32bit memcmp/strcmp/strncmp optimized for SSSE3/SSS4.2

1 files changed, 1905 insertions, 0 deletions

diff --git a/sysdeps/i386/i686/multiarch/memcmp-ssse3.S b/sysdeps/i386/i686/multiarch/memcmp-ssse3.S
new file mode 100644
index 0000000000..bfcf660729
--- /dev/null
+++ b/sysdeps/i386/i686/multiarch/memcmp-ssse3.S
@@ -0,0 +1,1905 @@
+/* memcmp with SSSE3
+   Copyright (C) 2010 Free Software Foundation, Inc.
+   Contributed by Intel Corporation.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, write to the Free
+   Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
+   02111-1307 USA.  */
+
+#ifndef NOT_IN_libc
+
+#include <sysdep.h>
+#include "asm-syntax.h"
+
+#ifndef MEMCMP
+# define MEMCMP		__memcmp_ssse3
+#endif
+
+#define CFI_PUSH(REG)						\
+  cfi_adjust_cfa_offset (4);					\
+  cfi_rel_offset (REG, 0)
+
+#define CFI_POP(REG)						\
+  cfi_adjust_cfa_offset (-4);					\
+  cfi_restore (REG)
+
+#define PUSH(REG)	pushl REG; CFI_PUSH (REG)
+#define POP(REG)	popl REG; CFI_POP (REG)
+
+#define PARMS		4
+#define BLK1		PARMS
+#define BLK2		BLK1+4
+#define LEN		BLK2+4
+#define RETURN_END	POP (%edi); POP (%esi); POP (%ebx); ret
+#define RETURN		RETURN_END; CFI_PUSH (%ebx); CFI_PUSH (%edi); \
+			CFI_PUSH (%esi)
+
+	.section .text.ssse3,"ax",@progbits
+ENTRY (MEMCMP)
+	movl	LEN(%esp), %ecx
+	movl	BLK1(%esp), %eax
+	cmp	$48, %ecx
+	movl	BLK2(%esp), %edx
+	jae	L(48bytesormore)
+	cmp	$1, %ecx
+	jbe	L(less1bytes)
+	PUSH (%ebx)
+	add	%ecx, %edx
+	add	%ecx, %eax
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(less1bytes):
+	jb	L(zero)
+	movb	(%eax), %cl
+	cmp	(%edx), %cl
+	je	L(zero)
+	mov	$1, %eax
+	ja	L(1bytesend)
+	neg	%eax
+L(1bytesend):
+	ret
+
+	ALIGN (4)
+L(zero):
+	mov	$0, %eax
+	ret
+	
+	ALIGN (4)
+L(48bytesormore):
+	PUSH (%ebx)
+	PUSH (%esi)
+	PUSH (%edi)
+	movdqu    (%eax), %xmm3
+	movdqu    (%edx), %xmm0
+	movl	%eax, %edi
+	movl	%edx, %esi
+	pcmpeqb   %xmm0, %xmm3
+	pmovmskb  %xmm3, %edx
+	lea	16(%edi), %edi
+
+	sub      $0xffff, %edx
+	lea	16(%esi), %esi
+	jnz	  L(less16bytes)
+	mov	%edi, %edx
+	and	$0xf, %edx
+	xor	%edx, %edi
+	sub	%edx, %esi
+	add	%edx, %ecx
+	mov	%esi, %edx
+	and	$0xf, %edx
+	jz	L(shr_0)
+	xor	%edx, %esi
+
+	cmp	$8, %edx
+	jae	L(next_unaligned_table)
+	cmp	$0, %edx
+	je	L(shr_0)
+	cmp	$1, %edx
+	je	L(shr_1)
+	cmp	$2, %edx
+	je	L(shr_2)
+	cmp	$3, %edx
+	je	L(shr_3)
+	cmp	$4, %edx
+	je	L(shr_4)
+	cmp	$5, %edx
+	je	L(shr_5)
+	cmp	$6, %edx
+	je	L(shr_6)
+	jmp	L(shr_7)
+
+	ALIGN (4)
+L(next_unaligned_table):
+	cmp	$8, %edx
+	je	L(shr_8)
+	cmp	$9, %edx
+	je	L(shr_9)
+	cmp	$10, %edx
+	je	L(shr_10)
+	cmp	$11, %edx
+	je	L(shr_11)
+	cmp	$12, %edx
+	je	L(shr_12)
+	cmp	$13, %edx
+	je	L(shr_13)
+	cmp	$14, %edx
+	je	L(shr_14)
+	jmp	L(shr_15)
+
+	ALIGN (4)
+L(shr_0):
+	cmp	$80, %ecx
+	jae	L(shr_0_gobble)
+	lea	-48(%ecx), %ecx
+	xor	%eax, %eax
+	movaps	(%esi), %xmm1
+	pcmpeqb	(%edi), %xmm1
+	movaps	16(%esi), %xmm2
+	pcmpeqb	16(%edi), %xmm2
+	pand	%xmm1, %xmm2
+	pmovmskb %xmm2, %edx
+	add	$32, %edi
+	add	$32, %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	
+
+	lea	(%ecx, %edi,1), %eax
+	lea	(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_0_gobble):
+	lea	-48(%ecx), %ecx
+	movdqa	(%esi), %xmm0
+	xor	%eax, %eax
+	pcmpeqb	(%edi), %xmm0
+	sub	$32, %ecx
+	movdqa	16(%esi), %xmm2
+	pcmpeqb	16(%edi), %xmm2
+L(shr_0_gobble_loop):
+	pand	%xmm0, %xmm2
+	sub	$32, %ecx
+	pmovmskb %xmm2, %edx
+	movdqa	%xmm0, %xmm1
+	movdqa	32(%esi), %xmm0
+	movdqa	48(%esi), %xmm2
+	sbb	$0xffff, %edx
+	pcmpeqb	32(%edi), %xmm0
+	pcmpeqb	48(%edi), %xmm2
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	jz	L(shr_0_gobble_loop)
+
+	pand	%xmm0, %xmm2
+	cmp	$0, %ecx
+	jge	L(shr_0_gobble_loop_next)
+	inc	%edx
+	add	$32, %ecx
+L(shr_0_gobble_loop_next):
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pmovmskb %xmm2, %edx
+	movdqa	%xmm0, %xmm1
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	lea	(%ecx, %edi,1), %eax
+	lea	(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_1):
+	cmp	$80, %ecx
+	lea	-48(%ecx), %ecx
+	mov	%edx, %eax
+	jae	L(shr_1_gobble)
+
+	movdqa	16(%esi), %xmm1
+	movdqa	%xmm1, %xmm2
+	palignr	$1,(%esi), %xmm1
+	pcmpeqb	(%edi), %xmm1
+
+	movdqa	32(%esi), %xmm3
+	palignr	$1,%xmm2, %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+	pand	%xmm1, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	lea	(%ecx, %edi,1), %eax
+	lea	1(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_1_gobble):
+	sub	$32, %ecx
+	movdqa	16(%esi), %xmm0
+	palignr	$1,(%esi), %xmm0
+	pcmpeqb	(%edi), %xmm0
+
+	movdqa	32(%esi), %xmm3
+	palignr	$1,16(%esi), %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+L(shr_1_gobble_loop):
+	pand	%xmm0, %xmm3
+	sub	$32, %ecx
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+
+	movdqa	64(%esi), %xmm3
+	palignr	$1,48(%esi), %xmm3
+	sbb	$0xffff, %edx
+	movdqa	48(%esi), %xmm0
+	palignr	$1,32(%esi), %xmm0
+	pcmpeqb	32(%edi), %xmm0
+	lea	32(%esi), %esi
+	pcmpeqb	48(%edi), %xmm3
+
+	lea	32(%edi), %edi
+	jz	L(shr_1_gobble_loop)
+
+	cmp	$0, %ecx
+	jge	L(shr_1_gobble_next)
+	inc	%edx
+	add	$32, %ecx
+L(shr_1_gobble_next):
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+
+	lea	(%ecx, %edi,1), %eax
+	lea	1(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+
+	ALIGN (4)
+L(shr_2):
+	cmp	$80, %ecx
+	lea	-48(%ecx), %ecx
+	mov	%edx, %eax
+	jae	L(shr_2_gobble)
+
+	movdqa	16(%esi), %xmm1
+	movdqa	%xmm1, %xmm2
+	palignr	$2,(%esi), %xmm1
+	pcmpeqb	(%edi), %xmm1
+
+	movdqa	32(%esi), %xmm3
+	palignr	$2,%xmm2, %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+	pand	%xmm1, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	lea	(%ecx, %edi,1), %eax
+	lea	2(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_2_gobble):
+	sub	$32, %ecx
+	movdqa	16(%esi), %xmm0
+	palignr	$2,(%esi), %xmm0
+	pcmpeqb	(%edi), %xmm0
+
+	movdqa	32(%esi), %xmm3
+	palignr	$2,16(%esi), %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+L(shr_2_gobble_loop):
+	pand	%xmm0, %xmm3
+	sub	$32, %ecx
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+
+	movdqa	64(%esi), %xmm3
+	palignr	$2,48(%esi), %xmm3
+	sbb	$0xffff, %edx
+	movdqa	48(%esi), %xmm0
+	palignr	$2,32(%esi), %xmm0
+	pcmpeqb	32(%edi), %xmm0
+	lea	32(%esi), %esi
+	pcmpeqb	48(%edi), %xmm3
+
+	lea	32(%edi), %edi
+	jz	L(shr_2_gobble_loop)
+
+	cmp	$0, %ecx
+	jge	L(shr_2_gobble_next)
+	inc	%edx
+	add	$32, %ecx
+L(shr_2_gobble_next):
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+
+	lea	(%ecx, %edi,1), %eax
+	lea	2(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_3):
+	cmp	$80, %ecx
+	lea	-48(%ecx), %ecx
+	mov	%edx, %eax
+	jae	L(shr_3_gobble)
+
+	movdqa	16(%esi), %xmm1
+	movdqa	%xmm1, %xmm2
+	palignr	$3,(%esi), %xmm1
+	pcmpeqb	(%edi), %xmm1
+
+	movdqa	32(%esi), %xmm3
+	palignr	$3,%xmm2, %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+	pand	%xmm1, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	lea	(%ecx, %edi,1), %eax
+	lea	3(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_3_gobble):
+	sub	$32, %ecx
+	movdqa	16(%esi), %xmm0
+	palignr	$3,(%esi), %xmm0
+	pcmpeqb	(%edi), %xmm0
+
+	movdqa	32(%esi), %xmm3
+	palignr	$3,16(%esi), %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+L(shr_3_gobble_loop):
+	pand	%xmm0, %xmm3
+	sub	$32, %ecx
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+
+	movdqa	64(%esi), %xmm3
+	palignr	$3,48(%esi), %xmm3
+	sbb	$0xffff, %edx
+	movdqa	48(%esi), %xmm0
+	palignr	$3,32(%esi), %xmm0
+	pcmpeqb	32(%edi), %xmm0
+	lea	32(%esi), %esi
+	pcmpeqb	48(%edi), %xmm3
+
+	lea	32(%edi), %edi
+	jz	L(shr_3_gobble_loop)
+
+	cmp	$0, %ecx
+	jge	L(shr_3_gobble_next)
+	inc	%edx
+	add	$32, %ecx
+L(shr_3_gobble_next):
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+
+	lea	(%ecx, %edi,1), %eax
+	lea	3(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_4):
+	cmp	$80, %ecx
+	lea	-48(%ecx), %ecx
+	mov	%edx, %eax
+	jae	L(shr_4_gobble)
+
+	movdqa	16(%esi), %xmm1
+	movdqa	%xmm1, %xmm2
+	palignr	$4,(%esi), %xmm1
+	pcmpeqb	(%edi), %xmm1
+
+	movdqa	32(%esi), %xmm3
+	palignr	$4,%xmm2, %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+	pand	%xmm1, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	lea	(%ecx, %edi,1), %eax
+	lea	4(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_4_gobble):
+	sub	$32, %ecx
+	movdqa	16(%esi), %xmm0
+	palignr	$4,(%esi), %xmm0
+	pcmpeqb	(%edi), %xmm0
+
+	movdqa	32(%esi), %xmm3
+	palignr	$4,16(%esi), %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+L(shr_4_gobble_loop):
+	pand	%xmm0, %xmm3
+	sub	$32, %ecx
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+
+	movdqa	64(%esi), %xmm3
+	palignr	$4,48(%esi), %xmm3
+	sbb	$0xffff, %edx
+	movdqa	48(%esi), %xmm0
+	palignr	$4,32(%esi), %xmm0
+	pcmpeqb	32(%edi), %xmm0
+	lea	32(%esi), %esi
+	pcmpeqb	48(%edi), %xmm3
+
+	lea	32(%edi), %edi
+	jz	L(shr_4_gobble_loop)
+
+	cmp	$0, %ecx
+	jge	L(shr_4_gobble_next)
+	inc	%edx
+	add	$32, %ecx
+L(shr_4_gobble_next):
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+
+	lea	(%ecx, %edi,1), %eax
+	lea	4(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_5):
+	cmp	$80, %ecx
+	lea	-48(%ecx), %ecx
+	mov	%edx, %eax
+	jae	L(shr_5_gobble)
+
+	movdqa	16(%esi), %xmm1
+	movdqa	%xmm1, %xmm2
+	palignr	$5,(%esi), %xmm1
+	pcmpeqb	(%edi), %xmm1
+
+	movdqa	32(%esi), %xmm3
+	palignr	$5,%xmm2, %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+	pand	%xmm1, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	lea	(%ecx, %edi,1), %eax
+	lea	5(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_5_gobble):
+	sub	$32, %ecx
+	movdqa	16(%esi), %xmm0
+	palignr	$5,(%esi), %xmm0
+	pcmpeqb	(%edi), %xmm0
+
+	movdqa	32(%esi), %xmm3
+	palignr	$5,16(%esi), %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+L(shr_5_gobble_loop):
+	pand	%xmm0, %xmm3
+	sub	$32, %ecx
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+
+	movdqa	64(%esi), %xmm3
+	palignr	$5,48(%esi), %xmm3
+	sbb	$0xffff, %edx
+	movdqa	48(%esi), %xmm0
+	palignr	$5,32(%esi), %xmm0
+	pcmpeqb	32(%edi), %xmm0
+	lea	32(%esi), %esi
+	pcmpeqb	48(%edi), %xmm3
+
+	lea	32(%edi), %edi
+	jz	L(shr_5_gobble_loop)
+
+	cmp	$0, %ecx
+	jge	L(shr_5_gobble_next)
+	inc	%edx
+	add	$32, %ecx
+L(shr_5_gobble_next):
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+
+	lea	(%ecx, %edi,1), %eax
+	lea	5(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_6):
+	cmp	$80, %ecx
+	lea	-48(%ecx), %ecx
+	mov	%edx, %eax
+	jae	L(shr_6_gobble)
+
+	movdqa	16(%esi), %xmm1
+	movdqa	%xmm1, %xmm2
+	palignr	$6,(%esi), %xmm1
+	pcmpeqb	(%edi), %xmm1
+
+	movdqa	32(%esi), %xmm3
+	palignr	$6,%xmm2, %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+	pand	%xmm1, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	lea	(%ecx, %edi,1), %eax
+	lea	6(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_6_gobble):
+	sub	$32, %ecx
+	movdqa	16(%esi), %xmm0
+	palignr	$6,(%esi), %xmm0
+	pcmpeqb	(%edi), %xmm0
+
+	movdqa	32(%esi), %xmm3
+	palignr	$6,16(%esi), %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+L(shr_6_gobble_loop):
+	pand	%xmm0, %xmm3
+	sub	$32, %ecx
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+
+	movdqa	64(%esi), %xmm3
+	palignr	$6,48(%esi), %xmm3
+	sbb	$0xffff, %edx
+	movdqa	48(%esi), %xmm0
+	palignr	$6,32(%esi), %xmm0
+	pcmpeqb	32(%edi), %xmm0
+	lea	32(%esi), %esi
+	pcmpeqb	48(%edi), %xmm3
+
+	lea	32(%edi), %edi
+	jz	L(shr_6_gobble_loop)
+
+	cmp	$0, %ecx
+	jge	L(shr_6_gobble_next)
+	inc	%edx
+	add	$32, %ecx
+L(shr_6_gobble_next):
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+
+	lea	(%ecx, %edi,1), %eax
+	lea	6(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_7):
+	cmp	$80, %ecx
+	lea	-48(%ecx), %ecx
+	mov	%edx, %eax
+	jae	L(shr_7_gobble)
+
+	movdqa	16(%esi), %xmm1
+	movdqa	%xmm1, %xmm2
+	palignr	$7,(%esi), %xmm1
+	pcmpeqb	(%edi), %xmm1
+
+	movdqa	32(%esi), %xmm3
+	palignr	$7,%xmm2, %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+	pand	%xmm1, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	lea	(%ecx, %edi,1), %eax
+	lea	7(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_7_gobble):
+	sub	$32, %ecx
+	movdqa	16(%esi), %xmm0
+	palignr	$7,(%esi), %xmm0
+	pcmpeqb	(%edi), %xmm0
+
+	movdqa	32(%esi), %xmm3
+	palignr	$7,16(%esi), %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+L(shr_7_gobble_loop):
+	pand	%xmm0, %xmm3
+	sub	$32, %ecx
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+
+	movdqa	64(%esi), %xmm3
+	palignr	$7,48(%esi), %xmm3
+	sbb	$0xffff, %edx
+	movdqa	48(%esi), %xmm0
+	palignr	$7,32(%esi), %xmm0
+	pcmpeqb	32(%edi), %xmm0
+	lea	32(%esi), %esi
+	pcmpeqb	48(%edi), %xmm3
+
+	lea	32(%edi), %edi
+	jz	L(shr_7_gobble_loop)
+
+	cmp	$0, %ecx
+	jge	L(shr_7_gobble_next)
+	inc	%edx
+	add	$32, %ecx
+L(shr_7_gobble_next):
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+
+	lea	(%ecx, %edi,1), %eax
+	lea	7(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_8):
+	cmp	$80, %ecx
+	lea	-48(%ecx), %ecx
+	mov	%edx, %eax
+	jae	L(shr_8_gobble)
+
+	movdqa	16(%esi), %xmm1
+	movdqa	%xmm1, %xmm2
+	palignr	$8,(%esi), %xmm1
+	pcmpeqb	(%edi), %xmm1
+
+	movdqa	32(%esi), %xmm3
+	palignr	$8,%xmm2, %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+	pand	%xmm1, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	lea	(%ecx, %edi,1), %eax
+	lea	8(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_8_gobble):
+	sub	$32, %ecx
+	movdqa	16(%esi), %xmm0
+	palignr	$8,(%esi), %xmm0
+	pcmpeqb	(%edi), %xmm0
+
+	movdqa	32(%esi), %xmm3
+	palignr	$8,16(%esi), %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+L(shr_8_gobble_loop):
+	pand	%xmm0, %xmm3
+	sub	$32, %ecx
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+
+	movdqa	64(%esi), %xmm3
+	palignr	$8,48(%esi), %xmm3
+	sbb	$0xffff, %edx
+	movdqa	48(%esi), %xmm0
+	palignr	$8,32(%esi), %xmm0
+	pcmpeqb	32(%edi), %xmm0
+	lea	32(%esi), %esi
+	pcmpeqb	48(%edi), %xmm3
+
+	lea	32(%edi), %edi
+	jz	L(shr_8_gobble_loop)
+
+	cmp	$0, %ecx
+	jge	L(shr_8_gobble_next)
+	inc	%edx
+	add	$32, %ecx
+L(shr_8_gobble_next):
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+
+	lea	(%ecx, %edi,1), %eax
+	lea	8(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_9):
+	cmp	$80, %ecx
+	lea	-48(%ecx), %ecx
+	mov	%edx, %eax
+	jae	L(shr_9_gobble)
+
+	movdqa	16(%esi), %xmm1
+	movdqa	%xmm1, %xmm2
+	palignr	$9,(%esi), %xmm1
+	pcmpeqb	(%edi), %xmm1
+
+	movdqa	32(%esi), %xmm3
+	palignr	$9,%xmm2, %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+	pand	%xmm1, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	lea	(%ecx, %edi,1), %eax
+	lea	9(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_9_gobble):
+	sub	$32, %ecx
+	movdqa	16(%esi), %xmm0
+	palignr	$9,(%esi), %xmm0
+	pcmpeqb	(%edi), %xmm0
+
+	movdqa	32(%esi), %xmm3
+	palignr	$9,16(%esi), %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+L(shr_9_gobble_loop):
+	pand	%xmm0, %xmm3
+	sub	$32, %ecx
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+
+	movdqa	64(%esi), %xmm3
+	palignr	$9,48(%esi), %xmm3
+	sbb	$0xffff, %edx
+	movdqa	48(%esi), %xmm0
+	palignr	$9,32(%esi), %xmm0
+	pcmpeqb	32(%edi), %xmm0
+	lea	32(%esi), %esi
+	pcmpeqb	48(%edi), %xmm3
+
+	lea	32(%edi), %edi
+	jz	L(shr_9_gobble_loop)
+
+	cmp	$0, %ecx
+	jge	L(shr_9_gobble_next)
+	inc	%edx
+	add	$32, %ecx
+L(shr_9_gobble_next):
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+
+	lea	(%ecx, %edi,1), %eax
+	lea	9(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_10):
+	cmp	$80, %ecx
+	lea	-48(%ecx), %ecx
+	mov	%edx, %eax
+	jae	L(shr_10_gobble)
+
+	movdqa	16(%esi), %xmm1
+	movdqa	%xmm1, %xmm2
+	palignr	$10, (%esi), %xmm1
+	pcmpeqb	(%edi), %xmm1
+
+	movdqa	32(%esi), %xmm3
+	palignr	$10,%xmm2, %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+	pand	%xmm1, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	lea	(%ecx, %edi,1), %eax
+	lea	10(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_10_gobble):
+	sub	$32, %ecx
+	movdqa	16(%esi), %xmm0
+	palignr	$10, (%esi), %xmm0
+	pcmpeqb	(%edi), %xmm0
+
+	movdqa	32(%esi), %xmm3
+	palignr	$10, 16(%esi), %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+L(shr_10_gobble_loop):
+	pand	%xmm0, %xmm3
+	sub	$32, %ecx
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+
+	movdqa	64(%esi), %xmm3
+	palignr	$10,48(%esi), %xmm3
+	sbb	$0xffff, %edx
+	movdqa	48(%esi), %xmm0
+	palignr	$10,32(%esi), %xmm0
+	pcmpeqb	32(%edi), %xmm0
+	lea	32(%esi), %esi
+	pcmpeqb	48(%edi), %xmm3
+
+	lea	32(%edi), %edi
+	jz	L(shr_10_gobble_loop)
+
+	cmp	$0, %ecx
+	jge	L(shr_10_gobble_next)
+	inc	%edx
+	add	$32, %ecx
+L(shr_10_gobble_next):
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+
+	lea	(%ecx, %edi,1), %eax
+	lea	10(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_11):
+	cmp	$80, %ecx
+	lea	-48(%ecx), %ecx
+	mov	%edx, %eax
+	jae	L(shr_11_gobble)
+
+	movdqa	16(%esi), %xmm1
+	movdqa	%xmm1, %xmm2
+	palignr	$11, (%esi), %xmm1
+	pcmpeqb	(%edi), %xmm1
+
+	movdqa	32(%esi), %xmm3
+	palignr	$11, %xmm2, %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+	pand	%xmm1, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	lea	(%ecx, %edi,1), %eax
+	lea	11(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_11_gobble):
+	sub	$32, %ecx
+	movdqa	16(%esi), %xmm0
+	palignr	$11, (%esi), %xmm0
+	pcmpeqb	(%edi), %xmm0
+
+	movdqa	32(%esi), %xmm3
+	palignr	$11, 16(%esi), %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+L(shr_11_gobble_loop):
+	pand	%xmm0, %xmm3
+	sub	$32, %ecx
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+
+	movdqa	64(%esi), %xmm3
+	palignr	$11,48(%esi), %xmm3
+	sbb	$0xffff, %edx
+	movdqa	48(%esi), %xmm0
+	palignr	$11,32(%esi), %xmm0
+	pcmpeqb	32(%edi), %xmm0
+	lea	32(%esi), %esi
+	pcmpeqb	48(%edi), %xmm3
+
+	lea	32(%edi), %edi
+	jz	L(shr_11_gobble_loop)
+
+	cmp	$0, %ecx
+	jge	L(shr_11_gobble_next)
+	inc	%edx
+	add	$32, %ecx
+L(shr_11_gobble_next):
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+
+	lea	(%ecx, %edi,1), %eax
+	lea	11(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_12):
+	cmp	$80, %ecx
+	lea	-48(%ecx), %ecx
+	mov	%edx, %eax
+	jae	L(shr_12_gobble)
+
+	movdqa	16(%esi), %xmm1
+	movdqa	%xmm1, %xmm2
+	palignr	$12, (%esi), %xmm1
+	pcmpeqb	(%edi), %xmm1
+
+	movdqa	32(%esi), %xmm3
+	palignr	$12, %xmm2, %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+	pand	%xmm1, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	lea	(%ecx, %edi,1), %eax
+	lea	12(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_12_gobble):
+	sub	$32, %ecx
+	movdqa	16(%esi), %xmm0
+	palignr	$12, (%esi), %xmm0
+	pcmpeqb	(%edi), %xmm0
+
+	movdqa	32(%esi), %xmm3
+	palignr	$12, 16(%esi), %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+L(shr_12_gobble_loop):
+	pand	%xmm0, %xmm3
+	sub	$32, %ecx
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+
+	movdqa	64(%esi), %xmm3
+	palignr	$12,48(%esi), %xmm3
+	sbb	$0xffff, %edx
+	movdqa	48(%esi), %xmm0
+	palignr	$12,32(%esi), %xmm0
+	pcmpeqb	32(%edi), %xmm0
+	lea	32(%esi), %esi
+	pcmpeqb	48(%edi), %xmm3
+
+	lea	32(%edi), %edi
+	jz	L(shr_12_gobble_loop)
+
+	cmp	$0, %ecx
+	jge	L(shr_12_gobble_next)
+	inc	%edx
+	add	$32, %ecx
+L(shr_12_gobble_next):
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+
+	lea	(%ecx, %edi,1), %eax
+	lea	12(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_13):
+	cmp	$80, %ecx
+	lea	-48(%ecx), %ecx
+	mov	%edx, %eax
+	jae	L(shr_13_gobble)
+
+	movdqa	16(%esi), %xmm1
+	movdqa	%xmm1, %xmm2
+	palignr	$13, (%esi), %xmm1
+	pcmpeqb	(%edi), %xmm1
+
+	movdqa	32(%esi), %xmm3
+	palignr	$13, %xmm2, %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+	pand	%xmm1, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	lea	(%ecx, %edi,1), %eax
+	lea	13(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_13_gobble):
+	sub	$32, %ecx
+	movdqa	16(%esi), %xmm0
+	palignr	$13, (%esi), %xmm0
+	pcmpeqb	(%edi), %xmm0
+
+	movdqa	32(%esi), %xmm3
+	palignr	$13, 16(%esi), %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+L(shr_13_gobble_loop):
+	pand	%xmm0, %xmm3
+	sub	$32, %ecx
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+
+	movdqa	64(%esi), %xmm3
+	palignr	$13,48(%esi), %xmm3
+	sbb	$0xffff, %edx
+	movdqa	48(%esi), %xmm0
+	palignr	$13,32(%esi), %xmm0
+	pcmpeqb	32(%edi), %xmm0
+	lea	32(%esi), %esi
+	pcmpeqb	48(%edi), %xmm3
+
+	lea	32(%edi), %edi
+	jz	L(shr_13_gobble_loop)
+
+	cmp	$0, %ecx
+	jge	L(shr_13_gobble_next)
+	inc	%edx
+	add	$32, %ecx
+L(shr_13_gobble_next):
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+
+	lea	(%ecx, %edi,1), %eax
+	lea	13(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_14):
+	cmp	$80, %ecx
+	lea	-48(%ecx), %ecx
+	mov	%edx, %eax
+	jae	L(shr_14_gobble)
+
+	movdqa	16(%esi), %xmm1
+	movdqa	%xmm1, %xmm2
+	palignr	$14, (%esi), %xmm1
+	pcmpeqb	(%edi), %xmm1
+
+	movdqa	32(%esi), %xmm3
+	palignr	$14, %xmm2, %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+	pand	%xmm1, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	lea	(%ecx, %edi,1), %eax
+	lea	14(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_14_gobble):
+	sub	$32, %ecx
+	movdqa	16(%esi), %xmm0
+	palignr	$14, (%esi), %xmm0
+	pcmpeqb	(%edi), %xmm0
+
+	movdqa	32(%esi), %xmm3
+	palignr	$14, 16(%esi), %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+L(shr_14_gobble_loop):
+	pand	%xmm0, %xmm3
+	sub	$32, %ecx
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+
+	movdqa	64(%esi), %xmm3
+	palignr	$14,48(%esi), %xmm3
+	sbb	$0xffff, %edx
+	movdqa	48(%esi), %xmm0
+	palignr	$14,32(%esi), %xmm0
+	pcmpeqb	32(%edi), %xmm0
+	lea	32(%esi), %esi
+	pcmpeqb	48(%edi), %xmm3
+
+	lea	32(%edi), %edi
+	jz	L(shr_14_gobble_loop)
+
+	cmp	$0, %ecx
+	jge	L(shr_14_gobble_next)
+	inc	%edx
+	add	$32, %ecx
+L(shr_14_gobble_next):
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+
+	lea	(%ecx, %edi,1), %eax
+	lea	14(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_15):
+	cmp	$80, %ecx
+	lea	-48(%ecx), %ecx
+	mov	%edx, %eax
+	jae	L(shr_15_gobble)
+
+	movdqa	16(%esi), %xmm1
+	movdqa	%xmm1, %xmm2
+	palignr	$15, (%esi), %xmm1
+	pcmpeqb	(%edi), %xmm1
+
+	movdqa	32(%esi), %xmm3
+	palignr	$15, %xmm2, %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+	pand	%xmm1, %xmm3
+	pmovmskb %xmm3, %edx
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+	lea	(%ecx, %edi,1), %eax
+	lea	15(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(shr_15_gobble):
+	sub	$32, %ecx
+	movdqa	16(%esi), %xmm0
+	palignr	$15, (%esi), %xmm0
+	pcmpeqb	(%edi), %xmm0
+
+	movdqa	32(%esi), %xmm3
+	palignr	$15, 16(%esi), %xmm3
+	pcmpeqb	16(%edi), %xmm3
+
+L(shr_15_gobble_loop):
+	pand	%xmm0, %xmm3
+	sub	$32, %ecx
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+
+	movdqa	64(%esi), %xmm3
+	palignr	$15,48(%esi), %xmm3
+	sbb	$0xffff, %edx
+	movdqa	48(%esi), %xmm0
+	palignr	$15,32(%esi), %xmm0
+	pcmpeqb	32(%edi), %xmm0
+	lea	32(%esi), %esi
+	pcmpeqb	48(%edi), %xmm3
+
+	lea	32(%edi), %edi
+	jz	L(shr_15_gobble_loop)
+
+	cmp	$0, %ecx
+	jge	L(shr_15_gobble_next)
+	inc	%edx
+	add	$32, %ecx
+L(shr_15_gobble_next):
+	test	%edx, %edx
+	jnz	L(exit)
+
+	pmovmskb %xmm3, %edx
+	movdqa	%xmm0, %xmm1
+	lea	32(%edi), %edi
+	lea	32(%esi), %esi
+	sub	$0xffff, %edx
+	jnz	L(exit)
+
+	lea	(%ecx, %edi,1), %eax
+	lea	15(%ecx, %esi,1), %edx
+	POP (%edi)
+	POP (%esi)
+	jmp	L(less48bytes)
+
+	ALIGN (4)
+L(exit):
+	pmovmskb %xmm1, %ebx
+	sub	$0xffff, %ebx
+	jz	L(first16bytes)
+	lea	-16(%esi), %esi
+	lea	-16(%edi), %edi
+	mov	%ebx, %edx
+L(first16bytes):
+	add	%eax, %esi
+L(less16bytes):
+	test	%dl, %dl
+	jz	L(next_24_bytes)
+
+	test	$0x01, %dl
+	jnz	L(Byte16)
+
+	test	$0x02, %dl
+	jnz	L(Byte17)
+
+	test	$0x04, %dl
+	jnz	L(Byte18)
+
+	test	$0x08, %dl
+	jnz	L(Byte19)
+
+	test	$0x10, %dl
+	jnz	L(Byte20)
+
+	test	$0x20, %dl
+	jnz	L(Byte21)
+
+	test	$0x40, %dl
+	jnz	L(Byte22)
+L(Byte23):
+	movzbl	 -9(%edi), %eax
+	movzbl	 -9(%esi), %edx
+	sub	%edx, %eax
+	RETURN
+
+	ALIGN (4)
+L(Byte16):
+	movzbl	 -16(%edi), %eax
+	movzbl	 -16(%esi), %edx
+	sub	%edx, %eax
+	RETURN
+
+	ALIGN (4)
+L(Byte17):
+	movzbl	 -15(%edi), %eax
+	movzbl	 -15(%esi), %edx
+	sub	%edx, %eax
+	RETURN
+
+	ALIGN (4)
+L(Byte18):
+	movzbl	 -14(%edi), %eax
+	movzbl	 -14(%esi), %edx
+	sub	%edx, %eax
+	RETURN
+
+	ALIGN (4)
+L(Byte19):
+	movzbl	 -13(%edi), %eax
+	movzbl	 -13(%esi), %edx
+	sub	%edx, %eax
+	RETURN
+
+	ALIGN (4)
+L(Byte20):
+	movzbl	 -12(%edi), %eax
+	movzbl	 -12(%esi), %edx
+	sub	%edx, %eax
+	RETURN
+
+	ALIGN (4)
+L(Byte21):
+	movzbl	 -11(%edi), %eax
+	movzbl	 -11(%esi), %edx
+	sub	%edx, %eax
+	RETURN
+
+	ALIGN (4)
+L(Byte22):
+	movzbl	 -10(%edi), %eax
+	movzbl	 -10(%esi), %edx
+	sub	%edx, %eax
+	RETURN
+
+	ALIGN (4)
+L(next_24_bytes):
+	lea	8(%edi), %edi
+	lea	8(%esi), %esi
+	test	$0x01, %dh
+	jnz	L(Byte16)
+
+	test	$0x02, %dh
+	jnz	L(Byte17)
+
+	test	$0x04, %dh
+	jnz	L(Byte18)
+
+	test	$0x08, %dh
+	jnz	L(Byte19)
+
+	test	$0x10, %dh
+	jnz	L(Byte20)
+
+	test	$0x20, %dh
+	jnz	L(Byte21)
+
+	test	$0x40, %dh
+	jnz	L(Byte22)
+
+	ALIGN (4)
+L(Byte31):
+	movzbl	 -9(%edi), %eax
+	movzbl	 -9(%esi), %edx
+	sub	%edx, %eax
+	RETURN
+
+	ALIGN (4)
+L(more8bytes):
+	cmp	$16, %ecx
+	jae	L(more16bytes)
+	cmp	$8, %ecx
+	je	L(8bytes)
+	cmp	$9, %ecx
+	je	L(9bytes)
+	cmp	$10, %ecx
+	je	L(10bytes)
+	cmp	$11, %ecx
+	je	L(11bytes)
+	cmp	$12, %ecx
+	je	L(12bytes)
+	cmp	$13, %ecx
+	je	L(13bytes)
+	cmp	$14, %ecx
+	je	L(14bytes)
+	jmp	L(15bytes)
+
+	ALIGN (4)
+L(more16bytes):
+	cmp	$24, %ecx
+	jae	L(more24bytes)
+	cmp	$16, %ecx
+	je	L(16bytes)
+	cmp	$17, %ecx
+	je	L(17bytes)
+	cmp	$18, %ecx
+	je	L(18bytes)
+	cmp	$19, %ecx
+	je	L(19bytes)
+	cmp	$20, %ecx
+	je	L(20bytes)
+	cmp	$21, %ecx
+	je	L(21bytes)
+	cmp	$22, %ecx
+	je	L(22bytes)
+	jmp	L(23bytes)
+
+	ALIGN (4)
+L(more24bytes):
+	cmp	$32, %ecx
+	jae	L(more32bytes)
+	cmp	$24, %ecx
+	je	L(24bytes)
+	cmp	$25, %ecx
+	je	L(25bytes)
+	cmp	$26, %ecx
+	je	L(26bytes)
+	cmp	$27, %ecx
+	je	L(27bytes)
+	cmp	$28, %ecx
+	je	L(28bytes)
+	cmp	$29, %ecx
+	je	L(29bytes)
+	cmp	$30, %ecx
+	je	L(30bytes)
+	jmp	L(31bytes)
+
+	ALIGN (4)
+L(more32bytes):
+	cmp	$40, %ecx
+	jae	L(more40bytes)
+	cmp	$32, %ecx
+	je	L(32bytes)
+	cmp	$33, %ecx
+	je	L(33bytes)
+	cmp	$34, %ecx
+	je	L(34bytes)
+	cmp	$35, %ecx
+	je	L(35bytes)
+	cmp	$36, %ecx
+	je	L(36bytes)
+	cmp	$37, %ecx
+	je	L(37bytes)
+	cmp	$38, %ecx
+	je	L(38bytes)
+	jmp	L(39bytes)
+
+	ALIGN (4)
+L(more40bytes):
+	cmp	$40, %ecx
+	je	L(40bytes)
+	cmp	$41, %ecx
+	je	L(41bytes)
+	cmp	$42, %ecx
+	je	L(42bytes)
+	cmp	$43, %ecx
+	je	L(43bytes)
+	cmp	$44, %ecx
+	je	L(44bytes)
+	cmp	$45, %ecx
+	je	L(45bytes)
+	cmp	$46, %ecx
+	je	L(46bytes)
+	jmp	L(47bytes)
+
+
+	ALIGN (4)
+L(less48bytes):
+	cmp	$8, %ecx
+	jae	L(more8bytes)
+	cmp	$2, %ecx
+	je	L(2bytes)
+	cmp	$3, %ecx
+	je	L(3bytes)
+	cmp	$4, %ecx
+	je	L(4bytes)
+	cmp	$5, %ecx
+	je	L(5bytes)
+	cmp	$6, %ecx
+	je	L(6bytes)
+	jmp	L(7bytes)
+
+
+
+
+	ALIGN (4)
+L(44bytes):
+	mov	-44(%eax), %ecx
+	mov	-44(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(40bytes):
+	mov	-40(%eax), %ecx
+	mov	-40(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(36bytes):
+	mov	-36(%eax), %ecx
+	mov	-36(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(32bytes):
+	mov	-32(%eax), %ecx
+	mov	-32(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(28bytes):
+	mov	-28(%eax), %ecx
+	mov	-28(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(24bytes):
+	mov	-24(%eax), %ecx
+	mov	-24(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(20bytes):
+	mov	-20(%eax), %ecx
+	mov	-20(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(16bytes):
+	mov	-16(%eax), %ecx
+	mov	-16(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(12bytes):
+	mov	-12(%eax), %ecx
+	mov	-12(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(8bytes):
+	mov	-8(%eax), %ecx
+	mov	-8(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(4bytes):
+	mov	-4(%eax), %ecx
+	mov	-4(%edx), %ebx
+	cmp	%ebx, %ecx
+	mov	$0, %eax
+	jne	L(find_diff)
+	POP (%ebx)
+	ret
+	CFI_PUSH (%ebx)
+
+	ALIGN (4)
+L(45bytes):
+	mov	-45(%eax), %ecx
+	mov	-45(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(41bytes):
+	mov	-41(%eax), %ecx
+	mov	-41(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(37bytes):
+	mov	-37(%eax), %ecx
+	mov	-37(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(33bytes):
+	mov	-33(%eax), %ecx
+	mov	-33(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(29bytes):
+	mov	-29(%eax), %ecx
+	mov	-29(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(25bytes):
+	mov	-25(%eax), %ecx
+	mov	-25(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(21bytes):
+	mov	-21(%eax), %ecx
+	mov	-21(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(17bytes):
+	mov	-17(%eax), %ecx
+	mov	-17(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(13bytes):
+	mov	-13(%eax), %ecx
+	mov	-13(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(9bytes):
+	mov	-9(%eax), %ecx
+	mov	-9(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(5bytes):
+	mov	-5(%eax), %ecx
+	mov	-5(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+	movzbl	-1(%eax), %ecx
+	cmp	-1(%edx), %cl
+	mov	$0, %eax
+	jne	L(end)
+	POP (%ebx)
+	ret
+	CFI_PUSH (%ebx)
+
+	ALIGN (4)
+L(46bytes):
+	mov	-46(%eax), %ecx
+	mov	-46(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(42bytes):
+	mov	-42(%eax), %ecx
+	mov	-42(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(38bytes):
+	mov	-38(%eax), %ecx
+	mov	-38(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(34bytes):
+	mov	-34(%eax), %ecx
+	mov	-34(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(30bytes):
+	mov	-30(%eax), %ecx
+	mov	-30(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(26bytes):
+	mov	-26(%eax), %ecx
+	mov	-26(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(22bytes):
+	mov	-22(%eax), %ecx
+	mov	-22(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(18bytes):
+	mov	-18(%eax), %ecx
+	mov	-18(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(14bytes):
+	mov	-14(%eax), %ecx
+	mov	-14(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(10bytes):
+	mov	-10(%eax), %ecx
+	mov	-10(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(6bytes):
+	mov	-6(%eax), %ecx
+	mov	-6(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(2bytes):
+	movzwl	-2(%eax), %ecx
+	movzwl	-2(%edx), %ebx
+	cmp	%bl, %cl
+	jne	L(end)
+	cmp	%bh, %ch
+	mov	$0, %eax
+	jne	L(end)
+	POP (%ebx)
+	ret
+	CFI_PUSH (%ebx)
+
+	ALIGN (4)
+L(47bytes):
+	movl	-47(%eax), %ecx
+	movl	-47(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(43bytes):
+	movl	-43(%eax), %ecx
+	movl	-43(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(39bytes):
+	movl	-39(%eax), %ecx
+	movl	-39(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(35bytes):
+	movl	-35(%eax), %ecx
+	movl	-35(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(31bytes):
+	movl	-31(%eax), %ecx
+	movl	-31(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(27bytes):
+	movl	-27(%eax), %ecx
+	movl	-27(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(23bytes):
+	movl	-23(%eax), %ecx
+	movl	-23(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(19bytes):
+	movl	-19(%eax), %ecx
+	movl	-19(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(15bytes):
+	movl	-15(%eax), %ecx
+	movl	-15(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(11bytes):
+	movl	-11(%eax), %ecx
+	movl	-11(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(7bytes):
+	movl	-7(%eax), %ecx
+	movl	-7(%edx), %ebx
+	cmp	%ebx, %ecx
+	jne	L(find_diff)
+L(3bytes):
+	movzwl	-3(%eax), %ecx
+	movzwl	-3(%edx), %ebx
+	cmpb	%bl, %cl
+	jne	L(end)
+	cmp	%bx, %cx
+	jne	L(end)
+	movzbl	-1(%eax), %eax
+	cmpb	-1(%edx), %al
+	mov	$0, %eax
+	jne	L(end)
+	POP (%ebx)
+	ret
+	CFI_PUSH (%ebx)
+
+	ALIGN (4)
+L(find_diff):
+	cmpb	%bl, %cl
+	jne	L(end)
+	cmp	%bx, %cx
+	jne	L(end)
+	shr	$16,%ecx
+	shr	$16,%ebx
+	cmp	%bl, %cl
+	jne	L(end)
+	cmp	%bx, %cx
+L(end):
+	POP (%ebx)
+	mov	$1, %eax
+	ja	L(bigger)
+	neg	%eax
+L(bigger):
+	ret
+
+END (MEMCMP)
+
+#endif