* sysdeps/x86_64/cacheinfo.c: Comment out code added in support of
new memset. too high for the improvements. Implement bzero unconditionally for use in libc.
This commit is contained in:
parent
ac1cb5da08
commit
406f28dbe5
|
@ -1,7 +1,10 @@
|
|||
2007-10-17 Ulrich Drepper <drepper@redhat.com>
|
||||
|
||||
* sysdeps/x86_64/cacheinfo.c: Comment out code added in support of
|
||||
new memset.
|
||||
* sysdeps/x86_64/memset.S: Revert to old version for now. The cost is
|
||||
too high for the improvements.
|
||||
too high for the improvements. Implement bzero unconditionally for
|
||||
use in libc.
|
||||
|
||||
2007-10-17 Ulrich Drepper <drepper@redhat.com>
|
||||
Jakub Jelinek <jakub@redhat.com>
|
||||
|
|
|
@ -404,10 +404,13 @@ long int __x86_64_data_cache_size_half attribute_hidden = 32 * 1024 / 2;
|
|||
/* Shared cache size for use in memory and string routines, typically
|
||||
L2 or L3 size. */
|
||||
long int __x86_64_shared_cache_size_half attribute_hidden = 1024 * 1024 / 2;
|
||||
#ifdef NOT_USED_RIGHT_NOW
|
||||
long int __x86_64_shared_cache_size attribute_hidden = 1024 * 1024;
|
||||
#endif
|
||||
/* PREFETCHW support flag for use in memory and string routines. */
|
||||
int __x86_64_prefetchw attribute_hidden;
|
||||
|
||||
#ifdef NOT_USED_RIGHT_NOW
|
||||
/* Instructions preferred for memory and string routines.
|
||||
|
||||
0: Regular instructions
|
||||
|
@ -417,6 +420,7 @@ int __x86_64_prefetchw attribute_hidden;
|
|||
|
||||
*/
|
||||
int __x86_64_preferred_memory_instruction attribute_hidden;
|
||||
#endif
|
||||
|
||||
|
||||
static void
|
||||
|
@ -459,12 +463,14 @@ init_cacheinfo (void)
|
|||
: "=a" (eax), "=b" (ebx), "=c" (ecx), "=d" (edx)
|
||||
: "0" (1));
|
||||
|
||||
#ifdef NOT_USED_RIGHT_NOW
|
||||
/* Intel prefers SSSE3 instructions for memory/string rountines
|
||||
if they are avaiable. */
|
||||
if ((ecx & 0x200))
|
||||
__x86_64_preferred_memory_instruction = 3;
|
||||
else
|
||||
__x86_64_preferred_memory_instruction = 2;
|
||||
#endif
|
||||
|
||||
/* Figure out the number of logical threads that share the
|
||||
highest cache level. */
|
||||
|
@ -570,6 +576,8 @@ init_cacheinfo (void)
|
|||
if (shared > 0)
|
||||
{
|
||||
__x86_64_shared_cache_size_half = shared / 2;
|
||||
#ifdef NOT_USED_RIGHT_NOW
|
||||
__x86_64_shared_cache_size = shared;
|
||||
#endif
|
||||
}
|
||||
}
|
||||
|
|
|
@ -1,6 +1,6 @@
|
|||
/* memset/bzero -- set memory area to CH/0
|
||||
Optimized version for x86-64.
|
||||
Copyright (C) 2002, 2003, 2004, 2005 Free Software Foundation, Inc.
|
||||
Copyright (C) 2002, 2003, 2004, 2005, 2007 Free Software Foundation, Inc.
|
||||
This file is part of the GNU C Library.
|
||||
Contributed by Andreas Jaeger <aj@suse.de>.
|
||||
|
||||
|
@ -24,37 +24,35 @@
|
|||
#include "bp-sym.h"
|
||||
#include "bp-asm.h"
|
||||
|
||||
/* BEWARE: `#ifdef memset' means that memset is redefined as `bzero' */
|
||||
#define BZERO_P (defined memset)
|
||||
|
||||
/* This is somehow experimental and could made dependend on the cache
|
||||
size. */
|
||||
#define LARGE $120000
|
||||
|
||||
.text
|
||||
#if !BZERO_P && defined PIC && !defined NOT_IN_libc
|
||||
#ifndef NOT_IN_libc
|
||||
ENTRY(bzero)
|
||||
mov %rsi,%rdx /* Adjust parameter. */
|
||||
xorl %esi,%esi /* Fill with 0s. */
|
||||
jmp L(memset_entry)
|
||||
END(bzero)
|
||||
#endif
|
||||
|
||||
#if defined PIC && !defined NOT_IN_libc
|
||||
ENTRY (__memset_chk)
|
||||
cmpq %rdx, %rcx
|
||||
jb HIDDEN_JUMPTARGET (__chk_fail)
|
||||
END (__memset_chk)
|
||||
#endif
|
||||
ENTRY (memset)
|
||||
#if BZERO_P
|
||||
mov %rsi,%rdx /* Adjust parameter. */
|
||||
xorl %esi,%esi /* Fill with 0s. */
|
||||
#endif
|
||||
L(memset_entry):
|
||||
cmp $0x7,%rdx /* Check for small length. */
|
||||
mov %rdi,%rcx /* Save ptr as return value. */
|
||||
jbe 7f
|
||||
|
||||
#if BZERO_P
|
||||
mov %rsi,%r8 /* Just copy 0. */
|
||||
#else
|
||||
/* Populate 8 bit data to full 64-bit. */
|
||||
movabs $0x0101010101010101,%r8
|
||||
movzbl %sil,%eax
|
||||
imul %rax,%r8
|
||||
#endif
|
||||
test $0x7,%edi /* Check for alignment. */
|
||||
je 2f
|
||||
|
||||
|
@ -109,12 +107,8 @@ ENTRY (memset)
|
|||
dec %rdx
|
||||
jne 8b
|
||||
9:
|
||||
#if BZERO_P
|
||||
nop
|
||||
#else
|
||||
/* Load result (only if used as memset). */
|
||||
mov %rdi,%rax /* start address of destination is result */
|
||||
#endif
|
||||
retq
|
||||
|
||||
.p2align 4
|
||||
|
@ -135,11 +129,9 @@ ENTRY (memset)
|
|||
jmp 4b
|
||||
|
||||
END (memset)
|
||||
#if !BZERO_P
|
||||
libc_hidden_builtin_def (memset)
|
||||
#endif
|
||||
|
||||
#if !BZERO_P && defined PIC && !defined NOT_IN_libc
|
||||
#if defined PIC && !defined NOT_IN_libc
|
||||
strong_alias (__memset_chk, __memset_zero_constant_len_parameter)
|
||||
.section .gnu.warning.__memset_zero_constant_len_parameter
|
||||
.string "memset used with constant zero length parameter; this could be due to transposed parameters"
|
||||
|
|
Loading…
Reference in New Issue