2010-01-12 19:22:03 +00:00
|
|
|
/* memset with SSE2
|
2021-01-02 19:32:25 +00:00
|
|
|
Copyright (C) 2010-2021 Free Software Foundation, Inc.
|
2010-01-12 19:22:03 +00:00
|
|
|
Contributed by Intel Corporation.
|
|
|
|
This file is part of the GNU C Library.
|
|
|
|
|
|
|
|
The GNU C Library is free software; you can redistribute it and/or
|
|
|
|
modify it under the terms of the GNU Lesser General Public
|
|
|
|
License as published by the Free Software Foundation; either
|
|
|
|
version 2.1 of the License, or (at your option) any later version.
|
|
|
|
|
|
|
|
The GNU C Library is distributed in the hope that it will be useful,
|
|
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
Lesser General Public License for more details.
|
|
|
|
|
|
|
|
You should have received a copy of the GNU Lesser General Public
|
2012-02-09 23:18:22 +00:00
|
|
|
License along with the GNU C Library; if not, see
|
Prefer https to http for gnu.org and fsf.org URLs
Also, change sources.redhat.com to sourceware.org.
This patch was automatically generated by running the following shell
script, which uses GNU sed, and which avoids modifying files imported
from upstream:
sed -ri '
s,(http|ftp)(://(.*\.)?(gnu|fsf|sourceware)\.org($|[^.]|\.[^a-z])),https\2,g
s,(http|ftp)(://(.*\.)?)sources\.redhat\.com($|[^.]|\.[^a-z]),https\2sourceware.org\4,g
' \
$(find $(git ls-files) -prune -type f \
! -name '*.po' \
! -name 'ChangeLog*' \
! -path COPYING ! -path COPYING.LIB \
! -path manual/fdl-1.3.texi ! -path manual/lgpl-2.1.texi \
! -path manual/texinfo.tex ! -path scripts/config.guess \
! -path scripts/config.sub ! -path scripts/install-sh \
! -path scripts/mkinstalldirs ! -path scripts/move-if-change \
! -path INSTALL ! -path locale/programs/charmap-kw.h \
! -path po/libc.pot ! -path sysdeps/gnu/errlist.c \
! '(' -name configure \
-execdir test -f configure.ac -o -f configure.in ';' ')' \
! '(' -name preconfigure \
-execdir test -f preconfigure.ac ';' ')' \
-print)
and then by running 'make dist-prepare' to regenerate files built
from the altered files, and then executing the following to cleanup:
chmod a+x sysdeps/unix/sysv/linux/riscv/configure
# Omit irrelevant whitespace and comment-only changes,
# perhaps from a slightly-different Autoconf version.
git checkout -f \
sysdeps/csky/configure \
sysdeps/hppa/configure \
sysdeps/riscv/configure \
sysdeps/unix/sysv/linux/csky/configure
# Omit changes that caused a pre-commit check to fail like this:
# remote: *** error: sysdeps/powerpc/powerpc64/ppc-mcount.S: trailing lines
git checkout -f \
sysdeps/powerpc/powerpc64/ppc-mcount.S \
sysdeps/unix/sysv/linux/s390/s390-64/syscall.S
# Omit change that caused a pre-commit check to fail like this:
# remote: *** error: sysdeps/sparc/sparc64/multiarch/memcpy-ultra3.S: last line does not end in newline
git checkout -f sysdeps/sparc/sparc64/multiarch/memcpy-ultra3.S
2019-09-07 05:40:42 +00:00
|
|
|
<https://www.gnu.org/licenses/>. */
|
2010-01-12 19:22:03 +00:00
|
|
|
|
2014-11-24 09:33:45 +00:00
|
|
|
#if IS_IN (libc)
|
2010-01-12 19:22:03 +00:00
|
|
|
|
|
|
|
#include <sysdep.h>
|
|
|
|
#include "asm-syntax.h"
|
|
|
|
|
|
|
|
#define CFI_PUSH(REG) \
|
|
|
|
cfi_adjust_cfa_offset (4); \
|
|
|
|
cfi_rel_offset (REG, 0)
|
|
|
|
|
|
|
|
#define CFI_POP(REG) \
|
|
|
|
cfi_adjust_cfa_offset (-4); \
|
|
|
|
cfi_restore (REG)
|
|
|
|
|
|
|
|
#define PUSH(REG) pushl REG; CFI_PUSH (REG)
|
|
|
|
#define POP(REG) popl REG; CFI_POP (REG)
|
|
|
|
|
|
|
|
#ifdef USE_AS_BZERO
|
|
|
|
# define DEST PARMS
|
|
|
|
# define LEN DEST+4
|
|
|
|
# define SETRTNVAL
|
|
|
|
#else
|
|
|
|
# define DEST PARMS
|
|
|
|
# define CHR DEST+4
|
|
|
|
# define LEN CHR+4
|
|
|
|
# define SETRTNVAL movl DEST(%esp), %eax
|
|
|
|
#endif
|
|
|
|
|
2017-08-02 17:25:11 +00:00
|
|
|
#ifdef PIC
|
2010-01-12 19:22:03 +00:00
|
|
|
# define ENTRANCE PUSH (%ebx);
|
|
|
|
# define RETURN_END POP (%ebx); ret
|
|
|
|
# define RETURN RETURN_END; CFI_PUSH (%ebx)
|
|
|
|
# define PARMS 8 /* Preserve EBX. */
|
|
|
|
# define JMPTBL(I, B) I - B
|
|
|
|
|
|
|
|
/* Load an entry in a jump table into EBX and branch to it. TABLE is a
|
|
|
|
jump table with relative offsets. */
|
|
|
|
# define BRANCH_TO_JMPTBL_ENTRY(TABLE) \
|
|
|
|
/* We first load PC into EBX. */ \
|
2012-02-03 23:22:53 +00:00
|
|
|
SETUP_PIC_REG(bx); \
|
2010-01-12 19:22:03 +00:00
|
|
|
/* Get the address of the jump table. */ \
|
|
|
|
add $(TABLE - .), %ebx; \
|
|
|
|
/* Get the entry and convert the relative offset to the \
|
|
|
|
absolute address. */ \
|
|
|
|
add (%ebx,%ecx,4), %ebx; \
|
|
|
|
add %ecx, %edx; \
|
2013-08-21 17:48:48 +00:00
|
|
|
/* We loaded the jump table and adjusted EDX. Go. */ \
|
2018-07-18 14:57:50 +00:00
|
|
|
_CET_NOTRACK jmp *%ebx
|
2010-01-12 19:22:03 +00:00
|
|
|
#else
|
|
|
|
# define ENTRANCE
|
|
|
|
# define RETURN_END ret
|
|
|
|
# define RETURN RETURN_END
|
|
|
|
# define PARMS 4
|
|
|
|
# define JMPTBL(I, B) I
|
|
|
|
|
|
|
|
/* Branch to an entry in a jump table. TABLE is a jump table with
|
|
|
|
absolute offsets. */
|
|
|
|
# define BRANCH_TO_JMPTBL_ENTRY(TABLE) \
|
|
|
|
add %ecx, %edx; \
|
2018-07-18 14:57:50 +00:00
|
|
|
_CET_NOTRACK jmp *TABLE(,%ecx,4)
|
2010-01-12 19:22:03 +00:00
|
|
|
#endif
|
|
|
|
|
|
|
|
.section .text.sse2,"ax",@progbits
|
2014-11-24 09:33:45 +00:00
|
|
|
#if defined SHARED && IS_IN (libc) && !defined USE_AS_BZERO
|
2010-01-12 19:22:03 +00:00
|
|
|
ENTRY (__memset_chk_sse2)
|
|
|
|
movl 12(%esp), %eax
|
|
|
|
cmpl %eax, 16(%esp)
|
|
|
|
jb HIDDEN_JUMPTARGET (__chk_fail)
|
|
|
|
END (__memset_chk_sse2)
|
|
|
|
#endif
|
|
|
|
ENTRY (__memset_sse2)
|
|
|
|
ENTRANCE
|
|
|
|
|
|
|
|
movl LEN(%esp), %ecx
|
|
|
|
#ifdef USE_AS_BZERO
|
|
|
|
xor %eax, %eax
|
|
|
|
#else
|
|
|
|
movzbl CHR(%esp), %eax
|
|
|
|
movb %al, %ah
|
|
|
|
/* Fill the whole EAX with pattern. */
|
|
|
|
movl %eax, %edx
|
|
|
|
shl $16, %eax
|
|
|
|
or %edx, %eax
|
|
|
|
#endif
|
|
|
|
movl DEST(%esp), %edx
|
|
|
|
cmp $32, %ecx
|
|
|
|
jae L(32bytesormore)
|
|
|
|
|
|
|
|
L(write_less32bytes):
|
|
|
|
BRANCH_TO_JMPTBL_ENTRY (L(table_less_32bytes))
|
|
|
|
|
|
|
|
|
|
|
|
.pushsection .rodata.sse2,"a",@progbits
|
|
|
|
ALIGN (2)
|
|
|
|
L(table_less_32bytes):
|
|
|
|
.int JMPTBL (L(write_0bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_1bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_2bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_3bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_4bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_5bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_6bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_7bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_8bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_9bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_10bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_11bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_12bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_13bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_14bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_15bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_16bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_17bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_18bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_19bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_20bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_21bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_22bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_23bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_24bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_25bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_26bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_27bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_28bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_29bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_30bytes), L(table_less_32bytes))
|
|
|
|
.int JMPTBL (L(write_31bytes), L(table_less_32bytes))
|
|
|
|
.popsection
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(write_28bytes):
|
|
|
|
movl %eax, -28(%edx)
|
|
|
|
L(write_24bytes):
|
|
|
|
movl %eax, -24(%edx)
|
|
|
|
L(write_20bytes):
|
|
|
|
movl %eax, -20(%edx)
|
|
|
|
L(write_16bytes):
|
|
|
|
movl %eax, -16(%edx)
|
|
|
|
L(write_12bytes):
|
|
|
|
movl %eax, -12(%edx)
|
|
|
|
L(write_8bytes):
|
|
|
|
movl %eax, -8(%edx)
|
|
|
|
L(write_4bytes):
|
|
|
|
movl %eax, -4(%edx)
|
|
|
|
L(write_0bytes):
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(write_29bytes):
|
|
|
|
movl %eax, -29(%edx)
|
|
|
|
L(write_25bytes):
|
|
|
|
movl %eax, -25(%edx)
|
|
|
|
L(write_21bytes):
|
|
|
|
movl %eax, -21(%edx)
|
|
|
|
L(write_17bytes):
|
|
|
|
movl %eax, -17(%edx)
|
|
|
|
L(write_13bytes):
|
|
|
|
movl %eax, -13(%edx)
|
|
|
|
L(write_9bytes):
|
|
|
|
movl %eax, -9(%edx)
|
|
|
|
L(write_5bytes):
|
|
|
|
movl %eax, -5(%edx)
|
|
|
|
L(write_1bytes):
|
|
|
|
movb %al, -1(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(write_30bytes):
|
|
|
|
movl %eax, -30(%edx)
|
|
|
|
L(write_26bytes):
|
|
|
|
movl %eax, -26(%edx)
|
|
|
|
L(write_22bytes):
|
|
|
|
movl %eax, -22(%edx)
|
|
|
|
L(write_18bytes):
|
|
|
|
movl %eax, -18(%edx)
|
|
|
|
L(write_14bytes):
|
|
|
|
movl %eax, -14(%edx)
|
|
|
|
L(write_10bytes):
|
|
|
|
movl %eax, -10(%edx)
|
|
|
|
L(write_6bytes):
|
|
|
|
movl %eax, -6(%edx)
|
|
|
|
L(write_2bytes):
|
|
|
|
movw %ax, -2(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(write_31bytes):
|
|
|
|
movl %eax, -31(%edx)
|
|
|
|
L(write_27bytes):
|
|
|
|
movl %eax, -27(%edx)
|
|
|
|
L(write_23bytes):
|
|
|
|
movl %eax, -23(%edx)
|
|
|
|
L(write_19bytes):
|
|
|
|
movl %eax, -19(%edx)
|
|
|
|
L(write_15bytes):
|
|
|
|
movl %eax, -15(%edx)
|
|
|
|
L(write_11bytes):
|
|
|
|
movl %eax, -11(%edx)
|
|
|
|
L(write_7bytes):
|
|
|
|
movl %eax, -7(%edx)
|
|
|
|
L(write_3bytes):
|
|
|
|
movw %ax, -3(%edx)
|
|
|
|
movb %al, -1(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
/* ECX > 32 and EDX is 4 byte aligned. */
|
|
|
|
L(32bytesormore):
|
|
|
|
/* Fill xmm0 with the pattern. */
|
|
|
|
#ifdef USE_AS_BZERO
|
|
|
|
pxor %xmm0, %xmm0
|
|
|
|
#else
|
|
|
|
movd %eax, %xmm0
|
|
|
|
pshufd $0, %xmm0, %xmm0
|
|
|
|
#endif
|
|
|
|
testl $0xf, %edx
|
|
|
|
jz L(aligned_16)
|
|
|
|
/* ECX > 32 and EDX is not 16 byte aligned. */
|
|
|
|
L(not_aligned_16):
|
|
|
|
movdqu %xmm0, (%edx)
|
|
|
|
movl %edx, %eax
|
|
|
|
and $-16, %edx
|
|
|
|
add $16, %edx
|
|
|
|
sub %edx, %eax
|
|
|
|
add %eax, %ecx
|
|
|
|
movd %xmm0, %eax
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16):
|
|
|
|
cmp $128, %ecx
|
2010-02-25 02:11:35 +00:00
|
|
|
jae L(128bytesormore)
|
2010-01-12 19:22:03 +00:00
|
|
|
|
|
|
|
L(aligned_16_less128bytes):
|
|
|
|
BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes))
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(128bytesormore):
|
|
|
|
#ifdef SHARED_CACHE_SIZE
|
|
|
|
PUSH (%ebx)
|
|
|
|
mov $SHARED_CACHE_SIZE, %ebx
|
|
|
|
#else
|
2017-08-02 17:25:11 +00:00
|
|
|
# ifdef PIC
|
2012-02-03 23:22:53 +00:00
|
|
|
SETUP_PIC_REG(bx)
|
2010-01-12 19:22:03 +00:00
|
|
|
add $_GLOBAL_OFFSET_TABLE_, %ebx
|
|
|
|
mov __x86_shared_cache_size@GOTOFF(%ebx), %ebx
|
|
|
|
# else
|
|
|
|
PUSH (%ebx)
|
|
|
|
mov __x86_shared_cache_size, %ebx
|
|
|
|
# endif
|
|
|
|
#endif
|
|
|
|
cmp %ebx, %ecx
|
|
|
|
jae L(128bytesormore_nt_start)
|
|
|
|
|
2010-01-12 19:28:17 +00:00
|
|
|
|
2010-01-12 19:22:03 +00:00
|
|
|
#ifdef DATA_CACHE_SIZE
|
|
|
|
POP (%ebx)
|
2010-02-25 02:11:35 +00:00
|
|
|
# define RESTORE_EBX_STATE CFI_PUSH (%ebx)
|
2010-01-12 19:22:03 +00:00
|
|
|
cmp $DATA_CACHE_SIZE, %ecx
|
|
|
|
#else
|
2017-08-02 17:25:11 +00:00
|
|
|
# ifdef PIC
|
2010-02-25 02:11:35 +00:00
|
|
|
# define RESTORE_EBX_STATE
|
2012-02-03 23:22:53 +00:00
|
|
|
SETUP_PIC_REG(bx)
|
2010-01-12 19:22:03 +00:00
|
|
|
add $_GLOBAL_OFFSET_TABLE_, %ebx
|
|
|
|
cmp __x86_data_cache_size@GOTOFF(%ebx), %ecx
|
|
|
|
# else
|
|
|
|
POP (%ebx)
|
2010-02-25 02:11:35 +00:00
|
|
|
# define RESTORE_EBX_STATE CFI_PUSH (%ebx)
|
2010-01-12 19:22:03 +00:00
|
|
|
cmp __x86_data_cache_size, %ecx
|
|
|
|
# endif
|
|
|
|
#endif
|
|
|
|
|
|
|
|
jae L(128bytes_L2_normal)
|
|
|
|
subl $128, %ecx
|
|
|
|
L(128bytesormore_normal):
|
|
|
|
sub $128, %ecx
|
|
|
|
movdqa %xmm0, (%edx)
|
|
|
|
movdqa %xmm0, 0x10(%edx)
|
|
|
|
movdqa %xmm0, 0x20(%edx)
|
|
|
|
movdqa %xmm0, 0x30(%edx)
|
|
|
|
movdqa %xmm0, 0x40(%edx)
|
|
|
|
movdqa %xmm0, 0x50(%edx)
|
|
|
|
movdqa %xmm0, 0x60(%edx)
|
|
|
|
movdqa %xmm0, 0x70(%edx)
|
|
|
|
lea 128(%edx), %edx
|
2010-02-25 02:11:35 +00:00
|
|
|
jb L(128bytesless_normal)
|
2010-01-12 19:22:03 +00:00
|
|
|
|
|
|
|
|
|
|
|
sub $128, %ecx
|
|
|
|
movdqa %xmm0, (%edx)
|
|
|
|
movdqa %xmm0, 0x10(%edx)
|
|
|
|
movdqa %xmm0, 0x20(%edx)
|
|
|
|
movdqa %xmm0, 0x30(%edx)
|
|
|
|
movdqa %xmm0, 0x40(%edx)
|
|
|
|
movdqa %xmm0, 0x50(%edx)
|
|
|
|
movdqa %xmm0, 0x60(%edx)
|
|
|
|
movdqa %xmm0, 0x70(%edx)
|
|
|
|
lea 128(%edx), %edx
|
2010-02-25 02:11:35 +00:00
|
|
|
jae L(128bytesormore_normal)
|
2010-01-12 19:22:03 +00:00
|
|
|
|
|
|
|
L(128bytesless_normal):
|
2010-02-25 02:11:35 +00:00
|
|
|
add $128, %ecx
|
2010-01-12 19:22:03 +00:00
|
|
|
BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes))
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(128bytes_L2_normal):
|
|
|
|
prefetcht0 0x380(%edx)
|
|
|
|
prefetcht0 0x3c0(%edx)
|
|
|
|
sub $128, %ecx
|
|
|
|
movdqa %xmm0, (%edx)
|
|
|
|
movaps %xmm0, 0x10(%edx)
|
|
|
|
movaps %xmm0, 0x20(%edx)
|
|
|
|
movaps %xmm0, 0x30(%edx)
|
|
|
|
movaps %xmm0, 0x40(%edx)
|
|
|
|
movaps %xmm0, 0x50(%edx)
|
|
|
|
movaps %xmm0, 0x60(%edx)
|
|
|
|
movaps %xmm0, 0x70(%edx)
|
|
|
|
add $128, %edx
|
2010-01-12 19:28:17 +00:00
|
|
|
cmp $128, %ecx
|
2010-02-25 02:11:35 +00:00
|
|
|
jae L(128bytes_L2_normal)
|
2010-01-12 19:22:03 +00:00
|
|
|
|
|
|
|
L(128bytesless_L2_normal):
|
|
|
|
BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes))
|
|
|
|
|
2010-02-25 02:11:35 +00:00
|
|
|
RESTORE_EBX_STATE
|
2010-01-12 19:22:03 +00:00
|
|
|
L(128bytesormore_nt_start):
|
|
|
|
sub %ebx, %ecx
|
|
|
|
ALIGN (4)
|
|
|
|
L(128bytesormore_shared_cache_loop):
|
|
|
|
prefetcht0 0x3c0(%edx)
|
|
|
|
prefetcht0 0x380(%edx)
|
|
|
|
sub $0x80, %ebx
|
|
|
|
movdqa %xmm0, (%edx)
|
|
|
|
movdqa %xmm0, 0x10(%edx)
|
|
|
|
movdqa %xmm0, 0x20(%edx)
|
|
|
|
movdqa %xmm0, 0x30(%edx)
|
|
|
|
movdqa %xmm0, 0x40(%edx)
|
|
|
|
movdqa %xmm0, 0x50(%edx)
|
|
|
|
movdqa %xmm0, 0x60(%edx)
|
|
|
|
movdqa %xmm0, 0x70(%edx)
|
|
|
|
add $0x80, %edx
|
|
|
|
cmp $0x80, %ebx
|
2010-02-25 02:11:35 +00:00
|
|
|
jae L(128bytesormore_shared_cache_loop)
|
2010-01-12 19:22:03 +00:00
|
|
|
cmp $0x80, %ecx
|
|
|
|
jb L(shared_cache_loop_end)
|
|
|
|
ALIGN (4)
|
|
|
|
L(128bytesormore_nt):
|
|
|
|
sub $0x80, %ecx
|
|
|
|
movntdq %xmm0, (%edx)
|
|
|
|
movntdq %xmm0, 0x10(%edx)
|
|
|
|
movntdq %xmm0, 0x20(%edx)
|
|
|
|
movntdq %xmm0, 0x30(%edx)
|
|
|
|
movntdq %xmm0, 0x40(%edx)
|
|
|
|
movntdq %xmm0, 0x50(%edx)
|
|
|
|
movntdq %xmm0, 0x60(%edx)
|
|
|
|
movntdq %xmm0, 0x70(%edx)
|
|
|
|
add $0x80, %edx
|
|
|
|
cmp $0x80, %ecx
|
2010-02-25 02:11:35 +00:00
|
|
|
jae L(128bytesormore_nt)
|
2010-01-12 19:22:03 +00:00
|
|
|
sfence
|
|
|
|
L(shared_cache_loop_end):
|
2017-08-02 17:25:11 +00:00
|
|
|
#if defined DATA_CACHE_SIZE || !defined PIC
|
2010-01-12 19:22:03 +00:00
|
|
|
POP (%ebx)
|
|
|
|
#endif
|
|
|
|
BRANCH_TO_JMPTBL_ENTRY (L(table_16_128bytes))
|
|
|
|
|
|
|
|
|
|
|
|
.pushsection .rodata.sse2,"a",@progbits
|
|
|
|
ALIGN (2)
|
|
|
|
L(table_16_128bytes):
|
|
|
|
.int JMPTBL (L(aligned_16_0bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_1bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_2bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_3bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_4bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_5bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_6bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_7bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_8bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_9bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_10bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_11bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_12bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_13bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_14bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_15bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_16bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_17bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_18bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_19bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_20bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_21bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_22bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_23bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_24bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_25bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_26bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_27bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_28bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_29bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_30bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_31bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_32bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_33bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_34bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_35bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_36bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_37bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_38bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_39bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_40bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_41bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_42bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_43bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_44bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_45bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_46bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_47bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_48bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_49bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_50bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_51bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_52bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_53bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_54bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_55bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_56bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_57bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_58bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_59bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_60bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_61bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_62bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_63bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_64bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_65bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_66bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_67bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_68bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_69bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_70bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_71bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_72bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_73bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_74bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_75bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_76bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_77bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_78bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_79bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_80bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_81bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_82bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_83bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_84bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_85bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_86bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_87bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_88bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_89bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_90bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_91bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_92bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_93bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_94bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_95bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_96bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_97bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_98bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_99bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_100bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_101bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_102bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_103bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_104bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_105bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_106bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_107bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_108bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_109bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_110bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_111bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_112bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_113bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_114bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_115bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_116bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_117bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_118bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_119bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_120bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_121bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_122bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_123bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_124bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_125bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_126bytes), L(table_16_128bytes))
|
|
|
|
.int JMPTBL (L(aligned_16_127bytes), L(table_16_128bytes))
|
|
|
|
.popsection
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16_112bytes):
|
|
|
|
movdqa %xmm0, -112(%edx)
|
|
|
|
L(aligned_16_96bytes):
|
|
|
|
movdqa %xmm0, -96(%edx)
|
|
|
|
L(aligned_16_80bytes):
|
|
|
|
movdqa %xmm0, -80(%edx)
|
|
|
|
L(aligned_16_64bytes):
|
|
|
|
movdqa %xmm0, -64(%edx)
|
|
|
|
L(aligned_16_48bytes):
|
|
|
|
movdqa %xmm0, -48(%edx)
|
|
|
|
L(aligned_16_32bytes):
|
|
|
|
movdqa %xmm0, -32(%edx)
|
|
|
|
L(aligned_16_16bytes):
|
|
|
|
movdqa %xmm0, -16(%edx)
|
|
|
|
L(aligned_16_0bytes):
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16_113bytes):
|
|
|
|
movdqa %xmm0, -113(%edx)
|
|
|
|
L(aligned_16_97bytes):
|
|
|
|
movdqa %xmm0, -97(%edx)
|
|
|
|
L(aligned_16_81bytes):
|
|
|
|
movdqa %xmm0, -81(%edx)
|
|
|
|
L(aligned_16_65bytes):
|
|
|
|
movdqa %xmm0, -65(%edx)
|
|
|
|
L(aligned_16_49bytes):
|
|
|
|
movdqa %xmm0, -49(%edx)
|
|
|
|
L(aligned_16_33bytes):
|
|
|
|
movdqa %xmm0, -33(%edx)
|
|
|
|
L(aligned_16_17bytes):
|
|
|
|
movdqa %xmm0, -17(%edx)
|
|
|
|
L(aligned_16_1bytes):
|
|
|
|
movb %al, -1(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16_114bytes):
|
|
|
|
movdqa %xmm0, -114(%edx)
|
|
|
|
L(aligned_16_98bytes):
|
|
|
|
movdqa %xmm0, -98(%edx)
|
|
|
|
L(aligned_16_82bytes):
|
|
|
|
movdqa %xmm0, -82(%edx)
|
|
|
|
L(aligned_16_66bytes):
|
|
|
|
movdqa %xmm0, -66(%edx)
|
|
|
|
L(aligned_16_50bytes):
|
|
|
|
movdqa %xmm0, -50(%edx)
|
|
|
|
L(aligned_16_34bytes):
|
|
|
|
movdqa %xmm0, -34(%edx)
|
|
|
|
L(aligned_16_18bytes):
|
|
|
|
movdqa %xmm0, -18(%edx)
|
|
|
|
L(aligned_16_2bytes):
|
|
|
|
movw %ax, -2(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16_115bytes):
|
|
|
|
movdqa %xmm0, -115(%edx)
|
|
|
|
L(aligned_16_99bytes):
|
|
|
|
movdqa %xmm0, -99(%edx)
|
|
|
|
L(aligned_16_83bytes):
|
|
|
|
movdqa %xmm0, -83(%edx)
|
|
|
|
L(aligned_16_67bytes):
|
|
|
|
movdqa %xmm0, -67(%edx)
|
|
|
|
L(aligned_16_51bytes):
|
|
|
|
movdqa %xmm0, -51(%edx)
|
|
|
|
L(aligned_16_35bytes):
|
|
|
|
movdqa %xmm0, -35(%edx)
|
|
|
|
L(aligned_16_19bytes):
|
|
|
|
movdqa %xmm0, -19(%edx)
|
|
|
|
L(aligned_16_3bytes):
|
|
|
|
movw %ax, -3(%edx)
|
|
|
|
movb %al, -1(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16_116bytes):
|
|
|
|
movdqa %xmm0, -116(%edx)
|
|
|
|
L(aligned_16_100bytes):
|
|
|
|
movdqa %xmm0, -100(%edx)
|
|
|
|
L(aligned_16_84bytes):
|
|
|
|
movdqa %xmm0, -84(%edx)
|
|
|
|
L(aligned_16_68bytes):
|
|
|
|
movdqa %xmm0, -68(%edx)
|
|
|
|
L(aligned_16_52bytes):
|
|
|
|
movdqa %xmm0, -52(%edx)
|
|
|
|
L(aligned_16_36bytes):
|
|
|
|
movdqa %xmm0, -36(%edx)
|
|
|
|
L(aligned_16_20bytes):
|
|
|
|
movdqa %xmm0, -20(%edx)
|
|
|
|
L(aligned_16_4bytes):
|
|
|
|
movl %eax, -4(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16_117bytes):
|
|
|
|
movdqa %xmm0, -117(%edx)
|
|
|
|
L(aligned_16_101bytes):
|
|
|
|
movdqa %xmm0, -101(%edx)
|
|
|
|
L(aligned_16_85bytes):
|
|
|
|
movdqa %xmm0, -85(%edx)
|
|
|
|
L(aligned_16_69bytes):
|
|
|
|
movdqa %xmm0, -69(%edx)
|
|
|
|
L(aligned_16_53bytes):
|
|
|
|
movdqa %xmm0, -53(%edx)
|
|
|
|
L(aligned_16_37bytes):
|
|
|
|
movdqa %xmm0, -37(%edx)
|
|
|
|
L(aligned_16_21bytes):
|
|
|
|
movdqa %xmm0, -21(%edx)
|
|
|
|
L(aligned_16_5bytes):
|
|
|
|
movl %eax, -5(%edx)
|
|
|
|
movb %al, -1(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16_118bytes):
|
|
|
|
movdqa %xmm0, -118(%edx)
|
|
|
|
L(aligned_16_102bytes):
|
|
|
|
movdqa %xmm0, -102(%edx)
|
|
|
|
L(aligned_16_86bytes):
|
|
|
|
movdqa %xmm0, -86(%edx)
|
|
|
|
L(aligned_16_70bytes):
|
|
|
|
movdqa %xmm0, -70(%edx)
|
|
|
|
L(aligned_16_54bytes):
|
|
|
|
movdqa %xmm0, -54(%edx)
|
|
|
|
L(aligned_16_38bytes):
|
|
|
|
movdqa %xmm0, -38(%edx)
|
|
|
|
L(aligned_16_22bytes):
|
|
|
|
movdqa %xmm0, -22(%edx)
|
|
|
|
L(aligned_16_6bytes):
|
|
|
|
movl %eax, -6(%edx)
|
|
|
|
movw %ax, -2(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16_119bytes):
|
|
|
|
movdqa %xmm0, -119(%edx)
|
|
|
|
L(aligned_16_103bytes):
|
|
|
|
movdqa %xmm0, -103(%edx)
|
|
|
|
L(aligned_16_87bytes):
|
|
|
|
movdqa %xmm0, -87(%edx)
|
|
|
|
L(aligned_16_71bytes):
|
|
|
|
movdqa %xmm0, -71(%edx)
|
|
|
|
L(aligned_16_55bytes):
|
|
|
|
movdqa %xmm0, -55(%edx)
|
|
|
|
L(aligned_16_39bytes):
|
|
|
|
movdqa %xmm0, -39(%edx)
|
|
|
|
L(aligned_16_23bytes):
|
|
|
|
movdqa %xmm0, -23(%edx)
|
|
|
|
L(aligned_16_7bytes):
|
|
|
|
movl %eax, -7(%edx)
|
|
|
|
movw %ax, -3(%edx)
|
|
|
|
movb %al, -1(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16_120bytes):
|
|
|
|
movdqa %xmm0, -120(%edx)
|
|
|
|
L(aligned_16_104bytes):
|
|
|
|
movdqa %xmm0, -104(%edx)
|
|
|
|
L(aligned_16_88bytes):
|
|
|
|
movdqa %xmm0, -88(%edx)
|
|
|
|
L(aligned_16_72bytes):
|
|
|
|
movdqa %xmm0, -72(%edx)
|
|
|
|
L(aligned_16_56bytes):
|
|
|
|
movdqa %xmm0, -56(%edx)
|
|
|
|
L(aligned_16_40bytes):
|
|
|
|
movdqa %xmm0, -40(%edx)
|
|
|
|
L(aligned_16_24bytes):
|
|
|
|
movdqa %xmm0, -24(%edx)
|
|
|
|
L(aligned_16_8bytes):
|
|
|
|
movq %xmm0, -8(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16_121bytes):
|
|
|
|
movdqa %xmm0, -121(%edx)
|
|
|
|
L(aligned_16_105bytes):
|
|
|
|
movdqa %xmm0, -105(%edx)
|
|
|
|
L(aligned_16_89bytes):
|
|
|
|
movdqa %xmm0, -89(%edx)
|
|
|
|
L(aligned_16_73bytes):
|
|
|
|
movdqa %xmm0, -73(%edx)
|
|
|
|
L(aligned_16_57bytes):
|
|
|
|
movdqa %xmm0, -57(%edx)
|
|
|
|
L(aligned_16_41bytes):
|
|
|
|
movdqa %xmm0, -41(%edx)
|
|
|
|
L(aligned_16_25bytes):
|
|
|
|
movdqa %xmm0, -25(%edx)
|
|
|
|
L(aligned_16_9bytes):
|
|
|
|
movq %xmm0, -9(%edx)
|
|
|
|
movb %al, -1(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16_122bytes):
|
|
|
|
movdqa %xmm0, -122(%edx)
|
|
|
|
L(aligned_16_106bytes):
|
|
|
|
movdqa %xmm0, -106(%edx)
|
|
|
|
L(aligned_16_90bytes):
|
|
|
|
movdqa %xmm0, -90(%edx)
|
|
|
|
L(aligned_16_74bytes):
|
|
|
|
movdqa %xmm0, -74(%edx)
|
|
|
|
L(aligned_16_58bytes):
|
|
|
|
movdqa %xmm0, -58(%edx)
|
|
|
|
L(aligned_16_42bytes):
|
|
|
|
movdqa %xmm0, -42(%edx)
|
|
|
|
L(aligned_16_26bytes):
|
|
|
|
movdqa %xmm0, -26(%edx)
|
|
|
|
L(aligned_16_10bytes):
|
|
|
|
movq %xmm0, -10(%edx)
|
|
|
|
movw %ax, -2(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16_123bytes):
|
|
|
|
movdqa %xmm0, -123(%edx)
|
|
|
|
L(aligned_16_107bytes):
|
|
|
|
movdqa %xmm0, -107(%edx)
|
|
|
|
L(aligned_16_91bytes):
|
|
|
|
movdqa %xmm0, -91(%edx)
|
|
|
|
L(aligned_16_75bytes):
|
|
|
|
movdqa %xmm0, -75(%edx)
|
|
|
|
L(aligned_16_59bytes):
|
|
|
|
movdqa %xmm0, -59(%edx)
|
|
|
|
L(aligned_16_43bytes):
|
|
|
|
movdqa %xmm0, -43(%edx)
|
|
|
|
L(aligned_16_27bytes):
|
|
|
|
movdqa %xmm0, -27(%edx)
|
|
|
|
L(aligned_16_11bytes):
|
|
|
|
movq %xmm0, -11(%edx)
|
|
|
|
movw %ax, -3(%edx)
|
|
|
|
movb %al, -1(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16_124bytes):
|
|
|
|
movdqa %xmm0, -124(%edx)
|
|
|
|
L(aligned_16_108bytes):
|
|
|
|
movdqa %xmm0, -108(%edx)
|
|
|
|
L(aligned_16_92bytes):
|
|
|
|
movdqa %xmm0, -92(%edx)
|
|
|
|
L(aligned_16_76bytes):
|
|
|
|
movdqa %xmm0, -76(%edx)
|
|
|
|
L(aligned_16_60bytes):
|
|
|
|
movdqa %xmm0, -60(%edx)
|
|
|
|
L(aligned_16_44bytes):
|
|
|
|
movdqa %xmm0, -44(%edx)
|
|
|
|
L(aligned_16_28bytes):
|
|
|
|
movdqa %xmm0, -28(%edx)
|
|
|
|
L(aligned_16_12bytes):
|
|
|
|
movq %xmm0, -12(%edx)
|
|
|
|
movl %eax, -4(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16_125bytes):
|
|
|
|
movdqa %xmm0, -125(%edx)
|
|
|
|
L(aligned_16_109bytes):
|
|
|
|
movdqa %xmm0, -109(%edx)
|
|
|
|
L(aligned_16_93bytes):
|
|
|
|
movdqa %xmm0, -93(%edx)
|
|
|
|
L(aligned_16_77bytes):
|
|
|
|
movdqa %xmm0, -77(%edx)
|
|
|
|
L(aligned_16_61bytes):
|
|
|
|
movdqa %xmm0, -61(%edx)
|
|
|
|
L(aligned_16_45bytes):
|
|
|
|
movdqa %xmm0, -45(%edx)
|
|
|
|
L(aligned_16_29bytes):
|
|
|
|
movdqa %xmm0, -29(%edx)
|
|
|
|
L(aligned_16_13bytes):
|
|
|
|
movq %xmm0, -13(%edx)
|
|
|
|
movl %eax, -5(%edx)
|
|
|
|
movb %al, -1(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16_126bytes):
|
|
|
|
movdqa %xmm0, -126(%edx)
|
|
|
|
L(aligned_16_110bytes):
|
|
|
|
movdqa %xmm0, -110(%edx)
|
|
|
|
L(aligned_16_94bytes):
|
|
|
|
movdqa %xmm0, -94(%edx)
|
|
|
|
L(aligned_16_78bytes):
|
|
|
|
movdqa %xmm0, -78(%edx)
|
|
|
|
L(aligned_16_62bytes):
|
|
|
|
movdqa %xmm0, -62(%edx)
|
|
|
|
L(aligned_16_46bytes):
|
|
|
|
movdqa %xmm0, -46(%edx)
|
|
|
|
L(aligned_16_30bytes):
|
|
|
|
movdqa %xmm0, -30(%edx)
|
|
|
|
L(aligned_16_14bytes):
|
|
|
|
movq %xmm0, -14(%edx)
|
|
|
|
movl %eax, -6(%edx)
|
|
|
|
movw %ax, -2(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN
|
|
|
|
|
|
|
|
ALIGN (4)
|
|
|
|
L(aligned_16_127bytes):
|
|
|
|
movdqa %xmm0, -127(%edx)
|
|
|
|
L(aligned_16_111bytes):
|
|
|
|
movdqa %xmm0, -111(%edx)
|
|
|
|
L(aligned_16_95bytes):
|
|
|
|
movdqa %xmm0, -95(%edx)
|
|
|
|
L(aligned_16_79bytes):
|
|
|
|
movdqa %xmm0, -79(%edx)
|
|
|
|
L(aligned_16_63bytes):
|
|
|
|
movdqa %xmm0, -63(%edx)
|
|
|
|
L(aligned_16_47bytes):
|
|
|
|
movdqa %xmm0, -47(%edx)
|
|
|
|
L(aligned_16_31bytes):
|
|
|
|
movdqa %xmm0, -31(%edx)
|
|
|
|
L(aligned_16_15bytes):
|
|
|
|
movq %xmm0, -15(%edx)
|
|
|
|
movl %eax, -7(%edx)
|
|
|
|
movw %ax, -3(%edx)
|
|
|
|
movb %al, -1(%edx)
|
|
|
|
SETRTNVAL
|
|
|
|
RETURN_END
|
|
|
|
|
|
|
|
END (__memset_sse2)
|
|
|
|
|
|
|
|
#endif
|