This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.
Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]
Other format:	[Raw text]
Re: [RFC v1.1] A memcpy ssse3/sse2 variant for older processors.

From: OndÅej BÃlka <neleai at seznam dot cz>
To: libc-alpha at sourceware dot org
Date: Thu, 29 Aug 2013 17:53:58 +0200
Subject: Re: [RFC v1.1] A memcpy ssse3/sse2 variant for older processors.
Authentication-results: sourceware.org; auth=none
References: <20130817221036 dot GA9150 at domone> <20130819084034 dot GA19541 at domone>
ping.

On Mon, Aug 19, 2013 at 10:41:14AM +0200, OndÅej BÃlka wrote:
> On Sun, Aug 18, 2013 at 12:10:36AM +0200, OndÅej BÃlka wrote:
> > Hi,
> > 
> > I now return to improve memcpy implementation. Based on experiment that
> > I done I created a implementation that is around 10% faster on older
> > processors. It combines a rep movsq trick for small lengths with
> > optimized ssse3/sse2 loop.
> 
> I noticed that I did not though that I must set direction for rep movsq
> to be used as memmove. I refactored that and as overhead of memmove
> checking was around 1% I decided to simply alias it to memmove.
> 
> The backward,forward loop that I use there are still not optimal, does
> somebody want to optimize them?
> 
> For large sizes I noticed that my measurements so far were flawed, we
> could assume that only interesting case is when source and destination
> are aligned. If user does not do that it will cause performance penalty.
> 
> This makes rep movsq even more attractive as it is around cycle faster
> per 16 byte block that any alternative for tested architectures.
> 
> I updated
> http://kam.mff.cuni.cz/~ondra/benchmark_string/memcpy_profile_loop.html
> generated by
> http://kam.mff.cuni.cz/~ondra/benchmark_string/memcpy_profile_loop190813.tar.bz2
> 
> ---
>  sysdeps/x86_64/multiarch/memcpy-ssse3.S  | 3316 ++----------------------------
>  sysdeps/x86_64/multiarch/memcpy.S        |    2 +
>  sysdeps/x86_64/multiarch/memmove-ssse3.S |    5 +-
>  sysdeps/x86_64/multiarch/mempcpy-ssse3.S |    5 +-
>  4 files changed, 232 insertions(+), 3096 deletions(-)
> 
> diff --git a/sysdeps/x86_64/multiarch/memcpy-ssse3.S b/sysdeps/x86_64/multiarch/memcpy-ssse3.S
> index 9642cee..b5c98e0 100644
> --- a/sysdeps/x86_64/multiarch/memcpy-ssse3.S
> +++ b/sysdeps/x86_64/multiarch/memcpy-ssse3.S
> @@ -19,10 +19,6 @@
>  
>  #include <sysdep.h>
>  
> -#if !defined NOT_IN_libc \
> -    && (defined SHARED \
> -        || defined USE_AS_MEMMOVE \
> -	|| !defined USE_MULTIARCH)
>  
>  #include "asm-syntax.h"
>  
> @@ -35,3104 +31,248 @@
>  # define ALIGN(n)	.p2align n
>  #endif
>  
> -#define JMPTBL(I, B)	I - B
> -
> -/* Branch to an entry in a jump table.  TABLE is a jump table with
> -   relative offsets.  INDEX is a register contains the index into the
> -   jump table.  SCALE is the scale of INDEX.  */
> -#define BRANCH_TO_JMPTBL_ENTRY(TABLE, INDEX, SCALE)		\
> -  lea		TABLE(%rip), %r11;				\
> -  movslq	(%r11, INDEX, SCALE), INDEX;			\
> -  lea		(%r11, INDEX), INDEX;				\
> -  jmp		*INDEX;						\
> -  ud2
> -
>  	.section .text.ssse3,"ax",@progbits
> -#if !defined USE_AS_BCOPY
> -ENTRY (MEMCPY_CHK)
> +ENTRY (__mempcpy_chk_ssse3)
>  	cmpq	%rdx, %rcx
>  	jb	HIDDEN_JUMPTARGET (__chk_fail)
> -END (MEMCPY_CHK)
> -#endif
> -
> -ENTRY (MEMCPY)
> -	mov	%rdi, %rax
> -#ifdef USE_AS_MEMPCPY
> -	add	%rdx, %rax
> -#endif
> -
> -#ifdef USE_AS_MEMMOVE
> -	cmp	%rsi, %rdi
> -	jb	L(copy_forward)
> -	je	L(write_0bytes)
> -	cmp	$79, %rdx
> -	jbe	L(copy_forward)
> -	jmp	L(copy_backward)
> -L(copy_forward):
> -#endif
> -	cmp	$79, %rdx
> -	lea     L(table_less_80bytes)(%rip), %r11
> -	ja	L(80bytesormore)
> -	movslq	(%r11, %rdx, 4), %r9
> -	add	%rdx, %rsi
> -	add	%rdx, %rdi
> -	add	%r11, %r9
> -	jmp	*%r9
> -	ud2
> -
> -	ALIGN (4)
> -L(80bytesormore):
> -#ifndef USE_AS_MEMMOVE
> -	cmp	%dil, %sil
> -	jle	L(copy_backward)
> -#endif
> -
> -	movdqu	(%rsi), %xmm0
> -	mov	%rdi, %rcx
> -	and	$-16, %rdi
> -	add	$16, %rdi
> -	mov	%rcx, %r8
> -	sub	%rdi, %rcx
> -	add	%rcx, %rdx
> -	sub	%rcx, %rsi
> +END (__mempcpy_chk_ssse3)
>  
> -#ifdef SHARED_CACHE_SIZE_HALF
> -	mov	$SHARED_CACHE_SIZE_HALF, %RCX_LP
> -#else
> -	mov	__x86_shared_cache_size_half(%rip), %RCX_LP
> -#endif
> -	cmp	%rcx, %rdx
> -	mov	%rsi, %r9
> -	ja	L(large_page_fwd)
> -	and	$0xf, %r9
> -	jz	L(shl_0)
> -#ifdef DATA_CACHE_SIZE_HALF
> -	mov	$DATA_CACHE_SIZE_HALF, %RCX_LP
> -#else
> -	mov	__x86_data_cache_size_half(%rip), %RCX_LP
> -#endif
> -	BRANCH_TO_JMPTBL_ENTRY (L(shl_table), %r9, 4)
> -
> -	ALIGN (4)
> -L(copy_backward):
> -	movdqu	-16(%rsi, %rdx), %xmm0
> -	add	%rdx, %rsi
> -	lea	-16(%rdi, %rdx), %r8
> -	add	%rdx, %rdi
> -
> -	mov	%rdi, %rcx
> -	and	$0xf, %rcx
> -	xor	%rcx, %rdi
> -	sub	%rcx, %rdx
> -	sub	%rcx, %rsi
> -
> -#ifdef SHARED_CACHE_SIZE_HALF
> -	mov	$SHARED_CACHE_SIZE_HALF, %RCX_LP
> -#else
> -	mov	__x86_shared_cache_size_half(%rip), %RCX_LP
> -#endif
> -
> -	cmp	%rcx, %rdx
> -	mov	%rsi, %r9
> -	ja	L(large_page_bwd)
> -	and	$0xf, %r9
> -	jz	L(shl_0_bwd)
> -#ifdef DATA_CACHE_SIZE_HALF
> -	mov	$DATA_CACHE_SIZE_HALF, %RCX_LP
> -#else
> -	mov	__x86_data_cache_size_half(%rip), %RCX_LP
> -#endif
> -	BRANCH_TO_JMPTBL_ENTRY (L(shl_table_bwd), %r9, 4)
> -
> -	ALIGN (4)
> -L(shl_0):
> -	sub	$16, %rdx
> -	movdqa	(%rsi), %xmm1
> -	add	$16, %rsi
> -	movdqa	%xmm1, (%rdi)
> -	add	$16, %rdi
> -	cmp	$128, %rdx
> -	movdqu	%xmm0, (%r8)
> -	ja	L(shl_0_gobble)
> -	cmp	$64, %rdx
> -	jb	L(shl_0_less_64bytes)
> -	movaps	(%rsi), %xmm4
> -	movaps	16(%rsi), %xmm1
> -	movaps	32(%rsi), %xmm2
> -	movaps	48(%rsi), %xmm3
> -	movaps	%xmm4, (%rdi)
> -	movaps	%xmm1, 16(%rdi)
> -	movaps	%xmm2, 32(%rdi)
> -	movaps	%xmm3, 48(%rdi)
> -	sub	$64, %rdx
> -	add	$64, %rsi
> -	add	$64, %rdi
> -L(shl_0_less_64bytes):
> -	add	%rdx, %rsi
> -	add	%rdx, %rdi
> -	BRANCH_TO_JMPTBL_ENTRY (L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_0_gobble):
> -#ifdef DATA_CACHE_SIZE_HALF
> -	cmp	$DATA_CACHE_SIZE_HALF, %RDX_LP
> -#else
> -	cmp	__x86_data_cache_size_half(%rip), %RDX_LP
> -#endif
> -	lea	-128(%rdx), %rdx
> -	jae	L(shl_0_gobble_mem_loop)
> -L(shl_0_gobble_cache_loop):
> -	movdqa	(%rsi), %xmm4
> -	movaps	0x10(%rsi), %xmm1
> -	movaps	0x20(%rsi), %xmm2
> -	movaps	0x30(%rsi), %xmm3
> +ENTRY (__mempcpy_ssse3)
> +	movq    %rdi, %rax
> +	addq	%rdx, %rax
> +	cmpq    $256, %rdx
> +	ja      L(more_256_bytes)
> +	jmp L(from_mempcpy)
> +END (__mempcpy_ssse3)
>  
> -	movdqa	%xmm4, (%rdi)
> -	movaps	%xmm1, 0x10(%rdi)
> -	movaps	%xmm2, 0x20(%rdi)
> -	movaps	%xmm3, 0x30(%rdi)
>  
> -	sub	$128, %rdx
> -	movaps	0x40(%rsi), %xmm4
> -	movaps	0x50(%rsi), %xmm5
> -	movaps	0x60(%rsi), %xmm6
> -	movaps	0x70(%rsi), %xmm7
> -	lea	0x80(%rsi), %rsi
> -	movaps	%xmm4, 0x40(%rdi)
> -	movaps	%xmm5, 0x50(%rdi)
> -	movaps	%xmm6, 0x60(%rdi)
> -	movaps	%xmm7, 0x70(%rdi)
> -	lea	0x80(%rdi), %rdi
>  
> -	jae	L(shl_0_gobble_cache_loop)
> -	cmp	$-0x40, %rdx
> -	lea	0x80(%rdx), %rdx
> -	jl	L(shl_0_cache_less_64bytes)
>  
> -	movdqa	(%rsi), %xmm4
> -	sub	$0x40, %rdx
> -	movdqa	0x10(%rsi), %xmm1
> -
> -	movdqa	%xmm4, (%rdi)
> -	movdqa	%xmm1, 0x10(%rdi)
> -
> -	movdqa	0x20(%rsi), %xmm4
> -	movdqa	0x30(%rsi), %xmm1
> -	add	$0x40, %rsi
> -
> -	movdqa	%xmm4, 0x20(%rdi)
> -	movdqa	%xmm1, 0x30(%rdi)
> -	add	$0x40, %rdi
> -L(shl_0_cache_less_64bytes):
> -	add	%rdx, %rsi
> -	add	%rdx, %rdi
> -	BRANCH_TO_JMPTBL_ENTRY (L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_0_gobble_mem_loop):
> -	prefetcht0 0x1c0(%rsi)
> -	prefetcht0 0x280(%rsi)
> -
> -	movdqa	(%rsi), %xmm0
> -	movdqa	0x10(%rsi), %xmm1
> -	movdqa	0x20(%rsi), %xmm2
> -	movdqa	0x30(%rsi), %xmm3
> -	movdqa	0x40(%rsi), %xmm4
> -	movdqa	0x50(%rsi), %xmm5
> -	movdqa	0x60(%rsi), %xmm6
> -	movdqa	0x70(%rsi), %xmm7
> -	lea	0x80(%rsi), %rsi
> -	sub	$0x80, %rdx
> -	movdqa	%xmm0, (%rdi)
> -	movdqa	%xmm1, 0x10(%rdi)
> -	movdqa	%xmm2, 0x20(%rdi)
> -	movdqa	%xmm3, 0x30(%rdi)
> -	movdqa	%xmm4, 0x40(%rdi)
> -	movdqa	%xmm5, 0x50(%rdi)
> -	movdqa	%xmm6, 0x60(%rdi)
> -	movdqa	%xmm7, 0x70(%rdi)
> -	lea	0x80(%rdi), %rdi
> -
> -	jae	L(shl_0_gobble_mem_loop)
> -	cmp	$-0x40, %rdx
> -	lea	0x80(%rdx), %rdx
> -	jl	L(shl_0_mem_less_64bytes)
> +ENTRY (MEMCPY_CHK)
> +	cmpq	%rdx, %rcx
> +	jb	HIDDEN_JUMPTARGET (__chk_fail)
> +END (MEMCPY_CHK)
>  
> -	movdqa	(%rsi), %xmm0
> -	sub	$0x40, %rdx
> -	movdqa	0x10(%rsi), %xmm1
> +ENTRY (MEMCPY)
>  
> -	movdqa	%xmm0, (%rdi)
> -	movdqa	%xmm1, 0x10(%rdi)
> +	cmpq	$256, %rdx
> +	movq	%rdi, %rax
> +L(from_mempcpy):
> +	ja	L(more_256_bytes)
> +	cmpq	$16, %rdx
> +	jbe	L(less_16_bytes)
>  
> -	movdqa	0x20(%rsi), %xmm0
> -	movdqa	0x30(%rsi), %xmm1
> -	add	$0x40, %rsi
> +	movq	%rsi, %rcx
> +	subq	%rdi, %rcx
> +	cmpq	%rdx, %rcx
> +	jb	L(fwd)
>  
> -	movdqa	%xmm0, 0x20(%rdi)
> -	movdqa	%xmm1, 0x30(%rdi)
> -	add	$0x40, %rdi
> -L(shl_0_mem_less_64bytes):
> -	cmp	$0x20, %rdx
> -	jb	L(shl_0_mem_less_32bytes)
> +	movq	%rdi, %rcx
> +	subq	%rsi, %rcx
> +	cmpq	%rdx, %rcx
> +	jb	L(bwd)
> +
> +	movq	(%rsi), %rcx
> +	movq	%rcx, (%rdi)
> +	movq	-8(%rsi, %rdx), %rcx
> +	movq	%rcx, -8(%rdi, %rdx)
> +	/* Do one iteration less for 8-byte aligned (%rdi, %rdx).  */
> +	leaq	-1(%rdi, %rdx), %rcx 
> +	movq	%rdi, %rdx
> +	leaq	8(%rdi), %rdi
> +	andq	$-8, %rdi
> +	subq	%rdi, %rdx
> +	subq	%rdx, %rsi
> +	subq	%rdi, %rcx
> +	shrq	$3, %rcx
> +	rep ; movsq
> +	ret
> +
> +	ALIGN (4)
> +L(more_256_bytes):
> +	movq	%rsi, %rcx
> +	subq	%rdi, %rcx
> +	cmpq	%rdx, %rcx
> +	jb	L(fwd)
> +	movq	%rdi, %rcx
> +	subq	%rsi, %rcx
> +	cmpq	%rdx, %rcx
> +	jb	L(bwd)
> +
> +	leaq	16(%rdi), %rcx
> +	movdqu	-64(%rsi, %rdx), %xmm0
> +	movdqu	-16(%rsi, %rdx), %xmm6
> +	andq	$-16, %rcx
> +	movdqu	-32(%rsi, %rdx), %xmm4
> +	movdqu	-48(%rsi, %rdx), %xmm2
> +	movdqu	(%rsi), %xmm7
> +	movdqu	%xmm7, (%rdi)
> +	movdqu	%xmm6, -16(%rdi, %rdx)
> +	movdqu	%xmm4, -32(%rdi, %rdx)
> +	movdqu	%xmm2, -48(%rdi, %rdx)
> +	movdqu	%xmm0, -64(%rdi, %rdx)
> +	movq	%rdi, %r8
> +	movq	%rcx, %rdi
> +	addq	%r8, %rdx
> +	subq	%r8, %rdi
> +	subq	%rcx, %rdx
> +	addq	%rdi, %rsi
> +	leaq	63(%rdx), %rdi
> +	movq	%rsi, %r9
> +	andl	$15, %r9d
> +	subq	%r9, %rsi
> +	addq	$16, %rsi
> +	leaq	.L100(%rip), %rdi
> +	movdqu	-16(%rsi), %xmm0
> +	shrq	$6, %rdx
> +	imul	$96, %r9
> +	subq	%r9, %rdi
> +	jmp	*%rdi
> +	
> +#define	LOOP(shift,label)\
> +	ALIGN (4);\
> +	label##:;\
> +	movdqa	(%rsi), %xmm3;\
> +	movdqa	%xmm3, %xmm4;\
> +	movdqa	16(%rsi), %xmm2;\
> +	palignr	shift, %xmm0, %xmm4;\
> +	movdqa	32(%rsi), %xmm1;\
> +	movdqa	%xmm4, (%rcx);\
> +	movdqa	%xmm2, %xmm4;\
> +	palignr	shift, %xmm3, %xmm4;\
> +	movdqa	%xmm1, %xmm3;\
> +	movdqa	48(%rsi), %xmm0;\
> +	addq	$64, %rsi;\
> +	palignr	shift, %xmm2, %xmm3;\
> +	movdqa	%xmm4, 16(%rcx);\
> +	movdqa	%xmm0, %xmm4;\
> +	movdqa	%xmm3, 32(%rcx);\
> +	palignr	shift, %xmm1, %xmm4;\
> +	movdqa	%xmm4, 48(%rcx);\
> +	addq	$64, %rcx;\
> +	subq	$1, %rdx;\
> +	jnz     label;\
> +	ret
> +
> +	LOOP($15, .L115)
> +	LOOP($14, .L114)
> +	LOOP($13, .L113)
> +	LOOP($12, .L112)
> +	LOOP($11, .L111)
> +	LOOP($10, .L110)
> +	LOOP($9, .L109)
> +	LOOP($8, .L108)
> +	LOOP($7, .L107)
> +	LOOP($6, .L106)
> +	LOOP($5, .L105)
> +	LOOP($4, .L104)
> +	LOOP($3, .L103)
> +	LOOP($2, .L102)
> +	LOOP($1, .L101)
> +
> +	ALIGN (4)
> +.L100:
> +	movdqa	-16(%rsi), %xmm0
> +	movdqa	%xmm0, (%rcx)
>  	movdqa	(%rsi), %xmm0
> -	sub	$0x20, %rdx
> -	movdqa	0x10(%rsi), %xmm1
> -	add	$0x20, %rsi
> -	movdqa	%xmm0, (%rdi)
> -	movdqa	%xmm1, 0x10(%rdi)
> -	add	$0x20, %rdi
> -L(shl_0_mem_less_32bytes):
> -	add	%rdx, %rdi
> -	add	%rdx, %rsi
> -	BRANCH_TO_JMPTBL_ENTRY (L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_0_bwd):
> -	sub	$16, %rdx
> -	movdqa	-0x10(%rsi), %xmm1
> -	sub	$16, %rsi
> -	movdqa	%xmm1, -0x10(%rdi)
> -	sub	$16, %rdi
> -	cmp	$0x80, %rdx
> -	movdqu	%xmm0, (%r8)
> -	ja	L(shl_0_gobble_bwd)
> -	cmp	$64, %rdx
> -	jb	L(shl_0_less_64bytes_bwd)
> -	movaps	-0x10(%rsi), %xmm0
> -	movaps	-0x20(%rsi), %xmm1
> -	movaps	-0x30(%rsi), %xmm2
> -	movaps	-0x40(%rsi), %xmm3
> -	movaps	%xmm0, -0x10(%rdi)
> -	movaps	%xmm1, -0x20(%rdi)
> -	movaps	%xmm2, -0x30(%rdi)
> -	movaps	%xmm3, -0x40(%rdi)
> -	sub	$64, %rdx
> -	sub	$0x40, %rsi
> -	sub	$0x40, %rdi
> -L(shl_0_less_64bytes_bwd):
> -	BRANCH_TO_JMPTBL_ENTRY (L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_0_gobble_bwd):
> -#ifdef DATA_CACHE_SIZE_HALF
> -	cmp	$DATA_CACHE_SIZE_HALF, %RDX_LP
> -#else
> -	cmp	__x86_data_cache_size_half(%rip), %RDX_LP
> -#endif
> -	lea	-128(%rdx), %rdx
> -	jae	L(shl_0_gobble_mem_bwd_loop)
> -L(shl_0_gobble_bwd_loop):
> -	movdqa	-0x10(%rsi), %xmm0
> -	movaps	-0x20(%rsi), %xmm1
> -	movaps	-0x30(%rsi), %xmm2
> -	movaps	-0x40(%rsi), %xmm3
> -
> -	movdqa	%xmm0, -0x10(%rdi)
> -	movaps	%xmm1, -0x20(%rdi)
> -	movaps	%xmm2, -0x30(%rdi)
> -	movaps	%xmm3, -0x40(%rdi)
> -
> -	sub	$0x80, %rdx
> -	movaps	-0x50(%rsi), %xmm4
> -	movaps	-0x60(%rsi), %xmm5
> -	movaps	-0x70(%rsi), %xmm6
> -	movaps	-0x80(%rsi), %xmm7
> -	lea	-0x80(%rsi), %rsi
> -	movaps	%xmm4, -0x50(%rdi)
> -	movaps	%xmm5, -0x60(%rdi)
> -	movaps	%xmm6, -0x70(%rdi)
> -	movaps	%xmm7, -0x80(%rdi)
> -	lea	-0x80(%rdi), %rdi
> -
> -	jae	L(shl_0_gobble_bwd_loop)
> -	cmp	$-0x40, %rdx
> -	lea	0x80(%rdx), %rdx
> -	jl	L(shl_0_gobble_bwd_less_64bytes)
> -
> -	movdqa	-0x10(%rsi), %xmm0
> -	sub	$0x40, %rdx
> -	movdqa	-0x20(%rsi), %xmm1
> -
> -	movdqa	%xmm0, -0x10(%rdi)
> -	movdqa	%xmm1, -0x20(%rdi)
> -
> -	movdqa	-0x30(%rsi), %xmm0
> -	movdqa	-0x40(%rsi), %xmm1
> -	sub	$0x40, %rsi
> -
> -	movdqa	%xmm0, -0x30(%rdi)
> -	movdqa	%xmm1, -0x40(%rdi)
> -	sub	$0x40, %rdi
> -L(shl_0_gobble_bwd_less_64bytes):
> -	BRANCH_TO_JMPTBL_ENTRY (L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_0_gobble_mem_bwd_loop):
> -	prefetcht0 -0x1c0(%rsi)
> -	prefetcht0 -0x280(%rsi)
> -	movdqa	-0x10(%rsi), %xmm0
> -	movdqa	-0x20(%rsi), %xmm1
> -	movdqa	-0x30(%rsi), %xmm2
> -	movdqa	-0x40(%rsi), %xmm3
> -	movdqa	-0x50(%rsi), %xmm4
> -	movdqa	-0x60(%rsi), %xmm5
> -	movdqa	-0x70(%rsi), %xmm6
> -	movdqa	-0x80(%rsi), %xmm7
> -	lea	-0x80(%rsi), %rsi
> -	sub	$0x80, %rdx
> -	movdqa	%xmm0, -0x10(%rdi)
> -	movdqa	%xmm1, -0x20(%rdi)
> -	movdqa	%xmm2, -0x30(%rdi)
> -	movdqa	%xmm3, -0x40(%rdi)
> -	movdqa	%xmm4, -0x50(%rdi)
> -	movdqa	%xmm5, -0x60(%rdi)
> -	movdqa	%xmm6, -0x70(%rdi)
> -	movdqa	%xmm7, -0x80(%rdi)
> -	lea	-0x80(%rdi), %rdi
> -
> -	jae	L(shl_0_gobble_mem_bwd_loop)
> -	cmp	$-0x40, %rdx
> -	lea	0x80(%rdx), %rdx
> -	jl	L(shl_0_mem_bwd_less_64bytes)
> -
> -	movdqa	-0x10(%rsi), %xmm0
> -	sub	$0x40, %rdx
> -	movdqa	-0x20(%rsi), %xmm1
> -
> -	movdqa	%xmm0, -0x10(%rdi)
> -	movdqa	%xmm1, -0x20(%rdi)
> -
> -	movdqa	-0x30(%rsi), %xmm0
> -	movdqa	-0x40(%rsi), %xmm1
> -	sub	$0x40, %rsi
> -
> -	movdqa	%xmm0, -0x30(%rdi)
> -	movdqa	%xmm1, -0x40(%rdi)
> -	sub	$0x40, %rdi
> -L(shl_0_mem_bwd_less_64bytes):
> -	cmp	$0x20, %rdx
> -	jb	L(shl_0_mem_bwd_less_32bytes)
> -	movdqa	-0x10(%rsi), %xmm0
> -	sub	$0x20, %rdx
> -	movdqa	-0x20(%rsi), %xmm1
> -	sub	$0x20, %rsi
> -	movdqa	%xmm0, -0x10(%rdi)
> -	movdqa	%xmm1, -0x20(%rdi)
> -	sub	$0x20, %rdi
> -L(shl_0_mem_bwd_less_32bytes):
> -	BRANCH_TO_JMPTBL_ENTRY (L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_1):
> -	lea	(L(shl_1_loop_L1)-L(shl_1))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x01(%rsi), %xmm1
> -	jb	L(L1_fwd)
> -	lea	(L(shl_1_loop_L2)-L(shl_1_loop_L1))(%r9), %r9
> -L(L1_fwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_1_loop_L2):
> -	prefetchnta 0x1c0(%rsi)
> -L(shl_1_loop_L1):
> -	sub	$64, %rdx
> -	movaps	0x0f(%rsi), %xmm2
> -	movaps	0x1f(%rsi), %xmm3
> -	movaps	0x2f(%rsi), %xmm4
> -	movaps	0x3f(%rsi), %xmm5
> -	movdqa	%xmm5, %xmm6
> -	palignr	$1, %xmm4, %xmm5
> -	lea	64(%rsi), %rsi
> -	palignr	$1, %xmm3, %xmm4
> -	palignr	$1, %xmm2, %xmm3
> -	lea	64(%rdi), %rdi
> -	palignr	$1, %xmm1, %xmm2
> -	movdqa	%xmm6, %xmm1
> -	movdqa	%xmm2, -0x40(%rdi)
> -	movaps	%xmm3, -0x30(%rdi)
> -	jb	L(shl_1_end)
> -	movaps	%xmm4, -0x20(%rdi)
> -	movaps	%xmm5, -0x10(%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_1_end):
> -	movaps	%xmm4, -0x20(%rdi)
> -	lea	64(%rdx), %rdx
> -	movaps	%xmm5, -0x10(%rdi)
> -	add	%rdx, %rdi
> -	movdqu	%xmm0, (%r8)
> -	add	%rdx, %rsi
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_1_bwd):
> -	lea	(L(shl_1_bwd_loop_L1)-L(shl_1_bwd))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x01(%rsi), %xmm1
> -	jb	L(L1_bwd)
> -	lea	(L(shl_1_bwd_loop_L2)-L(shl_1_bwd_loop_L1))(%r9), %r9
> -L(L1_bwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_1_bwd_loop_L2):
> -	prefetchnta -0x1c0(%rsi)
> -L(shl_1_bwd_loop_L1):
> -	movaps	-0x11(%rsi), %xmm2
> -	sub	$0x40, %rdx
> -	movaps	-0x21(%rsi), %xmm3
> -	movaps	-0x31(%rsi), %xmm4
> -	movaps	-0x41(%rsi), %xmm5
> -	lea	-0x40(%rsi), %rsi
> -	palignr	$1, %xmm2, %xmm1
> -	palignr	$1, %xmm3, %xmm2
> -	palignr	$1, %xmm4, %xmm3
> -	palignr	$1, %xmm5, %xmm4
> -
> -	movaps	%xmm1, -0x10(%rdi)
> -	movaps	%xmm5, %xmm1
> -
> -	movaps	%xmm2, -0x20(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -
> -	movaps	%xmm3, 0x10(%rdi)
> -	jb	L(shl_1_bwd_end)
> -	movaps	%xmm4, (%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_1_bwd_end):
> -	movaps	%xmm4, (%rdi)
> -	lea	64(%rdx), %rdx
> -	movdqu	%xmm0, (%r8)
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_2):
> -	lea	(L(shl_2_loop_L1)-L(shl_2))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x02(%rsi), %xmm1
> -	jb	L(L2_fwd)
> -	lea	(L(shl_2_loop_L2)-L(shl_2_loop_L1))(%r9), %r9
> -L(L2_fwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_2_loop_L2):
> -	prefetchnta 0x1c0(%rsi)
> -L(shl_2_loop_L1):
> -	sub	$64, %rdx
> -	movaps	0x0e(%rsi), %xmm2
> -	movaps	0x1e(%rsi), %xmm3
> -	movaps	0x2e(%rsi), %xmm4
> -	movaps	0x3e(%rsi), %xmm5
> -	movdqa	%xmm5, %xmm6
> -	palignr	$2, %xmm4, %xmm5
> -	lea	64(%rsi), %rsi
> -	palignr	$2, %xmm3, %xmm4
> -	palignr	$2, %xmm2, %xmm3
> -	lea	64(%rdi), %rdi
> -	palignr	$2, %xmm1, %xmm2
> -	movdqa	%xmm6, %xmm1
> -	movdqa	%xmm2, -0x40(%rdi)
> -	movaps	%xmm3, -0x30(%rdi)
> -	jb	L(shl_2_end)
> -	movaps	%xmm4, -0x20(%rdi)
> -	movaps	%xmm5, -0x10(%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_2_end):
> -	movaps	%xmm4, -0x20(%rdi)
> -	lea	64(%rdx), %rdx
> -	movaps	%xmm5, -0x10(%rdi)
> -	add	%rdx, %rdi
> -	movdqu	%xmm0, (%r8)
> -	add	%rdx, %rsi
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_2_bwd):
> -	lea	(L(shl_2_bwd_loop_L1)-L(shl_2_bwd))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x02(%rsi), %xmm1
> -	jb	L(L2_bwd)
> -	lea	(L(shl_2_bwd_loop_L2)-L(shl_2_bwd_loop_L1))(%r9), %r9
> -L(L2_bwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_2_bwd_loop_L2):
> -	prefetchnta -0x1c0(%rsi)
> -L(shl_2_bwd_loop_L1):
> -	movaps	-0x12(%rsi), %xmm2
> -	sub	$0x40, %rdx
> -	movaps	-0x22(%rsi), %xmm3
> -	movaps	-0x32(%rsi), %xmm4
> -	movaps	-0x42(%rsi), %xmm5
> -	lea	-0x40(%rsi), %rsi
> -	palignr	$2, %xmm2, %xmm1
> -	palignr	$2, %xmm3, %xmm2
> -	palignr	$2, %xmm4, %xmm3
> -	palignr	$2, %xmm5, %xmm4
> -
> -	movaps	%xmm1, -0x10(%rdi)
> -	movaps	%xmm5, %xmm1
> -
> -	movaps	%xmm2, -0x20(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -
> -	movaps	%xmm3, 0x10(%rdi)
> -	jb	L(shl_2_bwd_end)
> -	movaps	%xmm4, (%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_2_bwd_end):
> -	movaps	%xmm4, (%rdi)
> -	lea	64(%rdx), %rdx
> -	movdqu	%xmm0, (%r8)
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_3):
> -	lea	(L(shl_3_loop_L1)-L(shl_3))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x03(%rsi), %xmm1
> -	jb	L(L3_fwd)
> -	lea	(L(shl_3_loop_L2)-L(shl_3_loop_L1))(%r9), %r9
> -L(L3_fwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_3_loop_L2):
> -	prefetchnta 0x1c0(%rsi)
> -L(shl_3_loop_L1):
> -	sub	$64, %rdx
> -	movaps	0x0d(%rsi), %xmm2
> -	movaps	0x1d(%rsi), %xmm3
> -	movaps	0x2d(%rsi), %xmm4
> -	movaps	0x3d(%rsi), %xmm5
> -	movdqa	%xmm5, %xmm6
> -	palignr	$3, %xmm4, %xmm5
> -	lea	64(%rsi), %rsi
> -	palignr	$3, %xmm3, %xmm4
> -	palignr	$3, %xmm2, %xmm3
> -	lea	64(%rdi), %rdi
> -	palignr	$3, %xmm1, %xmm2
> -	movdqa	%xmm6, %xmm1
> -	movdqa	%xmm2, -0x40(%rdi)
> -	movaps	%xmm3, -0x30(%rdi)
> -	jb	L(shl_3_end)
> -	movaps	%xmm4, -0x20(%rdi)
> -	movaps	%xmm5, -0x10(%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_3_end):
> -	movaps	%xmm4, -0x20(%rdi)
> -	lea	64(%rdx), %rdx
> -	movaps	%xmm5, -0x10(%rdi)
> -	add	%rdx, %rdi
> -	movdqu	%xmm0, (%r8)
> -	add	%rdx, %rsi
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_3_bwd):
> -	lea	(L(shl_3_bwd_loop_L1)-L(shl_3_bwd))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x03(%rsi), %xmm1
> -	jb	L(L3_bwd)
> -	lea	(L(shl_3_bwd_loop_L2)-L(shl_3_bwd_loop_L1))(%r9), %r9
> -L(L3_bwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_3_bwd_loop_L2):
> -	prefetchnta -0x1c0(%rsi)
> -L(shl_3_bwd_loop_L1):
> -	movaps	-0x13(%rsi), %xmm2
> -	sub	$0x40, %rdx
> -	movaps	-0x23(%rsi), %xmm3
> -	movaps	-0x33(%rsi), %xmm4
> -	movaps	-0x43(%rsi), %xmm5
> -	lea	-0x40(%rsi), %rsi
> -	palignr	$3, %xmm2, %xmm1
> -	palignr	$3, %xmm3, %xmm2
> -	palignr	$3, %xmm4, %xmm3
> -	palignr	$3, %xmm5, %xmm4
> -
> -	movaps	%xmm1, -0x10(%rdi)
> -	movaps	%xmm5, %xmm1
> -
> -	movaps	%xmm2, -0x20(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -
> -	movaps	%xmm3, 0x10(%rdi)
> -	jb	L(shl_3_bwd_end)
> -	movaps	%xmm4, (%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_3_bwd_end):
> -	movaps	%xmm4, (%rdi)
> -	lea	64(%rdx), %rdx
> -	movdqu	%xmm0, (%r8)
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_4):
> -	lea	(L(shl_4_loop_L1)-L(shl_4))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x04(%rsi), %xmm1
> -	jb	L(L4_fwd)
> -	lea	(L(shl_4_loop_L2)-L(shl_4_loop_L1))(%r9), %r9
> -L(L4_fwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_4_loop_L2):
> -	prefetchnta 0x1c0(%rsi)
> -L(shl_4_loop_L1):
> -	sub	$64, %rdx
> -	movaps	0x0c(%rsi), %xmm2
> -	movaps	0x1c(%rsi), %xmm3
> -	movaps	0x2c(%rsi), %xmm4
> -	movaps	0x3c(%rsi), %xmm5
> -	movdqa	%xmm5, %xmm6
> -	palignr	$4, %xmm4, %xmm5
> -	lea	64(%rsi), %rsi
> -	palignr	$4, %xmm3, %xmm4
> -	palignr	$4, %xmm2, %xmm3
> -	lea	64(%rdi), %rdi
> -	palignr	$4, %xmm1, %xmm2
> -	movdqa	%xmm6, %xmm1
> -	movdqa	%xmm2, -0x40(%rdi)
> -	movaps	%xmm3, -0x30(%rdi)
> -	jb	L(shl_4_end)
> -	movaps	%xmm4, -0x20(%rdi)
> -	movaps	%xmm5, -0x10(%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_4_end):
> -	movaps	%xmm4, -0x20(%rdi)
> -	lea	64(%rdx), %rdx
> -	movaps	%xmm5, -0x10(%rdi)
> -	add	%rdx, %rdi
> -	movdqu	%xmm0, (%r8)
> -	add	%rdx, %rsi
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_4_bwd):
> -	lea	(L(shl_4_bwd_loop_L1)-L(shl_4_bwd))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x04(%rsi), %xmm1
> -	jb	L(L4_bwd)
> -	lea	(L(shl_4_bwd_loop_L2)-L(shl_4_bwd_loop_L1))(%r9), %r9
> -L(L4_bwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_4_bwd_loop_L2):
> -	prefetchnta -0x1c0(%rsi)
> -L(shl_4_bwd_loop_L1):
> -	movaps	-0x14(%rsi), %xmm2
> -	sub	$0x40, %rdx
> -	movaps	-0x24(%rsi), %xmm3
> -	movaps	-0x34(%rsi), %xmm4
> -	movaps	-0x44(%rsi), %xmm5
> -	lea	-0x40(%rsi), %rsi
> -	palignr	$4, %xmm2, %xmm1
> -	palignr	$4, %xmm3, %xmm2
> -	palignr	$4, %xmm4, %xmm3
> -	palignr	$4, %xmm5, %xmm4
> -
> -	movaps	%xmm1, -0x10(%rdi)
> -	movaps	%xmm5, %xmm1
> -
> -	movaps	%xmm2, -0x20(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -
> -	movaps	%xmm3, 0x10(%rdi)
> -	jb	L(shl_4_bwd_end)
> -	movaps	%xmm4, (%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_4_bwd_end):
> -	movaps	%xmm4, (%rdi)
> -	lea	64(%rdx), %rdx
> -	movdqu	%xmm0, (%r8)
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_5):
> -	lea	(L(shl_5_loop_L1)-L(shl_5))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x05(%rsi), %xmm1
> -	jb	L(L5_fwd)
> -	lea	(L(shl_5_loop_L2)-L(shl_5_loop_L1))(%r9), %r9
> -L(L5_fwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_5_loop_L2):
> -	prefetchnta 0x1c0(%rsi)
> -L(shl_5_loop_L1):
> -	sub	$64, %rdx
> -	movaps	0x0b(%rsi), %xmm2
> -	movaps	0x1b(%rsi), %xmm3
> -	movaps	0x2b(%rsi), %xmm4
> -	movaps	0x3b(%rsi), %xmm5
> -	movdqa	%xmm5, %xmm6
> -	palignr	$5, %xmm4, %xmm5
> -	lea	64(%rsi), %rsi
> -	palignr	$5, %xmm3, %xmm4
> -	palignr	$5, %xmm2, %xmm3
> -	lea	64(%rdi), %rdi
> -	palignr	$5, %xmm1, %xmm2
> -	movdqa	%xmm6, %xmm1
> -	movdqa	%xmm2, -0x40(%rdi)
> -	movaps	%xmm3, -0x30(%rdi)
> -	jb	L(shl_5_end)
> -	movaps	%xmm4, -0x20(%rdi)
> -	movaps	%xmm5, -0x10(%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_5_end):
> -	movaps	%xmm4, -0x20(%rdi)
> -	lea	64(%rdx), %rdx
> -	movaps	%xmm5, -0x10(%rdi)
> -	add	%rdx, %rdi
> -	movdqu	%xmm0, (%r8)
> -	add	%rdx, %rsi
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_5_bwd):
> -	lea	(L(shl_5_bwd_loop_L1)-L(shl_5_bwd))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x05(%rsi), %xmm1
> -	jb	L(L5_bwd)
> -	lea	(L(shl_5_bwd_loop_L2)-L(shl_5_bwd_loop_L1))(%r9), %r9
> -L(L5_bwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_5_bwd_loop_L2):
> -	prefetchnta -0x1c0(%rsi)
> -L(shl_5_bwd_loop_L1):
> -	movaps	-0x15(%rsi), %xmm2
> -	sub	$0x40, %rdx
> -	movaps	-0x25(%rsi), %xmm3
> -	movaps	-0x35(%rsi), %xmm4
> -	movaps	-0x45(%rsi), %xmm5
> -	lea	-0x40(%rsi), %rsi
> -	palignr	$5, %xmm2, %xmm1
> -	palignr	$5, %xmm3, %xmm2
> -	palignr	$5, %xmm4, %xmm3
> -	palignr	$5, %xmm5, %xmm4
> -
> -	movaps	%xmm1, -0x10(%rdi)
> -	movaps	%xmm5, %xmm1
> -
> -	movaps	%xmm2, -0x20(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -
> -	movaps	%xmm3, 0x10(%rdi)
> -	jb	L(shl_5_bwd_end)
> -	movaps	%xmm4, (%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_5_bwd_end):
> -	movaps	%xmm4, (%rdi)
> -	lea	64(%rdx), %rdx
> -	movdqu	%xmm0, (%r8)
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_6):
> -	lea	(L(shl_6_loop_L1)-L(shl_6))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x06(%rsi), %xmm1
> -	jb	L(L6_fwd)
> -	lea	(L(shl_6_loop_L2)-L(shl_6_loop_L1))(%r9), %r9
> -L(L6_fwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_6_loop_L2):
> -	prefetchnta 0x1c0(%rsi)
> -L(shl_6_loop_L1):
> -	sub	$64, %rdx
> -	movaps	0x0a(%rsi), %xmm2
> -	movaps	0x1a(%rsi), %xmm3
> -	movaps	0x2a(%rsi), %xmm4
> -	movaps	0x3a(%rsi), %xmm5
> -	movdqa	%xmm5, %xmm6
> -	palignr	$6, %xmm4, %xmm5
> -	lea	64(%rsi), %rsi
> -	palignr	$6, %xmm3, %xmm4
> -	palignr	$6, %xmm2, %xmm3
> -	lea	64(%rdi), %rdi
> -	palignr	$6, %xmm1, %xmm2
> -	movdqa	%xmm6, %xmm1
> -	movdqa	%xmm2, -0x40(%rdi)
> -	movaps	%xmm3, -0x30(%rdi)
> -	jb	L(shl_6_end)
> -	movaps	%xmm4, -0x20(%rdi)
> -	movaps	%xmm5, -0x10(%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_6_end):
> -	movaps	%xmm4, -0x20(%rdi)
> -	lea	64(%rdx), %rdx
> -	movaps	%xmm5, -0x10(%rdi)
> -	add	%rdx, %rdi
> -	movdqu	%xmm0, (%r8)
> -	add	%rdx, %rsi
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_6_bwd):
> -	lea	(L(shl_6_bwd_loop_L1)-L(shl_6_bwd))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x06(%rsi), %xmm1
> -	jb	L(L6_bwd)
> -	lea	(L(shl_6_bwd_loop_L2)-L(shl_6_bwd_loop_L1))(%r9), %r9
> -L(L6_bwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_6_bwd_loop_L2):
> -	prefetchnta -0x1c0(%rsi)
> -L(shl_6_bwd_loop_L1):
> -	movaps	-0x16(%rsi), %xmm2
> -	sub	$0x40, %rdx
> -	movaps	-0x26(%rsi), %xmm3
> -	movaps	-0x36(%rsi), %xmm4
> -	movaps	-0x46(%rsi), %xmm5
> -	lea	-0x40(%rsi), %rsi
> -	palignr	$6, %xmm2, %xmm1
> -	palignr	$6, %xmm3, %xmm2
> -	palignr	$6, %xmm4, %xmm3
> -	palignr	$6, %xmm5, %xmm4
> -
> -	movaps	%xmm1, -0x10(%rdi)
> -	movaps	%xmm5, %xmm1
> -
> -	movaps	%xmm2, -0x20(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -
> -	movaps	%xmm3, 0x10(%rdi)
> -	jb	L(shl_6_bwd_end)
> -	movaps	%xmm4, (%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_6_bwd_end):
> -	movaps	%xmm4, (%rdi)
> -	lea	64(%rdx), %rdx
> -	movdqu	%xmm0, (%r8)
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_7):
> -	lea	(L(shl_7_loop_L1)-L(shl_7))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x07(%rsi), %xmm1
> -	jb	L(L7_fwd)
> -	lea	(L(shl_7_loop_L2)-L(shl_7_loop_L1))(%r9), %r9
> -L(L7_fwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_7_loop_L2):
> -	prefetchnta 0x1c0(%rsi)
> -L(shl_7_loop_L1):
> -	sub	$64, %rdx
> -	movaps	0x09(%rsi), %xmm2
> -	movaps	0x19(%rsi), %xmm3
> -	movaps	0x29(%rsi), %xmm4
> -	movaps	0x39(%rsi), %xmm5
> -	movdqa	%xmm5, %xmm6
> -	palignr	$7, %xmm4, %xmm5
> -	lea	64(%rsi), %rsi
> -	palignr	$7, %xmm3, %xmm4
> -	palignr	$7, %xmm2, %xmm3
> -	lea	64(%rdi), %rdi
> -	palignr	$7, %xmm1, %xmm2
> -	movdqa	%xmm6, %xmm1
> -	movdqa	%xmm2, -0x40(%rdi)
> -	movaps	%xmm3, -0x30(%rdi)
> -	jb	L(shl_7_end)
> -	movaps	%xmm4, -0x20(%rdi)
> -	movaps	%xmm5, -0x10(%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_7_end):
> -	movaps	%xmm4, -0x20(%rdi)
> -	lea	64(%rdx), %rdx
> -	movaps	%xmm5, -0x10(%rdi)
> -	add	%rdx, %rdi
> -	movdqu	%xmm0, (%r8)
> -	add	%rdx, %rsi
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_7_bwd):
> -	lea	(L(shl_7_bwd_loop_L1)-L(shl_7_bwd))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x07(%rsi), %xmm1
> -	jb	L(L7_bwd)
> -	lea	(L(shl_7_bwd_loop_L2)-L(shl_7_bwd_loop_L1))(%r9), %r9
> -L(L7_bwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_7_bwd_loop_L2):
> -	prefetchnta -0x1c0(%rsi)
> -L(shl_7_bwd_loop_L1):
> -	movaps	-0x17(%rsi), %xmm2
> -	sub	$0x40, %rdx
> -	movaps	-0x27(%rsi), %xmm3
> -	movaps	-0x37(%rsi), %xmm4
> -	movaps	-0x47(%rsi), %xmm5
> -	lea	-0x40(%rsi), %rsi
> -	palignr	$7, %xmm2, %xmm1
> -	palignr	$7, %xmm3, %xmm2
> -	palignr	$7, %xmm4, %xmm3
> -	palignr	$7, %xmm5, %xmm4
> -
> -	movaps	%xmm1, -0x10(%rdi)
> -	movaps	%xmm5, %xmm1
> -
> -	movaps	%xmm2, -0x20(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -
> -	movaps	%xmm3, 0x10(%rdi)
> -	jb	L(shl_7_bwd_end)
> -	movaps	%xmm4, (%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_7_bwd_end):
> -	movaps	%xmm4, (%rdi)
> -	lea	64(%rdx), %rdx
> -	movdqu	%xmm0, (%r8)
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_8):
> -	lea	(L(shl_8_loop_L1)-L(shl_8))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x08(%rsi), %xmm1
> -	jb	L(L8_fwd)
> -	lea	(L(shl_8_loop_L2)-L(shl_8_loop_L1))(%r9), %r9
> -L(L8_fwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -L(shl_8_loop_L2):
> -	prefetchnta 0x1c0(%rsi)
> -L(shl_8_loop_L1):
> -	sub	$64, %rdx
> -	movaps	0x08(%rsi), %xmm2
> -	movaps	0x18(%rsi), %xmm3
> -	movaps	0x28(%rsi), %xmm4
> -	movaps	0x38(%rsi), %xmm5
> -	movdqa	%xmm5, %xmm6
> -	palignr	$8, %xmm4, %xmm5
> -	lea	64(%rsi), %rsi
> -	palignr	$8, %xmm3, %xmm4
> -	palignr	$8, %xmm2, %xmm3
> -	lea	64(%rdi), %rdi
> -	palignr	$8, %xmm1, %xmm2
> -	movdqa	%xmm6, %xmm1
> -	movdqa	%xmm2, -0x40(%rdi)
> -	movaps	%xmm3, -0x30(%rdi)
> -	jb	L(shl_8_end)
> -	movaps	%xmm4, -0x20(%rdi)
> -	movaps	%xmm5, -0x10(%rdi)
> -	jmp	*%r9
> -	ud2
> -	ALIGN (4)
> -L(shl_8_end):
> -	lea	64(%rdx), %rdx
> -	movaps	%xmm4, -0x20(%rdi)
> -	add	%rdx, %rsi
> -	movaps	%xmm5, -0x10(%rdi)
> -	add	%rdx, %rdi
> -	movdqu	%xmm0, (%r8)
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_8_bwd):
> -	lea	(L(shl_8_bwd_loop_L1)-L(shl_8_bwd))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x08(%rsi), %xmm1
> -	jb	L(L8_bwd)
> -	lea	(L(shl_8_bwd_loop_L2)-L(shl_8_bwd_loop_L1))(%r9), %r9
> -L(L8_bwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_8_bwd_loop_L2):
> -	prefetchnta -0x1c0(%rsi)
> -L(shl_8_bwd_loop_L1):
> -	movaps	-0x18(%rsi), %xmm2
> -	sub	$0x40, %rdx
> -	movaps	-0x28(%rsi), %xmm3
> -	movaps	-0x38(%rsi), %xmm4
> -	movaps	-0x48(%rsi), %xmm5
> -	lea	-0x40(%rsi), %rsi
> -	palignr	$8, %xmm2, %xmm1
> -	palignr	$8, %xmm3, %xmm2
> -	palignr	$8, %xmm4, %xmm3
> -	palignr	$8, %xmm5, %xmm4
> -
> -	movaps	%xmm1, -0x10(%rdi)
> -	movaps	%xmm5, %xmm1
> -
> -	movaps	%xmm2, -0x20(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -
> -	movaps	%xmm3, 0x10(%rdi)
> -	jb	L(shl_8_bwd_end)
> -	movaps	%xmm4, (%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_8_bwd_end):
> -	movaps	%xmm4, (%rdi)
> -	lea	64(%rdx), %rdx
> -	movdqu	%xmm0, (%r8)
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_9):
> -	lea	(L(shl_9_loop_L1)-L(shl_9))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x09(%rsi), %xmm1
> -	jb	L(L9_fwd)
> -	lea	(L(shl_9_loop_L2)-L(shl_9_loop_L1))(%r9), %r9
> -L(L9_fwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_9_loop_L2):
> -	prefetchnta 0x1c0(%rsi)
> -L(shl_9_loop_L1):
> -	sub	$64, %rdx
> -	movaps	0x07(%rsi), %xmm2
> -	movaps	0x17(%rsi), %xmm3
> -	movaps	0x27(%rsi), %xmm4
> -	movaps	0x37(%rsi), %xmm5
> -	movdqa	%xmm5, %xmm6
> -	palignr	$9, %xmm4, %xmm5
> -	lea	64(%rsi), %rsi
> -	palignr	$9, %xmm3, %xmm4
> -	palignr	$9, %xmm2, %xmm3
> -	lea	64(%rdi), %rdi
> -	palignr	$9, %xmm1, %xmm2
> -	movdqa	%xmm6, %xmm1
> -	movdqa	%xmm2, -0x40(%rdi)
> -	movaps	%xmm3, -0x30(%rdi)
> -	jb	L(shl_9_end)
> -	movaps	%xmm4, -0x20(%rdi)
> -	movaps	%xmm5, -0x10(%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_9_end):
> -	movaps	%xmm4, -0x20(%rdi)
> -	lea	64(%rdx), %rdx
> -	movaps	%xmm5, -0x10(%rdi)
> -	add	%rdx, %rdi
> -	movdqu	%xmm0, (%r8)
> -	add	%rdx, %rsi
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_9_bwd):
> -	lea	(L(shl_9_bwd_loop_L1)-L(shl_9_bwd))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x09(%rsi), %xmm1
> -	jb	L(L9_bwd)
> -	lea	(L(shl_9_bwd_loop_L2)-L(shl_9_bwd_loop_L1))(%r9), %r9
> -L(L9_bwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_9_bwd_loop_L2):
> -	prefetchnta -0x1c0(%rsi)
> -L(shl_9_bwd_loop_L1):
> -	movaps	-0x19(%rsi), %xmm2
> -	sub	$0x40, %rdx
> -	movaps	-0x29(%rsi), %xmm3
> -	movaps	-0x39(%rsi), %xmm4
> -	movaps	-0x49(%rsi), %xmm5
> -	lea	-0x40(%rsi), %rsi
> -	palignr	$9, %xmm2, %xmm1
> -	palignr	$9, %xmm3, %xmm2
> -	palignr	$9, %xmm4, %xmm3
> -	palignr	$9, %xmm5, %xmm4
> -
> -	movaps	%xmm1, -0x10(%rdi)
> -	movaps	%xmm5, %xmm1
> -
> -	movaps	%xmm2, -0x20(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -
> -	movaps	%xmm3, 0x10(%rdi)
> -	jb	L(shl_9_bwd_end)
> -	movaps	%xmm4, (%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_9_bwd_end):
> -	movaps	%xmm4, (%rdi)
> -	lea	64(%rdx), %rdx
> -	movdqu	%xmm0, (%r8)
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_10):
> -	lea	(L(shl_10_loop_L1)-L(shl_10))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x0a(%rsi), %xmm1
> -	jb	L(L10_fwd)
> -	lea	(L(shl_10_loop_L2)-L(shl_10_loop_L1))(%r9), %r9
> -L(L10_fwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_10_loop_L2):
> -	prefetchnta 0x1c0(%rsi)
> -L(shl_10_loop_L1):
> -	sub	$64, %rdx
> -	movaps	0x06(%rsi), %xmm2
> -	movaps	0x16(%rsi), %xmm3
> -	movaps	0x26(%rsi), %xmm4
> -	movaps	0x36(%rsi), %xmm5
> -	movdqa	%xmm5, %xmm6
> -	palignr	$10, %xmm4, %xmm5
> -	lea	64(%rsi), %rsi
> -	palignr	$10, %xmm3, %xmm4
> -	palignr	$10, %xmm2, %xmm3
> -	lea	64(%rdi), %rdi
> -	palignr	$10, %xmm1, %xmm2
> -	movdqa	%xmm6, %xmm1
> -	movdqa	%xmm2, -0x40(%rdi)
> -	movaps	%xmm3, -0x30(%rdi)
> -	jb	L(shl_10_end)
> -	movaps	%xmm4, -0x20(%rdi)
> -	movaps	%xmm5, -0x10(%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_10_end):
> -	movaps	%xmm4, -0x20(%rdi)
> -	lea	64(%rdx), %rdx
> -	movaps	%xmm5, -0x10(%rdi)
> -	add	%rdx, %rdi
> -	movdqu	%xmm0, (%r8)
> -	add	%rdx, %rsi
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_10_bwd):
> -	lea	(L(shl_10_bwd_loop_L1)-L(shl_10_bwd))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x0a(%rsi), %xmm1
> -	jb	L(L10_bwd)
> -	lea	(L(shl_10_bwd_loop_L2)-L(shl_10_bwd_loop_L1))(%r9), %r9
> -L(L10_bwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_10_bwd_loop_L2):
> -	prefetchnta -0x1c0(%rsi)
> -L(shl_10_bwd_loop_L1):
> -	movaps	-0x1a(%rsi), %xmm2
> -	sub	$0x40, %rdx
> -	movaps	-0x2a(%rsi), %xmm3
> -	movaps	-0x3a(%rsi), %xmm4
> -	movaps	-0x4a(%rsi), %xmm5
> -	lea	-0x40(%rsi), %rsi
> -	palignr	$10, %xmm2, %xmm1
> -	palignr	$10, %xmm3, %xmm2
> -	palignr	$10, %xmm4, %xmm3
> -	palignr	$10, %xmm5, %xmm4
> -
> -	movaps	%xmm1, -0x10(%rdi)
> -	movaps	%xmm5, %xmm1
> -
> -	movaps	%xmm2, -0x20(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -
> -	movaps	%xmm3, 0x10(%rdi)
> -	jb	L(shl_10_bwd_end)
> -	movaps	%xmm4, (%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_10_bwd_end):
> -	movaps	%xmm4, (%rdi)
> -	lea	64(%rdx), %rdx
> -	movdqu	%xmm0, (%r8)
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_11):
> -	lea	(L(shl_11_loop_L1)-L(shl_11))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x0b(%rsi), %xmm1
> -	jb	L(L11_fwd)
> -	lea	(L(shl_11_loop_L2)-L(shl_11_loop_L1))(%r9), %r9
> -L(L11_fwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_11_loop_L2):
> -	prefetchnta 0x1c0(%rsi)
> -L(shl_11_loop_L1):
> -	sub	$64, %rdx
> -	movaps	0x05(%rsi), %xmm2
> -	movaps	0x15(%rsi), %xmm3
> -	movaps	0x25(%rsi), %xmm4
> -	movaps	0x35(%rsi), %xmm5
> -	movdqa	%xmm5, %xmm6
> -	palignr	$11, %xmm4, %xmm5
> -	lea	64(%rsi), %rsi
> -	palignr	$11, %xmm3, %xmm4
> -	palignr	$11, %xmm2, %xmm3
> -	lea	64(%rdi), %rdi
> -	palignr	$11, %xmm1, %xmm2
> -	movdqa	%xmm6, %xmm1
> -	movdqa	%xmm2, -0x40(%rdi)
> -	movaps	%xmm3, -0x30(%rdi)
> -	jb	L(shl_11_end)
> -	movaps	%xmm4, -0x20(%rdi)
> -	movaps	%xmm5, -0x10(%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_11_end):
> -	movaps	%xmm4, -0x20(%rdi)
> -	lea	64(%rdx), %rdx
> -	movaps	%xmm5, -0x10(%rdi)
> -	add	%rdx, %rdi
> -	movdqu	%xmm0, (%r8)
> -	add	%rdx, %rsi
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_11_bwd):
> -	lea	(L(shl_11_bwd_loop_L1)-L(shl_11_bwd))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x0b(%rsi), %xmm1
> -	jb	L(L11_bwd)
> -	lea	(L(shl_11_bwd_loop_L2)-L(shl_11_bwd_loop_L1))(%r9), %r9
> -L(L11_bwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_11_bwd_loop_L2):
> -	prefetchnta -0x1c0(%rsi)
> -L(shl_11_bwd_loop_L1):
> -	movaps	-0x1b(%rsi), %xmm2
> -	sub	$0x40, %rdx
> -	movaps	-0x2b(%rsi), %xmm3
> -	movaps	-0x3b(%rsi), %xmm4
> -	movaps	-0x4b(%rsi), %xmm5
> -	lea	-0x40(%rsi), %rsi
> -	palignr	$11, %xmm2, %xmm1
> -	palignr	$11, %xmm3, %xmm2
> -	palignr	$11, %xmm4, %xmm3
> -	palignr	$11, %xmm5, %xmm4
> -
> -	movaps	%xmm1, -0x10(%rdi)
> -	movaps	%xmm5, %xmm1
> -
> -	movaps	%xmm2, -0x20(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -
> -	movaps	%xmm3, 0x10(%rdi)
> -	jb	L(shl_11_bwd_end)
> -	movaps	%xmm4, (%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_11_bwd_end):
> -	movaps	%xmm4, (%rdi)
> -	lea	64(%rdx), %rdx
> -	movdqu	%xmm0, (%r8)
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_12):
> -	lea	(L(shl_12_loop_L1)-L(shl_12))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x0c(%rsi), %xmm1
> -	jb	L(L12_fwd)
> -	lea	(L(shl_12_loop_L2)-L(shl_12_loop_L1))(%r9), %r9
> -L(L12_fwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_12_loop_L2):
> -	prefetchnta 0x1c0(%rsi)
> -L(shl_12_loop_L1):
> -	sub	$64, %rdx
> -	movaps	0x04(%rsi), %xmm2
> -	movaps	0x14(%rsi), %xmm3
> -	movaps	0x24(%rsi), %xmm4
> -	movaps	0x34(%rsi), %xmm5
> -	movdqa	%xmm5, %xmm6
> -	palignr	$12, %xmm4, %xmm5
> -	lea	64(%rsi), %rsi
> -	palignr	$12, %xmm3, %xmm4
> -	palignr	$12, %xmm2, %xmm3
> -	lea	64(%rdi), %rdi
> -	palignr	$12, %xmm1, %xmm2
> -	movdqa	%xmm6, %xmm1
> -	movdqa	%xmm2, -0x40(%rdi)
> -	movaps	%xmm3, -0x30(%rdi)
> -	jb	L(shl_12_end)
> -	movaps	%xmm4, -0x20(%rdi)
> -	movaps	%xmm5, -0x10(%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_12_end):
> -	movaps	%xmm4, -0x20(%rdi)
> -	lea	64(%rdx), %rdx
> -	movaps	%xmm5, -0x10(%rdi)
> -	add	%rdx, %rdi
> -	movdqu	%xmm0, (%r8)
> -	add	%rdx, %rsi
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_12_bwd):
> -	lea	(L(shl_12_bwd_loop_L1)-L(shl_12_bwd))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x0c(%rsi), %xmm1
> -	jb	L(L12_bwd)
> -	lea	(L(shl_12_bwd_loop_L2)-L(shl_12_bwd_loop_L1))(%r9), %r9
> -L(L12_bwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_12_bwd_loop_L2):
> -	prefetchnta -0x1c0(%rsi)
> -L(shl_12_bwd_loop_L1):
> -	movaps	-0x1c(%rsi), %xmm2
> -	sub	$0x40, %rdx
> -	movaps	-0x2c(%rsi), %xmm3
> -	movaps	-0x3c(%rsi), %xmm4
> -	movaps	-0x4c(%rsi), %xmm5
> -	lea	-0x40(%rsi), %rsi
> -	palignr	$12, %xmm2, %xmm1
> -	palignr	$12, %xmm3, %xmm2
> -	palignr	$12, %xmm4, %xmm3
> -	palignr	$12, %xmm5, %xmm4
> -
> -	movaps	%xmm1, -0x10(%rdi)
> -	movaps	%xmm5, %xmm1
> -
> -	movaps	%xmm2, -0x20(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -
> -	movaps	%xmm3, 0x10(%rdi)
> -	jb	L(shl_12_bwd_end)
> -	movaps	%xmm4, (%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_12_bwd_end):
> -	movaps	%xmm4, (%rdi)
> -	lea	64(%rdx), %rdx
> -	movdqu	%xmm0, (%r8)
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_13):
> -	lea	(L(shl_13_loop_L1)-L(shl_13))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x0d(%rsi), %xmm1
> -	jb	L(L13_fwd)
> -	lea	(L(shl_13_loop_L2)-L(shl_13_loop_L1))(%r9), %r9
> -L(L13_fwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_13_loop_L2):
> -	prefetchnta 0x1c0(%rsi)
> -L(shl_13_loop_L1):
> -	sub	$64, %rdx
> -	movaps	0x03(%rsi), %xmm2
> -	movaps	0x13(%rsi), %xmm3
> -	movaps	0x23(%rsi), %xmm4
> -	movaps	0x33(%rsi), %xmm5
> -	movdqa	%xmm5, %xmm6
> -	palignr	$13, %xmm4, %xmm5
> -	lea	64(%rsi), %rsi
> -	palignr	$13, %xmm3, %xmm4
> -	palignr	$13, %xmm2, %xmm3
> -	lea	64(%rdi), %rdi
> -	palignr	$13, %xmm1, %xmm2
> -	movdqa	%xmm6, %xmm1
> -	movdqa	%xmm2, -0x40(%rdi)
> -	movaps	%xmm3, -0x30(%rdi)
> -	jb	L(shl_13_end)
> -	movaps	%xmm4, -0x20(%rdi)
> -	movaps	%xmm5, -0x10(%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_13_end):
> -	movaps	%xmm4, -0x20(%rdi)
> -	lea	64(%rdx), %rdx
> -	movaps	%xmm5, -0x10(%rdi)
> -	add	%rdx, %rdi
> -	movdqu	%xmm0, (%r8)
> -	add	%rdx, %rsi
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_13_bwd):
> -	lea	(L(shl_13_bwd_loop_L1)-L(shl_13_bwd))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x0d(%rsi), %xmm1
> -	jb	L(L13_bwd)
> -	lea	(L(shl_13_bwd_loop_L2)-L(shl_13_bwd_loop_L1))(%r9), %r9
> -L(L13_bwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_13_bwd_loop_L2):
> -	prefetchnta -0x1c0(%rsi)
> -L(shl_13_bwd_loop_L1):
> -	movaps	-0x1d(%rsi), %xmm2
> -	sub	$0x40, %rdx
> -	movaps	-0x2d(%rsi), %xmm3
> -	movaps	-0x3d(%rsi), %xmm4
> -	movaps	-0x4d(%rsi), %xmm5
> -	lea	-0x40(%rsi), %rsi
> -	palignr	$13, %xmm2, %xmm1
> -	palignr	$13, %xmm3, %xmm2
> -	palignr	$13, %xmm4, %xmm3
> -	palignr	$13, %xmm5, %xmm4
> -
> -	movaps	%xmm1, -0x10(%rdi)
> -	movaps	%xmm5, %xmm1
> -
> -	movaps	%xmm2, -0x20(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -
> -	movaps	%xmm3, 0x10(%rdi)
> -	jb	L(shl_13_bwd_end)
> -	movaps	%xmm4, (%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_13_bwd_end):
> -	movaps	%xmm4, (%rdi)
> -	lea	64(%rdx), %rdx
> -	movdqu	%xmm0, (%r8)
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_14):
> -	lea	(L(shl_14_loop_L1)-L(shl_14))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x0e(%rsi), %xmm1
> -	jb	L(L14_fwd)
> -	lea	(L(shl_14_loop_L2)-L(shl_14_loop_L1))(%r9), %r9
> -L(L14_fwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_14_loop_L2):
> -	prefetchnta 0x1c0(%rsi)
> -L(shl_14_loop_L1):
> -	sub	$64, %rdx
> -	movaps	0x02(%rsi), %xmm2
> -	movaps	0x12(%rsi), %xmm3
> -	movaps	0x22(%rsi), %xmm4
> -	movaps	0x32(%rsi), %xmm5
> -	movdqa	%xmm5, %xmm6
> -	palignr	$14, %xmm4, %xmm5
> -	lea	64(%rsi), %rsi
> -	palignr	$14, %xmm3, %xmm4
> -	palignr	$14, %xmm2, %xmm3
> -	lea	64(%rdi), %rdi
> -	palignr	$14, %xmm1, %xmm2
> -	movdqa	%xmm6, %xmm1
> -	movdqa	%xmm2, -0x40(%rdi)
> -	movaps	%xmm3, -0x30(%rdi)
> -	jb	L(shl_14_end)
> -	movaps	%xmm4, -0x20(%rdi)
> -	movaps	%xmm5, -0x10(%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_14_end):
> -	movaps	%xmm4, -0x20(%rdi)
> -	lea	64(%rdx), %rdx
> -	movaps	%xmm5, -0x10(%rdi)
> -	add	%rdx, %rdi
> -	movdqu	%xmm0, (%r8)
> -	add	%rdx, %rsi
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_14_bwd):
> -	lea	(L(shl_14_bwd_loop_L1)-L(shl_14_bwd))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x0e(%rsi), %xmm1
> -	jb	L(L14_bwd)
> -	lea	(L(shl_14_bwd_loop_L2)-L(shl_14_bwd_loop_L1))(%r9), %r9
> -L(L14_bwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_14_bwd_loop_L2):
> -	prefetchnta -0x1c0(%rsi)
> -L(shl_14_bwd_loop_L1):
> -	movaps	-0x1e(%rsi), %xmm2
> -	sub	$0x40, %rdx
> -	movaps	-0x2e(%rsi), %xmm3
> -	movaps	-0x3e(%rsi), %xmm4
> -	movaps	-0x4e(%rsi), %xmm5
> -	lea	-0x40(%rsi), %rsi
> -	palignr	$14, %xmm2, %xmm1
> -	palignr	$14, %xmm3, %xmm2
> -	palignr	$14, %xmm4, %xmm3
> -	palignr	$14, %xmm5, %xmm4
> -
> -	movaps	%xmm1, -0x10(%rdi)
> -	movaps	%xmm5, %xmm1
> -
> -	movaps	%xmm2, -0x20(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -
> -	movaps	%xmm3, 0x10(%rdi)
> -	jb	L(shl_14_bwd_end)
> -	movaps	%xmm4, (%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_14_bwd_end):
> -	movaps	%xmm4, (%rdi)
> -	lea	64(%rdx), %rdx
> -	movdqu	%xmm0, (%r8)
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_15):
> -	lea	(L(shl_15_loop_L1)-L(shl_15))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x0f(%rsi), %xmm1
> -	jb	L(L15_fwd)
> -	lea	(L(shl_15_loop_L2)-L(shl_15_loop_L1))(%r9), %r9
> -L(L15_fwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_15_loop_L2):
> -	prefetchnta 0x1c0(%rsi)
> -L(shl_15_loop_L1):
> -	sub	$64, %rdx
> -	movaps	0x01(%rsi), %xmm2
> -	movaps	0x11(%rsi), %xmm3
> -	movaps	0x21(%rsi), %xmm4
> -	movaps	0x31(%rsi), %xmm5
> -	movdqa	%xmm5, %xmm6
> -	palignr	$15, %xmm4, %xmm5
> -	lea	64(%rsi), %rsi
> -	palignr	$15, %xmm3, %xmm4
> -	palignr	$15, %xmm2, %xmm3
> -	lea	64(%rdi), %rdi
> -	palignr	$15, %xmm1, %xmm2
> -	movdqa	%xmm6, %xmm1
> -	movdqa	%xmm2, -0x40(%rdi)
> -	movaps	%xmm3, -0x30(%rdi)
> -	jb	L(shl_15_end)
> -	movaps	%xmm4, -0x20(%rdi)
> -	movaps	%xmm5, -0x10(%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_15_end):
> -	movaps	%xmm4, -0x20(%rdi)
> -	lea	64(%rdx), %rdx
> -	movaps	%xmm5, -0x10(%rdi)
> -	add	%rdx, %rdi
> -	movdqu	%xmm0, (%r8)
> -	add	%rdx, %rsi
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(shl_15_bwd):
> -	lea	(L(shl_15_bwd_loop_L1)-L(shl_15_bwd))(%r9), %r9
> -	cmp	%rcx, %rdx
> -	movaps	-0x0f(%rsi), %xmm1
> -	jb	L(L15_bwd)
> -	lea	(L(shl_15_bwd_loop_L2)-L(shl_15_bwd_loop_L1))(%r9), %r9
> -L(L15_bwd):
> -	lea	-64(%rdx), %rdx
> -	jmp	*%r9
> -	ud2
> -L(shl_15_bwd_loop_L2):
> -	prefetchnta -0x1c0(%rsi)
> -L(shl_15_bwd_loop_L1):
> -	movaps	-0x1f(%rsi), %xmm2
> -	sub	$0x40, %rdx
> -	movaps	-0x2f(%rsi), %xmm3
> -	movaps	-0x3f(%rsi), %xmm4
> -	movaps	-0x4f(%rsi), %xmm5
> -	lea	-0x40(%rsi), %rsi
> -	palignr	$15, %xmm2, %xmm1
> -	palignr	$15, %xmm3, %xmm2
> -	palignr	$15, %xmm4, %xmm3
> -	palignr	$15, %xmm5, %xmm4
> -
> -	movaps	%xmm1, -0x10(%rdi)
> -	movaps	%xmm5, %xmm1
> -
> -	movaps	%xmm2, -0x20(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -
> -	movaps	%xmm3, 0x10(%rdi)
> -	jb	L(shl_15_bwd_end)
> -	movaps	%xmm4, (%rdi)
> -	jmp	*%r9
> -	ud2
> -L(shl_15_bwd_end):
> -	movaps	%xmm4, (%rdi)
> -	lea	64(%rdx), %rdx
> -	movdqu	%xmm0, (%r8)
> -	BRANCH_TO_JMPTBL_ENTRY(L(table_less_80bytes), %rdx, 4)
> -
> -	ALIGN (4)
> -L(write_72bytes):
> -	movdqu	-72(%rsi), %xmm0
> -	movdqu	-56(%rsi), %xmm1
> -	mov	-40(%rsi), %r8
> -	mov	-32(%rsi), %r9
> -	mov	-24(%rsi), %r10
> -	mov	-16(%rsi), %r11
> -	mov	-8(%rsi), %rcx
> -	movdqu	 %xmm0, -72(%rdi)
> -	movdqu	 %xmm1, -56(%rdi)
> -	mov	 %r8, -40(%rdi)
> -	mov	 %r9, -32(%rdi)
> -	mov	 %r10, -24(%rdi)
> -	mov	 %r11, -16(%rdi)
> -	mov	 %rcx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_64bytes):
> -	movdqu	-64(%rsi), %xmm0
> -	mov	-48(%rsi), %rcx
> -	mov	-40(%rsi), %r8
> -	mov	-32(%rsi), %r9
> -	mov	-24(%rsi), %r10
> -	mov	-16(%rsi), %r11
> -	mov	-8(%rsi), %rdx
> -	movdqu	 %xmm0, -64(%rdi)
> -	mov	 %rcx, -48(%rdi)
> -	mov	 %r8, -40(%rdi)
> -	mov	 %r9, -32(%rdi)
> -	mov	 %r10, -24(%rdi)
> -	mov	 %r11, -16(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_56bytes):
> -	movdqu	-56(%rsi), %xmm0
> -	mov	-40(%rsi), %r8
> -	mov	-32(%rsi), %r9
> -	mov	-24(%rsi), %r10
> -	mov	-16(%rsi), %r11
> -	mov	-8(%rsi), %rcx
> -	movdqu	 %xmm0, -56(%rdi)
> -	mov	 %r8, -40(%rdi)
> -	mov	 %r9, -32(%rdi)
> -	mov	 %r10, -24(%rdi)
> -	mov	 %r11, -16(%rdi)
> -	mov	 %rcx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_48bytes):
> -	mov	-48(%rsi), %rcx
> -	mov	-40(%rsi), %r8
> -	mov	-32(%rsi), %r9
> -	mov	-24(%rsi), %r10
> -	mov	-16(%rsi), %r11
> -	mov	-8(%rsi), %rdx
> -	mov	 %rcx, -48(%rdi)
> -	mov	 %r8, -40(%rdi)
> -	mov	 %r9, -32(%rdi)
> -	mov	 %r10, -24(%rdi)
> -	mov	 %r11, -16(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_40bytes):
> -	mov	-40(%rsi), %r8
> -	mov	-32(%rsi), %r9
> -	mov	-24(%rsi), %r10
> -	mov	-16(%rsi), %r11
> -	mov	-8(%rsi), %rdx
> -	mov	 %r8, -40(%rdi)
> -	mov	 %r9, -32(%rdi)
> -	mov	 %r10, -24(%rdi)
> -	mov	 %r11, -16(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_32bytes):
> -	mov	-32(%rsi), %r9
> -	mov	-24(%rsi), %r10
> -	mov	-16(%rsi), %r11
> -	mov	-8(%rsi), %rdx
> -	mov	 %r9, -32(%rdi)
> -	mov	 %r10, -24(%rdi)
> -	mov	 %r11, -16(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_24bytes):
> -	mov	-24(%rsi), %r10
> -	mov	-16(%rsi), %r11
> -	mov	-8(%rsi), %rdx
> -	mov	 %r10, -24(%rdi)
> -	mov	 %r11, -16(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_16bytes):
> -	mov	-16(%rsi), %r11
> -	mov	-8(%rsi), %rdx
> -	mov	 %r11, -16(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_8bytes):
> -	mov	-8(%rsi), %rdx
> -	mov	 %rdx, -8(%rdi)
> -L(write_0bytes):
> -	ret
> -
> -	ALIGN (4)
> -L(write_73bytes):
> -	movdqu	-73(%rsi), %xmm0
> -	movdqu	-57(%rsi), %xmm1
> -	mov	-41(%rsi), %rcx
> -	mov	-33(%rsi), %r9
> -	mov	-25(%rsi), %r10
> -	mov	-17(%rsi), %r11
> -	mov	-9(%rsi), %r8
> -	mov	-4(%rsi), %edx
> -	movdqu	 %xmm0, -73(%rdi)
> -	movdqu	 %xmm1, -57(%rdi)
> -	mov	 %rcx, -41(%rdi)
> -	mov	 %r9, -33(%rdi)
> -	mov	 %r10, -25(%rdi)
> -	mov	 %r11, -17(%rdi)
> -	mov	 %r8, -9(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_65bytes):
> -	movdqu	-65(%rsi), %xmm0
> -	movdqu	-49(%rsi), %xmm1
> -	mov	-33(%rsi), %r9
> -	mov	-25(%rsi), %r10
> -	mov	-17(%rsi), %r11
> -	mov	-9(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	movdqu	 %xmm0, -65(%rdi)
> -	movdqu	 %xmm1, -49(%rdi)
> -	mov	 %r9, -33(%rdi)
> -	mov	 %r10, -25(%rdi)
> -	mov	 %r11, -17(%rdi)
> -	mov	 %rcx, -9(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_57bytes):
> -	movdqu	-57(%rsi), %xmm0
> -	mov	-41(%rsi), %r8
> -	mov	-33(%rsi), %r9
> -	mov	-25(%rsi), %r10
> -	mov	-17(%rsi), %r11
> -	mov	-9(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	movdqu	 %xmm0, -57(%rdi)
> -	mov	 %r8, -41(%rdi)
> -	mov	 %r9, -33(%rdi)
> -	mov	 %r10, -25(%rdi)
> -	mov	 %r11, -17(%rdi)
> -	mov	 %rcx, -9(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_49bytes):
> -	movdqu	-49(%rsi), %xmm0
> -	mov	-33(%rsi), %r9
> -	mov	-25(%rsi), %r10
> -	mov	-17(%rsi), %r11
> -	mov	-9(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	movdqu	 %xmm0, -49(%rdi)
> -	mov	 %r9, -33(%rdi)
> -	mov	 %r10, -25(%rdi)
> -	mov	 %r11, -17(%rdi)
> -	mov	 %rcx, -9(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_41bytes):
> -	mov	-41(%rsi), %r8
> -	mov	-33(%rsi), %r9
> -	mov	-25(%rsi), %r10
> -	mov	-17(%rsi), %r11
> -	mov	-9(%rsi), %rcx
> -	mov	-1(%rsi), %dl
> -	mov	 %r8, -41(%rdi)
> -	mov	 %r9, -33(%rdi)
> -	mov	 %r10, -25(%rdi)
> -	mov	 %r11, -17(%rdi)
> -	mov	 %rcx, -9(%rdi)
> -	mov	 %dl, -1(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_33bytes):
> -	mov	-33(%rsi), %r9
> -	mov	-25(%rsi), %r10
> -	mov	-17(%rsi), %r11
> -	mov	-9(%rsi), %rcx
> -	mov	-1(%rsi), %dl
> -	mov	 %r9, -33(%rdi)
> -	mov	 %r10, -25(%rdi)
> -	mov	 %r11, -17(%rdi)
> -	mov	 %rcx, -9(%rdi)
> -	mov	 %dl, -1(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_25bytes):
> -	mov	-25(%rsi), %r10
> -	mov	-17(%rsi), %r11
> -	mov	-9(%rsi), %rcx
> -	mov	-1(%rsi), %dl
> -	mov	 %r10, -25(%rdi)
> -	mov	 %r11, -17(%rdi)
> -	mov	 %rcx, -9(%rdi)
> -	mov	 %dl, -1(%rdi)
> +	movdqa	%xmm0, 16(%rcx)
> +	movdqa	16(%rsi), %xmm0
> +	movdqa	%xmm0, 32(%rcx)
> +	movdqa	32(%rsi), %xmm0
> +	addq	$64, %rsi
> +	movdqa	%xmm0, 48(%rcx)
> +	addq	$64, %rcx
> +	subq	$1, %rdx
> +	jnz .L100
> +L(end):
> +	rep
> +	ret
> +
> +
> +
> +	ALIGN (4)
> +L(less_16_bytes):
> +	cmp	$8, %edx
> +	jae	.L60
> +	cmp	$4, %edx
> +	jae	.L61
> +	cmp	$1, %edx
> +	jbe	.L41
> +	movzwl	-2(%rsi, %rdx), %ecx
> +	movzwl	(%rsi), %esi
> +	movw	%si, (%rdi)
> +	movw	%cx, -2(%rdi, %rdx)
> +	ret
> +
> +	ALIGN (4)
> +.L60:
> +	movq	-8(%rsi, %rdx), %rcx
> +	movq	(%rsi), %rsi
> +	movq	%rsi, (%rdi)
> +	movq	%rcx, -8(%rdi, %rdx)
> +	ret
> +
> +	ALIGN (4)
> +.L61:
> +	movl	-4(%rsi, %rdx), %ecx
> +	movl	(%rsi), %esi
> +	movl	%esi, (%rdi)
> +	movl	%ecx, -4(%rdi, %rdx)
> +	ret
> +
> +	ALIGN (4)
> +.L41:
> +	je	.L62
> +	rep
> +	ret
> +.L62:
> +	movzbl	(%rsi), %edx
> +	movb	%dl, (%rdi)
> +	ret
> +
> +L(fwd):
> +	movq	-8(%rsi, %rdx), %r8
> +	leaq	-8(%rdi, %rdx), %r9
> +	movq	(%rsi), %r10
> +	movq	%rdi, %r11
> +
> +	leaq	-1(%rdi, %rdx), %rcx
> +	movq	%rdi, %rdx
> +	leaq	8(%rdi), %rdi
> +	andq	$-8, %rdi
> +	subq	%rdi, %rdx
> +	subq	%rdx, %rsi
> +	subq	%rdi, %rcx
> +	shrq	$3, %rcx
> +	cld
> +	rep ; movsq
> +	movq	%r8, (%r9)
> +	movq	%r10, (%r11)
> +	ret
> +
> +L(bwd):
> +	movq	-8(%rsi, %rdx), %r8
> +	leaq	-8(%rdi, %rdx), %r9
> +	movq	(%rsi), %r10
> +	movq	%rdi, %r11
> +	leaq	-1(%rdi, %rdx), %rcx
> +	movq	%rdi, %rdx
> +	leaq	8(%rdi), %rdi
> +	andq	$-8, %rdi
> +	subq	%rdi, %rdx
> +	subq	%rdx, %rsi
> +	subq	%rdi, %rcx
> +	shrq	$3, %rcx
> +	lea	-8(%rsi, %rcx, 8), %rsi
> +	lea	-8(%rdi, %rcx, 8), %rdi
> +	std
> +	rep ; movsq
> +	cld
> +	movq	%r8, (%r9)
> +	movq	%r10, (%r11)
>  	ret
> -
> -	ALIGN (4)
> -L(write_17bytes):
> -	mov	-17(%rsi), %r11
> -	mov	-9(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %r11, -17(%rdi)
> -	mov	 %rcx, -9(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_9bytes):
> -	mov	-9(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %rcx, -9(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_1bytes):
> -	mov	-1(%rsi), %dl
> -	mov	 %dl, -1(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_74bytes):
> -	movdqu	-74(%rsi), %xmm0
> -	movdqu	-58(%rsi), %xmm1
> -	mov	-42(%rsi), %r8
> -	mov	-34(%rsi), %r9
> -	mov	-26(%rsi), %r10
> -	mov	-18(%rsi), %r11
> -	mov	-10(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	movdqu	 %xmm0, -74(%rdi)
> -	movdqu	 %xmm1, -58(%rdi)
> -	mov	 %r8, -42(%rdi)
> -	mov	 %r9, -34(%rdi)
> -	mov	 %r10, -26(%rdi)
> -	mov	 %r11, -18(%rdi)
> -	mov	 %rcx, -10(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_66bytes):
> -	movdqu	-66(%rsi), %xmm0
> -	movdqu	-50(%rsi), %xmm1
> -	mov	-42(%rsi), %r8
> -	mov	-34(%rsi), %r9
> -	mov	-26(%rsi), %r10
> -	mov	-18(%rsi), %r11
> -	mov	-10(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	movdqu	 %xmm0, -66(%rdi)
> -	movdqu	 %xmm1, -50(%rdi)
> -	mov	 %r8, -42(%rdi)
> -	mov	 %r9, -34(%rdi)
> -	mov	 %r10, -26(%rdi)
> -	mov	 %r11, -18(%rdi)
> -	mov	 %rcx, -10(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_58bytes):
> -	movdqu	-58(%rsi), %xmm1
> -	mov	-42(%rsi), %r8
> -	mov	-34(%rsi), %r9
> -	mov	-26(%rsi), %r10
> -	mov	-18(%rsi), %r11
> -	mov	-10(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	movdqu	 %xmm1, -58(%rdi)
> -	mov	 %r8, -42(%rdi)
> -	mov	 %r9, -34(%rdi)
> -	mov	 %r10, -26(%rdi)
> -	mov	 %r11, -18(%rdi)
> -	mov	 %rcx, -10(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_50bytes):
> -	movdqu	-50(%rsi), %xmm0
> -	mov	-34(%rsi), %r9
> -	mov	-26(%rsi), %r10
> -	mov	-18(%rsi), %r11
> -	mov	-10(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	movdqu	 %xmm0, -50(%rdi)
> -	mov	 %r9, -34(%rdi)
> -	mov	 %r10, -26(%rdi)
> -	mov	 %r11, -18(%rdi)
> -	mov	 %rcx, -10(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_42bytes):
> -	mov	-42(%rsi), %r8
> -	mov	-34(%rsi), %r9
> -	mov	-26(%rsi), %r10
> -	mov	-18(%rsi), %r11
> -	mov	-10(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %r8, -42(%rdi)
> -	mov	 %r9, -34(%rdi)
> -	mov	 %r10, -26(%rdi)
> -	mov	 %r11, -18(%rdi)
> -	mov	 %rcx, -10(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_34bytes):
> -	mov	-34(%rsi), %r9
> -	mov	-26(%rsi), %r10
> -	mov	-18(%rsi), %r11
> -	mov	-10(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %r9, -34(%rdi)
> -	mov	 %r10, -26(%rdi)
> -	mov	 %r11, -18(%rdi)
> -	mov	 %rcx, -10(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_26bytes):
> -	mov	-26(%rsi), %r10
> -	mov	-18(%rsi), %r11
> -	mov	-10(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %r10, -26(%rdi)
> -	mov	 %r11, -18(%rdi)
> -	mov	 %rcx, -10(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_18bytes):
> -	mov	-18(%rsi), %r11
> -	mov	-10(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %r11, -18(%rdi)
> -	mov	 %rcx, -10(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_10bytes):
> -	mov	-10(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %rcx, -10(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_2bytes):
> -	mov	-2(%rsi), %dx
> -	mov	 %dx, -2(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_75bytes):
> -	movdqu	-75(%rsi), %xmm0
> -	movdqu	-59(%rsi), %xmm1
> -	mov	-43(%rsi), %r8
> -	mov	-35(%rsi), %r9
> -	mov	-27(%rsi), %r10
> -	mov	-19(%rsi), %r11
> -	mov	-11(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	movdqu	 %xmm0, -75(%rdi)
> -	movdqu	 %xmm1, -59(%rdi)
> -	mov	 %r8, -43(%rdi)
> -	mov	 %r9, -35(%rdi)
> -	mov	 %r10, -27(%rdi)
> -	mov	 %r11, -19(%rdi)
> -	mov	 %rcx, -11(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_67bytes):
> -	movdqu	-67(%rsi), %xmm0
> -	movdqu	-59(%rsi), %xmm1
> -	mov	-43(%rsi), %r8
> -	mov	-35(%rsi), %r9
> -	mov	-27(%rsi), %r10
> -	mov	-19(%rsi), %r11
> -	mov	-11(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	movdqu	 %xmm0, -67(%rdi)
> -	movdqu	 %xmm1, -59(%rdi)
> -	mov	 %r8, -43(%rdi)
> -	mov	 %r9, -35(%rdi)
> -	mov	 %r10, -27(%rdi)
> -	mov	 %r11, -19(%rdi)
> -	mov	 %rcx, -11(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_59bytes):
> -	movdqu	-59(%rsi), %xmm0
> -	mov	-43(%rsi), %r8
> -	mov	-35(%rsi), %r9
> -	mov	-27(%rsi), %r10
> -	mov	-19(%rsi), %r11
> -	mov	-11(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	movdqu	 %xmm0, -59(%rdi)
> -	mov	 %r8, -43(%rdi)
> -	mov	 %r9, -35(%rdi)
> -	mov	 %r10, -27(%rdi)
> -	mov	 %r11, -19(%rdi)
> -	mov	 %rcx, -11(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_51bytes):
> -	movdqu	-51(%rsi), %xmm0
> -	mov	-35(%rsi), %r9
> -	mov	-27(%rsi), %r10
> -	mov	-19(%rsi), %r11
> -	mov	-11(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	movdqu	 %xmm0, -51(%rdi)
> -	mov	 %r9, -35(%rdi)
> -	mov	 %r10, -27(%rdi)
> -	mov	 %r11, -19(%rdi)
> -	mov	 %rcx, -11(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_43bytes):
> -	mov	-43(%rsi), %r8
> -	mov	-35(%rsi), %r9
> -	mov	-27(%rsi), %r10
> -	mov	-19(%rsi), %r11
> -	mov	-11(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %r8, -43(%rdi)
> -	mov	 %r9, -35(%rdi)
> -	mov	 %r10, -27(%rdi)
> -	mov	 %r11, -19(%rdi)
> -	mov	 %rcx, -11(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_35bytes):
> -	mov	-35(%rsi), %r9
> -	mov	-27(%rsi), %r10
> -	mov	-19(%rsi), %r11
> -	mov	-11(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %r9, -35(%rdi)
> -	mov	 %r10, -27(%rdi)
> -	mov	 %r11, -19(%rdi)
> -	mov	 %rcx, -11(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_27bytes):
> -	mov	-27(%rsi), %r10
> -	mov	-19(%rsi), %r11
> -	mov	-11(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %r10, -27(%rdi)
> -	mov	 %r11, -19(%rdi)
> -	mov	 %rcx, -11(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_19bytes):
> -	mov	-19(%rsi), %r11
> -	mov	-11(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %r11, -19(%rdi)
> -	mov	 %rcx, -11(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_11bytes):
> -	mov	-11(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %rcx, -11(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_3bytes):
> -	mov	-3(%rsi), %dx
> -	mov	-2(%rsi), %cx
> -	mov	 %dx, -3(%rdi)
> -	mov	 %cx, -2(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_76bytes):
> -	movdqu	-76(%rsi), %xmm0
> -	movdqu	-60(%rsi), %xmm1
> -	mov	-44(%rsi), %r8
> -	mov	-36(%rsi), %r9
> -	mov	-28(%rsi), %r10
> -	mov	-20(%rsi), %r11
> -	mov	-12(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	movdqu	 %xmm0, -76(%rdi)
> -	movdqu	 %xmm1, -60(%rdi)
> -	mov	 %r8, -44(%rdi)
> -	mov	 %r9, -36(%rdi)
> -	mov	 %r10, -28(%rdi)
> -	mov	 %r11, -20(%rdi)
> -	mov	 %rcx, -12(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_68bytes):
> -	movdqu	-68(%rsi), %xmm0
> -	movdqu	-52(%rsi), %xmm1
> -	mov	-36(%rsi), %r9
> -	mov	-28(%rsi), %r10
> -	mov	-20(%rsi), %r11
> -	mov	-12(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	movdqu	 %xmm0, -68(%rdi)
> -	movdqu	 %xmm1, -52(%rdi)
> -	mov	 %r9, -36(%rdi)
> -	mov	 %r10, -28(%rdi)
> -	mov	 %r11, -20(%rdi)
> -	mov	 %rcx, -12(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_60bytes):
> -	movdqu	-60(%rsi), %xmm0
> -	mov	-44(%rsi), %r8
> -	mov	-36(%rsi), %r9
> -	mov	-28(%rsi), %r10
> -	mov	-20(%rsi), %r11
> -	mov	-12(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	movdqu	 %xmm0, -60(%rdi)
> -	mov	 %r8, -44(%rdi)
> -	mov	 %r9, -36(%rdi)
> -	mov	 %r10, -28(%rdi)
> -	mov	 %r11, -20(%rdi)
> -	mov	 %rcx, -12(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_52bytes):
> -	movdqu	-52(%rsi), %xmm0
> -	mov	-36(%rsi), %r9
> -	mov	-28(%rsi), %r10
> -	mov	-20(%rsi), %r11
> -	mov	-12(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	movdqu	 %xmm0, -52(%rdi)
> -	mov	 %r9, -36(%rdi)
> -	mov	 %r10, -28(%rdi)
> -	mov	 %r11, -20(%rdi)
> -	mov	 %rcx, -12(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_44bytes):
> -	mov	-44(%rsi), %r8
> -	mov	-36(%rsi), %r9
> -	mov	-28(%rsi), %r10
> -	mov	-20(%rsi), %r11
> -	mov	-12(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %r8, -44(%rdi)
> -	mov	 %r9, -36(%rdi)
> -	mov	 %r10, -28(%rdi)
> -	mov	 %r11, -20(%rdi)
> -	mov	 %rcx, -12(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_36bytes):
> -	mov	-36(%rsi), %r9
> -	mov	-28(%rsi), %r10
> -	mov	-20(%rsi), %r11
> -	mov	-12(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %r9, -36(%rdi)
> -	mov	 %r10, -28(%rdi)
> -	mov	 %r11, -20(%rdi)
> -	mov	 %rcx, -12(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_28bytes):
> -	mov	-28(%rsi), %r10
> -	mov	-20(%rsi), %r11
> -	mov	-12(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %r10, -28(%rdi)
> -	mov	 %r11, -20(%rdi)
> -	mov	 %rcx, -12(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_20bytes):
> -	mov	-20(%rsi), %r11
> -	mov	-12(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %r11, -20(%rdi)
> -	mov	 %rcx, -12(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_12bytes):
> -	mov	-12(%rsi), %rcx
> -	mov	-4(%rsi), %edx
> -	mov	 %rcx, -12(%rdi)
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_4bytes):
> -	mov	-4(%rsi), %edx
> -	mov	 %edx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_77bytes):
> -	movdqu	-77(%rsi), %xmm0
> -	movdqu	-61(%rsi), %xmm1
> -	mov	-45(%rsi), %r8
> -	mov	-37(%rsi), %r9
> -	mov	-29(%rsi), %r10
> -	mov	-21(%rsi), %r11
> -	mov	-13(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	movdqu	 %xmm0, -77(%rdi)
> -	movdqu	 %xmm1, -61(%rdi)
> -	mov	 %r8, -45(%rdi)
> -	mov	 %r9, -37(%rdi)
> -	mov	 %r10, -29(%rdi)
> -	mov	 %r11, -21(%rdi)
> -	mov	 %rcx, -13(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_69bytes):
> -	movdqu	-69(%rsi), %xmm0
> -	movdqu	-53(%rsi), %xmm1
> -	mov	-37(%rsi), %r9
> -	mov	-29(%rsi), %r10
> -	mov	-21(%rsi), %r11
> -	mov	-13(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	movdqu	 %xmm0, -69(%rdi)
> -	movdqu	 %xmm1, -53(%rdi)
> -	mov	 %r9, -37(%rdi)
> -	mov	 %r10, -29(%rdi)
> -	mov	 %r11, -21(%rdi)
> -	mov	 %rcx, -13(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_61bytes):
> -	movdqu	-61(%rsi), %xmm0
> -	mov	-45(%rsi), %r8
> -	mov	-37(%rsi), %r9
> -	mov	-29(%rsi), %r10
> -	mov	-21(%rsi), %r11
> -	mov	-13(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	movdqu	 %xmm0, -61(%rdi)
> -	mov	 %r8, -45(%rdi)
> -	mov	 %r9, -37(%rdi)
> -	mov	 %r10, -29(%rdi)
> -	mov	 %r11, -21(%rdi)
> -	mov	 %rcx, -13(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_53bytes):
> -	movdqu	-53(%rsi), %xmm0
> -	mov	-45(%rsi), %r8
> -	mov	-37(%rsi), %r9
> -	mov	-29(%rsi), %r10
> -	mov	-21(%rsi), %r11
> -	mov	-13(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	movdqu	 %xmm0, -53(%rdi)
> -	mov	 %r9, -37(%rdi)
> -	mov	 %r10, -29(%rdi)
> -	mov	 %r11, -21(%rdi)
> -	mov	 %rcx, -13(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_45bytes):
> -	mov	-45(%rsi), %r8
> -	mov	-37(%rsi), %r9
> -	mov	-29(%rsi), %r10
> -	mov	-21(%rsi), %r11
> -	mov	-13(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	mov	 %r8, -45(%rdi)
> -	mov	 %r9, -37(%rdi)
> -	mov	 %r10, -29(%rdi)
> -	mov	 %r11, -21(%rdi)
> -	mov	 %rcx, -13(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_37bytes):
> -	mov	-37(%rsi), %r9
> -	mov	-29(%rsi), %r10
> -	mov	-21(%rsi), %r11
> -	mov	-13(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	mov	 %r9, -37(%rdi)
> -	mov	 %r10, -29(%rdi)
> -	mov	 %r11, -21(%rdi)
> -	mov	 %rcx, -13(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_29bytes):
> -	mov	-29(%rsi), %r10
> -	mov	-21(%rsi), %r11
> -	mov	-13(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	mov	 %r10, -29(%rdi)
> -	mov	 %r11, -21(%rdi)
> -	mov	 %rcx, -13(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_21bytes):
> -	mov	-21(%rsi), %r11
> -	mov	-13(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	mov	 %r11, -21(%rdi)
> -	mov	 %rcx, -13(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_13bytes):
> -	mov	-13(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	mov	 %rcx, -13(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_5bytes):
> -	mov	-5(%rsi), %edx
> -	mov	-4(%rsi), %ecx
> -	mov	 %edx, -5(%rdi)
> -	mov	 %ecx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_78bytes):
> -	movdqu	-78(%rsi), %xmm0
> -	movdqu	-62(%rsi), %xmm1
> -	mov	-46(%rsi), %r8
> -	mov	-38(%rsi), %r9
> -	mov	-30(%rsi), %r10
> -	mov	-22(%rsi), %r11
> -	mov	-14(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	movdqu	 %xmm0, -78(%rdi)
> -	movdqu	 %xmm1, -62(%rdi)
> -	mov	 %r8, -46(%rdi)
> -	mov	 %r9, -38(%rdi)
> -	mov	 %r10, -30(%rdi)
> -	mov	 %r11, -22(%rdi)
> -	mov	 %rcx, -14(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_70bytes):
> -	movdqu	-70(%rsi), %xmm0
> -	movdqu	-54(%rsi), %xmm1
> -	mov	-38(%rsi), %r9
> -	mov	-30(%rsi), %r10
> -	mov	-22(%rsi), %r11
> -	mov	-14(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	movdqu	 %xmm0, -70(%rdi)
> -	movdqu	 %xmm1, -54(%rdi)
> -	mov	 %r9, -38(%rdi)
> -	mov	 %r10, -30(%rdi)
> -	mov	 %r11, -22(%rdi)
> -	mov	 %rcx, -14(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_62bytes):
> -	movdqu	-62(%rsi), %xmm0
> -	mov	-46(%rsi), %r8
> -	mov	-38(%rsi), %r9
> -	mov	-30(%rsi), %r10
> -	mov	-22(%rsi), %r11
> -	mov	-14(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	movdqu	 %xmm0, -62(%rdi)
> -	mov	 %r8, -46(%rdi)
> -	mov	 %r9, -38(%rdi)
> -	mov	 %r10, -30(%rdi)
> -	mov	 %r11, -22(%rdi)
> -	mov	 %rcx, -14(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_54bytes):
> -	movdqu	-54(%rsi), %xmm0
> -	mov	-38(%rsi), %r9
> -	mov	-30(%rsi), %r10
> -	mov	-22(%rsi), %r11
> -	mov	-14(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	movdqu	 %xmm0, -54(%rdi)
> -	mov	 %r9, -38(%rdi)
> -	mov	 %r10, -30(%rdi)
> -	mov	 %r11, -22(%rdi)
> -	mov	 %rcx, -14(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_46bytes):
> -	mov	-46(%rsi), %r8
> -	mov	-38(%rsi), %r9
> -	mov	-30(%rsi), %r10
> -	mov	-22(%rsi), %r11
> -	mov	-14(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	mov	 %r8, -46(%rdi)
> -	mov	 %r9, -38(%rdi)
> -	mov	 %r10, -30(%rdi)
> -	mov	 %r11, -22(%rdi)
> -	mov	 %rcx, -14(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_38bytes):
> -	mov	-38(%rsi), %r9
> -	mov	-30(%rsi), %r10
> -	mov	-22(%rsi), %r11
> -	mov	-14(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	mov	 %r9, -38(%rdi)
> -	mov	 %r10, -30(%rdi)
> -	mov	 %r11, -22(%rdi)
> -	mov	 %rcx, -14(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_30bytes):
> -	mov	-30(%rsi), %r10
> -	mov	-22(%rsi), %r11
> -	mov	-14(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	mov	 %r10, -30(%rdi)
> -	mov	 %r11, -22(%rdi)
> -	mov	 %rcx, -14(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_22bytes):
> -	mov	-22(%rsi), %r11
> -	mov	-14(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	mov	 %r11, -22(%rdi)
> -	mov	 %rcx, -14(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_14bytes):
> -	mov	-14(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	mov	 %rcx, -14(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_6bytes):
> -	mov	-6(%rsi), %edx
> -	mov	-4(%rsi), %ecx
> -	mov	 %edx, -6(%rdi)
> -	mov	 %ecx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_79bytes):
> -	movdqu	-79(%rsi), %xmm0
> -	movdqu	-63(%rsi), %xmm1
> -	mov	-47(%rsi), %r8
> -	mov	-39(%rsi), %r9
> -	mov	-31(%rsi), %r10
> -	mov	-23(%rsi), %r11
> -	mov	-15(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	movdqu	 %xmm0, -79(%rdi)
> -	movdqu	 %xmm1, -63(%rdi)
> -	mov	 %r8, -47(%rdi)
> -	mov	 %r9, -39(%rdi)
> -	mov	 %r10, -31(%rdi)
> -	mov	 %r11, -23(%rdi)
> -	mov	 %rcx, -15(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_71bytes):
> -	movdqu	-71(%rsi), %xmm0
> -	movdqu	-55(%rsi), %xmm1
> -	mov	-39(%rsi), %r9
> -	mov	-31(%rsi), %r10
> -	mov	-23(%rsi), %r11
> -	mov	-15(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	movdqu	 %xmm0, -71(%rdi)
> -	movdqu	 %xmm1, -55(%rdi)
> -	mov	 %r9, -39(%rdi)
> -	mov	 %r10, -31(%rdi)
> -	mov	 %r11, -23(%rdi)
> -	mov	 %rcx, -15(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_63bytes):
> -	movdqu	-63(%rsi), %xmm0
> -	mov	-47(%rsi), %r8
> -	mov	-39(%rsi), %r9
> -	mov	-31(%rsi), %r10
> -	mov	-23(%rsi), %r11
> -	mov	-15(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	movdqu	 %xmm0, -63(%rdi)
> -	mov	 %r8, -47(%rdi)
> -	mov	 %r9, -39(%rdi)
> -	mov	 %r10, -31(%rdi)
> -	mov	 %r11, -23(%rdi)
> -	mov	 %rcx, -15(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_55bytes):
> -	movdqu	-55(%rsi), %xmm0
> -	mov	-39(%rsi), %r9
> -	mov	-31(%rsi), %r10
> -	mov	-23(%rsi), %r11
> -	mov	-15(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	movdqu	 %xmm0, -55(%rdi)
> -	mov	 %r9, -39(%rdi)
> -	mov	 %r10, -31(%rdi)
> -	mov	 %r11, -23(%rdi)
> -	mov	 %rcx, -15(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_47bytes):
> -	mov	-47(%rsi), %r8
> -	mov	-39(%rsi), %r9
> -	mov	-31(%rsi), %r10
> -	mov	-23(%rsi), %r11
> -	mov	-15(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	mov	 %r8, -47(%rdi)
> -	mov	 %r9, -39(%rdi)
> -	mov	 %r10, -31(%rdi)
> -	mov	 %r11, -23(%rdi)
> -	mov	 %rcx, -15(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_39bytes):
> -	mov	-39(%rsi), %r9
> -	mov	-31(%rsi), %r10
> -	mov	-23(%rsi), %r11
> -	mov	-15(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	mov	 %r9, -39(%rdi)
> -	mov	 %r10, -31(%rdi)
> -	mov	 %r11, -23(%rdi)
> -	mov	 %rcx, -15(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_31bytes):
> -	mov	-31(%rsi), %r10
> -	mov	-23(%rsi), %r11
> -	mov	-15(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	mov	 %r10, -31(%rdi)
> -	mov	 %r11, -23(%rdi)
> -	mov	 %rcx, -15(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_23bytes):
> -	mov	-23(%rsi), %r11
> -	mov	-15(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	mov	 %r11, -23(%rdi)
> -	mov	 %rcx, -15(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_15bytes):
> -	mov	-15(%rsi), %rcx
> -	mov	-8(%rsi), %rdx
> -	mov	 %rcx, -15(%rdi)
> -	mov	 %rdx, -8(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(write_7bytes):
> -	mov	-7(%rsi), %edx
> -	mov	-4(%rsi), %ecx
> -	mov	 %edx, -7(%rdi)
> -	mov	 %ecx, -4(%rdi)
> -	ret
> -
> -	ALIGN (4)
> -L(large_page_fwd):
> -	movdqu	(%rsi), %xmm1
> -	lea	16(%rsi), %rsi
> -	movdqu	%xmm0, (%r8)
> -	movntdq	%xmm1, (%rdi)
> -	lea	16(%rdi), %rdi
> -	lea	-0x90(%rdx), %rdx
> -#ifdef USE_AS_MEMMOVE
> -	mov	%rsi, %r9
> -	sub	%rdi, %r9
> -	cmp	%rdx, %r9
> -	jae	L(memmove_is_memcpy_fwd)
> -	shl	$2, %rcx
> -	cmp	%rcx, %rdx
> -	jb	L(ll_cache_copy_fwd_start)
> -L(memmove_is_memcpy_fwd):
> -#endif
> -L(large_page_loop):
> -	movdqu	(%rsi), %xmm0
> -	movdqu	0x10(%rsi), %xmm1
> -	movdqu	0x20(%rsi), %xmm2
> -	movdqu	0x30(%rsi), %xmm3
> -	movdqu	0x40(%rsi), %xmm4
> -	movdqu	0x50(%rsi), %xmm5
> -	movdqu	0x60(%rsi), %xmm6
> -	movdqu	0x70(%rsi), %xmm7
> -	lea	0x80(%rsi), %rsi
> -
> -	sub	$0x80, %rdx
> -	movntdq	%xmm0, (%rdi)
> -	movntdq	%xmm1, 0x10(%rdi)
> -	movntdq	%xmm2, 0x20(%rdi)
> -	movntdq	%xmm3, 0x30(%rdi)
> -	movntdq	%xmm4, 0x40(%rdi)
> -	movntdq	%xmm5, 0x50(%rdi)
> -	movntdq	%xmm6, 0x60(%rdi)
> -	movntdq	%xmm7, 0x70(%rdi)
> -	lea	0x80(%rdi), %rdi
> -	jae	L(large_page_loop)
> -	cmp	$-0x40, %rdx
> -	lea	0x80(%rdx), %rdx
> -	jl	L(large_page_less_64bytes)
> -
> -	movdqu	(%rsi), %xmm0
> -	movdqu	0x10(%rsi), %xmm1
> -	movdqu	0x20(%rsi), %xmm2
> -	movdqu	0x30(%rsi), %xmm3
> -	lea	0x40(%rsi), %rsi
> -
> -	movntdq	%xmm0, (%rdi)
> -	movntdq	%xmm1, 0x10(%rdi)
> -	movntdq	%xmm2, 0x20(%rdi)
> -	movntdq	%xmm3, 0x30(%rdi)
> -	lea	0x40(%rdi), %rdi
> -	sub	$0x40, %rdx
> -L(large_page_less_64bytes):
> -	add	%rdx, %rsi
> -	add	%rdx, %rdi
> -	sfence
> -	BRANCH_TO_JMPTBL_ENTRY (L(table_less_80bytes), %rdx, 4)
> -
> -#ifdef USE_AS_MEMMOVE
> -	ALIGN (4)
> -L(ll_cache_copy_fwd_start):
> -	prefetcht0 0x1c0(%rsi)
> -	prefetcht0 0x200(%rsi)
> -	movdqu	(%rsi), %xmm0
> -	movdqu	0x10(%rsi), %xmm1
> -	movdqu	0x20(%rsi), %xmm2
> -	movdqu	0x30(%rsi), %xmm3
> -	movdqu	0x40(%rsi), %xmm4
> -	movdqu	0x50(%rsi), %xmm5
> -	movdqu	0x60(%rsi), %xmm6
> -	movdqu	0x70(%rsi), %xmm7
> -	lea	0x80(%rsi), %rsi
> -
> -	sub	$0x80, %rdx
> -	movaps	%xmm0, (%rdi)
> -	movaps	%xmm1, 0x10(%rdi)
> -	movaps	%xmm2, 0x20(%rdi)
> -	movaps	%xmm3, 0x30(%rdi)
> -	movaps	%xmm4, 0x40(%rdi)
> -	movaps	%xmm5, 0x50(%rdi)
> -	movaps	%xmm6, 0x60(%rdi)
> -	movaps	%xmm7, 0x70(%rdi)
> -	lea	0x80(%rdi), %rdi
> -	jae	L(ll_cache_copy_fwd_start)
> -	cmp	$-0x40, %rdx
> -	lea	0x80(%rdx), %rdx
> -	jl	L(large_page_ll_less_fwd_64bytes)
> -
> -	movdqu	(%rsi), %xmm0
> -	movdqu	0x10(%rsi), %xmm1
> -	movdqu	0x20(%rsi), %xmm2
> -	movdqu	0x30(%rsi), %xmm3
> -	lea	0x40(%rsi), %rsi
> -
> -	movaps	%xmm0, (%rdi)
> -	movaps	%xmm1, 0x10(%rdi)
> -	movaps	%xmm2, 0x20(%rdi)
> -	movaps	%xmm3, 0x30(%rdi)
> -	lea	0x40(%rdi), %rdi
> -	sub	$0x40, %rdx
> -L(large_page_ll_less_fwd_64bytes):
> -	add	%rdx, %rsi
> -	add	%rdx, %rdi
> -	BRANCH_TO_JMPTBL_ENTRY (L(table_less_80bytes), %rdx, 4)
> -
> -#endif
> -	ALIGN (4)
> -L(large_page_bwd):
> -	movdqu	-0x10(%rsi), %xmm1
> -	lea	-16(%rsi), %rsi
> -	movdqu	%xmm0, (%r8)
> -	movdqa	%xmm1, -0x10(%rdi)
> -	lea	-16(%rdi), %rdi
> -	lea	-0x90(%rdx), %rdx
> -#ifdef USE_AS_MEMMOVE
> -	mov	%rdi, %r9
> -	sub	%rsi, %r9
> -	cmp	%rdx, %r9
> -	jae	L(memmove_is_memcpy_bwd)
> -	cmp	%rcx, %r9
> -	jb	L(ll_cache_copy_bwd_start)
> -L(memmove_is_memcpy_bwd):
> -#endif
> -L(large_page_bwd_loop):
> -	movdqu	-0x10(%rsi), %xmm0
> -	movdqu	-0x20(%rsi), %xmm1
> -	movdqu	-0x30(%rsi), %xmm2
> -	movdqu	-0x40(%rsi), %xmm3
> -	movdqu	-0x50(%rsi), %xmm4
> -	movdqu	-0x60(%rsi), %xmm5
> -	movdqu	-0x70(%rsi), %xmm6
> -	movdqu	-0x80(%rsi), %xmm7
> -	lea	-0x80(%rsi), %rsi
> -
> -	sub	$0x80, %rdx
> -	movntdq	%xmm0, -0x10(%rdi)
> -	movntdq	%xmm1, -0x20(%rdi)
> -	movntdq	%xmm2, -0x30(%rdi)
> -	movntdq	%xmm3, -0x40(%rdi)
> -	movntdq	%xmm4, -0x50(%rdi)
> -	movntdq	%xmm5, -0x60(%rdi)
> -	movntdq	%xmm6, -0x70(%rdi)
> -	movntdq	%xmm7, -0x80(%rdi)
> -	lea	-0x80(%rdi), %rdi
> -	jae	L(large_page_bwd_loop)
> -	cmp	$-0x40, %rdx
> -	lea	0x80(%rdx), %rdx
> -	jl	L(large_page_less_bwd_64bytes)
> -
> -	movdqu	-0x10(%rsi), %xmm0
> -	movdqu	-0x20(%rsi), %xmm1
> -	movdqu	-0x30(%rsi), %xmm2
> -	movdqu	-0x40(%rsi), %xmm3
> -	lea	-0x40(%rsi), %rsi
> -
> -	movntdq	%xmm0, -0x10(%rdi)
> -	movntdq	%xmm1, -0x20(%rdi)
> -	movntdq	%xmm2, -0x30(%rdi)
> -	movntdq	%xmm3, -0x40(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -	sub	$0x40, %rdx
> -L(large_page_less_bwd_64bytes):
> -	sfence
> -	BRANCH_TO_JMPTBL_ENTRY (L(table_less_80bytes), %rdx, 4)
> -
> -#ifdef USE_AS_MEMMOVE
> -	ALIGN (4)
> -L(ll_cache_copy_bwd_start):
> -	prefetcht0 -0x1c0(%rsi)
> -	prefetcht0 -0x200(%rsi)
> -	movdqu	-0x10(%rsi), %xmm0
> -	movdqu	-0x20(%rsi), %xmm1
> -	movdqu	-0x30(%rsi), %xmm2
> -	movdqu	-0x40(%rsi), %xmm3
> -	movdqu	-0x50(%rsi), %xmm4
> -	movdqu	-0x60(%rsi), %xmm5
> -	movdqu	-0x70(%rsi), %xmm6
> -	movdqu	-0x80(%rsi), %xmm7
> -	lea	-0x80(%rsi), %rsi
> -
> -	sub	$0x80, %rdx
> -	movaps	%xmm0, -0x10(%rdi)
> -	movaps	%xmm1, -0x20(%rdi)
> -	movaps	%xmm2, -0x30(%rdi)
> -	movaps	%xmm3, -0x40(%rdi)
> -	movaps	%xmm4, -0x50(%rdi)
> -	movaps	%xmm5, -0x60(%rdi)
> -	movaps	%xmm6, -0x70(%rdi)
> -	movaps	%xmm7, -0x80(%rdi)
> -	lea	-0x80(%rdi), %rdi
> -	jae	L(ll_cache_copy_bwd_start)
> -	cmp	$-0x40, %rdx
> -	lea	0x80(%rdx), %rdx
> -	jl	L(large_page_ll_less_bwd_64bytes)
> -
> -	movdqu	-0x10(%rsi), %xmm0
> -	movdqu	-0x20(%rsi), %xmm1
> -	movdqu	-0x30(%rsi), %xmm2
> -	movdqu	-0x40(%rsi), %xmm3
> -	lea	-0x40(%rsi), %rsi
> -
> -	movaps	%xmm0, -0x10(%rdi)
> -	movaps	%xmm1, -0x20(%rdi)
> -	movaps	%xmm2, -0x30(%rdi)
> -	movaps	%xmm3, -0x40(%rdi)
> -	lea	-0x40(%rdi), %rdi
> -	sub	$0x40, %rdx
> -L(large_page_ll_less_bwd_64bytes):
> -	BRANCH_TO_JMPTBL_ENTRY (L(table_less_80bytes), %rdx, 4)
> -#endif
> -
>  END (MEMCPY)
>  
> -	.section .rodata.ssse3,"a",@progbits
> -	ALIGN (3)
> -L(table_less_80bytes):
> -	.int	JMPTBL (L(write_0bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_1bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_2bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_3bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_4bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_5bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_6bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_7bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_8bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_9bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_10bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_11bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_12bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_13bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_14bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_15bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_16bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_17bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_18bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_19bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_20bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_21bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_22bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_23bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_24bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_25bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_26bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_27bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_28bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_29bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_30bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_31bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_32bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_33bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_34bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_35bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_36bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_37bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_38bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_39bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_40bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_41bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_42bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_43bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_44bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_45bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_46bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_47bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_48bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_49bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_50bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_51bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_52bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_53bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_54bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_55bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_56bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_57bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_58bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_59bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_60bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_61bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_62bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_63bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_64bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_65bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_66bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_67bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_68bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_69bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_70bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_71bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_72bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_73bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_74bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_75bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_76bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_77bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_78bytes), L(table_less_80bytes))
> -	.int	JMPTBL (L(write_79bytes), L(table_less_80bytes))
> +strong_alias(__memcpy_chk_ssse3, __memmove_chk_ssse3)
> +strong_alias(__memcpy_ssse3, __memmove_ssse3)
>  
> -	ALIGN (3)
> -L(shl_table):
> -	.int	JMPTBL (L(shl_0), L(shl_table))
> -	.int	JMPTBL (L(shl_1), L(shl_table))
> -	.int	JMPTBL (L(shl_2), L(shl_table))
> -	.int	JMPTBL (L(shl_3), L(shl_table))
> -	.int	JMPTBL (L(shl_4), L(shl_table))
> -	.int	JMPTBL (L(shl_5), L(shl_table))
> -	.int	JMPTBL (L(shl_6), L(shl_table))
> -	.int	JMPTBL (L(shl_7), L(shl_table))
> -	.int	JMPTBL (L(shl_8), L(shl_table))
> -	.int	JMPTBL (L(shl_9), L(shl_table))
> -	.int	JMPTBL (L(shl_10), L(shl_table))
> -	.int	JMPTBL (L(shl_11), L(shl_table))
> -	.int	JMPTBL (L(shl_12), L(shl_table))
> -	.int	JMPTBL (L(shl_13), L(shl_table))
> -	.int	JMPTBL (L(shl_14), L(shl_table))
> -	.int	JMPTBL (L(shl_15), L(shl_table))
> -
> -	ALIGN (3)
> -L(shl_table_bwd):
> -	.int	JMPTBL (L(shl_0_bwd), L(shl_table_bwd))
> -	.int	JMPTBL (L(shl_1_bwd), L(shl_table_bwd))
> -	.int	JMPTBL (L(shl_2_bwd), L(shl_table_bwd))
> -	.int	JMPTBL (L(shl_3_bwd), L(shl_table_bwd))
> -	.int	JMPTBL (L(shl_4_bwd), L(shl_table_bwd))
> -	.int	JMPTBL (L(shl_5_bwd), L(shl_table_bwd))
> -	.int	JMPTBL (L(shl_6_bwd), L(shl_table_bwd))
> -	.int	JMPTBL (L(shl_7_bwd), L(shl_table_bwd))
> -	.int	JMPTBL (L(shl_8_bwd), L(shl_table_bwd))
> -	.int	JMPTBL (L(shl_9_bwd), L(shl_table_bwd))
> -	.int	JMPTBL (L(shl_10_bwd), L(shl_table_bwd))
> -	.int	JMPTBL (L(shl_11_bwd), L(shl_table_bwd))
> -	.int	JMPTBL (L(shl_12_bwd), L(shl_table_bwd))
> -	.int	JMPTBL (L(shl_13_bwd), L(shl_table_bwd))
> -	.int	JMPTBL (L(shl_14_bwd), L(shl_table_bwd))
> -	.int	JMPTBL (L(shl_15_bwd), L(shl_table_bwd))
> -
> -#endif
> diff --git a/sysdeps/x86_64/multiarch/memcpy.S b/sysdeps/x86_64/multiarch/memcpy.S
> index a1e5031..b4bde29 100644
> --- a/sysdeps/x86_64/multiarch/memcpy.S
> +++ b/sysdeps/x86_64/multiarch/memcpy.S
> @@ -35,6 +35,8 @@ ENTRY(__new_memcpy)
>  1:	leaq	__memcpy_sse2(%rip), %rax
>  	testl	$bit_Slow_BSF, __cpu_features+FEATURE_OFFSET+index_Slow_BSF(%rip)
>  	jnz	2f
> +	testl   $bit_Fast_Unaligned_Load, __cpu_features+FEATURE_OFFSET+index_Fast_Unaligned_Load(%rip)
> +	jz	2f
>  	leaq	__memcpy_sse2_unaligned(%rip), %rax
>  	ret
>  2:	testl   $bit_SSSE3, __cpu_features+CPUID_OFFSET+index_SSSE3(%rip)
> diff --git a/sysdeps/x86_64/multiarch/memmove-ssse3.S b/sysdeps/x86_64/multiarch/memmove-ssse3.S
> index 295430b..4969719 100644
> --- a/sysdeps/x86_64/multiarch/memmove-ssse3.S
> +++ b/sysdeps/x86_64/multiarch/memmove-ssse3.S
> @@ -1,4 +1 @@
> -#define USE_AS_MEMMOVE
> -#define MEMCPY		__memmove_ssse3
> -#define MEMCPY_CHK	__memmove_chk_ssse3
> -#include "memcpy-ssse3.S"
> +/* Moved to memcpy-ssse3.S.  */
> diff --git a/sysdeps/x86_64/multiarch/mempcpy-ssse3.S b/sysdeps/x86_64/multiarch/mempcpy-ssse3.S
> index 822d98e..98dcc32 100644
> --- a/sysdeps/x86_64/multiarch/mempcpy-ssse3.S
> +++ b/sysdeps/x86_64/multiarch/mempcpy-ssse3.S
> @@ -1,4 +1 @@
> -#define USE_AS_MEMPCPY
> -#define MEMCPY		__mempcpy_ssse3
> -#define MEMCPY_CHK	__mempcpy_chk_ssse3
> -#include "memcpy-ssse3.S"
> +/* Moved to memcpy-ssse.S.  */
> -- 
> 1.8.3.2

-- 

operators on strike due to broken coffee machine
References:
- [RFC] A memcpy ssse3/sse2 variant for older processors.
  - From: OndÅej BÃlka
- [RFC v1.1] A memcpy ssse3/sse2 variant for older processors.
  - From: OndÅej BÃlka
Index Nav:	[Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav:	[Date Prev] [Date Next]	[Thread Prev] [Thread Next]