - btMask;
+    U32 const windowLow = ZSTD_getLowestMatchIndex(ms, current, cParams->windowLog);
+    U32 const matchLow = windowLow ? windowLow : 1;
     U32* smallerPtr = bt + 2*(current&btMask);
     U32* largerPtr  = bt + 2*(current&btMask) + 1;
-    U32 matchEndIdx = current+8;
+    U32 matchEndIdx = current+8+1;   /* farthest referenced position of any match => detects repetitive patterns */
     U32 dummy32;   /* to be nullified at the end */
     U32 mnum = 0;
-
-    const U32 minMatch = (mls == 3) ? 3 : 4;
-    size_t bestLength = minMatchLen-1;
-
-    if (minMatch == 3) { /* HC3 match finder */
-        U32 const matchIndex3 = ZSTD_insertAndFindFirstIndexHash3 (zc, ip);
-        if (matchIndex3>windowLow && (current - matchIndex3 < (1<<18))) {
-            const BYTE* match;
-            size_t currentMl=0;
-            if ((!extDict) || matchIndex3 >= dictLimit) {
-                match = base + matchIndex3;
-                if (match[bestLength] == ip[bestLength]) currentMl = ZSTD_count(ip, match, iLimit);
+    U32 nbCompares = 1U << cParams->searchLog;
+
+    const ZSTD_matchState_t* dms    = dictMode == ZSTD_dictMatchState ? ms->dictMatchState : NULL;
+    const ZSTD_compressionParameters* const dmsCParams =
+                                      dictMode == ZSTD_dictMatchState ? &dms->cParams : NULL;
+    const BYTE* const dmsBase       = dictMode == ZSTD_dictMatchState ? dms->window.base : NULL;
+    const BYTE* const dmsEnd        = dictMode == ZSTD_dictMatchState ? dms->window.nextSrc : NULL;
+    U32         const dmsHighLimit  = dictMode == ZSTD_dictMatchState ? (U32)(dmsEnd - dmsBase) : 0;
+    U32         const dmsLowLimit   = dictMode == ZSTD_dictMatchState ? dms->window.lowLimit : 0;
+    U32         const dmsIndexDelta = dictMode == ZSTD_dictMatchState ? windowLow - dmsHighLimit : 0;
+    U32         const dmsHashLog    = dictMode == ZSTD_dictMatchState ? dmsCParams->hashLog : hashLog;
+    U32         const dmsBtLog      = dictMode == ZSTD_dictMatchState ? dmsCParams->chainLog - 1 : btLog;
+    U32         const dmsBtMask     = dictMode == ZSTD_dictMatchState ? (1U << dmsBtLog) - 1 : 0;
+    U32         const dmsBtLow      = dictMode == ZSTD_dictMatchState && dmsBtMask < dmsHighLimit - dmsLowLimit ? dmsHighLimit - dmsBtMask : dmsLowLimit;
+
+    size_t bestLength = lengthToBeat-1;
+    DEBUGLOG(8, "ZSTD_insertBtAndGetAllMatches: current=%u", current);
+
+    /* check repCode */
+    assert(ll0 <= 1);   /* necessarily 1 or 0 */
+    {   U32 const lastR = ZSTD_REP_NUM + ll0;
+        U32 repCode;
+        for (repCode = ll0; repCode < lastR; repCode++) {
+            U32 const repOffset = (repCode==ZSTD_REP_NUM) ? (rep[0] - 1) : rep[repCode];
+            U32 const repIndex = current - repOffset;
+            U32 repLen = 0;
+            assert(current >= dictLimit);
+            if (repOffset-1 /* intentional overflow, discards 0 and -1 */ < current-dictLimit) {  /* equivalent to `current > repIndex >= dictLimit` */
+                /* We must validate the repcode offset because when we're using a dictionary the
+                 * valid offset range shrinks when the dictionary goes out of bounds.
+                 */
+                if ((repIndex >= windowLow) & (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(ip - repOffset, minMatch))) {
+                    repLen = (U32)ZSTD_count(ip+minMatch, ip+minMatch-repOffset, iLimit) + minMatch;
+                }
+            } else {  /* repIndex < dictLimit || repIndex >= current */
+                const BYTE* const repMatch = dictMode == ZSTD_dictMatchState ?
+                                             dmsBase + repIndex - dmsIndexDelta :
+                                             dictBase + repIndex;
+                assert(current >= windowLow);
+                if ( dictMode == ZSTD_extDict
+                  && ( ((repOffset-1) /*intentional overflow*/ < current - windowLow)  /* equivalent to `current > repIndex >= windowLow` */
+                     & (((U32)((dictLimit-1) - repIndex) >= 3) ) /* intentional overflow : do not test positions overlapping 2 memory segments */)
+                  && (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
+                    repLen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iLimit, dictEnd, prefixStart) + minMatch;
+                }
+                if (dictMode == ZSTD_dictMatchState
+                  && ( ((repOffset-1) /*intentional overflow*/ < current - (dmsLowLimit + dmsIndexDelta))  /* equivalent to `current > repIndex >= dmsLowLimit` */
+                     & ((U32)((dictLimit-1) - repIndex) >= 3) ) /* intentional overflow : do not test positions overlapping 2 memory segments */
+                  && (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
+                    repLen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iLimit, dmsEnd, prefixStart) + minMatch;
+            }   }
+            /* save longer solution */
+            if (repLen > bestLength) {
+                DEBUGLOG(8, "found repCode %u (ll0:%u, offset:%u) of length %u",
+                            repCode, ll0, repOffset, repLen);
+                bestLength = repLen;
+                matches[mnum].off = repCode - ll0;
+                matches[mnum].len = (U32)repLen;
+                mnum++;
+                if ( (repLen > sufficient_len)
+                   | (ip+repLen == iLimit) ) {  /* best possible */
+                    return mnum;
+    }   }   }   }
+
+    /* HC3 match finder */
+    if ((mls == 3) /*static*/ && (bestLength < mls)) {
+        U32 const matchIndex3 = ZSTD_insertAndFindFirstIndexHash3(ms, nextToUpdate3, ip);
+        if ((matchIndex3 >= matchLow)
+          & (current - matchIndex3 < (1<<18)) /*heuristic : longer distance likely too expensive*/ ) {
+            size_t mlen;
+            if ((dictMode == ZSTD_noDict) /*static*/ || (dictMode == ZSTD_dictMatchState) /*static*/ || (matchIndex3 >= dictLimit)) {
+                const BYTE* const match = base + matchIndex3;
+                mlen = ZSTD_count(ip, match, iLimit);
             } else {
-                match = dictBase + matchIndex3;
-                if (ZSTD_readMINMATCH(match, MINMATCH) == ZSTD_readMINMATCH(ip, MINMATCH))    /* assumption : matchIndex3 <= dictLimit-4 (by table construction) */
-                    currentMl = ZSTD_count_2segments(ip+MINMATCH, match+MINMATCH, iLimit, dictEnd, prefixStart) + MINMATCH;
+                const BYTE* const match = dictBase + matchIndex3;
+                mlen = ZSTD_count_2segments(ip, match, iLimit, dictEnd, prefixStart);
             }
 
             /* save best solution */
-            if (currentMl > bestLength) {
-                bestLength = currentMl;
-                matches[mnum].off = ZSTD_REP_MOVE_OPT + current - matchIndex3;
-                matches[mnum].len = (U32)currentMl;
-                mnum++;
-                if (currentMl > ZSTD_OPT_NUM) goto update;
-                if (ip+currentMl == iLimit) goto update; /* best possible, and avoid read overflow*/
-            }
-        }
+            if (mlen >= mls /* == 3 > bestLength */) {
+                DEBUGLOG(8, "found small match with hlog3, of length %u",
+                            (U32)mlen);
+                bestLength = mlen;
+                assert(current > matchIndex3);
+                assert(mnum==0);  /* no prior solution */
+                matches[0].off = (current - matchIndex3) + ZSTD_REP_MOVE;
+                matches[0].len = (U32)mlen;
+                mnum = 1;
+                if ( (mlen > sufficient_len) |
+                     (ip+mlen == iLimit) ) {  /* best possible length */
+                    ms->nextToUpdate = current+1;  /* skip insertion */
+                    return 1;
+        }   }   }
+        /* no dictMatchState lookup: dicts don't have a populated HC3 table */
     }
 
     hashTable[h] = current;   /* Update Hash Table */
 
-    while (nbCompares-- && (matchIndex > windowLow)) {
-        U32* nextPtr = bt + 2*(matchIndex & btMask);
-        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+    while (nbCompares-- && (matchIndex >= matchLow)) {
+        U32* const nextPtr = bt + 2*(matchIndex & btMask);
         const BYTE* match;
+        size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+        assert(current > matchIndex);
 
-        if ((!extDict) || (matchIndex+matchLength >= dictLimit)) {
+        if ((dictMode == ZSTD_noDict) || (dictMode == ZSTD_dictMatchState) || (matchIndex+matchLength >= dictLimit)) {
+            assert(matchIndex+matchLength >= dictLimit);  /* ensure the condition is correct when !extDict */
             match = base + matchIndex;
-            if (match[matchLength] == ip[matchLength]) {
-                matchLength += ZSTD_count(ip+matchLength+1, match+matchLength+1, iLimit) +1;
-            }
+            if (matchIndex >= dictLimit) assert(memcmp(match, ip, matchLength) == 0);  /* ensure early section of match is equal as expected */
+            matchLength += ZSTD_count(ip+matchLength, match+matchLength, iLimit);
         } else {
             match = dictBase + matchIndex;
+            assert(memcmp(match, ip, matchLength) == 0);  /* ensure early section of match is equal as expected */
             matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iLimit, dictEnd, prefixStart);
             if (matchIndex+matchLength >= dictLimit)
-                match = base + matchIndex;   /* to prepare for next usage of match[matchLength] */
+                match = base + matchIndex;   /* prepare for match[matchLength] read */
         }
 
         if (matchLength > bestLength) {
-            if (matchLength > matchEndIdx - matchIndex) matchEndIdx = matchIndex + (U32)matchLength;
+            DEBUGLOG(8, "found match of length %u at distance %u (offCode=%u)",
+                    (U32)matchLength, current - matchIndex, current - matchIndex + ZSTD_REP_MOVE);
+            assert(matchEndIdx > matchIndex);
+            if (matchLength > matchEndIdx - matchIndex)
+                matchEndIdx = matchIndex + (U32)matchLength;
             bestLength = matchLength;
-            matches[mnum].off = ZSTD_REP_MOVE_OPT + current - matchIndex;
+            matches[mnum].off = (current - matchIndex) + ZSTD_REP_MOVE;
             matches[mnum].len = (U32)matchLength;
             mnum++;
-            if (matchLength > ZSTD_OPT_NUM) break;
-            if (ip+matchLength == iLimit)   /* equal : no way to know if inf or sup */
-                break;   /* drop, to guarantee consistency (miss a little bit of compression) */
+            if ( (matchLength > ZSTD_OPT_NUM)
+               | (ip+matchLength == iLimit) /* equal : no way to know if inf or sup */) {
+                if (dictMode == ZSTD_dictMatchState) nbCompares = 0; /* break should also skip searching dms */
+                break; /* drop, to preserve bt consistency (miss a little bit of compression) */
+            }
         }
 
         if (match[matchLength] < ip[matchLength]) {
-            /* match is smaller than current */
+            /* match smaller than current */
             *smallerPtr = matchIndex;             /* update smaller idx */
             commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
             if (matchIndex <= btLow) { smallerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
-            smallerPtr = nextPtr+1;               /* new "smaller" => larger of match */
-            matchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            smallerPtr = nextPtr+1;               /* new candidate => larger than match, which was smaller than current */
+            matchIndex = nextPtr[1];              /* new matchIndex, larger than previous, closer to current */
         } else {
-            /* match is larger than current */
             *largerPtr = matchIndex;
             commonLengthLarger = matchLength;
             if (matchIndex <= btLow) { largerPtr=&dummy32; break; }   /* beyond tree size, stop the search */
@@ -343,607 +692,509 @@ static U32 ZSTD_insertBtAndGetAllMatches (
 
     *smallerPtr = *largerPtr = 0;
 
-update:
-    zc->nextToUpdate = (matchEndIdx > current + 8) ? matchEndIdx - 8 : current+1;
-    return mnum;
-}
+    if (dictMode == ZSTD_dictMatchState && nbCompares) {
+        size_t const dmsH = ZSTD_hashPtr(ip, dmsHashLog, mls);
+        U32 dictMatchIndex = dms->hashTable[dmsH];
+        const U32* const dmsBt = dms->chainTable;
+        commonLengthSmaller = commonLengthLarger = 0;
+        while (nbCompares-- && (dictMatchIndex > dmsLowLimit)) {
+            const U32* const nextPtr = dmsBt + 2*(dictMatchIndex & dmsBtMask);
+            size_t matchLength = MIN(commonLengthSmaller, commonLengthLarger);   /* guaranteed minimum nb of common bytes */
+            const BYTE* match = dmsBase + dictMatchIndex;
+            matchLength += ZSTD_count_2segments(ip+matchLength, match+matchLength, iLimit, dmsEnd, prefixStart);
+            if (dictMatchIndex+matchLength >= dmsHighLimit)
+                match = base + dictMatchIndex + dmsIndexDelta;   /* to prepare for next usage of match[matchLength] */
+
+            if (matchLength > bestLength) {
+                matchIndex = dictMatchIndex + dmsIndexDelta;
+                DEBUGLOG(8, "found dms match of length %u at distance %u (offCode=%u)",
+                        (U32)matchLength, current - matchIndex, current - matchIndex + ZSTD_REP_MOVE);
+                if (matchLength > matchEndIdx - matchIndex)
+                    matchEndIdx = matchIndex + (U32)matchLength;
+                bestLength = matchLength;
+                matches[mnum].off = (current - matchIndex) + ZSTD_REP_MOVE;
+                matches[mnum].len = (U32)matchLength;
+                mnum++;
+                if ( (matchLength > ZSTD_OPT_NUM)
+                   | (ip+matchLength == iLimit) /* equal : no way to know if inf or sup */) {
+                    break;   /* drop, to guarantee consistency (miss a little bit of compression) */
+                }
+            }
 
+            if (dictMatchIndex <= dmsBtLow) { break; }   /* beyond tree size, stop the search */
+            if (match[matchLength] < ip[matchLength]) {
+                commonLengthSmaller = matchLength;    /* all smaller will now have at least this guaranteed common length */
+                dictMatchIndex = nextPtr[1];              /* new matchIndex larger than previous (closer to current) */
+            } else {
+                /* match is larger than current */
+                commonLengthLarger = matchLength;
+                dictMatchIndex = nextPtr[0];
+            }
+        }
+    }
 
-/** Tree updater, providing best match */
-static U32 ZSTD_BtGetAllMatches (
-                        ZSTD_CCtx* zc,
-                        const BYTE* const ip, const BYTE* const iLimit,
-                        const U32 maxNbAttempts, const U32 mls, ZSTD_match_t* matches, const U32 minMatchLen)
-{
-    if (ip < zc->base + zc->nextToUpdate) return 0;   /* skipped area */
-    ZSTD_updateTree(zc, ip, iLimit, maxNbAttempts, mls);
-    return ZSTD_insertBtAndGetAllMatches(zc, ip, iLimit, maxNbAttempts, mls, 0, matches, minMatchLen);
+    assert(matchEndIdx > current+8);
+    ms->nextToUpdate = matchEndIdx - 8;  /* skip repetitive patterns */
+    return mnum;
 }
 
 
-static U32 ZSTD_BtGetAllMatches_selectMLS (
-                        ZSTD_CCtx* zc,   /* Index table will be updated */
-                        const BYTE* ip, const BYTE* const iHighLimit,
-                        const U32 maxNbAttempts, const U32 matchLengthSearch, ZSTD_match_t* matches, const U32 minMatchLen)
+FORCE_INLINE_TEMPLATE U32 ZSTD_BtGetAllMatches (
+                        ZSTD_match_t* matches,   /* store result (match found, increasing size) in this table */
+                        ZSTD_matchState_t* ms,
+                        U32* nextToUpdate3,
+                        const BYTE* ip, const BYTE* const iHighLimit, const ZSTD_dictMode_e dictMode,
+                        const U32 rep[ZSTD_REP_NUM],
+                        U32 const ll0,
+                        U32 const lengthToBeat)
 {
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
+    U32 const matchLengthSearch = cParams->minMatch;
+    DEBUGLOG(8, "ZSTD_BtGetAllMatches");
+    if (ip < ms->window.base + ms->nextToUpdate) return 0;   /* skipped area */
+    ZSTD_updateTree_internal(ms, ip, iHighLimit, matchLengthSearch, dictMode);
     switch(matchLengthSearch)
     {
-    case 3 : return ZSTD_BtGetAllMatches(zc, ip, iHighLimit, maxNbAttempts, 3, matches, minMatchLen);
+    case 3 : return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, 3);
     default :
-    case 4 : return ZSTD_BtGetAllMatches(zc, ip, iHighLimit, maxNbAttempts, 4, matches, minMatchLen);
-    case 5 : return ZSTD_BtGetAllMatches(zc, ip, iHighLimit, maxNbAttempts, 5, matches, minMatchLen);
+    case 4 : return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, 4);
+    case 5 : return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, 5);
     case 7 :
-    case 6 : return ZSTD_BtGetAllMatches(zc, ip, iHighLimit, maxNbAttempts, 6, matches, minMatchLen);
+    case 6 : return ZSTD_insertBtAndGetAllMatches(matches, ms, nextToUpdate3, ip, iHighLimit, dictMode, rep, ll0, lengthToBeat, 6);
     }
 }
 
-/** Tree updater, providing best match */
-static U32 ZSTD_BtGetAllMatches_extDict (
-                        ZSTD_CCtx* zc,
-                        const BYTE* const ip, const BYTE* const iLimit,
-                        const U32 maxNbAttempts, const U32 mls, ZSTD_match_t* matches, const U32 minMatchLen)
+
+/*-*******************************
+*  Optimal parser
+*********************************/
+
+
+static U32 ZSTD_totalLen(ZSTD_optimal_t sol)
 {
-    if (ip < zc->base + zc->nextToUpdate) return 0;   /* skipped area */
-    ZSTD_updateTree_extDict(zc, ip, iLimit, maxNbAttempts, mls);
-    return ZSTD_insertBtAndGetAllMatches(zc, ip, iLimit, maxNbAttempts, mls, 1, matches, minMatchLen);
+    return sol.litlen + sol.mlen;
 }
 
+#if 0 /* debug */
 
-static U32 ZSTD_BtGetAllMatches_selectMLS_extDict (
-                        ZSTD_CCtx* zc,   /* Index table will be updated */
-                        const BYTE* ip, const BYTE* const iHighLimit,
-                        const U32 maxNbAttempts, const U32 matchLengthSearch, ZSTD_match_t* matches, const U32 minMatchLen)
+static void
+listStats(const U32* table, int lastEltID)
 {
-    switch(matchLengthSearch)
-    {
-    case 3 : return ZSTD_BtGetAllMatches_extDict(zc, ip, iHighLimit, maxNbAttempts, 3, matches, minMatchLen);
-    default :
-    case 4 : return ZSTD_BtGetAllMatches_extDict(zc, ip, iHighLimit, maxNbAttempts, 4, matches, minMatchLen);
-    case 5 : return ZSTD_BtGetAllMatches_extDict(zc, ip, iHighLimit, maxNbAttempts, 5, matches, minMatchLen);
-    case 7 :
-    case 6 : return ZSTD_BtGetAllMatches_extDict(zc, ip, iHighLimit, maxNbAttempts, 6, matches, minMatchLen);
+    int const nbElts = lastEltID + 1;
+    int enb;
+    for (enb=0; enb < nbElts; enb++) {
+        (void)table;
+        /* RAWLOG(2, "%3i:%3i,  ", enb, table[enb]); */
+        RAWLOG(2, "%4i,", table[enb]);
     }
+    RAWLOG(2, " \n");
 }
 
+#endif
 
-/*-*******************************
-*  Optimal parser
-*********************************/
-FORCE_INLINE_TEMPLATE
-size_t ZSTD_compressBlock_opt_generic(ZSTD_CCtx* ctx,
-                                      const void* src, size_t srcSize, const int ultra)
+FORCE_INLINE_TEMPLATE size_t
+ZSTD_compressBlock_opt_generic(ZSTD_matchState_t* ms,
+                               seqStore_t* seqStore,
+                               U32 rep[ZSTD_REP_NUM],
+                         const void* src, size_t srcSize,
+                         const int optLevel,
+                         const ZSTD_dictMode_e dictMode)
 {
-    seqStore_t* seqStorePtr = &(ctx->seqStore);
-    optState_t* optStatePtr = &(ctx->optState);
+    optState_t* const optStatePtr = &ms->opt;
     const BYTE* const istart = (const BYTE*)src;
     const BYTE* ip = istart;
     const BYTE* anchor = istart;
     const BYTE* const iend = istart + srcSize;
     const BYTE* const ilimit = iend - 8;
-    const BYTE* const base = ctx->base;
-    const BYTE* const prefixStart = base + ctx->dictLimit;
+    const BYTE* const base = ms->window.base;
+    const BYTE* const prefixStart = base + ms->window.dictLimit;
+    const ZSTD_compressionParameters* const cParams = &ms->cParams;
 
-    const U32 maxSearches = 1U << ctx->appliedParams.cParams.searchLog;
-    const U32 sufficient_len = ctx->appliedParams.cParams.targetLength;
-    const U32 mls = ctx->appliedParams.cParams.searchLength;
-    const U32 minMatch = (ctx->appliedParams.cParams.searchLength == 3) ? 3 : 4;
+    U32 const sufficient_len = MIN(cParams->targetLength, ZSTD_OPT_NUM -1);
+    U32 const minMatch = (cParams->minMatch == 3) ? 3 : 4;
+    U32 nextToUpdate3 = ms->nextToUpdate;
 
-    ZSTD_optimal_t* opt = optStatePtr->priceTable;
-    ZSTD_match_t* matches = optStatePtr->matchTable;
-    const BYTE* inr;
-    U32 offset, rep[ZSTD_REP_NUM];
+    ZSTD_optimal_t* const opt = optStatePtr->priceTable;
+    ZSTD_match_t* const matches = optStatePtr->matchTable;
+    ZSTD_optimal_t lastSequence;
 
     /* init */
-    ctx->nextToUpdate3 = ctx->nextToUpdate;
-    ZSTD_rescaleFreqs(optStatePtr, (const BYTE*)src, srcSize);
+    DEBUGLOG(5, "ZSTD_compressBlock_opt_generic: current=%u, prefix=%u, nextToUpdate=%u",
+                (U32)(ip - base), ms->window.dictLimit, ms->nextToUpdate);
+    assert(optLevel <= 2);
+    ZSTD_rescaleFreqs(optStatePtr, (const BYTE*)src, srcSize, optLevel);
     ip += (ip==prefixStart);
-    { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) rep[i]=seqStorePtr->rep[i]; }
 
     /* Match Loop */
     while (ip < ilimit) {
-        U32 cur, match_num, last_pos, litlen, price;
-        U32 u, mlen, best_mlen, best_off, litLength;
-        memset(opt, 0, sizeof(ZSTD_optimal_t));
-        last_pos = 0;
-        litlen = (U32)(ip - anchor);
-
-        /* check repCode */
-        {   U32 i, last_i = ZSTD_REP_CHECK + (ip==anchor);
-            for (i=(ip == anchor); i<last_i; i++) {
-                const S32 repCur = (i==ZSTD_REP_MOVE_OPT) ? (rep[0] - 1) : rep[i];
-                if ( (repCur > 0) && (repCur < (S32)(ip-prefixStart))
-                    && (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(ip - repCur, minMatch))) {
-                    mlen = (U32)ZSTD_count(ip+minMatch, ip+minMatch-repCur, iend) + minMatch;
-                    if (mlen > sufficient_len || mlen >= ZSTD_OPT_NUM) {
-                        best_mlen = mlen; best_off = i; cur = 0; last_pos = 1;
-                        goto _storeSequence;
-                    }
-                    best_off = i - (ip == anchor);
-                    do {
-                        price = ZSTD_getPrice(optStatePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
-                        if (mlen > last_pos || price < opt[mlen].price)
-                            SET_PRICE(mlen, mlen, i, litlen, price);   /* note : macro modifies last_pos */
-                        mlen--;
-                    } while (mlen >= minMatch);
-        }   }   }
-
-        match_num = ZSTD_BtGetAllMatches_selectMLS(ctx, ip, iend, maxSearches, mls, matches, minMatch);
-
-        if (!last_pos && !match_num) { ip++; continue; }
-
-        if (match_num && (matches[match_num-1].len > sufficient_len || matches[match_num-1].len >= ZSTD_OPT_NUM)) {
-            best_mlen = matches[match_num-1].len;
-            best_off = matches[match_num-1].off;
-            cur = 0;
-            last_pos = 1;
-            goto _storeSequence;
+        U32 cur, last_pos = 0;
+
+        /* find first match */
+        {   U32 const litlen = (U32)(ip - anchor);
+            U32 const ll0 = !litlen;
+            U32 const nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, ip, iend, dictMode, rep, ll0, minMatch);
+            if (!nbMatches) { ip++; continue; }
+
+            /* initialize opt[0] */
+            { U32 i ; for (i=0; i<ZSTD_REP_NUM; i++) opt[0].rep[i] = rep[i]; }
+            opt[0].mlen = 0;  /* means is_a_literal */
+            opt[0].litlen = litlen;
+            /* We don't need to include the actual price of the literals because
+             * it is static for the duration of the forward pass, and is included
+             * in every price. We include the literal length to avoid negative
+             * prices when we subtract the previous literal length.
+             */
+            opt[0].price = ZSTD_litLengthPrice(litlen, optStatePtr, optLevel);
+
+            /* large match -> immediate encoding */
+            {   U32 const maxML = matches[nbMatches-1].len;
+                U32 const maxOffset = matches[nbMatches-1].off;
+                DEBUGLOG(6, "found %u matches of maxLength=%u and maxOffCode=%u at cPos=%u => start new series",
+                            nbMatches, maxML, maxOffset, (U32)(ip-prefixStart));
+
+                if (maxML > sufficient_len) {
+                    lastSequence.litlen = litlen;
+                    lastSequence.mlen = maxML;
+                    lastSequence.off = maxOffset;
+                    DEBUGLOG(6, "large match (%u>%u), immediate encoding",
+                                maxML, sufficient_len);
+                    cur = 0;
+                    last_pos = ZSTD_totalLen(lastSequence);
+                    goto _shortestPath;
+            }   }
+
+            /* set prices for first matches starting position == 0 */
+            {   U32 const literalsPrice = opt[0].price + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
+                U32 pos;
+                U32 matchNb;
+                for (pos = 1; pos < minMatch; pos++) {
+                    opt[pos].price = ZSTD_MAX_PRICE;   /* mlen, litlen and price will be fixed during forward scanning */
+                }
+                for (matchNb = 0; matchNb < nbMatches; matchNb++) {
+                    U32 const offset = matches[matchNb].off;
+                    U32 const end = matches[matchNb].len;
+                    for ( ; pos <= end ; pos++ ) {
+                        U32 const matchPrice = ZSTD_getMatchPrice(offset, pos, optStatePtr, optLevel);
+                        U32 const sequencePrice = literalsPrice + matchPrice;
+                        DEBUGLOG(7, "rPos:%u => set initial price : %.2f",
+                                    pos, ZSTD_fCost(sequencePrice));
+                        opt[pos].mlen = pos;
+                        opt[pos].off = offset;
+                        opt[pos].litlen = litlen;
+                        opt[pos].price = sequencePrice;
+                }   }
+                last_pos = pos-1;
+            }
         }
 
-        /* set prices using matches at position = 0 */
-        best_mlen = (last_pos) ? last_pos : minMatch;
-        for (u = 0; u < match_num; u++) {
-            mlen = (u>0) ? matches[u-1].len+1 : best_mlen;
-            best_mlen = matches[u].len;
-            while (mlen <= best_mlen) {
-                price = ZSTD_getPrice(optStatePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
-                if (mlen > last_pos || price < opt[mlen].price)
-                    SET_PRICE(mlen, mlen, matches[u].off, litlen, price);   /* note : macro modifies last_pos */
-                mlen++;
-        }   }
-
-        if (last_pos < minMatch) { ip++; continue; }
-
-        /* initialize opt[0] */
-        { U32 i ; for (i=0; i<ZSTD_REP_NUM; i++) opt[0].rep[i] = rep[i]; }
-        opt[0].mlen = 1;
-        opt[0].litlen = litlen;
-
-         /* check further positions */
+        /* check further positions */
         for (cur = 1; cur <= last_pos; cur++) {
-           inr = ip + cur;
-
-           if (opt[cur-1].mlen == 1) {
-                litlen = opt[cur-1].litlen + 1;
-                if (cur > litlen) {
-                    price = opt[cur - litlen].price + ZSTD_getLiteralPrice(optStatePtr, litlen, inr-litlen);
-                } else
-                    price = ZSTD_getLiteralPrice(optStatePtr, litlen, anchor);
-           } else {
-                litlen = 1;
-                price = opt[cur - 1].price + ZSTD_getLiteralPrice(optStatePtr, litlen, inr-1);
-           }
-
-           if (cur > last_pos || price <= opt[cur].price)
-                SET_PRICE(cur, 1, 0, litlen, price);
-
-           if (cur == last_pos) break;
-
-           if (inr > ilimit)  /* last match must start at a minimum distance of 8 from oend */
-               continue;
-
-           mlen = opt[cur].mlen;
-           if (opt[cur].off > ZSTD_REP_MOVE_OPT) {
-                opt[cur].rep[2] = opt[cur-mlen].rep[1];
-                opt[cur].rep[1] = opt[cur-mlen].rep[0];
-                opt[cur].rep[0] = opt[cur].off - ZSTD_REP_MOVE_OPT;
-           } else {
-                opt[cur].rep[2] = (opt[cur].off > 1) ? opt[cur-mlen].rep[1] : opt[cur-mlen].rep[2];
-                opt[cur].rep[1] = (opt[cur].off > 0) ? opt[cur-mlen].rep[0] : opt[cur-mlen].rep[1];
-                opt[cur].rep[0] = ((opt[cur].off==ZSTD_REP_MOVE_OPT) && (mlen != 1)) ? (opt[cur-mlen].rep[0] - 1) : (opt[cur-mlen].rep[opt[cur].off]);
-           }
-
-            best_mlen = minMatch;
-            {   U32 i, last_i = ZSTD_REP_CHECK + (mlen != 1);
-                for (i=(opt[cur].mlen != 1); i<last_i; i++) {  /* check rep */
-                    const S32 repCur = (i==ZSTD_REP_MOVE_OPT) ? (opt[cur].rep[0] - 1) : opt[cur].rep[i];
-                    if ( (repCur > 0) && (repCur < (S32)(inr-prefixStart))
-                       && (ZSTD_readMINMATCH(inr, minMatch) == ZSTD_readMINMATCH(inr - repCur, minMatch))) {
-                       mlen = (U32)ZSTD_count(inr+minMatch, inr+minMatch - repCur, iend) + minMatch;
-
-                       if (mlen > sufficient_len || cur + mlen >= ZSTD_OPT_NUM) {
-                            best_mlen = mlen; best_off = i; last_pos = cur + 1;
-                            goto _storeSequence;
-                       }
-
-                       best_off = i - (opt[cur].mlen != 1);
-                       if (mlen > best_mlen) best_mlen = mlen;
-
-                       do {
-                           if (opt[cur].mlen == 1) {
-                                litlen = opt[cur].litlen;
-                                if (cur > litlen) {
-                                    price = opt[cur - litlen].price + ZSTD_getPrice(optStatePtr, litlen, inr-litlen, best_off, mlen - MINMATCH, ultra);
-                                } else
-                                    price = ZSTD_getPrice(optStatePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
-                            } else {
-                                litlen = 0;
-                                price = opt[cur].price + ZSTD_getPrice(optStatePtr, 0, NULL, best_off, mlen - MINMATCH, ultra);
-                            }
-
-                            if (cur + mlen > last_pos || price <= opt[cur + mlen].price)
-                                SET_PRICE(cur + mlen, mlen, i, litlen, price);
-                            mlen--;
-                        } while (mlen >= minMatch);
-            }   }   }
-
-            match_num = ZSTD_BtGetAllMatches_selectMLS(ctx, inr, iend, maxSearches, mls, matches, best_mlen);
+            const BYTE* const inr = ip + cur;
+            assert(cur < ZSTD_OPT_NUM);
+            DEBUGLOG(7, "cPos:%zi==rPos:%u", inr-istart, cur)
+
+            /* Fix current position with one literal if cheaper */
+            {   U32 const litlen = (opt[cur-1].mlen == 0) ? opt[cur-1].litlen + 1 : 1;
+                int const price = opt[cur-1].price
+                                + ZSTD_rawLiteralsCost(ip+cur-1, 1, optStatePtr, optLevel)
+                                + ZSTD_litLengthPrice(litlen, optStatePtr, optLevel)
+                                - ZSTD_litLengthPrice(litlen-1, optStatePtr, optLevel);
+                assert(price < 1000000000); /* overflow check */
+                if (price <= opt[cur].price) {
+                    DEBUGLOG(7, "cPos:%zi==rPos:%u : better price (%.2f<=%.2f) using literal (ll==%u) (hist:%u,%u,%u)",
+                                inr-istart, cur, ZSTD_fCost(price), ZSTD_fCost(opt[cur].price), litlen,
+                                opt[cur-1].rep[0], opt[cur-1].rep[1], opt[cur-1].rep[2]);
+                    opt[cur].mlen = 0;
+                    opt[cur].off = 0;
+                    opt[cur].litlen = litlen;
+                    opt[cur].price = price;
+                } else {
+                    DEBUGLOG(7, "cPos:%zi==rPos:%u : literal would cost more (%.2f>%.2f) (hist:%u,%u,%u)",
+                                inr-istart, cur, ZSTD_fCost(price), ZSTD_fCost(opt[cur].price),
+                                opt[cur].rep[0], opt[cur].rep[1], opt[cur].rep[2]);
+                }
+            }
 
-            if (match_num > 0 && (matches[match_num-1].len > sufficient_len || cur + matches[match_num-1].len >= ZSTD_OPT_NUM)) {
-                best_mlen = matches[match_num-1].len;
-                best_off = matches[match_num-1].off;
-                last_pos = cur + 1;
-                goto _storeSequence;
+            /* Set the repcodes of the current position. We must do it here
+             * because we rely on the repcodes of the 2nd to last sequence being
+             * correct to set the next chunks repcodes during the backward
+             * traversal.
+             */
+            ZSTD_STATIC_ASSERT(sizeof(opt[cur].rep) == sizeof(repcodes_t));
+            assert(cur >= opt[cur].mlen);
+            if (opt[cur].mlen != 0) {
+                U32 const prev = cur - opt[cur].mlen;
+                repcodes_t newReps = ZSTD_updateRep(opt[prev].rep, opt[cur].off, opt[cur].litlen==0);
+                memcpy(opt[cur].rep, &newReps, sizeof(repcodes_t));
+            } else {
+                memcpy(opt[cur].rep, opt[cur - 1].rep, sizeof(repcodes_t));
             }
 
-            /* set prices using matches at position = cur */
-            for (u = 0; u < match_num; u++) {
-                mlen = (u>0) ? matches[u-1].len+1 : best_mlen;
-                best_mlen = matches[u].len;
-
-                while (mlen <= best_mlen) {
-                    if (opt[cur].mlen == 1) {
-                        litlen = opt[cur].litlen;
-                        if (cur > litlen)
-                            price = opt[cur - litlen].price + ZSTD_getPrice(optStatePtr, litlen, ip+cur-litlen, matches[u].off-1, mlen - MINMATCH, ultra);
-                        else
-                            price = ZSTD_getPrice(optStatePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
-                    } else {
-                        litlen = 0;
-                        price = opt[cur].price + ZSTD_getPrice(optStatePtr, 0, NULL, matches[u].off-1, mlen - MINMATCH, ultra);
-                    }
+            /* last match must start at a minimum distance of 8 from oend */
+            if (inr > ilimit) continue;
 
-                    if (cur + mlen > last_pos || (price < opt[cur + mlen].price))
-                        SET_PRICE(cur + mlen, mlen, matches[u].off, litlen, price);
+            if (cur == last_pos) break;
 
-                    mlen++;
-        }   }   }
+            if ( (optLevel==0) /*static_test*/
+              && (opt[cur+1].price <= opt[cur].price + (BITCOST_MULTIPLIER/2)) ) {
+                DEBUGLOG(7, "move to next rPos:%u : price is <=", cur+1);
+                continue;  /* skip unpromising positions; about ~+6% speed, -0.01 ratio */
+            }
 
-        best_mlen = opt[last_pos].mlen;
-        best_off = opt[last_pos].off;
-        cur = last_pos - best_mlen;
-
-        /* store sequence */
-_storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
-        opt[0].mlen = 1;
-
-        while (1) {
-            mlen = opt[cur].mlen;
-            offset = opt[cur].off;
-            opt[cur].mlen = best_mlen;
-            opt[cur].off = best_off;
-            best_mlen = mlen;
-            best_off = offset;
-            if (mlen > cur) break;
-            cur -= mlen;
-        }
+            {   U32 const ll0 = (opt[cur].mlen != 0);
+                U32 const litlen = (opt[cur].mlen == 0) ? opt[cur].litlen : 0;
+                U32 const previousPrice = opt[cur].price;
+                U32 const basePrice = previousPrice + ZSTD_litLengthPrice(0, optStatePtr, optLevel);
+                U32 const nbMatches = ZSTD_BtGetAllMatches(matches, ms, &nextToUpdate3, inr, iend, dictMode, opt[cur].rep, ll0, minMatch);
+                U32 matchNb;
+                if (!nbMatches) {
+                    DEBUGLOG(7, "rPos:%u : no match found", cur);
+                    continue;
+                }
 
-        for (u = 0; u <= last_pos;) {
-            u += opt[u].mlen;
+                {   U32 const maxML = matches[nbMatches-1].len;
+                    DEBUGLOG(7, "cPos:%zi==rPos:%u, found %u matches, of maxLength=%u",
+                                inr-istart, cur, nbMatches, maxML);
+
+                    if ( (maxML > sufficient_len)
+                      || (cur + maxML >= ZSTD_OPT_NUM) ) {
+                        lastSequence.mlen = maxML;
+                        lastSequence.off = matches[nbMatches-1].off;
+                        lastSequence.litlen = litlen;
+                        cur -= (opt[cur].mlen==0) ? opt[cur].litlen : 0;  /* last sequence is actually only literals, fix cur to last match - note : may underflow, in which case, it's first sequence, and it's okay */
+                        last_pos = cur + ZSTD_totalLen(lastSequence);
+                        if (cur > ZSTD_OPT_NUM) cur = 0;   /* underflow => first match */
+                        goto _shortestPath;
+                }   }
+
+                /* set prices using matches found at position == cur */
+                for (matchNb = 0; matchNb < nbMatches; matchNb++) {
+                    U32 const offset = matches[matchNb].off;
+                    U32 const lastML = matches[matchNb].len;
+                    U32 const startML = (matchNb>0) ? matches[matchNb-1].len+1 : minMatch;
+                    U32 mlen;
+
+                    DEBUGLOG(7, "testing match %u => offCode=%4u, mlen=%2u, llen=%2u",
+                                matchNb, matches[matchNb].off, lastML, litlen);
+
+                    for (mlen = lastML; mlen >= startML; mlen--) {  /* scan downward */
+                        U32 const pos = cur + mlen;
+                        int const price = basePrice + ZSTD_getMatchPrice(offset, mlen, optStatePtr, optLevel);
+
+                        if ((pos > last_pos) || (price < opt[pos].price)) {
+                            DEBUGLOG(7, "rPos:%u (ml=%2u) => new better price (%.2f<%.2f)",
+                                        pos, mlen, ZSTD_fCost(price), ZSTD_fCost(opt[pos].price));
+                            while (last_pos < pos) { opt[last_pos+1].price = ZSTD_MAX_PRICE; last_pos++; }   /* fill empty positions */
+                            opt[pos].mlen = mlen;
+                            opt[pos].off = offset;
+                            opt[pos].litlen = litlen;
+                            opt[pos].price = price;
+                        } else {
+                            DEBUGLOG(7, "rPos:%u (ml=%2u) => new price is worse (%.2f>=%.2f)",
+                                        pos, mlen, ZSTD_fCost(price), ZSTD_fCost(opt[pos].price));
+                            if (optLevel==0) break;  /* early update abort; gets ~+10% speed for about -0.01 ratio loss */
+                        }
+            }   }   }
+        }  /* for (cur = 1; cur <= last_pos; cur++) */
+
+        lastSequence = opt[last_pos];
+        cur = last_pos > ZSTD_totalLen(lastSequence) ? last_pos - ZSTD_totalLen(lastSequence) : 0;  /* single sequence, and it starts before `ip` */
+        assert(cur < ZSTD_OPT_NUM);  /* control overflow*/
+
+_shortestPath:   /* cur, last_pos, best_mlen, best_off have to be set */
+        assert(opt[0].mlen == 0);
+
+        /* Set the next chunk's repcodes based on the repcodes of the beginning
+         * of the last match, and the last sequence. This avoids us having to
+         * update them while traversing the sequences.
+         */
+        if (lastSequence.mlen != 0) {
+            repcodes_t reps = ZSTD_updateRep(opt[cur].rep, lastSequence.off, lastSequence.litlen==0);
+            memcpy(rep, &reps, sizeof(reps));
+        } else {
+            memcpy(rep, opt[cur].rep, sizeof(repcodes_t));
         }
 
-        for (cur=0; cur < last_pos; ) {
-            mlen = opt[cur].mlen;
-            if (mlen == 1) { ip++; cur++; continue; }
-            offset = opt[cur].off;
-            cur += mlen;
-            litLength = (U32)(ip - anchor);
-
-            if (offset > ZSTD_REP_MOVE_OPT) {
-                rep[2] = rep[1];
-                rep[1] = rep[0];
-                rep[0] = offset - ZSTD_REP_MOVE_OPT;
-                offset--;
-            } else {
-                if (offset != 0) {
-                    best_off = (offset==ZSTD_REP_MOVE_OPT) ? (rep[0] - 1) : (rep[offset]);
-                    if (offset != 1) rep[2] = rep[1];
-                    rep[1] = rep[0];
-                    rep[0] = best_off;
-                }
-                if (litLength==0) offset--;
+        {   U32 const storeEnd = cur + 1;
+            U32 storeStart = storeEnd;
+            U32 seqPos = cur;
+
+            DEBUGLOG(6, "start reverse traversal (last_pos:%u, cur:%u)",
+                        last_pos, cur); (void)last_pos;
+            assert(storeEnd < ZSTD_OPT_NUM);
+            DEBUGLOG(6, "last sequence copied into pos=%u (llen=%u,mlen=%u,ofc=%u)",
+                        storeEnd, lastSequence.litlen, lastSequence.mlen, lastSequence.off);
+            opt[storeEnd] = lastSequence;
+            while (seqPos > 0) {
+                U32 const backDist = ZSTD_totalLen(opt[seqPos]);
+                storeStart--;
+                DEBUGLOG(6, "sequence from rPos=%u copied into pos=%u (llen=%u,mlen=%u,ofc=%u)",
+                            seqPos, storeStart, opt[seqPos].litlen, opt[seqPos].mlen, opt[seqPos].off);
+                opt[storeStart] = opt[seqPos];
+                seqPos = (seqPos > backDist) ? seqPos - backDist : 0;
             }
 
-            ZSTD_updatePrice(optStatePtr, litLength, anchor, offset, mlen-MINMATCH);
-            ZSTD_storeSeq(seqStorePtr, litLength, anchor, offset, mlen-MINMATCH);
-            anchor = ip = ip + mlen;
-    }    }   /* for (cur=0; cur < last_pos; ) */
+            /* save sequences */
+            DEBUGLOG(6, "sending selected sequences into seqStore")
+            {   U32 storePos;
+                for (storePos=storeStart; storePos <= storeEnd; storePos++) {
+                    U32 const llen = opt[storePos].litlen;
+                    U32 const mlen = opt[storePos].mlen;
+                    U32 const offCode = opt[storePos].off;
+                    U32 const advance = llen + mlen;
+                    DEBUGLOG(6, "considering seq starting at %zi, llen=%u, mlen=%u",
+                                anchor - istart, (unsigned)llen, (unsigned)mlen);
+
+                    if (mlen==0) {  /* only literals => must be last "sequence", actually starting a new stream of sequences */
+                        assert(storePos == storeEnd);   /* must be last sequence */
+                        ip = anchor + llen;     /* last "sequence" is a bunch of literals => don't progress anchor */
+                        continue;   /* will finish */
+                    }
 
-    /* Save reps for next block */
-    { int i; for (i=0; i<ZSTD_REP_NUM; i++) seqStorePtr->repToConfirm[i] = rep[i]; }
+                    assert(anchor + llen <= iend);
+                    ZSTD_updateStats(optStatePtr, llen, anchor, offCode, mlen);
+                    ZSTD_storeSeq(seqStore, llen, anchor, iend, offCode, mlen-MINMATCH);
+                    anchor += advance;
+                    ip = anchor;
+            }   }
+            ZSTD_setBasePrices(optStatePtr, optLevel);
+        }
+    }   /* while (ip < ilimit) */
 
     /* Return the last literals size */
-    return iend - anchor;
+    return (size_t)(iend - anchor);
 }
 
 
-size_t ZSTD_compressBlock_btopt(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+size_t ZSTD_compressBlock_btopt(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_opt_generic(ctx, src, srcSize, 0);
+    DEBUGLOG(5, "ZSTD_compressBlock_btopt");
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 0 /*optLevel*/, ZSTD_noDict);
 }
 
-size_t ZSTD_compressBlock_btultra(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+
+/* used in 2-pass strategy */
+static U32 ZSTD_upscaleStat(unsigned* table, U32 lastEltIndex, int bonus)
 {
-    return ZSTD_compressBlock_opt_generic(ctx, src, srcSize, 1);
+    U32 s, sum=0;
+    assert(ZSTD_FREQ_DIV+bonus >= 0);
+    for (s=0; s<lastEltIndex+1; s++) {
+        table[s] <<= ZSTD_FREQ_DIV+bonus;
+        table[s]--;
+        sum += table[s];
+    }
+    return sum;
 }
 
-
-FORCE_INLINE_TEMPLATE
-size_t ZSTD_compressBlock_opt_extDict_generic(ZSTD_CCtx* ctx,
-                                     const void* src, size_t srcSize, const int ultra)
+/* used in 2-pass strategy */
+MEM_STATIC void ZSTD_upscaleStats(optState_t* optPtr)
 {
-    seqStore_t* seqStorePtr = &(ctx->seqStore);
-    optState_t* optStatePtr = &(ctx->optState);
-    const BYTE* const istart = (const BYTE*)src;
-    const BYTE* ip = istart;
-    const BYTE* anchor = istart;
-    const BYTE* const iend = istart + srcSize;
-    const BYTE* const ilimit = iend - 8;
-    const BYTE* const base = ctx->base;
-    const U32 lowestIndex = ctx->lowLimit;
-    const U32 dictLimit = ctx->dictLimit;
-    const BYTE* const prefixStart = base + dictLimit;
-    const BYTE* const dictBase = ctx->dictBase;
-    const BYTE* const dictEnd  = dictBase + dictLimit;
-
-    const U32 maxSearches = 1U << ctx->appliedParams.cParams.searchLog;
-    const U32 sufficient_len = ctx->appliedParams.cParams.targetLength;
-    const U32 mls = ctx->appliedParams.cParams.searchLength;
-    const U32 minMatch = (ctx->appliedParams.cParams.searchLength == 3) ? 3 : 4;
-
-    ZSTD_optimal_t* opt = optStatePtr->priceTable;
-    ZSTD_match_t* matches = optStatePtr->matchTable;
-    const BYTE* inr;
-
-    /* init */
-    U32 offset, rep[ZSTD_REP_NUM];
-    { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) rep[i]=seqStorePtr->rep[i]; }
-
-    ctx->nextToUpdate3 = ctx->nextToUpdate;
-    ZSTD_rescaleFreqs(optStatePtr, (const BYTE*)src, srcSize);
-    ip += (ip==prefixStart);
-
-    /* Match Loop */
-    while (ip < ilimit) {
-        U32 cur, match_num, last_pos, litlen, price;
-        U32 u, mlen, best_mlen, best_off, litLength;
-        U32 current = (U32)(ip-base);
-        memset(opt, 0, sizeof(ZSTD_optimal_t));
-        last_pos = 0;
-        opt[0].litlen = (U32)(ip - anchor);
-
-        /* check repCode */
-        {   U32 i, last_i = ZSTD_REP_CHECK + (ip==anchor);
-            for (i = (ip==anchor); i<last_i; i++) {
-                const S32 repCur = (i==ZSTD_REP_MOVE_OPT) ? (rep[0] - 1) : rep[i];
-                const U32 repIndex = (U32)(current - repCur);
-                const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
-                const BYTE* const repMatch = repBase + repIndex;
-                if ( (repCur > 0 && repCur <= (S32)current)
-                   && (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex>lowestIndex))  /* intentional overflow */
-                   && (ZSTD_readMINMATCH(ip, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
-                    /* repcode detected we should take it */
-                    const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
-                    mlen = (U32)ZSTD_count_2segments(ip+minMatch, repMatch+minMatch, iend, repEnd, prefixStart) + minMatch;
-
-                    if (mlen > sufficient_len || mlen >= ZSTD_OPT_NUM) {
-                        best_mlen = mlen; best_off = i; cur = 0; last_pos = 1;
-                        goto _storeSequence;
-                    }
-
-                    best_off = i - (ip==anchor);
-                    litlen = opt[0].litlen;
-                    do {
-                        price = ZSTD_getPrice(optStatePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
-                        if (mlen > last_pos || price < opt[mlen].price)
-                            SET_PRICE(mlen, mlen, i, litlen, price);   /* note : macro modifies last_pos */
-                        mlen--;
-                    } while (mlen >= minMatch);
-        }   }   }
-
-        match_num = ZSTD_BtGetAllMatches_selectMLS_extDict(ctx, ip, iend, maxSearches, mls, matches, minMatch);  /* first search (depth 0) */
-
-        if (!last_pos && !match_num) { ip++; continue; }
-
-        { U32 i; for (i=0; i<ZSTD_REP_NUM; i++) opt[0].rep[i] = rep[i]; }
-        opt[0].mlen = 1;
-
-        if (match_num && (matches[match_num-1].len > sufficient_len || matches[match_num-1].len >= ZSTD_OPT_NUM)) {
-            best_mlen = matches[match_num-1].len;
-            best_off = matches[match_num-1].off;
-            cur = 0;
-            last_pos = 1;
-            goto _storeSequence;
-        }
-
-        best_mlen = (last_pos) ? last_pos : minMatch;
-
-        /* set prices using matches at position = 0 */
-        for (u = 0; u < match_num; u++) {
-            mlen = (u>0) ? matches[u-1].len+1 : best_mlen;
-            best_mlen = matches[u].len;
-            litlen = opt[0].litlen;
-            while (mlen <= best_mlen) {
-                price = ZSTD_getPrice(optStatePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
-                if (mlen > last_pos || price < opt[mlen].price)
-                    SET_PRICE(mlen, mlen, matches[u].off, litlen, price);
-                mlen++;
-        }   }
-
-        if (last_pos < minMatch) {
-            ip++; continue;
-        }
-
-        /* check further positions */
-        for (cur = 1; cur <= last_pos; cur++) {
-            inr = ip + cur;
-
-            if (opt[cur-1].mlen == 1) {
-                litlen = opt[cur-1].litlen + 1;
-                if (cur > litlen) {
-                    price = opt[cur - litlen].price + ZSTD_getLiteralPrice(optStatePtr, litlen, inr-litlen);
-                } else
-                    price = ZSTD_getLiteralPrice(optStatePtr, litlen, anchor);
-            } else {
-                litlen = 1;
-                price = opt[cur - 1].price + ZSTD_getLiteralPrice(optStatePtr, litlen, inr-1);
-            }
-
-            if (cur > last_pos || price <= opt[cur].price)
-                SET_PRICE(cur, 1, 0, litlen, price);
-
-            if (cur == last_pos) break;
-
-            if (inr > ilimit)  /* last match must start at a minimum distance of 8 from oend */
-                continue;
-
-            mlen = opt[cur].mlen;
-            if (opt[cur].off > ZSTD_REP_MOVE_OPT) {
-                opt[cur].rep[2] = opt[cur-mlen].rep[1];
-                opt[cur].rep[1] = opt[cur-mlen].rep[0];
-                opt[cur].rep[0] = opt[cur].off - ZSTD_REP_MOVE_OPT;
-            } else {
-                opt[cur].rep[2] = (opt[cur].off > 1) ? opt[cur-mlen].rep[1] : opt[cur-mlen].rep[2];
-                opt[cur].rep[1] = (opt[cur].off > 0) ? opt[cur-mlen].rep[0] : opt[cur-mlen].rep[1];
-                opt[cur].rep[0] = ((opt[cur].off==ZSTD_REP_MOVE_OPT) && (mlen != 1)) ? (opt[cur-mlen].rep[0] - 1) : (opt[cur-mlen].rep[opt[cur].off]);
-            }
-
-            best_mlen = minMatch;
-            {   U32 i, last_i = ZSTD_REP_CHECK + (mlen != 1);
-                for (i = (mlen != 1); i<last_i; i++) {
-                    const S32 repCur = (i==ZSTD_REP_MOVE_OPT) ? (opt[cur].rep[0] - 1) : opt[cur].rep[i];
-                    const U32 repIndex = (U32)(current+cur - repCur);
-                    const BYTE* const repBase = repIndex < dictLimit ? dictBase : base;
-                    const BYTE* const repMatch = repBase + repIndex;
-                    if ( (repCur > 0 && repCur <= (S32)(current+cur))
-                      && (((U32)((dictLimit-1) - repIndex) >= 3) & (repIndex>lowestIndex))  /* intentional overflow */
-                      && (ZSTD_readMINMATCH(inr, minMatch) == ZSTD_readMINMATCH(repMatch, minMatch)) ) {
-                        /* repcode detected */
-                        const BYTE* const repEnd = repIndex < dictLimit ? dictEnd : iend;
-                        mlen = (U32)ZSTD_count_2segments(inr+minMatch, repMatch+minMatch, iend, repEnd, prefixStart) + minMatch;
-
-                        if (mlen > sufficient_len || cur + mlen >= ZSTD_OPT_NUM) {
-                            best_mlen = mlen; best_off = i; last_pos = cur + 1;
-                            goto _storeSequence;
-                        }
-
-                        best_off = i - (opt[cur].mlen != 1);
-                        if (mlen > best_mlen) best_mlen = mlen;
-
-                        do {
-                            if (opt[cur].mlen == 1) {
-                                litlen = opt[cur].litlen;
-                                if (cur > litlen) {
-                                    price = opt[cur - litlen].price + ZSTD_getPrice(optStatePtr, litlen, inr-litlen, best_off, mlen - MINMATCH, ultra);
-                                } else
-                                    price = ZSTD_getPrice(optStatePtr, litlen, anchor, best_off, mlen - MINMATCH, ultra);
-                            } else {
-                                litlen = 0;
-                                price = opt[cur].price + ZSTD_getPrice(optStatePtr, 0, NULL, best_off, mlen - MINMATCH, ultra);
-                            }
-
-                            if (cur + mlen > last_pos || price <= opt[cur + mlen].price)
-                                SET_PRICE(cur + mlen, mlen, i, litlen, price);
-                            mlen--;
-                        } while (mlen >= minMatch);
-            }   }   }
-
-            match_num = ZSTD_BtGetAllMatches_selectMLS_extDict(ctx, inr, iend, maxSearches, mls, matches, minMatch);
-
-            if (match_num > 0 && (matches[match_num-1].len > sufficient_len || cur + matches[match_num-1].len >= ZSTD_OPT_NUM)) {
-                best_mlen = matches[match_num-1].len;
-                best_off = matches[match_num-1].off;
-                last_pos = cur + 1;
-                goto _storeSequence;
-            }
-
-            /* set prices using matches at position = cur */
-            for (u = 0; u < match_num; u++) {
-                mlen = (u>0) ? matches[u-1].len+1 : best_mlen;
-                best_mlen = matches[u].len;
-
-                while (mlen <= best_mlen) {
-                    if (opt[cur].mlen == 1) {
-                        litlen = opt[cur].litlen;
-                        if (cur > litlen)
-                            price = opt[cur - litlen].price + ZSTD_getPrice(optStatePtr, litlen, ip+cur-litlen, matches[u].off-1, mlen - MINMATCH, ultra);
-                        else
-                            price = ZSTD_getPrice(optStatePtr, litlen, anchor, matches[u].off-1, mlen - MINMATCH, ultra);
-                    } else {
-                        litlen = 0;
-                        price = opt[cur].price + ZSTD_getPrice(optStatePtr, 0, NULL, matches[u].off-1, mlen - MINMATCH, ultra);
-                    }
-
-                    if (cur + mlen > last_pos || (price < opt[cur + mlen].price))
-                        SET_PRICE(cur + mlen, mlen, matches[u].off, litlen, price);
-
-                    mlen++;
-        }   }   }   /* for (cur = 1; cur <= last_pos; cur++) */
-
-        best_mlen = opt[last_pos].mlen;
-        best_off = opt[last_pos].off;
-        cur = last_pos - best_mlen;
-
-        /* store sequence */
-_storeSequence:   /* cur, last_pos, best_mlen, best_off have to be set */
-        opt[0].mlen = 1;
-
-        while (1) {
-            mlen = opt[cur].mlen;
-            offset = opt[cur].off;
-            opt[cur].mlen = best_mlen;
-            opt[cur].off = best_off;
-            best_mlen = mlen;
-            best_off = offset;
-            if (mlen > cur) break;
-            cur -= mlen;
-        }
-
-        for (u = 0; u <= last_pos; ) {
-            u += opt[u].mlen;
-        }
+    if (ZSTD_compressedLiterals(optPtr))
+        optPtr->litSum = ZSTD_upscaleStat(optPtr->litFreq, MaxLit, 0);
+    optPtr->litLengthSum = ZSTD_upscaleStat(optPtr->litLengthFreq, MaxLL, 0);
+    optPtr->matchLengthSum = ZSTD_upscaleStat(optPtr->matchLengthFreq, MaxML, 0);
+    optPtr->offCodeSum = ZSTD_upscaleStat(optPtr->offCodeFreq, MaxOff, 0);
+}
 
-        for (cur=0; cur < last_pos; ) {
-            mlen = opt[cur].mlen;
-            if (mlen == 1) { ip++; cur++; continue; }
-            offset = opt[cur].off;
-            cur += mlen;
-            litLength = (U32)(ip - anchor);
-
-            if (offset > ZSTD_REP_MOVE_OPT) {
-                rep[2] = rep[1];
-                rep[1] = rep[0];
-                rep[0] = offset - ZSTD_REP_MOVE_OPT;
-                offset--;
-            } else {
-                if (offset != 0) {
-                    best_off = (offset==ZSTD_REP_MOVE_OPT) ? (rep[0] - 1) : (rep[offset]);
-                    if (offset != 1) rep[2] = rep[1];
-                    rep[1] = rep[0];
-                    rep[0] = best_off;
-                }
+/* ZSTD_initStats_ultra():
+ * make a first compression pass, just to seed stats with more accurate starting values.
+ * only works on first block, with no dictionary and no ldm.
+ * this function cannot error, hence its contract must be respected.
+ */
+static void
+ZSTD_initStats_ultra(ZSTD_matchState_t* ms,
+                     seqStore_t* seqStore,
+                     U32 rep[ZSTD_REP_NUM],
+               const void* src, size_t srcSize)
+{
+    U32 tmpRep[ZSTD_REP_NUM];  /* updated rep codes will sink here */
+    memcpy(tmpRep, rep, sizeof(tmpRep));
+
+    DEBUGLOG(4, "ZSTD_initStats_ultra (srcSize=%zu)", srcSize);
+    assert(ms->opt.litLengthSum == 0);    /* first block */
+    assert(seqStore->sequences == seqStore->sequencesStart);   /* no ldm */
+    assert(ms->window.dictLimit == ms->window.lowLimit);   /* no dictionary */
+    assert(ms->window.dictLimit - ms->nextToUpdate <= 1);  /* no prefix (note: intentional overflow, defined as 2-complement) */
+
+    ZSTD_compressBlock_opt_generic(ms, seqStore, tmpRep, src, srcSize, 2 /*optLevel*/, ZSTD_noDict);   /* generate stats into ms->opt*/
+
+    /* invalidate first scan from history */
+    ZSTD_resetSeqStore(seqStore);
+    ms->window.base -= srcSize;
+    ms->window.dictLimit += (U32)srcSize;
+    ms->window.lowLimit = ms->window.dictLimit;
+    ms->nextToUpdate = ms->window.dictLimit;
+
+    /* re-inforce weight of collected statistics */
+    ZSTD_upscaleStats(&ms->opt);
+}
 
-                if (litLength==0) offset--;
-            }
+size_t ZSTD_compressBlock_btultra(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    DEBUGLOG(5, "ZSTD_compressBlock_btultra (srcSize=%zu)", srcSize);
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 2 /*optLevel*/, ZSTD_noDict);
+}
 
-            ZSTD_updatePrice(optStatePtr, litLength, anchor, offset, mlen-MINMATCH);
-            ZSTD_storeSeq(seqStorePtr, litLength, anchor, offset, mlen-MINMATCH);
-            anchor = ip = ip + mlen;
-    }    }   /* for (cur=0; cur < last_pos; ) */
+size_t ZSTD_compressBlock_btultra2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    U32 const current = (U32)((const BYTE*)src - ms->window.base);
+    DEBUGLOG(5, "ZSTD_compressBlock_btultra2 (srcSize=%zu)", srcSize);
+
+    /* 2-pass strategy:
+     * this strategy makes a first pass over first block to collect statistics
+     * and seed next round's statistics with it.
+     * After 1st pass, function forgets everything, and starts a new block.
+     * Consequently, this can only work if no data has been previously loaded in tables,
+     * aka, no dictionary, no prefix, no ldm preprocessing.
+     * The compression ratio gain is generally small (~0.5% on first block),
+     * the cost is 2x cpu time on first block. */
+    assert(srcSize <= ZSTD_BLOCKSIZE_MAX);
+    if ( (ms->opt.litLengthSum==0)   /* first block */
+      && (seqStore->sequences == seqStore->sequencesStart)  /* no ldm */
+      && (ms->window.dictLimit == ms->window.lowLimit)   /* no dictionary */
+      && (current == ms->window.dictLimit)   /* start of frame, nothing already loaded nor skipped */
+      && (srcSize > ZSTD_PREDEF_THRESHOLD)
+      ) {
+        ZSTD_initStats_ultra(ms, seqStore, rep, src, srcSize);
+    }
 
-    /* Save reps for next block */
-    { int i; for (i=0; i<ZSTD_REP_NUM; i++) seqStorePtr->repToConfirm[i] = rep[i]; }
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 2 /*optLevel*/, ZSTD_noDict);
+}
 
-    /* Return the last literals size */
-    return iend - anchor;
+size_t ZSTD_compressBlock_btopt_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 0 /*optLevel*/, ZSTD_dictMatchState);
 }
 
+size_t ZSTD_compressBlock_btultra_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
+{
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 2 /*optLevel*/, ZSTD_dictMatchState);
+}
 
-size_t ZSTD_compressBlock_btopt_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+size_t ZSTD_compressBlock_btopt_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_opt_extDict_generic(ctx, src, srcSize, 0);
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 0 /*optLevel*/, ZSTD_extDict);
 }
 
-size_t ZSTD_compressBlock_btultra_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize)
+size_t ZSTD_compressBlock_btultra_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        const void* src, size_t srcSize)
 {
-    return ZSTD_compressBlock_opt_extDict_generic(ctx, src, srcSize, 1);
+    return ZSTD_compressBlock_opt_generic(ms, seqStore, rep, src, srcSize, 2 /*optLevel*/, ZSTD_extDict);
 }
+
+/* note : no btultra2 variant for extDict nor dictMatchState,
+ * because btultra2 is not meant to work with dictionaries
+ * and is only specific for the first block (no prefix) */
diff --git a/contrib/zstd/zstd_opt.h b/contrib/zstd/zstd_opt.h
index 816a1fabb..9aba8a901 100644
--- a/contrib/zstd/zstd_opt.h
+++ b/contrib/zstd/zstd_opt.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
+ * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc.
  * All rights reserved.
  *
  * This source code is licensed under both the BSD-style license (found in the
@@ -11,17 +11,43 @@
 #ifndef ZSTD_OPT_H
 #define ZSTD_OPT_H
 
-#include "zstd_compress.h"
-
 #if defined (__cplusplus)
 extern "C" {
 #endif
 
-size_t ZSTD_compressBlock_btopt(ZSTD_CCtx* ctx, const void* src, size_t srcSize);
-size_t ZSTD_compressBlock_btultra(ZSTD_CCtx* ctx, const void* src, size_t srcSize);
+#include "zstd_compress_internal.h"
+
+/* used in ZSTD_loadDictionaryContent() */
+void ZSTD_updateTree(ZSTD_matchState_t* ms, const BYTE* ip, const BYTE* iend);
+
+size_t ZSTD_compressBlock_btopt(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btultra(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btultra2(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+
+
+size_t ZSTD_compressBlock_btopt_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btultra_dictMatchState(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+
+size_t ZSTD_compressBlock_btopt_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
+size_t ZSTD_compressBlock_btultra_extDict(
+        ZSTD_matchState_t* ms, seqStore_t* seqStore, U32 rep[ZSTD_REP_NUM],
+        void const* src, size_t srcSize);
 
-size_t ZSTD_compressBlock_btopt_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize);
-size_t ZSTD_compressBlock_btultra_extDict(ZSTD_CCtx* ctx, const void* src, size_t srcSize);
+        /* note : no btultra2 variant for extDict nor dictMatchState,
+         * because btultra2 is not meant to work with dictionaries
+         * and is only specific for the first block (no prefix) */
 
 #if defined (__cplusplus)
 }
diff --git a/contrib/zstd/zstdmt_compress.c b/contrib/zstd/zstdmt_compress.c
deleted file mode 100644
index ecb799ab3..000000000
--- a/contrib/zstd/zstdmt_compress.c
+++ /dev/null
@@ -1,1089 +0,0 @@
-/*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
- * All rights reserved.
- *
- * This source code is licensed under both the BSD-style license (found in the
- * LICENSE file in the root directory of this source tree) and the GPLv2 (found
- * in the COPYING file in the root directory of this source tree).
- * You may select, at your option, one of the above-listed licenses.
- */
-
-
-/* ======   Tuning parameters   ====== */
-#define ZSTDMT_NBTHREADS_MAX 256
-#define ZSTDMT_OVERLAPLOG_DEFAULT 6
-
-
-/* ======   Compiler specifics   ====== */
-#if defined(_MSC_VER)
-#  pragma warning(disable : 4204)   /* disable: C4204: non-constant aggregate initializer */
-#endif
-
-
-/* ======   Dependencies   ====== */
-#include <string.h>      /* memcpy, memset */
-#include "pool.h"        /* threadpool */
-#include "threading.h"   /* mutex */
-#include "zstd_internal.h"  /* MIN, ERROR, ZSTD_*, ZSTD_highbit32 */
-#include "zstdmt_compress.h"
-
-
-/* ======   Debug   ====== */
-#if defined(ZSTD_DEBUG) && (ZSTD_DEBUG>=2)
-
-#  include <stdio.h>
-#  include <unistd.h>
-#  include <sys/times.h>
-#  define DEBUGLOGRAW(l, ...) if (l<=ZSTD_DEBUG) { fprintf(stderr, __VA_ARGS__); }
-
-#  define DEBUG_PRINTHEX(l,p,n) {            \
-    unsigned debug_u;                        \
-    for (debug_u=0; debug_u<(n); debug_u++)  \
-        DEBUGLOGRAW(l, "%02X ", ((const unsigned char*)(p))[debug_u]); \
-    DEBUGLOGRAW(l, " \n");                   \
-}
-
-static unsigned long long GetCurrentClockTimeMicroseconds(void)
-{
-   static clock_t _ticksPerSecond = 0;
-   if (_ticksPerSecond <= 0) _ticksPerSecond = sysconf(_SC_CLK_TCK);
-
-   { struct tms junk; clock_t newTicks = (clock_t) times(&junk);
-     return ((((unsigned long long)newTicks)*(1000000))/_ticksPerSecond); }
-}
-
-#define MUTEX_WAIT_TIME_DLEVEL 6
-#define PTHREAD_MUTEX_LOCK(mutex) {               \
-    if (ZSTD_DEBUG>=MUTEX_WAIT_TIME_DLEVEL) {   \
-        unsigned long long const beforeTime = GetCurrentClockTimeMicroseconds(); \
-        pthread_mutex_lock(mutex);                \
-        {   unsigned long long const afterTime = GetCurrentClockTimeMicroseconds(); \
-            unsigned long long const elapsedTime = (afterTime-beforeTime); \
-            if (elapsedTime > 1000) {  /* or whatever threshold you like; I'm using 1 millisecond here */ \
-                DEBUGLOG(MUTEX_WAIT_TIME_DLEVEL, "Thread took %llu microseconds to acquire mutex %s \n", \
-                   elapsedTime, #mutex);          \
-        }   }                                     \
-    } else pthread_mutex_lock(mutex);             \
-}
-
-#else
-
-#  define PTHREAD_MUTEX_LOCK(m) pthread_mutex_lock(m)
-#  define DEBUG_PRINTHEX(l,p,n) {}
-
-#endif
-
-
-/* =====   Buffer Pool   ===== */
-/* a single Buffer Pool can be invoked from multiple threads in parallel */
-
-typedef struct buffer_s {
-    void* start;
-    size_t size;
-} buffer_t;
-
-static const buffer_t g_nullBuffer = { NULL, 0 };
-
-typedef struct ZSTDMT_bufferPool_s {
-    pthread_mutex_t poolMutex;
-    size_t bufferSize;
-    unsigned totalBuffers;
-    unsigned nbBuffers;
-    ZSTD_customMem cMem;
-    buffer_t bTable[1];   /* variable size */
-} ZSTDMT_bufferPool;
-
-static ZSTDMT_bufferPool* ZSTDMT_createBufferPool(unsigned nbThreads, ZSTD_customMem cMem)
-{
-    unsigned const maxNbBuffers = 2*nbThreads + 3;
-    ZSTDMT_bufferPool* const bufPool = (ZSTDMT_bufferPool*)ZSTD_calloc(
-        sizeof(ZSTDMT_bufferPool) + (maxNbBuffers-1) * sizeof(buffer_t), cMem);
-    if (bufPool==NULL) return NULL;
-    if (pthread_mutex_init(&bufPool->poolMutex, NULL)) {
-        ZSTD_free(bufPool, cMem);
-        return NULL;
-    }
-    bufPool->bufferSize = 64 KB;
-    bufPool->totalBuffers = maxNbBuffers;
-    bufPool->nbBuffers = 0;
-    bufPool->cMem = cMem;
-    return bufPool;
-}
-
-static void ZSTDMT_freeBufferPool(ZSTDMT_bufferPool* bufPool)
-{
-    unsigned u;
-    if (!bufPool) return;   /* compatibility with free on NULL */
-    for (u=0; u<bufPool->totalBuffers; u++)
-        ZSTD_free(bufPool->bTable[u].start, bufPool->cMem);
-    pthread_mutex_destroy(&bufPool->poolMutex);
-    ZSTD_free(bufPool, bufPool->cMem);
-}
-
-/* only works at initialization, not during compression */
-static size_t ZSTDMT_sizeof_bufferPool(ZSTDMT_bufferPool* bufPool)
-{
-    size_t const poolSize = sizeof(*bufPool)
-                            + (bufPool->totalBuffers - 1) * sizeof(buffer_t);
-    unsigned u;
-    size_t totalBufferSize = 0;
-    pthread_mutex_lock(&bufPool->poolMutex);
-    for (u=0; u<bufPool->totalBuffers; u++)
-        totalBufferSize += bufPool->bTable[u].size;
-    pthread_mutex_unlock(&bufPool->poolMutex);
-
-    return poolSize + totalBufferSize;
-}
-
-static void ZSTDMT_setBufferSize(ZSTDMT_bufferPool* bufPool, size_t bSize)
-{
-    bufPool->bufferSize = bSize;
-}
-
-/** ZSTDMT_getBuffer() :
- *  assumption : bufPool must be valid */
-static buffer_t ZSTDMT_getBuffer(ZSTDMT_bufferPool* bufPool)
-{
-    size_t const bSize = bufPool->bufferSize;
-    DEBUGLOG(5, "ZSTDMT_getBuffer");
-    pthread_mutex_lock(&bufPool->poolMutex);
-    if (bufPool->nbBuffers) {   /* try to use an existing buffer */
-        buffer_t const buf = bufPool->bTable[--(bufPool->nbBuffers)];
-        size_t const availBufferSize = buf.size;
-        if ((availBufferSize >= bSize) & (availBufferSize <= 10*bSize)) {
-            /* large enough, but not too much */
-            pthread_mutex_unlock(&bufPool->poolMutex);
-            return buf;
-        }
-        /* size conditions not respected : scratch this buffer, create new one */
-        DEBUGLOG(5, "existing buffer does not meet size conditions => freeing");
-        ZSTD_free(buf.start, bufPool->cMem);
-    }
-    pthread_mutex_unlock(&bufPool->poolMutex);
-    /* create new buffer */
-    DEBUGLOG(5, "create a new buffer");
-    {   buffer_t buffer;
-        void* const start = ZSTD_malloc(bSize, bufPool->cMem);
-        buffer.start = start;   /* note : start can be NULL if malloc fails ! */
-        buffer.size = (start==NULL) ? 0 : bSize;
-        return buffer;
-    }
-}
-
-/* store buffer for later re-use, up to pool capacity */
-static void ZSTDMT_releaseBuffer(ZSTDMT_bufferPool* bufPool, buffer_t buf)
-{
-    if (buf.start == NULL) return;   /* compatible with release on NULL */
-    DEBUGLOG(5, "ZSTDMT_releaseBuffer");
-    pthread_mutex_lock(&bufPool->poolMutex);
-    if (bufPool->nbBuffers < bufPool->totalBuffers) {
-        bufPool->bTable[bufPool->nbBuffers++] = buf;  /* stored for later use */
-        pthread_mutex_unlock(&bufPool->poolMutex);
-        return;
-    }
-    pthread_mutex_unlock(&bufPool->poolMutex);
-    /* Reached bufferPool capacity (should not happen) */
-    DEBUGLOG(5, "buffer pool capacity reached => freeing ");
-    ZSTD_free(buf.start, bufPool->cMem);
-}
-
-/* Sets parameters relevant to the compression job, initializing others to
- * default values. Notably, nbThreads should probably be zero. */
-static ZSTD_CCtx_params ZSTDMT_makeJobCCtxParams(ZSTD_CCtx_params const params)
-{
-    ZSTD_CCtx_params jobParams;
-    memset(&jobParams, 0, sizeof(jobParams));
-
-    jobParams.cParams = params.cParams;
-    jobParams.fParams = params.fParams;
-    jobParams.compressionLevel = params.compressionLevel;
-
-    jobParams.ldmParams = params.ldmParams;
-    return jobParams;
-}
-
-/* =====   CCtx Pool   ===== */
-/* a single CCtx Pool can be invoked from multiple threads in parallel */
-
-typedef struct {
-    pthread_mutex_t poolMutex;
-    unsigned totalCCtx;
-    unsigned availCCtx;
-    ZSTD_customMem cMem;
-    ZSTD_CCtx* cctx[1];   /* variable size */
-} ZSTDMT_CCtxPool;
-
-/* note : all CCtx borrowed from the pool should be released back to the pool _before_ freeing the pool */
-static void ZSTDMT_freeCCtxPool(ZSTDMT_CCtxPool* pool)
-{
-    unsigned u;
-    for (u=0; u<pool->totalCCtx; u++)
-        ZSTD_freeCCtx(pool->cctx[u]);  /* note : compatible with free on NULL */
-    pthread_mutex_destroy(&pool->poolMutex);
-    ZSTD_free(pool, pool->cMem);
-}
-
-/* ZSTDMT_createCCtxPool() :
- * implies nbThreads >= 1 , checked by caller ZSTDMT_createCCtx() */
-static ZSTDMT_CCtxPool* ZSTDMT_createCCtxPool(unsigned nbThreads,
-                                              ZSTD_customMem cMem)
-{
-    ZSTDMT_CCtxPool* const cctxPool = (ZSTDMT_CCtxPool*) ZSTD_calloc(
-        sizeof(ZSTDMT_CCtxPool) + (nbThreads-1)*sizeof(ZSTD_CCtx*), cMem);
-    if (!cctxPool) return NULL;
-    if (pthread_mutex_init(&cctxPool->poolMutex, NULL)) {
-        ZSTD_free(cctxPool, cMem);
-        return NULL;
-    }
-    cctxPool->cMem = cMem;
-    cctxPool->totalCCtx = nbThreads;
-    cctxPool->availCCtx = 1;   /* at least one cctx for single-thread mode */
-    cctxPool->cctx[0] = ZSTD_createCCtx_advanced(cMem);
-    if (!cctxPool->cctx[0]) { ZSTDMT_freeCCtxPool(cctxPool); return NULL; }
-    DEBUGLOG(3, "cctxPool created, with %u threads", nbThreads);
-    return cctxPool;
-}
-
-/* only works during initialization phase, not during compression */
-static size_t ZSTDMT_sizeof_CCtxPool(ZSTDMT_CCtxPool* cctxPool)
-{
-    pthread_mutex_lock(&cctxPool->poolMutex);
-    {   unsigned const nbThreads = cctxPool->totalCCtx;
-        size_t const poolSize = sizeof(*cctxPool)
-                                + (nbThreads-1)*sizeof(ZSTD_CCtx*);
-        unsigned u;
-        size_t totalCCtxSize = 0;
-        for (u=0; u<nbThreads; u++) {
-            totalCCtxSize += ZSTD_sizeof_CCtx(cctxPool->cctx[u]);
-        }
-        pthread_mutex_unlock(&cctxPool->poolMutex);
-        return poolSize + totalCCtxSize;
-    }
-}
-
-static ZSTD_CCtx* ZSTDMT_getCCtx(ZSTDMT_CCtxPool* cctxPool)
-{
-    DEBUGLOG(5, "ZSTDMT_getCCtx");
-    pthread_mutex_lock(&cctxPool->poolMutex);
-    if (cctxPool->availCCtx) {
-        cctxPool->availCCtx--;
-        {   ZSTD_CCtx* const cctx = cctxPool->cctx[cctxPool->availCCtx];
-            pthread_mutex_unlock(&cctxPool->poolMutex);
-            return cctx;
-    }   }
-    pthread_mutex_unlock(&cctxPool->poolMutex);
-    DEBUGLOG(5, "create one more CCtx");
-    return ZSTD_createCCtx_advanced(cctxPool->cMem);   /* note : can be NULL, when creation fails ! */
-}
-
-static void ZSTDMT_releaseCCtx(ZSTDMT_CCtxPool* pool, ZSTD_CCtx* cctx)
-{
-    if (cctx==NULL) return;   /* compatibility with release on NULL */
-    pthread_mutex_lock(&pool->poolMutex);
-    if (pool->availCCtx < pool->totalCCtx)
-        pool->cctx[pool->availCCtx++] = cctx;
-    else {
-        /* pool overflow : should not happen, since totalCCtx==nbThreads */
-        DEBUGLOG(5, "CCtx pool overflow : free cctx");
-        ZSTD_freeCCtx(cctx);
-    }
-    pthread_mutex_unlock(&pool->poolMutex);
-}
-
-
-/* =====   Thread worker   ===== */
-
-typedef struct {
-    buffer_t src;
-    const void* srcStart;
-    size_t   dictSize;
-    size_t   srcSize;
-    buffer_t dstBuff;
-    size_t   cSize;
-    size_t   dstFlushed;
-    unsigned firstChunk;
-    unsigned lastChunk;
-    unsigned jobCompleted;
-    unsigned jobScanned;
-    pthread_mutex_t* jobCompleted_mutex;
-    pthread_cond_t* jobCompleted_cond;
-    ZSTD_CCtx_params params;
-    const ZSTD_CDict* cdict;
-    ZSTDMT_CCtxPool* cctxPool;
-    ZSTDMT_bufferPool* bufPool;
-    unsigned long long fullFrameSize;
-} ZSTDMT_jobDescription;
-
-/* ZSTDMT_compressChunk() : POOL_function type */
-void ZSTDMT_compressChunk(void* jobDescription)
-{
-    ZSTDMT_jobDescription* const job = (ZSTDMT_jobDescription*)jobDescription;
-    ZSTD_CCtx* cctx = ZSTDMT_getCCtx(job->cctxPool);
-    const void* const src = (const char*)job->srcStart + job->dictSize;
-    buffer_t dstBuff = job->dstBuff;
-    DEBUGLOG(5, "job (first:%u) (last:%u) : dictSize %u, srcSize %u",
-                 job->firstChunk, job->lastChunk, (U32)job->dictSize, (U32)job->srcSize);
-
-    if (cctx==NULL) {
-        job->cSize = ERROR(memory_allocation);
-        goto _endJob;
-    }
-
-    if (dstBuff.start == NULL) {
-        dstBuff = ZSTDMT_getBuffer(job->bufPool);
-        if (dstBuff.start==NULL) {
-            job->cSize = ERROR(memory_allocation);
-            goto _endJob;
-        }
-        job->dstBuff = dstBuff;
-    }
-
-    if (job->cdict) {  /* should only happen for first segment */
-        size_t const initError = ZSTD_compressBegin_usingCDict_advanced(cctx, job->cdict, job->params.fParams, job->fullFrameSize);
-        DEBUGLOG(5, "using CDict");
-        if (ZSTD_isError(initError)) { job->cSize = initError; goto _endJob; }
-    } else {  /* srcStart points at reloaded section */
-        if (!job->firstChunk) job->params.fParams.contentSizeFlag = 0;  /* ensure no srcSize control */
-        { ZSTD_CCtx_params jobParams = job->params;
-          size_t const forceWindowError =
-              ZSTD_CCtxParam_setParameter(&jobParams, ZSTD_p_forceMaxWindow, !job->firstChunk);
-          /* Force loading dictionary in "content-only" mode (no header analysis) */
-          size_t const initError = ZSTD_compressBegin_advanced_internal(cctx, job->srcStart, job->dictSize, ZSTD_dm_rawContent, jobParams, job->fullFrameSize);
-            if (ZSTD_isError(initError) || ZSTD_isError(forceWindowError)) {
-                job->cSize = initError;
-                goto _endJob;
-            }
-    }   }
-    if (!job->firstChunk) {  /* flush and overwrite frame header when it's not first segment */
-        size_t const hSize = ZSTD_compressContinue(cctx, dstBuff.start, dstBuff.size, src, 0);
-        if (ZSTD_isError(hSize)) { job->cSize = hSize; goto _endJob; }
-        ZSTD_invalidateRepCodes(cctx);
-    }
-
-    DEBUGLOG(5, "Compressing : ");
-    DEBUG_PRINTHEX(4, job->srcStart, 12);
-    job->cSize = (job->lastChunk) ?
-                 ZSTD_compressEnd     (cctx, dstBuff.start, dstBuff.size, src, job->srcSize) :
-                 ZSTD_compressContinue(cctx, dstBuff.start, dstBuff.size, src, job->srcSize);
-    DEBUGLOG(5, "compressed %u bytes into %u bytes   (first:%u) (last:%u)",
-                (unsigned)job->srcSize, (unsigned)job->cSize, job->firstChunk, job->lastChunk);
-    DEBUGLOG(5, "dstBuff.size : %u ; => %s", (U32)dstBuff.size, ZSTD_getErrorName(job->cSize));
-
-_endJob:
-    ZSTDMT_releaseCCtx(job->cctxPool, cctx);
-    ZSTDMT_releaseBuffer(job->bufPool, job->src);
-    job->src = g_nullBuffer; job->srcStart = NULL;
-    PTHREAD_MUTEX_LOCK(job->jobCompleted_mutex);
-    job->jobCompleted = 1;
-    job->jobScanned = 0;
-    pthread_cond_signal(job->jobCompleted_cond);
-    pthread_mutex_unlock(job->jobCompleted_mutex);
-}
-
-
-/* ------------------------------------------ */
-/* =====   Multi-threaded compression   ===== */
-/* ------------------------------------------ */
-
-typedef struct {
-    buffer_t buffer;
-    size_t filled;
-} inBuff_t;
-
-struct ZSTDMT_CCtx_s {
-    POOL_ctx* factory;
-    ZSTDMT_jobDescription* jobs;
-    ZSTDMT_bufferPool* bufPool;
-    ZSTDMT_CCtxPool* cctxPool;
-    pthread_mutex_t jobCompleted_mutex;
-    pthread_cond_t jobCompleted_cond;
-    size_t targetSectionSize;
-    size_t inBuffSize;
-    size_t dictSize;
-    size_t targetDictSize;
-    inBuff_t inBuff;
-    ZSTD_CCtx_params params;
-    XXH64_state_t xxhState;
-    unsigned jobIDMask;
-    unsigned doneJobID;
-    unsigned nextJobID;
-    unsigned frameEnded;
-    unsigned allJobsCompleted;
-    unsigned long long frameContentSize;
-    ZSTD_customMem cMem;
-    ZSTD_CDict* cdictLocal;
-    const ZSTD_CDict* cdict;
-};
-
-static ZSTDMT_jobDescription* ZSTDMT_allocJobsTable(U32* nbJobsPtr, ZSTD_customMem cMem)
-{
-    U32 const nbJobsLog2 = ZSTD_highbit32(*nbJobsPtr) + 1;
-    U32 const nbJobs = 1 << nbJobsLog2;
-    *nbJobsPtr = nbJobs;
-    return (ZSTDMT_jobDescription*) ZSTD_calloc(
-                            nbJobs * sizeof(ZSTDMT_jobDescription), cMem);
-}
-
-/* Internal only */
-size_t ZSTDMT_initializeCCtxParameters(ZSTD_CCtx_params* params, unsigned nbThreads)
-{
-    params->nbThreads = nbThreads;
-    params->overlapSizeLog = ZSTDMT_OVERLAPLOG_DEFAULT;
-    params->jobSize = 0;
-    return 0;
-}
-
-ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbThreads, ZSTD_customMem cMem)
-{
-    ZSTDMT_CCtx* mtctx;
-    U32 nbJobs = nbThreads + 2;
-    DEBUGLOG(3, "ZSTDMT_createCCtx_advanced");
-
-    if (nbThreads < 1) return NULL;
-    nbThreads = MIN(nbThreads , ZSTDMT_NBTHREADS_MAX);
-    if ((cMem.customAlloc!=NULL) ^ (cMem.customFree!=NULL))
-        /* invalid custom allocator */
-        return NULL;
-
-    mtctx = (ZSTDMT_CCtx*) ZSTD_calloc(sizeof(ZSTDMT_CCtx), cMem);
-    if (!mtctx) return NULL;
-    ZSTDMT_initializeCCtxParameters(&mtctx->params, nbThreads);
-    mtctx->cMem = cMem;
-    mtctx->allJobsCompleted = 1;
-    mtctx->factory = POOL_create_advanced(nbThreads, 0, cMem);
-    mtctx->jobs = ZSTDMT_allocJobsTable(&nbJobs, cMem);
-    mtctx->jobIDMask = nbJobs - 1;
-    mtctx->bufPool = ZSTDMT_createBufferPool(nbThreads, cMem);
-    mtctx->cctxPool = ZSTDMT_createCCtxPool(nbThreads, cMem);
-    if (!mtctx->factory | !mtctx->jobs | !mtctx->bufPool | !mtctx->cctxPool) {
-        ZSTDMT_freeCCtx(mtctx);
-        return NULL;
-    }
-    if (pthread_mutex_init(&mtctx->jobCompleted_mutex, NULL)) {
-        ZSTDMT_freeCCtx(mtctx);
-        return NULL;
-    }
-    if (pthread_cond_init(&mtctx->jobCompleted_cond, NULL)) {
-        ZSTDMT_freeCCtx(mtctx);
-        return NULL;
-    }
-    DEBUGLOG(3, "mt_cctx created, for %u threads", nbThreads);
-    return mtctx;
-}
-
-ZSTDMT_CCtx* ZSTDMT_createCCtx(unsigned nbThreads)
-{
-    return ZSTDMT_createCCtx_advanced(nbThreads, ZSTD_defaultCMem);
-}
-
-/* ZSTDMT_releaseAllJobResources() :
- * note : ensure all workers are killed first ! */
-static void ZSTDMT_releaseAllJobResources(ZSTDMT_CCtx* mtctx)
-{
-    unsigned jobID;
-    DEBUGLOG(3, "ZSTDMT_releaseAllJobResources");
-    for (jobID=0; jobID <= mtctx->jobIDMask; jobID++) {
-        ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].dstBuff);
-        mtctx->jobs[jobID].dstBuff = g_nullBuffer;
-        ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[jobID].src);
-        mtctx->jobs[jobID].src = g_nullBuffer;
-    }
-    memset(mtctx->jobs, 0, (mtctx->jobIDMask+1)*sizeof(ZSTDMT_jobDescription));
-    ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->inBuff.buffer);
-    mtctx->inBuff.buffer = g_nullBuffer;
-    mtctx->allJobsCompleted = 1;
-}
-
-size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx)
-{
-    if (mtctx==NULL) return 0;   /* compatible with free on NULL */
-    POOL_free(mtctx->factory);
-    if (!mtctx->allJobsCompleted) ZSTDMT_releaseAllJobResources(mtctx); /* stop workers first */
-    ZSTDMT_freeBufferPool(mtctx->bufPool);  /* release job resources into pools first */
-    ZSTD_free(mtctx->jobs, mtctx->cMem);
-    ZSTDMT_freeCCtxPool(mtctx->cctxPool);
-    ZSTD_freeCDict(mtctx->cdictLocal);
-    pthread_mutex_destroy(&mtctx->jobCompleted_mutex);
-    pthread_cond_destroy(&mtctx->jobCompleted_cond);
-    ZSTD_free(mtctx, mtctx->cMem);
-    return 0;
-}
-
-size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx)
-{
-    if (mtctx == NULL) return 0;   /* supports sizeof NULL */
-    return sizeof(*mtctx)
-            + POOL_sizeof(mtctx->factory)
-            + ZSTDMT_sizeof_bufferPool(mtctx->bufPool)
-            + (mtctx->jobIDMask+1) * sizeof(ZSTDMT_jobDescription)
-            + ZSTDMT_sizeof_CCtxPool(mtctx->cctxPool)
-            + ZSTD_sizeof_CDict(mtctx->cdictLocal);
-}
-
-/* Internal only */
-size_t ZSTDMT_CCtxParam_setMTCtxParameter(
-    ZSTD_CCtx_params* params, ZSTDMT_parameter parameter, unsigned value) {
-    switch(parameter)
-    {
-    case ZSTDMT_p_sectionSize :
-        params->jobSize = value;
-        return 0;
-    case ZSTDMT_p_overlapSectionLog :
-        DEBUGLOG(4, "ZSTDMT_p_overlapSectionLog : %u", value);
-        params->overlapSizeLog = (value >= 9) ? 9 : value;
-        return 0;
-    default :
-        return ERROR(parameter_unsupported);
-    }
-}
-
-size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, unsigned value)
-{
-    switch(parameter)
-    {
-    case ZSTDMT_p_sectionSize :
-        return ZSTDMT_CCtxParam_setMTCtxParameter(&mtctx->params, parameter, value);
-    case ZSTDMT_p_overlapSectionLog :
-        return ZSTDMT_CCtxParam_setMTCtxParameter(&mtctx->params, parameter, value);
-    default :
-        return ERROR(parameter_unsupported);
-    }
-}
-
-/* ------------------------------------------ */
-/* =====   Multi-threaded compression   ===== */
-/* ------------------------------------------ */
-
-static unsigned computeNbChunks(size_t srcSize, unsigned windowLog, unsigned nbThreads) {
-    size_t const chunkSizeTarget = (size_t)1 << (windowLog + 2);
-    size_t const chunkMaxSize = chunkSizeTarget << 2;
-    size_t const passSizeMax = chunkMaxSize * nbThreads;
-    unsigned const multiplier = (unsigned)(srcSize / passSizeMax) + 1;
-    unsigned const nbChunksLarge = multiplier * nbThreads;
-    unsigned const nbChunksMax = (unsigned)(srcSize / chunkSizeTarget) + 1;
-    unsigned const nbChunksSmall = MIN(nbChunksMax, nbThreads);
-    return (multiplier>1) ? nbChunksLarge : nbChunksSmall;
-}
-
-static size_t ZSTDMT_compress_advanced_internal(
-                ZSTDMT_CCtx* mtctx,
-                void* dst, size_t dstCapacity,
-          const void* src, size_t srcSize,
-          const ZSTD_CDict* cdict,
-                ZSTD_CCtx_params const params)
-{
-    ZSTD_CCtx_params const jobParams = ZSTDMT_makeJobCCtxParams(params);
-    unsigned const overlapRLog = (params.overlapSizeLog>9) ? 0 : 9-params.overlapSizeLog;
-    size_t const overlapSize = (overlapRLog>=9) ? 0 : (size_t)1 << (params.cParams.windowLog - overlapRLog);
-    unsigned nbChunks = computeNbChunks(srcSize, params.cParams.windowLog, params.nbThreads);
-    size_t const proposedChunkSize = (srcSize + (nbChunks-1)) / nbChunks;
-    size_t const avgChunkSize = ((proposedChunkSize & 0x1FFFF) < 0x7FFF) ? proposedChunkSize + 0xFFFF : proposedChunkSize;   /* avoid too small last block */
-    const char* const srcStart = (const char*)src;
-    size_t remainingSrcSize = srcSize;
-    unsigned const compressWithinDst = (dstCapacity >= ZSTD_compressBound(srcSize)) ? nbChunks : (unsigned)(dstCapacity / ZSTD_compressBound(avgChunkSize));  /* presumes avgChunkSize >= 256 KB, which should be the case */
-    size_t frameStartPos = 0, dstBufferPos = 0;
-    XXH64_state_t xxh64;
-    assert(jobParams.nbThreads == 0);
-    assert(mtctx->cctxPool->totalCCtx == params.nbThreads);
-
-    DEBUGLOG(4, "nbChunks  : %2u   (chunkSize : %u bytes)   ", nbChunks, (U32)avgChunkSize);
-    if (nbChunks==1) {   /* fallback to single-thread mode */
-        ZSTD_CCtx* const cctx = mtctx->cctxPool->cctx[0];
-        if (cdict) return ZSTD_compress_usingCDict_advanced(cctx, dst, dstCapacity, src, srcSize, cdict, jobParams.fParams);
-        return ZSTD_compress_advanced_internal(cctx, dst, dstCapacity, src, srcSize, NULL, 0, jobParams);
-    }
-    assert(avgChunkSize >= 256 KB);  /* condition for ZSTD_compressBound(A) + ZSTD_compressBound(B) <= ZSTD_compressBound(A+B), which is required for compressWithinDst */
-    ZSTDMT_setBufferSize(mtctx->bufPool, ZSTD_compressBound(avgChunkSize) );
-    XXH64_reset(&xxh64, 0);
-
-    if (nbChunks > mtctx->jobIDMask+1) {  /* enlarge job table */
-        U32 nbJobs = nbChunks;
-        ZSTD_free(mtctx->jobs, mtctx->cMem);
-        mtctx->jobIDMask = 0;
-        mtctx->jobs = ZSTDMT_allocJobsTable(&nbJobs, mtctx->cMem);
-        if (mtctx->jobs==NULL) return ERROR(memory_allocation);
-        mtctx->jobIDMask = nbJobs - 1;
-    }
-
-    {   unsigned u;
-        for (u=0; u<nbChunks; u++) {
-            size_t const chunkSize = MIN(remainingSrcSize, avgChunkSize);
-            size_t const dstBufferCapacity = ZSTD_compressBound(chunkSize);
-            buffer_t const dstAsBuffer = { (char*)dst + dstBufferPos, dstBufferCapacity };
-            buffer_t const dstBuffer = u < compressWithinDst ? dstAsBuffer : g_nullBuffer;
-            size_t dictSize = u ? overlapSize : 0;
-
-            mtctx->jobs[u].src = g_nullBuffer;
-            mtctx->jobs[u].srcStart = srcStart + frameStartPos - dictSize;
-            mtctx->jobs[u].dictSize = dictSize;
-            mtctx->jobs[u].srcSize = chunkSize;
-            mtctx->jobs[u].cdict = mtctx->nextJobID==0 ? cdict : NULL;
-            mtctx->jobs[u].fullFrameSize = srcSize;
-            mtctx->jobs[u].params = jobParams;
-            /* do not calculate checksum within sections, but write it in header for first section */
-            if (u!=0) mtctx->jobs[u].params.fParams.checksumFlag = 0;
-            mtctx->jobs[u].dstBuff = dstBuffer;
-            mtctx->jobs[u].cctxPool = mtctx->cctxPool;
-            mtctx->jobs[u].bufPool = mtctx->bufPool;
-            mtctx->jobs[u].firstChunk = (u==0);
-            mtctx->jobs[u].lastChunk = (u==nbChunks-1);
-            mtctx->jobs[u].jobCompleted = 0;
-            mtctx->jobs[u].jobCompleted_mutex = &mtctx->jobCompleted_mutex;
-            mtctx->jobs[u].jobCompleted_cond = &mtctx->jobCompleted_cond;
-
-            if (params.fParams.checksumFlag) {
-                XXH64_update(&xxh64, srcStart + frameStartPos, chunkSize);
-            }
-
-            DEBUGLOG(5, "posting job %u   (%u bytes)", u, (U32)chunkSize);
-            DEBUG_PRINTHEX(6, mtctx->jobs[u].srcStart, 12);
-            POOL_add(mtctx->factory, ZSTDMT_compressChunk, &mtctx->jobs[u]);
-
-            frameStartPos += chunkSize;
-            dstBufferPos += dstBufferCapacity;
-            remainingSrcSize -= chunkSize;
-    }   }
-
-    /* collect result */
-    {   size_t error = 0, dstPos = 0;
-        unsigned chunkID;
-        for (chunkID=0; chunkID<nbChunks; chunkID++) {
-            DEBUGLOG(5, "waiting for chunk %u ", chunkID);
-            PTHREAD_MUTEX_LOCK(&mtctx->jobCompleted_mutex);
-            while (mtctx->jobs[chunkID].jobCompleted==0) {
-                DEBUGLOG(5, "waiting for jobCompleted signal from chunk %u", chunkID);
-                pthread_cond_wait(&mtctx->jobCompleted_cond, &mtctx->jobCompleted_mutex);
-            }
-            pthread_mutex_unlock(&mtctx->jobCompleted_mutex);
-            DEBUGLOG(5, "ready to write chunk %u ", chunkID);
-
-            mtctx->jobs[chunkID].srcStart = NULL;
-            {   size_t const cSize = mtctx->jobs[chunkID].cSize;
-                if (ZSTD_isError(cSize)) error = cSize;
-                if ((!error) && (dstPos + cSize > dstCapacity)) error = ERROR(dstSize_tooSmall);
-                if (chunkID) {   /* note : chunk 0 is written directly at dst, which is correct position */
-                    if (!error)
-                        memmove((char*)dst + dstPos, mtctx->jobs[chunkID].dstBuff.start, cSize);  /* may overlap when chunk compressed within dst */
-                    if (chunkID >= compressWithinDst) {  /* chunk compressed into its own buffer, which must be released */
-                        DEBUGLOG(5, "releasing buffer %u>=%u", chunkID, compressWithinDst);
-                        ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->jobs[chunkID].dstBuff);
-                    }
-                    mtctx->jobs[chunkID].dstBuff = g_nullBuffer;
-                }
-                dstPos += cSize ;
-            }
-        }  /* for (chunkID=0; chunkID<nbChunks; chunkID++) */
-
-        DEBUGLOG(4, "checksumFlag : %u ", params.fParams.checksumFlag);
-        if (params.fParams.checksumFlag) {
-            U32 const checksum = (U32)XXH64_digest(&xxh64);
-            if (dstPos + 4 > dstCapacity) {
-                error = ERROR(dstSize_tooSmall);
-            } else {
-                DEBUGLOG(4, "writing checksum : %08X \n", checksum);
-                MEM_writeLE32((char*)dst + dstPos, checksum);
-                dstPos += 4;
-        }   }
-
-        if (!error) DEBUGLOG(4, "compressed size : %u  ", (U32)dstPos);
-        return error ? error : dstPos;
-    }
-}
-
-size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
-                               void* dst, size_t dstCapacity,
-                         const void* src, size_t srcSize,
-                         const ZSTD_CDict* cdict,
-                               ZSTD_parameters const params,
-                               unsigned overlapLog)
-{
-    ZSTD_CCtx_params cctxParams = mtctx->params;
-    cctxParams.cParams = params.cParams;
-    cctxParams.fParams = params.fParams;
-    cctxParams.overlapSizeLog = overlapLog;
-    return ZSTDMT_compress_advanced_internal(mtctx,
-                                             dst, dstCapacity,
-                                             src, srcSize,
-                                             cdict, cctxParams);
-}
-
-
-size_t ZSTDMT_compressCCtx(ZSTDMT_CCtx* mtctx,
-                           void* dst, size_t dstCapacity,
-                     const void* src, size_t srcSize,
-                           int compressionLevel)
-{
-    U32 const overlapLog = (compressionLevel >= ZSTD_maxCLevel()) ? 9 : ZSTDMT_OVERLAPLOG_DEFAULT;
-    ZSTD_parameters params = ZSTD_getParams(compressionLevel, srcSize, 0);
-    params.fParams.contentSizeFlag = 1;
-    return ZSTDMT_compress_advanced(mtctx, dst, dstCapacity, src, srcSize, NULL, params, overlapLog);
-}
-
-
-/* ====================================== */
-/* =======      Streaming API     ======= */
-/* ====================================== */
-
-static void ZSTDMT_waitForAllJobsCompleted(ZSTDMT_CCtx* zcs)
-{
-    DEBUGLOG(4, "ZSTDMT_waitForAllJobsCompleted");
-    while (zcs->doneJobID < zcs->nextJobID) {
-        unsigned const jobID = zcs->doneJobID & zcs->jobIDMask;
-        PTHREAD_MUTEX_LOCK(&zcs->jobCompleted_mutex);
-        while (zcs->jobs[jobID].jobCompleted==0) {
-            DEBUGLOG(5, "waiting for jobCompleted signal from chunk %u", zcs->doneJobID);   /* we want to block when waiting for data to flush */
-            pthread_cond_wait(&zcs->jobCompleted_cond, &zcs->jobCompleted_mutex);
-        }
-        pthread_mutex_unlock(&zcs->jobCompleted_mutex);
-        zcs->doneJobID++;
-    }
-}
-
-size_t ZSTDMT_initCStream_internal(
-        ZSTDMT_CCtx* zcs,
-        const void* dict, size_t dictSize, ZSTD_dictMode_e dictMode,
-        const ZSTD_CDict* cdict, ZSTD_CCtx_params params,
-        unsigned long long pledgedSrcSize)
-{
-    DEBUGLOG(4, "ZSTDMT_initCStream_internal");
-    /* params are supposed to be fully validated at this point */
-    assert(!ZSTD_isError(ZSTD_checkCParams(params.cParams)));
-    assert(!((dict) && (cdict)));  /* either dict or cdict, not both */
-    assert(zcs->cctxPool->totalCCtx == params.nbThreads);
-
-    if (params.nbThreads==1) {
-        ZSTD_CCtx_params const singleThreadParams = ZSTDMT_makeJobCCtxParams(params);
-        DEBUGLOG(4, "single thread mode");
-        assert(singleThreadParams.nbThreads == 0);
-        return ZSTD_initCStream_internal(zcs->cctxPool->cctx[0],
-                                         dict, dictSize, cdict,
-                                         singleThreadParams, pledgedSrcSize);
-    }
-
-    if (zcs->allJobsCompleted == 0) {   /* previous compression not correctly finished */
-        ZSTDMT_waitForAllJobsCompleted(zcs);
-        ZSTDMT_releaseAllJobResources(zcs);
-        zcs->allJobsCompleted = 1;
-    }
-
-    zcs->params = params;
-    zcs->frameContentSize = pledgedSrcSize;
-    if (dict) {
-        DEBUGLOG(4,"cdictLocal: %08X", (U32)(size_t)zcs->cdictLocal);
-        ZSTD_freeCDict(zcs->cdictLocal);
-        zcs->cdictLocal = ZSTD_createCDict_advanced(dict, dictSize,
-                                                    ZSTD_dlm_byCopy, dictMode, /* note : a loadPrefix becomes an internal CDict */
-                                                    params.cParams, zcs->cMem);
-        zcs->cdict = zcs->cdictLocal;
-        if (zcs->cdictLocal == NULL) return ERROR(memory_allocation);
-    } else {
-        DEBUGLOG(4,"cdictLocal: %08X", (U32)(size_t)zcs->cdictLocal);
-        ZSTD_freeCDict(zcs->cdictLocal);
-        zcs->cdictLocal = NULL;
-        zcs->cdict = cdict;
-    }
-
-    zcs->targetDictSize = (params.overlapSizeLog==0) ? 0 : (size_t)1 << (params.cParams.windowLog - (9 - params.overlapSizeLog));
-    DEBUGLOG(4, "overlapLog : %u ", params.overlapSizeLog);
-    DEBUGLOG(4, "overlap Size : %u KB", (U32)(zcs->targetDictSize>>10));
-    zcs->targetSectionSize = params.jobSize ? params.jobSize : (size_t)1 << (params.cParams.windowLog + 2);
-    zcs->targetSectionSize = MAX(ZSTDMT_SECTION_SIZE_MIN, zcs->targetSectionSize);
-    zcs->targetSectionSize = MAX(zcs->targetDictSize, zcs->targetSectionSize);
-    DEBUGLOG(4, "Section Size : %u KB", (U32)(zcs->targetSectionSize>>10));
-    zcs->inBuffSize = zcs->targetDictSize + zcs->targetSectionSize;
-    ZSTDMT_setBufferSize(zcs->bufPool, MAX(zcs->inBuffSize, ZSTD_compressBound(zcs->targetSectionSize)) );
-    zcs->inBuff.buffer = g_nullBuffer;
-    zcs->dictSize = 0;
-    zcs->doneJobID = 0;
-    zcs->nextJobID = 0;
-    zcs->frameEnded = 0;
-    zcs->allJobsCompleted = 0;
-    if (params.fParams.checksumFlag) XXH64_reset(&zcs->xxhState, 0);
-    return 0;
-}
-
-size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* mtctx,
-                             const void* dict, size_t dictSize,
-                                   ZSTD_parameters params,
-                                   unsigned long long pledgedSrcSize)
-{
-    ZSTD_CCtx_params cctxParams = mtctx->params;
-    DEBUGLOG(5, "ZSTDMT_initCStream_advanced");
-    cctxParams.cParams = params.cParams;
-    cctxParams.fParams = params.fParams;
-    return ZSTDMT_initCStream_internal(mtctx, dict, dictSize, ZSTD_dm_auto, NULL,
-                                       cctxParams, pledgedSrcSize);
-}
-
-size_t ZSTDMT_initCStream_usingCDict(ZSTDMT_CCtx* mtctx,
-                               const ZSTD_CDict* cdict,
-                                     ZSTD_frameParameters fParams,
-                                     unsigned long long pledgedSrcSize)
-{
-    ZSTD_CCtx_params cctxParams = mtctx->params;
-    cctxParams.cParams = ZSTD_getCParamsFromCDict(cdict);
-    cctxParams.fParams = fParams;
-    if (cdict==NULL) return ERROR(dictionary_wrong);   /* method incompatible with NULL cdict */
-    return ZSTDMT_initCStream_internal(mtctx, NULL, 0 /*dictSize*/, ZSTD_dm_auto, cdict,
-                                       cctxParams, pledgedSrcSize);
-}
-
-
-/* ZSTDMT_resetCStream() :
- * pledgedSrcSize is optional and can be zero == unknown */
-size_t ZSTDMT_resetCStream(ZSTDMT_CCtx* zcs, unsigned long long pledgedSrcSize)
-{
-    if (zcs->params.nbThreads==1)
-        return ZSTD_resetCStream(zcs->cctxPool->cctx[0], pledgedSrcSize);
-    return ZSTDMT_initCStream_internal(zcs, NULL, 0, ZSTD_dm_auto, 0, zcs->params,
-                                       pledgedSrcSize);
-}
-
-size_t ZSTDMT_initCStream(ZSTDMT_CCtx* zcs, int compressionLevel) {
-    ZSTD_parameters const params = ZSTD_getParams(compressionLevel, 0, 0);
-    ZSTD_CCtx_params cctxParams = zcs->params;
-    cctxParams.cParams = params.cParams;
-    cctxParams.fParams = params.fParams;
-    return ZSTDMT_initCStream_internal(zcs, NULL, 0, ZSTD_dm_auto, NULL, cctxParams, 0);
-}
-
-
-static size_t ZSTDMT_createCompressionJob(ZSTDMT_CCtx* zcs, size_t srcSize, unsigned endFrame)
-{
-    unsigned const jobID = zcs->nextJobID & zcs->jobIDMask;
-
-    DEBUGLOG(4, "preparing job %u to compress %u bytes with %u preload ",
-                zcs->nextJobID, (U32)srcSize, (U32)zcs->dictSize);
-    zcs->jobs[jobID].src = zcs->inBuff.buffer;
-    zcs->jobs[jobID].srcStart = zcs->inBuff.buffer.start;
-    zcs->jobs[jobID].srcSize = srcSize;
-    zcs->jobs[jobID].dictSize = zcs->dictSize;
-    assert(zcs->inBuff.filled >= srcSize + zcs->dictSize);
-    zcs->jobs[jobID].params = zcs->params;
-    /* do not calculate checksum within sections, but write it in header for first section */
-    if (zcs->nextJobID) zcs->jobs[jobID].params.fParams.checksumFlag = 0;
-    zcs->jobs[jobID].cdict = zcs->nextJobID==0 ? zcs->cdict : NULL;
-    zcs->jobs[jobID].fullFrameSize = zcs->frameContentSize;
-    zcs->jobs[jobID].dstBuff = g_nullBuffer;
-    zcs->jobs[jobID].cctxPool = zcs->cctxPool;
-    zcs->jobs[jobID].bufPool = zcs->bufPool;
-    zcs->jobs[jobID].firstChunk = (zcs->nextJobID==0);
-    zcs->jobs[jobID].lastChunk = endFrame;
-    zcs->jobs[jobID].jobCompleted = 0;
-    zcs->jobs[jobID].dstFlushed = 0;
-    zcs->jobs[jobID].jobCompleted_mutex = &zcs->jobCompleted_mutex;
-    zcs->jobs[jobID].jobCompleted_cond = &zcs->jobCompleted_cond;
-
-    if (zcs->params.fParams.checksumFlag)
-        XXH64_update(&zcs->xxhState, (const char*)zcs->inBuff.buffer.start + zcs->dictSize, srcSize);
-
-    /* get a new buffer for next input */
-    if (!endFrame) {
-        size_t const newDictSize = MIN(srcSize + zcs->dictSize, zcs->targetDictSize);
-        zcs->inBuff.buffer = ZSTDMT_getBuffer(zcs->bufPool);
-        if (zcs->inBuff.buffer.start == NULL) {   /* not enough memory to allocate next input buffer */
-            zcs->jobs[jobID].jobCompleted = 1;
-            zcs->nextJobID++;
-            ZSTDMT_waitForAllJobsCompleted(zcs);
-            ZSTDMT_releaseAllJobResources(zcs);
-            return ERROR(memory_allocation);
-        }
-        zcs->inBuff.filled -= srcSize + zcs->dictSize - newDictSize;
-        memmove(zcs->inBuff.buffer.start,
-            (const char*)zcs->jobs[jobID].srcStart + zcs->dictSize + srcSize - newDictSize,
-            zcs->inBuff.filled);
-        zcs->dictSize = newDictSize;
-    } else {   /* if (endFrame==1) */
-        zcs->inBuff.buffer = g_nullBuffer;
-        zcs->inBuff.filled = 0;
-        zcs->dictSize = 0;
-        zcs->frameEnded = 1;
-        if (zcs->nextJobID == 0) {
-            /* single chunk exception : checksum is calculated directly within worker thread */
-            zcs->params.fParams.checksumFlag = 0;
-    }   }
-
-    DEBUGLOG(4, "posting job %u : %u bytes  (end:%u) (note : doneJob = %u=>%u)",
-                zcs->nextJobID,
-                (U32)zcs->jobs[jobID].srcSize,
-                zcs->jobs[jobID].lastChunk,
-                zcs->doneJobID,
-                zcs->doneJobID & zcs->jobIDMask);
-    POOL_add(zcs->factory, ZSTDMT_compressChunk, &zcs->jobs[jobID]);   /* this call is blocking when thread worker pool is exhausted */
-    zcs->nextJobID++;
-    return 0;
-}
-
-
-/* ZSTDMT_flushNextJob() :
- * output : will be updated with amount of data flushed .
- * blockToFlush : if >0, the function will block and wait if there is no data available to flush .
- * @return : amount of data remaining within internal buffer, 1 if unknown but > 0, 0 if no more, or an error code */
-static size_t ZSTDMT_flushNextJob(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, unsigned blockToFlush)
-{
-    unsigned const wJobID = zcs->doneJobID & zcs->jobIDMask;
-    if (zcs->doneJobID == zcs->nextJobID) return 0;   /* all flushed ! */
-    PTHREAD_MUTEX_LOCK(&zcs->jobCompleted_mutex);
-    while (zcs->jobs[wJobID].jobCompleted==0) {
-        DEBUGLOG(5, "waiting for jobCompleted signal from job %u", zcs->doneJobID);
-        if (!blockToFlush) { pthread_mutex_unlock(&zcs->jobCompleted_mutex); return 0; }  /* nothing ready to be flushed => skip */
-        pthread_cond_wait(&zcs->jobCompleted_cond, &zcs->jobCompleted_mutex);  /* block when nothing available to flush */
-    }
-    pthread_mutex_unlock(&zcs->jobCompleted_mutex);
-    /* compression job completed : output can be flushed */
-    {   ZSTDMT_jobDescription job = zcs->jobs[wJobID];
-        if (!job.jobScanned) {
-            if (ZSTD_isError(job.cSize)) {
-                DEBUGLOG(5, "compression error detected ");
-                ZSTDMT_waitForAllJobsCompleted(zcs);
-                ZSTDMT_releaseAllJobResources(zcs);
-                return job.cSize;
-            }
-            DEBUGLOG(5, "zcs->params.fParams.checksumFlag : %u ", zcs->params.fParams.checksumFlag);
-            if (zcs->params.fParams.checksumFlag) {
-                if (zcs->frameEnded && (zcs->doneJobID+1 == zcs->nextJobID)) {  /* write checksum at end of last section */
-                    U32 const checksum = (U32)XXH64_digest(&zcs->xxhState);
-                    DEBUGLOG(5, "writing checksum : %08X \n", checksum);
-                    MEM_writeLE32((char*)job.dstBuff.start + job.cSize, checksum);
-                    job.cSize += 4;
-                    zcs->jobs[wJobID].cSize += 4;
-            }   }
-            zcs->jobs[wJobID].jobScanned = 1;
-        }
-        {   size_t const toWrite = MIN(job.cSize - job.dstFlushed, output->size - output->pos);
-            DEBUGLOG(5, "Flushing %u bytes from job %u ", (U32)toWrite, zcs->doneJobID);
-            memcpy((char*)output->dst + output->pos, (const char*)job.dstBuff.start + job.dstFlushed, toWrite);
-            output->pos += toWrite;
-            job.dstFlushed += toWrite;
-        }
-        if (job.dstFlushed == job.cSize) {   /* output buffer fully flushed => move to next one */
-            ZSTDMT_releaseBuffer(zcs->bufPool, job.dstBuff);
-            zcs->jobs[wJobID].dstBuff = g_nullBuffer;
-            zcs->jobs[wJobID].jobCompleted = 0;
-            zcs->doneJobID++;
-        } else {
-            zcs->jobs[wJobID].dstFlushed = job.dstFlushed;
-        }
-        /* return value : how many bytes left in buffer ; fake it to 1 if unknown but >0 */
-        if (job.cSize > job.dstFlushed) return (job.cSize - job.dstFlushed);
-        if (zcs->doneJobID < zcs->nextJobID) return 1;   /* still some buffer to flush */
-        zcs->allJobsCompleted = zcs->frameEnded;   /* frame completed and entirely flushed */
-        return 0;   /* everything flushed */
-}   }
-
-
-/** ZSTDMT_compressStream_generic() :
- *  internal use only
- *  assumption : output and input are valid (pos <= size)
- * @return : minimum amount of data remaining to flush, 0 if none */
-size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
-                                     ZSTD_outBuffer* output,
-                                     ZSTD_inBuffer* input,
-                                     ZSTD_EndDirective endOp)
-{
-    size_t const newJobThreshold = mtctx->dictSize + mtctx->targetSectionSize;
-    assert(output->pos <= output->size);
-    assert(input->pos  <= input->size);
-    if ((mtctx->frameEnded) && (endOp==ZSTD_e_continue)) {
-        /* current frame being ended. Only flush/end are allowed. Or start new frame with init */
-        return ERROR(stage_wrong);
-    }
-    if (mtctx->params.nbThreads==1) {  /* delegate to single-thread (synchronous) */
-        return ZSTD_compressStream_generic(mtctx->cctxPool->cctx[0], output, input, endOp);
-    }
-
-    /* single-pass shortcut (note : this is synchronous-mode) */
-    if ( (mtctx->nextJobID==0)      /* just started */
-      && (mtctx->inBuff.filled==0)  /* nothing buffered */
-      && (endOp==ZSTD_e_end)        /* end order */
-      && (output->size - output->pos >= ZSTD_compressBound(input->size - input->pos)) ) { /* enough room */
-        size_t const cSize = ZSTDMT_compress_advanced_internal(mtctx,
-                (char*)output->dst + output->pos, output->size - output->pos,
-                (const char*)input->src + input->pos, input->size - input->pos,
-                mtctx->cdict, mtctx->params);
-        if (ZSTD_isError(cSize)) return cSize;
-        input->pos = input->size;
-        output->pos += cSize;
-        ZSTDMT_releaseBuffer(mtctx->bufPool, mtctx->inBuff.buffer);  /* was allocated in initStream */
-        mtctx->allJobsCompleted = 1;
-        mtctx->frameEnded = 1;
-        return 0;
-    }
-
-    /* fill input buffer */
-    if (input->size > input->pos) {   /* support NULL input */
-        if (mtctx->inBuff.buffer.start == NULL) {
-            mtctx->inBuff.buffer = ZSTDMT_getBuffer(mtctx->bufPool);
-            if (mtctx->inBuff.buffer.start == NULL) return ERROR(memory_allocation);
-            mtctx->inBuff.filled = 0;
-        }
-        {   size_t const toLoad = MIN(input->size - input->pos, mtctx->inBuffSize - mtctx->inBuff.filled);
-            DEBUGLOG(5, "inBuff:%08X;  inBuffSize=%u;  ToCopy=%u", (U32)(size_t)mtctx->inBuff.buffer.start, (U32)mtctx->inBuffSize, (U32)toLoad);
-            memcpy((char*)mtctx->inBuff.buffer.start + mtctx->inBuff.filled, (const char*)input->src + input->pos, toLoad);
-            input->pos += toLoad;
-            mtctx->inBuff.filled += toLoad;
-    }   }
-
-    if ( (mtctx->inBuff.filled >= newJobThreshold)  /* filled enough : let's compress */
-      && (mtctx->nextJobID <= mtctx->doneJobID + mtctx->jobIDMask) ) {   /* avoid overwriting job round buffer */
-        CHECK_F( ZSTDMT_createCompressionJob(mtctx, mtctx->targetSectionSize, 0 /* endFrame */) );
-    }
-
-    /* check for potential compressed data ready to be flushed */
-    CHECK_F( ZSTDMT_flushNextJob(mtctx, output, (mtctx->inBuff.filled == mtctx->inBuffSize) /* blockToFlush */) ); /* block if it wasn't possible to create new job due to saturation */
-
-    if (input->pos < input->size)  /* input not consumed : do not flush yet */
-        endOp = ZSTD_e_continue;
-
-    switch(endOp)
-    {
-        case ZSTD_e_flush:
-            return ZSTDMT_flushStream(mtctx, output);
-        case ZSTD_e_end:
-            return ZSTDMT_endStream(mtctx, output);
-        case ZSTD_e_continue:
-            return 1;
-        default:
-            return ERROR(GENERIC);   /* invalid endDirective */
-    }
-}
-
-
-size_t ZSTDMT_compressStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, ZSTD_inBuffer* input)
-{
-    CHECK_F( ZSTDMT_compressStream_generic(zcs, output, input, ZSTD_e_continue) );
-
-    /* recommended next input size : fill current input buffer */
-    return zcs->inBuffSize - zcs->inBuff.filled;   /* note : could be zero when input buffer is fully filled and no more availability to create new job */
-}
-
-
-static size_t ZSTDMT_flushStream_internal(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output, unsigned endFrame)
-{
-    size_t const srcSize = zcs->inBuff.filled - zcs->dictSize;
-
-    if ( ((srcSize > 0) || (endFrame && !zcs->frameEnded))
-       && (zcs->nextJobID <= zcs->doneJobID + zcs->jobIDMask) ) {
-        CHECK_F( ZSTDMT_createCompressionJob(zcs, srcSize, endFrame) );
-    }
-
-    /* check if there is any data available to flush */
-    return ZSTDMT_flushNextJob(zcs, output, 1 /* blockToFlush */);
-}
-
-
-size_t ZSTDMT_flushStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output)
-{
-    DEBUGLOG(5, "ZSTDMT_flushStream");
-    if (zcs->params.nbThreads==1)
-        return ZSTD_flushStream(zcs->cctxPool->cctx[0], output);
-    return ZSTDMT_flushStream_internal(zcs, output, 0 /* endFrame */);
-}
-
-size_t ZSTDMT_endStream(ZSTDMT_CCtx* zcs, ZSTD_outBuffer* output)
-{
-    DEBUGLOG(4, "ZSTDMT_endStream");
-    if (zcs->params.nbThreads==1)
-        return ZSTD_endStream(zcs->cctxPool->cctx[0], output);
-    return ZSTDMT_flushStream_internal(zcs, output, 1 /* endFrame */);
-}
diff --git a/contrib/zstd/zstdmt_compress.h b/contrib/zstd/zstdmt_compress.h
deleted file mode 100644
index 5b3ddc9d1..000000000
--- a/contrib/zstd/zstdmt_compress.h
+++ /dev/null
@@ -1,134 +0,0 @@
-/*
- * Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
- * All rights reserved.
- *
- * This source code is licensed under both the BSD-style license (found in the
- * LICENSE file in the root directory of this source tree) and the GPLv2 (found
- * in the COPYING file in the root directory of this source tree).
- * You may select, at your option, one of the above-listed licenses.
- */
-
- #ifndef ZSTDMT_COMPRESS_H
- #define ZSTDMT_COMPRESS_H
-
- #if defined (__cplusplus)
- extern "C" {
- #endif
-
-
-/* Note : This is an internal API.
- *        Some methods are still exposed (ZSTDLIB_API),
- *        because it used to be the only way to invoke MT compression.
- *        Now, it's recommended to use ZSTD_compress_generic() instead.
- *        These methods will stop being exposed in a future version */
-
-/* ===   Dependencies   === */
-#include <stddef.h>                /* size_t */
-#ifndef ZSTD_STATIC_LINKING_ONLY
-#define ZSTD_STATIC_LINKING_ONLY   /* ZSTD_parameters */
-#endif
-#include "zstd.h"            /* ZSTD_inBuffer, ZSTD_outBuffer, ZSTDLIB_API */
-
-
-/* ===   Memory management   === */
-typedef struct ZSTDMT_CCtx_s ZSTDMT_CCtx;
-ZSTDLIB_API ZSTDMT_CCtx* ZSTDMT_createCCtx(unsigned nbThreads);
-ZSTDLIB_API ZSTDMT_CCtx* ZSTDMT_createCCtx_advanced(unsigned nbThreads,
-                                                    ZSTD_customMem cMem);
-ZSTDLIB_API size_t ZSTDMT_freeCCtx(ZSTDMT_CCtx* mtctx);
-
-ZSTDLIB_API size_t ZSTDMT_sizeof_CCtx(ZSTDMT_CCtx* mtctx);
-
-
-/* ===   Simple buffer-to-butter one-pass function   === */
-
-ZSTDLIB_API size_t ZSTDMT_compressCCtx(ZSTDMT_CCtx* mtctx,
-                                       void* dst, size_t dstCapacity,
-                                 const void* src, size_t srcSize,
-                                       int compressionLevel);
-
-
-
-/* ===   Streaming functions   === */
-
-ZSTDLIB_API size_t ZSTDMT_initCStream(ZSTDMT_CCtx* mtctx, int compressionLevel);
-ZSTDLIB_API size_t ZSTDMT_resetCStream(ZSTDMT_CCtx* mtctx, unsigned long long pledgedSrcSize);    /**< pledgedSrcSize is optional and can be zero == unknown */
-
-ZSTDLIB_API size_t ZSTDMT_compressStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output, ZSTD_inBuffer* input);
-
-ZSTDLIB_API size_t ZSTDMT_flushStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output);   /**< @return : 0 == all flushed; >0 : still some data to be flushed; or an error code (ZSTD_isError()) */
-ZSTDLIB_API size_t ZSTDMT_endStream(ZSTDMT_CCtx* mtctx, ZSTD_outBuffer* output);     /**< @return : 0 == all flushed; >0 : still some data to be flushed; or an error code (ZSTD_isError()) */
-
-
-/* ===   Advanced functions and parameters  === */
-
-#ifndef ZSTDMT_SECTION_SIZE_MIN
-#  define ZSTDMT_SECTION_SIZE_MIN (1U << 20)   /* 1 MB - Minimum size of each compression job */
-#endif
-
-ZSTDLIB_API size_t ZSTDMT_compress_advanced(ZSTDMT_CCtx* mtctx,
-                                           void* dst, size_t dstCapacity,
-                                     const void* src, size_t srcSize,
-                                     const ZSTD_CDict* cdict,
-                                           ZSTD_parameters const params,
-                                           unsigned overlapLog);
-
-ZSTDLIB_API size_t ZSTDMT_initCStream_advanced(ZSTDMT_CCtx* mtctx,
-                                        const void* dict, size_t dictSize,   /* dict can be released after init, a local copy is preserved within zcs */
-                                        ZSTD_parameters params,
-                                        unsigned long long pledgedSrcSize);  /* pledgedSrcSize is optional and can be zero == unknown */
-
-ZSTDLIB_API size_t ZSTDMT_initCStream_usingCDict(ZSTDMT_CCtx* mtctx,
-                                        const ZSTD_CDict* cdict,
-                                        ZSTD_frameParameters fparams,
-                                        unsigned long long pledgedSrcSize);  /* note : zero means empty */
-
-/* ZSTDMT_parameter :
- * List of parameters that can be set using ZSTDMT_setMTCtxParameter() */
-typedef enum {
-    ZSTDMT_p_sectionSize,        /* size of input "section". Each section is compressed in parallel. 0 means default, which is dynamically determined within compression functions */
-    ZSTDMT_p_overlapSectionLog   /* Log of overlapped section; 0 == no overlap, 6(default) == use 1/8th of window, >=9 == use full window */
-} ZSTDMT_parameter;
-
-/* ZSTDMT_setMTCtxParameter() :
- * allow setting individual parameters, one at a time, among a list of enums defined in ZSTDMT_parameter.
- * The function must be called typically after ZSTD_createCCtx().
- * Parameters not explicitly reset by ZSTDMT_init*() remain the same in consecutive compression sessions.
- * @return : 0, or an error code (which can be tested using ZSTD_isError()) */
-ZSTDLIB_API size_t ZSTDMT_setMTCtxParameter(ZSTDMT_CCtx* mtctx, ZSTDMT_parameter parameter, unsigned value);
-
-
-/*! ZSTDMT_compressStream_generic() :
- *  Combines ZSTDMT_compressStream() with ZSTDMT_flushStream() or ZSTDMT_endStream()
- *  depending on flush directive.
- * @return : minimum amount of data still to be flushed
- *           0 if fully flushed
- *           or an error code */
-ZSTDLIB_API size_t ZSTDMT_compressStream_generic(ZSTDMT_CCtx* mtctx,
-                                                ZSTD_outBuffer* output,
-                                                ZSTD_inBuffer* input,
-                                                ZSTD_EndDirective endOp);
-
-
-/* ===   Private definitions; never ever use directly  === */
-
-size_t ZSTDMT_CCtxParam_setMTCtxParameter(ZSTD_CCtx_params* params, ZSTDMT_parameter parameter, unsigned value);
-
-size_t ZSTDMT_initializeCCtxParameters(ZSTD_CCtx_params* params, unsigned nbThreads);
-
-/*! ZSTDMT_initCStream_internal() :
- *  Private use only. Init streaming operation.
- *  expects params to be valid.
- *  must receive dict, or cdict, or none, but not both.
- *  @return : 0, or an error code */
-size_t ZSTDMT_initCStream_internal(ZSTDMT_CCtx* zcs,
-                    const void* dict, size_t dictSize, ZSTD_dictMode_e dictMode,
-                    const ZSTD_CDict* cdict,
-                    ZSTD_CCtx_params params, unsigned long long pledgedSrcSize);
-
-
-#if defined (__cplusplus)
-}
-#endif
-
-#endif   /* ZSTDMT_COMPRESS_H */
diff --git a/src/client/rspamdclient.c b/src/client/rspamdclient.c
index e281816ed..bcb25672e 100644
--- a/src/client/rspamdclient.c
+++ b/src/client/rspamdclient.c
@@ -20,7 +20,6 @@
 #include "libserver/protocol_internal.h"
 #include "unix-std.h"
 #include "contrib/zstd/zstd.h"
-#include "contrib/zstd/zdict.h"
 
 #ifdef HAVE_FETCH_H
 #include <fetch.h>
@@ -391,7 +390,7 @@ rspamd_client_command (struct rspamd_client_connection *conn,
 					return FALSE;
 				}
 
-				dict_id = ZDICT_getDictID (comp_dictionary, dict_len);
+				dict_id = -1;
 
 				if (dict_id == 0) {
 					g_set_error (err, RCLIENT_ERROR, errno,
diff --git a/src/libserver/cfg_utils.c b/src/libserver/cfg_utils.c
index 7ef701aed..78e3fc79f 100644
--- a/src/libserver/cfg_utils.c
+++ b/src/libserver/cfg_utils.c
@@ -38,7 +38,6 @@
 
 #define ZSTD_STATIC_LINKING_ONLY
 #include "contrib/zstd/zstd.h"
-#include "contrib/zstd/zdict.h"
 
 #ifdef HAVE_OPENSSL
 #include <openssl/rand.h>
@@ -2764,7 +2763,7 @@ rspamd_open_zstd_dictionary (const char *path)
 		return NULL;
 	}
 
-	dict->id = ZDICT_getDictID (dict->dict, dict->size);
+	dict->id = -1;
 
 	if (dict->id == 0) {
 		g_free (dict);