Add AVX implementation generated by clang.

3 files changed, 344 insertions, 1 deletions

diff --git a/src/libcryptobox/siphash/avx.S b/src/libcryptobox/siphash/avx.S
new file mode 100644
index 000000000..72e18c7c1
--- /dev/null
+++ b/src/libcryptobox/siphash/avx.S
@@ -0,0 +1,332 @@
+#include "../macro.S"
+#include "constants.S"
+
+/*
+ * Generated by clang-3.7 with -mavx -Ofast from reference implementation
+ */
+
+SECTION_TEXT
+
+GLOBAL_HIDDEN_FN siphash_avx
+siphash_avx_local:
+	.cfi_startproc
+	pushq	%r15
+.Ltmp0:
+	.cfi_def_cfa_offset 16
+	pushq	%r14
+.Ltmp1:
+	.cfi_def_cfa_offset 24
+	pushq	%r12
+.Ltmp2:
+	.cfi_def_cfa_offset 32
+	pushq	%rbx
+.Ltmp3:
+	.cfi_def_cfa_offset 40
+.Ltmp4:
+	.cfi_offset %rbx, -40
+.Ltmp5:
+	.cfi_offset %r12, -32
+.Ltmp6:
+	.cfi_offset %r14, -24
+.Ltmp7:
+	.cfi_offset %r15, -16
+	movq	(%rdi), %rcx
+	movq	8(%rdi), %rbx
+	movq	%rdx, %r9
+	shlq	$56, %r9
+	movq	%r9, -8(%rsp)
+	movabsq	$8317987319222330741, %r12 # imm = 0x736F6D6570736575
+	xorq	%rcx, %r12
+	movabsq	$7237128888997146477, %rax # imm = 0x646F72616E646F6D
+	xorq	%rbx, %rax
+	movabsq	$7816392313619706465, %r8 # imm = 0x6C7967656E657261
+	xorq	%rcx, %r8
+	movabsq	$8387220255154660723, %rdi # imm = 0x7465646279746573
+	xorq	%rbx, %rdi
+	cmpq	$8, %rdx
+	jb	.LBB0_4
+# BB#1:                                 # %.lr.ph104
+	leaq	-8(%rdx), %r10
+	movq	%r10, %r11
+	andq	$-8, %r11
+	leaq	8(%r11), %r14
+	movq	%rsi, %rbx
+	.align	16, 0x90
+.LBB0_2:                                # =>This Inner Loop Header: Depth=1
+	movq	(%rbx), %r15
+	addq	$8, %rbx
+	xorq	%r15, %rdi
+	addq	%rax, %r12
+	addq	%rdi, %r8
+	#APP
+	shldq	$13, %rax, %rax
+	#NO_APP
+	#APP
+	shldq	$16, %rdi, %rdi
+	#NO_APP
+	xorq	%r12, %rax
+	xorq	%r8, %rdi
+	#APP
+	shldq	$32, %r12, %r12
+	#NO_APP
+	addq	%rax, %r8
+	addq	%rdi, %r12
+	#APP
+	shldq	$17, %rax, %rax
+	#NO_APP
+	#APP
+	shldq	$21, %rdi, %rdi
+	#NO_APP
+	xorq	%r8, %rax
+	xorq	%r12, %rdi
+	#APP
+	shldq	$32, %r8, %r8
+	#NO_APP
+	addq	%rax, %r12
+	addq	%rdi, %r8
+	#APP
+	shldq	$13, %rax, %rax
+	#NO_APP
+	#APP
+	shldq	$16, %rdi, %rdi
+	#NO_APP
+	xorq	%r12, %rax
+	xorq	%r8, %rdi
+	#APP
+	shldq	$32, %r12, %r12
+	#NO_APP
+	addq	%rax, %r8
+	addq	%rdi, %r12
+	#APP
+	shldq	$17, %rax, %rax
+	#NO_APP
+	#APP
+	shldq	$21, %rdi, %rdi
+	#NO_APP
+	xorq	%r8, %rax
+	xorq	%r12, %rdi
+	#APP
+	shldq	$32, %r8, %r8
+	#NO_APP
+	xorq	%r15, %r12
+	addq	$-8, %rdx
+	cmpq	$7, %rdx
+	ja	.LBB0_2
+# BB#3:                                 # %..preheader_crit_edge
+	subq	%r11, %r10
+	addq	%r14, %rsi
+	movq	%r10, %rdx
+.LBB0_4:                                # %.preheader
+	testq	%rdx, %rdx
+	je	.LBB0_13
+# BB#5:                                 # %overflow.checked
+	xorl	%ebx, %ebx
+	movq	%rdx, %r9
+	andq	$-128, %r9
+	je	.LBB0_9
+# BB#6:                                 # %vector.body.preheader
+	leaq	88(%rsp), %rbx
+	leaq	96(%rsi), %rcx
+	movq	%rdx, %r10
+	andq	$-128, %r10
+	.align	16, 0x90
+.LBB0_7:                                # %vector.body
+                                        # =>This Inner Loop Header: Depth=1
+	vmovups	-96(%rcx), %ymm0
+	vmovups	-64(%rcx), %ymm1
+	vmovups	-32(%rcx), %ymm2
+	vmovups	(%rcx), %ymm3
+	vmovups	%ymm0, -96(%rbx)
+	vmovups	%ymm1, -64(%rbx)
+	vmovups	%ymm2, -32(%rbx)
+	vmovups	%ymm3, (%rbx)
+	subq	$-128, %rbx
+	subq	$-128, %rcx
+	addq	$-128, %r10
+	jne	.LBB0_7
+# BB#8:
+	movq	%r9, %rbx
+.LBB0_9:                                # %middle.block
+	subq	%rbx, %rdx
+	je	.LBB0_12
+# BB#10:                                # %.lr.ph.preheader
+	leaq	-8(%rsp,%rbx), %rcx
+	addq	%rbx, %rsi
+	.align	16, 0x90
+.LBB0_11:                               # %.lr.ph
+                                        # =>This Inner Loop Header: Depth=1
+	movb	(%rsi), %bl
+	movb	%bl, (%rcx)
+	incq	%rcx
+	incq	%rsi
+	decq	%rdx
+	jne	.LBB0_11
+.LBB0_12:                               # %._crit_edge
+	movq	-8(%rsp), %r9
+.LBB0_13:
+	xorq	%r9, %rdi
+	addq	%rax, %r12
+	addq	%rdi, %r8
+	#APP
+	shldq	$13, %rax, %rax
+	#NO_APP
+	#APP
+	shldq	$16, %rdi, %rdi
+	#NO_APP
+	xorq	%r12, %rax
+	xorq	%r8, %rdi
+	#APP
+	shldq	$32, %r12, %r12
+	#NO_APP
+	addq	%rax, %r8
+	addq	%rdi, %r12
+	#APP
+	shldq	$17, %rax, %rax
+	#NO_APP
+	#APP
+	shldq	$21, %rdi, %rdi
+	#NO_APP
+	xorq	%r8, %rax
+	xorq	%r12, %rdi
+	#APP
+	shldq	$32, %r8, %r8
+	#NO_APP
+	addq	%rax, %r12
+	addq	%rdi, %r8
+	#APP
+	shldq	$13, %rax, %rax
+	#NO_APP
+	#APP
+	shldq	$16, %rdi, %rdi
+	#NO_APP
+	xorq	%r12, %rax
+	xorq	%r8, %rdi
+	#APP
+	shldq	$32, %r12, %r12
+	#NO_APP
+	addq	%rax, %r8
+	addq	%rdi, %r12
+	#APP
+	shldq	$17, %rax, %rax
+	#NO_APP
+	#APP
+	shldq	$21, %rdi, %rdi
+	#NO_APP
+	xorq	%r8, %rax
+	xorq	%r12, %rdi
+	#APP
+	shldq	$32, %r8, %r8
+	#NO_APP
+	xorq	%r9, %r12
+	xorq	$255, %r8
+	addq	%rax, %r12
+	addq	%rdi, %r8
+	#APP
+	shldq	$13, %rax, %rax
+	#NO_APP
+	#APP
+	shldq	$16, %rdi, %rdi
+	#NO_APP
+	xorq	%r12, %rax
+	xorq	%r8, %rdi
+	#APP
+	shldq	$32, %r12, %r12
+	#NO_APP
+	addq	%rax, %r8
+	addq	%rdi, %r12
+	#APP
+	shldq	$17, %rax, %rax
+	#NO_APP
+	#APP
+	shldq	$21, %rdi, %rdi
+	#NO_APP
+	xorq	%r8, %rax
+	xorq	%r12, %rdi
+	#APP
+	shldq	$32, %r8, %r8
+	#NO_APP
+	addq	%rax, %r12
+	addq	%rdi, %r8
+	#APP
+	shldq	$13, %rax, %rax
+	#NO_APP
+	#APP
+	shldq	$16, %rdi, %rdi
+	#NO_APP
+	xorq	%r12, %rax
+	xorq	%r8, %rdi
+	#APP
+	shldq	$32, %r12, %r12
+	#NO_APP
+	addq	%rax, %r8
+	addq	%rdi, %r12
+	#APP
+	shldq	$17, %rax, %rax
+	#NO_APP
+	#APP
+	shldq	$21, %rdi, %rdi
+	#NO_APP
+	xorq	%r8, %rax
+	xorq	%r12, %rdi
+	#APP
+	shldq	$32, %r8, %r8
+	#NO_APP
+	addq	%rax, %r12
+	addq	%rdi, %r8
+	#APP
+	shldq	$13, %rax, %rax
+	#NO_APP
+	#APP
+	shldq	$16, %rdi, %rdi
+	#NO_APP
+	xorq	%r12, %rax
+	xorq	%r8, %rdi
+	#APP
+	shldq	$32, %r12, %r12
+	#NO_APP
+	addq	%rax, %r8
+	addq	%rdi, %r12
+	#APP
+	shldq	$17, %rax, %rax
+	#NO_APP
+	#APP
+	shldq	$21, %rdi, %rdi
+	#NO_APP
+	xorq	%r8, %rax
+	xorq	%r12, %rdi
+	#APP
+	shldq	$32, %r8, %r8
+	#NO_APP
+	addq	%rax, %r12
+	addq	%rdi, %r8
+	#APP
+	shldq	$13, %rax, %rax
+	#NO_APP
+	#APP
+	shldq	$16, %rdi, %rdi
+	#NO_APP
+	xorq	%r12, %rax
+	xorq	%r8, %rdi
+	addq	%rax, %r8
+	#APP
+	shldq	$17, %rax, %rax
+	#NO_APP
+	#APP
+	shldq	$21, %rdi, %rdi
+	#NO_APP
+	xorq	%r8, %rax
+	#APP
+	shldq	$32, %r8, %r8
+	#NO_APP
+	xorq	%rdi, %rax
+	xorq	%r8, %rax
+	popq	%rbx
+	popq	%r12
+	popq	%r14
+	popq	%r15
+	vzeroupper
+	retq
+.Lfunc_end0:
+	.size	siphash_avx_local, .Lfunc_end0-siphash_avx_local
+	.cfi_endproc
+FN_END siphash_avx
diff --git a/src/libcryptobox/siphash/siphash.c b/src/libcryptobox/siphash/siphash.c
index f42456b7f..498609fc5 100644
--- a/src/libcryptobox/siphash/siphash.c
+++ b/src/libcryptobox/siphash/siphash.c
@@ -49,10 +49,17 @@ SIPHASH_DECLARE(ref)
 SIPHASH_DECLARE(sse41)
 #define SIPHASH_SSE41 SIPHASH_IMPL(CPUID_SSE41, "sse41", sse41)
 #endif
+#if defined(HAVE_AVX)
+SIPHASH_DECLARE(avx)
+#define SIPHASH_AVX SIPHASH_IMPL(CPUID_AVX, "avx", avx)
+#endif
 
 /* list implemenations from most optimized to least, with generic as the last entry */
 static const siphash_impl_t siphash_list[] = {
 		SIPHASH_GENERIC,
+#if defined(SIPHASH_AVX)
+		SIPHASH_AVX,
+#endif
 #if defined(SIPHASH_SSE41)
 		SIPHASH_SSE41,
 #endif
@@ -73,7 +80,6 @@ siphash_load(void)
 			}
 		}
 	}
-	fprintf(stderr, "selected %s\n", siphash_opt->desc);
 }
 
 void siphash24 (unsigned char *out, const unsigned char *in,
diff --git a/src/libcryptobox/siphash/sse41.S b/src/libcryptobox/siphash/sse41.S
index 58acfee8f..92c15671a 100644
--- a/src/libcryptobox/siphash/sse41.S
+++ b/src/libcryptobox/siphash/sse41.S
@@ -1,6 +1,11 @@
 #include "../macro.S"
 #include "constants.S"
 
+/*
+ * Generated by gcc-4.9 from siphash sse41 implementation written by
+ * Samuel Neves and submitted to supercop competition
+ */
+
 SECTION_TEXT
 
 GLOBAL_HIDDEN_FN siphash_sse41