Remove golang vendored directory (#18277)

* rm go vendor

* fix drone yaml

* add to gitignore

102 files changed, 0 insertions, 28395 deletions

diff --git a/vendor/github.com/klauspost/compress/LICENSE b/vendor/github.com/klauspost/compress/LICENSE
deleted file mode 100644
index 1eb75ef68e..0000000000
--- a/vendor/github.com/klauspost/compress/LICENSE
+++ /dev/null
@@ -1,28 +0,0 @@
-Copyright (c) 2012 The Go Authors. All rights reserved.
-Copyright (c) 2019 Klaus Post. All rights reserved.
-
-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.
-   * Neither the name of Google Inc. nor the names of its
-contributors may be used to endorse or promote products derived from
-this software without specific prior written permission.
-
-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.
diff --git a/vendor/github.com/klauspost/compress/flate/deflate.go b/vendor/github.com/klauspost/compress/flate/deflate.go
deleted file mode 100644
index 5283ac5a53..0000000000
--- a/vendor/github.com/klauspost/compress/flate/deflate.go
+++ /dev/null
@@ -1,820 +0,0 @@
-// Copyright 2009 The Go Authors. All rights reserved.
-// Copyright (c) 2015 Klaus Post
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package flate
-
-import (
-	"fmt"
-	"io"
-	"math"
-)
-
-const (
-	NoCompression      = 0
-	BestSpeed          = 1
-	BestCompression    = 9
-	DefaultCompression = -1
-
-	// HuffmanOnly disables Lempel-Ziv match searching and only performs Huffman
-	// entropy encoding. This mode is useful in compressing data that has
-	// already been compressed with an LZ style algorithm (e.g. Snappy or LZ4)
-	// that lacks an entropy encoder. Compression gains are achieved when
-	// certain bytes in the input stream occur more frequently than others.
-	//
-	// Note that HuffmanOnly produces a compressed output that is
-	// RFC 1951 compliant. That is, any valid DEFLATE decompressor will
-	// continue to be able to decompress this output.
-	HuffmanOnly         = -2
-	ConstantCompression = HuffmanOnly // compatibility alias.
-
-	logWindowSize    = 15
-	windowSize       = 1 << logWindowSize
-	windowMask       = windowSize - 1
-	logMaxOffsetSize = 15  // Standard DEFLATE
-	minMatchLength   = 4   // The smallest match that the compressor looks for
-	maxMatchLength   = 258 // The longest match for the compressor
-	minOffsetSize    = 1   // The shortest offset that makes any sense
-
-	// The maximum number of tokens we put into a single flat block, just too
-	// stop things from getting too large.
-	maxFlateBlockTokens = 1 << 14
-	maxStoreBlockSize   = 65535
-	hashBits            = 17 // After 17 performance degrades
-	hashSize            = 1 << hashBits
-	hashMask            = (1 << hashBits) - 1
-	hashShift           = (hashBits + minMatchLength - 1) / minMatchLength
-	maxHashOffset       = 1 << 24
-
-	skipNever = math.MaxInt32
-
-	debugDeflate = false
-)
-
-type compressionLevel struct {
-	good, lazy, nice, chain, fastSkipHashing, level int
-}
-
-// Compression levels have been rebalanced from zlib deflate defaults
-// to give a bigger spread in speed and compression.
-// See https://blog.klauspost.com/rebalancing-deflate-compression-levels/
-var levels = []compressionLevel{
-	{}, // 0
-	// Level 1-6 uses specialized algorithm - values not used
-	{0, 0, 0, 0, 0, 1},
-	{0, 0, 0, 0, 0, 2},
-	{0, 0, 0, 0, 0, 3},
-	{0, 0, 0, 0, 0, 4},
-	{0, 0, 0, 0, 0, 5},
-	{0, 0, 0, 0, 0, 6},
-	// Levels 7-9 use increasingly more lazy matching
-	// and increasingly stringent conditions for "good enough".
-	{8, 8, 24, 16, skipNever, 7},
-	{10, 16, 24, 64, skipNever, 8},
-	{32, 258, 258, 4096, skipNever, 9},
-}
-
-// advancedState contains state for the advanced levels, with bigger hash tables, etc.
-type advancedState struct {
-	// deflate state
-	length         int
-	offset         int
-	maxInsertIndex int
-
-	// Input hash chains
-	// hashHead[hashValue] contains the largest inputIndex with the specified hash value
-	// If hashHead[hashValue] is within the current window, then
-	// hashPrev[hashHead[hashValue] & windowMask] contains the previous index
-	// with the same hash value.
-	chainHead  int
-	hashHead   [hashSize]uint32
-	hashPrev   [windowSize]uint32
-	hashOffset int
-
-	// input window: unprocessed data is window[index:windowEnd]
-	index     int
-	hashMatch [maxMatchLength + minMatchLength]uint32
-
-	hash uint32
-	ii   uint16 // position of last match, intended to overflow to reset.
-}
-
-type compressor struct {
-	compressionLevel
-
-	w *huffmanBitWriter
-
-	// compression algorithm
-	fill func(*compressor, []byte) int // copy data to window
-	step func(*compressor)             // process window
-
-	window     []byte
-	windowEnd  int
-	blockStart int // window index where current tokens start
-	err        error
-
-	// queued output tokens
-	tokens tokens
-	fast   fastEnc
-	state  *advancedState
-
-	sync          bool // requesting flush
-	byteAvailable bool // if true, still need to process window[index-1].
-}
-
-func (d *compressor) fillDeflate(b []byte) int {
-	s := d.state
-	if s.index >= 2*windowSize-(minMatchLength+maxMatchLength) {
-		// shift the window by windowSize
-		copy(d.window[:], d.window[windowSize:2*windowSize])
-		s.index -= windowSize
-		d.windowEnd -= windowSize
-		if d.blockStart >= windowSize {
-			d.blockStart -= windowSize
-		} else {
-			d.blockStart = math.MaxInt32
-		}
-		s.hashOffset += windowSize
-		if s.hashOffset > maxHashOffset {
-			delta := s.hashOffset - 1
-			s.hashOffset -= delta
-			s.chainHead -= delta
-			// Iterate over slices instead of arrays to avoid copying
-			// the entire table onto the stack (Issue #18625).
-			for i, v := range s.hashPrev[:] {
-				if int(v) > delta {
-					s.hashPrev[i] = uint32(int(v) - delta)
-				} else {
-					s.hashPrev[i] = 0
-				}
-			}
-			for i, v := range s.hashHead[:] {
-				if int(v) > delta {
-					s.hashHead[i] = uint32(int(v) - delta)
-				} else {
-					s.hashHead[i] = 0
-				}
-			}
-		}
-	}
-	n := copy(d.window[d.windowEnd:], b)
-	d.windowEnd += n
-	return n
-}
-
-func (d *compressor) writeBlock(tok *tokens, index int, eof bool) error {
-	if index > 0 || eof {
-		var window []byte
-		if d.blockStart <= index {
-			window = d.window[d.blockStart:index]
-		}
-		d.blockStart = index
-		d.w.writeBlock(tok, eof, window)
-		return d.w.err
-	}
-	return nil
-}
-
-// writeBlockSkip writes the current block and uses the number of tokens
-// to determine if the block should be stored on no matches, or
-// only huffman encoded.
-func (d *compressor) writeBlockSkip(tok *tokens, index int, eof bool) error {
-	if index > 0 || eof {
-		if d.blockStart <= index {
-			window := d.window[d.blockStart:index]
-			// If we removed less than a 64th of all literals
-			// we huffman compress the block.
-			if int(tok.n) > len(window)-int(tok.n>>6) {
-				d.w.writeBlockHuff(eof, window, d.sync)
-			} else {
-				// Write a dynamic huffman block.
-				d.w.writeBlockDynamic(tok, eof, window, d.sync)
-			}
-		} else {
-			d.w.writeBlock(tok, eof, nil)
-		}
-		d.blockStart = index
-		return d.w.err
-	}
-	return nil
-}
-
-// fillWindow will fill the current window with the supplied
-// dictionary and calculate all hashes.
-// This is much faster than doing a full encode.
-// Should only be used after a start/reset.
-func (d *compressor) fillWindow(b []byte) {
-	// Do not fill window if we are in store-only or huffman mode.
-	if d.level <= 0 {
-		return
-	}
-	if d.fast != nil {
-		// encode the last data, but discard the result
-		if len(b) > maxMatchOffset {
-			b = b[len(b)-maxMatchOffset:]
-		}
-		d.fast.Encode(&d.tokens, b)
-		d.tokens.Reset()
-		return
-	}
-	s := d.state
-	// If we are given too much, cut it.
-	if len(b) > windowSize {
-		b = b[len(b)-windowSize:]
-	}
-	// Add all to window.
-	n := copy(d.window[d.windowEnd:], b)
-
-	// Calculate 256 hashes at the time (more L1 cache hits)
-	loops := (n + 256 - minMatchLength) / 256
-	for j := 0; j < loops; j++ {
-		startindex := j * 256
-		end := startindex + 256 + minMatchLength - 1
-		if end > n {
-			end = n
-		}
-		tocheck := d.window[startindex:end]
-		dstSize := len(tocheck) - minMatchLength + 1
-
-		if dstSize <= 0 {
-			continue
-		}
-
-		dst := s.hashMatch[:dstSize]
-		bulkHash4(tocheck, dst)
-		var newH uint32
-		for i, val := range dst {
-			di := i + startindex
-			newH = val & hashMask
-			// Get previous value with the same hash.
-			// Our chain should point to the previous value.
-			s.hashPrev[di&windowMask] = s.hashHead[newH]
-			// Set the head of the hash chain to us.
-			s.hashHead[newH] = uint32(di + s.hashOffset)
-		}
-		s.hash = newH
-	}
-	// Update window information.
-	d.windowEnd += n
-	s.index = n
-}
-
-// Try to find a match starting at index whose length is greater than prevSize.
-// We only look at chainCount possibilities before giving up.
-// pos = s.index, prevHead = s.chainHead-s.hashOffset, prevLength=minMatchLength-1, lookahead
-func (d *compressor) findMatch(pos int, prevHead int, prevLength int, lookahead int) (length, offset int, ok bool) {
-	minMatchLook := maxMatchLength
-	if lookahead < minMatchLook {
-		minMatchLook = lookahead
-	}
-
-	win := d.window[0 : pos+minMatchLook]
-
-	// We quit when we get a match that's at least nice long
-	nice := len(win) - pos
-	if d.nice < nice {
-		nice = d.nice
-	}
-
-	// If we've got a match that's good enough, only look in 1/4 the chain.
-	tries := d.chain
-	length = prevLength
-	if length >= d.good {
-		tries >>= 2
-	}
-
-	wEnd := win[pos+length]
-	wPos := win[pos:]
-	minIndex := pos - windowSize
-
-	for i := prevHead; tries > 0; tries-- {
-		if wEnd == win[i+length] {
-			n := matchLen(win[i:i+minMatchLook], wPos)
-
-			if n > length && (n > minMatchLength || pos-i <= 4096) {
-				length = n
-				offset = pos - i
-				ok = true
-				if n >= nice {
-					// The match is good enough that we don't try to find a better one.
-					break
-				}
-				wEnd = win[pos+n]
-			}
-		}
-		if i == minIndex {
-			// hashPrev[i & windowMask] has already been overwritten, so stop now.
-			break
-		}
-		i = int(d.state.hashPrev[i&windowMask]) - d.state.hashOffset
-		if i < minIndex || i < 0 {
-			break
-		}
-	}
-	return
-}
-
-func (d *compressor) writeStoredBlock(buf []byte) error {
-	if d.w.writeStoredHeader(len(buf), false); d.w.err != nil {
-		return d.w.err
-	}
-	d.w.writeBytes(buf)
-	return d.w.err
-}
-
-// hash4 returns a hash representation of the first 4 bytes
-// of the supplied slice.
-// The caller must ensure that len(b) >= 4.
-func hash4(b []byte) uint32 {
-	b = b[:4]
-	return hash4u(uint32(b[3])|uint32(b[2])<<8|uint32(b[1])<<16|uint32(b[0])<<24, hashBits)
-}
-
-// bulkHash4 will compute hashes using the same
-// algorithm as hash4
-func bulkHash4(b []byte, dst []uint32) {
-	if len(b) < 4 {
-		return
-	}
-	hb := uint32(b[3]) | uint32(b[2])<<8 | uint32(b[1])<<16 | uint32(b[0])<<24
-	dst[0] = hash4u(hb, hashBits)
-	end := len(b) - 4 + 1
-	for i := 1; i < end; i++ {
-		hb = (hb << 8) | uint32(b[i+3])
-		dst[i] = hash4u(hb, hashBits)
-	}
-}
-
-func (d *compressor) initDeflate() {
-	d.window = make([]byte, 2*windowSize)
-	d.byteAvailable = false
-	d.err = nil
-	if d.state == nil {
-		return
-	}
-	s := d.state
-	s.index = 0
-	s.hashOffset = 1
-	s.length = minMatchLength - 1
-	s.offset = 0
-	s.hash = 0
-	s.chainHead = -1
-}
-
-// deflateLazy is the same as deflate, but with d.fastSkipHashing == skipNever,
-// meaning it always has lazy matching on.
-func (d *compressor) deflateLazy() {
-	s := d.state
-	// Sanity enables additional runtime tests.
-	// It's intended to be used during development
-	// to supplement the currently ad-hoc unit tests.
-	const sanity = debugDeflate
-
-	if d.windowEnd-s.index < minMatchLength+maxMatchLength && !d.sync {
-		return
-	}
-
-	s.maxInsertIndex = d.windowEnd - (minMatchLength - 1)
-	if s.index < s.maxInsertIndex {
-		s.hash = hash4(d.window[s.index : s.index+minMatchLength])
-	}
-
-	for {
-		if sanity && s.index > d.windowEnd {
-			panic("index > windowEnd")
-		}
-		lookahead := d.windowEnd - s.index
-		if lookahead < minMatchLength+maxMatchLength {
-			if !d.sync {
-				return
-			}
-			if sanity && s.index > d.windowEnd {
-				panic("index > windowEnd")
-			}
-			if lookahead == 0 {
-				// Flush current output block if any.
-				if d.byteAvailable {
-					// There is still one pending token that needs to be flushed
-					d.tokens.AddLiteral(d.window[s.index-1])
-					d.byteAvailable = false
-				}
-				if d.tokens.n > 0 {
-					if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil {
-						return
-					}
-					d.tokens.Reset()
-				}
-				return
-			}
-		}
-		if s.index < s.maxInsertIndex {
-			// Update the hash
-			s.hash = hash4(d.window[s.index : s.index+minMatchLength])
-			ch := s.hashHead[s.hash&hashMask]
-			s.chainHead = int(ch)
-			s.hashPrev[s.index&windowMask] = ch
-			s.hashHead[s.hash&hashMask] = uint32(s.index + s.hashOffset)
-		}
-		prevLength := s.length
-		prevOffset := s.offset
-		s.length = minMatchLength - 1
-		s.offset = 0
-		minIndex := s.index - windowSize
-		if minIndex < 0 {
-			minIndex = 0
-		}
-
-		if s.chainHead-s.hashOffset >= minIndex && lookahead > prevLength && prevLength < d.lazy {
-			if newLength, newOffset, ok := d.findMatch(s.index, s.chainHead-s.hashOffset, minMatchLength-1, lookahead); ok {
-				s.length = newLength
-				s.offset = newOffset
-			}
-		}
-		if prevLength >= minMatchLength && s.length <= prevLength {
-			// There was a match at the previous step, and the current match is
-			// not better. Output the previous match.
-			d.tokens.AddMatch(uint32(prevLength-3), uint32(prevOffset-minOffsetSize))
-
-			// Insert in the hash table all strings up to the end of the match.
-			// index and index-1 are already inserted. If there is not enough
-			// lookahead, the last two strings are not inserted into the hash
-			// table.
-			newIndex := s.index + prevLength - 1
-			// Calculate missing hashes
-			end := newIndex
-			if end > s.maxInsertIndex {
-				end = s.maxInsertIndex
-			}
-			end += minMatchLength - 1
-			startindex := s.index + 1
-			if startindex > s.maxInsertIndex {
-				startindex = s.maxInsertIndex
-			}
-			tocheck := d.window[startindex:end]
-			dstSize := len(tocheck) - minMatchLength + 1
-			if dstSize > 0 {
-				dst := s.hashMatch[:dstSize]
-				bulkHash4(tocheck, dst)
-				var newH uint32
-				for i, val := range dst {
-					di := i + startindex
-					newH = val & hashMask
-					// Get previous value with the same hash.
-					// Our chain should point to the previous value.
-					s.hashPrev[di&windowMask] = s.hashHead[newH]
-					// Set the head of the hash chain to us.
-					s.hashHead[newH] = uint32(di + s.hashOffset)
-				}
-				s.hash = newH
-			}
-
-			s.index = newIndex
-			d.byteAvailable = false
-			s.length = minMatchLength - 1
-			if d.tokens.n == maxFlateBlockTokens {
-				// The block includes the current character
-				if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil {
-					return
-				}
-				d.tokens.Reset()
-			}
-		} else {
-			// Reset, if we got a match this run.
-			if s.length >= minMatchLength {
-				s.ii = 0
-			}
-			// We have a byte waiting. Emit it.
-			if d.byteAvailable {
-				s.ii++
-				d.tokens.AddLiteral(d.window[s.index-1])
-				if d.tokens.n == maxFlateBlockTokens {
-					if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil {
-						return
-					}
-					d.tokens.Reset()
-				}
-				s.index++
-
-				// If we have a long run of no matches, skip additional bytes
-				// Resets when s.ii overflows after 64KB.
-				if s.ii > 31 {
-					n := int(s.ii >> 5)
-					for j := 0; j < n; j++ {
-						if s.index >= d.windowEnd-1 {
-							break
-						}
-
-						d.tokens.AddLiteral(d.window[s.index-1])
-						if d.tokens.n == maxFlateBlockTokens {
-							if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil {
-								return
-							}
-							d.tokens.Reset()
-						}
-						s.index++
-					}
-					// Flush last byte
-					d.tokens.AddLiteral(d.window[s.index-1])
-					d.byteAvailable = false
-					// s.length = minMatchLength - 1 // not needed, since s.ii is reset above, so it should never be > minMatchLength
-					if d.tokens.n == maxFlateBlockTokens {
-						if d.err = d.writeBlock(&d.tokens, s.index, false); d.err != nil {
-							return
-						}
-						d.tokens.Reset()
-					}
-				}
-			} else {
-				s.index++
-				d.byteAvailable = true
-			}
-		}
-	}
-}
-
-func (d *compressor) store() {
-	if d.windowEnd > 0 && (d.windowEnd == maxStoreBlockSize || d.sync) {
-		d.err = d.writeStoredBlock(d.window[:d.windowEnd])
-		d.windowEnd = 0
-	}
-}
-
-// fillWindow will fill the buffer with data for huffman-only compression.
-// The number of bytes copied is returned.
-func (d *compressor) fillBlock(b []byte) int {
-	n := copy(d.window[d.windowEnd:], b)
-	d.windowEnd += n
-	return n
-}
-
-// storeHuff will compress and store the currently added data,
-// if enough has been accumulated or we at the end of the stream.
-// Any error that occurred will be in d.err
-func (d *compressor) storeHuff() {
-	if d.windowEnd < len(d.window) && !d.sync || d.windowEnd == 0 {
-		return
-	}
-	d.w.writeBlockHuff(false, d.window[:d.windowEnd], d.sync)
-	d.err = d.w.err
-	d.windowEnd = 0
-}
-
-// storeFast will compress and store the currently added data,
-// if enough has been accumulated or we at the end of the stream.
-// Any error that occurred will be in d.err
-func (d *compressor) storeFast() {
-	// We only compress if we have maxStoreBlockSize.
-	if d.windowEnd < len(d.window) {
-		if !d.sync {
-			return
-		}
-		// Handle extremely small sizes.
-		if d.windowEnd < 128 {
-			if d.windowEnd == 0 {
-				return
-			}
-			if d.windowEnd <= 32 {
-				d.err = d.writeStoredBlock(d.window[:d.windowEnd])
-			} else {
-				d.w.writeBlockHuff(false, d.window[:d.windowEnd], true)
-				d.err = d.w.err
-			}
-			d.tokens.Reset()
-			d.windowEnd = 0
-			d.fast.Reset()
-			return
-		}
-	}
-
-	d.fast.Encode(&d.tokens, d.window[:d.windowEnd])
-	// If we made zero matches, store the block as is.
-	if d.tokens.n == 0 {
-		d.err = d.writeStoredBlock(d.window[:d.windowEnd])
-		// If we removed less than 1/16th, huffman compress the block.
-	} else if int(d.tokens.n) > d.windowEnd-(d.windowEnd>>4) {
-		d.w.writeBlockHuff(false, d.window[:d.windowEnd], d.sync)
-		d.err = d.w.err
-	} else {
-		d.w.writeBlockDynamic(&d.tokens, false, d.window[:d.windowEnd], d.sync)
-		d.err = d.w.err
-	}
-	d.tokens.Reset()
-	d.windowEnd = 0
-}
-
-// write will add input byte to the stream.
-// Unless an error occurs all bytes will be consumed.
-func (d *compressor) write(b []byte) (n int, err error) {
-	if d.err != nil {
-		return 0, d.err
-	}
-	n = len(b)
-	for len(b) > 0 {
-		d.step(d)
-		b = b[d.fill(d, b):]
-		if d.err != nil {
-			return 0, d.err
-		}
-	}
-	return n, d.err
-}
-
-func (d *compressor) syncFlush() error {
-	d.sync = true
-	if d.err != nil {
-		return d.err
-	}
-	d.step(d)
-	if d.err == nil {
-		d.w.writeStoredHeader(0, false)
-		d.w.flush()
-		d.err = d.w.err
-	}
-	d.sync = false
-	return d.err
-}
-
-func (d *compressor) init(w io.Writer, level int) (err error) {
-	d.w = newHuffmanBitWriter(w)
-
-	switch {
-	case level == NoCompression:
-		d.window = make([]byte, maxStoreBlockSize)
-		d.fill = (*compressor).fillBlock
-		d.step = (*compressor).store
-	case level == ConstantCompression:
-		d.w.logNewTablePenalty = 10
-		d.window = make([]byte, 32<<10)
-		d.fill = (*compressor).fillBlock
-		d.step = (*compressor).storeHuff
-	case level == DefaultCompression:
-		level = 5
-		fallthrough
-	case level >= 1 && level <= 6:
-		d.w.logNewTablePenalty = 8
-		d.fast = newFastEnc(level)
-		d.window = make([]byte, maxStoreBlockSize)
-		d.fill = (*compressor).fillBlock
-		d.step = (*compressor).storeFast
-	case 7 <= level && level <= 9:
-		d.w.logNewTablePenalty = 10
-		d.state = &advancedState{}
-		d.compressionLevel = levels[level]
-		d.initDeflate()
-		d.fill = (*compressor).fillDeflate
-		d.step = (*compressor).deflateLazy
-	default:
-		return fmt.Errorf("flate: invalid compression level %d: want value in range [-2, 9]", level)
-	}
-	d.level = level
-	return nil
-}
-
-// reset the state of the compressor.
-func (d *compressor) reset(w io.Writer) {
-	d.w.reset(w)
-	d.sync = false
-	d.err = nil
-	// We only need to reset a few things for Snappy.
-	if d.fast != nil {
-		d.fast.Reset()
-		d.windowEnd = 0
-		d.tokens.Reset()
-		return
-	}
-	switch d.compressionLevel.chain {
-	case 0:
-		// level was NoCompression or ConstantCompresssion.
-		d.windowEnd = 0
-	default:
-		s := d.state
-		s.chainHead = -1
-		for i := range s.hashHead {
-			s.hashHead[i] = 0
-		}
-		for i := range s.hashPrev {
-			s.hashPrev[i] = 0
-		}
-		s.hashOffset = 1
-		s.index, d.windowEnd = 0, 0
-		d.blockStart, d.byteAvailable = 0, false
-		d.tokens.Reset()
-		s.length = minMatchLength - 1
-		s.offset = 0
-		s.hash = 0
-		s.ii = 0
-		s.maxInsertIndex = 0
-	}
-}
-
-func (d *compressor) close() error {
-	if d.err != nil {
-		return d.err
-	}
-	d.sync = true
-	d.step(d)
-	if d.err != nil {
-		return d.err
-	}
-	if d.w.writeStoredHeader(0, true); d.w.err != nil {
-		return d.w.err
-	}
-	d.w.flush()
-	d.w.reset(nil)
-	return d.w.err
-}
-
-// NewWriter returns a new Writer compressing data at the given level.
-// Following zlib, levels range from 1 (BestSpeed) to 9 (BestCompression);
-// higher levels typically run slower but compress more.
-// Level 0 (NoCompression) does not attempt any compression; it only adds the
-// necessary DEFLATE framing.
-// Level -1 (DefaultCompression) uses the default compression level.
-// Level -2 (ConstantCompression) will use Huffman compression only, giving
-// a very fast compression for all types of input, but sacrificing considerable
-// compression efficiency.
-//
-// If level is in the range [-2, 9] then the error returned will be nil.
-// Otherwise the error returned will be non-nil.
-func NewWriter(w io.Writer, level int) (*Writer, error) {
-	var dw Writer
-	if err := dw.d.init(w, level); err != nil {
-		return nil, err
-	}
-	return &dw, nil
-}
-
-// NewWriterDict is like NewWriter but initializes the new
-// Writer with a preset dictionary.  The returned Writer behaves
-// as if the dictionary had been written to it without producing
-// any compressed output.  The compressed data written to w
-// can only be decompressed by a Reader initialized with the
-// same dictionary.
-func NewWriterDict(w io.Writer, level int, dict []byte) (*Writer, error) {
-	zw, err := NewWriter(w, level)
-	if err != nil {
-		return nil, err
-	}
-	zw.d.fillWindow(dict)
-	zw.dict = append(zw.dict, dict...) // duplicate dictionary for Reset method.
-	return zw, err
-}
-
-// A Writer takes data written to it and writes the compressed
-// form of that data to an underlying writer (see NewWriter).
-type Writer struct {
-	d    compressor
-	dict []byte
-}
-
-// Write writes data to w, which will eventually write the
-// compressed form of data to its underlying writer.
-func (w *Writer) Write(data []byte) (n int, err error) {
-	return w.d.write(data)
-}
-
-// Flush flushes any pending data to the underlying writer.
-// It is useful mainly in compressed network protocols, to ensure that
-// a remote reader has enough data to reconstruct a packet.
-// Flush does not return until the data has been written.
-// Calling Flush when there is no pending data still causes the Writer
-// to emit a sync marker of at least 4 bytes.
-// If the underlying writer returns an error, Flush returns that error.
-//
-// In the terminology of the zlib library, Flush is equivalent to Z_SYNC_FLUSH.
-func (w *Writer) Flush() error {
-	// For more about flushing:
-	// http://www.bolet.org/~pornin/deflate-flush.html
-	return w.d.syncFlush()
-}
-
-// Close flushes and closes the writer.
-func (w *Writer) Close() error {
-	return w.d.close()
-}
-
-// Reset discards the writer's state and makes it equivalent to
-// the result of NewWriter or NewWriterDict called with dst
-// and w's level and dictionary.
-func (w *Writer) Reset(dst io.Writer) {
-	if len(w.dict) > 0 {
-		// w was created with NewWriterDict
-		w.d.reset(dst)
-		if dst != nil {
-			w.d.fillWindow(w.dict)
-		}
-	} else {
-		// w was created with NewWriter
-		w.d.reset(dst)
-	}
-}
-
-// ResetDict discards the writer's state and makes it equivalent to
-// the result of NewWriter or NewWriterDict called with dst
-// and w's level, but sets a specific dictionary.
-func (w *Writer) ResetDict(dst io.Writer, dict []byte) {
-	w.dict = dict
-	w.d.reset(dst)
-	w.d.fillWindow(w.dict)
-}
diff --git a/vendor/github.com/klauspost/compress/flate/dict_decoder.go b/vendor/github.com/klauspost/compress/flate/dict_decoder.go
deleted file mode 100644
index 71c75a065e..0000000000
--- a/vendor/github.com/klauspost/compress/flate/dict_decoder.go
+++ /dev/null
@@ -1,184 +0,0 @@
-// Copyright 2016 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package flate
-
-// dictDecoder implements the LZ77 sliding dictionary as used in decompression.
-// LZ77 decompresses data through sequences of two forms of commands:
-//
-//	* Literal insertions: Runs of one or more symbols are inserted into the data
-//	stream as is. This is accomplished through the writeByte method for a
-//	single symbol, or combinations of writeSlice/writeMark for multiple symbols.
-//	Any valid stream must start with a literal insertion if no preset dictionary
-//	is used.
-//
-//	* Backward copies: Runs of one or more symbols are copied from previously
-//	emitted data. Backward copies come as the tuple (dist, length) where dist
-//	determines how far back in the stream to copy from and length determines how
-//	many bytes to copy. Note that it is valid for the length to be greater than
-//	the distance. Since LZ77 uses forward copies, that situation is used to
-//	perform a form of run-length encoding on repeated runs of symbols.
-//	The writeCopy and tryWriteCopy are used to implement this command.
-//
-// For performance reasons, this implementation performs little to no sanity
-// checks about the arguments. As such, the invariants documented for each
-// method call must be respected.
-type dictDecoder struct {
-	hist []byte // Sliding window history
-
-	// Invariant: 0 <= rdPos <= wrPos <= len(hist)
-	wrPos int  // Current output position in buffer
-	rdPos int  // Have emitted hist[:rdPos] already
-	full  bool // Has a full window length been written yet?
-}
-
-// init initializes dictDecoder to have a sliding window dictionary of the given
-// size. If a preset dict is provided, it will initialize the dictionary with
-// the contents of dict.
-func (dd *dictDecoder) init(size int, dict []byte) {
-	*dd = dictDecoder{hist: dd.hist}
-
-	if cap(dd.hist) < size {
-		dd.hist = make([]byte, size)
-	}
-	dd.hist = dd.hist[:size]
-
-	if len(dict) > len(dd.hist) {
-		dict = dict[len(dict)-len(dd.hist):]
-	}
-	dd.wrPos = copy(dd.hist, dict)
-	if dd.wrPos == len(dd.hist) {
-		dd.wrPos = 0
-		dd.full = true
-	}
-	dd.rdPos = dd.wrPos
-}
-
-// histSize reports the total amount of historical data in the dictionary.
-func (dd *dictDecoder) histSize() int {
-	if dd.full {
-		return len(dd.hist)
-	}
-	return dd.wrPos
-}
-
-// availRead reports the number of bytes that can be flushed by readFlush.
-func (dd *dictDecoder) availRead() int {
-	return dd.wrPos - dd.rdPos
-}
-
-// availWrite reports the available amount of output buffer space.
-func (dd *dictDecoder) availWrite() int {
-	return len(dd.hist) - dd.wrPos
-}
-
-// writeSlice returns a slice of the available buffer to write data to.
-//
-// This invariant will be kept: len(s) <= availWrite()
-func (dd *dictDecoder) writeSlice() []byte {
-	return dd.hist[dd.wrPos:]
-}
-
-// writeMark advances the writer pointer by cnt.
-//
-// This invariant must be kept: 0 <= cnt <= availWrite()
-func (dd *dictDecoder) writeMark(cnt int) {
-	dd.wrPos += cnt
-}
-
-// writeByte writes a single byte to the dictionary.
-//
-// This invariant must be kept: 0 < availWrite()
-func (dd *dictDecoder) writeByte(c byte) {
-	dd.hist[dd.wrPos] = c
-	dd.wrPos++
-}
-
-// writeCopy copies a string at a given (dist, length) to the output.
-// This returns the number of bytes copied and may be less than the requested
-// length if the available space in the output buffer is too small.
-//
-// This invariant must be kept: 0 < dist <= histSize()
-func (dd *dictDecoder) writeCopy(dist, length int) int {
-	dstBase := dd.wrPos
-	dstPos := dstBase
-	srcPos := dstPos - dist
-	endPos := dstPos + length
-	if endPos > len(dd.hist) {
-		endPos = len(dd.hist)
-	}
-
-	// Copy non-overlapping section after destination position.
-	//
-	// This section is non-overlapping in that the copy length for this section
-	// is always less than or equal to the backwards distance. This can occur
-	// if a distance refers to data that wraps-around in the buffer.
-	// Thus, a backwards copy is performed here; that is, the exact bytes in
-	// the source prior to the copy is placed in the destination.
-	if srcPos < 0 {
-		srcPos += len(dd.hist)
-		dstPos += copy(dd.hist[dstPos:endPos], dd.hist[srcPos:])
-		srcPos = 0
-	}
-
-	// Copy possibly overlapping section before destination position.
-	//
-	// This section can overlap if the copy length for this section is larger
-	// than the backwards distance. This is allowed by LZ77 so that repeated
-	// strings can be succinctly represented using (dist, length) pairs.
-	// Thus, a forwards copy is performed here; that is, the bytes copied is
-	// possibly dependent on the resulting bytes in the destination as the copy
-	// progresses along. This is functionally equivalent to the following:
-	//
-	//	for i := 0; i < endPos-dstPos; i++ {
-	//		dd.hist[dstPos+i] = dd.hist[srcPos+i]
-	//	}
-	//	dstPos = endPos
-	//
-	for dstPos < endPos {
-		dstPos += copy(dd.hist[dstPos:endPos], dd.hist[srcPos:dstPos])
-	}
-
-	dd.wrPos = dstPos
-	return dstPos - dstBase
-}
-
-// tryWriteCopy tries to copy a string at a given (distance, length) to the
-// output. This specialized version is optimized for short distances.
-//
-// This method is designed to be inlined for performance reasons.
-//
-// This invariant must be kept: 0 < dist <= histSize()
-func (dd *dictDecoder) tryWriteCopy(dist, length int) int {
-	dstPos := dd.wrPos
-	endPos := dstPos + length
-	if dstPos < dist || endPos > len(dd.hist) {
-		return 0
-	}
-	dstBase := dstPos
-	srcPos := dstPos - dist
-
-	// Copy possibly overlapping section before destination position.
-loop:
-	dstPos += copy(dd.hist[dstPos:endPos], dd.hist[srcPos:dstPos])
-	if dstPos < endPos {
-		goto loop // Avoid for-loop so that this function can be inlined
-	}
-
-	dd.wrPos = dstPos
-	return dstPos - dstBase
-}
-
-// readFlush returns a slice of the historical buffer that is ready to be
-// emitted to the user. The data returned by readFlush must be fully consumed
-// before calling any other dictDecoder methods.
-func (dd *dictDecoder) readFlush() []byte {
-	toRead := dd.hist[dd.rdPos:dd.wrPos]
-	dd.rdPos = dd.wrPos
-	if dd.wrPos == len(dd.hist) {
-		dd.wrPos, dd.rdPos = 0, 0
-		dd.full = true
-	}
-	return toRead
-}
diff --git a/vendor/github.com/klauspost/compress/flate/fast_encoder.go b/vendor/github.com/klauspost/compress/flate/fast_encoder.go
deleted file mode 100644
index 347ac2c902..0000000000
--- a/vendor/github.com/klauspost/compress/flate/fast_encoder.go
+++ /dev/null
@@ -1,244 +0,0 @@
-// Copyright 2011 The Snappy-Go Authors. All rights reserved.
-// Modified for deflate by Klaus Post (c) 2015.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package flate
-
-import (
-	"encoding/binary"
-	"fmt"
-	"math/bits"
-)
-
-type fastEnc interface {
-	Encode(dst *tokens, src []byte)
-	Reset()
-}
-
-func newFastEnc(level int) fastEnc {
-	switch level {
-	case 1:
-		return &fastEncL1{fastGen: fastGen{cur: maxStoreBlockSize}}
-	case 2:
-		return &fastEncL2{fastGen: fastGen{cur: maxStoreBlockSize}}
-	case 3:
-		return &fastEncL3{fastGen: fastGen{cur: maxStoreBlockSize}}
-	case 4:
-		return &fastEncL4{fastGen: fastGen{cur: maxStoreBlockSize}}
-	case 5:
-		return &fastEncL5{fastGen: fastGen{cur: maxStoreBlockSize}}
-	case 6:
-		return &fastEncL6{fastGen: fastGen{cur: maxStoreBlockSize}}
-	default:
-		panic("invalid level specified")
-	}
-}
-
-const (
-	tableBits       = 15             // Bits used in the table
-	tableSize       = 1 << tableBits // Size of the table
-	tableShift      = 32 - tableBits // Right-shift to get the tableBits most significant bits of a uint32.
-	baseMatchOffset = 1              // The smallest match offset
-	baseMatchLength = 3              // The smallest match length per the RFC section 3.2.5
-	maxMatchOffset  = 1 << 15        // The largest match offset
-
-	bTableBits   = 17                                               // Bits used in the big tables
-	bTableSize   = 1 << bTableBits                                  // Size of the table
-	allocHistory = maxStoreBlockSize * 5                            // Size to preallocate for history.
-	bufferReset  = (1 << 31) - allocHistory - maxStoreBlockSize - 1 // Reset the buffer offset when reaching this.
-)
-
-const (
-	prime3bytes = 506832829
-	prime4bytes = 2654435761
-	prime5bytes = 889523592379
-	prime6bytes = 227718039650203
-	prime7bytes = 58295818150454627
-	prime8bytes = 0xcf1bbcdcb7a56463
-)
-
-func load32(b []byte, i int) uint32 {
-	// Help the compiler eliminate bounds checks on the read so it can be done in a single read.
-	b = b[i:]
-	b = b[:4]
-	return uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24
-}
-
-func load64(b []byte, i int) uint64 {
-	return binary.LittleEndian.Uint64(b[i:])
-}
-
-func load3232(b []byte, i int32) uint32 {
-	return binary.LittleEndian.Uint32(b[i:])
-}
-
-func load6432(b []byte, i int32) uint64 {
-	return binary.LittleEndian.Uint64(b[i:])
-}
-
-func hash(u uint32) uint32 {
-	return (u * 0x1e35a7bd) >> tableShift
-}
-
-type tableEntry struct {
-	offset int32
-}
-
-// fastGen maintains the table for matches,
-// and the previous byte block for level 2.
-// This is the generic implementation.
-type fastGen struct {
-	hist []byte
-	cur  int32
-}
-
-func (e *fastGen) addBlock(src []byte) int32 {
-	// check if we have space already
-	if len(e.hist)+len(src) > cap(e.hist) {
-		if cap(e.hist) == 0 {
-			e.hist = make([]byte, 0, allocHistory)
-		} else {
-			if cap(e.hist) < maxMatchOffset*2 {
-				panic("unexpected buffer size")
-			}
-			// Move down
-			offset := int32(len(e.hist)) - maxMatchOffset
-			copy(e.hist[0:maxMatchOffset], e.hist[offset:])
-			e.cur += offset
-			e.hist = e.hist[:maxMatchOffset]
-		}
-	}
-	s := int32(len(e.hist))
-	e.hist = append(e.hist, src...)
-	return s
-}
-
-// hash4 returns the hash of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <32.
-func hash4u(u uint32, h uint8) uint32 {
-	return (u * prime4bytes) >> ((32 - h) & reg8SizeMask32)
-}
-
-type tableEntryPrev struct {
-	Cur  tableEntry
-	Prev tableEntry
-}
-
-// hash4x64 returns the hash of the lowest 4 bytes of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <32.
-func hash4x64(u uint64, h uint8) uint32 {
-	return (uint32(u) * prime4bytes) >> ((32 - h) & reg8SizeMask32)
-}
-
-// hash7 returns the hash of the lowest 7 bytes of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <64.
-func hash7(u uint64, h uint8) uint32 {
-	return uint32(((u << (64 - 56)) * prime7bytes) >> ((64 - h) & reg8SizeMask64))
-}
-
-// hash8 returns the hash of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <64.
-func hash8(u uint64, h uint8) uint32 {
-	return uint32((u * prime8bytes) >> ((64 - h) & reg8SizeMask64))
-}
-
-// hash6 returns the hash of the lowest 6 bytes of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <64.
-func hash6(u uint64, h uint8) uint32 {
-	return uint32(((u << (64 - 48)) * prime6bytes) >> ((64 - h) & reg8SizeMask64))
-}
-
-// matchlen will return the match length between offsets and t in src.
-// The maximum length returned is maxMatchLength - 4.
-// It is assumed that s > t, that t >=0 and s < len(src).
-func (e *fastGen) matchlen(s, t int32, src []byte) int32 {
-	if debugDecode {
-		if t >= s {
-			panic(fmt.Sprint("t >=s:", t, s))
-		}
-		if int(s) >= len(src) {
-			panic(fmt.Sprint("s >= len(src):", s, len(src)))
-		}
-		if t < 0 {
-			panic(fmt.Sprint("t < 0:", t))
-		}
-		if s-t > maxMatchOffset {
-			panic(fmt.Sprint(s, "-", t, "(", s-t, ") > maxMatchLength (", maxMatchOffset, ")"))
-		}
-	}
-	s1 := int(s) + maxMatchLength - 4
-	if s1 > len(src) {
-		s1 = len(src)
-	}
-
-	// Extend the match to be as long as possible.
-	return int32(matchLen(src[s:s1], src[t:]))
-}
-
-// matchlenLong will return the match length between offsets and t in src.
-// It is assumed that s > t, that t >=0 and s < len(src).
-func (e *fastGen) matchlenLong(s, t int32, src []byte) int32 {
-	if debugDecode {
-		if t >= s {
-			panic(fmt.Sprint("t >=s:", t, s))
-		}
-		if int(s) >= len(src) {
-			panic(fmt.Sprint("s >= len(src):", s, len(src)))
-		}
-		if t < 0 {
-			panic(fmt.Sprint("t < 0:", t))
-		}
-		if s-t > maxMatchOffset {
-			panic(fmt.Sprint(s, "-", t, "(", s-t, ") > maxMatchLength (", maxMatchOffset, ")"))
-		}
-	}
-	// Extend the match to be as long as possible.
-	return int32(matchLen(src[s:], src[t:]))
-}
-
-// Reset the encoding table.
-func (e *fastGen) Reset() {
-	if cap(e.hist) < allocHistory {
-		e.hist = make([]byte, 0, allocHistory)
-	}
-	// We offset current position so everything will be out of reach.
-	// If we are above the buffer reset it will be cleared anyway since len(hist) == 0.
-	if e.cur <= bufferReset {
-		e.cur += maxMatchOffset + int32(len(e.hist))
-	}
-	e.hist = e.hist[:0]
-}
-
-// matchLen returns the maximum length.
-// 'a' must be the shortest of the two.
-func matchLen(a, b []byte) int {
-	b = b[:len(a)]
-	var checked int
-	if len(a) >= 4 {
-		// Try 4 bytes first
-		if diff := binary.LittleEndian.Uint32(a) ^ binary.LittleEndian.Uint32(b); diff != 0 {
-			return bits.TrailingZeros32(diff) >> 3
-		}
-		// Switch to 8 byte matching.
-		checked = 4
-		a = a[4:]
-		b = b[4:]
-		for len(a) >= 8 {
-			b = b[:len(a)]
-			if diff := binary.LittleEndian.Uint64(a) ^ binary.LittleEndian.Uint64(b); diff != 0 {
-				return checked + (bits.TrailingZeros64(diff) >> 3)
-			}
-			checked += 8
-			a = a[8:]
-			b = b[8:]
-		}
-	}
-	b = b[:len(a)]
-	for i := range a {
-		if a[i] != b[i] {
-			return i + checked
-		}
-	}
-	return len(a) + checked
-}
diff --git a/vendor/github.com/klauspost/compress/flate/huffman_bit_writer.go b/vendor/github.com/klauspost/compress/flate/huffman_bit_writer.go
deleted file mode 100644
index 3ad5e98072..0000000000
--- a/vendor/github.com/klauspost/compress/flate/huffman_bit_writer.go
+++ /dev/null
@@ -1,1034 +0,0 @@
-// Copyright 2009 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package flate
-
-import (
-	"encoding/binary"
-	"fmt"
-	"io"
-)
-
-const (
-	// The largest offset code.
-	offsetCodeCount = 30
-
-	// The special code used to mark the end of a block.
-	endBlockMarker = 256
-
-	// The first length code.
-	lengthCodesStart = 257
-
-	// The number of codegen codes.
-	codegenCodeCount = 19
-	badCode          = 255
-
-	// bufferFlushSize indicates the buffer size
-	// after which bytes are flushed to the writer.
-	// Should preferably be a multiple of 6, since
-	// we accumulate 6 bytes between writes to the buffer.
-	bufferFlushSize = 246
-
-	// bufferSize is the actual output byte buffer size.
-	// It must have additional headroom for a flush
-	// which can contain up to 8 bytes.
-	bufferSize = bufferFlushSize + 8
-)
-
-// The number of extra bits needed by length code X - LENGTH_CODES_START.
-var lengthExtraBits = [32]int8{
-	/* 257 */ 0, 0, 0,
-	/* 260 */ 0, 0, 0, 0, 0, 1, 1, 1, 1, 2,
-	/* 270 */ 2, 2, 2, 3, 3, 3, 3, 4, 4, 4,
-	/* 280 */ 4, 5, 5, 5, 5, 0,
-}
-
-// The length indicated by length code X - LENGTH_CODES_START.
-var lengthBase = [32]uint8{
-	0, 1, 2, 3, 4, 5, 6, 7, 8, 10,
-	12, 14, 16, 20, 24, 28, 32, 40, 48, 56,
-	64, 80, 96, 112, 128, 160, 192, 224, 255,
-}
-
-// offset code word extra bits.
-var offsetExtraBits = [64]int8{
-	0, 0, 0, 0, 1, 1, 2, 2, 3, 3,
-	4, 4, 5, 5, 6, 6, 7, 7, 8, 8,
-	9, 9, 10, 10, 11, 11, 12, 12, 13, 13,
-	/* extended window */
-	14, 14, 15, 15, 16, 16, 17, 17, 18, 18, 19, 19, 20, 20,
-}
-
-var offsetCombined = [32]uint32{}
-
-func init() {
-	var offsetBase = [64]uint32{
-		/* normal deflate */
-		0x000000, 0x000001, 0x000002, 0x000003, 0x000004,
-		0x000006, 0x000008, 0x00000c, 0x000010, 0x000018,
-		0x000020, 0x000030, 0x000040, 0x000060, 0x000080,
-		0x0000c0, 0x000100, 0x000180, 0x000200, 0x000300,
-		0x000400, 0x000600, 0x000800, 0x000c00, 0x001000,
-		0x001800, 0x002000, 0x003000, 0x004000, 0x006000,
-
-		/* extended window */
-		0x008000, 0x00c000, 0x010000, 0x018000, 0x020000,
-		0x030000, 0x040000, 0x060000, 0x080000, 0x0c0000,
-		0x100000, 0x180000, 0x200000, 0x300000,
-	}
-
-	for i := range offsetCombined[:] {
-		// Don't use extended window values...
-		if offsetBase[i] > 0x006000 {
-			continue
-		}
-		offsetCombined[i] = uint32(offsetExtraBits[i])<<16 | (offsetBase[i])
-	}
-}
-
-// The odd order in which the codegen code sizes are written.
-var codegenOrder = []uint32{16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15}
-
-type huffmanBitWriter struct {
-	// writer is the underlying writer.
-	// Do not use it directly; use the write method, which ensures
-	// that Write errors are sticky.
-	writer io.Writer
-
-	// Data waiting to be written is bytes[0:nbytes]
-	// and then the low nbits of bits.
-	bits            uint64
-	nbits           uint16
-	nbytes          uint8
-	lastHuffMan     bool
-	literalEncoding *huffmanEncoder
-	tmpLitEncoding  *huffmanEncoder
-	offsetEncoding  *huffmanEncoder
-	codegenEncoding *huffmanEncoder
-	err             error
-	lastHeader      int
-	// Set between 0 (reused block can be up to 2x the size)
-	logNewTablePenalty uint
-	bytes              [256 + 8]byte
-	literalFreq        [lengthCodesStart + 32]uint16
-	offsetFreq         [32]uint16
-	codegenFreq        [codegenCodeCount]uint16
-
-	// codegen must have an extra space for the final symbol.
-	codegen [literalCount + offsetCodeCount + 1]uint8
-}
-
-// Huffman reuse.
-//
-// The huffmanBitWriter supports reusing huffman tables and thereby combining block sections.
-//
-// This is controlled by several variables:
-//
-// If lastHeader is non-zero the Huffman table can be reused.
-// This also indicates that a Huffman table has been generated that can output all
-// possible symbols.
-// It also indicates that an EOB has not yet been emitted, so if a new tabel is generated
-// an EOB with the previous table must be written.
-//
-// If lastHuffMan is set, a table for outputting literals has been generated and offsets are invalid.
-//
-// An incoming block estimates the output size of a new table using a 'fresh' by calculating the
-// optimal size and adding a penalty in 'logNewTablePenalty'.
-// A Huffman table is not optimal, which is why we add a penalty, and generating a new table
-// is slower both for compression and decompression.
-
-func newHuffmanBitWriter(w io.Writer) *huffmanBitWriter {
-	return &huffmanBitWriter{
-		writer:          w,
-		literalEncoding: newHuffmanEncoder(literalCount),
-		tmpLitEncoding:  newHuffmanEncoder(literalCount),
-		codegenEncoding: newHuffmanEncoder(codegenCodeCount),
-		offsetEncoding:  newHuffmanEncoder(offsetCodeCount),
-	}
-}
-
-func (w *huffmanBitWriter) reset(writer io.Writer) {
-	w.writer = writer
-	w.bits, w.nbits, w.nbytes, w.err = 0, 0, 0, nil
-	w.lastHeader = 0
-	w.lastHuffMan = false
-}
-
-func (w *huffmanBitWriter) canReuse(t *tokens) (offsets, lits bool) {
-	offsets, lits = true, true
-	a := t.offHist[:offsetCodeCount]
-	b := w.offsetFreq[:len(a)]
-	for i := range a {
-		if b[i] == 0 && a[i] != 0 {
-			offsets = false
-			break
-		}
-	}
-
-	a = t.extraHist[:literalCount-256]
-	b = w.literalFreq[256:literalCount]
-	b = b[:len(a)]
-	for i := range a {
-		if b[i] == 0 && a[i] != 0 {
-			lits = false
-			break
-		}
-	}
-	if lits {
-		a = t.litHist[:]
-		b = w.literalFreq[:len(a)]
-		for i := range a {
-			if b[i] == 0 && a[i] != 0 {
-				lits = false
-				break
-			}
-		}
-	}
-	return
-}
-
-func (w *huffmanBitWriter) flush() {
-	if w.err != nil {
-		w.nbits = 0
-		return
-	}
-	if w.lastHeader > 0 {
-		// We owe an EOB
-		w.writeCode(w.literalEncoding.codes[endBlockMarker])
-		w.lastHeader = 0
-	}
-	n := w.nbytes
-	for w.nbits != 0 {
-		w.bytes[n] = byte(w.bits)
-		w.bits >>= 8
-		if w.nbits > 8 { // Avoid underflow
-			w.nbits -= 8
-		} else {
-			w.nbits = 0
-		}
-		n++
-	}
-	w.bits = 0
-	w.write(w.bytes[:n])
-	w.nbytes = 0
-}
-
-func (w *huffmanBitWriter) write(b []byte) {
-	if w.err != nil {
-		return
-	}
-	_, w.err = w.writer.Write(b)
-}
-
-func (w *huffmanBitWriter) writeBits(b int32, nb uint16) {
-	w.bits |= uint64(b) << w.nbits
-	w.nbits += nb
-	if w.nbits >= 48 {
-		w.writeOutBits()
-	}
-}
-
-func (w *huffmanBitWriter) writeBytes(bytes []byte) {
-	if w.err != nil {
-		return
-	}
-	n := w.nbytes
-	if w.nbits&7 != 0 {
-		w.err = InternalError("writeBytes with unfinished bits")
-		return
-	}
-	for w.nbits != 0 {
-		w.bytes[n] = byte(w.bits)
-		w.bits >>= 8
-		w.nbits -= 8
-		n++
-	}
-	if n != 0 {
-		w.write(w.bytes[:n])
-	}
-	w.nbytes = 0
-	w.write(bytes)
-}
-
-// RFC 1951 3.2.7 specifies a special run-length encoding for specifying
-// the literal and offset lengths arrays (which are concatenated into a single
-// array).  This method generates that run-length encoding.
-//
-// The result is written into the codegen array, and the frequencies
-// of each code is written into the codegenFreq array.
-// Codes 0-15 are single byte codes. Codes 16-18 are followed by additional
-// information. Code badCode is an end marker
-//
-//  numLiterals      The number of literals in literalEncoding
-//  numOffsets       The number of offsets in offsetEncoding
-//  litenc, offenc   The literal and offset encoder to use
-func (w *huffmanBitWriter) generateCodegen(numLiterals int, numOffsets int, litEnc, offEnc *huffmanEncoder) {
-	for i := range w.codegenFreq {
-		w.codegenFreq[i] = 0
-	}
-	// Note that we are using codegen both as a temporary variable for holding
-	// a copy of the frequencies, and as the place where we put the result.
-	// This is fine because the output is always shorter than the input used
-	// so far.
-	codegen := w.codegen[:] // cache
-	// Copy the concatenated code sizes to codegen. Put a marker at the end.
-	cgnl := codegen[:numLiterals]
-	for i := range cgnl {
-		cgnl[i] = uint8(litEnc.codes[i].len)
-	}
-
-	cgnl = codegen[numLiterals : numLiterals+numOffsets]
-	for i := range cgnl {
-		cgnl[i] = uint8(offEnc.codes[i].len)
-	}
-	codegen[numLiterals+numOffsets] = badCode
-
-	size := codegen[0]
-	count := 1
-	outIndex := 0
-	for inIndex := 1; size != badCode; inIndex++ {
-		// INVARIANT: We have seen "count" copies of size that have not yet
-		// had output generated for them.
-		nextSize := codegen[inIndex]
-		if nextSize == size {
-			count++
-			continue
-		}
-		// We need to generate codegen indicating "count" of size.
-		if size != 0 {
-			codegen[outIndex] = size
-			outIndex++
-			w.codegenFreq[size]++
-			count--
-			for count >= 3 {
-				n := 6
-				if n > count {
-					n = count
-				}
-				codegen[outIndex] = 16
-				outIndex++
-				codegen[outIndex] = uint8(n - 3)
-				outIndex++
-				w.codegenFreq[16]++
-				count -= n
-			}
-		} else {
-			for count >= 11 {
-				n := 138
-				if n > count {
-					n = count
-				}
-				codegen[outIndex] = 18
-				outIndex++
-				codegen[outIndex] = uint8(n - 11)
-				outIndex++
-				w.codegenFreq[18]++
-				count -= n
-			}
-			if count >= 3 {
-				// count >= 3 && count <= 10
-				codegen[outIndex] = 17
-				outIndex++
-				codegen[outIndex] = uint8(count - 3)
-				outIndex++
-				w.codegenFreq[17]++
-				count = 0
-			}
-		}
-		count--
-		for ; count >= 0; count-- {
-			codegen[outIndex] = size
-			outIndex++
-			w.codegenFreq[size]++
-		}
-		// Set up invariant for next time through the loop.
-		size = nextSize
-		count = 1
-	}
-	// Marker indicating the end of the codegen.
-	codegen[outIndex] = badCode
-}
-
-func (w *huffmanBitWriter) codegens() int {
-	numCodegens := len(w.codegenFreq)
-	for numCodegens > 4 && w.codegenFreq[codegenOrder[numCodegens-1]] == 0 {
-		numCodegens--
-	}
-	return numCodegens
-}
-
-func (w *huffmanBitWriter) headerSize() (size, numCodegens int) {
-	numCodegens = len(w.codegenFreq)
-	for numCodegens > 4 && w.codegenFreq[codegenOrder[numCodegens-1]] == 0 {
-		numCodegens--
-	}
-	return 3 + 5 + 5 + 4 + (3 * numCodegens) +
-		w.codegenEncoding.bitLength(w.codegenFreq[:]) +
-		int(w.codegenFreq[16])*2 +
-		int(w.codegenFreq[17])*3 +
-		int(w.codegenFreq[18])*7, numCodegens
-}
-
-// dynamicSize returns the size of dynamically encoded data in bits.
-func (w *huffmanBitWriter) dynamicReuseSize(litEnc, offEnc *huffmanEncoder) (size int) {
-	size = litEnc.bitLength(w.literalFreq[:]) +
-		offEnc.bitLength(w.offsetFreq[:])
-	return size
-}
-
-// dynamicSize returns the size of dynamically encoded data in bits.
-func (w *huffmanBitWriter) dynamicSize(litEnc, offEnc *huffmanEncoder, extraBits int) (size, numCodegens int) {
-	header, numCodegens := w.headerSize()
-	size = header +
-		litEnc.bitLength(w.literalFreq[:]) +
-		offEnc.bitLength(w.offsetFreq[:]) +
-		extraBits
-	return size, numCodegens
-}
-
-// extraBitSize will return the number of bits that will be written
-// as "extra" bits on matches.
-func (w *huffmanBitWriter) extraBitSize() int {
-	total := 0
-	for i, n := range w.literalFreq[257:literalCount] {
-		total += int(n) * int(lengthExtraBits[i&31])
-	}
-	for i, n := range w.offsetFreq[:offsetCodeCount] {
-		total += int(n) * int(offsetExtraBits[i&31])
-	}
-	return total
-}
-
-// fixedSize returns the size of dynamically encoded data in bits.
-func (w *huffmanBitWriter) fixedSize(extraBits int) int {
-	return 3 +
-		fixedLiteralEncoding.bitLength(w.literalFreq[:]) +
-		fixedOffsetEncoding.bitLength(w.offsetFreq[:]) +
-		extraBits
-}
-
-// storedSize calculates the stored size, including header.
-// The function returns the size in bits and whether the block
-// fits inside a single block.
-func (w *huffmanBitWriter) storedSize(in []byte) (int, bool) {
-	if in == nil {
-		return 0, false
-	}
-	if len(in) <= maxStoreBlockSize {
-		return (len(in) + 5) * 8, true
-	}
-	return 0, false
-}
-
-func (w *huffmanBitWriter) writeCode(c hcode) {
-	// The function does not get inlined if we "& 63" the shift.
-	w.bits |= uint64(c.code) << w.nbits
-	w.nbits += c.len
-	if w.nbits >= 48 {
-		w.writeOutBits()
-	}
-}
-
-// writeOutBits will write bits to the buffer.
-func (w *huffmanBitWriter) writeOutBits() {
-	bits := w.bits
-	w.bits >>= 48
-	w.nbits -= 48
-	n := w.nbytes
-
-	// We over-write, but faster...
-	binary.LittleEndian.PutUint64(w.bytes[n:], bits)
-	n += 6
-
-	if n >= bufferFlushSize {
-		if w.err != nil {
-			n = 0
-			return
-		}
-		w.write(w.bytes[:n])
-		n = 0
-	}
-
-	w.nbytes = n
-}
-
-// Write the header of a dynamic Huffman block to the output stream.
-//
-//  numLiterals  The number of literals specified in codegen
-//  numOffsets   The number of offsets specified in codegen
-//  numCodegens  The number of codegens used in codegen
-func (w *huffmanBitWriter) writeDynamicHeader(numLiterals int, numOffsets int, numCodegens int, isEof bool) {
-	if w.err != nil {
-		return
-	}
-	var firstBits int32 = 4
-	if isEof {
-		firstBits = 5
-	}
-	w.writeBits(firstBits, 3)
-	w.writeBits(int32(numLiterals-257), 5)
-	w.writeBits(int32(numOffsets-1), 5)
-	w.writeBits(int32(numCodegens-4), 4)
-
-	for i := 0; i < numCodegens; i++ {
-		value := uint(w.codegenEncoding.codes[codegenOrder[i]].len)
-		w.writeBits(int32(value), 3)
-	}
-
-	i := 0
-	for {
-		var codeWord = uint32(w.codegen[i])
-		i++
-		if codeWord == badCode {
-			break
-		}
-		w.writeCode(w.codegenEncoding.codes[codeWord])
-
-		switch codeWord {
-		case 16:
-			w.writeBits(int32(w.codegen[i]), 2)
-			i++
-		case 17:
-			w.writeBits(int32(w.codegen[i]), 3)
-			i++
-		case 18:
-			w.writeBits(int32(w.codegen[i]), 7)
-			i++
-		}
-	}
-}
-
-// writeStoredHeader will write a stored header.
-// If the stored block is only used for EOF,
-// it is replaced with a fixed huffman block.
-func (w *huffmanBitWriter) writeStoredHeader(length int, isEof bool) {
-	if w.err != nil {
-		return
-	}
-	if w.lastHeader > 0 {
-		// We owe an EOB
-		w.writeCode(w.literalEncoding.codes[endBlockMarker])
-		w.lastHeader = 0
-	}
-
-	// To write EOF, use a fixed encoding block. 10 bits instead of 5 bytes.
-	if length == 0 && isEof {
-		w.writeFixedHeader(isEof)
-		// EOB: 7 bits, value: 0
-		w.writeBits(0, 7)
-		w.flush()
-		return
-	}
-
-	var flag int32
-	if isEof {
-		flag = 1
-	}
-	w.writeBits(flag, 3)
-	w.flush()
-	w.writeBits(int32(length), 16)
-	w.writeBits(int32(^uint16(length)), 16)
-}
-
-func (w *huffmanBitWriter) writeFixedHeader(isEof bool) {
-	if w.err != nil {
-		return
-	}
-	if w.lastHeader > 0 {
-		// We owe an EOB
-		w.writeCode(w.literalEncoding.codes[endBlockMarker])
-		w.lastHeader = 0
-	}
-
-	// Indicate that we are a fixed Huffman block
-	var value int32 = 2
-	if isEof {
-		value = 3
-	}
-	w.writeBits(value, 3)
-}
-
-// writeBlock will write a block of tokens with the smallest encoding.
-// The original input can be supplied, and if the huffman encoded data
-// is larger than the original bytes, the data will be written as a
-// stored block.
-// If the input is nil, the tokens will always be Huffman encoded.
-func (w *huffmanBitWriter) writeBlock(tokens *tokens, eof bool, input []byte) {
-	if w.err != nil {
-		return
-	}
-
-	tokens.AddEOB()
-	if w.lastHeader > 0 {
-		// We owe an EOB
-		w.writeCode(w.literalEncoding.codes[endBlockMarker])
-		w.lastHeader = 0
-	}
-	numLiterals, numOffsets := w.indexTokens(tokens, false)
-	w.generate(tokens)
-	var extraBits int
-	storedSize, storable := w.storedSize(input)
-	if storable {
-		extraBits = w.extraBitSize()
-	}
-
-	// Figure out smallest code.
-	// Fixed Huffman baseline.
-	var literalEncoding = fixedLiteralEncoding
-	var offsetEncoding = fixedOffsetEncoding
-	var size = w.fixedSize(extraBits)
-
-	// Dynamic Huffman?
-	var numCodegens int
-
-	// Generate codegen and codegenFrequencies, which indicates how to encode
-	// the literalEncoding and the offsetEncoding.
-	w.generateCodegen(numLiterals, numOffsets, w.literalEncoding, w.offsetEncoding)
-	w.codegenEncoding.generate(w.codegenFreq[:], 7)
-	dynamicSize, numCodegens := w.dynamicSize(w.literalEncoding, w.offsetEncoding, extraBits)
-
-	if dynamicSize < size {
-		size = dynamicSize
-		literalEncoding = w.literalEncoding
-		offsetEncoding = w.offsetEncoding
-	}
-
-	// Stored bytes?
-	if storable && storedSize < size {
-		w.writeStoredHeader(len(input), eof)
-		w.writeBytes(input)
-		return
-	}
-
-	// Huffman.
-	if literalEncoding == fixedLiteralEncoding {
-		w.writeFixedHeader(eof)
-	} else {
-		w.writeDynamicHeader(numLiterals, numOffsets, numCodegens, eof)
-	}
-
-	// Write the tokens.
-	w.writeTokens(tokens.Slice(), literalEncoding.codes, offsetEncoding.codes)
-}
-
-// writeBlockDynamic encodes a block using a dynamic Huffman table.
-// This should be used if the symbols used have a disproportionate
-// histogram distribution.
-// If input is supplied and the compression savings are below 1/16th of the
-// input size the block is stored.
-func (w *huffmanBitWriter) writeBlockDynamic(tokens *tokens, eof bool, input []byte, sync bool) {
-	if w.err != nil {
-		return
-	}
-
-	sync = sync || eof
-	if sync {
-		tokens.AddEOB()
-	}
-
-	// We cannot reuse pure huffman table, and must mark as EOF.
-	if (w.lastHuffMan || eof) && w.lastHeader > 0 {
-		// We will not try to reuse.
-		w.writeCode(w.literalEncoding.codes[endBlockMarker])
-		w.lastHeader = 0
-		w.lastHuffMan = false
-	}
-	if !sync {
-		tokens.Fill()
-	}
-	numLiterals, numOffsets := w.indexTokens(tokens, !sync)
-
-	var size int
-	// Check if we should reuse.
-	if w.lastHeader > 0 {
-		// Estimate size for using a new table.
-		// Use the previous header size as the best estimate.
-		newSize := w.lastHeader + tokens.EstimatedBits()
-		newSize += newSize >> w.logNewTablePenalty
-
-		// The estimated size is calculated as an optimal table.
-		// We add a penalty to make it more realistic and re-use a bit more.
-		reuseSize := w.dynamicReuseSize(w.literalEncoding, w.offsetEncoding) + w.extraBitSize()
-
-		// Check if a new table is better.
-		if newSize < reuseSize {
-			// Write the EOB we owe.
-			w.writeCode(w.literalEncoding.codes[endBlockMarker])
-			size = newSize
-			w.lastHeader = 0
-		} else {
-			size = reuseSize
-		}
-		// Check if we get a reasonable size decrease.
-		if ssize, storable := w.storedSize(input); storable && ssize < (size+size>>4) {
-			w.writeStoredHeader(len(input), eof)
-			w.writeBytes(input)
-			w.lastHeader = 0
-			return
-		}
-	}
-
-	// We want a new block/table
-	if w.lastHeader == 0 {
-		w.generate(tokens)
-		// Generate codegen and codegenFrequencies, which indicates how to encode
-		// the literalEncoding and the offsetEncoding.
-		w.generateCodegen(numLiterals, numOffsets, w.literalEncoding, w.offsetEncoding)
-		w.codegenEncoding.generate(w.codegenFreq[:], 7)
-		var numCodegens int
-		size, numCodegens = w.dynamicSize(w.literalEncoding, w.offsetEncoding, w.extraBitSize())
-		// Store bytes, if we don't get a reasonable improvement.
-		if ssize, storable := w.storedSize(input); storable && ssize < (size+size>>4) {
-			w.writeStoredHeader(len(input), eof)
-			w.writeBytes(input)
-			w.lastHeader = 0
-			return
-		}
-
-		// Write Huffman table.
-		w.writeDynamicHeader(numLiterals, numOffsets, numCodegens, eof)
-		w.lastHeader, _ = w.headerSize()
-		w.lastHuffMan = false
-	}
-
-	if sync {
-		w.lastHeader = 0
-	}
-	// Write the tokens.
-	w.writeTokens(tokens.Slice(), w.literalEncoding.codes, w.offsetEncoding.codes)
-}
-
-// indexTokens indexes a slice of tokens, and updates
-// literalFreq and offsetFreq, and generates literalEncoding
-// and offsetEncoding.
-// The number of literal and offset tokens is returned.
-func (w *huffmanBitWriter) indexTokens(t *tokens, filled bool) (numLiterals, numOffsets int) {
-	copy(w.literalFreq[:], t.litHist[:])
-	copy(w.literalFreq[256:], t.extraHist[:])
-	copy(w.offsetFreq[:], t.offHist[:offsetCodeCount])
-
-	if t.n == 0 {
-		return
-	}
-	if filled {
-		return maxNumLit, maxNumDist
-	}
-	// get the number of literals
-	numLiterals = len(w.literalFreq)
-	for w.literalFreq[numLiterals-1] == 0 {
-		numLiterals--
-	}
-	// get the number of offsets
-	numOffsets = len(w.offsetFreq)
-	for numOffsets > 0 && w.offsetFreq[numOffsets-1] == 0 {
-		numOffsets--
-	}
-	if numOffsets == 0 {
-		// We haven't found a single match. If we want to go with the dynamic encoding,
-		// we should count at least one offset to be sure that the offset huffman tree could be encoded.
-		w.offsetFreq[0] = 1
-		numOffsets = 1
-	}
-	return
-}
-
-func (w *huffmanBitWriter) generate(t *tokens) {
-	w.literalEncoding.generate(w.literalFreq[:literalCount], 15)
-	w.offsetEncoding.generate(w.offsetFreq[:offsetCodeCount], 15)
-}
-
-// writeTokens writes a slice of tokens to the output.
-// codes for literal and offset encoding must be supplied.
-func (w *huffmanBitWriter) writeTokens(tokens []token, leCodes, oeCodes []hcode) {
-	if w.err != nil {
-		return
-	}
-	if len(tokens) == 0 {
-		return
-	}
-
-	// Only last token should be endBlockMarker.
-	var deferEOB bool
-	if tokens[len(tokens)-1] == endBlockMarker {
-		tokens = tokens[:len(tokens)-1]
-		deferEOB = true
-	}
-
-	// Create slices up to the next power of two to avoid bounds checks.
-	lits := leCodes[:256]
-	offs := oeCodes[:32]
-	lengths := leCodes[lengthCodesStart:]
-	lengths = lengths[:32]
-
-	// Go 1.16 LOVES having these on stack.
-	bits, nbits, nbytes := w.bits, w.nbits, w.nbytes
-
-	for _, t := range tokens {
-		if t < matchType {
-			//w.writeCode(lits[t.literal()])
-			c := lits[t.literal()]
-			bits |= uint64(c.code) << nbits
-			nbits += c.len
-			if nbits >= 48 {
-				binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits)
-				//*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits
-				bits >>= 48
-				nbits -= 48
-				nbytes += 6
-				if nbytes >= bufferFlushSize {
-					if w.err != nil {
-						nbytes = 0
-						return
-					}
-					_, w.err = w.writer.Write(w.bytes[:nbytes])
-					nbytes = 0
-				}
-			}
-			continue
-		}
-
-		// Write the length
-		length := t.length()
-		lengthCode := lengthCode(length)
-		if false {
-			w.writeCode(lengths[lengthCode&31])
-		} else {
-			// inlined
-			c := lengths[lengthCode&31]
-			bits |= uint64(c.code) << nbits
-			nbits += c.len
-			if nbits >= 48 {
-				binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits)
-				//*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits
-				bits >>= 48
-				nbits -= 48
-				nbytes += 6
-				if nbytes >= bufferFlushSize {
-					if w.err != nil {
-						nbytes = 0
-						return
-					}
-					_, w.err = w.writer.Write(w.bytes[:nbytes])
-					nbytes = 0
-				}
-			}
-		}
-
-		extraLengthBits := uint16(lengthExtraBits[lengthCode&31])
-		if extraLengthBits > 0 {
-			//w.writeBits(extraLength, extraLengthBits)
-			extraLength := int32(length - lengthBase[lengthCode&31])
-			bits |= uint64(extraLength) << nbits
-			nbits += extraLengthBits
-			if nbits >= 48 {
-				binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits)
-				//*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits
-				bits >>= 48
-				nbits -= 48
-				nbytes += 6
-				if nbytes >= bufferFlushSize {
-					if w.err != nil {
-						nbytes = 0
-						return
-					}
-					_, w.err = w.writer.Write(w.bytes[:nbytes])
-					nbytes = 0
-				}
-			}
-		}
-		// Write the offset
-		offset := t.offset()
-		offsetCode := offset >> 16
-		offset &= matchOffsetOnlyMask
-		if false {
-			w.writeCode(offs[offsetCode&31])
-		} else {
-			// inlined
-			c := offs[offsetCode]
-			bits |= uint64(c.code) << nbits
-			nbits += c.len
-			if nbits >= 48 {
-				binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits)
-				//*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits
-				bits >>= 48
-				nbits -= 48
-				nbytes += 6
-				if nbytes >= bufferFlushSize {
-					if w.err != nil {
-						nbytes = 0
-						return
-					}
-					_, w.err = w.writer.Write(w.bytes[:nbytes])
-					nbytes = 0
-				}
-			}
-		}
-		offsetComb := offsetCombined[offsetCode]
-		if offsetComb > 1<<16 {
-			//w.writeBits(extraOffset, extraOffsetBits)
-			bits |= uint64(offset&matchOffsetOnlyMask-(offsetComb&0xffff)) << nbits
-			nbits += uint16(offsetComb >> 16)
-			if nbits >= 48 {
-				binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits)
-				//*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits
-				bits >>= 48
-				nbits -= 48
-				nbytes += 6
-				if nbytes >= bufferFlushSize {
-					if w.err != nil {
-						nbytes = 0
-						return
-					}
-					_, w.err = w.writer.Write(w.bytes[:nbytes])
-					nbytes = 0
-				}
-			}
-		}
-	}
-	// Restore...
-	w.bits, w.nbits, w.nbytes = bits, nbits, nbytes
-
-	if deferEOB {
-		w.writeCode(leCodes[endBlockMarker])
-	}
-}
-
-// huffOffset is a static offset encoder used for huffman only encoding.
-// It can be reused since we will not be encoding offset values.
-var huffOffset *huffmanEncoder
-
-func init() {
-	w := newHuffmanBitWriter(nil)
-	w.offsetFreq[0] = 1
-	huffOffset = newHuffmanEncoder(offsetCodeCount)
-	huffOffset.generate(w.offsetFreq[:offsetCodeCount], 15)
-}
-
-// writeBlockHuff encodes a block of bytes as either
-// Huffman encoded literals or uncompressed bytes if the
-// results only gains very little from compression.
-func (w *huffmanBitWriter) writeBlockHuff(eof bool, input []byte, sync bool) {
-	if w.err != nil {
-		return
-	}
-
-	// Clear histogram
-	for i := range w.literalFreq[:] {
-		w.literalFreq[i] = 0
-	}
-	if !w.lastHuffMan {
-		for i := range w.offsetFreq[:] {
-			w.offsetFreq[i] = 0
-		}
-	}
-
-	// Fill is rarely better...
-	const fill = false
-	const numLiterals = endBlockMarker + 1
-	const numOffsets = 1
-
-	// Add everything as literals
-	// We have to estimate the header size.
-	// Assume header is around 70 bytes:
-	// https://stackoverflow.com/a/25454430
-	const guessHeaderSizeBits = 70 * 8
-	histogram(input, w.literalFreq[:numLiterals], fill)
-	w.literalFreq[endBlockMarker] = 1
-	w.tmpLitEncoding.generate(w.literalFreq[:numLiterals], 15)
-	if fill {
-		// Clear fill...
-		for i := range w.literalFreq[:numLiterals] {
-			w.literalFreq[i] = 0
-		}
-		histogram(input, w.literalFreq[:numLiterals], false)
-	}
-	estBits := w.tmpLitEncoding.canReuseBits(w.literalFreq[:numLiterals])
-	estBits += w.lastHeader
-	if w.lastHeader == 0 {
-		estBits += guessHeaderSizeBits
-	}
-	estBits += estBits >> w.logNewTablePenalty
-
-	// Store bytes, if we don't get a reasonable improvement.
-	ssize, storable := w.storedSize(input)
-	if storable && ssize <= estBits {
-		w.writeStoredHeader(len(input), eof)
-		w.writeBytes(input)
-		return
-	}
-
-	if w.lastHeader > 0 {
-		reuseSize := w.literalEncoding.canReuseBits(w.literalFreq[:256])
-
-		if estBits < reuseSize {
-			if debugDeflate {
-				//fmt.Println("not reusing, reuse:", reuseSize/8, "> new:", estBits/8, "- header est:", w.lastHeader/8)
-			}
-			// We owe an EOB
-			w.writeCode(w.literalEncoding.codes[endBlockMarker])
-			w.lastHeader = 0
-		} else if debugDeflate {
-			fmt.Println("reusing, reuse:", reuseSize/8, "> new:", estBits/8, "- header est:", w.lastHeader/8)
-		}
-	}
-
-	count := 0
-	if w.lastHeader == 0 {
-		// Use the temp encoding, so swap.
-		w.literalEncoding, w.tmpLitEncoding = w.tmpLitEncoding, w.literalEncoding
-		// Generate codegen and codegenFrequencies, which indicates how to encode
-		// the literalEncoding and the offsetEncoding.
-		w.generateCodegen(numLiterals, numOffsets, w.literalEncoding, huffOffset)
-		w.codegenEncoding.generate(w.codegenFreq[:], 7)
-		numCodegens := w.codegens()
-
-		// Huffman.
-		w.writeDynamicHeader(numLiterals, numOffsets, numCodegens, eof)
-		w.lastHuffMan = true
-		w.lastHeader, _ = w.headerSize()
-		if debugDeflate {
-			count += w.lastHeader
-			fmt.Println("header:", count/8)
-		}
-	}
-
-	encoding := w.literalEncoding.codes[:256]
-	// Go 1.16 LOVES having these on stack. At least 1.5x the speed.
-	bits, nbits, nbytes := w.bits, w.nbits, w.nbytes
-	for _, t := range input {
-		// Bitwriting inlined, ~30% speedup
-		c := encoding[t]
-		bits |= uint64(c.code) << nbits
-		nbits += c.len
-		if debugDeflate {
-			count += int(c.len)
-		}
-		if nbits >= 48 {
-			binary.LittleEndian.PutUint64(w.bytes[nbytes:], bits)
-			//*(*uint64)(unsafe.Pointer(&w.bytes[nbytes])) = bits
-			bits >>= 48
-			nbits -= 48
-			nbytes += 6
-			if nbytes >= bufferFlushSize {
-				if w.err != nil {
-					nbytes = 0
-					return
-				}
-				_, w.err = w.writer.Write(w.bytes[:nbytes])
-				nbytes = 0
-			}
-		}
-	}
-	// Restore...
-	w.bits, w.nbits, w.nbytes = bits, nbits, nbytes
-
-	if debugDeflate {
-		fmt.Println("wrote", count/8, "bytes")
-	}
-	if eof || sync {
-		w.writeCode(w.literalEncoding.codes[endBlockMarker])
-		w.lastHeader = 0
-		w.lastHuffMan = false
-	}
-}
diff --git a/vendor/github.com/klauspost/compress/flate/huffman_code.go b/vendor/github.com/klauspost/compress/flate/huffman_code.go
deleted file mode 100644
index 67b2b38728..0000000000
--- a/vendor/github.com/klauspost/compress/flate/huffman_code.go
+++ /dev/null
@@ -1,374 +0,0 @@
-// Copyright 2009 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package flate
-
-import (
-	"math"
-	"math/bits"
-)
-
-const (
-	maxBitsLimit = 16
-	// number of valid literals
-	literalCount = 286
-)
-
-// hcode is a huffman code with a bit code and bit length.
-type hcode struct {
-	code, len uint16
-}
-
-type huffmanEncoder struct {
-	codes    []hcode
-	bitCount [17]int32
-
-	// Allocate a reusable buffer with the longest possible frequency table.
-	// Possible lengths are codegenCodeCount, offsetCodeCount and literalCount.
-	// The largest of these is literalCount, so we allocate for that case.
-	freqcache [literalCount + 1]literalNode
-}
-
-type literalNode struct {
-	literal uint16
-	freq    uint16
-}
-
-// A levelInfo describes the state of the constructed tree for a given depth.
-type levelInfo struct {
-	// Our level.  for better printing
-	level int32
-
-	// The frequency of the last node at this level
-	lastFreq int32
-
-	// The frequency of the next character to add to this level
-	nextCharFreq int32
-
-	// The frequency of the next pair (from level below) to add to this level.
-	// Only valid if the "needed" value of the next lower level is 0.
-	nextPairFreq int32
-
-	// The number of chains remaining to generate for this level before moving
-	// up to the next level
-	needed int32
-}
-
-// set sets the code and length of an hcode.
-func (h *hcode) set(code uint16, length uint16) {
-	h.len = length
-	h.code = code
-}
-
-func reverseBits(number uint16, bitLength byte) uint16 {
-	return bits.Reverse16(number << ((16 - bitLength) & 15))
-}
-
-func maxNode() literalNode { return literalNode{math.MaxUint16, math.MaxUint16} }
-
-func newHuffmanEncoder(size int) *huffmanEncoder {
-	// Make capacity to next power of two.
-	c := uint(bits.Len32(uint32(size - 1)))
-	return &huffmanEncoder{codes: make([]hcode, size, 1<<c)}
-}
-
-// Generates a HuffmanCode corresponding to the fixed literal table
-func generateFixedLiteralEncoding() *huffmanEncoder {
-	h := newHuffmanEncoder(literalCount)
-	codes := h.codes
-	var ch uint16
-	for ch = 0; ch < literalCount; ch++ {
-		var bits uint16
-		var size uint16
-		switch {
-		case ch < 144:
-			// size 8, 000110000  .. 10111111
-			bits = ch + 48
-			size = 8
-		case ch < 256:
-			// size 9, 110010000 .. 111111111
-			bits = ch + 400 - 144
-			size = 9
-		case ch < 280:
-			// size 7, 0000000 .. 0010111
-			bits = ch - 256
-			size = 7
-		default:
-			// size 8, 11000000 .. 11000111
-			bits = ch + 192 - 280
-			size = 8
-		}
-		codes[ch] = hcode{code: reverseBits(bits, byte(size)), len: size}
-	}
-	return h
-}
-
-func generateFixedOffsetEncoding() *huffmanEncoder {
-	h := newHuffmanEncoder(30)
-	codes := h.codes
-	for ch := range codes {
-		codes[ch] = hcode{code: reverseBits(uint16(ch), 5), len: 5}
-	}
-	return h
-}
-
-var fixedLiteralEncoding = generateFixedLiteralEncoding()
-var fixedOffsetEncoding = generateFixedOffsetEncoding()
-
-func (h *huffmanEncoder) bitLength(freq []uint16) int {
-	var total int
-	for i, f := range freq {
-		if f != 0 {
-			total += int(f) * int(h.codes[i].len)
-		}
-	}
-	return total
-}
-
-func (h *huffmanEncoder) bitLengthRaw(b []byte) int {
-	var total int
-	for _, f := range b {
-		if f != 0 {
-			total += int(h.codes[f].len)
-		}
-	}
-	return total
-}
-
-// canReuseBits returns the number of bits or math.MaxInt32 if the encoder cannot be reused.
-func (h *huffmanEncoder) canReuseBits(freq []uint16) int {
-	var total int
-	for i, f := range freq {
-		if f != 0 {
-			code := h.codes[i]
-			if code.len == 0 {
-				return math.MaxInt32
-			}
-			total += int(f) * int(code.len)
-		}
-	}
-	return total
-}
-
-// Return the number of literals assigned to each bit size in the Huffman encoding
-//
-// This method is only called when list.length >= 3
-// The cases of 0, 1, and 2 literals are handled by special case code.
-//
-// list  An array of the literals with non-zero frequencies
-//             and their associated frequencies. The array is in order of increasing
-//             frequency, and has as its last element a special element with frequency
-//             MaxInt32
-// maxBits     The maximum number of bits that should be used to encode any literal.
-//             Must be less than 16.
-// return      An integer array in which array[i] indicates the number of literals
-//             that should be encoded in i bits.
-func (h *huffmanEncoder) bitCounts(list []literalNode, maxBits int32) []int32 {
-	if maxBits >= maxBitsLimit {
-		panic("flate: maxBits too large")
-	}
-	n := int32(len(list))
-	list = list[0 : n+1]
-	list[n] = maxNode()
-
-	// The tree can't have greater depth than n - 1, no matter what. This
-	// saves a little bit of work in some small cases
-	if maxBits > n-1 {
-		maxBits = n - 1
-	}
-
-	// Create information about each of the levels.
-	// A bogus "Level 0" whose sole purpose is so that
-	// level1.prev.needed==0.  This makes level1.nextPairFreq
-	// be a legitimate value that never gets chosen.
-	var levels [maxBitsLimit]levelInfo
-	// leafCounts[i] counts the number of literals at the left
-	// of ancestors of the rightmost node at level i.
-	// leafCounts[i][j] is the number of literals at the left
-	// of the level j ancestor.
-	var leafCounts [maxBitsLimit][maxBitsLimit]int32
-
-	for level := int32(1); level <= maxBits; level++ {
-		// For every level, the first two items are the first two characters.
-		// We initialize the levels as if we had already figured this out.
-		levels[level] = levelInfo{
-			level:        level,
-			lastFreq:     int32(list[1].freq),
-			nextCharFreq: int32(list[2].freq),
-			nextPairFreq: int32(list[0].freq) + int32(list[1].freq),
-		}
-		leafCounts[level][level] = 2
-		if level == 1 {
-			levels[level].nextPairFreq = math.MaxInt32
-		}
-	}
-
-	// We need a total of 2*n - 2 items at top level and have already generated 2.
-	levels[maxBits].needed = 2*n - 4
-
-	level := maxBits
-	for {
-		l := &levels[level]
-		if l.nextPairFreq == math.MaxInt32 && l.nextCharFreq == math.MaxInt32 {
-			// We've run out of both leafs and pairs.
-			// End all calculations for this level.
-			// To make sure we never come back to this level or any lower level,
-			// set nextPairFreq impossibly large.
-			l.needed = 0
-			levels[level+1].nextPairFreq = math.MaxInt32
-			level++
-			continue
-		}
-
-		prevFreq := l.lastFreq
-		if l.nextCharFreq < l.nextPairFreq {
-			// The next item on this row is a leaf node.
-			n := leafCounts[level][level] + 1
-			l.lastFreq = l.nextCharFreq
-			// Lower leafCounts are the same of the previous node.
-			leafCounts[level][level] = n
-			e := list[n]
-			if e.literal < math.MaxUint16 {
-				l.nextCharFreq = int32(e.freq)
-			} else {
-				l.nextCharFreq = math.MaxInt32
-			}
-		} else {
-			// The next item on this row is a pair from the previous row.
-			// nextPairFreq isn't valid until we generate two
-			// more values in the level below
-			l.lastFreq = l.nextPairFreq
-			// Take leaf counts from the lower level, except counts[level] remains the same.
-			copy(leafCounts[level][:level], leafCounts[level-1][:level])
-			levels[l.level-1].needed = 2
-		}
-
-		if l.needed--; l.needed == 0 {
-			// We've done everything we need to do for this level.
-			// Continue calculating one level up. Fill in nextPairFreq
-			// of that level with the sum of the two nodes we've just calculated on
-			// this level.
-			if l.level == maxBits {
-				// All done!
-				break
-			}
-			levels[l.level+1].nextPairFreq = prevFreq + l.lastFreq
-			level++
-		} else {
-			// If we stole from below, move down temporarily to replenish it.
-			for levels[level-1].needed > 0 {
-				level--
-			}
-		}
-	}
-
-	// Somethings is wrong if at the end, the top level is null or hasn't used
-	// all of the leaves.
-	if leafCounts[maxBits][maxBits] != n {
-		panic("leafCounts[maxBits][maxBits] != n")
-	}
-
-	bitCount := h.bitCount[:maxBits+1]
-	bits := 1
-	counts := &leafCounts[maxBits]
-	for level := maxBits; level > 0; level-- {
-		// chain.leafCount gives the number of literals requiring at least "bits"
-		// bits to encode.
-		bitCount[bits] = counts[level] - counts[level-1]
-		bits++
-	}
-	return bitCount
-}
-
-// Look at the leaves and assign them a bit count and an encoding as specified
-// in RFC 1951 3.2.2
-func (h *huffmanEncoder) assignEncodingAndSize(bitCount []int32, list []literalNode) {
-	code := uint16(0)
-	for n, bits := range bitCount {
-		code <<= 1
-		if n == 0 || bits == 0 {
-			continue
-		}
-		// The literals list[len(list)-bits] .. list[len(list)-bits]
-		// are encoded using "bits" bits, and get the values
-		// code, code + 1, ....  The code values are
-		// assigned in literal order (not frequency order).
-		chunk := list[len(list)-int(bits):]
-
-		sortByLiteral(chunk)
-		for _, node := range chunk {
-			h.codes[node.literal] = hcode{code: reverseBits(code, uint8(n)), len: uint16(n)}
-			code++
-		}
-		list = list[0 : len(list)-int(bits)]
-	}
-}
-
-// Update this Huffman Code object to be the minimum code for the specified frequency count.
-//
-// freq  An array of frequencies, in which frequency[i] gives the frequency of literal i.
-// maxBits  The maximum number of bits to use for any literal.
-func (h *huffmanEncoder) generate(freq []uint16, maxBits int32) {
-	list := h.freqcache[:len(freq)+1]
-	// Number of non-zero literals
-	count := 0
-	// Set list to be the set of all non-zero literals and their frequencies
-	for i, f := range freq {
-		if f != 0 {
-			list[count] = literalNode{uint16(i), f}
-			count++
-		} else {
-			list[count] = literalNode{}
-			h.codes[i].len = 0
-		}
-	}
-	list[len(freq)] = literalNode{}
-
-	list = list[:count]
-	if count <= 2 {
-		// Handle the small cases here, because they are awkward for the general case code. With
-		// two or fewer literals, everything has bit length 1.
-		for i, node := range list {
-			// "list" is in order of increasing literal value.
-			h.codes[node.literal].set(uint16(i), 1)
-		}
-		return
-	}
-	sortByFreq(list)
-
-	// Get the number of literals for each bit count
-	bitCount := h.bitCounts(list, maxBits)
-	// And do the assignment
-	h.assignEncodingAndSize(bitCount, list)
-}
-
-// atLeastOne clamps the result between 1 and 15.
-func atLeastOne(v float32) float32 {
-	if v < 1 {
-		return 1
-	}
-	if v > 15 {
-		return 15
-	}
-	return v
-}
-
-// Unassigned values are assigned '1' in the histogram.
-func fillHist(b []uint16) {
-	for i, v := range b {
-		if v == 0 {
-			b[i] = 1
-		}
-	}
-}
-
-func histogram(b []byte, h []uint16, fill bool) {
-	h = h[:256]
-	for _, t := range b {
-		h[t]++
-	}
-	if fill {
-		fillHist(h)
-	}
-}
diff --git a/vendor/github.com/klauspost/compress/flate/huffman_sortByFreq.go b/vendor/github.com/klauspost/compress/flate/huffman_sortByFreq.go
deleted file mode 100644
index 2077802990..0000000000
--- a/vendor/github.com/klauspost/compress/flate/huffman_sortByFreq.go
+++ /dev/null
@@ -1,178 +0,0 @@
-// Copyright 2009 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package flate
-
-// Sort sorts data.
-// It makes one call to data.Len to determine n, and O(n*log(n)) calls to
-// data.Less and data.Swap. The sort is not guaranteed to be stable.
-func sortByFreq(data []literalNode) {
-	n := len(data)
-	quickSortByFreq(data, 0, n, maxDepth(n))
-}
-
-func quickSortByFreq(data []literalNode, a, b, maxDepth int) {
-	for b-a > 12 { // Use ShellSort for slices <= 12 elements
-		if maxDepth == 0 {
-			heapSort(data, a, b)
-			return
-		}
-		maxDepth--
-		mlo, mhi := doPivotByFreq(data, a, b)
-		// Avoiding recursion on the larger subproblem guarantees
-		// a stack depth of at most lg(b-a).
-		if mlo-a < b-mhi {
-			quickSortByFreq(data, a, mlo, maxDepth)
-			a = mhi // i.e., quickSortByFreq(data, mhi, b)
-		} else {
-			quickSortByFreq(data, mhi, b, maxDepth)
-			b = mlo // i.e., quickSortByFreq(data, a, mlo)
-		}
-	}
-	if b-a > 1 {
-		// Do ShellSort pass with gap 6
-		// It could be written in this simplified form cause b-a <= 12
-		for i := a + 6; i < b; i++ {
-			if data[i].freq == data[i-6].freq && data[i].literal < data[i-6].literal || data[i].freq < data[i-6].freq {
-				data[i], data[i-6] = data[i-6], data[i]
-			}
-		}
-		insertionSortByFreq(data, a, b)
-	}
-}
-
-// siftDownByFreq implements the heap property on data[lo, hi).
-// first is an offset into the array where the root of the heap lies.
-func siftDownByFreq(data []literalNode, lo, hi, first int) {
-	root := lo
-	for {
-		child := 2*root + 1
-		if child >= hi {
-			break
-		}
-		if child+1 < hi && (data[first+child].freq == data[first+child+1].freq && data[first+child].literal < data[first+child+1].literal || data[first+child].freq < data[first+child+1].freq) {
-			child++
-		}
-		if data[first+root].freq == data[first+child].freq && data[first+root].literal > data[first+child].literal || data[first+root].freq > data[first+child].freq {
-			return
-		}
-		data[first+root], data[first+child] = data[first+child], data[first+root]
-		root = child
-	}
-}
-func doPivotByFreq(data []literalNode, lo, hi int) (midlo, midhi int) {
-	m := int(uint(lo+hi) >> 1) // Written like this to avoid integer overflow.
-	if hi-lo > 40 {
-		// Tukey's ``Ninther,'' median of three medians of three.
-		s := (hi - lo) / 8
-		medianOfThreeSortByFreq(data, lo, lo+s, lo+2*s)
-		medianOfThreeSortByFreq(data, m, m-s, m+s)
-		medianOfThreeSortByFreq(data, hi-1, hi-1-s, hi-1-2*s)
-	}
-	medianOfThreeSortByFreq(data, lo, m, hi-1)
-
-	// Invariants are:
-	//	data[lo] = pivot (set up by ChoosePivot)
-	//	data[lo < i < a] < pivot
-	//	data[a <= i < b] <= pivot
-	//	data[b <= i < c] unexamined
-	//	data[c <= i < hi-1] > pivot
-	//	data[hi-1] >= pivot
-	pivot := lo
-	a, c := lo+1, hi-1
-
-	for ; a < c && (data[a].freq == data[pivot].freq && data[a].literal < data[pivot].literal || data[a].freq < data[pivot].freq); a++ {
-	}
-	b := a
-	for {
-		for ; b < c && (data[pivot].freq == data[b].freq && data[pivot].literal > data[b].literal || data[pivot].freq > data[b].freq); b++ { // data[b] <= pivot
-		}
-		for ; b < c && (data[pivot].freq == data[c-1].freq && data[pivot].literal < data[c-1].literal || data[pivot].freq < data[c-1].freq); c-- { // data[c-1] > pivot
-		}
-		if b >= c {
-			break
-		}
-		// data[b] > pivot; data[c-1] <= pivot
-		data[b], data[c-1] = data[c-1], data[b]
-		b++
-		c--
-	}
-	// If hi-c<3 then there are duplicates (by property of median of nine).
-	// Let's be a bit more conservative, and set border to 5.
-	protect := hi-c < 5
-	if !protect && hi-c < (hi-lo)/4 {
-		// Lets test some points for equality to pivot
-		dups := 0
-		if data[pivot].freq == data[hi-1].freq && data[pivot].literal > data[hi-1].literal || data[pivot].freq > data[hi-1].freq { // data[hi-1] = pivot
-			data[c], data[hi-1] = data[hi-1], data[c]
-			c++
-			dups++
-		}
-		if data[b-1].freq == data[pivot].freq && data[b-1].literal > data[pivot].literal || data[b-1].freq > data[pivot].freq { // data[b-1] = pivot
-			b--
-			dups++
-		}
-		// m-lo = (hi-lo)/2 > 6
-		// b-lo > (hi-lo)*3/4-1 > 8
-		// ==> m < b ==> data[m] <= pivot
-		if data[m].freq == data[pivot].freq && data[m].literal > data[pivot].literal || data[m].freq > data[pivot].freq { // data[m] = pivot
-			data[m], data[b-1] = data[b-1], data[m]
-			b--
-			dups++
-		}
-		// if at least 2 points are equal to pivot, assume skewed distribution
-		protect = dups > 1
-	}
-	if protect {
-		// Protect against a lot of duplicates
-		// Add invariant:
-		//	data[a <= i < b] unexamined
-		//	data[b <= i < c] = pivot
-		for {
-			for ; a < b && (data[b-1].freq == data[pivot].freq && data[b-1].literal > data[pivot].literal || data[b-1].freq > data[pivot].freq); b-- { // data[b] == pivot
-			}
-			for ; a < b && (data[a].freq == data[pivot].freq && data[a].literal < data[pivot].literal || data[a].freq < data[pivot].freq); a++ { // data[a] < pivot
-			}
-			if a >= b {
-				break
-			}
-			// data[a] == pivot; data[b-1] < pivot
-			data[a], data[b-1] = data[b-1], data[a]
-			a++
-			b--
-		}
-	}
-	// Swap pivot into middle
-	data[pivot], data[b-1] = data[b-1], data[pivot]
-	return b - 1, c
-}
-
-// Insertion sort
-func insertionSortByFreq(data []literalNode, a, b int) {
-	for i := a + 1; i < b; i++ {
-		for j := i; j > a && (data[j].freq == data[j-1].freq && data[j].literal < data[j-1].literal || data[j].freq < data[j-1].freq); j-- {
-			data[j], data[j-1] = data[j-1], data[j]
-		}
-	}
-}
-
-// quickSortByFreq, loosely following Bentley and McIlroy,
-// ``Engineering a Sort Function,'' SP&E November 1993.
-
-// medianOfThreeSortByFreq moves the median of the three values data[m0], data[m1], data[m2] into data[m1].
-func medianOfThreeSortByFreq(data []literalNode, m1, m0, m2 int) {
-	// sort 3 elements
-	if data[m1].freq == data[m0].freq && data[m1].literal < data[m0].literal || data[m1].freq < data[m0].freq {
-		data[m1], data[m0] = data[m0], data[m1]
-	}
-	// data[m0] <= data[m1]
-	if data[m2].freq == data[m1].freq && data[m2].literal < data[m1].literal || data[m2].freq < data[m1].freq {
-		data[m2], data[m1] = data[m1], data[m2]
-		// data[m0] <= data[m2] && data[m1] < data[m2]
-		if data[m1].freq == data[m0].freq && data[m1].literal < data[m0].literal || data[m1].freq < data[m0].freq {
-			data[m1], data[m0] = data[m0], data[m1]
-		}
-	}
-	// now data[m0] <= data[m1] <= data[m2]
-}
diff --git a/vendor/github.com/klauspost/compress/flate/huffman_sortByLiteral.go b/vendor/github.com/klauspost/compress/flate/huffman_sortByLiteral.go
deleted file mode 100644
index 93f1aea109..0000000000
--- a/vendor/github.com/klauspost/compress/flate/huffman_sortByLiteral.go
+++ /dev/null
@@ -1,201 +0,0 @@
-// Copyright 2009 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package flate
-
-// Sort sorts data.
-// It makes one call to data.Len to determine n, and O(n*log(n)) calls to
-// data.Less and data.Swap. The sort is not guaranteed to be stable.
-func sortByLiteral(data []literalNode) {
-	n := len(data)
-	quickSort(data, 0, n, maxDepth(n))
-}
-
-func quickSort(data []literalNode, a, b, maxDepth int) {
-	for b-a > 12 { // Use ShellSort for slices <= 12 elements
-		if maxDepth == 0 {
-			heapSort(data, a, b)
-			return
-		}
-		maxDepth--
-		mlo, mhi := doPivot(data, a, b)
-		// Avoiding recursion on the larger subproblem guarantees
-		// a stack depth of at most lg(b-a).
-		if mlo-a < b-mhi {
-			quickSort(data, a, mlo, maxDepth)
-			a = mhi // i.e., quickSort(data, mhi, b)
-		} else {
-			quickSort(data, mhi, b, maxDepth)
-			b = mlo // i.e., quickSort(data, a, mlo)
-		}
-	}
-	if b-a > 1 {
-		// Do ShellSort pass with gap 6
-		// It could be written in this simplified form cause b-a <= 12
-		for i := a + 6; i < b; i++ {
-			if data[i].literal < data[i-6].literal {
-				data[i], data[i-6] = data[i-6], data[i]
-			}
-		}
-		insertionSort(data, a, b)
-	}
-}
-func heapSort(data []literalNode, a, b int) {
-	first := a
-	lo := 0
-	hi := b - a
-
-	// Build heap with greatest element at top.
-	for i := (hi - 1) / 2; i >= 0; i-- {
-		siftDown(data, i, hi, first)
-	}
-
-	// Pop elements, largest first, into end of data.
-	for i := hi - 1; i >= 0; i-- {
-		data[first], data[first+i] = data[first+i], data[first]
-		siftDown(data, lo, i, first)
-	}
-}
-
-// siftDown implements the heap property on data[lo, hi).
-// first is an offset into the array where the root of the heap lies.
-func siftDown(data []literalNode, lo, hi, first int) {
-	root := lo
-	for {
-		child := 2*root + 1
-		if child >= hi {
-			break
-		}
-		if child+1 < hi && data[first+child].literal < data[first+child+1].literal {
-			child++
-		}
-		if data[first+root].literal > data[first+child].literal {
-			return
-		}
-		data[first+root], data[first+child] = data[first+child], data[first+root]
-		root = child
-	}
-}
-func doPivot(data []literalNode, lo, hi int) (midlo, midhi int) {
-	m := int(uint(lo+hi) >> 1) // Written like this to avoid integer overflow.
-	if hi-lo > 40 {
-		// Tukey's ``Ninther,'' median of three medians of three.
-		s := (hi - lo) / 8
-		medianOfThree(data, lo, lo+s, lo+2*s)
-		medianOfThree(data, m, m-s, m+s)
-		medianOfThree(data, hi-1, hi-1-s, hi-1-2*s)
-	}
-	medianOfThree(data, lo, m, hi-1)
-
-	// Invariants are:
-	//	data[lo] = pivot (set up by ChoosePivot)
-	//	data[lo < i < a] < pivot
-	//	data[a <= i < b] <= pivot
-	//	data[b <= i < c] unexamined
-	//	data[c <= i < hi-1] > pivot
-	//	data[hi-1] >= pivot
-	pivot := lo
-	a, c := lo+1, hi-1
-
-	for ; a < c && data[a].literal < data[pivot].literal; a++ {
-	}
-	b := a
-	for {
-		for ; b < c && data[pivot].literal > data[b].literal; b++ { // data[b] <= pivot
-		}
-		for ; b < c && data[pivot].literal < data[c-1].literal; c-- { // data[c-1] > pivot
-		}
-		if b >= c {
-			break
-		}
-		// data[b] > pivot; data[c-1] <= pivot
-		data[b], data[c-1] = data[c-1], data[b]
-		b++
-		c--
-	}
-	// If hi-c<3 then there are duplicates (by property of median of nine).
-	// Let's be a bit more conservative, and set border to 5.
-	protect := hi-c < 5
-	if !protect && hi-c < (hi-lo)/4 {
-		// Lets test some points for equality to pivot
-		dups := 0
-		if data[pivot].literal > data[hi-1].literal { // data[hi-1] = pivot
-			data[c], data[hi-1] = data[hi-1], data[c]
-			c++
-			dups++
-		}
-		if data[b-1].literal > data[pivot].literal { // data[b-1] = pivot
-			b--
-			dups++
-		}
-		// m-lo = (hi-lo)/2 > 6
-		// b-lo > (hi-lo)*3/4-1 > 8
-		// ==> m < b ==> data[m] <= pivot
-		if data[m].literal > data[pivot].literal { // data[m] = pivot
-			data[m], data[b-1] = data[b-1], data[m]
-			b--
-			dups++
-		}
-		// if at least 2 points are equal to pivot, assume skewed distribution
-		protect = dups > 1
-	}
-	if protect {
-		// Protect against a lot of duplicates
-		// Add invariant:
-		//	data[a <= i < b] unexamined
-		//	data[b <= i < c] = pivot
-		for {
-			for ; a < b && data[b-1].literal > data[pivot].literal; b-- { // data[b] == pivot
-			}
-			for ; a < b && data[a].literal < data[pivot].literal; a++ { // data[a] < pivot
-			}
-			if a >= b {
-				break
-			}
-			// data[a] == pivot; data[b-1] < pivot
-			data[a], data[b-1] = data[b-1], data[a]
-			a++
-			b--
-		}
-	}
-	// Swap pivot into middle
-	data[pivot], data[b-1] = data[b-1], data[pivot]
-	return b - 1, c
-}
-
-// Insertion sort
-func insertionSort(data []literalNode, a, b int) {
-	for i := a + 1; i < b; i++ {
-		for j := i; j > a && data[j].literal < data[j-1].literal; j-- {
-			data[j], data[j-1] = data[j-1], data[j]
-		}
-	}
-}
-
-// maxDepth returns a threshold at which quicksort should switch
-// to heapsort. It returns 2*ceil(lg(n+1)).
-func maxDepth(n int) int {
-	var depth int
-	for i := n; i > 0; i >>= 1 {
-		depth++
-	}
-	return depth * 2
-}
-
-// medianOfThree moves the median of the three values data[m0], data[m1], data[m2] into data[m1].
-func medianOfThree(data []literalNode, m1, m0, m2 int) {
-	// sort 3 elements
-	if data[m1].literal < data[m0].literal {
-		data[m1], data[m0] = data[m0], data[m1]
-	}
-	// data[m0] <= data[m1]
-	if data[m2].literal < data[m1].literal {
-		data[m2], data[m1] = data[m1], data[m2]
-		// data[m0] <= data[m2] && data[m1] < data[m2]
-		if data[m1].literal < data[m0].literal {
-			data[m1], data[m0] = data[m0], data[m1]
-		}
-	}
-	// now data[m0] <= data[m1] <= data[m2]
-}
diff --git a/vendor/github.com/klauspost/compress/flate/inflate.go b/vendor/github.com/klauspost/compress/flate/inflate.go
deleted file mode 100644
index 16bc51408e..0000000000
--- a/vendor/github.com/klauspost/compress/flate/inflate.go
+++ /dev/null
@@ -1,1010 +0,0 @@
-// Copyright 2009 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// Package flate implements the DEFLATE compressed data format, described in
-// RFC 1951.  The gzip and zlib packages implement access to DEFLATE-based file
-// formats.
-package flate
-
-import (
-	"bufio"
-	"fmt"
-	"io"
-	"math/bits"
-	"strconv"
-	"sync"
-)
-
-const (
-	maxCodeLen     = 16 // max length of Huffman code
-	maxCodeLenMask = 15 // mask for max length of Huffman code
-	// The next three numbers come from the RFC section 3.2.7, with the
-	// additional proviso in section 3.2.5 which implies that distance codes
-	// 30 and 31 should never occur in compressed data.
-	maxNumLit  = 286
-	maxNumDist = 30
-	numCodes   = 19 // number of codes in Huffman meta-code
-
-	debugDecode = false
-)
-
-// Value of length - 3 and extra bits.
-type lengthExtra struct {
-	length, extra uint8
-}
-
-var decCodeToLen = [32]lengthExtra{{length: 0x0, extra: 0x0}, {length: 0x1, extra: 0x0}, {length: 0x2, extra: 0x0}, {length: 0x3, extra: 0x0}, {length: 0x4, extra: 0x0}, {length: 0x5, extra: 0x0}, {length: 0x6, extra: 0x0}, {length: 0x7, extra: 0x0}, {length: 0x8, extra: 0x1}, {length: 0xa, extra: 0x1}, {length: 0xc, extra: 0x1}, {length: 0xe, extra: 0x1}, {length: 0x10, extra: 0x2}, {length: 0x14, extra: 0x2}, {length: 0x18, extra: 0x2}, {length: 0x1c, extra: 0x2}, {length: 0x20, extra: 0x3}, {length: 0x28, extra: 0x3}, {length: 0x30, extra: 0x3}, {length: 0x38, extra: 0x3}, {length: 0x40, extra: 0x4}, {length: 0x50, extra: 0x4}, {length: 0x60, extra: 0x4}, {length: 0x70, extra: 0x4}, {length: 0x80, extra: 0x5}, {length: 0xa0, extra: 0x5}, {length: 0xc0, extra: 0x5}, {length: 0xe0, extra: 0x5}, {length: 0xff, extra: 0x0}, {length: 0x0, extra: 0x0}, {length: 0x0, extra: 0x0}, {length: 0x0, extra: 0x0}}
-
-// Initialize the fixedHuffmanDecoder only once upon first use.
-var fixedOnce sync.Once
-var fixedHuffmanDecoder huffmanDecoder
-
-// A CorruptInputError reports the presence of corrupt input at a given offset.
-type CorruptInputError int64
-
-func (e CorruptInputError) Error() string {
-	return "flate: corrupt input before offset " + strconv.FormatInt(int64(e), 10)
-}
-
-// An InternalError reports an error in the flate code itself.
-type InternalError string
-
-func (e InternalError) Error() string { return "flate: internal error: " + string(e) }
-
-// A ReadError reports an error encountered while reading input.
-//
-// Deprecated: No longer returned.
-type ReadError struct {
-	Offset int64 // byte offset where error occurred
-	Err    error // error returned by underlying Read
-}
-
-func (e *ReadError) Error() string {
-	return "flate: read error at offset " + strconv.FormatInt(e.Offset, 10) + ": " + e.Err.Error()
-}
-
-// A WriteError reports an error encountered while writing output.
-//
-// Deprecated: No longer returned.
-type WriteError struct {
-	Offset int64 // byte offset where error occurred
-	Err    error // error returned by underlying Write
-}
-
-func (e *WriteError) Error() string {
-	return "flate: write error at offset " + strconv.FormatInt(e.Offset, 10) + ": " + e.Err.Error()
-}
-
-// Resetter resets a ReadCloser returned by NewReader or NewReaderDict to
-// to switch to a new underlying Reader. This permits reusing a ReadCloser
-// instead of allocating a new one.
-type Resetter interface {
-	// Reset discards any buffered data and resets the Resetter as if it was
-	// newly initialized with the given reader.
-	Reset(r io.Reader, dict []byte) error
-}
-
-// The data structure for decoding Huffman tables is based on that of
-// zlib. There is a lookup table of a fixed bit width (huffmanChunkBits),
-// For codes smaller than the table width, there are multiple entries
-// (each combination of trailing bits has the same value). For codes
-// larger than the table width, the table contains a link to an overflow
-// table. The width of each entry in the link table is the maximum code
-// size minus the chunk width.
-//
-// Note that you can do a lookup in the table even without all bits
-// filled. Since the extra bits are zero, and the DEFLATE Huffman codes
-// have the property that shorter codes come before longer ones, the
-// bit length estimate in the result is a lower bound on the actual
-// number of bits.
-//
-// See the following:
-//	http://www.gzip.org/algorithm.txt
-
-// chunk & 15 is number of bits
-// chunk >> 4 is value, including table link
-
-const (
-	huffmanChunkBits  = 9
-	huffmanNumChunks  = 1 << huffmanChunkBits
-	huffmanCountMask  = 15
-	huffmanValueShift = 4
-)
-
-type huffmanDecoder struct {
-	maxRead  int                       // the maximum number of bits we can read and not overread
-	chunks   *[huffmanNumChunks]uint16 // chunks as described above
-	links    [][]uint16                // overflow links
-	linkMask uint32                    // mask the width of the link table
-}
-
-// Initialize Huffman decoding tables from array of code lengths.
-// Following this function, h is guaranteed to be initialized into a complete
-// tree (i.e., neither over-subscribed nor under-subscribed). The exception is a
-// degenerate case where the tree has only a single symbol with length 1. Empty
-// trees are permitted.
-func (h *huffmanDecoder) init(lengths []int) bool {
-	// Sanity enables additional runtime tests during Huffman
-	// table construction. It's intended to be used during
-	// development to supplement the currently ad-hoc unit tests.
-	const sanity = false
-
-	if h.chunks == nil {
-		h.chunks = &[huffmanNumChunks]uint16{}
-	}
-	if h.maxRead != 0 {
-		*h = huffmanDecoder{chunks: h.chunks, links: h.links}
-	}
-
-	// Count number of codes of each length,
-	// compute maxRead and max length.
-	var count [maxCodeLen]int
-	var min, max int
-	for _, n := range lengths {
-		if n == 0 {
-			continue
-		}
-		if min == 0 || n < min {
-			min = n
-		}
-		if n > max {
-			max = n
-		}
-		count[n&maxCodeLenMask]++
-	}
-
-	// Empty tree. The decompressor.huffSym function will fail later if the tree
-	// is used. Technically, an empty tree is only valid for the HDIST tree and
-	// not the HCLEN and HLIT tree. However, a stream with an empty HCLEN tree
-	// is guaranteed to fail since it will attempt to use the tree to decode the
-	// codes for the HLIT and HDIST trees. Similarly, an empty HLIT tree is
-	// guaranteed to fail later since the compressed data section must be
-	// composed of at least one symbol (the end-of-block marker).
-	if max == 0 {
-		return true
-	}
-
-	code := 0
-	var nextcode [maxCodeLen]int
-	for i := min; i <= max; i++ {
-		code <<= 1
-		nextcode[i&maxCodeLenMask] = code
-		code += count[i&maxCodeLenMask]
-	}
-
-	// Check that the coding is complete (i.e., that we've
-	// assigned all 2-to-the-max possible bit sequences).
-	// Exception: To be compatible with zlib, we also need to
-	// accept degenerate single-code codings. See also
-	// TestDegenerateHuffmanCoding.
-	if code != 1<<uint(max) && !(code == 1 && max == 1) {
-		if debugDecode {
-			fmt.Println("coding failed, code, max:", code, max, code == 1<<uint(max), code == 1 && max == 1, "(one should be true)")
-		}
-		return false
-	}
-
-	h.maxRead = min
-	chunks := h.chunks[:]
-	for i := range chunks {
-		chunks[i] = 0
-	}
-
-	if max > huffmanChunkBits {
-		numLinks := 1 << (uint(max) - huffmanChunkBits)
-		h.linkMask = uint32(numLinks - 1)
-
-		// create link tables
-		link := nextcode[huffmanChunkBits+1] >> 1
-		if cap(h.links) < huffmanNumChunks-link {
-			h.links = make([][]uint16, huffmanNumChunks-link)
-		} else {
-			h.links = h.links[:huffmanNumChunks-link]
-		}
-		for j := uint(link); j < huffmanNumChunks; j++ {
-			reverse := int(bits.Reverse16(uint16(j)))
-			reverse >>= uint(16 - huffmanChunkBits)
-			off := j - uint(link)
-			if sanity && h.chunks[reverse] != 0 {
-				panic("impossible: overwriting existing chunk")
-			}
-			h.chunks[reverse] = uint16(off<<huffmanValueShift | (huffmanChunkBits + 1))
-			if cap(h.links[off]) < numLinks {
-				h.links[off] = make([]uint16, numLinks)
-			} else {
-				links := h.links[off][:0]
-				h.links[off] = links[:numLinks]
-			}
-		}
-	} else {
-		h.links = h.links[:0]
-	}
-
-	for i, n := range lengths {
-		if n == 0 {
-			continue
-		}
-		code := nextcode[n]
-		nextcode[n]++
-		chunk := uint16(i<<huffmanValueShift | n)
-		reverse := int(bits.Reverse16(uint16(code)))
-		reverse >>= uint(16 - n)
-		if n <= huffmanChunkBits {
-			for off := reverse; off < len(h.chunks); off += 1 << uint(n) {
-				// We should never need to overwrite
-				// an existing chunk. Also, 0 is
-				// never a valid chunk, because the
-				// lower 4 "count" bits should be
-				// between 1 and 15.
-				if sanity && h.chunks[off] != 0 {
-					panic("impossible: overwriting existing chunk")
-				}
-				h.chunks[off] = chunk
-			}
-		} else {
-			j := reverse & (huffmanNumChunks - 1)
-			if sanity && h.chunks[j]&huffmanCountMask != huffmanChunkBits+1 {
-				// Longer codes should have been
-				// associated with a link table above.
-				panic("impossible: not an indirect chunk")
-			}
-			value := h.chunks[j] >> huffmanValueShift
-			linktab := h.links[value]
-			reverse >>= huffmanChunkBits
-			for off := reverse; off < len(linktab); off += 1 << uint(n-huffmanChunkBits) {
-				if sanity && linktab[off] != 0 {
-					panic("impossible: overwriting existing chunk")
-				}
-				linktab[off] = chunk
-			}
-		}
-	}
-
-	if sanity {
-		// Above we've sanity checked that we never overwrote
-		// an existing entry. Here we additionally check that
-		// we filled the tables completely.
-		for i, chunk := range h.chunks {
-			if chunk == 0 {
-				// As an exception, in the degenerate
-				// single-code case, we allow odd
-				// chunks to be missing.
-				if code == 1 && i%2 == 1 {
-					continue
-				}
-				panic("impossible: missing chunk")
-			}
-		}
-		for _, linktab := range h.links {
-			for _, chunk := range linktab {
-				if chunk == 0 {
-					panic("impossible: missing chunk")
-				}
-			}
-		}
-	}
-
-	return true
-}
-
-// The actual read interface needed by NewReader.
-// If the passed in io.Reader does not also have ReadByte,
-// the NewReader will introduce its own buffering.
-type Reader interface {
-	io.Reader
-	io.ByteReader
-}
-
-// Decompress state.
-type decompressor struct {
-	// Input source.
-	r       Reader
-	roffset int64
-
-	// Huffman decoders for literal/length, distance.
-	h1, h2 huffmanDecoder
-
-	// Length arrays used to define Huffman codes.
-	bits     *[maxNumLit + maxNumDist]int
-	codebits *[numCodes]int
-
-	// Output history, buffer.
-	dict dictDecoder
-
-	// Next step in the decompression,
-	// and decompression state.
-	step      func(*decompressor)
-	stepState int
-	err       error
-	toRead    []byte
-	hl, hd    *huffmanDecoder
-	copyLen   int
-	copyDist  int
-
-	// Temporary buffer (avoids repeated allocation).
-	buf [4]byte
-
-	// Input bits, in top of b.
-	b uint32
-
-	nb    uint
-	final bool
-}
-
-func (f *decompressor) nextBlock() {
-	for f.nb < 1+2 {
-		if f.err = f.moreBits(); f.err != nil {
-			return
-		}
-	}
-	f.final = f.b&1 == 1
-	f.b >>= 1
-	typ := f.b & 3
-	f.b >>= 2
-	f.nb -= 1 + 2
-	switch typ {
-	case 0:
-		f.dataBlock()
-	case 1:
-		// compressed, fixed Huffman tables
-		f.hl = &fixedHuffmanDecoder
-		f.hd = nil
-		f.huffmanBlockDecoder()()
-	case 2:
-		// compressed, dynamic Huffman tables
-		if f.err = f.readHuffman(); f.err != nil {
-			break
-		}
-		f.hl = &f.h1
-		f.hd = &f.h2
-		f.huffmanBlockDecoder()()
-	default:
-		// 3 is reserved.
-		if debugDecode {
-			fmt.Println("reserved data block encountered")
-		}
-		f.err = CorruptInputError(f.roffset)
-	}
-}
-
-func (f *decompressor) Read(b []byte) (int, error) {
-	for {
-		if len(f.toRead) > 0 {
-			n := copy(b, f.toRead)
-			f.toRead = f.toRead[n:]
-			if len(f.toRead) == 0 {
-				return n, f.err
-			}
-			return n, nil
-		}
-		if f.err != nil {
-			return 0, f.err
-		}
-		f.step(f)
-		if f.err != nil && len(f.toRead) == 0 {
-			f.toRead = f.dict.readFlush() // Flush what's left in case of error
-		}
-	}
-}
-
-// Support the io.WriteTo interface for io.Copy and friends.
-func (f *decompressor) WriteTo(w io.Writer) (int64, error) {
-	total := int64(0)
-	flushed := false
-	for {
-		if len(f.toRead) > 0 {
-			n, err := w.Write(f.toRead)
-			total += int64(n)
-			if err != nil {
-				f.err = err
-				return total, err
-			}
-			if n != len(f.toRead) {
-				return total, io.ErrShortWrite
-			}
-			f.toRead = f.toRead[:0]
-		}
-		if f.err != nil && flushed {
-			if f.err == io.EOF {
-				return total, nil
-			}
-			return total, f.err
-		}
-		if f.err == nil {
-			f.step(f)
-		}
-		if len(f.toRead) == 0 && f.err != nil && !flushed {
-			f.toRead = f.dict.readFlush() // Flush what's left in case of error
-			flushed = true
-		}
-	}
-}
-
-func (f *decompressor) Close() error {
-	if f.err == io.EOF {
-		return nil
-	}
-	return f.err
-}
-
-// RFC 1951 section 3.2.7.
-// Compression with dynamic Huffman codes
-
-var codeOrder = [...]int{16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15}
-
-func (f *decompressor) readHuffman() error {
-	// HLIT[5], HDIST[5], HCLEN[4].
-	for f.nb < 5+5+4 {
-		if err := f.moreBits(); err != nil {
-			return err
-		}
-	}
-	nlit := int(f.b&0x1F) + 257
-	if nlit > maxNumLit {
-		if debugDecode {
-			fmt.Println("nlit > maxNumLit", nlit)
-		}
-		return CorruptInputError(f.roffset)
-	}
-	f.b >>= 5
-	ndist := int(f.b&0x1F) + 1
-	if ndist > maxNumDist {
-		if debugDecode {
-			fmt.Println("ndist > maxNumDist", ndist)
-		}
-		return CorruptInputError(f.roffset)
-	}
-	f.b >>= 5
-	nclen := int(f.b&0xF) + 4
-	// numCodes is 19, so nclen is always valid.
-	f.b >>= 4
-	f.nb -= 5 + 5 + 4
-
-	// (HCLEN+4)*3 bits: code lengths in the magic codeOrder order.
-	for i := 0; i < nclen; i++ {
-		for f.nb < 3 {
-			if err := f.moreBits(); err != nil {
-				return err
-			}
-		}
-		f.codebits[codeOrder[i]] = int(f.b & 0x7)
-		f.b >>= 3
-		f.nb -= 3
-	}
-	for i := nclen; i < len(codeOrder); i++ {
-		f.codebits[codeOrder[i]] = 0
-	}
-	if !f.h1.init(f.codebits[0:]) {
-		if debugDecode {
-			fmt.Println("init codebits failed")
-		}
-		return CorruptInputError(f.roffset)
-	}
-
-	// HLIT + 257 code lengths, HDIST + 1 code lengths,
-	// using the code length Huffman code.
-	for i, n := 0, nlit+ndist; i < n; {
-		x, err := f.huffSym(&f.h1)
-		if err != nil {
-			return err
-		}
-		if x < 16 {
-			// Actual length.
-			f.bits[i] = x
-			i++
-			continue
-		}
-		// Repeat previous length or zero.
-		var rep int
-		var nb uint
-		var b int
-		switch x {
-		default:
-			return InternalError("unexpected length code")
-		case 16:
-			rep = 3
-			nb = 2
-			if i == 0 {
-				if debugDecode {
-					fmt.Println("i==0")
-				}
-				return CorruptInputError(f.roffset)
-			}
-			b = f.bits[i-1]
-		case 17:
-			rep = 3
-			nb = 3
-			b = 0
-		case 18:
-			rep = 11
-			nb = 7
-			b = 0
-		}
-		for f.nb < nb {
-			if err := f.moreBits(); err != nil {
-				if debugDecode {
-					fmt.Println("morebits:", err)
-				}
-				return err
-			}
-		}
-		rep += int(f.b & uint32(1<<(nb&regSizeMaskUint32)-1))
-		f.b >>= nb & regSizeMaskUint32
-		f.nb -= nb
-		if i+rep > n {
-			if debugDecode {
-				fmt.Println("i+rep > n", i, rep, n)
-			}
-			return CorruptInputError(f.roffset)
-		}
-		for j := 0; j < rep; j++ {
-			f.bits[i] = b
-			i++
-		}
-	}
-
-	if !f.h1.init(f.bits[0:nlit]) || !f.h2.init(f.bits[nlit:nlit+ndist]) {
-		if debugDecode {
-			fmt.Println("init2 failed")
-		}
-		return CorruptInputError(f.roffset)
-	}
-
-	// As an optimization, we can initialize the maxRead bits to read at a time
-	// for the HLIT tree to the length of the EOB marker since we know that
-	// every block must terminate with one. This preserves the property that
-	// we never read any extra bytes after the end of the DEFLATE stream.
-	if f.h1.maxRead < f.bits[endBlockMarker] {
-		f.h1.maxRead = f.bits[endBlockMarker]
-	}
-	if !f.final {
-		// If not the final block, the smallest block possible is
-		// a predefined table, BTYPE=01, with a single EOB marker.
-		// This will take up 3 + 7 bits.
-		f.h1.maxRead += 10
-	}
-
-	return nil
-}
-
-// Decode a single Huffman block from f.
-// hl and hd are the Huffman states for the lit/length values
-// and the distance values, respectively. If hd == nil, using the
-// fixed distance encoding associated with fixed Huffman blocks.
-func (f *decompressor) huffmanBlockGeneric() {
-	const (
-		stateInit = iota // Zero value must be stateInit
-		stateDict
-	)
-
-	switch f.stepState {
-	case stateInit:
-		goto readLiteral
-	case stateDict:
-		goto copyHistory
-	}
-
-readLiteral:
-	// Read literal and/or (length, distance) according to RFC section 3.2.3.
-	{
-		var v int
-		{
-			// Inlined v, err := f.huffSym(f.hl)
-			// Since a huffmanDecoder can be empty or be composed of a degenerate tree
-			// with single element, huffSym must error on these two edge cases. In both
-			// cases, the chunks slice will be 0 for the invalid sequence, leading it
-			// satisfy the n == 0 check below.
-			n := uint(f.hl.maxRead)
-			// Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers,
-			// but is smart enough to keep local variables in registers, so use nb and b,
-			// inline call to moreBits and reassign b,nb back to f on return.
-			nb, b := f.nb, f.b
-			for {
-				for nb < n {
-					c, err := f.r.ReadByte()
-					if err != nil {
-						f.b = b
-						f.nb = nb
-						f.err = noEOF(err)
-						return
-					}
-					f.roffset++
-					b |= uint32(c) << (nb & regSizeMaskUint32)
-					nb += 8
-				}
-				chunk := f.hl.chunks[b&(huffmanNumChunks-1)]
-				n = uint(chunk & huffmanCountMask)
-				if n > huffmanChunkBits {
-					chunk = f.hl.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hl.linkMask]
-					n = uint(chunk & huffmanCountMask)
-				}
-				if n <= nb {
-					if n == 0 {
-						f.b = b
-						f.nb = nb
-						if debugDecode {
-							fmt.Println("huffsym: n==0")
-						}
-						f.err = CorruptInputError(f.roffset)
-						return
-					}
-					f.b = b >> (n & regSizeMaskUint32)
-					f.nb = nb - n
-					v = int(chunk >> huffmanValueShift)
-					break
-				}
-			}
-		}
-
-		var n uint // number of bits extra
-		var length int
-		var err error
-		switch {
-		case v < 256:
-			f.dict.writeByte(byte(v))
-			if f.dict.availWrite() == 0 {
-				f.toRead = f.dict.readFlush()
-				f.step = (*decompressor).huffmanBlockGeneric
-				f.stepState = stateInit
-				return
-			}
-			goto readLiteral
-		case v == 256:
-			f.finishBlock()
-			return
-		// otherwise, reference to older data
-		case v < 265:
-			length = v - (257 - 3)
-			n = 0
-		case v < 269:
-			length = v*2 - (265*2 - 11)
-			n = 1
-		case v < 273:
-			length = v*4 - (269*4 - 19)
-			n = 2
-		case v < 277:
-			length = v*8 - (273*8 - 35)
-			n = 3
-		case v < 281:
-			length = v*16 - (277*16 - 67)
-			n = 4
-		case v < 285:
-			length = v*32 - (281*32 - 131)
-			n = 5
-		case v < maxNumLit:
-			length = 258
-			n = 0
-		default:
-			if debugDecode {
-				fmt.Println(v, ">= maxNumLit")
-			}
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-		if n > 0 {
-			for f.nb < n {
-				if err = f.moreBits(); err != nil {
-					if debugDecode {
-						fmt.Println("morebits n>0:", err)
-					}
-					f.err = err
-					return
-				}
-			}
-			length += int(f.b & uint32(1<<(n&regSizeMaskUint32)-1))
-			f.b >>= n & regSizeMaskUint32
-			f.nb -= n
-		}
-
-		var dist uint32
-		if f.hd == nil {
-			for f.nb < 5 {
-				if err = f.moreBits(); err != nil {
-					if debugDecode {
-						fmt.Println("morebits f.nb<5:", err)
-					}
-					f.err = err
-					return
-				}
-			}
-			dist = uint32(bits.Reverse8(uint8(f.b & 0x1F << 3)))
-			f.b >>= 5
-			f.nb -= 5
-		} else {
-			sym, err := f.huffSym(f.hd)
-			if err != nil {
-				if debugDecode {
-					fmt.Println("huffsym:", err)
-				}
-				f.err = err
-				return
-			}
-			dist = uint32(sym)
-		}
-
-		switch {
-		case dist < 4:
-			dist++
-		case dist < maxNumDist:
-			nb := uint(dist-2) >> 1
-			// have 1 bit in bottom of dist, need nb more.
-			extra := (dist & 1) << (nb & regSizeMaskUint32)
-			for f.nb < nb {
-				if err = f.moreBits(); err != nil {
-					if debugDecode {
-						fmt.Println("morebits f.nb<nb:", err)
-					}
-					f.err = err
-					return
-				}
-			}
-			extra |= f.b & uint32(1<<(nb&regSizeMaskUint32)-1)
-			f.b >>= nb & regSizeMaskUint32
-			f.nb -= nb
-			dist = 1<<((nb+1)&regSizeMaskUint32) + 1 + extra
-		default:
-			if debugDecode {
-				fmt.Println("dist too big:", dist, maxNumDist)
-			}
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-
-		// No check on length; encoding can be prescient.
-		if dist > uint32(f.dict.histSize()) {
-			if debugDecode {
-				fmt.Println("dist > f.dict.histSize():", dist, f.dict.histSize())
-			}
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-
-		f.copyLen, f.copyDist = length, int(dist)
-		goto copyHistory
-	}
-
-copyHistory:
-	// Perform a backwards copy according to RFC section 3.2.3.
-	{
-		cnt := f.dict.tryWriteCopy(f.copyDist, f.copyLen)
-		if cnt == 0 {
-			cnt = f.dict.writeCopy(f.copyDist, f.copyLen)
-		}
-		f.copyLen -= cnt
-
-		if f.dict.availWrite() == 0 || f.copyLen > 0 {
-			f.toRead = f.dict.readFlush()
-			f.step = (*decompressor).huffmanBlockGeneric // We need to continue this work
-			f.stepState = stateDict
-			return
-		}
-		goto readLiteral
-	}
-}
-
-// Copy a single uncompressed data block from input to output.
-func (f *decompressor) dataBlock() {
-	// Uncompressed.
-	// Discard current half-byte.
-	left := (f.nb) & 7
-	f.nb -= left
-	f.b >>= left
-
-	offBytes := f.nb >> 3
-	// Unfilled values will be overwritten.
-	f.buf[0] = uint8(f.b)
-	f.buf[1] = uint8(f.b >> 8)
-	f.buf[2] = uint8(f.b >> 16)
-	f.buf[3] = uint8(f.b >> 24)
-
-	f.roffset += int64(offBytes)
-	f.nb, f.b = 0, 0
-
-	// Length then ones-complement of length.
-	nr, err := io.ReadFull(f.r, f.buf[offBytes:4])
-	f.roffset += int64(nr)
-	if err != nil {
-		f.err = noEOF(err)
-		return
-	}
-	n := uint16(f.buf[0]) | uint16(f.buf[1])<<8
-	nn := uint16(f.buf[2]) | uint16(f.buf[3])<<8
-	if nn != ^n {
-		if debugDecode {
-			ncomp := ^n
-			fmt.Println("uint16(nn) != uint16(^n)", nn, ncomp)
-		}
-		f.err = CorruptInputError(f.roffset)
-		return
-	}
-
-	if n == 0 {
-		f.toRead = f.dict.readFlush()
-		f.finishBlock()
-		return
-	}
-
-	f.copyLen = int(n)
-	f.copyData()
-}
-
-// copyData copies f.copyLen bytes from the underlying reader into f.hist.
-// It pauses for reads when f.hist is full.
-func (f *decompressor) copyData() {
-	buf := f.dict.writeSlice()
-	if len(buf) > f.copyLen {
-		buf = buf[:f.copyLen]
-	}
-
-	cnt, err := io.ReadFull(f.r, buf)
-	f.roffset += int64(cnt)
-	f.copyLen -= cnt
-	f.dict.writeMark(cnt)
-	if err != nil {
-		f.err = noEOF(err)
-		return
-	}
-
-	if f.dict.availWrite() == 0 || f.copyLen > 0 {
-		f.toRead = f.dict.readFlush()
-		f.step = (*decompressor).copyData
-		return
-	}
-	f.finishBlock()
-}
-
-func (f *decompressor) finishBlock() {
-	if f.final {
-		if f.dict.availRead() > 0 {
-			f.toRead = f.dict.readFlush()
-		}
-		f.err = io.EOF
-	}
-	f.step = (*decompressor).nextBlock
-}
-
-// noEOF returns err, unless err == io.EOF, in which case it returns io.ErrUnexpectedEOF.
-func noEOF(e error) error {
-	if e == io.EOF {
-		return io.ErrUnexpectedEOF
-	}
-	return e
-}
-
-func (f *decompressor) moreBits() error {
-	c, err := f.r.ReadByte()
-	if err != nil {
-		return noEOF(err)
-	}
-	f.roffset++
-	f.b |= uint32(c) << (f.nb & regSizeMaskUint32)
-	f.nb += 8
-	return nil
-}
-
-// Read the next Huffman-encoded symbol from f according to h.
-func (f *decompressor) huffSym(h *huffmanDecoder) (int, error) {
-	// Since a huffmanDecoder can be empty or be composed of a degenerate tree
-	// with single element, huffSym must error on these two edge cases. In both
-	// cases, the chunks slice will be 0 for the invalid sequence, leading it
-	// satisfy the n == 0 check below.
-	n := uint(h.maxRead)
-	// Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers,
-	// but is smart enough to keep local variables in registers, so use nb and b,
-	// inline call to moreBits and reassign b,nb back to f on return.
-	nb, b := f.nb, f.b
-	for {
-		for nb < n {
-			c, err := f.r.ReadByte()
-			if err != nil {
-				f.b = b
-				f.nb = nb
-				return 0, noEOF(err)
-			}
-			f.roffset++
-			b |= uint32(c) << (nb & regSizeMaskUint32)
-			nb += 8
-		}
-		chunk := h.chunks[b&(huffmanNumChunks-1)]
-		n = uint(chunk & huffmanCountMask)
-		if n > huffmanChunkBits {
-			chunk = h.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&h.linkMask]
-			n = uint(chunk & huffmanCountMask)
-		}
-		if n <= nb {
-			if n == 0 {
-				f.b = b
-				f.nb = nb
-				if debugDecode {
-					fmt.Println("huffsym: n==0")
-				}
-				f.err = CorruptInputError(f.roffset)
-				return 0, f.err
-			}
-			f.b = b >> (n & regSizeMaskUint32)
-			f.nb = nb - n
-			return int(chunk >> huffmanValueShift), nil
-		}
-	}
-}
-
-func makeReader(r io.Reader) Reader {
-	if rr, ok := r.(Reader); ok {
-		return rr
-	}
-	return bufio.NewReader(r)
-}
-
-func fixedHuffmanDecoderInit() {
-	fixedOnce.Do(func() {
-		// These come from the RFC section 3.2.6.
-		var bits [288]int
-		for i := 0; i < 144; i++ {
-			bits[i] = 8
-		}
-		for i := 144; i < 256; i++ {
-			bits[i] = 9
-		}
-		for i := 256; i < 280; i++ {
-			bits[i] = 7
-		}
-		for i := 280; i < 288; i++ {
-			bits[i] = 8
-		}
-		fixedHuffmanDecoder.init(bits[:])
-	})
-}
-
-func (f *decompressor) Reset(r io.Reader, dict []byte) error {
-	*f = decompressor{
-		r:        makeReader(r),
-		bits:     f.bits,
-		codebits: f.codebits,
-		h1:       f.h1,
-		h2:       f.h2,
-		dict:     f.dict,
-		step:     (*decompressor).nextBlock,
-	}
-	f.dict.init(maxMatchOffset, dict)
-	return nil
-}
-
-// NewReader returns a new ReadCloser that can be used
-// to read the uncompressed version of r.
-// If r does not also implement io.ByteReader,
-// the decompressor may read more data than necessary from r.
-// It is the caller's responsibility to call Close on the ReadCloser
-// when finished reading.
-//
-// The ReadCloser returned by NewReader also implements Resetter.
-func NewReader(r io.Reader) io.ReadCloser {
-	fixedHuffmanDecoderInit()
-
-	var f decompressor
-	f.r = makeReader(r)
-	f.bits = new([maxNumLit + maxNumDist]int)
-	f.codebits = new([numCodes]int)
-	f.step = (*decompressor).nextBlock
-	f.dict.init(maxMatchOffset, nil)
-	return &f
-}
-
-// NewReaderDict is like NewReader but initializes the reader
-// with a preset dictionary. The returned Reader behaves as if
-// the uncompressed data stream started with the given dictionary,
-// which has already been read. NewReaderDict is typically used
-// to read data compressed by NewWriterDict.
-//
-// The ReadCloser returned by NewReader also implements Resetter.
-func NewReaderDict(r io.Reader, dict []byte) io.ReadCloser {
-	fixedHuffmanDecoderInit()
-
-	var f decompressor
-	f.r = makeReader(r)
-	f.bits = new([maxNumLit + maxNumDist]int)
-	f.codebits = new([numCodes]int)
-	f.step = (*decompressor).nextBlock
-	f.dict.init(maxMatchOffset, dict)
-	return &f
-}
diff --git a/vendor/github.com/klauspost/compress/flate/inflate_gen.go b/vendor/github.com/klauspost/compress/flate/inflate_gen.go
deleted file mode 100644
index cc6db27925..0000000000
--- a/vendor/github.com/klauspost/compress/flate/inflate_gen.go
+++ /dev/null
@@ -1,1002 +0,0 @@
-// Code generated by go generate gen_inflate.go. DO NOT EDIT.
-
-package flate
-
-import (
-	"bufio"
-	"bytes"
-	"fmt"
-	"math/bits"
-	"strings"
-)
-
-// Decode a single Huffman block from f.
-// hl and hd are the Huffman states for the lit/length values
-// and the distance values, respectively. If hd == nil, using the
-// fixed distance encoding associated with fixed Huffman blocks.
-func (f *decompressor) huffmanBytesBuffer() {
-	const (
-		stateInit = iota // Zero value must be stateInit
-		stateDict
-	)
-	fr := f.r.(*bytes.Buffer)
-
-	switch f.stepState {
-	case stateInit:
-		goto readLiteral
-	case stateDict:
-		goto copyHistory
-	}
-
-readLiteral:
-	// Read literal and/or (length, distance) according to RFC section 3.2.3.
-	{
-		var v int
-		{
-			// Inlined v, err := f.huffSym(f.hl)
-			// Since a huffmanDecoder can be empty or be composed of a degenerate tree
-			// with single element, huffSym must error on these two edge cases. In both
-			// cases, the chunks slice will be 0 for the invalid sequence, leading it
-			// satisfy the n == 0 check below.
-			n := uint(f.hl.maxRead)
-			// Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers,
-			// but is smart enough to keep local variables in registers, so use nb and b,
-			// inline call to moreBits and reassign b,nb back to f on return.
-			nb, b := f.nb, f.b
-			for {
-				for nb < n {
-					c, err := fr.ReadByte()
-					if err != nil {
-						f.b = b
-						f.nb = nb
-						f.err = noEOF(err)
-						return
-					}
-					f.roffset++
-					b |= uint32(c) << (nb & regSizeMaskUint32)
-					nb += 8
-				}
-				chunk := f.hl.chunks[b&(huffmanNumChunks-1)]
-				n = uint(chunk & huffmanCountMask)
-				if n > huffmanChunkBits {
-					chunk = f.hl.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hl.linkMask]
-					n = uint(chunk & huffmanCountMask)
-				}
-				if n <= nb {
-					if n == 0 {
-						f.b = b
-						f.nb = nb
-						if debugDecode {
-							fmt.Println("huffsym: n==0")
-						}
-						f.err = CorruptInputError(f.roffset)
-						return
-					}
-					f.b = b >> (n & regSizeMaskUint32)
-					f.nb = nb - n
-					v = int(chunk >> huffmanValueShift)
-					break
-				}
-			}
-		}
-
-		var length int
-		switch {
-		case v < 256:
-			f.dict.writeByte(byte(v))
-			if f.dict.availWrite() == 0 {
-				f.toRead = f.dict.readFlush()
-				f.step = (*decompressor).huffmanBytesBuffer
-				f.stepState = stateInit
-				return
-			}
-			goto readLiteral
-		case v == 256:
-			f.finishBlock()
-			return
-		// otherwise, reference to older data
-		case v < 265:
-			length = v - (257 - 3)
-		case v < maxNumLit:
-			val := decCodeToLen[(v - 257)]
-			length = int(val.length) + 3
-			n := uint(val.extra)
-			for f.nb < n {
-				c, err := fr.ReadByte()
-				if err != nil {
-					if debugDecode {
-						fmt.Println("morebits n>0:", err)
-					}
-					f.err = err
-					return
-				}
-				f.roffset++
-				f.b |= uint32(c) << f.nb
-				f.nb += 8
-			}
-			length += int(f.b & uint32(1<<(n&regSizeMaskUint32)-1))
-			f.b >>= n & regSizeMaskUint32
-			f.nb -= n
-		default:
-			if debugDecode {
-				fmt.Println(v, ">= maxNumLit")
-			}
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-
-		var dist uint32
-		if f.hd == nil {
-			for f.nb < 5 {
-				c, err := fr.ReadByte()
-				if err != nil {
-					if debugDecode {
-						fmt.Println("morebits f.nb<5:", err)
-					}
-					f.err = err
-					return
-				}
-				f.roffset++
-				f.b |= uint32(c) << f.nb
-				f.nb += 8
-			}
-			dist = uint32(bits.Reverse8(uint8(f.b & 0x1F << 3)))
-			f.b >>= 5
-			f.nb -= 5
-		} else {
-			// Since a huffmanDecoder can be empty or be composed of a degenerate tree
-			// with single element, huffSym must error on these two edge cases. In both
-			// cases, the chunks slice will be 0 for the invalid sequence, leading it
-			// satisfy the n == 0 check below.
-			n := uint(f.hd.maxRead)
-			// Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers,
-			// but is smart enough to keep local variables in registers, so use nb and b,
-			// inline call to moreBits and reassign b,nb back to f on return.
-			nb, b := f.nb, f.b
-			for {
-				for nb < n {
-					c, err := fr.ReadByte()
-					if err != nil {
-						f.b = b
-						f.nb = nb
-						f.err = noEOF(err)
-						return
-					}
-					f.roffset++
-					b |= uint32(c) << (nb & regSizeMaskUint32)
-					nb += 8
-				}
-				chunk := f.hd.chunks[b&(huffmanNumChunks-1)]
-				n = uint(chunk & huffmanCountMask)
-				if n > huffmanChunkBits {
-					chunk = f.hd.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hd.linkMask]
-					n = uint(chunk & huffmanCountMask)
-				}
-				if n <= nb {
-					if n == 0 {
-						f.b = b
-						f.nb = nb
-						if debugDecode {
-							fmt.Println("huffsym: n==0")
-						}
-						f.err = CorruptInputError(f.roffset)
-						return
-					}
-					f.b = b >> (n & regSizeMaskUint32)
-					f.nb = nb - n
-					dist = uint32(chunk >> huffmanValueShift)
-					break
-				}
-			}
-		}
-
-		switch {
-		case dist < 4:
-			dist++
-		case dist < maxNumDist:
-			nb := uint(dist-2) >> 1
-			// have 1 bit in bottom of dist, need nb more.
-			extra := (dist & 1) << (nb & regSizeMaskUint32)
-			for f.nb < nb {
-				c, err := fr.ReadByte()
-				if err != nil {
-					if debugDecode {
-						fmt.Println("morebits f.nb<nb:", err)
-					}
-					f.err = err
-					return
-				}
-				f.roffset++
-				f.b |= uint32(c) << f.nb
-				f.nb += 8
-			}
-			extra |= f.b & uint32(1<<(nb&regSizeMaskUint32)-1)
-			f.b >>= nb & regSizeMaskUint32
-			f.nb -= nb
-			dist = 1<<((nb+1)&regSizeMaskUint32) + 1 + extra
-		default:
-			if debugDecode {
-				fmt.Println("dist too big:", dist, maxNumDist)
-			}
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-
-		// No check on length; encoding can be prescient.
-		if dist > uint32(f.dict.histSize()) {
-			if debugDecode {
-				fmt.Println("dist > f.dict.histSize():", dist, f.dict.histSize())
-			}
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-
-		f.copyLen, f.copyDist = length, int(dist)
-		goto copyHistory
-	}
-
-copyHistory:
-	// Perform a backwards copy according to RFC section 3.2.3.
-	{
-		cnt := f.dict.tryWriteCopy(f.copyDist, f.copyLen)
-		if cnt == 0 {
-			cnt = f.dict.writeCopy(f.copyDist, f.copyLen)
-		}
-		f.copyLen -= cnt
-
-		if f.dict.availWrite() == 0 || f.copyLen > 0 {
-			f.toRead = f.dict.readFlush()
-			f.step = (*decompressor).huffmanBytesBuffer // We need to continue this work
-			f.stepState = stateDict
-			return
-		}
-		goto readLiteral
-	}
-}
-
-// Decode a single Huffman block from f.
-// hl and hd are the Huffman states for the lit/length values
-// and the distance values, respectively. If hd == nil, using the
-// fixed distance encoding associated with fixed Huffman blocks.
-func (f *decompressor) huffmanBytesReader() {
-	const (
-		stateInit = iota // Zero value must be stateInit
-		stateDict
-	)
-	fr := f.r.(*bytes.Reader)
-
-	switch f.stepState {
-	case stateInit:
-		goto readLiteral
-	case stateDict:
-		goto copyHistory
-	}
-
-readLiteral:
-	// Read literal and/or (length, distance) according to RFC section 3.2.3.
-	{
-		var v int
-		{
-			// Inlined v, err := f.huffSym(f.hl)
-			// Since a huffmanDecoder can be empty or be composed of a degenerate tree
-			// with single element, huffSym must error on these two edge cases. In both
-			// cases, the chunks slice will be 0 for the invalid sequence, leading it
-			// satisfy the n == 0 check below.
-			n := uint(f.hl.maxRead)
-			// Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers,
-			// but is smart enough to keep local variables in registers, so use nb and b,
-			// inline call to moreBits and reassign b,nb back to f on return.
-			nb, b := f.nb, f.b
-			for {
-				for nb < n {
-					c, err := fr.ReadByte()
-					if err != nil {
-						f.b = b
-						f.nb = nb
-						f.err = noEOF(err)
-						return
-					}
-					f.roffset++
-					b |= uint32(c) << (nb & regSizeMaskUint32)
-					nb += 8
-				}
-				chunk := f.hl.chunks[b&(huffmanNumChunks-1)]
-				n = uint(chunk & huffmanCountMask)
-				if n > huffmanChunkBits {
-					chunk = f.hl.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hl.linkMask]
-					n = uint(chunk & huffmanCountMask)
-				}
-				if n <= nb {
-					if n == 0 {
-						f.b = b
-						f.nb = nb
-						if debugDecode {
-							fmt.Println("huffsym: n==0")
-						}
-						f.err = CorruptInputError(f.roffset)
-						return
-					}
-					f.b = b >> (n & regSizeMaskUint32)
-					f.nb = nb - n
-					v = int(chunk >> huffmanValueShift)
-					break
-				}
-			}
-		}
-
-		var length int
-		switch {
-		case v < 256:
-			f.dict.writeByte(byte(v))
-			if f.dict.availWrite() == 0 {
-				f.toRead = f.dict.readFlush()
-				f.step = (*decompressor).huffmanBytesReader
-				f.stepState = stateInit
-				return
-			}
-			goto readLiteral
-		case v == 256:
-			f.finishBlock()
-			return
-		// otherwise, reference to older data
-		case v < 265:
-			length = v - (257 - 3)
-		case v < maxNumLit:
-			val := decCodeToLen[(v - 257)]
-			length = int(val.length) + 3
-			n := uint(val.extra)
-			for f.nb < n {
-				c, err := fr.ReadByte()
-				if err != nil {
-					if debugDecode {
-						fmt.Println("morebits n>0:", err)
-					}
-					f.err = err
-					return
-				}
-				f.roffset++
-				f.b |= uint32(c) << f.nb
-				f.nb += 8
-			}
-			length += int(f.b & uint32(1<<(n&regSizeMaskUint32)-1))
-			f.b >>= n & regSizeMaskUint32
-			f.nb -= n
-		default:
-			if debugDecode {
-				fmt.Println(v, ">= maxNumLit")
-			}
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-
-		var dist uint32
-		if f.hd == nil {
-			for f.nb < 5 {
-				c, err := fr.ReadByte()
-				if err != nil {
-					if debugDecode {
-						fmt.Println("morebits f.nb<5:", err)
-					}
-					f.err = err
-					return
-				}
-				f.roffset++
-				f.b |= uint32(c) << f.nb
-				f.nb += 8
-			}
-			dist = uint32(bits.Reverse8(uint8(f.b & 0x1F << 3)))
-			f.b >>= 5
-			f.nb -= 5
-		} else {
-			// Since a huffmanDecoder can be empty or be composed of a degenerate tree
-			// with single element, huffSym must error on these two edge cases. In both
-			// cases, the chunks slice will be 0 for the invalid sequence, leading it
-			// satisfy the n == 0 check below.
-			n := uint(f.hd.maxRead)
-			// Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers,
-			// but is smart enough to keep local variables in registers, so use nb and b,
-			// inline call to moreBits and reassign b,nb back to f on return.
-			nb, b := f.nb, f.b
-			for {
-				for nb < n {
-					c, err := fr.ReadByte()
-					if err != nil {
-						f.b = b
-						f.nb = nb
-						f.err = noEOF(err)
-						return
-					}
-					f.roffset++
-					b |= uint32(c) << (nb & regSizeMaskUint32)
-					nb += 8
-				}
-				chunk := f.hd.chunks[b&(huffmanNumChunks-1)]
-				n = uint(chunk & huffmanCountMask)
-				if n > huffmanChunkBits {
-					chunk = f.hd.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hd.linkMask]
-					n = uint(chunk & huffmanCountMask)
-				}
-				if n <= nb {
-					if n == 0 {
-						f.b = b
-						f.nb = nb
-						if debugDecode {
-							fmt.Println("huffsym: n==0")
-						}
-						f.err = CorruptInputError(f.roffset)
-						return
-					}
-					f.b = b >> (n & regSizeMaskUint32)
-					f.nb = nb - n
-					dist = uint32(chunk >> huffmanValueShift)
-					break
-				}
-			}
-		}
-
-		switch {
-		case dist < 4:
-			dist++
-		case dist < maxNumDist:
-			nb := uint(dist-2) >> 1
-			// have 1 bit in bottom of dist, need nb more.
-			extra := (dist & 1) << (nb & regSizeMaskUint32)
-			for f.nb < nb {
-				c, err := fr.ReadByte()
-				if err != nil {
-					if debugDecode {
-						fmt.Println("morebits f.nb<nb:", err)
-					}
-					f.err = err
-					return
-				}
-				f.roffset++
-				f.b |= uint32(c) << f.nb
-				f.nb += 8
-			}
-			extra |= f.b & uint32(1<<(nb&regSizeMaskUint32)-1)
-			f.b >>= nb & regSizeMaskUint32
-			f.nb -= nb
-			dist = 1<<((nb+1)&regSizeMaskUint32) + 1 + extra
-		default:
-			if debugDecode {
-				fmt.Println("dist too big:", dist, maxNumDist)
-			}
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-
-		// No check on length; encoding can be prescient.
-		if dist > uint32(f.dict.histSize()) {
-			if debugDecode {
-				fmt.Println("dist > f.dict.histSize():", dist, f.dict.histSize())
-			}
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-
-		f.copyLen, f.copyDist = length, int(dist)
-		goto copyHistory
-	}
-
-copyHistory:
-	// Perform a backwards copy according to RFC section 3.2.3.
-	{
-		cnt := f.dict.tryWriteCopy(f.copyDist, f.copyLen)
-		if cnt == 0 {
-			cnt = f.dict.writeCopy(f.copyDist, f.copyLen)
-		}
-		f.copyLen -= cnt
-
-		if f.dict.availWrite() == 0 || f.copyLen > 0 {
-			f.toRead = f.dict.readFlush()
-			f.step = (*decompressor).huffmanBytesReader // We need to continue this work
-			f.stepState = stateDict
-			return
-		}
-		goto readLiteral
-	}
-}
-
-// Decode a single Huffman block from f.
-// hl and hd are the Huffman states for the lit/length values
-// and the distance values, respectively. If hd == nil, using the
-// fixed distance encoding associated with fixed Huffman blocks.
-func (f *decompressor) huffmanBufioReader() {
-	const (
-		stateInit = iota // Zero value must be stateInit
-		stateDict
-	)
-	fr := f.r.(*bufio.Reader)
-
-	switch f.stepState {
-	case stateInit:
-		goto readLiteral
-	case stateDict:
-		goto copyHistory
-	}
-
-readLiteral:
-	// Read literal and/or (length, distance) according to RFC section 3.2.3.
-	{
-		var v int
-		{
-			// Inlined v, err := f.huffSym(f.hl)
-			// Since a huffmanDecoder can be empty or be composed of a degenerate tree
-			// with single element, huffSym must error on these two edge cases. In both
-			// cases, the chunks slice will be 0 for the invalid sequence, leading it
-			// satisfy the n == 0 check below.
-			n := uint(f.hl.maxRead)
-			// Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers,
-			// but is smart enough to keep local variables in registers, so use nb and b,
-			// inline call to moreBits and reassign b,nb back to f on return.
-			nb, b := f.nb, f.b
-			for {
-				for nb < n {
-					c, err := fr.ReadByte()
-					if err != nil {
-						f.b = b
-						f.nb = nb
-						f.err = noEOF(err)
-						return
-					}
-					f.roffset++
-					b |= uint32(c) << (nb & regSizeMaskUint32)
-					nb += 8
-				}
-				chunk := f.hl.chunks[b&(huffmanNumChunks-1)]
-				n = uint(chunk & huffmanCountMask)
-				if n > huffmanChunkBits {
-					chunk = f.hl.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hl.linkMask]
-					n = uint(chunk & huffmanCountMask)
-				}
-				if n <= nb {
-					if n == 0 {
-						f.b = b
-						f.nb = nb
-						if debugDecode {
-							fmt.Println("huffsym: n==0")
-						}
-						f.err = CorruptInputError(f.roffset)
-						return
-					}
-					f.b = b >> (n & regSizeMaskUint32)
-					f.nb = nb - n
-					v = int(chunk >> huffmanValueShift)
-					break
-				}
-			}
-		}
-
-		var length int
-		switch {
-		case v < 256:
-			f.dict.writeByte(byte(v))
-			if f.dict.availWrite() == 0 {
-				f.toRead = f.dict.readFlush()
-				f.step = (*decompressor).huffmanBufioReader
-				f.stepState = stateInit
-				return
-			}
-			goto readLiteral
-		case v == 256:
-			f.finishBlock()
-			return
-		// otherwise, reference to older data
-		case v < 265:
-			length = v - (257 - 3)
-		case v < maxNumLit:
-			val := decCodeToLen[(v - 257)]
-			length = int(val.length) + 3
-			n := uint(val.extra)
-			for f.nb < n {
-				c, err := fr.ReadByte()
-				if err != nil {
-					if debugDecode {
-						fmt.Println("morebits n>0:", err)
-					}
-					f.err = err
-					return
-				}
-				f.roffset++
-				f.b |= uint32(c) << f.nb
-				f.nb += 8
-			}
-			length += int(f.b & uint32(1<<(n&regSizeMaskUint32)-1))
-			f.b >>= n & regSizeMaskUint32
-			f.nb -= n
-		default:
-			if debugDecode {
-				fmt.Println(v, ">= maxNumLit")
-			}
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-
-		var dist uint32
-		if f.hd == nil {
-			for f.nb < 5 {
-				c, err := fr.ReadByte()
-				if err != nil {
-					if debugDecode {
-						fmt.Println("morebits f.nb<5:", err)
-					}
-					f.err = err
-					return
-				}
-				f.roffset++
-				f.b |= uint32(c) << f.nb
-				f.nb += 8
-			}
-			dist = uint32(bits.Reverse8(uint8(f.b & 0x1F << 3)))
-			f.b >>= 5
-			f.nb -= 5
-		} else {
-			// Since a huffmanDecoder can be empty or be composed of a degenerate tree
-			// with single element, huffSym must error on these two edge cases. In both
-			// cases, the chunks slice will be 0 for the invalid sequence, leading it
-			// satisfy the n == 0 check below.
-			n := uint(f.hd.maxRead)
-			// Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers,
-			// but is smart enough to keep local variables in registers, so use nb and b,
-			// inline call to moreBits and reassign b,nb back to f on return.
-			nb, b := f.nb, f.b
-			for {
-				for nb < n {
-					c, err := fr.ReadByte()
-					if err != nil {
-						f.b = b
-						f.nb = nb
-						f.err = noEOF(err)
-						return
-					}
-					f.roffset++
-					b |= uint32(c) << (nb & regSizeMaskUint32)
-					nb += 8
-				}
-				chunk := f.hd.chunks[b&(huffmanNumChunks-1)]
-				n = uint(chunk & huffmanCountMask)
-				if n > huffmanChunkBits {
-					chunk = f.hd.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hd.linkMask]
-					n = uint(chunk & huffmanCountMask)
-				}
-				if n <= nb {
-					if n == 0 {
-						f.b = b
-						f.nb = nb
-						if debugDecode {
-							fmt.Println("huffsym: n==0")
-						}
-						f.err = CorruptInputError(f.roffset)
-						return
-					}
-					f.b = b >> (n & regSizeMaskUint32)
-					f.nb = nb - n
-					dist = uint32(chunk >> huffmanValueShift)
-					break
-				}
-			}
-		}
-
-		switch {
-		case dist < 4:
-			dist++
-		case dist < maxNumDist:
-			nb := uint(dist-2) >> 1
-			// have 1 bit in bottom of dist, need nb more.
-			extra := (dist & 1) << (nb & regSizeMaskUint32)
-			for f.nb < nb {
-				c, err := fr.ReadByte()
-				if err != nil {
-					if debugDecode {
-						fmt.Println("morebits f.nb<nb:", err)
-					}
-					f.err = err
-					return
-				}
-				f.roffset++
-				f.b |= uint32(c) << f.nb
-				f.nb += 8
-			}
-			extra |= f.b & uint32(1<<(nb&regSizeMaskUint32)-1)
-			f.b >>= nb & regSizeMaskUint32
-			f.nb -= nb
-			dist = 1<<((nb+1)&regSizeMaskUint32) + 1 + extra
-		default:
-			if debugDecode {
-				fmt.Println("dist too big:", dist, maxNumDist)
-			}
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-
-		// No check on length; encoding can be prescient.
-		if dist > uint32(f.dict.histSize()) {
-			if debugDecode {
-				fmt.Println("dist > f.dict.histSize():", dist, f.dict.histSize())
-			}
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-
-		f.copyLen, f.copyDist = length, int(dist)
-		goto copyHistory
-	}
-
-copyHistory:
-	// Perform a backwards copy according to RFC section 3.2.3.
-	{
-		cnt := f.dict.tryWriteCopy(f.copyDist, f.copyLen)
-		if cnt == 0 {
-			cnt = f.dict.writeCopy(f.copyDist, f.copyLen)
-		}
-		f.copyLen -= cnt
-
-		if f.dict.availWrite() == 0 || f.copyLen > 0 {
-			f.toRead = f.dict.readFlush()
-			f.step = (*decompressor).huffmanBufioReader // We need to continue this work
-			f.stepState = stateDict
-			return
-		}
-		goto readLiteral
-	}
-}
-
-// Decode a single Huffman block from f.
-// hl and hd are the Huffman states for the lit/length values
-// and the distance values, respectively. If hd == nil, using the
-// fixed distance encoding associated with fixed Huffman blocks.
-func (f *decompressor) huffmanStringsReader() {
-	const (
-		stateInit = iota // Zero value must be stateInit
-		stateDict
-	)
-	fr := f.r.(*strings.Reader)
-
-	switch f.stepState {
-	case stateInit:
-		goto readLiteral
-	case stateDict:
-		goto copyHistory
-	}
-
-readLiteral:
-	// Read literal and/or (length, distance) according to RFC section 3.2.3.
-	{
-		var v int
-		{
-			// Inlined v, err := f.huffSym(f.hl)
-			// Since a huffmanDecoder can be empty or be composed of a degenerate tree
-			// with single element, huffSym must error on these two edge cases. In both
-			// cases, the chunks slice will be 0 for the invalid sequence, leading it
-			// satisfy the n == 0 check below.
-			n := uint(f.hl.maxRead)
-			// Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers,
-			// but is smart enough to keep local variables in registers, so use nb and b,
-			// inline call to moreBits and reassign b,nb back to f on return.
-			nb, b := f.nb, f.b
-			for {
-				for nb < n {
-					c, err := fr.ReadByte()
-					if err != nil {
-						f.b = b
-						f.nb = nb
-						f.err = noEOF(err)
-						return
-					}
-					f.roffset++
-					b |= uint32(c) << (nb & regSizeMaskUint32)
-					nb += 8
-				}
-				chunk := f.hl.chunks[b&(huffmanNumChunks-1)]
-				n = uint(chunk & huffmanCountMask)
-				if n > huffmanChunkBits {
-					chunk = f.hl.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hl.linkMask]
-					n = uint(chunk & huffmanCountMask)
-				}
-				if n <= nb {
-					if n == 0 {
-						f.b = b
-						f.nb = nb
-						if debugDecode {
-							fmt.Println("huffsym: n==0")
-						}
-						f.err = CorruptInputError(f.roffset)
-						return
-					}
-					f.b = b >> (n & regSizeMaskUint32)
-					f.nb = nb - n
-					v = int(chunk >> huffmanValueShift)
-					break
-				}
-			}
-		}
-
-		var length int
-		switch {
-		case v < 256:
-			f.dict.writeByte(byte(v))
-			if f.dict.availWrite() == 0 {
-				f.toRead = f.dict.readFlush()
-				f.step = (*decompressor).huffmanStringsReader
-				f.stepState = stateInit
-				return
-			}
-			goto readLiteral
-		case v == 256:
-			f.finishBlock()
-			return
-		// otherwise, reference to older data
-		case v < 265:
-			length = v - (257 - 3)
-		case v < maxNumLit:
-			val := decCodeToLen[(v - 257)]
-			length = int(val.length) + 3
-			n := uint(val.extra)
-			for f.nb < n {
-				c, err := fr.ReadByte()
-				if err != nil {
-					if debugDecode {
-						fmt.Println("morebits n>0:", err)
-					}
-					f.err = err
-					return
-				}
-				f.roffset++
-				f.b |= uint32(c) << f.nb
-				f.nb += 8
-			}
-			length += int(f.b & uint32(1<<(n&regSizeMaskUint32)-1))
-			f.b >>= n & regSizeMaskUint32
-			f.nb -= n
-		default:
-			if debugDecode {
-				fmt.Println(v, ">= maxNumLit")
-			}
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-
-		var dist uint32
-		if f.hd == nil {
-			for f.nb < 5 {
-				c, err := fr.ReadByte()
-				if err != nil {
-					if debugDecode {
-						fmt.Println("morebits f.nb<5:", err)
-					}
-					f.err = err
-					return
-				}
-				f.roffset++
-				f.b |= uint32(c) << f.nb
-				f.nb += 8
-			}
-			dist = uint32(bits.Reverse8(uint8(f.b & 0x1F << 3)))
-			f.b >>= 5
-			f.nb -= 5
-		} else {
-			// Since a huffmanDecoder can be empty or be composed of a degenerate tree
-			// with single element, huffSym must error on these two edge cases. In both
-			// cases, the chunks slice will be 0 for the invalid sequence, leading it
-			// satisfy the n == 0 check below.
-			n := uint(f.hd.maxRead)
-			// Optimization. Compiler isn't smart enough to keep f.b,f.nb in registers,
-			// but is smart enough to keep local variables in registers, so use nb and b,
-			// inline call to moreBits and reassign b,nb back to f on return.
-			nb, b := f.nb, f.b
-			for {
-				for nb < n {
-					c, err := fr.ReadByte()
-					if err != nil {
-						f.b = b
-						f.nb = nb
-						f.err = noEOF(err)
-						return
-					}
-					f.roffset++
-					b |= uint32(c) << (nb & regSizeMaskUint32)
-					nb += 8
-				}
-				chunk := f.hd.chunks[b&(huffmanNumChunks-1)]
-				n = uint(chunk & huffmanCountMask)
-				if n > huffmanChunkBits {
-					chunk = f.hd.links[chunk>>huffmanValueShift][(b>>huffmanChunkBits)&f.hd.linkMask]
-					n = uint(chunk & huffmanCountMask)
-				}
-				if n <= nb {
-					if n == 0 {
-						f.b = b
-						f.nb = nb
-						if debugDecode {
-							fmt.Println("huffsym: n==0")
-						}
-						f.err = CorruptInputError(f.roffset)
-						return
-					}
-					f.b = b >> (n & regSizeMaskUint32)
-					f.nb = nb - n
-					dist = uint32(chunk >> huffmanValueShift)
-					break
-				}
-			}
-		}
-
-		switch {
-		case dist < 4:
-			dist++
-		case dist < maxNumDist:
-			nb := uint(dist-2) >> 1
-			// have 1 bit in bottom of dist, need nb more.
-			extra := (dist & 1) << (nb & regSizeMaskUint32)
-			for f.nb < nb {
-				c, err := fr.ReadByte()
-				if err != nil {
-					if debugDecode {
-						fmt.Println("morebits f.nb<nb:", err)
-					}
-					f.err = err
-					return
-				}
-				f.roffset++
-				f.b |= uint32(c) << f.nb
-				f.nb += 8
-			}
-			extra |= f.b & uint32(1<<(nb&regSizeMaskUint32)-1)
-			f.b >>= nb & regSizeMaskUint32
-			f.nb -= nb
-			dist = 1<<((nb+1)&regSizeMaskUint32) + 1 + extra
-		default:
-			if debugDecode {
-				fmt.Println("dist too big:", dist, maxNumDist)
-			}
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-
-		// No check on length; encoding can be prescient.
-		if dist > uint32(f.dict.histSize()) {
-			if debugDecode {
-				fmt.Println("dist > f.dict.histSize():", dist, f.dict.histSize())
-			}
-			f.err = CorruptInputError(f.roffset)
-			return
-		}
-
-		f.copyLen, f.copyDist = length, int(dist)
-		goto copyHistory
-	}
-
-copyHistory:
-	// Perform a backwards copy according to RFC section 3.2.3.
-	{
-		cnt := f.dict.tryWriteCopy(f.copyDist, f.copyLen)
-		if cnt == 0 {
-			cnt = f.dict.writeCopy(f.copyDist, f.copyLen)
-		}
-		f.copyLen -= cnt
-
-		if f.dict.availWrite() == 0 || f.copyLen > 0 {
-			f.toRead = f.dict.readFlush()
-			f.step = (*decompressor).huffmanStringsReader // We need to continue this work
-			f.stepState = stateDict
-			return
-		}
-		goto readLiteral
-	}
-}
-
-func (f *decompressor) huffmanBlockDecoder() func() {
-	switch f.r.(type) {
-	case *bytes.Buffer:
-		return f.huffmanBytesBuffer
-	case *bytes.Reader:
-		return f.huffmanBytesReader
-	case *bufio.Reader:
-		return f.huffmanBufioReader
-	case *strings.Reader:
-		return f.huffmanStringsReader
-	default:
-		return f.huffmanBlockGeneric
-	}
-}
diff --git a/vendor/github.com/klauspost/compress/flate/level1.go b/vendor/github.com/klauspost/compress/flate/level1.go
deleted file mode 100644
index 1e5eea3968..0000000000
--- a/vendor/github.com/klauspost/compress/flate/level1.go
+++ /dev/null
@@ -1,179 +0,0 @@
-package flate
-
-import "fmt"
-
-// fastGen maintains the table for matches,
-// and the previous byte block for level 2.
-// This is the generic implementation.
-type fastEncL1 struct {
-	fastGen
-	table [tableSize]tableEntry
-}
-
-// EncodeL1 uses a similar algorithm to level 1
-func (e *fastEncL1) Encode(dst *tokens, src []byte) {
-	const (
-		inputMargin            = 12 - 1
-		minNonLiteralBlockSize = 1 + 1 + inputMargin
-	)
-	if debugDeflate && e.cur < 0 {
-		panic(fmt.Sprint("e.cur < 0: ", e.cur))
-	}
-
-	// Protect against e.cur wraparound.
-	for e.cur >= bufferReset {
-		if len(e.hist) == 0 {
-			for i := range e.table[:] {
-				e.table[i] = tableEntry{}
-			}
-			e.cur = maxMatchOffset
-			break
-		}
-		// Shift down everything in the table that isn't already too far away.
-		minOff := e.cur + int32(len(e.hist)) - maxMatchOffset
-		for i := range e.table[:] {
-			v := e.table[i].offset
-			if v <= minOff {
-				v = 0
-			} else {
-				v = v - e.cur + maxMatchOffset
-			}
-			e.table[i].offset = v
-		}
-		e.cur = maxMatchOffset
-	}
-
-	s := e.addBlock(src)
-
-	// This check isn't in the Snappy implementation, but there, the caller
-	// instead of the callee handles this case.
-	if len(src) < minNonLiteralBlockSize {
-		// We do not fill the token table.
-		// This will be picked up by caller.
-		dst.n = uint16(len(src))
-		return
-	}
-
-	// Override src
-	src = e.hist
-	nextEmit := s
-
-	// sLimit is when to stop looking for offset/length copies. The inputMargin
-	// lets us use a fast path for emitLiteral in the main loop, while we are
-	// looking for copies.
-	sLimit := int32(len(src) - inputMargin)
-
-	// nextEmit is where in src the next emitLiteral should start from.
-	cv := load3232(src, s)
-
-	for {
-		const skipLog = 5
-		const doEvery = 2
-
-		nextS := s
-		var candidate tableEntry
-		for {
-			nextHash := hash(cv)
-			candidate = e.table[nextHash]
-			nextS = s + doEvery + (s-nextEmit)>>skipLog
-			if nextS > sLimit {
-				goto emitRemainder
-			}
-
-			now := load6432(src, nextS)
-			e.table[nextHash] = tableEntry{offset: s + e.cur}
-			nextHash = hash(uint32(now))
-
-			offset := s - (candidate.offset - e.cur)
-			if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) {
-				e.table[nextHash] = tableEntry{offset: nextS + e.cur}
-				break
-			}
-
-			// Do one right away...
-			cv = uint32(now)
-			s = nextS
-			nextS++
-			candidate = e.table[nextHash]
-			now >>= 8
-			e.table[nextHash] = tableEntry{offset: s + e.cur}
-
-			offset = s - (candidate.offset - e.cur)
-			if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) {
-				e.table[nextHash] = tableEntry{offset: nextS + e.cur}
-				break
-			}
-			cv = uint32(now)
-			s = nextS
-		}
-
-		// A 4-byte match has been found. We'll later see if more than 4 bytes
-		// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
-		// them as literal bytes.
-		for {
-			// Invariant: we have a 4-byte match at s, and no need to emit any
-			// literal bytes prior to s.
-
-			// Extend the 4-byte match as long as possible.
-			t := candidate.offset - e.cur
-			l := e.matchlenLong(s+4, t+4, src) + 4
-
-			// Extend backwards
-			for t > 0 && s > nextEmit && src[t-1] == src[s-1] {
-				s--
-				t--
-				l++
-			}
-			if nextEmit < s {
-				emitLiteral(dst, src[nextEmit:s])
-			}
-
-			// Save the match found
-			dst.AddMatchLong(l, uint32(s-t-baseMatchOffset))
-			s += l
-			nextEmit = s
-			if nextS >= s {
-				s = nextS + 1
-			}
-			if s >= sLimit {
-				// Index first pair after match end.
-				if int(s+l+4) < len(src) {
-					cv := load3232(src, s)
-					e.table[hash(cv)] = tableEntry{offset: s + e.cur}
-				}
-				goto emitRemainder
-			}
-
-			// We could immediately start working at s now, but to improve
-			// compression we first update the hash table at s-2 and at s. If
-			// another emitCopy is not our next move, also calculate nextHash
-			// at s+1. At least on GOARCH=amd64, these three hash calculations
-			// are faster as one load64 call (with some shifts) instead of
-			// three load32 calls.
-			x := load6432(src, s-2)
-			o := e.cur + s - 2
-			prevHash := hash(uint32(x))
-			e.table[prevHash] = tableEntry{offset: o}
-			x >>= 16
-			currHash := hash(uint32(x))
-			candidate = e.table[currHash]
-			e.table[currHash] = tableEntry{offset: o + 2}
-
-			offset := s - (candidate.offset - e.cur)
-			if offset > maxMatchOffset || uint32(x) != load3232(src, candidate.offset-e.cur) {
-				cv = uint32(x >> 8)
-				s++
-				break
-			}
-		}
-	}
-
-emitRemainder:
-	if int(nextEmit) < len(src) {
-		// If nothing was added, don't encode literals.
-		if dst.n == 0 {
-			return
-		}
-		emitLiteral(dst, src[nextEmit:])
-	}
-}
diff --git a/vendor/github.com/klauspost/compress/flate/level2.go b/vendor/github.com/klauspost/compress/flate/level2.go
deleted file mode 100644
index 234c4389ab..0000000000
--- a/vendor/github.com/klauspost/compress/flate/level2.go
+++ /dev/null
@@ -1,205 +0,0 @@
-package flate
-
-import "fmt"
-
-// fastGen maintains the table for matches,
-// and the previous byte block for level 2.
-// This is the generic implementation.
-type fastEncL2 struct {
-	fastGen
-	table [bTableSize]tableEntry
-}
-
-// EncodeL2 uses a similar algorithm to level 1, but is capable
-// of matching across blocks giving better compression at a small slowdown.
-func (e *fastEncL2) Encode(dst *tokens, src []byte) {
-	const (
-		inputMargin            = 12 - 1
-		minNonLiteralBlockSize = 1 + 1 + inputMargin
-	)
-
-	if debugDeflate && e.cur < 0 {
-		panic(fmt.Sprint("e.cur < 0: ", e.cur))
-	}
-
-	// Protect against e.cur wraparound.
-	for e.cur >= bufferReset {
-		if len(e.hist) == 0 {
-			for i := range e.table[:] {
-				e.table[i] = tableEntry{}
-			}
-			e.cur = maxMatchOffset
-			break
-		}
-		// Shift down everything in the table that isn't already too far away.
-		minOff := e.cur + int32(len(e.hist)) - maxMatchOffset
-		for i := range e.table[:] {
-			v := e.table[i].offset
-			if v <= minOff {
-				v = 0
-			} else {
-				v = v - e.cur + maxMatchOffset
-			}
-			e.table[i].offset = v
-		}
-		e.cur = maxMatchOffset
-	}
-
-	s := e.addBlock(src)
-
-	// This check isn't in the Snappy implementation, but there, the caller
-	// instead of the callee handles this case.
-	if len(src) < minNonLiteralBlockSize {
-		// We do not fill the token table.
-		// This will be picked up by caller.
-		dst.n = uint16(len(src))
-		return
-	}
-
-	// Override src
-	src = e.hist
-	nextEmit := s
-
-	// sLimit is when to stop looking for offset/length copies. The inputMargin
-	// lets us use a fast path for emitLiteral in the main loop, while we are
-	// looking for copies.
-	sLimit := int32(len(src) - inputMargin)
-
-	// nextEmit is where in src the next emitLiteral should start from.
-	cv := load3232(src, s)
-	for {
-		// When should we start skipping if we haven't found matches in a long while.
-		const skipLog = 5
-		const doEvery = 2
-
-		nextS := s
-		var candidate tableEntry
-		for {
-			nextHash := hash4u(cv, bTableBits)
-			s = nextS
-			nextS = s + doEvery + (s-nextEmit)>>skipLog
-			if nextS > sLimit {
-				goto emitRemainder
-			}
-			candidate = e.table[nextHash]
-			now := load6432(src, nextS)
-			e.table[nextHash] = tableEntry{offset: s + e.cur}
-			nextHash = hash4u(uint32(now), bTableBits)
-
-			offset := s - (candidate.offset - e.cur)
-			if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) {
-				e.table[nextHash] = tableEntry{offset: nextS + e.cur}
-				break
-			}
-
-			// Do one right away...
-			cv = uint32(now)
-			s = nextS
-			nextS++
-			candidate = e.table[nextHash]
-			now >>= 8
-			e.table[nextHash] = tableEntry{offset: s + e.cur}
-
-			offset = s - (candidate.offset - e.cur)
-			if offset < maxMatchOffset && cv == load3232(src, candidate.offset-e.cur) {
-				break
-			}
-			cv = uint32(now)
-		}
-
-		// A 4-byte match has been found. We'll later see if more than 4 bytes
-		// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
-		// them as literal bytes.
-
-		// Call emitCopy, and then see if another emitCopy could be our next
-		// move. Repeat until we find no match for the input immediately after
-		// what was consumed by the last emitCopy call.
-		//
-		// If we exit this loop normally then we need to call emitLiteral next,
-		// though we don't yet know how big the literal will be. We handle that
-		// by proceeding to the next iteration of the main loop. We also can
-		// exit this loop via goto if we get close to exhausting the input.
-		for {
-			// Invariant: we have a 4-byte match at s, and no need to emit any
-			// literal bytes prior to s.
-
-			// Extend the 4-byte match as long as possible.
-			t := candidate.offset - e.cur
-			l := e.matchlenLong(s+4, t+4, src) + 4
-
-			// Extend backwards
-			for t > 0 && s > nextEmit && src[t-1] == src[s-1] {
-				s--
-				t--
-				l++
-			}
-			if nextEmit < s {
-				emitLiteral(dst, src[nextEmit:s])
-			}
-
-			dst.AddMatchLong(l, uint32(s-t-baseMatchOffset))
-			s += l
-			nextEmit = s
-			if nextS >= s {
-				s = nextS + 1
-			}
-
-			if s >= sLimit {
-				// Index first pair after match end.
-				if int(s+l+4) < len(src) {
-					cv := load3232(src, s)
-					e.table[hash4u(cv, bTableBits)] = tableEntry{offset: s + e.cur}
-				}
-				goto emitRemainder
-			}
-
-			// Store every second hash in-between, but offset by 1.
-			for i := s - l + 2; i < s-5; i += 7 {
-				x := load6432(src, i)
-				nextHash := hash4u(uint32(x), bTableBits)
-				e.table[nextHash] = tableEntry{offset: e.cur + i}
-				// Skip one
-				x >>= 16
-				nextHash = hash4u(uint32(x), bTableBits)
-				e.table[nextHash] = tableEntry{offset: e.cur + i + 2}
-				// Skip one
-				x >>= 16
-				nextHash = hash4u(uint32(x), bTableBits)
-				e.table[nextHash] = tableEntry{offset: e.cur + i + 4}
-			}
-
-			// We could immediately start working at s now, but to improve
-			// compression we first update the hash table at s-2 to s. If
-			// another emitCopy is not our next move, also calculate nextHash
-			// at s+1. At least on GOARCH=amd64, these three hash calculations
-			// are faster as one load64 call (with some shifts) instead of
-			// three load32 calls.
-			x := load6432(src, s-2)
-			o := e.cur + s - 2
-			prevHash := hash4u(uint32(x), bTableBits)
-			prevHash2 := hash4u(uint32(x>>8), bTableBits)
-			e.table[prevHash] = tableEntry{offset: o}
-			e.table[prevHash2] = tableEntry{offset: o + 1}
-			currHash := hash4u(uint32(x>>16), bTableBits)
-			candidate = e.table[currHash]
-			e.table[currHash] = tableEntry{offset: o + 2}
-
-			offset := s - (candidate.offset - e.cur)
-			if offset > maxMatchOffset || uint32(x>>16) != load3232(src, candidate.offset-e.cur) {
-				cv = uint32(x >> 24)
-				s++
-				break
-			}
-		}
-	}
-
-emitRemainder:
-	if int(nextEmit) < len(src) {
-		// If nothing was added, don't encode literals.
-		if dst.n == 0 {
-			return
-		}
-
-		emitLiteral(dst, src[nextEmit:])
-	}
-}
diff --git a/vendor/github.com/klauspost/compress/flate/level3.go b/vendor/github.com/klauspost/compress/flate/level3.go
deleted file mode 100644
index c22b4244a5..0000000000
--- a/vendor/github.com/klauspost/compress/flate/level3.go
+++ /dev/null
@@ -1,229 +0,0 @@
-package flate
-
-import "fmt"
-
-// fastEncL3
-type fastEncL3 struct {
-	fastGen
-	table [tableSize]tableEntryPrev
-}
-
-// Encode uses a similar algorithm to level 2, will check up to two candidates.
-func (e *fastEncL3) Encode(dst *tokens, src []byte) {
-	const (
-		inputMargin            = 8 - 1
-		minNonLiteralBlockSize = 1 + 1 + inputMargin
-	)
-
-	if debugDeflate && e.cur < 0 {
-		panic(fmt.Sprint("e.cur < 0: ", e.cur))
-	}
-
-	// Protect against e.cur wraparound.
-	for e.cur >= bufferReset {
-		if len(e.hist) == 0 {
-			for i := range e.table[:] {
-				e.table[i] = tableEntryPrev{}
-			}
-			e.cur = maxMatchOffset
-			break
-		}
-		// Shift down everything in the table that isn't already too far away.
-		minOff := e.cur + int32(len(e.hist)) - maxMatchOffset
-		for i := range e.table[:] {
-			v := e.table[i]
-			if v.Cur.offset <= minOff {
-				v.Cur.offset = 0
-			} else {
-				v.Cur.offset = v.Cur.offset - e.cur + maxMatchOffset
-			}
-			if v.Prev.offset <= minOff {
-				v.Prev.offset = 0
-			} else {
-				v.Prev.offset = v.Prev.offset - e.cur + maxMatchOffset
-			}
-			e.table[i] = v
-		}
-		e.cur = maxMatchOffset
-	}
-
-	s := e.addBlock(src)
-
-	// Skip if too small.
-	if len(src) < minNonLiteralBlockSize {
-		// We do not fill the token table.
-		// This will be picked up by caller.
-		dst.n = uint16(len(src))
-		return
-	}
-
-	// Override src
-	src = e.hist
-	nextEmit := s
-
-	// sLimit is when to stop looking for offset/length copies. The inputMargin
-	// lets us use a fast path for emitLiteral in the main loop, while we are
-	// looking for copies.
-	sLimit := int32(len(src) - inputMargin)
-
-	// nextEmit is where in src the next emitLiteral should start from.
-	cv := load3232(src, s)
-	for {
-		const skipLog = 6
-		nextS := s
-		var candidate tableEntry
-		for {
-			nextHash := hash(cv)
-			s = nextS
-			nextS = s + 1 + (s-nextEmit)>>skipLog
-			if nextS > sLimit {
-				goto emitRemainder
-			}
-			candidates := e.table[nextHash]
-			now := load3232(src, nextS)
-
-			// Safe offset distance until s + 4...
-			minOffset := e.cur + s - (maxMatchOffset - 4)
-			e.table[nextHash] = tableEntryPrev{Prev: candidates.Cur, Cur: tableEntry{offset: s + e.cur}}
-
-			// Check both candidates
-			candidate = candidates.Cur
-			if candidate.offset < minOffset {
-				cv = now
-				// Previous will also be invalid, we have nothing.
-				continue
-			}
-
-			if cv == load3232(src, candidate.offset-e.cur) {
-				if candidates.Prev.offset < minOffset || cv != load3232(src, candidates.Prev.offset-e.cur) {
-					break
-				}
-				// Both match and are valid, pick longest.
-				offset := s - (candidate.offset - e.cur)
-				o2 := s - (candidates.Prev.offset - e.cur)
-				l1, l2 := matchLen(src[s+4:], src[s-offset+4:]), matchLen(src[s+4:], src[s-o2+4:])
-				if l2 > l1 {
-					candidate = candidates.Prev
-				}
-				break
-			} else {
-				// We only check if value mismatches.
-				// Offset will always be invalid in other cases.
-				candidate = candidates.Prev
-				if candidate.offset > minOffset && cv == load3232(src, candidate.offset-e.cur) {
-					break
-				}
-			}
-			cv = now
-		}
-
-		// Call emitCopy, and then see if another emitCopy could be our next
-		// move. Repeat until we find no match for the input immediately after
-		// what was consumed by the last emitCopy call.
-		//
-		// If we exit this loop normally then we need to call emitLiteral next,
-		// though we don't yet know how big the literal will be. We handle that
-		// by proceeding to the next iteration of the main loop. We also can
-		// exit this loop via goto if we get close to exhausting the input.
-		for {
-			// Invariant: we have a 4-byte match at s, and no need to emit any
-			// literal bytes prior to s.
-
-			// Extend the 4-byte match as long as possible.
-			//
-			t := candidate.offset - e.cur
-			l := e.matchlenLong(s+4, t+4, src) + 4
-
-			// Extend backwards
-			for t > 0 && s > nextEmit && src[t-1] == src[s-1] {
-				s--
-				t--
-				l++
-			}
-			if nextEmit < s {
-				emitLiteral(dst, src[nextEmit:s])
-			}
-
-			dst.AddMatchLong(l, uint32(s-t-baseMatchOffset))
-			s += l
-			nextEmit = s
-			if nextS >= s {
-				s = nextS + 1
-			}
-
-			if s >= sLimit {
-				t += l
-				// Index first pair after match end.
-				if int(t+4) < len(src) && t > 0 {
-					cv := load3232(src, t)
-					nextHash := hash(cv)
-					e.table[nextHash] = tableEntryPrev{
-						Prev: e.table[nextHash].Cur,
-						Cur:  tableEntry{offset: e.cur + t},
-					}
-				}
-				goto emitRemainder
-			}
-
-			// We could immediately start working at s now, but to improve
-			// compression we first update the hash table at s-3 to s.
-			x := load6432(src, s-3)
-			prevHash := hash(uint32(x))
-			e.table[prevHash] = tableEntryPrev{
-				Prev: e.table[prevHash].Cur,
-				Cur:  tableEntry{offset: e.cur + s - 3},
-			}
-			x >>= 8
-			prevHash = hash(uint32(x))
-
-			e.table[prevHash] = tableEntryPrev{
-				Prev: e.table[prevHash].Cur,
-				Cur:  tableEntry{offset: e.cur + s - 2},
-			}
-			x >>= 8
-			prevHash = hash(uint32(x))
-
-			e.table[prevHash] = tableEntryPrev{
-				Prev: e.table[prevHash].Cur,
-				Cur:  tableEntry{offset: e.cur + s - 1},
-			}
-			x >>= 8
-			currHash := hash(uint32(x))
-			candidates := e.table[currHash]
-			cv = uint32(x)
-			e.table[currHash] = tableEntryPrev{
-				Prev: candidates.Cur,
-				Cur:  tableEntry{offset: s + e.cur},
-			}
-
-			// Check both candidates
-			candidate = candidates.Cur
-			minOffset := e.cur + s - (maxMatchOffset - 4)
-
-			if candidate.offset > minOffset && cv != load3232(src, candidate.offset-e.cur) {
-				// We only check if value mismatches.
-				// Offset will always be invalid in other cases.
-				candidate = candidates.Prev
-				if candidate.offset > minOffset && cv == load3232(src, candidate.offset-e.cur) {
-					offset := s - (candidate.offset - e.cur)
-					if offset <= maxMatchOffset {
-						continue
-					}
-				}
-			}
-			cv = uint32(x >> 8)
-			s++
-			break
-		}
-	}
-
-emitRemainder:
-	if int(nextEmit) < len(src) {
-		// If nothing was added, don't encode literals.
-		if dst.n == 0 {
-			return
-		}
-
-		emitLiteral(dst, src[nextEmit:])
-	}
-}
diff --git a/vendor/github.com/klauspost/compress/flate/level4.go b/vendor/github.com/klauspost/compress/flate/level4.go
deleted file mode 100644
index e62f0c02b1..0000000000
--- a/vendor/github.com/klauspost/compress/flate/level4.go
+++ /dev/null
@@ -1,212 +0,0 @@
-package flate
-
-import "fmt"
-
-type fastEncL4 struct {
-	fastGen
-	table  [tableSize]tableEntry
-	bTable [tableSize]tableEntry
-}
-
-func (e *fastEncL4) Encode(dst *tokens, src []byte) {
-	const (
-		inputMargin            = 12 - 1
-		minNonLiteralBlockSize = 1 + 1 + inputMargin
-	)
-	if debugDeflate && e.cur < 0 {
-		panic(fmt.Sprint("e.cur < 0: ", e.cur))
-	}
-	// Protect against e.cur wraparound.
-	for e.cur >= bufferReset {
-		if len(e.hist) == 0 {
-			for i := range e.table[:] {
-				e.table[i] = tableEntry{}
-			}
-			for i := range e.bTable[:] {
-				e.bTable[i] = tableEntry{}
-			}
-			e.cur = maxMatchOffset
-			break
-		}
-		// Shift down everything in the table that isn't already too far away.
-		minOff := e.cur + int32(len(e.hist)) - maxMatchOffset
-		for i := range e.table[:] {
-			v := e.table[i].offset
-			if v <= minOff {
-				v = 0
-			} else {
-				v = v - e.cur + maxMatchOffset
-			}
-			e.table[i].offset = v
-		}
-		for i := range e.bTable[:] {
-			v := e.bTable[i].offset
-			if v <= minOff {
-				v = 0
-			} else {
-				v = v - e.cur + maxMatchOffset
-			}
-			e.bTable[i].offset = v
-		}
-		e.cur = maxMatchOffset
-	}
-
-	s := e.addBlock(src)
-
-	// This check isn't in the Snappy implementation, but there, the caller
-	// instead of the callee handles this case.
-	if len(src) < minNonLiteralBlockSize {
-		// We do not fill the token table.
-		// This will be picked up by caller.
-		dst.n = uint16(len(src))
-		return
-	}
-
-	// Override src
-	src = e.hist
-	nextEmit := s
-
-	// sLimit is when to stop looking for offset/length copies. The inputMargin
-	// lets us use a fast path for emitLiteral in the main loop, while we are
-	// looking for copies.
-	sLimit := int32(len(src) - inputMargin)
-
-	// nextEmit is where in src the next emitLiteral should start from.
-	cv := load6432(src, s)
-	for {
-		const skipLog = 6
-		const doEvery = 1
-
-		nextS := s
-		var t int32
-		for {
-			nextHashS := hash4x64(cv, tableBits)
-			nextHashL := hash7(cv, tableBits)
-
-			s = nextS
-			nextS = s + doEvery + (s-nextEmit)>>skipLog
-			if nextS > sLimit {
-				goto emitRemainder
-			}
-			// Fetch a short+long candidate
-			sCandidate := e.table[nextHashS]
-			lCandidate := e.bTable[nextHashL]
-			next := load6432(src, nextS)
-			entry := tableEntry{offset: s + e.cur}
-			e.table[nextHashS] = entry
-			e.bTable[nextHashL] = entry
-
-			t = lCandidate.offset - e.cur
-			if s-t < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.offset-e.cur) {
-				// We got a long match. Use that.
-				break
-			}
-
-			t = sCandidate.offset - e.cur
-			if s-t < maxMatchOffset && uint32(cv) == load3232(src, sCandidate.offset-e.cur) {
-				// Found a 4 match...
-				lCandidate = e.bTable[hash7(next, tableBits)]
-
-				// If the next long is a candidate, check if we should use that instead...
-				lOff := nextS - (lCandidate.offset - e.cur)
-				if lOff < maxMatchOffset && load3232(src, lCandidate.offset-e.cur) == uint32(next) {
-					l1, l2 := matchLen(src[s+4:], src[t+4:]), matchLen(src[nextS+4:], src[nextS-lOff+4:])
-					if l2 > l1 {
-						s = nextS
-						t = lCandidate.offset - e.cur
-					}
-				}
-				break
-			}
-			cv = next
-		}
-
-		// A 4-byte match has been found. We'll later see if more than 4 bytes
-		// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
-		// them as literal bytes.
-
-		// Extend the 4-byte match as long as possible.
-		l := e.matchlenLong(s+4, t+4, src) + 4
-
-		// Extend backwards
-		for t > 0 && s > nextEmit && src[t-1] == src[s-1] {
-			s--
-			t--
-			l++
-		}
-		if nextEmit < s {
-			emitLiteral(dst, src[nextEmit:s])
-		}
-		if debugDeflate {
-			if t >= s {
-				panic("s-t")
-			}
-			if (s - t) > maxMatchOffset {
-				panic(fmt.Sprintln("mmo", t))
-			}
-			if l < baseMatchLength {
-				panic("bml")
-			}
-		}
-
-		dst.AddMatchLong(l, uint32(s-t-baseMatchOffset))
-		s += l
-		nextEmit = s
-		if nextS >= s {
-			s = nextS + 1
-		}
-
-		if s >= sLimit {
-			// Index first pair after match end.
-			if int(s+8) < len(src) {
-				cv := load6432(src, s)
-				e.table[hash4x64(cv, tableBits)] = tableEntry{offset: s + e.cur}
-				e.bTable[hash7(cv, tableBits)] = tableEntry{offset: s + e.cur}
-			}
-			goto emitRemainder
-		}
-
-		// Store every 3rd hash in-between
-		if true {
-			i := nextS
-			if i < s-1 {
-				cv := load6432(src, i)
-				t := tableEntry{offset: i + e.cur}
-				t2 := tableEntry{offset: t.offset + 1}
-				e.bTable[hash7(cv, tableBits)] = t
-				e.bTable[hash7(cv>>8, tableBits)] = t2
-				e.table[hash4u(uint32(cv>>8), tableBits)] = t2
-
-				i += 3
-				for ; i < s-1; i += 3 {
-					cv := load6432(src, i)
-					t := tableEntry{offset: i + e.cur}
-					t2 := tableEntry{offset: t.offset + 1}
-					e.bTable[hash7(cv, tableBits)] = t
-					e.bTable[hash7(cv>>8, tableBits)] = t2
-					e.table[hash4u(uint32(cv>>8), tableBits)] = t2
-				}
-			}
-		}
-
-		// We could immediately start working at s now, but to improve
-		// compression we first update the hash table at s-1 and at s.
-		x := load6432(src, s-1)
-		o := e.cur + s - 1
-		prevHashS := hash4x64(x, tableBits)
-		prevHashL := hash7(x, tableBits)
-		e.table[prevHashS] = tableEntry{offset: o}
-		e.bTable[prevHashL] = tableEntry{offset: o}
-		cv = x >> 8
-	}
-
-emitRemainder:
-	if int(nextEmit) < len(src) {
-		// If nothing was added, don't encode literals.
-		if dst.n == 0 {
-			return
-		}
-
-		emitLiteral(dst, src[nextEmit:])
-	}
-}
diff --git a/vendor/github.com/klauspost/compress/flate/level5.go b/vendor/github.com/klauspost/compress/flate/level5.go
deleted file mode 100644
index 293a3a320b..0000000000
--- a/vendor/github.com/klauspost/compress/flate/level5.go
+++ /dev/null
@@ -1,294 +0,0 @@
-package flate
-
-import "fmt"
-
-type fastEncL5 struct {
-	fastGen
-	table  [tableSize]tableEntry
-	bTable [tableSize]tableEntryPrev
-}
-
-func (e *fastEncL5) Encode(dst *tokens, src []byte) {
-	const (
-		inputMargin            = 12 - 1
-		minNonLiteralBlockSize = 1 + 1 + inputMargin
-	)
-	if debugDeflate && e.cur < 0 {
-		panic(fmt.Sprint("e.cur < 0: ", e.cur))
-	}
-
-	// Protect against e.cur wraparound.
-	for e.cur >= bufferReset {
-		if len(e.hist) == 0 {
-			for i := range e.table[:] {
-				e.table[i] = tableEntry{}
-			}
-			for i := range e.bTable[:] {
-				e.bTable[i] = tableEntryPrev{}
-			}
-			e.cur = maxMatchOffset
-			break
-		}
-		// Shift down everything in the table that isn't already too far away.
-		minOff := e.cur + int32(len(e.hist)) - maxMatchOffset
-		for i := range e.table[:] {
-			v := e.table[i].offset
-			if v <= minOff {
-				v = 0
-			} else {
-				v = v - e.cur + maxMatchOffset
-			}
-			e.table[i].offset = v
-		}
-		for i := range e.bTable[:] {
-			v := e.bTable[i]
-			if v.Cur.offset <= minOff {
-				v.Cur.offset = 0
-				v.Prev.offset = 0
-			} else {
-				v.Cur.offset = v.Cur.offset - e.cur + maxMatchOffset
-				if v.Prev.offset <= minOff {
-					v.Prev.offset = 0
-				} else {
-					v.Prev.offset = v.Prev.offset - e.cur + maxMatchOffset
-				}
-			}
-			e.bTable[i] = v
-		}
-		e.cur = maxMatchOffset
-	}
-
-	s := e.addBlock(src)
-
-	// This check isn't in the Snappy implementation, but there, the caller
-	// instead of the callee handles this case.
-	if len(src) < minNonLiteralBlockSize {
-		// We do not fill the token table.
-		// This will be picked up by caller.
-		dst.n = uint16(len(src))
-		return
-	}
-
-	// Override src
-	src = e.hist
-	nextEmit := s
-
-	// sLimit is when to stop looking for offset/length copies. The inputMargin
-	// lets us use a fast path for emitLiteral in the main loop, while we are
-	// looking for copies.
-	sLimit := int32(len(src) - inputMargin)
-
-	// nextEmit is where in src the next emitLiteral should start from.
-	cv := load6432(src, s)
-	for {
-		const skipLog = 6
-		const doEvery = 1
-
-		nextS := s
-		var l int32
-		var t int32
-		for {
-			nextHashS := hash4x64(cv, tableBits)
-			nextHashL := hash7(cv, tableBits)
-
-			s = nextS
-			nextS = s + doEvery + (s-nextEmit)>>skipLog
-			if nextS > sLimit {
-				goto emitRemainder
-			}
-			// Fetch a short+long candidate
-			sCandidate := e.table[nextHashS]
-			lCandidate := e.bTable[nextHashL]
-			next := load6432(src, nextS)
-			entry := tableEntry{offset: s + e.cur}
-			e.table[nextHashS] = entry
-			eLong := &e.bTable[nextHashL]
-			eLong.Cur, eLong.Prev = entry, eLong.Cur
-
-			nextHashS = hash4x64(next, tableBits)
-			nextHashL = hash7(next, tableBits)
-
-			t = lCandidate.Cur.offset - e.cur
-			if s-t < maxMatchOffset {
-				if uint32(cv) == load3232(src, lCandidate.Cur.offset-e.cur) {
-					// Store the next match
-					e.table[nextHashS] = tableEntry{offset: nextS + e.cur}
-					eLong := &e.bTable[nextHashL]
-					eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur
-
-					t2 := lCandidate.Prev.offset - e.cur
-					if s-t2 < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.Prev.offset-e.cur) {
-						l = e.matchlen(s+4, t+4, src) + 4
-						ml1 := e.matchlen(s+4, t2+4, src) + 4
-						if ml1 > l {
-							t = t2
-							l = ml1
-							break
-						}
-					}
-					break
-				}
-				t = lCandidate.Prev.offset - e.cur
-				if s-t < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.Prev.offset-e.cur) {
-					// Store the next match
-					e.table[nextHashS] = tableEntry{offset: nextS + e.cur}
-					eLong := &e.bTable[nextHashL]
-					eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur
-					break
-				}
-			}
-
-			t = sCandidate.offset - e.cur
-			if s-t < maxMatchOffset && uint32(cv) == load3232(src, sCandidate.offset-e.cur) {
-				// Found a 4 match...
-				l = e.matchlen(s+4, t+4, src) + 4
-				lCandidate = e.bTable[nextHashL]
-				// Store the next match
-
-				e.table[nextHashS] = tableEntry{offset: nextS + e.cur}
-				eLong := &e.bTable[nextHashL]
-				eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur
-
-				// If the next long is a candidate, use that...
-				t2 := lCandidate.Cur.offset - e.cur
-				if nextS-t2 < maxMatchOffset {
-					if load3232(src, lCandidate.Cur.offset-e.cur) == uint32(next) {
-						ml := e.matchlen(nextS+4, t2+4, src) + 4
-						if ml > l {
-							t = t2
-							s = nextS
-							l = ml
-							break
-						}
-					}
-					// If the previous long is a candidate, use that...
-					t2 = lCandidate.Prev.offset - e.cur
-					if nextS-t2 < maxMatchOffset && load3232(src, lCandidate.Prev.offset-e.cur) == uint32(next) {
-						ml := e.matchlen(nextS+4, t2+4, src) + 4
-						if ml > l {
-							t = t2
-							s = nextS
-							l = ml
-							break
-						}
-					}
-				}
-				break
-			}
-			cv = next
-		}
-
-		// A 4-byte match has been found. We'll later see if more than 4 bytes
-		// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
-		// them as literal bytes.
-
-		if l == 0 {
-			// Extend the 4-byte match as long as possible.
-			l = e.matchlenLong(s+4, t+4, src) + 4
-		} else if l == maxMatchLength {
-			l += e.matchlenLong(s+l, t+l, src)
-		}
-
-		// Try to locate a better match by checking the end of best match...
-		if sAt := s + l; l < 30 && sAt < sLimit {
-			eLong := e.bTable[hash7(load6432(src, sAt), tableBits)].Cur.offset
-			// Test current
-			t2 := eLong - e.cur - l
-			off := s - t2
-			if t2 >= 0 && off < maxMatchOffset && off > 0 {
-				if l2 := e.matchlenLong(s, t2, src); l2 > l {
-					t = t2
-					l = l2
-				}
-			}
-		}
-
-		// Extend backwards
-		for t > 0 && s > nextEmit && src[t-1] == src[s-1] {
-			s--
-			t--
-			l++
-		}
-		if nextEmit < s {
-			emitLiteral(dst, src[nextEmit:s])
-		}
-		if debugDeflate {
-			if t >= s {
-				panic(fmt.Sprintln("s-t", s, t))
-			}
-			if (s - t) > maxMatchOffset {
-				panic(fmt.Sprintln("mmo", s-t))
-			}
-			if l < baseMatchLength {
-				panic("bml")
-			}
-		}
-
-		dst.AddMatchLong(l, uint32(s-t-baseMatchOffset))
-		s += l
-		nextEmit = s
-		if nextS >= s {
-			s = nextS + 1
-		}
-
-		if s >= sLimit {
-			goto emitRemainder
-		}
-
-		// Store every 3rd hash in-between.
-		if true {
-			const hashEvery = 3
-			i := s - l + 1
-			if i < s-1 {
-				cv := load6432(src, i)
-				t := tableEntry{offset: i + e.cur}
-				e.table[hash4x64(cv, tableBits)] = t
-				eLong := &e.bTable[hash7(cv, tableBits)]
-				eLong.Cur, eLong.Prev = t, eLong.Cur
-
-				// Do an long at i+1
-				cv >>= 8
-				t = tableEntry{offset: t.offset + 1}
-				eLong = &e.bTable[hash7(cv, tableBits)]
-				eLong.Cur, eLong.Prev = t, eLong.Cur
-
-				// We only have enough bits for a short entry at i+2
-				cv >>= 8
-				t = tableEntry{offset: t.offset + 1}
-				e.table[hash4x64(cv, tableBits)] = t
-
-				// Skip one - otherwise we risk hitting 's'
-				i += 4
-				for ; i < s-1; i += hashEvery {
-					cv := load6432(src, i)
-					t := tableEntry{offset: i + e.cur}
-					t2 := tableEntry{offset: t.offset + 1}
-					eLong := &e.bTable[hash7(cv, tableBits)]
-					eLong.Cur, eLong.Prev = t, eLong.Cur
-					e.table[hash4u(uint32(cv>>8), tableBits)] = t2
-				}
-			}
-		}
-
-		// We could immediately start working at s now, but to improve
-		// compression we first update the hash table at s-1 and at s.
-		x := load6432(src, s-1)
-		o := e.cur + s - 1
-		prevHashS := hash4x64(x, tableBits)
-		prevHashL := hash7(x, tableBits)
-		e.table[prevHashS] = tableEntry{offset: o}
-		eLong := &e.bTable[prevHashL]
-		eLong.Cur, eLong.Prev = tableEntry{offset: o}, eLong.Cur
-		cv = x >> 8
-	}
-
-emitRemainder:
-	if int(nextEmit) < len(src) {
-		// If nothing was added, don't encode literals.
-		if dst.n == 0 {
-			return
-		}
-
-		emitLiteral(dst, src[nextEmit:])
-	}
-}
diff --git a/vendor/github.com/klauspost/compress/flate/level6.go b/vendor/github.com/klauspost/compress/flate/level6.go
deleted file mode 100644
index a709977ec4..0000000000
--- a/vendor/github.com/klauspost/compress/flate/level6.go
+++ /dev/null
@@ -1,307 +0,0 @@
-package flate
-
-import "fmt"
-
-type fastEncL6 struct {
-	fastGen
-	table  [tableSize]tableEntry
-	bTable [tableSize]tableEntryPrev
-}
-
-func (e *fastEncL6) Encode(dst *tokens, src []byte) {
-	const (
-		inputMargin            = 12 - 1
-		minNonLiteralBlockSize = 1 + 1 + inputMargin
-	)
-	if debugDeflate && e.cur < 0 {
-		panic(fmt.Sprint("e.cur < 0: ", e.cur))
-	}
-
-	// Protect against e.cur wraparound.
-	for e.cur >= bufferReset {
-		if len(e.hist) == 0 {
-			for i := range e.table[:] {
-				e.table[i] = tableEntry{}
-			}
-			for i := range e.bTable[:] {
-				e.bTable[i] = tableEntryPrev{}
-			}
-			e.cur = maxMatchOffset
-			break
-		}
-		// Shift down everything in the table that isn't already too far away.
-		minOff := e.cur + int32(len(e.hist)) - maxMatchOffset
-		for i := range e.table[:] {
-			v := e.table[i].offset
-			if v <= minOff {
-				v = 0
-			} else {
-				v = v - e.cur + maxMatchOffset
-			}
-			e.table[i].offset = v
-		}
-		for i := range e.bTable[:] {
-			v := e.bTable[i]
-			if v.Cur.offset <= minOff {
-				v.Cur.offset = 0
-				v.Prev.offset = 0
-			} else {
-				v.Cur.offset = v.Cur.offset - e.cur + maxMatchOffset
-				if v.Prev.offset <= minOff {
-					v.Prev.offset = 0
-				} else {
-					v.Prev.offset = v.Prev.offset - e.cur + maxMatchOffset
-				}
-			}
-			e.bTable[i] = v
-		}
-		e.cur = maxMatchOffset
-	}
-
-	s := e.addBlock(src)
-
-	// This check isn't in the Snappy implementation, but there, the caller
-	// instead of the callee handles this case.
-	if len(src) < minNonLiteralBlockSize {
-		// We do not fill the token table.
-		// This will be picked up by caller.
-		dst.n = uint16(len(src))
-		return
-	}
-
-	// Override src
-	src = e.hist
-	nextEmit := s
-
-	// sLimit is when to stop looking for offset/length copies. The inputMargin
-	// lets us use a fast path for emitLiteral in the main loop, while we are
-	// looking for copies.
-	sLimit := int32(len(src) - inputMargin)
-
-	// nextEmit is where in src the next emitLiteral should start from.
-	cv := load6432(src, s)
-	// Repeat MUST be > 1 and within range
-	repeat := int32(1)
-	for {
-		const skipLog = 7
-		const doEvery = 1
-
-		nextS := s
-		var l int32
-		var t int32
-		for {
-			nextHashS := hash4x64(cv, tableBits)
-			nextHashL := hash7(cv, tableBits)
-			s = nextS
-			nextS = s + doEvery + (s-nextEmit)>>skipLog
-			if nextS > sLimit {
-				goto emitRemainder
-			}
-			// Fetch a short+long candidate
-			sCandidate := e.table[nextHashS]
-			lCandidate := e.bTable[nextHashL]
-			next := load6432(src, nextS)
-			entry := tableEntry{offset: s + e.cur}
-			e.table[nextHashS] = entry
-			eLong := &e.bTable[nextHashL]
-			eLong.Cur, eLong.Prev = entry, eLong.Cur
-
-			// Calculate hashes of 'next'
-			nextHashS = hash4x64(next, tableBits)
-			nextHashL = hash7(next, tableBits)
-
-			t = lCandidate.Cur.offset - e.cur
-			if s-t < maxMatchOffset {
-				if uint32(cv) == load3232(src, lCandidate.Cur.offset-e.cur) {
-					// Long candidate matches at least 4 bytes.
-
-					// Store the next match
-					e.table[nextHashS] = tableEntry{offset: nextS + e.cur}
-					eLong := &e.bTable[nextHashL]
-					eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur
-
-					// Check the previous long candidate as well.
-					t2 := lCandidate.Prev.offset - e.cur
-					if s-t2 < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.Prev.offset-e.cur) {
-						l = e.matchlen(s+4, t+4, src) + 4
-						ml1 := e.matchlen(s+4, t2+4, src) + 4
-						if ml1 > l {
-							t = t2
-							l = ml1
-							break
-						}
-					}
-					break
-				}
-				// Current value did not match, but check if previous long value does.
-				t = lCandidate.Prev.offset - e.cur
-				if s-t < maxMatchOffset && uint32(cv) == load3232(src, lCandidate.Prev.offset-e.cur) {
-					// Store the next match
-					e.table[nextHashS] = tableEntry{offset: nextS + e.cur}
-					eLong := &e.bTable[nextHashL]
-					eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur
-					break
-				}
-			}
-
-			t = sCandidate.offset - e.cur
-			if s-t < maxMatchOffset && uint32(cv) == load3232(src, sCandidate.offset-e.cur) {
-				// Found a 4 match...
-				l = e.matchlen(s+4, t+4, src) + 4
-
-				// Look up next long candidate (at nextS)
-				lCandidate = e.bTable[nextHashL]
-
-				// Store the next match
-				e.table[nextHashS] = tableEntry{offset: nextS + e.cur}
-				eLong := &e.bTable[nextHashL]
-				eLong.Cur, eLong.Prev = tableEntry{offset: nextS + e.cur}, eLong.Cur
-
-				// Check repeat at s + repOff
-				const repOff = 1
-				t2 := s - repeat + repOff
-				if load3232(src, t2) == uint32(cv>>(8*repOff)) {
-					ml := e.matchlen(s+4+repOff, t2+4, src) + 4
-					if ml > l {
-						t = t2
-						l = ml
-						s += repOff
-						// Not worth checking more.
-						break
-					}
-				}
-
-				// If the next long is a candidate, use that...
-				t2 = lCandidate.Cur.offset - e.cur
-				if nextS-t2 < maxMatchOffset {
-					if load3232(src, lCandidate.Cur.offset-e.cur) == uint32(next) {
-						ml := e.matchlen(nextS+4, t2+4, src) + 4
-						if ml > l {
-							t = t2
-							s = nextS
-							l = ml
-							// This is ok, but check previous as well.
-						}
-					}
-					// If the previous long is a candidate, use that...
-					t2 = lCandidate.Prev.offset - e.cur
-					if nextS-t2 < maxMatchOffset && load3232(src, lCandidate.Prev.offset-e.cur) == uint32(next) {
-						ml := e.matchlen(nextS+4, t2+4, src) + 4
-						if ml > l {
-							t = t2
-							s = nextS
-							l = ml
-							break
-						}
-					}
-				}
-				break
-			}
-			cv = next
-		}
-
-		// A 4-byte match has been found. We'll later see if more than 4 bytes
-		// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
-		// them as literal bytes.
-
-		// Extend the 4-byte match as long as possible.
-		if l == 0 {
-			l = e.matchlenLong(s+4, t+4, src) + 4
-		} else if l == maxMatchLength {
-			l += e.matchlenLong(s+l, t+l, src)
-		}
-
-		// Try to locate a better match by checking the end-of-match...
-		if sAt := s + l; sAt < sLimit {
-			eLong := &e.bTable[hash7(load6432(src, sAt), tableBits)]
-			// Test current
-			t2 := eLong.Cur.offset - e.cur - l
-			off := s - t2
-			if off < maxMatchOffset {
-				if off > 0 && t2 >= 0 {
-					if l2 := e.matchlenLong(s, t2, src); l2 > l {
-						t = t2
-						l = l2
-					}
-				}
-				// Test next:
-				t2 = eLong.Prev.offset - e.cur - l
-				off := s - t2
-				if off > 0 && off < maxMatchOffset && t2 >= 0 {
-					if l2 := e.matchlenLong(s, t2, src); l2 > l {
-						t = t2
-						l = l2
-					}
-				}
-			}
-		}
-
-		// Extend backwards
-		for t > 0 && s > nextEmit && src[t-1] == src[s-1] {
-			s--
-			t--
-			l++
-		}
-		if nextEmit < s {
-			emitLiteral(dst, src[nextEmit:s])
-		}
-		if false {
-			if t >= s {
-				panic(fmt.Sprintln("s-t", s, t))
-			}
-			if (s - t) > maxMatchOffset {
-				panic(fmt.Sprintln("mmo", s-t))
-			}
-			if l < baseMatchLength {
-				panic("bml")
-			}
-		}
-
-		dst.AddMatchLong(l, uint32(s-t-baseMatchOffset))
-		repeat = s - t
-		s += l
-		nextEmit = s
-		if nextS >= s {
-			s = nextS + 1
-		}
-
-		if s >= sLimit {
-			// Index after match end.
-			for i := nextS + 1; i < int32(len(src))-8; i += 2 {
-				cv := load6432(src, i)
-				e.table[hash4x64(cv, tableBits)] = tableEntry{offset: i + e.cur}
-				eLong := &e.bTable[hash7(cv, tableBits)]
-				eLong.Cur, eLong.Prev = tableEntry{offset: i + e.cur}, eLong.Cur
-			}
-			goto emitRemainder
-		}
-
-		// Store every long hash in-between and every second short.
-		if true {
-			for i := nextS + 1; i < s-1; i += 2 {
-				cv := load6432(src, i)
-				t := tableEntry{offset: i + e.cur}
-				t2 := tableEntry{offset: t.offset + 1}
-				eLong := &e.bTable[hash7(cv, tableBits)]
-				eLong2 := &e.bTable[hash7(cv>>8, tableBits)]
-				e.table[hash4x64(cv, tableBits)] = t
-				eLong.Cur, eLong.Prev = t, eLong.Cur
-				eLong2.Cur, eLong2.Prev = t2, eLong2.Cur
-			}
-		}
-
-		// We could immediately start working at s now, but to improve
-		// compression we first update the hash table at s-1 and at s.
-		cv = load6432(src, s)
-	}
-
-emitRemainder:
-	if int(nextEmit) < len(src) {
-		// If nothing was added, don't encode literals.
-		if dst.n == 0 {
-			return
-		}
-
-		emitLiteral(dst, src[nextEmit:])
-	}
-}
diff --git a/vendor/github.com/klauspost/compress/flate/regmask_amd64.go b/vendor/github.com/klauspost/compress/flate/regmask_amd64.go
deleted file mode 100644
index 6ed28061b2..0000000000
--- a/vendor/github.com/klauspost/compress/flate/regmask_amd64.go
+++ /dev/null
@@ -1,37 +0,0 @@
-package flate
-
-const (
-	// Masks for shifts with register sizes of the shift value.
-	// This can be used to work around the x86 design of shifting by mod register size.
-	// It can be used when a variable shift is always smaller than the register size.
-
-	// reg8SizeMaskX - shift value is 8 bits, shifted is X
-	reg8SizeMask8  = 7
-	reg8SizeMask16 = 15
-	reg8SizeMask32 = 31
-	reg8SizeMask64 = 63
-
-	// reg16SizeMaskX - shift value is 16 bits, shifted is X
-	reg16SizeMask8  = reg8SizeMask8
-	reg16SizeMask16 = reg8SizeMask16
-	reg16SizeMask32 = reg8SizeMask32
-	reg16SizeMask64 = reg8SizeMask64
-
-	// reg32SizeMaskX - shift value is 32 bits, shifted is X
-	reg32SizeMask8  = reg8SizeMask8
-	reg32SizeMask16 = reg8SizeMask16
-	reg32SizeMask32 = reg8SizeMask32
-	reg32SizeMask64 = reg8SizeMask64
-
-	// reg64SizeMaskX - shift value is 64 bits, shifted is X
-	reg64SizeMask8  = reg8SizeMask8
-	reg64SizeMask16 = reg8SizeMask16
-	reg64SizeMask32 = reg8SizeMask32
-	reg64SizeMask64 = reg8SizeMask64
-
-	// regSizeMaskUintX - shift value is uint, shifted is X
-	regSizeMaskUint8  = reg8SizeMask8
-	regSizeMaskUint16 = reg8SizeMask16
-	regSizeMaskUint32 = reg8SizeMask32
-	regSizeMaskUint64 = reg8SizeMask64
-)
diff --git a/vendor/github.com/klauspost/compress/flate/regmask_other.go b/vendor/github.com/klauspost/compress/flate/regmask_other.go
deleted file mode 100644
index f477a5d6e5..0000000000
--- a/vendor/github.com/klauspost/compress/flate/regmask_other.go
+++ /dev/null
@@ -1,39 +0,0 @@
-//+build !amd64
-
-package flate
-
-const (
-	// Masks for shifts with register sizes of the shift value.
-	// This can be used to work around the x86 design of shifting by mod register size.
-	// It can be used when a variable shift is always smaller than the register size.
-
-	// reg8SizeMaskX - shift value is 8 bits, shifted is X
-	reg8SizeMask8  = 0xff
-	reg8SizeMask16 = 0xff
-	reg8SizeMask32 = 0xff
-	reg8SizeMask64 = 0xff
-
-	// reg16SizeMaskX - shift value is 16 bits, shifted is X
-	reg16SizeMask8  = 0xffff
-	reg16SizeMask16 = 0xffff
-	reg16SizeMask32 = 0xffff
-	reg16SizeMask64 = 0xffff
-
-	// reg32SizeMaskX - shift value is 32 bits, shifted is X
-	reg32SizeMask8  = 0xffffffff
-	reg32SizeMask16 = 0xffffffff
-	reg32SizeMask32 = 0xffffffff
-	reg32SizeMask64 = 0xffffffff
-
-	// reg64SizeMaskX - shift value is 64 bits, shifted is X
-	reg64SizeMask8  = 0xffffffffffffffff
-	reg64SizeMask16 = 0xffffffffffffffff
-	reg64SizeMask32 = 0xffffffffffffffff
-	reg64SizeMask64 = 0xffffffffffffffff
-
-	// regSizeMaskUintX - shift value is uint, shifted is X
-	regSizeMaskUint8  = ^uint(0)
-	regSizeMaskUint16 = ^uint(0)
-	regSizeMaskUint32 = ^uint(0)
-	regSizeMaskUint64 = ^uint(0)
-)
diff --git a/vendor/github.com/klauspost/compress/flate/stateless.go b/vendor/github.com/klauspost/compress/flate/stateless.go
deleted file mode 100644
index 53e8991246..0000000000
--- a/vendor/github.com/klauspost/compress/flate/stateless.go
+++ /dev/null
@@ -1,297 +0,0 @@
-package flate
-
-import (
-	"io"
-	"math"
-	"sync"
-)
-
-const (
-	maxStatelessBlock = math.MaxInt16
-	// dictionary will be taken from maxStatelessBlock, so limit it.
-	maxStatelessDict = 8 << 10
-
-	slTableBits  = 13
-	slTableSize  = 1 << slTableBits
-	slTableShift = 32 - slTableBits
-)
-
-type statelessWriter struct {
-	dst    io.Writer
-	closed bool
-}
-
-func (s *statelessWriter) Close() error {
-	if s.closed {
-		return nil
-	}
-	s.closed = true
-	// Emit EOF block
-	return StatelessDeflate(s.dst, nil, true, nil)
-}
-
-func (s *statelessWriter) Write(p []byte) (n int, err error) {
-	err = StatelessDeflate(s.dst, p, false, nil)
-	if err != nil {
-		return 0, err
-	}
-	return len(p), nil
-}
-
-func (s *statelessWriter) Reset(w io.Writer) {
-	s.dst = w
-	s.closed = false
-}
-
-// NewStatelessWriter will do compression but without maintaining any state
-// between Write calls.
-// There will be no memory kept between Write calls,
-// but compression and speed will be suboptimal.
-// Because of this, the size of actual Write calls will affect output size.
-func NewStatelessWriter(dst io.Writer) io.WriteCloser {
-	return &statelessWriter{dst: dst}
-}
-
-// bitWriterPool contains bit writers that can be reused.
-var bitWriterPool = sync.Pool{
-	New: func() interface{} {
-		return newHuffmanBitWriter(nil)
-	},
-}
-
-// StatelessDeflate allows to compress directly to a Writer without retaining state.
-// When returning everything will be flushed.
-// Up to 8KB of an optional dictionary can be given which is presumed to presumed to precede the block.
-// Longer dictionaries will be truncated and will still produce valid output.
-// Sending nil dictionary is perfectly fine.
-func StatelessDeflate(out io.Writer, in []byte, eof bool, dict []byte) error {
-	var dst tokens
-	bw := bitWriterPool.Get().(*huffmanBitWriter)
-	bw.reset(out)
-	defer func() {
-		// don't keep a reference to our output
-		bw.reset(nil)
-		bitWriterPool.Put(bw)
-	}()
-	if eof && len(in) == 0 {
-		// Just write an EOF block.
-		// Could be faster...
-		bw.writeStoredHeader(0, true)
-		bw.flush()
-		return bw.err
-	}
-
-	// Truncate dict
-	if len(dict) > maxStatelessDict {
-		dict = dict[len(dict)-maxStatelessDict:]
-	}
-
-	for len(in) > 0 {
-		todo := in
-		if len(todo) > maxStatelessBlock-len(dict) {
-			todo = todo[:maxStatelessBlock-len(dict)]
-		}
-		in = in[len(todo):]
-		uncompressed := todo
-		if len(dict) > 0 {
-			// combine dict and source
-			bufLen := len(todo) + len(dict)
-			combined := make([]byte, bufLen)
-			copy(combined, dict)
-			copy(combined[len(dict):], todo)
-			todo = combined
-		}
-		// Compress
-		statelessEnc(&dst, todo, int16(len(dict)))
-		isEof := eof && len(in) == 0
-
-		if dst.n == 0 {
-			bw.writeStoredHeader(len(uncompressed), isEof)
-			if bw.err != nil {
-				return bw.err
-			}
-			bw.writeBytes(uncompressed)
-		} else if int(dst.n) > len(uncompressed)-len(uncompressed)>>4 {
-			// If we removed less than 1/16th, huffman compress the block.
-			bw.writeBlockHuff(isEof, uncompressed, len(in) == 0)
-		} else {
-			bw.writeBlockDynamic(&dst, isEof, uncompressed, len(in) == 0)
-		}
-		if len(in) > 0 {
-			// Retain a dict if we have more
-			dict = todo[len(todo)-maxStatelessDict:]
-			dst.Reset()
-		}
-		if bw.err != nil {
-			return bw.err
-		}
-	}
-	if !eof {
-		// Align, only a stored block can do that.
-		bw.writeStoredHeader(0, false)
-	}
-	bw.flush()
-	return bw.err
-}
-
-func hashSL(u uint32) uint32 {
-	return (u * 0x1e35a7bd) >> slTableShift
-}
-
-func load3216(b []byte, i int16) uint32 {
-	// Help the compiler eliminate bounds checks on the read so it can be done in a single read.
-	b = b[i:]
-	b = b[:4]
-	return uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24
-}
-
-func load6416(b []byte, i int16) uint64 {
-	// Help the compiler eliminate bounds checks on the read so it can be done in a single read.
-	b = b[i:]
-	b = b[:8]
-	return uint64(b[0]) | uint64(b[1])<<8 | uint64(b[2])<<16 | uint64(b[3])<<24 |
-		uint64(b[4])<<32 | uint64(b[5])<<40 | uint64(b[6])<<48 | uint64(b[7])<<56
-}
-
-func statelessEnc(dst *tokens, src []byte, startAt int16) {
-	const (
-		inputMargin            = 12 - 1
-		minNonLiteralBlockSize = 1 + 1 + inputMargin
-	)
-
-	type tableEntry struct {
-		offset int16
-	}
-
-	var table [slTableSize]tableEntry
-
-	// This check isn't in the Snappy implementation, but there, the caller
-	// instead of the callee handles this case.
-	if len(src)-int(startAt) < minNonLiteralBlockSize {
-		// We do not fill the token table.
-		// This will be picked up by caller.
-		dst.n = 0
-		return
-	}
-	// Index until startAt
-	if startAt > 0 {
-		cv := load3232(src, 0)
-		for i := int16(0); i < startAt; i++ {
-			table[hashSL(cv)] = tableEntry{offset: i}
-			cv = (cv >> 8) | (uint32(src[i+4]) << 24)
-		}
-	}
-
-	s := startAt + 1
-	nextEmit := startAt
-	// sLimit is when to stop looking for offset/length copies. The inputMargin
-	// lets us use a fast path for emitLiteral in the main loop, while we are
-	// looking for copies.
-	sLimit := int16(len(src) - inputMargin)
-
-	// nextEmit is where in src the next emitLiteral should start from.
-	cv := load3216(src, s)
-
-	for {
-		const skipLog = 5
-		const doEvery = 2
-
-		nextS := s
-		var candidate tableEntry
-		for {
-			nextHash := hashSL(cv)
-			candidate = table[nextHash]
-			nextS = s + doEvery + (s-nextEmit)>>skipLog
-			if nextS > sLimit || nextS <= 0 {
-				goto emitRemainder
-			}
-
-			now := load6416(src, nextS)
-			table[nextHash] = tableEntry{offset: s}
-			nextHash = hashSL(uint32(now))
-
-			if cv == load3216(src, candidate.offset) {
-				table[nextHash] = tableEntry{offset: nextS}
-				break
-			}
-
-			// Do one right away...
-			cv = uint32(now)
-			s = nextS
-			nextS++
-			candidate = table[nextHash]
-			now >>= 8
-			table[nextHash] = tableEntry{offset: s}
-
-			if cv == load3216(src, candidate.offset) {
-				table[nextHash] = tableEntry{offset: nextS}
-				break
-			}
-			cv = uint32(now)
-			s = nextS
-		}
-
-		// A 4-byte match has been found. We'll later see if more than 4 bytes
-		// match. But, prior to the match, src[nextEmit:s] are unmatched. Emit
-		// them as literal bytes.
-		for {
-			// Invariant: we have a 4-byte match at s, and no need to emit any
-			// literal bytes prior to s.
-
-			// Extend the 4-byte match as long as possible.
-			t := candidate.offset
-			l := int16(matchLen(src[s+4:], src[t+4:]) + 4)
-
-			// Extend backwards
-			for t > 0 && s > nextEmit && src[t-1] == src[s-1] {
-				s--
-				t--
-				l++
-			}
-			if nextEmit < s {
-				emitLiteral(dst, src[nextEmit:s])
-			}
-
-			// Save the match found
-			dst.AddMatchLong(int32(l), uint32(s-t-baseMatchOffset))
-			s += l
-			nextEmit = s
-			if nextS >= s {
-				s = nextS + 1
-			}
-			if s >= sLimit {
-				goto emitRemainder
-			}
-
-			// We could immediately start working at s now, but to improve
-			// compression we first update the hash table at s-2 and at s. If
-			// another emitCopy is not our next move, also calculate nextHash
-			// at s+1. At least on GOARCH=amd64, these three hash calculations
-			// are faster as one load64 call (with some shifts) instead of
-			// three load32 calls.
-			x := load6416(src, s-2)
-			o := s - 2
-			prevHash := hashSL(uint32(x))
-			table[prevHash] = tableEntry{offset: o}
-			x >>= 16
-			currHash := hashSL(uint32(x))
-			candidate = table[currHash]
-			table[currHash] = tableEntry{offset: o + 2}
-
-			if uint32(x) != load3216(src, candidate.offset) {
-				cv = uint32(x >> 8)
-				s++
-				break
-			}
-		}
-	}
-
-emitRemainder:
-	if int(nextEmit) < len(src) {
-		// If nothing was added, don't encode literals.
-		if dst.n == 0 {
-			return
-		}
-		emitLiteral(dst, src[nextEmit:])
-	}
-}
diff --git a/vendor/github.com/klauspost/compress/flate/token.go b/vendor/github.com/klauspost/compress/flate/token.go
deleted file mode 100644
index eb862d7a92..0000000000
--- a/vendor/github.com/klauspost/compress/flate/token.go
+++ /dev/null
@@ -1,380 +0,0 @@
-// Copyright 2009 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package flate
-
-import (
-	"bytes"
-	"encoding/binary"
-	"fmt"
-	"io"
-	"math"
-)
-
-const (
-	// From top
-	// 2 bits:   type   0 = literal  1=EOF  2=Match   3=Unused
-	// 8 bits:   xlength = length - MIN_MATCH_LENGTH
-	// 5 bits    offsetcode
-	// 16 bits   xoffset = offset - MIN_OFFSET_SIZE, or literal
-	lengthShift         = 22
-	offsetMask          = 1<<lengthShift - 1
-	typeMask            = 3 << 30
-	literalType         = 0 << 30
-	matchType           = 1 << 30
-	matchOffsetOnlyMask = 0xffff
-)
-
-// The length code for length X (MIN_MATCH_LENGTH <= X <= MAX_MATCH_LENGTH)
-// is lengthCodes[length - MIN_MATCH_LENGTH]
-var lengthCodes = [256]uint8{
-	0, 1, 2, 3, 4, 5, 6, 7, 8, 8,
-	9, 9, 10, 10, 11, 11, 12, 12, 12, 12,
-	13, 13, 13, 13, 14, 14, 14, 14, 15, 15,
-	15, 15, 16, 16, 16, 16, 16, 16, 16, 16,
-	17, 17, 17, 17, 17, 17, 17, 17, 18, 18,
-	18, 18, 18, 18, 18, 18, 19, 19, 19, 19,
-	19, 19, 19, 19, 20, 20, 20, 20, 20, 20,
-	20, 20, 20, 20, 20, 20, 20, 20, 20, 20,
-	21, 21, 21, 21, 21, 21, 21, 21, 21, 21,
-	21, 21, 21, 21, 21, 21, 22, 22, 22, 22,
-	22, 22, 22, 22, 22, 22, 22, 22, 22, 22,
-	22, 22, 23, 23, 23, 23, 23, 23, 23, 23,
-	23, 23, 23, 23, 23, 23, 23, 23, 24, 24,
-	24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
-	24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
-	24, 24, 24, 24, 24, 24, 24, 24, 24, 24,
-	25, 25, 25, 25, 25, 25, 25, 25, 25, 25,
-	25, 25, 25, 25, 25, 25, 25, 25, 25, 25,
-	25, 25, 25, 25, 25, 25, 25, 25, 25, 25,
-	25, 25, 26, 26, 26, 26, 26, 26, 26, 26,
-	26, 26, 26, 26, 26, 26, 26, 26, 26, 26,
-	26, 26, 26, 26, 26, 26, 26, 26, 26, 26,
-	26, 26, 26, 26, 27, 27, 27, 27, 27, 27,
-	27, 27, 27, 27, 27, 27, 27, 27, 27, 27,
-	27, 27, 27, 27, 27, 27, 27, 27, 27, 27,
-	27, 27, 27, 27, 27, 28,
-}
-
-// lengthCodes1 is length codes, but starting at 1.
-var lengthCodes1 = [256]uint8{
-	1, 2, 3, 4, 5, 6, 7, 8, 9, 9,
-	10, 10, 11, 11, 12, 12, 13, 13, 13, 13,
-	14, 14, 14, 14, 15, 15, 15, 15, 16, 16,
-	16, 16, 17, 17, 17, 17, 17, 17, 17, 17,
-	18, 18, 18, 18, 18, 18, 18, 18, 19, 19,
-	19, 19, 19, 19, 19, 19, 20, 20, 20, 20,
-	20, 20, 20, 20, 21, 21, 21, 21, 21, 21,
-	21, 21, 21, 21, 21, 21, 21, 21, 21, 21,
-	22, 22, 22, 22, 22, 22, 22, 22, 22, 22,
-	22, 22, 22, 22, 22, 22, 23, 23, 23, 23,
-	23, 23, 23, 23, 23, 23, 23, 23, 23, 23,
-	23, 23, 24, 24, 24, 24, 24, 24, 24, 24,
-	24, 24, 24, 24, 24, 24, 24, 24, 25, 25,
-	25, 25, 25, 25, 25, 25, 25, 25, 25, 25,
-	25, 25, 25, 25, 25, 25, 25, 25, 25, 25,
-	25, 25, 25, 25, 25, 25, 25, 25, 25, 25,
-	26, 26, 26, 26, 26, 26, 26, 26, 26, 26,
-	26, 26, 26, 26, 26, 26, 26, 26, 26, 26,
-	26, 26, 26, 26, 26, 26, 26, 26, 26, 26,
-	26, 26, 27, 27, 27, 27, 27, 27, 27, 27,
-	27, 27, 27, 27, 27, 27, 27, 27, 27, 27,
-	27, 27, 27, 27, 27, 27, 27, 27, 27, 27,
-	27, 27, 27, 27, 28, 28, 28, 28, 28, 28,
-	28, 28, 28, 28, 28, 28, 28, 28, 28, 28,
-	28, 28, 28, 28, 28, 28, 28, 28, 28, 28,
-	28, 28, 28, 28, 28, 29,
-}
-
-var offsetCodes = [256]uint32{
-	0, 1, 2, 3, 4, 4, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7,
-	8, 8, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 9, 9,
-	10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10,
-	11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11,
-	12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12,
-	12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12,
-	13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13,
-	13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13,
-	14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14,
-	14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14,
-	14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14,
-	14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14,
-	15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
-	15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
-	15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
-	15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15,
-}
-
-// offsetCodes14 are offsetCodes, but with 14 added.
-var offsetCodes14 = [256]uint32{
-	14, 15, 16, 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,
-	25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25, 25,
-	26, 26, 26, 26, 26, 26, 26, 26, 26, 26, 26, 26, 26, 26, 26, 26,
-	26, 26, 26, 26, 26, 26, 26, 26, 26, 26, 26, 26, 26, 26, 26, 26,
-	27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27,
-	27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27, 27,
-	28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28,
-	28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28,
-	28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28,
-	28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28, 28,
-	29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29,
-	29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29,
-	29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29,
-	29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29, 29,
-}
-
-type token uint32
-
-type tokens struct {
-	nLits     int
-	extraHist [32]uint16  // codes 256->maxnumlit
-	offHist   [32]uint16  // offset codes
-	litHist   [256]uint16 // codes 0->255
-	n         uint16      // Must be able to contain maxStoreBlockSize
-	tokens    [maxStoreBlockSize + 1]token
-}
-
-func (t *tokens) Reset() {
-	if t.n == 0 {
-		return
-	}
-	t.n = 0
-	t.nLits = 0
-	for i := range t.litHist[:] {
-		t.litHist[i] = 0
-	}
-	for i := range t.extraHist[:] {
-		t.extraHist[i] = 0
-	}
-	for i := range t.offHist[:] {
-		t.offHist[i] = 0
-	}
-}
-
-func (t *tokens) Fill() {
-	if t.n == 0 {
-		return
-	}
-	for i, v := range t.litHist[:] {
-		if v == 0 {
-			t.litHist[i] = 1
-			t.nLits++
-		}
-	}
-	for i, v := range t.extraHist[:literalCount-256] {
-		if v == 0 {
-			t.nLits++
-			t.extraHist[i] = 1
-		}
-	}
-	for i, v := range t.offHist[:offsetCodeCount] {
-		if v == 0 {
-			t.offHist[i] = 1
-		}
-	}
-}
-
-func indexTokens(in []token) tokens {
-	var t tokens
-	t.indexTokens(in)
-	return t
-}
-
-func (t *tokens) indexTokens(in []token) {
-	t.Reset()
-	for _, tok := range in {
-		if tok < matchType {
-			t.AddLiteral(tok.literal())
-			continue
-		}
-		t.AddMatch(uint32(tok.length()), tok.offset()&matchOffsetOnlyMask)
-	}
-}
-
-// emitLiteral writes a literal chunk and returns the number of bytes written.
-func emitLiteral(dst *tokens, lit []byte) {
-	ol := int(dst.n)
-	for i, v := range lit {
-		dst.tokens[(i+ol)&maxStoreBlockSize] = token(v)
-		dst.litHist[v]++
-	}
-	dst.n += uint16(len(lit))
-	dst.nLits += len(lit)
-}
-
-func (t *tokens) AddLiteral(lit byte) {
-	t.tokens[t.n] = token(lit)
-	t.litHist[lit]++
-	t.n++
-	t.nLits++
-}
-
-// from https://stackoverflow.com/a/28730362
-func mFastLog2(val float32) float32 {
-	ux := int32(math.Float32bits(val))
-	log2 := (float32)(((ux >> 23) & 255) - 128)
-	ux &= -0x7f800001
-	ux += 127 << 23
-	uval := math.Float32frombits(uint32(ux))
-	log2 += ((-0.34484843)*uval+2.02466578)*uval - 0.67487759
-	return log2
-}
-
-// EstimatedBits will return an minimum size estimated by an *optimal*
-// compression of the block.
-// The size of the block
-func (t *tokens) EstimatedBits() int {
-	shannon := float32(0)
-	bits := int(0)
-	nMatches := 0
-	if t.nLits > 0 {
-		invTotal := 1.0 / float32(t.nLits)
-		for _, v := range t.litHist[:] {
-			if v > 0 {
-				n := float32(v)
-				shannon += atLeastOne(-mFastLog2(n*invTotal)) * n
-			}
-		}
-		// Just add 15 for EOB
-		shannon += 15
-		for i, v := range t.extraHist[1 : literalCount-256] {
-			if v > 0 {
-				n := float32(v)
-				shannon += atLeastOne(-mFastLog2(n*invTotal)) * n
-				bits += int(lengthExtraBits[i&31]) * int(v)
-				nMatches += int(v)
-			}
-		}
-	}
-	if nMatches > 0 {
-		invTotal := 1.0 / float32(nMatches)
-		for i, v := range t.offHist[:offsetCodeCount] {
-			if v > 0 {
-				n := float32(v)
-				shannon += atLeastOne(-mFastLog2(n*invTotal)) * n
-				bits += int(offsetExtraBits[i&31]) * int(v)
-			}
-		}
-	}
-	return int(shannon) + bits
-}
-
-// AddMatch adds a match to the tokens.
-// This function is very sensitive to inlining and right on the border.
-func (t *tokens) AddMatch(xlength uint32, xoffset uint32) {
-	if debugDeflate {
-		if xlength >= maxMatchLength+baseMatchLength {
-			panic(fmt.Errorf("invalid length: %v", xlength))
-		}
-		if xoffset >= maxMatchOffset+baseMatchOffset {
-			panic(fmt.Errorf("invalid offset: %v", xoffset))
-		}
-	}
-	oCode := offsetCode(xoffset)
-	xoffset |= oCode << 16
-	t.nLits++
-
-	t.extraHist[lengthCodes1[uint8(xlength)]]++
-	t.offHist[oCode]++
-	t.tokens[t.n] = token(matchType | xlength<<lengthShift | xoffset)
-	t.n++
-}
-
-// AddMatchLong adds a match to the tokens, potentially longer than max match length.
-// Length should NOT have the base subtracted, only offset should.
-func (t *tokens) AddMatchLong(xlength int32, xoffset uint32) {
-	if debugDeflate {
-		if xoffset >= maxMatchOffset+baseMatchOffset {
-			panic(fmt.Errorf("invalid offset: %v", xoffset))
-		}
-	}
-	oc := offsetCode(xoffset)
-	xoffset |= oc << 16
-	for xlength > 0 {
-		xl := xlength
-		if xl > 258 {
-			// We need to have at least baseMatchLength left over for next loop.
-			xl = 258 - baseMatchLength
-		}
-		xlength -= xl
-		xl -= baseMatchLength
-		t.nLits++
-		t.extraHist[lengthCodes1[uint8(xl)]]++
-		t.offHist[oc]++
-		t.tokens[t.n] = token(matchType | uint32(xl)<<lengthShift | xoffset)
-		t.n++
-	}
-}
-
-func (t *tokens) AddEOB() {
-	t.tokens[t.n] = token(endBlockMarker)
-	t.extraHist[0]++
-	t.n++
-}
-
-func (t *tokens) Slice() []token {
-	return t.tokens[:t.n]
-}
-
-// VarInt returns the tokens as varint encoded bytes.
-func (t *tokens) VarInt() []byte {
-	var b = make([]byte, binary.MaxVarintLen32*int(t.n))
-	var off int
-	for _, v := range t.tokens[:t.n] {
-		off += binary.PutUvarint(b[off:], uint64(v))
-	}
-	return b[:off]
-}
-
-// FromVarInt restores t to the varint encoded tokens provided.
-// Any data in t is removed.
-func (t *tokens) FromVarInt(b []byte) error {
-	var buf = bytes.NewReader(b)
-	var toks []token
-	for {
-		r, err := binary.ReadUvarint(buf)
-		if err == io.EOF {
-			break
-		}
-		if err != nil {
-			return err
-		}
-		toks = append(toks, token(r))
-	}
-	t.indexTokens(toks)
-	return nil
-}
-
-// Returns the type of a token
-func (t token) typ() uint32 { return uint32(t) & typeMask }
-
-// Returns the literal of a literal token
-func (t token) literal() uint8 { return uint8(t) }
-
-// Returns the extra offset of a match token
-func (t token) offset() uint32 { return uint32(t) & offsetMask }
-
-func (t token) length() uint8 { return uint8(t >> lengthShift) }
-
-// The code is never more than 8 bits, but is returned as uint32 for convenience.
-func lengthCode(len uint8) uint32 { return uint32(lengthCodes[len]) }
-
-// Returns the offset code corresponding to a specific offset
-func offsetCode(off uint32) uint32 {
-	if false {
-		if off < uint32(len(offsetCodes)) {
-			return offsetCodes[off&255]
-		} else if off>>7 < uint32(len(offsetCodes)) {
-			return offsetCodes[(off>>7)&255] + 14
-		} else {
-			return offsetCodes[(off>>14)&255] + 28
-		}
-	}
-	if off < uint32(len(offsetCodes)) {
-		return offsetCodes[uint8(off)]
-	}
-	return offsetCodes14[uint8(off>>7)]
-}
diff --git a/vendor/github.com/klauspost/compress/fse/README.md b/vendor/github.com/klauspost/compress/fse/README.md
deleted file mode 100644
index ea7324da67..0000000000
--- a/vendor/github.com/klauspost/compress/fse/README.md
+++ /dev/null
@@ -1,79 +0,0 @@
-# Finite State Entropy

-

-This package provides Finite State Entropy encoding and decoding.

-            

-Finite State Entropy (also referenced as [tANS](https://en.wikipedia.org/wiki/Asymmetric_numeral_systems#tANS)) 

-encoding provides a fast near-optimal symbol encoding/decoding

-for byte blocks as implemented in [zstandard](https://github.com/facebook/zstd).

-

-This can be used for compressing input with a lot of similar input values to the smallest number of bytes.

-This does not perform any multi-byte [dictionary coding](https://en.wikipedia.org/wiki/Dictionary_coder) as LZ coders,

-but it can be used as a secondary step to compressors (like Snappy) that does not do entropy encoding. 

-

-* [Godoc documentation](https://godoc.org/github.com/klauspost/compress/fse)

-

-## News

-

- * Feb 2018: First implementation released. Consider this beta software for now.

-

-# Usage

-

-This package provides a low level interface that allows to compress single independent blocks. 

-

-Each block is separate, and there is no built in integrity checks. 

-This means that the caller should keep track of block sizes and also do checksums if needed.  

-

-Compressing a block is done via the [`Compress`](https://godoc.org/github.com/klauspost/compress/fse#Compress) function.

-You must provide input and will receive the output and maybe an error.

-

-These error values can be returned:

-

-| Error               | Description                                                                 |

-|---------------------|-----------------------------------------------------------------------------|

-| `<nil>`             | Everything ok, output is returned                                           |

-| `ErrIncompressible` | Returned when input is judged to be too hard to compress                    |

-| `ErrUseRLE`         | Returned from the compressor when the input is a single byte value repeated |

-| `(error)`           | An internal error occurred.                                                 |

-

-As can be seen above there are errors that will be returned even under normal operation so it is important to handle these.

-

-To reduce allocations you can provide a [`Scratch`](https://godoc.org/github.com/klauspost/compress/fse#Scratch) object 

-that can be re-used for successive calls. Both compression and decompression accepts a `Scratch` object, and the same 

-object can be used for both.   

-

-Be aware, that when re-using a `Scratch` object that the *output* buffer is also re-used, so if you are still using this

-you must set the `Out` field in the scratch to nil. The same buffer is used for compression and decompression output.

-

-Decompressing is done by calling the [`Decompress`](https://godoc.org/github.com/klauspost/compress/fse#Decompress) function.

-You must provide the output from the compression stage, at exactly the size you got back. If you receive an error back

-your input was likely corrupted. 

-

-It is important to note that a successful decoding does *not* mean your output matches your original input. 

-There are no integrity checks, so relying on errors from the decompressor does not assure your data is valid.

-

-For more detailed usage, see examples in the [godoc documentation](https://godoc.org/github.com/klauspost/compress/fse#pkg-examples).

-

-# Performance

-

-A lot of factors are affecting speed. Block sizes and compressibility of the material are primary factors.  

-All compression functions are currently only running on the calling goroutine so only one core will be used per block.  

-

-The compressor is significantly faster if symbols are kept as small as possible. The highest byte value of the input

-is used to reduce some of the processing, so if all your input is above byte value 64 for instance, it may be 

-beneficial to transpose all your input values down by 64.   

-

-With moderate block sizes around 64k speed are typically 200MB/s per core for compression and 

-around 300MB/s decompression speed. 

-

-The same hardware typically does Huffman (deflate) encoding at 125MB/s and decompression at 100MB/s. 

-

-# Plans

-

-At one point, more internals will be exposed to facilitate more "expert" usage of the components. 

-

-A streaming interface is also likely to be implemented. Likely compatible with [FSE stream format](https://github.com/Cyan4973/FiniteStateEntropy/blob/dev/programs/fileio.c#L261).  

-

-# Contributing

-

-Contributions are always welcome. Be aware that adding public functions will require good justification and breaking 

-changes will likely not be accepted. If in doubt open an issue before writing the PR.  
\ No newline at end of file
diff --git a/vendor/github.com/klauspost/compress/fse/bitreader.go b/vendor/github.com/klauspost/compress/fse/bitreader.go
deleted file mode 100644
index f65eb3909c..0000000000
--- a/vendor/github.com/klauspost/compress/fse/bitreader.go
+++ /dev/null
@@ -1,122 +0,0 @@
-// Copyright 2018 Klaus Post. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-// Based on work Copyright (c) 2013, Yann Collet, released under BSD License.
-
-package fse
-
-import (
-	"encoding/binary"
-	"errors"
-	"io"
-)
-
-// bitReader reads a bitstream in reverse.
-// The last set bit indicates the start of the stream and is used
-// for aligning the input.
-type bitReader struct {
-	in       []byte
-	off      uint // next byte to read is at in[off - 1]
-	value    uint64
-	bitsRead uint8
-}
-
-// init initializes and resets the bit reader.
-func (b *bitReader) init(in []byte) error {
-	if len(in) < 1 {
-		return errors.New("corrupt stream: too short")
-	}
-	b.in = in
-	b.off = uint(len(in))
-	// The highest bit of the last byte indicates where to start
-	v := in[len(in)-1]
-	if v == 0 {
-		return errors.New("corrupt stream, did not find end of stream")
-	}
-	b.bitsRead = 64
-	b.value = 0
-	if len(in) >= 8 {
-		b.fillFastStart()
-	} else {
-		b.fill()
-		b.fill()
-	}
-	b.bitsRead += 8 - uint8(highBits(uint32(v)))
-	return nil
-}
-
-// getBits will return n bits. n can be 0.
-func (b *bitReader) getBits(n uint8) uint16 {
-	if n == 0 || b.bitsRead >= 64 {
-		return 0
-	}
-	return b.getBitsFast(n)
-}
-
-// getBitsFast requires that at least one bit is requested every time.
-// There are no checks if the buffer is filled.
-func (b *bitReader) getBitsFast(n uint8) uint16 {
-	const regMask = 64 - 1
-	v := uint16((b.value << (b.bitsRead & regMask)) >> ((regMask + 1 - n) & regMask))
-	b.bitsRead += n
-	return v
-}
-
-// fillFast() will make sure at least 32 bits are available.
-// There must be at least 4 bytes available.
-func (b *bitReader) fillFast() {
-	if b.bitsRead < 32 {
-		return
-	}
-	// 2 bounds checks.
-	v := b.in[b.off-4:]
-	v = v[:4]
-	low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-	b.value = (b.value << 32) | uint64(low)
-	b.bitsRead -= 32
-	b.off -= 4
-}
-
-// fill() will make sure at least 32 bits are available.
-func (b *bitReader) fill() {
-	if b.bitsRead < 32 {
-		return
-	}
-	if b.off > 4 {
-		v := b.in[b.off-4:]
-		v = v[:4]
-		low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-		b.value = (b.value << 32) | uint64(low)
-		b.bitsRead -= 32
-		b.off -= 4
-		return
-	}
-	for b.off > 0 {
-		b.value = (b.value << 8) | uint64(b.in[b.off-1])
-		b.bitsRead -= 8
-		b.off--
-	}
-}
-
-// fillFastStart() assumes the bitreader is empty and there is at least 8 bytes to read.
-func (b *bitReader) fillFastStart() {
-	// Do single re-slice to avoid bounds checks.
-	b.value = binary.LittleEndian.Uint64(b.in[b.off-8:])
-	b.bitsRead = 0
-	b.off -= 8
-}
-
-// finished returns true if all bits have been read from the bit stream.
-func (b *bitReader) finished() bool {
-	return b.bitsRead >= 64 && b.off == 0
-}
-
-// close the bitstream and returns an error if out-of-buffer reads occurred.
-func (b *bitReader) close() error {
-	// Release reference.
-	b.in = nil
-	if b.bitsRead > 64 {
-		return io.ErrUnexpectedEOF
-	}
-	return nil
-}
diff --git a/vendor/github.com/klauspost/compress/fse/bitwriter.go b/vendor/github.com/klauspost/compress/fse/bitwriter.go
deleted file mode 100644
index 43e463611b..0000000000
--- a/vendor/github.com/klauspost/compress/fse/bitwriter.go
+++ /dev/null
@@ -1,168 +0,0 @@
-// Copyright 2018 Klaus Post. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-// Based on work Copyright (c) 2013, Yann Collet, released under BSD License.
-
-package fse
-
-import "fmt"
-
-// bitWriter will write bits.
-// First bit will be LSB of the first byte of output.
-type bitWriter struct {
-	bitContainer uint64
-	nBits        uint8
-	out          []byte
-}
-
-// bitMask16 is bitmasks. Has extra to avoid bounds check.
-var bitMask16 = [32]uint16{
-	0, 1, 3, 7, 0xF, 0x1F,
-	0x3F, 0x7F, 0xFF, 0x1FF, 0x3FF, 0x7FF,
-	0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, 0xFFFF,
-	0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF,
-	0xFFFF, 0xFFFF} /* up to 16 bits */
-
-// addBits16NC will add up to 16 bits.
-// It will not check if there is space for them,
-// so the caller must ensure that it has flushed recently.
-func (b *bitWriter) addBits16NC(value uint16, bits uint8) {
-	b.bitContainer |= uint64(value&bitMask16[bits&31]) << (b.nBits & 63)
-	b.nBits += bits
-}
-
-// addBits16Clean will add up to 16 bits. value may not contain more set bits than indicated.
-// It will not check if there is space for them, so the caller must ensure that it has flushed recently.
-func (b *bitWriter) addBits16Clean(value uint16, bits uint8) {
-	b.bitContainer |= uint64(value) << (b.nBits & 63)
-	b.nBits += bits
-}
-
-// addBits16ZeroNC will add up to 16 bits.
-// It will not check if there is space for them,
-// so the caller must ensure that it has flushed recently.
-// This is fastest if bits can be zero.
-func (b *bitWriter) addBits16ZeroNC(value uint16, bits uint8) {
-	if bits == 0 {
-		return
-	}
-	value <<= (16 - bits) & 15
-	value >>= (16 - bits) & 15
-	b.bitContainer |= uint64(value) << (b.nBits & 63)
-	b.nBits += bits
-}
-
-// flush will flush all pending full bytes.
-// There will be at least 56 bits available for writing when this has been called.
-// Using flush32 is faster, but leaves less space for writing.
-func (b *bitWriter) flush() {
-	v := b.nBits >> 3
-	switch v {
-	case 0:
-	case 1:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-		)
-	case 2:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-		)
-	case 3:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-		)
-	case 4:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-		)
-	case 5:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-		)
-	case 6:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-		)
-	case 7:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-			byte(b.bitContainer>>48),
-		)
-	case 8:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-			byte(b.bitContainer>>48),
-			byte(b.bitContainer>>56),
-		)
-	default:
-		panic(fmt.Errorf("bits (%d) > 64", b.nBits))
-	}
-	b.bitContainer >>= v << 3
-	b.nBits &= 7
-}
-
-// flush32 will flush out, so there are at least 32 bits available for writing.
-func (b *bitWriter) flush32() {
-	if b.nBits < 32 {
-		return
-	}
-	b.out = append(b.out,
-		byte(b.bitContainer),
-		byte(b.bitContainer>>8),
-		byte(b.bitContainer>>16),
-		byte(b.bitContainer>>24))
-	b.nBits -= 32
-	b.bitContainer >>= 32
-}
-
-// flushAlign will flush remaining full bytes and align to next byte boundary.
-func (b *bitWriter) flushAlign() {
-	nbBytes := (b.nBits + 7) >> 3
-	for i := uint8(0); i < nbBytes; i++ {
-		b.out = append(b.out, byte(b.bitContainer>>(i*8)))
-	}
-	b.nBits = 0
-	b.bitContainer = 0
-}
-
-// close will write the alignment bit and write the final byte(s)
-// to the output.
-func (b *bitWriter) close() error {
-	// End mark
-	b.addBits16Clean(1, 1)
-	// flush until next byte.
-	b.flushAlign()
-	return nil
-}
-
-// reset and continue writing by appending to out.
-func (b *bitWriter) reset(out []byte) {
-	b.bitContainer = 0
-	b.nBits = 0
-	b.out = out
-}
diff --git a/vendor/github.com/klauspost/compress/fse/bytereader.go b/vendor/github.com/klauspost/compress/fse/bytereader.go
deleted file mode 100644
index abade2d605..0000000000
--- a/vendor/github.com/klauspost/compress/fse/bytereader.go
+++ /dev/null
@@ -1,47 +0,0 @@
-// Copyright 2018 Klaus Post. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-// Based on work Copyright (c) 2013, Yann Collet, released under BSD License.
-
-package fse
-
-// byteReader provides a byte reader that reads
-// little endian values from a byte stream.
-// The input stream is manually advanced.
-// The reader performs no bounds checks.
-type byteReader struct {
-	b   []byte
-	off int
-}
-
-// init will initialize the reader and set the input.
-func (b *byteReader) init(in []byte) {
-	b.b = in
-	b.off = 0
-}
-
-// advance the stream b n bytes.
-func (b *byteReader) advance(n uint) {
-	b.off += int(n)
-}
-
-// Uint32 returns a little endian uint32 starting at current offset.
-func (b byteReader) Uint32() uint32 {
-	b2 := b.b[b.off:]
-	b2 = b2[:4]
-	v3 := uint32(b2[3])
-	v2 := uint32(b2[2])
-	v1 := uint32(b2[1])
-	v0 := uint32(b2[0])
-	return v0 | (v1 << 8) | (v2 << 16) | (v3 << 24)
-}
-
-// unread returns the unread portion of the input.
-func (b byteReader) unread() []byte {
-	return b.b[b.off:]
-}
-
-// remain will return the number of bytes remaining.
-func (b byteReader) remain() int {
-	return len(b.b) - b.off
-}
diff --git a/vendor/github.com/klauspost/compress/fse/compress.go b/vendor/github.com/klauspost/compress/fse/compress.go
deleted file mode 100644
index 6f341914c6..0000000000
--- a/vendor/github.com/klauspost/compress/fse/compress.go
+++ /dev/null
@@ -1,683 +0,0 @@
-// Copyright 2018 Klaus Post. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-// Based on work Copyright (c) 2013, Yann Collet, released under BSD License.
-
-package fse
-
-import (
-	"errors"
-	"fmt"
-)
-
-// Compress the input bytes. Input must be < 2GB.
-// Provide a Scratch buffer to avoid memory allocations.
-// Note that the output is also kept in the scratch buffer.
-// If input is too hard to compress, ErrIncompressible is returned.
-// If input is a single byte value repeated ErrUseRLE is returned.
-func Compress(in []byte, s *Scratch) ([]byte, error) {
-	if len(in) <= 1 {
-		return nil, ErrIncompressible
-	}
-	if len(in) > (2<<30)-1 {
-		return nil, errors.New("input too big, must be < 2GB")
-	}
-	s, err := s.prepare(in)
-	if err != nil {
-		return nil, err
-	}
-
-	// Create histogram, if none was provided.
-	maxCount := s.maxCount
-	if maxCount == 0 {
-		maxCount = s.countSimple(in)
-	}
-	// Reset for next run.
-	s.clearCount = true
-	s.maxCount = 0
-	if maxCount == len(in) {
-		// One symbol, use RLE
-		return nil, ErrUseRLE
-	}
-	if maxCount == 1 || maxCount < (len(in)>>7) {
-		// Each symbol present maximum once or too well distributed.
-		return nil, ErrIncompressible
-	}
-	s.optimalTableLog()
-	err = s.normalizeCount()
-	if err != nil {
-		return nil, err
-	}
-	err = s.writeCount()
-	if err != nil {
-		return nil, err
-	}
-
-	if false {
-		err = s.validateNorm()
-		if err != nil {
-			return nil, err
-		}
-	}
-
-	err = s.buildCTable()
-	if err != nil {
-		return nil, err
-	}
-	err = s.compress(in)
-	if err != nil {
-		return nil, err
-	}
-	s.Out = s.bw.out
-	// Check if we compressed.
-	if len(s.Out) >= len(in) {
-		return nil, ErrIncompressible
-	}
-	return s.Out, nil
-}
-
-// cState contains the compression state of a stream.
-type cState struct {
-	bw         *bitWriter
-	stateTable []uint16
-	state      uint16
-}
-
-// init will initialize the compression state to the first symbol of the stream.
-func (c *cState) init(bw *bitWriter, ct *cTable, tableLog uint8, first symbolTransform) {
-	c.bw = bw
-	c.stateTable = ct.stateTable
-
-	nbBitsOut := (first.deltaNbBits + (1 << 15)) >> 16
-	im := int32((nbBitsOut << 16) - first.deltaNbBits)
-	lu := (im >> nbBitsOut) + first.deltaFindState
-	c.state = c.stateTable[lu]
-}
-
-// encode the output symbol provided and write it to the bitstream.
-func (c *cState) encode(symbolTT symbolTransform) {
-	nbBitsOut := (uint32(c.state) + symbolTT.deltaNbBits) >> 16
-	dstState := int32(c.state>>(nbBitsOut&15)) + symbolTT.deltaFindState
-	c.bw.addBits16NC(c.state, uint8(nbBitsOut))
-	c.state = c.stateTable[dstState]
-}
-
-// encode the output symbol provided and write it to the bitstream.
-func (c *cState) encodeZero(symbolTT symbolTransform) {
-	nbBitsOut := (uint32(c.state) + symbolTT.deltaNbBits) >> 16
-	dstState := int32(c.state>>(nbBitsOut&15)) + symbolTT.deltaFindState
-	c.bw.addBits16ZeroNC(c.state, uint8(nbBitsOut))
-	c.state = c.stateTable[dstState]
-}
-
-// flush will write the tablelog to the output and flush the remaining full bytes.
-func (c *cState) flush(tableLog uint8) {
-	c.bw.flush32()
-	c.bw.addBits16NC(c.state, tableLog)
-	c.bw.flush()
-}
-
-// compress is the main compression loop that will encode the input from the last byte to the first.
-func (s *Scratch) compress(src []byte) error {
-	if len(src) <= 2 {
-		return errors.New("compress: src too small")
-	}
-	tt := s.ct.symbolTT[:256]
-	s.bw.reset(s.Out)
-
-	// Our two states each encodes every second byte.
-	// Last byte encoded (first byte decoded) will always be encoded by c1.
-	var c1, c2 cState
-
-	// Encode so remaining size is divisible by 4.
-	ip := len(src)
-	if ip&1 == 1 {
-		c1.init(&s.bw, &s.ct, s.actualTableLog, tt[src[ip-1]])
-		c2.init(&s.bw, &s.ct, s.actualTableLog, tt[src[ip-2]])
-		c1.encodeZero(tt[src[ip-3]])
-		ip -= 3
-	} else {
-		c2.init(&s.bw, &s.ct, s.actualTableLog, tt[src[ip-1]])
-		c1.init(&s.bw, &s.ct, s.actualTableLog, tt[src[ip-2]])
-		ip -= 2
-	}
-	if ip&2 != 0 {
-		c2.encodeZero(tt[src[ip-1]])
-		c1.encodeZero(tt[src[ip-2]])
-		ip -= 2
-	}
-
-	// Main compression loop.
-	switch {
-	case !s.zeroBits && s.actualTableLog <= 8:
-		// We can encode 4 symbols without requiring a flush.
-		// We do not need to check if any output is 0 bits.
-		for ip >= 4 {
-			s.bw.flush32()
-			v3, v2, v1, v0 := src[ip-4], src[ip-3], src[ip-2], src[ip-1]
-			c2.encode(tt[v0])
-			c1.encode(tt[v1])
-			c2.encode(tt[v2])
-			c1.encode(tt[v3])
-			ip -= 4
-		}
-	case !s.zeroBits:
-		// We do not need to check if any output is 0 bits.
-		for ip >= 4 {
-			s.bw.flush32()
-			v3, v2, v1, v0 := src[ip-4], src[ip-3], src[ip-2], src[ip-1]
-			c2.encode(tt[v0])
-			c1.encode(tt[v1])
-			s.bw.flush32()
-			c2.encode(tt[v2])
-			c1.encode(tt[v3])
-			ip -= 4
-		}
-	case s.actualTableLog <= 8:
-		// We can encode 4 symbols without requiring a flush
-		for ip >= 4 {
-			s.bw.flush32()
-			v3, v2, v1, v0 := src[ip-4], src[ip-3], src[ip-2], src[ip-1]
-			c2.encodeZero(tt[v0])
-			c1.encodeZero(tt[v1])
-			c2.encodeZero(tt[v2])
-			c1.encodeZero(tt[v3])
-			ip -= 4
-		}
-	default:
-		for ip >= 4 {
-			s.bw.flush32()
-			v3, v2, v1, v0 := src[ip-4], src[ip-3], src[ip-2], src[ip-1]
-			c2.encodeZero(tt[v0])
-			c1.encodeZero(tt[v1])
-			s.bw.flush32()
-			c2.encodeZero(tt[v2])
-			c1.encodeZero(tt[v3])
-			ip -= 4
-		}
-	}
-
-	// Flush final state.
-	// Used to initialize state when decoding.
-	c2.flush(s.actualTableLog)
-	c1.flush(s.actualTableLog)
-
-	return s.bw.close()
-}
-
-// writeCount will write the normalized histogram count to header.
-// This is read back by readNCount.
-func (s *Scratch) writeCount() error {
-	var (
-		tableLog  = s.actualTableLog
-		tableSize = 1 << tableLog
-		previous0 bool
-		charnum   uint16
-
-		maxHeaderSize = ((int(s.symbolLen) * int(tableLog)) >> 3) + 3
-
-		// Write Table Size
-		bitStream = uint32(tableLog - minTablelog)
-		bitCount  = uint(4)
-		remaining = int16(tableSize + 1) /* +1 for extra accuracy */
-		threshold = int16(tableSize)
-		nbBits    = uint(tableLog + 1)
-	)
-	if cap(s.Out) < maxHeaderSize {
-		s.Out = make([]byte, 0, s.br.remain()+maxHeaderSize)
-	}
-	outP := uint(0)
-	out := s.Out[:maxHeaderSize]
-
-	// stops at 1
-	for remaining > 1 {
-		if previous0 {
-			start := charnum
-			for s.norm[charnum] == 0 {
-				charnum++
-			}
-			for charnum >= start+24 {
-				start += 24
-				bitStream += uint32(0xFFFF) << bitCount
-				out[outP] = byte(bitStream)
-				out[outP+1] = byte(bitStream >> 8)
-				outP += 2
-				bitStream >>= 16
-			}
-			for charnum >= start+3 {
-				start += 3
-				bitStream += 3 << bitCount
-				bitCount += 2
-			}
-			bitStream += uint32(charnum-start) << bitCount
-			bitCount += 2
-			if bitCount > 16 {
-				out[outP] = byte(bitStream)
-				out[outP+1] = byte(bitStream >> 8)
-				outP += 2
-				bitStream >>= 16
-				bitCount -= 16
-			}
-		}
-
-		count := s.norm[charnum]
-		charnum++
-		max := (2*threshold - 1) - remaining
-		if count < 0 {
-			remaining += count
-		} else {
-			remaining -= count
-		}
-		count++ // +1 for extra accuracy
-		if count >= threshold {
-			count += max // [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[
-		}
-		bitStream += uint32(count) << bitCount
-		bitCount += nbBits
-		if count < max {
-			bitCount--
-		}
-
-		previous0 = count == 1
-		if remaining < 1 {
-			return errors.New("internal error: remaining<1")
-		}
-		for remaining < threshold {
-			nbBits--
-			threshold >>= 1
-		}
-
-		if bitCount > 16 {
-			out[outP] = byte(bitStream)
-			out[outP+1] = byte(bitStream >> 8)
-			outP += 2
-			bitStream >>= 16
-			bitCount -= 16
-		}
-	}
-
-	out[outP] = byte(bitStream)
-	out[outP+1] = byte(bitStream >> 8)
-	outP += (bitCount + 7) / 8
-
-	if charnum > s.symbolLen {
-		return errors.New("internal error: charnum > s.symbolLen")
-	}
-	s.Out = out[:outP]
-	return nil
-}
-
-// symbolTransform contains the state transform for a symbol.
-type symbolTransform struct {
-	deltaFindState int32
-	deltaNbBits    uint32
-}
-
-// String prints values as a human readable string.
-func (s symbolTransform) String() string {
-	return fmt.Sprintf("dnbits: %08x, fs:%d", s.deltaNbBits, s.deltaFindState)
-}
-
-// cTable contains tables used for compression.
-type cTable struct {
-	tableSymbol []byte
-	stateTable  []uint16
-	symbolTT    []symbolTransform
-}
-
-// allocCtable will allocate tables needed for compression.
-// If existing tables a re big enough, they are simply re-used.
-func (s *Scratch) allocCtable() {
-	tableSize := 1 << s.actualTableLog
-	// get tableSymbol that is big enough.
-	if cap(s.ct.tableSymbol) < tableSize {
-		s.ct.tableSymbol = make([]byte, tableSize)
-	}
-	s.ct.tableSymbol = s.ct.tableSymbol[:tableSize]
-
-	ctSize := tableSize
-	if cap(s.ct.stateTable) < ctSize {
-		s.ct.stateTable = make([]uint16, ctSize)
-	}
-	s.ct.stateTable = s.ct.stateTable[:ctSize]
-
-	if cap(s.ct.symbolTT) < 256 {
-		s.ct.symbolTT = make([]symbolTransform, 256)
-	}
-	s.ct.symbolTT = s.ct.symbolTT[:256]
-}
-
-// buildCTable will populate the compression table so it is ready to be used.
-func (s *Scratch) buildCTable() error {
-	tableSize := uint32(1 << s.actualTableLog)
-	highThreshold := tableSize - 1
-	var cumul [maxSymbolValue + 2]int16
-
-	s.allocCtable()
-	tableSymbol := s.ct.tableSymbol[:tableSize]
-	// symbol start positions
-	{
-		cumul[0] = 0
-		for ui, v := range s.norm[:s.symbolLen-1] {
-			u := byte(ui) // one less than reference
-			if v == -1 {
-				// Low proba symbol
-				cumul[u+1] = cumul[u] + 1
-				tableSymbol[highThreshold] = u
-				highThreshold--
-			} else {
-				cumul[u+1] = cumul[u] + v
-			}
-		}
-		// Encode last symbol separately to avoid overflowing u
-		u := int(s.symbolLen - 1)
-		v := s.norm[s.symbolLen-1]
-		if v == -1 {
-			// Low proba symbol
-			cumul[u+1] = cumul[u] + 1
-			tableSymbol[highThreshold] = byte(u)
-			highThreshold--
-		} else {
-			cumul[u+1] = cumul[u] + v
-		}
-		if uint32(cumul[s.symbolLen]) != tableSize {
-			return fmt.Errorf("internal error: expected cumul[s.symbolLen] (%d) == tableSize (%d)", cumul[s.symbolLen], tableSize)
-		}
-		cumul[s.symbolLen] = int16(tableSize) + 1
-	}
-	// Spread symbols
-	s.zeroBits = false
-	{
-		step := tableStep(tableSize)
-		tableMask := tableSize - 1
-		var position uint32
-		// if any symbol > largeLimit, we may have 0 bits output.
-		largeLimit := int16(1 << (s.actualTableLog - 1))
-		for ui, v := range s.norm[:s.symbolLen] {
-			symbol := byte(ui)
-			if v > largeLimit {
-				s.zeroBits = true
-			}
-			for nbOccurrences := int16(0); nbOccurrences < v; nbOccurrences++ {
-				tableSymbol[position] = symbol
-				position = (position + step) & tableMask
-				for position > highThreshold {
-					position = (position + step) & tableMask
-				} /* Low proba area */
-			}
-		}
-
-		// Check if we have gone through all positions
-		if position != 0 {
-			return errors.New("position!=0")
-		}
-	}
-
-	// Build table
-	table := s.ct.stateTable
-	{
-		tsi := int(tableSize)
-		for u, v := range tableSymbol {
-			// TableU16 : sorted by symbol order; gives next state value
-			table[cumul[v]] = uint16(tsi + u)
-			cumul[v]++
-		}
-	}
-
-	// Build Symbol Transformation Table
-	{
-		total := int16(0)
-		symbolTT := s.ct.symbolTT[:s.symbolLen]
-		tableLog := s.actualTableLog
-		tl := (uint32(tableLog) << 16) - (1 << tableLog)
-		for i, v := range s.norm[:s.symbolLen] {
-			switch v {
-			case 0:
-			case -1, 1:
-				symbolTT[i].deltaNbBits = tl
-				symbolTT[i].deltaFindState = int32(total - 1)
-				total++
-			default:
-				maxBitsOut := uint32(tableLog) - highBits(uint32(v-1))
-				minStatePlus := uint32(v) << maxBitsOut
-				symbolTT[i].deltaNbBits = (maxBitsOut << 16) - minStatePlus
-				symbolTT[i].deltaFindState = int32(total - v)
-				total += v
-			}
-		}
-		if total != int16(tableSize) {
-			return fmt.Errorf("total mismatch %d (got) != %d (want)", total, tableSize)
-		}
-	}
-	return nil
-}
-
-// countSimple will create a simple histogram in s.count.
-// Returns the biggest count.
-// Does not update s.clearCount.
-func (s *Scratch) countSimple(in []byte) (max int) {
-	for _, v := range in {
-		s.count[v]++
-	}
-	m := uint32(0)
-	for i, v := range s.count[:] {
-		if v > m {
-			m = v
-		}
-		if v > 0 {
-			s.symbolLen = uint16(i) + 1
-		}
-	}
-	return int(m)
-}
-
-// minTableLog provides the minimum logSize to safely represent a distribution.
-func (s *Scratch) minTableLog() uint8 {
-	minBitsSrc := highBits(uint32(s.br.remain()-1)) + 1
-	minBitsSymbols := highBits(uint32(s.symbolLen-1)) + 2
-	if minBitsSrc < minBitsSymbols {
-		return uint8(minBitsSrc)
-	}
-	return uint8(minBitsSymbols)
-}
-
-// optimalTableLog calculates and sets the optimal tableLog in s.actualTableLog
-func (s *Scratch) optimalTableLog() {
-	tableLog := s.TableLog
-	minBits := s.minTableLog()
-	maxBitsSrc := uint8(highBits(uint32(s.br.remain()-1))) - 2
-	if maxBitsSrc < tableLog {
-		// Accuracy can be reduced
-		tableLog = maxBitsSrc
-	}
-	if minBits > tableLog {
-		tableLog = minBits
-	}
-	// Need a minimum to safely represent all symbol values
-	if tableLog < minTablelog {
-		tableLog = minTablelog
-	}
-	if tableLog > maxTableLog {
-		tableLog = maxTableLog
-	}
-	s.actualTableLog = tableLog
-}
-
-var rtbTable = [...]uint32{0, 473195, 504333, 520860, 550000, 700000, 750000, 830000}
-
-// normalizeCount will normalize the count of the symbols so
-// the total is equal to the table size.
-func (s *Scratch) normalizeCount() error {
-	var (
-		tableLog          = s.actualTableLog
-		scale             = 62 - uint64(tableLog)
-		step              = (1 << 62) / uint64(s.br.remain())
-		vStep             = uint64(1) << (scale - 20)
-		stillToDistribute = int16(1 << tableLog)
-		largest           int
-		largestP          int16
-		lowThreshold      = (uint32)(s.br.remain() >> tableLog)
-	)
-
-	for i, cnt := range s.count[:s.symbolLen] {
-		// already handled
-		// if (count[s] == s.length) return 0;   /* rle special case */
-
-		if cnt == 0 {
-			s.norm[i] = 0
-			continue
-		}
-		if cnt <= lowThreshold {
-			s.norm[i] = -1
-			stillToDistribute--
-		} else {
-			proba := (int16)((uint64(cnt) * step) >> scale)
-			if proba < 8 {
-				restToBeat := vStep * uint64(rtbTable[proba])
-				v := uint64(cnt)*step - (uint64(proba) << scale)
-				if v > restToBeat {
-					proba++
-				}
-			}
-			if proba > largestP {
-				largestP = proba
-				largest = i
-			}
-			s.norm[i] = proba
-			stillToDistribute -= proba
-		}
-	}
-
-	if -stillToDistribute >= (s.norm[largest] >> 1) {
-		// corner case, need another normalization method
-		return s.normalizeCount2()
-	}
-	s.norm[largest] += stillToDistribute
-	return nil
-}
-
-// Secondary normalization method.
-// To be used when primary method fails.
-func (s *Scratch) normalizeCount2() error {
-	const notYetAssigned = -2
-	var (
-		distributed  uint32
-		total        = uint32(s.br.remain())
-		tableLog     = s.actualTableLog
-		lowThreshold = total >> tableLog
-		lowOne       = (total * 3) >> (tableLog + 1)
-	)
-	for i, cnt := range s.count[:s.symbolLen] {
-		if cnt == 0 {
-			s.norm[i] = 0
-			continue
-		}
-		if cnt <= lowThreshold {
-			s.norm[i] = -1
-			distributed++
-			total -= cnt
-			continue
-		}
-		if cnt <= lowOne {
-			s.norm[i] = 1
-			distributed++
-			total -= cnt
-			continue
-		}
-		s.norm[i] = notYetAssigned
-	}
-	toDistribute := (1 << tableLog) - distributed
-
-	if (total / toDistribute) > lowOne {
-		// risk of rounding to zero
-		lowOne = (total * 3) / (toDistribute * 2)
-		for i, cnt := range s.count[:s.symbolLen] {
-			if (s.norm[i] == notYetAssigned) && (cnt <= lowOne) {
-				s.norm[i] = 1
-				distributed++
-				total -= cnt
-				continue
-			}
-		}
-		toDistribute = (1 << tableLog) - distributed
-	}
-	if distributed == uint32(s.symbolLen)+1 {
-		// all values are pretty poor;
-		//   probably incompressible data (should have already been detected);
-		//   find max, then give all remaining points to max
-		var maxV int
-		var maxC uint32
-		for i, cnt := range s.count[:s.symbolLen] {
-			if cnt > maxC {
-				maxV = i
-				maxC = cnt
-			}
-		}
-		s.norm[maxV] += int16(toDistribute)
-		return nil
-	}
-
-	if total == 0 {
-		// all of the symbols were low enough for the lowOne or lowThreshold
-		for i := uint32(0); toDistribute > 0; i = (i + 1) % (uint32(s.symbolLen)) {
-			if s.norm[i] > 0 {
-				toDistribute--
-				s.norm[i]++
-			}
-		}
-		return nil
-	}
-
-	var (
-		vStepLog = 62 - uint64(tableLog)
-		mid      = uint64((1 << (vStepLog - 1)) - 1)
-		rStep    = (((1 << vStepLog) * uint64(toDistribute)) + mid) / uint64(total) // scale on remaining
-		tmpTotal = mid
-	)
-	for i, cnt := range s.count[:s.symbolLen] {
-		if s.norm[i] == notYetAssigned {
-			var (
-				end    = tmpTotal + uint64(cnt)*rStep
-				sStart = uint32(tmpTotal >> vStepLog)
-				sEnd   = uint32(end >> vStepLog)
-				weight = sEnd - sStart
-			)
-			if weight < 1 {
-				return errors.New("weight < 1")
-			}
-			s.norm[i] = int16(weight)
-			tmpTotal = end
-		}
-	}
-	return nil
-}
-
-// validateNorm validates the normalized histogram table.
-func (s *Scratch) validateNorm() (err error) {
-	var total int
-	for _, v := range s.norm[:s.symbolLen] {
-		if v >= 0 {
-			total += int(v)
-		} else {
-			total -= int(v)
-		}
-	}
-	defer func() {
-		if err == nil {
-			return
-		}
-		fmt.Printf("selected TableLog: %d, Symbol length: %d\n", s.actualTableLog, s.symbolLen)
-		for i, v := range s.norm[:s.symbolLen] {
-			fmt.Printf("%3d: %5d -> %4d \n", i, s.count[i], v)
-		}
-	}()
-	if total != (1 << s.actualTableLog) {
-		return fmt.Errorf("warning: Total == %d != %d", total, 1<<s.actualTableLog)
-	}
-	for i, v := range s.count[s.symbolLen:] {
-		if v != 0 {
-			return fmt.Errorf("warning: Found symbol out of range, %d after cut", i)
-		}
-	}
-	return nil
-}
diff --git a/vendor/github.com/klauspost/compress/fse/decompress.go b/vendor/github.com/klauspost/compress/fse/decompress.go
deleted file mode 100644
index 926f5f1535..0000000000
--- a/vendor/github.com/klauspost/compress/fse/decompress.go
+++ /dev/null
@@ -1,374 +0,0 @@
-package fse
-
-import (
-	"errors"
-	"fmt"
-)
-
-const (
-	tablelogAbsoluteMax = 15
-)
-
-// Decompress a block of data.
-// You can provide a scratch buffer to avoid allocations.
-// If nil is provided a temporary one will be allocated.
-// It is possible, but by no way guaranteed that corrupt data will
-// return an error.
-// It is up to the caller to verify integrity of the returned data.
-// Use a predefined Scrach to set maximum acceptable output size.
-func Decompress(b []byte, s *Scratch) ([]byte, error) {
-	s, err := s.prepare(b)
-	if err != nil {
-		return nil, err
-	}
-	s.Out = s.Out[:0]
-	err = s.readNCount()
-	if err != nil {
-		return nil, err
-	}
-	err = s.buildDtable()
-	if err != nil {
-		return nil, err
-	}
-	err = s.decompress()
-	if err != nil {
-		return nil, err
-	}
-
-	return s.Out, nil
-}
-
-// readNCount will read the symbol distribution so decoding tables can be constructed.
-func (s *Scratch) readNCount() error {
-	var (
-		charnum   uint16
-		previous0 bool
-		b         = &s.br
-	)
-	iend := b.remain()
-	if iend < 4 {
-		return errors.New("input too small")
-	}
-	bitStream := b.Uint32()
-	nbBits := uint((bitStream & 0xF) + minTablelog) // extract tableLog
-	if nbBits > tablelogAbsoluteMax {
-		return errors.New("tableLog too large")
-	}
-	bitStream >>= 4
-	bitCount := uint(4)
-
-	s.actualTableLog = uint8(nbBits)
-	remaining := int32((1 << nbBits) + 1)
-	threshold := int32(1 << nbBits)
-	gotTotal := int32(0)
-	nbBits++
-
-	for remaining > 1 {
-		if previous0 {
-			n0 := charnum
-			for (bitStream & 0xFFFF) == 0xFFFF {
-				n0 += 24
-				if b.off < iend-5 {
-					b.advance(2)
-					bitStream = b.Uint32() >> bitCount
-				} else {
-					bitStream >>= 16
-					bitCount += 16
-				}
-			}
-			for (bitStream & 3) == 3 {
-				n0 += 3
-				bitStream >>= 2
-				bitCount += 2
-			}
-			n0 += uint16(bitStream & 3)
-			bitCount += 2
-			if n0 > maxSymbolValue {
-				return errors.New("maxSymbolValue too small")
-			}
-			for charnum < n0 {
-				s.norm[charnum&0xff] = 0
-				charnum++
-			}
-
-			if b.off <= iend-7 || b.off+int(bitCount>>3) <= iend-4 {
-				b.advance(bitCount >> 3)
-				bitCount &= 7
-				bitStream = b.Uint32() >> bitCount
-			} else {
-				bitStream >>= 2
-			}
-		}
-
-		max := (2*(threshold) - 1) - (remaining)
-		var count int32
-
-		if (int32(bitStream) & (threshold - 1)) < max {
-			count = int32(bitStream) & (threshold - 1)
-			bitCount += nbBits - 1
-		} else {
-			count = int32(bitStream) & (2*threshold - 1)
-			if count >= threshold {
-				count -= max
-			}
-			bitCount += nbBits
-		}
-
-		count-- // extra accuracy
-		if count < 0 {
-			// -1 means +1
-			remaining += count
-			gotTotal -= count
-		} else {
-			remaining -= count
-			gotTotal += count
-		}
-		s.norm[charnum&0xff] = int16(count)
-		charnum++
-		previous0 = count == 0
-		for remaining < threshold {
-			nbBits--
-			threshold >>= 1
-		}
-		if b.off <= iend-7 || b.off+int(bitCount>>3) <= iend-4 {
-			b.advance(bitCount >> 3)
-			bitCount &= 7
-		} else {
-			bitCount -= (uint)(8 * (len(b.b) - 4 - b.off))
-			b.off = len(b.b) - 4
-		}
-		bitStream = b.Uint32() >> (bitCount & 31)
-	}
-	s.symbolLen = charnum
-
-	if s.symbolLen <= 1 {
-		return fmt.Errorf("symbolLen (%d) too small", s.symbolLen)
-	}
-	if s.symbolLen > maxSymbolValue+1 {
-		return fmt.Errorf("symbolLen (%d) too big", s.symbolLen)
-	}
-	if remaining != 1 {
-		return fmt.Errorf("corruption detected (remaining %d != 1)", remaining)
-	}
-	if bitCount > 32 {
-		return fmt.Errorf("corruption detected (bitCount %d > 32)", bitCount)
-	}
-	if gotTotal != 1<<s.actualTableLog {
-		return fmt.Errorf("corruption detected (total %d != %d)", gotTotal, 1<<s.actualTableLog)
-	}
-	b.advance((bitCount + 7) >> 3)
-	return nil
-}
-
-// decSymbol contains information about a state entry,
-// Including the state offset base, the output symbol and
-// the number of bits to read for the low part of the destination state.
-type decSymbol struct {
-	newState uint16
-	symbol   uint8
-	nbBits   uint8
-}
-
-// allocDtable will allocate decoding tables if they are not big enough.
-func (s *Scratch) allocDtable() {
-	tableSize := 1 << s.actualTableLog
-	if cap(s.decTable) < tableSize {
-		s.decTable = make([]decSymbol, tableSize)
-	}
-	s.decTable = s.decTable[:tableSize]
-
-	if cap(s.ct.tableSymbol) < 256 {
-		s.ct.tableSymbol = make([]byte, 256)
-	}
-	s.ct.tableSymbol = s.ct.tableSymbol[:256]
-
-	if cap(s.ct.stateTable) < 256 {
-		s.ct.stateTable = make([]uint16, 256)
-	}
-	s.ct.stateTable = s.ct.stateTable[:256]
-}
-
-// buildDtable will build the decoding table.
-func (s *Scratch) buildDtable() error {
-	tableSize := uint32(1 << s.actualTableLog)
-	highThreshold := tableSize - 1
-	s.allocDtable()
-	symbolNext := s.ct.stateTable[:256]
-
-	// Init, lay down lowprob symbols
-	s.zeroBits = false
-	{
-		largeLimit := int16(1 << (s.actualTableLog - 1))
-		for i, v := range s.norm[:s.symbolLen] {
-			if v == -1 {
-				s.decTable[highThreshold].symbol = uint8(i)
-				highThreshold--
-				symbolNext[i] = 1
-			} else {
-				if v >= largeLimit {
-					s.zeroBits = true
-				}
-				symbolNext[i] = uint16(v)
-			}
-		}
-	}
-	// Spread symbols
-	{
-		tableMask := tableSize - 1
-		step := tableStep(tableSize)
-		position := uint32(0)
-		for ss, v := range s.norm[:s.symbolLen] {
-			for i := 0; i < int(v); i++ {
-				s.decTable[position].symbol = uint8(ss)
-				position = (position + step) & tableMask
-				for position > highThreshold {
-					// lowprob area
-					position = (position + step) & tableMask
-				}
-			}
-		}
-		if position != 0 {
-			// position must reach all cells once, otherwise normalizedCounter is incorrect
-			return errors.New("corrupted input (position != 0)")
-		}
-	}
-
-	// Build Decoding table
-	{
-		tableSize := uint16(1 << s.actualTableLog)
-		for u, v := range s.decTable {
-			symbol := v.symbol
-			nextState := symbolNext[symbol]
-			symbolNext[symbol] = nextState + 1
-			nBits := s.actualTableLog - byte(highBits(uint32(nextState)))
-			s.decTable[u].nbBits = nBits
-			newState := (nextState << nBits) - tableSize
-			if newState >= tableSize {
-				return fmt.Errorf("newState (%d) outside table size (%d)", newState, tableSize)
-			}
-			if newState == uint16(u) && nBits == 0 {
-				// Seems weird that this is possible with nbits > 0.
-				return fmt.Errorf("newState (%d) == oldState (%d) and no bits", newState, u)
-			}
-			s.decTable[u].newState = newState
-		}
-	}
-	return nil
-}
-
-// decompress will decompress the bitstream.
-// If the buffer is over-read an error is returned.
-func (s *Scratch) decompress() error {
-	br := &s.bits
-	br.init(s.br.unread())
-
-	var s1, s2 decoder
-	// Initialize and decode first state and symbol.
-	s1.init(br, s.decTable, s.actualTableLog)
-	s2.init(br, s.decTable, s.actualTableLog)
-
-	// Use temp table to avoid bound checks/append penalty.
-	var tmp = s.ct.tableSymbol[:256]
-	var off uint8
-
-	// Main part
-	if !s.zeroBits {
-		for br.off >= 8 {
-			br.fillFast()
-			tmp[off+0] = s1.nextFast()
-			tmp[off+1] = s2.nextFast()
-			br.fillFast()
-			tmp[off+2] = s1.nextFast()
-			tmp[off+3] = s2.nextFast()
-			off += 4
-			// When off is 0, we have overflowed and should write.
-			if off == 0 {
-				s.Out = append(s.Out, tmp...)
-				if len(s.Out) >= s.DecompressLimit {
-					return fmt.Errorf("output size (%d) > DecompressLimit (%d)", len(s.Out), s.DecompressLimit)
-				}
-			}
-		}
-	} else {
-		for br.off >= 8 {
-			br.fillFast()
-			tmp[off+0] = s1.next()
-			tmp[off+1] = s2.next()
-			br.fillFast()
-			tmp[off+2] = s1.next()
-			tmp[off+3] = s2.next()
-			off += 4
-			if off == 0 {
-				s.Out = append(s.Out, tmp...)
-				// When off is 0, we have overflowed and should write.
-				if len(s.Out) >= s.DecompressLimit {
-					return fmt.Errorf("output size (%d) > DecompressLimit (%d)", len(s.Out), s.DecompressLimit)
-				}
-			}
-		}
-	}
-	s.Out = append(s.Out, tmp[:off]...)
-
-	// Final bits, a bit more expensive check
-	for {
-		if s1.finished() {
-			s.Out = append(s.Out, s1.final(), s2.final())
-			break
-		}
-		br.fill()
-		s.Out = append(s.Out, s1.next())
-		if s2.finished() {
-			s.Out = append(s.Out, s2.final(), s1.final())
-			break
-		}
-		s.Out = append(s.Out, s2.next())
-		if len(s.Out) >= s.DecompressLimit {
-			return fmt.Errorf("output size (%d) > DecompressLimit (%d)", len(s.Out), s.DecompressLimit)
-		}
-	}
-	return br.close()
-}
-
-// decoder keeps track of the current state and updates it from the bitstream.
-type decoder struct {
-	state uint16
-	br    *bitReader
-	dt    []decSymbol
-}
-
-// init will initialize the decoder and read the first state from the stream.
-func (d *decoder) init(in *bitReader, dt []decSymbol, tableLog uint8) {
-	d.dt = dt
-	d.br = in
-	d.state = in.getBits(tableLog)
-}
-
-// next returns the next symbol and sets the next state.
-// At least tablelog bits must be available in the bit reader.
-func (d *decoder) next() uint8 {
-	n := &d.dt[d.state]
-	lowBits := d.br.getBits(n.nbBits)
-	d.state = n.newState + lowBits
-	return n.symbol
-}
-
-// finished returns true if all bits have been read from the bitstream
-// and the next state would require reading bits from the input.
-func (d *decoder) finished() bool {
-	return d.br.finished() && d.dt[d.state].nbBits > 0
-}
-
-// final returns the current state symbol without decoding the next.
-func (d *decoder) final() uint8 {
-	return d.dt[d.state].symbol
-}
-
-// nextFast returns the next symbol and sets the next state.
-// This can only be used if no symbols are 0 bits.
-// At least tablelog bits must be available in the bit reader.
-func (d *decoder) nextFast() uint8 {
-	n := d.dt[d.state]
-	lowBits := d.br.getBitsFast(n.nbBits)
-	d.state = n.newState + lowBits
-	return n.symbol
-}
diff --git a/vendor/github.com/klauspost/compress/fse/fse.go b/vendor/github.com/klauspost/compress/fse/fse.go
deleted file mode 100644
index 535cbadfde..0000000000
--- a/vendor/github.com/klauspost/compress/fse/fse.go
+++ /dev/null
@@ -1,144 +0,0 @@
-// Copyright 2018 Klaus Post. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-// Based on work Copyright (c) 2013, Yann Collet, released under BSD License.
-
-// Package fse provides Finite State Entropy encoding and decoding.
-//
-// Finite State Entropy encoding provides a fast near-optimal symbol encoding/decoding
-// for byte blocks as implemented in zstd.
-//
-// See https://github.com/klauspost/compress/tree/master/fse for more information.
-package fse
-
-import (
-	"errors"
-	"fmt"
-	"math/bits"
-)
-
-const (
-	/*!MEMORY_USAGE :
-	 *  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
-	 *  Increasing memory usage improves compression ratio
-	 *  Reduced memory usage can improve speed, due to cache effect
-	 *  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
-	maxMemoryUsage     = 14
-	defaultMemoryUsage = 13
-
-	maxTableLog     = maxMemoryUsage - 2
-	maxTablesize    = 1 << maxTableLog
-	defaultTablelog = defaultMemoryUsage - 2
-	minTablelog     = 5
-	maxSymbolValue  = 255
-)
-
-var (
-	// ErrIncompressible is returned when input is judged to be too hard to compress.
-	ErrIncompressible = errors.New("input is not compressible")
-
-	// ErrUseRLE is returned from the compressor when the input is a single byte value repeated.
-	ErrUseRLE = errors.New("input is single value repeated")
-)
-
-// Scratch provides temporary storage for compression and decompression.
-type Scratch struct {
-	// Private
-	count    [maxSymbolValue + 1]uint32
-	norm     [maxSymbolValue + 1]int16
-	br       byteReader
-	bits     bitReader
-	bw       bitWriter
-	ct       cTable      // Compression tables.
-	decTable []decSymbol // Decompression table.
-	maxCount int         // count of the most probable symbol
-
-	// Per block parameters.
-	// These can be used to override compression parameters of the block.
-	// Do not touch, unless you know what you are doing.
-
-	// Out is output buffer.
-	// If the scratch is re-used before the caller is done processing the output,
-	// set this field to nil.
-	// Otherwise the output buffer will be re-used for next Compression/Decompression step
-	// and allocation will be avoided.
-	Out []byte
-
-	// DecompressLimit limits the maximum decoded size acceptable.
-	// If > 0 decompression will stop when approximately this many bytes
-	// has been decoded.
-	// If 0, maximum size will be 2GB.
-	DecompressLimit int
-
-	symbolLen      uint16 // Length of active part of the symbol table.
-	actualTableLog uint8  // Selected tablelog.
-	zeroBits       bool   // no bits has prob > 50%.
-	clearCount     bool   // clear count
-
-	// MaxSymbolValue will override the maximum symbol value of the next block.
-	MaxSymbolValue uint8
-
-	// TableLog will attempt to override the tablelog for the next block.
-	TableLog uint8
-}
-
-// Histogram allows to populate the histogram and skip that step in the compression,
-// It otherwise allows to inspect the histogram when compression is done.
-// To indicate that you have populated the histogram call HistogramFinished
-// with the value of the highest populated symbol, as well as the number of entries
-// in the most populated entry. These are accepted at face value.
-// The returned slice will always be length 256.
-func (s *Scratch) Histogram() []uint32 {
-	return s.count[:]
-}
-
-// HistogramFinished can be called to indicate that the histogram has been populated.
-// maxSymbol is the index of the highest set symbol of the next data segment.
-// maxCount is the number of entries in the most populated entry.
-// These are accepted at face value.
-func (s *Scratch) HistogramFinished(maxSymbol uint8, maxCount int) {
-	s.maxCount = maxCount
-	s.symbolLen = uint16(maxSymbol) + 1
-	s.clearCount = maxCount != 0
-}
-
-// prepare will prepare and allocate scratch tables used for both compression and decompression.
-func (s *Scratch) prepare(in []byte) (*Scratch, error) {
-	if s == nil {
-		s = &Scratch{}
-	}
-	if s.MaxSymbolValue == 0 {
-		s.MaxSymbolValue = 255
-	}
-	if s.TableLog == 0 {
-		s.TableLog = defaultTablelog
-	}
-	if s.TableLog > maxTableLog {
-		return nil, fmt.Errorf("tableLog (%d) > maxTableLog (%d)", s.TableLog, maxTableLog)
-	}
-	if cap(s.Out) == 0 {
-		s.Out = make([]byte, 0, len(in))
-	}
-	if s.clearCount && s.maxCount == 0 {
-		for i := range s.count {
-			s.count[i] = 0
-		}
-		s.clearCount = false
-	}
-	s.br.init(in)
-	if s.DecompressLimit == 0 {
-		// Max size 2GB.
-		s.DecompressLimit = (2 << 30) - 1
-	}
-
-	return s, nil
-}
-
-// tableStep returns the next table index.
-func tableStep(tableSize uint32) uint32 {
-	return (tableSize >> 1) + (tableSize >> 3) + 3
-}
-
-func highBits(val uint32) (n uint32) {
-	return uint32(bits.Len32(val) - 1)
-}
diff --git a/vendor/github.com/klauspost/compress/gzip/gunzip.go b/vendor/github.com/klauspost/compress/gzip/gunzip.go
deleted file mode 100644
index 21e768b360..0000000000
--- a/vendor/github.com/klauspost/compress/gzip/gunzip.go
+++ /dev/null
@@ -1,351 +0,0 @@
-// Copyright 2009 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// Package gzip implements reading and writing of gzip format compressed files,
-// as specified in RFC 1952.
-package gzip
-
-import (
-	"bufio"
-	"encoding/binary"
-	"errors"
-	"hash/crc32"
-	"io"
-	"time"
-
-	"github.com/klauspost/compress/flate"
-)
-
-const (
-	gzipID1     = 0x1f
-	gzipID2     = 0x8b
-	gzipDeflate = 8
-	flagText    = 1 << 0
-	flagHdrCrc  = 1 << 1
-	flagExtra   = 1 << 2
-	flagName    = 1 << 3
-	flagComment = 1 << 4
-)
-
-var (
-	// ErrChecksum is returned when reading GZIP data that has an invalid checksum.
-	ErrChecksum = errors.New("gzip: invalid checksum")
-	// ErrHeader is returned when reading GZIP data that has an invalid header.
-	ErrHeader = errors.New("gzip: invalid header")
-)
-
-var le = binary.LittleEndian
-
-// noEOF converts io.EOF to io.ErrUnexpectedEOF.
-func noEOF(err error) error {
-	if err == io.EOF {
-		return io.ErrUnexpectedEOF
-	}
-	return err
-}
-
-// The gzip file stores a header giving metadata about the compressed file.
-// That header is exposed as the fields of the Writer and Reader structs.
-//
-// Strings must be UTF-8 encoded and may only contain Unicode code points
-// U+0001 through U+00FF, due to limitations of the GZIP file format.
-type Header struct {
-	Comment string    // comment
-	Extra   []byte    // "extra data"
-	ModTime time.Time // modification time
-	Name    string    // file name
-	OS      byte      // operating system type
-}
-
-// A Reader is an io.Reader that can be read to retrieve
-// uncompressed data from a gzip-format compressed file.
-//
-// In general, a gzip file can be a concatenation of gzip files,
-// each with its own header. Reads from the Reader
-// return the concatenation of the uncompressed data of each.
-// Only the first header is recorded in the Reader fields.
-//
-// Gzip files store a length and checksum of the uncompressed data.
-// The Reader will return a ErrChecksum when Read
-// reaches the end of the uncompressed data if it does not
-// have the expected length or checksum. Clients should treat data
-// returned by Read as tentative until they receive the io.EOF
-// marking the end of the data.
-type Reader struct {
-	Header       // valid after NewReader or Reader.Reset
-	r            flate.Reader
-	br           *bufio.Reader
-	decompressor io.ReadCloser
-	digest       uint32 // CRC-32, IEEE polynomial (section 8)
-	size         uint32 // Uncompressed size (section 2.3.1)
-	buf          [512]byte
-	err          error
-	multistream  bool
-}
-
-// NewReader creates a new Reader reading the given reader.
-// If r does not also implement io.ByteReader,
-// the decompressor may read more data than necessary from r.
-//
-// It is the caller's responsibility to call Close on the Reader when done.
-//
-// The Reader.Header fields will be valid in the Reader returned.
-func NewReader(r io.Reader) (*Reader, error) {
-	z := new(Reader)
-	if err := z.Reset(r); err != nil {
-		return nil, err
-	}
-	return z, nil
-}
-
-// Reset discards the Reader z's state and makes it equivalent to the
-// result of its original state from NewReader, but reading from r instead.
-// This permits reusing a Reader rather than allocating a new one.
-func (z *Reader) Reset(r io.Reader) error {
-	*z = Reader{
-		decompressor: z.decompressor,
-		multistream:  true,
-	}
-	if rr, ok := r.(flate.Reader); ok {
-		z.r = rr
-	} else {
-		// Reuse if we can.
-		if z.br != nil {
-			z.br.Reset(r)
-		} else {
-			z.br = bufio.NewReader(r)
-		}
-		z.r = z.br
-	}
-	z.Header, z.err = z.readHeader()
-	return z.err
-}
-
-// Multistream controls whether the reader supports multistream files.
-//
-// If enabled (the default), the Reader expects the input to be a sequence
-// of individually gzipped data streams, each with its own header and
-// trailer, ending at EOF. The effect is that the concatenation of a sequence
-// of gzipped files is treated as equivalent to the gzip of the concatenation
-// of the sequence. This is standard behavior for gzip readers.
-//
-// Calling Multistream(false) disables this behavior; disabling the behavior
-// can be useful when reading file formats that distinguish individual gzip
-// data streams or mix gzip data streams with other data streams.
-// In this mode, when the Reader reaches the end of the data stream,
-// Read returns io.EOF. If the underlying reader implements io.ByteReader,
-// it will be left positioned just after the gzip stream.
-// To start the next stream, call z.Reset(r) followed by z.Multistream(false).
-// If there is no next stream, z.Reset(r) will return io.EOF.
-func (z *Reader) Multistream(ok bool) {
-	z.multistream = ok
-}
-
-// readString reads a NUL-terminated string from z.r.
-// It treats the bytes read as being encoded as ISO 8859-1 (Latin-1) and
-// will output a string encoded using UTF-8.
-// This method always updates z.digest with the data read.
-func (z *Reader) readString() (string, error) {
-	var err error
-	needConv := false
-	for i := 0; ; i++ {
-		if i >= len(z.buf) {
-			return "", ErrHeader
-		}
-		z.buf[i], err = z.r.ReadByte()
-		if err != nil {
-			return "", err
-		}
-		if z.buf[i] > 0x7f {
-			needConv = true
-		}
-		if z.buf[i] == 0 {
-			// Digest covers the NUL terminator.
-			z.digest = crc32.Update(z.digest, crc32.IEEETable, z.buf[:i+1])
-
-			// Strings are ISO 8859-1, Latin-1 (RFC 1952, section 2.3.1).
-			if needConv {
-				s := make([]rune, 0, i)
-				for _, v := range z.buf[:i] {
-					s = append(s, rune(v))
-				}
-				return string(s), nil
-			}
-			return string(z.buf[:i]), nil
-		}
-	}
-}
-
-// readHeader reads the GZIP header according to section 2.3.1.
-// This method does not set z.err.
-func (z *Reader) readHeader() (hdr Header, err error) {
-	if _, err = io.ReadFull(z.r, z.buf[:10]); err != nil {
-		// RFC 1952, section 2.2, says the following:
-		//	A gzip file consists of a series of "members" (compressed data sets).
-		//
-		// Other than this, the specification does not clarify whether a
-		// "series" is defined as "one or more" or "zero or more". To err on the
-		// side of caution, Go interprets this to mean "zero or more".
-		// Thus, it is okay to return io.EOF here.
-		return hdr, err
-	}
-	if z.buf[0] != gzipID1 || z.buf[1] != gzipID2 || z.buf[2] != gzipDeflate {
-		return hdr, ErrHeader
-	}
-	flg := z.buf[3]
-	hdr.ModTime = time.Unix(int64(le.Uint32(z.buf[4:8])), 0)
-	// z.buf[8] is XFL and is currently ignored.
-	hdr.OS = z.buf[9]
-	z.digest = crc32.ChecksumIEEE(z.buf[:10])
-
-	if flg&flagExtra != 0 {
-		if _, err = io.ReadFull(z.r, z.buf[:2]); err != nil {
-			return hdr, noEOF(err)
-		}
-		z.digest = crc32.Update(z.digest, crc32.IEEETable, z.buf[:2])
-		data := make([]byte, le.Uint16(z.buf[:2]))
-		if _, err = io.ReadFull(z.r, data); err != nil {
-			return hdr, noEOF(err)
-		}
-		z.digest = crc32.Update(z.digest, crc32.IEEETable, data)
-		hdr.Extra = data
-	}
-
-	var s string
-	if flg&flagName != 0 {
-		if s, err = z.readString(); err != nil {
-			return hdr, err
-		}
-		hdr.Name = s
-	}
-
-	if flg&flagComment != 0 {
-		if s, err = z.readString(); err != nil {
-			return hdr, err
-		}
-		hdr.Comment = s
-	}
-
-	if flg&flagHdrCrc != 0 {
-		if _, err = io.ReadFull(z.r, z.buf[:2]); err != nil {
-			return hdr, noEOF(err)
-		}
-		digest := le.Uint16(z.buf[:2])
-		if digest != uint16(z.digest) {
-			return hdr, ErrHeader
-		}
-	}
-
-	z.digest = 0
-	if z.decompressor == nil {
-		z.decompressor = flate.NewReader(z.r)
-	} else {
-		z.decompressor.(flate.Resetter).Reset(z.r, nil)
-	}
-	return hdr, nil
-}
-
-// Read implements io.Reader, reading uncompressed bytes from its underlying Reader.
-func (z *Reader) Read(p []byte) (n int, err error) {
-	if z.err != nil {
-		return 0, z.err
-	}
-
-	n, z.err = z.decompressor.Read(p)
-	z.digest = crc32.Update(z.digest, crc32.IEEETable, p[:n])
-	z.size += uint32(n)
-	if z.err != io.EOF {
-		// In the normal case we return here.
-		return n, z.err
-	}
-
-	// Finished file; check checksum and size.
-	if _, err := io.ReadFull(z.r, z.buf[:8]); err != nil {
-		z.err = noEOF(err)
-		return n, z.err
-	}
-	digest := le.Uint32(z.buf[:4])
-	size := le.Uint32(z.buf[4:8])
-	if digest != z.digest || size != z.size {
-		z.err = ErrChecksum
-		return n, z.err
-	}
-	z.digest, z.size = 0, 0
-
-	// File is ok; check if there is another.
-	if !z.multistream {
-		return n, io.EOF
-	}
-	z.err = nil // Remove io.EOF
-
-	if _, z.err = z.readHeader(); z.err != nil {
-		return n, z.err
-	}
-
-	// Read from next file, if necessary.
-	if n > 0 {
-		return n, nil
-	}
-	return z.Read(p)
-}
-
-// Support the io.WriteTo interface for io.Copy and friends.
-func (z *Reader) WriteTo(w io.Writer) (int64, error) {
-	total := int64(0)
-	crcWriter := crc32.NewIEEE()
-	for {
-		if z.err != nil {
-			if z.err == io.EOF {
-				return total, nil
-			}
-			return total, z.err
-		}
-
-		// We write both to output and digest.
-		mw := io.MultiWriter(w, crcWriter)
-		n, err := z.decompressor.(io.WriterTo).WriteTo(mw)
-		total += n
-		z.size += uint32(n)
-		if err != nil {
-			z.err = err
-			return total, z.err
-		}
-
-		// Finished file; check checksum + size.
-		if _, err := io.ReadFull(z.r, z.buf[0:8]); err != nil {
-			if err == io.EOF {
-				err = io.ErrUnexpectedEOF
-			}
-			z.err = err
-			return total, err
-		}
-		z.digest = crcWriter.Sum32()
-		digest := le.Uint32(z.buf[:4])
-		size := le.Uint32(z.buf[4:8])
-		if digest != z.digest || size != z.size {
-			z.err = ErrChecksum
-			return total, z.err
-		}
-		z.digest, z.size = 0, 0
-
-		// File is ok; check if there is another.
-		if !z.multistream {
-			return total, nil
-		}
-		crcWriter.Reset()
-		z.err = nil // Remove io.EOF
-
-		if _, z.err = z.readHeader(); z.err != nil {
-			if z.err == io.EOF {
-				return total, nil
-			}
-			return total, z.err
-		}
-	}
-}
-
-// Close closes the Reader. It does not close the underlying io.Reader.
-// In order for the GZIP checksum to be verified, the reader must be
-// fully consumed until the io.EOF.
-func (z *Reader) Close() error { return z.decompressor.Close() }
diff --git a/vendor/github.com/klauspost/compress/gzip/gzip.go b/vendor/github.com/klauspost/compress/gzip/gzip.go
deleted file mode 100644
index 26203851bd..0000000000
--- a/vendor/github.com/klauspost/compress/gzip/gzip.go
+++ /dev/null
@@ -1,269 +0,0 @@
-// Copyright 2010 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package gzip
-
-import (
-	"errors"
-	"fmt"
-	"hash/crc32"
-	"io"
-
-	"github.com/klauspost/compress/flate"
-)
-
-// These constants are copied from the flate package, so that code that imports
-// "compress/gzip" does not also have to import "compress/flate".
-const (
-	NoCompression       = flate.NoCompression
-	BestSpeed           = flate.BestSpeed
-	BestCompression     = flate.BestCompression
-	DefaultCompression  = flate.DefaultCompression
-	ConstantCompression = flate.ConstantCompression
-	HuffmanOnly         = flate.HuffmanOnly
-
-	// StatelessCompression will do compression but without maintaining any state
-	// between Write calls.
-	// There will be no memory kept between Write calls,
-	// but compression and speed will be suboptimal.
-	// Because of this, the size of actual Write calls will affect output size.
-	StatelessCompression = -3
-)
-
-// A Writer is an io.WriteCloser.
-// Writes to a Writer are compressed and written to w.
-type Writer struct {
-	Header      // written at first call to Write, Flush, or Close
-	w           io.Writer
-	level       int
-	err         error
-	compressor  *flate.Writer
-	digest      uint32 // CRC-32, IEEE polynomial (section 8)
-	size        uint32 // Uncompressed size (section 2.3.1)
-	wroteHeader bool
-	closed      bool
-	buf         [10]byte
-}
-
-// NewWriter returns a new Writer.
-// Writes to the returned writer are compressed and written to w.
-//
-// It is the caller's responsibility to call Close on the WriteCloser when done.
-// Writes may be buffered and not flushed until Close.
-//
-// Callers that wish to set the fields in Writer.Header must do so before
-// the first call to Write, Flush, or Close.
-func NewWriter(w io.Writer) *Writer {
-	z, _ := NewWriterLevel(w, DefaultCompression)
-	return z
-}
-
-// NewWriterLevel is like NewWriter but specifies the compression level instead
-// of assuming DefaultCompression.
-//
-// The compression level can be DefaultCompression, NoCompression, or any
-// integer value between BestSpeed and BestCompression inclusive. The error
-// returned will be nil if the level is valid.
-func NewWriterLevel(w io.Writer, level int) (*Writer, error) {
-	if level < StatelessCompression || level > BestCompression {
-		return nil, fmt.Errorf("gzip: invalid compression level: %d", level)
-	}
-	z := new(Writer)
-	z.init(w, level)
-	return z, nil
-}
-
-func (z *Writer) init(w io.Writer, level int) {
-	compressor := z.compressor
-	if level != StatelessCompression {
-		if compressor != nil {
-			compressor.Reset(w)
-		}
-	}
-
-	*z = Writer{
-		Header: Header{
-			OS: 255, // unknown
-		},
-		w:          w,
-		level:      level,
-		compressor: compressor,
-	}
-}
-
-// Reset discards the Writer z's state and makes it equivalent to the
-// result of its original state from NewWriter or NewWriterLevel, but
-// writing to w instead. This permits reusing a Writer rather than
-// allocating a new one.
-func (z *Writer) Reset(w io.Writer) {
-	z.init(w, z.level)
-}
-
-// writeBytes writes a length-prefixed byte slice to z.w.
-func (z *Writer) writeBytes(b []byte) error {
-	if len(b) > 0xffff {
-		return errors.New("gzip.Write: Extra data is too large")
-	}
-	le.PutUint16(z.buf[:2], uint16(len(b)))
-	_, err := z.w.Write(z.buf[:2])
-	if err != nil {
-		return err
-	}
-	_, err = z.w.Write(b)
-	return err
-}
-
-// writeString writes a UTF-8 string s in GZIP's format to z.w.
-// GZIP (RFC 1952) specifies that strings are NUL-terminated ISO 8859-1 (Latin-1).
-func (z *Writer) writeString(s string) (err error) {
-	// GZIP stores Latin-1 strings; error if non-Latin-1; convert if non-ASCII.
-	needconv := false
-	for _, v := range s {
-		if v == 0 || v > 0xff {
-			return errors.New("gzip.Write: non-Latin-1 header string")
-		}
-		if v > 0x7f {
-			needconv = true
-		}
-	}
-	if needconv {
-		b := make([]byte, 0, len(s))
-		for _, v := range s {
-			b = append(b, byte(v))
-		}
-		_, err = z.w.Write(b)
-	} else {
-		_, err = io.WriteString(z.w, s)
-	}
-	if err != nil {
-		return err
-	}
-	// GZIP strings are NUL-terminated.
-	z.buf[0] = 0
-	_, err = z.w.Write(z.buf[:1])
-	return err
-}
-
-// Write writes a compressed form of p to the underlying io.Writer. The
-// compressed bytes are not necessarily flushed until the Writer is closed.
-func (z *Writer) Write(p []byte) (int, error) {
-	if z.err != nil {
-		return 0, z.err
-	}
-	var n int
-	// Write the GZIP header lazily.
-	if !z.wroteHeader {
-		z.wroteHeader = true
-		z.buf[0] = gzipID1
-		z.buf[1] = gzipID2
-		z.buf[2] = gzipDeflate
-		z.buf[3] = 0
-		if z.Extra != nil {
-			z.buf[3] |= 0x04
-		}
-		if z.Name != "" {
-			z.buf[3] |= 0x08
-		}
-		if z.Comment != "" {
-			z.buf[3] |= 0x10
-		}
-		le.PutUint32(z.buf[4:8], uint32(z.ModTime.Unix()))
-		if z.level == BestCompression {
-			z.buf[8] = 2
-		} else if z.level == BestSpeed {
-			z.buf[8] = 4
-		} else {
-			z.buf[8] = 0
-		}
-		z.buf[9] = z.OS
-		n, z.err = z.w.Write(z.buf[:10])
-		if z.err != nil {
-			return n, z.err
-		}
-		if z.Extra != nil {
-			z.err = z.writeBytes(z.Extra)
-			if z.err != nil {
-				return n, z.err
-			}
-		}
-		if z.Name != "" {
-			z.err = z.writeString(z.Name)
-			if z.err != nil {
-				return n, z.err
-			}
-		}
-		if z.Comment != "" {
-			z.err = z.writeString(z.Comment)
-			if z.err != nil {
-				return n, z.err
-			}
-		}
-
-		if z.compressor == nil && z.level != StatelessCompression {
-			z.compressor, _ = flate.NewWriter(z.w, z.level)
-		}
-	}
-	z.size += uint32(len(p))
-	z.digest = crc32.Update(z.digest, crc32.IEEETable, p)
-	if z.level == StatelessCompression {
-		return len(p), flate.StatelessDeflate(z.w, p, false, nil)
-	}
-	n, z.err = z.compressor.Write(p)
-	return n, z.err
-}
-
-// Flush flushes any pending compressed data to the underlying writer.
-//
-// It is useful mainly in compressed network protocols, to ensure that
-// a remote reader has enough data to reconstruct a packet. Flush does
-// not return until the data has been written. If the underlying
-// writer returns an error, Flush returns that error.
-//
-// In the terminology of the zlib library, Flush is equivalent to Z_SYNC_FLUSH.
-func (z *Writer) Flush() error {
-	if z.err != nil {
-		return z.err
-	}
-	if z.closed || z.level == StatelessCompression {
-		return nil
-	}
-	if !z.wroteHeader {
-		z.Write(nil)
-		if z.err != nil {
-			return z.err
-		}
-	}
-	z.err = z.compressor.Flush()
-	return z.err
-}
-
-// Close closes the Writer, flushing any unwritten data to the underlying
-// io.Writer, but does not close the underlying io.Writer.
-func (z *Writer) Close() error {
-	if z.err != nil {
-		return z.err
-	}
-	if z.closed {
-		return nil
-	}
-	z.closed = true
-	if !z.wroteHeader {
-		z.Write(nil)
-		if z.err != nil {
-			return z.err
-		}
-	}
-	if z.level == StatelessCompression {
-		z.err = flate.StatelessDeflate(z.w, nil, true, nil)
-	} else {
-		z.err = z.compressor.Close()
-	}
-	if z.err != nil {
-		return z.err
-	}
-	le.PutUint32(z.buf[:4], z.digest)
-	le.PutUint32(z.buf[4:8], z.size)
-	_, z.err = z.w.Write(z.buf[:8])
-	return z.err
-}
diff --git a/vendor/github.com/klauspost/compress/huff0/.gitignore b/vendor/github.com/klauspost/compress/huff0/.gitignore
deleted file mode 100644
index b3d262958f..0000000000
--- a/vendor/github.com/klauspost/compress/huff0/.gitignore
+++ /dev/null
@@ -1 +0,0 @@
-/huff0-fuzz.zip
diff --git a/vendor/github.com/klauspost/compress/huff0/README.md b/vendor/github.com/klauspost/compress/huff0/README.md
deleted file mode 100644
index 8b6e5c6638..0000000000
--- a/vendor/github.com/klauspost/compress/huff0/README.md
+++ /dev/null
@@ -1,89 +0,0 @@
-# Huff0 entropy compression

-

-This package provides Huff0 encoding and decoding as used in zstd.

-            

-[Huff0](https://github.com/Cyan4973/FiniteStateEntropy#new-generation-entropy-coders), 

-a Huffman codec designed for modern CPU, featuring OoO (Out of Order) operations on multiple ALU 

-(Arithmetic Logic Unit), achieving extremely fast compression and decompression speeds.

-

-This can be used for compressing input with a lot of similar input values to the smallest number of bytes.

-This does not perform any multi-byte [dictionary coding](https://en.wikipedia.org/wiki/Dictionary_coder) as LZ coders,

-but it can be used as a secondary step to compressors (like Snappy) that does not do entropy encoding. 

-

-* [Godoc documentation](https://godoc.org/github.com/klauspost/compress/huff0)

-

-## News

-

-This is used as part of the [zstandard](https://github.com/klauspost/compress/tree/master/zstd#zstd) compression and decompression package.

-

-This ensures that most functionality is well tested.

-

-# Usage

-

-This package provides a low level interface that allows to compress single independent blocks. 

-

-Each block is separate, and there is no built in integrity checks. 

-This means that the caller should keep track of block sizes and also do checksums if needed.  

-

-Compressing a block is done via the [`Compress1X`](https://godoc.org/github.com/klauspost/compress/huff0#Compress1X) and 

-[`Compress4X`](https://godoc.org/github.com/klauspost/compress/huff0#Compress4X) functions.

-You must provide input and will receive the output and maybe an error.

-

-These error values can be returned:

-

-| Error               | Description                                                                 |

-|---------------------|-----------------------------------------------------------------------------|

-| `<nil>`             | Everything ok, output is returned                                           |

-| `ErrIncompressible` | Returned when input is judged to be too hard to compress                    |

-| `ErrUseRLE`         | Returned from the compressor when the input is a single byte value repeated |

-| `ErrTooBig`         | Returned if the input block exceeds the maximum allowed size (128 Kib)      |

-| `(error)`           | An internal error occurred.                                                 |

-

-

-As can be seen above some of there are errors that will be returned even under normal operation so it is important to handle these.

-

-To reduce allocations you can provide a [`Scratch`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch) object 

-that can be re-used for successive calls. Both compression and decompression accepts a `Scratch` object, and the same 

-object can be used for both.   

-

-Be aware, that when re-using a `Scratch` object that the *output* buffer is also re-used, so if you are still using this

-you must set the `Out` field in the scratch to nil. The same buffer is used for compression and decompression output.

-

-The `Scratch` object will retain state that allows to re-use previous tables for encoding and decoding.  

-

-## Tables and re-use

-

-Huff0 allows for reusing tables from the previous block to save space if that is expected to give better/faster results. 

-

-The Scratch object allows you to set a [`ReusePolicy`](https://godoc.org/github.com/klauspost/compress/huff0#ReusePolicy) 

-that controls this behaviour. See the documentation for details. This can be altered between each block.

-

-Do however note that this information is *not* stored in the output block and it is up to the users of the package to

-record whether [`ReadTable`](https://godoc.org/github.com/klauspost/compress/huff0#ReadTable) should be called,

-based on the boolean reported back from the CompressXX call. 

-

-If you want to store the table separate from the data, you can access them as `OutData` and `OutTable` on the 

-[`Scratch`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch) object.

-

-## Decompressing

-

-The first part of decoding is to initialize the decoding table through [`ReadTable`](https://godoc.org/github.com/klauspost/compress/huff0#ReadTable).

-This will initialize the decoding tables. 

-You can supply the complete block to `ReadTable` and it will return the data part of the block 

-which can be given to the decompressor. 

-

-Decompressing is done by calling the [`Decompress1X`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch.Decompress1X) 

-or [`Decompress4X`](https://godoc.org/github.com/klauspost/compress/huff0#Scratch.Decompress4X) function.

-

-For concurrently decompressing content with a fixed table a stateless [`Decoder`](https://godoc.org/github.com/klauspost/compress/huff0#Decoder) can be requested which will remain correct as long as the scratch is unchanged. The capacity of the provided slice indicates the expected output size.

-

-You must provide the output from the compression stage, at exactly the size you got back. If you receive an error back

-your input was likely corrupted. 

-

-It is important to note that a successful decoding does *not* mean your output matches your original input. 

-There are no integrity checks, so relying on errors from the decompressor does not assure your data is valid.

-

-# Contributing

-

-Contributions are always welcome. Be aware that adding public functions will require good justification and breaking 

-changes will likely not be accepted. If in doubt open an issue before writing the PR.

diff --git a/vendor/github.com/klauspost/compress/huff0/bitreader.go b/vendor/github.com/klauspost/compress/huff0/bitreader.go
deleted file mode 100644
index a4979e8868..0000000000
--- a/vendor/github.com/klauspost/compress/huff0/bitreader.go
+++ /dev/null
@@ -1,329 +0,0 @@
-// Copyright 2018 Klaus Post. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-// Based on work Copyright (c) 2013, Yann Collet, released under BSD License.
-
-package huff0
-
-import (
-	"encoding/binary"
-	"errors"
-	"io"
-)
-
-// bitReader reads a bitstream in reverse.
-// The last set bit indicates the start of the stream and is used
-// for aligning the input.
-type bitReader struct {
-	in       []byte
-	off      uint // next byte to read is at in[off - 1]
-	value    uint64
-	bitsRead uint8
-}
-
-// init initializes and resets the bit reader.
-func (b *bitReader) init(in []byte) error {
-	if len(in) < 1 {
-		return errors.New("corrupt stream: too short")
-	}
-	b.in = in
-	b.off = uint(len(in))
-	// The highest bit of the last byte indicates where to start
-	v := in[len(in)-1]
-	if v == 0 {
-		return errors.New("corrupt stream, did not find end of stream")
-	}
-	b.bitsRead = 64
-	b.value = 0
-	if len(in) >= 8 {
-		b.fillFastStart()
-	} else {
-		b.fill()
-		b.fill()
-	}
-	b.bitsRead += 8 - uint8(highBit32(uint32(v)))
-	return nil
-}
-
-// peekBitsFast requires that at least one bit is requested every time.
-// There are no checks if the buffer is filled.
-func (b *bitReader) peekBitsFast(n uint8) uint16 {
-	const regMask = 64 - 1
-	v := uint16((b.value << (b.bitsRead & regMask)) >> ((regMask + 1 - n) & regMask))
-	return v
-}
-
-// fillFast() will make sure at least 32 bits are available.
-// There must be at least 4 bytes available.
-func (b *bitReader) fillFast() {
-	if b.bitsRead < 32 {
-		return
-	}
-
-	// 2 bounds checks.
-	v := b.in[b.off-4 : b.off]
-	v = v[:4]
-	low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-	b.value = (b.value << 32) | uint64(low)
-	b.bitsRead -= 32
-	b.off -= 4
-}
-
-func (b *bitReader) advance(n uint8) {
-	b.bitsRead += n
-}
-
-// fillFastStart() assumes the bitreader is empty and there is at least 8 bytes to read.
-func (b *bitReader) fillFastStart() {
-	// Do single re-slice to avoid bounds checks.
-	b.value = binary.LittleEndian.Uint64(b.in[b.off-8:])
-	b.bitsRead = 0
-	b.off -= 8
-}
-
-// fill() will make sure at least 32 bits are available.
-func (b *bitReader) fill() {
-	if b.bitsRead < 32 {
-		return
-	}
-	if b.off > 4 {
-		v := b.in[b.off-4:]
-		v = v[:4]
-		low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-		b.value = (b.value << 32) | uint64(low)
-		b.bitsRead -= 32
-		b.off -= 4
-		return
-	}
-	for b.off > 0 {
-		b.value = (b.value << 8) | uint64(b.in[b.off-1])
-		b.bitsRead -= 8
-		b.off--
-	}
-}
-
-// finished returns true if all bits have been read from the bit stream.
-func (b *bitReader) finished() bool {
-	return b.off == 0 && b.bitsRead >= 64
-}
-
-// close the bitstream and returns an error if out-of-buffer reads occurred.
-func (b *bitReader) close() error {
-	// Release reference.
-	b.in = nil
-	if b.bitsRead > 64 {
-		return io.ErrUnexpectedEOF
-	}
-	return nil
-}
-
-// bitReader reads a bitstream in reverse.
-// The last set bit indicates the start of the stream and is used
-// for aligning the input.
-type bitReaderBytes struct {
-	in       []byte
-	off      uint // next byte to read is at in[off - 1]
-	value    uint64
-	bitsRead uint8
-}
-
-// init initializes and resets the bit reader.
-func (b *bitReaderBytes) init(in []byte) error {
-	if len(in) < 1 {
-		return errors.New("corrupt stream: too short")
-	}
-	b.in = in
-	b.off = uint(len(in))
-	// The highest bit of the last byte indicates where to start
-	v := in[len(in)-1]
-	if v == 0 {
-		return errors.New("corrupt stream, did not find end of stream")
-	}
-	b.bitsRead = 64
-	b.value = 0
-	if len(in) >= 8 {
-		b.fillFastStart()
-	} else {
-		b.fill()
-		b.fill()
-	}
-	b.advance(8 - uint8(highBit32(uint32(v))))
-	return nil
-}
-
-// peekBitsFast requires that at least one bit is requested every time.
-// There are no checks if the buffer is filled.
-func (b *bitReaderBytes) peekByteFast() uint8 {
-	got := uint8(b.value >> 56)
-	return got
-}
-
-func (b *bitReaderBytes) advance(n uint8) {
-	b.bitsRead += n
-	b.value <<= n & 63
-}
-
-// fillFast() will make sure at least 32 bits are available.
-// There must be at least 4 bytes available.
-func (b *bitReaderBytes) fillFast() {
-	if b.bitsRead < 32 {
-		return
-	}
-
-	// 2 bounds checks.
-	v := b.in[b.off-4 : b.off]
-	v = v[:4]
-	low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-	b.value |= uint64(low) << (b.bitsRead - 32)
-	b.bitsRead -= 32
-	b.off -= 4
-}
-
-// fillFastStart() assumes the bitReaderBytes is empty and there is at least 8 bytes to read.
-func (b *bitReaderBytes) fillFastStart() {
-	// Do single re-slice to avoid bounds checks.
-	b.value = binary.LittleEndian.Uint64(b.in[b.off-8:])
-	b.bitsRead = 0
-	b.off -= 8
-}
-
-// fill() will make sure at least 32 bits are available.
-func (b *bitReaderBytes) fill() {
-	if b.bitsRead < 32 {
-		return
-	}
-	if b.off > 4 {
-		v := b.in[b.off-4:]
-		v = v[:4]
-		low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-		b.value |= uint64(low) << (b.bitsRead - 32)
-		b.bitsRead -= 32
-		b.off -= 4
-		return
-	}
-	for b.off > 0 {
-		b.value |= uint64(b.in[b.off-1]) << (b.bitsRead - 8)
-		b.bitsRead -= 8
-		b.off--
-	}
-}
-
-// finished returns true if all bits have been read from the bit stream.
-func (b *bitReaderBytes) finished() bool {
-	return b.off == 0 && b.bitsRead >= 64
-}
-
-// close the bitstream and returns an error if out-of-buffer reads occurred.
-func (b *bitReaderBytes) close() error {
-	// Release reference.
-	b.in = nil
-	if b.bitsRead > 64 {
-		return io.ErrUnexpectedEOF
-	}
-	return nil
-}
-
-// bitReaderShifted reads a bitstream in reverse.
-// The last set bit indicates the start of the stream and is used
-// for aligning the input.
-type bitReaderShifted struct {
-	in       []byte
-	off      uint // next byte to read is at in[off - 1]
-	value    uint64
-	bitsRead uint8
-}
-
-// init initializes and resets the bit reader.
-func (b *bitReaderShifted) init(in []byte) error {
-	if len(in) < 1 {
-		return errors.New("corrupt stream: too short")
-	}
-	b.in = in
-	b.off = uint(len(in))
-	// The highest bit of the last byte indicates where to start
-	v := in[len(in)-1]
-	if v == 0 {
-		return errors.New("corrupt stream, did not find end of stream")
-	}
-	b.bitsRead = 64
-	b.value = 0
-	if len(in) >= 8 {
-		b.fillFastStart()
-	} else {
-		b.fill()
-		b.fill()
-	}
-	b.advance(8 - uint8(highBit32(uint32(v))))
-	return nil
-}
-
-// peekBitsFast requires that at least one bit is requested every time.
-// There are no checks if the buffer is filled.
-func (b *bitReaderShifted) peekBitsFast(n uint8) uint16 {
-	return uint16(b.value >> ((64 - n) & 63))
-}
-
-func (b *bitReaderShifted) advance(n uint8) {
-	b.bitsRead += n
-	b.value <<= n & 63
-}
-
-// fillFast() will make sure at least 32 bits are available.
-// There must be at least 4 bytes available.
-func (b *bitReaderShifted) fillFast() {
-	if b.bitsRead < 32 {
-		return
-	}
-
-	// 2 bounds checks.
-	v := b.in[b.off-4 : b.off]
-	v = v[:4]
-	low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-	b.value |= uint64(low) << ((b.bitsRead - 32) & 63)
-	b.bitsRead -= 32
-	b.off -= 4
-}
-
-// fillFastStart() assumes the bitReaderShifted is empty and there is at least 8 bytes to read.
-func (b *bitReaderShifted) fillFastStart() {
-	// Do single re-slice to avoid bounds checks.
-	b.value = binary.LittleEndian.Uint64(b.in[b.off-8:])
-	b.bitsRead = 0
-	b.off -= 8
-}
-
-// fill() will make sure at least 32 bits are available.
-func (b *bitReaderShifted) fill() {
-	if b.bitsRead < 32 {
-		return
-	}
-	if b.off > 4 {
-		v := b.in[b.off-4:]
-		v = v[:4]
-		low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-		b.value |= uint64(low) << ((b.bitsRead - 32) & 63)
-		b.bitsRead -= 32
-		b.off -= 4
-		return
-	}
-	for b.off > 0 {
-		b.value |= uint64(b.in[b.off-1]) << ((b.bitsRead - 8) & 63)
-		b.bitsRead -= 8
-		b.off--
-	}
-}
-
-// finished returns true if all bits have been read from the bit stream.
-func (b *bitReaderShifted) finished() bool {
-	return b.off == 0 && b.bitsRead >= 64
-}
-
-// close the bitstream and returns an error if out-of-buffer reads occurred.
-func (b *bitReaderShifted) close() error {
-	// Release reference.
-	b.in = nil
-	if b.bitsRead > 64 {
-		return io.ErrUnexpectedEOF
-	}
-	return nil
-}
diff --git a/vendor/github.com/klauspost/compress/huff0/bitwriter.go b/vendor/github.com/klauspost/compress/huff0/bitwriter.go
deleted file mode 100644
index 6bce4e87d4..0000000000
--- a/vendor/github.com/klauspost/compress/huff0/bitwriter.go
+++ /dev/null
@@ -1,210 +0,0 @@
-// Copyright 2018 Klaus Post. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-// Based on work Copyright (c) 2013, Yann Collet, released under BSD License.
-
-package huff0
-
-import "fmt"
-
-// bitWriter will write bits.
-// First bit will be LSB of the first byte of output.
-type bitWriter struct {
-	bitContainer uint64
-	nBits        uint8
-	out          []byte
-}
-
-// bitMask16 is bitmasks. Has extra to avoid bounds check.
-var bitMask16 = [32]uint16{
-	0, 1, 3, 7, 0xF, 0x1F,
-	0x3F, 0x7F, 0xFF, 0x1FF, 0x3FF, 0x7FF,
-	0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, 0xFFFF,
-	0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF,
-	0xFFFF, 0xFFFF} /* up to 16 bits */
-
-// addBits16NC will add up to 16 bits.
-// It will not check if there is space for them,
-// so the caller must ensure that it has flushed recently.
-func (b *bitWriter) addBits16NC(value uint16, bits uint8) {
-	b.bitContainer |= uint64(value&bitMask16[bits&31]) << (b.nBits & 63)
-	b.nBits += bits
-}
-
-// addBits16Clean will add up to 16 bits. value may not contain more set bits than indicated.
-// It will not check if there is space for them, so the caller must ensure that it has flushed recently.
-func (b *bitWriter) addBits16Clean(value uint16, bits uint8) {
-	b.bitContainer |= uint64(value) << (b.nBits & 63)
-	b.nBits += bits
-}
-
-// encSymbol will add up to 16 bits. value may not contain more set bits than indicated.
-// It will not check if there is space for them, so the caller must ensure that it has flushed recently.
-func (b *bitWriter) encSymbol(ct cTable, symbol byte) {
-	enc := ct[symbol]
-	b.bitContainer |= uint64(enc.val) << (b.nBits & 63)
-	if false {
-		if enc.nBits == 0 {
-			panic("nbits 0")
-		}
-	}
-	b.nBits += enc.nBits
-}
-
-// encTwoSymbols will add up to 32 bits. value may not contain more set bits than indicated.
-// It will not check if there is space for them, so the caller must ensure that it has flushed recently.
-func (b *bitWriter) encTwoSymbols(ct cTable, av, bv byte) {
-	encA := ct[av]
-	encB := ct[bv]
-	sh := b.nBits & 63
-	combined := uint64(encA.val) | (uint64(encB.val) << (encA.nBits & 63))
-	b.bitContainer |= combined << sh
-	if false {
-		if encA.nBits == 0 {
-			panic("nbitsA 0")
-		}
-		if encB.nBits == 0 {
-			panic("nbitsB 0")
-		}
-	}
-	b.nBits += encA.nBits + encB.nBits
-}
-
-// addBits16ZeroNC will add up to 16 bits.
-// It will not check if there is space for them,
-// so the caller must ensure that it has flushed recently.
-// This is fastest if bits can be zero.
-func (b *bitWriter) addBits16ZeroNC(value uint16, bits uint8) {
-	if bits == 0 {
-		return
-	}
-	value <<= (16 - bits) & 15
-	value >>= (16 - bits) & 15
-	b.bitContainer |= uint64(value) << (b.nBits & 63)
-	b.nBits += bits
-}
-
-// flush will flush all pending full bytes.
-// There will be at least 56 bits available for writing when this has been called.
-// Using flush32 is faster, but leaves less space for writing.
-func (b *bitWriter) flush() {
-	v := b.nBits >> 3
-	switch v {
-	case 0:
-		return
-	case 1:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-		)
-		b.bitContainer >>= 1 << 3
-	case 2:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-		)
-		b.bitContainer >>= 2 << 3
-	case 3:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-		)
-		b.bitContainer >>= 3 << 3
-	case 4:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-		)
-		b.bitContainer >>= 4 << 3
-	case 5:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-		)
-		b.bitContainer >>= 5 << 3
-	case 6:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-		)
-		b.bitContainer >>= 6 << 3
-	case 7:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-			byte(b.bitContainer>>48),
-		)
-		b.bitContainer >>= 7 << 3
-	case 8:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-			byte(b.bitContainer>>48),
-			byte(b.bitContainer>>56),
-		)
-		b.bitContainer = 0
-		b.nBits = 0
-		return
-	default:
-		panic(fmt.Errorf("bits (%d) > 64", b.nBits))
-	}
-	b.nBits &= 7
-}
-
-// flush32 will flush out, so there are at least 32 bits available for writing.
-func (b *bitWriter) flush32() {
-	if b.nBits < 32 {
-		return
-	}
-	b.out = append(b.out,
-		byte(b.bitContainer),
-		byte(b.bitContainer>>8),
-		byte(b.bitContainer>>16),
-		byte(b.bitContainer>>24))
-	b.nBits -= 32
-	b.bitContainer >>= 32
-}
-
-// flushAlign will flush remaining full bytes and align to next byte boundary.
-func (b *bitWriter) flushAlign() {
-	nbBytes := (b.nBits + 7) >> 3
-	for i := uint8(0); i < nbBytes; i++ {
-		b.out = append(b.out, byte(b.bitContainer>>(i*8)))
-	}
-	b.nBits = 0
-	b.bitContainer = 0
-}
-
-// close will write the alignment bit and write the final byte(s)
-// to the output.
-func (b *bitWriter) close() error {
-	// End mark
-	b.addBits16Clean(1, 1)
-	// flush until next byte.
-	b.flushAlign()
-	return nil
-}
-
-// reset and continue writing by appending to out.
-func (b *bitWriter) reset(out []byte) {
-	b.bitContainer = 0
-	b.nBits = 0
-	b.out = out
-}
diff --git a/vendor/github.com/klauspost/compress/huff0/bytereader.go b/vendor/github.com/klauspost/compress/huff0/bytereader.go
deleted file mode 100644
index 50bcdf6ea9..0000000000
--- a/vendor/github.com/klauspost/compress/huff0/bytereader.go
+++ /dev/null
@@ -1,54 +0,0 @@
-// Copyright 2018 Klaus Post. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-// Based on work Copyright (c) 2013, Yann Collet, released under BSD License.
-
-package huff0
-
-// byteReader provides a byte reader that reads
-// little endian values from a byte stream.
-// The input stream is manually advanced.
-// The reader performs no bounds checks.
-type byteReader struct {
-	b   []byte
-	off int
-}
-
-// init will initialize the reader and set the input.
-func (b *byteReader) init(in []byte) {
-	b.b = in
-	b.off = 0
-}
-
-// advance the stream b n bytes.
-func (b *byteReader) advance(n uint) {
-	b.off += int(n)
-}
-
-// Int32 returns a little endian int32 starting at current offset.
-func (b byteReader) Int32() int32 {
-	v3 := int32(b.b[b.off+3])
-	v2 := int32(b.b[b.off+2])
-	v1 := int32(b.b[b.off+1])
-	v0 := int32(b.b[b.off])
-	return (v3 << 24) | (v2 << 16) | (v1 << 8) | v0
-}
-
-// Uint32 returns a little endian uint32 starting at current offset.
-func (b byteReader) Uint32() uint32 {
-	v3 := uint32(b.b[b.off+3])
-	v2 := uint32(b.b[b.off+2])
-	v1 := uint32(b.b[b.off+1])
-	v0 := uint32(b.b[b.off])
-	return (v3 << 24) | (v2 << 16) | (v1 << 8) | v0
-}
-
-// unread returns the unread portion of the input.
-func (b byteReader) unread() []byte {
-	return b.b[b.off:]
-}
-
-// remain will return the number of bytes remaining.
-func (b byteReader) remain() int {
-	return len(b.b) - b.off
-}
diff --git a/vendor/github.com/klauspost/compress/huff0/compress.go b/vendor/github.com/klauspost/compress/huff0/compress.go
deleted file mode 100644
index 0823c928ce..0000000000
--- a/vendor/github.com/klauspost/compress/huff0/compress.go
+++ /dev/null
@@ -1,656 +0,0 @@
-package huff0
-
-import (
-	"fmt"
-	"runtime"
-	"sync"
-)
-
-// Compress1X will compress the input.
-// The output can be decoded using Decompress1X.
-// Supply a Scratch object. The scratch object contains state about re-use,
-// So when sharing across independent encodes, be sure to set the re-use policy.
-func Compress1X(in []byte, s *Scratch) (out []byte, reUsed bool, err error) {
-	s, err = s.prepare(in)
-	if err != nil {
-		return nil, false, err
-	}
-	return compress(in, s, s.compress1X)
-}
-
-// Compress4X will compress the input. The input is split into 4 independent blocks
-// and compressed similar to Compress1X.
-// The output can be decoded using Decompress4X.
-// Supply a Scratch object. The scratch object contains state about re-use,
-// So when sharing across independent encodes, be sure to set the re-use policy.
-func Compress4X(in []byte, s *Scratch) (out []byte, reUsed bool, err error) {
-	s, err = s.prepare(in)
-	if err != nil {
-		return nil, false, err
-	}
-	if false {
-		// TODO: compress4Xp only slightly faster.
-		const parallelThreshold = 8 << 10
-		if len(in) < parallelThreshold || runtime.GOMAXPROCS(0) == 1 {
-			return compress(in, s, s.compress4X)
-		}
-		return compress(in, s, s.compress4Xp)
-	}
-	return compress(in, s, s.compress4X)
-}
-
-func compress(in []byte, s *Scratch, compressor func(src []byte) ([]byte, error)) (out []byte, reUsed bool, err error) {
-	// Nuke previous table if we cannot reuse anyway.
-	if s.Reuse == ReusePolicyNone {
-		s.prevTable = s.prevTable[:0]
-	}
-
-	// Create histogram, if none was provided.
-	maxCount := s.maxCount
-	var canReuse = false
-	if maxCount == 0 {
-		maxCount, canReuse = s.countSimple(in)
-	} else {
-		canReuse = s.canUseTable(s.prevTable)
-	}
-
-	// We want the output size to be less than this:
-	wantSize := len(in)
-	if s.WantLogLess > 0 {
-		wantSize -= wantSize >> s.WantLogLess
-	}
-
-	// Reset for next run.
-	s.clearCount = true
-	s.maxCount = 0
-	if maxCount >= len(in) {
-		if maxCount > len(in) {
-			return nil, false, fmt.Errorf("maxCount (%d) > length (%d)", maxCount, len(in))
-		}
-		if len(in) == 1 {
-			return nil, false, ErrIncompressible
-		}
-		// One symbol, use RLE
-		return nil, false, ErrUseRLE
-	}
-	if maxCount == 1 || maxCount < (len(in)>>7) {
-		// Each symbol present maximum once or too well distributed.
-		return nil, false, ErrIncompressible
-	}
-	if s.Reuse == ReusePolicyMust && !canReuse {
-		// We must reuse, but we can't.
-		return nil, false, ErrIncompressible
-	}
-	if (s.Reuse == ReusePolicyPrefer || s.Reuse == ReusePolicyMust) && canReuse {
-		keepTable := s.cTable
-		keepTL := s.actualTableLog
-		s.cTable = s.prevTable
-		s.actualTableLog = s.prevTableLog
-		s.Out, err = compressor(in)
-		s.cTable = keepTable
-		s.actualTableLog = keepTL
-		if err == nil && len(s.Out) < wantSize {
-			s.OutData = s.Out
-			return s.Out, true, nil
-		}
-		if s.Reuse == ReusePolicyMust {
-			return nil, false, ErrIncompressible
-		}
-		// Do not attempt to re-use later.
-		s.prevTable = s.prevTable[:0]
-	}
-
-	// Calculate new table.
-	err = s.buildCTable()
-	if err != nil {
-		return nil, false, err
-	}
-
-	if false && !s.canUseTable(s.cTable) {
-		panic("invalid table generated")
-	}
-
-	if s.Reuse == ReusePolicyAllow && canReuse {
-		hSize := len(s.Out)
-		oldSize := s.prevTable.estimateSize(s.count[:s.symbolLen])
-		newSize := s.cTable.estimateSize(s.count[:s.symbolLen])
-		if oldSize <= hSize+newSize || hSize+12 >= wantSize {
-			// Retain cTable even if we re-use.
-			keepTable := s.cTable
-			keepTL := s.actualTableLog
-
-			s.cTable = s.prevTable
-			s.actualTableLog = s.prevTableLog
-			s.Out, err = compressor(in)
-
-			// Restore ctable.
-			s.cTable = keepTable
-			s.actualTableLog = keepTL
-			if err != nil {
-				return nil, false, err
-			}
-			if len(s.Out) >= wantSize {
-				return nil, false, ErrIncompressible
-			}
-			s.OutData = s.Out
-			return s.Out, true, nil
-		}
-	}
-
-	// Use new table
-	err = s.cTable.write(s)
-	if err != nil {
-		s.OutTable = nil
-		return nil, false, err
-	}
-	s.OutTable = s.Out
-
-	// Compress using new table
-	s.Out, err = compressor(in)
-	if err != nil {
-		s.OutTable = nil
-		return nil, false, err
-	}
-	if len(s.Out) >= wantSize {
-		s.OutTable = nil
-		return nil, false, ErrIncompressible
-	}
-	// Move current table into previous.
-	s.prevTable, s.prevTableLog, s.cTable = s.cTable, s.actualTableLog, s.prevTable[:0]
-	s.OutData = s.Out[len(s.OutTable):]
-	return s.Out, false, nil
-}
-
-func (s *Scratch) compress1X(src []byte) ([]byte, error) {
-	return s.compress1xDo(s.Out, src)
-}
-
-func (s *Scratch) compress1xDo(dst, src []byte) ([]byte, error) {
-	var bw = bitWriter{out: dst}
-
-	// N is length divisible by 4.
-	n := len(src)
-	n -= n & 3
-	cTable := s.cTable[:256]
-
-	// Encode last bytes.
-	for i := len(src) & 3; i > 0; i-- {
-		bw.encSymbol(cTable, src[n+i-1])
-	}
-	n -= 4
-	if s.actualTableLog <= 8 {
-		for ; n >= 0; n -= 4 {
-			tmp := src[n : n+4]
-			// tmp should be len 4
-			bw.flush32()
-			bw.encTwoSymbols(cTable, tmp[3], tmp[2])
-			bw.encTwoSymbols(cTable, tmp[1], tmp[0])
-		}
-	} else {
-		for ; n >= 0; n -= 4 {
-			tmp := src[n : n+4]
-			// tmp should be len 4
-			bw.flush32()
-			bw.encTwoSymbols(cTable, tmp[3], tmp[2])
-			bw.flush32()
-			bw.encTwoSymbols(cTable, tmp[1], tmp[0])
-		}
-	}
-	err := bw.close()
-	return bw.out, err
-}
-
-var sixZeros [6]byte
-
-func (s *Scratch) compress4X(src []byte) ([]byte, error) {
-	if len(src) < 12 {
-		return nil, ErrIncompressible
-	}
-	segmentSize := (len(src) + 3) / 4
-
-	// Add placeholder for output length
-	offsetIdx := len(s.Out)
-	s.Out = append(s.Out, sixZeros[:]...)
-
-	for i := 0; i < 4; i++ {
-		toDo := src
-		if len(toDo) > segmentSize {
-			toDo = toDo[:segmentSize]
-		}
-		src = src[len(toDo):]
-
-		var err error
-		idx := len(s.Out)
-		s.Out, err = s.compress1xDo(s.Out, toDo)
-		if err != nil {
-			return nil, err
-		}
-		// Write compressed length as little endian before block.
-		if i < 3 {
-			// Last length is not written.
-			length := len(s.Out) - idx
-			s.Out[i*2+offsetIdx] = byte(length)
-			s.Out[i*2+offsetIdx+1] = byte(length >> 8)
-		}
-	}
-
-	return s.Out, nil
-}
-
-// compress4Xp will compress 4 streams using separate goroutines.
-func (s *Scratch) compress4Xp(src []byte) ([]byte, error) {
-	if len(src) < 12 {
-		return nil, ErrIncompressible
-	}
-	// Add placeholder for output length
-	s.Out = s.Out[:6]
-
-	segmentSize := (len(src) + 3) / 4
-	var wg sync.WaitGroup
-	var errs [4]error
-	wg.Add(4)
-	for i := 0; i < 4; i++ {
-		toDo := src
-		if len(toDo) > segmentSize {
-			toDo = toDo[:segmentSize]
-		}
-		src = src[len(toDo):]
-
-		// Separate goroutine for each block.
-		go func(i int) {
-			s.tmpOut[i], errs[i] = s.compress1xDo(s.tmpOut[i][:0], toDo)
-			wg.Done()
-		}(i)
-	}
-	wg.Wait()
-	for i := 0; i < 4; i++ {
-		if errs[i] != nil {
-			return nil, errs[i]
-		}
-		o := s.tmpOut[i]
-		// Write compressed length as little endian before block.
-		if i < 3 {
-			// Last length is not written.
-			s.Out[i*2] = byte(len(o))
-			s.Out[i*2+1] = byte(len(o) >> 8)
-		}
-
-		// Write output.
-		s.Out = append(s.Out, o...)
-	}
-	return s.Out, nil
-}
-
-// countSimple will create a simple histogram in s.count.
-// Returns the biggest count.
-// Does not update s.clearCount.
-func (s *Scratch) countSimple(in []byte) (max int, reuse bool) {
-	reuse = true
-	for _, v := range in {
-		s.count[v]++
-	}
-	m := uint32(0)
-	if len(s.prevTable) > 0 {
-		for i, v := range s.count[:] {
-			if v > m {
-				m = v
-			}
-			if v > 0 {
-				s.symbolLen = uint16(i) + 1
-				if i >= len(s.prevTable) {
-					reuse = false
-				} else {
-					if s.prevTable[i].nBits == 0 {
-						reuse = false
-					}
-				}
-			}
-		}
-		return int(m), reuse
-	}
-	for i, v := range s.count[:] {
-		if v > m {
-			m = v
-		}
-		if v > 0 {
-			s.symbolLen = uint16(i) + 1
-		}
-	}
-	return int(m), false
-}
-
-func (s *Scratch) canUseTable(c cTable) bool {
-	if len(c) < int(s.symbolLen) {
-		return false
-	}
-	for i, v := range s.count[:s.symbolLen] {
-		if v != 0 && c[i].nBits == 0 {
-			return false
-		}
-	}
-	return true
-}
-
-func (s *Scratch) validateTable(c cTable) bool {
-	if len(c) < int(s.symbolLen) {
-		return false
-	}
-	for i, v := range s.count[:s.symbolLen] {
-		if v != 0 {
-			if c[i].nBits == 0 {
-				return false
-			}
-			if c[i].nBits > s.actualTableLog {
-				return false
-			}
-		}
-	}
-	return true
-}
-
-// minTableLog provides the minimum logSize to safely represent a distribution.
-func (s *Scratch) minTableLog() uint8 {
-	minBitsSrc := highBit32(uint32(s.br.remain())) + 1
-	minBitsSymbols := highBit32(uint32(s.symbolLen-1)) + 2
-	if minBitsSrc < minBitsSymbols {
-		return uint8(minBitsSrc)
-	}
-	return uint8(minBitsSymbols)
-}
-
-// optimalTableLog calculates and sets the optimal tableLog in s.actualTableLog
-func (s *Scratch) optimalTableLog() {
-	tableLog := s.TableLog
-	minBits := s.minTableLog()
-	maxBitsSrc := uint8(highBit32(uint32(s.br.remain()-1))) - 1
-	if maxBitsSrc < tableLog {
-		// Accuracy can be reduced
-		tableLog = maxBitsSrc
-	}
-	if minBits > tableLog {
-		tableLog = minBits
-	}
-	// Need a minimum to safely represent all symbol values
-	if tableLog < minTablelog {
-		tableLog = minTablelog
-	}
-	if tableLog > tableLogMax {
-		tableLog = tableLogMax
-	}
-	s.actualTableLog = tableLog
-}
-
-type cTableEntry struct {
-	val   uint16
-	nBits uint8
-	// We have 8 bits extra
-}
-
-const huffNodesMask = huffNodesLen - 1
-
-func (s *Scratch) buildCTable() error {
-	s.optimalTableLog()
-	s.huffSort()
-	if cap(s.cTable) < maxSymbolValue+1 {
-		s.cTable = make([]cTableEntry, s.symbolLen, maxSymbolValue+1)
-	} else {
-		s.cTable = s.cTable[:s.symbolLen]
-		for i := range s.cTable {
-			s.cTable[i] = cTableEntry{}
-		}
-	}
-
-	var startNode = int16(s.symbolLen)
-	nonNullRank := s.symbolLen - 1
-
-	nodeNb := startNode
-	huffNode := s.nodes[1 : huffNodesLen+1]
-
-	// This overlays the slice above, but allows "-1" index lookups.
-	// Different from reference implementation.
-	huffNode0 := s.nodes[0 : huffNodesLen+1]
-
-	for huffNode[nonNullRank].count == 0 {
-		nonNullRank--
-	}
-
-	lowS := int16(nonNullRank)
-	nodeRoot := nodeNb + lowS - 1
-	lowN := nodeNb
-	huffNode[nodeNb].count = huffNode[lowS].count + huffNode[lowS-1].count
-	huffNode[lowS].parent, huffNode[lowS-1].parent = uint16(nodeNb), uint16(nodeNb)
-	nodeNb++
-	lowS -= 2
-	for n := nodeNb; n <= nodeRoot; n++ {
-		huffNode[n].count = 1 << 30
-	}
-	// fake entry, strong barrier
-	huffNode0[0].count = 1 << 31
-
-	// create parents
-	for nodeNb <= nodeRoot {
-		var n1, n2 int16
-		if huffNode0[lowS+1].count < huffNode0[lowN+1].count {
-			n1 = lowS
-			lowS--
-		} else {
-			n1 = lowN
-			lowN++
-		}
-		if huffNode0[lowS+1].count < huffNode0[lowN+1].count {
-			n2 = lowS
-			lowS--
-		} else {
-			n2 = lowN
-			lowN++
-		}
-
-		huffNode[nodeNb].count = huffNode0[n1+1].count + huffNode0[n2+1].count
-		huffNode0[n1+1].parent, huffNode0[n2+1].parent = uint16(nodeNb), uint16(nodeNb)
-		nodeNb++
-	}
-
-	// distribute weights (unlimited tree height)
-	huffNode[nodeRoot].nbBits = 0
-	for n := nodeRoot - 1; n >= startNode; n-- {
-		huffNode[n].nbBits = huffNode[huffNode[n].parent].nbBits + 1
-	}
-	for n := uint16(0); n <= nonNullRank; n++ {
-		huffNode[n].nbBits = huffNode[huffNode[n].parent].nbBits + 1
-	}
-	s.actualTableLog = s.setMaxHeight(int(nonNullRank))
-	maxNbBits := s.actualTableLog
-
-	// fill result into tree (val, nbBits)
-	if maxNbBits > tableLogMax {
-		return fmt.Errorf("internal error: maxNbBits (%d) > tableLogMax (%d)", maxNbBits, tableLogMax)
-	}
-	var nbPerRank [tableLogMax + 1]uint16
-	var valPerRank [16]uint16
-	for _, v := range huffNode[:nonNullRank+1] {
-		nbPerRank[v.nbBits]++
-	}
-	// determine stating value per rank
-	{
-		min := uint16(0)
-		for n := maxNbBits; n > 0; n-- {
-			// get starting value within each rank
-			valPerRank[n] = min
-			min += nbPerRank[n]
-			min >>= 1
-		}
-	}
-
-	// push nbBits per symbol, symbol order
-	for _, v := range huffNode[:nonNullRank+1] {
-		s.cTable[v.symbol].nBits = v.nbBits
-	}
-
-	// assign value within rank, symbol order
-	t := s.cTable[:s.symbolLen]
-	for n, val := range t {
-		nbits := val.nBits & 15
-		v := valPerRank[nbits]
-		t[n].val = v
-		valPerRank[nbits] = v + 1
-	}
-
-	return nil
-}
-
-// huffSort will sort symbols, decreasing order.
-func (s *Scratch) huffSort() {
-	type rankPos struct {
-		base    uint32
-		current uint32
-	}
-
-	// Clear nodes
-	nodes := s.nodes[:huffNodesLen+1]
-	s.nodes = nodes
-	nodes = nodes[1 : huffNodesLen+1]
-
-	// Sort into buckets based on length of symbol count.
-	var rank [32]rankPos
-	for _, v := range s.count[:s.symbolLen] {
-		r := highBit32(v+1) & 31
-		rank[r].base++
-	}
-	// maxBitLength is log2(BlockSizeMax) + 1
-	const maxBitLength = 18 + 1
-	for n := maxBitLength; n > 0; n-- {
-		rank[n-1].base += rank[n].base
-	}
-	for n := range rank[:maxBitLength] {
-		rank[n].current = rank[n].base
-	}
-	for n, c := range s.count[:s.symbolLen] {
-		r := (highBit32(c+1) + 1) & 31
-		pos := rank[r].current
-		rank[r].current++
-		prev := nodes[(pos-1)&huffNodesMask]
-		for pos > rank[r].base && c > prev.count {
-			nodes[pos&huffNodesMask] = prev
-			pos--
-			prev = nodes[(pos-1)&huffNodesMask]
-		}
-		nodes[pos&huffNodesMask] = nodeElt{count: c, symbol: byte(n)}
-	}
-}
-
-func (s *Scratch) setMaxHeight(lastNonNull int) uint8 {
-	maxNbBits := s.actualTableLog
-	huffNode := s.nodes[1 : huffNodesLen+1]
-	//huffNode = huffNode[: huffNodesLen]
-
-	largestBits := huffNode[lastNonNull].nbBits
-
-	// early exit : no elt > maxNbBits
-	if largestBits <= maxNbBits {
-		return largestBits
-	}
-	totalCost := int(0)
-	baseCost := int(1) << (largestBits - maxNbBits)
-	n := uint32(lastNonNull)
-
-	for huffNode[n].nbBits > maxNbBits {
-		totalCost += baseCost - (1 << (largestBits - huffNode[n].nbBits))
-		huffNode[n].nbBits = maxNbBits
-		n--
-	}
-	// n stops at huffNode[n].nbBits <= maxNbBits
-
-	for huffNode[n].nbBits == maxNbBits {
-		n--
-	}
-	// n end at index of smallest symbol using < maxNbBits
-
-	// renorm totalCost
-	totalCost >>= largestBits - maxNbBits /* note : totalCost is necessarily a multiple of baseCost */
-
-	// repay normalized cost
-	{
-		const noSymbol = 0xF0F0F0F0
-		var rankLast [tableLogMax + 2]uint32
-
-		for i := range rankLast[:] {
-			rankLast[i] = noSymbol
-		}
-
-		// Get pos of last (smallest) symbol per rank
-		{
-			currentNbBits := maxNbBits
-			for pos := int(n); pos >= 0; pos-- {
-				if huffNode[pos].nbBits >= currentNbBits {
-					continue
-				}
-				currentNbBits = huffNode[pos].nbBits // < maxNbBits
-				rankLast[maxNbBits-currentNbBits] = uint32(pos)
-			}
-		}
-
-		for totalCost > 0 {
-			nBitsToDecrease := uint8(highBit32(uint32(totalCost))) + 1
-
-			for ; nBitsToDecrease > 1; nBitsToDecrease-- {
-				highPos := rankLast[nBitsToDecrease]
-				lowPos := rankLast[nBitsToDecrease-1]
-				if highPos == noSymbol {
-					continue
-				}
-				if lowPos == noSymbol {
-					break
-				}
-				highTotal := huffNode[highPos].count
-				lowTotal := 2 * huffNode[lowPos].count
-				if highTotal <= lowTotal {
-					break
-				}
-			}
-			// only triggered when no more rank 1 symbol left => find closest one (note : there is necessarily at least one !)
-			// HUF_MAX_TABLELOG test just to please gcc 5+; but it should not be necessary
-			// FIXME: try to remove
-			for (nBitsToDecrease <= tableLogMax) && (rankLast[nBitsToDecrease] == noSymbol) {
-				nBitsToDecrease++
-			}
-			totalCost -= 1 << (nBitsToDecrease - 1)
-			if rankLast[nBitsToDecrease-1] == noSymbol {
-				// this rank is no longer empty
-				rankLast[nBitsToDecrease-1] = rankLast[nBitsToDecrease]
-			}
-			huffNode[rankLast[nBitsToDecrease]].nbBits++
-			if rankLast[nBitsToDecrease] == 0 {
-				/* special case, reached largest symbol */
-				rankLast[nBitsToDecrease] = noSymbol
-			} else {
-				rankLast[nBitsToDecrease]--
-				if huffNode[rankLast[nBitsToDecrease]].nbBits != maxNbBits-nBitsToDecrease {
-					rankLast[nBitsToDecrease] = noSymbol /* this rank is now empty */
-				}
-			}
-		}
-
-		for totalCost < 0 { /* Sometimes, cost correction overshoot */
-			if rankLast[1] == noSymbol { /* special case : no rank 1 symbol (using maxNbBits-1); let's create one from largest rank 0 (using maxNbBits) */
-				for huffNode[n].nbBits == maxNbBits {
-					n--
-				}
-				huffNode[n+1].nbBits--
-				rankLast[1] = n + 1
-				totalCost++
-				continue
-			}
-			huffNode[rankLast[1]+1].nbBits--
-			rankLast[1]++
-			totalCost++
-		}
-	}
-	return maxNbBits
-}
-
-type nodeElt struct {
-	count  uint32
-	parent uint16
-	symbol byte
-	nbBits uint8
-}
diff --git a/vendor/github.com/klauspost/compress/huff0/decompress.go b/vendor/github.com/klauspost/compress/huff0/decompress.go
deleted file mode 100644
index 41703bba4d..0000000000
--- a/vendor/github.com/klauspost/compress/huff0/decompress.go
+++ /dev/null
@@ -1,1164 +0,0 @@
-package huff0
-
-import (
-	"errors"
-	"fmt"
-	"io"
-
-	"github.com/klauspost/compress/fse"
-)
-
-type dTable struct {
-	single []dEntrySingle
-	double []dEntryDouble
-}
-
-// single-symbols decoding
-type dEntrySingle struct {
-	entry uint16
-}
-
-// double-symbols decoding
-type dEntryDouble struct {
-	seq   uint16
-	nBits uint8
-	len   uint8
-}
-
-// Uses special code for all tables that are < 8 bits.
-const use8BitTables = true
-
-// ReadTable will read a table from the input.
-// The size of the input may be larger than the table definition.
-// Any content remaining after the table definition will be returned.
-// If no Scratch is provided a new one is allocated.
-// The returned Scratch can be used for encoding or decoding input using this table.
-func ReadTable(in []byte, s *Scratch) (s2 *Scratch, remain []byte, err error) {
-	s, err = s.prepare(in)
-	if err != nil {
-		return s, nil, err
-	}
-	if len(in) <= 1 {
-		return s, nil, errors.New("input too small for table")
-	}
-	iSize := in[0]
-	in = in[1:]
-	if iSize >= 128 {
-		// Uncompressed
-		oSize := iSize - 127
-		iSize = (oSize + 1) / 2
-		if int(iSize) > len(in) {
-			return s, nil, errors.New("input too small for table")
-		}
-		for n := uint8(0); n < oSize; n += 2 {
-			v := in[n/2]
-			s.huffWeight[n] = v >> 4
-			s.huffWeight[n+1] = v & 15
-		}
-		s.symbolLen = uint16(oSize)
-		in = in[iSize:]
-	} else {
-		if len(in) < int(iSize) {
-			return s, nil, fmt.Errorf("input too small for table, want %d bytes, have %d", iSize, len(in))
-		}
-		// FSE compressed weights
-		s.fse.DecompressLimit = 255
-		hw := s.huffWeight[:]
-		s.fse.Out = hw
-		b, err := fse.Decompress(in[:iSize], s.fse)
-		s.fse.Out = nil
-		if err != nil {
-			return s, nil, err
-		}
-		if len(b) > 255 {
-			return s, nil, errors.New("corrupt input: output table too large")
-		}
-		s.symbolLen = uint16(len(b))
-		in = in[iSize:]
-	}
-
-	// collect weight stats
-	var rankStats [16]uint32
-	weightTotal := uint32(0)
-	for _, v := range s.huffWeight[:s.symbolLen] {
-		if v > tableLogMax {
-			return s, nil, errors.New("corrupt input: weight too large")
-		}
-		v2 := v & 15
-		rankStats[v2]++
-		// (1 << (v2-1)) is slower since the compiler cannot prove that v2 isn't 0.
-		weightTotal += (1 << v2) >> 1
-	}
-	if weightTotal == 0 {
-		return s, nil, errors.New("corrupt input: weights zero")
-	}
-
-	// get last non-null symbol weight (implied, total must be 2^n)
-	{
-		tableLog := highBit32(weightTotal) + 1
-		if tableLog > tableLogMax {
-			return s, nil, errors.New("corrupt input: tableLog too big")
-		}
-		s.actualTableLog = uint8(tableLog)
-		// determine last weight
-		{
-			total := uint32(1) << tableLog
-			rest := total - weightTotal
-			verif := uint32(1) << highBit32(rest)
-			lastWeight := highBit32(rest) + 1
-			if verif != rest {
-				// last value must be a clean power of 2
-				return s, nil, errors.New("corrupt input: last value not power of two")
-			}
-			s.huffWeight[s.symbolLen] = uint8(lastWeight)
-			s.symbolLen++
-			rankStats[lastWeight]++
-		}
-	}
-
-	if (rankStats[1] < 2) || (rankStats[1]&1 != 0) {
-		// by construction : at least 2 elts of rank 1, must be even
-		return s, nil, errors.New("corrupt input: min elt size, even check failed ")
-	}
-
-	// TODO: Choose between single/double symbol decoding
-
-	// Calculate starting value for each rank
-	{
-		var nextRankStart uint32
-		for n := uint8(1); n < s.actualTableLog+1; n++ {
-			current := nextRankStart
-			nextRankStart += rankStats[n] << (n - 1)
-			rankStats[n] = current
-		}
-	}
-
-	// fill DTable (always full size)
-	tSize := 1 << tableLogMax
-	if len(s.dt.single) != tSize {
-		s.dt.single = make([]dEntrySingle, tSize)
-	}
-	cTable := s.prevTable
-	if cap(cTable) < maxSymbolValue+1 {
-		cTable = make([]cTableEntry, 0, maxSymbolValue+1)
-	}
-	cTable = cTable[:maxSymbolValue+1]
-	s.prevTable = cTable[:s.symbolLen]
-	s.prevTableLog = s.actualTableLog
-
-	for n, w := range s.huffWeight[:s.symbolLen] {
-		if w == 0 {
-			cTable[n] = cTableEntry{
-				val:   0,
-				nBits: 0,
-			}
-			continue
-		}
-		length := (uint32(1) << w) >> 1
-		d := dEntrySingle{
-			entry: uint16(s.actualTableLog+1-w) | (uint16(n) << 8),
-		}
-
-		rank := &rankStats[w]
-		cTable[n] = cTableEntry{
-			val:   uint16(*rank >> (w - 1)),
-			nBits: uint8(d.entry),
-		}
-
-		single := s.dt.single[*rank : *rank+length]
-		for i := range single {
-			single[i] = d
-		}
-		*rank += length
-	}
-
-	return s, in, nil
-}
-
-// Decompress1X will decompress a 1X encoded stream.
-// The length of the supplied input must match the end of a block exactly.
-// Before this is called, the table must be initialized with ReadTable unless
-// the encoder re-used the table.
-// deprecated: Use the stateless Decoder() to get a concurrent version.
-func (s *Scratch) Decompress1X(in []byte) (out []byte, err error) {
-	if cap(s.Out) < s.MaxDecodedSize {
-		s.Out = make([]byte, s.MaxDecodedSize)
-	}
-	s.Out = s.Out[:0:s.MaxDecodedSize]
-	s.Out, err = s.Decoder().Decompress1X(s.Out, in)
-	return s.Out, err
-}
-
-// Decompress4X will decompress a 4X encoded stream.
-// Before this is called, the table must be initialized with ReadTable unless
-// the encoder re-used the table.
-// The length of the supplied input must match the end of a block exactly.
-// The destination size of the uncompressed data must be known and provided.
-// deprecated: Use the stateless Decoder() to get a concurrent version.
-func (s *Scratch) Decompress4X(in []byte, dstSize int) (out []byte, err error) {
-	if dstSize > s.MaxDecodedSize {
-		return nil, ErrMaxDecodedSizeExceeded
-	}
-	if cap(s.Out) < dstSize {
-		s.Out = make([]byte, s.MaxDecodedSize)
-	}
-	s.Out = s.Out[:0:dstSize]
-	s.Out, err = s.Decoder().Decompress4X(s.Out, in)
-	return s.Out, err
-}
-
-// Decoder will return a stateless decoder that can be used by multiple
-// decompressors concurrently.
-// Before this is called, the table must be initialized with ReadTable.
-// The Decoder is still linked to the scratch buffer so that cannot be reused.
-// However, it is safe to discard the scratch.
-func (s *Scratch) Decoder() *Decoder {
-	return &Decoder{
-		dt:             s.dt,
-		actualTableLog: s.actualTableLog,
-	}
-}
-
-// Decoder provides stateless decoding.
-type Decoder struct {
-	dt             dTable
-	actualTableLog uint8
-}
-
-// Decompress1X will decompress a 1X encoded stream.
-// The cap of the output buffer will be the maximum decompressed size.
-// The length of the supplied input must match the end of a block exactly.
-func (d *Decoder) Decompress1X(dst, src []byte) ([]byte, error) {
-	if len(d.dt.single) == 0 {
-		return nil, errors.New("no table loaded")
-	}
-	if use8BitTables && d.actualTableLog <= 8 {
-		return d.decompress1X8Bit(dst, src)
-	}
-	var br bitReaderShifted
-	err := br.init(src)
-	if err != nil {
-		return dst, err
-	}
-	maxDecodedSize := cap(dst)
-	dst = dst[:0]
-
-	// Avoid bounds check by always having full sized table.
-	const tlSize = 1 << tableLogMax
-	const tlMask = tlSize - 1
-	dt := d.dt.single[:tlSize]
-
-	// Use temp table to avoid bound checks/append penalty.
-	var buf [256]byte
-	var off uint8
-
-	for br.off >= 8 {
-		br.fillFast()
-		v := dt[br.peekBitsFast(d.actualTableLog)&tlMask]
-		br.advance(uint8(v.entry))
-		buf[off+0] = uint8(v.entry >> 8)
-
-		v = dt[br.peekBitsFast(d.actualTableLog)&tlMask]
-		br.advance(uint8(v.entry))
-		buf[off+1] = uint8(v.entry >> 8)
-
-		// Refill
-		br.fillFast()
-
-		v = dt[br.peekBitsFast(d.actualTableLog)&tlMask]
-		br.advance(uint8(v.entry))
-		buf[off+2] = uint8(v.entry >> 8)
-
-		v = dt[br.peekBitsFast(d.actualTableLog)&tlMask]
-		br.advance(uint8(v.entry))
-		buf[off+3] = uint8(v.entry >> 8)
-
-		off += 4
-		if off == 0 {
-			if len(dst)+256 > maxDecodedSize {
-				br.close()
-				return nil, ErrMaxDecodedSizeExceeded
-			}
-			dst = append(dst, buf[:]...)
-		}
-	}
-
-	if len(dst)+int(off) > maxDecodedSize {
-		br.close()
-		return nil, ErrMaxDecodedSizeExceeded
-	}
-	dst = append(dst, buf[:off]...)
-
-	// br < 8, so uint8 is fine
-	bitsLeft := uint8(br.off)*8 + 64 - br.bitsRead
-	for bitsLeft > 0 {
-		br.fill()
-		if false && br.bitsRead >= 32 {
-			if br.off >= 4 {
-				v := br.in[br.off-4:]
-				v = v[:4]
-				low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-				br.value = (br.value << 32) | uint64(low)
-				br.bitsRead -= 32
-				br.off -= 4
-			} else {
-				for br.off > 0 {
-					br.value = (br.value << 8) | uint64(br.in[br.off-1])
-					br.bitsRead -= 8
-					br.off--
-				}
-			}
-		}
-		if len(dst) >= maxDecodedSize {
-			br.close()
-			return nil, ErrMaxDecodedSizeExceeded
-		}
-		v := d.dt.single[br.peekBitsFast(d.actualTableLog)&tlMask]
-		nBits := uint8(v.entry)
-		br.advance(nBits)
-		bitsLeft -= nBits
-		dst = append(dst, uint8(v.entry>>8))
-	}
-	return dst, br.close()
-}
-
-// decompress1X8Bit will decompress a 1X encoded stream with tablelog <= 8.
-// The cap of the output buffer will be the maximum decompressed size.
-// The length of the supplied input must match the end of a block exactly.
-func (d *Decoder) decompress1X8Bit(dst, src []byte) ([]byte, error) {
-	if d.actualTableLog == 8 {
-		return d.decompress1X8BitExactly(dst, src)
-	}
-	var br bitReaderBytes
-	err := br.init(src)
-	if err != nil {
-		return dst, err
-	}
-	maxDecodedSize := cap(dst)
-	dst = dst[:0]
-
-	// Avoid bounds check by always having full sized table.
-	dt := d.dt.single[:256]
-
-	// Use temp table to avoid bound checks/append penalty.
-	var buf [256]byte
-	var off uint8
-
-	shift := (8 - d.actualTableLog) & 7
-
-	//fmt.Printf("mask: %b, tl:%d\n", mask, d.actualTableLog)
-	for br.off >= 4 {
-		br.fillFast()
-		v := dt[br.peekByteFast()>>shift]
-		br.advance(uint8(v.entry))
-		buf[off+0] = uint8(v.entry >> 8)
-
-		v = dt[br.peekByteFast()>>shift]
-		br.advance(uint8(v.entry))
-		buf[off+1] = uint8(v.entry >> 8)
-
-		v = dt[br.peekByteFast()>>shift]
-		br.advance(uint8(v.entry))
-		buf[off+2] = uint8(v.entry >> 8)
-
-		v = dt[br.peekByteFast()>>shift]
-		br.advance(uint8(v.entry))
-		buf[off+3] = uint8(v.entry >> 8)
-
-		off += 4
-		if off == 0 {
-			if len(dst)+256 > maxDecodedSize {
-				br.close()
-				return nil, ErrMaxDecodedSizeExceeded
-			}
-			dst = append(dst, buf[:]...)
-		}
-	}
-
-	if len(dst)+int(off) > maxDecodedSize {
-		br.close()
-		return nil, ErrMaxDecodedSizeExceeded
-	}
-	dst = append(dst, buf[:off]...)
-
-	// br < 4, so uint8 is fine
-	bitsLeft := int8(uint8(br.off)*8 + (64 - br.bitsRead))
-	for bitsLeft > 0 {
-		if br.bitsRead >= 64-8 {
-			for br.off > 0 {
-				br.value |= uint64(br.in[br.off-1]) << (br.bitsRead - 8)
-				br.bitsRead -= 8
-				br.off--
-			}
-		}
-		if len(dst) >= maxDecodedSize {
-			br.close()
-			return nil, ErrMaxDecodedSizeExceeded
-		}
-		v := dt[br.peekByteFast()>>shift]
-		nBits := uint8(v.entry)
-		br.advance(nBits)
-		bitsLeft -= int8(nBits)
-		dst = append(dst, uint8(v.entry>>8))
-	}
-	return dst, br.close()
-}
-
-// decompress1X8Bit will decompress a 1X encoded stream with tablelog <= 8.
-// The cap of the output buffer will be the maximum decompressed size.
-// The length of the supplied input must match the end of a block exactly.
-func (d *Decoder) decompress1X8BitExactly(dst, src []byte) ([]byte, error) {
-	var br bitReaderBytes
-	err := br.init(src)
-	if err != nil {
-		return dst, err
-	}
-	maxDecodedSize := cap(dst)
-	dst = dst[:0]
-
-	// Avoid bounds check by always having full sized table.
-	dt := d.dt.single[:256]
-
-	// Use temp table to avoid bound checks/append penalty.
-	var buf [256]byte
-	var off uint8
-
-	const shift = 0
-
-	//fmt.Printf("mask: %b, tl:%d\n", mask, d.actualTableLog)
-	for br.off >= 4 {
-		br.fillFast()
-		v := dt[br.peekByteFast()>>shift]
-		br.advance(uint8(v.entry))
-		buf[off+0] = uint8(v.entry >> 8)
-
-		v = dt[br.peekByteFast()>>shift]
-		br.advance(uint8(v.entry))
-		buf[off+1] = uint8(v.entry >> 8)
-
-		v = dt[br.peekByteFast()>>shift]
-		br.advance(uint8(v.entry))
-		buf[off+2] = uint8(v.entry >> 8)
-
-		v = dt[br.peekByteFast()>>shift]
-		br.advance(uint8(v.entry))
-		buf[off+3] = uint8(v.entry >> 8)
-
-		off += 4
-		if off == 0 {
-			if len(dst)+256 > maxDecodedSize {
-				br.close()
-				return nil, ErrMaxDecodedSizeExceeded
-			}
-			dst = append(dst, buf[:]...)
-		}
-	}
-
-	if len(dst)+int(off) > maxDecodedSize {
-		br.close()
-		return nil, ErrMaxDecodedSizeExceeded
-	}
-	dst = append(dst, buf[:off]...)
-
-	// br < 4, so uint8 is fine
-	bitsLeft := int8(uint8(br.off)*8 + (64 - br.bitsRead))
-	for bitsLeft > 0 {
-		if br.bitsRead >= 64-8 {
-			for br.off > 0 {
-				br.value |= uint64(br.in[br.off-1]) << (br.bitsRead - 8)
-				br.bitsRead -= 8
-				br.off--
-			}
-		}
-		if len(dst) >= maxDecodedSize {
-			br.close()
-			return nil, ErrMaxDecodedSizeExceeded
-		}
-		v := dt[br.peekByteFast()>>shift]
-		nBits := uint8(v.entry)
-		br.advance(nBits)
-		bitsLeft -= int8(nBits)
-		dst = append(dst, uint8(v.entry>>8))
-	}
-	return dst, br.close()
-}
-
-// Decompress4X will decompress a 4X encoded stream.
-// The length of the supplied input must match the end of a block exactly.
-// The *capacity* of the dst slice must match the destination size of
-// the uncompressed data exactly.
-func (d *Decoder) Decompress4X(dst, src []byte) ([]byte, error) {
-	if len(d.dt.single) == 0 {
-		return nil, errors.New("no table loaded")
-	}
-	if len(src) < 6+(4*1) {
-		return nil, errors.New("input too small")
-	}
-	if use8BitTables && d.actualTableLog <= 8 {
-		return d.decompress4X8bit(dst, src)
-	}
-
-	var br [4]bitReaderShifted
-	start := 6
-	for i := 0; i < 3; i++ {
-		length := int(src[i*2]) | (int(src[i*2+1]) << 8)
-		if start+length >= len(src) {
-			return nil, errors.New("truncated input (or invalid offset)")
-		}
-		err := br[i].init(src[start : start+length])
-		if err != nil {
-			return nil, err
-		}
-		start += length
-	}
-	err := br[3].init(src[start:])
-	if err != nil {
-		return nil, err
-	}
-
-	// destination, offset to match first output
-	dstSize := cap(dst)
-	dst = dst[:dstSize]
-	out := dst
-	dstEvery := (dstSize + 3) / 4
-
-	const tlSize = 1 << tableLogMax
-	const tlMask = tlSize - 1
-	single := d.dt.single[:tlSize]
-
-	// Use temp table to avoid bound checks/append penalty.
-	var buf [256]byte
-	var off uint8
-	var decoded int
-
-	// Decode 2 values from each decoder/loop.
-	const bufoff = 256 / 4
-	for {
-		if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 {
-			break
-		}
-
-		{
-			const stream = 0
-			const stream2 = 1
-			br[stream].fillFast()
-			br[stream2].fillFast()
-
-			val := br[stream].peekBitsFast(d.actualTableLog)
-			v := single[val&tlMask]
-			br[stream].advance(uint8(v.entry))
-			buf[off+bufoff*stream] = uint8(v.entry >> 8)
-
-			val2 := br[stream2].peekBitsFast(d.actualTableLog)
-			v2 := single[val2&tlMask]
-			br[stream2].advance(uint8(v2.entry))
-			buf[off+bufoff*stream2] = uint8(v2.entry >> 8)
-
-			val = br[stream].peekBitsFast(d.actualTableLog)
-			v = single[val&tlMask]
-			br[stream].advance(uint8(v.entry))
-			buf[off+bufoff*stream+1] = uint8(v.entry >> 8)
-
-			val2 = br[stream2].peekBitsFast(d.actualTableLog)
-			v2 = single[val2&tlMask]
-			br[stream2].advance(uint8(v2.entry))
-			buf[off+bufoff*stream2+1] = uint8(v2.entry >> 8)
-		}
-
-		{
-			const stream = 2
-			const stream2 = 3
-			br[stream].fillFast()
-			br[stream2].fillFast()
-
-			val := br[stream].peekBitsFast(d.actualTableLog)
-			v := single[val&tlMask]
-			br[stream].advance(uint8(v.entry))
-			buf[off+bufoff*stream] = uint8(v.entry >> 8)
-
-			val2 := br[stream2].peekBitsFast(d.actualTableLog)
-			v2 := single[val2&tlMask]
-			br[stream2].advance(uint8(v2.entry))
-			buf[off+bufoff*stream2] = uint8(v2.entry >> 8)
-
-			val = br[stream].peekBitsFast(d.actualTableLog)
-			v = single[val&tlMask]
-			br[stream].advance(uint8(v.entry))
-			buf[off+bufoff*stream+1] = uint8(v.entry >> 8)
-
-			val2 = br[stream2].peekBitsFast(d.actualTableLog)
-			v2 = single[val2&tlMask]
-			br[stream2].advance(uint8(v2.entry))
-			buf[off+bufoff*stream2+1] = uint8(v2.entry >> 8)
-		}
-
-		off += 2
-
-		if off == bufoff {
-			if bufoff > dstEvery {
-				return nil, errors.New("corruption detected: stream overrun 1")
-			}
-			copy(out, buf[:bufoff])
-			copy(out[dstEvery:], buf[bufoff:bufoff*2])
-			copy(out[dstEvery*2:], buf[bufoff*2:bufoff*3])
-			copy(out[dstEvery*3:], buf[bufoff*3:bufoff*4])
-			off = 0
-			out = out[bufoff:]
-			decoded += 256
-			// There must at least be 3 buffers left.
-			if len(out) < dstEvery*3 {
-				return nil, errors.New("corruption detected: stream overrun 2")
-			}
-		}
-	}
-	if off > 0 {
-		ioff := int(off)
-		if len(out) < dstEvery*3+ioff {
-			return nil, errors.New("corruption detected: stream overrun 3")
-		}
-		copy(out, buf[:off])
-		copy(out[dstEvery:dstEvery+ioff], buf[bufoff:bufoff*2])
-		copy(out[dstEvery*2:dstEvery*2+ioff], buf[bufoff*2:bufoff*3])
-		copy(out[dstEvery*3:dstEvery*3+ioff], buf[bufoff*3:bufoff*4])
-		decoded += int(off) * 4
-		out = out[off:]
-	}
-
-	// Decode remaining.
-	for i := range br {
-		offset := dstEvery * i
-		br := &br[i]
-		bitsLeft := br.off*8 + uint(64-br.bitsRead)
-		for bitsLeft > 0 {
-			br.fill()
-			if false && br.bitsRead >= 32 {
-				if br.off >= 4 {
-					v := br.in[br.off-4:]
-					v = v[:4]
-					low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-					br.value = (br.value << 32) | uint64(low)
-					br.bitsRead -= 32
-					br.off -= 4
-				} else {
-					for br.off > 0 {
-						br.value = (br.value << 8) | uint64(br.in[br.off-1])
-						br.bitsRead -= 8
-						br.off--
-					}
-				}
-			}
-			// end inline...
-			if offset >= len(out) {
-				return nil, errors.New("corruption detected: stream overrun 4")
-			}
-
-			// Read value and increment offset.
-			val := br.peekBitsFast(d.actualTableLog)
-			v := single[val&tlMask].entry
-			nBits := uint8(v)
-			br.advance(nBits)
-			bitsLeft -= uint(nBits)
-			out[offset] = uint8(v >> 8)
-			offset++
-		}
-		decoded += offset - dstEvery*i
-		err = br.close()
-		if err != nil {
-			return nil, err
-		}
-	}
-	if dstSize != decoded {
-		return nil, errors.New("corruption detected: short output block")
-	}
-	return dst, nil
-}
-
-// Decompress4X will decompress a 4X encoded stream.
-// The length of the supplied input must match the end of a block exactly.
-// The *capacity* of the dst slice must match the destination size of
-// the uncompressed data exactly.
-func (d *Decoder) decompress4X8bit(dst, src []byte) ([]byte, error) {
-	if d.actualTableLog == 8 {
-		return d.decompress4X8bitExactly(dst, src)
-	}
-
-	var br [4]bitReaderBytes
-	start := 6
-	for i := 0; i < 3; i++ {
-		length := int(src[i*2]) | (int(src[i*2+1]) << 8)
-		if start+length >= len(src) {
-			return nil, errors.New("truncated input (or invalid offset)")
-		}
-		err := br[i].init(src[start : start+length])
-		if err != nil {
-			return nil, err
-		}
-		start += length
-	}
-	err := br[3].init(src[start:])
-	if err != nil {
-		return nil, err
-	}
-
-	// destination, offset to match first output
-	dstSize := cap(dst)
-	dst = dst[:dstSize]
-	out := dst
-	dstEvery := (dstSize + 3) / 4
-
-	shift := (8 - d.actualTableLog) & 7
-
-	const tlSize = 1 << 8
-	const tlMask = tlSize - 1
-	single := d.dt.single[:tlSize]
-
-	// Use temp table to avoid bound checks/append penalty.
-	var buf [256]byte
-	var off uint8
-	var decoded int
-
-	// Decode 4 values from each decoder/loop.
-	const bufoff = 256 / 4
-	for {
-		if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 {
-			break
-		}
-
-		{
-			// Interleave 2 decodes.
-			const stream = 0
-			const stream2 = 1
-			br[stream].fillFast()
-			br[stream2].fillFast()
-
-			v := single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 := single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+1] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+1] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+2] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+3] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+3] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-		}
-
-		{
-			const stream = 2
-			const stream2 = 3
-			br[stream].fillFast()
-			br[stream2].fillFast()
-
-			v := single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 := single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+1] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+1] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+2] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+3] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+3] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-		}
-
-		off += 4
-
-		if off == bufoff {
-			if bufoff > dstEvery {
-				return nil, errors.New("corruption detected: stream overrun 1")
-			}
-			copy(out, buf[:bufoff])
-			copy(out[dstEvery:], buf[bufoff:bufoff*2])
-			copy(out[dstEvery*2:], buf[bufoff*2:bufoff*3])
-			copy(out[dstEvery*3:], buf[bufoff*3:bufoff*4])
-			off = 0
-			out = out[bufoff:]
-			decoded += 256
-			// There must at least be 3 buffers left.
-			if len(out) < dstEvery*3 {
-				return nil, errors.New("corruption detected: stream overrun 2")
-			}
-		}
-	}
-	if off > 0 {
-		ioff := int(off)
-		if len(out) < dstEvery*3+ioff {
-			return nil, errors.New("corruption detected: stream overrun 3")
-		}
-		copy(out, buf[:off])
-		copy(out[dstEvery:dstEvery+ioff], buf[bufoff:bufoff*2])
-		copy(out[dstEvery*2:dstEvery*2+ioff], buf[bufoff*2:bufoff*3])
-		copy(out[dstEvery*3:dstEvery*3+ioff], buf[bufoff*3:bufoff*4])
-		decoded += int(off) * 4
-		out = out[off:]
-	}
-
-	// Decode remaining.
-	for i := range br {
-		offset := dstEvery * i
-		br := &br[i]
-		bitsLeft := int(br.off*8) + int(64-br.bitsRead)
-		for bitsLeft > 0 {
-			if br.finished() {
-				return nil, io.ErrUnexpectedEOF
-			}
-			if br.bitsRead >= 56 {
-				if br.off >= 4 {
-					v := br.in[br.off-4:]
-					v = v[:4]
-					low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-					br.value |= uint64(low) << (br.bitsRead - 32)
-					br.bitsRead -= 32
-					br.off -= 4
-				} else {
-					for br.off > 0 {
-						br.value |= uint64(br.in[br.off-1]) << (br.bitsRead - 8)
-						br.bitsRead -= 8
-						br.off--
-					}
-				}
-			}
-			// end inline...
-			if offset >= len(out) {
-				return nil, errors.New("corruption detected: stream overrun 4")
-			}
-
-			// Read value and increment offset.
-			v := single[br.peekByteFast()>>shift].entry
-			nBits := uint8(v)
-			br.advance(nBits)
-			bitsLeft -= int(nBits)
-			out[offset] = uint8(v >> 8)
-			offset++
-		}
-		decoded += offset - dstEvery*i
-		err = br.close()
-		if err != nil {
-			return nil, err
-		}
-	}
-	if dstSize != decoded {
-		return nil, errors.New("corruption detected: short output block")
-	}
-	return dst, nil
-}
-
-// Decompress4X will decompress a 4X encoded stream.
-// The length of the supplied input must match the end of a block exactly.
-// The *capacity* of the dst slice must match the destination size of
-// the uncompressed data exactly.
-func (d *Decoder) decompress4X8bitExactly(dst, src []byte) ([]byte, error) {
-	var br [4]bitReaderBytes
-	start := 6
-	for i := 0; i < 3; i++ {
-		length := int(src[i*2]) | (int(src[i*2+1]) << 8)
-		if start+length >= len(src) {
-			return nil, errors.New("truncated input (or invalid offset)")
-		}
-		err := br[i].init(src[start : start+length])
-		if err != nil {
-			return nil, err
-		}
-		start += length
-	}
-	err := br[3].init(src[start:])
-	if err != nil {
-		return nil, err
-	}
-
-	// destination, offset to match first output
-	dstSize := cap(dst)
-	dst = dst[:dstSize]
-	out := dst
-	dstEvery := (dstSize + 3) / 4
-
-	const shift = 0
-	const tlSize = 1 << 8
-	const tlMask = tlSize - 1
-	single := d.dt.single[:tlSize]
-
-	// Use temp table to avoid bound checks/append penalty.
-	var buf [256]byte
-	var off uint8
-	var decoded int
-
-	// Decode 4 values from each decoder/loop.
-	const bufoff = 256 / 4
-	for {
-		if br[0].off < 4 || br[1].off < 4 || br[2].off < 4 || br[3].off < 4 {
-			break
-		}
-
-		{
-			// Interleave 2 decodes.
-			const stream = 0
-			const stream2 = 1
-			br[stream].fillFast()
-			br[stream2].fillFast()
-
-			v := single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 := single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+1] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+1] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+2] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+3] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+3] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-		}
-
-		{
-			const stream = 2
-			const stream2 = 3
-			br[stream].fillFast()
-			br[stream2].fillFast()
-
-			v := single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 := single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+1] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+1] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+2] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+2] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-
-			v = single[br[stream].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream+3] = uint8(v >> 8)
-			br[stream].advance(uint8(v))
-
-			v2 = single[br[stream2].peekByteFast()>>shift].entry
-			buf[off+bufoff*stream2+3] = uint8(v2 >> 8)
-			br[stream2].advance(uint8(v2))
-		}
-
-		off += 4
-
-		if off == bufoff {
-			if bufoff > dstEvery {
-				return nil, errors.New("corruption detected: stream overrun 1")
-			}
-			copy(out, buf[:bufoff])
-			copy(out[dstEvery:], buf[bufoff:bufoff*2])
-			copy(out[dstEvery*2:], buf[bufoff*2:bufoff*3])
-			copy(out[dstEvery*3:], buf[bufoff*3:bufoff*4])
-			off = 0
-			out = out[bufoff:]
-			decoded += 256
-			// There must at least be 3 buffers left.
-			if len(out) < dstEvery*3 {
-				return nil, errors.New("corruption detected: stream overrun 2")
-			}
-		}
-	}
-	if off > 0 {
-		ioff := int(off)
-		if len(out) < dstEvery*3+ioff {
-			return nil, errors.New("corruption detected: stream overrun 3")
-		}
-		copy(out, buf[:off])
-		copy(out[dstEvery:dstEvery+ioff], buf[bufoff:bufoff*2])
-		copy(out[dstEvery*2:dstEvery*2+ioff], buf[bufoff*2:bufoff*3])
-		copy(out[dstEvery*3:dstEvery*3+ioff], buf[bufoff*3:bufoff*4])
-		decoded += int(off) * 4
-		out = out[off:]
-	}
-
-	// Decode remaining.
-	for i := range br {
-		offset := dstEvery * i
-		br := &br[i]
-		bitsLeft := int(br.off*8) + int(64-br.bitsRead)
-		for bitsLeft > 0 {
-			if br.finished() {
-				return nil, io.ErrUnexpectedEOF
-			}
-			if br.bitsRead >= 56 {
-				if br.off >= 4 {
-					v := br.in[br.off-4:]
-					v = v[:4]
-					low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-					br.value |= uint64(low) << (br.bitsRead - 32)
-					br.bitsRead -= 32
-					br.off -= 4
-				} else {
-					for br.off > 0 {
-						br.value |= uint64(br.in[br.off-1]) << (br.bitsRead - 8)
-						br.bitsRead -= 8
-						br.off--
-					}
-				}
-			}
-			// end inline...
-			if offset >= len(out) {
-				return nil, errors.New("corruption detected: stream overrun 4")
-			}
-
-			// Read value and increment offset.
-			v := single[br.peekByteFast()>>shift].entry
-			nBits := uint8(v)
-			br.advance(nBits)
-			bitsLeft -= int(nBits)
-			out[offset] = uint8(v >> 8)
-			offset++
-		}
-		decoded += offset - dstEvery*i
-		err = br.close()
-		if err != nil {
-			return nil, err
-		}
-	}
-	if dstSize != decoded {
-		return nil, errors.New("corruption detected: short output block")
-	}
-	return dst, nil
-}
-
-// matches will compare a decoding table to a coding table.
-// Errors are written to the writer.
-// Nothing will be written if table is ok.
-func (s *Scratch) matches(ct cTable, w io.Writer) {
-	if s == nil || len(s.dt.single) == 0 {
-		return
-	}
-	dt := s.dt.single[:1<<s.actualTableLog]
-	tablelog := s.actualTableLog
-	ok := 0
-	broken := 0
-	for sym, enc := range ct {
-		errs := 0
-		broken++
-		if enc.nBits == 0 {
-			for _, dec := range dt {
-				if uint8(dec.entry>>8) == byte(sym) {
-					fmt.Fprintf(w, "symbol %x has decoder, but no encoder\n", sym)
-					errs++
-					break
-				}
-			}
-			if errs == 0 {
-				broken--
-			}
-			continue
-		}
-		// Unused bits in input
-		ub := tablelog - enc.nBits
-		top := enc.val << ub
-		// decoder looks at top bits.
-		dec := dt[top]
-		if uint8(dec.entry) != enc.nBits {
-			fmt.Fprintf(w, "symbol 0x%x bit size mismatch (enc: %d, dec:%d).\n", sym, enc.nBits, uint8(dec.entry))
-			errs++
-		}
-		if uint8(dec.entry>>8) != uint8(sym) {
-			fmt.Fprintf(w, "symbol 0x%x decoder output mismatch (enc: %d, dec:%d).\n", sym, sym, uint8(dec.entry>>8))
-			errs++
-		}
-		if errs > 0 {
-			fmt.Fprintf(w, "%d errros in base, stopping\n", errs)
-			continue
-		}
-		// Ensure that all combinations are covered.
-		for i := uint16(0); i < (1 << ub); i++ {
-			vval := top | i
-			dec := dt[vval]
-			if uint8(dec.entry) != enc.nBits {
-				fmt.Fprintf(w, "symbol 0x%x bit size mismatch (enc: %d, dec:%d).\n", vval, enc.nBits, uint8(dec.entry))
-				errs++
-			}
-			if uint8(dec.entry>>8) != uint8(sym) {
-				fmt.Fprintf(w, "symbol 0x%x decoder output mismatch (enc: %d, dec:%d).\n", vval, sym, uint8(dec.entry>>8))
-				errs++
-			}
-			if errs > 20 {
-				fmt.Fprintf(w, "%d errros, stopping\n", errs)
-				break
-			}
-		}
-		if errs == 0 {
-			ok++
-			broken--
-		}
-	}
-	if broken > 0 {
-		fmt.Fprintf(w, "%d broken, %d ok\n", broken, ok)
-	}
-}
diff --git a/vendor/github.com/klauspost/compress/huff0/huff0.go b/vendor/github.com/klauspost/compress/huff0/huff0.go
deleted file mode 100644
index 7ec2022b65..0000000000
--- a/vendor/github.com/klauspost/compress/huff0/huff0.go
+++ /dev/null
@@ -1,273 +0,0 @@
-// Package huff0 provides fast huffman encoding as used in zstd.
-//
-// See README.md at https://github.com/klauspost/compress/tree/master/huff0 for details.
-package huff0
-
-import (
-	"errors"
-	"fmt"
-	"math"
-	"math/bits"
-
-	"github.com/klauspost/compress/fse"
-)
-
-const (
-	maxSymbolValue = 255
-
-	// zstandard limits tablelog to 11, see:
-	// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#huffman-tree-description
-	tableLogMax     = 11
-	tableLogDefault = 11
-	minTablelog     = 5
-	huffNodesLen    = 512
-
-	// BlockSizeMax is maximum input size for a single block uncompressed.
-	BlockSizeMax = 1<<18 - 1
-)
-
-var (
-	// ErrIncompressible is returned when input is judged to be too hard to compress.
-	ErrIncompressible = errors.New("input is not compressible")
-
-	// ErrUseRLE is returned from the compressor when the input is a single byte value repeated.
-	ErrUseRLE = errors.New("input is single value repeated")
-
-	// ErrTooBig is return if input is too large for a single block.
-	ErrTooBig = errors.New("input too big")
-
-	// ErrMaxDecodedSizeExceeded is return if input is too large for a single block.
-	ErrMaxDecodedSizeExceeded = errors.New("maximum output size exceeded")
-)
-
-type ReusePolicy uint8
-
-const (
-	// ReusePolicyAllow will allow reuse if it produces smaller output.
-	ReusePolicyAllow ReusePolicy = iota
-
-	// ReusePolicyPrefer will re-use aggressively if possible.
-	// This will not check if a new table will produce smaller output,
-	// except if the current table is impossible to use or
-	// compressed output is bigger than input.
-	ReusePolicyPrefer
-
-	// ReusePolicyNone will disable re-use of tables.
-	// This is slightly faster than ReusePolicyAllow but may produce larger output.
-	ReusePolicyNone
-
-	// ReusePolicyMust must allow reuse and produce smaller output.
-	ReusePolicyMust
-)
-
-type Scratch struct {
-	count [maxSymbolValue + 1]uint32
-
-	// Per block parameters.
-	// These can be used to override compression parameters of the block.
-	// Do not touch, unless you know what you are doing.
-
-	// Out is output buffer.
-	// If the scratch is re-used before the caller is done processing the output,
-	// set this field to nil.
-	// Otherwise the output buffer will be re-used for next Compression/Decompression step
-	// and allocation will be avoided.
-	Out []byte
-
-	// OutTable will contain the table data only, if a new table has been generated.
-	// Slice of the returned data.
-	OutTable []byte
-
-	// OutData will contain the compressed data.
-	// Slice of the returned data.
-	OutData []byte
-
-	// MaxDecodedSize will set the maximum allowed output size.
-	// This value will automatically be set to BlockSizeMax if not set.
-	// Decoders will return ErrMaxDecodedSizeExceeded is this limit is exceeded.
-	MaxDecodedSize int
-
-	br byteReader
-
-	// MaxSymbolValue will override the maximum symbol value of the next block.
-	MaxSymbolValue uint8
-
-	// TableLog will attempt to override the tablelog for the next block.
-	// Must be <= 11 and >= 5.
-	TableLog uint8
-
-	// Reuse will specify the reuse policy
-	Reuse ReusePolicy
-
-	// WantLogLess allows to specify a log 2 reduction that should at least be achieved,
-	// otherwise the block will be returned as incompressible.
-	// The reduction should then at least be (input size >> WantLogLess)
-	// If WantLogLess == 0 any improvement will do.
-	WantLogLess uint8
-
-	symbolLen      uint16 // Length of active part of the symbol table.
-	maxCount       int    // count of the most probable symbol
-	clearCount     bool   // clear count
-	actualTableLog uint8  // Selected tablelog.
-	prevTableLog   uint8  // Tablelog for previous table
-	prevTable      cTable // Table used for previous compression.
-	cTable         cTable // compression table
-	dt             dTable // decompression table
-	nodes          []nodeElt
-	tmpOut         [4][]byte
-	fse            *fse.Scratch
-	huffWeight     [maxSymbolValue + 1]byte
-}
-
-// TransferCTable will transfer the previously used compression table.
-func (s *Scratch) TransferCTable(src *Scratch) {
-	if cap(s.prevTable) < len(src.prevTable) {
-		s.prevTable = make(cTable, 0, maxSymbolValue+1)
-	}
-	s.prevTable = s.prevTable[:len(src.prevTable)]
-	copy(s.prevTable, src.prevTable)
-	s.prevTableLog = src.prevTableLog
-}
-
-func (s *Scratch) prepare(in []byte) (*Scratch, error) {
-	if len(in) > BlockSizeMax {
-		return nil, ErrTooBig
-	}
-	if s == nil {
-		s = &Scratch{}
-	}
-	if s.MaxSymbolValue == 0 {
-		s.MaxSymbolValue = maxSymbolValue
-	}
-	if s.TableLog == 0 {
-		s.TableLog = tableLogDefault
-	}
-	if s.TableLog > tableLogMax || s.TableLog < minTablelog {
-		return nil, fmt.Errorf(" invalid tableLog %d (%d -> %d)", s.TableLog, minTablelog, tableLogMax)
-	}
-	if s.MaxDecodedSize <= 0 || s.MaxDecodedSize > BlockSizeMax {
-		s.MaxDecodedSize = BlockSizeMax
-	}
-	if s.clearCount && s.maxCount == 0 {
-		for i := range s.count {
-			s.count[i] = 0
-		}
-		s.clearCount = false
-	}
-	if cap(s.Out) == 0 {
-		s.Out = make([]byte, 0, len(in))
-	}
-	s.Out = s.Out[:0]
-
-	s.OutTable = nil
-	s.OutData = nil
-	if cap(s.nodes) < huffNodesLen+1 {
-		s.nodes = make([]nodeElt, 0, huffNodesLen+1)
-	}
-	s.nodes = s.nodes[:0]
-	if s.fse == nil {
-		s.fse = &fse.Scratch{}
-	}
-	s.br.init(in)
-
-	return s, nil
-}
-
-type cTable []cTableEntry
-
-func (c cTable) write(s *Scratch) error {
-	var (
-		// precomputed conversion table
-		bitsToWeight [tableLogMax + 1]byte
-		huffLog      = s.actualTableLog
-		// last weight is not saved.
-		maxSymbolValue = uint8(s.symbolLen - 1)
-		huffWeight     = s.huffWeight[:256]
-	)
-	const (
-		maxFSETableLog = 6
-	)
-	// convert to weight
-	bitsToWeight[0] = 0
-	for n := uint8(1); n < huffLog+1; n++ {
-		bitsToWeight[n] = huffLog + 1 - n
-	}
-
-	// Acquire histogram for FSE.
-	hist := s.fse.Histogram()
-	hist = hist[:256]
-	for i := range hist[:16] {
-		hist[i] = 0
-	}
-	for n := uint8(0); n < maxSymbolValue; n++ {
-		v := bitsToWeight[c[n].nBits] & 15
-		huffWeight[n] = v
-		hist[v]++
-	}
-
-	// FSE compress if feasible.
-	if maxSymbolValue >= 2 {
-		huffMaxCnt := uint32(0)
-		huffMax := uint8(0)
-		for i, v := range hist[:16] {
-			if v == 0 {
-				continue
-			}
-			huffMax = byte(i)
-			if v > huffMaxCnt {
-				huffMaxCnt = v
-			}
-		}
-		s.fse.HistogramFinished(huffMax, int(huffMaxCnt))
-		s.fse.TableLog = maxFSETableLog
-		b, err := fse.Compress(huffWeight[:maxSymbolValue], s.fse)
-		if err == nil && len(b) < int(s.symbolLen>>1) {
-			s.Out = append(s.Out, uint8(len(b)))
-			s.Out = append(s.Out, b...)
-			return nil
-		}
-		// Unable to compress (RLE/uncompressible)
-	}
-	// write raw values as 4-bits (max : 15)
-	if maxSymbolValue > (256 - 128) {
-		// should not happen : likely means source cannot be compressed
-		return ErrIncompressible
-	}
-	op := s.Out
-	// special case, pack weights 4 bits/weight.
-	op = append(op, 128|(maxSymbolValue-1))
-	// be sure it doesn't cause msan issue in final combination
-	huffWeight[maxSymbolValue] = 0
-	for n := uint16(0); n < uint16(maxSymbolValue); n += 2 {
-		op = append(op, (huffWeight[n]<<4)|huffWeight[n+1])
-	}
-	s.Out = op
-	return nil
-}
-
-// estimateSize returns the estimated size in bytes of the input represented in the
-// histogram supplied.
-func (c cTable) estimateSize(hist []uint32) int {
-	nbBits := uint32(7)
-	for i, v := range c[:len(hist)] {
-		nbBits += uint32(v.nBits) * hist[i]
-	}
-	return int(nbBits >> 3)
-}
-
-// minSize returns the minimum possible size considering the shannon limit.
-func (s *Scratch) minSize(total int) int {
-	nbBits := float64(7)
-	fTotal := float64(total)
-	for _, v := range s.count[:s.symbolLen] {
-		n := float64(v)
-		if n > 0 {
-			nbBits += math.Log2(fTotal/n) * n
-		}
-	}
-	return int(nbBits) >> 3
-}
-
-func highBit32(val uint32) (n uint32) {
-	return uint32(bits.Len32(val) - 1)
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/README.md b/vendor/github.com/klauspost/compress/zstd/README.md
deleted file mode 100644
index 787813fa9e..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/README.md
+++ /dev/null
@@ -1,431 +0,0 @@
-# zstd 
-
-[Zstandard](https://facebook.github.io/zstd/) is a real-time compression algorithm, providing high compression ratios. 
-It offers a very wide range of compression / speed trade-off, while being backed by a very fast decoder.
-A high performance compression algorithm is implemented. For now focused on speed. 
-
-This package provides [compression](#Compressor) to and [decompression](#Decompressor) of Zstandard content. 
-
-This package is pure Go and without use of "unsafe". 
-
-The `zstd` package is provided as open source software using a Go standard license.
-
-Currently the package is heavily optimized for 64 bit processors and will be significantly slower on 32 bit processors.
-
-## Installation
-
-Install using `go get -u github.com/klauspost/compress`. The package is located in `github.com/klauspost/compress/zstd`.
-
-[![Go Reference](https://pkg.go.dev/badge/github.com/klauspost/compress/zstd.svg)](https://pkg.go.dev/github.com/klauspost/compress/zstd)
-
-## Compressor
-
-### Status: 
-
-STABLE - there may always be subtle bugs, a wide variety of content has been tested and the library is actively 
-used by several projects. This library is being [fuzz-tested](https://github.com/klauspost/compress-fuzz) for all updates.
-
-There may still be specific combinations of data types/size/settings that could lead to edge cases, 
-so as always, testing is recommended.  
-
-For now, a high speed (fastest) and medium-fast (default) compressor has been implemented. 
-
-* The "Fastest" compression ratio is roughly equivalent to zstd level 1. 
-* The "Default" compression ratio is roughly equivalent to zstd level 3 (default).
-* The "Better" compression ratio is roughly equivalent to zstd level 7.
-* The "Best" compression ratio is roughly equivalent to zstd level 11.
-
-In terms of speed, it is typically 2x as fast as the stdlib deflate/gzip in its fastest mode. 
-The compression ratio compared to stdlib is around level 3, but usually 3x as fast.
-
- 
-### Usage
-
-An Encoder can be used for either compressing a stream via the
-`io.WriteCloser` interface supported by the Encoder or as multiple independent
-tasks via the `EncodeAll` function.
-Smaller encodes are encouraged to use the EncodeAll function.
-Use `NewWriter` to create a new instance that can be used for both.
-
-To create a writer with default options, do like this:
-
-```Go
-// Compress input to output.
-func Compress(in io.Reader, out io.Writer) error {
-    enc, err := zstd.NewWriter(out)
-    if err != nil {
-        return err
-    }
-    _, err = io.Copy(enc, in)
-    if err != nil {
-        enc.Close()
-        return err
-    }
-    return enc.Close()
-}
-```
-
-Now you can encode by writing data to `enc`. The output will be finished writing when `Close()` is called.
-Even if your encode fails, you should still call `Close()` to release any resources that may be held up.  
-
-The above is fine for big encodes. However, whenever possible try to *reuse* the writer.
-
-To reuse the encoder, you can use the `Reset(io.Writer)` function to change to another output. 
-This will allow the encoder to reuse all resources and avoid wasteful allocations. 
-
-Currently stream encoding has 'light' concurrency, meaning up to 2 goroutines can be working on part 
-of a stream. This is independent of the `WithEncoderConcurrency(n)`, but that is likely to change 
-in the future. So if you want to limit concurrency for future updates, specify the concurrency
-you would like.
-
-You can specify your desired compression level using `WithEncoderLevel()` option. Currently only pre-defined 
-compression settings can be specified.
-
-#### Future Compatibility Guarantees
-
-This will be an evolving project. When using this package it is important to note that both the compression efficiency and speed may change.
-
-The goal will be to keep the default efficiency at the default zstd (level 3). 
-However the encoding should never be assumed to remain the same, 
-and you should not use hashes of compressed output for similarity checks.
-
-The Encoder can be assumed to produce the same output from the exact same code version.
-However, the may be modes in the future that break this, 
-although they will not be enabled without an explicit option.   
-
-This encoder is not designed to (and will probably never) output the exact same bitstream as the reference encoder.
-
-Also note, that the cgo decompressor currently does not [report all errors on invalid input](https://github.com/DataDog/zstd/issues/59),
-[omits error checks](https://github.com/DataDog/zstd/issues/61), [ignores checksums](https://github.com/DataDog/zstd/issues/43) 
-and seems to ignore concatenated streams, even though [it is part of the spec](https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#frames).
-
-#### Blocks
-
-For compressing small blocks, the returned encoder has a function called `EncodeAll(src, dst []byte) []byte`.
-
-`EncodeAll` will encode all input in src and append it to dst.
-This function can be called concurrently, but each call will only run on a single goroutine.
-
-Encoded blocks can be concatenated and the result will be the combined input stream.
-Data compressed with EncodeAll can be decoded with the Decoder, using either a stream or `DecodeAll`.
-
-Especially when encoding blocks you should take special care to reuse the encoder. 
-This will effectively make it run without allocations after a warmup period. 
-To make it run completely without allocations, supply a destination buffer with space for all content.   
-
-```Go
-import "github.com/klauspost/compress/zstd"
-
-// Create a writer that caches compressors.
-// For this operation type we supply a nil Reader.
-var encoder, _ = zstd.NewWriter(nil)
-
-// Compress a buffer. 
-// If you have a destination buffer, the allocation in the call can also be eliminated.
-func Compress(src []byte) []byte {
-    return encoder.EncodeAll(src, make([]byte, 0, len(src)))
-} 
-```
-
-You can control the maximum number of concurrent encodes using the `WithEncoderConcurrency(n)` 
-option when creating the writer.
-
-Using the Encoder for both a stream and individual blocks concurrently is safe. 
-
-### Performance
-
-I have collected some speed examples to compare speed and compression against other compressors.
-
-* `file` is the input file.
-* `out` is the compressor used. `zskp` is this package. `zstd` is the Datadog cgo library. `gzstd/gzkp` is gzip standard and this library.
-* `level` is the compression level used. For `zskp` level 1 is "fastest", level 2 is "default"; 3 is "better", 4 is "best".
-* `insize`/`outsize` is the input/output size.
-* `millis` is the number of milliseconds used for compression.
-* `mb/s` is megabytes (2^20 bytes) per second.
-
-```
-Silesia Corpus:
-http://sun.aei.polsl.pl/~sdeor/corpus/silesia.zip
-
-This package:
-file    out     level   insize      outsize     millis  mb/s
-silesia.tar zskp    1   211947520   73101992    643     313.87
-silesia.tar zskp    2   211947520   67504318    969     208.38
-silesia.tar zskp    3   211947520   64595893    2007    100.68
-silesia.tar zskp    4   211947520   60995370    7691    26.28
-
-cgo zstd:
-silesia.tar zstd    1   211947520   73605392    543     371.56
-silesia.tar zstd    3   211947520   66793289    864     233.68
-silesia.tar zstd    6   211947520   62916450    1913    105.66
-silesia.tar zstd    9   211947520   60212393    5063    39.92
-
-gzip, stdlib/this package:
-silesia.tar gzstd   1   211947520   80007735    1654    122.21
-silesia.tar gzkp    1   211947520   80369488    1168    173.06
-
-GOB stream of binary data. Highly compressible.
-https://files.klauspost.com/compress/gob-stream.7z
-
-file        out     level   insize  outsize     millis  mb/s
-gob-stream  zskp    1   1911399616  235022249   3088    590.30
-gob-stream  zskp    2   1911399616  205669791   3786    481.34
-gob-stream  zskp    3   1911399616  175034659   9636    189.17
-gob-stream  zskp    4   1911399616  167273881   29337   62.13
-gob-stream  zstd    1   1911399616  249810424   2637    691.26
-gob-stream  zstd    3   1911399616  208192146   3490    522.31
-gob-stream  zstd    6   1911399616  193632038   6687    272.56
-gob-stream  zstd    9   1911399616  177620386   16175   112.70
-gob-stream  gzstd   1   1911399616  357382641   10251   177.82
-gob-stream  gzkp    1   1911399616  362156523   5695    320.08
-
-The test data for the Large Text Compression Benchmark is the first
-10^9 bytes of the English Wikipedia dump on Mar. 3, 2006.
-http://mattmahoney.net/dc/textdata.html
-
-file    out level   insize      outsize     millis  mb/s
-enwik9  zskp    1   1000000000  343848582   3609    264.18
-enwik9  zskp    2   1000000000  317276632   5746    165.97
-enwik9  zskp    3   1000000000  292243069   12162   78.41
-enwik9  zskp    4   1000000000  275241169   36430   26.18
-enwik9  zstd    1   1000000000  358072021   3110    306.65
-enwik9  zstd    3   1000000000  313734672   4784    199.35
-enwik9  zstd    6   1000000000  295138875   10290   92.68
-enwik9  zstd    9   1000000000  278348700   28549   33.40
-enwik9  gzstd   1   1000000000  382578136   9604    99.30
-enwik9  gzkp    1   1000000000  383825945   6544    145.73
-
-Highly compressible JSON file.
-https://files.klauspost.com/compress/github-june-2days-2019.json.zst
-
-file                        out level   insize      outsize     millis  mb/s
-github-june-2days-2019.json zskp    1   6273951764  699045015   10620   563.40
-github-june-2days-2019.json zskp    2   6273951764  617881763   11687   511.96
-github-june-2days-2019.json zskp    3   6273951764  524340691   34043   175.75
-github-june-2days-2019.json zskp    4   6273951764  503314661   93811   63.78
-github-june-2days-2019.json zstd    1   6273951764  766284037   8450    708.00
-github-june-2days-2019.json zstd    3   6273951764  661889476   10927   547.57
-github-june-2days-2019.json zstd    6   6273951764  642756859   22996   260.18
-github-june-2days-2019.json zstd    9   6273951764  601974523   52413   114.16
-github-june-2days-2019.json gzstd   1   6273951764  1164400847  29948   199.79
-github-june-2days-2019.json gzkp    1   6273951764  1128755542  19236   311.03
-
-VM Image, Linux mint with a few installed applications:
-https://files.klauspost.com/compress/rawstudio-mint14.7z
-
-file                    out level   insize      outsize     millis  mb/s
-rawstudio-mint14.tar    zskp    1   8558382592  3667489370  20210   403.84
-rawstudio-mint14.tar    zskp    2   8558382592  3364592300  31873   256.07
-rawstudio-mint14.tar    zskp    3   8558382592  3158085214  77675   105.08
-rawstudio-mint14.tar    zskp    4   8558382592  3020370044  404956  20.16
-rawstudio-mint14.tar    zstd    1   8558382592  3609250104  17136   476.27
-rawstudio-mint14.tar    zstd    3   8558382592  3341679997  29262   278.92
-rawstudio-mint14.tar    zstd    6   8558382592  3235846406  77904   104.77
-rawstudio-mint14.tar    zstd    9   8558382592  3160778861  140946  57.91
-rawstudio-mint14.tar    gzstd   1   8558382592  3926257486  57722   141.40
-rawstudio-mint14.tar    gzkp    1   8558382592  3970463184  41749   195.49
-
-CSV data:
-https://files.klauspost.com/compress/nyc-taxi-data-10M.csv.zst
-
-file                    out level   insize      outsize     millis  mb/s
-nyc-taxi-data-10M.csv   zskp    1   3325605752  641339945   8925    355.35
-nyc-taxi-data-10M.csv   zskp    2   3325605752  591748091   11268   281.44
-nyc-taxi-data-10M.csv   zskp    3   3325605752  530289687   25239   125.66
-nyc-taxi-data-10M.csv   zskp    4   3325605752  490907191   65939   48.10
-nyc-taxi-data-10M.csv   zstd    1   3325605752  687399637   8233    385.18
-nyc-taxi-data-10M.csv   zstd    3   3325605752  598514411   10065   315.07
-nyc-taxi-data-10M.csv   zstd    6   3325605752  570522953   20038   158.27
-nyc-taxi-data-10M.csv   zstd    9   3325605752  517554797   64565   49.12
-nyc-taxi-data-10M.csv   gzstd   1   3325605752  928656485   23876   132.83
-nyc-taxi-data-10M.csv   gzkp    1   3325605752  924718719   16388   193.53
-```
-
-## Decompressor
-
-Staus: STABLE - there may still be subtle bugs, but a wide variety of content has been tested.
-
-This library is being continuously [fuzz-tested](https://github.com/klauspost/compress-fuzz),
-kindly supplied by [fuzzit.dev](https://fuzzit.dev/). 
-The main purpose of the fuzz testing is to ensure that it is not possible to crash the decoder, 
-or run it past its limits with ANY input provided.  
- 
-### Usage
-
-The package has been designed for two main usages, big streams of data and smaller in-memory buffers. 
-There are two main usages of the package for these. Both of them are accessed by creating a `Decoder`.
-
-For streaming use a simple setup could look like this:
-
-```Go
-import "github.com/klauspost/compress/zstd"
-
-func Decompress(in io.Reader, out io.Writer) error {
-    d, err := zstd.NewReader(in)
-    if err != nil {
-        return err
-    }
-    defer d.Close()
-    
-    // Copy content...
-    _, err = io.Copy(out, d)
-    return err
-}
-```
-
-It is important to use the "Close" function when you no longer need the Reader to stop running goroutines. 
-See "Allocation-less operation" below.
-
-For decoding buffers, it could look something like this:
-
-```Go
-import "github.com/klauspost/compress/zstd"
-
-// Create a reader that caches decompressors.
-// For this operation type we supply a nil Reader.
-var decoder, _ = zstd.NewReader(nil)
-
-// Decompress a buffer. We don't supply a destination buffer,
-// so it will be allocated by the decoder.
-func Decompress(src []byte) ([]byte, error) {
-    return decoder.DecodeAll(src, nil)
-} 
-```
-
-Both of these cases should provide the functionality needed. 
-The decoder can be used for *concurrent* decompression of multiple buffers. 
-It will only allow a certain number of concurrent operations to run. 
-To tweak that yourself use the `WithDecoderConcurrency(n)` option when creating the decoder.   
-
-### Dictionaries
-
-Data compressed with [dictionaries](https://github.com/facebook/zstd#the-case-for-small-data-compression) can be decompressed.
-
-Dictionaries are added individually to Decoders.
-Dictionaries are generated by the `zstd --train` command and contains an initial state for the decoder.
-To add a dictionary use the `WithDecoderDicts(dicts ...[]byte)` option with the dictionary data.
-Several dictionaries can be added at once.
-
-The dictionary will be used automatically for the data that specifies them.
-A re-used Decoder will still contain the dictionaries registered.
-
-When registering multiple dictionaries with the same ID, the last one will be used.
-
-It is possible to use dictionaries when compressing data.
-
-To enable a dictionary use `WithEncoderDict(dict []byte)`. Here only one dictionary will be used 
-and it will likely be used even if it doesn't improve compression. 
-
-The used dictionary must be used to decompress the content.
-
-For any real gains, the dictionary should be built with similar data. 
-If an unsuitable dictionary is used the output may be slightly larger than using no dictionary.
-Use the [zstd commandline tool](https://github.com/facebook/zstd/releases) to build a dictionary from sample data.
-For information see [zstd dictionary information](https://github.com/facebook/zstd#the-case-for-small-data-compression). 
-
-For now there is a fixed startup performance penalty for compressing content with dictionaries. 
-This will likely be improved over time. Just be aware to test performance when implementing.  
-
-### Allocation-less operation
-
-The decoder has been designed to operate without allocations after a warmup. 
-
-This means that you should *store* the decoder for best performance. 
-To re-use a stream decoder, use the `Reset(r io.Reader) error` to switch to another stream.
-A decoder can safely be re-used even if the previous stream failed.
-
-To release the resources, you must call the `Close()` function on a decoder.
-After this it can *no longer be reused*, but all running goroutines will be stopped.
-So you *must* use this if you will no longer need the Reader.
-
-For decompressing smaller buffers a single decoder can be used.
-When decoding buffers, you can supply a destination slice with length 0 and your expected capacity.
-In this case no unneeded allocations should be made. 
-
-### Concurrency
-
-The buffer decoder does everything on the same goroutine and does nothing concurrently.
-It can however decode several buffers concurrently. Use `WithDecoderConcurrency(n)` to limit that.
-
-The stream decoder operates on
-
-* One goroutine reads input and splits the input to several block decoders.
-* A number of decoders will decode blocks.
-* A goroutine coordinates these blocks and sends history from one to the next.
-
-So effectively this also means the decoder will "read ahead" and prepare data to always be available for output.
-
-Since "blocks" are quite dependent on the output of the previous block stream decoding will only have limited concurrency.
-
-In practice this means that concurrency is often limited to utilizing about 2 cores effectively.
- 
- 
-### Benchmarks
-
-These are some examples of performance compared to [datadog cgo library](https://github.com/DataDog/zstd).
-
-The first two are streaming decodes and the last are smaller inputs. 
- 
-```
-BenchmarkDecoderSilesia-8                          3     385000067 ns/op     550.51 MB/s        5498 B/op          8 allocs/op
-BenchmarkDecoderSilesiaCgo-8                       6     197666567 ns/op    1072.25 MB/s      270672 B/op          8 allocs/op
-
-BenchmarkDecoderEnwik9-8                           1    2027001600 ns/op     493.34 MB/s       10496 B/op         18 allocs/op
-BenchmarkDecoderEnwik9Cgo-8                        2     979499200 ns/op    1020.93 MB/s      270672 B/op          8 allocs/op
-
-Concurrent performance:
-
-BenchmarkDecoder_DecodeAllParallel/kppkn.gtb.zst-16                28915         42469 ns/op    4340.07 MB/s         114 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/geo.protodata.zst-16           116505          9965 ns/op    11900.16 MB/s         16 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/plrabn12.txt.zst-16              8952        134272 ns/op    3588.70 MB/s         915 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/lcet10.txt.zst-16               11820        102538 ns/op    4161.90 MB/s         594 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/asyoulik.txt.zst-16             34782         34184 ns/op    3661.88 MB/s          60 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/alice29.txt.zst-16              27712         43447 ns/op    3500.58 MB/s          99 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/html_x_4.zst-16                 62826         18750 ns/op    21845.10 MB/s        104 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/paper-100k.pdf.zst-16          631545          1794 ns/op    57078.74 MB/s          2 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/fireworks.jpeg.zst-16         1690140           712 ns/op    172938.13 MB/s         1 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/urls.10K.zst-16                 10432        113593 ns/op    6180.73 MB/s        1143 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/html.zst-16                    113206         10671 ns/op    9596.27 MB/s          15 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallel/comp-data.bin.zst-16          1530615           779 ns/op    5229.49 MB/s           0 B/op          0 allocs/op
-
-BenchmarkDecoder_DecodeAllParallelCgo/kppkn.gtb.zst-16             65217         16192 ns/op    11383.34 MB/s         46 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/geo.protodata.zst-16        292671          4039 ns/op    29363.19 MB/s          6 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/plrabn12.txt.zst-16          26314         46021 ns/op    10470.43 MB/s        293 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/lcet10.txt.zst-16            33897         34900 ns/op    12227.96 MB/s        205 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/asyoulik.txt.zst-16         104348         11433 ns/op    10949.01 MB/s         20 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/alice29.txt.zst-16           75949         15510 ns/op    9805.60 MB/s          32 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/html_x_4.zst-16             173910          6756 ns/op    60624.29 MB/s         37 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/paper-100k.pdf.zst-16       923076          1339 ns/op    76474.87 MB/s          1 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/fireworks.jpeg.zst-16       922920          1351 ns/op    91102.57 MB/s          2 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/urls.10K.zst-16              27649         43618 ns/op    16096.19 MB/s        407 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/html.zst-16                 279073          4160 ns/op    24614.18 MB/s          6 B/op          0 allocs/op
-BenchmarkDecoder_DecodeAllParallelCgo/comp-data.bin.zst-16        749938          1579 ns/op    2581.71 MB/s           0 B/op          0 allocs/op
-```
-
-This reflects the performance around May 2020, but this may be out of date.
-
-## Zstd inside ZIP files
-
-It is possible to use zstandard to compress individual files inside zip archives.
-While this isn't widely supported it can be useful for internal files.
-
-To support the compression and decompression of these files you must register a compressor and decompressor.
-
-It is highly recommended registering the (de)compressors on individual zip Reader/Writer and NOT
-use the global registration functions. The main reason for this is that 2 registrations from 
-different packages will result in a panic.
-
-It is a good idea to only have a single compressor and decompressor, since they can be used for multiple zip
-files concurrently, and using a single instance will allow reusing some resources.
-
-See [this example](https://pkg.go.dev/github.com/klauspost/compress/zstd#example-ZipCompressor) for 
-how to compress and decompress files inside zip archives.
-
-# Contributions
-
-Contributions are always welcome. 
-For new features/fixes, remember to add tests and for performance enhancements include benchmarks.
-
-For general feedback and experience reports, feel free to open an issue or write me on [Twitter](https://twitter.com/sh0dan).
-
-This package includes the excellent [`github.com/cespare/xxhash`](https://github.com/cespare/xxhash) package Copyright (c) 2016 Caleb Spare.
diff --git a/vendor/github.com/klauspost/compress/zstd/bitreader.go b/vendor/github.com/klauspost/compress/zstd/bitreader.go
deleted file mode 100644
index 8544585371..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/bitreader.go
+++ /dev/null
@@ -1,136 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"encoding/binary"
-	"errors"
-	"io"
-	"math/bits"
-)
-
-// bitReader reads a bitstream in reverse.
-// The last set bit indicates the start of the stream and is used
-// for aligning the input.
-type bitReader struct {
-	in       []byte
-	off      uint   // next byte to read is at in[off - 1]
-	value    uint64 // Maybe use [16]byte, but shifting is awkward.
-	bitsRead uint8
-}
-
-// init initializes and resets the bit reader.
-func (b *bitReader) init(in []byte) error {
-	if len(in) < 1 {
-		return errors.New("corrupt stream: too short")
-	}
-	b.in = in
-	b.off = uint(len(in))
-	// The highest bit of the last byte indicates where to start
-	v := in[len(in)-1]
-	if v == 0 {
-		return errors.New("corrupt stream, did not find end of stream")
-	}
-	b.bitsRead = 64
-	b.value = 0
-	if len(in) >= 8 {
-		b.fillFastStart()
-	} else {
-		b.fill()
-		b.fill()
-	}
-	b.bitsRead += 8 - uint8(highBits(uint32(v)))
-	return nil
-}
-
-// getBits will return n bits. n can be 0.
-func (b *bitReader) getBits(n uint8) int {
-	if n == 0 /*|| b.bitsRead >= 64 */ {
-		return 0
-	}
-	return b.getBitsFast(n)
-}
-
-// getBitsFast requires that at least one bit is requested every time.
-// There are no checks if the buffer is filled.
-func (b *bitReader) getBitsFast(n uint8) int {
-	const regMask = 64 - 1
-	v := uint32((b.value << (b.bitsRead & regMask)) >> ((regMask + 1 - n) & regMask))
-	b.bitsRead += n
-	return int(v)
-}
-
-// fillFast() will make sure at least 32 bits are available.
-// There must be at least 4 bytes available.
-func (b *bitReader) fillFast() {
-	if b.bitsRead < 32 {
-		return
-	}
-	// 2 bounds checks.
-	v := b.in[b.off-4:]
-	v = v[:4]
-	low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-	b.value = (b.value << 32) | uint64(low)
-	b.bitsRead -= 32
-	b.off -= 4
-}
-
-// fillFastStart() assumes the bitreader is empty and there is at least 8 bytes to read.
-func (b *bitReader) fillFastStart() {
-	// Do single re-slice to avoid bounds checks.
-	b.value = binary.LittleEndian.Uint64(b.in[b.off-8:])
-	b.bitsRead = 0
-	b.off -= 8
-}
-
-// fill() will make sure at least 32 bits are available.
-func (b *bitReader) fill() {
-	if b.bitsRead < 32 {
-		return
-	}
-	if b.off >= 4 {
-		v := b.in[b.off-4:]
-		v = v[:4]
-		low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
-		b.value = (b.value << 32) | uint64(low)
-		b.bitsRead -= 32
-		b.off -= 4
-		return
-	}
-	for b.off > 0 {
-		b.value = (b.value << 8) | uint64(b.in[b.off-1])
-		b.bitsRead -= 8
-		b.off--
-	}
-}
-
-// finished returns true if all bits have been read from the bit stream.
-func (b *bitReader) finished() bool {
-	return b.off == 0 && b.bitsRead >= 64
-}
-
-// overread returns true if more bits have been requested than is on the stream.
-func (b *bitReader) overread() bool {
-	return b.bitsRead > 64
-}
-
-// remain returns the number of bits remaining.
-func (b *bitReader) remain() uint {
-	return b.off*8 + 64 - uint(b.bitsRead)
-}
-
-// close the bitstream and returns an error if out-of-buffer reads occurred.
-func (b *bitReader) close() error {
-	// Release reference.
-	b.in = nil
-	if b.bitsRead > 64 {
-		return io.ErrUnexpectedEOF
-	}
-	return nil
-}
-
-func highBits(val uint32) (n uint32) {
-	return uint32(bits.Len32(val) - 1)
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/bitwriter.go b/vendor/github.com/klauspost/compress/zstd/bitwriter.go
deleted file mode 100644
index 303ae90f94..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/bitwriter.go
+++ /dev/null
@@ -1,169 +0,0 @@
-// Copyright 2018 Klaus Post. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-// Based on work Copyright (c) 2013, Yann Collet, released under BSD License.
-
-package zstd
-
-import "fmt"
-
-// bitWriter will write bits.
-// First bit will be LSB of the first byte of output.
-type bitWriter struct {
-	bitContainer uint64
-	nBits        uint8
-	out          []byte
-}
-
-// bitMask16 is bitmasks. Has extra to avoid bounds check.
-var bitMask16 = [32]uint16{
-	0, 1, 3, 7, 0xF, 0x1F,
-	0x3F, 0x7F, 0xFF, 0x1FF, 0x3FF, 0x7FF,
-	0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF, 0xFFFF,
-	0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF,
-	0xFFFF, 0xFFFF} /* up to 16 bits */
-
-var bitMask32 = [32]uint32{
-	0, 1, 3, 7, 0xF, 0x1F, 0x3F, 0x7F, 0xFF,
-	0x1FF, 0x3FF, 0x7FF, 0xFFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF,
-	0x1ffff, 0x3ffff, 0x7FFFF, 0xfFFFF, 0x1fFFFF, 0x3fFFFF, 0x7fFFFF, 0xffFFFF,
-	0x1ffFFFF, 0x3ffFFFF, 0x7ffFFFF, 0xfffFFFF, 0x1fffFFFF, 0x3fffFFFF, 0x7fffFFFF,
-} // up to 32 bits
-
-// addBits16NC will add up to 16 bits.
-// It will not check if there is space for them,
-// so the caller must ensure that it has flushed recently.
-func (b *bitWriter) addBits16NC(value uint16, bits uint8) {
-	b.bitContainer |= uint64(value&bitMask16[bits&31]) << (b.nBits & 63)
-	b.nBits += bits
-}
-
-// addBits32NC will add up to 32 bits.
-// It will not check if there is space for them,
-// so the caller must ensure that it has flushed recently.
-func (b *bitWriter) addBits32NC(value uint32, bits uint8) {
-	b.bitContainer |= uint64(value&bitMask32[bits&31]) << (b.nBits & 63)
-	b.nBits += bits
-}
-
-// addBits16Clean will add up to 16 bits. value may not contain more set bits than indicated.
-// It will not check if there is space for them, so the caller must ensure that it has flushed recently.
-func (b *bitWriter) addBits16Clean(value uint16, bits uint8) {
-	b.bitContainer |= uint64(value) << (b.nBits & 63)
-	b.nBits += bits
-}
-
-// flush will flush all pending full bytes.
-// There will be at least 56 bits available for writing when this has been called.
-// Using flush32 is faster, but leaves less space for writing.
-func (b *bitWriter) flush() {
-	v := b.nBits >> 3
-	switch v {
-	case 0:
-	case 1:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-		)
-	case 2:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-		)
-	case 3:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-		)
-	case 4:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-		)
-	case 5:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-		)
-	case 6:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-		)
-	case 7:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-			byte(b.bitContainer>>48),
-		)
-	case 8:
-		b.out = append(b.out,
-			byte(b.bitContainer),
-			byte(b.bitContainer>>8),
-			byte(b.bitContainer>>16),
-			byte(b.bitContainer>>24),
-			byte(b.bitContainer>>32),
-			byte(b.bitContainer>>40),
-			byte(b.bitContainer>>48),
-			byte(b.bitContainer>>56),
-		)
-	default:
-		panic(fmt.Errorf("bits (%d) > 64", b.nBits))
-	}
-	b.bitContainer >>= v << 3
-	b.nBits &= 7
-}
-
-// flush32 will flush out, so there are at least 32 bits available for writing.
-func (b *bitWriter) flush32() {
-	if b.nBits < 32 {
-		return
-	}
-	b.out = append(b.out,
-		byte(b.bitContainer),
-		byte(b.bitContainer>>8),
-		byte(b.bitContainer>>16),
-		byte(b.bitContainer>>24))
-	b.nBits -= 32
-	b.bitContainer >>= 32
-}
-
-// flushAlign will flush remaining full bytes and align to next byte boundary.
-func (b *bitWriter) flushAlign() {
-	nbBytes := (b.nBits + 7) >> 3
-	for i := uint8(0); i < nbBytes; i++ {
-		b.out = append(b.out, byte(b.bitContainer>>(i*8)))
-	}
-	b.nBits = 0
-	b.bitContainer = 0
-}
-
-// close will write the alignment bit and write the final byte(s)
-// to the output.
-func (b *bitWriter) close() error {
-	// End mark
-	b.addBits16Clean(1, 1)
-	// flush until next byte.
-	b.flushAlign()
-	return nil
-}
-
-// reset and continue writing by appending to out.
-func (b *bitWriter) reset(out []byte) {
-	b.bitContainer = 0
-	b.nBits = 0
-	b.out = out
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/blockdec.go b/vendor/github.com/klauspost/compress/zstd/blockdec.go
deleted file mode 100644
index 8a98c4562e..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/blockdec.go
+++ /dev/null
@@ -1,736 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"errors"
-	"fmt"
-	"io"
-	"sync"
-
-	"github.com/klauspost/compress/huff0"
-	"github.com/klauspost/compress/zstd/internal/xxhash"
-)
-
-type blockType uint8
-
-//go:generate stringer -type=blockType,literalsBlockType,seqCompMode,tableIndex
-
-const (
-	blockTypeRaw blockType = iota
-	blockTypeRLE
-	blockTypeCompressed
-	blockTypeReserved
-)
-
-type literalsBlockType uint8
-
-const (
-	literalsBlockRaw literalsBlockType = iota
-	literalsBlockRLE
-	literalsBlockCompressed
-	literalsBlockTreeless
-)
-
-const (
-	// maxCompressedBlockSize is the biggest allowed compressed block size (128KB)
-	maxCompressedBlockSize = 128 << 10
-
-	// Maximum possible block size (all Raw+Uncompressed).
-	maxBlockSize = (1 << 21) - 1
-
-	// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#literals_section_header
-	maxCompressedLiteralSize = 1 << 18
-	maxRLELiteralSize        = 1 << 20
-	maxMatchLen              = 131074
-	maxSequences             = 0x7f00 + 0xffff
-
-	// We support slightly less than the reference decoder to be able to
-	// use ints on 32 bit archs.
-	maxOffsetBits = 30
-)
-
-var (
-	huffDecoderPool = sync.Pool{New: func() interface{} {
-		return &huff0.Scratch{}
-	}}
-
-	fseDecoderPool = sync.Pool{New: func() interface{} {
-		return &fseDecoder{}
-	}}
-)
-
-type blockDec struct {
-	// Raw source data of the block.
-	data        []byte
-	dataStorage []byte
-
-	// Destination of the decoded data.
-	dst []byte
-
-	// Buffer for literals data.
-	literalBuf []byte
-
-	// Window size of the block.
-	WindowSize uint64
-
-	history     chan *history
-	input       chan struct{}
-	result      chan decodeOutput
-	sequenceBuf []seq
-	err         error
-	decWG       sync.WaitGroup
-
-	// Frame to use for singlethreaded decoding.
-	// Should not be used by the decoder itself since parent may be another frame.
-	localFrame *frameDec
-
-	// Block is RLE, this is the size.
-	RLESize uint32
-	tmp     [4]byte
-
-	Type blockType
-
-	// Is this the last block of a frame?
-	Last bool
-
-	// Use less memory
-	lowMem bool
-}
-
-func (b *blockDec) String() string {
-	if b == nil {
-		return "<nil>"
-	}
-	return fmt.Sprintf("Steam Size: %d, Type: %v, Last: %t, Window: %d", len(b.data), b.Type, b.Last, b.WindowSize)
-}
-
-func newBlockDec(lowMem bool) *blockDec {
-	b := blockDec{
-		lowMem:  lowMem,
-		result:  make(chan decodeOutput, 1),
-		input:   make(chan struct{}, 1),
-		history: make(chan *history, 1),
-	}
-	b.decWG.Add(1)
-	go b.startDecoder()
-	return &b
-}
-
-// reset will reset the block.
-// Input must be a start of a block and will be at the end of the block when returned.
-func (b *blockDec) reset(br byteBuffer, windowSize uint64) error {
-	b.WindowSize = windowSize
-	tmp, err := br.readSmall(3)
-	if err != nil {
-		println("Reading block header:", err)
-		return err
-	}
-	bh := uint32(tmp[0]) | (uint32(tmp[1]) << 8) | (uint32(tmp[2]) << 16)
-	b.Last = bh&1 != 0
-	b.Type = blockType((bh >> 1) & 3)
-	// find size.
-	cSize := int(bh >> 3)
-	maxSize := maxBlockSize
-	switch b.Type {
-	case blockTypeReserved:
-		return ErrReservedBlockType
-	case blockTypeRLE:
-		b.RLESize = uint32(cSize)
-		if b.lowMem {
-			maxSize = cSize
-		}
-		cSize = 1
-	case blockTypeCompressed:
-		if debugDecoder {
-			println("Data size on stream:", cSize)
-		}
-		b.RLESize = 0
-		maxSize = maxCompressedBlockSize
-		if windowSize < maxCompressedBlockSize && b.lowMem {
-			maxSize = int(windowSize)
-		}
-		if cSize > maxCompressedBlockSize || uint64(cSize) > b.WindowSize {
-			if debugDecoder {
-				printf("compressed block too big: csize:%d block: %+v\n", uint64(cSize), b)
-			}
-			return ErrCompressedSizeTooBig
-		}
-	case blockTypeRaw:
-		b.RLESize = 0
-		// We do not need a destination for raw blocks.
-		maxSize = -1
-	default:
-		panic("Invalid block type")
-	}
-
-	// Read block data.
-	if cap(b.dataStorage) < cSize {
-		if b.lowMem || cSize > maxCompressedBlockSize {
-			b.dataStorage = make([]byte, 0, cSize)
-		} else {
-			b.dataStorage = make([]byte, 0, maxCompressedBlockSize)
-		}
-	}
-	if cap(b.dst) <= maxSize {
-		b.dst = make([]byte, 0, maxSize+1)
-	}
-	b.data, err = br.readBig(cSize, b.dataStorage)
-	if err != nil {
-		if debugDecoder {
-			println("Reading block:", err, "(", cSize, ")", len(b.data))
-			printf("%T", br)
-		}
-		return err
-	}
-	return nil
-}
-
-// sendEOF will make the decoder send EOF on this frame.
-func (b *blockDec) sendErr(err error) {
-	b.Last = true
-	b.Type = blockTypeReserved
-	b.err = err
-	b.input <- struct{}{}
-}
-
-// Close will release resources.
-// Closed blockDec cannot be reset.
-func (b *blockDec) Close() {
-	close(b.input)
-	close(b.history)
-	close(b.result)
-	b.decWG.Wait()
-}
-
-// decodeAsync will prepare decoding the block when it receives input.
-// This will separate output and history.
-func (b *blockDec) startDecoder() {
-	defer b.decWG.Done()
-	for range b.input {
-		//println("blockDec: Got block input")
-		switch b.Type {
-		case blockTypeRLE:
-			if cap(b.dst) < int(b.RLESize) {
-				if b.lowMem {
-					b.dst = make([]byte, b.RLESize)
-				} else {
-					b.dst = make([]byte, maxBlockSize)
-				}
-			}
-			o := decodeOutput{
-				d:   b,
-				b:   b.dst[:b.RLESize],
-				err: nil,
-			}
-			v := b.data[0]
-			for i := range o.b {
-				o.b[i] = v
-			}
-			hist := <-b.history
-			hist.append(o.b)
-			b.result <- o
-		case blockTypeRaw:
-			o := decodeOutput{
-				d:   b,
-				b:   b.data,
-				err: nil,
-			}
-			hist := <-b.history
-			hist.append(o.b)
-			b.result <- o
-		case blockTypeCompressed:
-			b.dst = b.dst[:0]
-			err := b.decodeCompressed(nil)
-			o := decodeOutput{
-				d:   b,
-				b:   b.dst,
-				err: err,
-			}
-			if debugDecoder {
-				println("Decompressed to", len(b.dst), "bytes, error:", err)
-			}
-			b.result <- o
-		case blockTypeReserved:
-			// Used for returning errors.
-			<-b.history
-			b.result <- decodeOutput{
-				d:   b,
-				b:   nil,
-				err: b.err,
-			}
-		default:
-			panic("Invalid block type")
-		}
-		if debugDecoder {
-			println("blockDec: Finished block")
-		}
-	}
-}
-
-// decodeAsync will prepare decoding the block when it receives the history.
-// If history is provided, it will not fetch it from the channel.
-func (b *blockDec) decodeBuf(hist *history) error {
-	switch b.Type {
-	case blockTypeRLE:
-		if cap(b.dst) < int(b.RLESize) {
-			if b.lowMem {
-				b.dst = make([]byte, b.RLESize)
-			} else {
-				b.dst = make([]byte, maxBlockSize)
-			}
-		}
-		b.dst = b.dst[:b.RLESize]
-		v := b.data[0]
-		for i := range b.dst {
-			b.dst[i] = v
-		}
-		hist.appendKeep(b.dst)
-		return nil
-	case blockTypeRaw:
-		hist.appendKeep(b.data)
-		return nil
-	case blockTypeCompressed:
-		saved := b.dst
-		b.dst = hist.b
-		hist.b = nil
-		err := b.decodeCompressed(hist)
-		if debugDecoder {
-			println("Decompressed to total", len(b.dst), "bytes, hash:", xxhash.Sum64(b.dst), "error:", err)
-		}
-		hist.b = b.dst
-		b.dst = saved
-		return err
-	case blockTypeReserved:
-		// Used for returning errors.
-		return b.err
-	default:
-		panic("Invalid block type")
-	}
-}
-
-// decodeCompressed will start decompressing a block.
-// If no history is supplied the decoder will decodeAsync as much as possible
-// before fetching from blockDec.history
-func (b *blockDec) decodeCompressed(hist *history) error {
-	in := b.data
-	delayedHistory := hist == nil
-
-	if delayedHistory {
-		// We must always grab history.
-		defer func() {
-			if hist == nil {
-				<-b.history
-			}
-		}()
-	}
-	// There must be at least one byte for Literals_Block_Type and one for Sequences_Section_Header
-	if len(in) < 2 {
-		return ErrBlockTooSmall
-	}
-	litType := literalsBlockType(in[0] & 3)
-	var litRegenSize int
-	var litCompSize int
-	sizeFormat := (in[0] >> 2) & 3
-	var fourStreams bool
-	switch litType {
-	case literalsBlockRaw, literalsBlockRLE:
-		switch sizeFormat {
-		case 0, 2:
-			// Regenerated_Size uses 5 bits (0-31). Literals_Section_Header uses 1 byte.
-			litRegenSize = int(in[0] >> 3)
-			in = in[1:]
-		case 1:
-			// Regenerated_Size uses 12 bits (0-4095). Literals_Section_Header uses 2 bytes.
-			litRegenSize = int(in[0]>>4) + (int(in[1]) << 4)
-			in = in[2:]
-		case 3:
-			//  Regenerated_Size uses 20 bits (0-1048575). Literals_Section_Header uses 3 bytes.
-			if len(in) < 3 {
-				println("too small: litType:", litType, " sizeFormat", sizeFormat, len(in))
-				return ErrBlockTooSmall
-			}
-			litRegenSize = int(in[0]>>4) + (int(in[1]) << 4) + (int(in[2]) << 12)
-			in = in[3:]
-		}
-	case literalsBlockCompressed, literalsBlockTreeless:
-		switch sizeFormat {
-		case 0, 1:
-			// Both Regenerated_Size and Compressed_Size use 10 bits (0-1023).
-			if len(in) < 3 {
-				println("too small: litType:", litType, " sizeFormat", sizeFormat, len(in))
-				return ErrBlockTooSmall
-			}
-			n := uint64(in[0]>>4) + (uint64(in[1]) << 4) + (uint64(in[2]) << 12)
-			litRegenSize = int(n & 1023)
-			litCompSize = int(n >> 10)
-			fourStreams = sizeFormat == 1
-			in = in[3:]
-		case 2:
-			fourStreams = true
-			if len(in) < 4 {
-				println("too small: litType:", litType, " sizeFormat", sizeFormat, len(in))
-				return ErrBlockTooSmall
-			}
-			n := uint64(in[0]>>4) + (uint64(in[1]) << 4) + (uint64(in[2]) << 12) + (uint64(in[3]) << 20)
-			litRegenSize = int(n & 16383)
-			litCompSize = int(n >> 14)
-			in = in[4:]
-		case 3:
-			fourStreams = true
-			if len(in) < 5 {
-				println("too small: litType:", litType, " sizeFormat", sizeFormat, len(in))
-				return ErrBlockTooSmall
-			}
-			n := uint64(in[0]>>4) + (uint64(in[1]) << 4) + (uint64(in[2]) << 12) + (uint64(in[3]) << 20) + (uint64(in[4]) << 28)
-			litRegenSize = int(n & 262143)
-			litCompSize = int(n >> 18)
-			in = in[5:]
-		}
-	}
-	if debugDecoder {
-		println("literals type:", litType, "litRegenSize:", litRegenSize, "litCompSize:", litCompSize, "sizeFormat:", sizeFormat, "4X:", fourStreams)
-	}
-	var literals []byte
-	var huff *huff0.Scratch
-	switch litType {
-	case literalsBlockRaw:
-		if len(in) < litRegenSize {
-			println("too small: litType:", litType, " sizeFormat", sizeFormat, "remain:", len(in), "want:", litRegenSize)
-			return ErrBlockTooSmall
-		}
-		literals = in[:litRegenSize]
-		in = in[litRegenSize:]
-		//printf("Found %d uncompressed literals\n", litRegenSize)
-	case literalsBlockRLE:
-		if len(in) < 1 {
-			println("too small: litType:", litType, " sizeFormat", sizeFormat, "remain:", len(in), "want:", 1)
-			return ErrBlockTooSmall
-		}
-		if cap(b.literalBuf) < litRegenSize {
-			if b.lowMem {
-				b.literalBuf = make([]byte, litRegenSize)
-			} else {
-				if litRegenSize > maxCompressedLiteralSize {
-					// Exceptional
-					b.literalBuf = make([]byte, litRegenSize)
-				} else {
-					b.literalBuf = make([]byte, litRegenSize, maxCompressedLiteralSize)
-
-				}
-			}
-		}
-		literals = b.literalBuf[:litRegenSize]
-		v := in[0]
-		for i := range literals {
-			literals[i] = v
-		}
-		in = in[1:]
-		if debugDecoder {
-			printf("Found %d RLE compressed literals\n", litRegenSize)
-		}
-	case literalsBlockTreeless:
-		if len(in) < litCompSize {
-			println("too small: litType:", litType, " sizeFormat", sizeFormat, "remain:", len(in), "want:", litCompSize)
-			return ErrBlockTooSmall
-		}
-		// Store compressed literals, so we defer decoding until we get history.
-		literals = in[:litCompSize]
-		in = in[litCompSize:]
-		if debugDecoder {
-			printf("Found %d compressed literals\n", litCompSize)
-		}
-	case literalsBlockCompressed:
-		if len(in) < litCompSize {
-			println("too small: litType:", litType, " sizeFormat", sizeFormat, "remain:", len(in), "want:", litCompSize)
-			return ErrBlockTooSmall
-		}
-		literals = in[:litCompSize]
-		in = in[litCompSize:]
-		huff = huffDecoderPool.Get().(*huff0.Scratch)
-		var err error
-		// Ensure we have space to store it.
-		if cap(b.literalBuf) < litRegenSize {
-			if b.lowMem {
-				b.literalBuf = make([]byte, 0, litRegenSize)
-			} else {
-				b.literalBuf = make([]byte, 0, maxCompressedLiteralSize)
-			}
-		}
-		if huff == nil {
-			huff = &huff0.Scratch{}
-		}
-		huff, literals, err = huff0.ReadTable(literals, huff)
-		if err != nil {
-			println("reading huffman table:", err)
-			return err
-		}
-		// Use our out buffer.
-		if fourStreams {
-			literals, err = huff.Decoder().Decompress4X(b.literalBuf[:0:litRegenSize], literals)
-		} else {
-			literals, err = huff.Decoder().Decompress1X(b.literalBuf[:0:litRegenSize], literals)
-		}
-		if err != nil {
-			println("decoding compressed literals:", err)
-			return err
-		}
-		// Make sure we don't leak our literals buffer
-		if len(literals) != litRegenSize {
-			return fmt.Errorf("literal output size mismatch want %d, got %d", litRegenSize, len(literals))
-		}
-		if debugDecoder {
-			printf("Decompressed %d literals into %d bytes\n", litCompSize, litRegenSize)
-		}
-	}
-
-	// Decode Sequences
-	// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#sequences-section
-	if len(in) < 1 {
-		return ErrBlockTooSmall
-	}
-	seqHeader := in[0]
-	nSeqs := 0
-	switch {
-	case seqHeader == 0:
-		in = in[1:]
-	case seqHeader < 128:
-		nSeqs = int(seqHeader)
-		in = in[1:]
-	case seqHeader < 255:
-		if len(in) < 2 {
-			return ErrBlockTooSmall
-		}
-		nSeqs = int(seqHeader-128)<<8 | int(in[1])
-		in = in[2:]
-	case seqHeader == 255:
-		if len(in) < 3 {
-			return ErrBlockTooSmall
-		}
-		nSeqs = 0x7f00 + int(in[1]) + (int(in[2]) << 8)
-		in = in[3:]
-	}
-	// Allocate sequences
-	if cap(b.sequenceBuf) < nSeqs {
-		if b.lowMem {
-			b.sequenceBuf = make([]seq, nSeqs)
-		} else {
-			// Allocate max
-			b.sequenceBuf = make([]seq, nSeqs, maxSequences)
-		}
-	} else {
-		// Reuse buffer
-		b.sequenceBuf = b.sequenceBuf[:nSeqs]
-	}
-	var seqs = &sequenceDecs{}
-	if nSeqs > 0 {
-		if len(in) < 1 {
-			return ErrBlockTooSmall
-		}
-		br := byteReader{b: in, off: 0}
-		compMode := br.Uint8()
-		br.advance(1)
-		if debugDecoder {
-			printf("Compression modes: 0b%b", compMode)
-		}
-		for i := uint(0); i < 3; i++ {
-			mode := seqCompMode((compMode >> (6 - i*2)) & 3)
-			if debugDecoder {
-				println("Table", tableIndex(i), "is", mode)
-			}
-			var seq *sequenceDec
-			switch tableIndex(i) {
-			case tableLiteralLengths:
-				seq = &seqs.litLengths
-			case tableOffsets:
-				seq = &seqs.offsets
-			case tableMatchLengths:
-				seq = &seqs.matchLengths
-			default:
-				panic("unknown table")
-			}
-			switch mode {
-			case compModePredefined:
-				seq.fse = &fsePredef[i]
-			case compModeRLE:
-				if br.remain() < 1 {
-					return ErrBlockTooSmall
-				}
-				v := br.Uint8()
-				br.advance(1)
-				dec := fseDecoderPool.Get().(*fseDecoder)
-				symb, err := decSymbolValue(v, symbolTableX[i])
-				if err != nil {
-					printf("RLE Transform table (%v) error: %v", tableIndex(i), err)
-					return err
-				}
-				dec.setRLE(symb)
-				seq.fse = dec
-				if debugDecoder {
-					printf("RLE set to %+v, code: %v", symb, v)
-				}
-			case compModeFSE:
-				println("Reading table for", tableIndex(i))
-				dec := fseDecoderPool.Get().(*fseDecoder)
-				err := dec.readNCount(&br, uint16(maxTableSymbol[i]))
-				if err != nil {
-					println("Read table error:", err)
-					return err
-				}
-				err = dec.transform(symbolTableX[i])
-				if err != nil {
-					println("Transform table error:", err)
-					return err
-				}
-				if debugDecoder {
-					println("Read table ok", "symbolLen:", dec.symbolLen)
-				}
-				seq.fse = dec
-			case compModeRepeat:
-				seq.repeat = true
-			}
-			if br.overread() {
-				return io.ErrUnexpectedEOF
-			}
-		}
-		in = br.unread()
-	}
-
-	// Wait for history.
-	// All time spent after this is critical since it is strictly sequential.
-	if hist == nil {
-		hist = <-b.history
-		if hist.error {
-			return ErrDecoderClosed
-		}
-	}
-
-	// Decode treeless literal block.
-	if litType == literalsBlockTreeless {
-		// TODO: We could send the history early WITHOUT the stream history.
-		//   This would allow decoding treeless literals before the byte history is available.
-		//   Silencia stats: Treeless 4393, with: 32775, total: 37168, 11% treeless.
-		//   So not much obvious gain here.
-
-		if hist.huffTree == nil {
-			return errors.New("literal block was treeless, but no history was defined")
-		}
-		// Ensure we have space to store it.
-		if cap(b.literalBuf) < litRegenSize {
-			if b.lowMem {
-				b.literalBuf = make([]byte, 0, litRegenSize)
-			} else {
-				b.literalBuf = make([]byte, 0, maxCompressedLiteralSize)
-			}
-		}
-		var err error
-		// Use our out buffer.
-		huff = hist.huffTree
-		if fourStreams {
-			literals, err = huff.Decoder().Decompress4X(b.literalBuf[:0:litRegenSize], literals)
-		} else {
-			literals, err = huff.Decoder().Decompress1X(b.literalBuf[:0:litRegenSize], literals)
-		}
-		// Make sure we don't leak our literals buffer
-		if err != nil {
-			println("decompressing literals:", err)
-			return err
-		}
-		if len(literals) != litRegenSize {
-			return fmt.Errorf("literal output size mismatch want %d, got %d", litRegenSize, len(literals))
-		}
-	} else {
-		if hist.huffTree != nil && huff != nil {
-			if hist.dict == nil || hist.dict.litEnc != hist.huffTree {
-				huffDecoderPool.Put(hist.huffTree)
-			}
-			hist.huffTree = nil
-		}
-	}
-	if huff != nil {
-		hist.huffTree = huff
-	}
-	if debugDecoder {
-		println("Final literals:", len(literals), "hash:", xxhash.Sum64(literals), "and", nSeqs, "sequences.")
-	}
-
-	if nSeqs == 0 {
-		// Decompressed content is defined entirely as Literals Section content.
-		b.dst = append(b.dst, literals...)
-		if delayedHistory {
-			hist.append(literals)
-		}
-		return nil
-	}
-
-	seqs, err := seqs.mergeHistory(&hist.decoders)
-	if err != nil {
-		return err
-	}
-	if debugDecoder {
-		println("History merged ok")
-	}
-	br := &bitReader{}
-	if err := br.init(in); err != nil {
-		return err
-	}
-
-	// TODO: Investigate if sending history without decoders are faster.
-	//   This would allow the sequences to be decoded async and only have to construct stream history.
-	//   If only recent offsets were not transferred, this would be an obvious win.
-	// 	 Also, if first 3 sequences don't reference recent offsets, all sequences can be decoded.
-
-	hbytes := hist.b
-	if len(hbytes) > hist.windowSize {
-		hbytes = hbytes[len(hbytes)-hist.windowSize:]
-		// We do not need history any more.
-		if hist.dict != nil {
-			hist.dict.content = nil
-		}
-	}
-
-	if err := seqs.initialize(br, hist, literals, b.dst); err != nil {
-		println("initializing sequences:", err)
-		return err
-	}
-
-	err = seqs.decode(nSeqs, br, hbytes)
-	if err != nil {
-		return err
-	}
-	if !br.finished() {
-		return fmt.Errorf("%d extra bits on block, should be 0", br.remain())
-	}
-
-	err = br.close()
-	if err != nil {
-		printf("Closing sequences: %v, %+v\n", err, *br)
-	}
-	if len(b.data) > maxCompressedBlockSize {
-		return fmt.Errorf("compressed block size too large (%d)", len(b.data))
-	}
-	// Set output and release references.
-	b.dst = seqs.out
-	seqs.out, seqs.literals, seqs.hist = nil, nil, nil
-
-	if !delayedHistory {
-		// If we don't have delayed history, no need to update.
-		hist.recentOffsets = seqs.prevOffset
-		return nil
-	}
-	if b.Last {
-		// if last block we don't care about history.
-		println("Last block, no history returned")
-		hist.b = hist.b[:0]
-		return nil
-	}
-	hist.append(b.dst)
-	hist.recentOffsets = seqs.prevOffset
-	if debugDecoder {
-		println("Finished block with literals:", len(literals), "and", nSeqs, "sequences.")
-	}
-
-	return nil
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/blockenc.go b/vendor/github.com/klauspost/compress/zstd/blockenc.go
deleted file mode 100644
index 3df185ee46..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/blockenc.go
+++ /dev/null
@@ -1,871 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"errors"
-	"fmt"
-	"math"
-	"math/bits"
-
-	"github.com/klauspost/compress/huff0"
-)
-
-type blockEnc struct {
-	size       int
-	literals   []byte
-	sequences  []seq
-	coders     seqCoders
-	litEnc     *huff0.Scratch
-	dictLitEnc *huff0.Scratch
-	wr         bitWriter
-
-	extraLits         int
-	output            []byte
-	recentOffsets     [3]uint32
-	prevRecentOffsets [3]uint32
-
-	last   bool
-	lowMem bool
-}
-
-// init should be used once the block has been created.
-// If called more than once, the effect is the same as calling reset.
-func (b *blockEnc) init() {
-	if b.lowMem {
-		// 1K literals
-		if cap(b.literals) < 1<<10 {
-			b.literals = make([]byte, 0, 1<<10)
-		}
-		const defSeqs = 20
-		if cap(b.sequences) < defSeqs {
-			b.sequences = make([]seq, 0, defSeqs)
-		}
-		// 1K
-		if cap(b.output) < 1<<10 {
-			b.output = make([]byte, 0, 1<<10)
-		}
-	} else {
-		if cap(b.literals) < maxCompressedBlockSize {
-			b.literals = make([]byte, 0, maxCompressedBlockSize)
-		}
-		const defSeqs = 200
-		if cap(b.sequences) < defSeqs {
-			b.sequences = make([]seq, 0, defSeqs)
-		}
-		if cap(b.output) < maxCompressedBlockSize {
-			b.output = make([]byte, 0, maxCompressedBlockSize)
-		}
-	}
-
-	if b.coders.mlEnc == nil {
-		b.coders.mlEnc = &fseEncoder{}
-		b.coders.mlPrev = &fseEncoder{}
-		b.coders.ofEnc = &fseEncoder{}
-		b.coders.ofPrev = &fseEncoder{}
-		b.coders.llEnc = &fseEncoder{}
-		b.coders.llPrev = &fseEncoder{}
-	}
-	b.litEnc = &huff0.Scratch{WantLogLess: 4}
-	b.reset(nil)
-}
-
-// initNewEncode can be used to reset offsets and encoders to the initial state.
-func (b *blockEnc) initNewEncode() {
-	b.recentOffsets = [3]uint32{1, 4, 8}
-	b.litEnc.Reuse = huff0.ReusePolicyNone
-	b.coders.setPrev(nil, nil, nil)
-}
-
-// reset will reset the block for a new encode, but in the same stream,
-// meaning that state will be carried over, but the block content is reset.
-// If a previous block is provided, the recent offsets are carried over.
-func (b *blockEnc) reset(prev *blockEnc) {
-	b.extraLits = 0
-	b.literals = b.literals[:0]
-	b.size = 0
-	b.sequences = b.sequences[:0]
-	b.output = b.output[:0]
-	b.last = false
-	if prev != nil {
-		b.recentOffsets = prev.prevRecentOffsets
-	}
-	b.dictLitEnc = nil
-}
-
-// reset will reset the block for a new encode, but in the same stream,
-// meaning that state will be carried over, but the block content is reset.
-// If a previous block is provided, the recent offsets are carried over.
-func (b *blockEnc) swapEncoders(prev *blockEnc) {
-	b.coders.swap(&prev.coders)
-	b.litEnc, prev.litEnc = prev.litEnc, b.litEnc
-}
-
-// blockHeader contains the information for a block header.
-type blockHeader uint32
-
-// setLast sets the 'last' indicator on a block.
-func (h *blockHeader) setLast(b bool) {
-	if b {
-		*h = *h | 1
-	} else {
-		const mask = (1 << 24) - 2
-		*h = *h & mask
-	}
-}
-
-// setSize will store the compressed size of a block.
-func (h *blockHeader) setSize(v uint32) {
-	const mask = 7
-	*h = (*h)&mask | blockHeader(v<<3)
-}
-
-// setType sets the block type.
-func (h *blockHeader) setType(t blockType) {
-	const mask = 1 | (((1 << 24) - 1) ^ 7)
-	*h = (*h & mask) | blockHeader(t<<1)
-}
-
-// appendTo will append the block header to a slice.
-func (h blockHeader) appendTo(b []byte) []byte {
-	return append(b, uint8(h), uint8(h>>8), uint8(h>>16))
-}
-
-// String returns a string representation of the block.
-func (h blockHeader) String() string {
-	return fmt.Sprintf("Type: %d, Size: %d, Last:%t", (h>>1)&3, h>>3, h&1 == 1)
-}
-
-// literalsHeader contains literals header information.
-type literalsHeader uint64
-
-// setType can be used to set the type of literal block.
-func (h *literalsHeader) setType(t literalsBlockType) {
-	const mask = math.MaxUint64 - 3
-	*h = (*h & mask) | literalsHeader(t)
-}
-
-// setSize can be used to set a single size, for uncompressed and RLE content.
-func (h *literalsHeader) setSize(regenLen int) {
-	inBits := bits.Len32(uint32(regenLen))
-	// Only retain 2 bits
-	const mask = 3
-	lh := uint64(*h & mask)
-	switch {
-	case inBits < 5:
-		lh |= (uint64(regenLen) << 3) | (1 << 60)
-		if debugEncoder {
-			got := int(lh>>3) & 0xff
-			if got != regenLen {
-				panic(fmt.Sprint("litRegenSize = ", regenLen, "(want) != ", got, "(got)"))
-			}
-		}
-	case inBits < 12:
-		lh |= (1 << 2) | (uint64(regenLen) << 4) | (2 << 60)
-	case inBits < 20:
-		lh |= (3 << 2) | (uint64(regenLen) << 4) | (3 << 60)
-	default:
-		panic(fmt.Errorf("internal error: block too big (%d)", regenLen))
-	}
-	*h = literalsHeader(lh)
-}
-
-// setSizes will set the size of a compressed literals section and the input length.
-func (h *literalsHeader) setSizes(compLen, inLen int, single bool) {
-	compBits, inBits := bits.Len32(uint32(compLen)), bits.Len32(uint32(inLen))
-	// Only retain 2 bits
-	const mask = 3
-	lh := uint64(*h & mask)
-	switch {
-	case compBits <= 10 && inBits <= 10:
-		if !single {
-			lh |= 1 << 2
-		}
-		lh |= (uint64(inLen) << 4) | (uint64(compLen) << (10 + 4)) | (3 << 60)
-		if debugEncoder {
-			const mmask = (1 << 24) - 1
-			n := (lh >> 4) & mmask
-			if int(n&1023) != inLen {
-				panic(fmt.Sprint("regensize:", int(n&1023), "!=", inLen, inBits))
-			}
-			if int(n>>10) != compLen {
-				panic(fmt.Sprint("compsize:", int(n>>10), "!=", compLen, compBits))
-			}
-		}
-	case compBits <= 14 && inBits <= 14:
-		lh |= (2 << 2) | (uint64(inLen) << 4) | (uint64(compLen) << (14 + 4)) | (4 << 60)
-		if single {
-			panic("single stream used with more than 10 bits length.")
-		}
-	case compBits <= 18 && inBits <= 18:
-		lh |= (3 << 2) | (uint64(inLen) << 4) | (uint64(compLen) << (18 + 4)) | (5 << 60)
-		if single {
-			panic("single stream used with more than 10 bits length.")
-		}
-	default:
-		panic("internal error: block too big")
-	}
-	*h = literalsHeader(lh)
-}
-
-// appendTo will append the literals header to a byte slice.
-func (h literalsHeader) appendTo(b []byte) []byte {
-	size := uint8(h >> 60)
-	switch size {
-	case 1:
-		b = append(b, uint8(h))
-	case 2:
-		b = append(b, uint8(h), uint8(h>>8))
-	case 3:
-		b = append(b, uint8(h), uint8(h>>8), uint8(h>>16))
-	case 4:
-		b = append(b, uint8(h), uint8(h>>8), uint8(h>>16), uint8(h>>24))
-	case 5:
-		b = append(b, uint8(h), uint8(h>>8), uint8(h>>16), uint8(h>>24), uint8(h>>32))
-	default:
-		panic(fmt.Errorf("internal error: literalsHeader has invalid size (%d)", size))
-	}
-	return b
-}
-
-// size returns the output size with currently set values.
-func (h literalsHeader) size() int {
-	return int(h >> 60)
-}
-
-func (h literalsHeader) String() string {
-	return fmt.Sprintf("Type: %d, SizeFormat: %d, Size: 0x%d, Bytes:%d", literalsBlockType(h&3), (h>>2)&3, h&((1<<60)-1)>>4, h>>60)
-}
-
-// pushOffsets will push the recent offsets to the backup store.
-func (b *blockEnc) pushOffsets() {
-	b.prevRecentOffsets = b.recentOffsets
-}
-
-// pushOffsets will push the recent offsets to the backup store.
-func (b *blockEnc) popOffsets() {
-	b.recentOffsets = b.prevRecentOffsets
-}
-
-// matchOffset will adjust recent offsets and return the adjusted one,
-// if it matches a previous offset.
-func (b *blockEnc) matchOffset(offset, lits uint32) uint32 {
-	// Check if offset is one of the recent offsets.
-	// Adjusts the output offset accordingly.
-	// Gives a tiny bit of compression, typically around 1%.
-	if true {
-		if lits > 0 {
-			switch offset {
-			case b.recentOffsets[0]:
-				offset = 1
-			case b.recentOffsets[1]:
-				b.recentOffsets[1] = b.recentOffsets[0]
-				b.recentOffsets[0] = offset
-				offset = 2
-			case b.recentOffsets[2]:
-				b.recentOffsets[2] = b.recentOffsets[1]
-				b.recentOffsets[1] = b.recentOffsets[0]
-				b.recentOffsets[0] = offset
-				offset = 3
-			default:
-				b.recentOffsets[2] = b.recentOffsets[1]
-				b.recentOffsets[1] = b.recentOffsets[0]
-				b.recentOffsets[0] = offset
-				offset += 3
-			}
-		} else {
-			switch offset {
-			case b.recentOffsets[1]:
-				b.recentOffsets[1] = b.recentOffsets[0]
-				b.recentOffsets[0] = offset
-				offset = 1
-			case b.recentOffsets[2]:
-				b.recentOffsets[2] = b.recentOffsets[1]
-				b.recentOffsets[1] = b.recentOffsets[0]
-				b.recentOffsets[0] = offset
-				offset = 2
-			case b.recentOffsets[0] - 1:
-				b.recentOffsets[2] = b.recentOffsets[1]
-				b.recentOffsets[1] = b.recentOffsets[0]
-				b.recentOffsets[0] = offset
-				offset = 3
-			default:
-				b.recentOffsets[2] = b.recentOffsets[1]
-				b.recentOffsets[1] = b.recentOffsets[0]
-				b.recentOffsets[0] = offset
-				offset += 3
-			}
-		}
-	} else {
-		offset += 3
-	}
-	return offset
-}
-
-// encodeRaw can be used to set the output to a raw representation of supplied bytes.
-func (b *blockEnc) encodeRaw(a []byte) {
-	var bh blockHeader
-	bh.setLast(b.last)
-	bh.setSize(uint32(len(a)))
-	bh.setType(blockTypeRaw)
-	b.output = bh.appendTo(b.output[:0])
-	b.output = append(b.output, a...)
-	if debugEncoder {
-		println("Adding RAW block, length", len(a), "last:", b.last)
-	}
-}
-
-// encodeRaw can be used to set the output to a raw representation of supplied bytes.
-func (b *blockEnc) encodeRawTo(dst, src []byte) []byte {
-	var bh blockHeader
-	bh.setLast(b.last)
-	bh.setSize(uint32(len(src)))
-	bh.setType(blockTypeRaw)
-	dst = bh.appendTo(dst)
-	dst = append(dst, src...)
-	if debugEncoder {
-		println("Adding RAW block, length", len(src), "last:", b.last)
-	}
-	return dst
-}
-
-// encodeLits can be used if the block is only litLen.
-func (b *blockEnc) encodeLits(lits []byte, raw bool) error {
-	var bh blockHeader
-	bh.setLast(b.last)
-	bh.setSize(uint32(len(lits)))
-
-	// Don't compress extremely small blocks
-	if len(lits) < 8 || (len(lits) < 32 && b.dictLitEnc == nil) || raw {
-		if debugEncoder {
-			println("Adding RAW block, length", len(lits), "last:", b.last)
-		}
-		bh.setType(blockTypeRaw)
-		b.output = bh.appendTo(b.output)
-		b.output = append(b.output, lits...)
-		return nil
-	}
-
-	var (
-		out            []byte
-		reUsed, single bool
-		err            error
-	)
-	if b.dictLitEnc != nil {
-		b.litEnc.TransferCTable(b.dictLitEnc)
-		b.litEnc.Reuse = huff0.ReusePolicyAllow
-		b.dictLitEnc = nil
-	}
-	if len(lits) >= 1024 {
-		// Use 4 Streams.
-		out, reUsed, err = huff0.Compress4X(lits, b.litEnc)
-	} else if len(lits) > 32 {
-		// Use 1 stream
-		single = true
-		out, reUsed, err = huff0.Compress1X(lits, b.litEnc)
-	} else {
-		err = huff0.ErrIncompressible
-	}
-
-	switch err {
-	case huff0.ErrIncompressible:
-		if debugEncoder {
-			println("Adding RAW block, length", len(lits), "last:", b.last)
-		}
-		bh.setType(blockTypeRaw)
-		b.output = bh.appendTo(b.output)
-		b.output = append(b.output, lits...)
-		return nil
-	case huff0.ErrUseRLE:
-		if debugEncoder {
-			println("Adding RLE block, length", len(lits))
-		}
-		bh.setType(blockTypeRLE)
-		b.output = bh.appendTo(b.output)
-		b.output = append(b.output, lits[0])
-		return nil
-	case nil:
-	default:
-		return err
-	}
-	// Compressed...
-	// Now, allow reuse
-	b.litEnc.Reuse = huff0.ReusePolicyAllow
-	bh.setType(blockTypeCompressed)
-	var lh literalsHeader
-	if reUsed {
-		if debugEncoder {
-			println("Reused tree, compressed to", len(out))
-		}
-		lh.setType(literalsBlockTreeless)
-	} else {
-		if debugEncoder {
-			println("New tree, compressed to", len(out), "tree size:", len(b.litEnc.OutTable))
-		}
-		lh.setType(literalsBlockCompressed)
-	}
-	// Set sizes
-	lh.setSizes(len(out), len(lits), single)
-	bh.setSize(uint32(len(out) + lh.size() + 1))
-
-	// Write block headers.
-	b.output = bh.appendTo(b.output)
-	b.output = lh.appendTo(b.output)
-	// Add compressed data.
-	b.output = append(b.output, out...)
-	// No sequences.
-	b.output = append(b.output, 0)
-	return nil
-}
-
-// fuzzFseEncoder can be used to fuzz the FSE encoder.
-func fuzzFseEncoder(data []byte) int {
-	if len(data) > maxSequences || len(data) < 2 {
-		return 0
-	}
-	enc := fseEncoder{}
-	hist := enc.Histogram()[:256]
-	maxSym := uint8(0)
-	for i, v := range data {
-		v = v & 63
-		data[i] = v
-		hist[v]++
-		if v > maxSym {
-			maxSym = v
-		}
-	}
-	if maxSym == 0 {
-		// All 0
-		return 0
-	}
-	maxCount := func(a []uint32) int {
-		var max uint32
-		for _, v := range a {
-			if v > max {
-				max = v
-			}
-		}
-		return int(max)
-	}
-	cnt := maxCount(hist[:maxSym])
-	if cnt == len(data) {
-		// RLE
-		return 0
-	}
-	enc.HistogramFinished(maxSym, cnt)
-	err := enc.normalizeCount(len(data))
-	if err != nil {
-		return 0
-	}
-	_, err = enc.writeCount(nil)
-	if err != nil {
-		panic(err)
-	}
-	return 1
-}
-
-// encode will encode the block and append the output in b.output.
-// Previous offset codes must be pushed if more blocks are expected.
-func (b *blockEnc) encode(org []byte, raw, rawAllLits bool) error {
-	if len(b.sequences) == 0 {
-		return b.encodeLits(b.literals, rawAllLits)
-	}
-	// We want some difference to at least account for the headers.
-	saved := b.size - len(b.literals) - (b.size >> 5)
-	if saved < 16 {
-		if org == nil {
-			return errIncompressible
-		}
-		b.popOffsets()
-		return b.encodeLits(org, rawAllLits)
-	}
-
-	var bh blockHeader
-	var lh literalsHeader
-	bh.setLast(b.last)
-	bh.setType(blockTypeCompressed)
-	// Store offset of the block header. Needed when we know the size.
-	bhOffset := len(b.output)
-	b.output = bh.appendTo(b.output)
-
-	var (
-		out            []byte
-		reUsed, single bool
-		err            error
-	)
-	if b.dictLitEnc != nil {
-		b.litEnc.TransferCTable(b.dictLitEnc)
-		b.litEnc.Reuse = huff0.ReusePolicyAllow
-		b.dictLitEnc = nil
-	}
-	if len(b.literals) >= 1024 && !raw {
-		// Use 4 Streams.
-		out, reUsed, err = huff0.Compress4X(b.literals, b.litEnc)
-	} else if len(b.literals) > 32 && !raw {
-		// Use 1 stream
-		single = true
-		out, reUsed, err = huff0.Compress1X(b.literals, b.litEnc)
-	} else {
-		err = huff0.ErrIncompressible
-	}
-
-	switch err {
-	case huff0.ErrIncompressible:
-		lh.setType(literalsBlockRaw)
-		lh.setSize(len(b.literals))
-		b.output = lh.appendTo(b.output)
-		b.output = append(b.output, b.literals...)
-		if debugEncoder {
-			println("Adding literals RAW, length", len(b.literals))
-		}
-	case huff0.ErrUseRLE:
-		lh.setType(literalsBlockRLE)
-		lh.setSize(len(b.literals))
-		b.output = lh.appendTo(b.output)
-		b.output = append(b.output, b.literals[0])
-		if debugEncoder {
-			println("Adding literals RLE")
-		}
-	case nil:
-		// Compressed litLen...
-		if reUsed {
-			if debugEncoder {
-				println("reused tree")
-			}
-			lh.setType(literalsBlockTreeless)
-		} else {
-			if debugEncoder {
-				println("new tree, size:", len(b.litEnc.OutTable))
-			}
-			lh.setType(literalsBlockCompressed)
-			if debugEncoder {
-				_, _, err := huff0.ReadTable(out, nil)
-				if err != nil {
-					panic(err)
-				}
-			}
-		}
-		lh.setSizes(len(out), len(b.literals), single)
-		if debugEncoder {
-			printf("Compressed %d literals to %d bytes", len(b.literals), len(out))
-			println("Adding literal header:", lh)
-		}
-		b.output = lh.appendTo(b.output)
-		b.output = append(b.output, out...)
-		b.litEnc.Reuse = huff0.ReusePolicyAllow
-		if debugEncoder {
-			println("Adding literals compressed")
-		}
-	default:
-		if debugEncoder {
-			println("Adding literals ERROR:", err)
-		}
-		return err
-	}
-	// Sequence compression
-
-	// Write the number of sequences
-	switch {
-	case len(b.sequences) < 128:
-		b.output = append(b.output, uint8(len(b.sequences)))
-	case len(b.sequences) < 0x7f00: // TODO: this could be wrong
-		n := len(b.sequences)
-		b.output = append(b.output, 128+uint8(n>>8), uint8(n))
-	default:
-		n := len(b.sequences) - 0x7f00
-		b.output = append(b.output, 255, uint8(n), uint8(n>>8))
-	}
-	if debugEncoder {
-		println("Encoding", len(b.sequences), "sequences")
-	}
-	b.genCodes()
-	llEnc := b.coders.llEnc
-	ofEnc := b.coders.ofEnc
-	mlEnc := b.coders.mlEnc
-	err = llEnc.normalizeCount(len(b.sequences))
-	if err != nil {
-		return err
-	}
-	err = ofEnc.normalizeCount(len(b.sequences))
-	if err != nil {
-		return err
-	}
-	err = mlEnc.normalizeCount(len(b.sequences))
-	if err != nil {
-		return err
-	}
-
-	// Choose the best compression mode for each type.
-	// Will evaluate the new vs predefined and previous.
-	chooseComp := func(cur, prev, preDef *fseEncoder) (*fseEncoder, seqCompMode) {
-		// See if predefined/previous is better
-		hist := cur.count[:cur.symbolLen]
-		nSize := cur.approxSize(hist) + cur.maxHeaderSize()
-		predefSize := preDef.approxSize(hist)
-		prevSize := prev.approxSize(hist)
-
-		// Add a small penalty for new encoders.
-		// Don't bother with extremely small (<2 byte gains).
-		nSize = nSize + (nSize+2*8*16)>>4
-		switch {
-		case predefSize <= prevSize && predefSize <= nSize || forcePreDef:
-			if debugEncoder {
-				println("Using predefined", predefSize>>3, "<=", nSize>>3)
-			}
-			return preDef, compModePredefined
-		case prevSize <= nSize:
-			if debugEncoder {
-				println("Using previous", prevSize>>3, "<=", nSize>>3)
-			}
-			return prev, compModeRepeat
-		default:
-			if debugEncoder {
-				println("Using new, predef", predefSize>>3, ". previous:", prevSize>>3, ">", nSize>>3, "header max:", cur.maxHeaderSize()>>3, "bytes")
-				println("tl:", cur.actualTableLog, "symbolLen:", cur.symbolLen, "norm:", cur.norm[:cur.symbolLen], "hist", cur.count[:cur.symbolLen])
-			}
-			return cur, compModeFSE
-		}
-	}
-
-	// Write compression mode
-	var mode uint8
-	if llEnc.useRLE {
-		mode |= uint8(compModeRLE) << 6
-		llEnc.setRLE(b.sequences[0].llCode)
-		if debugEncoder {
-			println("llEnc.useRLE")
-		}
-	} else {
-		var m seqCompMode
-		llEnc, m = chooseComp(llEnc, b.coders.llPrev, &fsePredefEnc[tableLiteralLengths])
-		mode |= uint8(m) << 6
-	}
-	if ofEnc.useRLE {
-		mode |= uint8(compModeRLE) << 4
-		ofEnc.setRLE(b.sequences[0].ofCode)
-		if debugEncoder {
-			println("ofEnc.useRLE")
-		}
-	} else {
-		var m seqCompMode
-		ofEnc, m = chooseComp(ofEnc, b.coders.ofPrev, &fsePredefEnc[tableOffsets])
-		mode |= uint8(m) << 4
-	}
-
-	if mlEnc.useRLE {
-		mode |= uint8(compModeRLE) << 2
-		mlEnc.setRLE(b.sequences[0].mlCode)
-		if debugEncoder {
-			println("mlEnc.useRLE, code: ", b.sequences[0].mlCode, "value", b.sequences[0].matchLen)
-		}
-	} else {
-		var m seqCompMode
-		mlEnc, m = chooseComp(mlEnc, b.coders.mlPrev, &fsePredefEnc[tableMatchLengths])
-		mode |= uint8(m) << 2
-	}
-	b.output = append(b.output, mode)
-	if debugEncoder {
-		printf("Compression modes: 0b%b", mode)
-	}
-	b.output, err = llEnc.writeCount(b.output)
-	if err != nil {
-		return err
-	}
-	start := len(b.output)
-	b.output, err = ofEnc.writeCount(b.output)
-	if err != nil {
-		return err
-	}
-	if false {
-		println("block:", b.output[start:], "tablelog", ofEnc.actualTableLog, "maxcount:", ofEnc.maxCount)
-		fmt.Printf("selected TableLog: %d, Symbol length: %d\n", ofEnc.actualTableLog, ofEnc.symbolLen)
-		for i, v := range ofEnc.norm[:ofEnc.symbolLen] {
-			fmt.Printf("%3d: %5d -> %4d \n", i, ofEnc.count[i], v)
-		}
-	}
-	b.output, err = mlEnc.writeCount(b.output)
-	if err != nil {
-		return err
-	}
-
-	// Maybe in block?
-	wr := &b.wr
-	wr.reset(b.output)
-
-	var ll, of, ml cState
-
-	// Current sequence
-	seq := len(b.sequences) - 1
-	s := b.sequences[seq]
-	llEnc.setBits(llBitsTable[:])
-	mlEnc.setBits(mlBitsTable[:])
-	ofEnc.setBits(nil)
-
-	llTT, ofTT, mlTT := llEnc.ct.symbolTT[:256], ofEnc.ct.symbolTT[:256], mlEnc.ct.symbolTT[:256]
-
-	// We have 3 bounds checks here (and in the loop).
-	// Since we are iterating backwards it is kinda hard to avoid.
-	llB, ofB, mlB := llTT[s.llCode], ofTT[s.ofCode], mlTT[s.mlCode]
-	ll.init(wr, &llEnc.ct, llB)
-	of.init(wr, &ofEnc.ct, ofB)
-	wr.flush32()
-	ml.init(wr, &mlEnc.ct, mlB)
-
-	// Each of these lookups also generates a bounds check.
-	wr.addBits32NC(s.litLen, llB.outBits)
-	wr.addBits32NC(s.matchLen, mlB.outBits)
-	wr.flush32()
-	wr.addBits32NC(s.offset, ofB.outBits)
-	if debugSequences {
-		println("Encoded seq", seq, s, "codes:", s.llCode, s.mlCode, s.ofCode, "states:", ll.state, ml.state, of.state, "bits:", llB, mlB, ofB)
-	}
-	seq--
-	if llEnc.maxBits+mlEnc.maxBits+ofEnc.maxBits <= 32 {
-		// No need to flush (common)
-		for seq >= 0 {
-			s = b.sequences[seq]
-			wr.flush32()
-			llB, ofB, mlB := llTT[s.llCode], ofTT[s.ofCode], mlTT[s.mlCode]
-			// tabelog max is 8 for all.
-			of.encode(ofB)
-			ml.encode(mlB)
-			ll.encode(llB)
-			wr.flush32()
-
-			// We checked that all can stay within 32 bits
-			wr.addBits32NC(s.litLen, llB.outBits)
-			wr.addBits32NC(s.matchLen, mlB.outBits)
-			wr.addBits32NC(s.offset, ofB.outBits)
-
-			if debugSequences {
-				println("Encoded seq", seq, s)
-			}
-
-			seq--
-		}
-	} else {
-		for seq >= 0 {
-			s = b.sequences[seq]
-			wr.flush32()
-			llB, ofB, mlB := llTT[s.llCode], ofTT[s.ofCode], mlTT[s.mlCode]
-			// tabelog max is below 8 for each.
-			of.encode(ofB)
-			ml.encode(mlB)
-			ll.encode(llB)
-			wr.flush32()
-
-			// ml+ll = max 32 bits total
-			wr.addBits32NC(s.litLen, llB.outBits)
-			wr.addBits32NC(s.matchLen, mlB.outBits)
-			wr.flush32()
-			wr.addBits32NC(s.offset, ofB.outBits)
-
-			if debugSequences {
-				println("Encoded seq", seq, s)
-			}
-
-			seq--
-		}
-	}
-	ml.flush(mlEnc.actualTableLog)
-	of.flush(ofEnc.actualTableLog)
-	ll.flush(llEnc.actualTableLog)
-	err = wr.close()
-	if err != nil {
-		return err
-	}
-	b.output = wr.out
-
-	if len(b.output)-3-bhOffset >= b.size {
-		// Maybe even add a bigger margin.
-		b.litEnc.Reuse = huff0.ReusePolicyNone
-		return errIncompressible
-	}
-
-	// Size is output minus block header.
-	bh.setSize(uint32(len(b.output)-bhOffset) - 3)
-	if debugEncoder {
-		println("Rewriting block header", bh)
-	}
-	_ = bh.appendTo(b.output[bhOffset:bhOffset])
-	b.coders.setPrev(llEnc, mlEnc, ofEnc)
-	return nil
-}
-
-var errIncompressible = errors.New("incompressible")
-
-func (b *blockEnc) genCodes() {
-	if len(b.sequences) == 0 {
-		// nothing to do
-		return
-	}
-
-	if len(b.sequences) > math.MaxUint16 {
-		panic("can only encode up to 64K sequences")
-	}
-	// No bounds checks after here:
-	llH := b.coders.llEnc.Histogram()[:256]
-	ofH := b.coders.ofEnc.Histogram()[:256]
-	mlH := b.coders.mlEnc.Histogram()[:256]
-	for i := range llH {
-		llH[i] = 0
-	}
-	for i := range ofH {
-		ofH[i] = 0
-	}
-	for i := range mlH {
-		mlH[i] = 0
-	}
-
-	var llMax, ofMax, mlMax uint8
-	for i, seq := range b.sequences {
-		v := llCode(seq.litLen)
-		seq.llCode = v
-		llH[v]++
-		if v > llMax {
-			llMax = v
-		}
-
-		v = ofCode(seq.offset)
-		seq.ofCode = v
-		ofH[v]++
-		if v > ofMax {
-			ofMax = v
-		}
-
-		v = mlCode(seq.matchLen)
-		seq.mlCode = v
-		mlH[v]++
-		if v > mlMax {
-			mlMax = v
-			if debugAsserts && mlMax > maxMatchLengthSymbol {
-				panic(fmt.Errorf("mlMax > maxMatchLengthSymbol (%d), matchlen: %d", mlMax, seq.matchLen))
-			}
-		}
-		b.sequences[i] = seq
-	}
-	maxCount := func(a []uint32) int {
-		var max uint32
-		for _, v := range a {
-			if v > max {
-				max = v
-			}
-		}
-		return int(max)
-	}
-	if debugAsserts && mlMax > maxMatchLengthSymbol {
-		panic(fmt.Errorf("mlMax > maxMatchLengthSymbol (%d)", mlMax))
-	}
-	if debugAsserts && ofMax > maxOffsetBits {
-		panic(fmt.Errorf("ofMax > maxOffsetBits (%d)", ofMax))
-	}
-	if debugAsserts && llMax > maxLiteralLengthSymbol {
-		panic(fmt.Errorf("llMax > maxLiteralLengthSymbol (%d)", llMax))
-	}
-
-	b.coders.mlEnc.HistogramFinished(mlMax, maxCount(mlH[:mlMax+1]))
-	b.coders.ofEnc.HistogramFinished(ofMax, maxCount(ofH[:ofMax+1]))
-	b.coders.llEnc.HistogramFinished(llMax, maxCount(llH[:llMax+1]))
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/blocktype_string.go b/vendor/github.com/klauspost/compress/zstd/blocktype_string.go
deleted file mode 100644
index 01a01e486e..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/blocktype_string.go
+++ /dev/null
@@ -1,85 +0,0 @@
-// Code generated by "stringer -type=blockType,literalsBlockType,seqCompMode,tableIndex"; DO NOT EDIT.
-
-package zstd
-
-import "strconv"
-
-func _() {
-	// An "invalid array index" compiler error signifies that the constant values have changed.
-	// Re-run the stringer command to generate them again.
-	var x [1]struct{}
-	_ = x[blockTypeRaw-0]
-	_ = x[blockTypeRLE-1]
-	_ = x[blockTypeCompressed-2]
-	_ = x[blockTypeReserved-3]
-}
-
-const _blockType_name = "blockTypeRawblockTypeRLEblockTypeCompressedblockTypeReserved"
-
-var _blockType_index = [...]uint8{0, 12, 24, 43, 60}
-
-func (i blockType) String() string {
-	if i >= blockType(len(_blockType_index)-1) {
-		return "blockType(" + strconv.FormatInt(int64(i), 10) + ")"
-	}
-	return _blockType_name[_blockType_index[i]:_blockType_index[i+1]]
-}
-func _() {
-	// An "invalid array index" compiler error signifies that the constant values have changed.
-	// Re-run the stringer command to generate them again.
-	var x [1]struct{}
-	_ = x[literalsBlockRaw-0]
-	_ = x[literalsBlockRLE-1]
-	_ = x[literalsBlockCompressed-2]
-	_ = x[literalsBlockTreeless-3]
-}
-
-const _literalsBlockType_name = "literalsBlockRawliteralsBlockRLEliteralsBlockCompressedliteralsBlockTreeless"
-
-var _literalsBlockType_index = [...]uint8{0, 16, 32, 55, 76}
-
-func (i literalsBlockType) String() string {
-	if i >= literalsBlockType(len(_literalsBlockType_index)-1) {
-		return "literalsBlockType(" + strconv.FormatInt(int64(i), 10) + ")"
-	}
-	return _literalsBlockType_name[_literalsBlockType_index[i]:_literalsBlockType_index[i+1]]
-}
-func _() {
-	// An "invalid array index" compiler error signifies that the constant values have changed.
-	// Re-run the stringer command to generate them again.
-	var x [1]struct{}
-	_ = x[compModePredefined-0]
-	_ = x[compModeRLE-1]
-	_ = x[compModeFSE-2]
-	_ = x[compModeRepeat-3]
-}
-
-const _seqCompMode_name = "compModePredefinedcompModeRLEcompModeFSEcompModeRepeat"
-
-var _seqCompMode_index = [...]uint8{0, 18, 29, 40, 54}
-
-func (i seqCompMode) String() string {
-	if i >= seqCompMode(len(_seqCompMode_index)-1) {
-		return "seqCompMode(" + strconv.FormatInt(int64(i), 10) + ")"
-	}
-	return _seqCompMode_name[_seqCompMode_index[i]:_seqCompMode_index[i+1]]
-}
-func _() {
-	// An "invalid array index" compiler error signifies that the constant values have changed.
-	// Re-run the stringer command to generate them again.
-	var x [1]struct{}
-	_ = x[tableLiteralLengths-0]
-	_ = x[tableOffsets-1]
-	_ = x[tableMatchLengths-2]
-}
-
-const _tableIndex_name = "tableLiteralLengthstableOffsetstableMatchLengths"
-
-var _tableIndex_index = [...]uint8{0, 19, 31, 48}
-
-func (i tableIndex) String() string {
-	if i >= tableIndex(len(_tableIndex_index)-1) {
-		return "tableIndex(" + strconv.FormatInt(int64(i), 10) + ")"
-	}
-	return _tableIndex_name[_tableIndex_index[i]:_tableIndex_index[i+1]]
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/bytebuf.go b/vendor/github.com/klauspost/compress/zstd/bytebuf.go
deleted file mode 100644
index aab71c6cf8..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/bytebuf.go
+++ /dev/null
@@ -1,130 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"fmt"
-	"io"
-	"io/ioutil"
-)
-
-type byteBuffer interface {
-	// Read up to 8 bytes.
-	// Returns io.ErrUnexpectedEOF if this cannot be satisfied.
-	readSmall(n int) ([]byte, error)
-
-	// Read >8 bytes.
-	// MAY use the destination slice.
-	readBig(n int, dst []byte) ([]byte, error)
-
-	// Read a single byte.
-	readByte() (byte, error)
-
-	// Skip n bytes.
-	skipN(n int) error
-}
-
-// in-memory buffer
-type byteBuf []byte
-
-func (b *byteBuf) readSmall(n int) ([]byte, error) {
-	if debugAsserts && n > 8 {
-		panic(fmt.Errorf("small read > 8 (%d). use readBig", n))
-	}
-	bb := *b
-	if len(bb) < n {
-		return nil, io.ErrUnexpectedEOF
-	}
-	r := bb[:n]
-	*b = bb[n:]
-	return r, nil
-}
-
-func (b *byteBuf) readBig(n int, dst []byte) ([]byte, error) {
-	bb := *b
-	if len(bb) < n {
-		return nil, io.ErrUnexpectedEOF
-	}
-	r := bb[:n]
-	*b = bb[n:]
-	return r, nil
-}
-
-func (b *byteBuf) remain() []byte {
-	return *b
-}
-
-func (b *byteBuf) readByte() (byte, error) {
-	bb := *b
-	if len(bb) < 1 {
-		return 0, nil
-	}
-	r := bb[0]
-	*b = bb[1:]
-	return r, nil
-}
-
-func (b *byteBuf) skipN(n int) error {
-	bb := *b
-	if len(bb) < n {
-		return io.ErrUnexpectedEOF
-	}
-	*b = bb[n:]
-	return nil
-}
-
-// wrapper around a reader.
-type readerWrapper struct {
-	r   io.Reader
-	tmp [8]byte
-}
-
-func (r *readerWrapper) readSmall(n int) ([]byte, error) {
-	if debugAsserts && n > 8 {
-		panic(fmt.Errorf("small read > 8 (%d). use readBig", n))
-	}
-	n2, err := io.ReadFull(r.r, r.tmp[:n])
-	// We only really care about the actual bytes read.
-	if err != nil {
-		if err == io.EOF {
-			return nil, io.ErrUnexpectedEOF
-		}
-		if debugDecoder {
-			println("readSmall: got", n2, "want", n, "err", err)
-		}
-		return nil, err
-	}
-	return r.tmp[:n], nil
-}
-
-func (r *readerWrapper) readBig(n int, dst []byte) ([]byte, error) {
-	if cap(dst) < n {
-		dst = make([]byte, n)
-	}
-	n2, err := io.ReadFull(r.r, dst[:n])
-	if err == io.EOF && n > 0 {
-		err = io.ErrUnexpectedEOF
-	}
-	return dst[:n2], err
-}
-
-func (r *readerWrapper) readByte() (byte, error) {
-	n2, err := r.r.Read(r.tmp[:1])
-	if err != nil {
-		return 0, err
-	}
-	if n2 != 1 {
-		return 0, io.ErrUnexpectedEOF
-	}
-	return r.tmp[0], nil
-}
-
-func (r *readerWrapper) skipN(n int) error {
-	n2, err := io.CopyN(ioutil.Discard, r.r, int64(n))
-	if n2 != int64(n) {
-		err = io.ErrUnexpectedEOF
-	}
-	return err
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/bytereader.go b/vendor/github.com/klauspost/compress/zstd/bytereader.go
deleted file mode 100644
index 2c4fca17fa..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/bytereader.go
+++ /dev/null
@@ -1,88 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-// byteReader provides a byte reader that reads
-// little endian values from a byte stream.
-// The input stream is manually advanced.
-// The reader performs no bounds checks.
-type byteReader struct {
-	b   []byte
-	off int
-}
-
-// init will initialize the reader and set the input.
-func (b *byteReader) init(in []byte) {
-	b.b = in
-	b.off = 0
-}
-
-// advance the stream b n bytes.
-func (b *byteReader) advance(n uint) {
-	b.off += int(n)
-}
-
-// overread returns whether we have advanced too far.
-func (b *byteReader) overread() bool {
-	return b.off > len(b.b)
-}
-
-// Int32 returns a little endian int32 starting at current offset.
-func (b byteReader) Int32() int32 {
-	b2 := b.b[b.off:]
-	b2 = b2[:4]
-	v3 := int32(b2[3])
-	v2 := int32(b2[2])
-	v1 := int32(b2[1])
-	v0 := int32(b2[0])
-	return v0 | (v1 << 8) | (v2 << 16) | (v3 << 24)
-}
-
-// Uint8 returns the next byte
-func (b *byteReader) Uint8() uint8 {
-	v := b.b[b.off]
-	return v
-}
-
-// Uint32 returns a little endian uint32 starting at current offset.
-func (b byteReader) Uint32() uint32 {
-	if r := b.remain(); r < 4 {
-		// Very rare
-		v := uint32(0)
-		for i := 1; i <= r; i++ {
-			v = (v << 8) | uint32(b.b[len(b.b)-i])
-		}
-		return v
-	}
-	b2 := b.b[b.off:]
-	b2 = b2[:4]
-	v3 := uint32(b2[3])
-	v2 := uint32(b2[2])
-	v1 := uint32(b2[1])
-	v0 := uint32(b2[0])
-	return v0 | (v1 << 8) | (v2 << 16) | (v3 << 24)
-}
-
-// Uint32NC returns a little endian uint32 starting at current offset.
-// The caller must be sure if there are at least 4 bytes left.
-func (b byteReader) Uint32NC() uint32 {
-	b2 := b.b[b.off:]
-	b2 = b2[:4]
-	v3 := uint32(b2[3])
-	v2 := uint32(b2[2])
-	v1 := uint32(b2[1])
-	v0 := uint32(b2[0])
-	return v0 | (v1 << 8) | (v2 << 16) | (v3 << 24)
-}
-
-// unread returns the unread portion of the input.
-func (b byteReader) unread() []byte {
-	return b.b[b.off:]
-}
-
-// remain will return the number of bytes remaining.
-func (b byteReader) remain() int {
-	return len(b.b) - b.off
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/decodeheader.go b/vendor/github.com/klauspost/compress/zstd/decodeheader.go
deleted file mode 100644
index 69736e8d4b..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/decodeheader.go
+++ /dev/null
@@ -1,202 +0,0 @@
-// Copyright 2020+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-
-package zstd
-
-import (
-	"bytes"
-	"errors"
-	"io"
-)
-
-// HeaderMaxSize is the maximum size of a Frame and Block Header.
-// If less is sent to Header.Decode it *may* still contain enough information.
-const HeaderMaxSize = 14 + 3
-
-// Header contains information about the first frame and block within that.
-type Header struct {
-	// Window Size the window of data to keep while decoding.
-	// Will only be set if HasFCS is false.
-	WindowSize uint64
-
-	// Frame content size.
-	// Expected size of the entire frame.
-	FrameContentSize uint64
-
-	// Dictionary ID.
-	// If 0, no dictionary.
-	DictionaryID uint32
-
-	// First block information.
-	FirstBlock struct {
-		// OK will be set if first block could be decoded.
-		OK bool
-
-		// Is this the last block of a frame?
-		Last bool
-
-		// Is the data compressed?
-		// If true CompressedSize will be populated.
-		// Unfortunately DecompressedSize cannot be determined
-		// without decoding the blocks.
-		Compressed bool
-
-		// DecompressedSize is the expected decompressed size of the block.
-		// Will be 0 if it cannot be determined.
-		DecompressedSize int
-
-		// CompressedSize of the data in the block.
-		// Does not include the block header.
-		// Will be equal to DecompressedSize if not Compressed.
-		CompressedSize int
-	}
-
-	// Skippable will be true if the frame is meant to be skipped.
-	// No other information will be populated.
-	Skippable bool
-
-	// If set there is a checksum present for the block content.
-	HasCheckSum bool
-
-	// If this is true FrameContentSize will have a valid value
-	HasFCS bool
-
-	SingleSegment bool
-}
-
-// Decode the header from the beginning of the stream.
-// This will decode the frame header and the first block header if enough bytes are provided.
-// It is recommended to provide at least HeaderMaxSize bytes.
-// If the frame header cannot be read an error will be returned.
-// If there isn't enough input, io.ErrUnexpectedEOF is returned.
-// The FirstBlock.OK will indicate if enough information was available to decode the first block header.
-func (h *Header) Decode(in []byte) error {
-	if len(in) < 4 {
-		return io.ErrUnexpectedEOF
-	}
-	b, in := in[:4], in[4:]
-	if !bytes.Equal(b, frameMagic) {
-		if !bytes.Equal(b[1:4], skippableFrameMagic) || b[0]&0xf0 != 0x50 {
-			return ErrMagicMismatch
-		}
-		*h = Header{Skippable: true}
-		return nil
-	}
-	if len(in) < 1 {
-		return io.ErrUnexpectedEOF
-	}
-
-	// Clear output
-	*h = Header{}
-	fhd, in := in[0], in[1:]
-	h.SingleSegment = fhd&(1<<5) != 0
-	h.HasCheckSum = fhd&(1<<2) != 0
-
-	if fhd&(1<<3) != 0 {
-		return errors.New("reserved bit set on frame header")
-	}
-
-	// Read Window_Descriptor
-	// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#window_descriptor
-	if !h.SingleSegment {
-		if len(in) < 1 {
-			return io.ErrUnexpectedEOF
-		}
-		var wd byte
-		wd, in = in[0], in[1:]
-		windowLog := 10 + (wd >> 3)
-		windowBase := uint64(1) << windowLog
-		windowAdd := (windowBase / 8) * uint64(wd&0x7)
-		h.WindowSize = windowBase + windowAdd
-	}
-
-	// Read Dictionary_ID
-	// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#dictionary_id
-	if size := fhd & 3; size != 0 {
-		if size == 3 {
-			size = 4
-		}
-		if len(in) < int(size) {
-			return io.ErrUnexpectedEOF
-		}
-		b, in = in[:size], in[size:]
-		if b == nil {
-			return io.ErrUnexpectedEOF
-		}
-		switch size {
-		case 1:
-			h.DictionaryID = uint32(b[0])
-		case 2:
-			h.DictionaryID = uint32(b[0]) | (uint32(b[1]) << 8)
-		case 4:
-			h.DictionaryID = uint32(b[0]) | (uint32(b[1]) << 8) | (uint32(b[2]) << 16) | (uint32(b[3]) << 24)
-		}
-	}
-
-	// Read Frame_Content_Size
-	// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#frame_content_size
-	var fcsSize int
-	v := fhd >> 6
-	switch v {
-	case 0:
-		if h.SingleSegment {
-			fcsSize = 1
-		}
-	default:
-		fcsSize = 1 << v
-	}
-
-	if fcsSize > 0 {
-		h.HasFCS = true
-		if len(in) < fcsSize {
-			return io.ErrUnexpectedEOF
-		}
-		b, in = in[:fcsSize], in[fcsSize:]
-		if b == nil {
-			return io.ErrUnexpectedEOF
-		}
-		switch fcsSize {
-		case 1:
-			h.FrameContentSize = uint64(b[0])
-		case 2:
-			// When FCS_Field_Size is 2, the offset of 256 is added.
-			h.FrameContentSize = uint64(b[0]) | (uint64(b[1]) << 8) + 256
-		case 4:
-			h.FrameContentSize = uint64(b[0]) | (uint64(b[1]) << 8) | (uint64(b[2]) << 16) | (uint64(b[3]) << 24)
-		case 8:
-			d1 := uint32(b[0]) | (uint32(b[1]) << 8) | (uint32(b[2]) << 16) | (uint32(b[3]) << 24)
-			d2 := uint32(b[4]) | (uint32(b[5]) << 8) | (uint32(b[6]) << 16) | (uint32(b[7]) << 24)
-			h.FrameContentSize = uint64(d1) | (uint64(d2) << 32)
-		}
-	}
-
-	// Frame Header done, we will not fail from now on.
-	if len(in) < 3 {
-		return nil
-	}
-	tmp := in[:3]
-	bh := uint32(tmp[0]) | (uint32(tmp[1]) << 8) | (uint32(tmp[2]) << 16)
-	h.FirstBlock.Last = bh&1 != 0
-	blockType := blockType((bh >> 1) & 3)
-	// find size.
-	cSize := int(bh >> 3)
-	switch blockType {
-	case blockTypeReserved:
-		return nil
-	case blockTypeRLE:
-		h.FirstBlock.Compressed = true
-		h.FirstBlock.DecompressedSize = cSize
-		h.FirstBlock.CompressedSize = 1
-	case blockTypeCompressed:
-		h.FirstBlock.Compressed = true
-		h.FirstBlock.CompressedSize = cSize
-	case blockTypeRaw:
-		h.FirstBlock.DecompressedSize = cSize
-		h.FirstBlock.CompressedSize = cSize
-	default:
-		panic("Invalid block type")
-	}
-
-	h.FirstBlock.OK = true
-	return nil
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/decoder.go b/vendor/github.com/klauspost/compress/zstd/decoder.go
deleted file mode 100644
index 4d984c3b26..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/decoder.go
+++ /dev/null
@@ -1,554 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"errors"
-	"io"
-	"sync"
-)
-
-// Decoder provides decoding of zstandard streams.
-// The decoder has been designed to operate without allocations after a warmup.
-// This means that you should store the decoder for best performance.
-// To re-use a stream decoder, use the Reset(r io.Reader) error to switch to another stream.
-// A decoder can safely be re-used even if the previous stream failed.
-// To release the resources, you must call the Close() function on a decoder.
-type Decoder struct {
-	o decoderOptions
-
-	// Unreferenced decoders, ready for use.
-	decoders chan *blockDec
-
-	// Streams ready to be decoded.
-	stream chan decodeStream
-
-	// Current read position used for Reader functionality.
-	current decoderState
-
-	// Custom dictionaries.
-	// Always uses copies.
-	dicts map[uint32]dict
-
-	// streamWg is the waitgroup for all streams
-	streamWg sync.WaitGroup
-}
-
-// decoderState is used for maintaining state when the decoder
-// is used for streaming.
-type decoderState struct {
-	// current block being written to stream.
-	decodeOutput
-
-	// output in order to be written to stream.
-	output chan decodeOutput
-
-	// cancel remaining output.
-	cancel chan struct{}
-
-	flushed bool
-}
-
-var (
-	// Check the interfaces we want to support.
-	_ = io.WriterTo(&Decoder{})
-	_ = io.Reader(&Decoder{})
-)
-
-// NewReader creates a new decoder.
-// A nil Reader can be provided in which case Reset can be used to start a decode.
-//
-// A Decoder can be used in two modes:
-//
-// 1) As a stream, or
-// 2) For stateless decoding using DecodeAll.
-//
-// Only a single stream can be decoded concurrently, but the same decoder
-// can run multiple concurrent stateless decodes. It is even possible to
-// use stateless decodes while a stream is being decoded.
-//
-// The Reset function can be used to initiate a new stream, which is will considerably
-// reduce the allocations normally caused by NewReader.
-func NewReader(r io.Reader, opts ...DOption) (*Decoder, error) {
-	initPredefined()
-	var d Decoder
-	d.o.setDefault()
-	for _, o := range opts {
-		err := o(&d.o)
-		if err != nil {
-			return nil, err
-		}
-	}
-	d.current.output = make(chan decodeOutput, d.o.concurrent)
-	d.current.flushed = true
-
-	if r == nil {
-		d.current.err = ErrDecoderNilInput
-	}
-
-	// Transfer option dicts.
-	d.dicts = make(map[uint32]dict, len(d.o.dicts))
-	for _, dc := range d.o.dicts {
-		d.dicts[dc.id] = dc
-	}
-	d.o.dicts = nil
-
-	// Create decoders
-	d.decoders = make(chan *blockDec, d.o.concurrent)
-	for i := 0; i < d.o.concurrent; i++ {
-		dec := newBlockDec(d.o.lowMem)
-		dec.localFrame = newFrameDec(d.o)
-		d.decoders <- dec
-	}
-
-	if r == nil {
-		return &d, nil
-	}
-	return &d, d.Reset(r)
-}
-
-// Read bytes from the decompressed stream into p.
-// Returns the number of bytes written and any error that occurred.
-// When the stream is done, io.EOF will be returned.
-func (d *Decoder) Read(p []byte) (int, error) {
-	var n int
-	for {
-		if len(d.current.b) > 0 {
-			filled := copy(p, d.current.b)
-			p = p[filled:]
-			d.current.b = d.current.b[filled:]
-			n += filled
-		}
-		if len(p) == 0 {
-			break
-		}
-		if len(d.current.b) == 0 {
-			// We have an error and no more data
-			if d.current.err != nil {
-				break
-			}
-			if !d.nextBlock(n == 0) {
-				return n, nil
-			}
-		}
-	}
-	if len(d.current.b) > 0 {
-		if debugDecoder {
-			println("returning", n, "still bytes left:", len(d.current.b))
-		}
-		// Only return error at end of block
-		return n, nil
-	}
-	if d.current.err != nil {
-		d.drainOutput()
-	}
-	if debugDecoder {
-		println("returning", n, d.current.err, len(d.decoders))
-	}
-	return n, d.current.err
-}
-
-// Reset will reset the decoder the supplied stream after the current has finished processing.
-// Note that this functionality cannot be used after Close has been called.
-// Reset can be called with a nil reader to release references to the previous reader.
-// After being called with a nil reader, no other operations than Reset or DecodeAll or Close
-// should be used.
-func (d *Decoder) Reset(r io.Reader) error {
-	if d.current.err == ErrDecoderClosed {
-		return d.current.err
-	}
-
-	d.drainOutput()
-
-	if r == nil {
-		d.current.err = ErrDecoderNilInput
-		if len(d.current.b) > 0 {
-			d.current.b = d.current.b[:0]
-		}
-		d.current.flushed = true
-		return nil
-	}
-
-	// If bytes buffer and < 5MB, do sync decoding anyway.
-	if bb, ok := r.(byter); ok && bb.Len() < 5<<20 {
-		bb2 := bb
-		if debugDecoder {
-			println("*bytes.Buffer detected, doing sync decode, len:", bb.Len())
-		}
-		b := bb2.Bytes()
-		var dst []byte
-		if cap(d.current.b) > 0 {
-			dst = d.current.b
-		}
-
-		dst, err := d.DecodeAll(b, dst[:0])
-		if err == nil {
-			err = io.EOF
-		}
-		d.current.b = dst
-		d.current.err = err
-		d.current.flushed = true
-		if debugDecoder {
-			println("sync decode to", len(dst), "bytes, err:", err)
-		}
-		return nil
-	}
-
-	if d.stream == nil {
-		d.stream = make(chan decodeStream, 1)
-		d.streamWg.Add(1)
-		go d.startStreamDecoder(d.stream)
-	}
-
-	// Remove current block.
-	d.current.decodeOutput = decodeOutput{}
-	d.current.err = nil
-	d.current.cancel = make(chan struct{})
-	d.current.flushed = false
-	d.current.d = nil
-
-	d.stream <- decodeStream{
-		r:      r,
-		output: d.current.output,
-		cancel: d.current.cancel,
-	}
-	return nil
-}
-
-// drainOutput will drain the output until errEndOfStream is sent.
-func (d *Decoder) drainOutput() {
-	if d.current.cancel != nil {
-		println("cancelling current")
-		close(d.current.cancel)
-		d.current.cancel = nil
-	}
-	if d.current.d != nil {
-		if debugDecoder {
-			printf("re-adding current decoder %p, decoders: %d", d.current.d, len(d.decoders))
-		}
-		d.decoders <- d.current.d
-		d.current.d = nil
-		d.current.b = nil
-	}
-	if d.current.output == nil || d.current.flushed {
-		println("current already flushed")
-		return
-	}
-	for v := range d.current.output {
-		if v.d != nil {
-			if debugDecoder {
-				printf("re-adding decoder %p", v.d)
-			}
-			d.decoders <- v.d
-		}
-		if v.err == errEndOfStream {
-			println("current flushed")
-			d.current.flushed = true
-			return
-		}
-	}
-}
-
-// WriteTo writes data to w until there's no more data to write or when an error occurs.
-// The return value n is the number of bytes written.
-// Any error encountered during the write is also returned.
-func (d *Decoder) WriteTo(w io.Writer) (int64, error) {
-	var n int64
-	for {
-		if len(d.current.b) > 0 {
-			n2, err2 := w.Write(d.current.b)
-			n += int64(n2)
-			if err2 != nil && d.current.err == nil {
-				d.current.err = err2
-				break
-			}
-		}
-		if d.current.err != nil {
-			break
-		}
-		d.nextBlock(true)
-	}
-	err := d.current.err
-	if err != nil {
-		d.drainOutput()
-	}
-	if err == io.EOF {
-		err = nil
-	}
-	return n, err
-}
-
-// DecodeAll allows stateless decoding of a blob of bytes.
-// Output will be appended to dst, so if the destination size is known
-// you can pre-allocate the destination slice to avoid allocations.
-// DecodeAll can be used concurrently.
-// The Decoder concurrency limits will be respected.
-func (d *Decoder) DecodeAll(input, dst []byte) ([]byte, error) {
-	if d.current.err == ErrDecoderClosed {
-		return dst, ErrDecoderClosed
-	}
-
-	// Grab a block decoder and frame decoder.
-	block := <-d.decoders
-	frame := block.localFrame
-	defer func() {
-		if debugDecoder {
-			printf("re-adding decoder: %p", block)
-		}
-		frame.rawInput = nil
-		frame.bBuf = nil
-		d.decoders <- block
-	}()
-	frame.bBuf = input
-
-	for {
-		frame.history.reset()
-		err := frame.reset(&frame.bBuf)
-		if err == io.EOF {
-			if debugDecoder {
-				println("frame reset return EOF")
-			}
-			return dst, nil
-		}
-		if frame.DictionaryID != nil {
-			dict, ok := d.dicts[*frame.DictionaryID]
-			if !ok {
-				return nil, ErrUnknownDictionary
-			}
-			frame.history.setDict(&dict)
-		}
-		if err != nil {
-			return dst, err
-		}
-		if frame.FrameContentSize > d.o.maxDecodedSize-uint64(len(dst)) {
-			return dst, ErrDecoderSizeExceeded
-		}
-		if frame.FrameContentSize > 0 && frame.FrameContentSize < 1<<30 {
-			// Never preallocate moe than 1 GB up front.
-			if cap(dst)-len(dst) < int(frame.FrameContentSize) {
-				dst2 := make([]byte, len(dst), len(dst)+int(frame.FrameContentSize))
-				copy(dst2, dst)
-				dst = dst2
-			}
-		}
-		if cap(dst) == 0 {
-			// Allocate len(input) * 2 by default if nothing is provided
-			// and we didn't get frame content size.
-			size := len(input) * 2
-			// Cap to 1 MB.
-			if size > 1<<20 {
-				size = 1 << 20
-			}
-			if uint64(size) > d.o.maxDecodedSize {
-				size = int(d.o.maxDecodedSize)
-			}
-			dst = make([]byte, 0, size)
-		}
-
-		dst, err = frame.runDecoder(dst, block)
-		if err != nil {
-			return dst, err
-		}
-		if len(frame.bBuf) == 0 {
-			if debugDecoder {
-				println("frame dbuf empty")
-			}
-			break
-		}
-	}
-	return dst, nil
-}
-
-// nextBlock returns the next block.
-// If an error occurs d.err will be set.
-// Optionally the function can block for new output.
-// If non-blocking mode is used the returned boolean will be false
-// if no data was available without blocking.
-func (d *Decoder) nextBlock(blocking bool) (ok bool) {
-	if d.current.d != nil {
-		if debugDecoder {
-			printf("re-adding current decoder %p", d.current.d)
-		}
-		d.decoders <- d.current.d
-		d.current.d = nil
-	}
-	if d.current.err != nil {
-		// Keep error state.
-		return blocking
-	}
-
-	if blocking {
-		d.current.decodeOutput = <-d.current.output
-	} else {
-		select {
-		case d.current.decodeOutput = <-d.current.output:
-		default:
-			return false
-		}
-	}
-	if debugDecoder {
-		println("got", len(d.current.b), "bytes, error:", d.current.err)
-	}
-	return true
-}
-
-// Close will release all resources.
-// It is NOT possible to reuse the decoder after this.
-func (d *Decoder) Close() {
-	if d.current.err == ErrDecoderClosed {
-		return
-	}
-	d.drainOutput()
-	if d.stream != nil {
-		close(d.stream)
-		d.streamWg.Wait()
-		d.stream = nil
-	}
-	if d.decoders != nil {
-		close(d.decoders)
-		for dec := range d.decoders {
-			dec.Close()
-		}
-		d.decoders = nil
-	}
-	if d.current.d != nil {
-		d.current.d.Close()
-		d.current.d = nil
-	}
-	d.current.err = ErrDecoderClosed
-}
-
-// IOReadCloser returns the decoder as an io.ReadCloser for convenience.
-// Any changes to the decoder will be reflected, so the returned ReadCloser
-// can be reused along with the decoder.
-// io.WriterTo is also supported by the returned ReadCloser.
-func (d *Decoder) IOReadCloser() io.ReadCloser {
-	return closeWrapper{d: d}
-}
-
-// closeWrapper wraps a function call as a closer.
-type closeWrapper struct {
-	d *Decoder
-}
-
-// WriteTo forwards WriteTo calls to the decoder.
-func (c closeWrapper) WriteTo(w io.Writer) (n int64, err error) {
-	return c.d.WriteTo(w)
-}
-
-// Read forwards read calls to the decoder.
-func (c closeWrapper) Read(p []byte) (n int, err error) {
-	return c.d.Read(p)
-}
-
-// Close closes the decoder.
-func (c closeWrapper) Close() error {
-	c.d.Close()
-	return nil
-}
-
-type decodeOutput struct {
-	d   *blockDec
-	b   []byte
-	err error
-}
-
-type decodeStream struct {
-	r io.Reader
-
-	// Blocks ready to be written to output.
-	output chan decodeOutput
-
-	// cancel reading from the input
-	cancel chan struct{}
-}
-
-// errEndOfStream indicates that everything from the stream was read.
-var errEndOfStream = errors.New("end-of-stream")
-
-// Create Decoder:
-// Spawn n block decoders. These accept tasks to decode a block.
-// Create goroutine that handles stream processing, this will send history to decoders as they are available.
-// Decoders update the history as they decode.
-// When a block is returned:
-// 		a) history is sent to the next decoder,
-// 		b) content written to CRC.
-// 		c) return data to WRITER.
-// 		d) wait for next block to return data.
-// Once WRITTEN, the decoders reused by the writer frame decoder for re-use.
-func (d *Decoder) startStreamDecoder(inStream chan decodeStream) {
-	defer d.streamWg.Done()
-	frame := newFrameDec(d.o)
-	for stream := range inStream {
-		if debugDecoder {
-			println("got new stream")
-		}
-		br := readerWrapper{r: stream.r}
-	decodeStream:
-		for {
-			frame.history.reset()
-			err := frame.reset(&br)
-			if debugDecoder && err != nil {
-				println("Frame decoder returned", err)
-			}
-			if err == nil && frame.DictionaryID != nil {
-				dict, ok := d.dicts[*frame.DictionaryID]
-				if !ok {
-					err = ErrUnknownDictionary
-				} else {
-					frame.history.setDict(&dict)
-				}
-			}
-			if err != nil {
-				stream.output <- decodeOutput{
-					err: err,
-				}
-				break
-			}
-			if debugDecoder {
-				println("starting frame decoder")
-			}
-
-			// This goroutine will forward history between frames.
-			frame.frameDone.Add(1)
-			frame.initAsync()
-
-			go frame.startDecoder(stream.output)
-		decodeFrame:
-			// Go through all blocks of the frame.
-			for {
-				dec := <-d.decoders
-				select {
-				case <-stream.cancel:
-					if !frame.sendErr(dec, io.EOF) {
-						// To not let the decoder dangle, send it back.
-						stream.output <- decodeOutput{d: dec}
-					}
-					break decodeStream
-				default:
-				}
-				err := frame.next(dec)
-				switch err {
-				case io.EOF:
-					// End of current frame, no error
-					println("EOF on next block")
-					break decodeFrame
-				case nil:
-					continue
-				default:
-					println("block decoder returned", err)
-					break decodeStream
-				}
-			}
-			// All blocks have started decoding, check if there are more frames.
-			println("waiting for done")
-			frame.frameDone.Wait()
-			println("done waiting...")
-		}
-		frame.frameDone.Wait()
-		println("Sending EOS")
-		stream.output <- decodeOutput{err: errEndOfStream}
-	}
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/decoder_options.go b/vendor/github.com/klauspost/compress/zstd/decoder_options.go
deleted file mode 100644
index 95cc9b8b81..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/decoder_options.go
+++ /dev/null
@@ -1,102 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"errors"
-	"runtime"
-)
-
-// DOption is an option for creating a decoder.
-type DOption func(*decoderOptions) error
-
-// options retains accumulated state of multiple options.
-type decoderOptions struct {
-	lowMem         bool
-	concurrent     int
-	maxDecodedSize uint64
-	maxWindowSize  uint64
-	dicts          []dict
-}
-
-func (o *decoderOptions) setDefault() {
-	*o = decoderOptions{
-		// use less ram: true for now, but may change.
-		lowMem:        true,
-		concurrent:    runtime.GOMAXPROCS(0),
-		maxWindowSize: MaxWindowSize,
-	}
-	o.maxDecodedSize = 1 << 63
-}
-
-// WithDecoderLowmem will set whether to use a lower amount of memory,
-// but possibly have to allocate more while running.
-func WithDecoderLowmem(b bool) DOption {
-	return func(o *decoderOptions) error { o.lowMem = b; return nil }
-}
-
-// WithDecoderConcurrency will set the concurrency,
-// meaning the maximum number of decoders to run concurrently.
-// The value supplied must be at least 1.
-// By default this will be set to GOMAXPROCS.
-func WithDecoderConcurrency(n int) DOption {
-	return func(o *decoderOptions) error {
-		if n <= 0 {
-			return errors.New("concurrency must be at least 1")
-		}
-		o.concurrent = n
-		return nil
-	}
-}
-
-// WithDecoderMaxMemory allows to set a maximum decoded size for in-memory
-// non-streaming operations or maximum window size for streaming operations.
-// This can be used to control memory usage of potentially hostile content.
-// Maximum and default is 1 << 63 bytes.
-func WithDecoderMaxMemory(n uint64) DOption {
-	return func(o *decoderOptions) error {
-		if n == 0 {
-			return errors.New("WithDecoderMaxMemory must be at least 1")
-		}
-		if n > 1<<63 {
-			return errors.New("WithDecoderMaxmemory must be less than 1 << 63")
-		}
-		o.maxDecodedSize = n
-		return nil
-	}
-}
-
-// WithDecoderDicts allows to register one or more dictionaries for the decoder.
-// If several dictionaries with the same ID is provided the last one will be used.
-func WithDecoderDicts(dicts ...[]byte) DOption {
-	return func(o *decoderOptions) error {
-		for _, b := range dicts {
-			d, err := loadDict(b)
-			if err != nil {
-				return err
-			}
-			o.dicts = append(o.dicts, *d)
-		}
-		return nil
-	}
-}
-
-// WithDecoderMaxWindow allows to set a maximum window size for decodes.
-// This allows rejecting packets that will cause big memory usage.
-// The Decoder will likely allocate more memory based on the WithDecoderLowmem setting.
-// If WithDecoderMaxMemory is set to a lower value, that will be used.
-// Default is 512MB, Maximum is ~3.75 TB as per zstandard spec.
-func WithDecoderMaxWindow(size uint64) DOption {
-	return func(o *decoderOptions) error {
-		if size < MinWindowSize {
-			return errors.New("WithMaxWindowSize must be at least 1KB, 1024 bytes")
-		}
-		if size > (1<<41)+7*(1<<38) {
-			return errors.New("WithMaxWindowSize must be less than (1<<41) + 7*(1<<38) ~ 3.75TB")
-		}
-		o.maxWindowSize = size
-		return nil
-	}
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/dict.go b/vendor/github.com/klauspost/compress/zstd/dict.go
deleted file mode 100644
index a36ae83ef5..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/dict.go
+++ /dev/null
@@ -1,122 +0,0 @@
-package zstd
-
-import (
-	"bytes"
-	"encoding/binary"
-	"errors"
-	"fmt"
-	"io"
-
-	"github.com/klauspost/compress/huff0"
-)
-
-type dict struct {
-	id uint32
-
-	litEnc              *huff0.Scratch
-	llDec, ofDec, mlDec sequenceDec
-	//llEnc, ofEnc, mlEnc []*fseEncoder
-	offsets [3]int
-	content []byte
-}
-
-var dictMagic = [4]byte{0x37, 0xa4, 0x30, 0xec}
-
-// ID returns the dictionary id or 0 if d is nil.
-func (d *dict) ID() uint32 {
-	if d == nil {
-		return 0
-	}
-	return d.id
-}
-
-// DictContentSize returns the dictionary content size or 0 if d is nil.
-func (d *dict) DictContentSize() int {
-	if d == nil {
-		return 0
-	}
-	return len(d.content)
-}
-
-// Load a dictionary as described in
-// https://github.com/facebook/zstd/blob/master/doc/zstd_compression_format.md#dictionary-format
-func loadDict(b []byte) (*dict, error) {
-	// Check static field size.
-	if len(b) <= 8+(3*4) {
-		return nil, io.ErrUnexpectedEOF
-	}
-	d := dict{
-		llDec: sequenceDec{fse: &fseDecoder{}},
-		ofDec: sequenceDec{fse: &fseDecoder{}},
-		mlDec: sequenceDec{fse: &fseDecoder{}},
-	}
-	if !bytes.Equal(b[:4], dictMagic[:]) {
-		return nil, ErrMagicMismatch
-	}
-	d.id = binary.LittleEndian.Uint32(b[4:8])
-	if d.id == 0 {
-		return nil, errors.New("dictionaries cannot have ID 0")
-	}
-
-	// Read literal table
-	var err error
-	d.litEnc, b, err = huff0.ReadTable(b[8:], nil)
-	if err != nil {
-		return nil, err
-	}
-	d.litEnc.Reuse = huff0.ReusePolicyMust
-
-	br := byteReader{
-		b:   b,
-		off: 0,
-	}
-	readDec := func(i tableIndex, dec *fseDecoder) error {
-		if err := dec.readNCount(&br, uint16(maxTableSymbol[i])); err != nil {
-			return err
-		}
-		if br.overread() {
-			return io.ErrUnexpectedEOF
-		}
-		err = dec.transform(symbolTableX[i])
-		if err != nil {
-			println("Transform table error:", err)
-			return err
-		}
-		if debugDecoder || debugEncoder {
-			println("Read table ok", "symbolLen:", dec.symbolLen)
-		}
-		// Set decoders as predefined so they aren't reused.
-		dec.preDefined = true
-		return nil
-	}
-
-	if err := readDec(tableOffsets, d.ofDec.fse); err != nil {
-		return nil, err
-	}
-	if err := readDec(tableMatchLengths, d.mlDec.fse); err != nil {
-		return nil, err
-	}
-	if err := readDec(tableLiteralLengths, d.llDec.fse); err != nil {
-		return nil, err
-	}
-	if br.remain() < 12 {
-		return nil, io.ErrUnexpectedEOF
-	}
-
-	d.offsets[0] = int(br.Uint32())
-	br.advance(4)
-	d.offsets[1] = int(br.Uint32())
-	br.advance(4)
-	d.offsets[2] = int(br.Uint32())
-	br.advance(4)
-	if d.offsets[0] <= 0 || d.offsets[1] <= 0 || d.offsets[2] <= 0 {
-		return nil, errors.New("invalid offset in dictionary")
-	}
-	d.content = make([]byte, br.remain())
-	copy(d.content, br.unread())
-	if d.offsets[0] > len(d.content) || d.offsets[1] > len(d.content) || d.offsets[2] > len(d.content) {
-		return nil, fmt.Errorf("initial offset bigger than dictionary content size %d, offsets: %v", len(d.content), d.offsets)
-	}
-
-	return &d, nil
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/enc_base.go b/vendor/github.com/klauspost/compress/zstd/enc_base.go
deleted file mode 100644
index 60f2986486..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/enc_base.go
+++ /dev/null
@@ -1,178 +0,0 @@
-package zstd
-
-import (
-	"fmt"
-	"math/bits"
-
-	"github.com/klauspost/compress/zstd/internal/xxhash"
-)
-
-const (
-	dictShardBits = 6
-)
-
-type fastBase struct {
-	// cur is the offset at the start of hist
-	cur int32
-	// maximum offset. Should be at least 2x block size.
-	maxMatchOff int32
-	hist        []byte
-	crc         *xxhash.Digest
-	tmp         [8]byte
-	blk         *blockEnc
-	lastDictID  uint32
-	lowMem      bool
-}
-
-// CRC returns the underlying CRC writer.
-func (e *fastBase) CRC() *xxhash.Digest {
-	return e.crc
-}
-
-// AppendCRC will append the CRC to the destination slice and return it.
-func (e *fastBase) AppendCRC(dst []byte) []byte {
-	crc := e.crc.Sum(e.tmp[:0])
-	dst = append(dst, crc[7], crc[6], crc[5], crc[4])
-	return dst
-}
-
-// WindowSize returns the window size of the encoder,
-// or a window size small enough to contain the input size, if > 0.
-func (e *fastBase) WindowSize(size int) int32 {
-	if size > 0 && size < int(e.maxMatchOff) {
-		b := int32(1) << uint(bits.Len(uint(size)))
-		// Keep minimum window.
-		if b < 1024 {
-			b = 1024
-		}
-		return b
-	}
-	return e.maxMatchOff
-}
-
-// Block returns the current block.
-func (e *fastBase) Block() *blockEnc {
-	return e.blk
-}
-
-func (e *fastBase) addBlock(src []byte) int32 {
-	if debugAsserts && e.cur > bufferReset {
-		panic(fmt.Sprintf("ecur (%d) > buffer reset (%d)", e.cur, bufferReset))
-	}
-	// check if we have space already
-	if len(e.hist)+len(src) > cap(e.hist) {
-		if cap(e.hist) == 0 {
-			e.ensureHist(len(src))
-		} else {
-			if cap(e.hist) < int(e.maxMatchOff+maxCompressedBlockSize) {
-				panic(fmt.Errorf("unexpected buffer cap %d, want at least %d with window %d", cap(e.hist), e.maxMatchOff+maxCompressedBlockSize, e.maxMatchOff))
-			}
-			// Move down
-			offset := int32(len(e.hist)) - e.maxMatchOff
-			copy(e.hist[0:e.maxMatchOff], e.hist[offset:])
-			e.cur += offset
-			e.hist = e.hist[:e.maxMatchOff]
-		}
-	}
-	s := int32(len(e.hist))
-	e.hist = append(e.hist, src...)
-	return s
-}
-
-// ensureHist will ensure that history can keep at least this many bytes.
-func (e *fastBase) ensureHist(n int) {
-	if cap(e.hist) >= n {
-		return
-	}
-	l := e.maxMatchOff
-	if (e.lowMem && e.maxMatchOff > maxCompressedBlockSize) || e.maxMatchOff <= maxCompressedBlockSize {
-		l += maxCompressedBlockSize
-	} else {
-		l += e.maxMatchOff
-	}
-	// Make it at least 1MB.
-	if l < 1<<20 && !e.lowMem {
-		l = 1 << 20
-	}
-	// Make it at least the requested size.
-	if l < int32(n) {
-		l = int32(n)
-	}
-	e.hist = make([]byte, 0, l)
-}
-
-// useBlock will replace the block with the provided one,
-// but transfer recent offsets from the previous.
-func (e *fastBase) UseBlock(enc *blockEnc) {
-	enc.reset(e.blk)
-	e.blk = enc
-}
-
-func (e *fastBase) matchlenNoHist(s, t int32, src []byte) int32 {
-	// Extend the match to be as long as possible.
-	return int32(matchLen(src[s:], src[t:]))
-}
-
-func (e *fastBase) matchlen(s, t int32, src []byte) int32 {
-	if debugAsserts {
-		if s < 0 {
-			err := fmt.Sprintf("s (%d) < 0", s)
-			panic(err)
-		}
-		if t < 0 {
-			err := fmt.Sprintf("s (%d) < 0", s)
-			panic(err)
-		}
-		if s-t > e.maxMatchOff {
-			err := fmt.Sprintf("s (%d) - t (%d) > maxMatchOff (%d)", s, t, e.maxMatchOff)
-			panic(err)
-		}
-		if len(src)-int(s) > maxCompressedBlockSize {
-			panic(fmt.Sprintf("len(src)-s (%d) > maxCompressedBlockSize (%d)", len(src)-int(s), maxCompressedBlockSize))
-		}
-	}
-
-	// Extend the match to be as long as possible.
-	return int32(matchLen(src[s:], src[t:]))
-}
-
-// Reset the encoding table.
-func (e *fastBase) resetBase(d *dict, singleBlock bool) {
-	if e.blk == nil {
-		e.blk = &blockEnc{lowMem: e.lowMem}
-		e.blk.init()
-	} else {
-		e.blk.reset(nil)
-	}
-	e.blk.initNewEncode()
-	if e.crc == nil {
-		e.crc = xxhash.New()
-	} else {
-		e.crc.Reset()
-	}
-	if d != nil {
-		low := e.lowMem
-		if singleBlock {
-			e.lowMem = true
-		}
-		e.ensureHist(d.DictContentSize() + maxCompressedBlockSize)
-		e.lowMem = low
-	}
-
-	// We offset current position so everything will be out of reach.
-	// If above reset line, history will be purged.
-	if e.cur < bufferReset {
-		e.cur += e.maxMatchOff + int32(len(e.hist))
-	}
-	e.hist = e.hist[:0]
-	if d != nil {
-		// Set offsets (currently not used)
-		for i, off := range d.offsets {
-			e.blk.recentOffsets[i] = uint32(off)
-			e.blk.prevRecentOffsets[i] = e.blk.recentOffsets[i]
-		}
-		// Transfer litenc.
-		e.blk.dictLitEnc = d.litEnc
-		e.hist = append(e.hist, d.content...)
-	}
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/enc_best.go b/vendor/github.com/klauspost/compress/zstd/enc_best.go
deleted file mode 100644
index b7d4b90047..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/enc_best.go
+++ /dev/null
@@ -1,501 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"fmt"
-	"math/bits"
-)
-
-const (
-	bestLongTableBits = 20                     // Bits used in the long match table
-	bestLongTableSize = 1 << bestLongTableBits // Size of the table
-
-	// Note: Increasing the short table bits or making the hash shorter
-	// can actually lead to compression degradation since it will 'steal' more from the
-	// long match table and match offsets are quite big.
-	// This greatly depends on the type of input.
-	bestShortTableBits = 16                      // Bits used in the short match table
-	bestShortTableSize = 1 << bestShortTableBits // Size of the table
-)
-
-// bestFastEncoder uses 2 tables, one for short matches (5 bytes) and one for long matches.
-// The long match table contains the previous entry with the same hash,
-// effectively making it a "chain" of length 2.
-// When we find a long match we choose between the two values and select the longest.
-// When we find a short match, after checking the long, we check if we can find a long at n+1
-// and that it is longer (lazy matching).
-type bestFastEncoder struct {
-	fastBase
-	table         [bestShortTableSize]prevEntry
-	longTable     [bestLongTableSize]prevEntry
-	dictTable     []prevEntry
-	dictLongTable []prevEntry
-}
-
-// Encode improves compression...
-func (e *bestFastEncoder) Encode(blk *blockEnc, src []byte) {
-	const (
-		// Input margin is the number of bytes we read (8)
-		// and the maximum we will read ahead (2)
-		inputMargin            = 8 + 4
-		minNonLiteralBlockSize = 16
-	)
-
-	// Protect against e.cur wraparound.
-	for e.cur >= bufferReset {
-		if len(e.hist) == 0 {
-			for i := range e.table[:] {
-				e.table[i] = prevEntry{}
-			}
-			for i := range e.longTable[:] {
-				e.longTable[i] = prevEntry{}
-			}
-			e.cur = e.maxMatchOff
-			break
-		}
-		// Shift down everything in the table that isn't already too far away.
-		minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff
-		for i := range e.table[:] {
-			v := e.table[i].offset
-			v2 := e.table[i].prev
-			if v < minOff {
-				v = 0
-				v2 = 0
-			} else {
-				v = v - e.cur + e.maxMatchOff
-				if v2 < minOff {
-					v2 = 0
-				} else {
-					v2 = v2 - e.cur + e.maxMatchOff
-				}
-			}
-			e.table[i] = prevEntry{
-				offset: v,
-				prev:   v2,
-			}
-		}
-		for i := range e.longTable[:] {
-			v := e.longTable[i].offset
-			v2 := e.longTable[i].prev
-			if v < minOff {
-				v = 0
-				v2 = 0
-			} else {
-				v = v - e.cur + e.maxMatchOff
-				if v2 < minOff {
-					v2 = 0
-				} else {
-					v2 = v2 - e.cur + e.maxMatchOff
-				}
-			}
-			e.longTable[i] = prevEntry{
-				offset: v,
-				prev:   v2,
-			}
-		}
-		e.cur = e.maxMatchOff
-		break
-	}
-
-	s := e.addBlock(src)
-	blk.size = len(src)
-	if len(src) < minNonLiteralBlockSize {
-		blk.extraLits = len(src)
-		blk.literals = blk.literals[:len(src)]
-		copy(blk.literals, src)
-		return
-	}
-
-	// Override src
-	src = e.hist
-	sLimit := int32(len(src)) - inputMargin
-	const kSearchStrength = 10
-
-	// nextEmit is where in src the next emitLiteral should start from.
-	nextEmit := s
-	cv := load6432(src, s)
-
-	// Relative offsets
-	offset1 := int32(blk.recentOffsets[0])
-	offset2 := int32(blk.recentOffsets[1])
-	offset3 := int32(blk.recentOffsets[2])
-
-	addLiterals := func(s *seq, until int32) {
-		if until == nextEmit {
-			return
-		}
-		blk.literals = append(blk.literals, src[nextEmit:until]...)
-		s.litLen = uint32(until - nextEmit)
-	}
-	_ = addLiterals
-
-	if debugEncoder {
-		println("recent offsets:", blk.recentOffsets)
-	}
-
-encodeLoop:
-	for {
-		// We allow the encoder to optionally turn off repeat offsets across blocks
-		canRepeat := len(blk.sequences) > 2
-
-		if debugAsserts && canRepeat && offset1 == 0 {
-			panic("offset0 was 0")
-		}
-
-		type match struct {
-			offset int32
-			s      int32
-			length int32
-			rep    int32
-		}
-		matchAt := func(offset int32, s int32, first uint32, rep int32) match {
-			if s-offset >= e.maxMatchOff || load3232(src, offset) != first {
-				return match{offset: offset, s: s}
-			}
-			return match{offset: offset, s: s, length: 4 + e.matchlen(s+4, offset+4, src), rep: rep}
-		}
-
-		bestOf := func(a, b match) match {
-			aScore := b.s - a.s + a.length
-			bScore := a.s - b.s + b.length
-			if a.rep < 0 {
-				aScore = aScore - int32(bits.Len32(uint32(a.offset)))/8
-			}
-			if b.rep < 0 {
-				bScore = bScore - int32(bits.Len32(uint32(b.offset)))/8
-			}
-			if aScore >= bScore {
-				return a
-			}
-			return b
-		}
-		const goodEnough = 100
-
-		nextHashL := hash8(cv, bestLongTableBits)
-		nextHashS := hash4x64(cv, bestShortTableBits)
-		candidateL := e.longTable[nextHashL]
-		candidateS := e.table[nextHashS]
-
-		best := bestOf(matchAt(candidateL.offset-e.cur, s, uint32(cv), -1), matchAt(candidateL.prev-e.cur, s, uint32(cv), -1))
-		best = bestOf(best, matchAt(candidateS.offset-e.cur, s, uint32(cv), -1))
-		best = bestOf(best, matchAt(candidateS.prev-e.cur, s, uint32(cv), -1))
-		if canRepeat && best.length < goodEnough {
-			best = bestOf(best, matchAt(s-offset1+1, s+1, uint32(cv>>8), 1))
-			best = bestOf(best, matchAt(s-offset2+1, s+1, uint32(cv>>8), 2))
-			best = bestOf(best, matchAt(s-offset3+1, s+1, uint32(cv>>8), 3))
-			if best.length > 0 {
-				best = bestOf(best, matchAt(s-offset1+3, s+3, uint32(cv>>24), 1))
-				best = bestOf(best, matchAt(s-offset2+3, s+3, uint32(cv>>24), 2))
-				best = bestOf(best, matchAt(s-offset3+3, s+3, uint32(cv>>24), 3))
-			}
-		}
-		// Load next and check...
-		e.longTable[nextHashL] = prevEntry{offset: s + e.cur, prev: candidateL.offset}
-		e.table[nextHashS] = prevEntry{offset: s + e.cur, prev: candidateS.offset}
-
-		// Look far ahead, unless we have a really long match already...
-		if best.length < goodEnough {
-			// No match found, move forward on input, no need to check forward...
-			if best.length < 4 {
-				s += 1 + (s-nextEmit)>>(kSearchStrength-1)
-				if s >= sLimit {
-					break encodeLoop
-				}
-				cv = load6432(src, s)
-				continue
-			}
-
-			s++
-			candidateS = e.table[hash4x64(cv>>8, bestShortTableBits)]
-			cv = load6432(src, s)
-			cv2 := load6432(src, s+1)
-			candidateL = e.longTable[hash8(cv, bestLongTableBits)]
-			candidateL2 := e.longTable[hash8(cv2, bestLongTableBits)]
-
-			best = bestOf(best, matchAt(candidateS.offset-e.cur, s, uint32(cv), -1))
-			best = bestOf(best, matchAt(candidateL.offset-e.cur, s, uint32(cv), -1))
-			best = bestOf(best, matchAt(candidateL.prev-e.cur, s, uint32(cv), -1))
-			best = bestOf(best, matchAt(candidateL2.offset-e.cur, s+1, uint32(cv2), -1))
-			best = bestOf(best, matchAt(candidateL2.prev-e.cur, s+1, uint32(cv2), -1))
-
-			// See if we can find a better match by checking where the current best ends.
-			// Use that offset to see if we can find a better full match.
-			if sAt := best.s + best.length; sAt < sLimit {
-				nextHashL := hash8(load6432(src, sAt), bestLongTableBits)
-				candidateEnd := e.longTable[nextHashL]
-				if pos := candidateEnd.offset - e.cur - best.length; pos >= 0 {
-					bestEnd := bestOf(best, matchAt(pos, best.s, load3232(src, best.s), -1))
-					if pos := candidateEnd.prev - e.cur - best.length; pos >= 0 {
-						bestEnd = bestOf(bestEnd, matchAt(pos, best.s, load3232(src, best.s), -1))
-					}
-					best = bestEnd
-				}
-			}
-		}
-
-		// We have a match, we can store the forward value
-		if best.rep > 0 {
-			s = best.s
-			var seq seq
-			seq.matchLen = uint32(best.length - zstdMinMatch)
-
-			// We might be able to match backwards.
-			// Extend as long as we can.
-			start := best.s
-			// We end the search early, so we don't risk 0 literals
-			// and have to do special offset treatment.
-			startLimit := nextEmit + 1
-
-			tMin := s - e.maxMatchOff
-			if tMin < 0 {
-				tMin = 0
-			}
-			repIndex := best.offset
-			for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 {
-				repIndex--
-				start--
-				seq.matchLen++
-			}
-			addLiterals(&seq, start)
-
-			// rep 0
-			seq.offset = uint32(best.rep)
-			if debugSequences {
-				println("repeat sequence", seq, "next s:", s)
-			}
-			blk.sequences = append(blk.sequences, seq)
-
-			// Index match start+1 (long) -> s - 1
-			index0 := s
-			s = best.s + best.length
-
-			nextEmit = s
-			if s >= sLimit {
-				if debugEncoder {
-					println("repeat ended", s, best.length)
-
-				}
-				break encodeLoop
-			}
-			// Index skipped...
-			off := index0 + e.cur
-			for index0 < s-1 {
-				cv0 := load6432(src, index0)
-				h0 := hash8(cv0, bestLongTableBits)
-				h1 := hash4x64(cv0, bestShortTableBits)
-				e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset}
-				e.table[h1] = prevEntry{offset: off, prev: e.table[h1].offset}
-				off++
-				index0++
-			}
-			switch best.rep {
-			case 2:
-				offset1, offset2 = offset2, offset1
-			case 3:
-				offset1, offset2, offset3 = offset3, offset1, offset2
-			}
-			cv = load6432(src, s)
-			continue
-		}
-
-		// A 4-byte match has been found. Update recent offsets.
-		// We'll later see if more than 4 bytes.
-		s = best.s
-		t := best.offset
-		offset1, offset2, offset3 = s-t, offset1, offset2
-
-		if debugAsserts && s <= t {
-			panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-		}
-
-		if debugAsserts && canRepeat && int(offset1) > len(src) {
-			panic("invalid offset")
-		}
-
-		// Extend the n-byte match as long as possible.
-		l := best.length
-
-		// Extend backwards
-		tMin := s - e.maxMatchOff
-		if tMin < 0 {
-			tMin = 0
-		}
-		for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength {
-			s--
-			t--
-			l++
-		}
-
-		// Write our sequence
-		var seq seq
-		seq.litLen = uint32(s - nextEmit)
-		seq.matchLen = uint32(l - zstdMinMatch)
-		if seq.litLen > 0 {
-			blk.literals = append(blk.literals, src[nextEmit:s]...)
-		}
-		seq.offset = uint32(s-t) + 3
-		s += l
-		if debugSequences {
-			println("sequence", seq, "next s:", s)
-		}
-		blk.sequences = append(blk.sequences, seq)
-		nextEmit = s
-		if s >= sLimit {
-			break encodeLoop
-		}
-
-		// Index match start+1 (long) -> s - 1
-		index0 := s - l + 1
-		// every entry
-		for index0 < s-1 {
-			cv0 := load6432(src, index0)
-			h0 := hash8(cv0, bestLongTableBits)
-			h1 := hash4x64(cv0, bestShortTableBits)
-			off := index0 + e.cur
-			e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset}
-			e.table[h1] = prevEntry{offset: off, prev: e.table[h1].offset}
-			index0++
-		}
-
-		cv = load6432(src, s)
-		if !canRepeat {
-			continue
-		}
-
-		// Check offset 2
-		for {
-			o2 := s - offset2
-			if load3232(src, o2) != uint32(cv) {
-				// Do regular search
-				break
-			}
-
-			// Store this, since we have it.
-			nextHashS := hash4x64(cv, bestShortTableBits)
-			nextHashL := hash8(cv, bestLongTableBits)
-
-			// We have at least 4 byte match.
-			// No need to check backwards. We come straight from a match
-			l := 4 + e.matchlen(s+4, o2+4, src)
-
-			e.longTable[nextHashL] = prevEntry{offset: s + e.cur, prev: e.longTable[nextHashL].offset}
-			e.table[nextHashS] = prevEntry{offset: s + e.cur, prev: e.table[nextHashS].offset}
-			seq.matchLen = uint32(l) - zstdMinMatch
-			seq.litLen = 0
-
-			// Since litlen is always 0, this is offset 1.
-			seq.offset = 1
-			s += l
-			nextEmit = s
-			if debugSequences {
-				println("sequence", seq, "next s:", s)
-			}
-			blk.sequences = append(blk.sequences, seq)
-
-			// Swap offset 1 and 2.
-			offset1, offset2 = offset2, offset1
-			if s >= sLimit {
-				// Finished
-				break encodeLoop
-			}
-			cv = load6432(src, s)
-		}
-	}
-
-	if int(nextEmit) < len(src) {
-		blk.literals = append(blk.literals, src[nextEmit:]...)
-		blk.extraLits = len(src) - int(nextEmit)
-	}
-	blk.recentOffsets[0] = uint32(offset1)
-	blk.recentOffsets[1] = uint32(offset2)
-	blk.recentOffsets[2] = uint32(offset3)
-	if debugEncoder {
-		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
-	}
-}
-
-// EncodeNoHist will encode a block with no history and no following blocks.
-// Most notable difference is that src will not be copied for history and
-// we do not need to check for max match length.
-func (e *bestFastEncoder) EncodeNoHist(blk *blockEnc, src []byte) {
-	e.ensureHist(len(src))
-	e.Encode(blk, src)
-}
-
-// ResetDict will reset and set a dictionary if not nil
-func (e *bestFastEncoder) Reset(d *dict, singleBlock bool) {
-	e.resetBase(d, singleBlock)
-	if d == nil {
-		return
-	}
-	// Init or copy dict table
-	if len(e.dictTable) != len(e.table) || d.id != e.lastDictID {
-		if len(e.dictTable) != len(e.table) {
-			e.dictTable = make([]prevEntry, len(e.table))
-		}
-		end := int32(len(d.content)) - 8 + e.maxMatchOff
-		for i := e.maxMatchOff; i < end; i += 4 {
-			const hashLog = bestShortTableBits
-
-			cv := load6432(d.content, i-e.maxMatchOff)
-			nextHash := hash4x64(cv, hashLog)      // 0 -> 4
-			nextHash1 := hash4x64(cv>>8, hashLog)  // 1 -> 5
-			nextHash2 := hash4x64(cv>>16, hashLog) // 2 -> 6
-			nextHash3 := hash4x64(cv>>24, hashLog) // 3 -> 7
-			e.dictTable[nextHash] = prevEntry{
-				prev:   e.dictTable[nextHash].offset,
-				offset: i,
-			}
-			e.dictTable[nextHash1] = prevEntry{
-				prev:   e.dictTable[nextHash1].offset,
-				offset: i + 1,
-			}
-			e.dictTable[nextHash2] = prevEntry{
-				prev:   e.dictTable[nextHash2].offset,
-				offset: i + 2,
-			}
-			e.dictTable[nextHash3] = prevEntry{
-				prev:   e.dictTable[nextHash3].offset,
-				offset: i + 3,
-			}
-		}
-		e.lastDictID = d.id
-	}
-
-	// Init or copy dict table
-	if len(e.dictLongTable) != len(e.longTable) || d.id != e.lastDictID {
-		if len(e.dictLongTable) != len(e.longTable) {
-			e.dictLongTable = make([]prevEntry, len(e.longTable))
-		}
-		if len(d.content) >= 8 {
-			cv := load6432(d.content, 0)
-			h := hash8(cv, bestLongTableBits)
-			e.dictLongTable[h] = prevEntry{
-				offset: e.maxMatchOff,
-				prev:   e.dictLongTable[h].offset,
-			}
-
-			end := int32(len(d.content)) - 8 + e.maxMatchOff
-			off := 8 // First to read
-			for i := e.maxMatchOff + 1; i < end; i++ {
-				cv = cv>>8 | (uint64(d.content[off]) << 56)
-				h := hash8(cv, bestLongTableBits)
-				e.dictLongTable[h] = prevEntry{
-					offset: i,
-					prev:   e.dictLongTable[h].offset,
-				}
-				off++
-			}
-		}
-		e.lastDictID = d.id
-	}
-	// Reset table to initial state
-	copy(e.longTable[:], e.dictLongTable)
-
-	e.cur = e.maxMatchOff
-	// Reset table to initial state
-	copy(e.table[:], e.dictTable)
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/enc_better.go b/vendor/github.com/klauspost/compress/zstd/enc_better.go
deleted file mode 100644
index eab7b5083e..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/enc_better.go
+++ /dev/null
@@ -1,1235 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import "fmt"
-
-const (
-	betterLongTableBits = 19                       // Bits used in the long match table
-	betterLongTableSize = 1 << betterLongTableBits // Size of the table
-
-	// Note: Increasing the short table bits or making the hash shorter
-	// can actually lead to compression degradation since it will 'steal' more from the
-	// long match table and match offsets are quite big.
-	// This greatly depends on the type of input.
-	betterShortTableBits = 13                        // Bits used in the short match table
-	betterShortTableSize = 1 << betterShortTableBits // Size of the table
-
-	betterLongTableShardCnt  = 1 << (betterLongTableBits - dictShardBits)    // Number of shards in the table
-	betterLongTableShardSize = betterLongTableSize / betterLongTableShardCnt // Size of an individual shard
-
-	betterShortTableShardCnt  = 1 << (betterShortTableBits - dictShardBits)     // Number of shards in the table
-	betterShortTableShardSize = betterShortTableSize / betterShortTableShardCnt // Size of an individual shard
-)
-
-type prevEntry struct {
-	offset int32
-	prev   int32
-}
-
-// betterFastEncoder uses 2 tables, one for short matches (5 bytes) and one for long matches.
-// The long match table contains the previous entry with the same hash,
-// effectively making it a "chain" of length 2.
-// When we find a long match we choose between the two values and select the longest.
-// When we find a short match, after checking the long, we check if we can find a long at n+1
-// and that it is longer (lazy matching).
-type betterFastEncoder struct {
-	fastBase
-	table     [betterShortTableSize]tableEntry
-	longTable [betterLongTableSize]prevEntry
-}
-
-type betterFastEncoderDict struct {
-	betterFastEncoder
-	dictTable            []tableEntry
-	dictLongTable        []prevEntry
-	shortTableShardDirty [betterShortTableShardCnt]bool
-	longTableShardDirty  [betterLongTableShardCnt]bool
-	allDirty             bool
-}
-
-// Encode improves compression...
-func (e *betterFastEncoder) Encode(blk *blockEnc, src []byte) {
-	const (
-		// Input margin is the number of bytes we read (8)
-		// and the maximum we will read ahead (2)
-		inputMargin            = 8 + 2
-		minNonLiteralBlockSize = 16
-	)
-
-	// Protect against e.cur wraparound.
-	for e.cur >= bufferReset {
-		if len(e.hist) == 0 {
-			for i := range e.table[:] {
-				e.table[i] = tableEntry{}
-			}
-			for i := range e.longTable[:] {
-				e.longTable[i] = prevEntry{}
-			}
-			e.cur = e.maxMatchOff
-			break
-		}
-		// Shift down everything in the table that isn't already too far away.
-		minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff
-		for i := range e.table[:] {
-			v := e.table[i].offset
-			if v < minOff {
-				v = 0
-			} else {
-				v = v - e.cur + e.maxMatchOff
-			}
-			e.table[i].offset = v
-		}
-		for i := range e.longTable[:] {
-			v := e.longTable[i].offset
-			v2 := e.longTable[i].prev
-			if v < minOff {
-				v = 0
-				v2 = 0
-			} else {
-				v = v - e.cur + e.maxMatchOff
-				if v2 < minOff {
-					v2 = 0
-				} else {
-					v2 = v2 - e.cur + e.maxMatchOff
-				}
-			}
-			e.longTable[i] = prevEntry{
-				offset: v,
-				prev:   v2,
-			}
-		}
-		e.cur = e.maxMatchOff
-		break
-	}
-
-	s := e.addBlock(src)
-	blk.size = len(src)
-	if len(src) < minNonLiteralBlockSize {
-		blk.extraLits = len(src)
-		blk.literals = blk.literals[:len(src)]
-		copy(blk.literals, src)
-		return
-	}
-
-	// Override src
-	src = e.hist
-	sLimit := int32(len(src)) - inputMargin
-	// stepSize is the number of bytes to skip on every main loop iteration.
-	// It should be >= 1.
-	const stepSize = 1
-
-	const kSearchStrength = 9
-
-	// nextEmit is where in src the next emitLiteral should start from.
-	nextEmit := s
-	cv := load6432(src, s)
-
-	// Relative offsets
-	offset1 := int32(blk.recentOffsets[0])
-	offset2 := int32(blk.recentOffsets[1])
-
-	addLiterals := func(s *seq, until int32) {
-		if until == nextEmit {
-			return
-		}
-		blk.literals = append(blk.literals, src[nextEmit:until]...)
-		s.litLen = uint32(until - nextEmit)
-	}
-	if debugEncoder {
-		println("recent offsets:", blk.recentOffsets)
-	}
-
-encodeLoop:
-	for {
-		var t int32
-		// We allow the encoder to optionally turn off repeat offsets across blocks
-		canRepeat := len(blk.sequences) > 2
-		var matched int32
-
-		for {
-			if debugAsserts && canRepeat && offset1 == 0 {
-				panic("offset0 was 0")
-			}
-
-			nextHashS := hash5(cv, betterShortTableBits)
-			nextHashL := hash8(cv, betterLongTableBits)
-			candidateL := e.longTable[nextHashL]
-			candidateS := e.table[nextHashS]
-
-			const repOff = 1
-			repIndex := s - offset1 + repOff
-			off := s + e.cur
-			e.longTable[nextHashL] = prevEntry{offset: off, prev: candidateL.offset}
-			e.table[nextHashS] = tableEntry{offset: off, val: uint32(cv)}
-
-			if canRepeat {
-				if repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>(repOff*8)) {
-					// Consider history as well.
-					var seq seq
-					lenght := 4 + e.matchlen(s+4+repOff, repIndex+4, src)
-
-					seq.matchLen = uint32(lenght - zstdMinMatch)
-
-					// We might be able to match backwards.
-					// Extend as long as we can.
-					start := s + repOff
-					// We end the search early, so we don't risk 0 literals
-					// and have to do special offset treatment.
-					startLimit := nextEmit + 1
-
-					tMin := s - e.maxMatchOff
-					if tMin < 0 {
-						tMin = 0
-					}
-					for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 {
-						repIndex--
-						start--
-						seq.matchLen++
-					}
-					addLiterals(&seq, start)
-
-					// rep 0
-					seq.offset = 1
-					if debugSequences {
-						println("repeat sequence", seq, "next s:", s)
-					}
-					blk.sequences = append(blk.sequences, seq)
-
-					// Index match start+1 (long) -> s - 1
-					index0 := s + repOff
-					s += lenght + repOff
-
-					nextEmit = s
-					if s >= sLimit {
-						if debugEncoder {
-							println("repeat ended", s, lenght)
-
-						}
-						break encodeLoop
-					}
-					// Index skipped...
-					for index0 < s-1 {
-						cv0 := load6432(src, index0)
-						cv1 := cv0 >> 8
-						h0 := hash8(cv0, betterLongTableBits)
-						off := index0 + e.cur
-						e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset}
-						e.table[hash5(cv1, betterShortTableBits)] = tableEntry{offset: off + 1, val: uint32(cv1)}
-						index0 += 2
-					}
-					cv = load6432(src, s)
-					continue
-				}
-				const repOff2 = 1
-
-				// We deviate from the reference encoder and also check offset 2.
-				// Still slower and not much better, so disabled.
-				// repIndex = s - offset2 + repOff2
-				if false && repIndex >= 0 && load6432(src, repIndex) == load6432(src, s+repOff) {
-					// Consider history as well.
-					var seq seq
-					lenght := 8 + e.matchlen(s+8+repOff2, repIndex+8, src)
-
-					seq.matchLen = uint32(lenght - zstdMinMatch)
-
-					// We might be able to match backwards.
-					// Extend as long as we can.
-					start := s + repOff2
-					// We end the search early, so we don't risk 0 literals
-					// and have to do special offset treatment.
-					startLimit := nextEmit + 1
-
-					tMin := s - e.maxMatchOff
-					if tMin < 0 {
-						tMin = 0
-					}
-					for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 {
-						repIndex--
-						start--
-						seq.matchLen++
-					}
-					addLiterals(&seq, start)
-
-					// rep 2
-					seq.offset = 2
-					if debugSequences {
-						println("repeat sequence 2", seq, "next s:", s)
-					}
-					blk.sequences = append(blk.sequences, seq)
-
-					index0 := s + repOff2
-					s += lenght + repOff2
-					nextEmit = s
-					if s >= sLimit {
-						if debugEncoder {
-							println("repeat ended", s, lenght)
-
-						}
-						break encodeLoop
-					}
-
-					// Index skipped...
-					for index0 < s-1 {
-						cv0 := load6432(src, index0)
-						cv1 := cv0 >> 8
-						h0 := hash8(cv0, betterLongTableBits)
-						off := index0 + e.cur
-						e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset}
-						e.table[hash5(cv1, betterShortTableBits)] = tableEntry{offset: off + 1, val: uint32(cv1)}
-						index0 += 2
-					}
-					cv = load6432(src, s)
-					// Swap offsets
-					offset1, offset2 = offset2, offset1
-					continue
-				}
-			}
-			// Find the offsets of our two matches.
-			coffsetL := candidateL.offset - e.cur
-			coffsetLP := candidateL.prev - e.cur
-
-			// Check if we have a long match.
-			if s-coffsetL < e.maxMatchOff && cv == load6432(src, coffsetL) {
-				// Found a long match, at least 8 bytes.
-				matched = e.matchlen(s+8, coffsetL+8, src) + 8
-				t = coffsetL
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				if debugMatches {
-					println("long match")
-				}
-
-				if s-coffsetLP < e.maxMatchOff && cv == load6432(src, coffsetLP) {
-					// Found a long match, at least 8 bytes.
-					prevMatch := e.matchlen(s+8, coffsetLP+8, src) + 8
-					if prevMatch > matched {
-						matched = prevMatch
-						t = coffsetLP
-					}
-					if debugAsserts && s <= t {
-						panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-					}
-					if debugAsserts && s-t > e.maxMatchOff {
-						panic("s - t >e.maxMatchOff")
-					}
-					if debugMatches {
-						println("long match")
-					}
-				}
-				break
-			}
-
-			// Check if we have a long match on prev.
-			if s-coffsetLP < e.maxMatchOff && cv == load6432(src, coffsetLP) {
-				// Found a long match, at least 8 bytes.
-				matched = e.matchlen(s+8, coffsetLP+8, src) + 8
-				t = coffsetLP
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				if debugMatches {
-					println("long match")
-				}
-				break
-			}
-
-			coffsetS := candidateS.offset - e.cur
-
-			// Check if we have a short match.
-			if s-coffsetS < e.maxMatchOff && uint32(cv) == candidateS.val {
-				// found a regular match
-				matched = e.matchlen(s+4, coffsetS+4, src) + 4
-
-				// See if we can find a long match at s+1
-				const checkAt = 1
-				cv := load6432(src, s+checkAt)
-				nextHashL = hash8(cv, betterLongTableBits)
-				candidateL = e.longTable[nextHashL]
-				coffsetL = candidateL.offset - e.cur
-
-				// We can store it, since we have at least a 4 byte match.
-				e.longTable[nextHashL] = prevEntry{offset: s + checkAt + e.cur, prev: candidateL.offset}
-				if s-coffsetL < e.maxMatchOff && cv == load6432(src, coffsetL) {
-					// Found a long match, at least 8 bytes.
-					matchedNext := e.matchlen(s+8+checkAt, coffsetL+8, src) + 8
-					if matchedNext > matched {
-						t = coffsetL
-						s += checkAt
-						matched = matchedNext
-						if debugMatches {
-							println("long match (after short)")
-						}
-						break
-					}
-				}
-
-				// Check prev long...
-				coffsetL = candidateL.prev - e.cur
-				if s-coffsetL < e.maxMatchOff && cv == load6432(src, coffsetL) {
-					// Found a long match, at least 8 bytes.
-					matchedNext := e.matchlen(s+8+checkAt, coffsetL+8, src) + 8
-					if matchedNext > matched {
-						t = coffsetL
-						s += checkAt
-						matched = matchedNext
-						if debugMatches {
-							println("prev long match (after short)")
-						}
-						break
-					}
-				}
-				t = coffsetS
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				if debugAsserts && t < 0 {
-					panic("t<0")
-				}
-				if debugMatches {
-					println("short match")
-				}
-				break
-			}
-
-			// No match found, move forward in input.
-			s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1))
-			if s >= sLimit {
-				break encodeLoop
-			}
-			cv = load6432(src, s)
-		}
-
-		// Try to find a better match by searching for a long match at the end of the current best match
-		if true && s+matched < sLimit {
-			nextHashL := hash8(load6432(src, s+matched), betterLongTableBits)
-			cv := load3232(src, s)
-			candidateL := e.longTable[nextHashL]
-			coffsetL := candidateL.offset - e.cur - matched
-			if coffsetL >= 0 && coffsetL < s && s-coffsetL < e.maxMatchOff && cv == load3232(src, coffsetL) {
-				// Found a long match, at least 4 bytes.
-				matchedNext := e.matchlen(s+4, coffsetL+4, src) + 4
-				if matchedNext > matched {
-					t = coffsetL
-					matched = matchedNext
-					if debugMatches {
-						println("long match at end-of-match")
-					}
-				}
-			}
-
-			// Check prev long...
-			if true {
-				coffsetL = candidateL.prev - e.cur - matched
-				if coffsetL >= 0 && coffsetL < s && s-coffsetL < e.maxMatchOff && cv == load3232(src, coffsetL) {
-					// Found a long match, at least 4 bytes.
-					matchedNext := e.matchlen(s+4, coffsetL+4, src) + 4
-					if matchedNext > matched {
-						t = coffsetL
-						matched = matchedNext
-						if debugMatches {
-							println("prev long match at end-of-match")
-						}
-					}
-				}
-			}
-		}
-		// A match has been found. Update recent offsets.
-		offset2 = offset1
-		offset1 = s - t
-
-		if debugAsserts && s <= t {
-			panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-		}
-
-		if debugAsserts && canRepeat && int(offset1) > len(src) {
-			panic("invalid offset")
-		}
-
-		// Extend the n-byte match as long as possible.
-		l := matched
-
-		// Extend backwards
-		tMin := s - e.maxMatchOff
-		if tMin < 0 {
-			tMin = 0
-		}
-		for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength {
-			s--
-			t--
-			l++
-		}
-
-		// Write our sequence
-		var seq seq
-		seq.litLen = uint32(s - nextEmit)
-		seq.matchLen = uint32(l - zstdMinMatch)
-		if seq.litLen > 0 {
-			blk.literals = append(blk.literals, src[nextEmit:s]...)
-		}
-		seq.offset = uint32(s-t) + 3
-		s += l
-		if debugSequences {
-			println("sequence", seq, "next s:", s)
-		}
-		blk.sequences = append(blk.sequences, seq)
-		nextEmit = s
-		if s >= sLimit {
-			break encodeLoop
-		}
-
-		// Index match start+1 (long) -> s - 1
-		index0 := s - l + 1
-		for index0 < s-1 {
-			cv0 := load6432(src, index0)
-			cv1 := cv0 >> 8
-			h0 := hash8(cv0, betterLongTableBits)
-			off := index0 + e.cur
-			e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset}
-			e.table[hash5(cv1, betterShortTableBits)] = tableEntry{offset: off + 1, val: uint32(cv1)}
-			index0 += 2
-		}
-
-		cv = load6432(src, s)
-		if !canRepeat {
-			continue
-		}
-
-		// Check offset 2
-		for {
-			o2 := s - offset2
-			if load3232(src, o2) != uint32(cv) {
-				// Do regular search
-				break
-			}
-
-			// Store this, since we have it.
-			nextHashS := hash5(cv, betterShortTableBits)
-			nextHashL := hash8(cv, betterLongTableBits)
-
-			// We have at least 4 byte match.
-			// No need to check backwards. We come straight from a match
-			l := 4 + e.matchlen(s+4, o2+4, src)
-
-			e.longTable[nextHashL] = prevEntry{offset: s + e.cur, prev: e.longTable[nextHashL].offset}
-			e.table[nextHashS] = tableEntry{offset: s + e.cur, val: uint32(cv)}
-			seq.matchLen = uint32(l) - zstdMinMatch
-			seq.litLen = 0
-
-			// Since litlen is always 0, this is offset 1.
-			seq.offset = 1
-			s += l
-			nextEmit = s
-			if debugSequences {
-				println("sequence", seq, "next s:", s)
-			}
-			blk.sequences = append(blk.sequences, seq)
-
-			// Swap offset 1 and 2.
-			offset1, offset2 = offset2, offset1
-			if s >= sLimit {
-				// Finished
-				break encodeLoop
-			}
-			cv = load6432(src, s)
-		}
-	}
-
-	if int(nextEmit) < len(src) {
-		blk.literals = append(blk.literals, src[nextEmit:]...)
-		blk.extraLits = len(src) - int(nextEmit)
-	}
-	blk.recentOffsets[0] = uint32(offset1)
-	blk.recentOffsets[1] = uint32(offset2)
-	if debugEncoder {
-		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
-	}
-}
-
-// EncodeNoHist will encode a block with no history and no following blocks.
-// Most notable difference is that src will not be copied for history and
-// we do not need to check for max match length.
-func (e *betterFastEncoder) EncodeNoHist(blk *blockEnc, src []byte) {
-	e.ensureHist(len(src))
-	e.Encode(blk, src)
-}
-
-// Encode improves compression...
-func (e *betterFastEncoderDict) Encode(blk *blockEnc, src []byte) {
-	const (
-		// Input margin is the number of bytes we read (8)
-		// and the maximum we will read ahead (2)
-		inputMargin            = 8 + 2
-		minNonLiteralBlockSize = 16
-	)
-
-	// Protect against e.cur wraparound.
-	for e.cur >= bufferReset {
-		if len(e.hist) == 0 {
-			for i := range e.table[:] {
-				e.table[i] = tableEntry{}
-			}
-			for i := range e.longTable[:] {
-				e.longTable[i] = prevEntry{}
-			}
-			e.cur = e.maxMatchOff
-			e.allDirty = true
-			break
-		}
-		// Shift down everything in the table that isn't already too far away.
-		minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff
-		for i := range e.table[:] {
-			v := e.table[i].offset
-			if v < minOff {
-				v = 0
-			} else {
-				v = v - e.cur + e.maxMatchOff
-			}
-			e.table[i].offset = v
-		}
-		for i := range e.longTable[:] {
-			v := e.longTable[i].offset
-			v2 := e.longTable[i].prev
-			if v < minOff {
-				v = 0
-				v2 = 0
-			} else {
-				v = v - e.cur + e.maxMatchOff
-				if v2 < minOff {
-					v2 = 0
-				} else {
-					v2 = v2 - e.cur + e.maxMatchOff
-				}
-			}
-			e.longTable[i] = prevEntry{
-				offset: v,
-				prev:   v2,
-			}
-		}
-		e.allDirty = true
-		e.cur = e.maxMatchOff
-		break
-	}
-
-	s := e.addBlock(src)
-	blk.size = len(src)
-	if len(src) < minNonLiteralBlockSize {
-		blk.extraLits = len(src)
-		blk.literals = blk.literals[:len(src)]
-		copy(blk.literals, src)
-		return
-	}
-
-	// Override src
-	src = e.hist
-	sLimit := int32(len(src)) - inputMargin
-	// stepSize is the number of bytes to skip on every main loop iteration.
-	// It should be >= 1.
-	const stepSize = 1
-
-	const kSearchStrength = 9
-
-	// nextEmit is where in src the next emitLiteral should start from.
-	nextEmit := s
-	cv := load6432(src, s)
-
-	// Relative offsets
-	offset1 := int32(blk.recentOffsets[0])
-	offset2 := int32(blk.recentOffsets[1])
-
-	addLiterals := func(s *seq, until int32) {
-		if until == nextEmit {
-			return
-		}
-		blk.literals = append(blk.literals, src[nextEmit:until]...)
-		s.litLen = uint32(until - nextEmit)
-	}
-	if debugEncoder {
-		println("recent offsets:", blk.recentOffsets)
-	}
-
-encodeLoop:
-	for {
-		var t int32
-		// We allow the encoder to optionally turn off repeat offsets across blocks
-		canRepeat := len(blk.sequences) > 2
-		var matched int32
-
-		for {
-			if debugAsserts && canRepeat && offset1 == 0 {
-				panic("offset0 was 0")
-			}
-
-			nextHashS := hash5(cv, betterShortTableBits)
-			nextHashL := hash8(cv, betterLongTableBits)
-			candidateL := e.longTable[nextHashL]
-			candidateS := e.table[nextHashS]
-
-			const repOff = 1
-			repIndex := s - offset1 + repOff
-			off := s + e.cur
-			e.longTable[nextHashL] = prevEntry{offset: off, prev: candidateL.offset}
-			e.markLongShardDirty(nextHashL)
-			e.table[nextHashS] = tableEntry{offset: off, val: uint32(cv)}
-			e.markShortShardDirty(nextHashS)
-
-			if canRepeat {
-				if repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>(repOff*8)) {
-					// Consider history as well.
-					var seq seq
-					lenght := 4 + e.matchlen(s+4+repOff, repIndex+4, src)
-
-					seq.matchLen = uint32(lenght - zstdMinMatch)
-
-					// We might be able to match backwards.
-					// Extend as long as we can.
-					start := s + repOff
-					// We end the search early, so we don't risk 0 literals
-					// and have to do special offset treatment.
-					startLimit := nextEmit + 1
-
-					tMin := s - e.maxMatchOff
-					if tMin < 0 {
-						tMin = 0
-					}
-					for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 {
-						repIndex--
-						start--
-						seq.matchLen++
-					}
-					addLiterals(&seq, start)
-
-					// rep 0
-					seq.offset = 1
-					if debugSequences {
-						println("repeat sequence", seq, "next s:", s)
-					}
-					blk.sequences = append(blk.sequences, seq)
-
-					// Index match start+1 (long) -> s - 1
-					index0 := s + repOff
-					s += lenght + repOff
-
-					nextEmit = s
-					if s >= sLimit {
-						if debugEncoder {
-							println("repeat ended", s, lenght)
-
-						}
-						break encodeLoop
-					}
-					// Index skipped...
-					for index0 < s-1 {
-						cv0 := load6432(src, index0)
-						cv1 := cv0 >> 8
-						h0 := hash8(cv0, betterLongTableBits)
-						off := index0 + e.cur
-						e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset}
-						e.markLongShardDirty(h0)
-						h1 := hash5(cv1, betterShortTableBits)
-						e.table[h1] = tableEntry{offset: off + 1, val: uint32(cv1)}
-						e.markShortShardDirty(h1)
-						index0 += 2
-					}
-					cv = load6432(src, s)
-					continue
-				}
-				const repOff2 = 1
-
-				// We deviate from the reference encoder and also check offset 2.
-				// Still slower and not much better, so disabled.
-				// repIndex = s - offset2 + repOff2
-				if false && repIndex >= 0 && load6432(src, repIndex) == load6432(src, s+repOff) {
-					// Consider history as well.
-					var seq seq
-					lenght := 8 + e.matchlen(s+8+repOff2, repIndex+8, src)
-
-					seq.matchLen = uint32(lenght - zstdMinMatch)
-
-					// We might be able to match backwards.
-					// Extend as long as we can.
-					start := s + repOff2
-					// We end the search early, so we don't risk 0 literals
-					// and have to do special offset treatment.
-					startLimit := nextEmit + 1
-
-					tMin := s - e.maxMatchOff
-					if tMin < 0 {
-						tMin = 0
-					}
-					for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 {
-						repIndex--
-						start--
-						seq.matchLen++
-					}
-					addLiterals(&seq, start)
-
-					// rep 2
-					seq.offset = 2
-					if debugSequences {
-						println("repeat sequence 2", seq, "next s:", s)
-					}
-					blk.sequences = append(blk.sequences, seq)
-
-					index0 := s + repOff2
-					s += lenght + repOff2
-					nextEmit = s
-					if s >= sLimit {
-						if debugEncoder {
-							println("repeat ended", s, lenght)
-
-						}
-						break encodeLoop
-					}
-
-					// Index skipped...
-					for index0 < s-1 {
-						cv0 := load6432(src, index0)
-						cv1 := cv0 >> 8
-						h0 := hash8(cv0, betterLongTableBits)
-						off := index0 + e.cur
-						e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset}
-						e.markLongShardDirty(h0)
-						h1 := hash5(cv1, betterShortTableBits)
-						e.table[h1] = tableEntry{offset: off + 1, val: uint32(cv1)}
-						e.markShortShardDirty(h1)
-						index0 += 2
-					}
-					cv = load6432(src, s)
-					// Swap offsets
-					offset1, offset2 = offset2, offset1
-					continue
-				}
-			}
-			// Find the offsets of our two matches.
-			coffsetL := candidateL.offset - e.cur
-			coffsetLP := candidateL.prev - e.cur
-
-			// Check if we have a long match.
-			if s-coffsetL < e.maxMatchOff && cv == load6432(src, coffsetL) {
-				// Found a long match, at least 8 bytes.
-				matched = e.matchlen(s+8, coffsetL+8, src) + 8
-				t = coffsetL
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				if debugMatches {
-					println("long match")
-				}
-
-				if s-coffsetLP < e.maxMatchOff && cv == load6432(src, coffsetLP) {
-					// Found a long match, at least 8 bytes.
-					prevMatch := e.matchlen(s+8, coffsetLP+8, src) + 8
-					if prevMatch > matched {
-						matched = prevMatch
-						t = coffsetLP
-					}
-					if debugAsserts && s <= t {
-						panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-					}
-					if debugAsserts && s-t > e.maxMatchOff {
-						panic("s - t >e.maxMatchOff")
-					}
-					if debugMatches {
-						println("long match")
-					}
-				}
-				break
-			}
-
-			// Check if we have a long match on prev.
-			if s-coffsetLP < e.maxMatchOff && cv == load6432(src, coffsetLP) {
-				// Found a long match, at least 8 bytes.
-				matched = e.matchlen(s+8, coffsetLP+8, src) + 8
-				t = coffsetLP
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				if debugMatches {
-					println("long match")
-				}
-				break
-			}
-
-			coffsetS := candidateS.offset - e.cur
-
-			// Check if we have a short match.
-			if s-coffsetS < e.maxMatchOff && uint32(cv) == candidateS.val {
-				// found a regular match
-				matched = e.matchlen(s+4, coffsetS+4, src) + 4
-
-				// See if we can find a long match at s+1
-				const checkAt = 1
-				cv := load6432(src, s+checkAt)
-				nextHashL = hash8(cv, betterLongTableBits)
-				candidateL = e.longTable[nextHashL]
-				coffsetL = candidateL.offset - e.cur
-
-				// We can store it, since we have at least a 4 byte match.
-				e.longTable[nextHashL] = prevEntry{offset: s + checkAt + e.cur, prev: candidateL.offset}
-				e.markLongShardDirty(nextHashL)
-				if s-coffsetL < e.maxMatchOff && cv == load6432(src, coffsetL) {
-					// Found a long match, at least 8 bytes.
-					matchedNext := e.matchlen(s+8+checkAt, coffsetL+8, src) + 8
-					if matchedNext > matched {
-						t = coffsetL
-						s += checkAt
-						matched = matchedNext
-						if debugMatches {
-							println("long match (after short)")
-						}
-						break
-					}
-				}
-
-				// Check prev long...
-				coffsetL = candidateL.prev - e.cur
-				if s-coffsetL < e.maxMatchOff && cv == load6432(src, coffsetL) {
-					// Found a long match, at least 8 bytes.
-					matchedNext := e.matchlen(s+8+checkAt, coffsetL+8, src) + 8
-					if matchedNext > matched {
-						t = coffsetL
-						s += checkAt
-						matched = matchedNext
-						if debugMatches {
-							println("prev long match (after short)")
-						}
-						break
-					}
-				}
-				t = coffsetS
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				if debugAsserts && t < 0 {
-					panic("t<0")
-				}
-				if debugMatches {
-					println("short match")
-				}
-				break
-			}
-
-			// No match found, move forward in input.
-			s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1))
-			if s >= sLimit {
-				break encodeLoop
-			}
-			cv = load6432(src, s)
-		}
-		// Try to find a better match by searching for a long match at the end of the current best match
-		if s+matched < sLimit {
-			nextHashL := hash8(load6432(src, s+matched), betterLongTableBits)
-			cv := load3232(src, s)
-			candidateL := e.longTable[nextHashL]
-			coffsetL := candidateL.offset - e.cur - matched
-			if coffsetL >= 0 && coffsetL < s && s-coffsetL < e.maxMatchOff && cv == load3232(src, coffsetL) {
-				// Found a long match, at least 4 bytes.
-				matchedNext := e.matchlen(s+4, coffsetL+4, src) + 4
-				if matchedNext > matched {
-					t = coffsetL
-					matched = matchedNext
-					if debugMatches {
-						println("long match at end-of-match")
-					}
-				}
-			}
-
-			// Check prev long...
-			if true {
-				coffsetL = candidateL.prev - e.cur - matched
-				if coffsetL >= 0 && coffsetL < s && s-coffsetL < e.maxMatchOff && cv == load3232(src, coffsetL) {
-					// Found a long match, at least 4 bytes.
-					matchedNext := e.matchlen(s+4, coffsetL+4, src) + 4
-					if matchedNext > matched {
-						t = coffsetL
-						matched = matchedNext
-						if debugMatches {
-							println("prev long match at end-of-match")
-						}
-					}
-				}
-			}
-		}
-		// A match has been found. Update recent offsets.
-		offset2 = offset1
-		offset1 = s - t
-
-		if debugAsserts && s <= t {
-			panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-		}
-
-		if debugAsserts && canRepeat && int(offset1) > len(src) {
-			panic("invalid offset")
-		}
-
-		// Extend the n-byte match as long as possible.
-		l := matched
-
-		// Extend backwards
-		tMin := s - e.maxMatchOff
-		if tMin < 0 {
-			tMin = 0
-		}
-		for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength {
-			s--
-			t--
-			l++
-		}
-
-		// Write our sequence
-		var seq seq
-		seq.litLen = uint32(s - nextEmit)
-		seq.matchLen = uint32(l - zstdMinMatch)
-		if seq.litLen > 0 {
-			blk.literals = append(blk.literals, src[nextEmit:s]...)
-		}
-		seq.offset = uint32(s-t) + 3
-		s += l
-		if debugSequences {
-			println("sequence", seq, "next s:", s)
-		}
-		blk.sequences = append(blk.sequences, seq)
-		nextEmit = s
-		if s >= sLimit {
-			break encodeLoop
-		}
-
-		// Index match start+1 (long) -> s - 1
-		index0 := s - l + 1
-		for index0 < s-1 {
-			cv0 := load6432(src, index0)
-			cv1 := cv0 >> 8
-			h0 := hash8(cv0, betterLongTableBits)
-			off := index0 + e.cur
-			e.longTable[h0] = prevEntry{offset: off, prev: e.longTable[h0].offset}
-			e.markLongShardDirty(h0)
-			h1 := hash5(cv1, betterShortTableBits)
-			e.table[h1] = tableEntry{offset: off + 1, val: uint32(cv1)}
-			e.markShortShardDirty(h1)
-			index0 += 2
-		}
-
-		cv = load6432(src, s)
-		if !canRepeat {
-			continue
-		}
-
-		// Check offset 2
-		for {
-			o2 := s - offset2
-			if load3232(src, o2) != uint32(cv) {
-				// Do regular search
-				break
-			}
-
-			// Store this, since we have it.
-			nextHashS := hash5(cv, betterShortTableBits)
-			nextHashL := hash8(cv, betterLongTableBits)
-
-			// We have at least 4 byte match.
-			// No need to check backwards. We come straight from a match
-			l := 4 + e.matchlen(s+4, o2+4, src)
-
-			e.longTable[nextHashL] = prevEntry{offset: s + e.cur, prev: e.longTable[nextHashL].offset}
-			e.markLongShardDirty(nextHashL)
-			e.table[nextHashS] = tableEntry{offset: s + e.cur, val: uint32(cv)}
-			e.markShortShardDirty(nextHashS)
-			seq.matchLen = uint32(l) - zstdMinMatch
-			seq.litLen = 0
-
-			// Since litlen is always 0, this is offset 1.
-			seq.offset = 1
-			s += l
-			nextEmit = s
-			if debugSequences {
-				println("sequence", seq, "next s:", s)
-			}
-			blk.sequences = append(blk.sequences, seq)
-
-			// Swap offset 1 and 2.
-			offset1, offset2 = offset2, offset1
-			if s >= sLimit {
-				// Finished
-				break encodeLoop
-			}
-			cv = load6432(src, s)
-		}
-	}
-
-	if int(nextEmit) < len(src) {
-		blk.literals = append(blk.literals, src[nextEmit:]...)
-		blk.extraLits = len(src) - int(nextEmit)
-	}
-	blk.recentOffsets[0] = uint32(offset1)
-	blk.recentOffsets[1] = uint32(offset2)
-	if debugEncoder {
-		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
-	}
-}
-
-// ResetDict will reset and set a dictionary if not nil
-func (e *betterFastEncoder) Reset(d *dict, singleBlock bool) {
-	e.resetBase(d, singleBlock)
-	if d != nil {
-		panic("betterFastEncoder: Reset with dict")
-	}
-}
-
-// ResetDict will reset and set a dictionary if not nil
-func (e *betterFastEncoderDict) Reset(d *dict, singleBlock bool) {
-	e.resetBase(d, singleBlock)
-	if d == nil {
-		return
-	}
-	// Init or copy dict table
-	if len(e.dictTable) != len(e.table) || d.id != e.lastDictID {
-		if len(e.dictTable) != len(e.table) {
-			e.dictTable = make([]tableEntry, len(e.table))
-		}
-		end := int32(len(d.content)) - 8 + e.maxMatchOff
-		for i := e.maxMatchOff; i < end; i += 4 {
-			const hashLog = betterShortTableBits
-
-			cv := load6432(d.content, i-e.maxMatchOff)
-			nextHash := hash5(cv, hashLog)      // 0 -> 4
-			nextHash1 := hash5(cv>>8, hashLog)  // 1 -> 5
-			nextHash2 := hash5(cv>>16, hashLog) // 2 -> 6
-			nextHash3 := hash5(cv>>24, hashLog) // 3 -> 7
-			e.dictTable[nextHash] = tableEntry{
-				val:    uint32(cv),
-				offset: i,
-			}
-			e.dictTable[nextHash1] = tableEntry{
-				val:    uint32(cv >> 8),
-				offset: i + 1,
-			}
-			e.dictTable[nextHash2] = tableEntry{
-				val:    uint32(cv >> 16),
-				offset: i + 2,
-			}
-			e.dictTable[nextHash3] = tableEntry{
-				val:    uint32(cv >> 24),
-				offset: i + 3,
-			}
-		}
-		e.lastDictID = d.id
-		e.allDirty = true
-	}
-
-	// Init or copy dict table
-	if len(e.dictLongTable) != len(e.longTable) || d.id != e.lastDictID {
-		if len(e.dictLongTable) != len(e.longTable) {
-			e.dictLongTable = make([]prevEntry, len(e.longTable))
-		}
-		if len(d.content) >= 8 {
-			cv := load6432(d.content, 0)
-			h := hash8(cv, betterLongTableBits)
-			e.dictLongTable[h] = prevEntry{
-				offset: e.maxMatchOff,
-				prev:   e.dictLongTable[h].offset,
-			}
-
-			end := int32(len(d.content)) - 8 + e.maxMatchOff
-			off := 8 // First to read
-			for i := e.maxMatchOff + 1; i < end; i++ {
-				cv = cv>>8 | (uint64(d.content[off]) << 56)
-				h := hash8(cv, betterLongTableBits)
-				e.dictLongTable[h] = prevEntry{
-					offset: i,
-					prev:   e.dictLongTable[h].offset,
-				}
-				off++
-			}
-		}
-		e.lastDictID = d.id
-		e.allDirty = true
-	}
-
-	// Reset table to initial state
-	{
-		dirtyShardCnt := 0
-		if !e.allDirty {
-			for i := range e.shortTableShardDirty {
-				if e.shortTableShardDirty[i] {
-					dirtyShardCnt++
-				}
-			}
-		}
-		const shardCnt = betterShortTableShardCnt
-		const shardSize = betterShortTableShardSize
-		if e.allDirty || dirtyShardCnt > shardCnt*4/6 {
-			copy(e.table[:], e.dictTable)
-			for i := range e.shortTableShardDirty {
-				e.shortTableShardDirty[i] = false
-			}
-		} else {
-			for i := range e.shortTableShardDirty {
-				if !e.shortTableShardDirty[i] {
-					continue
-				}
-
-				copy(e.table[i*shardSize:(i+1)*shardSize], e.dictTable[i*shardSize:(i+1)*shardSize])
-				e.shortTableShardDirty[i] = false
-			}
-		}
-	}
-	{
-		dirtyShardCnt := 0
-		if !e.allDirty {
-			for i := range e.shortTableShardDirty {
-				if e.shortTableShardDirty[i] {
-					dirtyShardCnt++
-				}
-			}
-		}
-		const shardCnt = betterLongTableShardCnt
-		const shardSize = betterLongTableShardSize
-		if e.allDirty || dirtyShardCnt > shardCnt*4/6 {
-			copy(e.longTable[:], e.dictLongTable)
-			for i := range e.longTableShardDirty {
-				e.longTableShardDirty[i] = false
-			}
-		} else {
-			for i := range e.longTableShardDirty {
-				if !e.longTableShardDirty[i] {
-					continue
-				}
-
-				copy(e.longTable[i*shardSize:(i+1)*shardSize], e.dictLongTable[i*shardSize:(i+1)*shardSize])
-				e.longTableShardDirty[i] = false
-			}
-		}
-	}
-	e.cur = e.maxMatchOff
-	e.allDirty = false
-}
-
-func (e *betterFastEncoderDict) markLongShardDirty(entryNum uint32) {
-	e.longTableShardDirty[entryNum/betterLongTableShardSize] = true
-}
-
-func (e *betterFastEncoderDict) markShortShardDirty(entryNum uint32) {
-	e.shortTableShardDirty[entryNum/betterShortTableShardSize] = true
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/enc_dfast.go b/vendor/github.com/klauspost/compress/zstd/enc_dfast.go
deleted file mode 100644
index 96b21b90e8..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/enc_dfast.go
+++ /dev/null
@@ -1,1121 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import "fmt"
-
-const (
-	dFastLongTableBits = 17                      // Bits used in the long match table
-	dFastLongTableSize = 1 << dFastLongTableBits // Size of the table
-	dFastLongTableMask = dFastLongTableSize - 1  // Mask for table indices. Redundant, but can eliminate bounds checks.
-
-	dLongTableShardCnt  = 1 << (dFastLongTableBits - dictShardBits) // Number of shards in the table
-	dLongTableShardSize = dFastLongTableSize / tableShardCnt        // Size of an individual shard
-
-	dFastShortTableBits = tableBits                // Bits used in the short match table
-	dFastShortTableSize = 1 << dFastShortTableBits // Size of the table
-	dFastShortTableMask = dFastShortTableSize - 1  // Mask for table indices. Redundant, but can eliminate bounds checks.
-)
-
-type doubleFastEncoder struct {
-	fastEncoder
-	longTable [dFastLongTableSize]tableEntry
-}
-
-type doubleFastEncoderDict struct {
-	fastEncoderDict
-	longTable           [dFastLongTableSize]tableEntry
-	dictLongTable       []tableEntry
-	longTableShardDirty [dLongTableShardCnt]bool
-}
-
-// Encode mimmics functionality in zstd_dfast.c
-func (e *doubleFastEncoder) Encode(blk *blockEnc, src []byte) {
-	const (
-		// Input margin is the number of bytes we read (8)
-		// and the maximum we will read ahead (2)
-		inputMargin            = 8 + 2
-		minNonLiteralBlockSize = 16
-	)
-
-	// Protect against e.cur wraparound.
-	for e.cur >= bufferReset {
-		if len(e.hist) == 0 {
-			for i := range e.table[:] {
-				e.table[i] = tableEntry{}
-			}
-			for i := range e.longTable[:] {
-				e.longTable[i] = tableEntry{}
-			}
-			e.cur = e.maxMatchOff
-			break
-		}
-		// Shift down everything in the table that isn't already too far away.
-		minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff
-		for i := range e.table[:] {
-			v := e.table[i].offset
-			if v < minOff {
-				v = 0
-			} else {
-				v = v - e.cur + e.maxMatchOff
-			}
-			e.table[i].offset = v
-		}
-		for i := range e.longTable[:] {
-			v := e.longTable[i].offset
-			if v < minOff {
-				v = 0
-			} else {
-				v = v - e.cur + e.maxMatchOff
-			}
-			e.longTable[i].offset = v
-		}
-		e.cur = e.maxMatchOff
-		break
-	}
-
-	s := e.addBlock(src)
-	blk.size = len(src)
-	if len(src) < minNonLiteralBlockSize {
-		blk.extraLits = len(src)
-		blk.literals = blk.literals[:len(src)]
-		copy(blk.literals, src)
-		return
-	}
-
-	// Override src
-	src = e.hist
-	sLimit := int32(len(src)) - inputMargin
-	// stepSize is the number of bytes to skip on every main loop iteration.
-	// It should be >= 1.
-	const stepSize = 1
-
-	const kSearchStrength = 8
-
-	// nextEmit is where in src the next emitLiteral should start from.
-	nextEmit := s
-	cv := load6432(src, s)
-
-	// Relative offsets
-	offset1 := int32(blk.recentOffsets[0])
-	offset2 := int32(blk.recentOffsets[1])
-
-	addLiterals := func(s *seq, until int32) {
-		if until == nextEmit {
-			return
-		}
-		blk.literals = append(blk.literals, src[nextEmit:until]...)
-		s.litLen = uint32(until - nextEmit)
-	}
-	if debugEncoder {
-		println("recent offsets:", blk.recentOffsets)
-	}
-
-encodeLoop:
-	for {
-		var t int32
-		// We allow the encoder to optionally turn off repeat offsets across blocks
-		canRepeat := len(blk.sequences) > 2
-
-		for {
-			if debugAsserts && canRepeat && offset1 == 0 {
-				panic("offset0 was 0")
-			}
-
-			nextHashS := hash5(cv, dFastShortTableBits)
-			nextHashL := hash8(cv, dFastLongTableBits)
-			candidateL := e.longTable[nextHashL]
-			candidateS := e.table[nextHashS]
-
-			const repOff = 1
-			repIndex := s - offset1 + repOff
-			entry := tableEntry{offset: s + e.cur, val: uint32(cv)}
-			e.longTable[nextHashL] = entry
-			e.table[nextHashS] = entry
-
-			if canRepeat {
-				if repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>(repOff*8)) {
-					// Consider history as well.
-					var seq seq
-					lenght := 4 + e.matchlen(s+4+repOff, repIndex+4, src)
-
-					seq.matchLen = uint32(lenght - zstdMinMatch)
-
-					// We might be able to match backwards.
-					// Extend as long as we can.
-					start := s + repOff
-					// We end the search early, so we don't risk 0 literals
-					// and have to do special offset treatment.
-					startLimit := nextEmit + 1
-
-					tMin := s - e.maxMatchOff
-					if tMin < 0 {
-						tMin = 0
-					}
-					for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 {
-						repIndex--
-						start--
-						seq.matchLen++
-					}
-					addLiterals(&seq, start)
-
-					// rep 0
-					seq.offset = 1
-					if debugSequences {
-						println("repeat sequence", seq, "next s:", s)
-					}
-					blk.sequences = append(blk.sequences, seq)
-					s += lenght + repOff
-					nextEmit = s
-					if s >= sLimit {
-						if debugEncoder {
-							println("repeat ended", s, lenght)
-
-						}
-						break encodeLoop
-					}
-					cv = load6432(src, s)
-					continue
-				}
-			}
-			// Find the offsets of our two matches.
-			coffsetL := s - (candidateL.offset - e.cur)
-			coffsetS := s - (candidateS.offset - e.cur)
-
-			// Check if we have a long match.
-			if coffsetL < e.maxMatchOff && uint32(cv) == candidateL.val {
-				// Found a long match, likely at least 8 bytes.
-				// Reference encoder checks all 8 bytes, we only check 4,
-				// but the likelihood of both the first 4 bytes and the hash matching should be enough.
-				t = candidateL.offset - e.cur
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				if debugMatches {
-					println("long match")
-				}
-				break
-			}
-
-			// Check if we have a short match.
-			if coffsetS < e.maxMatchOff && uint32(cv) == candidateS.val {
-				// found a regular match
-				// See if we can find a long match at s+1
-				const checkAt = 1
-				cv := load6432(src, s+checkAt)
-				nextHashL = hash8(cv, dFastLongTableBits)
-				candidateL = e.longTable[nextHashL]
-				coffsetL = s - (candidateL.offset - e.cur) + checkAt
-
-				// We can store it, since we have at least a 4 byte match.
-				e.longTable[nextHashL] = tableEntry{offset: s + checkAt + e.cur, val: uint32(cv)}
-				if coffsetL < e.maxMatchOff && uint32(cv) == candidateL.val {
-					// Found a long match, likely at least 8 bytes.
-					// Reference encoder checks all 8 bytes, we only check 4,
-					// but the likelihood of both the first 4 bytes and the hash matching should be enough.
-					t = candidateL.offset - e.cur
-					s += checkAt
-					if debugMatches {
-						println("long match (after short)")
-					}
-					break
-				}
-
-				t = candidateS.offset - e.cur
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				if debugAsserts && t < 0 {
-					panic("t<0")
-				}
-				if debugMatches {
-					println("short match")
-				}
-				break
-			}
-
-			// No match found, move forward in input.
-			s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1))
-			if s >= sLimit {
-				break encodeLoop
-			}
-			cv = load6432(src, s)
-		}
-
-		// A 4-byte match has been found. Update recent offsets.
-		// We'll later see if more than 4 bytes.
-		offset2 = offset1
-		offset1 = s - t
-
-		if debugAsserts && s <= t {
-			panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-		}
-
-		if debugAsserts && canRepeat && int(offset1) > len(src) {
-			panic("invalid offset")
-		}
-
-		// Extend the 4-byte match as long as possible.
-		l := e.matchlen(s+4, t+4, src) + 4
-
-		// Extend backwards
-		tMin := s - e.maxMatchOff
-		if tMin < 0 {
-			tMin = 0
-		}
-		for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength {
-			s--
-			t--
-			l++
-		}
-
-		// Write our sequence
-		var seq seq
-		seq.litLen = uint32(s - nextEmit)
-		seq.matchLen = uint32(l - zstdMinMatch)
-		if seq.litLen > 0 {
-			blk.literals = append(blk.literals, src[nextEmit:s]...)
-		}
-		seq.offset = uint32(s-t) + 3
-		s += l
-		if debugSequences {
-			println("sequence", seq, "next s:", s)
-		}
-		blk.sequences = append(blk.sequences, seq)
-		nextEmit = s
-		if s >= sLimit {
-			break encodeLoop
-		}
-
-		// Index match start+1 (long) and start+2 (short)
-		index0 := s - l + 1
-		// Index match end-2 (long) and end-1 (short)
-		index1 := s - 2
-
-		cv0 := load6432(src, index0)
-		cv1 := load6432(src, index1)
-		te0 := tableEntry{offset: index0 + e.cur, val: uint32(cv0)}
-		te1 := tableEntry{offset: index1 + e.cur, val: uint32(cv1)}
-		e.longTable[hash8(cv0, dFastLongTableBits)] = te0
-		e.longTable[hash8(cv1, dFastLongTableBits)] = te1
-		cv0 >>= 8
-		cv1 >>= 8
-		te0.offset++
-		te1.offset++
-		te0.val = uint32(cv0)
-		te1.val = uint32(cv1)
-		e.table[hash5(cv0, dFastShortTableBits)] = te0
-		e.table[hash5(cv1, dFastShortTableBits)] = te1
-
-		cv = load6432(src, s)
-
-		if !canRepeat {
-			continue
-		}
-
-		// Check offset 2
-		for {
-			o2 := s - offset2
-			if load3232(src, o2) != uint32(cv) {
-				// Do regular search
-				break
-			}
-
-			// Store this, since we have it.
-			nextHashS := hash5(cv, dFastShortTableBits)
-			nextHashL := hash8(cv, dFastLongTableBits)
-
-			// We have at least 4 byte match.
-			// No need to check backwards. We come straight from a match
-			l := 4 + e.matchlen(s+4, o2+4, src)
-
-			entry := tableEntry{offset: s + e.cur, val: uint32(cv)}
-			e.longTable[nextHashL] = entry
-			e.table[nextHashS] = entry
-			seq.matchLen = uint32(l) - zstdMinMatch
-			seq.litLen = 0
-
-			// Since litlen is always 0, this is offset 1.
-			seq.offset = 1
-			s += l
-			nextEmit = s
-			if debugSequences {
-				println("sequence", seq, "next s:", s)
-			}
-			blk.sequences = append(blk.sequences, seq)
-
-			// Swap offset 1 and 2.
-			offset1, offset2 = offset2, offset1
-			if s >= sLimit {
-				// Finished
-				break encodeLoop
-			}
-			cv = load6432(src, s)
-		}
-	}
-
-	if int(nextEmit) < len(src) {
-		blk.literals = append(blk.literals, src[nextEmit:]...)
-		blk.extraLits = len(src) - int(nextEmit)
-	}
-	blk.recentOffsets[0] = uint32(offset1)
-	blk.recentOffsets[1] = uint32(offset2)
-	if debugEncoder {
-		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
-	}
-}
-
-// EncodeNoHist will encode a block with no history and no following blocks.
-// Most notable difference is that src will not be copied for history and
-// we do not need to check for max match length.
-func (e *doubleFastEncoder) EncodeNoHist(blk *blockEnc, src []byte) {
-	const (
-		// Input margin is the number of bytes we read (8)
-		// and the maximum we will read ahead (2)
-		inputMargin            = 8 + 2
-		minNonLiteralBlockSize = 16
-	)
-
-	// Protect against e.cur wraparound.
-	if e.cur >= bufferReset {
-		for i := range e.table[:] {
-			e.table[i] = tableEntry{}
-		}
-		for i := range e.longTable[:] {
-			e.longTable[i] = tableEntry{}
-		}
-		e.cur = e.maxMatchOff
-	}
-
-	s := int32(0)
-	blk.size = len(src)
-	if len(src) < minNonLiteralBlockSize {
-		blk.extraLits = len(src)
-		blk.literals = blk.literals[:len(src)]
-		copy(blk.literals, src)
-		return
-	}
-
-	// Override src
-	sLimit := int32(len(src)) - inputMargin
-	// stepSize is the number of bytes to skip on every main loop iteration.
-	// It should be >= 1.
-	const stepSize = 1
-
-	const kSearchStrength = 8
-
-	// nextEmit is where in src the next emitLiteral should start from.
-	nextEmit := s
-	cv := load6432(src, s)
-
-	// Relative offsets
-	offset1 := int32(blk.recentOffsets[0])
-	offset2 := int32(blk.recentOffsets[1])
-
-	addLiterals := func(s *seq, until int32) {
-		if until == nextEmit {
-			return
-		}
-		blk.literals = append(blk.literals, src[nextEmit:until]...)
-		s.litLen = uint32(until - nextEmit)
-	}
-	if debugEncoder {
-		println("recent offsets:", blk.recentOffsets)
-	}
-
-encodeLoop:
-	for {
-		var t int32
-		for {
-
-			nextHashS := hash5(cv, dFastShortTableBits)
-			nextHashL := hash8(cv, dFastLongTableBits)
-			candidateL := e.longTable[nextHashL]
-			candidateS := e.table[nextHashS]
-
-			const repOff = 1
-			repIndex := s - offset1 + repOff
-			entry := tableEntry{offset: s + e.cur, val: uint32(cv)}
-			e.longTable[nextHashL] = entry
-			e.table[nextHashS] = entry
-
-			if len(blk.sequences) > 2 {
-				if load3232(src, repIndex) == uint32(cv>>(repOff*8)) {
-					// Consider history as well.
-					var seq seq
-					//length := 4 + e.matchlen(s+4+repOff, repIndex+4, src)
-					length := 4 + int32(matchLen(src[s+4+repOff:], src[repIndex+4:]))
-
-					seq.matchLen = uint32(length - zstdMinMatch)
-
-					// We might be able to match backwards.
-					// Extend as long as we can.
-					start := s + repOff
-					// We end the search early, so we don't risk 0 literals
-					// and have to do special offset treatment.
-					startLimit := nextEmit + 1
-
-					tMin := s - e.maxMatchOff
-					if tMin < 0 {
-						tMin = 0
-					}
-					for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] {
-						repIndex--
-						start--
-						seq.matchLen++
-					}
-					addLiterals(&seq, start)
-
-					// rep 0
-					seq.offset = 1
-					if debugSequences {
-						println("repeat sequence", seq, "next s:", s)
-					}
-					blk.sequences = append(blk.sequences, seq)
-					s += length + repOff
-					nextEmit = s
-					if s >= sLimit {
-						if debugEncoder {
-							println("repeat ended", s, length)
-
-						}
-						break encodeLoop
-					}
-					cv = load6432(src, s)
-					continue
-				}
-			}
-			// Find the offsets of our two matches.
-			coffsetL := s - (candidateL.offset - e.cur)
-			coffsetS := s - (candidateS.offset - e.cur)
-
-			// Check if we have a long match.
-			if coffsetL < e.maxMatchOff && uint32(cv) == candidateL.val {
-				// Found a long match, likely at least 8 bytes.
-				// Reference encoder checks all 8 bytes, we only check 4,
-				// but the likelihood of both the first 4 bytes and the hash matching should be enough.
-				t = candidateL.offset - e.cur
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d). cur: %d", s, t, e.cur))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				if debugMatches {
-					println("long match")
-				}
-				break
-			}
-
-			// Check if we have a short match.
-			if coffsetS < e.maxMatchOff && uint32(cv) == candidateS.val {
-				// found a regular match
-				// See if we can find a long match at s+1
-				const checkAt = 1
-				cv := load6432(src, s+checkAt)
-				nextHashL = hash8(cv, dFastLongTableBits)
-				candidateL = e.longTable[nextHashL]
-				coffsetL = s - (candidateL.offset - e.cur) + checkAt
-
-				// We can store it, since we have at least a 4 byte match.
-				e.longTable[nextHashL] = tableEntry{offset: s + checkAt + e.cur, val: uint32(cv)}
-				if coffsetL < e.maxMatchOff && uint32(cv) == candidateL.val {
-					// Found a long match, likely at least 8 bytes.
-					// Reference encoder checks all 8 bytes, we only check 4,
-					// but the likelihood of both the first 4 bytes and the hash matching should be enough.
-					t = candidateL.offset - e.cur
-					s += checkAt
-					if debugMatches {
-						println("long match (after short)")
-					}
-					break
-				}
-
-				t = candidateS.offset - e.cur
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				if debugAsserts && t < 0 {
-					panic("t<0")
-				}
-				if debugMatches {
-					println("short match")
-				}
-				break
-			}
-
-			// No match found, move forward in input.
-			s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1))
-			if s >= sLimit {
-				break encodeLoop
-			}
-			cv = load6432(src, s)
-		}
-
-		// A 4-byte match has been found. Update recent offsets.
-		// We'll later see if more than 4 bytes.
-		offset2 = offset1
-		offset1 = s - t
-
-		if debugAsserts && s <= t {
-			panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-		}
-
-		// Extend the 4-byte match as long as possible.
-		//l := e.matchlen(s+4, t+4, src) + 4
-		l := int32(matchLen(src[s+4:], src[t+4:])) + 4
-
-		// Extend backwards
-		tMin := s - e.maxMatchOff
-		if tMin < 0 {
-			tMin = 0
-		}
-		for t > tMin && s > nextEmit && src[t-1] == src[s-1] {
-			s--
-			t--
-			l++
-		}
-
-		// Write our sequence
-		var seq seq
-		seq.litLen = uint32(s - nextEmit)
-		seq.matchLen = uint32(l - zstdMinMatch)
-		if seq.litLen > 0 {
-			blk.literals = append(blk.literals, src[nextEmit:s]...)
-		}
-		seq.offset = uint32(s-t) + 3
-		s += l
-		if debugSequences {
-			println("sequence", seq, "next s:", s)
-		}
-		blk.sequences = append(blk.sequences, seq)
-		nextEmit = s
-		if s >= sLimit {
-			break encodeLoop
-		}
-
-		// Index match start+1 (long) and start+2 (short)
-		index0 := s - l + 1
-		// Index match end-2 (long) and end-1 (short)
-		index1 := s - 2
-
-		cv0 := load6432(src, index0)
-		cv1 := load6432(src, index1)
-		te0 := tableEntry{offset: index0 + e.cur, val: uint32(cv0)}
-		te1 := tableEntry{offset: index1 + e.cur, val: uint32(cv1)}
-		e.longTable[hash8(cv0, dFastLongTableBits)] = te0
-		e.longTable[hash8(cv1, dFastLongTableBits)] = te1
-		cv0 >>= 8
-		cv1 >>= 8
-		te0.offset++
-		te1.offset++
-		te0.val = uint32(cv0)
-		te1.val = uint32(cv1)
-		e.table[hash5(cv0, dFastShortTableBits)] = te0
-		e.table[hash5(cv1, dFastShortTableBits)] = te1
-
-		cv = load6432(src, s)
-
-		if len(blk.sequences) <= 2 {
-			continue
-		}
-
-		// Check offset 2
-		for {
-			o2 := s - offset2
-			if load3232(src, o2) != uint32(cv) {
-				// Do regular search
-				break
-			}
-
-			// Store this, since we have it.
-			nextHashS := hash5(cv1>>8, dFastShortTableBits)
-			nextHashL := hash8(cv, dFastLongTableBits)
-
-			// We have at least 4 byte match.
-			// No need to check backwards. We come straight from a match
-			//l := 4 + e.matchlen(s+4, o2+4, src)
-			l := 4 + int32(matchLen(src[s+4:], src[o2+4:]))
-
-			entry := tableEntry{offset: s + e.cur, val: uint32(cv)}
-			e.longTable[nextHashL] = entry
-			e.table[nextHashS] = entry
-			seq.matchLen = uint32(l) - zstdMinMatch
-			seq.litLen = 0
-
-			// Since litlen is always 0, this is offset 1.
-			seq.offset = 1
-			s += l
-			nextEmit = s
-			if debugSequences {
-				println("sequence", seq, "next s:", s)
-			}
-			blk.sequences = append(blk.sequences, seq)
-
-			// Swap offset 1 and 2.
-			offset1, offset2 = offset2, offset1
-			if s >= sLimit {
-				// Finished
-				break encodeLoop
-			}
-			cv = load6432(src, s)
-		}
-	}
-
-	if int(nextEmit) < len(src) {
-		blk.literals = append(blk.literals, src[nextEmit:]...)
-		blk.extraLits = len(src) - int(nextEmit)
-	}
-	if debugEncoder {
-		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
-	}
-
-	// We do not store history, so we must offset e.cur to avoid false matches for next user.
-	if e.cur < bufferReset {
-		e.cur += int32(len(src))
-	}
-}
-
-// Encode will encode the content, with a dictionary if initialized for it.
-func (e *doubleFastEncoderDict) Encode(blk *blockEnc, src []byte) {
-	const (
-		// Input margin is the number of bytes we read (8)
-		// and the maximum we will read ahead (2)
-		inputMargin            = 8 + 2
-		minNonLiteralBlockSize = 16
-	)
-
-	// Protect against e.cur wraparound.
-	for e.cur >= bufferReset {
-		if len(e.hist) == 0 {
-			for i := range e.table[:] {
-				e.table[i] = tableEntry{}
-			}
-			for i := range e.longTable[:] {
-				e.longTable[i] = tableEntry{}
-			}
-			e.markAllShardsDirty()
-			e.cur = e.maxMatchOff
-			break
-		}
-		// Shift down everything in the table that isn't already too far away.
-		minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff
-		for i := range e.table[:] {
-			v := e.table[i].offset
-			if v < minOff {
-				v = 0
-			} else {
-				v = v - e.cur + e.maxMatchOff
-			}
-			e.table[i].offset = v
-		}
-		for i := range e.longTable[:] {
-			v := e.longTable[i].offset
-			if v < minOff {
-				v = 0
-			} else {
-				v = v - e.cur + e.maxMatchOff
-			}
-			e.longTable[i].offset = v
-		}
-		e.markAllShardsDirty()
-		e.cur = e.maxMatchOff
-		break
-	}
-
-	s := e.addBlock(src)
-	blk.size = len(src)
-	if len(src) < minNonLiteralBlockSize {
-		blk.extraLits = len(src)
-		blk.literals = blk.literals[:len(src)]
-		copy(blk.literals, src)
-		return
-	}
-
-	// Override src
-	src = e.hist
-	sLimit := int32(len(src)) - inputMargin
-	// stepSize is the number of bytes to skip on every main loop iteration.
-	// It should be >= 1.
-	const stepSize = 1
-
-	const kSearchStrength = 8
-
-	// nextEmit is where in src the next emitLiteral should start from.
-	nextEmit := s
-	cv := load6432(src, s)
-
-	// Relative offsets
-	offset1 := int32(blk.recentOffsets[0])
-	offset2 := int32(blk.recentOffsets[1])
-
-	addLiterals := func(s *seq, until int32) {
-		if until == nextEmit {
-			return
-		}
-		blk.literals = append(blk.literals, src[nextEmit:until]...)
-		s.litLen = uint32(until - nextEmit)
-	}
-	if debugEncoder {
-		println("recent offsets:", blk.recentOffsets)
-	}
-
-encodeLoop:
-	for {
-		var t int32
-		// We allow the encoder to optionally turn off repeat offsets across blocks
-		canRepeat := len(blk.sequences) > 2
-
-		for {
-			if debugAsserts && canRepeat && offset1 == 0 {
-				panic("offset0 was 0")
-			}
-
-			nextHashS := hash5(cv, dFastShortTableBits)
-			nextHashL := hash8(cv, dFastLongTableBits)
-			candidateL := e.longTable[nextHashL]
-			candidateS := e.table[nextHashS]
-
-			const repOff = 1
-			repIndex := s - offset1 + repOff
-			entry := tableEntry{offset: s + e.cur, val: uint32(cv)}
-			e.longTable[nextHashL] = entry
-			e.markLongShardDirty(nextHashL)
-			e.table[nextHashS] = entry
-			e.markShardDirty(nextHashS)
-
-			if canRepeat {
-				if repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>(repOff*8)) {
-					// Consider history as well.
-					var seq seq
-					lenght := 4 + e.matchlen(s+4+repOff, repIndex+4, src)
-
-					seq.matchLen = uint32(lenght - zstdMinMatch)
-
-					// We might be able to match backwards.
-					// Extend as long as we can.
-					start := s + repOff
-					// We end the search early, so we don't risk 0 literals
-					// and have to do special offset treatment.
-					startLimit := nextEmit + 1
-
-					tMin := s - e.maxMatchOff
-					if tMin < 0 {
-						tMin = 0
-					}
-					for repIndex > tMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch-1 {
-						repIndex--
-						start--
-						seq.matchLen++
-					}
-					addLiterals(&seq, start)
-
-					// rep 0
-					seq.offset = 1
-					if debugSequences {
-						println("repeat sequence", seq, "next s:", s)
-					}
-					blk.sequences = append(blk.sequences, seq)
-					s += lenght + repOff
-					nextEmit = s
-					if s >= sLimit {
-						if debugEncoder {
-							println("repeat ended", s, lenght)
-
-						}
-						break encodeLoop
-					}
-					cv = load6432(src, s)
-					continue
-				}
-			}
-			// Find the offsets of our two matches.
-			coffsetL := s - (candidateL.offset - e.cur)
-			coffsetS := s - (candidateS.offset - e.cur)
-
-			// Check if we have a long match.
-			if coffsetL < e.maxMatchOff && uint32(cv) == candidateL.val {
-				// Found a long match, likely at least 8 bytes.
-				// Reference encoder checks all 8 bytes, we only check 4,
-				// but the likelihood of both the first 4 bytes and the hash matching should be enough.
-				t = candidateL.offset - e.cur
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				if debugMatches {
-					println("long match")
-				}
-				break
-			}
-
-			// Check if we have a short match.
-			if coffsetS < e.maxMatchOff && uint32(cv) == candidateS.val {
-				// found a regular match
-				// See if we can find a long match at s+1
-				const checkAt = 1
-				cv := load6432(src, s+checkAt)
-				nextHashL = hash8(cv, dFastLongTableBits)
-				candidateL = e.longTable[nextHashL]
-				coffsetL = s - (candidateL.offset - e.cur) + checkAt
-
-				// We can store it, since we have at least a 4 byte match.
-				e.longTable[nextHashL] = tableEntry{offset: s + checkAt + e.cur, val: uint32(cv)}
-				e.markLongShardDirty(nextHashL)
-				if coffsetL < e.maxMatchOff && uint32(cv) == candidateL.val {
-					// Found a long match, likely at least 8 bytes.
-					// Reference encoder checks all 8 bytes, we only check 4,
-					// but the likelihood of both the first 4 bytes and the hash matching should be enough.
-					t = candidateL.offset - e.cur
-					s += checkAt
-					if debugMatches {
-						println("long match (after short)")
-					}
-					break
-				}
-
-				t = candidateS.offset - e.cur
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				if debugAsserts && t < 0 {
-					panic("t<0")
-				}
-				if debugMatches {
-					println("short match")
-				}
-				break
-			}
-
-			// No match found, move forward in input.
-			s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1))
-			if s >= sLimit {
-				break encodeLoop
-			}
-			cv = load6432(src, s)
-		}
-
-		// A 4-byte match has been found. Update recent offsets.
-		// We'll later see if more than 4 bytes.
-		offset2 = offset1
-		offset1 = s - t
-
-		if debugAsserts && s <= t {
-			panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-		}
-
-		if debugAsserts && canRepeat && int(offset1) > len(src) {
-			panic("invalid offset")
-		}
-
-		// Extend the 4-byte match as long as possible.
-		l := e.matchlen(s+4, t+4, src) + 4
-
-		// Extend backwards
-		tMin := s - e.maxMatchOff
-		if tMin < 0 {
-			tMin = 0
-		}
-		for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength {
-			s--
-			t--
-			l++
-		}
-
-		// Write our sequence
-		var seq seq
-		seq.litLen = uint32(s - nextEmit)
-		seq.matchLen = uint32(l - zstdMinMatch)
-		if seq.litLen > 0 {
-			blk.literals = append(blk.literals, src[nextEmit:s]...)
-		}
-		seq.offset = uint32(s-t) + 3
-		s += l
-		if debugSequences {
-			println("sequence", seq, "next s:", s)
-		}
-		blk.sequences = append(blk.sequences, seq)
-		nextEmit = s
-		if s >= sLimit {
-			break encodeLoop
-		}
-
-		// Index match start+1 (long) and start+2 (short)
-		index0 := s - l + 1
-		// Index match end-2 (long) and end-1 (short)
-		index1 := s - 2
-
-		cv0 := load6432(src, index0)
-		cv1 := load6432(src, index1)
-		te0 := tableEntry{offset: index0 + e.cur, val: uint32(cv0)}
-		te1 := tableEntry{offset: index1 + e.cur, val: uint32(cv1)}
-		longHash1 := hash8(cv0, dFastLongTableBits)
-		longHash2 := hash8(cv0, dFastLongTableBits)
-		e.longTable[longHash1] = te0
-		e.longTable[longHash2] = te1
-		e.markLongShardDirty(longHash1)
-		e.markLongShardDirty(longHash2)
-		cv0 >>= 8
-		cv1 >>= 8
-		te0.offset++
-		te1.offset++
-		te0.val = uint32(cv0)
-		te1.val = uint32(cv1)
-		hashVal1 := hash5(cv0, dFastShortTableBits)
-		hashVal2 := hash5(cv1, dFastShortTableBits)
-		e.table[hashVal1] = te0
-		e.markShardDirty(hashVal1)
-		e.table[hashVal2] = te1
-		e.markShardDirty(hashVal2)
-
-		cv = load6432(src, s)
-
-		if !canRepeat {
-			continue
-		}
-
-		// Check offset 2
-		for {
-			o2 := s - offset2
-			if load3232(src, o2) != uint32(cv) {
-				// Do regular search
-				break
-			}
-
-			// Store this, since we have it.
-			nextHashS := hash5(cv, dFastShortTableBits)
-			nextHashL := hash8(cv, dFastLongTableBits)
-
-			// We have at least 4 byte match.
-			// No need to check backwards. We come straight from a match
-			l := 4 + e.matchlen(s+4, o2+4, src)
-
-			entry := tableEntry{offset: s + e.cur, val: uint32(cv)}
-			e.longTable[nextHashL] = entry
-			e.markLongShardDirty(nextHashL)
-			e.table[nextHashS] = entry
-			e.markShardDirty(nextHashS)
-			seq.matchLen = uint32(l) - zstdMinMatch
-			seq.litLen = 0
-
-			// Since litlen is always 0, this is offset 1.
-			seq.offset = 1
-			s += l
-			nextEmit = s
-			if debugSequences {
-				println("sequence", seq, "next s:", s)
-			}
-			blk.sequences = append(blk.sequences, seq)
-
-			// Swap offset 1 and 2.
-			offset1, offset2 = offset2, offset1
-			if s >= sLimit {
-				// Finished
-				break encodeLoop
-			}
-			cv = load6432(src, s)
-		}
-	}
-
-	if int(nextEmit) < len(src) {
-		blk.literals = append(blk.literals, src[nextEmit:]...)
-		blk.extraLits = len(src) - int(nextEmit)
-	}
-	blk.recentOffsets[0] = uint32(offset1)
-	blk.recentOffsets[1] = uint32(offset2)
-	if debugEncoder {
-		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
-	}
-	// If we encoded more than 64K mark all dirty.
-	if len(src) > 64<<10 {
-		e.markAllShardsDirty()
-	}
-}
-
-// ResetDict will reset and set a dictionary if not nil
-func (e *doubleFastEncoder) Reset(d *dict, singleBlock bool) {
-	e.fastEncoder.Reset(d, singleBlock)
-	if d != nil {
-		panic("doubleFastEncoder: Reset with dict not supported")
-	}
-}
-
-// ResetDict will reset and set a dictionary if not nil
-func (e *doubleFastEncoderDict) Reset(d *dict, singleBlock bool) {
-	allDirty := e.allDirty
-	e.fastEncoderDict.Reset(d, singleBlock)
-	if d == nil {
-		return
-	}
-
-	// Init or copy dict table
-	if len(e.dictLongTable) != len(e.longTable) || d.id != e.lastDictID {
-		if len(e.dictLongTable) != len(e.longTable) {
-			e.dictLongTable = make([]tableEntry, len(e.longTable))
-		}
-		if len(d.content) >= 8 {
-			cv := load6432(d.content, 0)
-			e.dictLongTable[hash8(cv, dFastLongTableBits)] = tableEntry{
-				val:    uint32(cv),
-				offset: e.maxMatchOff,
-			}
-			end := int32(len(d.content)) - 8 + e.maxMatchOff
-			for i := e.maxMatchOff + 1; i < end; i++ {
-				cv = cv>>8 | (uint64(d.content[i-e.maxMatchOff+7]) << 56)
-				e.dictLongTable[hash8(cv, dFastLongTableBits)] = tableEntry{
-					val:    uint32(cv),
-					offset: i,
-				}
-			}
-		}
-		e.lastDictID = d.id
-		e.allDirty = true
-	}
-	// Reset table to initial state
-	e.cur = e.maxMatchOff
-
-	dirtyShardCnt := 0
-	if !allDirty {
-		for i := range e.longTableShardDirty {
-			if e.longTableShardDirty[i] {
-				dirtyShardCnt++
-			}
-		}
-	}
-
-	if allDirty || dirtyShardCnt > dLongTableShardCnt/2 {
-		copy(e.longTable[:], e.dictLongTable)
-		for i := range e.longTableShardDirty {
-			e.longTableShardDirty[i] = false
-		}
-		return
-	}
-	for i := range e.longTableShardDirty {
-		if !e.longTableShardDirty[i] {
-			continue
-		}
-
-		copy(e.longTable[i*dLongTableShardSize:(i+1)*dLongTableShardSize], e.dictLongTable[i*dLongTableShardSize:(i+1)*dLongTableShardSize])
-		e.longTableShardDirty[i] = false
-	}
-}
-
-func (e *doubleFastEncoderDict) markLongShardDirty(entryNum uint32) {
-	e.longTableShardDirty[entryNum/dLongTableShardSize] = true
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/enc_fast.go b/vendor/github.com/klauspost/compress/zstd/enc_fast.go
deleted file mode 100644
index 2246d286dc..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/enc_fast.go
+++ /dev/null
@@ -1,1018 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"fmt"
-	"math"
-	"math/bits"
-)
-
-const (
-	tableBits      = 15                               // Bits used in the table
-	tableSize      = 1 << tableBits                   // Size of the table
-	tableShardCnt  = 1 << (tableBits - dictShardBits) // Number of shards in the table
-	tableShardSize = tableSize / tableShardCnt        // Size of an individual shard
-	tableMask      = tableSize - 1                    // Mask for table indices. Redundant, but can eliminate bounds checks.
-	maxMatchLength = 131074
-)
-
-type tableEntry struct {
-	val    uint32
-	offset int32
-}
-
-type fastEncoder struct {
-	fastBase
-	table [tableSize]tableEntry
-}
-
-type fastEncoderDict struct {
-	fastEncoder
-	dictTable       []tableEntry
-	tableShardDirty [tableShardCnt]bool
-	allDirty        bool
-}
-
-// Encode mimmics functionality in zstd_fast.c
-func (e *fastEncoder) Encode(blk *blockEnc, src []byte) {
-	const (
-		inputMargin            = 8
-		minNonLiteralBlockSize = 1 + 1 + inputMargin
-	)
-
-	// Protect against e.cur wraparound.
-	for e.cur >= bufferReset {
-		if len(e.hist) == 0 {
-			for i := range e.table[:] {
-				e.table[i] = tableEntry{}
-			}
-			e.cur = e.maxMatchOff
-			break
-		}
-		// Shift down everything in the table that isn't already too far away.
-		minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff
-		for i := range e.table[:] {
-			v := e.table[i].offset
-			if v < minOff {
-				v = 0
-			} else {
-				v = v - e.cur + e.maxMatchOff
-			}
-			e.table[i].offset = v
-		}
-		e.cur = e.maxMatchOff
-		break
-	}
-
-	s := e.addBlock(src)
-	blk.size = len(src)
-	if len(src) < minNonLiteralBlockSize {
-		blk.extraLits = len(src)
-		blk.literals = blk.literals[:len(src)]
-		copy(blk.literals, src)
-		return
-	}
-
-	// Override src
-	src = e.hist
-	sLimit := int32(len(src)) - inputMargin
-	// stepSize is the number of bytes to skip on every main loop iteration.
-	// It should be >= 2.
-	const stepSize = 2
-
-	// TEMPLATE
-	const hashLog = tableBits
-	// seems global, but would be nice to tweak.
-	const kSearchStrength = 7
-
-	// nextEmit is where in src the next emitLiteral should start from.
-	nextEmit := s
-	cv := load6432(src, s)
-
-	// Relative offsets
-	offset1 := int32(blk.recentOffsets[0])
-	offset2 := int32(blk.recentOffsets[1])
-
-	addLiterals := func(s *seq, until int32) {
-		if until == nextEmit {
-			return
-		}
-		blk.literals = append(blk.literals, src[nextEmit:until]...)
-		s.litLen = uint32(until - nextEmit)
-	}
-	if debugEncoder {
-		println("recent offsets:", blk.recentOffsets)
-	}
-
-encodeLoop:
-	for {
-		// t will contain the match offset when we find one.
-		// When existing the search loop, we have already checked 4 bytes.
-		var t int32
-
-		// We will not use repeat offsets across blocks.
-		// By not using them for the first 3 matches
-		canRepeat := len(blk.sequences) > 2
-
-		for {
-			if debugAsserts && canRepeat && offset1 == 0 {
-				panic("offset0 was 0")
-			}
-
-			nextHash := hash6(cv, hashLog)
-			nextHash2 := hash6(cv>>8, hashLog)
-			candidate := e.table[nextHash]
-			candidate2 := e.table[nextHash2]
-			repIndex := s - offset1 + 2
-
-			e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)}
-			e.table[nextHash2] = tableEntry{offset: s + e.cur + 1, val: uint32(cv >> 8)}
-
-			if canRepeat && repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>16) {
-				// Consider history as well.
-				var seq seq
-				var length int32
-				// length = 4 + e.matchlen(s+6, repIndex+4, src)
-				{
-					a := src[s+6:]
-					b := src[repIndex+4:]
-					endI := len(a) & (math.MaxInt32 - 7)
-					length = int32(endI) + 4
-					for i := 0; i < endI; i += 8 {
-						if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-							length = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-							break
-						}
-					}
-				}
-
-				seq.matchLen = uint32(length - zstdMinMatch)
-
-				// We might be able to match backwards.
-				// Extend as long as we can.
-				start := s + 2
-				// We end the search early, so we don't risk 0 literals
-				// and have to do special offset treatment.
-				startLimit := nextEmit + 1
-
-				sMin := s - e.maxMatchOff
-				if sMin < 0 {
-					sMin = 0
-				}
-				for repIndex > sMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch {
-					repIndex--
-					start--
-					seq.matchLen++
-				}
-				addLiterals(&seq, start)
-
-				// rep 0
-				seq.offset = 1
-				if debugSequences {
-					println("repeat sequence", seq, "next s:", s)
-				}
-				blk.sequences = append(blk.sequences, seq)
-				s += length + 2
-				nextEmit = s
-				if s >= sLimit {
-					if debugEncoder {
-						println("repeat ended", s, length)
-
-					}
-					break encodeLoop
-				}
-				cv = load6432(src, s)
-				continue
-			}
-			coffset0 := s - (candidate.offset - e.cur)
-			coffset1 := s - (candidate2.offset - e.cur) + 1
-			if coffset0 < e.maxMatchOff && uint32(cv) == candidate.val {
-				// found a regular match
-				t = candidate.offset - e.cur
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				break
-			}
-
-			if coffset1 < e.maxMatchOff && uint32(cv>>8) == candidate2.val {
-				// found a regular match
-				t = candidate2.offset - e.cur
-				s++
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				if debugAsserts && t < 0 {
-					panic("t<0")
-				}
-				break
-			}
-			s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1))
-			if s >= sLimit {
-				break encodeLoop
-			}
-			cv = load6432(src, s)
-		}
-		// A 4-byte match has been found. We'll later see if more than 4 bytes.
-		offset2 = offset1
-		offset1 = s - t
-
-		if debugAsserts && s <= t {
-			panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-		}
-
-		if debugAsserts && canRepeat && int(offset1) > len(src) {
-			panic("invalid offset")
-		}
-
-		// Extend the 4-byte match as long as possible.
-		//l := e.matchlen(s+4, t+4, src) + 4
-		var l int32
-		{
-			a := src[s+4:]
-			b := src[t+4:]
-			endI := len(a) & (math.MaxInt32 - 7)
-			l = int32(endI) + 4
-			for i := 0; i < endI; i += 8 {
-				if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-					l = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-					break
-				}
-			}
-		}
-
-		// Extend backwards
-		tMin := s - e.maxMatchOff
-		if tMin < 0 {
-			tMin = 0
-		}
-		for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength {
-			s--
-			t--
-			l++
-		}
-
-		// Write our sequence.
-		var seq seq
-		seq.litLen = uint32(s - nextEmit)
-		seq.matchLen = uint32(l - zstdMinMatch)
-		if seq.litLen > 0 {
-			blk.literals = append(blk.literals, src[nextEmit:s]...)
-		}
-		// Don't use repeat offsets
-		seq.offset = uint32(s-t) + 3
-		s += l
-		if debugSequences {
-			println("sequence", seq, "next s:", s)
-		}
-		blk.sequences = append(blk.sequences, seq)
-		nextEmit = s
-		if s >= sLimit {
-			break encodeLoop
-		}
-		cv = load6432(src, s)
-
-		// Check offset 2
-		if o2 := s - offset2; canRepeat && load3232(src, o2) == uint32(cv) {
-			// We have at least 4 byte match.
-			// No need to check backwards. We come straight from a match
-			//l := 4 + e.matchlen(s+4, o2+4, src)
-			var l int32
-			{
-				a := src[s+4:]
-				b := src[o2+4:]
-				endI := len(a) & (math.MaxInt32 - 7)
-				l = int32(endI) + 4
-				for i := 0; i < endI; i += 8 {
-					if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-						l = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-						break
-					}
-				}
-			}
-
-			// Store this, since we have it.
-			nextHash := hash6(cv, hashLog)
-			e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)}
-			seq.matchLen = uint32(l) - zstdMinMatch
-			seq.litLen = 0
-			// Since litlen is always 0, this is offset 1.
-			seq.offset = 1
-			s += l
-			nextEmit = s
-			if debugSequences {
-				println("sequence", seq, "next s:", s)
-			}
-			blk.sequences = append(blk.sequences, seq)
-
-			// Swap offset 1 and 2.
-			offset1, offset2 = offset2, offset1
-			if s >= sLimit {
-				break encodeLoop
-			}
-			// Prepare next loop.
-			cv = load6432(src, s)
-		}
-	}
-
-	if int(nextEmit) < len(src) {
-		blk.literals = append(blk.literals, src[nextEmit:]...)
-		blk.extraLits = len(src) - int(nextEmit)
-	}
-	blk.recentOffsets[0] = uint32(offset1)
-	blk.recentOffsets[1] = uint32(offset2)
-	if debugEncoder {
-		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
-	}
-}
-
-// EncodeNoHist will encode a block with no history and no following blocks.
-// Most notable difference is that src will not be copied for history and
-// we do not need to check for max match length.
-func (e *fastEncoder) EncodeNoHist(blk *blockEnc, src []byte) {
-	const (
-		inputMargin            = 8
-		minNonLiteralBlockSize = 1 + 1 + inputMargin
-	)
-	if debugEncoder {
-		if len(src) > maxBlockSize {
-			panic("src too big")
-		}
-	}
-
-	// Protect against e.cur wraparound.
-	if e.cur >= bufferReset {
-		for i := range e.table[:] {
-			e.table[i] = tableEntry{}
-		}
-		e.cur = e.maxMatchOff
-	}
-
-	s := int32(0)
-	blk.size = len(src)
-	if len(src) < minNonLiteralBlockSize {
-		blk.extraLits = len(src)
-		blk.literals = blk.literals[:len(src)]
-		copy(blk.literals, src)
-		return
-	}
-
-	sLimit := int32(len(src)) - inputMargin
-	// stepSize is the number of bytes to skip on every main loop iteration.
-	// It should be >= 2.
-	const stepSize = 2
-
-	// TEMPLATE
-	const hashLog = tableBits
-	// seems global, but would be nice to tweak.
-	const kSearchStrength = 8
-
-	// nextEmit is where in src the next emitLiteral should start from.
-	nextEmit := s
-	cv := load6432(src, s)
-
-	// Relative offsets
-	offset1 := int32(blk.recentOffsets[0])
-	offset2 := int32(blk.recentOffsets[1])
-
-	addLiterals := func(s *seq, until int32) {
-		if until == nextEmit {
-			return
-		}
-		blk.literals = append(blk.literals, src[nextEmit:until]...)
-		s.litLen = uint32(until - nextEmit)
-	}
-	if debugEncoder {
-		println("recent offsets:", blk.recentOffsets)
-	}
-
-encodeLoop:
-	for {
-		// t will contain the match offset when we find one.
-		// When existing the search loop, we have already checked 4 bytes.
-		var t int32
-
-		// We will not use repeat offsets across blocks.
-		// By not using them for the first 3 matches
-
-		for {
-			nextHash := hash6(cv, hashLog)
-			nextHash2 := hash6(cv>>8, hashLog)
-			candidate := e.table[nextHash]
-			candidate2 := e.table[nextHash2]
-			repIndex := s - offset1 + 2
-
-			e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)}
-			e.table[nextHash2] = tableEntry{offset: s + e.cur + 1, val: uint32(cv >> 8)}
-
-			if len(blk.sequences) > 2 && load3232(src, repIndex) == uint32(cv>>16) {
-				// Consider history as well.
-				var seq seq
-				// length := 4 + e.matchlen(s+6, repIndex+4, src)
-				// length := 4 + int32(matchLen(src[s+6:], src[repIndex+4:]))
-				var length int32
-				{
-					a := src[s+6:]
-					b := src[repIndex+4:]
-					endI := len(a) & (math.MaxInt32 - 7)
-					length = int32(endI) + 4
-					for i := 0; i < endI; i += 8 {
-						if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-							length = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-							break
-						}
-					}
-				}
-
-				seq.matchLen = uint32(length - zstdMinMatch)
-
-				// We might be able to match backwards.
-				// Extend as long as we can.
-				start := s + 2
-				// We end the search early, so we don't risk 0 literals
-				// and have to do special offset treatment.
-				startLimit := nextEmit + 1
-
-				sMin := s - e.maxMatchOff
-				if sMin < 0 {
-					sMin = 0
-				}
-				for repIndex > sMin && start > startLimit && src[repIndex-1] == src[start-1] {
-					repIndex--
-					start--
-					seq.matchLen++
-				}
-				addLiterals(&seq, start)
-
-				// rep 0
-				seq.offset = 1
-				if debugSequences {
-					println("repeat sequence", seq, "next s:", s)
-				}
-				blk.sequences = append(blk.sequences, seq)
-				s += length + 2
-				nextEmit = s
-				if s >= sLimit {
-					if debugEncoder {
-						println("repeat ended", s, length)
-
-					}
-					break encodeLoop
-				}
-				cv = load6432(src, s)
-				continue
-			}
-			coffset0 := s - (candidate.offset - e.cur)
-			coffset1 := s - (candidate2.offset - e.cur) + 1
-			if coffset0 < e.maxMatchOff && uint32(cv) == candidate.val {
-				// found a regular match
-				t = candidate.offset - e.cur
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				if debugAsserts && t < 0 {
-					panic(fmt.Sprintf("t (%d) < 0, candidate.offset: %d, e.cur: %d, coffset0: %d, e.maxMatchOff: %d", t, candidate.offset, e.cur, coffset0, e.maxMatchOff))
-				}
-				break
-			}
-
-			if coffset1 < e.maxMatchOff && uint32(cv>>8) == candidate2.val {
-				// found a regular match
-				t = candidate2.offset - e.cur
-				s++
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				if debugAsserts && t < 0 {
-					panic("t<0")
-				}
-				break
-			}
-			s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1))
-			if s >= sLimit {
-				break encodeLoop
-			}
-			cv = load6432(src, s)
-		}
-		// A 4-byte match has been found. We'll later see if more than 4 bytes.
-		offset2 = offset1
-		offset1 = s - t
-
-		if debugAsserts && s <= t {
-			panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-		}
-
-		if debugAsserts && t < 0 {
-			panic(fmt.Sprintf("t (%d) < 0 ", t))
-		}
-		// Extend the 4-byte match as long as possible.
-		//l := e.matchlenNoHist(s+4, t+4, src) + 4
-		// l := int32(matchLen(src[s+4:], src[t+4:])) + 4
-		var l int32
-		{
-			a := src[s+4:]
-			b := src[t+4:]
-			endI := len(a) & (math.MaxInt32 - 7)
-			l = int32(endI) + 4
-			for i := 0; i < endI; i += 8 {
-				if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-					l = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-					break
-				}
-			}
-		}
-
-		// Extend backwards
-		tMin := s - e.maxMatchOff
-		if tMin < 0 {
-			tMin = 0
-		}
-		for t > tMin && s > nextEmit && src[t-1] == src[s-1] {
-			s--
-			t--
-			l++
-		}
-
-		// Write our sequence.
-		var seq seq
-		seq.litLen = uint32(s - nextEmit)
-		seq.matchLen = uint32(l - zstdMinMatch)
-		if seq.litLen > 0 {
-			blk.literals = append(blk.literals, src[nextEmit:s]...)
-		}
-		// Don't use repeat offsets
-		seq.offset = uint32(s-t) + 3
-		s += l
-		if debugSequences {
-			println("sequence", seq, "next s:", s)
-		}
-		blk.sequences = append(blk.sequences, seq)
-		nextEmit = s
-		if s >= sLimit {
-			break encodeLoop
-		}
-		cv = load6432(src, s)
-
-		// Check offset 2
-		if o2 := s - offset2; len(blk.sequences) > 2 && load3232(src, o2) == uint32(cv) {
-			// We have at least 4 byte match.
-			// No need to check backwards. We come straight from a match
-			//l := 4 + e.matchlenNoHist(s+4, o2+4, src)
-			// l := 4 + int32(matchLen(src[s+4:], src[o2+4:]))
-			var l int32
-			{
-				a := src[s+4:]
-				b := src[o2+4:]
-				endI := len(a) & (math.MaxInt32 - 7)
-				l = int32(endI) + 4
-				for i := 0; i < endI; i += 8 {
-					if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-						l = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-						break
-					}
-				}
-			}
-
-			// Store this, since we have it.
-			nextHash := hash6(cv, hashLog)
-			e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)}
-			seq.matchLen = uint32(l) - zstdMinMatch
-			seq.litLen = 0
-			// Since litlen is always 0, this is offset 1.
-			seq.offset = 1
-			s += l
-			nextEmit = s
-			if debugSequences {
-				println("sequence", seq, "next s:", s)
-			}
-			blk.sequences = append(blk.sequences, seq)
-
-			// Swap offset 1 and 2.
-			offset1, offset2 = offset2, offset1
-			if s >= sLimit {
-				break encodeLoop
-			}
-			// Prepare next loop.
-			cv = load6432(src, s)
-		}
-	}
-
-	if int(nextEmit) < len(src) {
-		blk.literals = append(blk.literals, src[nextEmit:]...)
-		blk.extraLits = len(src) - int(nextEmit)
-	}
-	if debugEncoder {
-		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
-	}
-	// We do not store history, so we must offset e.cur to avoid false matches for next user.
-	if e.cur < bufferReset {
-		e.cur += int32(len(src))
-	}
-}
-
-// Encode will encode the content, with a dictionary if initialized for it.
-func (e *fastEncoderDict) Encode(blk *blockEnc, src []byte) {
-	const (
-		inputMargin            = 8
-		minNonLiteralBlockSize = 1 + 1 + inputMargin
-	)
-	if e.allDirty || len(src) > 32<<10 {
-		e.fastEncoder.Encode(blk, src)
-		e.allDirty = true
-		return
-	}
-	// Protect against e.cur wraparound.
-	for e.cur >= bufferReset {
-		if len(e.hist) == 0 {
-			for i := range e.table[:] {
-				e.table[i] = tableEntry{}
-			}
-			e.cur = e.maxMatchOff
-			break
-		}
-		// Shift down everything in the table that isn't already too far away.
-		minOff := e.cur + int32(len(e.hist)) - e.maxMatchOff
-		for i := range e.table[:] {
-			v := e.table[i].offset
-			if v < minOff {
-				v = 0
-			} else {
-				v = v - e.cur + e.maxMatchOff
-			}
-			e.table[i].offset = v
-		}
-		e.cur = e.maxMatchOff
-		break
-	}
-
-	s := e.addBlock(src)
-	blk.size = len(src)
-	if len(src) < minNonLiteralBlockSize {
-		blk.extraLits = len(src)
-		blk.literals = blk.literals[:len(src)]
-		copy(blk.literals, src)
-		return
-	}
-
-	// Override src
-	src = e.hist
-	sLimit := int32(len(src)) - inputMargin
-	// stepSize is the number of bytes to skip on every main loop iteration.
-	// It should be >= 2.
-	const stepSize = 2
-
-	// TEMPLATE
-	const hashLog = tableBits
-	// seems global, but would be nice to tweak.
-	const kSearchStrength = 7
-
-	// nextEmit is where in src the next emitLiteral should start from.
-	nextEmit := s
-	cv := load6432(src, s)
-
-	// Relative offsets
-	offset1 := int32(blk.recentOffsets[0])
-	offset2 := int32(blk.recentOffsets[1])
-
-	addLiterals := func(s *seq, until int32) {
-		if until == nextEmit {
-			return
-		}
-		blk.literals = append(blk.literals, src[nextEmit:until]...)
-		s.litLen = uint32(until - nextEmit)
-	}
-	if debugEncoder {
-		println("recent offsets:", blk.recentOffsets)
-	}
-
-encodeLoop:
-	for {
-		// t will contain the match offset when we find one.
-		// When existing the search loop, we have already checked 4 bytes.
-		var t int32
-
-		// We will not use repeat offsets across blocks.
-		// By not using them for the first 3 matches
-		canRepeat := len(blk.sequences) > 2
-
-		for {
-			if debugAsserts && canRepeat && offset1 == 0 {
-				panic("offset0 was 0")
-			}
-
-			nextHash := hash6(cv, hashLog)
-			nextHash2 := hash6(cv>>8, hashLog)
-			candidate := e.table[nextHash]
-			candidate2 := e.table[nextHash2]
-			repIndex := s - offset1 + 2
-
-			e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)}
-			e.markShardDirty(nextHash)
-			e.table[nextHash2] = tableEntry{offset: s + e.cur + 1, val: uint32(cv >> 8)}
-			e.markShardDirty(nextHash2)
-
-			if canRepeat && repIndex >= 0 && load3232(src, repIndex) == uint32(cv>>16) {
-				// Consider history as well.
-				var seq seq
-				var length int32
-				// length = 4 + e.matchlen(s+6, repIndex+4, src)
-				{
-					a := src[s+6:]
-					b := src[repIndex+4:]
-					endI := len(a) & (math.MaxInt32 - 7)
-					length = int32(endI) + 4
-					for i := 0; i < endI; i += 8 {
-						if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-							length = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-							break
-						}
-					}
-				}
-
-				seq.matchLen = uint32(length - zstdMinMatch)
-
-				// We might be able to match backwards.
-				// Extend as long as we can.
-				start := s + 2
-				// We end the search early, so we don't risk 0 literals
-				// and have to do special offset treatment.
-				startLimit := nextEmit + 1
-
-				sMin := s - e.maxMatchOff
-				if sMin < 0 {
-					sMin = 0
-				}
-				for repIndex > sMin && start > startLimit && src[repIndex-1] == src[start-1] && seq.matchLen < maxMatchLength-zstdMinMatch {
-					repIndex--
-					start--
-					seq.matchLen++
-				}
-				addLiterals(&seq, start)
-
-				// rep 0
-				seq.offset = 1
-				if debugSequences {
-					println("repeat sequence", seq, "next s:", s)
-				}
-				blk.sequences = append(blk.sequences, seq)
-				s += length + 2
-				nextEmit = s
-				if s >= sLimit {
-					if debugEncoder {
-						println("repeat ended", s, length)
-
-					}
-					break encodeLoop
-				}
-				cv = load6432(src, s)
-				continue
-			}
-			coffset0 := s - (candidate.offset - e.cur)
-			coffset1 := s - (candidate2.offset - e.cur) + 1
-			if coffset0 < e.maxMatchOff && uint32(cv) == candidate.val {
-				// found a regular match
-				t = candidate.offset - e.cur
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				break
-			}
-
-			if coffset1 < e.maxMatchOff && uint32(cv>>8) == candidate2.val {
-				// found a regular match
-				t = candidate2.offset - e.cur
-				s++
-				if debugAsserts && s <= t {
-					panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-				}
-				if debugAsserts && s-t > e.maxMatchOff {
-					panic("s - t >e.maxMatchOff")
-				}
-				if debugAsserts && t < 0 {
-					panic("t<0")
-				}
-				break
-			}
-			s += stepSize + ((s - nextEmit) >> (kSearchStrength - 1))
-			if s >= sLimit {
-				break encodeLoop
-			}
-			cv = load6432(src, s)
-		}
-		// A 4-byte match has been found. We'll later see if more than 4 bytes.
-		offset2 = offset1
-		offset1 = s - t
-
-		if debugAsserts && s <= t {
-			panic(fmt.Sprintf("s (%d) <= t (%d)", s, t))
-		}
-
-		if debugAsserts && canRepeat && int(offset1) > len(src) {
-			panic("invalid offset")
-		}
-
-		// Extend the 4-byte match as long as possible.
-		//l := e.matchlen(s+4, t+4, src) + 4
-		var l int32
-		{
-			a := src[s+4:]
-			b := src[t+4:]
-			endI := len(a) & (math.MaxInt32 - 7)
-			l = int32(endI) + 4
-			for i := 0; i < endI; i += 8 {
-				if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-					l = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-					break
-				}
-			}
-		}
-
-		// Extend backwards
-		tMin := s - e.maxMatchOff
-		if tMin < 0 {
-			tMin = 0
-		}
-		for t > tMin && s > nextEmit && src[t-1] == src[s-1] && l < maxMatchLength {
-			s--
-			t--
-			l++
-		}
-
-		// Write our sequence.
-		var seq seq
-		seq.litLen = uint32(s - nextEmit)
-		seq.matchLen = uint32(l - zstdMinMatch)
-		if seq.litLen > 0 {
-			blk.literals = append(blk.literals, src[nextEmit:s]...)
-		}
-		// Don't use repeat offsets
-		seq.offset = uint32(s-t) + 3
-		s += l
-		if debugSequences {
-			println("sequence", seq, "next s:", s)
-		}
-		blk.sequences = append(blk.sequences, seq)
-		nextEmit = s
-		if s >= sLimit {
-			break encodeLoop
-		}
-		cv = load6432(src, s)
-
-		// Check offset 2
-		if o2 := s - offset2; canRepeat && load3232(src, o2) == uint32(cv) {
-			// We have at least 4 byte match.
-			// No need to check backwards. We come straight from a match
-			//l := 4 + e.matchlen(s+4, o2+4, src)
-			var l int32
-			{
-				a := src[s+4:]
-				b := src[o2+4:]
-				endI := len(a) & (math.MaxInt32 - 7)
-				l = int32(endI) + 4
-				for i := 0; i < endI; i += 8 {
-					if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-						l = int32(i+bits.TrailingZeros64(diff)>>3) + 4
-						break
-					}
-				}
-			}
-
-			// Store this, since we have it.
-			nextHash := hash6(cv, hashLog)
-			e.table[nextHash] = tableEntry{offset: s + e.cur, val: uint32(cv)}
-			e.markShardDirty(nextHash)
-			seq.matchLen = uint32(l) - zstdMinMatch
-			seq.litLen = 0
-			// Since litlen is always 0, this is offset 1.
-			seq.offset = 1
-			s += l
-			nextEmit = s
-			if debugSequences {
-				println("sequence", seq, "next s:", s)
-			}
-			blk.sequences = append(blk.sequences, seq)
-
-			// Swap offset 1 and 2.
-			offset1, offset2 = offset2, offset1
-			if s >= sLimit {
-				break encodeLoop
-			}
-			// Prepare next loop.
-			cv = load6432(src, s)
-		}
-	}
-
-	if int(nextEmit) < len(src) {
-		blk.literals = append(blk.literals, src[nextEmit:]...)
-		blk.extraLits = len(src) - int(nextEmit)
-	}
-	blk.recentOffsets[0] = uint32(offset1)
-	blk.recentOffsets[1] = uint32(offset2)
-	if debugEncoder {
-		println("returning, recent offsets:", blk.recentOffsets, "extra literals:", blk.extraLits)
-	}
-}
-
-// ResetDict will reset and set a dictionary if not nil
-func (e *fastEncoder) Reset(d *dict, singleBlock bool) {
-	e.resetBase(d, singleBlock)
-	if d != nil {
-		panic("fastEncoder: Reset with dict")
-	}
-}
-
-// ResetDict will reset and set a dictionary if not nil
-func (e *fastEncoderDict) Reset(d *dict, singleBlock bool) {
-	e.resetBase(d, singleBlock)
-	if d == nil {
-		return
-	}
-
-	// Init or copy dict table
-	if len(e.dictTable) != len(e.table) || d.id != e.lastDictID {
-		if len(e.dictTable) != len(e.table) {
-			e.dictTable = make([]tableEntry, len(e.table))
-		}
-		if true {
-			end := e.maxMatchOff + int32(len(d.content)) - 8
-			for i := e.maxMatchOff; i < end; i += 3 {
-				const hashLog = tableBits
-
-				cv := load6432(d.content, i-e.maxMatchOff)
-				nextHash := hash6(cv, hashLog)      // 0 -> 5
-				nextHash1 := hash6(cv>>8, hashLog)  // 1 -> 6
-				nextHash2 := hash6(cv>>16, hashLog) // 2 -> 7
-				e.dictTable[nextHash] = tableEntry{
-					val:    uint32(cv),
-					offset: i,
-				}
-				e.dictTable[nextHash1] = tableEntry{
-					val:    uint32(cv >> 8),
-					offset: i + 1,
-				}
-				e.dictTable[nextHash2] = tableEntry{
-					val:    uint32(cv >> 16),
-					offset: i + 2,
-				}
-			}
-		}
-		e.lastDictID = d.id
-		e.allDirty = true
-	}
-
-	e.cur = e.maxMatchOff
-	dirtyShardCnt := 0
-	if !e.allDirty {
-		for i := range e.tableShardDirty {
-			if e.tableShardDirty[i] {
-				dirtyShardCnt++
-			}
-		}
-	}
-
-	const shardCnt = tableShardCnt
-	const shardSize = tableShardSize
-	if e.allDirty || dirtyShardCnt > shardCnt*4/6 {
-		copy(e.table[:], e.dictTable)
-		for i := range e.tableShardDirty {
-			e.tableShardDirty[i] = false
-		}
-		e.allDirty = false
-		return
-	}
-	for i := range e.tableShardDirty {
-		if !e.tableShardDirty[i] {
-			continue
-		}
-
-		copy(e.table[i*shardSize:(i+1)*shardSize], e.dictTable[i*shardSize:(i+1)*shardSize])
-		e.tableShardDirty[i] = false
-	}
-	e.allDirty = false
-}
-
-func (e *fastEncoderDict) markAllShardsDirty() {
-	e.allDirty = true
-}
-
-func (e *fastEncoderDict) markShardDirty(entryNum uint32) {
-	e.tableShardDirty[entryNum/tableShardSize] = true
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/encoder.go b/vendor/github.com/klauspost/compress/zstd/encoder.go
deleted file mode 100644
index ea85548fc9..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/encoder.go
+++ /dev/null
@@ -1,576 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"crypto/rand"
-	"fmt"
-	"io"
-	rdebug "runtime/debug"
-	"sync"
-
-	"github.com/klauspost/compress/zstd/internal/xxhash"
-)
-
-// Encoder provides encoding to Zstandard.
-// An Encoder can be used for either compressing a stream via the
-// io.WriteCloser interface supported by the Encoder or as multiple independent
-// tasks via the EncodeAll function.
-// Smaller encodes are encouraged to use the EncodeAll function.
-// Use NewWriter to create a new instance.
-type Encoder struct {
-	o        encoderOptions
-	encoders chan encoder
-	state    encoderState
-	init     sync.Once
-}
-
-type encoder interface {
-	Encode(blk *blockEnc, src []byte)
-	EncodeNoHist(blk *blockEnc, src []byte)
-	Block() *blockEnc
-	CRC() *xxhash.Digest
-	AppendCRC([]byte) []byte
-	WindowSize(size int) int32
-	UseBlock(*blockEnc)
-	Reset(d *dict, singleBlock bool)
-}
-
-type encoderState struct {
-	w                io.Writer
-	filling          []byte
-	current          []byte
-	previous         []byte
-	encoder          encoder
-	writing          *blockEnc
-	err              error
-	writeErr         error
-	nWritten         int64
-	headerWritten    bool
-	eofWritten       bool
-	fullFrameWritten bool
-
-	// This waitgroup indicates an encode is running.
-	wg sync.WaitGroup
-	// This waitgroup indicates we have a block encoding/writing.
-	wWg sync.WaitGroup
-}
-
-// NewWriter will create a new Zstandard encoder.
-// If the encoder will be used for encoding blocks a nil writer can be used.
-func NewWriter(w io.Writer, opts ...EOption) (*Encoder, error) {
-	initPredefined()
-	var e Encoder
-	e.o.setDefault()
-	for _, o := range opts {
-		err := o(&e.o)
-		if err != nil {
-			return nil, err
-		}
-	}
-	if w != nil {
-		e.Reset(w)
-	}
-	return &e, nil
-}
-
-func (e *Encoder) initialize() {
-	if e.o.concurrent == 0 {
-		e.o.setDefault()
-	}
-	e.encoders = make(chan encoder, e.o.concurrent)
-	for i := 0; i < e.o.concurrent; i++ {
-		enc := e.o.encoder()
-		e.encoders <- enc
-	}
-}
-
-// Reset will re-initialize the writer and new writes will encode to the supplied writer
-// as a new, independent stream.
-func (e *Encoder) Reset(w io.Writer) {
-	s := &e.state
-	s.wg.Wait()
-	s.wWg.Wait()
-	if cap(s.filling) == 0 {
-		s.filling = make([]byte, 0, e.o.blockSize)
-	}
-	if cap(s.current) == 0 {
-		s.current = make([]byte, 0, e.o.blockSize)
-	}
-	if cap(s.previous) == 0 {
-		s.previous = make([]byte, 0, e.o.blockSize)
-	}
-	if s.encoder == nil {
-		s.encoder = e.o.encoder()
-	}
-	if s.writing == nil {
-		s.writing = &blockEnc{lowMem: e.o.lowMem}
-		s.writing.init()
-	}
-	s.writing.initNewEncode()
-	s.filling = s.filling[:0]
-	s.current = s.current[:0]
-	s.previous = s.previous[:0]
-	s.encoder.Reset(e.o.dict, false)
-	s.headerWritten = false
-	s.eofWritten = false
-	s.fullFrameWritten = false
-	s.w = w
-	s.err = nil
-	s.nWritten = 0
-	s.writeErr = nil
-}
-
-// Write data to the encoder.
-// Input data will be buffered and as the buffer fills up
-// content will be compressed and written to the output.
-// When done writing, use Close to flush the remaining output
-// and write CRC if requested.
-func (e *Encoder) Write(p []byte) (n int, err error) {
-	s := &e.state
-	for len(p) > 0 {
-		if len(p)+len(s.filling) < e.o.blockSize {
-			if e.o.crc {
-				_, _ = s.encoder.CRC().Write(p)
-			}
-			s.filling = append(s.filling, p...)
-			return n + len(p), nil
-		}
-		add := p
-		if len(p)+len(s.filling) > e.o.blockSize {
-			add = add[:e.o.blockSize-len(s.filling)]
-		}
-		if e.o.crc {
-			_, _ = s.encoder.CRC().Write(add)
-		}
-		s.filling = append(s.filling, add...)
-		p = p[len(add):]
-		n += len(add)
-		if len(s.filling) < e.o.blockSize {
-			return n, nil
-		}
-		err := e.nextBlock(false)
-		if err != nil {
-			return n, err
-		}
-		if debugAsserts && len(s.filling) > 0 {
-			panic(len(s.filling))
-		}
-	}
-	return n, nil
-}
-
-// nextBlock will synchronize and start compressing input in e.state.filling.
-// If an error has occurred during encoding it will be returned.
-func (e *Encoder) nextBlock(final bool) error {
-	s := &e.state
-	// Wait for current block.
-	s.wg.Wait()
-	if s.err != nil {
-		return s.err
-	}
-	if len(s.filling) > e.o.blockSize {
-		return fmt.Errorf("block > maxStoreBlockSize")
-	}
-	if !s.headerWritten {
-		// If we have a single block encode, do a sync compression.
-		if final && len(s.filling) == 0 && !e.o.fullZero {
-			s.headerWritten = true
-			s.fullFrameWritten = true
-			s.eofWritten = true
-			return nil
-		}
-		if final && len(s.filling) > 0 {
-			s.current = e.EncodeAll(s.filling, s.current[:0])
-			var n2 int
-			n2, s.err = s.w.Write(s.current)
-			if s.err != nil {
-				return s.err
-			}
-			s.nWritten += int64(n2)
-			s.current = s.current[:0]
-			s.filling = s.filling[:0]
-			s.headerWritten = true
-			s.fullFrameWritten = true
-			s.eofWritten = true
-			return nil
-		}
-
-		var tmp [maxHeaderSize]byte
-		fh := frameHeader{
-			ContentSize:   0,
-			WindowSize:    uint32(s.encoder.WindowSize(0)),
-			SingleSegment: false,
-			Checksum:      e.o.crc,
-			DictID:        e.o.dict.ID(),
-		}
-
-		dst, err := fh.appendTo(tmp[:0])
-		if err != nil {
-			return err
-		}
-		s.headerWritten = true
-		s.wWg.Wait()
-		var n2 int
-		n2, s.err = s.w.Write(dst)
-		if s.err != nil {
-			return s.err
-		}
-		s.nWritten += int64(n2)
-	}
-	if s.eofWritten {
-		// Ensure we only write it once.
-		final = false
-	}
-
-	if len(s.filling) == 0 {
-		// Final block, but no data.
-		if final {
-			enc := s.encoder
-			blk := enc.Block()
-			blk.reset(nil)
-			blk.last = true
-			blk.encodeRaw(nil)
-			s.wWg.Wait()
-			_, s.err = s.w.Write(blk.output)
-			s.nWritten += int64(len(blk.output))
-			s.eofWritten = true
-		}
-		return s.err
-	}
-
-	// Move blocks forward.
-	s.filling, s.current, s.previous = s.previous[:0], s.filling, s.current
-	s.wg.Add(1)
-	go func(src []byte) {
-		if debugEncoder {
-			println("Adding block,", len(src), "bytes, final:", final)
-		}
-		defer func() {
-			if r := recover(); r != nil {
-				s.err = fmt.Errorf("panic while encoding: %v", r)
-				rdebug.PrintStack()
-			}
-			s.wg.Done()
-		}()
-		enc := s.encoder
-		blk := enc.Block()
-		enc.Encode(blk, src)
-		blk.last = final
-		if final {
-			s.eofWritten = true
-		}
-		// Wait for pending writes.
-		s.wWg.Wait()
-		if s.writeErr != nil {
-			s.err = s.writeErr
-			return
-		}
-		// Transfer encoders from previous write block.
-		blk.swapEncoders(s.writing)
-		// Transfer recent offsets to next.
-		enc.UseBlock(s.writing)
-		s.writing = blk
-		s.wWg.Add(1)
-		go func() {
-			defer func() {
-				if r := recover(); r != nil {
-					s.writeErr = fmt.Errorf("panic while encoding/writing: %v", r)
-					rdebug.PrintStack()
-				}
-				s.wWg.Done()
-			}()
-			err := errIncompressible
-			// If we got the exact same number of literals as input,
-			// assume the literals cannot be compressed.
-			if len(src) != len(blk.literals) || len(src) != e.o.blockSize {
-				err = blk.encode(src, e.o.noEntropy, !e.o.allLitEntropy)
-			}
-			switch err {
-			case errIncompressible:
-				if debugEncoder {
-					println("Storing incompressible block as raw")
-				}
-				blk.encodeRaw(src)
-				// In fast mode, we do not transfer offsets, so we don't have to deal with changing the.
-			case nil:
-			default:
-				s.writeErr = err
-				return
-			}
-			_, s.writeErr = s.w.Write(blk.output)
-			s.nWritten += int64(len(blk.output))
-		}()
-	}(s.current)
-	return nil
-}
-
-// ReadFrom reads data from r until EOF or error.
-// The return value n is the number of bytes read.
-// Any error except io.EOF encountered during the read is also returned.
-//
-// The Copy function uses ReaderFrom if available.
-func (e *Encoder) ReadFrom(r io.Reader) (n int64, err error) {
-	if debugEncoder {
-		println("Using ReadFrom")
-	}
-
-	// Flush any current writes.
-	if len(e.state.filling) > 0 {
-		if err := e.nextBlock(false); err != nil {
-			return 0, err
-		}
-	}
-	e.state.filling = e.state.filling[:e.o.blockSize]
-	src := e.state.filling
-	for {
-		n2, err := r.Read(src)
-		if e.o.crc {
-			_, _ = e.state.encoder.CRC().Write(src[:n2])
-		}
-		// src is now the unfilled part...
-		src = src[n2:]
-		n += int64(n2)
-		switch err {
-		case io.EOF:
-			e.state.filling = e.state.filling[:len(e.state.filling)-len(src)]
-			if debugEncoder {
-				println("ReadFrom: got EOF final block:", len(e.state.filling))
-			}
-			return n, nil
-		case nil:
-		default:
-			if debugEncoder {
-				println("ReadFrom: got error:", err)
-			}
-			e.state.err = err
-			return n, err
-		}
-		if len(src) > 0 {
-			if debugEncoder {
-				println("ReadFrom: got space left in source:", len(src))
-			}
-			continue
-		}
-		err = e.nextBlock(false)
-		if err != nil {
-			return n, err
-		}
-		e.state.filling = e.state.filling[:e.o.blockSize]
-		src = e.state.filling
-	}
-}
-
-// Flush will send the currently written data to output
-// and block until everything has been written.
-// This should only be used on rare occasions where pushing the currently queued data is critical.
-func (e *Encoder) Flush() error {
-	s := &e.state
-	if len(s.filling) > 0 {
-		err := e.nextBlock(false)
-		if err != nil {
-			return err
-		}
-	}
-	s.wg.Wait()
-	s.wWg.Wait()
-	if s.err != nil {
-		return s.err
-	}
-	return s.writeErr
-}
-
-// Close will flush the final output and close the stream.
-// The function will block until everything has been written.
-// The Encoder can still be re-used after calling this.
-func (e *Encoder) Close() error {
-	s := &e.state
-	if s.encoder == nil {
-		return nil
-	}
-	err := e.nextBlock(true)
-	if err != nil {
-		return err
-	}
-	if e.state.fullFrameWritten {
-		return s.err
-	}
-	s.wg.Wait()
-	s.wWg.Wait()
-
-	if s.err != nil {
-		return s.err
-	}
-	if s.writeErr != nil {
-		return s.writeErr
-	}
-
-	// Write CRC
-	if e.o.crc && s.err == nil {
-		// heap alloc.
-		var tmp [4]byte
-		_, s.err = s.w.Write(s.encoder.AppendCRC(tmp[:0]))
-		s.nWritten += 4
-	}
-
-	// Add padding with content from crypto/rand.Reader
-	if s.err == nil && e.o.pad > 0 {
-		add := calcSkippableFrame(s.nWritten, int64(e.o.pad))
-		frame, err := skippableFrame(s.filling[:0], add, rand.Reader)
-		if err != nil {
-			return err
-		}
-		_, s.err = s.w.Write(frame)
-	}
-	return s.err
-}
-
-// EncodeAll will encode all input in src and append it to dst.
-// This function can be called concurrently, but each call will only run on a single goroutine.
-// If empty input is given, nothing is returned, unless WithZeroFrames is specified.
-// Encoded blocks can be concatenated and the result will be the combined input stream.
-// Data compressed with EncodeAll can be decoded with the Decoder,
-// using either a stream or DecodeAll.
-func (e *Encoder) EncodeAll(src, dst []byte) []byte {
-	if len(src) == 0 {
-		if e.o.fullZero {
-			// Add frame header.
-			fh := frameHeader{
-				ContentSize:   0,
-				WindowSize:    MinWindowSize,
-				SingleSegment: true,
-				// Adding a checksum would be a waste of space.
-				Checksum: false,
-				DictID:   0,
-			}
-			dst, _ = fh.appendTo(dst)
-
-			// Write raw block as last one only.
-			var blk blockHeader
-			blk.setSize(0)
-			blk.setType(blockTypeRaw)
-			blk.setLast(true)
-			dst = blk.appendTo(dst)
-		}
-		return dst
-	}
-	e.init.Do(e.initialize)
-	enc := <-e.encoders
-	defer func() {
-		// Release encoder reference to last block.
-		// If a non-single block is needed the encoder will reset again.
-		e.encoders <- enc
-	}()
-	// Use single segments when above minimum window and below 1MB.
-	single := len(src) < 1<<20 && len(src) > MinWindowSize
-	if e.o.single != nil {
-		single = *e.o.single
-	}
-	fh := frameHeader{
-		ContentSize:   uint64(len(src)),
-		WindowSize:    uint32(enc.WindowSize(len(src))),
-		SingleSegment: single,
-		Checksum:      e.o.crc,
-		DictID:        e.o.dict.ID(),
-	}
-
-	// If less than 1MB, allocate a buffer up front.
-	if len(dst) == 0 && cap(dst) == 0 && len(src) < 1<<20 && !e.o.lowMem {
-		dst = make([]byte, 0, len(src))
-	}
-	dst, err := fh.appendTo(dst)
-	if err != nil {
-		panic(err)
-	}
-
-	// If we can do everything in one block, prefer that.
-	if len(src) <= maxCompressedBlockSize {
-		enc.Reset(e.o.dict, true)
-		// Slightly faster with no history and everything in one block.
-		if e.o.crc {
-			_, _ = enc.CRC().Write(src)
-		}
-		blk := enc.Block()
-		blk.last = true
-		if e.o.dict == nil {
-			enc.EncodeNoHist(blk, src)
-		} else {
-			enc.Encode(blk, src)
-		}
-
-		// If we got the exact same number of literals as input,
-		// assume the literals cannot be compressed.
-		err := errIncompressible
-		oldout := blk.output
-		if len(blk.literals) != len(src) || len(src) != e.o.blockSize {
-			// Output directly to dst
-			blk.output = dst
-			err = blk.encode(src, e.o.noEntropy, !e.o.allLitEntropy)
-		}
-
-		switch err {
-		case errIncompressible:
-			if debugEncoder {
-				println("Storing incompressible block as raw")
-			}
-			dst = blk.encodeRawTo(dst, src)
-		case nil:
-			dst = blk.output
-		default:
-			panic(err)
-		}
-		blk.output = oldout
-	} else {
-		enc.Reset(e.o.dict, false)
-		blk := enc.Block()
-		for len(src) > 0 {
-			todo := src
-			if len(todo) > e.o.blockSize {
-				todo = todo[:e.o.blockSize]
-			}
-			src = src[len(todo):]
-			if e.o.crc {
-				_, _ = enc.CRC().Write(todo)
-			}
-			blk.pushOffsets()
-			enc.Encode(blk, todo)
-			if len(src) == 0 {
-				blk.last = true
-			}
-			err := errIncompressible
-			// If we got the exact same number of literals as input,
-			// assume the literals cannot be compressed.
-			if len(blk.literals) != len(todo) || len(todo) != e.o.blockSize {
-				err = blk.encode(todo, e.o.noEntropy, !e.o.allLitEntropy)
-			}
-
-			switch err {
-			case errIncompressible:
-				if debugEncoder {
-					println("Storing incompressible block as raw")
-				}
-				dst = blk.encodeRawTo(dst, todo)
-				blk.popOffsets()
-			case nil:
-				dst = append(dst, blk.output...)
-			default:
-				panic(err)
-			}
-			blk.reset(nil)
-		}
-	}
-	if e.o.crc {
-		dst = enc.AppendCRC(dst)
-	}
-	// Add padding with content from crypto/rand.Reader
-	if e.o.pad > 0 {
-		add := calcSkippableFrame(int64(len(dst)), int64(e.o.pad))
-		dst, err = skippableFrame(dst, add, rand.Reader)
-		if err != nil {
-			panic(err)
-		}
-	}
-	return dst
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/encoder_options.go b/vendor/github.com/klauspost/compress/zstd/encoder_options.go
deleted file mode 100644
index 16d4ab63c1..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/encoder_options.go
+++ /dev/null
@@ -1,312 +0,0 @@
-package zstd
-
-import (
-	"errors"
-	"fmt"
-	"runtime"
-	"strings"
-)
-
-// EOption is an option for creating a encoder.
-type EOption func(*encoderOptions) error
-
-// options retains accumulated state of multiple options.
-type encoderOptions struct {
-	concurrent      int
-	level           EncoderLevel
-	single          *bool
-	pad             int
-	blockSize       int
-	windowSize      int
-	crc             bool
-	fullZero        bool
-	noEntropy       bool
-	allLitEntropy   bool
-	customWindow    bool
-	customALEntropy bool
-	lowMem          bool
-	dict            *dict
-}
-
-func (o *encoderOptions) setDefault() {
-	*o = encoderOptions{
-		concurrent:    runtime.GOMAXPROCS(0),
-		crc:           true,
-		single:        nil,
-		blockSize:     1 << 16,
-		windowSize:    8 << 20,
-		level:         SpeedDefault,
-		allLitEntropy: true,
-		lowMem:        false,
-	}
-}
-
-// encoder returns an encoder with the selected options.
-func (o encoderOptions) encoder() encoder {
-	switch o.level {
-	case SpeedFastest:
-		if o.dict != nil {
-			return &fastEncoderDict{fastEncoder: fastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize), lowMem: o.lowMem}}}
-		}
-		return &fastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize), lowMem: o.lowMem}}
-
-	case SpeedDefault:
-		if o.dict != nil {
-			return &doubleFastEncoderDict{fastEncoderDict: fastEncoderDict{fastEncoder: fastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize), lowMem: o.lowMem}}}}
-		}
-		return &doubleFastEncoder{fastEncoder: fastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize), lowMem: o.lowMem}}}
-	case SpeedBetterCompression:
-		if o.dict != nil {
-			return &betterFastEncoderDict{betterFastEncoder: betterFastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize), lowMem: o.lowMem}}}
-		}
-		return &betterFastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize), lowMem: o.lowMem}}
-	case SpeedBestCompression:
-		return &bestFastEncoder{fastBase: fastBase{maxMatchOff: int32(o.windowSize), lowMem: o.lowMem}}
-	}
-	panic("unknown compression level")
-}
-
-// WithEncoderCRC will add CRC value to output.
-// Output will be 4 bytes larger.
-func WithEncoderCRC(b bool) EOption {
-	return func(o *encoderOptions) error { o.crc = b; return nil }
-}
-
-// WithEncoderConcurrency will set the concurrency,
-// meaning the maximum number of encoders to run concurrently.
-// The value supplied must be at least 1.
-// By default this will be set to GOMAXPROCS.
-func WithEncoderConcurrency(n int) EOption {
-	return func(o *encoderOptions) error {
-		if n <= 0 {
-			return fmt.Errorf("concurrency must be at least 1")
-		}
-		o.concurrent = n
-		return nil
-	}
-}
-
-// WithWindowSize will set the maximum allowed back-reference distance.
-// The value must be a power of two between MinWindowSize and MaxWindowSize.
-// A larger value will enable better compression but allocate more memory and,
-// for above-default values, take considerably longer.
-// The default value is determined by the compression level.
-func WithWindowSize(n int) EOption {
-	return func(o *encoderOptions) error {
-		switch {
-		case n < MinWindowSize:
-			return fmt.Errorf("window size must be at least %d", MinWindowSize)
-		case n > MaxWindowSize:
-			return fmt.Errorf("window size must be at most %d", MaxWindowSize)
-		case (n & (n - 1)) != 0:
-			return errors.New("window size must be a power of 2")
-		}
-
-		o.windowSize = n
-		o.customWindow = true
-		if o.blockSize > o.windowSize {
-			o.blockSize = o.windowSize
-		}
-		return nil
-	}
-}
-
-// WithEncoderPadding will add padding to all output so the size will be a multiple of n.
-// This can be used to obfuscate the exact output size or make blocks of a certain size.
-// The contents will be a skippable frame, so it will be invisible by the decoder.
-// n must be > 0 and <= 1GB, 1<<30 bytes.
-// The padded area will be filled with data from crypto/rand.Reader.
-// If `EncodeAll` is used with data already in the destination, the total size will be multiple of this.
-func WithEncoderPadding(n int) EOption {
-	return func(o *encoderOptions) error {
-		if n <= 0 {
-			return fmt.Errorf("padding must be at least 1")
-		}
-		// No need to waste our time.
-		if n == 1 {
-			o.pad = 0
-		}
-		if n > 1<<30 {
-			return fmt.Errorf("padding must less than 1GB (1<<30 bytes) ")
-		}
-		o.pad = n
-		return nil
-	}
-}
-
-// EncoderLevel predefines encoder compression levels.
-// Only use the constants made available, since the actual mapping
-// of these values are very likely to change and your compression could change
-// unpredictably when upgrading the library.
-type EncoderLevel int
-
-const (
-	speedNotSet EncoderLevel = iota
-
-	// SpeedFastest will choose the fastest reasonable compression.
-	// This is roughly equivalent to the fastest Zstandard mode.
-	SpeedFastest
-
-	// SpeedDefault is the default "pretty fast" compression option.
-	// This is roughly equivalent to the default Zstandard mode (level 3).
-	SpeedDefault
-
-	// SpeedBetterCompression will yield better compression than the default.
-	// Currently it is about zstd level 7-8 with ~ 2x-3x the default CPU usage.
-	// By using this, notice that CPU usage may go up in the future.
-	SpeedBetterCompression
-
-	// SpeedBestCompression will choose the best available compression option.
-	// This will offer the best compression no matter the CPU cost.
-	SpeedBestCompression
-
-	// speedLast should be kept as the last actual compression option.
-	// The is not for external usage, but is used to keep track of the valid options.
-	speedLast
-)
-
-// EncoderLevelFromString will convert a string representation of an encoding level back
-// to a compression level. The compare is not case sensitive.
-// If the string wasn't recognized, (false, SpeedDefault) will be returned.
-func EncoderLevelFromString(s string) (bool, EncoderLevel) {
-	for l := speedNotSet + 1; l < speedLast; l++ {
-		if strings.EqualFold(s, l.String()) {
-			return true, l
-		}
-	}
-	return false, SpeedDefault
-}
-
-// EncoderLevelFromZstd will return an encoder level that closest matches the compression
-// ratio of a specific zstd compression level.
-// Many input values will provide the same compression level.
-func EncoderLevelFromZstd(level int) EncoderLevel {
-	switch {
-	case level < 3:
-		return SpeedFastest
-	case level >= 3 && level < 6:
-		return SpeedDefault
-	case level >= 6 && level < 10:
-		return SpeedBetterCompression
-	case level >= 10:
-		return SpeedBetterCompression
-	}
-	return SpeedDefault
-}
-
-// String provides a string representation of the compression level.
-func (e EncoderLevel) String() string {
-	switch e {
-	case SpeedFastest:
-		return "fastest"
-	case SpeedDefault:
-		return "default"
-	case SpeedBetterCompression:
-		return "better"
-	case SpeedBestCompression:
-		return "best"
-	default:
-		return "invalid"
-	}
-}
-
-// WithEncoderLevel specifies a predefined compression level.
-func WithEncoderLevel(l EncoderLevel) EOption {
-	return func(o *encoderOptions) error {
-		switch {
-		case l <= speedNotSet || l >= speedLast:
-			return fmt.Errorf("unknown encoder level")
-		}
-		o.level = l
-		if !o.customWindow {
-			switch o.level {
-			case SpeedFastest:
-				o.windowSize = 4 << 20
-			case SpeedDefault:
-				o.windowSize = 8 << 20
-			case SpeedBetterCompression:
-				o.windowSize = 16 << 20
-			case SpeedBestCompression:
-				o.windowSize = 32 << 20
-			}
-		}
-		if !o.customALEntropy {
-			o.allLitEntropy = l > SpeedFastest
-		}
-
-		return nil
-	}
-}
-
-// WithZeroFrames will encode 0 length input as full frames.
-// This can be needed for compatibility with zstandard usage,
-// but is not needed for this package.
-func WithZeroFrames(b bool) EOption {
-	return func(o *encoderOptions) error {
-		o.fullZero = b
-		return nil
-	}
-}
-
-// WithAllLitEntropyCompression will apply entropy compression if no matches are found.
-// Disabling this will skip incompressible data faster, but in cases with no matches but
-// skewed character distribution compression is lost.
-// Default value depends on the compression level selected.
-func WithAllLitEntropyCompression(b bool) EOption {
-	return func(o *encoderOptions) error {
-		o.customALEntropy = true
-		o.allLitEntropy = b
-		return nil
-	}
-}
-
-// WithNoEntropyCompression will always skip entropy compression of literals.
-// This can be useful if content has matches, but unlikely to benefit from entropy
-// compression. Usually the slight speed improvement is not worth enabling this.
-func WithNoEntropyCompression(b bool) EOption {
-	return func(o *encoderOptions) error {
-		o.noEntropy = b
-		return nil
-	}
-}
-
-// WithSingleSegment will set the "single segment" flag when EncodeAll is used.
-// If this flag is set, data must be regenerated within a single continuous memory segment.
-// In this case, Window_Descriptor byte is skipped, but Frame_Content_Size is necessarily present.
-// As a consequence, the decoder must allocate a memory segment of size equal or larger than size of your content.
-// In order to preserve the decoder from unreasonable memory requirements,
-// a decoder is allowed to reject a compressed frame which requests a memory size beyond decoder's authorized range.
-// For broader compatibility, decoders are recommended to support memory sizes of at least 8 MB.
-// This is only a recommendation, each decoder is free to support higher or lower limits, depending on local limitations.
-// If this is not specified, block encodes will automatically choose this based on the input size.
-// This setting has no effect on streamed encodes.
-func WithSingleSegment(b bool) EOption {
-	return func(o *encoderOptions) error {
-		o.single = &b
-		return nil
-	}
-}
-
-// WithLowerEncoderMem will trade in some memory cases trade less memory usage for
-// slower encoding speed.
-// This will not change the window size which is the primary function for reducing
-// memory usage. See WithWindowSize.
-func WithLowerEncoderMem(b bool) EOption {
-	return func(o *encoderOptions) error {
-		o.lowMem = b
-		return nil
-	}
-}
-
-// WithEncoderDict allows to register a dictionary that will be used for the encode.
-// The encoder *may* choose to use no dictionary instead for certain payloads.
-func WithEncoderDict(dict []byte) EOption {
-	return func(o *encoderOptions) error {
-		d, err := loadDict(dict)
-		if err != nil {
-			return err
-		}
-		o.dict = d
-		return nil
-	}
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/framedec.go b/vendor/github.com/klauspost/compress/zstd/framedec.go
deleted file mode 100644
index 989c79f8c3..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/framedec.go
+++ /dev/null
@@ -1,521 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"bytes"
-	"encoding/hex"
-	"errors"
-	"hash"
-	"io"
-	"sync"
-
-	"github.com/klauspost/compress/zstd/internal/xxhash"
-)
-
-type frameDec struct {
-	o      decoderOptions
-	crc    hash.Hash64
-	offset int64
-
-	WindowSize uint64
-
-	// In order queue of blocks being decoded.
-	decoding chan *blockDec
-
-	// Frame history passed between blocks
-	history history
-
-	rawInput byteBuffer
-
-	// Byte buffer that can be reused for small input blocks.
-	bBuf byteBuf
-
-	FrameContentSize uint64
-	frameDone        sync.WaitGroup
-
-	DictionaryID  *uint32
-	HasCheckSum   bool
-	SingleSegment bool
-
-	// asyncRunning indicates whether the async routine processes input on 'decoding'.
-	asyncRunningMu sync.Mutex
-	asyncRunning   bool
-}
-
-const (
-	// MinWindowSize is the minimum Window Size, which is 1 KB.
-	MinWindowSize = 1 << 10
-
-	// MaxWindowSize is the maximum encoder window size
-	// and the default decoder maximum window size.
-	MaxWindowSize = 1 << 29
-)
-
-var (
-	frameMagic          = []byte{0x28, 0xb5, 0x2f, 0xfd}
-	skippableFrameMagic = []byte{0x2a, 0x4d, 0x18}
-)
-
-func newFrameDec(o decoderOptions) *frameDec {
-	if o.maxWindowSize > o.maxDecodedSize {
-		o.maxWindowSize = o.maxDecodedSize
-	}
-	d := frameDec{
-		o: o,
-	}
-	return &d
-}
-
-// reset will read the frame header and prepare for block decoding.
-// If nothing can be read from the input, io.EOF will be returned.
-// Any other error indicated that the stream contained data, but
-// there was a problem.
-func (d *frameDec) reset(br byteBuffer) error {
-	d.HasCheckSum = false
-	d.WindowSize = 0
-	var signature [4]byte
-	for {
-		var err error
-		// Check if we can read more...
-		b, err := br.readSmall(1)
-		switch err {
-		case io.EOF, io.ErrUnexpectedEOF:
-			return io.EOF
-		default:
-			return err
-		case nil:
-			signature[0] = b[0]
-		}
-		// Read the rest, don't allow io.ErrUnexpectedEOF
-		b, err = br.readSmall(3)
-		switch err {
-		case io.EOF:
-			return io.EOF
-		default:
-			return err
-		case nil:
-			copy(signature[1:], b)
-		}
-
-		if !bytes.Equal(signature[1:4], skippableFrameMagic) || signature[0]&0xf0 != 0x50 {
-			if debugDecoder {
-				println("Not skippable", hex.EncodeToString(signature[:]), hex.EncodeToString(skippableFrameMagic))
-			}
-			// Break if not skippable frame.
-			break
-		}
-		// Read size to skip
-		b, err = br.readSmall(4)
-		if err != nil {
-			if debugDecoder {
-				println("Reading Frame Size", err)
-			}
-			return err
-		}
-		n := uint32(b[0]) | (uint32(b[1]) << 8) | (uint32(b[2]) << 16) | (uint32(b[3]) << 24)
-		println("Skipping frame with", n, "bytes.")
-		err = br.skipN(int(n))
-		if err != nil {
-			if debugDecoder {
-				println("Reading discarded frame", err)
-			}
-			return err
-		}
-	}
-	if !bytes.Equal(signature[:], frameMagic) {
-		if debugDecoder {
-			println("Got magic numbers: ", signature, "want:", frameMagic)
-		}
-		return ErrMagicMismatch
-	}
-
-	// Read Frame_Header_Descriptor
-	fhd, err := br.readByte()
-	if err != nil {
-		if debugDecoder {
-			println("Reading Frame_Header_Descriptor", err)
-		}
-		return err
-	}
-	d.SingleSegment = fhd&(1<<5) != 0
-
-	if fhd&(1<<3) != 0 {
-		return errors.New("reserved bit set on frame header")
-	}
-
-	// Read Window_Descriptor
-	// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#window_descriptor
-	d.WindowSize = 0
-	if !d.SingleSegment {
-		wd, err := br.readByte()
-		if err != nil {
-			if debugDecoder {
-				println("Reading Window_Descriptor", err)
-			}
-			return err
-		}
-		printf("raw: %x, mantissa: %d, exponent: %d\n", wd, wd&7, wd>>3)
-		windowLog := 10 + (wd >> 3)
-		windowBase := uint64(1) << windowLog
-		windowAdd := (windowBase / 8) * uint64(wd&0x7)
-		d.WindowSize = windowBase + windowAdd
-	}
-
-	// Read Dictionary_ID
-	// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#dictionary_id
-	d.DictionaryID = nil
-	if size := fhd & 3; size != 0 {
-		if size == 3 {
-			size = 4
-		}
-
-		b, err := br.readSmall(int(size))
-		if err != nil {
-			println("Reading Dictionary_ID", err)
-			return err
-		}
-		var id uint32
-		switch size {
-		case 1:
-			id = uint32(b[0])
-		case 2:
-			id = uint32(b[0]) | (uint32(b[1]) << 8)
-		case 4:
-			id = uint32(b[0]) | (uint32(b[1]) << 8) | (uint32(b[2]) << 16) | (uint32(b[3]) << 24)
-		}
-		if debugDecoder {
-			println("Dict size", size, "ID:", id)
-		}
-		if id > 0 {
-			// ID 0 means "sorry, no dictionary anyway".
-			// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#dictionary-format
-			d.DictionaryID = &id
-		}
-	}
-
-	// Read Frame_Content_Size
-	// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#frame_content_size
-	var fcsSize int
-	v := fhd >> 6
-	switch v {
-	case 0:
-		if d.SingleSegment {
-			fcsSize = 1
-		}
-	default:
-		fcsSize = 1 << v
-	}
-	d.FrameContentSize = 0
-	if fcsSize > 0 {
-		b, err := br.readSmall(fcsSize)
-		if err != nil {
-			println("Reading Frame content", err)
-			return err
-		}
-		switch fcsSize {
-		case 1:
-			d.FrameContentSize = uint64(b[0])
-		case 2:
-			// When FCS_Field_Size is 2, the offset of 256 is added.
-			d.FrameContentSize = uint64(b[0]) | (uint64(b[1]) << 8) + 256
-		case 4:
-			d.FrameContentSize = uint64(b[0]) | (uint64(b[1]) << 8) | (uint64(b[2]) << 16) | (uint64(b[3]) << 24)
-		case 8:
-			d1 := uint32(b[0]) | (uint32(b[1]) << 8) | (uint32(b[2]) << 16) | (uint32(b[3]) << 24)
-			d2 := uint32(b[4]) | (uint32(b[5]) << 8) | (uint32(b[6]) << 16) | (uint32(b[7]) << 24)
-			d.FrameContentSize = uint64(d1) | (uint64(d2) << 32)
-		}
-		if debugDecoder {
-			println("field size bits:", v, "fcsSize:", fcsSize, "FrameContentSize:", d.FrameContentSize, hex.EncodeToString(b[:fcsSize]), "singleseg:", d.SingleSegment, "window:", d.WindowSize)
-		}
-	}
-	// Move this to shared.
-	d.HasCheckSum = fhd&(1<<2) != 0
-	if d.HasCheckSum {
-		if d.crc == nil {
-			d.crc = xxhash.New()
-		}
-		d.crc.Reset()
-	}
-
-	if d.WindowSize == 0 && d.SingleSegment {
-		// We may not need window in this case.
-		d.WindowSize = d.FrameContentSize
-		if d.WindowSize < MinWindowSize {
-			d.WindowSize = MinWindowSize
-		}
-	}
-
-	if d.WindowSize > uint64(d.o.maxWindowSize) {
-		if debugDecoder {
-			printf("window size %d > max %d\n", d.WindowSize, d.o.maxWindowSize)
-		}
-		return ErrWindowSizeExceeded
-	}
-	// The minimum Window_Size is 1 KB.
-	if d.WindowSize < MinWindowSize {
-		if debugDecoder {
-			println("got window size: ", d.WindowSize)
-		}
-		return ErrWindowSizeTooSmall
-	}
-	d.history.windowSize = int(d.WindowSize)
-	if d.o.lowMem && d.history.windowSize < maxBlockSize {
-		d.history.maxSize = d.history.windowSize * 2
-	} else {
-		d.history.maxSize = d.history.windowSize + maxBlockSize
-	}
-	// history contains input - maybe we do something
-	d.rawInput = br
-	return nil
-}
-
-// next will start decoding the next block from stream.
-func (d *frameDec) next(block *blockDec) error {
-	if debugDecoder {
-		printf("decoding new block %p:%p", block, block.data)
-	}
-	err := block.reset(d.rawInput, d.WindowSize)
-	if err != nil {
-		println("block error:", err)
-		// Signal the frame decoder we have a problem.
-		d.sendErr(block, err)
-		return err
-	}
-	block.input <- struct{}{}
-	if debugDecoder {
-		println("next block:", block)
-	}
-	d.asyncRunningMu.Lock()
-	defer d.asyncRunningMu.Unlock()
-	if !d.asyncRunning {
-		return nil
-	}
-	if block.Last {
-		// We indicate the frame is done by sending io.EOF
-		d.decoding <- block
-		return io.EOF
-	}
-	d.decoding <- block
-	return nil
-}
-
-// sendEOF will queue an error block on the frame.
-// This will cause the frame decoder to return when it encounters the block.
-// Returns true if the decoder was added.
-func (d *frameDec) sendErr(block *blockDec, err error) bool {
-	d.asyncRunningMu.Lock()
-	defer d.asyncRunningMu.Unlock()
-	if !d.asyncRunning {
-		return false
-	}
-
-	println("sending error", err.Error())
-	block.sendErr(err)
-	d.decoding <- block
-	return true
-}
-
-// checkCRC will check the checksum if the frame has one.
-// Will return ErrCRCMismatch if crc check failed, otherwise nil.
-func (d *frameDec) checkCRC() error {
-	if !d.HasCheckSum {
-		return nil
-	}
-	var tmp [4]byte
-	got := d.crc.Sum64()
-	// Flip to match file order.
-	tmp[0] = byte(got >> 0)
-	tmp[1] = byte(got >> 8)
-	tmp[2] = byte(got >> 16)
-	tmp[3] = byte(got >> 24)
-
-	// We can overwrite upper tmp now
-	want, err := d.rawInput.readSmall(4)
-	if err != nil {
-		println("CRC missing?", err)
-		return err
-	}
-
-	if !bytes.Equal(tmp[:], want) {
-		if debugDecoder {
-			println("CRC Check Failed:", tmp[:], "!=", want)
-		}
-		return ErrCRCMismatch
-	}
-	if debugDecoder {
-		println("CRC ok", tmp[:])
-	}
-	return nil
-}
-
-func (d *frameDec) initAsync() {
-	if !d.o.lowMem && !d.SingleSegment {
-		// set max extra size history to 2MB.
-		d.history.maxSize = d.history.windowSize + maxBlockSize
-	}
-	// re-alloc if more than one extra block size.
-	if d.o.lowMem && cap(d.history.b) > d.history.maxSize+maxBlockSize {
-		d.history.b = make([]byte, 0, d.history.maxSize)
-	}
-	if cap(d.history.b) < d.history.maxSize {
-		d.history.b = make([]byte, 0, d.history.maxSize)
-	}
-	if cap(d.decoding) < d.o.concurrent {
-		d.decoding = make(chan *blockDec, d.o.concurrent)
-	}
-	if debugDecoder {
-		h := d.history
-		printf("history init. len: %d, cap: %d", len(h.b), cap(h.b))
-	}
-	d.asyncRunningMu.Lock()
-	d.asyncRunning = true
-	d.asyncRunningMu.Unlock()
-}
-
-// startDecoder will start decoding blocks and write them to the writer.
-// The decoder will stop as soon as an error occurs or at end of frame.
-// When the frame has finished decoding the *bufio.Reader
-// containing the remaining input will be sent on frameDec.frameDone.
-func (d *frameDec) startDecoder(output chan decodeOutput) {
-	written := int64(0)
-
-	defer func() {
-		d.asyncRunningMu.Lock()
-		d.asyncRunning = false
-		d.asyncRunningMu.Unlock()
-
-		// Drain the currently decoding.
-		d.history.error = true
-	flushdone:
-		for {
-			select {
-			case b := <-d.decoding:
-				b.history <- &d.history
-				output <- <-b.result
-			default:
-				break flushdone
-			}
-		}
-		println("frame decoder done, signalling done")
-		d.frameDone.Done()
-	}()
-	// Get decoder for first block.
-	block := <-d.decoding
-	block.history <- &d.history
-	for {
-		var next *blockDec
-		// Get result
-		r := <-block.result
-		if r.err != nil {
-			println("Result contained error", r.err)
-			output <- r
-			return
-		}
-		if debugDecoder {
-			println("got result, from ", d.offset, "to", d.offset+int64(len(r.b)))
-			d.offset += int64(len(r.b))
-		}
-		if !block.Last {
-			// Send history to next block
-			select {
-			case next = <-d.decoding:
-				if debugDecoder {
-					println("Sending ", len(d.history.b), "bytes as history")
-				}
-				next.history <- &d.history
-			default:
-				// Wait until we have sent the block, so
-				// other decoders can potentially get the decoder.
-				next = nil
-			}
-		}
-
-		// Add checksum, async to decoding.
-		if d.HasCheckSum {
-			n, err := d.crc.Write(r.b)
-			if err != nil {
-				r.err = err
-				if n != len(r.b) {
-					r.err = io.ErrShortWrite
-				}
-				output <- r
-				return
-			}
-		}
-		written += int64(len(r.b))
-		if d.SingleSegment && uint64(written) > d.FrameContentSize {
-			println("runDecoder: single segment and", uint64(written), ">", d.FrameContentSize)
-			r.err = ErrFrameSizeExceeded
-			output <- r
-			return
-		}
-		if block.Last {
-			r.err = d.checkCRC()
-			output <- r
-			return
-		}
-		output <- r
-		if next == nil {
-			// There was no decoder available, we wait for one now that we have sent to the writer.
-			if debugDecoder {
-				println("Sending ", len(d.history.b), " bytes as history")
-			}
-			next = <-d.decoding
-			next.history <- &d.history
-		}
-		block = next
-	}
-}
-
-// runDecoder will create a sync decoder that will decode a block of data.
-func (d *frameDec) runDecoder(dst []byte, dec *blockDec) ([]byte, error) {
-	saved := d.history.b
-
-	// We use the history for output to avoid copying it.
-	d.history.b = dst
-	// Store input length, so we only check new data.
-	crcStart := len(dst)
-	var err error
-	for {
-		err = dec.reset(d.rawInput, d.WindowSize)
-		if err != nil {
-			break
-		}
-		if debugDecoder {
-			println("next block:", dec)
-		}
-		err = dec.decodeBuf(&d.history)
-		if err != nil || dec.Last {
-			break
-		}
-		if uint64(len(d.history.b)) > d.o.maxDecodedSize {
-			err = ErrDecoderSizeExceeded
-			break
-		}
-		if d.SingleSegment && uint64(len(d.history.b)) > d.o.maxDecodedSize {
-			println("runDecoder: single segment and", uint64(len(d.history.b)), ">", d.o.maxDecodedSize)
-			err = ErrFrameSizeExceeded
-			break
-		}
-	}
-	dst = d.history.b
-	if err == nil {
-		if d.HasCheckSum {
-			var n int
-			n, err = d.crc.Write(dst[crcStart:])
-			if err == nil {
-				if n != len(dst)-crcStart {
-					err = io.ErrShortWrite
-				} else {
-					err = d.checkCRC()
-				}
-			}
-		}
-	}
-	d.history.b = saved
-	return dst, err
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/frameenc.go b/vendor/github.com/klauspost/compress/zstd/frameenc.go
deleted file mode 100644
index 4ef7f5a3e3..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/frameenc.go
+++ /dev/null
@@ -1,137 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"encoding/binary"
-	"fmt"
-	"io"
-	"math"
-	"math/bits"
-)
-
-type frameHeader struct {
-	ContentSize   uint64
-	WindowSize    uint32
-	SingleSegment bool
-	Checksum      bool
-	DictID        uint32
-}
-
-const maxHeaderSize = 14
-
-func (f frameHeader) appendTo(dst []byte) ([]byte, error) {
-	dst = append(dst, frameMagic...)
-	var fhd uint8
-	if f.Checksum {
-		fhd |= 1 << 2
-	}
-	if f.SingleSegment {
-		fhd |= 1 << 5
-	}
-
-	var dictIDContent []byte
-	if f.DictID > 0 {
-		var tmp [4]byte
-		if f.DictID < 256 {
-			fhd |= 1
-			tmp[0] = uint8(f.DictID)
-			dictIDContent = tmp[:1]
-		} else if f.DictID < 1<<16 {
-			fhd |= 2
-			binary.LittleEndian.PutUint16(tmp[:2], uint16(f.DictID))
-			dictIDContent = tmp[:2]
-		} else {
-			fhd |= 3
-			binary.LittleEndian.PutUint32(tmp[:4], f.DictID)
-			dictIDContent = tmp[:4]
-		}
-	}
-	var fcs uint8
-	if f.ContentSize >= 256 {
-		fcs++
-	}
-	if f.ContentSize >= 65536+256 {
-		fcs++
-	}
-	if f.ContentSize >= 0xffffffff {
-		fcs++
-	}
-
-	fhd |= fcs << 6
-
-	dst = append(dst, fhd)
-	if !f.SingleSegment {
-		const winLogMin = 10
-		windowLog := (bits.Len32(f.WindowSize-1) - winLogMin) << 3
-		dst = append(dst, uint8(windowLog))
-	}
-	if f.DictID > 0 {
-		dst = append(dst, dictIDContent...)
-	}
-	switch fcs {
-	case 0:
-		if f.SingleSegment {
-			dst = append(dst, uint8(f.ContentSize))
-		}
-		// Unless SingleSegment is set, framessizes < 256 are nto stored.
-	case 1:
-		f.ContentSize -= 256
-		dst = append(dst, uint8(f.ContentSize), uint8(f.ContentSize>>8))
-	case 2:
-		dst = append(dst, uint8(f.ContentSize), uint8(f.ContentSize>>8), uint8(f.ContentSize>>16), uint8(f.ContentSize>>24))
-	case 3:
-		dst = append(dst, uint8(f.ContentSize), uint8(f.ContentSize>>8), uint8(f.ContentSize>>16), uint8(f.ContentSize>>24),
-			uint8(f.ContentSize>>32), uint8(f.ContentSize>>40), uint8(f.ContentSize>>48), uint8(f.ContentSize>>56))
-	default:
-		panic("invalid fcs")
-	}
-	return dst, nil
-}
-
-const skippableFrameHeader = 4 + 4
-
-// calcSkippableFrame will return a total size to be added for written
-// to be divisible by multiple.
-// The value will always be > skippableFrameHeader.
-// The function will panic if written < 0 or wantMultiple <= 0.
-func calcSkippableFrame(written, wantMultiple int64) int {
-	if wantMultiple <= 0 {
-		panic("wantMultiple <= 0")
-	}
-	if written < 0 {
-		panic("written < 0")
-	}
-	leftOver := written % wantMultiple
-	if leftOver == 0 {
-		return 0
-	}
-	toAdd := wantMultiple - leftOver
-	for toAdd < skippableFrameHeader {
-		toAdd += wantMultiple
-	}
-	return int(toAdd)
-}
-
-// skippableFrame will add a skippable frame with a total size of bytes.
-// total should be >= skippableFrameHeader and < math.MaxUint32.
-func skippableFrame(dst []byte, total int, r io.Reader) ([]byte, error) {
-	if total == 0 {
-		return dst, nil
-	}
-	if total < skippableFrameHeader {
-		return dst, fmt.Errorf("requested skippable frame (%d) < 8", total)
-	}
-	if int64(total) > math.MaxUint32 {
-		return dst, fmt.Errorf("requested skippable frame (%d) > max uint32", total)
-	}
-	dst = append(dst, 0x50, 0x2a, 0x4d, 0x18)
-	f := uint32(total - skippableFrameHeader)
-	dst = append(dst, uint8(f), uint8(f>>8), uint8(f>>16), uint8(f>>24))
-	start := len(dst)
-	dst = append(dst, make([]byte, f)...)
-	_, err := io.ReadFull(r, dst[start:])
-	return dst, err
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/fse_decoder.go b/vendor/github.com/klauspost/compress/zstd/fse_decoder.go
deleted file mode 100644
index e6d3d49b39..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/fse_decoder.go
+++ /dev/null
@@ -1,385 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"errors"
-	"fmt"
-)
-
-const (
-	tablelogAbsoluteMax = 9
-)
-
-const (
-	/*!MEMORY_USAGE :
-	 *  Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
-	 *  Increasing memory usage improves compression ratio
-	 *  Reduced memory usage can improve speed, due to cache effect
-	 *  Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */
-	maxMemoryUsage = tablelogAbsoluteMax + 2
-
-	maxTableLog    = maxMemoryUsage - 2
-	maxTablesize   = 1 << maxTableLog
-	maxTableMask   = (1 << maxTableLog) - 1
-	minTablelog    = 5
-	maxSymbolValue = 255
-)
-
-// fseDecoder provides temporary storage for compression and decompression.
-type fseDecoder struct {
-	dt             [maxTablesize]decSymbol // Decompression table.
-	symbolLen      uint16                  // Length of active part of the symbol table.
-	actualTableLog uint8                   // Selected tablelog.
-	maxBits        uint8                   // Maximum number of additional bits
-
-	// used for table creation to avoid allocations.
-	stateTable [256]uint16
-	norm       [maxSymbolValue + 1]int16
-	preDefined bool
-}
-
-// tableStep returns the next table index.
-func tableStep(tableSize uint32) uint32 {
-	return (tableSize >> 1) + (tableSize >> 3) + 3
-}
-
-// readNCount will read the symbol distribution so decoding tables can be constructed.
-func (s *fseDecoder) readNCount(b *byteReader, maxSymbol uint16) error {
-	var (
-		charnum   uint16
-		previous0 bool
-	)
-	if b.remain() < 4 {
-		return errors.New("input too small")
-	}
-	bitStream := b.Uint32NC()
-	nbBits := uint((bitStream & 0xF) + minTablelog) // extract tableLog
-	if nbBits > tablelogAbsoluteMax {
-		println("Invalid tablelog:", nbBits)
-		return errors.New("tableLog too large")
-	}
-	bitStream >>= 4
-	bitCount := uint(4)
-
-	s.actualTableLog = uint8(nbBits)
-	remaining := int32((1 << nbBits) + 1)
-	threshold := int32(1 << nbBits)
-	gotTotal := int32(0)
-	nbBits++
-
-	for remaining > 1 && charnum <= maxSymbol {
-		if previous0 {
-			//println("prev0")
-			n0 := charnum
-			for (bitStream & 0xFFFF) == 0xFFFF {
-				//println("24 x 0")
-				n0 += 24
-				if r := b.remain(); r > 5 {
-					b.advance(2)
-					// The check above should make sure we can read 32 bits
-					bitStream = b.Uint32NC() >> bitCount
-				} else {
-					// end of bit stream
-					bitStream >>= 16
-					bitCount += 16
-				}
-			}
-			//printf("bitstream: %d, 0b%b", bitStream&3, bitStream)
-			for (bitStream & 3) == 3 {
-				n0 += 3
-				bitStream >>= 2
-				bitCount += 2
-			}
-			n0 += uint16(bitStream & 3)
-			bitCount += 2
-
-			if n0 > maxSymbolValue {
-				return errors.New("maxSymbolValue too small")
-			}
-			//println("inserting ", n0-charnum, "zeroes from idx", charnum, "ending before", n0)
-			for charnum < n0 {
-				s.norm[uint8(charnum)] = 0
-				charnum++
-			}
-
-			if r := b.remain(); r >= 7 || r-int(bitCount>>3) >= 4 {
-				b.advance(bitCount >> 3)
-				bitCount &= 7
-				// The check above should make sure we can read 32 bits
-				bitStream = b.Uint32NC() >> bitCount
-			} else {
-				bitStream >>= 2
-			}
-		}
-
-		max := (2*threshold - 1) - remaining
-		var count int32
-
-		if int32(bitStream)&(threshold-1) < max {
-			count = int32(bitStream) & (threshold - 1)
-			if debugAsserts && nbBits < 1 {
-				panic("nbBits underflow")
-			}
-			bitCount += nbBits - 1
-		} else {
-			count = int32(bitStream) & (2*threshold - 1)
-			if count >= threshold {
-				count -= max
-			}
-			bitCount += nbBits
-		}
-
-		// extra accuracy
-		count--
-		if count < 0 {
-			// -1 means +1
-			remaining += count
-			gotTotal -= count
-		} else {
-			remaining -= count
-			gotTotal += count
-		}
-		s.norm[charnum&0xff] = int16(count)
-		charnum++
-		previous0 = count == 0
-		for remaining < threshold {
-			nbBits--
-			threshold >>= 1
-		}
-
-		if r := b.remain(); r >= 7 || r-int(bitCount>>3) >= 4 {
-			b.advance(bitCount >> 3)
-			bitCount &= 7
-			// The check above should make sure we can read 32 bits
-			bitStream = b.Uint32NC() >> (bitCount & 31)
-		} else {
-			bitCount -= (uint)(8 * (len(b.b) - 4 - b.off))
-			b.off = len(b.b) - 4
-			bitStream = b.Uint32() >> (bitCount & 31)
-		}
-	}
-	s.symbolLen = charnum
-	if s.symbolLen <= 1 {
-		return fmt.Errorf("symbolLen (%d) too small", s.symbolLen)
-	}
-	if s.symbolLen > maxSymbolValue+1 {
-		return fmt.Errorf("symbolLen (%d) too big", s.symbolLen)
-	}
-	if remaining != 1 {
-		return fmt.Errorf("corruption detected (remaining %d != 1)", remaining)
-	}
-	if bitCount > 32 {
-		return fmt.Errorf("corruption detected (bitCount %d > 32)", bitCount)
-	}
-	if gotTotal != 1<<s.actualTableLog {
-		return fmt.Errorf("corruption detected (total %d != %d)", gotTotal, 1<<s.actualTableLog)
-	}
-	b.advance((bitCount + 7) >> 3)
-	// println(s.norm[:s.symbolLen], s.symbolLen)
-	return s.buildDtable()
-}
-
-// decSymbol contains information about a state entry,
-// Including the state offset base, the output symbol and
-// the number of bits to read for the low part of the destination state.
-// Using a composite uint64 is faster than a struct with separate members.
-type decSymbol uint64
-
-func newDecSymbol(nbits, addBits uint8, newState uint16, baseline uint32) decSymbol {
-	return decSymbol(nbits) | (decSymbol(addBits) << 8) | (decSymbol(newState) << 16) | (decSymbol(baseline) << 32)
-}
-
-func (d decSymbol) nbBits() uint8 {
-	return uint8(d)
-}
-
-func (d decSymbol) addBits() uint8 {
-	return uint8(d >> 8)
-}
-
-func (d decSymbol) newState() uint16 {
-	return uint16(d >> 16)
-}
-
-func (d decSymbol) baseline() uint32 {
-	return uint32(d >> 32)
-}
-
-func (d decSymbol) baselineInt() int {
-	return int(d >> 32)
-}
-
-func (d *decSymbol) set(nbits, addBits uint8, newState uint16, baseline uint32) {
-	*d = decSymbol(nbits) | (decSymbol(addBits) << 8) | (decSymbol(newState) << 16) | (decSymbol(baseline) << 32)
-}
-
-func (d *decSymbol) setNBits(nBits uint8) {
-	const mask = 0xffffffffffffff00
-	*d = (*d & mask) | decSymbol(nBits)
-}
-
-func (d *decSymbol) setAddBits(addBits uint8) {
-	const mask = 0xffffffffffff00ff
-	*d = (*d & mask) | (decSymbol(addBits) << 8)
-}
-
-func (d *decSymbol) setNewState(state uint16) {
-	const mask = 0xffffffff0000ffff
-	*d = (*d & mask) | decSymbol(state)<<16
-}
-
-func (d *decSymbol) setBaseline(baseline uint32) {
-	const mask = 0xffffffff
-	*d = (*d & mask) | decSymbol(baseline)<<32
-}
-
-func (d *decSymbol) setExt(addBits uint8, baseline uint32) {
-	const mask = 0xffff00ff
-	*d = (*d & mask) | (decSymbol(addBits) << 8) | (decSymbol(baseline) << 32)
-}
-
-// decSymbolValue returns the transformed decSymbol for the given symbol.
-func decSymbolValue(symb uint8, t []baseOffset) (decSymbol, error) {
-	if int(symb) >= len(t) {
-		return 0, fmt.Errorf("rle symbol %d >= max %d", symb, len(t))
-	}
-	lu := t[symb]
-	return newDecSymbol(0, lu.addBits, 0, lu.baseLine), nil
-}
-
-// setRLE will set the decoder til RLE mode.
-func (s *fseDecoder) setRLE(symbol decSymbol) {
-	s.actualTableLog = 0
-	s.maxBits = symbol.addBits()
-	s.dt[0] = symbol
-}
-
-// buildDtable will build the decoding table.
-func (s *fseDecoder) buildDtable() error {
-	tableSize := uint32(1 << s.actualTableLog)
-	highThreshold := tableSize - 1
-	symbolNext := s.stateTable[:256]
-
-	// Init, lay down lowprob symbols
-	{
-		for i, v := range s.norm[:s.symbolLen] {
-			if v == -1 {
-				s.dt[highThreshold].setAddBits(uint8(i))
-				highThreshold--
-				symbolNext[i] = 1
-			} else {
-				symbolNext[i] = uint16(v)
-			}
-		}
-	}
-	// Spread symbols
-	{
-		tableMask := tableSize - 1
-		step := tableStep(tableSize)
-		position := uint32(0)
-		for ss, v := range s.norm[:s.symbolLen] {
-			for i := 0; i < int(v); i++ {
-				s.dt[position].setAddBits(uint8(ss))
-				position = (position + step) & tableMask
-				for position > highThreshold {
-					// lowprob area
-					position = (position + step) & tableMask
-				}
-			}
-		}
-		if position != 0 {
-			// position must reach all cells once, otherwise normalizedCounter is incorrect
-			return errors.New("corrupted input (position != 0)")
-		}
-	}
-
-	// Build Decoding table
-	{
-		tableSize := uint16(1 << s.actualTableLog)
-		for u, v := range s.dt[:tableSize] {
-			symbol := v.addBits()
-			nextState := symbolNext[symbol]
-			symbolNext[symbol] = nextState + 1
-			nBits := s.actualTableLog - byte(highBits(uint32(nextState)))
-			s.dt[u&maxTableMask].setNBits(nBits)
-			newState := (nextState << nBits) - tableSize
-			if newState > tableSize {
-				return fmt.Errorf("newState (%d) outside table size (%d)", newState, tableSize)
-			}
-			if newState == uint16(u) && nBits == 0 {
-				// Seems weird that this is possible with nbits > 0.
-				return fmt.Errorf("newState (%d) == oldState (%d) and no bits", newState, u)
-			}
-			s.dt[u&maxTableMask].setNewState(newState)
-		}
-	}
-	return nil
-}
-
-// transform will transform the decoder table into a table usable for
-// decoding without having to apply the transformation while decoding.
-// The state will contain the base value and the number of bits to read.
-func (s *fseDecoder) transform(t []baseOffset) error {
-	tableSize := uint16(1 << s.actualTableLog)
-	s.maxBits = 0
-	for i, v := range s.dt[:tableSize] {
-		add := v.addBits()
-		if int(add) >= len(t) {
-			return fmt.Errorf("invalid decoding table entry %d, symbol %d >= max (%d)", i, v.addBits(), len(t))
-		}
-		lu := t[add]
-		if lu.addBits > s.maxBits {
-			s.maxBits = lu.addBits
-		}
-		v.setExt(lu.addBits, lu.baseLine)
-		s.dt[i] = v
-	}
-	return nil
-}
-
-type fseState struct {
-	dt    []decSymbol
-	state decSymbol
-}
-
-// Initialize and decodeAsync first state and symbol.
-func (s *fseState) init(br *bitReader, tableLog uint8, dt []decSymbol) {
-	s.dt = dt
-	br.fill()
-	s.state = dt[br.getBits(tableLog)]
-}
-
-// next returns the current symbol and sets the next state.
-// At least tablelog bits must be available in the bit reader.
-func (s *fseState) next(br *bitReader) {
-	lowBits := uint16(br.getBits(s.state.nbBits()))
-	s.state = s.dt[s.state.newState()+lowBits]
-}
-
-// finished returns true if all bits have been read from the bitstream
-// and the next state would require reading bits from the input.
-func (s *fseState) finished(br *bitReader) bool {
-	return br.finished() && s.state.nbBits() > 0
-}
-
-// final returns the current state symbol without decoding the next.
-func (s *fseState) final() (int, uint8) {
-	return s.state.baselineInt(), s.state.addBits()
-}
-
-// final returns the current state symbol without decoding the next.
-func (s decSymbol) final() (int, uint8) {
-	return s.baselineInt(), s.addBits()
-}
-
-// nextFast returns the next symbol and sets the next state.
-// This can only be used if no symbols are 0 bits.
-// At least tablelog bits must be available in the bit reader.
-func (s *fseState) nextFast(br *bitReader) (uint32, uint8) {
-	lowBits := uint16(br.getBitsFast(s.state.nbBits()))
-	s.state = s.dt[s.state.newState()+lowBits]
-	return s.state.baseline(), s.state.addBits()
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/fse_encoder.go b/vendor/github.com/klauspost/compress/zstd/fse_encoder.go
deleted file mode 100644
index b4757ee3f0..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/fse_encoder.go
+++ /dev/null
@@ -1,725 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"errors"
-	"fmt"
-	"math"
-)
-
-const (
-	// For encoding we only support up to
-	maxEncTableLog    = 8
-	maxEncTablesize   = 1 << maxTableLog
-	maxEncTableMask   = (1 << maxTableLog) - 1
-	minEncTablelog    = 5
-	maxEncSymbolValue = maxMatchLengthSymbol
-)
-
-// Scratch provides temporary storage for compression and decompression.
-type fseEncoder struct {
-	symbolLen      uint16 // Length of active part of the symbol table.
-	actualTableLog uint8  // Selected tablelog.
-	ct             cTable // Compression tables.
-	maxCount       int    // count of the most probable symbol
-	zeroBits       bool   // no bits has prob > 50%.
-	clearCount     bool   // clear count
-	useRLE         bool   // This encoder is for RLE
-	preDefined     bool   // This encoder is predefined.
-	reUsed         bool   // Set to know when the encoder has been reused.
-	rleVal         uint8  // RLE Symbol
-	maxBits        uint8  // Maximum output bits after transform.
-
-	// TODO: Technically zstd should be fine with 64 bytes.
-	count [256]uint32
-	norm  [256]int16
-}
-
-// cTable contains tables used for compression.
-type cTable struct {
-	tableSymbol []byte
-	stateTable  []uint16
-	symbolTT    []symbolTransform
-}
-
-// symbolTransform contains the state transform for a symbol.
-type symbolTransform struct {
-	deltaNbBits    uint32
-	deltaFindState int16
-	outBits        uint8
-}
-
-// String prints values as a human readable string.
-func (s symbolTransform) String() string {
-	return fmt.Sprintf("{deltabits: %08x, findstate:%d outbits:%d}", s.deltaNbBits, s.deltaFindState, s.outBits)
-}
-
-// Histogram allows to populate the histogram and skip that step in the compression,
-// It otherwise allows to inspect the histogram when compression is done.
-// To indicate that you have populated the histogram call HistogramFinished
-// with the value of the highest populated symbol, as well as the number of entries
-// in the most populated entry. These are accepted at face value.
-// The returned slice will always be length 256.
-func (s *fseEncoder) Histogram() []uint32 {
-	return s.count[:]
-}
-
-// HistogramFinished can be called to indicate that the histogram has been populated.
-// maxSymbol is the index of the highest set symbol of the next data segment.
-// maxCount is the number of entries in the most populated entry.
-// These are accepted at face value.
-func (s *fseEncoder) HistogramFinished(maxSymbol uint8, maxCount int) {
-	s.maxCount = maxCount
-	s.symbolLen = uint16(maxSymbol) + 1
-	s.clearCount = maxCount != 0
-}
-
-// prepare will prepare and allocate scratch tables used for both compression and decompression.
-func (s *fseEncoder) prepare() (*fseEncoder, error) {
-	if s == nil {
-		s = &fseEncoder{}
-	}
-	s.useRLE = false
-	if s.clearCount && s.maxCount == 0 {
-		for i := range s.count {
-			s.count[i] = 0
-		}
-		s.clearCount = false
-	}
-	return s, nil
-}
-
-// allocCtable will allocate tables needed for compression.
-// If existing tables a re big enough, they are simply re-used.
-func (s *fseEncoder) allocCtable() {
-	tableSize := 1 << s.actualTableLog
-	// get tableSymbol that is big enough.
-	if cap(s.ct.tableSymbol) < tableSize {
-		s.ct.tableSymbol = make([]byte, tableSize)
-	}
-	s.ct.tableSymbol = s.ct.tableSymbol[:tableSize]
-
-	ctSize := tableSize
-	if cap(s.ct.stateTable) < ctSize {
-		s.ct.stateTable = make([]uint16, ctSize)
-	}
-	s.ct.stateTable = s.ct.stateTable[:ctSize]
-
-	if cap(s.ct.symbolTT) < 256 {
-		s.ct.symbolTT = make([]symbolTransform, 256)
-	}
-	s.ct.symbolTT = s.ct.symbolTT[:256]
-}
-
-// buildCTable will populate the compression table so it is ready to be used.
-func (s *fseEncoder) buildCTable() error {
-	tableSize := uint32(1 << s.actualTableLog)
-	highThreshold := tableSize - 1
-	var cumul [256]int16
-
-	s.allocCtable()
-	tableSymbol := s.ct.tableSymbol[:tableSize]
-	// symbol start positions
-	{
-		cumul[0] = 0
-		for ui, v := range s.norm[:s.symbolLen-1] {
-			u := byte(ui) // one less than reference
-			if v == -1 {
-				// Low proba symbol
-				cumul[u+1] = cumul[u] + 1
-				tableSymbol[highThreshold] = u
-				highThreshold--
-			} else {
-				cumul[u+1] = cumul[u] + v
-			}
-		}
-		// Encode last symbol separately to avoid overflowing u
-		u := int(s.symbolLen - 1)
-		v := s.norm[s.symbolLen-1]
-		if v == -1 {
-			// Low proba symbol
-			cumul[u+1] = cumul[u] + 1
-			tableSymbol[highThreshold] = byte(u)
-			highThreshold--
-		} else {
-			cumul[u+1] = cumul[u] + v
-		}
-		if uint32(cumul[s.symbolLen]) != tableSize {
-			return fmt.Errorf("internal error: expected cumul[s.symbolLen] (%d) == tableSize (%d)", cumul[s.symbolLen], tableSize)
-		}
-		cumul[s.symbolLen] = int16(tableSize) + 1
-	}
-	// Spread symbols
-	s.zeroBits = false
-	{
-		step := tableStep(tableSize)
-		tableMask := tableSize - 1
-		var position uint32
-		// if any symbol > largeLimit, we may have 0 bits output.
-		largeLimit := int16(1 << (s.actualTableLog - 1))
-		for ui, v := range s.norm[:s.symbolLen] {
-			symbol := byte(ui)
-			if v > largeLimit {
-				s.zeroBits = true
-			}
-			for nbOccurrences := int16(0); nbOccurrences < v; nbOccurrences++ {
-				tableSymbol[position] = symbol
-				position = (position + step) & tableMask
-				for position > highThreshold {
-					position = (position + step) & tableMask
-				} /* Low proba area */
-			}
-		}
-
-		// Check if we have gone through all positions
-		if position != 0 {
-			return errors.New("position!=0")
-		}
-	}
-
-	// Build table
-	table := s.ct.stateTable
-	{
-		tsi := int(tableSize)
-		for u, v := range tableSymbol {
-			// TableU16 : sorted by symbol order; gives next state value
-			table[cumul[v]] = uint16(tsi + u)
-			cumul[v]++
-		}
-	}
-
-	// Build Symbol Transformation Table
-	{
-		total := int16(0)
-		symbolTT := s.ct.symbolTT[:s.symbolLen]
-		tableLog := s.actualTableLog
-		tl := (uint32(tableLog) << 16) - (1 << tableLog)
-		for i, v := range s.norm[:s.symbolLen] {
-			switch v {
-			case 0:
-			case -1, 1:
-				symbolTT[i].deltaNbBits = tl
-				symbolTT[i].deltaFindState = total - 1
-				total++
-			default:
-				maxBitsOut := uint32(tableLog) - highBit(uint32(v-1))
-				minStatePlus := uint32(v) << maxBitsOut
-				symbolTT[i].deltaNbBits = (maxBitsOut << 16) - minStatePlus
-				symbolTT[i].deltaFindState = total - v
-				total += v
-			}
-		}
-		if total != int16(tableSize) {
-			return fmt.Errorf("total mismatch %d (got) != %d (want)", total, tableSize)
-		}
-	}
-	return nil
-}
-
-var rtbTable = [...]uint32{0, 473195, 504333, 520860, 550000, 700000, 750000, 830000}
-
-func (s *fseEncoder) setRLE(val byte) {
-	s.allocCtable()
-	s.actualTableLog = 0
-	s.ct.stateTable = s.ct.stateTable[:1]
-	s.ct.symbolTT[val] = symbolTransform{
-		deltaFindState: 0,
-		deltaNbBits:    0,
-	}
-	if debugEncoder {
-		println("setRLE: val", val, "symbolTT", s.ct.symbolTT[val])
-	}
-	s.rleVal = val
-	s.useRLE = true
-}
-
-// setBits will set output bits for the transform.
-// if nil is provided, the number of bits is equal to the index.
-func (s *fseEncoder) setBits(transform []byte) {
-	if s.reUsed || s.preDefined {
-		return
-	}
-	if s.useRLE {
-		if transform == nil {
-			s.ct.symbolTT[s.rleVal].outBits = s.rleVal
-			s.maxBits = s.rleVal
-			return
-		}
-		s.maxBits = transform[s.rleVal]
-		s.ct.symbolTT[s.rleVal].outBits = s.maxBits
-		return
-	}
-	if transform == nil {
-		for i := range s.ct.symbolTT[:s.symbolLen] {
-			s.ct.symbolTT[i].outBits = uint8(i)
-		}
-		s.maxBits = uint8(s.symbolLen - 1)
-		return
-	}
-	s.maxBits = 0
-	for i, v := range transform[:s.symbolLen] {
-		s.ct.symbolTT[i].outBits = v
-		if v > s.maxBits {
-			// We could assume bits always going up, but we play safe.
-			s.maxBits = v
-		}
-	}
-}
-
-// normalizeCount will normalize the count of the symbols so
-// the total is equal to the table size.
-// If successful, compression tables will also be made ready.
-func (s *fseEncoder) normalizeCount(length int) error {
-	if s.reUsed {
-		return nil
-	}
-	s.optimalTableLog(length)
-	var (
-		tableLog          = s.actualTableLog
-		scale             = 62 - uint64(tableLog)
-		step              = (1 << 62) / uint64(length)
-		vStep             = uint64(1) << (scale - 20)
-		stillToDistribute = int16(1 << tableLog)
-		largest           int
-		largestP          int16
-		lowThreshold      = (uint32)(length >> tableLog)
-	)
-	if s.maxCount == length {
-		s.useRLE = true
-		return nil
-	}
-	s.useRLE = false
-	for i, cnt := range s.count[:s.symbolLen] {
-		// already handled
-		// if (count[s] == s.length) return 0;   /* rle special case */
-
-		if cnt == 0 {
-			s.norm[i] = 0
-			continue
-		}
-		if cnt <= lowThreshold {
-			s.norm[i] = -1
-			stillToDistribute--
-		} else {
-			proba := (int16)((uint64(cnt) * step) >> scale)
-			if proba < 8 {
-				restToBeat := vStep * uint64(rtbTable[proba])
-				v := uint64(cnt)*step - (uint64(proba) << scale)
-				if v > restToBeat {
-					proba++
-				}
-			}
-			if proba > largestP {
-				largestP = proba
-				largest = i
-			}
-			s.norm[i] = proba
-			stillToDistribute -= proba
-		}
-	}
-
-	if -stillToDistribute >= (s.norm[largest] >> 1) {
-		// corner case, need another normalization method
-		err := s.normalizeCount2(length)
-		if err != nil {
-			return err
-		}
-		if debugAsserts {
-			err = s.validateNorm()
-			if err != nil {
-				return err
-			}
-		}
-		return s.buildCTable()
-	}
-	s.norm[largest] += stillToDistribute
-	if debugAsserts {
-		err := s.validateNorm()
-		if err != nil {
-			return err
-		}
-	}
-	return s.buildCTable()
-}
-
-// Secondary normalization method.
-// To be used when primary method fails.
-func (s *fseEncoder) normalizeCount2(length int) error {
-	const notYetAssigned = -2
-	var (
-		distributed  uint32
-		total        = uint32(length)
-		tableLog     = s.actualTableLog
-		lowThreshold = total >> tableLog
-		lowOne       = (total * 3) >> (tableLog + 1)
-	)
-	for i, cnt := range s.count[:s.symbolLen] {
-		if cnt == 0 {
-			s.norm[i] = 0
-			continue
-		}
-		if cnt <= lowThreshold {
-			s.norm[i] = -1
-			distributed++
-			total -= cnt
-			continue
-		}
-		if cnt <= lowOne {
-			s.norm[i] = 1
-			distributed++
-			total -= cnt
-			continue
-		}
-		s.norm[i] = notYetAssigned
-	}
-	toDistribute := (1 << tableLog) - distributed
-
-	if (total / toDistribute) > lowOne {
-		// risk of rounding to zero
-		lowOne = (total * 3) / (toDistribute * 2)
-		for i, cnt := range s.count[:s.symbolLen] {
-			if (s.norm[i] == notYetAssigned) && (cnt <= lowOne) {
-				s.norm[i] = 1
-				distributed++
-				total -= cnt
-				continue
-			}
-		}
-		toDistribute = (1 << tableLog) - distributed
-	}
-	if distributed == uint32(s.symbolLen)+1 {
-		// all values are pretty poor;
-		//   probably incompressible data (should have already been detected);
-		//   find max, then give all remaining points to max
-		var maxV int
-		var maxC uint32
-		for i, cnt := range s.count[:s.symbolLen] {
-			if cnt > maxC {
-				maxV = i
-				maxC = cnt
-			}
-		}
-		s.norm[maxV] += int16(toDistribute)
-		return nil
-	}
-
-	if total == 0 {
-		// all of the symbols were low enough for the lowOne or lowThreshold
-		for i := uint32(0); toDistribute > 0; i = (i + 1) % (uint32(s.symbolLen)) {
-			if s.norm[i] > 0 {
-				toDistribute--
-				s.norm[i]++
-			}
-		}
-		return nil
-	}
-
-	var (
-		vStepLog = 62 - uint64(tableLog)
-		mid      = uint64((1 << (vStepLog - 1)) - 1)
-		rStep    = (((1 << vStepLog) * uint64(toDistribute)) + mid) / uint64(total) // scale on remaining
-		tmpTotal = mid
-	)
-	for i, cnt := range s.count[:s.symbolLen] {
-		if s.norm[i] == notYetAssigned {
-			var (
-				end    = tmpTotal + uint64(cnt)*rStep
-				sStart = uint32(tmpTotal >> vStepLog)
-				sEnd   = uint32(end >> vStepLog)
-				weight = sEnd - sStart
-			)
-			if weight < 1 {
-				return errors.New("weight < 1")
-			}
-			s.norm[i] = int16(weight)
-			tmpTotal = end
-		}
-	}
-	return nil
-}
-
-// optimalTableLog calculates and sets the optimal tableLog in s.actualTableLog
-func (s *fseEncoder) optimalTableLog(length int) {
-	tableLog := uint8(maxEncTableLog)
-	minBitsSrc := highBit(uint32(length)) + 1
-	minBitsSymbols := highBit(uint32(s.symbolLen-1)) + 2
-	minBits := uint8(minBitsSymbols)
-	if minBitsSrc < minBitsSymbols {
-		minBits = uint8(minBitsSrc)
-	}
-
-	maxBitsSrc := uint8(highBit(uint32(length-1))) - 2
-	if maxBitsSrc < tableLog {
-		// Accuracy can be reduced
-		tableLog = maxBitsSrc
-	}
-	if minBits > tableLog {
-		tableLog = minBits
-	}
-	// Need a minimum to safely represent all symbol values
-	if tableLog < minEncTablelog {
-		tableLog = minEncTablelog
-	}
-	if tableLog > maxEncTableLog {
-		tableLog = maxEncTableLog
-	}
-	s.actualTableLog = tableLog
-}
-
-// validateNorm validates the normalized histogram table.
-func (s *fseEncoder) validateNorm() (err error) {
-	var total int
-	for _, v := range s.norm[:s.symbolLen] {
-		if v >= 0 {
-			total += int(v)
-		} else {
-			total -= int(v)
-		}
-	}
-	defer func() {
-		if err == nil {
-			return
-		}
-		fmt.Printf("selected TableLog: %d, Symbol length: %d\n", s.actualTableLog, s.symbolLen)
-		for i, v := range s.norm[:s.symbolLen] {
-			fmt.Printf("%3d: %5d -> %4d \n", i, s.count[i], v)
-		}
-	}()
-	if total != (1 << s.actualTableLog) {
-		return fmt.Errorf("warning: Total == %d != %d", total, 1<<s.actualTableLog)
-	}
-	for i, v := range s.count[s.symbolLen:] {
-		if v != 0 {
-			return fmt.Errorf("warning: Found symbol out of range, %d after cut", i)
-		}
-	}
-	return nil
-}
-
-// writeCount will write the normalized histogram count to header.
-// This is read back by readNCount.
-func (s *fseEncoder) writeCount(out []byte) ([]byte, error) {
-	if s.useRLE {
-		return append(out, s.rleVal), nil
-	}
-	if s.preDefined || s.reUsed {
-		// Never write predefined.
-		return out, nil
-	}
-
-	var (
-		tableLog  = s.actualTableLog
-		tableSize = 1 << tableLog
-		previous0 bool
-		charnum   uint16
-
-		// maximum header size plus 2 extra bytes for final output if bitCount == 0.
-		maxHeaderSize = ((int(s.symbolLen) * int(tableLog)) >> 3) + 3 + 2
-
-		// Write Table Size
-		bitStream = uint32(tableLog - minEncTablelog)
-		bitCount  = uint(4)
-		remaining = int16(tableSize + 1) /* +1 for extra accuracy */
-		threshold = int16(tableSize)
-		nbBits    = uint(tableLog + 1)
-		outP      = len(out)
-	)
-	if cap(out) < outP+maxHeaderSize {
-		out = append(out, make([]byte, maxHeaderSize*3)...)
-		out = out[:len(out)-maxHeaderSize*3]
-	}
-	out = out[:outP+maxHeaderSize]
-
-	// stops at 1
-	for remaining > 1 {
-		if previous0 {
-			start := charnum
-			for s.norm[charnum] == 0 {
-				charnum++
-			}
-			for charnum >= start+24 {
-				start += 24
-				bitStream += uint32(0xFFFF) << bitCount
-				out[outP] = byte(bitStream)
-				out[outP+1] = byte(bitStream >> 8)
-				outP += 2
-				bitStream >>= 16
-			}
-			for charnum >= start+3 {
-				start += 3
-				bitStream += 3 << bitCount
-				bitCount += 2
-			}
-			bitStream += uint32(charnum-start) << bitCount
-			bitCount += 2
-			if bitCount > 16 {
-				out[outP] = byte(bitStream)
-				out[outP+1] = byte(bitStream >> 8)
-				outP += 2
-				bitStream >>= 16
-				bitCount -= 16
-			}
-		}
-
-		count := s.norm[charnum]
-		charnum++
-		max := (2*threshold - 1) - remaining
-		if count < 0 {
-			remaining += count
-		} else {
-			remaining -= count
-		}
-		count++ // +1 for extra accuracy
-		if count >= threshold {
-			count += max // [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[
-		}
-		bitStream += uint32(count) << bitCount
-		bitCount += nbBits
-		if count < max {
-			bitCount--
-		}
-
-		previous0 = count == 1
-		if remaining < 1 {
-			return nil, errors.New("internal error: remaining < 1")
-		}
-		for remaining < threshold {
-			nbBits--
-			threshold >>= 1
-		}
-
-		if bitCount > 16 {
-			out[outP] = byte(bitStream)
-			out[outP+1] = byte(bitStream >> 8)
-			outP += 2
-			bitStream >>= 16
-			bitCount -= 16
-		}
-	}
-
-	if outP+2 > len(out) {
-		return nil, fmt.Errorf("internal error: %d > %d, maxheader: %d, sl: %d, tl: %d, normcount: %v", outP+2, len(out), maxHeaderSize, s.symbolLen, int(tableLog), s.norm[:s.symbolLen])
-	}
-	out[outP] = byte(bitStream)
-	out[outP+1] = byte(bitStream >> 8)
-	outP += int((bitCount + 7) / 8)
-
-	if charnum > s.symbolLen {
-		return nil, errors.New("internal error: charnum > s.symbolLen")
-	}
-	return out[:outP], nil
-}
-
-// 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 *
-func (s *fseEncoder) bitCost(symbolValue uint8, accuracyLog uint32) uint32 {
-	minNbBits := s.ct.symbolTT[symbolValue].deltaNbBits >> 16
-	threshold := (minNbBits + 1) << 16
-	if debugAsserts {
-		if !(s.actualTableLog < 16) {
-			panic("!s.actualTableLog < 16")
-		}
-		// ensure enough room for renormalization double shift
-		if !(uint8(accuracyLog) < 31-s.actualTableLog) {
-			panic("!uint8(accuracyLog) < 31-s.actualTableLog")
-		}
-	}
-	tableSize := uint32(1) << s.actualTableLog
-	deltaFromThreshold := threshold - (s.ct.symbolTT[symbolValue].deltaNbBits + tableSize)
-	// linear interpolation (very approximate)
-	normalizedDeltaFromThreshold := (deltaFromThreshold << accuracyLog) >> s.actualTableLog
-	bitMultiplier := uint32(1) << accuracyLog
-	if debugAsserts {
-		if s.ct.symbolTT[symbolValue].deltaNbBits+tableSize > threshold {
-			panic("s.ct.symbolTT[symbolValue].deltaNbBits+tableSize > threshold")
-		}
-		if normalizedDeltaFromThreshold > bitMultiplier {
-			panic("normalizedDeltaFromThreshold > bitMultiplier")
-		}
-	}
-	return (minNbBits+1)*bitMultiplier - normalizedDeltaFromThreshold
-}
-
-// Returns the cost in bits of encoding the distribution in count using ctable.
-// Histogram should only be up to the last non-zero symbol.
-// Returns an -1 if ctable cannot represent all the symbols in count.
-func (s *fseEncoder) approxSize(hist []uint32) uint32 {
-	if int(s.symbolLen) < len(hist) {
-		// More symbols than we have.
-		return math.MaxUint32
-	}
-	if s.useRLE {
-		// We will never reuse RLE encoders.
-		return math.MaxUint32
-	}
-	const kAccuracyLog = 8
-	badCost := (uint32(s.actualTableLog) + 1) << kAccuracyLog
-	var cost uint32
-	for i, v := range hist {
-		if v == 0 {
-			continue
-		}
-		if s.norm[i] == 0 {
-			return math.MaxUint32
-		}
-		bitCost := s.bitCost(uint8(i), kAccuracyLog)
-		if bitCost > badCost {
-			return math.MaxUint32
-		}
-		cost += v * bitCost
-	}
-	return cost >> kAccuracyLog
-}
-
-// maxHeaderSize returns the maximum header size in bits.
-// This is not exact size, but we want a penalty for new tables anyway.
-func (s *fseEncoder) maxHeaderSize() uint32 {
-	if s.preDefined {
-		return 0
-	}
-	if s.useRLE {
-		return 8
-	}
-	return (((uint32(s.symbolLen) * uint32(s.actualTableLog)) >> 3) + 3) * 8
-}
-
-// cState contains the compression state of a stream.
-type cState struct {
-	bw         *bitWriter
-	stateTable []uint16
-	state      uint16
-}
-
-// init will initialize the compression state to the first symbol of the stream.
-func (c *cState) init(bw *bitWriter, ct *cTable, first symbolTransform) {
-	c.bw = bw
-	c.stateTable = ct.stateTable
-	if len(c.stateTable) == 1 {
-		// RLE
-		c.stateTable[0] = uint16(0)
-		c.state = 0
-		return
-	}
-	nbBitsOut := (first.deltaNbBits + (1 << 15)) >> 16
-	im := int32((nbBitsOut << 16) - first.deltaNbBits)
-	lu := (im >> nbBitsOut) + int32(first.deltaFindState)
-	c.state = c.stateTable[lu]
-}
-
-// encode the output symbol provided and write it to the bitstream.
-func (c *cState) encode(symbolTT symbolTransform) {
-	nbBitsOut := (uint32(c.state) + symbolTT.deltaNbBits) >> 16
-	dstState := int32(c.state>>(nbBitsOut&15)) + int32(symbolTT.deltaFindState)
-	c.bw.addBits16NC(c.state, uint8(nbBitsOut))
-	c.state = c.stateTable[dstState]
-}
-
-// flush will write the tablelog to the output and flush the remaining full bytes.
-func (c *cState) flush(tableLog uint8) {
-	c.bw.flush32()
-	c.bw.addBits16NC(c.state, tableLog)
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/fse_predefined.go b/vendor/github.com/klauspost/compress/zstd/fse_predefined.go
deleted file mode 100644
index 474cb77d2b..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/fse_predefined.go
+++ /dev/null
@@ -1,158 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"fmt"
-	"math"
-	"sync"
-)
-
-var (
-	// fsePredef are the predefined fse tables as defined here:
-	// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#default-distributions
-	// These values are already transformed.
-	fsePredef [3]fseDecoder
-
-	// fsePredefEnc are the predefined encoder based on fse tables as defined here:
-	// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#default-distributions
-	// These values are already transformed.
-	fsePredefEnc [3]fseEncoder
-
-	// symbolTableX contain the transformations needed for each type as defined in
-	// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#the-codes-for-literals-lengths-match-lengths-and-offsets
-	symbolTableX [3][]baseOffset
-
-	// maxTableSymbol is the biggest supported symbol for each table type
-	// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#the-codes-for-literals-lengths-match-lengths-and-offsets
-	maxTableSymbol = [3]uint8{tableLiteralLengths: maxLiteralLengthSymbol, tableOffsets: maxOffsetLengthSymbol, tableMatchLengths: maxMatchLengthSymbol}
-
-	// bitTables is the bits table for each table.
-	bitTables = [3][]byte{tableLiteralLengths: llBitsTable[:], tableOffsets: nil, tableMatchLengths: mlBitsTable[:]}
-)
-
-type tableIndex uint8
-
-const (
-	// indexes for fsePredef and symbolTableX
-	tableLiteralLengths tableIndex = 0
-	tableOffsets        tableIndex = 1
-	tableMatchLengths   tableIndex = 2
-
-	maxLiteralLengthSymbol = 35
-	maxOffsetLengthSymbol  = 30
-	maxMatchLengthSymbol   = 52
-)
-
-// baseOffset is used for calculating transformations.
-type baseOffset struct {
-	baseLine uint32
-	addBits  uint8
-}
-
-// fillBase will precalculate base offsets with the given bit distributions.
-func fillBase(dst []baseOffset, base uint32, bits ...uint8) {
-	if len(bits) != len(dst) {
-		panic(fmt.Sprintf("len(dst) (%d) != len(bits) (%d)", len(dst), len(bits)))
-	}
-	for i, bit := range bits {
-		if base > math.MaxInt32 {
-			panic("invalid decoding table, base overflows int32")
-		}
-
-		dst[i] = baseOffset{
-			baseLine: base,
-			addBits:  bit,
-		}
-		base += 1 << bit
-	}
-}
-
-var predef sync.Once
-
-func initPredefined() {
-	predef.Do(func() {
-		// Literals length codes
-		tmp := make([]baseOffset, 36)
-		for i := range tmp[:16] {
-			tmp[i] = baseOffset{
-				baseLine: uint32(i),
-				addBits:  0,
-			}
-		}
-		fillBase(tmp[16:], 16, 1, 1, 1, 1, 2, 2, 3, 3, 4, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16)
-		symbolTableX[tableLiteralLengths] = tmp
-
-		// Match length codes
-		tmp = make([]baseOffset, 53)
-		for i := range tmp[:32] {
-			tmp[i] = baseOffset{
-				// The transformation adds the 3 length.
-				baseLine: uint32(i) + 3,
-				addBits:  0,
-			}
-		}
-		fillBase(tmp[32:], 35, 1, 1, 1, 1, 2, 2, 3, 3, 4, 4, 5, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16)
-		symbolTableX[tableMatchLengths] = tmp
-
-		// Offset codes
-		tmp = make([]baseOffset, maxOffsetBits+1)
-		tmp[1] = baseOffset{
-			baseLine: 1,
-			addBits:  1,
-		}
-		fillBase(tmp[2:], 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)
-		symbolTableX[tableOffsets] = tmp
-
-		// Fill predefined tables and transform them.
-		// https://github.com/facebook/zstd/blob/dev/doc/zstd_compression_format.md#default-distributions
-		for i := range fsePredef[:] {
-			f := &fsePredef[i]
-			switch tableIndex(i) {
-			case tableLiteralLengths:
-				// https://github.com/facebook/zstd/blob/ededcfca57366461021c922720878c81a5854a0a/lib/decompress/zstd_decompress_block.c#L243
-				f.actualTableLog = 6
-				copy(f.norm[:], []int16{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})
-				f.symbolLen = 36
-			case tableOffsets:
-				// https://github.com/facebook/zstd/blob/ededcfca57366461021c922720878c81a5854a0a/lib/decompress/zstd_decompress_block.c#L281
-				f.actualTableLog = 5
-				copy(f.norm[:], []int16{
-					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})
-				f.symbolLen = 29
-			case tableMatchLengths:
-				//https://github.com/facebook/zstd/blob/ededcfca57366461021c922720878c81a5854a0a/lib/decompress/zstd_decompress_block.c#L304
-				f.actualTableLog = 6
-				copy(f.norm[:], []int16{
-					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})
-				f.symbolLen = 53
-			}
-			if err := f.buildDtable(); err != nil {
-				panic(fmt.Errorf("building table %v: %v", tableIndex(i), err))
-			}
-			if err := f.transform(symbolTableX[i]); err != nil {
-				panic(fmt.Errorf("building table %v: %v", tableIndex(i), err))
-			}
-			f.preDefined = true
-
-			// Create encoder as well
-			enc := &fsePredefEnc[i]
-			copy(enc.norm[:], f.norm[:])
-			enc.symbolLen = f.symbolLen
-			enc.actualTableLog = f.actualTableLog
-			if err := enc.buildCTable(); err != nil {
-				panic(fmt.Errorf("building encoding table %v: %v", tableIndex(i), err))
-			}
-			enc.setBits(bitTables[i])
-			enc.preDefined = true
-		}
-	})
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/hash.go b/vendor/github.com/klauspost/compress/zstd/hash.go
deleted file mode 100644
index 4a752067fc..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/hash.go
+++ /dev/null
@@ -1,77 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-const (
-	prime3bytes = 506832829
-	prime4bytes = 2654435761
-	prime5bytes = 889523592379
-	prime6bytes = 227718039650203
-	prime7bytes = 58295818150454627
-	prime8bytes = 0xcf1bbcdcb7a56463
-)
-
-// hashLen returns a hash of the lowest l bytes of u for a size size of h bytes.
-// l must be >=4 and <=8. Any other value will return hash for 4 bytes.
-// h should always be <32.
-// Preferably h and l should be a constant.
-// FIXME: This does NOT get resolved, if 'mls' is constant,
-//  so this cannot be used.
-func hashLen(u uint64, hashLog, mls uint8) uint32 {
-	switch mls {
-	case 5:
-		return hash5(u, hashLog)
-	case 6:
-		return hash6(u, hashLog)
-	case 7:
-		return hash7(u, hashLog)
-	case 8:
-		return hash8(u, hashLog)
-	default:
-		return hash4x64(u, hashLog)
-	}
-}
-
-// hash3 returns the hash of the lower 3 bytes of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <32.
-func hash3(u uint32, h uint8) uint32 {
-	return ((u << (32 - 24)) * prime3bytes) >> ((32 - h) & 31)
-}
-
-// hash4 returns the hash of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <32.
-func hash4(u uint32, h uint8) uint32 {
-	return (u * prime4bytes) >> ((32 - h) & 31)
-}
-
-// hash4x64 returns the hash of the lowest 4 bytes of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <32.
-func hash4x64(u uint64, h uint8) uint32 {
-	return (uint32(u) * prime4bytes) >> ((32 - h) & 31)
-}
-
-// hash5 returns the hash of the lowest 5 bytes of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <64.
-func hash5(u uint64, h uint8) uint32 {
-	return uint32(((u << (64 - 40)) * prime5bytes) >> ((64 - h) & 63))
-}
-
-// hash6 returns the hash of the lowest 6 bytes of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <64.
-func hash6(u uint64, h uint8) uint32 {
-	return uint32(((u << (64 - 48)) * prime6bytes) >> ((64 - h) & 63))
-}
-
-// hash7 returns the hash of the lowest 7 bytes of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <64.
-func hash7(u uint64, h uint8) uint32 {
-	return uint32(((u << (64 - 56)) * prime7bytes) >> ((64 - h) & 63))
-}
-
-// hash8 returns the hash of u to fit in a hash table with h bits.
-// Preferably h should be a constant and should always be <64.
-func hash8(u uint64, h uint8) uint32 {
-	return uint32((u * prime8bytes) >> ((64 - h) & 63))
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/history.go b/vendor/github.com/klauspost/compress/zstd/history.go
deleted file mode 100644
index f783e32d25..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/history.go
+++ /dev/null
@@ -1,89 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"github.com/klauspost/compress/huff0"
-)
-
-// history contains the information transferred between blocks.
-type history struct {
-	b             []byte
-	huffTree      *huff0.Scratch
-	recentOffsets [3]int
-	decoders      sequenceDecs
-	windowSize    int
-	maxSize       int
-	error         bool
-	dict          *dict
-}
-
-// reset will reset the history to initial state of a frame.
-// The history must already have been initialized to the desired size.
-func (h *history) reset() {
-	h.b = h.b[:0]
-	h.error = false
-	h.recentOffsets = [3]int{1, 4, 8}
-	if f := h.decoders.litLengths.fse; f != nil && !f.preDefined {
-		fseDecoderPool.Put(f)
-	}
-	if f := h.decoders.offsets.fse; f != nil && !f.preDefined {
-		fseDecoderPool.Put(f)
-	}
-	if f := h.decoders.matchLengths.fse; f != nil && !f.preDefined {
-		fseDecoderPool.Put(f)
-	}
-	h.decoders = sequenceDecs{}
-	if h.huffTree != nil {
-		if h.dict == nil || h.dict.litEnc != h.huffTree {
-			huffDecoderPool.Put(h.huffTree)
-		}
-	}
-	h.huffTree = nil
-	h.dict = nil
-	//printf("history created: %+v (l: %d, c: %d)", *h, len(h.b), cap(h.b))
-}
-
-func (h *history) setDict(dict *dict) {
-	if dict == nil {
-		return
-	}
-	h.dict = dict
-	h.decoders.litLengths = dict.llDec
-	h.decoders.offsets = dict.ofDec
-	h.decoders.matchLengths = dict.mlDec
-	h.recentOffsets = dict.offsets
-	h.huffTree = dict.litEnc
-}
-
-// append bytes to history.
-// This function will make sure there is space for it,
-// if the buffer has been allocated with enough extra space.
-func (h *history) append(b []byte) {
-	if len(b) >= h.windowSize {
-		// Discard all history by simply overwriting
-		h.b = h.b[:h.windowSize]
-		copy(h.b, b[len(b)-h.windowSize:])
-		return
-	}
-
-	// If there is space, append it.
-	if len(b) < cap(h.b)-len(h.b) {
-		h.b = append(h.b, b...)
-		return
-	}
-
-	// Move data down so we only have window size left.
-	// We know we have less than window size in b at this point.
-	discard := len(b) + len(h.b) - h.windowSize
-	copy(h.b, h.b[discard:])
-	h.b = h.b[:h.windowSize]
-	copy(h.b[h.windowSize-len(b):], b)
-}
-
-// append bytes to history without ever discarding anything.
-func (h *history) appendKeep(b []byte) {
-	h.b = append(h.b, b...)
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/LICENSE.txt b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/LICENSE.txt
deleted file mode 100644
index 24b53065f4..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/LICENSE.txt
+++ /dev/null
@@ -1,22 +0,0 @@
-Copyright (c) 2016 Caleb Spare
-
-MIT License
-
-Permission is hereby granted, free of charge, to any person obtaining
-a copy of this software and associated documentation files (the
-"Software"), to deal in the Software without restriction, including
-without limitation the rights to use, copy, modify, merge, publish,
-distribute, sublicense, and/or sell copies of the Software, and to
-permit persons to whom the Software is furnished to do so, subject to
-the following conditions:
-
-The above copyright notice and this permission notice shall be
-included in all copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
-EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
-MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
-NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
-LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
-OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
-WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/README.md b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/README.md
deleted file mode 100644
index 69aa3bb587..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/README.md
+++ /dev/null
@@ -1,58 +0,0 @@
-# xxhash
-
-VENDORED: Go to [github.com/cespare/xxhash](https://github.com/cespare/xxhash) for original package.
-
-
-[![GoDoc](https://godoc.org/github.com/cespare/xxhash?status.svg)](https://godoc.org/github.com/cespare/xxhash)
-[![Build Status](https://travis-ci.org/cespare/xxhash.svg?branch=master)](https://travis-ci.org/cespare/xxhash)
-
-xxhash is a Go implementation of the 64-bit
-[xxHash](http://cyan4973.github.io/xxHash/) algorithm, XXH64. This is a
-high-quality hashing algorithm that is much faster than anything in the Go
-standard library.
-
-This package provides a straightforward API:
-
-```
-func Sum64(b []byte) uint64
-func Sum64String(s string) uint64
-type Digest struct{ ... }
-    func New() *Digest
-```
-
-The `Digest` type implements hash.Hash64. Its key methods are:
-
-```
-func (*Digest) Write([]byte) (int, error)
-func (*Digest) WriteString(string) (int, error)
-func (*Digest) Sum64() uint64
-```
-
-This implementation provides a fast pure-Go implementation and an even faster
-assembly implementation for amd64.
-
-## Benchmarks
-
-Here are some quick benchmarks comparing the pure-Go and assembly
-implementations of Sum64.
-
-| input size | purego | asm |
-| --- | --- | --- |
-| 5 B   |  979.66 MB/s |  1291.17 MB/s  |
-| 100 B | 7475.26 MB/s | 7973.40 MB/s  |
-| 4 KB  | 17573.46 MB/s | 17602.65 MB/s |
-| 10 MB | 17131.46 MB/s | 17142.16 MB/s |
-
-These numbers were generated on Ubuntu 18.04 with an Intel i7-8700K CPU using
-the following commands under Go 1.11.2:
-
-```
-$ go test -tags purego -benchtime 10s -bench '/xxhash,direct,bytes'
-$ go test -benchtime 10s -bench '/xxhash,direct,bytes'
-```
-
-## Projects using this package
-
-- [InfluxDB](https://github.com/influxdata/influxdb)
-- [Prometheus](https://github.com/prometheus/prometheus)
-- [FreeCache](https://github.com/coocood/freecache)
diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash.go b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash.go
deleted file mode 100644
index 426b9cac78..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash.go
+++ /dev/null
@@ -1,238 +0,0 @@
-// Package xxhash implements the 64-bit variant of xxHash (XXH64) as described
-// at http://cyan4973.github.io/xxHash/.
-// THIS IS VENDORED: Go to github.com/cespare/xxhash for original package.
-
-package xxhash
-
-import (
-	"encoding/binary"
-	"errors"
-	"math/bits"
-)
-
-const (
-	prime1 uint64 = 11400714785074694791
-	prime2 uint64 = 14029467366897019727
-	prime3 uint64 = 1609587929392839161
-	prime4 uint64 = 9650029242287828579
-	prime5 uint64 = 2870177450012600261
-)
-
-// NOTE(caleb): I'm using both consts and vars of the primes. Using consts where
-// possible in the Go code is worth a small (but measurable) performance boost
-// by avoiding some MOVQs. Vars are needed for the asm and also are useful for
-// convenience in the Go code in a few places where we need to intentionally
-// avoid constant arithmetic (e.g., v1 := prime1 + prime2 fails because the
-// result overflows a uint64).
-var (
-	prime1v = prime1
-	prime2v = prime2
-	prime3v = prime3
-	prime4v = prime4
-	prime5v = prime5
-)
-
-// Digest implements hash.Hash64.
-type Digest struct {
-	v1    uint64
-	v2    uint64
-	v3    uint64
-	v4    uint64
-	total uint64
-	mem   [32]byte
-	n     int // how much of mem is used
-}
-
-// New creates a new Digest that computes the 64-bit xxHash algorithm.
-func New() *Digest {
-	var d Digest
-	d.Reset()
-	return &d
-}
-
-// Reset clears the Digest's state so that it can be reused.
-func (d *Digest) Reset() {
-	d.v1 = prime1v + prime2
-	d.v2 = prime2
-	d.v3 = 0
-	d.v4 = -prime1v
-	d.total = 0
-	d.n = 0
-}
-
-// Size always returns 8 bytes.
-func (d *Digest) Size() int { return 8 }
-
-// BlockSize always returns 32 bytes.
-func (d *Digest) BlockSize() int { return 32 }
-
-// Write adds more data to d. It always returns len(b), nil.
-func (d *Digest) Write(b []byte) (n int, err error) {
-	n = len(b)
-	d.total += uint64(n)
-
-	if d.n+n < 32 {
-		// This new data doesn't even fill the current block.
-		copy(d.mem[d.n:], b)
-		d.n += n
-		return
-	}
-
-	if d.n > 0 {
-		// Finish off the partial block.
-		copy(d.mem[d.n:], b)
-		d.v1 = round(d.v1, u64(d.mem[0:8]))
-		d.v2 = round(d.v2, u64(d.mem[8:16]))
-		d.v3 = round(d.v3, u64(d.mem[16:24]))
-		d.v4 = round(d.v4, u64(d.mem[24:32]))
-		b = b[32-d.n:]
-		d.n = 0
-	}
-
-	if len(b) >= 32 {
-		// One or more full blocks left.
-		nw := writeBlocks(d, b)
-		b = b[nw:]
-	}
-
-	// Store any remaining partial block.
-	copy(d.mem[:], b)
-	d.n = len(b)
-
-	return
-}
-
-// Sum appends the current hash to b and returns the resulting slice.
-func (d *Digest) Sum(b []byte) []byte {
-	s := d.Sum64()
-	return append(
-		b,
-		byte(s>>56),
-		byte(s>>48),
-		byte(s>>40),
-		byte(s>>32),
-		byte(s>>24),
-		byte(s>>16),
-		byte(s>>8),
-		byte(s),
-	)
-}
-
-// Sum64 returns the current hash.
-func (d *Digest) Sum64() uint64 {
-	var h uint64
-
-	if d.total >= 32 {
-		v1, v2, v3, v4 := d.v1, d.v2, d.v3, d.v4
-		h = rol1(v1) + rol7(v2) + rol12(v3) + rol18(v4)
-		h = mergeRound(h, v1)
-		h = mergeRound(h, v2)
-		h = mergeRound(h, v3)
-		h = mergeRound(h, v4)
-	} else {
-		h = d.v3 + prime5
-	}
-
-	h += d.total
-
-	i, end := 0, d.n
-	for ; i+8 <= end; i += 8 {
-		k1 := round(0, u64(d.mem[i:i+8]))
-		h ^= k1
-		h = rol27(h)*prime1 + prime4
-	}
-	if i+4 <= end {
-		h ^= uint64(u32(d.mem[i:i+4])) * prime1
-		h = rol23(h)*prime2 + prime3
-		i += 4
-	}
-	for i < end {
-		h ^= uint64(d.mem[i]) * prime5
-		h = rol11(h) * prime1
-		i++
-	}
-
-	h ^= h >> 33
-	h *= prime2
-	h ^= h >> 29
-	h *= prime3
-	h ^= h >> 32
-
-	return h
-}
-
-const (
-	magic         = "xxh\x06"
-	marshaledSize = len(magic) + 8*5 + 32
-)
-
-// MarshalBinary implements the encoding.BinaryMarshaler interface.
-func (d *Digest) MarshalBinary() ([]byte, error) {
-	b := make([]byte, 0, marshaledSize)
-	b = append(b, magic...)
-	b = appendUint64(b, d.v1)
-	b = appendUint64(b, d.v2)
-	b = appendUint64(b, d.v3)
-	b = appendUint64(b, d.v4)
-	b = appendUint64(b, d.total)
-	b = append(b, d.mem[:d.n]...)
-	b = b[:len(b)+len(d.mem)-d.n]
-	return b, nil
-}
-
-// UnmarshalBinary implements the encoding.BinaryUnmarshaler interface.
-func (d *Digest) UnmarshalBinary(b []byte) error {
-	if len(b) < len(magic) || string(b[:len(magic)]) != magic {
-		return errors.New("xxhash: invalid hash state identifier")
-	}
-	if len(b) != marshaledSize {
-		return errors.New("xxhash: invalid hash state size")
-	}
-	b = b[len(magic):]
-	b, d.v1 = consumeUint64(b)
-	b, d.v2 = consumeUint64(b)
-	b, d.v3 = consumeUint64(b)
-	b, d.v4 = consumeUint64(b)
-	b, d.total = consumeUint64(b)
-	copy(d.mem[:], b)
-	b = b[len(d.mem):]
-	d.n = int(d.total % uint64(len(d.mem)))
-	return nil
-}
-
-func appendUint64(b []byte, x uint64) []byte {
-	var a [8]byte
-	binary.LittleEndian.PutUint64(a[:], x)
-	return append(b, a[:]...)
-}
-
-func consumeUint64(b []byte) ([]byte, uint64) {
-	x := u64(b)
-	return b[8:], x
-}
-
-func u64(b []byte) uint64 { return binary.LittleEndian.Uint64(b) }
-func u32(b []byte) uint32 { return binary.LittleEndian.Uint32(b) }
-
-func round(acc, input uint64) uint64 {
-	acc += input * prime2
-	acc = rol31(acc)
-	acc *= prime1
-	return acc
-}
-
-func mergeRound(acc, val uint64) uint64 {
-	val = round(0, val)
-	acc ^= val
-	acc = acc*prime1 + prime4
-	return acc
-}
-
-func rol1(x uint64) uint64  { return bits.RotateLeft64(x, 1) }
-func rol7(x uint64) uint64  { return bits.RotateLeft64(x, 7) }
-func rol11(x uint64) uint64 { return bits.RotateLeft64(x, 11) }
-func rol12(x uint64) uint64 { return bits.RotateLeft64(x, 12) }
-func rol18(x uint64) uint64 { return bits.RotateLeft64(x, 18) }
-func rol23(x uint64) uint64 { return bits.RotateLeft64(x, 23) }
-func rol27(x uint64) uint64 { return bits.RotateLeft64(x, 27) }
-func rol31(x uint64) uint64 { return bits.RotateLeft64(x, 31) }
diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.go b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.go
deleted file mode 100644
index 35318d7c46..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.go
+++ /dev/null
@@ -1,13 +0,0 @@
-// +build !appengine
-// +build gc
-// +build !purego
-
-package xxhash
-
-// Sum64 computes the 64-bit xxHash digest of b.
-//
-//go:noescape
-func Sum64(b []byte) uint64
-
-//go:noescape
-func writeBlocks(*Digest, []byte) int
diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.s b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.s
deleted file mode 100644
index 2c9c5357a1..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_amd64.s
+++ /dev/null
@@ -1,215 +0,0 @@
-// +build !appengine
-// +build gc
-// +build !purego
-
-#include "textflag.h"
-
-// Register allocation:
-// AX	h
-// CX	pointer to advance through b
-// DX	n
-// BX	loop end
-// R8	v1, k1
-// R9	v2
-// R10	v3
-// R11	v4
-// R12	tmp
-// R13	prime1v
-// R14	prime2v
-// R15	prime4v
-
-// round reads from and advances the buffer pointer in CX.
-// It assumes that R13 has prime1v and R14 has prime2v.
-#define round(r) \
-	MOVQ  (CX), R12 \
-	ADDQ  $8, CX    \
-	IMULQ R14, R12  \
-	ADDQ  R12, r    \
-	ROLQ  $31, r    \
-	IMULQ R13, r
-
-// mergeRound applies a merge round on the two registers acc and val.
-// It assumes that R13 has prime1v, R14 has prime2v, and R15 has prime4v.
-#define mergeRound(acc, val) \
-	IMULQ R14, val \
-	ROLQ  $31, val \
-	IMULQ R13, val \
-	XORQ  val, acc \
-	IMULQ R13, acc \
-	ADDQ  R15, acc
-
-// func Sum64(b []byte) uint64
-TEXT ·Sum64(SB), NOSPLIT, $0-32
-	// Load fixed primes.
-	MOVQ ·prime1v(SB), R13
-	MOVQ ·prime2v(SB), R14
-	MOVQ ·prime4v(SB), R15
-
-	// Load slice.
-	MOVQ b_base+0(FP), CX
-	MOVQ b_len+8(FP), DX
-	LEAQ (CX)(DX*1), BX
-
-	// The first loop limit will be len(b)-32.
-	SUBQ $32, BX
-
-	// Check whether we have at least one block.
-	CMPQ DX, $32
-	JLT  noBlocks
-
-	// Set up initial state (v1, v2, v3, v4).
-	MOVQ R13, R8
-	ADDQ R14, R8
-	MOVQ R14, R9
-	XORQ R10, R10
-	XORQ R11, R11
-	SUBQ R13, R11
-
-	// Loop until CX > BX.
-blockLoop:
-	round(R8)
-	round(R9)
-	round(R10)
-	round(R11)
-
-	CMPQ CX, BX
-	JLE  blockLoop
-
-	MOVQ R8, AX
-	ROLQ $1, AX
-	MOVQ R9, R12
-	ROLQ $7, R12
-	ADDQ R12, AX
-	MOVQ R10, R12
-	ROLQ $12, R12
-	ADDQ R12, AX
-	MOVQ R11, R12
-	ROLQ $18, R12
-	ADDQ R12, AX
-
-	mergeRound(AX, R8)
-	mergeRound(AX, R9)
-	mergeRound(AX, R10)
-	mergeRound(AX, R11)
-
-	JMP afterBlocks
-
-noBlocks:
-	MOVQ ·prime5v(SB), AX
-
-afterBlocks:
-	ADDQ DX, AX
-
-	// Right now BX has len(b)-32, and we want to loop until CX > len(b)-8.
-	ADDQ $24, BX
-
-	CMPQ CX, BX
-	JG   fourByte
-
-wordLoop:
-	// Calculate k1.
-	MOVQ  (CX), R8
-	ADDQ  $8, CX
-	IMULQ R14, R8
-	ROLQ  $31, R8
-	IMULQ R13, R8
-
-	XORQ  R8, AX
-	ROLQ  $27, AX
-	IMULQ R13, AX
-	ADDQ  R15, AX
-
-	CMPQ CX, BX
-	JLE  wordLoop
-
-fourByte:
-	ADDQ $4, BX
-	CMPQ CX, BX
-	JG   singles
-
-	MOVL  (CX), R8
-	ADDQ  $4, CX
-	IMULQ R13, R8
-	XORQ  R8, AX
-
-	ROLQ  $23, AX
-	IMULQ R14, AX
-	ADDQ  ·prime3v(SB), AX
-
-singles:
-	ADDQ $4, BX
-	CMPQ CX, BX
-	JGE  finalize
-
-singlesLoop:
-	MOVBQZX (CX), R12
-	ADDQ    $1, CX
-	IMULQ   ·prime5v(SB), R12
-	XORQ    R12, AX
-
-	ROLQ  $11, AX
-	IMULQ R13, AX
-
-	CMPQ CX, BX
-	JL   singlesLoop
-
-finalize:
-	MOVQ  AX, R12
-	SHRQ  $33, R12
-	XORQ  R12, AX
-	IMULQ R14, AX
-	MOVQ  AX, R12
-	SHRQ  $29, R12
-	XORQ  R12, AX
-	IMULQ ·prime3v(SB), AX
-	MOVQ  AX, R12
-	SHRQ  $32, R12
-	XORQ  R12, AX
-
-	MOVQ AX, ret+24(FP)
-	RET
-
-// writeBlocks uses the same registers as above except that it uses AX to store
-// the d pointer.
-
-// func writeBlocks(d *Digest, b []byte) int
-TEXT ·writeBlocks(SB), NOSPLIT, $0-40
-	// Load fixed primes needed for round.
-	MOVQ ·prime1v(SB), R13
-	MOVQ ·prime2v(SB), R14
-
-	// Load slice.
-	MOVQ arg1_base+8(FP), CX
-	MOVQ arg1_len+16(FP), DX
-	LEAQ (CX)(DX*1), BX
-	SUBQ $32, BX
-
-	// Load vN from d.
-	MOVQ arg+0(FP), AX
-	MOVQ 0(AX), R8   // v1
-	MOVQ 8(AX), R9   // v2
-	MOVQ 16(AX), R10 // v3
-	MOVQ 24(AX), R11 // v4
-
-	// We don't need to check the loop condition here; this function is
-	// always called with at least one block of data to process.
-blockLoop:
-	round(R8)
-	round(R9)
-	round(R10)
-	round(R11)
-
-	CMPQ CX, BX
-	JLE  blockLoop
-
-	// Copy vN back to d.
-	MOVQ R8, 0(AX)
-	MOVQ R9, 8(AX)
-	MOVQ R10, 16(AX)
-	MOVQ R11, 24(AX)
-
-	// The number of bytes written is CX minus the old base pointer.
-	SUBQ arg1_base+8(FP), CX
-	MOVQ CX, ret+32(FP)
-
-	RET
diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_other.go b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_other.go
deleted file mode 100644
index 4a5a821603..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_other.go
+++ /dev/null
@@ -1,76 +0,0 @@
-// +build !amd64 appengine !gc purego
-
-package xxhash
-
-// Sum64 computes the 64-bit xxHash digest of b.
-func Sum64(b []byte) uint64 {
-	// A simpler version would be
-	//   d := New()
-	//   d.Write(b)
-	//   return d.Sum64()
-	// but this is faster, particularly for small inputs.
-
-	n := len(b)
-	var h uint64
-
-	if n >= 32 {
-		v1 := prime1v + prime2
-		v2 := prime2
-		v3 := uint64(0)
-		v4 := -prime1v
-		for len(b) >= 32 {
-			v1 = round(v1, u64(b[0:8:len(b)]))
-			v2 = round(v2, u64(b[8:16:len(b)]))
-			v3 = round(v3, u64(b[16:24:len(b)]))
-			v4 = round(v4, u64(b[24:32:len(b)]))
-			b = b[32:len(b):len(b)]
-		}
-		h = rol1(v1) + rol7(v2) + rol12(v3) + rol18(v4)
-		h = mergeRound(h, v1)
-		h = mergeRound(h, v2)
-		h = mergeRound(h, v3)
-		h = mergeRound(h, v4)
-	} else {
-		h = prime5
-	}
-
-	h += uint64(n)
-
-	i, end := 0, len(b)
-	for ; i+8 <= end; i += 8 {
-		k1 := round(0, u64(b[i:i+8:len(b)]))
-		h ^= k1
-		h = rol27(h)*prime1 + prime4
-	}
-	if i+4 <= end {
-		h ^= uint64(u32(b[i:i+4:len(b)])) * prime1
-		h = rol23(h)*prime2 + prime3
-		i += 4
-	}
-	for ; i < end; i++ {
-		h ^= uint64(b[i]) * prime5
-		h = rol11(h) * prime1
-	}
-
-	h ^= h >> 33
-	h *= prime2
-	h ^= h >> 29
-	h *= prime3
-	h ^= h >> 32
-
-	return h
-}
-
-func writeBlocks(d *Digest, b []byte) int {
-	v1, v2, v3, v4 := d.v1, d.v2, d.v3, d.v4
-	n := len(b)
-	for len(b) >= 32 {
-		v1 = round(v1, u64(b[0:8:len(b)]))
-		v2 = round(v2, u64(b[8:16:len(b)]))
-		v3 = round(v3, u64(b[16:24:len(b)]))
-		v4 = round(v4, u64(b[24:32:len(b)]))
-		b = b[32:len(b):len(b)]
-	}
-	d.v1, d.v2, d.v3, d.v4 = v1, v2, v3, v4
-	return n - len(b)
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_safe.go b/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_safe.go
deleted file mode 100644
index 6f3b0cb102..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/internal/xxhash/xxhash_safe.go
+++ /dev/null
@@ -1,11 +0,0 @@
-package xxhash
-
-// Sum64String computes the 64-bit xxHash digest of s.
-func Sum64String(s string) uint64 {
-	return Sum64([]byte(s))
-}
-
-// WriteString adds more data to d. It always returns len(s), nil.
-func (d *Digest) WriteString(s string) (n int, err error) {
-	return d.Write([]byte(s))
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/seqdec.go b/vendor/github.com/klauspost/compress/zstd/seqdec.go
deleted file mode 100644
index 1dd39e63b7..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/seqdec.go
+++ /dev/null
@@ -1,492 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"errors"
-	"fmt"
-	"io"
-)
-
-type seq struct {
-	litLen   uint32
-	matchLen uint32
-	offset   uint32
-
-	// Codes are stored here for the encoder
-	// so they only have to be looked up once.
-	llCode, mlCode, ofCode uint8
-}
-
-func (s seq) String() string {
-	if s.offset <= 3 {
-		if s.offset == 0 {
-			return fmt.Sprint("litLen:", s.litLen, ", matchLen:", s.matchLen+zstdMinMatch, ", offset: INVALID (0)")
-		}
-		return fmt.Sprint("litLen:", s.litLen, ", matchLen:", s.matchLen+zstdMinMatch, ", offset:", s.offset, " (repeat)")
-	}
-	return fmt.Sprint("litLen:", s.litLen, ", matchLen:", s.matchLen+zstdMinMatch, ", offset:", s.offset-3, " (new)")
-}
-
-type seqCompMode uint8
-
-const (
-	compModePredefined seqCompMode = iota
-	compModeRLE
-	compModeFSE
-	compModeRepeat
-)
-
-type sequenceDec struct {
-	// decoder keeps track of the current state and updates it from the bitstream.
-	fse    *fseDecoder
-	state  fseState
-	repeat bool
-}
-
-// init the state of the decoder with input from stream.
-func (s *sequenceDec) init(br *bitReader) error {
-	if s.fse == nil {
-		return errors.New("sequence decoder not defined")
-	}
-	s.state.init(br, s.fse.actualTableLog, s.fse.dt[:1<<s.fse.actualTableLog])
-	return nil
-}
-
-// sequenceDecs contains all 3 sequence decoders and their state.
-type sequenceDecs struct {
-	litLengths   sequenceDec
-	offsets      sequenceDec
-	matchLengths sequenceDec
-	prevOffset   [3]int
-	hist         []byte
-	dict         []byte
-	literals     []byte
-	out          []byte
-	windowSize   int
-	maxBits      uint8
-}
-
-// initialize all 3 decoders from the stream input.
-func (s *sequenceDecs) initialize(br *bitReader, hist *history, literals, out []byte) error {
-	if err := s.litLengths.init(br); err != nil {
-		return errors.New("litLengths:" + err.Error())
-	}
-	if err := s.offsets.init(br); err != nil {
-		return errors.New("offsets:" + err.Error())
-	}
-	if err := s.matchLengths.init(br); err != nil {
-		return errors.New("matchLengths:" + err.Error())
-	}
-	s.literals = literals
-	s.hist = hist.b
-	s.prevOffset = hist.recentOffsets
-	s.maxBits = s.litLengths.fse.maxBits + s.offsets.fse.maxBits + s.matchLengths.fse.maxBits
-	s.windowSize = hist.windowSize
-	s.out = out
-	s.dict = nil
-	if hist.dict != nil {
-		s.dict = hist.dict.content
-	}
-	return nil
-}
-
-// decode sequences from the stream with the provided history.
-func (s *sequenceDecs) decode(seqs int, br *bitReader, hist []byte) error {
-	startSize := len(s.out)
-	// Grab full sizes tables, to avoid bounds checks.
-	llTable, mlTable, ofTable := s.litLengths.fse.dt[:maxTablesize], s.matchLengths.fse.dt[:maxTablesize], s.offsets.fse.dt[:maxTablesize]
-	llState, mlState, ofState := s.litLengths.state.state, s.matchLengths.state.state, s.offsets.state.state
-
-	for i := seqs - 1; i >= 0; i-- {
-		if br.overread() {
-			printf("reading sequence %d, exceeded available data\n", seqs-i)
-			return io.ErrUnexpectedEOF
-		}
-		var ll, mo, ml int
-		if br.off > 4+((maxOffsetBits+16+16)>>3) {
-			// inlined function:
-			// ll, mo, ml = s.nextFast(br, llState, mlState, ofState)
-
-			// Final will not read from stream.
-			var llB, mlB, moB uint8
-			ll, llB = llState.final()
-			ml, mlB = mlState.final()
-			mo, moB = ofState.final()
-
-			// extra bits are stored in reverse order.
-			br.fillFast()
-			mo += br.getBits(moB)
-			if s.maxBits > 32 {
-				br.fillFast()
-			}
-			ml += br.getBits(mlB)
-			ll += br.getBits(llB)
-
-			if moB > 1 {
-				s.prevOffset[2] = s.prevOffset[1]
-				s.prevOffset[1] = s.prevOffset[0]
-				s.prevOffset[0] = mo
-			} else {
-				// mo = s.adjustOffset(mo, ll, moB)
-				// Inlined for rather big speedup
-				if ll == 0 {
-					// There is an exception though, when current sequence's literals_length = 0.
-					// In this case, repeated offsets are shifted by one, so an offset_value of 1 means Repeated_Offset2,
-					// an offset_value of 2 means Repeated_Offset3, and an offset_value of 3 means Repeated_Offset1 - 1_byte.
-					mo++
-				}
-
-				if mo == 0 {
-					mo = s.prevOffset[0]
-				} else {
-					var temp int
-					if mo == 3 {
-						temp = s.prevOffset[0] - 1
-					} else {
-						temp = s.prevOffset[mo]
-					}
-
-					if temp == 0 {
-						// 0 is not valid; input is corrupted; force offset to 1
-						println("temp was 0")
-						temp = 1
-					}
-
-					if mo != 1 {
-						s.prevOffset[2] = s.prevOffset[1]
-					}
-					s.prevOffset[1] = s.prevOffset[0]
-					s.prevOffset[0] = temp
-					mo = temp
-				}
-			}
-			br.fillFast()
-		} else {
-			ll, mo, ml = s.next(br, llState, mlState, ofState)
-			br.fill()
-		}
-
-		if debugSequences {
-			println("Seq", seqs-i-1, "Litlen:", ll, "mo:", mo, "(abs) ml:", ml)
-		}
-
-		if ll > len(s.literals) {
-			return fmt.Errorf("unexpected literal count, want %d bytes, but only %d is available", ll, len(s.literals))
-		}
-		size := ll + ml + len(s.out)
-		if size-startSize > maxBlockSize {
-			return fmt.Errorf("output (%d) bigger than max block size", size)
-		}
-		if size > cap(s.out) {
-			// Not enough size, which can happen under high volume block streaming conditions
-			// but could be if destination slice is too small for sync operations.
-			// over-allocating here can create a large amount of GC pressure so we try to keep
-			// it as contained as possible
-			used := len(s.out) - startSize
-			addBytes := 256 + ll + ml + used>>2
-			// Clamp to max block size.
-			if used+addBytes > maxBlockSize {
-				addBytes = maxBlockSize - used
-			}
-			s.out = append(s.out, make([]byte, addBytes)...)
-			s.out = s.out[:len(s.out)-addBytes]
-		}
-		if ml > maxMatchLen {
-			return fmt.Errorf("match len (%d) bigger than max allowed length", ml)
-		}
-
-		// Add literals
-		s.out = append(s.out, s.literals[:ll]...)
-		s.literals = s.literals[ll:]
-		out := s.out
-
-		if mo == 0 && ml > 0 {
-			return fmt.Errorf("zero matchoff and matchlen (%d) > 0", ml)
-		}
-
-		if mo > len(s.out)+len(hist) || mo > s.windowSize {
-			if len(s.dict) == 0 {
-				return fmt.Errorf("match offset (%d) bigger than current history (%d)", mo, len(s.out)+len(hist))
-			}
-
-			// we may be in dictionary.
-			dictO := len(s.dict) - (mo - (len(s.out) + len(hist)))
-			if dictO < 0 || dictO >= len(s.dict) {
-				return fmt.Errorf("match offset (%d) bigger than current history (%d)", mo, len(s.out)+len(hist))
-			}
-			end := dictO + ml
-			if end > len(s.dict) {
-				out = append(out, s.dict[dictO:]...)
-				mo -= len(s.dict) - dictO
-				ml -= len(s.dict) - dictO
-			} else {
-				out = append(out, s.dict[dictO:end]...)
-				mo = 0
-				ml = 0
-			}
-		}
-
-		// Copy from history.
-		// TODO: Blocks without history could be made to ignore this completely.
-		if v := mo - len(s.out); v > 0 {
-			// v is the start position in history from end.
-			start := len(s.hist) - v
-			if ml > v {
-				// Some goes into current block.
-				// Copy remainder of history
-				out = append(out, s.hist[start:]...)
-				mo -= v
-				ml -= v
-			} else {
-				out = append(out, s.hist[start:start+ml]...)
-				ml = 0
-			}
-		}
-		// We must be in current buffer now
-		if ml > 0 {
-			start := len(s.out) - mo
-			if ml <= len(s.out)-start {
-				// No overlap
-				out = append(out, s.out[start:start+ml]...)
-			} else {
-				// Overlapping copy
-				// Extend destination slice and copy one byte at the time.
-				out = out[:len(out)+ml]
-				src := out[start : start+ml]
-				// Destination is the space we just added.
-				dst := out[len(out)-ml:]
-				dst = dst[:len(src)]
-				for i := range src {
-					dst[i] = src[i]
-				}
-			}
-		}
-		s.out = out
-		if i == 0 {
-			// This is the last sequence, so we shouldn't update state.
-			break
-		}
-
-		// Manually inlined, ~ 5-20% faster
-		// Update all 3 states at once. Approx 20% faster.
-		nBits := llState.nbBits() + mlState.nbBits() + ofState.nbBits()
-		if nBits == 0 {
-			llState = llTable[llState.newState()&maxTableMask]
-			mlState = mlTable[mlState.newState()&maxTableMask]
-			ofState = ofTable[ofState.newState()&maxTableMask]
-		} else {
-			bits := br.getBitsFast(nBits)
-			lowBits := uint16(bits >> ((ofState.nbBits() + mlState.nbBits()) & 31))
-			llState = llTable[(llState.newState()+lowBits)&maxTableMask]
-
-			lowBits = uint16(bits >> (ofState.nbBits() & 31))
-			lowBits &= bitMask[mlState.nbBits()&15]
-			mlState = mlTable[(mlState.newState()+lowBits)&maxTableMask]
-
-			lowBits = uint16(bits) & bitMask[ofState.nbBits()&15]
-			ofState = ofTable[(ofState.newState()+lowBits)&maxTableMask]
-		}
-	}
-
-	// Add final literals
-	s.out = append(s.out, s.literals...)
-	return nil
-}
-
-// update states, at least 27 bits must be available.
-func (s *sequenceDecs) update(br *bitReader) {
-	// Max 8 bits
-	s.litLengths.state.next(br)
-	// Max 9 bits
-	s.matchLengths.state.next(br)
-	// Max 8 bits
-	s.offsets.state.next(br)
-}
-
-var bitMask [16]uint16
-
-func init() {
-	for i := range bitMask[:] {
-		bitMask[i] = uint16((1 << uint(i)) - 1)
-	}
-}
-
-// update states, at least 27 bits must be available.
-func (s *sequenceDecs) updateAlt(br *bitReader) {
-	// Update all 3 states at once. Approx 20% faster.
-	a, b, c := s.litLengths.state.state, s.matchLengths.state.state, s.offsets.state.state
-
-	nBits := a.nbBits() + b.nbBits() + c.nbBits()
-	if nBits == 0 {
-		s.litLengths.state.state = s.litLengths.state.dt[a.newState()]
-		s.matchLengths.state.state = s.matchLengths.state.dt[b.newState()]
-		s.offsets.state.state = s.offsets.state.dt[c.newState()]
-		return
-	}
-	bits := br.getBitsFast(nBits)
-	lowBits := uint16(bits >> ((c.nbBits() + b.nbBits()) & 31))
-	s.litLengths.state.state = s.litLengths.state.dt[a.newState()+lowBits]
-
-	lowBits = uint16(bits >> (c.nbBits() & 31))
-	lowBits &= bitMask[b.nbBits()&15]
-	s.matchLengths.state.state = s.matchLengths.state.dt[b.newState()+lowBits]
-
-	lowBits = uint16(bits) & bitMask[c.nbBits()&15]
-	s.offsets.state.state = s.offsets.state.dt[c.newState()+lowBits]
-}
-
-// nextFast will return new states when there are at least 4 unused bytes left on the stream when done.
-func (s *sequenceDecs) nextFast(br *bitReader, llState, mlState, ofState decSymbol) (ll, mo, ml int) {
-	// Final will not read from stream.
-	ll, llB := llState.final()
-	ml, mlB := mlState.final()
-	mo, moB := ofState.final()
-
-	// extra bits are stored in reverse order.
-	br.fillFast()
-	mo += br.getBits(moB)
-	if s.maxBits > 32 {
-		br.fillFast()
-	}
-	ml += br.getBits(mlB)
-	ll += br.getBits(llB)
-
-	if moB > 1 {
-		s.prevOffset[2] = s.prevOffset[1]
-		s.prevOffset[1] = s.prevOffset[0]
-		s.prevOffset[0] = mo
-		return
-	}
-	// mo = s.adjustOffset(mo, ll, moB)
-	// Inlined for rather big speedup
-	if ll == 0 {
-		// There is an exception though, when current sequence's literals_length = 0.
-		// In this case, repeated offsets are shifted by one, so an offset_value of 1 means Repeated_Offset2,
-		// an offset_value of 2 means Repeated_Offset3, and an offset_value of 3 means Repeated_Offset1 - 1_byte.
-		mo++
-	}
-
-	if mo == 0 {
-		mo = s.prevOffset[0]
-		return
-	}
-	var temp int
-	if mo == 3 {
-		temp = s.prevOffset[0] - 1
-	} else {
-		temp = s.prevOffset[mo]
-	}
-
-	if temp == 0 {
-		// 0 is not valid; input is corrupted; force offset to 1
-		println("temp was 0")
-		temp = 1
-	}
-
-	if mo != 1 {
-		s.prevOffset[2] = s.prevOffset[1]
-	}
-	s.prevOffset[1] = s.prevOffset[0]
-	s.prevOffset[0] = temp
-	mo = temp
-	return
-}
-
-func (s *sequenceDecs) next(br *bitReader, llState, mlState, ofState decSymbol) (ll, mo, ml int) {
-	// Final will not read from stream.
-	ll, llB := llState.final()
-	ml, mlB := mlState.final()
-	mo, moB := ofState.final()
-
-	// extra bits are stored in reverse order.
-	br.fill()
-	if s.maxBits <= 32 {
-		mo += br.getBits(moB)
-		ml += br.getBits(mlB)
-		ll += br.getBits(llB)
-	} else {
-		mo += br.getBits(moB)
-		br.fill()
-		// matchlength+literal length, max 32 bits
-		ml += br.getBits(mlB)
-		ll += br.getBits(llB)
-
-	}
-	mo = s.adjustOffset(mo, ll, moB)
-	return
-}
-
-func (s *sequenceDecs) adjustOffset(offset, litLen int, offsetB uint8) int {
-	if offsetB > 1 {
-		s.prevOffset[2] = s.prevOffset[1]
-		s.prevOffset[1] = s.prevOffset[0]
-		s.prevOffset[0] = offset
-		return offset
-	}
-
-	if litLen == 0 {
-		// There is an exception though, when current sequence's literals_length = 0.
-		// In this case, repeated offsets are shifted by one, so an offset_value of 1 means Repeated_Offset2,
-		// an offset_value of 2 means Repeated_Offset3, and an offset_value of 3 means Repeated_Offset1 - 1_byte.
-		offset++
-	}
-
-	if offset == 0 {
-		return s.prevOffset[0]
-	}
-	var temp int
-	if offset == 3 {
-		temp = s.prevOffset[0] - 1
-	} else {
-		temp = s.prevOffset[offset]
-	}
-
-	if temp == 0 {
-		// 0 is not valid; input is corrupted; force offset to 1
-		println("temp was 0")
-		temp = 1
-	}
-
-	if offset != 1 {
-		s.prevOffset[2] = s.prevOffset[1]
-	}
-	s.prevOffset[1] = s.prevOffset[0]
-	s.prevOffset[0] = temp
-	return temp
-}
-
-// mergeHistory will merge history.
-func (s *sequenceDecs) mergeHistory(hist *sequenceDecs) (*sequenceDecs, error) {
-	for i := uint(0); i < 3; i++ {
-		var sNew, sHist *sequenceDec
-		switch i {
-		default:
-			// same as "case 0":
-			sNew = &s.litLengths
-			sHist = &hist.litLengths
-		case 1:
-			sNew = &s.offsets
-			sHist = &hist.offsets
-		case 2:
-			sNew = &s.matchLengths
-			sHist = &hist.matchLengths
-		}
-		if sNew.repeat {
-			if sHist.fse == nil {
-				return nil, fmt.Errorf("sequence stream %d, repeat requested, but no history", i)
-			}
-			continue
-		}
-		if sNew.fse == nil {
-			return nil, fmt.Errorf("sequence stream %d, no fse found", i)
-		}
-		if sHist.fse != nil && !sHist.fse.preDefined {
-			fseDecoderPool.Put(sHist.fse)
-		}
-		sHist.fse = sNew.fse
-	}
-	return hist, nil
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/seqenc.go b/vendor/github.com/klauspost/compress/zstd/seqenc.go
deleted file mode 100644
index 8014174a77..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/seqenc.go
+++ /dev/null
@@ -1,114 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import "math/bits"
-
-type seqCoders struct {
-	llEnc, ofEnc, mlEnc    *fseEncoder
-	llPrev, ofPrev, mlPrev *fseEncoder
-}
-
-// swap coders with another (block).
-func (s *seqCoders) swap(other *seqCoders) {
-	*s, *other = *other, *s
-}
-
-// setPrev will update the previous encoders to the actually used ones
-// and make sure a fresh one is in the main slot.
-func (s *seqCoders) setPrev(ll, ml, of *fseEncoder) {
-	compareSwap := func(used *fseEncoder, current, prev **fseEncoder) {
-		// We used the new one, more current to history and reuse the previous history
-		if *current == used {
-			*prev, *current = *current, *prev
-			c := *current
-			p := *prev
-			c.reUsed = false
-			p.reUsed = true
-			return
-		}
-		if used == *prev {
-			return
-		}
-		// Ensure we cannot reuse by accident
-		prevEnc := *prev
-		prevEnc.symbolLen = 0
-	}
-	compareSwap(ll, &s.llEnc, &s.llPrev)
-	compareSwap(ml, &s.mlEnc, &s.mlPrev)
-	compareSwap(of, &s.ofEnc, &s.ofPrev)
-}
-
-func highBit(val uint32) (n uint32) {
-	return uint32(bits.Len32(val) - 1)
-}
-
-var llCodeTable = [64]byte{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}
-
-// Up to 6 bits
-const maxLLCode = 35
-
-// llBitsTable translates from ll code to number of bits.
-var llBitsTable = [maxLLCode + 1]byte{
-	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}
-
-// llCode returns the code that represents the literal length requested.
-func llCode(litLength uint32) uint8 {
-	const llDeltaCode = 19
-	if litLength <= 63 {
-		// Compiler insists on bounds check (Go 1.12)
-		return llCodeTable[litLength&63]
-	}
-	return uint8(highBit(litLength)) + llDeltaCode
-}
-
-var mlCodeTable = [128]byte{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}
-
-// Up to 6 bits
-const maxMLCode = 52
-
-// mlBitsTable translates from ml code to number of bits.
-var mlBitsTable = [maxMLCode + 1]byte{
-	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}
-
-// note : mlBase = matchLength - MINMATCH;
-// because it's the format it's stored in seqStore->sequences
-func mlCode(mlBase uint32) uint8 {
-	const mlDeltaCode = 36
-	if mlBase <= 127 {
-		// Compiler insists on bounds check (Go 1.12)
-		return mlCodeTable[mlBase&127]
-	}
-	return uint8(highBit(mlBase)) + mlDeltaCode
-}
-
-func ofCode(offset uint32) uint8 {
-	// A valid offset will always be > 0.
-	return uint8(bits.Len32(offset) - 1)
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/snappy.go b/vendor/github.com/klauspost/compress/zstd/snappy.go
deleted file mode 100644
index 0372b1714a..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/snappy.go
+++ /dev/null
@@ -1,435 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-// Based on work by Yann Collet, released under BSD License.
-
-package zstd
-
-import (
-	"encoding/binary"
-	"errors"
-	"hash/crc32"
-	"io"
-
-	"github.com/golang/snappy"
-	"github.com/klauspost/compress/huff0"
-)
-
-const (
-	snappyTagLiteral = 0x00
-	snappyTagCopy1   = 0x01
-	snappyTagCopy2   = 0x02
-	snappyTagCopy4   = 0x03
-)
-
-const (
-	snappyChecksumSize = 4
-	snappyMagicBody    = "sNaPpY"
-
-	// snappyMaxBlockSize is the maximum size of the input to encodeBlock. It is not
-	// part of the wire format per se, but some parts of the encoder assume
-	// that an offset fits into a uint16.
-	//
-	// Also, for the framing format (Writer type instead of Encode function),
-	// https://github.com/google/snappy/blob/master/framing_format.txt says
-	// that "the uncompressed data in a chunk must be no longer than 65536
-	// bytes".
-	snappyMaxBlockSize = 65536
-
-	// snappyMaxEncodedLenOfMaxBlockSize equals MaxEncodedLen(snappyMaxBlockSize), but is
-	// hard coded to be a const instead of a variable, so that obufLen can also
-	// be a const. Their equivalence is confirmed by
-	// TestMaxEncodedLenOfMaxBlockSize.
-	snappyMaxEncodedLenOfMaxBlockSize = 76490
-)
-
-const (
-	chunkTypeCompressedData   = 0x00
-	chunkTypeUncompressedData = 0x01
-	chunkTypePadding          = 0xfe
-	chunkTypeStreamIdentifier = 0xff
-)
-
-var (
-	// ErrSnappyCorrupt reports that the input is invalid.
-	ErrSnappyCorrupt = errors.New("snappy: corrupt input")
-	// ErrSnappyTooLarge reports that the uncompressed length is too large.
-	ErrSnappyTooLarge = errors.New("snappy: decoded block is too large")
-	// ErrSnappyUnsupported reports that the input isn't supported.
-	ErrSnappyUnsupported = errors.New("snappy: unsupported input")
-
-	errUnsupportedLiteralLength = errors.New("snappy: unsupported literal length")
-)
-
-// SnappyConverter can read SnappyConverter-compressed streams and convert them to zstd.
-// Conversion is done by converting the stream directly from Snappy without intermediate
-// full decoding.
-// Therefore the compression ratio is much less than what can be done by a full decompression
-// and compression, and a faulty Snappy stream may lead to a faulty Zstandard stream without
-// any errors being generated.
-// No CRC value is being generated and not all CRC values of the Snappy stream are checked.
-// However, it provides really fast recompression of Snappy streams.
-// The converter can be reused to avoid allocations, even after errors.
-type SnappyConverter struct {
-	r     io.Reader
-	err   error
-	buf   []byte
-	block *blockEnc
-}
-
-// Convert the Snappy stream supplied in 'in' and write the zStandard stream to 'w'.
-// If any error is detected on the Snappy stream it is returned.
-// The number of bytes written is returned.
-func (r *SnappyConverter) Convert(in io.Reader, w io.Writer) (int64, error) {
-	initPredefined()
-	r.err = nil
-	r.r = in
-	if r.block == nil {
-		r.block = &blockEnc{}
-		r.block.init()
-	}
-	r.block.initNewEncode()
-	if len(r.buf) != snappyMaxEncodedLenOfMaxBlockSize+snappyChecksumSize {
-		r.buf = make([]byte, snappyMaxEncodedLenOfMaxBlockSize+snappyChecksumSize)
-	}
-	r.block.litEnc.Reuse = huff0.ReusePolicyNone
-	var written int64
-	var readHeader bool
-	{
-		var header []byte
-		var n int
-		header, r.err = frameHeader{WindowSize: snappyMaxBlockSize}.appendTo(r.buf[:0])
-
-		n, r.err = w.Write(header)
-		if r.err != nil {
-			return written, r.err
-		}
-		written += int64(n)
-	}
-
-	for {
-		if !r.readFull(r.buf[:4], true) {
-			// Add empty last block
-			r.block.reset(nil)
-			r.block.last = true
-			err := r.block.encodeLits(r.block.literals, false)
-			if err != nil {
-				return written, err
-			}
-			n, err := w.Write(r.block.output)
-			if err != nil {
-				return written, err
-			}
-			written += int64(n)
-
-			return written, r.err
-		}
-		chunkType := r.buf[0]
-		if !readHeader {
-			if chunkType != chunkTypeStreamIdentifier {
-				println("chunkType != chunkTypeStreamIdentifier", chunkType)
-				r.err = ErrSnappyCorrupt
-				return written, r.err
-			}
-			readHeader = true
-		}
-		chunkLen := int(r.buf[1]) | int(r.buf[2])<<8 | int(r.buf[3])<<16
-		if chunkLen > len(r.buf) {
-			println("chunkLen > len(r.buf)", chunkType)
-			r.err = ErrSnappyUnsupported
-			return written, r.err
-		}
-
-		// The chunk types are specified at
-		// https://github.com/google/snappy/blob/master/framing_format.txt
-		switch chunkType {
-		case chunkTypeCompressedData:
-			// Section 4.2. Compressed data (chunk type 0x00).
-			if chunkLen < snappyChecksumSize {
-				println("chunkLen < snappyChecksumSize", chunkLen, snappyChecksumSize)
-				r.err = ErrSnappyCorrupt
-				return written, r.err
-			}
-			buf := r.buf[:chunkLen]
-			if !r.readFull(buf, false) {
-				return written, r.err
-			}
-			//checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24
-			buf = buf[snappyChecksumSize:]
-
-			n, hdr, err := snappyDecodedLen(buf)
-			if err != nil {
-				r.err = err
-				return written, r.err
-			}
-			buf = buf[hdr:]
-			if n > snappyMaxBlockSize {
-				println("n > snappyMaxBlockSize", n, snappyMaxBlockSize)
-				r.err = ErrSnappyCorrupt
-				return written, r.err
-			}
-			r.block.reset(nil)
-			r.block.pushOffsets()
-			if err := decodeSnappy(r.block, buf); err != nil {
-				r.err = err
-				return written, r.err
-			}
-			if r.block.size+r.block.extraLits != n {
-				printf("invalid size, want %d, got %d\n", n, r.block.size+r.block.extraLits)
-				r.err = ErrSnappyCorrupt
-				return written, r.err
-			}
-			err = r.block.encode(nil, false, false)
-			switch err {
-			case errIncompressible:
-				r.block.popOffsets()
-				r.block.reset(nil)
-				r.block.literals, err = snappy.Decode(r.block.literals[:n], r.buf[snappyChecksumSize:chunkLen])
-				if err != nil {
-					return written, err
-				}
-				err = r.block.encodeLits(r.block.literals, false)
-				if err != nil {
-					return written, err
-				}
-			case nil:
-			default:
-				return written, err
-			}
-
-			n, r.err = w.Write(r.block.output)
-			if r.err != nil {
-				return written, err
-			}
-			written += int64(n)
-			continue
-		case chunkTypeUncompressedData:
-			if debugEncoder {
-				println("Uncompressed, chunklen", chunkLen)
-			}
-			// Section 4.3. Uncompressed data (chunk type 0x01).
-			if chunkLen < snappyChecksumSize {
-				println("chunkLen < snappyChecksumSize", chunkLen, snappyChecksumSize)
-				r.err = ErrSnappyCorrupt
-				return written, r.err
-			}
-			r.block.reset(nil)
-			buf := r.buf[:snappyChecksumSize]
-			if !r.readFull(buf, false) {
-				return written, r.err
-			}
-			checksum := uint32(buf[0]) | uint32(buf[1])<<8 | uint32(buf[2])<<16 | uint32(buf[3])<<24
-			// Read directly into r.decoded instead of via r.buf.
-			n := chunkLen - snappyChecksumSize
-			if n > snappyMaxBlockSize {
-				println("n > snappyMaxBlockSize", n, snappyMaxBlockSize)
-				r.err = ErrSnappyCorrupt
-				return written, r.err
-			}
-			r.block.literals = r.block.literals[:n]
-			if !r.readFull(r.block.literals, false) {
-				return written, r.err
-			}
-			if snappyCRC(r.block.literals) != checksum {
-				println("literals crc mismatch")
-				r.err = ErrSnappyCorrupt
-				return written, r.err
-			}
-			err := r.block.encodeLits(r.block.literals, false)
-			if err != nil {
-				return written, err
-			}
-			n, r.err = w.Write(r.block.output)
-			if r.err != nil {
-				return written, err
-			}
-			written += int64(n)
-			continue
-
-		case chunkTypeStreamIdentifier:
-			if debugEncoder {
-				println("stream id", chunkLen, len(snappyMagicBody))
-			}
-			// Section 4.1. Stream identifier (chunk type 0xff).
-			if chunkLen != len(snappyMagicBody) {
-				println("chunkLen != len(snappyMagicBody)", chunkLen, len(snappyMagicBody))
-				r.err = ErrSnappyCorrupt
-				return written, r.err
-			}
-			if !r.readFull(r.buf[:len(snappyMagicBody)], false) {
-				return written, r.err
-			}
-			for i := 0; i < len(snappyMagicBody); i++ {
-				if r.buf[i] != snappyMagicBody[i] {
-					println("r.buf[i] != snappyMagicBody[i]", r.buf[i], snappyMagicBody[i], i)
-					r.err = ErrSnappyCorrupt
-					return written, r.err
-				}
-			}
-			continue
-		}
-
-		if chunkType <= 0x7f {
-			// Section 4.5. Reserved unskippable chunks (chunk types 0x02-0x7f).
-			println("chunkType <= 0x7f")
-			r.err = ErrSnappyUnsupported
-			return written, r.err
-		}
-		// Section 4.4 Padding (chunk type 0xfe).
-		// Section 4.6. Reserved skippable chunks (chunk types 0x80-0xfd).
-		if !r.readFull(r.buf[:chunkLen], false) {
-			return written, r.err
-		}
-	}
-}
-
-// decodeSnappy writes the decoding of src to dst. It assumes that the varint-encoded
-// length of the decompressed bytes has already been read.
-func decodeSnappy(blk *blockEnc, src []byte) error {
-	//decodeRef(make([]byte, snappyMaxBlockSize), src)
-	var s, length int
-	lits := blk.extraLits
-	var offset uint32
-	for s < len(src) {
-		switch src[s] & 0x03 {
-		case snappyTagLiteral:
-			x := uint32(src[s] >> 2)
-			switch {
-			case x < 60:
-				s++
-			case x == 60:
-				s += 2
-				if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
-					println("uint(s) > uint(len(src)", s, src)
-					return ErrSnappyCorrupt
-				}
-				x = uint32(src[s-1])
-			case x == 61:
-				s += 3
-				if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
-					println("uint(s) > uint(len(src)", s, src)
-					return ErrSnappyCorrupt
-				}
-				x = uint32(src[s-2]) | uint32(src[s-1])<<8
-			case x == 62:
-				s += 4
-				if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
-					println("uint(s) > uint(len(src)", s, src)
-					return ErrSnappyCorrupt
-				}
-				x = uint32(src[s-3]) | uint32(src[s-2])<<8 | uint32(src[s-1])<<16
-			case x == 63:
-				s += 5
-				if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
-					println("uint(s) > uint(len(src)", s, src)
-					return ErrSnappyCorrupt
-				}
-				x = uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24
-			}
-			if x > snappyMaxBlockSize {
-				println("x > snappyMaxBlockSize", x, snappyMaxBlockSize)
-				return ErrSnappyCorrupt
-			}
-			length = int(x) + 1
-			if length <= 0 {
-				println("length <= 0 ", length)
-
-				return errUnsupportedLiteralLength
-			}
-			//if length > snappyMaxBlockSize-d || uint32(length) > len(src)-s {
-			//	return ErrSnappyCorrupt
-			//}
-
-			blk.literals = append(blk.literals, src[s:s+length]...)
-			//println(length, "litLen")
-			lits += length
-			s += length
-			continue
-
-		case snappyTagCopy1:
-			s += 2
-			if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
-				println("uint(s) > uint(len(src)", s, len(src))
-				return ErrSnappyCorrupt
-			}
-			length = 4 + int(src[s-2])>>2&0x7
-			offset = uint32(src[s-2])&0xe0<<3 | uint32(src[s-1])
-
-		case snappyTagCopy2:
-			s += 3
-			if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
-				println("uint(s) > uint(len(src)", s, len(src))
-				return ErrSnappyCorrupt
-			}
-			length = 1 + int(src[s-3])>>2
-			offset = uint32(src[s-2]) | uint32(src[s-1])<<8
-
-		case snappyTagCopy4:
-			s += 5
-			if uint(s) > uint(len(src)) { // The uint conversions catch overflow from the previous line.
-				println("uint(s) > uint(len(src)", s, len(src))
-				return ErrSnappyCorrupt
-			}
-			length = 1 + int(src[s-5])>>2
-			offset = uint32(src[s-4]) | uint32(src[s-3])<<8 | uint32(src[s-2])<<16 | uint32(src[s-1])<<24
-		}
-
-		if offset <= 0 || blk.size+lits < int(offset) /*|| length > len(blk)-d */ {
-			println("offset <= 0 || blk.size+lits < int(offset)", offset, blk.size+lits, int(offset), blk.size, lits)
-
-			return ErrSnappyCorrupt
-		}
-
-		// Check if offset is one of the recent offsets.
-		// Adjusts the output offset accordingly.
-		// Gives a tiny bit of compression, typically around 1%.
-		if false {
-			offset = blk.matchOffset(offset, uint32(lits))
-		} else {
-			offset += 3
-		}
-
-		blk.sequences = append(blk.sequences, seq{
-			litLen:   uint32(lits),
-			offset:   offset,
-			matchLen: uint32(length) - zstdMinMatch,
-		})
-		blk.size += length + lits
-		lits = 0
-	}
-	blk.extraLits = lits
-	return nil
-}
-
-func (r *SnappyConverter) readFull(p []byte, allowEOF bool) (ok bool) {
-	if _, r.err = io.ReadFull(r.r, p); r.err != nil {
-		if r.err == io.ErrUnexpectedEOF || (r.err == io.EOF && !allowEOF) {
-			r.err = ErrSnappyCorrupt
-		}
-		return false
-	}
-	return true
-}
-
-var crcTable = crc32.MakeTable(crc32.Castagnoli)
-
-// crc implements the checksum specified in section 3 of
-// https://github.com/google/snappy/blob/master/framing_format.txt
-func snappyCRC(b []byte) uint32 {
-	c := crc32.Update(0, crcTable, b)
-	return c>>15 | c<<17 + 0xa282ead8
-}
-
-// snappyDecodedLen returns the length of the decoded block and the number of bytes
-// that the length header occupied.
-func snappyDecodedLen(src []byte) (blockLen, headerLen int, err error) {
-	v, n := binary.Uvarint(src)
-	if n <= 0 || v > 0xffffffff {
-		return 0, 0, ErrSnappyCorrupt
-	}
-
-	const wordSize = 32 << (^uint(0) >> 32 & 1)
-	if wordSize == 32 && v > 0x7fffffff {
-		return 0, 0, ErrSnappyTooLarge
-	}
-	return int(v), n, nil
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/zip.go b/vendor/github.com/klauspost/compress/zstd/zip.go
deleted file mode 100644
index 9325b928ae..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/zip.go
+++ /dev/null
@@ -1,121 +0,0 @@
-// Copyright 2019+ Klaus Post. All rights reserved.
-// License information can be found in the LICENSE file.
-
-package zstd
-
-import (
-	"errors"
-	"io"
-	"sync"
-)
-
-// ZipMethodWinZip is the method for Zstandard compressed data inside Zip files for WinZip.
-// See https://www.winzip.com/win/en/comp_info.html
-const ZipMethodWinZip = 93
-
-// ZipMethodPKWare is the original method number used by PKWARE to indicate Zstandard compression.
-// Deprecated: This has been deprecated by PKWARE, use ZipMethodWinZip instead for compression.
-// See https://pkware.cachefly.net/webdocs/APPNOTE/APPNOTE-6.3.9.TXT
-const ZipMethodPKWare = 20
-
-var zipReaderPool sync.Pool
-
-// newZipReader cannot be used since we would leak goroutines...
-func newZipReader(r io.Reader) io.ReadCloser {
-	dec, ok := zipReaderPool.Get().(*Decoder)
-	if ok {
-		dec.Reset(r)
-	} else {
-		d, err := NewReader(r, WithDecoderConcurrency(1), WithDecoderLowmem(true))
-		if err != nil {
-			panic(err)
-		}
-		dec = d
-	}
-	return &pooledZipReader{dec: dec}
-}
-
-type pooledZipReader struct {
-	mu  sync.Mutex // guards Close and Read
-	dec *Decoder
-}
-
-func (r *pooledZipReader) Read(p []byte) (n int, err error) {
-	r.mu.Lock()
-	defer r.mu.Unlock()
-	if r.dec == nil {
-		return 0, errors.New("Read after Close")
-	}
-	dec, err := r.dec.Read(p)
-
-	return dec, err
-}
-
-func (r *pooledZipReader) Close() error {
-	r.mu.Lock()
-	defer r.mu.Unlock()
-	var err error
-	if r.dec != nil {
-		err = r.dec.Reset(nil)
-		zipReaderPool.Put(r.dec)
-		r.dec = nil
-	}
-	return err
-}
-
-type pooledZipWriter struct {
-	mu  sync.Mutex // guards Close and Read
-	enc *Encoder
-}
-
-func (w *pooledZipWriter) Write(p []byte) (n int, err error) {
-	w.mu.Lock()
-	defer w.mu.Unlock()
-	if w.enc == nil {
-		return 0, errors.New("Write after Close")
-	}
-	return w.enc.Write(p)
-}
-
-func (w *pooledZipWriter) Close() error {
-	w.mu.Lock()
-	defer w.mu.Unlock()
-	var err error
-	if w.enc != nil {
-		err = w.enc.Close()
-		zipReaderPool.Put(w.enc)
-		w.enc = nil
-	}
-	return err
-}
-
-// ZipCompressor returns a compressor that can be registered with zip libraries.
-// The provided encoder options will be used on all encodes.
-func ZipCompressor(opts ...EOption) func(w io.Writer) (io.WriteCloser, error) {
-	var pool sync.Pool
-	return func(w io.Writer) (io.WriteCloser, error) {
-		enc, ok := pool.Get().(*Encoder)
-		if ok {
-			enc.Reset(w)
-		} else {
-			var err error
-			enc, err = NewWriter(w, opts...)
-			if err != nil {
-				return nil, err
-			}
-		}
-		return &pooledZipWriter{enc: enc}, nil
-	}
-}
-
-// ZipDecompressor returns a decompressor that can be registered with zip libraries.
-// See ZipCompressor for example.
-func ZipDecompressor() func(r io.Reader) io.ReadCloser {
-	return func(r io.Reader) io.ReadCloser {
-		d, err := NewReader(r, WithDecoderConcurrency(1), WithDecoderLowmem(true))
-		if err != nil {
-			panic(err)
-		}
-		return d.IOReadCloser()
-	}
-}
diff --git a/vendor/github.com/klauspost/compress/zstd/zstd.go b/vendor/github.com/klauspost/compress/zstd/zstd.go
deleted file mode 100644
index ef1d49a009..0000000000
--- a/vendor/github.com/klauspost/compress/zstd/zstd.go
+++ /dev/null
@@ -1,152 +0,0 @@
-// Package zstd provides decompression of zstandard files.
-//
-// For advanced usage and examples, go to the README: https://github.com/klauspost/compress/tree/master/zstd#zstd
-package zstd
-
-import (
-	"bytes"
-	"encoding/binary"
-	"errors"
-	"log"
-	"math"
-	"math/bits"
-)
-
-// enable debug printing
-const debug = false
-
-// enable encoding debug printing
-const debugEncoder = debug
-
-// enable decoding debug printing
-const debugDecoder = debug
-
-// Enable extra assertions.
-const debugAsserts = debug || false
-
-// print sequence details
-const debugSequences = false
-
-// print detailed matching information
-const debugMatches = false
-
-// force encoder to use predefined tables.
-const forcePreDef = false
-
-// zstdMinMatch is the minimum zstd match length.
-const zstdMinMatch = 3
-
-// Reset the buffer offset when reaching this.
-const bufferReset = math.MaxInt32 - MaxWindowSize
-
-var (
-	// ErrReservedBlockType is returned when a reserved block type is found.
-	// Typically this indicates wrong or corrupted input.
-	ErrReservedBlockType = errors.New("invalid input: reserved block type encountered")
-
-	// ErrCompressedSizeTooBig is returned when a block is bigger than allowed.
-	// Typically this indicates wrong or corrupted input.
-	ErrCompressedSizeTooBig = errors.New("invalid input: compressed size too big")
-
-	// ErrBlockTooSmall is returned when a block is too small to be decoded.
-	// Typically returned on invalid input.
-	ErrBlockTooSmall = errors.New("block too small")
-
-	// ErrMagicMismatch is returned when a "magic" number isn't what is expected.
-	// Typically this indicates wrong or corrupted input.
-	ErrMagicMismatch = errors.New("invalid input: magic number mismatch")
-
-	// ErrWindowSizeExceeded is returned when a reference exceeds the valid window size.
-	// Typically this indicates wrong or corrupted input.
-	ErrWindowSizeExceeded = errors.New("window size exceeded")
-
-	// ErrWindowSizeTooSmall is returned when no window size is specified.
-	// Typically this indicates wrong or corrupted input.
-	ErrWindowSizeTooSmall = errors.New("invalid input: window size was too small")
-
-	// ErrDecoderSizeExceeded is returned if decompressed size exceeds the configured limit.
-	ErrDecoderSizeExceeded = errors.New("decompressed size exceeds configured limit")
-
-	// ErrUnknownDictionary is returned if the dictionary ID is unknown.
-	// For the time being dictionaries are not supported.
-	ErrUnknownDictionary = errors.New("unknown dictionary")
-
-	// ErrFrameSizeExceeded is returned if the stated frame size is exceeded.
-	// This is only returned if SingleSegment is specified on the frame.
-	ErrFrameSizeExceeded = errors.New("frame size exceeded")
-
-	// ErrCRCMismatch is returned if CRC mismatches.
-	ErrCRCMismatch = errors.New("CRC check failed")
-
-	// ErrDecoderClosed will be returned if the Decoder was used after
-	// Close has been called.
-	ErrDecoderClosed = errors.New("decoder used after Close")
-
-	// ErrDecoderNilInput is returned when a nil Reader was provided
-	// and an operation other than Reset/DecodeAll/Close was attempted.
-	ErrDecoderNilInput = errors.New("nil input provided as reader")
-)
-
-func println(a ...interface{}) {
-	if debug || debugDecoder || debugEncoder {
-		log.Println(a...)
-	}
-}
-
-func printf(format string, a ...interface{}) {
-	if debug || debugDecoder || debugEncoder {
-		log.Printf(format, a...)
-	}
-}
-
-// matchLenFast does matching, but will not match the last up to 7 bytes.
-func matchLenFast(a, b []byte) int {
-	endI := len(a) & (math.MaxInt32 - 7)
-	for i := 0; i < endI; i += 8 {
-		if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-			return i + bits.TrailingZeros64(diff)>>3
-		}
-	}
-	return endI
-}
-
-// matchLen returns the maximum length.
-// a must be the shortest of the two.
-// The function also returns whether all bytes matched.
-func matchLen(a, b []byte) int {
-	b = b[:len(a)]
-	for i := 0; i < len(a)-7; i += 8 {
-		if diff := load64(a, i) ^ load64(b, i); diff != 0 {
-			return i + (bits.TrailingZeros64(diff) >> 3)
-		}
-	}
-
-	checked := (len(a) >> 3) << 3
-	a = a[checked:]
-	b = b[checked:]
-	for i := range a {
-		if a[i] != b[i] {
-			return i + checked
-		}
-	}
-	return len(a) + checked
-}
-
-func load3232(b []byte, i int32) uint32 {
-	return binary.LittleEndian.Uint32(b[i:])
-}
-
-func load6432(b []byte, i int32) uint64 {
-	return binary.LittleEndian.Uint64(b[i:])
-}
-
-func load64(b []byte, i int) uint64 {
-	return binary.LittleEndian.Uint64(b[i:])
-}
-
-type byter interface {
-	Bytes() []byte
-	Len() int
-}
-
-var _ byter = &bytes.Buffer{}
diff --git a/vendor/github.com/klauspost/cpuid/v2/.gitignore b/vendor/github.com/klauspost/cpuid/v2/.gitignore
deleted file mode 100644
index daf913b1b3..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/.gitignore
+++ /dev/null
@@ -1,24 +0,0 @@
-# Compiled Object files, Static and Dynamic libs (Shared Objects)
-*.o
-*.a
-*.so
-
-# Folders
-_obj
-_test
-
-# Architecture specific extensions/prefixes
-*.[568vq]
-[568vq].out
-
-*.cgo1.go
-*.cgo2.c
-_cgo_defun.c
-_cgo_gotypes.go
-_cgo_export.*
-
-_testmain.go
-
-*.exe
-*.test
-*.prof
diff --git a/vendor/github.com/klauspost/cpuid/v2/.goreleaser.yml b/vendor/github.com/klauspost/cpuid/v2/.goreleaser.yml
deleted file mode 100644
index 944cc00075..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/.goreleaser.yml
+++ /dev/null
@@ -1,74 +0,0 @@
-# This is an example goreleaser.yaml file with some sane defaults.
-# Make sure to check the documentation at http://goreleaser.com
-
-builds:
-  -
-    id: "cpuid"
-    binary: cpuid
-    main: ./cmd/cpuid/main.go
-    env:
-      - CGO_ENABLED=0
-    flags:
-      - -ldflags=-s -w
-    goos:
-      - aix
-      - linux
-      - freebsd
-      - netbsd
-      - windows
-      - darwin
-    goarch:
-      - 386
-      - amd64
-      - arm64
-    goarm:
-      - 7
-
-archives:
-  -
-    id: cpuid
-    name_template: "cpuid-{{ .Os }}_{{ .Arch }}_{{ .Version }}"
-    replacements:
-      aix: AIX
-      darwin: OSX
-      linux: Linux
-      windows: Windows
-      386: i386
-      amd64: x86_64
-      freebsd: FreeBSD
-      netbsd: NetBSD
-    format_overrides:
-      - goos: windows
-        format: zip
-    files:
-      - LICENSE
-checksum:
-  name_template: 'checksums.txt'
-snapshot:
-  name_template: "{{ .Tag }}-next"
-changelog:
-  sort: asc
-  filters:
-    exclude:
-    - '^doc:'
-    - '^docs:'
-    - '^test:'
-    - '^tests:'
-    - '^Update\sREADME.md'
-
-nfpms:
-  -
-    file_name_template: "cpuid_package_{{ .Version }}_{{ .Os }}_{{ .Arch }}"
-    vendor: Klaus Post
-    homepage: https://github.com/klauspost/cpuid
-    maintainer: Klaus Post <klauspost@gmail.com>
-    description: CPUID Tool
-    license: BSD 3-Clause
-    formats:
-      - deb
-      - rpm
-    replacements:
-      darwin: Darwin
-      linux: Linux
-      freebsd: FreeBSD
-      amd64: x86_64
diff --git a/vendor/github.com/klauspost/cpuid/v2/CONTRIBUTING.txt b/vendor/github.com/klauspost/cpuid/v2/CONTRIBUTING.txt
deleted file mode 100644
index 2ef4714f71..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/CONTRIBUTING.txt
+++ /dev/null
@@ -1,35 +0,0 @@
-Developer Certificate of Origin

-Version 1.1

-

-Copyright (C) 2015- Klaus Post & Contributors.

-Email: klauspost@gmail.com

-

-Everyone is permitted to copy and distribute verbatim copies of this

-license document, but changing it is not allowed.

-

-

-Developer's Certificate of Origin 1.1

-

-By making a contribution to this project, I certify that:

-

-(a) The contribution was created in whole or in part by me and I

-    have the right to submit it under the open source license

-    indicated in the file; or

-

-(b) The contribution is based upon previous work that, to the best

-    of my knowledge, is covered under an appropriate open source

-    license and I have the right under that license to submit that

-    work with modifications, whether created in whole or in part

-    by me, under the same open source license (unless I am

-    permitted to submit under a different license), as indicated

-    in the file; or

-

-(c) The contribution was provided directly to me by some other

-    person who certified (a), (b) or (c) and I have not modified

-    it.

-

-(d) I understand and agree that this project and the contribution

-    are public and that a record of the contribution (including all

-    personal information I submit with it, including my sign-off) is

-    maintained indefinitely and may be redistributed consistent with

-    this project or the open source license(s) involved.

diff --git a/vendor/github.com/klauspost/cpuid/v2/LICENSE b/vendor/github.com/klauspost/cpuid/v2/LICENSE
deleted file mode 100644
index 5cec7ee949..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/LICENSE
+++ /dev/null
@@ -1,22 +0,0 @@
-The MIT License (MIT)
-
-Copyright (c) 2015 Klaus Post
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
-
diff --git a/vendor/github.com/klauspost/cpuid/v2/README.md b/vendor/github.com/klauspost/cpuid/v2/README.md
deleted file mode 100644
index 465f4b77cb..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/README.md
+++ /dev/null
@@ -1,137 +0,0 @@
-# cpuid
-Package cpuid provides information about the CPU running the current program.
-
-CPU features are detected on startup, and kept for fast access through the life of the application.
-Currently x86 / x64 (AMD64/i386) and ARM (ARM64) is supported, and no external C (cgo) code is used, which should make the library very easy to use.
-
-You can access the CPU information by accessing the shared CPU variable of the cpuid library.
-
-Package home: https://github.com/klauspost/cpuid
-
-[![PkgGoDev](https://pkg.go.dev/badge/github.com/klauspost/cpuid)](https://pkg.go.dev/github.com/klauspost/cpuid/v2)
-[![Build Status][3]][4]
-
-[3]: https://travis-ci.org/klauspost/cpuid.svg?branch=master
-[4]: https://travis-ci.org/klauspost/cpuid
-
-## installing
-
-`go get -u github.com/klauspost/cpuid/v2` using modules. 
-
-Drop `v2` for others.
-
-## example
-
-```Go
-package main
-
-import (
-	"fmt"
-	"strings"
-
-	. "github.com/klauspost/cpuid/v2"
-)
-
-func main() {
-	// Print basic CPU information:
-	fmt.Println("Name:", CPU.BrandName)
-	fmt.Println("PhysicalCores:", CPU.PhysicalCores)
-	fmt.Println("ThreadsPerCore:", CPU.ThreadsPerCore)
-	fmt.Println("LogicalCores:", CPU.LogicalCores)
-	fmt.Println("Family", CPU.Family, "Model:", CPU.Model, "Vendor ID:", CPU.VendorID)
-	fmt.Println("Features:", fmt.Sprintf(strings.Join(CPU.FeatureSet(), ",")))
-	fmt.Println("Cacheline bytes:", CPU.CacheLine)
-	fmt.Println("L1 Data Cache:", CPU.Cache.L1D, "bytes")
-	fmt.Println("L1 Instruction Cache:", CPU.Cache.L1D, "bytes")
-	fmt.Println("L2 Cache:", CPU.Cache.L2, "bytes")
-	fmt.Println("L3 Cache:", CPU.Cache.L3, "bytes")
-	fmt.Println("Frequency", CPU.Hz, "hz")
-
-	// Test if we have these specific features:
-	if CPU.Supports(SSE, SSE2) {
-		fmt.Println("We have Streaming SIMD 2 Extensions")
-	}
-}
-```
-
-Sample output:
-```
->go run main.go
-Name: AMD Ryzen 9 3950X 16-Core Processor
-PhysicalCores: 16
-ThreadsPerCore: 2
-LogicalCores: 32
-Family 23 Model: 113 Vendor ID: AMD
-Features: ADX,AESNI,AVX,AVX2,BMI1,BMI2,CLMUL,CMOV,CX16,F16C,FMA3,HTT,HYPERVISOR,LZCNT,MMX,MMXEXT,NX,POPCNT,RDRAND,RDSEED,RDTSCP,SHA,SSE,SSE2,SSE3,SSE4,SSE42,SSE4A,SSSE3
-Cacheline bytes: 64
-L1 Data Cache: 32768 bytes
-L1 Instruction Cache: 32768 bytes
-L2 Cache: 524288 bytes
-L3 Cache: 16777216 bytes
-Frequency 0 hz
-We have Streaming SIMD 2 Extensions
-```
-
-# usage
-
-The `cpuid.CPU` provides access to CPU features. Use `cpuid.CPU.Supports()` to check for CPU features.
-A faster `cpuid.CPU.Has()` is provided which will usually be inlined by the gc compiler.  
-
-Note that for some cpu/os combinations some features will not be detected.
-`amd64` has rather good support and should work reliably on all platforms.
-
-Note that hypervisors may not pass through all CPU features.
-
-## arm64 feature detection
-
-Not all operating systems provide ARM features directly 
-and there is no safe way to do so for the rest.
-
-Currently `arm64/linux` and `arm64/freebsd` should be quite reliable. 
-`arm64/darwin` adds features expected from the M1 processor, but a lot remains undetected.
-
-A `DetectARM()` can be used if you are able to control your deployment,
-it will detect CPU features, but may crash if the OS doesn't intercept the calls.
-A `-cpu.arm` flag for detecting unsafe ARM features can be added. See below.
- 
-Note that currently only features are detected on ARM, 
-no additional information is currently available. 
-
-## flags
-
-It is possible to add flags that affects cpu detection.
-
-For this the `Flags()` command is provided.
-
-This must be called *before* `flag.Parse()` AND after the flags have been parsed `Detect()` must be called.
-
-This means that any detection used in `init()` functions will not contain these flags.
-
-Example:
-
-```Go
-package main
-
-import (
-	"flag"
-	"fmt"
-	"strings"
-
-	"github.com/klauspost/cpuid/v2"
-)
-
-func main() {
-	cpuid.Flags()
-	flag.Parse()
-	cpuid.Detect()
-
-	// Test if we have these specific features:
-	if cpuid.CPU.Supports(cpuid.SSE, cpuid.SSE2) {
-		fmt.Println("We have Streaming SIMD 2 Extensions")
-	}
-}
-```
-
-# license
-
-This code is published under an MIT license. See LICENSE file for more information.
diff --git a/vendor/github.com/klauspost/cpuid/v2/cpuid.go b/vendor/github.com/klauspost/cpuid/v2/cpuid.go
deleted file mode 100644
index 1d88736b68..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/cpuid.go
+++ /dev/null
@@ -1,1070 +0,0 @@
-// Copyright (c) 2015 Klaus Post, released under MIT License. See LICENSE file.
-
-// Package cpuid provides information about the CPU running the current program.
-//
-// CPU features are detected on startup, and kept for fast access through the life of the application.
-// Currently x86 / x64 (AMD64) as well as arm64 is supported.
-//
-// You can access the CPU information by accessing the shared CPU variable of the cpuid library.
-//
-// Package home: https://github.com/klauspost/cpuid
-package cpuid
-
-import (
-	"flag"
-	"fmt"
-	"math"
-	"os"
-	"runtime"
-	"strings"
-)
-
-// AMD refererence: https://www.amd.com/system/files/TechDocs/25481.pdf
-// and Processor Programming Reference (PPR)
-
-// Vendor is a representation of a CPU vendor.
-type Vendor int
-
-const (
-	VendorUnknown Vendor = iota
-	Intel
-	AMD
-	VIA
-	Transmeta
-	NSC
-	KVM  // Kernel-based Virtual Machine
-	MSVM // Microsoft Hyper-V or Windows Virtual PC
-	VMware
-	XenHVM
-	Bhyve
-	Hygon
-	SiS
-	RDC
-
-	Ampere
-	ARM
-	Broadcom
-	Cavium
-	DEC
-	Fujitsu
-	Infineon
-	Motorola
-	NVIDIA
-	AMCC
-	Qualcomm
-	Marvell
-
-	lastVendor
-)
-
-//go:generate stringer -type=FeatureID,Vendor
-
-// FeatureID is the ID of a specific cpu feature.
-type FeatureID int
-
-const (
-	// Keep index -1 as unknown
-	UNKNOWN = -1
-
-	// Add features
-	ADX                FeatureID = iota // Intel ADX (Multi-Precision Add-Carry Instruction Extensions)
-	AESNI                               // Advanced Encryption Standard New Instructions
-	AMD3DNOW                            // AMD 3DNOW
-	AMD3DNOWEXT                         // AMD 3DNowExt
-	AMXBF16                             // Tile computational operations on BFLOAT16 numbers
-	AMXINT8                             // Tile computational operations on 8-bit integers
-	AMXTILE                             // Tile architecture
-	AVX                                 // AVX functions
-	AVX2                                // AVX2 functions
-	AVX512BF16                          // AVX-512 BFLOAT16 Instructions
-	AVX512BITALG                        // AVX-512 Bit Algorithms
-	AVX512BW                            // AVX-512 Byte and Word Instructions
-	AVX512CD                            // AVX-512 Conflict Detection Instructions
-	AVX512DQ                            // AVX-512 Doubleword and Quadword Instructions
-	AVX512ER                            // AVX-512 Exponential and Reciprocal Instructions
-	AVX512F                             // AVX-512 Foundation
-	AVX512FP16                          // AVX-512 FP16 Instructions
-	AVX512IFMA                          // AVX-512 Integer Fused Multiply-Add Instructions
-	AVX512PF                            // AVX-512 Prefetch Instructions
-	AVX512VBMI                          // AVX-512 Vector Bit Manipulation Instructions
-	AVX512VBMI2                         // AVX-512 Vector Bit Manipulation Instructions, Version 2
-	AVX512VL                            // AVX-512 Vector Length Extensions
-	AVX512VNNI                          // AVX-512 Vector Neural Network Instructions
-	AVX512VP2INTERSECT                  // AVX-512 Intersect for D/Q
-	AVX512VPOPCNTDQ                     // AVX-512 Vector Population Count Doubleword and Quadword
-	AVXSLOW                             // Indicates the CPU performs 2 128 bit operations instead of one.
-	BMI1                                // Bit Manipulation Instruction Set 1
-	BMI2                                // Bit Manipulation Instruction Set 2
-	CLDEMOTE                            // Cache Line Demote
-	CLMUL                               // Carry-less Multiplication
-	CLZERO                              // CLZERO instruction supported
-	CMOV                                // i686 CMOV
-	CPBOOST                             // Core Performance Boost
-	CX16                                // CMPXCHG16B Instruction
-	ENQCMD                              // Enqueue Command
-	ERMS                                // Enhanced REP MOVSB/STOSB
-	F16C                                // Half-precision floating-point conversion
-	FMA3                                // Intel FMA 3. Does not imply AVX.
-	FMA4                                // Bulldozer FMA4 functions
-	GFNI                                // Galois Field New Instructions
-	HLE                                 // Hardware Lock Elision
-	HTT                                 // Hyperthreading (enabled)
-	HWA                                 // Hardware assert supported. Indicates support for MSRC001_10
-	HYPERVISOR                          // This bit has been reserved by Intel & AMD for use by hypervisors
-	IBPB                                // Indirect Branch Restricted Speculation (IBRS) and Indirect Branch Predictor Barrier (IBPB)
-	IBS                                 // Instruction Based Sampling (AMD)
-	IBSBRNTRGT                          // Instruction Based Sampling Feature (AMD)
-	IBSFETCHSAM                         // Instruction Based Sampling Feature (AMD)
-	IBSFFV                              // Instruction Based Sampling Feature (AMD)
-	IBSOPCNT                            // Instruction Based Sampling Feature (AMD)
-	IBSOPCNTEXT                         // Instruction Based Sampling Feature (AMD)
-	IBSOPSAM                            // Instruction Based Sampling Feature (AMD)
-	IBSRDWROPCNT                        // Instruction Based Sampling Feature (AMD)
-	IBSRIPINVALIDCHK                    // Instruction Based Sampling Feature (AMD)
-	INT_WBINVD                          // WBINVD/WBNOINVD are interruptible.
-	INVLPGB                             // NVLPGB and TLBSYNC instruction supported
-	LZCNT                               // LZCNT instruction
-	MCAOVERFLOW                         // MCA overflow recovery support.
-	MCOMMIT                             // MCOMMIT instruction supported
-	MMX                                 // standard MMX
-	MMXEXT                              // SSE integer functions or AMD MMX ext
-	MOVDIR64B                           // Move 64 Bytes as Direct Store
-	MOVDIRI                             // Move Doubleword as Direct Store
-	MPX                                 // Intel MPX (Memory Protection Extensions)
-	MSRIRC                              // Instruction Retired Counter MSR available
-	NX                                  // NX (No-Execute) bit
-	POPCNT                              // POPCNT instruction
-	RDPRU                               // RDPRU instruction supported
-	RDRAND                              // RDRAND instruction is available
-	RDSEED                              // RDSEED instruction is available
-	RDTSCP                              // RDTSCP Instruction
-	RTM                                 // Restricted Transactional Memory
-	RTM_ALWAYS_ABORT                    // Indicates that the loaded microcode is forcing RTM abort.
-	SERIALIZE                           // Serialize Instruction Execution
-	SGX                                 // Software Guard Extensions
-	SGXLC                               // Software Guard Extensions Launch Control
-	SHA                                 // Intel SHA Extensions
-	SSE                                 // SSE functions
-	SSE2                                // P4 SSE functions
-	SSE3                                // Prescott SSE3 functions
-	SSE4                                // Penryn SSE4.1 functions
-	SSE42                               // Nehalem SSE4.2 functions
-	SSE4A                               // AMD Barcelona microarchitecture SSE4a instructions
-	SSSE3                               // Conroe SSSE3 functions
-	STIBP                               // Single Thread Indirect Branch Predictors
-	SUCCOR                              // Software uncorrectable error containment and recovery capability.
-	TBM                                 // AMD Trailing Bit Manipulation
-	TSXLDTRK                            // Intel TSX Suspend Load Address Tracking
-	VAES                                // Vector AES
-	VMX                                 // Virtual Machine Extensions
-	VPCLMULQDQ                          // Carry-Less Multiplication Quadword
-	WAITPKG                             // TPAUSE, UMONITOR, UMWAIT
-	WBNOINVD                            // Write Back and Do Not Invalidate Cache
-	XOP                                 // Bulldozer XOP functions
-
-	// ARM features:
-	AESARM   // AES instructions
-	ARMCPUID // Some CPU ID registers readable at user-level
-	ASIMD    // Advanced SIMD
-	ASIMDDP  // SIMD Dot Product
-	ASIMDHP  // Advanced SIMD half-precision floating point
-	ASIMDRDM // Rounding Double Multiply Accumulate/Subtract (SQRDMLAH/SQRDMLSH)
-	ATOMICS  // Large System Extensions (LSE)
-	CRC32    // CRC32/CRC32C instructions
-	DCPOP    // Data cache clean to Point of Persistence (DC CVAP)
-	EVTSTRM  // Generic timer
-	FCMA     // Floatin point complex number addition and multiplication
-	FP       // Single-precision and double-precision floating point
-	FPHP     // Half-precision floating point
-	GPA      // Generic Pointer Authentication
-	JSCVT    // Javascript-style double->int convert (FJCVTZS)
-	LRCPC    // Weaker release consistency (LDAPR, etc)
-	PMULL    // Polynomial Multiply instructions (PMULL/PMULL2)
-	SHA1     // SHA-1 instructions (SHA1C, etc)
-	SHA2     // SHA-2 instructions (SHA256H, etc)
-	SHA3     // SHA-3 instructions (EOR3, RAXI, XAR, BCAX)
-	SHA512   // SHA512 instructions
-	SM3      // SM3 instructions
-	SM4      // SM4 instructions
-	SVE      // Scalable Vector Extension
-
-	// Keep it last. It automatically defines the size of []flagSet
-	lastID
-
-	firstID FeatureID = UNKNOWN + 1
-)
-
-// CPUInfo contains information about the detected system CPU.
-type CPUInfo struct {
-	BrandName      string  // Brand name reported by the CPU
-	VendorID       Vendor  // Comparable CPU vendor ID
-	VendorString   string  // Raw vendor string.
-	featureSet     flagSet // Features of the CPU
-	PhysicalCores  int     // Number of physical processor cores in your CPU. Will be 0 if undetectable.
-	ThreadsPerCore int     // Number of threads per physical core. Will be 1 if undetectable.
-	LogicalCores   int     // Number of physical cores times threads that can run on each core through the use of hyperthreading. Will be 0 if undetectable.
-	Family         int     // CPU family number
-	Model          int     // CPU model number
-	CacheLine      int     // Cache line size in bytes. Will be 0 if undetectable.
-	Hz             int64   // Clock speed, if known, 0 otherwise. Will attempt to contain base clock speed.
-	BoostFreq      int64   // Max clock speed, if known, 0 otherwise
-	Cache          struct {
-		L1I int // L1 Instruction Cache (per core or shared). Will be -1 if undetected
-		L1D int // L1 Data Cache (per core or shared). Will be -1 if undetected
-		L2  int // L2 Cache (per core or shared). Will be -1 if undetected
-		L3  int // L3 Cache (per core, per ccx or shared). Will be -1 if undetected
-	}
-	SGX       SGXSupport
-	maxFunc   uint32
-	maxExFunc uint32
-}
-
-var cpuid func(op uint32) (eax, ebx, ecx, edx uint32)
-var cpuidex func(op, op2 uint32) (eax, ebx, ecx, edx uint32)
-var xgetbv func(index uint32) (eax, edx uint32)
-var rdtscpAsm func() (eax, ebx, ecx, edx uint32)
-var darwinHasAVX512 = func() bool { return false }
-
-// CPU contains information about the CPU as detected on startup,
-// or when Detect last was called.
-//
-// Use this as the primary entry point to you data.
-var CPU CPUInfo
-
-func init() {
-	initCPU()
-	Detect()
-}
-
-// Detect will re-detect current CPU info.
-// This will replace the content of the exported CPU variable.
-//
-// Unless you expect the CPU to change while you are running your program
-// you should not need to call this function.
-// If you call this, you must ensure that no other goroutine is accessing the
-// exported CPU variable.
-func Detect() {
-	// Set defaults
-	CPU.ThreadsPerCore = 1
-	CPU.Cache.L1I = -1
-	CPU.Cache.L1D = -1
-	CPU.Cache.L2 = -1
-	CPU.Cache.L3 = -1
-	safe := true
-	if detectArmFlag != nil {
-		safe = !*detectArmFlag
-	}
-	addInfo(&CPU, safe)
-	if displayFeats != nil && *displayFeats {
-		fmt.Println("cpu features:", strings.Join(CPU.FeatureSet(), ","))
-		// Exit with non-zero so tests will print value.
-		os.Exit(1)
-	}
-	if disableFlag != nil {
-		s := strings.Split(*disableFlag, ",")
-		for _, feat := range s {
-			feat := ParseFeature(strings.TrimSpace(feat))
-			if feat != UNKNOWN {
-				CPU.featureSet.unset(feat)
-			}
-		}
-	}
-}
-
-// DetectARM will detect ARM64 features.
-// This is NOT done automatically since it can potentially crash
-// if the OS does not handle the command.
-// If in the future this can be done safely this function may not
-// do anything.
-func DetectARM() {
-	addInfo(&CPU, false)
-}
-
-var detectArmFlag *bool
-var displayFeats *bool
-var disableFlag *string
-
-// Flags will enable flags.
-// This must be called *before* flag.Parse AND
-// Detect must be called after the flags have been parsed.
-// Note that this means that any detection used in init() functions
-// will not contain these flags.
-func Flags() {
-	disableFlag = flag.String("cpu.disable", "", "disable cpu features; comma separated list")
-	displayFeats = flag.Bool("cpu.features", false, "lists cpu features and exits")
-	detectArmFlag = flag.Bool("cpu.arm", false, "allow ARM features to be detected; can potentially crash")
-}
-
-// Supports returns whether the CPU supports all of the requested features.
-func (c CPUInfo) Supports(ids ...FeatureID) bool {
-	for _, id := range ids {
-		if !c.featureSet.inSet(id) {
-			return false
-		}
-	}
-	return true
-}
-
-// Has allows for checking a single feature.
-// Should be inlined by the compiler.
-func (c CPUInfo) Has(id FeatureID) bool {
-	return c.featureSet.inSet(id)
-}
-
-// Disable will disable one or several features.
-func (c *CPUInfo) Disable(ids ...FeatureID) bool {
-	for _, id := range ids {
-		c.featureSet.unset(id)
-	}
-	return true
-}
-
-// Enable will disable one or several features even if they were undetected.
-// This is of course not recommended for obvious reasons.
-func (c *CPUInfo) Enable(ids ...FeatureID) bool {
-	for _, id := range ids {
-		c.featureSet.set(id)
-	}
-	return true
-}
-
-// IsVendor returns true if vendor is recognized as Intel
-func (c CPUInfo) IsVendor(v Vendor) bool {
-	return c.VendorID == v
-}
-
-func (c CPUInfo) FeatureSet() []string {
-	s := make([]string, 0)
-	for _, f := range c.featureSet.Strings() {
-		s = append(s, f)
-	}
-	return s
-}
-
-// RTCounter returns the 64-bit time-stamp counter
-// Uses the RDTSCP instruction. The value 0 is returned
-// if the CPU does not support the instruction.
-func (c CPUInfo) RTCounter() uint64 {
-	if !c.Supports(RDTSCP) {
-		return 0
-	}
-	a, _, _, d := rdtscpAsm()
-	return uint64(a) | (uint64(d) << 32)
-}
-
-// Ia32TscAux returns the IA32_TSC_AUX part of the RDTSCP.
-// This variable is OS dependent, but on Linux contains information
-// about the current cpu/core the code is running on.
-// If the RDTSCP instruction isn't supported on the CPU, the value 0 is returned.
-func (c CPUInfo) Ia32TscAux() uint32 {
-	if !c.Supports(RDTSCP) {
-		return 0
-	}
-	_, _, ecx, _ := rdtscpAsm()
-	return ecx
-}
-
-// LogicalCPU will return the Logical CPU the code is currently executing on.
-// This is likely to change when the OS re-schedules the running thread
-// to another CPU.
-// If the current core cannot be detected, -1 will be returned.
-func (c CPUInfo) LogicalCPU() int {
-	if c.maxFunc < 1 {
-		return -1
-	}
-	_, ebx, _, _ := cpuid(1)
-	return int(ebx >> 24)
-}
-
-// frequencies tries to compute the clock speed of the CPU. If leaf 15 is
-// supported, use it, otherwise parse the brand string. Yes, really.
-func (c *CPUInfo) frequencies() {
-	c.Hz, c.BoostFreq = 0, 0
-	mfi := maxFunctionID()
-	if mfi >= 0x15 {
-		eax, ebx, ecx, _ := cpuid(0x15)
-		if eax != 0 && ebx != 0 && ecx != 0 {
-			c.Hz = (int64(ecx) * int64(ebx)) / int64(eax)
-		}
-	}
-	if mfi >= 0x16 {
-		a, b, _, _ := cpuid(0x16)
-		// Base...
-		if a&0xffff > 0 {
-			c.Hz = int64(a&0xffff) * 1_000_000
-		}
-		// Boost...
-		if b&0xffff > 0 {
-			c.BoostFreq = int64(b&0xffff) * 1_000_000
-		}
-	}
-	if c.Hz > 0 {
-		return
-	}
-
-	// computeHz determines the official rated speed of a CPU from its brand
-	// string. This insanity is *actually the official documented way to do
-	// this according to Intel*, prior to leaf 0x15 existing. The official
-	// documentation only shows this working for exactly `x.xx` or `xxxx`
-	// cases, e.g., `2.50GHz` or `1300MHz`; this parser will accept other
-	// sizes.
-	model := c.BrandName
-	hz := strings.LastIndex(model, "Hz")
-	if hz < 3 {
-		return
-	}
-	var multiplier int64
-	switch model[hz-1] {
-	case 'M':
-		multiplier = 1000 * 1000
-	case 'G':
-		multiplier = 1000 * 1000 * 1000
-	case 'T':
-		multiplier = 1000 * 1000 * 1000 * 1000
-	}
-	if multiplier == 0 {
-		return
-	}
-	freq := int64(0)
-	divisor := int64(0)
-	decimalShift := int64(1)
-	var i int
-	for i = hz - 2; i >= 0 && model[i] != ' '; i-- {
-		if model[i] >= '0' && model[i] <= '9' {
-			freq += int64(model[i]-'0') * decimalShift
-			decimalShift *= 10
-		} else if model[i] == '.' {
-			if divisor != 0 {
-				return
-			}
-			divisor = decimalShift
-		} else {
-			return
-		}
-	}
-	// we didn't find a space
-	if i < 0 {
-		return
-	}
-	if divisor != 0 {
-		c.Hz = (freq * multiplier) / divisor
-		return
-	}
-	c.Hz = freq * multiplier
-}
-
-// VM Will return true if the cpu id indicates we are in
-// a virtual machine.
-func (c CPUInfo) VM() bool {
-	return CPU.featureSet.inSet(HYPERVISOR)
-}
-
-// flags contains detected cpu features and characteristics
-type flags uint64
-
-// log2(bits_in_uint64)
-const flagBitsLog2 = 6
-const flagBits = 1 << flagBitsLog2
-const flagMask = flagBits - 1
-
-// flagSet contains detected cpu features and characteristics in an array of flags
-type flagSet [(lastID + flagMask) / flagBits]flags
-
-func (s flagSet) inSet(feat FeatureID) bool {
-	return s[feat>>flagBitsLog2]&(1<<(feat&flagMask)) != 0
-}
-
-func (s *flagSet) set(feat FeatureID) {
-	s[feat>>flagBitsLog2] |= 1 << (feat & flagMask)
-}
-
-// setIf will set a feature if boolean is true.
-func (s *flagSet) setIf(cond bool, features ...FeatureID) {
-	if cond {
-		for _, offset := range features {
-			s[offset>>flagBitsLog2] |= 1 << (offset & flagMask)
-		}
-	}
-}
-
-func (s *flagSet) unset(offset FeatureID) {
-	bit := flags(1 << (offset & flagMask))
-	s[offset>>flagBitsLog2] = s[offset>>flagBitsLog2] & ^bit
-}
-
-// or with another flagset.
-func (s *flagSet) or(other flagSet) {
-	for i, v := range other[:] {
-		s[i] |= v
-	}
-}
-
-// ParseFeature will parse the string and return the ID of the matching feature.
-// Will return UNKNOWN if not found.
-func ParseFeature(s string) FeatureID {
-	s = strings.ToUpper(s)
-	for i := firstID; i < lastID; i++ {
-		if i.String() == s {
-			return i
-		}
-	}
-	return UNKNOWN
-}
-
-// Strings returns an array of the detected features for FlagsSet.
-func (s flagSet) Strings() []string {
-	if len(s) == 0 {
-		return []string{""}
-	}
-	r := make([]string, 0)
-	for i := firstID; i < lastID; i++ {
-		if s.inSet(i) {
-			r = append(r, i.String())
-		}
-	}
-	return r
-}
-
-func maxExtendedFunction() uint32 {
-	eax, _, _, _ := cpuid(0x80000000)
-	return eax
-}
-
-func maxFunctionID() uint32 {
-	a, _, _, _ := cpuid(0)
-	return a
-}
-
-func brandName() string {
-	if maxExtendedFunction() >= 0x80000004 {
-		v := make([]uint32, 0, 48)
-		for i := uint32(0); i < 3; i++ {
-			a, b, c, d := cpuid(0x80000002 + i)
-			v = append(v, a, b, c, d)
-		}
-		return strings.Trim(string(valAsString(v...)), " ")
-	}
-	return "unknown"
-}
-
-func threadsPerCore() int {
-	mfi := maxFunctionID()
-	vend, _ := vendorID()
-
-	if mfi < 0x4 || (vend != Intel && vend != AMD) {
-		return 1
-	}
-
-	if mfi < 0xb {
-		if vend != Intel {
-			return 1
-		}
-		_, b, _, d := cpuid(1)
-		if (d & (1 << 28)) != 0 {
-			// v will contain logical core count
-			v := (b >> 16) & 255
-			if v > 1 {
-				a4, _, _, _ := cpuid(4)
-				// physical cores
-				v2 := (a4 >> 26) + 1
-				if v2 > 0 {
-					return int(v) / int(v2)
-				}
-			}
-		}
-		return 1
-	}
-	_, b, _, _ := cpuidex(0xb, 0)
-	if b&0xffff == 0 {
-		if vend == AMD {
-			// Workaround for AMD returning 0, assume 2 if >= Zen 2
-			// It will be more correct than not.
-			fam, _ := familyModel()
-			_, _, _, d := cpuid(1)
-			if (d&(1<<28)) != 0 && fam >= 23 {
-				return 2
-			}
-		}
-		return 1
-	}
-	return int(b & 0xffff)
-}
-
-func logicalCores() int {
-	mfi := maxFunctionID()
-	v, _ := vendorID()
-	switch v {
-	case Intel:
-		// Use this on old Intel processors
-		if mfi < 0xb {
-			if mfi < 1 {
-				return 0
-			}
-			// CPUID.1:EBX[23:16] represents the maximum number of addressable IDs (initial APIC ID)
-			// that can be assigned to logical processors in a physical package.
-			// The value may not be the same as the number of logical processors that are present in the hardware of a physical package.
-			_, ebx, _, _ := cpuid(1)
-			logical := (ebx >> 16) & 0xff
-			return int(logical)
-		}
-		_, b, _, _ := cpuidex(0xb, 1)
-		return int(b & 0xffff)
-	case AMD, Hygon:
-		_, b, _, _ := cpuid(1)
-		return int((b >> 16) & 0xff)
-	default:
-		return 0
-	}
-}
-
-func familyModel() (int, int) {
-	if maxFunctionID() < 0x1 {
-		return 0, 0
-	}
-	eax, _, _, _ := cpuid(1)
-	family := ((eax >> 8) & 0xf) + ((eax >> 20) & 0xff)
-	model := ((eax >> 4) & 0xf) + ((eax >> 12) & 0xf0)
-	return int(family), int(model)
-}
-
-func physicalCores() int {
-	v, _ := vendorID()
-	switch v {
-	case Intel:
-		return logicalCores() / threadsPerCore()
-	case AMD, Hygon:
-		lc := logicalCores()
-		tpc := threadsPerCore()
-		if lc > 0 && tpc > 0 {
-			return lc / tpc
-		}
-
-		// The following is inaccurate on AMD EPYC 7742 64-Core Processor
-		if maxExtendedFunction() >= 0x80000008 {
-			_, _, c, _ := cpuid(0x80000008)
-			if c&0xff > 0 {
-				return int(c&0xff) + 1
-			}
-		}
-	}
-	return 0
-}
-
-// Except from http://en.wikipedia.org/wiki/CPUID#EAX.3D0:_Get_vendor_ID
-var vendorMapping = map[string]Vendor{
-	"AMDisbetter!": AMD,
-	"AuthenticAMD": AMD,
-	"CentaurHauls": VIA,
-	"GenuineIntel": Intel,
-	"TransmetaCPU": Transmeta,
-	"GenuineTMx86": Transmeta,
-	"Geode by NSC": NSC,
-	"VIA VIA VIA ": VIA,
-	"KVMKVMKVMKVM": KVM,
-	"Microsoft Hv": MSVM,
-	"VMwareVMware": VMware,
-	"XenVMMXenVMM": XenHVM,
-	"bhyve bhyve ": Bhyve,
-	"HygonGenuine": Hygon,
-	"Vortex86 SoC": SiS,
-	"SiS SiS SiS ": SiS,
-	"RiseRiseRise": SiS,
-	"Genuine  RDC": RDC,
-}
-
-func vendorID() (Vendor, string) {
-	_, b, c, d := cpuid(0)
-	v := string(valAsString(b, d, c))
-	vend, ok := vendorMapping[v]
-	if !ok {
-		return VendorUnknown, v
-	}
-	return vend, v
-}
-
-func cacheLine() int {
-	if maxFunctionID() < 0x1 {
-		return 0
-	}
-
-	_, ebx, _, _ := cpuid(1)
-	cache := (ebx & 0xff00) >> 5 // cflush size
-	if cache == 0 && maxExtendedFunction() >= 0x80000006 {
-		_, _, ecx, _ := cpuid(0x80000006)
-		cache = ecx & 0xff // cacheline size
-	}
-	// TODO: Read from Cache and TLB Information
-	return int(cache)
-}
-
-func (c *CPUInfo) cacheSize() {
-	c.Cache.L1D = -1
-	c.Cache.L1I = -1
-	c.Cache.L2 = -1
-	c.Cache.L3 = -1
-	vendor, _ := vendorID()
-	switch vendor {
-	case Intel:
-		if maxFunctionID() < 4 {
-			return
-		}
-		for i := uint32(0); ; i++ {
-			eax, ebx, ecx, _ := cpuidex(4, i)
-			cacheType := eax & 15
-			if cacheType == 0 {
-				break
-			}
-			cacheLevel := (eax >> 5) & 7
-			coherency := int(ebx&0xfff) + 1
-			partitions := int((ebx>>12)&0x3ff) + 1
-			associativity := int((ebx>>22)&0x3ff) + 1
-			sets := int(ecx) + 1
-			size := associativity * partitions * coherency * sets
-			switch cacheLevel {
-			case 1:
-				if cacheType == 1 {
-					// 1 = Data Cache
-					c.Cache.L1D = size
-				} else if cacheType == 2 {
-					// 2 = Instruction Cache
-					c.Cache.L1I = size
-				} else {
-					if c.Cache.L1D < 0 {
-						c.Cache.L1I = size
-					}
-					if c.Cache.L1I < 0 {
-						c.Cache.L1I = size
-					}
-				}
-			case 2:
-				c.Cache.L2 = size
-			case 3:
-				c.Cache.L3 = size
-			}
-		}
-	case AMD, Hygon:
-		// Untested.
-		if maxExtendedFunction() < 0x80000005 {
-			return
-		}
-		_, _, ecx, edx := cpuid(0x80000005)
-		c.Cache.L1D = int(((ecx >> 24) & 0xFF) * 1024)
-		c.Cache.L1I = int(((edx >> 24) & 0xFF) * 1024)
-
-		if maxExtendedFunction() < 0x80000006 {
-			return
-		}
-		_, _, ecx, _ = cpuid(0x80000006)
-		c.Cache.L2 = int(((ecx >> 16) & 0xFFFF) * 1024)
-
-		// CPUID Fn8000_001D_EAX_x[N:0] Cache Properties
-		if maxExtendedFunction() < 0x8000001D {
-			return
-		}
-		for i := uint32(0); i < math.MaxUint32; i++ {
-			eax, ebx, ecx, _ := cpuidex(0x8000001D, i)
-
-			level := (eax >> 5) & 7
-			cacheNumSets := ecx + 1
-			cacheLineSize := 1 + (ebx & 2047)
-			cachePhysPartitions := 1 + ((ebx >> 12) & 511)
-			cacheNumWays := 1 + ((ebx >> 22) & 511)
-
-			typ := eax & 15
-			size := int(cacheNumSets * cacheLineSize * cachePhysPartitions * cacheNumWays)
-			if typ == 0 {
-				return
-			}
-
-			switch level {
-			case 1:
-				switch typ {
-				case 1:
-					// Data cache
-					c.Cache.L1D = size
-				case 2:
-					// Inst cache
-					c.Cache.L1I = size
-				default:
-					if c.Cache.L1D < 0 {
-						c.Cache.L1I = size
-					}
-					if c.Cache.L1I < 0 {
-						c.Cache.L1I = size
-					}
-				}
-			case 2:
-				c.Cache.L2 = size
-			case 3:
-				c.Cache.L3 = size
-			}
-		}
-	}
-
-	return
-}
-
-type SGXEPCSection struct {
-	BaseAddress uint64
-	EPCSize     uint64
-}
-
-type SGXSupport struct {
-	Available           bool
-	LaunchControl       bool
-	SGX1Supported       bool
-	SGX2Supported       bool
-	MaxEnclaveSizeNot64 int64
-	MaxEnclaveSize64    int64
-	EPCSections         []SGXEPCSection
-}
-
-func hasSGX(available, lc bool) (rval SGXSupport) {
-	rval.Available = available
-
-	if !available {
-		return
-	}
-
-	rval.LaunchControl = lc
-
-	a, _, _, d := cpuidex(0x12, 0)
-	rval.SGX1Supported = a&0x01 != 0
-	rval.SGX2Supported = a&0x02 != 0
-	rval.MaxEnclaveSizeNot64 = 1 << (d & 0xFF)     // pow 2
-	rval.MaxEnclaveSize64 = 1 << ((d >> 8) & 0xFF) // pow 2
-	rval.EPCSections = make([]SGXEPCSection, 0)
-
-	for subleaf := uint32(2); subleaf < 2+8; subleaf++ {
-		eax, ebx, ecx, edx := cpuidex(0x12, subleaf)
-		leafType := eax & 0xf
-
-		if leafType == 0 {
-			// Invalid subleaf, stop iterating
-			break
-		} else if leafType == 1 {
-			// EPC Section subleaf
-			baseAddress := uint64(eax&0xfffff000) + (uint64(ebx&0x000fffff) << 32)
-			size := uint64(ecx&0xfffff000) + (uint64(edx&0x000fffff) << 32)
-
-			section := SGXEPCSection{BaseAddress: baseAddress, EPCSize: size}
-			rval.EPCSections = append(rval.EPCSections, section)
-		}
-	}
-
-	return
-}
-
-func support() flagSet {
-	var fs flagSet
-	mfi := maxFunctionID()
-	vend, _ := vendorID()
-	if mfi < 0x1 {
-		return fs
-	}
-	family, model := familyModel()
-
-	_, _, c, d := cpuid(1)
-	fs.setIf((d&(1<<15)) != 0, CMOV)
-	fs.setIf((d&(1<<23)) != 0, MMX)
-	fs.setIf((d&(1<<25)) != 0, MMXEXT)
-	fs.setIf((d&(1<<25)) != 0, SSE)
-	fs.setIf((d&(1<<26)) != 0, SSE2)
-	fs.setIf((c&1) != 0, SSE3)
-	fs.setIf((c&(1<<5)) != 0, VMX)
-	fs.setIf((c&0x00000200) != 0, SSSE3)
-	fs.setIf((c&0x00080000) != 0, SSE4)
-	fs.setIf((c&0x00100000) != 0, SSE42)
-	fs.setIf((c&(1<<25)) != 0, AESNI)
-	fs.setIf((c&(1<<1)) != 0, CLMUL)
-	fs.setIf(c&(1<<23) != 0, POPCNT)
-	fs.setIf(c&(1<<30) != 0, RDRAND)
-
-	// This bit has been reserved by Intel & AMD for use by hypervisors,
-	// and indicates the presence of a hypervisor.
-	fs.setIf(c&(1<<31) != 0, HYPERVISOR)
-	fs.setIf(c&(1<<29) != 0, F16C)
-	fs.setIf(c&(1<<13) != 0, CX16)
-
-	if vend == Intel && (d&(1<<28)) != 0 && mfi >= 4 {
-		fs.setIf(threadsPerCore() > 1, HTT)
-	}
-	if vend == AMD && (d&(1<<28)) != 0 && mfi >= 4 {
-		fs.setIf(threadsPerCore() > 1, HTT)
-	}
-	// Check XGETBV/XSAVE (26), OXSAVE (27) and AVX (28) bits
-	const avxCheck = 1<<26 | 1<<27 | 1<<28
-	if c&avxCheck == avxCheck {
-		// Check for OS support
-		eax, _ := xgetbv(0)
-		if (eax & 0x6) == 0x6 {
-			fs.set(AVX)
-			switch vend {
-			case Intel:
-				// Older than Haswell.
-				fs.setIf(family == 6 && model < 60, AVXSLOW)
-			case AMD:
-				// Older than Zen 2
-				fs.setIf(family < 23 || (family == 23 && model < 49), AVXSLOW)
-			}
-		}
-	}
-	// FMA3 can be used with SSE registers, so no OS support is strictly needed.
-	// fma3 and OSXSAVE needed.
-	const fma3Check = 1<<12 | 1<<27
-	fs.setIf(c&fma3Check == fma3Check, FMA3)
-
-	// Check AVX2, AVX2 requires OS support, but BMI1/2 don't.
-	if mfi >= 7 {
-		_, ebx, ecx, edx := cpuidex(7, 0)
-		eax1, _, _, _ := cpuidex(7, 1)
-		if fs.inSet(AVX) && (ebx&0x00000020) != 0 {
-			fs.set(AVX2)
-		}
-		// CPUID.(EAX=7, ECX=0).EBX
-		if (ebx & 0x00000008) != 0 {
-			fs.set(BMI1)
-			fs.setIf((ebx&0x00000100) != 0, BMI2)
-		}
-		fs.setIf(ebx&(1<<2) != 0, SGX)
-		fs.setIf(ebx&(1<<4) != 0, HLE)
-		fs.setIf(ebx&(1<<9) != 0, ERMS)
-		fs.setIf(ebx&(1<<11) != 0, RTM)
-		fs.setIf(ebx&(1<<14) != 0, MPX)
-		fs.setIf(ebx&(1<<18) != 0, RDSEED)
-		fs.setIf(ebx&(1<<19) != 0, ADX)
-		fs.setIf(ebx&(1<<29) != 0, SHA)
-		// CPUID.(EAX=7, ECX=0).ECX
-		fs.setIf(ecx&(1<<5) != 0, WAITPKG)
-		fs.setIf(ecx&(1<<25) != 0, CLDEMOTE)
-		fs.setIf(ecx&(1<<27) != 0, MOVDIRI)
-		fs.setIf(ecx&(1<<28) != 0, MOVDIR64B)
-		fs.setIf(ecx&(1<<29) != 0, ENQCMD)
-		fs.setIf(ecx&(1<<30) != 0, SGXLC)
-		// CPUID.(EAX=7, ECX=0).EDX
-		fs.setIf(edx&(1<<11) != 0, RTM_ALWAYS_ABORT)
-		fs.setIf(edx&(1<<14) != 0, SERIALIZE)
-		fs.setIf(edx&(1<<16) != 0, TSXLDTRK)
-		fs.setIf(edx&(1<<26) != 0, IBPB)
-		fs.setIf(edx&(1<<27) != 0, STIBP)
-
-		// Only detect AVX-512 features if XGETBV is supported
-		if c&((1<<26)|(1<<27)) == (1<<26)|(1<<27) {
-			// Check for OS support
-			eax, _ := xgetbv(0)
-
-			// Verify that XCR0[7:5] = ‘111b’ (OPMASK state, upper 256-bit of ZMM0-ZMM15 and
-			// ZMM16-ZMM31 state are enabled by OS)
-			/// and that XCR0[2:1] = ‘11b’ (XMM state and YMM state are enabled by OS).
-			hasAVX512 := (eax>>5)&7 == 7 && (eax>>1)&3 == 3
-			if runtime.GOOS == "darwin" {
-				hasAVX512 = fs.inSet(AVX) && darwinHasAVX512()
-			}
-			if hasAVX512 {
-				fs.setIf(ebx&(1<<16) != 0, AVX512F)
-				fs.setIf(ebx&(1<<17) != 0, AVX512DQ)
-				fs.setIf(ebx&(1<<21) != 0, AVX512IFMA)
-				fs.setIf(ebx&(1<<26) != 0, AVX512PF)
-				fs.setIf(ebx&(1<<27) != 0, AVX512ER)
-				fs.setIf(ebx&(1<<28) != 0, AVX512CD)
-				fs.setIf(ebx&(1<<30) != 0, AVX512BW)
-				fs.setIf(ebx&(1<<31) != 0, AVX512VL)
-				// ecx
-				fs.setIf(ecx&(1<<1) != 0, AVX512VBMI)
-				fs.setIf(ecx&(1<<6) != 0, AVX512VBMI2)
-				fs.setIf(ecx&(1<<8) != 0, GFNI)
-				fs.setIf(ecx&(1<<9) != 0, VAES)
-				fs.setIf(ecx&(1<<10) != 0, VPCLMULQDQ)
-				fs.setIf(ecx&(1<<11) != 0, AVX512VNNI)
-				fs.setIf(ecx&(1<<12) != 0, AVX512BITALG)
-				fs.setIf(ecx&(1<<14) != 0, AVX512VPOPCNTDQ)
-				// edx
-				fs.setIf(edx&(1<<8) != 0, AVX512VP2INTERSECT)
-				fs.setIf(edx&(1<<22) != 0, AMXBF16)
-				fs.setIf(edx&(1<<23) != 0, AVX512FP16)
-				fs.setIf(edx&(1<<24) != 0, AMXTILE)
-				fs.setIf(edx&(1<<25) != 0, AMXINT8)
-				// eax1 = CPUID.(EAX=7, ECX=1).EAX
-				fs.setIf(eax1&(1<<5) != 0, AVX512BF16)
-			}
-		}
-	}
-
-	if maxExtendedFunction() >= 0x80000001 {
-		_, _, c, d := cpuid(0x80000001)
-		if (c & (1 << 5)) != 0 {
-			fs.set(LZCNT)
-			fs.set(POPCNT)
-		}
-		fs.setIf((c&(1<<10)) != 0, IBS)
-		fs.setIf((d&(1<<31)) != 0, AMD3DNOW)
-		fs.setIf((d&(1<<30)) != 0, AMD3DNOWEXT)
-		fs.setIf((d&(1<<23)) != 0, MMX)
-		fs.setIf((d&(1<<22)) != 0, MMXEXT)
-		fs.setIf((c&(1<<6)) != 0, SSE4A)
-		fs.setIf(d&(1<<20) != 0, NX)
-		fs.setIf(d&(1<<27) != 0, RDTSCP)
-
-		/* XOP and FMA4 use the AVX instruction coding scheme, so they can't be
-		 * used unless the OS has AVX support. */
-		if fs.inSet(AVX) {
-			fs.setIf((c&0x00000800) != 0, XOP)
-			fs.setIf((c&0x00010000) != 0, FMA4)
-		}
-
-	}
-	if maxExtendedFunction() >= 0x80000007 {
-		_, b, _, d := cpuid(0x80000007)
-		fs.setIf((b&(1<<0)) != 0, MCAOVERFLOW)
-		fs.setIf((b&(1<<1)) != 0, SUCCOR)
-		fs.setIf((b&(1<<2)) != 0, HWA)
-		fs.setIf((d&(1<<9)) != 0, CPBOOST)
-	}
-
-	if maxExtendedFunction() >= 0x80000008 {
-		_, b, _, _ := cpuid(0x80000008)
-		fs.setIf((b&(1<<9)) != 0, WBNOINVD)
-		fs.setIf((b&(1<<8)) != 0, MCOMMIT)
-		fs.setIf((b&(1<<13)) != 0, INT_WBINVD)
-		fs.setIf((b&(1<<4)) != 0, RDPRU)
-		fs.setIf((b&(1<<3)) != 0, INVLPGB)
-		fs.setIf((b&(1<<1)) != 0, MSRIRC)
-		fs.setIf((b&(1<<0)) != 0, CLZERO)
-	}
-
-	if maxExtendedFunction() >= 0x8000001b && fs.inSet(IBS) {
-		eax, _, _, _ := cpuid(0x8000001b)
-		fs.setIf((eax>>0)&1 == 1, IBSFFV)
-		fs.setIf((eax>>1)&1 == 1, IBSFETCHSAM)
-		fs.setIf((eax>>2)&1 == 1, IBSOPSAM)
-		fs.setIf((eax>>3)&1 == 1, IBSRDWROPCNT)
-		fs.setIf((eax>>4)&1 == 1, IBSOPCNT)
-		fs.setIf((eax>>5)&1 == 1, IBSBRNTRGT)
-		fs.setIf((eax>>6)&1 == 1, IBSOPCNTEXT)
-		fs.setIf((eax>>7)&1 == 1, IBSRIPINVALIDCHK)
-	}
-
-	return fs
-}
-
-func valAsString(values ...uint32) []byte {
-	r := make([]byte, 4*len(values))
-	for i, v := range values {
-		dst := r[i*4:]
-		dst[0] = byte(v & 0xff)
-		dst[1] = byte((v >> 8) & 0xff)
-		dst[2] = byte((v >> 16) & 0xff)
-		dst[3] = byte((v >> 24) & 0xff)
-		switch {
-		case dst[0] == 0:
-			return r[:i*4]
-		case dst[1] == 0:
-			return r[:i*4+1]
-		case dst[2] == 0:
-			return r[:i*4+2]
-		case dst[3] == 0:
-			return r[:i*4+3]
-		}
-	}
-	return r
-}
diff --git a/vendor/github.com/klauspost/cpuid/v2/cpuid_386.s b/vendor/github.com/klauspost/cpuid/v2/cpuid_386.s
deleted file mode 100644
index 8587c3a1fc..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/cpuid_386.s
+++ /dev/null
@@ -1,47 +0,0 @@
-// Copyright (c) 2015 Klaus Post, released under MIT License. See LICENSE file.
-
-//+build 386,!gccgo,!noasm,!appengine
-
-// func asmCpuid(op uint32) (eax, ebx, ecx, edx uint32)
-TEXT ·asmCpuid(SB), 7, $0
-	XORL CX, CX
-	MOVL op+0(FP), AX
-	CPUID
-	MOVL AX, eax+4(FP)
-	MOVL BX, ebx+8(FP)
-	MOVL CX, ecx+12(FP)
-	MOVL DX, edx+16(FP)
-	RET
-
-// func asmCpuidex(op, op2 uint32) (eax, ebx, ecx, edx uint32)
-TEXT ·asmCpuidex(SB), 7, $0
-	MOVL op+0(FP), AX
-	MOVL op2+4(FP), CX
-	CPUID
-	MOVL AX, eax+8(FP)
-	MOVL BX, ebx+12(FP)
-	MOVL CX, ecx+16(FP)
-	MOVL DX, edx+20(FP)
-	RET
-
-// func xgetbv(index uint32) (eax, edx uint32)
-TEXT ·asmXgetbv(SB), 7, $0
-	MOVL index+0(FP), CX
-	BYTE $0x0f; BYTE $0x01; BYTE $0xd0 // XGETBV
-	MOVL AX, eax+4(FP)
-	MOVL DX, edx+8(FP)
-	RET
-
-// func asmRdtscpAsm() (eax, ebx, ecx, edx uint32)
-TEXT ·asmRdtscpAsm(SB), 7, $0
-	BYTE $0x0F; BYTE $0x01; BYTE $0xF9 // RDTSCP
-	MOVL AX, eax+0(FP)
-	MOVL BX, ebx+4(FP)
-	MOVL CX, ecx+8(FP)
-	MOVL DX, edx+12(FP)
-	RET
-
-// func asmDarwinHasAVX512() bool
-TEXT ·asmDarwinHasAVX512(SB), 7, $0
-	MOVL $0, eax+0(FP)
-	RET
diff --git a/vendor/github.com/klauspost/cpuid/v2/cpuid_amd64.s b/vendor/github.com/klauspost/cpuid/v2/cpuid_amd64.s
deleted file mode 100644
index bc11f89421..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/cpuid_amd64.s
+++ /dev/null
@@ -1,72 +0,0 @@
-// Copyright (c) 2015 Klaus Post, released under MIT License. See LICENSE file.
-
-//+build amd64,!gccgo,!noasm,!appengine
-
-// func asmCpuid(op uint32) (eax, ebx, ecx, edx uint32)
-TEXT ·asmCpuid(SB), 7, $0
-	XORQ CX, CX
-	MOVL op+0(FP), AX
-	CPUID
-	MOVL AX, eax+8(FP)
-	MOVL BX, ebx+12(FP)
-	MOVL CX, ecx+16(FP)
-	MOVL DX, edx+20(FP)
-	RET
-
-// func asmCpuidex(op, op2 uint32) (eax, ebx, ecx, edx uint32)
-TEXT ·asmCpuidex(SB), 7, $0
-	MOVL op+0(FP), AX
-	MOVL op2+4(FP), CX
-	CPUID
-	MOVL AX, eax+8(FP)
-	MOVL BX, ebx+12(FP)
-	MOVL CX, ecx+16(FP)
-	MOVL DX, edx+20(FP)
-	RET
-
-// func asmXgetbv(index uint32) (eax, edx uint32)
-TEXT ·asmXgetbv(SB), 7, $0
-	MOVL index+0(FP), CX
-	BYTE $0x0f; BYTE $0x01; BYTE $0xd0 // XGETBV
-	MOVL AX, eax+8(FP)
-	MOVL DX, edx+12(FP)
-	RET
-
-// func asmRdtscpAsm() (eax, ebx, ecx, edx uint32)
-TEXT ·asmRdtscpAsm(SB), 7, $0
-	BYTE $0x0F; BYTE $0x01; BYTE $0xF9 // RDTSCP
-	MOVL AX, eax+0(FP)
-	MOVL BX, ebx+4(FP)
-	MOVL CX, ecx+8(FP)
-	MOVL DX, edx+12(FP)
-	RET
-
-// From https://go-review.googlesource.com/c/sys/+/285572/
-// func asmDarwinHasAVX512() bool
-TEXT ·asmDarwinHasAVX512(SB), 7, $0-1
-	MOVB $0, ret+0(FP) // default to false
-
-#ifdef GOOS_darwin // return if not darwin
-#ifdef GOARCH_amd64 // return if not amd64
-// These values from:
-// https://github.com/apple/darwin-xnu/blob/xnu-4570.1.46/osfmk/i386/cpu_capabilities.h
-#define commpage64_base_address         0x00007fffffe00000
-#define commpage64_cpu_capabilities64   (commpage64_base_address+0x010)
-#define commpage64_version              (commpage64_base_address+0x01E)
-#define hasAVX512F                      0x0000004000000000
-	MOVQ $commpage64_version, BX
-	MOVW (BX), AX
-	CMPW AX, $13                            // versions < 13 do not support AVX512
-	JL   no_avx512
-	MOVQ $commpage64_cpu_capabilities64, BX
-	MOVQ (BX), AX
-	MOVQ $hasAVX512F, CX
-	ANDQ CX, AX
-	JZ   no_avx512
-	MOVB $1, ret+0(FP)
-
-no_avx512:
-#endif
-#endif
-	RET
-
diff --git a/vendor/github.com/klauspost/cpuid/v2/cpuid_arm64.s b/vendor/github.com/klauspost/cpuid/v2/cpuid_arm64.s
deleted file mode 100644
index b31d6aec43..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/cpuid_arm64.s
+++ /dev/null
@@ -1,26 +0,0 @@
-// Copyright (c) 2015 Klaus Post, released under MIT License. See LICENSE file.
-
-//+build arm64,!gccgo,!noasm,!appengine
-
-// See https://www.kernel.org/doc/Documentation/arm64/cpu-feature-registers.txt
-
-// func getMidr
-TEXT ·getMidr(SB), 7, $0
-	WORD $0xd5380000    // mrs x0, midr_el1         /* Main ID Register */
-	MOVD R0, midr+0(FP)
-	RET
-
-// func getProcFeatures
-TEXT ·getProcFeatures(SB), 7, $0
-	WORD $0xd5380400            // mrs x0, id_aa64pfr0_el1  /* Processor Feature Register 0 */
-	MOVD R0, procFeatures+0(FP)
-	RET
-
-// func getInstAttributes
-TEXT ·getInstAttributes(SB), 7, $0
-	WORD $0xd5380600            // mrs x0, id_aa64isar0_el1 /* Instruction Set Attribute Register 0 */
-	WORD $0xd5380621            // mrs x1, id_aa64isar1_el1 /* Instruction Set Attribute Register 1 */
-	MOVD R0, instAttrReg0+0(FP)
-	MOVD R1, instAttrReg1+8(FP)
-	RET
-
diff --git a/vendor/github.com/klauspost/cpuid/v2/detect_arm64.go b/vendor/github.com/klauspost/cpuid/v2/detect_arm64.go
deleted file mode 100644
index 9bf9f77f37..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/detect_arm64.go
+++ /dev/null
@@ -1,246 +0,0 @@
-// Copyright (c) 2015 Klaus Post, released under MIT License. See LICENSE file.
-
-//+build arm64,!gccgo,!noasm,!appengine
-
-package cpuid
-
-import "runtime"
-
-func getMidr() (midr uint64)
-func getProcFeatures() (procFeatures uint64)
-func getInstAttributes() (instAttrReg0, instAttrReg1 uint64)
-
-func initCPU() {
-	cpuid = func(uint32) (a, b, c, d uint32) { return 0, 0, 0, 0 }
-	cpuidex = func(x, y uint32) (a, b, c, d uint32) { return 0, 0, 0, 0 }
-	xgetbv = func(uint32) (a, b uint32) { return 0, 0 }
-	rdtscpAsm = func() (a, b, c, d uint32) { return 0, 0, 0, 0 }
-}
-
-func addInfo(c *CPUInfo, safe bool) {
-	// Seems to be safe to assume on ARM64
-	c.CacheLine = 64
-	detectOS(c)
-
-	// ARM64 disabled since it may crash if interrupt is not intercepted by OS.
-	if safe && !c.Supports(ARMCPUID) && runtime.GOOS != "freebsd" {
-		return
-	}
-	midr := getMidr()
-
-	// MIDR_EL1 - Main ID Register
-	// https://developer.arm.com/docs/ddi0595/h/aarch64-system-registers/midr_el1
-	//  x--------------------------------------------------x
-	//  | Name                         |  bits   | visible |
-	//  |--------------------------------------------------|
-	//  | Implementer                  | [31-24] |    y    |
-	//  |--------------------------------------------------|
-	//  | Variant                      | [23-20] |    y    |
-	//  |--------------------------------------------------|
-	//  | Architecture                 | [19-16] |    y    |
-	//  |--------------------------------------------------|
-	//  | PartNum                      | [15-4]  |    y    |
-	//  |--------------------------------------------------|
-	//  | Revision                     | [3-0]   |    y    |
-	//  x--------------------------------------------------x
-
-	switch (midr >> 24) & 0xff {
-	case 0xC0:
-		c.VendorString = "Ampere Computing"
-		c.VendorID = Ampere
-	case 0x41:
-		c.VendorString = "Arm Limited"
-		c.VendorID = ARM
-	case 0x42:
-		c.VendorString = "Broadcom Corporation"
-		c.VendorID = Broadcom
-	case 0x43:
-		c.VendorString = "Cavium Inc"
-		c.VendorID = Cavium
-	case 0x44:
-		c.VendorString = "Digital Equipment Corporation"
-		c.VendorID = DEC
-	case 0x46:
-		c.VendorString = "Fujitsu Ltd"
-		c.VendorID = Fujitsu
-	case 0x49:
-		c.VendorString = "Infineon Technologies AG"
-		c.VendorID = Infineon
-	case 0x4D:
-		c.VendorString = "Motorola or Freescale Semiconductor Inc"
-		c.VendorID = Motorola
-	case 0x4E:
-		c.VendorString = "NVIDIA Corporation"
-		c.VendorID = NVIDIA
-	case 0x50:
-		c.VendorString = "Applied Micro Circuits Corporation"
-		c.VendorID = AMCC
-	case 0x51:
-		c.VendorString = "Qualcomm Inc"
-		c.VendorID = Qualcomm
-	case 0x56:
-		c.VendorString = "Marvell International Ltd"
-		c.VendorID = Marvell
-	case 0x69:
-		c.VendorString = "Intel Corporation"
-		c.VendorID = Intel
-	}
-
-	// Lower 4 bits: Architecture
-	// Architecture	Meaning
-	// 0b0001		Armv4.
-	// 0b0010		Armv4T.
-	// 0b0011		Armv5 (obsolete).
-	// 0b0100		Armv5T.
-	// 0b0101		Armv5TE.
-	// 0b0110		Armv5TEJ.
-	// 0b0111		Armv6.
-	// 0b1111		Architectural features are individually identified in the ID_* registers, see 'ID registers'.
-	// Upper 4 bit: Variant
-	// An IMPLEMENTATION DEFINED variant number.
-	// Typically, this field is used to distinguish between different product variants, or major revisions of a product.
-	c.Family = int(midr>>16) & 0xff
-
-	// PartNum, bits [15:4]
-	// An IMPLEMENTATION DEFINED primary part number for the device.
-	// On processors implemented by Arm, if the top four bits of the primary
-	// part number are 0x0 or 0x7, the variant and architecture are encoded differently.
-	// Revision, bits [3:0]
-	// An IMPLEMENTATION DEFINED revision number for the device.
-	c.Model = int(midr) & 0xffff
-
-	procFeatures := getProcFeatures()
-
-	// ID_AA64PFR0_EL1 - Processor Feature Register 0
-	// x--------------------------------------------------x
-	// | Name                         |  bits   | visible |
-	// |--------------------------------------------------|
-	// | DIT                          | [51-48] |    y    |
-	// |--------------------------------------------------|
-	// | SVE                          | [35-32] |    y    |
-	// |--------------------------------------------------|
-	// | GIC                          | [27-24] |    n    |
-	// |--------------------------------------------------|
-	// | AdvSIMD                      | [23-20] |    y    |
-	// |--------------------------------------------------|
-	// | FP                           | [19-16] |    y    |
-	// |--------------------------------------------------|
-	// | EL3                          | [15-12] |    n    |
-	// |--------------------------------------------------|
-	// | EL2                          | [11-8]  |    n    |
-	// |--------------------------------------------------|
-	// | EL1                          | [7-4]   |    n    |
-	// |--------------------------------------------------|
-	// | EL0                          | [3-0]   |    n    |
-	// x--------------------------------------------------x
-
-	var f flagSet
-	// if procFeatures&(0xf<<48) != 0 {
-	// 	fmt.Println("DIT")
-	// }
-	f.setIf(procFeatures&(0xf<<32) != 0, SVE)
-	if procFeatures&(0xf<<20) != 15<<20 {
-		f.set(ASIMD)
-		// https://developer.arm.com/docs/ddi0595/b/aarch64-system-registers/id_aa64pfr0_el1
-		// 0b0001 --> As for 0b0000, and also includes support for half-precision floating-point arithmetic.
-		f.setIf(procFeatures&(0xf<<20) == 1<<20, FPHP, ASIMDHP)
-	}
-	f.setIf(procFeatures&(0xf<<16) != 0, FP)
-
-	instAttrReg0, instAttrReg1 := getInstAttributes()
-
-	// https://developer.arm.com/docs/ddi0595/b/aarch64-system-registers/id_aa64isar0_el1
-	//
-	// ID_AA64ISAR0_EL1 - Instruction Set Attribute Register 0
-	// x--------------------------------------------------x
-	// | Name                         |  bits   | visible |
-	// |--------------------------------------------------|
-	// | TS                           | [55-52] |    y    |
-	// |--------------------------------------------------|
-	// | FHM                          | [51-48] |    y    |
-	// |--------------------------------------------------|
-	// | DP                           | [47-44] |    y    |
-	// |--------------------------------------------------|
-	// | SM4                          | [43-40] |    y    |
-	// |--------------------------------------------------|
-	// | SM3                          | [39-36] |    y    |
-	// |--------------------------------------------------|
-	// | SHA3                         | [35-32] |    y    |
-	// |--------------------------------------------------|
-	// | RDM                          | [31-28] |    y    |
-	// |--------------------------------------------------|
-	// | ATOMICS                      | [23-20] |    y    |
-	// |--------------------------------------------------|
-	// | CRC32                        | [19-16] |    y    |
-	// |--------------------------------------------------|
-	// | SHA2                         | [15-12] |    y    |
-	// |--------------------------------------------------|
-	// | SHA1                         | [11-8]  |    y    |
-	// |--------------------------------------------------|
-	// | AES                          | [7-4]   |    y    |
-	// x--------------------------------------------------x
-
-	// if instAttrReg0&(0xf<<52) != 0 {
-	// 	fmt.Println("TS")
-	// }
-	// if instAttrReg0&(0xf<<48) != 0 {
-	// 	fmt.Println("FHM")
-	// }
-	f.setIf(instAttrReg0&(0xf<<44) != 0, ASIMDDP)
-	f.setIf(instAttrReg0&(0xf<<40) != 0, SM4)
-	f.setIf(instAttrReg0&(0xf<<36) != 0, SM3)
-	f.setIf(instAttrReg0&(0xf<<32) != 0, SHA3)
-	f.setIf(instAttrReg0&(0xf<<28) != 0, ASIMDRDM)
-	f.setIf(instAttrReg0&(0xf<<20) != 0, ATOMICS)
-	f.setIf(instAttrReg0&(0xf<<16) != 0, CRC32)
-	f.setIf(instAttrReg0&(0xf<<12) != 0, SHA2)
-	// https://developer.arm.com/docs/ddi0595/b/aarch64-system-registers/id_aa64isar0_el1
-	// 0b0010 --> As 0b0001, plus SHA512H, SHA512H2, SHA512SU0, and SHA512SU1 instructions implemented.
-	f.setIf(instAttrReg0&(0xf<<12) == 2<<12, SHA512)
-	f.setIf(instAttrReg0&(0xf<<8) != 0, SHA1)
-	f.setIf(instAttrReg0&(0xf<<4) != 0, AESARM)
-	// https://developer.arm.com/docs/ddi0595/b/aarch64-system-registers/id_aa64isar0_el1
-	// 0b0010 --> As for 0b0001, plus PMULL/PMULL2 instructions operating on 64-bit data quantities.
-	f.setIf(instAttrReg0&(0xf<<4) == 2<<4, PMULL)
-
-	// https://developer.arm.com/docs/ddi0595/b/aarch64-system-registers/id_aa64isar1_el1
-	//
-	// ID_AA64ISAR1_EL1 - Instruction set attribute register 1
-	// x--------------------------------------------------x
-	// | Name                         |  bits   | visible |
-	// |--------------------------------------------------|
-	// | GPI                          | [31-28] |    y    |
-	// |--------------------------------------------------|
-	// | GPA                          | [27-24] |    y    |
-	// |--------------------------------------------------|
-	// | LRCPC                        | [23-20] |    y    |
-	// |--------------------------------------------------|
-	// | FCMA                         | [19-16] |    y    |
-	// |--------------------------------------------------|
-	// | JSCVT                        | [15-12] |    y    |
-	// |--------------------------------------------------|
-	// | API                          | [11-8]  |    y    |
-	// |--------------------------------------------------|
-	// | APA                          | [7-4]   |    y    |
-	// |--------------------------------------------------|
-	// | DPB                          | [3-0]   |    y    |
-	// x--------------------------------------------------x
-
-	// if instAttrReg1&(0xf<<28) != 0 {
-	// 	fmt.Println("GPI")
-	// }
-	f.setIf(instAttrReg1&(0xf<<28) != 24, GPA)
-	f.setIf(instAttrReg1&(0xf<<20) != 0, LRCPC)
-	f.setIf(instAttrReg1&(0xf<<16) != 0, FCMA)
-	f.setIf(instAttrReg1&(0xf<<12) != 0, JSCVT)
-	// if instAttrReg1&(0xf<<8) != 0 {
-	// 	fmt.Println("API")
-	// }
-	// if instAttrReg1&(0xf<<4) != 0 {
-	// 	fmt.Println("APA")
-	// }
-	f.setIf(instAttrReg1&(0xf<<0) != 0, DCPOP)
-
-	// Store
-	c.featureSet.or(f)
-}
diff --git a/vendor/github.com/klauspost/cpuid/v2/detect_ref.go b/vendor/github.com/klauspost/cpuid/v2/detect_ref.go
deleted file mode 100644
index e9c8606ab9..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/detect_ref.go
+++ /dev/null
@@ -1,14 +0,0 @@
-// Copyright (c) 2015 Klaus Post, released under MIT License. See LICENSE file.
-
-//+build !amd64,!386,!arm64 gccgo noasm appengine
-
-package cpuid
-
-func initCPU() {
-	cpuid = func(uint32) (a, b, c, d uint32) { return 0, 0, 0, 0 }
-	cpuidex = func(x, y uint32) (a, b, c, d uint32) { return 0, 0, 0, 0 }
-	xgetbv = func(uint32) (a, b uint32) { return 0, 0 }
-	rdtscpAsm = func() (a, b, c, d uint32) { return 0, 0, 0, 0 }
-}
-
-func addInfo(info *CPUInfo, safe bool) {}
diff --git a/vendor/github.com/klauspost/cpuid/v2/detect_x86.go b/vendor/github.com/klauspost/cpuid/v2/detect_x86.go
deleted file mode 100644
index 367c35c88c..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/detect_x86.go
+++ /dev/null
@@ -1,35 +0,0 @@
-// Copyright (c) 2015 Klaus Post, released under MIT License. See LICENSE file.
-
-//+build 386,!gccgo,!noasm,!appengine amd64,!gccgo,!noasm,!appengine
-
-package cpuid
-
-func asmCpuid(op uint32) (eax, ebx, ecx, edx uint32)
-func asmCpuidex(op, op2 uint32) (eax, ebx, ecx, edx uint32)
-func asmXgetbv(index uint32) (eax, edx uint32)
-func asmRdtscpAsm() (eax, ebx, ecx, edx uint32)
-func asmDarwinHasAVX512() bool
-
-func initCPU() {
-	cpuid = asmCpuid
-	cpuidex = asmCpuidex
-	xgetbv = asmXgetbv
-	rdtscpAsm = asmRdtscpAsm
-	darwinHasAVX512 = asmDarwinHasAVX512
-}
-
-func addInfo(c *CPUInfo, safe bool) {
-	c.maxFunc = maxFunctionID()
-	c.maxExFunc = maxExtendedFunction()
-	c.BrandName = brandName()
-	c.CacheLine = cacheLine()
-	c.Family, c.Model = familyModel()
-	c.featureSet = support()
-	c.SGX = hasSGX(c.featureSet.inSet(SGX), c.featureSet.inSet(SGXLC))
-	c.ThreadsPerCore = threadsPerCore()
-	c.LogicalCores = logicalCores()
-	c.PhysicalCores = physicalCores()
-	c.VendorID, c.VendorString = vendorID()
-	c.cacheSize()
-	c.frequencies()
-}
diff --git a/vendor/github.com/klauspost/cpuid/v2/featureid_string.go b/vendor/github.com/klauspost/cpuid/v2/featureid_string.go
deleted file mode 100644
index b1fe42e467..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/featureid_string.go
+++ /dev/null
@@ -1,185 +0,0 @@
-// Code generated by "stringer -type=FeatureID,Vendor"; DO NOT EDIT.
-
-package cpuid
-
-import "strconv"
-
-func _() {
-	// An "invalid array index" compiler error signifies that the constant values have changed.
-	// Re-run the stringer command to generate them again.
-	var x [1]struct{}
-	_ = x[ADX-1]
-	_ = x[AESNI-2]
-	_ = x[AMD3DNOW-3]
-	_ = x[AMD3DNOWEXT-4]
-	_ = x[AMXBF16-5]
-	_ = x[AMXINT8-6]
-	_ = x[AMXTILE-7]
-	_ = x[AVX-8]
-	_ = x[AVX2-9]
-	_ = x[AVX512BF16-10]
-	_ = x[AVX512BITALG-11]
-	_ = x[AVX512BW-12]
-	_ = x[AVX512CD-13]
-	_ = x[AVX512DQ-14]
-	_ = x[AVX512ER-15]
-	_ = x[AVX512F-16]
-	_ = x[AVX512FP16-17]
-	_ = x[AVX512IFMA-18]
-	_ = x[AVX512PF-19]
-	_ = x[AVX512VBMI-20]
-	_ = x[AVX512VBMI2-21]
-	_ = x[AVX512VL-22]
-	_ = x[AVX512VNNI-23]
-	_ = x[AVX512VP2INTERSECT-24]
-	_ = x[AVX512VPOPCNTDQ-25]
-	_ = x[AVXSLOW-26]
-	_ = x[BMI1-27]
-	_ = x[BMI2-28]
-	_ = x[CLDEMOTE-29]
-	_ = x[CLMUL-30]
-	_ = x[CLZERO-31]
-	_ = x[CMOV-32]
-	_ = x[CPBOOST-33]
-	_ = x[CX16-34]
-	_ = x[ENQCMD-35]
-	_ = x[ERMS-36]
-	_ = x[F16C-37]
-	_ = x[FMA3-38]
-	_ = x[FMA4-39]
-	_ = x[GFNI-40]
-	_ = x[HLE-41]
-	_ = x[HTT-42]
-	_ = x[HWA-43]
-	_ = x[HYPERVISOR-44]
-	_ = x[IBPB-45]
-	_ = x[IBS-46]
-	_ = x[IBSBRNTRGT-47]
-	_ = x[IBSFETCHSAM-48]
-	_ = x[IBSFFV-49]
-	_ = x[IBSOPCNT-50]
-	_ = x[IBSOPCNTEXT-51]
-	_ = x[IBSOPSAM-52]
-	_ = x[IBSRDWROPCNT-53]
-	_ = x[IBSRIPINVALIDCHK-54]
-	_ = x[INT_WBINVD-55]
-	_ = x[INVLPGB-56]
-	_ = x[LZCNT-57]
-	_ = x[MCAOVERFLOW-58]
-	_ = x[MCOMMIT-59]
-	_ = x[MMX-60]
-	_ = x[MMXEXT-61]
-	_ = x[MOVDIR64B-62]
-	_ = x[MOVDIRI-63]
-	_ = x[MPX-64]
-	_ = x[MSRIRC-65]
-	_ = x[NX-66]
-	_ = x[POPCNT-67]
-	_ = x[RDPRU-68]
-	_ = x[RDRAND-69]
-	_ = x[RDSEED-70]
-	_ = x[RDTSCP-71]
-	_ = x[RTM-72]
-	_ = x[RTM_ALWAYS_ABORT-73]
-	_ = x[SERIALIZE-74]
-	_ = x[SGX-75]
-	_ = x[SGXLC-76]
-	_ = x[SHA-77]
-	_ = x[SSE-78]
-	_ = x[SSE2-79]
-	_ = x[SSE3-80]
-	_ = x[SSE4-81]
-	_ = x[SSE42-82]
-	_ = x[SSE4A-83]
-	_ = x[SSSE3-84]
-	_ = x[STIBP-85]
-	_ = x[SUCCOR-86]
-	_ = x[TBM-87]
-	_ = x[TSXLDTRK-88]
-	_ = x[VAES-89]
-	_ = x[VMX-90]
-	_ = x[VPCLMULQDQ-91]
-	_ = x[WAITPKG-92]
-	_ = x[WBNOINVD-93]
-	_ = x[XOP-94]
-	_ = x[AESARM-95]
-	_ = x[ARMCPUID-96]
-	_ = x[ASIMD-97]
-	_ = x[ASIMDDP-98]
-	_ = x[ASIMDHP-99]
-	_ = x[ASIMDRDM-100]
-	_ = x[ATOMICS-101]
-	_ = x[CRC32-102]
-	_ = x[DCPOP-103]
-	_ = x[EVTSTRM-104]
-	_ = x[FCMA-105]
-	_ = x[FP-106]
-	_ = x[FPHP-107]
-	_ = x[GPA-108]
-	_ = x[JSCVT-109]
-	_ = x[LRCPC-110]
-	_ = x[PMULL-111]
-	_ = x[SHA1-112]
-	_ = x[SHA2-113]
-	_ = x[SHA3-114]
-	_ = x[SHA512-115]
-	_ = x[SM3-116]
-	_ = x[SM4-117]
-	_ = x[SVE-118]
-	_ = x[lastID-119]
-	_ = x[firstID-0]
-}
-
-const _FeatureID_name = "firstIDADXAESNIAMD3DNOWAMD3DNOWEXTAMXBF16AMXINT8AMXTILEAVXAVX2AVX512BF16AVX512BITALGAVX512BWAVX512CDAVX512DQAVX512ERAVX512FAVX512FP16AVX512IFMAAVX512PFAVX512VBMIAVX512VBMI2AVX512VLAVX512VNNIAVX512VP2INTERSECTAVX512VPOPCNTDQAVXSLOWBMI1BMI2CLDEMOTECLMULCLZEROCMOVCPBOOSTCX16ENQCMDERMSF16CFMA3FMA4GFNIHLEHTTHWAHYPERVISORIBPBIBSIBSBRNTRGTIBSFETCHSAMIBSFFVIBSOPCNTIBSOPCNTEXTIBSOPSAMIBSRDWROPCNTIBSRIPINVALIDCHKINT_WBINVDINVLPGBLZCNTMCAOVERFLOWMCOMMITMMXMMXEXTMOVDIR64BMOVDIRIMPXMSRIRCNXPOPCNTRDPRURDRANDRDSEEDRDTSCPRTMRTM_ALWAYS_ABORTSERIALIZESGXSGXLCSHASSESSE2SSE3SSE4SSE42SSE4ASSSE3STIBPSUCCORTBMTSXLDTRKVAESVMXVPCLMULQDQWAITPKGWBNOINVDXOPAESARMARMCPUIDASIMDASIMDDPASIMDHPASIMDRDMATOMICSCRC32DCPOPEVTSTRMFCMAFPFPHPGPAJSCVTLRCPCPMULLSHA1SHA2SHA3SHA512SM3SM4SVElastID"
-
-var _FeatureID_index = [...]uint16{0, 7, 10, 15, 23, 34, 41, 48, 55, 58, 62, 72, 84, 92, 100, 108, 116, 123, 133, 143, 151, 161, 172, 180, 190, 208, 223, 230, 234, 238, 246, 251, 257, 261, 268, 272, 278, 282, 286, 290, 294, 298, 301, 304, 307, 317, 321, 324, 334, 345, 351, 359, 370, 378, 390, 406, 416, 423, 428, 439, 446, 449, 455, 464, 471, 474, 480, 482, 488, 493, 499, 505, 511, 514, 530, 539, 542, 547, 550, 553, 557, 561, 565, 570, 575, 580, 585, 591, 594, 602, 606, 609, 619, 626, 634, 637, 643, 651, 656, 663, 670, 678, 685, 690, 695, 702, 706, 708, 712, 715, 720, 725, 730, 734, 738, 742, 748, 751, 754, 757, 763}
-
-func (i FeatureID) String() string {
-	if i < 0 || i >= FeatureID(len(_FeatureID_index)-1) {
-		return "FeatureID(" + strconv.FormatInt(int64(i), 10) + ")"
-	}
-	return _FeatureID_name[_FeatureID_index[i]:_FeatureID_index[i+1]]
-}
-func _() {
-	// An "invalid array index" compiler error signifies that the constant values have changed.
-	// Re-run the stringer command to generate them again.
-	var x [1]struct{}
-	_ = x[VendorUnknown-0]
-	_ = x[Intel-1]
-	_ = x[AMD-2]
-	_ = x[VIA-3]
-	_ = x[Transmeta-4]
-	_ = x[NSC-5]
-	_ = x[KVM-6]
-	_ = x[MSVM-7]
-	_ = x[VMware-8]
-	_ = x[XenHVM-9]
-	_ = x[Bhyve-10]
-	_ = x[Hygon-11]
-	_ = x[SiS-12]
-	_ = x[RDC-13]
-	_ = x[Ampere-14]
-	_ = x[ARM-15]
-	_ = x[Broadcom-16]
-	_ = x[Cavium-17]
-	_ = x[DEC-18]
-	_ = x[Fujitsu-19]
-	_ = x[Infineon-20]
-	_ = x[Motorola-21]
-	_ = x[NVIDIA-22]
-	_ = x[AMCC-23]
-	_ = x[Qualcomm-24]
-	_ = x[Marvell-25]
-	_ = x[lastVendor-26]
-}
-
-const _Vendor_name = "VendorUnknownIntelAMDVIATransmetaNSCKVMMSVMVMwareXenHVMBhyveHygonSiSRDCAmpereARMBroadcomCaviumDECFujitsuInfineonMotorolaNVIDIAAMCCQualcommMarvelllastVendor"
-
-var _Vendor_index = [...]uint8{0, 13, 18, 21, 24, 33, 36, 39, 43, 49, 55, 60, 65, 68, 71, 77, 80, 88, 94, 97, 104, 112, 120, 126, 130, 138, 145, 155}
-
-func (i Vendor) String() string {
-	if i < 0 || i >= Vendor(len(_Vendor_index)-1) {
-		return "Vendor(" + strconv.FormatInt(int64(i), 10) + ")"
-	}
-	return _Vendor_name[_Vendor_index[i]:_Vendor_index[i+1]]
-}
diff --git a/vendor/github.com/klauspost/cpuid/v2/go.mod b/vendor/github.com/klauspost/cpuid/v2/go.mod
deleted file mode 100644
index 2afac8eb28..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/go.mod
+++ /dev/null
@@ -1,3 +0,0 @@
-module github.com/klauspost/cpuid/v2
-
-go 1.13
diff --git a/vendor/github.com/klauspost/cpuid/v2/os_darwin_arm64.go b/vendor/github.com/klauspost/cpuid/v2/os_darwin_arm64.go
deleted file mode 100644
index 8d2cb0368b..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/os_darwin_arm64.go
+++ /dev/null
@@ -1,19 +0,0 @@
-// Copyright (c) 2020 Klaus Post, released under MIT License. See LICENSE file.
-
-package cpuid
-
-import "runtime"
-
-func detectOS(c *CPUInfo) bool {
-	// There are no hw.optional sysctl values for the below features on Mac OS 11.0
-	// to detect their supported state dynamically. Assume the CPU features that
-	// Apple Silicon M1 supports to be available as a minimal set of features
-	// to all Go programs running on darwin/arm64.
-	// TODO: Add more if we know them.
-	c.featureSet.setIf(runtime.GOOS != "ios", AESARM, PMULL, SHA1, SHA2)
-	c.PhysicalCores = runtime.NumCPU()
-	// For now assuming 1 thread per core...
-	c.ThreadsPerCore = 1
-	c.LogicalCores = c.PhysicalCores
-	return true
-}
diff --git a/vendor/github.com/klauspost/cpuid/v2/os_linux_arm64.go b/vendor/github.com/klauspost/cpuid/v2/os_linux_arm64.go
deleted file mode 100644
index ee278b9e4b..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/os_linux_arm64.go
+++ /dev/null
@@ -1,130 +0,0 @@
-// Copyright (c) 2020 Klaus Post, released under MIT License. See LICENSE file.
-
-// Copyright 2018 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file located
-// here https://github.com/golang/sys/blob/master/LICENSE
-
-package cpuid
-
-import (
-	"encoding/binary"
-	"io/ioutil"
-	"runtime"
-)
-
-// HWCAP bits.
-const (
-	hwcap_FP       = 1 << 0
-	hwcap_ASIMD    = 1 << 1
-	hwcap_EVTSTRM  = 1 << 2
-	hwcap_AES      = 1 << 3
-	hwcap_PMULL    = 1 << 4
-	hwcap_SHA1     = 1 << 5
-	hwcap_SHA2     = 1 << 6
-	hwcap_CRC32    = 1 << 7
-	hwcap_ATOMICS  = 1 << 8
-	hwcap_FPHP     = 1 << 9
-	hwcap_ASIMDHP  = 1 << 10
-	hwcap_CPUID    = 1 << 11
-	hwcap_ASIMDRDM = 1 << 12
-	hwcap_JSCVT    = 1 << 13
-	hwcap_FCMA     = 1 << 14
-	hwcap_LRCPC    = 1 << 15
-	hwcap_DCPOP    = 1 << 16
-	hwcap_SHA3     = 1 << 17
-	hwcap_SM3      = 1 << 18
-	hwcap_SM4      = 1 << 19
-	hwcap_ASIMDDP  = 1 << 20
-	hwcap_SHA512   = 1 << 21
-	hwcap_SVE      = 1 << 22
-	hwcap_ASIMDFHM = 1 << 23
-)
-
-func detectOS(c *CPUInfo) bool {
-	// For now assuming no hyperthreading is reasonable.
-	c.LogicalCores = runtime.NumCPU()
-	c.PhysicalCores = c.LogicalCores
-	c.ThreadsPerCore = 1
-	if hwcap == 0 {
-		// We did not get values from the runtime.
-		// Try reading /proc/self/auxv
-
-		// From https://github.com/golang/sys
-		const (
-			_AT_HWCAP  = 16
-			_AT_HWCAP2 = 26
-
-			uintSize = int(32 << (^uint(0) >> 63))
-		)
-
-		buf, err := ioutil.ReadFile("/proc/self/auxv")
-		if err != nil {
-			// e.g. on android /proc/self/auxv is not accessible, so silently
-			// ignore the error and leave Initialized = false. On some
-			// architectures (e.g. arm64) doinit() implements a fallback
-			// readout and will set Initialized = true again.
-			return false
-		}
-		bo := binary.LittleEndian
-		for len(buf) >= 2*(uintSize/8) {
-			var tag, val uint
-			switch uintSize {
-			case 32:
-				tag = uint(bo.Uint32(buf[0:]))
-				val = uint(bo.Uint32(buf[4:]))
-				buf = buf[8:]
-			case 64:
-				tag = uint(bo.Uint64(buf[0:]))
-				val = uint(bo.Uint64(buf[8:]))
-				buf = buf[16:]
-			}
-			switch tag {
-			case _AT_HWCAP:
-				hwcap = val
-			case _AT_HWCAP2:
-				// Not used
-			}
-		}
-		if hwcap == 0 {
-			return false
-		}
-	}
-
-	// HWCap was populated by the runtime from the auxiliary vector.
-	// Use HWCap information since reading aarch64 system registers
-	// is not supported in user space on older linux kernels.
-	c.featureSet.setIf(isSet(hwcap, hwcap_AES), AESARM)
-	c.featureSet.setIf(isSet(hwcap, hwcap_ASIMD), ASIMD)
-	c.featureSet.setIf(isSet(hwcap, hwcap_ASIMDDP), ASIMDDP)
-	c.featureSet.setIf(isSet(hwcap, hwcap_ASIMDHP), ASIMDHP)
-	c.featureSet.setIf(isSet(hwcap, hwcap_ASIMDRDM), ASIMDRDM)
-	c.featureSet.setIf(isSet(hwcap, hwcap_CPUID), ARMCPUID)
-	c.featureSet.setIf(isSet(hwcap, hwcap_CRC32), CRC32)
-	c.featureSet.setIf(isSet(hwcap, hwcap_DCPOP), DCPOP)
-	c.featureSet.setIf(isSet(hwcap, hwcap_EVTSTRM), EVTSTRM)
-	c.featureSet.setIf(isSet(hwcap, hwcap_FCMA), FCMA)
-	c.featureSet.setIf(isSet(hwcap, hwcap_FP), FP)
-	c.featureSet.setIf(isSet(hwcap, hwcap_FPHP), FPHP)
-	c.featureSet.setIf(isSet(hwcap, hwcap_JSCVT), JSCVT)
-	c.featureSet.setIf(isSet(hwcap, hwcap_LRCPC), LRCPC)
-	c.featureSet.setIf(isSet(hwcap, hwcap_PMULL), PMULL)
-	c.featureSet.setIf(isSet(hwcap, hwcap_SHA1), SHA1)
-	c.featureSet.setIf(isSet(hwcap, hwcap_SHA2), SHA2)
-	c.featureSet.setIf(isSet(hwcap, hwcap_SHA3), SHA3)
-	c.featureSet.setIf(isSet(hwcap, hwcap_SHA512), SHA512)
-	c.featureSet.setIf(isSet(hwcap, hwcap_SM3), SM3)
-	c.featureSet.setIf(isSet(hwcap, hwcap_SM4), SM4)
-	c.featureSet.setIf(isSet(hwcap, hwcap_SVE), SVE)
-
-	// The Samsung S9+ kernel reports support for atomics, but not all cores
-	// actually support them, resulting in SIGILL. See issue #28431.
-	// TODO(elias.naur): Only disable the optimization on bad chipsets on android.
-	c.featureSet.setIf(isSet(hwcap, hwcap_ATOMICS) && runtime.GOOS != "android", ATOMICS)
-
-	return true
-}
-
-func isSet(hwc uint, value uint) bool {
-	return hwc&value != 0
-}
diff --git a/vendor/github.com/klauspost/cpuid/v2/os_other_arm64.go b/vendor/github.com/klauspost/cpuid/v2/os_other_arm64.go
deleted file mode 100644
index 1a951e6ca0..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/os_other_arm64.go
+++ /dev/null
@@ -1,17 +0,0 @@
-// Copyright (c) 2020 Klaus Post, released under MIT License. See LICENSE file.
-
-// +build arm64
-// +build !linux
-// +build !darwin
-
-package cpuid
-
-import "runtime"
-
-func detectOS(c *CPUInfo) bool {
-	c.PhysicalCores = runtime.NumCPU()
-	// For now assuming 1 thread per core...
-	c.ThreadsPerCore = 1
-	c.LogicalCores = c.PhysicalCores
-	return false
-}
diff --git a/vendor/github.com/klauspost/cpuid/v2/os_safe_linux_arm64.go b/vendor/github.com/klauspost/cpuid/v2/os_safe_linux_arm64.go
deleted file mode 100644
index 4d0b8b465b..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/os_safe_linux_arm64.go
+++ /dev/null
@@ -1,7 +0,0 @@
-// Copyright (c) 2021 Klaus Post, released under MIT License. See LICENSE file.
-
-//+build nounsafe
-
-package cpuid
-
-var hwcap uint
diff --git a/vendor/github.com/klauspost/cpuid/v2/os_unsafe_linux_arm64.go b/vendor/github.com/klauspost/cpuid/v2/os_unsafe_linux_arm64.go
deleted file mode 100644
index 329800286e..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/os_unsafe_linux_arm64.go
+++ /dev/null
@@ -1,10 +0,0 @@
-// Copyright (c) 2021 Klaus Post, released under MIT License. See LICENSE file.
-
-//+build !nounsafe
-
-package cpuid
-
-import _ "unsafe" // needed for go:linkname
-
-//go:linkname hwcap internal/cpu.HWCap
-var hwcap uint
diff --git a/vendor/github.com/klauspost/cpuid/v2/test-architectures.sh b/vendor/github.com/klauspost/cpuid/v2/test-architectures.sh
deleted file mode 100644
index 471d986d24..0000000000
--- a/vendor/github.com/klauspost/cpuid/v2/test-architectures.sh
+++ /dev/null
@@ -1,15 +0,0 @@
-#!/bin/sh
-
-set -e
-
-go tool dist list | while IFS=/ read os arch; do
-    echo "Checking $os/$arch..."
-    echo " normal"
-    GOARCH=$arch GOOS=$os go build -o /dev/null .
-    echo " noasm"
-    GOARCH=$arch GOOS=$os go build -tags noasm -o /dev/null .
-    echo " appengine"
-    GOARCH=$arch GOOS=$os go build -tags appengine -o /dev/null .
-    echo " noasm,appengine"
-    GOARCH=$arch GOOS=$os go build -tags 'appengine noasm' -o /dev/null .
-done
diff --git a/vendor/github.com/klauspost/pgzip/.gitignore b/vendor/github.com/klauspost/pgzip/.gitignore
deleted file mode 100644
index daf913b1b3..0000000000
--- a/vendor/github.com/klauspost/pgzip/.gitignore
+++ /dev/null
@@ -1,24 +0,0 @@
-# Compiled Object files, Static and Dynamic libs (Shared Objects)
-*.o
-*.a
-*.so
-
-# Folders
-_obj
-_test
-
-# Architecture specific extensions/prefixes
-*.[568vq]
-[568vq].out
-
-*.cgo1.go
-*.cgo2.c
-_cgo_defun.c
-_cgo_gotypes.go
-_cgo_export.*
-
-_testmain.go
-
-*.exe
-*.test
-*.prof
diff --git a/vendor/github.com/klauspost/pgzip/.travis.yml b/vendor/github.com/klauspost/pgzip/.travis.yml
deleted file mode 100644
index acfec4bb09..0000000000
--- a/vendor/github.com/klauspost/pgzip/.travis.yml
+++ /dev/null
@@ -1,24 +0,0 @@
-language: go
-
-os:
-  - linux
-  - osx
-
-go:
-  - 1.13.x
-  - 1.14.x
-  - 1.15.x
-  - master
-
-env:
-  - GO111MODULE=off
-
-script:
-  - diff <(gofmt -d .) <(printf "")
-  - go test -v -cpu=1,2,4 .
-  - go test -v -cpu=2 -race -short .
-
-matrix:
-  allow_failures:
-    - go: 'master'
-  fast_finish: true 
diff --git a/vendor/github.com/klauspost/pgzip/GO_LICENSE b/vendor/github.com/klauspost/pgzip/GO_LICENSE
deleted file mode 100644
index 7448756763..0000000000
--- a/vendor/github.com/klauspost/pgzip/GO_LICENSE
+++ /dev/null
@@ -1,27 +0,0 @@
-Copyright (c) 2012 The Go Authors. All rights reserved.
-
-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.
-   * Neither the name of Google Inc. nor the names of its
-contributors may be used to endorse or promote products derived from
-this software without specific prior written permission.
-
-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.
diff --git a/vendor/github.com/klauspost/pgzip/LICENSE b/vendor/github.com/klauspost/pgzip/LICENSE
deleted file mode 100644
index 3909da4103..0000000000
--- a/vendor/github.com/klauspost/pgzip/LICENSE
+++ /dev/null
@@ -1,21 +0,0 @@
-MIT License
-
-Copyright (c) 2014 Klaus Post
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
diff --git a/vendor/github.com/klauspost/pgzip/README.md b/vendor/github.com/klauspost/pgzip/README.md
deleted file mode 100644
index 171b978fdc..0000000000
--- a/vendor/github.com/klauspost/pgzip/README.md
+++ /dev/null
@@ -1,135 +0,0 @@
-pgzip
-=====
-
-Go parallel gzip compression/decompression. This is a fully gzip compatible drop in replacement for "compress/gzip".
-
-This will split compression into blocks that are compressed in parallel. 
-This can be useful for compressing big amounts of data. The output is a standard gzip file.
-
-The gzip decompression is modified so it decompresses ahead of the current reader. 
-This means that reads will be non-blocking if the decompressor can keep ahead of your code reading from it. 
-CRC calculation also takes place in a separate goroutine.
-
-You should only use this if you are (de)compressing big amounts of data, 
-say **more than 1MB** at the time, otherwise you will not see any benefit, 
-and it will likely be faster to use the internal gzip library 
-or [this package](https://github.com/klauspost/compress).
-
-It is important to note that this library creates and reads *standard gzip files*. 
-You do not have to match the compressor/decompressor to get the described speedups, 
-and the gzip files are fully compatible with other gzip readers/writers.
-
-A golang variant of this is [bgzf](https://godoc.org/github.com/biogo/hts/bgzf), 
-which has the same feature, as well as seeking in the resulting file. 
-The only drawback is a slightly bigger overhead compared to this and pure gzip. 
-See a comparison below.
-
-[![GoDoc][1]][2] [![Build Status][3]][4]
-
-[1]: https://godoc.org/github.com/klauspost/pgzip?status.svg
-[2]: https://godoc.org/github.com/klauspost/pgzip
-[3]: https://travis-ci.org/klauspost/pgzip.svg
-[4]: https://travis-ci.org/klauspost/pgzip
-
-Installation
-====
-```go get github.com/klauspost/pgzip/...```
-
-You might need to get/update the dependencies:
-
-```
-go get -u github.com/klauspost/compress
-```
-
-Usage
-====
-[Godoc Doumentation](https://godoc.org/github.com/klauspost/pgzip)
-
-To use as a replacement for gzip, exchange 
-
-```import "compress/gzip"``` 
-with 
-```import gzip "github.com/klauspost/pgzip"```.
-
-# Changes
-
-* Oct 6, 2016: Fixed an issue if the destination writer returned an error.
-* Oct 6, 2016: Better buffer reuse, should now generate less garbage.
-* Oct 6, 2016: Output does not change based on write sizes.
-* Dec 8, 2015: Decoder now supports the io.WriterTo interface, giving a speedup and less GC pressure.
-* Oct 9, 2015: Reduced allocations by ~35 by using sync.Pool. ~15% overall speedup.
-
-Changes in [github.com/klauspost/compress](https://github.com/klauspost/compress#changelog) are also carried over, so see that for more changes.
-
-## Compression
-The simplest way to use this is to simply do the same as you would when using [compress/gzip](http://golang.org/pkg/compress/gzip). 
-
-To change the block size, use the added (*pgzip.Writer).SetConcurrency(blockSize, blocks int) function. With this you can control the approximate size of your blocks, as well as how many you want to be processing in parallel. Default values for this is SetConcurrency(1MB, runtime.GOMAXPROCS(0)), meaning blocks are split at 1 MB and up to the number of CPU threads blocks can be processing at once before the writer blocks.
-
-
-Example:
-```
-var b bytes.Buffer
-w := gzip.NewWriter(&b)
-w.SetConcurrency(100000, 10)
-w.Write([]byte("hello, world\n"))
-w.Close()
-```
-
-To get any performance gains, you should at least be compressing more than 1 megabyte of data at the time.
-
-You should at least have a block size of 100k and at least a number of blocks that match the number of cores your would like to utilize, but about twice the number of blocks would be the best.
-
-Another side effect of this is, that it is likely to speed up your other code, since writes to the compressor only blocks if the compressor is already compressing the number of blocks you have specified. This also means you don't have worry about buffering input to the compressor.
-
-## Decompression
-
-Decompression works similar to compression. That means that you simply call pgzip the same way as you would call [compress/gzip](http://golang.org/pkg/compress/gzip). 
-
-The only difference is that if you want to specify your own readahead, you have to use `pgzip.NewReaderN(r io.Reader, blockSize, blocks int)` to get a reader with your custom blocksizes. The `blockSize` is the size of each block decoded, and `blocks` is the maximum number of blocks that is decoded ahead.
-
-See [Example on playground](http://play.golang.org/p/uHv1B5NbDh)
-
-Performance
-====
-## Compression
-
-See my blog post in [Benchmarks of Golang Gzip](https://blog.klauspost.com/go-gzipdeflate-benchmarks/).
-
-Compression cost is usually about 0.2% with default settings with a block size of 250k.
-
-Example with GOMAXPROC set to 32 (16 core CPU)
-
-Content is [Matt Mahoneys 10GB corpus](http://mattmahoney.net/dc/10gb.html). Compression level 6.
-
-Compressor  | MB/sec   | speedup | size | size overhead (lower=better)
-------------|----------|---------|------|---------
-[gzip](http://golang.org/pkg/compress/gzip) (golang) | 15.44MB/s (1 thread) | 1.0x | 4781329307 | 0%
-[gzip](http://github.com/klauspost/compress/gzip) (klauspost) | 135.04MB/s (1 thread) | 8.74x | 4894858258 | +2.37%
-[pgzip](https://github.com/klauspost/pgzip) (klauspost) | 1573.23MB/s| 101.9x | 4902285651 | +2.53%
-[bgzf](https://godoc.org/github.com/biogo/hts/bgzf) (biogo) | 361.40MB/s | 23.4x | 4869686090 | +1.85%
-[pargzip](https://godoc.org/github.com/golang/build/pargzip) (builder) | 306.01MB/s | 19.8x | 4786890417 | +0.12%
-
-pgzip also contains a [linear time compression](https://github.com/klauspost/compress#linear-time-compression-huffman-only) mode, that will allow compression at ~250MB per core per second, independent of the content.
-
-See the [complete sheet](https://docs.google.com/spreadsheets/d/1nuNE2nPfuINCZJRMt6wFWhKpToF95I47XjSsc-1rbPQ/edit?usp=sharing) for different content types and compression settings.
-
-## Decompression
-
-The decompression speedup is there because it allows you to do other work while the decompression is taking place.
-
-In the example above, the numbers are as follows on a 4 CPU machine:
-
-Decompressor | Time | Speedup
--------------|------|--------
-[gzip](http://golang.org/pkg/compress/gzip) (golang) | 1m28.85s | 0%
-[pgzip](https://github.com/klauspost/pgzip) (golang) | 43.48s | 104%
-
-But wait, since gzip decompression is inherently singlethreaded (aside from CRC calculation) how can it be more than 100% faster?  Because pgzip due to its design also acts as a buffer. When using unbuffered gzip, you are also waiting for io when you are decompressing. If the gzip decoder can keep up, it will always have data ready for your reader, and you will not be waiting for input to the gzip decompressor to complete.
-
-This is pretty much an optimal situation for pgzip, but it reflects most common usecases for CPU intensive gzip usage.
-
-I haven't included [bgzf](https://godoc.org/github.com/biogo/hts/bgzf) in this comparison, since it only can decompress files created by a compatible encoder, and therefore cannot be considered a generic gzip decompressor. But if you are able to compress your files with a bgzf compatible program, you can expect it to scale beyond 100%.
-
-# License
-This contains large portions of code from the go repository - see GO_LICENSE for more information. The changes are released under MIT License. See LICENSE for more information.
diff --git a/vendor/github.com/klauspost/pgzip/gunzip.go b/vendor/github.com/klauspost/pgzip/gunzip.go
deleted file mode 100644
index d1ae730b25..0000000000
--- a/vendor/github.com/klauspost/pgzip/gunzip.go
+++ /dev/null
@@ -1,584 +0,0 @@
-// Copyright 2009 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// Package pgzip implements reading and writing of gzip format compressed files,
-// as specified in RFC 1952.
-//
-// This is a drop in replacement for "compress/gzip".
-// This will split compression into blocks that are compressed in parallel.
-// This can be useful for compressing big amounts of data.
-// The gzip decompression has not been modified, but remains in the package,
-// so you can use it as a complete replacement for "compress/gzip".
-//
-// See more at https://github.com/klauspost/pgzip
-package pgzip
-
-import (
-	"bufio"
-	"errors"
-	"hash"
-	"hash/crc32"
-	"io"
-	"sync"
-	"time"
-
-	"github.com/klauspost/compress/flate"
-)
-
-const (
-	gzipID1     = 0x1f
-	gzipID2     = 0x8b
-	gzipDeflate = 8
-	flagText    = 1 << 0
-	flagHdrCrc  = 1 << 1
-	flagExtra   = 1 << 2
-	flagName    = 1 << 3
-	flagComment = 1 << 4
-)
-
-func makeReader(r io.Reader) flate.Reader {
-	if rr, ok := r.(flate.Reader); ok {
-		return rr
-	}
-	return bufio.NewReader(r)
-}
-
-var (
-	// ErrChecksum is returned when reading GZIP data that has an invalid checksum.
-	ErrChecksum = errors.New("gzip: invalid checksum")
-	// ErrHeader is returned when reading GZIP data that has an invalid header.
-	ErrHeader = errors.New("gzip: invalid header")
-)
-
-// The gzip file stores a header giving metadata about the compressed file.
-// That header is exposed as the fields of the Writer and Reader structs.
-type Header struct {
-	Comment string    // comment
-	Extra   []byte    // "extra data"
-	ModTime time.Time // modification time
-	Name    string    // file name
-	OS      byte      // operating system type
-}
-
-// A Reader is an io.Reader that can be read to retrieve
-// uncompressed data from a gzip-format compressed file.
-//
-// In general, a gzip file can be a concatenation of gzip files,
-// each with its own header.  Reads from the Reader
-// return the concatenation of the uncompressed data of each.
-// Only the first header is recorded in the Reader fields.
-//
-// Gzip files store a length and checksum of the uncompressed data.
-// The Reader will return a ErrChecksum when Read
-// reaches the end of the uncompressed data if it does not
-// have the expected length or checksum.  Clients should treat data
-// returned by Read as tentative until they receive the io.EOF
-// marking the end of the data.
-type Reader struct {
-	Header
-	r            flate.Reader
-	decompressor io.ReadCloser
-	digest       hash.Hash32
-	size         uint32
-	flg          byte
-	buf          [512]byte
-	err          error
-	closeErr     chan error
-	multistream  bool
-
-	readAhead   chan read
-	roff        int // read offset
-	current     []byte
-	closeReader chan struct{}
-	lastBlock   bool
-	blockSize   int
-	blocks      int
-
-	activeRA bool       // Indication if readahead is active
-	mu       sync.Mutex // Lock for above
-
-	blockPool chan []byte
-}
-
-type read struct {
-	b   []byte
-	err error
-}
-
-// NewReader creates a new Reader reading the given reader.
-// The implementation buffers input and may read more data than necessary from r.
-// It is the caller's responsibility to call Close on the Reader when done.
-func NewReader(r io.Reader) (*Reader, error) {
-	z := new(Reader)
-	z.blocks = defaultBlocks
-	z.blockSize = defaultBlockSize
-	z.r = makeReader(r)
-	z.digest = crc32.NewIEEE()
-	z.multistream = true
-	z.blockPool = make(chan []byte, z.blocks)
-	for i := 0; i < z.blocks; i++ {
-		z.blockPool <- make([]byte, z.blockSize)
-	}
-	if err := z.readHeader(true); err != nil {
-		return nil, err
-	}
-	return z, nil
-}
-
-// NewReaderN creates a new Reader reading the given reader.
-// The implementation buffers input and may read more data than necessary from r.
-// It is the caller's responsibility to call Close on the Reader when done.
-//
-// With this you can control the approximate size of your blocks,
-// as well as how many blocks you want to have prefetched.
-//
-// Default values for this is blockSize = 250000, blocks = 16,
-// meaning up to 16 blocks of maximum 250000 bytes will be
-// prefetched.
-func NewReaderN(r io.Reader, blockSize, blocks int) (*Reader, error) {
-	z := new(Reader)
-	z.blocks = blocks
-	z.blockSize = blockSize
-	z.r = makeReader(r)
-	z.digest = crc32.NewIEEE()
-	z.multistream = true
-
-	// Account for too small values
-	if z.blocks <= 0 {
-		z.blocks = defaultBlocks
-	}
-	if z.blockSize <= 512 {
-		z.blockSize = defaultBlockSize
-	}
-	z.blockPool = make(chan []byte, z.blocks)
-	for i := 0; i < z.blocks; i++ {
-		z.blockPool <- make([]byte, z.blockSize)
-	}
-	if err := z.readHeader(true); err != nil {
-		return nil, err
-	}
-	return z, nil
-}
-
-// Reset discards the Reader z's state and makes it equivalent to the
-// result of its original state from NewReader, but reading from r instead.
-// This permits reusing a Reader rather than allocating a new one.
-func (z *Reader) Reset(r io.Reader) error {
-	z.killReadAhead()
-	z.r = makeReader(r)
-	z.digest = crc32.NewIEEE()
-	z.size = 0
-	z.err = nil
-	z.multistream = true
-
-	// Account for uninitialized values
-	if z.blocks <= 0 {
-		z.blocks = defaultBlocks
-	}
-	if z.blockSize <= 512 {
-		z.blockSize = defaultBlockSize
-	}
-
-	if z.blockPool == nil {
-		z.blockPool = make(chan []byte, z.blocks)
-		for i := 0; i < z.blocks; i++ {
-			z.blockPool <- make([]byte, z.blockSize)
-		}
-	}
-
-	return z.readHeader(true)
-}
-
-// Multistream controls whether the reader supports multistream files.
-//
-// If enabled (the default), the Reader expects the input to be a sequence
-// of individually gzipped data streams, each with its own header and
-// trailer, ending at EOF. The effect is that the concatenation of a sequence
-// of gzipped files is treated as equivalent to the gzip of the concatenation
-// of the sequence. This is standard behavior for gzip readers.
-//
-// Calling Multistream(false) disables this behavior; disabling the behavior
-// can be useful when reading file formats that distinguish individual gzip
-// data streams or mix gzip data streams with other data streams.
-// In this mode, when the Reader reaches the end of the data stream,
-// Read returns io.EOF. If the underlying reader implements io.ByteReader,
-// it will be left positioned just after the gzip stream.
-// To start the next stream, call z.Reset(r) followed by z.Multistream(false).
-// If there is no next stream, z.Reset(r) will return io.EOF.
-func (z *Reader) Multistream(ok bool) {
-	z.multistream = ok
-}
-
-// GZIP (RFC 1952) is little-endian, unlike ZLIB (RFC 1950).
-func get4(p []byte) uint32 {
-	return uint32(p[0]) | uint32(p[1])<<8 | uint32(p[2])<<16 | uint32(p[3])<<24
-}
-
-func (z *Reader) readString() (string, error) {
-	var err error
-	needconv := false
-	for i := 0; ; i++ {
-		if i >= len(z.buf) {
-			return "", ErrHeader
-		}
-		z.buf[i], err = z.r.ReadByte()
-		if err != nil {
-			return "", err
-		}
-		if z.buf[i] > 0x7f {
-			needconv = true
-		}
-		if z.buf[i] == 0 {
-			// GZIP (RFC 1952) specifies that strings are NUL-terminated ISO 8859-1 (Latin-1).
-			if needconv {
-				s := make([]rune, 0, i)
-				for _, v := range z.buf[0:i] {
-					s = append(s, rune(v))
-				}
-				return string(s), nil
-			}
-			return string(z.buf[0:i]), nil
-		}
-	}
-}
-
-func (z *Reader) read2() (uint32, error) {
-	_, err := io.ReadFull(z.r, z.buf[0:2])
-	if err != nil {
-		return 0, err
-	}
-	return uint32(z.buf[0]) | uint32(z.buf[1])<<8, nil
-}
-
-func (z *Reader) readHeader(save bool) error {
-	z.killReadAhead()
-
-	_, err := io.ReadFull(z.r, z.buf[0:10])
-	if err != nil {
-		return err
-	}
-	if z.buf[0] != gzipID1 || z.buf[1] != gzipID2 || z.buf[2] != gzipDeflate {
-		return ErrHeader
-	}
-	z.flg = z.buf[3]
-	if save {
-		z.ModTime = time.Unix(int64(get4(z.buf[4:8])), 0)
-		// z.buf[8] is xfl, ignored
-		z.OS = z.buf[9]
-	}
-	z.digest.Reset()
-	z.digest.Write(z.buf[0:10])
-
-	if z.flg&flagExtra != 0 {
-		n, err := z.read2()
-		if err != nil {
-			return err
-		}
-		data := make([]byte, n)
-		if _, err = io.ReadFull(z.r, data); err != nil {
-			return err
-		}
-		if save {
-			z.Extra = data
-		}
-	}
-
-	var s string
-	if z.flg&flagName != 0 {
-		if s, err = z.readString(); err != nil {
-			return err
-		}
-		if save {
-			z.Name = s
-		}
-	}
-
-	if z.flg&flagComment != 0 {
-		if s, err = z.readString(); err != nil {
-			return err
-		}
-		if save {
-			z.Comment = s
-		}
-	}
-
-	if z.flg&flagHdrCrc != 0 {
-		n, err := z.read2()
-		if err != nil {
-			return err
-		}
-		sum := z.digest.Sum32() & 0xFFFF
-		if n != sum {
-			return ErrHeader
-		}
-	}
-
-	z.digest.Reset()
-	z.decompressor = flate.NewReader(z.r)
-	z.doReadAhead()
-	return nil
-}
-
-func (z *Reader) killReadAhead() error {
-	z.mu.Lock()
-	defer z.mu.Unlock()
-	if z.activeRA {
-		if z.closeReader != nil {
-			close(z.closeReader)
-		}
-
-		// Wait for decompressor to be closed and return error, if any.
-		e, ok := <-z.closeErr
-		z.activeRA = false
-
-		for blk := range z.readAhead {
-			if blk.b != nil {
-				z.blockPool <- blk.b
-			}
-		}
-		if cap(z.current) > 0 {
-			z.blockPool <- z.current
-			z.current = nil
-		}
-		if !ok {
-			// Channel is closed, so if there was any error it has already been returned.
-			return nil
-		}
-		return e
-	}
-	return nil
-}
-
-// Starts readahead.
-// Will return on error (including io.EOF)
-// or when z.closeReader is closed.
-func (z *Reader) doReadAhead() {
-	z.mu.Lock()
-	defer z.mu.Unlock()
-	z.activeRA = true
-
-	if z.blocks <= 0 {
-		z.blocks = defaultBlocks
-	}
-	if z.blockSize <= 512 {
-		z.blockSize = defaultBlockSize
-	}
-	ra := make(chan read, z.blocks)
-	z.readAhead = ra
-	closeReader := make(chan struct{}, 0)
-	z.closeReader = closeReader
-	z.lastBlock = false
-	closeErr := make(chan error, 1)
-	z.closeErr = closeErr
-	z.size = 0
-	z.roff = 0
-	z.current = nil
-	decomp := z.decompressor
-
-	go func() {
-		defer func() {
-			closeErr <- decomp.Close()
-			close(closeErr)
-			close(ra)
-		}()
-
-		// We hold a local reference to digest, since
-		// it way be changed by reset.
-		digest := z.digest
-		var wg sync.WaitGroup
-		for {
-			var buf []byte
-			select {
-			case buf = <-z.blockPool:
-			case <-closeReader:
-				return
-			}
-			buf = buf[0:z.blockSize]
-			// Try to fill the buffer
-			n, err := io.ReadFull(decomp, buf)
-			if err == io.ErrUnexpectedEOF {
-				if n > 0 {
-					err = nil
-				} else {
-					// If we got zero bytes, we need to establish if
-					// we reached end of stream or truncated stream.
-					_, err = decomp.Read([]byte{})
-					if err == io.EOF {
-						err = nil
-					}
-				}
-			}
-			if n < len(buf) {
-				buf = buf[0:n]
-			}
-			wg.Wait()
-			wg.Add(1)
-			go func() {
-				digest.Write(buf)
-				wg.Done()
-			}()
-			z.size += uint32(n)
-
-			// If we return any error, out digest must be ready
-			if err != nil {
-				wg.Wait()
-			}
-			select {
-			case z.readAhead <- read{b: buf, err: err}:
-			case <-closeReader:
-				// Sent on close, we don't care about the next results
-				z.blockPool <- buf
-				return
-			}
-			if err != nil {
-				return
-			}
-		}
-	}()
-}
-
-func (z *Reader) Read(p []byte) (n int, err error) {
-	if z.err != nil {
-		return 0, z.err
-	}
-	if len(p) == 0 {
-		return 0, nil
-	}
-
-	for {
-		if len(z.current) == 0 && !z.lastBlock {
-			read := <-z.readAhead
-
-			if read.err != nil {
-				// If not nil, the reader will have exited
-				z.closeReader = nil
-
-				if read.err != io.EOF {
-					z.err = read.err
-					return
-				}
-				if read.err == io.EOF {
-					z.lastBlock = true
-					err = nil
-				}
-			}
-			z.current = read.b
-			z.roff = 0
-		}
-		avail := z.current[z.roff:]
-		if len(p) >= len(avail) {
-			// If len(p) >= len(current), return all content of current
-			n = copy(p, avail)
-			z.blockPool <- z.current
-			z.current = nil
-			if z.lastBlock {
-				err = io.EOF
-				break
-			}
-		} else {
-			// We copy as much as there is space for
-			n = copy(p, avail)
-			z.roff += n
-		}
-		return
-	}
-
-	// Finished file; check checksum + size.
-	if _, err := io.ReadFull(z.r, z.buf[0:8]); err != nil {
-		z.err = err
-		return 0, err
-	}
-	crc32, isize := get4(z.buf[0:4]), get4(z.buf[4:8])
-	sum := z.digest.Sum32()
-	if sum != crc32 || isize != z.size {
-		z.err = ErrChecksum
-		return 0, z.err
-	}
-
-	// File is ok; should we attempt reading one more?
-	if !z.multistream {
-		return 0, io.EOF
-	}
-
-	// Is there another?
-	if err = z.readHeader(false); err != nil {
-		z.err = err
-		return
-	}
-
-	// Yes.  Reset and read from it.
-	return z.Read(p)
-}
-
-func (z *Reader) WriteTo(w io.Writer) (n int64, err error) {
-	total := int64(0)
-	for {
-		if z.err != nil {
-			return total, z.err
-		}
-		// We write both to output and digest.
-		for {
-			// Read from input
-			read := <-z.readAhead
-			if read.err != nil {
-				// If not nil, the reader will have exited
-				z.closeReader = nil
-
-				if read.err != io.EOF {
-					z.err = read.err
-					return total, z.err
-				}
-				if read.err == io.EOF {
-					z.lastBlock = true
-					err = nil
-				}
-			}
-			// Write what we got
-			n, err := w.Write(read.b)
-			if n != len(read.b) {
-				return total, io.ErrShortWrite
-			}
-			total += int64(n)
-			if err != nil {
-				return total, err
-			}
-			// Put block back
-			z.blockPool <- read.b
-			if z.lastBlock {
-				break
-			}
-		}
-
-		// Finished file; check checksum + size.
-		if _, err := io.ReadFull(z.r, z.buf[0:8]); err != nil {
-			z.err = err
-			return total, err
-		}
-		crc32, isize := get4(z.buf[0:4]), get4(z.buf[4:8])
-		sum := z.digest.Sum32()
-		if sum != crc32 || isize != z.size {
-			z.err = ErrChecksum
-			return total, z.err
-		}
-		// File is ok; should we attempt reading one more?
-		if !z.multistream {
-			return total, nil
-		}
-
-		// Is there another?
-		err = z.readHeader(false)
-		if err == io.EOF {
-			return total, nil
-		}
-		if err != nil {
-			z.err = err
-			return total, err
-		}
-	}
-}
-
-// Close closes the Reader. It does not close the underlying io.Reader.
-func (z *Reader) Close() error {
-	return z.killReadAhead()
-}
diff --git a/vendor/github.com/klauspost/pgzip/gzip.go b/vendor/github.com/klauspost/pgzip/gzip.go
deleted file mode 100644
index 257c4d299f..0000000000
--- a/vendor/github.com/klauspost/pgzip/gzip.go
+++ /dev/null
@@ -1,519 +0,0 @@
-// Copyright 2010 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-package pgzip
-
-import (
-	"bytes"
-	"errors"
-	"fmt"
-	"hash"
-	"hash/crc32"
-	"io"
-	"runtime"
-	"sync"
-	"time"
-
-	"github.com/klauspost/compress/flate"
-)
-
-const (
-	defaultBlockSize = 1 << 20
-	tailSize         = 16384
-	defaultBlocks    = 4
-)
-
-// These constants are copied from the flate package, so that code that imports
-// "compress/gzip" does not also have to import "compress/flate".
-const (
-	NoCompression       = flate.NoCompression
-	BestSpeed           = flate.BestSpeed
-	BestCompression     = flate.BestCompression
-	DefaultCompression  = flate.DefaultCompression
-	ConstantCompression = flate.ConstantCompression
-	HuffmanOnly         = flate.HuffmanOnly
-)
-
-// A Writer is an io.WriteCloser.
-// Writes to a Writer are compressed and written to w.
-type Writer struct {
-	Header
-	w             io.Writer
-	level         int
-	wroteHeader   bool
-	blockSize     int
-	blocks        int
-	currentBuffer []byte
-	prevTail      []byte
-	digest        hash.Hash32
-	size          int
-	closed        bool
-	buf           [10]byte
-	errMu         sync.RWMutex
-	err           error
-	pushedErr     chan struct{}
-	results       chan result
-	dictFlatePool sync.Pool
-	dstPool       sync.Pool
-	wg            sync.WaitGroup
-}
-
-type result struct {
-	result        chan []byte
-	notifyWritten chan struct{}
-}
-
-// Use SetConcurrency to finetune the concurrency level if needed.
-//
-// With this you can control the approximate size of your blocks,
-// as well as how many you want to be processing in parallel.
-//
-// Default values for this is SetConcurrency(defaultBlockSize, runtime.GOMAXPROCS(0)),
-// meaning blocks are split at 1 MB and up to the number of CPU threads
-// can be processing at once before the writer blocks.
-func (z *Writer) SetConcurrency(blockSize, blocks int) error {
-	if blockSize <= tailSize {
-		return fmt.Errorf("gzip: block size cannot be less than or equal to %d", tailSize)
-	}
-	if blocks <= 0 {
-		return errors.New("gzip: blocks cannot be zero or less")
-	}
-	if blockSize == z.blockSize && blocks == z.blocks {
-		return nil
-	}
-	z.blockSize = blockSize
-	z.results = make(chan result, blocks)
-	z.blocks = blocks
-	z.dstPool.New = func() interface{} { return make([]byte, 0, blockSize+(blockSize)>>4) }
-	return nil
-}
-
-// NewWriter returns a new Writer.
-// Writes to the returned writer are compressed and written to w.
-//
-// It is the caller's responsibility to call Close on the WriteCloser when done.
-// Writes may be buffered and not flushed until Close.
-//
-// Callers that wish to set the fields in Writer.Header must do so before
-// the first call to Write or Close. The Comment and Name header fields are
-// UTF-8 strings in Go, but the underlying format requires NUL-terminated ISO
-// 8859-1 (Latin-1). NUL or non-Latin-1 runes in those strings will lead to an
-// error on Write.
-func NewWriter(w io.Writer) *Writer {
-	z, _ := NewWriterLevel(w, DefaultCompression)
-	return z
-}
-
-// NewWriterLevel is like NewWriter but specifies the compression level instead
-// of assuming DefaultCompression.
-//
-// The compression level can be DefaultCompression, NoCompression, or any
-// integer value between BestSpeed and BestCompression inclusive. The error
-// returned will be nil if the level is valid.
-func NewWriterLevel(w io.Writer, level int) (*Writer, error) {
-	if level < ConstantCompression || level > BestCompression {
-		return nil, fmt.Errorf("gzip: invalid compression level: %d", level)
-	}
-	z := new(Writer)
-	z.SetConcurrency(defaultBlockSize, runtime.GOMAXPROCS(0))
-	z.init(w, level)
-	return z, nil
-}
-
-// This function must be used by goroutines to set an
-// error condition, since z.err access is restricted
-// to the callers goruotine.
-func (z *Writer) pushError(err error) {
-	z.errMu.Lock()
-	if z.err != nil {
-		z.errMu.Unlock()
-		return
-	}
-	z.err = err
-	close(z.pushedErr)
-	z.errMu.Unlock()
-}
-
-func (z *Writer) init(w io.Writer, level int) {
-	z.wg.Wait()
-	digest := z.digest
-	if digest != nil {
-		digest.Reset()
-	} else {
-		digest = crc32.NewIEEE()
-	}
-	z.Header = Header{OS: 255}
-	z.w = w
-	z.level = level
-	z.digest = digest
-	z.pushedErr = make(chan struct{}, 0)
-	z.results = make(chan result, z.blocks)
-	z.err = nil
-	z.closed = false
-	z.Comment = ""
-	z.Extra = nil
-	z.ModTime = time.Time{}
-	z.wroteHeader = false
-	z.currentBuffer = nil
-	z.buf = [10]byte{}
-	z.prevTail = nil
-	z.size = 0
-	if z.dictFlatePool.New == nil {
-		z.dictFlatePool.New = func() interface{} {
-			f, _ := flate.NewWriterDict(w, level, nil)
-			return f
-		}
-	}
-}
-
-// Reset discards the Writer z's state and makes it equivalent to the
-// result of its original state from NewWriter or NewWriterLevel, but
-// writing to w instead. This permits reusing a Writer rather than
-// allocating a new one.
-func (z *Writer) Reset(w io.Writer) {
-	if z.results != nil && !z.closed {
-		close(z.results)
-	}
-	z.SetConcurrency(defaultBlockSize, runtime.GOMAXPROCS(0))
-	z.init(w, z.level)
-}
-
-// GZIP (RFC 1952) is little-endian, unlike ZLIB (RFC 1950).
-func put2(p []byte, v uint16) {
-	p[0] = uint8(v >> 0)
-	p[1] = uint8(v >> 8)
-}
-
-func put4(p []byte, v uint32) {
-	p[0] = uint8(v >> 0)
-	p[1] = uint8(v >> 8)
-	p[2] = uint8(v >> 16)
-	p[3] = uint8(v >> 24)
-}
-
-// writeBytes writes a length-prefixed byte slice to z.w.
-func (z *Writer) writeBytes(b []byte) error {
-	if len(b) > 0xffff {
-		return errors.New("gzip.Write: Extra data is too large")
-	}
-	put2(z.buf[0:2], uint16(len(b)))
-	_, err := z.w.Write(z.buf[0:2])
-	if err != nil {
-		return err
-	}
-	_, err = z.w.Write(b)
-	return err
-}
-
-// writeString writes a UTF-8 string s in GZIP's format to z.w.
-// GZIP (RFC 1952) specifies that strings are NUL-terminated ISO 8859-1 (Latin-1).
-func (z *Writer) writeString(s string) (err error) {
-	// GZIP stores Latin-1 strings; error if non-Latin-1; convert if non-ASCII.
-	needconv := false
-	for _, v := range s {
-		if v == 0 || v > 0xff {
-			return errors.New("gzip.Write: non-Latin-1 header string")
-		}
-		if v > 0x7f {
-			needconv = true
-		}
-	}
-	if needconv {
-		b := make([]byte, 0, len(s))
-		for _, v := range s {
-			b = append(b, byte(v))
-		}
-		_, err = z.w.Write(b)
-	} else {
-		_, err = io.WriteString(z.w, s)
-	}
-	if err != nil {
-		return err
-	}
-	// GZIP strings are NUL-terminated.
-	z.buf[0] = 0
-	_, err = z.w.Write(z.buf[0:1])
-	return err
-}
-
-// compressCurrent will compress the data currently buffered
-// This should only be called from the main writer/flush/closer
-func (z *Writer) compressCurrent(flush bool) {
-	c := z.currentBuffer
-	if len(c) > z.blockSize {
-		// This can never happen through the public interface.
-		panic("len(z.currentBuffer) > z.blockSize (most likely due to concurrent Write race)")
-	}
-
-	r := result{}
-	r.result = make(chan []byte, 1)
-	r.notifyWritten = make(chan struct{}, 0)
-	// Reserve a result slot
-	select {
-	case z.results <- r:
-	case <-z.pushedErr:
-		return
-	}
-
-	z.wg.Add(1)
-	tail := z.prevTail
-	if len(c) > tailSize {
-		buf := z.dstPool.Get().([]byte) // Put in .compressBlock
-		// Copy tail from current buffer before handing the buffer over to the
-		// compressBlock goroutine.
-		buf = append(buf[:0], c[len(c)-tailSize:]...)
-		z.prevTail = buf
-	} else {
-		z.prevTail = nil
-	}
-	go z.compressBlock(c, tail, r, z.closed)
-
-	z.currentBuffer = z.dstPool.Get().([]byte) // Put in .compressBlock
-	z.currentBuffer = z.currentBuffer[:0]
-
-	// Wait if flushing
-	if flush {
-		<-r.notifyWritten
-	}
-}
-
-// Returns an error if it has been set.
-// Cannot be used by functions that are from internal goroutines.
-func (z *Writer) checkError() error {
-	z.errMu.RLock()
-	err := z.err
-	z.errMu.RUnlock()
-	return err
-}
-
-// Write writes a compressed form of p to the underlying io.Writer. The
-// compressed bytes are not necessarily flushed to output until
-// the Writer is closed or Flush() is called.
-//
-// The function will return quickly, if there are unused buffers.
-// The sent slice (p) is copied, and the caller is free to re-use the buffer
-// when the function returns.
-//
-// Errors that occur during compression will be reported later, and a nil error
-// does not signify that the compression succeeded (since it is most likely still running)
-// That means that the call that returns an error may not be the call that caused it.
-// Only Flush and Close functions are guaranteed to return any errors up to that point.
-func (z *Writer) Write(p []byte) (int, error) {
-	if err := z.checkError(); err != nil {
-		return 0, err
-	}
-	// Write the GZIP header lazily.
-	if !z.wroteHeader {
-		z.wroteHeader = true
-		z.buf[0] = gzipID1
-		z.buf[1] = gzipID2
-		z.buf[2] = gzipDeflate
-		z.buf[3] = 0
-		if z.Extra != nil {
-			z.buf[3] |= 0x04
-		}
-		if z.Name != "" {
-			z.buf[3] |= 0x08
-		}
-		if z.Comment != "" {
-			z.buf[3] |= 0x10
-		}
-		put4(z.buf[4:8], uint32(z.ModTime.Unix()))
-		if z.level == BestCompression {
-			z.buf[8] = 2
-		} else if z.level == BestSpeed {
-			z.buf[8] = 4
-		} else {
-			z.buf[8] = 0
-		}
-		z.buf[9] = z.OS
-		var n int
-		var err error
-		n, err = z.w.Write(z.buf[0:10])
-		if err != nil {
-			z.pushError(err)
-			return n, err
-		}
-		if z.Extra != nil {
-			err = z.writeBytes(z.Extra)
-			if err != nil {
-				z.pushError(err)
-				return n, err
-			}
-		}
-		if z.Name != "" {
-			err = z.writeString(z.Name)
-			if err != nil {
-				z.pushError(err)
-				return n, err
-			}
-		}
-		if z.Comment != "" {
-			err = z.writeString(z.Comment)
-			if err != nil {
-				z.pushError(err)
-				return n, err
-			}
-		}
-		// Start receiving data from compressors
-		go func() {
-			listen := z.results
-			var failed bool
-			for {
-				r, ok := <-listen
-				// If closed, we are finished.
-				if !ok {
-					return
-				}
-				if failed {
-					close(r.notifyWritten)
-					continue
-				}
-				buf := <-r.result
-				n, err := z.w.Write(buf)
-				if err != nil {
-					z.pushError(err)
-					close(r.notifyWritten)
-					failed = true
-					continue
-				}
-				if n != len(buf) {
-					z.pushError(fmt.Errorf("gzip: short write %d should be %d", n, len(buf)))
-					failed = true
-					close(r.notifyWritten)
-					continue
-				}
-				z.dstPool.Put(buf)
-				close(r.notifyWritten)
-			}
-		}()
-		z.currentBuffer = z.dstPool.Get().([]byte)
-		z.currentBuffer = z.currentBuffer[:0]
-	}
-	q := p
-	for len(q) > 0 {
-		length := len(q)
-		if length+len(z.currentBuffer) > z.blockSize {
-			length = z.blockSize - len(z.currentBuffer)
-		}
-		z.digest.Write(q[:length])
-		z.currentBuffer = append(z.currentBuffer, q[:length]...)
-		if len(z.currentBuffer) > z.blockSize {
-			panic("z.currentBuffer too large (most likely due to concurrent Write race)")
-		}
-		if len(z.currentBuffer) == z.blockSize {
-			z.compressCurrent(false)
-			if err := z.checkError(); err != nil {
-				return len(p) - len(q), err
-			}
-		}
-		z.size += length
-		q = q[length:]
-	}
-	return len(p), z.checkError()
-}
-
-// Step 1: compresses buffer to buffer
-// Step 2: send writer to channel
-// Step 3: Close result channel to indicate we are done
-func (z *Writer) compressBlock(p, prevTail []byte, r result, closed bool) {
-	defer func() {
-		close(r.result)
-		z.wg.Done()
-	}()
-	buf := z.dstPool.Get().([]byte) // Corresponding Put in .Write's result writer
-	dest := bytes.NewBuffer(buf[:0])
-
-	compressor := z.dictFlatePool.Get().(*flate.Writer) // Put below
-	compressor.ResetDict(dest, prevTail)
-	compressor.Write(p)
-	z.dstPool.Put(p) // Corresponding Get in .Write and .compressCurrent
-
-	err := compressor.Flush()
-	if err != nil {
-		z.pushError(err)
-		return
-	}
-	if closed {
-		err = compressor.Close()
-		if err != nil {
-			z.pushError(err)
-			return
-		}
-	}
-	z.dictFlatePool.Put(compressor) // Get above
-
-	if prevTail != nil {
-		z.dstPool.Put(prevTail) // Get in .compressCurrent
-	}
-
-	// Read back buffer
-	buf = dest.Bytes()
-	r.result <- buf
-}
-
-// Flush flushes any pending compressed data to the underlying writer.
-//
-// It is useful mainly in compressed network protocols, to ensure that
-// a remote reader has enough data to reconstruct a packet. Flush does
-// not return until the data has been written. If the underlying
-// writer returns an error, Flush returns that error.
-//
-// In the terminology of the zlib library, Flush is equivalent to Z_SYNC_FLUSH.
-func (z *Writer) Flush() error {
-	if err := z.checkError(); err != nil {
-		return err
-	}
-	if z.closed {
-		return nil
-	}
-	if !z.wroteHeader {
-		_, err := z.Write(nil)
-		if err != nil {
-			return err
-		}
-	}
-	// We send current block to compression
-	z.compressCurrent(true)
-
-	return z.checkError()
-}
-
-// UncompressedSize will return the number of bytes written.
-// pgzip only, not a function in the official gzip package.
-func (z *Writer) UncompressedSize() int {
-	return z.size
-}
-
-// Close closes the Writer, flushing any unwritten data to the underlying
-// io.Writer, but does not close the underlying io.Writer.
-func (z *Writer) Close() error {
-	if err := z.checkError(); err != nil {
-		return err
-	}
-	if z.closed {
-		return nil
-	}
-
-	z.closed = true
-	if !z.wroteHeader {
-		z.Write(nil)
-		if err := z.checkError(); err != nil {
-			return err
-		}
-	}
-	z.compressCurrent(true)
-	if err := z.checkError(); err != nil {
-		return err
-	}
-	close(z.results)
-	put4(z.buf[0:4], z.digest.Sum32())
-	put4(z.buf[4:8], uint32(z.size))
-	_, err := z.w.Write(z.buf[0:8])
-	if err != nil {
-		z.pushError(err)
-		return err
-	}
-	return nil
-}