This is the mail archive of the
glibc-cvs@sourceware.org
mailing list for the glibc project.
GNU C Library master sources branch hjl/memcpy/unaligned/master created. glibc-2.23-74-gecfd2bd
- From: hjl at sourceware dot org
- To: glibc-cvs at sourceware dot org
- Date: 7 Mar 2016 14:42:47 -0000
- Subject: GNU C Library master sources branch hjl/memcpy/unaligned/master created. glibc-2.23-74-gecfd2bd
This is an automated email from the git hooks/post-receive script. It was
generated because a ref change was pushed to the repository containing
the project "GNU C Library master sources".
The branch, hjl/memcpy/unaligned/master has been created
at ecfd2bdc35628916e14ca10d73155bd1317c79f4 (commit)
- Log -----------------------------------------------------------------
http://sourceware.org/git/gitweb.cgi?p=glibc.git;a=commitdiff;h=ecfd2bdc35628916e14ca10d73155bd1317c79f4
commit ecfd2bdc35628916e14ca10d73155bd1317c79f4
Author: H.J. Lu <hjl.tools@gmail.com>
Date: Sun Mar 6 16:42:41 2016 -0800
Make __memcpy_sse2_unaligned as the default
In libc.a and ld.so, choose __memcpy_sse2_unaligned which is optimized
for current Intel and AMD x86-64 processors.
* sysdeps/x86_64/multiarch/memcpy-sse2-unaligned.S: Moved to ...
* sysdeps/x86_64/memcpy.S: Here. Add __GI_memcpy, __GI_mempcpy
and __GI___mempcpy. Support symbol versioning.
* sysdeps/x86_64/memcpy.S: Moved to ...
* sysdeps/x86_64/multiarch/memcpy-sse2.S: Here.
(__memcpy_chk): Renamed to ...
(__memcpy_chk_sse2): This.
(memcpy): Renamed to ...
(__memcpy_sse2): This.
* sysdeps/x86_64/mempcpy.S: Make it dummy.
* sysdeps/x86_64/mempcpy_chk.S: Check SHARED instead of PIC.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Replace
memcpy-sse2-unaligned with memcpy-sse2. Add mempcpy-sse2.
* sysdeps/x86_64/multiarch/memcpy.S (ENTRY): Removed.
(END): Likewise.
(ENTRY_CHK): Likewise.
(END_CHK): Likewise.
(libc_hidden_builtin_def): Likewise.
(__GI_memcpy): New alias.
(__GI_mempcpy): Likewise.
(__GI___mempcpy): Likewise.
* sysdeps/x86_64/multiarch/mempcpy-sse2.S: New file.
* sysdeps/x86_64/multiarch/mempcpy.S (ENTRY): Removed.
(END): Likewise.
(ENTRY_CHK): Likewise.
(END_CHK): Likewise.
(libc_hidden_builtin_def): Likewise.
diff --git a/sysdeps/x86_64/memcpy.S b/sysdeps/x86_64/memcpy.S
index f6e3d93..f2a84c9 100644
--- a/sysdeps/x86_64/memcpy.S
+++ b/sysdeps/x86_64/memcpy.S
@@ -1,9 +1,5 @@
-/*
- Optimized memcpy for x86-64.
-
- Copyright (C) 2007-2016 Free Software Foundation, Inc.
- Contributed by Evandro Menezes <evandro.menezes@amd.com>, 2007.
-
+/* Optimized memcpy for x86-64
+ Copyright (C) 2013-2016 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
@@ -18,565 +14,140 @@
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
- <http://www.gnu.org/licenses/>.
-*/
+ <http://www.gnu.org/licenses/>. */
#include <sysdep.h>
-#include "asm-syntax.h"
-/* Stack slots in the red-zone. */
+#include "asm-syntax.h"
-#ifdef USE_AS_MEMPCPY
-# define RETVAL (0)
-#else
-# define RETVAL (-8)
-# if defined SHARED && !defined USE_MULTIARCH && IS_IN (libc)
-# define memcpy __memcpy
-# undef libc_hidden_builtin_def
-# define libc_hidden_builtin_def(name) \
- .globl __GI_memcpy; __GI_memcpy = __memcpy
+#if IS_IN (libc)
+# ifdef SHARED
+# ifndef USE_MULTIARCH
+# define memcpy __memcpy
# endif
-#endif
-#define SAVE0 (RETVAL - 8)
-#define SAVE1 (SAVE0 - 8)
-#define SAVE2 (SAVE1 - 8)
-#define SAVE3 (SAVE2 - 8)
- .text
+ENTRY(__mempcpy_chk)
+ cmpq %rdx, %rcx
+ jb HIDDEN_JUMPTARGET (__chk_fail)
+END(__mempcpy_chk)
+# endif
-#if defined PIC && IS_IN (libc)
-ENTRY_CHK (__memcpy_chk)
+ENTRY(__mempcpy)
+ mov %rdi, %rax
+ add %rdx, %rax
+ jmp L(start)
+END (__mempcpy)
+# ifdef SHARED
+ENTRY(__memcpy_chk)
cmpq %rdx, %rcx
jb HIDDEN_JUMPTARGET (__chk_fail)
-
-END_CHK (__memcpy_chk)
+END(__memcpy_chk)
+# endif
#endif
-ENTRY(memcpy) /* (void *, const void*, size_t) */
-
-/* Handle tiny blocks. */
-
-L(1try): /* up to 32B */
+ENTRY(memcpy)
+ movq %rdi, %rax
+L(start):
+ testq %rdx, %rdx
+ je L(return)
+ cmpq $16, %rdx
+ jbe L(less_16)
+ movdqu (%rsi), %xmm8
cmpq $32, %rdx
-#ifndef USE_AS_MEMPCPY
- movq %rdi, %rax /* save return value */
-#endif
- jae L(1after)
-
-L(1): /* 1-byte once */
- testb $1, %dl
- jz L(1a)
-
- movzbl (%rsi), %ecx
- movb %cl, (%rdi)
-
- incq %rsi
- incq %rdi
-
- .p2align 4,, 4
-
-L(1a): /* 2-byte once */
- testb $2, %dl
- jz L(1b)
-
- movzwl (%rsi), %ecx
- movw %cx, (%rdi)
-
- addq $2, %rsi
- addq $2, %rdi
-
- .p2align 4,, 4
-
-L(1b): /* 4-byte once */
+ movdqu %xmm8, (%rdi)
+ movdqu -16(%rsi,%rdx), %xmm8
+ movdqu %xmm8, -16(%rdi,%rdx)
+ ja .L31
+L(return):
+ ret
+ .p2align 4,,10
+ .p2align 4
+.L31:
+ movdqu 16(%rsi), %xmm8
+ cmpq $64, %rdx
+ movdqu %xmm8, 16(%rdi)
+ movdqu -32(%rsi,%rdx), %xmm8
+ movdqu %xmm8, -32(%rdi,%rdx)
+ jbe L(return)
+ movdqu 32(%rsi), %xmm8
+ cmpq $128, %rdx
+ movdqu %xmm8, 32(%rdi)
+ movdqu -48(%rsi,%rdx), %xmm8
+ movdqu %xmm8, -48(%rdi,%rdx)
+ movdqu 48(%rsi), %xmm8
+ movdqu %xmm8, 48(%rdi)
+ movdqu -64(%rsi,%rdx), %xmm8
+ movdqu %xmm8, -64(%rdi,%rdx)
+ jbe L(return)
+ leaq 64(%rdi), %rcx
+ addq %rdi, %rdx
+ andq $-64, %rdx
+ andq $-64, %rcx
+ movq %rcx, %r11
+ subq %rdi, %r11
+ addq %r11, %rsi
+ cmpq %rdx, %rcx
+ je L(return)
+ movq %rsi, %r10
+ subq %rcx, %r10
+ leaq 16(%r10), %r9
+ leaq 32(%r10), %r8
+ leaq 48(%r10), %r11
+ .p2align 4,,10
+ .p2align 4
+L(loop):
+ movdqu (%rcx,%r10), %xmm8
+ movdqa %xmm8, (%rcx)
+ movdqu (%rcx,%r9), %xmm8
+ movdqa %xmm8, 16(%rcx)
+ movdqu (%rcx,%r8), %xmm8
+ movdqa %xmm8, 32(%rcx)
+ movdqu (%rcx,%r11), %xmm8
+ movdqa %xmm8, 48(%rcx)
+ addq $64, %rcx
+ cmpq %rcx, %rdx
+ jne L(loop)
+ ret
+L(less_16):
+ testb $24, %dl
+ jne L(between_9_16)
testb $4, %dl
- jz L(1c)
-
- movl (%rsi), %ecx
- movl %ecx, (%rdi)
-
- addq $4, %rsi
- addq $4, %rdi
-
- .p2align 4,, 4
-
-L(1c): /* 8-byte once */
- testb $8, %dl
- jz L(1d)
-
+ .p2align 4,,5
+ jne L(between_5_8)
+ testq %rdx, %rdx
+ .p2align 4,,2
+ je L(return)
+ movzbl (%rsi), %ecx
+ testb $2, %dl
+ movb %cl, (%rdi)
+ je L(return)
+ movzwl -2(%rsi,%rdx), %ecx
+ movw %cx, -2(%rdi,%rdx)
+ ret
+L(between_9_16):
movq (%rsi), %rcx
movq %rcx, (%rdi)
-
- addq $8, %rsi
- addq $8, %rdi
-
- .p2align 4,, 4
-
-L(1d): /* 16-byte loop */
- andl $0xf0, %edx
- jz L(exit)
-
- .p2align 4
-
-L(1loop):
- movq (%rsi), %rcx
- movq 8(%rsi), %r8
- movq %rcx, (%rdi)
- movq %r8, 8(%rdi)
-
- subl $16, %edx
-
- leaq 16(%rsi), %rsi
- leaq 16(%rdi), %rdi
-
- jnz L(1loop)
-
- .p2align 4,, 4
-
-L(exit): /* exit */
-#ifdef USE_AS_MEMPCPY
- movq %rdi, %rax /* return value */
-#else
- rep
-#endif
- retq
-
- .p2align 4
-
-L(1after):
-#ifndef USE_AS_MEMPCPY
- movq %rax, RETVAL(%rsp) /* save return value */
-#endif
-
-/* Align to the natural word size. */
-
-L(aligntry):
- movl %esi, %ecx /* align by source */
-
- andl $7, %ecx
- jz L(alignafter) /* already aligned */
-
-L(align): /* align */
- leaq -8(%rcx, %rdx), %rdx /* calculate remaining bytes */
- subl $8, %ecx
-
- .p2align 4
-
-L(alignloop): /* 1-byte alignment loop */
- movzbl (%rsi), %eax
- movb %al, (%rdi)
-
- incl %ecx
-
- leaq 1(%rsi), %rsi
- leaq 1(%rdi), %rdi
-
- jnz L(alignloop)
-
- .p2align 4
-
-L(alignafter):
-
-/* Handle mid-sized blocks. */
-
-L(32try): /* up to 1KB */
- cmpq $1024, %rdx
- ja L(32after)
-
-L(32): /* 32-byte loop */
- movl %edx, %ecx
- shrl $5, %ecx
- jz L(32skip)
-
- .p2align 4
-
-L(32loop):
- decl %ecx
-
- movq (%rsi), %rax
- movq 8(%rsi), %r8
- movq 16(%rsi), %r9
- movq 24(%rsi), %r10
-
- movq %rax, (%rdi)
- movq %r8, 8(%rdi)
- movq %r9, 16(%rdi)
- movq %r10, 24(%rdi)
-
- leaq 32(%rsi), %rsi
- leaq 32(%rdi), %rdi
-
- jz L(32skip) /* help out smaller blocks */
-
- decl %ecx
-
- movq (%rsi), %rax
- movq 8(%rsi), %r8
- movq 16(%rsi), %r9
- movq 24(%rsi), %r10
-
- movq %rax, (%rdi)
- movq %r8, 8(%rdi)
- movq %r9, 16(%rdi)
- movq %r10, 24(%rdi)
-
- leaq 32(%rsi), %rsi
- leaq 32(%rdi), %rdi
-
- jnz L(32loop)
-
- .p2align 4
-
-L(32skip):
- andl $31, %edx /* check for left overs */
-#ifdef USE_AS_MEMPCPY
- jnz L(1)
-
- movq %rdi, %rax
-#else
- movq RETVAL(%rsp), %rax
- jnz L(1)
-
- rep
-#endif
- retq /* exit */
-
- .p2align 4
-
-L(32after):
-
-/*
- In order to minimize code-size in RTLD, algorithms specific for
- larger blocks are excluded when building for RTLD.
-*/
-
-/* Handle blocks smaller than 1/2 L1. */
-
-L(fasttry): /* first 1/2 L1 */
-#if IS_IN (libc) /* only up to this algorithm outside of libc.so */
- mov __x86_data_cache_size_half(%rip), %R11_LP
- cmpq %rdx, %r11 /* calculate the smaller of */
- cmovaq %rdx, %r11 /* remaining bytes and 1/2 L1 */
-#endif
-
-L(fast): /* good ol' MOVS */
-#if IS_IN (libc)
- movq %r11, %rcx
- andq $-8, %r11
-#else
- movq %rdx, %rcx
-#endif
- shrq $3, %rcx
- jz L(fastskip)
-
- rep
- movsq
-
- .p2align 4,, 4
-
-L(fastskip):
-#if IS_IN (libc)
- subq %r11, %rdx /* check for more */
- testq $-8, %rdx
- jnz L(fastafter)
-#endif
-
- andl $7, %edx /* check for left overs */
-#ifdef USE_AS_MEMPCPY
- jnz L(1)
-
- movq %rdi, %rax
-#else
- movq RETVAL(%rsp), %rax
- jnz L(1)
-
- rep
-#endif
- retq /* exit */
-
-#if IS_IN (libc) /* none of the algorithms below for RTLD */
-
- .p2align 4
-
-L(fastafter):
-
-/* Handle large blocks smaller than 1/2 L2. */
-
-L(pretry): /* first 1/2 L2 */
- mov __x86_shared_cache_size_half (%rip), %R8_LP
- cmpq %rdx, %r8 /* calculate the lesser of */
- cmovaq %rdx, %r8 /* remaining bytes and 1/2 L2 */
-
-L(pre): /* 64-byte with prefetching */
- movq %r8, %rcx
- andq $-64, %r8
- shrq $6, %rcx
- jz L(preskip)
-
- movq %r14, SAVE0(%rsp)
- cfi_rel_offset (%r14, SAVE0)
- movq %r13, SAVE1(%rsp)
- cfi_rel_offset (%r13, SAVE1)
- movq %r12, SAVE2(%rsp)
- cfi_rel_offset (%r12, SAVE2)
- movq %rbx, SAVE3(%rsp)
- cfi_rel_offset (%rbx, SAVE3)
-
- cmpl $0, __x86_prefetchw(%rip)
- jz L(preloop) /* check if PREFETCHW OK */
-
- .p2align 4
-
-/* ... when PREFETCHW is available (less cache-probe traffic in MP systems). */
-
-L(prewloop): /* cache-line in state M */
- decq %rcx
-
- movq (%rsi), %rax
- movq 8 (%rsi), %rbx
- movq 16 (%rsi), %r9
- movq 24 (%rsi), %r10
- movq 32 (%rsi), %r11
- movq 40 (%rsi), %r12
- movq 48 (%rsi), %r13
- movq 56 (%rsi), %r14
-
- prefetcht0 0 + 896 (%rsi)
- prefetcht0 64 + 896 (%rsi)
-
- movq %rax, (%rdi)
- movq %rbx, 8(%rdi)
- movq %r9, 16(%rdi)
- movq %r10, 24(%rdi)
- movq %r11, 32(%rdi)
- movq %r12, 40(%rdi)
- movq %r13, 48(%rdi)
- movq %r14, 56(%rdi)
-
- leaq 64(%rsi), %rsi
- leaq 64(%rdi), %rdi
-
- jz L(prebail)
-
- decq %rcx
-
- movq (%rsi), %rax
- movq 8(%rsi), %rbx
- movq 16(%rsi), %r9
- movq 24(%rsi), %r10
- movq 32(%rsi), %r11
- movq 40(%rsi), %r12
- movq 48(%rsi), %r13
- movq 56(%rsi), %r14
-
- movq %rax, (%rdi)
- movq %rbx, 8(%rdi)
- movq %r9, 16(%rdi)
- movq %r10, 24(%rdi)
- movq %r11, 32(%rdi)
- movq %r12, 40(%rdi)
- movq %r13, 48(%rdi)
- movq %r14, 56(%rdi)
-
- prefetchw 896 - 64(%rdi)
- prefetchw 896 - 0(%rdi)
-
- leaq 64(%rsi), %rsi
- leaq 64(%rdi), %rdi
-
- jnz L(prewloop)
- jmp L(prebail)
-
- .p2align 4
-
-/* ... when PREFETCHW is not available. */
-
-L(preloop): /* cache-line in state E */
- decq %rcx
-
- movq (%rsi), %rax
- movq 8(%rsi), %rbx
- movq 16(%rsi), %r9
- movq 24(%rsi), %r10
- movq 32(%rsi), %r11
- movq 40(%rsi), %r12
- movq 48(%rsi), %r13
- movq 56(%rsi), %r14
-
- prefetcht0 896 + 0(%rsi)
- prefetcht0 896 + 64(%rsi)
-
- movq %rax, (%rdi)
- movq %rbx, 8(%rdi)
- movq %r9, 16(%rdi)
- movq %r10, 24(%rdi)
- movq %r11, 32(%rdi)
- movq %r12, 40(%rdi)
- movq %r13, 48(%rdi)
- movq %r14, 56(%rdi)
-
- leaq 64 (%rsi), %rsi
- leaq 64 (%rdi), %rdi
-
- jz L(prebail)
-
- decq %rcx
-
- movq (%rsi), %rax
- movq 8(%rsi), %rbx
- movq 16(%rsi), %r9
- movq 24(%rsi), %r10
- movq 32(%rsi), %r11
- movq 40(%rsi), %r12
- movq 48(%rsi), %r13
- movq 56(%rsi), %r14
-
- prefetcht0 896 - 64(%rdi)
- prefetcht0 896 - 0(%rdi)
-
- movq %rax, (%rdi)
- movq %rbx, 8(%rdi)
- movq %r9, 16(%rdi)
- movq %r10, 24(%rdi)
- movq %r11, 32(%rdi)
- movq %r12, 40(%rdi)
- movq %r13, 48(%rdi)
- movq %r14, 56(%rdi)
-
- leaq 64(%rsi), %rsi
- leaq 64(%rdi), %rdi
-
- jnz L(preloop)
-
-L(prebail):
- movq SAVE3(%rsp), %rbx
- cfi_restore (%rbx)
- movq SAVE2(%rsp), %r12
- cfi_restore (%r12)
- movq SAVE1(%rsp), %r13
- cfi_restore (%r13)
- movq SAVE0(%rsp), %r14
- cfi_restore (%r14)
-
-/* .p2align 4 */
-
-L(preskip):
- subq %r8, %rdx /* check for more */
- testq $-64, %rdx
- jnz L(preafter)
-
- andl $63, %edx /* check for left overs */
-#ifdef USE_AS_MEMPCPY
- jnz L(1)
-
- movq %rdi, %rax
-#else
- movq RETVAL(%rsp), %rax
- jnz L(1)
-
- rep
-#endif
- retq /* exit */
-
- .p2align 4
-
-L(preafter):
-
-/* Handle huge blocks. */
-
-L(NTtry):
-
-L(NT): /* non-temporal 128-byte */
- movq %rdx, %rcx
- shrq $7, %rcx
- jz L(NTskip)
-
- movq %r14, SAVE0(%rsp)
- cfi_rel_offset (%r14, SAVE0)
- movq %r13, SAVE1(%rsp)
- cfi_rel_offset (%r13, SAVE1)
- movq %r12, SAVE2(%rsp)
- cfi_rel_offset (%r12, SAVE2)
-
- .p2align 4
-
-L(NTloop):
- prefetchnta 768(%rsi)
- prefetchnta 832(%rsi)
-
- decq %rcx
-
- movq (%rsi), %rax
- movq 8(%rsi), %r8
- movq 16(%rsi), %r9
- movq 24(%rsi), %r10
- movq 32(%rsi), %r11
- movq 40(%rsi), %r12
- movq 48(%rsi), %r13
- movq 56(%rsi), %r14
-
- movntiq %rax, (%rdi)
- movntiq %r8, 8(%rdi)
- movntiq %r9, 16(%rdi)
- movntiq %r10, 24(%rdi)
- movntiq %r11, 32(%rdi)
- movntiq %r12, 40(%rdi)
- movntiq %r13, 48(%rdi)
- movntiq %r14, 56(%rdi)
-
- movq 64(%rsi), %rax
- movq 72(%rsi), %r8
- movq 80(%rsi), %r9
- movq 88(%rsi), %r10
- movq 96(%rsi), %r11
- movq 104(%rsi), %r12
- movq 112(%rsi), %r13
- movq 120(%rsi), %r14
-
- movntiq %rax, 64(%rdi)
- movntiq %r8, 72(%rdi)
- movntiq %r9, 80(%rdi)
- movntiq %r10, 88(%rdi)
- movntiq %r11, 96(%rdi)
- movntiq %r12, 104(%rdi)
- movntiq %r13, 112(%rdi)
- movntiq %r14, 120(%rdi)
-
- leaq 128(%rsi), %rsi
- leaq 128(%rdi), %rdi
-
- jnz L(NTloop)
-
- sfence /* serialize memory stores */
-
- movq SAVE2(%rsp), %r12
- cfi_restore (%r12)
- movq SAVE1(%rsp), %r13
- cfi_restore (%r13)
- movq SAVE0(%rsp), %r14
- cfi_restore (%r14)
-
-L(NTskip):
- andl $127, %edx /* check for left overs */
-#ifdef USE_AS_MEMPCPY
- jnz L(1)
-
- movq %rdi, %rax
-#else
- movq RETVAL(%rsp), %rax
- jnz L(1)
-
- rep
-#endif
- retq /* exit */
-
-#endif /* IS_IN (libc) */
-
+ movq -8(%rsi,%rdx), %rcx
+ movq %rcx, -8(%rdi,%rdx)
+ ret
+L(between_5_8):
+ movl (%rsi), %ecx
+ movl %ecx, (%rdi)
+ movl -4(%rsi,%rdx), %ecx
+ movl %ecx, -4(%rdi,%rdx)
+ ret
END(memcpy)
-#ifndef USE_AS_MEMPCPY
-libc_hidden_builtin_def (memcpy)
-# if defined SHARED && !defined USE_MULTIARCH && IS_IN (libc)
+#if IS_IN (libc)
+weak_alias (__mempcpy, mempcpy)
+# if defined SHARED && !defined USE_MULTIARCH
+ .globl __GI_memcpy
+ __GI_memcpy = __memcpy
+ .globl __GI_mempcpy
+ __GI_mempcpy = __mempcpy
+ .globl __GI___mempcpy
+ __GI___mempcpy = __mempcpy
# undef memcpy
# include <shlib-compat.h>
versioned_symbol (libc, __memcpy, memcpy, GLIBC_2_14);
diff --git a/sysdeps/x86_64/mempcpy.S b/sysdeps/x86_64/mempcpy.S
index acee5e5..21fd03d 100644
--- a/sysdeps/x86_64/mempcpy.S
+++ b/sysdeps/x86_64/mempcpy.S
@@ -1,8 +1 @@
-#define USE_AS_MEMPCPY
-#define memcpy __mempcpy
-#define __memcpy_chk __mempcpy_chk
-#include <sysdeps/x86_64/memcpy.S>
-
-libc_hidden_def (__mempcpy)
-weak_alias (__mempcpy, mempcpy)
-libc_hidden_builtin_def (mempcpy)
+/* mempcpy is implemented in memcpy.S. */
diff --git a/sysdeps/x86_64/mempcpy_chk.S b/sysdeps/x86_64/mempcpy_chk.S
index 390abc6..4ed9c91 100644
--- a/sysdeps/x86_64/mempcpy_chk.S
+++ b/sysdeps/x86_64/mempcpy_chk.S
@@ -19,7 +19,7 @@
#include <sysdep.h>
#include "asm-syntax.h"
-#ifndef PIC
+#ifndef SHARED
/* For libc.so this is defined in memcpy.S.
For libc.a, this is a separate source to avoid
mempcpy bringing in __chk_fail and all routines
diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile
index d234f4a..a624bdb 100644
--- a/sysdeps/x86_64/multiarch/Makefile
+++ b/sysdeps/x86_64/multiarch/Makefile
@@ -7,10 +7,10 @@ ifeq ($(subdir),string)
sysdep_routines += strncat-c stpncpy-c strncpy-c strcmp-ssse3 \
strcmp-sse2-unaligned strncmp-ssse3 \
- memcmp-sse4 memcpy-ssse3 memcpy-sse2-unaligned \
+ memcmp-sse4 memcpy-ssse3 memcpy-sse2 \
memcpy-avx512-no-vzeroupper mempcpy-ssse3 memmove-ssse3 \
memcpy-ssse3-back mempcpy-ssse3-back memmove-avx-unaligned \
- memcpy-avx-unaligned mempcpy-avx-unaligned \
+ mempcpy-sse2 memcpy-avx-unaligned mempcpy-avx-unaligned \
mempcpy-avx512-no-vzeroupper memmove-ssse3-back \
memmove-avx512-no-vzeroupper strcasecmp_l-ssse3 \
strncase_l-ssse3 strcat-ssse3 strncat-ssse3\
diff --git a/sysdeps/x86_64/multiarch/memcpy-sse2-unaligned.S b/sysdeps/x86_64/multiarch/memcpy-sse2-unaligned.S
deleted file mode 100644
index 947c50f..0000000
--- a/sysdeps/x86_64/multiarch/memcpy-sse2-unaligned.S
+++ /dev/null
@@ -1,138 +0,0 @@
-/* memcpy with unaliged loads
- Copyright (C) 2013-2016 Free Software Foundation, Inc.
- 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, see
- <http://www.gnu.org/licenses/>. */
-
-#if IS_IN (libc)
-
-#include <sysdep.h>
-
-#include "asm-syntax.h"
-
-# ifdef SHARED
-ENTRY (__mempcpy_chk_sse2_unaligned)
- cmpq %rdx, %rcx
- jb HIDDEN_JUMPTARGET (__chk_fail)
-END (__mempcpy_chk_sse2_unaligned)
-# endif
-
-ENTRY (__mempcpy_sse2_unaligned)
- mov %rdi, %rax
- add %rdx, %rax
- jmp L(start)
-END (__mempcpy_sse2_unaligned)
-
-# ifdef SHARED
-ENTRY (__memcpy_chk_sse2_unaligned)
- cmpq %rdx, %rcx
- jb HIDDEN_JUMPTARGET (__chk_fail)
-END (__memcpy_chk_sse2_unaligned)
-# endif
-
-ENTRY(__memcpy_sse2_unaligned)
- movq %rdi, %rax
-L(start):
- testq %rdx, %rdx
- je L(return)
- cmpq $16, %rdx
- jbe L(less_16)
- movdqu (%rsi), %xmm8
- cmpq $32, %rdx
- movdqu %xmm8, (%rdi)
- movdqu -16(%rsi,%rdx), %xmm8
- movdqu %xmm8, -16(%rdi,%rdx)
- ja .L31
-L(return):
- ret
- .p2align 4,,10
- .p2align 4
-.L31:
- movdqu 16(%rsi), %xmm8
- cmpq $64, %rdx
- movdqu %xmm8, 16(%rdi)
- movdqu -32(%rsi,%rdx), %xmm8
- movdqu %xmm8, -32(%rdi,%rdx)
- jbe L(return)
- movdqu 32(%rsi), %xmm8
- cmpq $128, %rdx
- movdqu %xmm8, 32(%rdi)
- movdqu -48(%rsi,%rdx), %xmm8
- movdqu %xmm8, -48(%rdi,%rdx)
- movdqu 48(%rsi), %xmm8
- movdqu %xmm8, 48(%rdi)
- movdqu -64(%rsi,%rdx), %xmm8
- movdqu %xmm8, -64(%rdi,%rdx)
- jbe L(return)
- leaq 64(%rdi), %rcx
- addq %rdi, %rdx
- andq $-64, %rdx
- andq $-64, %rcx
- movq %rcx, %r11
- subq %rdi, %r11
- addq %r11, %rsi
- cmpq %rdx, %rcx
- je L(return)
- movq %rsi, %r10
- subq %rcx, %r10
- leaq 16(%r10), %r9
- leaq 32(%r10), %r8
- leaq 48(%r10), %r11
- .p2align 4,,10
- .p2align 4
-L(loop):
- movdqu (%rcx,%r10), %xmm8
- movdqa %xmm8, (%rcx)
- movdqu (%rcx,%r9), %xmm8
- movdqa %xmm8, 16(%rcx)
- movdqu (%rcx,%r8), %xmm8
- movdqa %xmm8, 32(%rcx)
- movdqu (%rcx,%r11), %xmm8
- movdqa %xmm8, 48(%rcx)
- addq $64, %rcx
- cmpq %rcx, %rdx
- jne L(loop)
- ret
-L(less_16):
- testb $24, %dl
- jne L(between_9_16)
- testb $4, %dl
- .p2align 4,,5
- jne L(between_5_8)
- testq %rdx, %rdx
- .p2align 4,,2
- je L(return)
- movzbl (%rsi), %ecx
- testb $2, %dl
- movb %cl, (%rdi)
- je L(return)
- movzwl -2(%rsi,%rdx), %ecx
- movw %cx, -2(%rdi,%rdx)
- ret
-L(between_9_16):
- movq (%rsi), %rcx
- movq %rcx, (%rdi)
- movq -8(%rsi,%rdx), %rcx
- movq %rcx, -8(%rdi,%rdx)
- ret
-L(between_5_8):
- movl (%rsi), %ecx
- movl %ecx, (%rdi)
- movl -4(%rsi,%rdx), %ecx
- movl %ecx, -4(%rdi,%rdx)
- ret
-END(__memcpy_sse2_unaligned)
-
-#endif
diff --git a/sysdeps/x86_64/memcpy.S b/sysdeps/x86_64/multiarch/memcpy-sse2.S
similarity index 94%
copy from sysdeps/x86_64/memcpy.S
copy to sysdeps/x86_64/multiarch/memcpy-sse2.S
index f6e3d93..faf8d59 100644
--- a/sysdeps/x86_64/memcpy.S
+++ b/sysdeps/x86_64/multiarch/memcpy-sse2.S
@@ -30,12 +30,6 @@
# define RETVAL (0)
#else
# define RETVAL (-8)
-# if defined SHARED && !defined USE_MULTIARCH && IS_IN (libc)
-# define memcpy __memcpy
-# undef libc_hidden_builtin_def
-# define libc_hidden_builtin_def(name) \
- .globl __GI_memcpy; __GI_memcpy = __memcpy
-# endif
#endif
#define SAVE0 (RETVAL - 8)
#define SAVE1 (SAVE0 - 8)
@@ -44,16 +38,16 @@
.text
-#if defined PIC && IS_IN (libc)
-ENTRY_CHK (__memcpy_chk)
+#if defined SHARED && IS_IN (libc)
+ENTRY_CHK (__memcpy_chk_sse2)
cmpq %rdx, %rcx
jb HIDDEN_JUMPTARGET (__chk_fail)
-END_CHK (__memcpy_chk)
+END_CHK (__memcpy_chk_sse2)
#endif
-ENTRY(memcpy) /* (void *, const void*, size_t) */
+ENTRY(__memcpy_sse2) /* (void *, const void*, size_t) */
/* Handle tiny blocks. */
@@ -572,13 +566,4 @@ L(NTskip):
#endif /* IS_IN (libc) */
-END(memcpy)
-
-#ifndef USE_AS_MEMPCPY
-libc_hidden_builtin_def (memcpy)
-# if defined SHARED && !defined USE_MULTIARCH && IS_IN (libc)
-# undef memcpy
-# include <shlib-compat.h>
-versioned_symbol (libc, __memcpy, memcpy, GLIBC_2_14);
-# endif
-#endif
+END(__memcpy_sse2)
diff --git a/sysdeps/x86_64/multiarch/memcpy.S b/sysdeps/x86_64/multiarch/memcpy.S
index 8882590..efa028e 100644
--- a/sysdeps/x86_64/multiarch/memcpy.S
+++ b/sysdeps/x86_64/multiarch/memcpy.S
@@ -54,37 +54,25 @@ ENTRY(__new_memcpy)
2: ret
END(__new_memcpy)
-# undef ENTRY
-# define ENTRY(name) \
- .type __memcpy_sse2, @function; \
- .globl __memcpy_sse2; \
- .hidden __memcpy_sse2; \
- .p2align 4; \
- __memcpy_sse2: cfi_startproc; \
- CALL_MCOUNT
-# undef END
-# define END(name) \
- cfi_endproc; .size __memcpy_sse2, .-__memcpy_sse2
+# define memcpy __memcpy_sse2_unaligned
+# define __memcpy_chk __memcpy_chk_sse2_unaligned
+# define __mempcpy __mempcpy_sse2_unaligned
+# define __mempcpy_chk __mempcpy_chk_sse2_unaligned
-# undef ENTRY_CHK
-# define ENTRY_CHK(name) \
- .type __memcpy_chk_sse2, @function; \
- .globl __memcpy_chk_sse2; \
- .p2align 4; \
- __memcpy_chk_sse2: cfi_startproc; \
- CALL_MCOUNT
-# undef END_CHK
-# define END_CHK(name) \
- cfi_endproc; .size __memcpy_chk_sse2, .-__memcpy_chk_sse2
-
-# undef libc_hidden_builtin_def
/* It doesn't make sense to send libc-internal memcpy calls through a PLT.
The speedup we get from using SSSE3 instruction is likely eaten away
by the indirect call in the PLT. */
-# define libc_hidden_builtin_def(name) \
- .globl __GI_memcpy; __GI_memcpy = __memcpy_sse2
-
-versioned_symbol (libc, __new_memcpy, memcpy, GLIBC_2_14);
+ .globl __GI_memcpy
+ __GI_memcpy = __memcpy_sse2_unaligned
+ .globl __GI_mempcpy
+ __GI_mempcpy = __mempcpy_sse2_unaligned
+ .globl __GI___mempcpy
+ __GI___mempcpy = __mempcpy_sse2_unaligned
#endif
#include "../memcpy.S"
+
+#if defined SHARED && IS_IN (libc)
+# undef memcpy
+versioned_symbol (libc, __new_memcpy, memcpy, GLIBC_2_14);
+#endif
diff --git a/sysdeps/x86_64/multiarch/mempcpy-sse2.S b/sysdeps/x86_64/multiarch/mempcpy-sse2.S
new file mode 100644
index 0000000..e8bde29
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/mempcpy-sse2.S
@@ -0,0 +1,4 @@
+#define USE_AS_MEMPCPY
+#define __memcpy_sse2 __mempcpy_sse2
+#define __memcpy_chk_sse2 __mempcpy_chk_sse2
+#include "memcpy-sse2.S"
diff --git a/sysdeps/x86_64/multiarch/mempcpy.S b/sysdeps/x86_64/multiarch/mempcpy.S
index 1314d76..055c52c 100644
--- a/sysdeps/x86_64/multiarch/mempcpy.S
+++ b/sysdeps/x86_64/multiarch/mempcpy.S
@@ -51,39 +51,4 @@ ENTRY(__mempcpy)
lea __mempcpy_ssse3(%rip), %RAX_LP
2: ret
END(__mempcpy)
-
-# undef ENTRY
-# define ENTRY(name) \
- .type __mempcpy_sse2, @function; \
- .p2align 4; \
- .globl __mempcpy_sse2; \
- .hidden __mempcpy_sse2; \
- __mempcpy_sse2: cfi_startproc; \
- CALL_MCOUNT
-# undef END
-# define END(name) \
- cfi_endproc; .size __mempcpy_sse2, .-__mempcpy_sse2
-
-# undef ENTRY_CHK
-# define ENTRY_CHK(name) \
- .type __mempcpy_chk_sse2, @function; \
- .globl __mempcpy_chk_sse2; \
- .p2align 4; \
- __mempcpy_chk_sse2: cfi_startproc; \
- CALL_MCOUNT
-# undef END_CHK
-# define END_CHK(name) \
- cfi_endproc; .size __mempcpy_chk_sse2, .-__mempcpy_chk_sse2
-
-# undef libc_hidden_def
-# undef libc_hidden_builtin_def
-/* It doesn't make sense to send libc-internal mempcpy calls through a PLT.
- The speedup we get from using SSSE3 instruction is likely eaten away
- by the indirect call in the PLT. */
-# define libc_hidden_def(name) \
- .globl __GI_mempcpy; __GI_mempcpy = __mempcpy_sse2
-# define libc_hidden_builtin_def(name) \
- .globl __GI___mempcpy; __GI___mempcpy = __mempcpy_sse2
#endif
-
-#include "../mempcpy.S"
-----------------------------------------------------------------------
hooks/post-receive
--
GNU C Library master sources