2012-10-10 14:49:12 +00:00
|
|
|
/* Enumerate available IFUNC implementations of a function. x86-64 version.
|
2022-01-01 18:54:23 +00:00
|
|
|
Copyright (C) 2012-2022 Free Software Foundation, Inc.
|
2012-10-10 14:49:12 +00:00
|
|
|
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
|
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/>. */
|
2012-10-10 14:49:12 +00:00
|
|
|
|
|
|
|
|
#include <assert.h>
|
|
|
|
|
#include <string.h>
|
|
|
|
|
#include <wchar.h>
|
|
|
|
|
#include <ifunc-impl-list.h>
|
2015-12-18 23:47:28 +00:00
|
|
|
#include <sysdep.h>
|
2012-10-10 14:49:12 +00:00
|
|
|
#include "init-arch.h"
|
|
|
|
|
|
|
|
|
|
/* Fill ARRAY of MAX elements with IFUNC implementations for function
|
|
|
|
|
NAME supported on target machine and return the number of valid
|
|
|
|
|
entries. */
|
|
|
|
|
|
|
|
|
|
size_t
|
|
|
|
|
__libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
|
|
size_t max)
|
|
|
|
|
{
|
2022-06-10 16:13:29 +00:00
|
|
|
size_t i = max;
|
2012-10-10 14:49:12 +00:00
|
|
|
|
2021-10-27 00:43:18 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/memcmpeq.c. */
|
|
|
|
|
IFUNC_IMPL (i, name, __memcmpeq,
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __memcmpeq,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__memcmpeq_avx2)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __memcmpeq,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__memcmpeq_avx2_rtm)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __memcmpeq,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__memcmpeq_evex)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __memcmpeq, 1, __memcmpeq_sse2))
|
|
|
|
|
|
2017-06-09 12:53:45 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/memchr.c. */
|
2017-06-09 12:13:15 +00:00
|
|
|
IFUNC_IMPL (i, name, memchr,
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, memchr,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
2017-06-09 12:13:15 +00:00
|
|
|
__memchr_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__memchr_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__memchr_evex)
|
x86: Add EVEX optimized memchr family not safe for RTM
No bug.
This commit adds a new implementation for EVEX memchr that is not safe
for RTM because it uses vzeroupper. The benefit is that by using
ymm0-ymm15 it can use vpcmpeq and vpternlogd in the 4x loop which is
faster than the RTM safe version which cannot use vpcmpeq because
there is no EVEX encoding for the instruction. All parts of the
implementation aside from the 4x loop are the same for the two
versions and the optimization is only relevant for large sizes.
Tigerlake:
size , algn , Pos , Cur T , New T , Win , Dif
512 , 6 , 192 , 9.2 , 9.04 , no-RTM , 0.16
512 , 7 , 224 , 9.19 , 8.98 , no-RTM , 0.21
2048 , 0 , 256 , 10.74 , 10.54 , no-RTM , 0.2
2048 , 0 , 512 , 14.81 , 14.87 , RTM , 0.06
2048 , 0 , 1024 , 22.97 , 22.57 , no-RTM , 0.4
2048 , 0 , 2048 , 37.49 , 34.51 , no-RTM , 2.98 <--
Icelake:
size , algn , Pos , Cur T , New T , Win , Dif
512 , 6 , 192 , 7.6 , 7.3 , no-RTM , 0.3
512 , 7 , 224 , 7.63 , 7.27 , no-RTM , 0.36
2048 , 0 , 256 , 8.48 , 8.38 , no-RTM , 0.1
2048 , 0 , 512 , 11.57 , 11.42 , no-RTM , 0.15
2048 , 0 , 1024 , 17.92 , 17.38 , no-RTM , 0.54
2048 , 0 , 2048 , 30.37 , 27.34 , no-RTM , 3.03 <--
test-memchr, test-wmemchr, and test-rawmemchr are all passing.
Signed-off-by: Noah Goldstein <goldstein.w.n@gmail.com>
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
2021-05-04 23:02:40 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__memchr_evex_rtm)
|
2017-06-09 12:13:15 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memchr, 1, __memchr_sse2))
|
|
|
|
|
|
2017-06-15 15:49:43 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/memcmp.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, memcmp,
|
2017-06-05 19:52:41 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memcmp,
|
2020-06-29 23:36:08 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
2021-05-17 17:56:52 +00:00
|
|
|
&& CPU_FEATURE_USABLE (BMI2)
|
2020-06-29 23:36:08 +00:00
|
|
|
&& CPU_FEATURE_USABLE (MOVBE)),
|
2017-06-05 19:52:41 +00:00
|
|
|
__memcmp_avx2_movbe)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memcmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
2021-05-17 17:56:52 +00:00
|
|
|
&& CPU_FEATURE_USABLE (BMI2)
|
2021-03-05 15:26:42 +00:00
|
|
|
&& CPU_FEATURE_USABLE (MOVBE)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__memcmp_avx2_movbe_rtm)
|
2021-03-05 15:20:28 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memcmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
2021-05-17 17:56:52 +00:00
|
|
|
&& CPU_FEATURE_USABLE (BMI2)
|
2021-03-05 15:20:28 +00:00
|
|
|
&& CPU_FEATURE_USABLE (MOVBE)),
|
|
|
|
|
__memcmp_evex_movbe)
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memcmp, 1, __memcmp_sse2))
|
|
|
|
|
|
2017-07-10 11:44:24 +00:00
|
|
|
#ifdef SHARED
|
2017-06-14 19:10:57 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/memmove_chk.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, __memmove_chk,
|
2016-01-15 21:49:45 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memmove_chk,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512F),
|
2016-01-15 21:49:45 +00:00
|
|
|
__memmove_chk_avx512_no_vzeroupper)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memmove_chk,
|
2021-03-07 17:45:23 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
2016-06-08 20:57:50 +00:00
|
|
|
__memmove_chk_avx512_unaligned)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memmove_chk,
|
2021-03-07 17:45:23 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
__memmove_chk_avx512_unaligned_erms)
|
2015-08-13 10:38:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memmove_chk,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX),
|
2014-07-14 04:02:52 +00:00
|
|
|
__memmove_chk_avx_unaligned)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memmove_chk,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX),
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
__memmove_chk_avx_unaligned_erms)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memmove_chk,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__memmove_chk_avx_unaligned_rtm)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __memmove_chk,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__memmove_chk_avx_unaligned_erms_rtm)
|
2021-03-05 14:46:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memmove_chk,
|
|
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
|
|
|
|
__memmove_chk_evex_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __memmove_chk,
|
|
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
|
|
|
|
__memmove_chk_evex_unaligned_erms)
|
2015-08-13 10:38:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memmove_chk,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (SSSE3),
|
2012-10-10 14:49:12 +00:00
|
|
|
__memmove_chk_ssse3)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memmove_chk, 1,
|
2016-06-08 20:57:50 +00:00
|
|
|
__memmove_chk_sse2_unaligned)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memmove_chk, 1,
|
2017-06-14 19:10:57 +00:00
|
|
|
__memmove_chk_sse2_unaligned_erms)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __memmove_chk, 1,
|
|
|
|
|
__memmove_chk_erms))
|
2017-07-10 11:44:24 +00:00
|
|
|
#endif
|
2012-10-10 14:49:12 +00:00
|
|
|
|
2017-06-14 19:10:57 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/memmove.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, memmove,
|
2015-08-13 10:38:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memmove,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX),
|
2014-07-14 04:02:52 +00:00
|
|
|
__memmove_avx_unaligned)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memmove,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX),
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
__memmove_avx_unaligned_erms)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memmove,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__memmove_avx_unaligned_rtm)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, memmove,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__memmove_avx_unaligned_erms_rtm)
|
2021-03-05 14:46:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memmove,
|
|
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
|
|
|
|
__memmove_evex_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, memmove,
|
|
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
|
|
|
|
__memmove_evex_unaligned_erms)
|
2016-01-15 21:49:45 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memmove,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512F),
|
2016-01-15 21:49:45 +00:00
|
|
|
__memmove_avx512_no_vzeroupper)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memmove,
|
2021-03-07 17:45:23 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
2016-06-08 20:57:50 +00:00
|
|
|
__memmove_avx512_unaligned)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memmove,
|
2021-03-07 17:45:23 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
__memmove_avx512_unaligned_erms)
|
2020-06-29 23:36:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memmove, CPU_FEATURE_USABLE (SSSE3),
|
2012-10-10 14:49:12 +00:00
|
|
|
__memmove_ssse3)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memmove, 1, __memmove_erms)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, memmove, 1,
|
2016-06-08 20:57:50 +00:00
|
|
|
__memmove_sse2_unaligned)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memmove, 1,
|
2016-06-08 20:57:50 +00:00
|
|
|
__memmove_sse2_unaligned_erms))
|
2012-10-10 14:49:12 +00:00
|
|
|
|
2017-06-09 12:53:45 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/memrchr.c. */
|
2017-06-09 12:44:30 +00:00
|
|
|
IFUNC_IMPL (i, name, memrchr,
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, memrchr,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
2017-06-09 12:44:30 +00:00
|
|
|
__memrchr_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memrchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__memrchr_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memrchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
|
|
__memrchr_evex)
|
|
|
|
|
|
2017-06-09 12:44:30 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memrchr, 1, __memrchr_sse2))
|
|
|
|
|
|
2017-07-10 11:44:24 +00:00
|
|
|
#ifdef SHARED
|
2017-06-15 15:33:25 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/memset_chk.c. */
|
2014-06-20 21:52:29 +00:00
|
|
|
IFUNC_IMPL (i, name, __memset_chk,
|
2017-06-15 15:33:25 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memset_chk, 1,
|
|
|
|
|
__memset_chk_erms)
|
Add x86-64 memset with unaligned store and rep stosb
Implement x86-64 memset with unaligned store and rep movsb. Support
16-byte, 32-byte and 64-byte vector register sizes. A single file
provides 2 implementations of memset, one with rep stosb and the other
without rep stosb. They share the same codes when size is between 2
times of vector register size and REP_STOSB_THRESHOLD which defaults
to 2KB.
Key features:
1. Use overlapping store to avoid branch.
2. For size <= 4 times of vector register size, fully unroll the loop.
3. For size > 4 times of vector register size, store 4 times of vector
register size at a time.
[BZ #19881]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memset-sse2-unaligned-erms, memset-avx2-unaligned-erms and
memset-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test __memset_chk_sse2_unaligned,
__memset_chk_sse2_unaligned_erms, __memset_chk_avx2_unaligned,
__memset_chk_avx2_unaligned_erms, __memset_chk_avx512_unaligned,
__memset_chk_avx512_unaligned_erms, __memset_sse2_unaligned,
__memset_sse2_unaligned_erms, __memset_erms,
__memset_avx2_unaligned, __memset_avx2_unaligned_erms,
__memset_avx512_unaligned_erms and __memset_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-sse2-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:
Likewise.
2016-03-31 17:05:51 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memset_chk, 1,
|
|
|
|
|
__memset_chk_sse2_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __memset_chk, 1,
|
|
|
|
|
__memset_chk_sse2_unaligned_erms)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __memset_chk,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
Add x86-64 memset with unaligned store and rep stosb
Implement x86-64 memset with unaligned store and rep movsb. Support
16-byte, 32-byte and 64-byte vector register sizes. A single file
provides 2 implementations of memset, one with rep stosb and the other
without rep stosb. They share the same codes when size is between 2
times of vector register size and REP_STOSB_THRESHOLD which defaults
to 2KB.
Key features:
1. Use overlapping store to avoid branch.
2. For size <= 4 times of vector register size, fully unroll the loop.
3. For size > 4 times of vector register size, store 4 times of vector
register size at a time.
[BZ #19881]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memset-sse2-unaligned-erms, memset-avx2-unaligned-erms and
memset-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test __memset_chk_sse2_unaligned,
__memset_chk_sse2_unaligned_erms, __memset_chk_avx2_unaligned,
__memset_chk_avx2_unaligned_erms, __memset_chk_avx512_unaligned,
__memset_chk_avx512_unaligned_erms, __memset_sse2_unaligned,
__memset_sse2_unaligned_erms, __memset_erms,
__memset_avx2_unaligned, __memset_avx2_unaligned_erms,
__memset_avx512_unaligned_erms and __memset_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-sse2-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:
Likewise.
2016-03-31 17:05:51 +00:00
|
|
|
__memset_chk_avx2_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __memset_chk,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
Add x86-64 memset with unaligned store and rep stosb
Implement x86-64 memset with unaligned store and rep movsb. Support
16-byte, 32-byte and 64-byte vector register sizes. A single file
provides 2 implementations of memset, one with rep stosb and the other
without rep stosb. They share the same codes when size is between 2
times of vector register size and REP_STOSB_THRESHOLD which defaults
to 2KB.
Key features:
1. Use overlapping store to avoid branch.
2. For size <= 4 times of vector register size, fully unroll the loop.
3. For size > 4 times of vector register size, store 4 times of vector
register size at a time.
[BZ #19881]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memset-sse2-unaligned-erms, memset-avx2-unaligned-erms and
memset-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test __memset_chk_sse2_unaligned,
__memset_chk_sse2_unaligned_erms, __memset_chk_avx2_unaligned,
__memset_chk_avx2_unaligned_erms, __memset_chk_avx512_unaligned,
__memset_chk_avx512_unaligned_erms, __memset_sse2_unaligned,
__memset_sse2_unaligned_erms, __memset_erms,
__memset_avx2_unaligned, __memset_avx2_unaligned_erms,
__memset_avx512_unaligned_erms and __memset_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-sse2-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:
Likewise.
2016-03-31 17:05:51 +00:00
|
|
|
__memset_chk_avx2_unaligned_erms)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memset_chk,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__memset_chk_avx2_unaligned_rtm)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __memset_chk,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__memset_chk_avx2_unaligned_erms_rtm)
|
2021-03-05 15:15:03 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memset_chk,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
2021-04-19 21:48:10 +00:00
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
2021-03-05 15:15:03 +00:00
|
|
|
__memset_chk_evex_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __memset_chk,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
2021-04-19 21:48:10 +00:00
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
2021-03-05 15:15:03 +00:00
|
|
|
__memset_chk_evex_unaligned_erms)
|
Add x86-64 memset with unaligned store and rep stosb
Implement x86-64 memset with unaligned store and rep movsb. Support
16-byte, 32-byte and 64-byte vector register sizes. A single file
provides 2 implementations of memset, one with rep stosb and the other
without rep stosb. They share the same codes when size is between 2
times of vector register size and REP_STOSB_THRESHOLD which defaults
to 2KB.
Key features:
1. Use overlapping store to avoid branch.
2. For size <= 4 times of vector register size, fully unroll the loop.
3. For size > 4 times of vector register size, store 4 times of vector
register size at a time.
[BZ #19881]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memset-sse2-unaligned-erms, memset-avx2-unaligned-erms and
memset-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test __memset_chk_sse2_unaligned,
__memset_chk_sse2_unaligned_erms, __memset_chk_avx2_unaligned,
__memset_chk_avx2_unaligned_erms, __memset_chk_avx512_unaligned,
__memset_chk_avx512_unaligned_erms, __memset_sse2_unaligned,
__memset_sse2_unaligned_erms, __memset_erms,
__memset_avx2_unaligned, __memset_avx2_unaligned_erms,
__memset_avx512_unaligned_erms and __memset_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-sse2-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:
Likewise.
2016-03-31 17:05:51 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memset_chk,
|
2021-03-07 17:44:18 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
2021-04-19 21:48:10 +00:00
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
Add x86-64 memset with unaligned store and rep stosb
Implement x86-64 memset with unaligned store and rep movsb. Support
16-byte, 32-byte and 64-byte vector register sizes. A single file
provides 2 implementations of memset, one with rep stosb and the other
without rep stosb. They share the same codes when size is between 2
times of vector register size and REP_STOSB_THRESHOLD which defaults
to 2KB.
Key features:
1. Use overlapping store to avoid branch.
2. For size <= 4 times of vector register size, fully unroll the loop.
3. For size > 4 times of vector register size, store 4 times of vector
register size at a time.
[BZ #19881]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memset-sse2-unaligned-erms, memset-avx2-unaligned-erms and
memset-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test __memset_chk_sse2_unaligned,
__memset_chk_sse2_unaligned_erms, __memset_chk_avx2_unaligned,
__memset_chk_avx2_unaligned_erms, __memset_chk_avx512_unaligned,
__memset_chk_avx512_unaligned_erms, __memset_sse2_unaligned,
__memset_sse2_unaligned_erms, __memset_erms,
__memset_avx2_unaligned, __memset_avx2_unaligned_erms,
__memset_avx512_unaligned_erms and __memset_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-sse2-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:
Likewise.
2016-03-31 17:05:51 +00:00
|
|
|
__memset_chk_avx512_unaligned_erms)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __memset_chk,
|
2021-03-07 17:44:18 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
2021-04-19 21:48:10 +00:00
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
Add x86-64 memset with unaligned store and rep stosb
Implement x86-64 memset with unaligned store and rep movsb. Support
16-byte, 32-byte and 64-byte vector register sizes. A single file
provides 2 implementations of memset, one with rep stosb and the other
without rep stosb. They share the same codes when size is between 2
times of vector register size and REP_STOSB_THRESHOLD which defaults
to 2KB.
Key features:
1. Use overlapping store to avoid branch.
2. For size <= 4 times of vector register size, fully unroll the loop.
3. For size > 4 times of vector register size, store 4 times of vector
register size at a time.
[BZ #19881]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memset-sse2-unaligned-erms, memset-avx2-unaligned-erms and
memset-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test __memset_chk_sse2_unaligned,
__memset_chk_sse2_unaligned_erms, __memset_chk_avx2_unaligned,
__memset_chk_avx2_unaligned_erms, __memset_chk_avx512_unaligned,
__memset_chk_avx512_unaligned_erms, __memset_sse2_unaligned,
__memset_sse2_unaligned_erms, __memset_erms,
__memset_avx2_unaligned, __memset_avx2_unaligned_erms,
__memset_avx512_unaligned_erms and __memset_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-sse2-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:
Likewise.
2016-03-31 17:05:51 +00:00
|
|
|
__memset_chk_avx512_unaligned)
|
2015-12-18 23:47:28 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memset_chk,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512F),
|
2015-12-18 23:47:28 +00:00
|
|
|
__memset_chk_avx512_no_vzeroupper)
|
|
|
|
|
)
|
2017-07-10 11:44:24 +00:00
|
|
|
#endif
|
2014-06-20 21:52:29 +00:00
|
|
|
|
2017-06-15 15:33:25 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/memset.c. */
|
2014-06-20 21:52:29 +00:00
|
|
|
IFUNC_IMPL (i, name, memset,
|
Add x86-64 memset with unaligned store and rep stosb
Implement x86-64 memset with unaligned store and rep movsb. Support
16-byte, 32-byte and 64-byte vector register sizes. A single file
provides 2 implementations of memset, one with rep stosb and the other
without rep stosb. They share the same codes when size is between 2
times of vector register size and REP_STOSB_THRESHOLD which defaults
to 2KB.
Key features:
1. Use overlapping store to avoid branch.
2. For size <= 4 times of vector register size, fully unroll the loop.
3. For size > 4 times of vector register size, store 4 times of vector
register size at a time.
[BZ #19881]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memset-sse2-unaligned-erms, memset-avx2-unaligned-erms and
memset-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test __memset_chk_sse2_unaligned,
__memset_chk_sse2_unaligned_erms, __memset_chk_avx2_unaligned,
__memset_chk_avx2_unaligned_erms, __memset_chk_avx512_unaligned,
__memset_chk_avx512_unaligned_erms, __memset_sse2_unaligned,
__memset_sse2_unaligned_erms, __memset_erms,
__memset_avx2_unaligned, __memset_avx2_unaligned_erms,
__memset_avx512_unaligned_erms and __memset_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-sse2-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:
Likewise.
2016-03-31 17:05:51 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memset, 1,
|
|
|
|
|
__memset_sse2_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, memset, 1,
|
|
|
|
|
__memset_sse2_unaligned_erms)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, memset, 1, __memset_erms)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, memset,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
Add x86-64 memset with unaligned store and rep stosb
Implement x86-64 memset with unaligned store and rep movsb. Support
16-byte, 32-byte and 64-byte vector register sizes. A single file
provides 2 implementations of memset, one with rep stosb and the other
without rep stosb. They share the same codes when size is between 2
times of vector register size and REP_STOSB_THRESHOLD which defaults
to 2KB.
Key features:
1. Use overlapping store to avoid branch.
2. For size <= 4 times of vector register size, fully unroll the loop.
3. For size > 4 times of vector register size, store 4 times of vector
register size at a time.
[BZ #19881]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memset-sse2-unaligned-erms, memset-avx2-unaligned-erms and
memset-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test __memset_chk_sse2_unaligned,
__memset_chk_sse2_unaligned_erms, __memset_chk_avx2_unaligned,
__memset_chk_avx2_unaligned_erms, __memset_chk_avx512_unaligned,
__memset_chk_avx512_unaligned_erms, __memset_sse2_unaligned,
__memset_sse2_unaligned_erms, __memset_erms,
__memset_avx2_unaligned, __memset_avx2_unaligned_erms,
__memset_avx512_unaligned_erms and __memset_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-sse2-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:
Likewise.
2016-03-31 17:05:51 +00:00
|
|
|
__memset_avx2_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, memset,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
Add x86-64 memset with unaligned store and rep stosb
Implement x86-64 memset with unaligned store and rep movsb. Support
16-byte, 32-byte and 64-byte vector register sizes. A single file
provides 2 implementations of memset, one with rep stosb and the other
without rep stosb. They share the same codes when size is between 2
times of vector register size and REP_STOSB_THRESHOLD which defaults
to 2KB.
Key features:
1. Use overlapping store to avoid branch.
2. For size <= 4 times of vector register size, fully unroll the loop.
3. For size > 4 times of vector register size, store 4 times of vector
register size at a time.
[BZ #19881]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memset-sse2-unaligned-erms, memset-avx2-unaligned-erms and
memset-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test __memset_chk_sse2_unaligned,
__memset_chk_sse2_unaligned_erms, __memset_chk_avx2_unaligned,
__memset_chk_avx2_unaligned_erms, __memset_chk_avx512_unaligned,
__memset_chk_avx512_unaligned_erms, __memset_sse2_unaligned,
__memset_sse2_unaligned_erms, __memset_erms,
__memset_avx2_unaligned, __memset_avx2_unaligned_erms,
__memset_avx512_unaligned_erms and __memset_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-sse2-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:
Likewise.
2016-03-31 17:05:51 +00:00
|
|
|
__memset_avx2_unaligned_erms)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memset,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__memset_avx2_unaligned_rtm)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, memset,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__memset_avx2_unaligned_erms_rtm)
|
2021-03-05 15:15:03 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memset,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
2021-04-19 21:48:10 +00:00
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
2021-03-05 15:15:03 +00:00
|
|
|
__memset_evex_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, memset,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
2021-04-19 21:48:10 +00:00
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
2021-03-05 15:15:03 +00:00
|
|
|
__memset_evex_unaligned_erms)
|
Add x86-64 memset with unaligned store and rep stosb
Implement x86-64 memset with unaligned store and rep movsb. Support
16-byte, 32-byte and 64-byte vector register sizes. A single file
provides 2 implementations of memset, one with rep stosb and the other
without rep stosb. They share the same codes when size is between 2
times of vector register size and REP_STOSB_THRESHOLD which defaults
to 2KB.
Key features:
1. Use overlapping store to avoid branch.
2. For size <= 4 times of vector register size, fully unroll the loop.
3. For size > 4 times of vector register size, store 4 times of vector
register size at a time.
[BZ #19881]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memset-sse2-unaligned-erms, memset-avx2-unaligned-erms and
memset-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test __memset_chk_sse2_unaligned,
__memset_chk_sse2_unaligned_erms, __memset_chk_avx2_unaligned,
__memset_chk_avx2_unaligned_erms, __memset_chk_avx512_unaligned,
__memset_chk_avx512_unaligned_erms, __memset_sse2_unaligned,
__memset_sse2_unaligned_erms, __memset_erms,
__memset_avx2_unaligned, __memset_avx2_unaligned_erms,
__memset_avx512_unaligned_erms and __memset_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-sse2-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:
Likewise.
2016-03-31 17:05:51 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memset,
|
2021-03-07 17:44:18 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
2021-04-19 21:48:10 +00:00
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
Add x86-64 memset with unaligned store and rep stosb
Implement x86-64 memset with unaligned store and rep movsb. Support
16-byte, 32-byte and 64-byte vector register sizes. A single file
provides 2 implementations of memset, one with rep stosb and the other
without rep stosb. They share the same codes when size is between 2
times of vector register size and REP_STOSB_THRESHOLD which defaults
to 2KB.
Key features:
1. Use overlapping store to avoid branch.
2. For size <= 4 times of vector register size, fully unroll the loop.
3. For size > 4 times of vector register size, store 4 times of vector
register size at a time.
[BZ #19881]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memset-sse2-unaligned-erms, memset-avx2-unaligned-erms and
memset-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test __memset_chk_sse2_unaligned,
__memset_chk_sse2_unaligned_erms, __memset_chk_avx2_unaligned,
__memset_chk_avx2_unaligned_erms, __memset_chk_avx512_unaligned,
__memset_chk_avx512_unaligned_erms, __memset_sse2_unaligned,
__memset_sse2_unaligned_erms, __memset_erms,
__memset_avx2_unaligned, __memset_avx2_unaligned_erms,
__memset_avx512_unaligned_erms and __memset_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-sse2-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:
Likewise.
2016-03-31 17:05:51 +00:00
|
|
|
__memset_avx512_unaligned_erms)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, memset,
|
2021-03-07 17:44:18 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
2021-04-19 21:48:10 +00:00
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
Add x86-64 memset with unaligned store and rep stosb
Implement x86-64 memset with unaligned store and rep movsb. Support
16-byte, 32-byte and 64-byte vector register sizes. A single file
provides 2 implementations of memset, one with rep stosb and the other
without rep stosb. They share the same codes when size is between 2
times of vector register size and REP_STOSB_THRESHOLD which defaults
to 2KB.
Key features:
1. Use overlapping store to avoid branch.
2. For size <= 4 times of vector register size, fully unroll the loop.
3. For size > 4 times of vector register size, store 4 times of vector
register size at a time.
[BZ #19881]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memset-sse2-unaligned-erms, memset-avx2-unaligned-erms and
memset-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test __memset_chk_sse2_unaligned,
__memset_chk_sse2_unaligned_erms, __memset_chk_avx2_unaligned,
__memset_chk_avx2_unaligned_erms, __memset_chk_avx512_unaligned,
__memset_chk_avx512_unaligned_erms, __memset_sse2_unaligned,
__memset_sse2_unaligned_erms, __memset_erms,
__memset_avx2_unaligned, __memset_avx2_unaligned_erms,
__memset_avx512_unaligned_erms and __memset_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-sse2-unaligned-erms.S:
Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S:
Likewise.
2016-03-31 17:05:51 +00:00
|
|
|
__memset_avx512_unaligned)
|
2015-12-18 23:47:28 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memset,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512F),
|
2015-12-18 23:47:28 +00:00
|
|
|
__memset_avx512_no_vzeroupper)
|
|
|
|
|
)
|
2014-06-20 21:52:29 +00:00
|
|
|
|
2017-06-09 12:53:45 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/rawmemchr.c. */
|
2017-06-09 12:13:15 +00:00
|
|
|
IFUNC_IMPL (i, name, rawmemchr,
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, rawmemchr,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
2017-06-09 12:13:15 +00:00
|
|
|
__rawmemchr_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, rawmemchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__rawmemchr_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, rawmemchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__rawmemchr_evex)
|
x86: Add EVEX optimized memchr family not safe for RTM
No bug.
This commit adds a new implementation for EVEX memchr that is not safe
for RTM because it uses vzeroupper. The benefit is that by using
ymm0-ymm15 it can use vpcmpeq and vpternlogd in the 4x loop which is
faster than the RTM safe version which cannot use vpcmpeq because
there is no EVEX encoding for the instruction. All parts of the
implementation aside from the 4x loop are the same for the two
versions and the optimization is only relevant for large sizes.
Tigerlake:
size , algn , Pos , Cur T , New T , Win , Dif
512 , 6 , 192 , 9.2 , 9.04 , no-RTM , 0.16
512 , 7 , 224 , 9.19 , 8.98 , no-RTM , 0.21
2048 , 0 , 256 , 10.74 , 10.54 , no-RTM , 0.2
2048 , 0 , 512 , 14.81 , 14.87 , RTM , 0.06
2048 , 0 , 1024 , 22.97 , 22.57 , no-RTM , 0.4
2048 , 0 , 2048 , 37.49 , 34.51 , no-RTM , 2.98 <--
Icelake:
size , algn , Pos , Cur T , New T , Win , Dif
512 , 6 , 192 , 7.6 , 7.3 , no-RTM , 0.3
512 , 7 , 224 , 7.63 , 7.27 , no-RTM , 0.36
2048 , 0 , 256 , 8.48 , 8.38 , no-RTM , 0.1
2048 , 0 , 512 , 11.57 , 11.42 , no-RTM , 0.15
2048 , 0 , 1024 , 17.92 , 17.38 , no-RTM , 0.54
2048 , 0 , 2048 , 30.37 , 27.34 , no-RTM , 3.03 <--
test-memchr, test-wmemchr, and test-rawmemchr are all passing.
Signed-off-by: Noah Goldstein <goldstein.w.n@gmail.com>
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
2021-05-04 23:02:40 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, rawmemchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__rawmemchr_evex_rtm)
|
2017-06-09 12:13:15 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, rawmemchr, 1, __rawmemchr_sse2))
|
|
|
|
|
|
2017-06-09 12:53:45 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strlen.c. */
|
x86-64: Optimize strlen/strnlen/wcslen/wcsnlen with AVX2
Optimize strlen/strnlen/wcslen/wcsnlen with AVX2 to check 32 bytes with
a single vector compare instruction. It is as fast as SSE2 versions for
size <= 16 bytes and up to 1X faster for or size > 16 bytes on Haswell.
Select AVX2 version on AVX2 machines where vzeroupper is preferred and
AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strlen-sse2, strnlen-sse2, strlen-avx2, strnlen-avx2,
wcslen-sse2, wcslen-avx2 and wcsnlen-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strlen_avx2,
__strlen_sse2, __strnlen_avx2, __strnlen_sse2, __wcslen_avx2,
__wcslen_sse2 and __wcsnlen_avx2.
* sysdeps/x86_64/multiarch/strlen-avx2.S: New file.
* sysdeps/x86_64/multiarch/strlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strlen.c: Likewise.
* sysdeps/x86_64/multiarch/strnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen.c: Likewise.
* sysdeps/x86_64/multiarch/wcslen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen.c: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen.c (OPTIMIZE (avx2)): New.
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX2 machines where
vzeroupper is preferred and AVX unaligned load is fast.
2017-06-09 12:18:03 +00:00
|
|
|
IFUNC_IMPL (i, name, strlen,
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, strlen,
|
2021-04-19 23:36:07 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
x86-64: Optimize strlen/strnlen/wcslen/wcsnlen with AVX2
Optimize strlen/strnlen/wcslen/wcsnlen with AVX2 to check 32 bytes with
a single vector compare instruction. It is as fast as SSE2 versions for
size <= 16 bytes and up to 1X faster for or size > 16 bytes on Haswell.
Select AVX2 version on AVX2 machines where vzeroupper is preferred and
AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strlen-sse2, strnlen-sse2, strlen-avx2, strnlen-avx2,
wcslen-sse2, wcslen-avx2 and wcsnlen-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strlen_avx2,
__strlen_sse2, __strnlen_avx2, __strnlen_sse2, __wcslen_avx2,
__wcslen_sse2 and __wcsnlen_avx2.
* sysdeps/x86_64/multiarch/strlen-avx2.S: New file.
* sysdeps/x86_64/multiarch/strlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strlen.c: Likewise.
* sysdeps/x86_64/multiarch/strnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen.c: Likewise.
* sysdeps/x86_64/multiarch/wcslen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen.c: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen.c (OPTIMIZE (avx2)): New.
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX2 machines where
vzeroupper is preferred and AVX unaligned load is fast.
2017-06-09 12:18:03 +00:00
|
|
|
__strlen_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strlen,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
2021-04-19 23:36:07 +00:00
|
|
|
&& CPU_FEATURE_USABLE (BMI2)
|
2021-03-05 15:26:42 +00:00
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__strlen_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strlen,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
2021-04-19 14:07:21 +00:00
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
2021-03-05 14:24:52 +00:00
|
|
|
__strlen_evex)
|
2022-02-28 00:39:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strlen,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__strlen_evex512)
|
x86-64: Optimize strlen/strnlen/wcslen/wcsnlen with AVX2
Optimize strlen/strnlen/wcslen/wcsnlen with AVX2 to check 32 bytes with
a single vector compare instruction. It is as fast as SSE2 versions for
size <= 16 bytes and up to 1X faster for or size > 16 bytes on Haswell.
Select AVX2 version on AVX2 machines where vzeroupper is preferred and
AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strlen-sse2, strnlen-sse2, strlen-avx2, strnlen-avx2,
wcslen-sse2, wcslen-avx2 and wcsnlen-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strlen_avx2,
__strlen_sse2, __strnlen_avx2, __strnlen_sse2, __wcslen_avx2,
__wcslen_sse2 and __wcsnlen_avx2.
* sysdeps/x86_64/multiarch/strlen-avx2.S: New file.
* sysdeps/x86_64/multiarch/strlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strlen.c: Likewise.
* sysdeps/x86_64/multiarch/strnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen.c: Likewise.
* sysdeps/x86_64/multiarch/wcslen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen.c: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen.c (OPTIMIZE (avx2)): New.
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX2 machines where
vzeroupper is preferred and AVX unaligned load is fast.
2017-06-09 12:18:03 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strlen, 1, __strlen_sse2))
|
|
|
|
|
|
2017-06-09 12:53:45 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strnlen.c. */
|
x86-64: Optimize strlen/strnlen/wcslen/wcsnlen with AVX2
Optimize strlen/strnlen/wcslen/wcsnlen with AVX2 to check 32 bytes with
a single vector compare instruction. It is as fast as SSE2 versions for
size <= 16 bytes and up to 1X faster for or size > 16 bytes on Haswell.
Select AVX2 version on AVX2 machines where vzeroupper is preferred and
AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strlen-sse2, strnlen-sse2, strlen-avx2, strnlen-avx2,
wcslen-sse2, wcslen-avx2 and wcsnlen-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strlen_avx2,
__strlen_sse2, __strnlen_avx2, __strnlen_sse2, __wcslen_avx2,
__wcslen_sse2 and __wcsnlen_avx2.
* sysdeps/x86_64/multiarch/strlen-avx2.S: New file.
* sysdeps/x86_64/multiarch/strlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strlen.c: Likewise.
* sysdeps/x86_64/multiarch/strnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen.c: Likewise.
* sysdeps/x86_64/multiarch/wcslen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen.c: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen.c (OPTIMIZE (avx2)): New.
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX2 machines where
vzeroupper is preferred and AVX unaligned load is fast.
2017-06-09 12:18:03 +00:00
|
|
|
IFUNC_IMPL (i, name, strnlen,
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, strnlen,
|
2021-04-19 23:36:07 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
x86-64: Optimize strlen/strnlen/wcslen/wcsnlen with AVX2
Optimize strlen/strnlen/wcslen/wcsnlen with AVX2 to check 32 bytes with
a single vector compare instruction. It is as fast as SSE2 versions for
size <= 16 bytes and up to 1X faster for or size > 16 bytes on Haswell.
Select AVX2 version on AVX2 machines where vzeroupper is preferred and
AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strlen-sse2, strnlen-sse2, strlen-avx2, strnlen-avx2,
wcslen-sse2, wcslen-avx2 and wcsnlen-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strlen_avx2,
__strlen_sse2, __strnlen_avx2, __strnlen_sse2, __wcslen_avx2,
__wcslen_sse2 and __wcsnlen_avx2.
* sysdeps/x86_64/multiarch/strlen-avx2.S: New file.
* sysdeps/x86_64/multiarch/strlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strlen.c: Likewise.
* sysdeps/x86_64/multiarch/strnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen.c: Likewise.
* sysdeps/x86_64/multiarch/wcslen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen.c: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen.c (OPTIMIZE (avx2)): New.
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX2 machines where
vzeroupper is preferred and AVX unaligned load is fast.
2017-06-09 12:18:03 +00:00
|
|
|
__strnlen_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strnlen,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
2021-04-19 23:36:07 +00:00
|
|
|
&& CPU_FEATURE_USABLE (BMI2)
|
2021-03-05 15:26:42 +00:00
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__strnlen_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strnlen,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
2021-04-19 14:07:21 +00:00
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
2021-03-05 14:24:52 +00:00
|
|
|
__strnlen_evex)
|
2022-02-28 00:39:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strnlen,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__strnlen_evex512)
|
x86-64: Optimize strlen/strnlen/wcslen/wcsnlen with AVX2
Optimize strlen/strnlen/wcslen/wcsnlen with AVX2 to check 32 bytes with
a single vector compare instruction. It is as fast as SSE2 versions for
size <= 16 bytes and up to 1X faster for or size > 16 bytes on Haswell.
Select AVX2 version on AVX2 machines where vzeroupper is preferred and
AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strlen-sse2, strnlen-sse2, strlen-avx2, strnlen-avx2,
wcslen-sse2, wcslen-avx2 and wcsnlen-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strlen_avx2,
__strlen_sse2, __strnlen_avx2, __strnlen_sse2, __wcslen_avx2,
__wcslen_sse2 and __wcsnlen_avx2.
* sysdeps/x86_64/multiarch/strlen-avx2.S: New file.
* sysdeps/x86_64/multiarch/strlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strlen.c: Likewise.
* sysdeps/x86_64/multiarch/strnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen.c: Likewise.
* sysdeps/x86_64/multiarch/wcslen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen.c: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen.c (OPTIMIZE (avx2)): New.
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX2 machines where
vzeroupper is preferred and AVX unaligned load is fast.
2017-06-09 12:18:03 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strnlen, 1, __strnlen_sse2))
|
|
|
|
|
|
2017-07-09 18:38:37 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/stpncpy.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, stpncpy,
|
2020-06-29 23:36:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, stpncpy, CPU_FEATURE_USABLE (AVX2),
|
x86-64: Optimize strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2
Optimize x86-64 strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2.
It uses vector comparison as much as possible. In general, the larger the
source string, the greater performance gain observed, reaching speedups of
1.6x compared to SSE2 unaligned routines. Select AVX2 strcat/strncat,
strcpy/strncpy and stpcpy/stpncpy on AVX2 machines where vzeroupper is
preferred and AVX unaligned load is fast.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strcat-avx2, strncat-avx2, strcpy-avx2, strncpy-avx2,
stpcpy-avx2 and stpncpy-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c:
(__libc_ifunc_impl_list): Add tests for __strcat_avx2,
__strncat_avx2, __strcpy_avx2, __strncpy_avx2, __stpcpy_avx2
and __stpncpy_avx2.
* sysdeps/x86_64/multiarch/{ifunc-unaligned-ssse3.h =>
ifunc-strcpy.h}: rename header for a more generic name.
* sysdeps/x86_64/multiarch/ifunc-strcpy.h:
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
AVX unaligned load is fast and vzeroupper is preferred.
* sysdeps/x86_64/multiarch/stpcpy-avx2.S: New file
* sysdeps/x86_64/multiarch/stpncpy-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strcat-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strcpy-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strncat-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strncpy-avx2.S: Likewise
2018-10-08 13:59:50 +00:00
|
|
|
__stpncpy_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, stpncpy,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__stpncpy_avx2_rtm)
|
2021-03-05 14:36:50 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, stpncpy,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
|
|
__stpncpy_evex)
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, stpncpy, 1,
|
|
|
|
|
__stpncpy_sse2_unaligned)
|
2022-06-16 22:07:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, stpncpy, 1, __stpncpy_generic))
|
2012-10-10 14:49:12 +00:00
|
|
|
|
2017-07-09 18:38:37 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/stpcpy.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, stpcpy,
|
2020-06-29 23:36:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, stpcpy, CPU_FEATURE_USABLE (AVX2),
|
x86-64: Optimize strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2
Optimize x86-64 strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2.
It uses vector comparison as much as possible. In general, the larger the
source string, the greater performance gain observed, reaching speedups of
1.6x compared to SSE2 unaligned routines. Select AVX2 strcat/strncat,
strcpy/strncpy and stpcpy/stpncpy on AVX2 machines where vzeroupper is
preferred and AVX unaligned load is fast.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strcat-avx2, strncat-avx2, strcpy-avx2, strncpy-avx2,
stpcpy-avx2 and stpncpy-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c:
(__libc_ifunc_impl_list): Add tests for __strcat_avx2,
__strncat_avx2, __strcpy_avx2, __strncpy_avx2, __stpcpy_avx2
and __stpncpy_avx2.
* sysdeps/x86_64/multiarch/{ifunc-unaligned-ssse3.h =>
ifunc-strcpy.h}: rename header for a more generic name.
* sysdeps/x86_64/multiarch/ifunc-strcpy.h:
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
AVX unaligned load is fast and vzeroupper is preferred.
* sysdeps/x86_64/multiarch/stpcpy-avx2.S: New file
* sysdeps/x86_64/multiarch/stpncpy-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strcat-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strcpy-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strncat-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strncpy-avx2.S: Likewise
2018-10-08 13:59:50 +00:00
|
|
|
__stpcpy_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, stpcpy,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__stpcpy_avx2_rtm)
|
2021-03-05 14:36:50 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, stpcpy,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
|
|
__stpcpy_evex)
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, stpcpy, 1, __stpcpy_sse2_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, stpcpy, 1, __stpcpy_sse2))
|
|
|
|
|
|
2017-07-09 18:38:37 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strcasecmp_l.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, strcasecmp,
|
2022-03-24 23:56:13 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcasecmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
|
|
__strcasecmp_evex)
|
2022-03-24 23:56:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcasecmp,
|
|
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
|
|
|
|
__strcasecmp_avx2)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, strcasecmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__strcasecmp_avx2_rtm)
|
2015-08-13 10:38:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcasecmp,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (SSE4_2),
|
2012-10-10 14:49:12 +00:00
|
|
|
__strcasecmp_sse42)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, strcasecmp, 1, __strcasecmp_sse2))
|
|
|
|
|
|
2017-07-09 18:38:37 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strcasecmp_l.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, strcasecmp_l,
|
2022-03-24 23:56:13 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcasecmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
|
|
__strcasecmp_l_evex)
|
2022-03-24 23:56:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcasecmp,
|
|
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
|
|
|
|
__strcasecmp_l_avx2)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, strcasecmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__strcasecmp_l_avx2_rtm)
|
2015-08-13 10:38:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcasecmp_l,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (SSE4_2),
|
2012-10-10 14:49:12 +00:00
|
|
|
__strcasecmp_l_sse42)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, strcasecmp_l, 1,
|
|
|
|
|
__strcasecmp_l_sse2))
|
|
|
|
|
|
2017-07-09 18:38:37 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strcat.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, strcat,
|
2020-06-29 23:36:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcat, CPU_FEATURE_USABLE (AVX2),
|
x86-64: Optimize strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2
Optimize x86-64 strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2.
It uses vector comparison as much as possible. In general, the larger the
source string, the greater performance gain observed, reaching speedups of
1.6x compared to SSE2 unaligned routines. Select AVX2 strcat/strncat,
strcpy/strncpy and stpcpy/stpncpy on AVX2 machines where vzeroupper is
preferred and AVX unaligned load is fast.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strcat-avx2, strncat-avx2, strcpy-avx2, strncpy-avx2,
stpcpy-avx2 and stpncpy-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c:
(__libc_ifunc_impl_list): Add tests for __strcat_avx2,
__strncat_avx2, __strcpy_avx2, __strncpy_avx2, __stpcpy_avx2
and __stpncpy_avx2.
* sysdeps/x86_64/multiarch/{ifunc-unaligned-ssse3.h =>
ifunc-strcpy.h}: rename header for a more generic name.
* sysdeps/x86_64/multiarch/ifunc-strcpy.h:
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
AVX unaligned load is fast and vzeroupper is preferred.
* sysdeps/x86_64/multiarch/stpcpy-avx2.S: New file
* sysdeps/x86_64/multiarch/stpncpy-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strcat-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strcpy-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strncat-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strncpy-avx2.S: Likewise
2018-10-08 13:59:50 +00:00
|
|
|
__strcat_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcat,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__strcat_avx2_rtm)
|
2021-03-05 14:36:50 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcat,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
|
|
__strcat_evex)
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcat, 1, __strcat_sse2_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, strcat, 1, __strcat_sse2))
|
|
|
|
|
|
2017-06-09 12:42:16 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strchr.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, strchr,
|
2017-06-09 12:42:16 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strchr,
|
2021-04-19 17:45:07 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
2017-06-09 12:42:16 +00:00
|
|
|
__strchr_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
2021-04-19 17:45:07 +00:00
|
|
|
&& CPU_FEATURE_USABLE (BMI2)
|
2021-03-05 15:26:42 +00:00
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__strchr_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__strchr_evex)
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strchr, 1, __strchr_sse2_no_bsf)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, strchr, 1, __strchr_sse2))
|
|
|
|
|
|
2017-06-09 12:42:16 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strchrnul.c. */
|
|
|
|
|
IFUNC_IMPL (i, name, strchrnul,
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, strchrnul,
|
2021-04-19 17:45:07 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
2017-06-09 12:42:16 +00:00
|
|
|
__strchrnul_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strchrnul,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
2021-04-19 17:45:07 +00:00
|
|
|
&& CPU_FEATURE_USABLE (BMI2)
|
2021-03-05 15:26:42 +00:00
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__strchrnul_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strchrnul,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__strchrnul_evex)
|
2017-06-09 12:42:16 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strchrnul, 1, __strchrnul_sse2))
|
|
|
|
|
|
2017-06-09 12:45:43 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strrchr.c. */
|
|
|
|
|
IFUNC_IMPL (i, name, strrchr,
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, strrchr,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
2017-06-09 12:45:43 +00:00
|
|
|
__strrchr_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strrchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__strrchr_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strrchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
|
|
__strrchr_evex)
|
2017-06-09 12:45:43 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strrchr, 1, __strrchr_sse2))
|
|
|
|
|
|
2017-07-09 18:38:37 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strcmp.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, strcmp,
|
x86-64: Optimize strcmp/wcscmp and strncmp/wcsncmp with AVX2
Optimize x86-64 strcmp/wcscmp and strncmp/wcsncmp with AVX2. It uses vector
comparison as much as possible. Peak performance observed on a SkyLake
machine: 9x, 3x, 2.5x and 5.5x for strcmp, strncmp, wcscmp and wcsncmp,
respectively. The larger the comparison length, the more benefit using
avx2 functions, except on the strcmp, where peak is observed at length
== 32 bytes. Select AVX2 strcmp/wcscmp on AVX2 machines where vzeroupper
is preferred and AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strcmp-avx2, strncmp-avx2, wcscmp-avx2, wcscmp-sse2, wcsncmp-avx2 and
wcsncmp-sse2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strcmp_avx2,
__strncmp_avx2, __wcscmp_avx2, __wcsncmp_avx2, __wcscmp_sse2
and __wcsncmp_sse2.
* sysdeps/x86_64/multiarch/strcmp.c (OPTIMIZE (avx2)):
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
AVX unaligned load is fast and vzeroupper is preferred.
* sysdeps/x86_64/multiarch/strncmp.c: Likewise.
* sysdeps/x86_64/multiarch/strcmp-avx2.S: New file.
* sysdeps/x86_64/multiarch/strncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-sse2.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp.c: Likewise.
* sysdeps/x86_64/wcscmp.S (__wcscmp): Add alias only if __wcscmp
is undefined.
2018-05-03 16:09:30 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcmp,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
x86-64: Optimize strcmp/wcscmp and strncmp/wcsncmp with AVX2
Optimize x86-64 strcmp/wcscmp and strncmp/wcsncmp with AVX2. It uses vector
comparison as much as possible. Peak performance observed on a SkyLake
machine: 9x, 3x, 2.5x and 5.5x for strcmp, strncmp, wcscmp and wcsncmp,
respectively. The larger the comparison length, the more benefit using
avx2 functions, except on the strcmp, where peak is observed at length
== 32 bytes. Select AVX2 strcmp/wcscmp on AVX2 machines where vzeroupper
is preferred and AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strcmp-avx2, strncmp-avx2, wcscmp-avx2, wcscmp-sse2, wcsncmp-avx2 and
wcsncmp-sse2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strcmp_avx2,
__strncmp_avx2, __wcscmp_avx2, __wcsncmp_avx2, __wcscmp_sse2
and __wcsncmp_sse2.
* sysdeps/x86_64/multiarch/strcmp.c (OPTIMIZE (avx2)):
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
AVX unaligned load is fast and vzeroupper is preferred.
* sysdeps/x86_64/multiarch/strncmp.c: Likewise.
* sysdeps/x86_64/multiarch/strcmp-avx2.S: New file.
* sysdeps/x86_64/multiarch/strncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-sse2.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp.c: Likewise.
* sysdeps/x86_64/wcscmp.S (__wcscmp): Add alias only if __wcscmp
is undefined.
2018-05-03 16:09:30 +00:00
|
|
|
__strcmp_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__strcmp_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__strcmp_evex)
|
2020-06-29 23:36:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcmp, CPU_FEATURE_USABLE (SSE4_2),
|
2015-08-13 10:38:47 +00:00
|
|
|
__strcmp_sse42)
|
2013-09-03 14:21:38 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcmp, 1, __strcmp_sse2_unaligned)
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcmp, 1, __strcmp_sse2))
|
|
|
|
|
|
2017-07-09 18:38:37 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strcpy.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, strcpy,
|
2020-06-29 23:36:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcpy, CPU_FEATURE_USABLE (AVX2),
|
x86-64: Optimize strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2
Optimize x86-64 strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2.
It uses vector comparison as much as possible. In general, the larger the
source string, the greater performance gain observed, reaching speedups of
1.6x compared to SSE2 unaligned routines. Select AVX2 strcat/strncat,
strcpy/strncpy and stpcpy/stpncpy on AVX2 machines where vzeroupper is
preferred and AVX unaligned load is fast.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strcat-avx2, strncat-avx2, strcpy-avx2, strncpy-avx2,
stpcpy-avx2 and stpncpy-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c:
(__libc_ifunc_impl_list): Add tests for __strcat_avx2,
__strncat_avx2, __strcpy_avx2, __strncpy_avx2, __stpcpy_avx2
and __stpncpy_avx2.
* sysdeps/x86_64/multiarch/{ifunc-unaligned-ssse3.h =>
ifunc-strcpy.h}: rename header for a more generic name.
* sysdeps/x86_64/multiarch/ifunc-strcpy.h:
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
AVX unaligned load is fast and vzeroupper is preferred.
* sysdeps/x86_64/multiarch/stpcpy-avx2.S: New file
* sysdeps/x86_64/multiarch/stpncpy-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strcat-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strcpy-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strncat-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strncpy-avx2.S: Likewise
2018-10-08 13:59:50 +00:00
|
|
|
__strcpy_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcpy,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__strcpy_avx2_rtm)
|
2021-03-05 14:36:50 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcpy,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
|
|
__strcpy_evex)
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcpy, 1, __strcpy_sse2_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, strcpy, 1, __strcpy_sse2))
|
|
|
|
|
|
2017-07-09 18:38:37 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strcspn.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, strcspn,
|
2020-06-29 23:36:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcspn, CPU_FEATURE_USABLE (SSE4_2),
|
2012-10-10 14:49:12 +00:00
|
|
|
__strcspn_sse42)
|
2022-06-16 22:07:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strcspn, 1, __strcspn_generic))
|
2012-10-10 14:49:12 +00:00
|
|
|
|
2017-07-09 18:38:37 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strncase_l.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, strncasecmp,
|
2022-03-24 23:56:13 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncasecmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
|
|
__strncasecmp_evex)
|
2022-03-24 23:56:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncasecmp,
|
|
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
|
|
|
|
__strncasecmp_avx2)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, strncasecmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__strncasecmp_avx2_rtm)
|
2015-08-13 10:38:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncasecmp,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (SSE4_2),
|
2012-10-10 14:49:12 +00:00
|
|
|
__strncasecmp_sse42)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, strncasecmp, 1,
|
|
|
|
|
__strncasecmp_sse2))
|
|
|
|
|
|
2017-07-09 18:38:37 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strncase_l.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, strncasecmp_l,
|
2022-03-24 23:56:13 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncasecmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
|
|
__strncasecmp_l_evex)
|
2022-03-24 23:56:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncasecmp,
|
|
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
|
|
|
|
__strncasecmp_l_avx2)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, strncasecmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__strncasecmp_l_avx2_rtm)
|
2015-08-13 10:38:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncasecmp_l,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (SSE4_2),
|
2012-10-10 14:49:12 +00:00
|
|
|
__strncasecmp_l_sse42)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, strncasecmp_l, 1,
|
|
|
|
|
__strncasecmp_l_sse2))
|
|
|
|
|
|
2017-07-09 18:38:37 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strncat.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, strncat,
|
2020-06-29 23:36:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncat, CPU_FEATURE_USABLE (AVX2),
|
x86-64: Optimize strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2
Optimize x86-64 strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2.
It uses vector comparison as much as possible. In general, the larger the
source string, the greater performance gain observed, reaching speedups of
1.6x compared to SSE2 unaligned routines. Select AVX2 strcat/strncat,
strcpy/strncpy and stpcpy/stpncpy on AVX2 machines where vzeroupper is
preferred and AVX unaligned load is fast.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strcat-avx2, strncat-avx2, strcpy-avx2, strncpy-avx2,
stpcpy-avx2 and stpncpy-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c:
(__libc_ifunc_impl_list): Add tests for __strcat_avx2,
__strncat_avx2, __strcpy_avx2, __strncpy_avx2, __stpcpy_avx2
and __stpncpy_avx2.
* sysdeps/x86_64/multiarch/{ifunc-unaligned-ssse3.h =>
ifunc-strcpy.h}: rename header for a more generic name.
* sysdeps/x86_64/multiarch/ifunc-strcpy.h:
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
AVX unaligned load is fast and vzeroupper is preferred.
* sysdeps/x86_64/multiarch/stpcpy-avx2.S: New file
* sysdeps/x86_64/multiarch/stpncpy-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strcat-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strcpy-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strncat-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strncpy-avx2.S: Likewise
2018-10-08 13:59:50 +00:00
|
|
|
__strncat_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncat,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__strncat_avx2_rtm)
|
2021-03-05 14:36:50 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncat,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
|
|
__strncat_evex)
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncat, 1,
|
|
|
|
|
__strncat_sse2_unaligned)
|
2022-06-16 22:07:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncat, 1, __strncat_generic))
|
2012-10-10 14:49:12 +00:00
|
|
|
|
2017-07-09 18:38:37 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strncpy.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, strncpy,
|
2020-06-29 23:36:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncpy, CPU_FEATURE_USABLE (AVX2),
|
x86-64: Optimize strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2
Optimize x86-64 strcat/strncat, strcpy/strncpy and stpcpy/stpncpy with AVX2.
It uses vector comparison as much as possible. In general, the larger the
source string, the greater performance gain observed, reaching speedups of
1.6x compared to SSE2 unaligned routines. Select AVX2 strcat/strncat,
strcpy/strncpy and stpcpy/stpncpy on AVX2 machines where vzeroupper is
preferred and AVX unaligned load is fast.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strcat-avx2, strncat-avx2, strcpy-avx2, strncpy-avx2,
stpcpy-avx2 and stpncpy-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c:
(__libc_ifunc_impl_list): Add tests for __strcat_avx2,
__strncat_avx2, __strcpy_avx2, __strncpy_avx2, __stpcpy_avx2
and __stpncpy_avx2.
* sysdeps/x86_64/multiarch/{ifunc-unaligned-ssse3.h =>
ifunc-strcpy.h}: rename header for a more generic name.
* sysdeps/x86_64/multiarch/ifunc-strcpy.h:
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
AVX unaligned load is fast and vzeroupper is preferred.
* sysdeps/x86_64/multiarch/stpcpy-avx2.S: New file
* sysdeps/x86_64/multiarch/stpncpy-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strcat-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strcpy-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strncat-avx2.S: Likewise
* sysdeps/x86_64/multiarch/strncpy-avx2.S: Likewise
2018-10-08 13:59:50 +00:00
|
|
|
__strncpy_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncpy,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__strncpy_avx2_rtm)
|
2021-03-05 14:36:50 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncpy,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
|
|
__strncpy_evex)
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncpy, 1,
|
|
|
|
|
__strncpy_sse2_unaligned)
|
2022-06-16 22:07:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncpy, 1, __strncpy_generic))
|
2012-10-10 14:49:12 +00:00
|
|
|
|
2017-07-09 18:38:37 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strpbrk.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, strpbrk,
|
2020-06-29 23:36:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strpbrk, CPU_FEATURE_USABLE (SSE4_2),
|
2012-10-10 14:49:12 +00:00
|
|
|
__strpbrk_sse42)
|
2022-06-16 22:07:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strpbrk, 1, __strpbrk_generic))
|
2012-10-10 14:49:12 +00:00
|
|
|
|
|
|
|
|
|
2017-07-09 18:38:37 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strspn.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, strspn,
|
2020-06-29 23:36:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strspn, CPU_FEATURE_USABLE (SSE4_2),
|
2015-08-13 10:38:47 +00:00
|
|
|
__strspn_sse42)
|
2022-06-16 22:07:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strspn, 1, __strspn_generic))
|
2012-10-10 14:49:12 +00:00
|
|
|
|
2013-12-16 02:44:14 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strstr.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, strstr,
|
2022-06-06 19:17:43 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strstr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512DQ)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__strstr_avx512)
|
2013-12-14 18:33:56 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strstr, 1, __strstr_sse2_unaligned)
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strstr, 1, __strstr_sse2))
|
|
|
|
|
|
2017-06-09 12:42:16 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/wcschr.c. */
|
|
|
|
|
IFUNC_IMPL (i, name, wcschr,
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, wcschr,
|
2021-04-19 17:45:07 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
2017-06-09 12:42:16 +00:00
|
|
|
__wcschr_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcschr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
2021-04-19 17:45:07 +00:00
|
|
|
&& CPU_FEATURE_USABLE (BMI2)
|
2021-03-05 15:26:42 +00:00
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__wcschr_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcschr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__wcschr_evex)
|
2017-06-09 12:42:16 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcschr, 1, __wcschr_sse2))
|
|
|
|
|
|
2017-06-09 12:45:43 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/wcsrchr.c. */
|
|
|
|
|
IFUNC_IMPL (i, name, wcsrchr,
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, wcsrchr,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
2017-06-09 12:45:43 +00:00
|
|
|
__wcsrchr_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcsrchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__wcsrchr_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcsrchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__wcsrchr_evex)
|
2017-06-09 12:45:43 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcsrchr, 1, __wcsrchr_sse2))
|
|
|
|
|
|
x86-64: Optimize strcmp/wcscmp and strncmp/wcsncmp with AVX2
Optimize x86-64 strcmp/wcscmp and strncmp/wcsncmp with AVX2. It uses vector
comparison as much as possible. Peak performance observed on a SkyLake
machine: 9x, 3x, 2.5x and 5.5x for strcmp, strncmp, wcscmp and wcsncmp,
respectively. The larger the comparison length, the more benefit using
avx2 functions, except on the strcmp, where peak is observed at length
== 32 bytes. Select AVX2 strcmp/wcscmp on AVX2 machines where vzeroupper
is preferred and AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strcmp-avx2, strncmp-avx2, wcscmp-avx2, wcscmp-sse2, wcsncmp-avx2 and
wcsncmp-sse2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strcmp_avx2,
__strncmp_avx2, __wcscmp_avx2, __wcsncmp_avx2, __wcscmp_sse2
and __wcsncmp_sse2.
* sysdeps/x86_64/multiarch/strcmp.c (OPTIMIZE (avx2)):
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
AVX unaligned load is fast and vzeroupper is preferred.
* sysdeps/x86_64/multiarch/strncmp.c: Likewise.
* sysdeps/x86_64/multiarch/strcmp-avx2.S: New file.
* sysdeps/x86_64/multiarch/strncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-sse2.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp.c: Likewise.
* sysdeps/x86_64/wcscmp.S (__wcscmp): Add alias only if __wcscmp
is undefined.
2018-05-03 16:09:30 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/wcscmp.c. */
|
|
|
|
|
IFUNC_IMPL (i, name, wcscmp,
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, wcscmp,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
x86-64: Optimize strcmp/wcscmp and strncmp/wcsncmp with AVX2
Optimize x86-64 strcmp/wcscmp and strncmp/wcsncmp with AVX2. It uses vector
comparison as much as possible. Peak performance observed on a SkyLake
machine: 9x, 3x, 2.5x and 5.5x for strcmp, strncmp, wcscmp and wcsncmp,
respectively. The larger the comparison length, the more benefit using
avx2 functions, except on the strcmp, where peak is observed at length
== 32 bytes. Select AVX2 strcmp/wcscmp on AVX2 machines where vzeroupper
is preferred and AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strcmp-avx2, strncmp-avx2, wcscmp-avx2, wcscmp-sse2, wcsncmp-avx2 and
wcsncmp-sse2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strcmp_avx2,
__strncmp_avx2, __wcscmp_avx2, __wcsncmp_avx2, __wcscmp_sse2
and __wcsncmp_sse2.
* sysdeps/x86_64/multiarch/strcmp.c (OPTIMIZE (avx2)):
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
AVX unaligned load is fast and vzeroupper is preferred.
* sysdeps/x86_64/multiarch/strncmp.c: Likewise.
* sysdeps/x86_64/multiarch/strcmp-avx2.S: New file.
* sysdeps/x86_64/multiarch/strncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-sse2.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp.c: Likewise.
* sysdeps/x86_64/wcscmp.S (__wcscmp): Add alias only if __wcscmp
is undefined.
2018-05-03 16:09:30 +00:00
|
|
|
__wcscmp_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcscmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__wcscmp_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcscmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__wcscmp_evex)
|
x86-64: Optimize strcmp/wcscmp and strncmp/wcsncmp with AVX2
Optimize x86-64 strcmp/wcscmp and strncmp/wcsncmp with AVX2. It uses vector
comparison as much as possible. Peak performance observed on a SkyLake
machine: 9x, 3x, 2.5x and 5.5x for strcmp, strncmp, wcscmp and wcsncmp,
respectively. The larger the comparison length, the more benefit using
avx2 functions, except on the strcmp, where peak is observed at length
== 32 bytes. Select AVX2 strcmp/wcscmp on AVX2 machines where vzeroupper
is preferred and AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strcmp-avx2, strncmp-avx2, wcscmp-avx2, wcscmp-sse2, wcsncmp-avx2 and
wcsncmp-sse2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strcmp_avx2,
__strncmp_avx2, __wcscmp_avx2, __wcsncmp_avx2, __wcscmp_sse2
and __wcsncmp_sse2.
* sysdeps/x86_64/multiarch/strcmp.c (OPTIMIZE (avx2)):
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
AVX unaligned load is fast and vzeroupper is preferred.
* sysdeps/x86_64/multiarch/strncmp.c: Likewise.
* sysdeps/x86_64/multiarch/strcmp-avx2.S: New file.
* sysdeps/x86_64/multiarch/strncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-sse2.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp.c: Likewise.
* sysdeps/x86_64/wcscmp.S (__wcscmp): Add alias only if __wcscmp
is undefined.
2018-05-03 16:09:30 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcscmp, 1, __wcscmp_sse2))
|
|
|
|
|
|
|
|
|
|
/* Support sysdeps/x86_64/multiarch/wcsncmp.c. */
|
|
|
|
|
IFUNC_IMPL (i, name, wcsncmp,
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, wcsncmp,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
x86-64: Optimize strcmp/wcscmp and strncmp/wcsncmp with AVX2
Optimize x86-64 strcmp/wcscmp and strncmp/wcsncmp with AVX2. It uses vector
comparison as much as possible. Peak performance observed on a SkyLake
machine: 9x, 3x, 2.5x and 5.5x for strcmp, strncmp, wcscmp and wcsncmp,
respectively. The larger the comparison length, the more benefit using
avx2 functions, except on the strcmp, where peak is observed at length
== 32 bytes. Select AVX2 strcmp/wcscmp on AVX2 machines where vzeroupper
is preferred and AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strcmp-avx2, strncmp-avx2, wcscmp-avx2, wcscmp-sse2, wcsncmp-avx2 and
wcsncmp-sse2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strcmp_avx2,
__strncmp_avx2, __wcscmp_avx2, __wcsncmp_avx2, __wcscmp_sse2
and __wcsncmp_sse2.
* sysdeps/x86_64/multiarch/strcmp.c (OPTIMIZE (avx2)):
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
AVX unaligned load is fast and vzeroupper is preferred.
* sysdeps/x86_64/multiarch/strncmp.c: Likewise.
* sysdeps/x86_64/multiarch/strcmp-avx2.S: New file.
* sysdeps/x86_64/multiarch/strncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-sse2.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp.c: Likewise.
* sysdeps/x86_64/wcscmp.S (__wcscmp): Add alias only if __wcscmp
is undefined.
2018-05-03 16:09:30 +00:00
|
|
|
__wcsncmp_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcsncmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__wcsncmp_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcsncmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__wcsncmp_evex)
|
2022-06-16 22:07:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcsncmp, 1, __wcsncmp_generic))
|
x86-64: Optimize strcmp/wcscmp and strncmp/wcsncmp with AVX2
Optimize x86-64 strcmp/wcscmp and strncmp/wcsncmp with AVX2. It uses vector
comparison as much as possible. Peak performance observed on a SkyLake
machine: 9x, 3x, 2.5x and 5.5x for strcmp, strncmp, wcscmp and wcsncmp,
respectively. The larger the comparison length, the more benefit using
avx2 functions, except on the strcmp, where peak is observed at length
== 32 bytes. Select AVX2 strcmp/wcscmp on AVX2 machines where vzeroupper
is preferred and AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strcmp-avx2, strncmp-avx2, wcscmp-avx2, wcscmp-sse2, wcsncmp-avx2 and
wcsncmp-sse2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strcmp_avx2,
__strncmp_avx2, __wcscmp_avx2, __wcsncmp_avx2, __wcscmp_sse2
and __wcsncmp_sse2.
* sysdeps/x86_64/multiarch/strcmp.c (OPTIMIZE (avx2)):
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
AVX unaligned load is fast and vzeroupper is preferred.
* sysdeps/x86_64/multiarch/strncmp.c: Likewise.
* sysdeps/x86_64/multiarch/strcmp-avx2.S: New file.
* sysdeps/x86_64/multiarch/strncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-sse2.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp.c: Likewise.
* sysdeps/x86_64/wcscmp.S (__wcscmp): Add alias only if __wcscmp
is undefined.
2018-05-03 16:09:30 +00:00
|
|
|
|
2017-07-09 18:38:37 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/wcscpy.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, wcscpy,
|
2020-06-29 23:36:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcscpy, CPU_FEATURE_USABLE (SSSE3),
|
2015-08-13 10:38:47 +00:00
|
|
|
__wcscpy_ssse3)
|
2022-06-16 22:07:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcscpy, 1, __wcscpy_generic))
|
2012-10-10 14:49:12 +00:00
|
|
|
|
x86-64: Optimize strlen/strnlen/wcslen/wcsnlen with AVX2
Optimize strlen/strnlen/wcslen/wcsnlen with AVX2 to check 32 bytes with
a single vector compare instruction. It is as fast as SSE2 versions for
size <= 16 bytes and up to 1X faster for or size > 16 bytes on Haswell.
Select AVX2 version on AVX2 machines where vzeroupper is preferred and
AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strlen-sse2, strnlen-sse2, strlen-avx2, strnlen-avx2,
wcslen-sse2, wcslen-avx2 and wcsnlen-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strlen_avx2,
__strlen_sse2, __strnlen_avx2, __strnlen_sse2, __wcslen_avx2,
__wcslen_sse2 and __wcsnlen_avx2.
* sysdeps/x86_64/multiarch/strlen-avx2.S: New file.
* sysdeps/x86_64/multiarch/strlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strlen.c: Likewise.
* sysdeps/x86_64/multiarch/strnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen.c: Likewise.
* sysdeps/x86_64/multiarch/wcslen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen.c: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen.c (OPTIMIZE (avx2)): New.
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX2 machines where
vzeroupper is preferred and AVX unaligned load is fast.
2017-06-09 12:18:03 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/wcslen.c. */
|
|
|
|
|
IFUNC_IMPL (i, name, wcslen,
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, wcslen,
|
2021-04-19 23:36:07 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
x86-64: Optimize strlen/strnlen/wcslen/wcsnlen with AVX2
Optimize strlen/strnlen/wcslen/wcsnlen with AVX2 to check 32 bytes with
a single vector compare instruction. It is as fast as SSE2 versions for
size <= 16 bytes and up to 1X faster for or size > 16 bytes on Haswell.
Select AVX2 version on AVX2 machines where vzeroupper is preferred and
AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strlen-sse2, strnlen-sse2, strlen-avx2, strnlen-avx2,
wcslen-sse2, wcslen-avx2 and wcsnlen-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strlen_avx2,
__strlen_sse2, __strnlen_avx2, __strnlen_sse2, __wcslen_avx2,
__wcslen_sse2 and __wcsnlen_avx2.
* sysdeps/x86_64/multiarch/strlen-avx2.S: New file.
* sysdeps/x86_64/multiarch/strlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strlen.c: Likewise.
* sysdeps/x86_64/multiarch/strnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen.c: Likewise.
* sysdeps/x86_64/multiarch/wcslen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen.c: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen.c (OPTIMIZE (avx2)): New.
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX2 machines where
vzeroupper is preferred and AVX unaligned load is fast.
2017-06-09 12:18:03 +00:00
|
|
|
__wcslen_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcslen,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
2021-04-19 23:36:07 +00:00
|
|
|
&& CPU_FEATURE_USABLE (BMI2)
|
2021-03-05 15:26:42 +00:00
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__wcslen_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcslen,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__wcslen_evex)
|
2022-02-28 00:39:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcslen,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__wcslen_evex512)
|
2021-07-08 20:13:19 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcslen,
|
2021-06-23 05:19:34 +00:00
|
|
|
CPU_FEATURE_USABLE (SSE4_1),
|
2021-07-08 20:13:19 +00:00
|
|
|
__wcslen_sse4_1)
|
x86-64: Optimize strlen/strnlen/wcslen/wcsnlen with AVX2
Optimize strlen/strnlen/wcslen/wcsnlen with AVX2 to check 32 bytes with
a single vector compare instruction. It is as fast as SSE2 versions for
size <= 16 bytes and up to 1X faster for or size > 16 bytes on Haswell.
Select AVX2 version on AVX2 machines where vzeroupper is preferred and
AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strlen-sse2, strnlen-sse2, strlen-avx2, strnlen-avx2,
wcslen-sse2, wcslen-avx2 and wcsnlen-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strlen_avx2,
__strlen_sse2, __strnlen_avx2, __strnlen_sse2, __wcslen_avx2,
__wcslen_sse2 and __wcsnlen_avx2.
* sysdeps/x86_64/multiarch/strlen-avx2.S: New file.
* sysdeps/x86_64/multiarch/strlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strlen.c: Likewise.
* sysdeps/x86_64/multiarch/strnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen.c: Likewise.
* sysdeps/x86_64/multiarch/wcslen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen.c: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen.c (OPTIMIZE (avx2)): New.
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX2 machines where
vzeroupper is preferred and AVX unaligned load is fast.
2017-06-09 12:18:03 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcslen, 1, __wcslen_sse2))
|
|
|
|
|
|
2017-06-06 13:12:22 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/wcsnlen.c. */
|
|
|
|
|
IFUNC_IMPL (i, name, wcsnlen,
|
x86-64: Optimize strlen/strnlen/wcslen/wcsnlen with AVX2
Optimize strlen/strnlen/wcslen/wcsnlen with AVX2 to check 32 bytes with
a single vector compare instruction. It is as fast as SSE2 versions for
size <= 16 bytes and up to 1X faster for or size > 16 bytes on Haswell.
Select AVX2 version on AVX2 machines where vzeroupper is preferred and
AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strlen-sse2, strnlen-sse2, strlen-avx2, strnlen-avx2,
wcslen-sse2, wcslen-avx2 and wcsnlen-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strlen_avx2,
__strlen_sse2, __strnlen_avx2, __strnlen_sse2, __wcslen_avx2,
__wcslen_sse2 and __wcsnlen_avx2.
* sysdeps/x86_64/multiarch/strlen-avx2.S: New file.
* sysdeps/x86_64/multiarch/strlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strlen.c: Likewise.
* sysdeps/x86_64/multiarch/strnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen.c: Likewise.
* sysdeps/x86_64/multiarch/wcslen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen.c: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen.c (OPTIMIZE (avx2)): New.
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX2 machines where
vzeroupper is preferred and AVX unaligned load is fast.
2017-06-09 12:18:03 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcsnlen,
|
2021-04-19 23:36:07 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
x86-64: Optimize strlen/strnlen/wcslen/wcsnlen with AVX2
Optimize strlen/strnlen/wcslen/wcsnlen with AVX2 to check 32 bytes with
a single vector compare instruction. It is as fast as SSE2 versions for
size <= 16 bytes and up to 1X faster for or size > 16 bytes on Haswell.
Select AVX2 version on AVX2 machines where vzeroupper is preferred and
AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strlen-sse2, strnlen-sse2, strlen-avx2, strnlen-avx2,
wcslen-sse2, wcslen-avx2 and wcsnlen-avx2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strlen_avx2,
__strlen_sse2, __strnlen_avx2, __strnlen_sse2, __wcslen_avx2,
__wcslen_sse2 and __wcsnlen_avx2.
* sysdeps/x86_64/multiarch/strlen-avx2.S: New file.
* sysdeps/x86_64/multiarch/strlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strlen.c: Likewise.
* sysdeps/x86_64/multiarch/strnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen.c: Likewise.
* sysdeps/x86_64/multiarch/wcslen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen.c: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen.c (OPTIMIZE (avx2)): New.
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX2 machines where
vzeroupper is preferred and AVX unaligned load is fast.
2017-06-09 12:18:03 +00:00
|
|
|
__wcsnlen_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcsnlen,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
2021-04-19 23:36:07 +00:00
|
|
|
&& CPU_FEATURE_USABLE (BMI2)
|
2021-03-05 15:26:42 +00:00
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__wcsnlen_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcsnlen,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__wcsnlen_evex)
|
2022-02-28 00:39:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcsnlen,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__wcsnlen_evex512)
|
2017-06-06 13:12:22 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcsnlen,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (SSE4_1),
|
2017-06-06 13:12:22 +00:00
|
|
|
__wcsnlen_sse4_1)
|
2022-06-16 22:07:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wcsnlen, 1, __wcsnlen_generic))
|
2017-06-06 13:12:22 +00:00
|
|
|
|
2017-06-09 12:13:15 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/wmemchr.c. */
|
|
|
|
|
IFUNC_IMPL (i, name, wmemchr,
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, wmemchr,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
2017-06-09 12:13:15 +00:00
|
|
|
__wmemchr_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wmemchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__wmemchr_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wmemchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__wmemchr_evex)
|
x86: Add EVEX optimized memchr family not safe for RTM
No bug.
This commit adds a new implementation for EVEX memchr that is not safe
for RTM because it uses vzeroupper. The benefit is that by using
ymm0-ymm15 it can use vpcmpeq and vpternlogd in the 4x loop which is
faster than the RTM safe version which cannot use vpcmpeq because
there is no EVEX encoding for the instruction. All parts of the
implementation aside from the 4x loop are the same for the two
versions and the optimization is only relevant for large sizes.
Tigerlake:
size , algn , Pos , Cur T , New T , Win , Dif
512 , 6 , 192 , 9.2 , 9.04 , no-RTM , 0.16
512 , 7 , 224 , 9.19 , 8.98 , no-RTM , 0.21
2048 , 0 , 256 , 10.74 , 10.54 , no-RTM , 0.2
2048 , 0 , 512 , 14.81 , 14.87 , RTM , 0.06
2048 , 0 , 1024 , 22.97 , 22.57 , no-RTM , 0.4
2048 , 0 , 2048 , 37.49 , 34.51 , no-RTM , 2.98 <--
Icelake:
size , algn , Pos , Cur T , New T , Win , Dif
512 , 6 , 192 , 7.6 , 7.3 , no-RTM , 0.3
512 , 7 , 224 , 7.63 , 7.27 , no-RTM , 0.36
2048 , 0 , 256 , 8.48 , 8.38 , no-RTM , 0.1
2048 , 0 , 512 , 11.57 , 11.42 , no-RTM , 0.15
2048 , 0 , 1024 , 17.92 , 17.38 , no-RTM , 0.54
2048 , 0 , 2048 , 30.37 , 27.34 , no-RTM , 3.03 <--
test-memchr, test-wmemchr, and test-rawmemchr are all passing.
Signed-off-by: Noah Goldstein <goldstein.w.n@gmail.com>
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
2021-05-04 23:02:40 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wmemchr,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
|
|
__wmemchr_evex_rtm)
|
2017-06-09 12:13:15 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wmemchr, 1, __wmemchr_sse2))
|
|
|
|
|
|
2017-06-15 15:49:43 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/wmemcmp.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, wmemcmp,
|
2017-06-05 19:52:41 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wmemcmp,
|
2020-06-29 23:36:08 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
2021-05-17 17:56:52 +00:00
|
|
|
&& CPU_FEATURE_USABLE (BMI2)
|
2020-06-29 23:36:08 +00:00
|
|
|
&& CPU_FEATURE_USABLE (MOVBE)),
|
2017-06-05 19:52:41 +00:00
|
|
|
__wmemcmp_avx2_movbe)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wmemcmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
2021-05-17 17:56:52 +00:00
|
|
|
&& CPU_FEATURE_USABLE (BMI2)
|
2021-03-05 15:26:42 +00:00
|
|
|
&& CPU_FEATURE_USABLE (MOVBE)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__wmemcmp_avx2_movbe_rtm)
|
2021-03-05 15:20:28 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wmemcmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
2021-05-17 17:56:52 +00:00
|
|
|
&& CPU_FEATURE_USABLE (BMI2)
|
2021-03-05 15:20:28 +00:00
|
|
|
&& CPU_FEATURE_USABLE (MOVBE)),
|
|
|
|
|
__wmemcmp_evex_movbe)
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wmemcmp, 1, __wmemcmp_sse2))
|
|
|
|
|
|
2017-06-07 21:09:59 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/wmemset.c. */
|
x86-64: Optimize wmemset with SSE2/AVX2/AVX512
The difference between memset and wmemset is byte vs int. Add stubs
to SSE2/AVX2/AVX512 memset for wmemset with updated constant and size:
SSE2 wmemset:
shl $0x2,%rdx
movd %esi,%xmm0
mov %rdi,%rax
pshufd $0x0,%xmm0,%xmm0
jmp entry_from_wmemset
SSE2 memset:
movd %esi,%xmm0
mov %rdi,%rax
punpcklbw %xmm0,%xmm0
punpcklwd %xmm0,%xmm0
pshufd $0x0,%xmm0,%xmm0
entry_from_wmemset:
Since the ERMS versions of wmemset requires "rep stosl" instead of
"rep stosb", only the vector store stubs of SSE2/AVX2/AVX512 wmemset
are added. The SSE2 wmemset is about 3X faster and the AVX2 wmemset
is about 6X faster on Haswell.
* include/wchar.h (__wmemset_chk): New.
* sysdeps/x86_64/memset.S (VDUP_TO_VEC0_AND_SET_RETURN): Renamed
to MEMSET_VDUP_TO_VEC0_AND_SET_RETURN.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_CHK_SYMBOL): Likewise.
(WMEMSET_SYMBOL): Likewise.
(__wmemset): Add hidden definition.
(wmemset): Add weak hidden definition.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
wmemset_chk-nonshared.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add __wmemset_sse2_unaligned,
__wmemset_avx2_unaligned, __wmemset_avx512_unaligned,
__wmemset_chk_sse2_unaligned, __wmemset_chk_avx2_unaligned
and __wmemset_chk_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S: Updated.
(WMEMSET_CHK_SYMBOL): New.
(WMEMSET_CHK_SYMBOL (__wmemset_chk, unaligned)): Likewise.
(WMEMSET_SYMBOL (__wmemset, unaligned)): Likewise.
* sysdeps/x86_64/multiarch/memset.S (WMEMSET_SYMBOL): New.
(libc_hidden_builtin_def): Also define __GI_wmemset and
__GI___wmemset.
(weak_alias): New.
* sysdeps/x86_64/multiarch/wmemset.c: New file.
* sysdeps/x86_64/multiarch/wmemset.h: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk-nonshared.S: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk.c: Likewise.
* sysdeps/x86_64/wmemset.c: Likewise.
* sysdeps/x86_64/wmemset_chk.c: Likewise.
2017-06-05 18:09:48 +00:00
|
|
|
IFUNC_IMPL (i, name, wmemset,
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, wmemset, 1,
|
|
|
|
|
__wmemset_sse2_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, wmemset,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
x86-64: Optimize wmemset with SSE2/AVX2/AVX512
The difference between memset and wmemset is byte vs int. Add stubs
to SSE2/AVX2/AVX512 memset for wmemset with updated constant and size:
SSE2 wmemset:
shl $0x2,%rdx
movd %esi,%xmm0
mov %rdi,%rax
pshufd $0x0,%xmm0,%xmm0
jmp entry_from_wmemset
SSE2 memset:
movd %esi,%xmm0
mov %rdi,%rax
punpcklbw %xmm0,%xmm0
punpcklwd %xmm0,%xmm0
pshufd $0x0,%xmm0,%xmm0
entry_from_wmemset:
Since the ERMS versions of wmemset requires "rep stosl" instead of
"rep stosb", only the vector store stubs of SSE2/AVX2/AVX512 wmemset
are added. The SSE2 wmemset is about 3X faster and the AVX2 wmemset
is about 6X faster on Haswell.
* include/wchar.h (__wmemset_chk): New.
* sysdeps/x86_64/memset.S (VDUP_TO_VEC0_AND_SET_RETURN): Renamed
to MEMSET_VDUP_TO_VEC0_AND_SET_RETURN.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_CHK_SYMBOL): Likewise.
(WMEMSET_SYMBOL): Likewise.
(__wmemset): Add hidden definition.
(wmemset): Add weak hidden definition.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
wmemset_chk-nonshared.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add __wmemset_sse2_unaligned,
__wmemset_avx2_unaligned, __wmemset_avx512_unaligned,
__wmemset_chk_sse2_unaligned, __wmemset_chk_avx2_unaligned
and __wmemset_chk_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S: Updated.
(WMEMSET_CHK_SYMBOL): New.
(WMEMSET_CHK_SYMBOL (__wmemset_chk, unaligned)): Likewise.
(WMEMSET_SYMBOL (__wmemset, unaligned)): Likewise.
* sysdeps/x86_64/multiarch/memset.S (WMEMSET_SYMBOL): New.
(libc_hidden_builtin_def): Also define __GI_wmemset and
__GI___wmemset.
(weak_alias): New.
* sysdeps/x86_64/multiarch/wmemset.c: New file.
* sysdeps/x86_64/multiarch/wmemset.h: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk-nonshared.S: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk.c: Likewise.
* sysdeps/x86_64/wmemset.c: Likewise.
* sysdeps/x86_64/wmemset_chk.c: Likewise.
2017-06-05 18:09:48 +00:00
|
|
|
__wmemset_avx2_unaligned)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wmemset,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__wmemset_avx2_unaligned_rtm)
|
2021-03-05 15:15:03 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wmemset,
|
2021-04-19 21:48:10 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
2021-03-05 15:15:03 +00:00
|
|
|
__wmemset_evex_unaligned)
|
x86-64: Optimize wmemset with SSE2/AVX2/AVX512
The difference between memset and wmemset is byte vs int. Add stubs
to SSE2/AVX2/AVX512 memset for wmemset with updated constant and size:
SSE2 wmemset:
shl $0x2,%rdx
movd %esi,%xmm0
mov %rdi,%rax
pshufd $0x0,%xmm0,%xmm0
jmp entry_from_wmemset
SSE2 memset:
movd %esi,%xmm0
mov %rdi,%rax
punpcklbw %xmm0,%xmm0
punpcklwd %xmm0,%xmm0
pshufd $0x0,%xmm0,%xmm0
entry_from_wmemset:
Since the ERMS versions of wmemset requires "rep stosl" instead of
"rep stosb", only the vector store stubs of SSE2/AVX2/AVX512 wmemset
are added. The SSE2 wmemset is about 3X faster and the AVX2 wmemset
is about 6X faster on Haswell.
* include/wchar.h (__wmemset_chk): New.
* sysdeps/x86_64/memset.S (VDUP_TO_VEC0_AND_SET_RETURN): Renamed
to MEMSET_VDUP_TO_VEC0_AND_SET_RETURN.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_CHK_SYMBOL): Likewise.
(WMEMSET_SYMBOL): Likewise.
(__wmemset): Add hidden definition.
(wmemset): Add weak hidden definition.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
wmemset_chk-nonshared.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add __wmemset_sse2_unaligned,
__wmemset_avx2_unaligned, __wmemset_avx512_unaligned,
__wmemset_chk_sse2_unaligned, __wmemset_chk_avx2_unaligned
and __wmemset_chk_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S: Updated.
(WMEMSET_CHK_SYMBOL): New.
(WMEMSET_CHK_SYMBOL (__wmemset_chk, unaligned)): Likewise.
(WMEMSET_SYMBOL (__wmemset, unaligned)): Likewise.
* sysdeps/x86_64/multiarch/memset.S (WMEMSET_SYMBOL): New.
(libc_hidden_builtin_def): Also define __GI_wmemset and
__GI___wmemset.
(weak_alias): New.
* sysdeps/x86_64/multiarch/wmemset.c: New file.
* sysdeps/x86_64/multiarch/wmemset.h: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk-nonshared.S: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk.c: Likewise.
* sysdeps/x86_64/wmemset.c: Likewise.
* sysdeps/x86_64/wmemset_chk.c: Likewise.
2017-06-05 18:09:48 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, wmemset,
|
2021-04-19 21:48:10 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
x86-64: Optimize wmemset with SSE2/AVX2/AVX512
The difference between memset and wmemset is byte vs int. Add stubs
to SSE2/AVX2/AVX512 memset for wmemset with updated constant and size:
SSE2 wmemset:
shl $0x2,%rdx
movd %esi,%xmm0
mov %rdi,%rax
pshufd $0x0,%xmm0,%xmm0
jmp entry_from_wmemset
SSE2 memset:
movd %esi,%xmm0
mov %rdi,%rax
punpcklbw %xmm0,%xmm0
punpcklwd %xmm0,%xmm0
pshufd $0x0,%xmm0,%xmm0
entry_from_wmemset:
Since the ERMS versions of wmemset requires "rep stosl" instead of
"rep stosb", only the vector store stubs of SSE2/AVX2/AVX512 wmemset
are added. The SSE2 wmemset is about 3X faster and the AVX2 wmemset
is about 6X faster on Haswell.
* include/wchar.h (__wmemset_chk): New.
* sysdeps/x86_64/memset.S (VDUP_TO_VEC0_AND_SET_RETURN): Renamed
to MEMSET_VDUP_TO_VEC0_AND_SET_RETURN.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_CHK_SYMBOL): Likewise.
(WMEMSET_SYMBOL): Likewise.
(__wmemset): Add hidden definition.
(wmemset): Add weak hidden definition.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
wmemset_chk-nonshared.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add __wmemset_sse2_unaligned,
__wmemset_avx2_unaligned, __wmemset_avx512_unaligned,
__wmemset_chk_sse2_unaligned, __wmemset_chk_avx2_unaligned
and __wmemset_chk_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S: Updated.
(WMEMSET_CHK_SYMBOL): New.
(WMEMSET_CHK_SYMBOL (__wmemset_chk, unaligned)): Likewise.
(WMEMSET_SYMBOL (__wmemset, unaligned)): Likewise.
* sysdeps/x86_64/multiarch/memset.S (WMEMSET_SYMBOL): New.
(libc_hidden_builtin_def): Also define __GI_wmemset and
__GI___wmemset.
(weak_alias): New.
* sysdeps/x86_64/multiarch/wmemset.c: New file.
* sysdeps/x86_64/multiarch/wmemset.h: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk-nonshared.S: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk.c: Likewise.
* sysdeps/x86_64/wmemset.c: Likewise.
* sysdeps/x86_64/wmemset_chk.c: Likewise.
2017-06-05 18:09:48 +00:00
|
|
|
__wmemset_avx512_unaligned))
|
|
|
|
|
|
2012-10-10 14:49:12 +00:00
|
|
|
#ifdef SHARED
|
2017-06-14 19:10:57 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/memcpy_chk.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, __memcpy_chk,
|
2016-01-15 21:49:45 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memcpy_chk,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512F),
|
2016-01-15 21:49:45 +00:00
|
|
|
__memcpy_chk_avx512_no_vzeroupper)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memcpy_chk,
|
2021-03-07 17:45:23 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
2016-06-08 20:57:50 +00:00
|
|
|
__memcpy_chk_avx512_unaligned)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memcpy_chk,
|
2021-03-07 17:45:23 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
__memcpy_chk_avx512_unaligned_erms)
|
2015-08-13 10:38:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memcpy_chk,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX),
|
2014-07-14 04:02:52 +00:00
|
|
|
__memcpy_chk_avx_unaligned)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memcpy_chk,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX),
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
__memcpy_chk_avx_unaligned_erms)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memcpy_chk,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__memcpy_chk_avx_unaligned_rtm)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __memcpy_chk,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__memcpy_chk_avx_unaligned_erms_rtm)
|
2021-03-05 14:46:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memcpy_chk,
|
|
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
|
|
|
|
__memcpy_chk_evex_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __memcpy_chk,
|
|
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
|
|
|
|
__memcpy_chk_evex_unaligned_erms)
|
2015-08-13 10:38:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memcpy_chk,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (SSSE3),
|
2012-10-10 14:49:12 +00:00
|
|
|
__memcpy_chk_ssse3)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memcpy_chk, 1,
|
2016-06-08 20:57:50 +00:00
|
|
|
__memcpy_chk_sse2_unaligned)
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __memcpy_chk, 1,
|
2017-06-14 19:10:57 +00:00
|
|
|
__memcpy_chk_sse2_unaligned_erms)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __memcpy_chk, 1,
|
|
|
|
|
__memcpy_chk_erms))
|
2017-08-04 19:27:00 +00:00
|
|
|
#endif
|
2012-10-10 14:49:12 +00:00
|
|
|
|
2017-06-14 19:10:57 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/memcpy.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, memcpy,
|
2015-08-13 10:38:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memcpy,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX),
|
2014-07-14 04:02:52 +00:00
|
|
|
__memcpy_avx_unaligned)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memcpy,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX),
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
__memcpy_avx_unaligned_erms)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memcpy,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__memcpy_avx_unaligned_rtm)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, memcpy,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__memcpy_avx_unaligned_erms_rtm)
|
2021-03-05 14:46:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memcpy,
|
|
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
|
|
|
|
__memcpy_evex_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, memcpy,
|
|
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
|
|
|
|
__memcpy_evex_unaligned_erms)
|
2020-06-29 23:36:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memcpy, CPU_FEATURE_USABLE (SSSE3),
|
2015-08-13 10:38:47 +00:00
|
|
|
__memcpy_ssse3)
|
2016-01-15 21:49:45 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memcpy,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512F),
|
2016-01-15 21:49:45 +00:00
|
|
|
__memcpy_avx512_no_vzeroupper)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memcpy,
|
2021-03-07 17:45:23 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
2016-06-08 20:57:50 +00:00
|
|
|
__memcpy_avx512_unaligned)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memcpy,
|
2021-03-07 17:45:23 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
__memcpy_avx512_unaligned_erms)
|
2013-05-20 06:20:00 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memcpy, 1, __memcpy_sse2_unaligned)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memcpy, 1,
|
|
|
|
|
__memcpy_sse2_unaligned_erms)
|
2016-06-08 20:57:50 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, memcpy, 1, __memcpy_erms))
|
2012-10-10 14:49:12 +00:00
|
|
|
|
2017-08-04 19:27:00 +00:00
|
|
|
#ifdef SHARED
|
2017-06-14 19:10:57 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/mempcpy_chk.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, __mempcpy_chk,
|
2016-01-15 21:49:45 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512F),
|
2016-01-15 21:49:45 +00:00
|
|
|
__mempcpy_chk_avx512_no_vzeroupper)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
|
2021-03-07 17:45:23 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
2016-06-08 20:57:50 +00:00
|
|
|
__mempcpy_chk_avx512_unaligned)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
|
2021-03-07 17:45:23 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
__mempcpy_chk_avx512_unaligned_erms)
|
2015-08-13 10:38:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX),
|
2014-07-14 04:02:52 +00:00
|
|
|
__mempcpy_chk_avx_unaligned)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX),
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
__mempcpy_chk_avx_unaligned_erms)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__mempcpy_chk_avx_unaligned_rtm)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__mempcpy_chk_avx_unaligned_erms_rtm)
|
2021-03-05 14:46:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
|
|
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
|
|
|
|
__mempcpy_chk_evex_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
|
|
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
|
|
|
|
__mempcpy_chk_evex_unaligned_erms)
|
2015-08-13 10:38:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (SSSE3),
|
2012-10-10 14:49:12 +00:00
|
|
|
__mempcpy_chk_ssse3)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __mempcpy_chk, 1,
|
2016-06-08 20:57:50 +00:00
|
|
|
__mempcpy_chk_sse2_unaligned)
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __mempcpy_chk, 1,
|
2017-06-14 19:10:57 +00:00
|
|
|
__mempcpy_chk_sse2_unaligned_erms)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __mempcpy_chk, 1,
|
|
|
|
|
__mempcpy_chk_erms))
|
2017-08-04 19:27:00 +00:00
|
|
|
#endif
|
2012-10-10 14:49:12 +00:00
|
|
|
|
2017-06-14 19:10:57 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/mempcpy.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, mempcpy,
|
2016-01-15 21:49:45 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, mempcpy,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512F),
|
2016-01-15 21:49:45 +00:00
|
|
|
__mempcpy_avx512_no_vzeroupper)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, mempcpy,
|
2021-03-07 17:45:23 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
2016-06-08 20:57:50 +00:00
|
|
|
__mempcpy_avx512_unaligned)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, mempcpy,
|
2021-03-07 17:45:23 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
__mempcpy_avx512_unaligned_erms)
|
2015-08-13 10:38:47 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, mempcpy,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX),
|
2014-07-14 04:02:52 +00:00
|
|
|
__mempcpy_avx_unaligned)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, mempcpy,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX),
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
__mempcpy_avx_unaligned_erms)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, mempcpy,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__mempcpy_avx_unaligned_rtm)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, mempcpy,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__mempcpy_avx_unaligned_erms_rtm)
|
2021-03-05 14:46:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, mempcpy,
|
|
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
|
|
|
|
__mempcpy_evex_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, mempcpy,
|
|
|
|
|
CPU_FEATURE_USABLE (AVX512VL),
|
|
|
|
|
__mempcpy_evex_unaligned_erms)
|
2020-06-29 23:36:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, mempcpy, CPU_FEATURE_USABLE (SSSE3),
|
2012-10-10 14:49:12 +00:00
|
|
|
__mempcpy_ssse3)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, mempcpy, 1,
|
2016-06-08 20:57:50 +00:00
|
|
|
__mempcpy_sse2_unaligned)
|
Add x86-64 memmove with unaligned load/store and rep movsb
Implement x86-64 memmove with unaligned load/store and rep movsb.
Support 16-byte, 32-byte and 64-byte vector register sizes. When
size <= 8 times of vector register size, there is no check for
address overlap bewteen source and destination. Since overhead for
overlap check is small when size > 8 times of vector register size,
memcpy is an alias of memmove.
A single file provides 2 implementations of memmove, one with rep movsb
and the other without rep movsb. They share the same codes when size is
between 2 times of vector register size and REP_MOVSB_THRESHOLD which
is 2KB for 16-byte vector register size and scaled up by large vector
register size.
Key features:
1. Use overlapping load and store to avoid branch.
2. For size <= 8 times of vector register size, load all sources into
registers and store them together.
3. If there is no address overlap bewteen source and destination, copy
from both ends with 4 times of vector register size at a time.
4. If address of destination > address of source, backward copy 8 times
of vector register size at a time.
5. Otherwise, forward copy 8 times of vector register size at a time.
6. Use rep movsb only for forward copy. Avoid slow backward rep movsb
by fallbacking to backward copy 8 times of vector register size at a
time.
7. Skip when address of destination == address of source.
[BZ #19776]
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
memmove-sse2-unaligned-erms, memmove-avx-unaligned-erms and
memmove-avx512-unaligned-erms.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Test
__memmove_chk_avx512_unaligned_2,
__memmove_chk_avx512_unaligned_erms,
__memmove_chk_avx_unaligned_2, __memmove_chk_avx_unaligned_erms,
__memmove_chk_sse2_unaligned_2,
__memmove_chk_sse2_unaligned_erms, __memmove_avx_unaligned_2,
__memmove_avx_unaligned_erms, __memmove_avx512_unaligned_2,
__memmove_avx512_unaligned_erms, __memmove_erms,
__memmove_sse2_unaligned_2, __memmove_sse2_unaligned_erms,
__memcpy_chk_avx512_unaligned_2,
__memcpy_chk_avx512_unaligned_erms,
__memcpy_chk_avx_unaligned_2, __memcpy_chk_avx_unaligned_erms,
__memcpy_chk_sse2_unaligned_2, __memcpy_chk_sse2_unaligned_erms,
__memcpy_avx_unaligned_2, __memcpy_avx_unaligned_erms,
__memcpy_avx512_unaligned_2, __memcpy_avx512_unaligned_erms,
__memcpy_sse2_unaligned_2, __memcpy_sse2_unaligned_erms,
__memcpy_erms, __mempcpy_chk_avx512_unaligned_2,
__mempcpy_chk_avx512_unaligned_erms,
__mempcpy_chk_avx_unaligned_2, __mempcpy_chk_avx_unaligned_erms,
__mempcpy_chk_sse2_unaligned_2, __mempcpy_chk_sse2_unaligned_erms,
__mempcpy_avx512_unaligned_2, __mempcpy_avx512_unaligned_erms,
__mempcpy_avx_unaligned_2, __mempcpy_avx_unaligned_erms,
__mempcpy_sse2_unaligned_2, __mempcpy_sse2_unaligned_erms and
__mempcpy_erms.
* sysdeps/x86_64/multiarch/memmove-avx-unaligned-erms.S: New
file.
* sysdeps/x86_64/multiarch/memmove-avx512-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-sse2-unaligned-erms.S:
Likwise.
* sysdeps/x86_64/multiarch/memmove-vec-unaligned-erms.S:
Likwise.
2016-03-31 17:04:26 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, mempcpy, 1,
|
|
|
|
|
__mempcpy_sse2_unaligned_erms)
|
2016-06-08 20:57:50 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, mempcpy, 1, __mempcpy_erms))
|
2012-10-10 14:49:12 +00:00
|
|
|
|
2017-07-09 18:38:37 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/strncmp.c. */
|
2012-10-10 14:49:12 +00:00
|
|
|
IFUNC_IMPL (i, name, strncmp,
|
x86-64: Optimize strcmp/wcscmp and strncmp/wcsncmp with AVX2
Optimize x86-64 strcmp/wcscmp and strncmp/wcsncmp with AVX2. It uses vector
comparison as much as possible. Peak performance observed on a SkyLake
machine: 9x, 3x, 2.5x and 5.5x for strcmp, strncmp, wcscmp and wcsncmp,
respectively. The larger the comparison length, the more benefit using
avx2 functions, except on the strcmp, where peak is observed at length
== 32 bytes. Select AVX2 strcmp/wcscmp on AVX2 machines where vzeroupper
is preferred and AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strcmp-avx2, strncmp-avx2, wcscmp-avx2, wcscmp-sse2, wcsncmp-avx2 and
wcsncmp-sse2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strcmp_avx2,
__strncmp_avx2, __wcscmp_avx2, __wcsncmp_avx2, __wcscmp_sse2
and __wcsncmp_sse2.
* sysdeps/x86_64/multiarch/strcmp.c (OPTIMIZE (avx2)):
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
AVX unaligned load is fast and vzeroupper is preferred.
* sysdeps/x86_64/multiarch/strncmp.c: Likewise.
* sysdeps/x86_64/multiarch/strcmp-avx2.S: New file.
* sysdeps/x86_64/multiarch/strncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-sse2.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp.c: Likewise.
* sysdeps/x86_64/wcscmp.S (__wcscmp): Add alias only if __wcscmp
is undefined.
2018-05-03 16:09:30 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncmp,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
x86-64: Optimize strcmp/wcscmp and strncmp/wcsncmp with AVX2
Optimize x86-64 strcmp/wcscmp and strncmp/wcsncmp with AVX2. It uses vector
comparison as much as possible. Peak performance observed on a SkyLake
machine: 9x, 3x, 2.5x and 5.5x for strcmp, strncmp, wcscmp and wcsncmp,
respectively. The larger the comparison length, the more benefit using
avx2 functions, except on the strcmp, where peak is observed at length
== 32 bytes. Select AVX2 strcmp/wcscmp on AVX2 machines where vzeroupper
is preferred and AVX unaligned load is fast.
NB: It uses TZCNT instead of BSF since TZCNT produces the same result
as BSF for non-zero input. TZCNT is faster than BSF and is executed
as BSF if machine doesn't support TZCNT.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strcmp-avx2, strncmp-avx2, wcscmp-avx2, wcscmp-sse2, wcsncmp-avx2 and
wcsncmp-sse2.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add tests for __strcmp_avx2,
__strncmp_avx2, __wcscmp_avx2, __wcsncmp_avx2, __wcscmp_sse2
and __wcsncmp_sse2.
* sysdeps/x86_64/multiarch/strcmp.c (OPTIMIZE (avx2)):
(IFUNC_SELECTOR): Return OPTIMIZE (avx2) on AVX 2 machines if
AVX unaligned load is fast and vzeroupper is preferred.
* sysdeps/x86_64/multiarch/strncmp.c: Likewise.
* sysdeps/x86_64/multiarch/strcmp-avx2.S: New file.
* sysdeps/x86_64/multiarch/strncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcscmp.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp-sse2.c: Likewise.
* sysdeps/x86_64/multiarch/wcsncmp.c: Likewise.
* sysdeps/x86_64/wcscmp.S (__wcscmp): Add alias only if __wcscmp
is undefined.
2018-05-03 16:09:30 +00:00
|
|
|
__strncmp_avx2)
|
2021-03-05 15:26:42 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX2)
|
|
|
|
|
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
|
|
__strncmp_avx2_rtm)
|
2021-03-05 14:24:52 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncmp,
|
|
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
|
|
__strncmp_evex)
|
2020-06-29 23:36:08 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, strncmp, CPU_FEATURE_USABLE (SSE4_2),
|
2012-10-10 14:49:12 +00:00
|
|
|
__strncmp_sse42)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, strncmp, 1, __strncmp_sse2))
|
x86-64: Optimize wmemset with SSE2/AVX2/AVX512
The difference between memset and wmemset is byte vs int. Add stubs
to SSE2/AVX2/AVX512 memset for wmemset with updated constant and size:
SSE2 wmemset:
shl $0x2,%rdx
movd %esi,%xmm0
mov %rdi,%rax
pshufd $0x0,%xmm0,%xmm0
jmp entry_from_wmemset
SSE2 memset:
movd %esi,%xmm0
mov %rdi,%rax
punpcklbw %xmm0,%xmm0
punpcklwd %xmm0,%xmm0
pshufd $0x0,%xmm0,%xmm0
entry_from_wmemset:
Since the ERMS versions of wmemset requires "rep stosl" instead of
"rep stosb", only the vector store stubs of SSE2/AVX2/AVX512 wmemset
are added. The SSE2 wmemset is about 3X faster and the AVX2 wmemset
is about 6X faster on Haswell.
* include/wchar.h (__wmemset_chk): New.
* sysdeps/x86_64/memset.S (VDUP_TO_VEC0_AND_SET_RETURN): Renamed
to MEMSET_VDUP_TO_VEC0_AND_SET_RETURN.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_CHK_SYMBOL): Likewise.
(WMEMSET_SYMBOL): Likewise.
(__wmemset): Add hidden definition.
(wmemset): Add weak hidden definition.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
wmemset_chk-nonshared.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add __wmemset_sse2_unaligned,
__wmemset_avx2_unaligned, __wmemset_avx512_unaligned,
__wmemset_chk_sse2_unaligned, __wmemset_chk_avx2_unaligned
and __wmemset_chk_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S: Updated.
(WMEMSET_CHK_SYMBOL): New.
(WMEMSET_CHK_SYMBOL (__wmemset_chk, unaligned)): Likewise.
(WMEMSET_SYMBOL (__wmemset, unaligned)): Likewise.
* sysdeps/x86_64/multiarch/memset.S (WMEMSET_SYMBOL): New.
(libc_hidden_builtin_def): Also define __GI_wmemset and
__GI___wmemset.
(weak_alias): New.
* sysdeps/x86_64/multiarch/wmemset.c: New file.
* sysdeps/x86_64/multiarch/wmemset.h: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk-nonshared.S: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk.c: Likewise.
* sysdeps/x86_64/wmemset.c: Likewise.
* sysdeps/x86_64/wmemset_chk.c: Likewise.
2017-06-05 18:09:48 +00:00
|
|
|
|
2017-08-04 19:27:00 +00:00
|
|
|
#ifdef SHARED
|
2017-06-07 21:09:59 +00:00
|
|
|
/* Support sysdeps/x86_64/multiarch/wmemset_chk.c. */
|
x86-64: Optimize wmemset with SSE2/AVX2/AVX512
The difference between memset and wmemset is byte vs int. Add stubs
to SSE2/AVX2/AVX512 memset for wmemset with updated constant and size:
SSE2 wmemset:
shl $0x2,%rdx
movd %esi,%xmm0
mov %rdi,%rax
pshufd $0x0,%xmm0,%xmm0
jmp entry_from_wmemset
SSE2 memset:
movd %esi,%xmm0
mov %rdi,%rax
punpcklbw %xmm0,%xmm0
punpcklwd %xmm0,%xmm0
pshufd $0x0,%xmm0,%xmm0
entry_from_wmemset:
Since the ERMS versions of wmemset requires "rep stosl" instead of
"rep stosb", only the vector store stubs of SSE2/AVX2/AVX512 wmemset
are added. The SSE2 wmemset is about 3X faster and the AVX2 wmemset
is about 6X faster on Haswell.
* include/wchar.h (__wmemset_chk): New.
* sysdeps/x86_64/memset.S (VDUP_TO_VEC0_AND_SET_RETURN): Renamed
to MEMSET_VDUP_TO_VEC0_AND_SET_RETURN.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_CHK_SYMBOL): Likewise.
(WMEMSET_SYMBOL): Likewise.
(__wmemset): Add hidden definition.
(wmemset): Add weak hidden definition.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
wmemset_chk-nonshared.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add __wmemset_sse2_unaligned,
__wmemset_avx2_unaligned, __wmemset_avx512_unaligned,
__wmemset_chk_sse2_unaligned, __wmemset_chk_avx2_unaligned
and __wmemset_chk_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S: Updated.
(WMEMSET_CHK_SYMBOL): New.
(WMEMSET_CHK_SYMBOL (__wmemset_chk, unaligned)): Likewise.
(WMEMSET_SYMBOL (__wmemset, unaligned)): Likewise.
* sysdeps/x86_64/multiarch/memset.S (WMEMSET_SYMBOL): New.
(libc_hidden_builtin_def): Also define __GI_wmemset and
__GI___wmemset.
(weak_alias): New.
* sysdeps/x86_64/multiarch/wmemset.c: New file.
* sysdeps/x86_64/multiarch/wmemset.h: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk-nonshared.S: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk.c: Likewise.
* sysdeps/x86_64/wmemset.c: Likewise.
* sysdeps/x86_64/wmemset_chk.c: Likewise.
2017-06-05 18:09:48 +00:00
|
|
|
IFUNC_IMPL (i, name, __wmemset_chk,
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __wmemset_chk, 1,
|
|
|
|
|
__wmemset_chk_sse2_unaligned)
|
|
|
|
|
IFUNC_IMPL_ADD (array, i, __wmemset_chk,
|
2020-06-29 23:36:08 +00:00
|
|
|
CPU_FEATURE_USABLE (AVX2),
|
x86-64: Optimize wmemset with SSE2/AVX2/AVX512
The difference between memset and wmemset is byte vs int. Add stubs
to SSE2/AVX2/AVX512 memset for wmemset with updated constant and size:
SSE2 wmemset:
shl $0x2,%rdx
movd %esi,%xmm0
mov %rdi,%rax
pshufd $0x0,%xmm0,%xmm0
jmp entry_from_wmemset
SSE2 memset:
movd %esi,%xmm0
mov %rdi,%rax
punpcklbw %xmm0,%xmm0
punpcklwd %xmm0,%xmm0
pshufd $0x0,%xmm0,%xmm0
entry_from_wmemset:
Since the ERMS versions of wmemset requires "rep stosl" instead of
"rep stosb", only the vector store stubs of SSE2/AVX2/AVX512 wmemset
are added. The SSE2 wmemset is about 3X faster and the AVX2 wmemset
is about 6X faster on Haswell.
* include/wchar.h (__wmemset_chk): New.
* sysdeps/x86_64/memset.S (VDUP_TO_VEC0_AND_SET_RETURN): Renamed
to MEMSET_VDUP_TO_VEC0_AND_SET_RETURN.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_CHK_SYMBOL): Likewise.
(WMEMSET_SYMBOL): Likewise.
(__wmemset): Add hidden definition.
(wmemset): Add weak hidden definition.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
wmemset_chk-nonshared.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add __wmemset_sse2_unaligned,
__wmemset_avx2_unaligned, __wmemset_avx512_unaligned,
__wmemset_chk_sse2_unaligned, __wmemset_chk_avx2_unaligned
and __wmemset_chk_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S: Updated.
(WMEMSET_CHK_SYMBOL): New.
(WMEMSET_CHK_SYMBOL (__wmemset_chk, unaligned)): Likewise.
(WMEMSET_SYMBOL (__wmemset, unaligned)): Likewise.
* sysdeps/x86_64/multiarch/memset.S (WMEMSET_SYMBOL): New.
(libc_hidden_builtin_def): Also define __GI_wmemset and
__GI___wmemset.
(weak_alias): New.
* sysdeps/x86_64/multiarch/wmemset.c: New file.
* sysdeps/x86_64/multiarch/wmemset.h: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk-nonshared.S: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk.c: Likewise.
* sysdeps/x86_64/wmemset.c: Likewise.
* sysdeps/x86_64/wmemset_chk.c: Likewise.
2017-06-05 18:09:48 +00:00
|
|
|
__wmemset_chk_avx2_unaligned)
|
2021-03-05 15:15:03 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __wmemset_chk,
|
2021-04-19 21:48:10 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
2021-03-05 15:15:03 +00:00
|
|
|
__wmemset_chk_evex_unaligned)
|
x86-64: Optimize wmemset with SSE2/AVX2/AVX512
The difference between memset and wmemset is byte vs int. Add stubs
to SSE2/AVX2/AVX512 memset for wmemset with updated constant and size:
SSE2 wmemset:
shl $0x2,%rdx
movd %esi,%xmm0
mov %rdi,%rax
pshufd $0x0,%xmm0,%xmm0
jmp entry_from_wmemset
SSE2 memset:
movd %esi,%xmm0
mov %rdi,%rax
punpcklbw %xmm0,%xmm0
punpcklwd %xmm0,%xmm0
pshufd $0x0,%xmm0,%xmm0
entry_from_wmemset:
Since the ERMS versions of wmemset requires "rep stosl" instead of
"rep stosb", only the vector store stubs of SSE2/AVX2/AVX512 wmemset
are added. The SSE2 wmemset is about 3X faster and the AVX2 wmemset
is about 6X faster on Haswell.
* include/wchar.h (__wmemset_chk): New.
* sysdeps/x86_64/memset.S (VDUP_TO_VEC0_AND_SET_RETURN): Renamed
to MEMSET_VDUP_TO_VEC0_AND_SET_RETURN.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_CHK_SYMBOL): Likewise.
(WMEMSET_SYMBOL): Likewise.
(__wmemset): Add hidden definition.
(wmemset): Add weak hidden definition.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
wmemset_chk-nonshared.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add __wmemset_sse2_unaligned,
__wmemset_avx2_unaligned, __wmemset_avx512_unaligned,
__wmemset_chk_sse2_unaligned, __wmemset_chk_avx2_unaligned
and __wmemset_chk_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S: Updated.
(WMEMSET_CHK_SYMBOL): New.
(WMEMSET_CHK_SYMBOL (__wmemset_chk, unaligned)): Likewise.
(WMEMSET_SYMBOL (__wmemset, unaligned)): Likewise.
* sysdeps/x86_64/multiarch/memset.S (WMEMSET_SYMBOL): New.
(libc_hidden_builtin_def): Also define __GI_wmemset and
__GI___wmemset.
(weak_alias): New.
* sysdeps/x86_64/multiarch/wmemset.c: New file.
* sysdeps/x86_64/multiarch/wmemset.h: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk-nonshared.S: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk.c: Likewise.
* sysdeps/x86_64/wmemset.c: Likewise.
* sysdeps/x86_64/wmemset_chk.c: Likewise.
2017-06-05 18:09:48 +00:00
|
|
|
IFUNC_IMPL_ADD (array, i, __wmemset_chk,
|
2021-04-19 21:48:10 +00:00
|
|
|
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
|
|
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
|
|
&& CPU_FEATURE_USABLE (BMI2)),
|
x86-64: Optimize wmemset with SSE2/AVX2/AVX512
The difference between memset and wmemset is byte vs int. Add stubs
to SSE2/AVX2/AVX512 memset for wmemset with updated constant and size:
SSE2 wmemset:
shl $0x2,%rdx
movd %esi,%xmm0
mov %rdi,%rax
pshufd $0x0,%xmm0,%xmm0
jmp entry_from_wmemset
SSE2 memset:
movd %esi,%xmm0
mov %rdi,%rax
punpcklbw %xmm0,%xmm0
punpcklwd %xmm0,%xmm0
pshufd $0x0,%xmm0,%xmm0
entry_from_wmemset:
Since the ERMS versions of wmemset requires "rep stosl" instead of
"rep stosb", only the vector store stubs of SSE2/AVX2/AVX512 wmemset
are added. The SSE2 wmemset is about 3X faster and the AVX2 wmemset
is about 6X faster on Haswell.
* include/wchar.h (__wmemset_chk): New.
* sysdeps/x86_64/memset.S (VDUP_TO_VEC0_AND_SET_RETURN): Renamed
to MEMSET_VDUP_TO_VEC0_AND_SET_RETURN.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_CHK_SYMBOL): Likewise.
(WMEMSET_SYMBOL): Likewise.
(__wmemset): Add hidden definition.
(wmemset): Add weak hidden definition.
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
wmemset_chk-nonshared.
* sysdeps/x86_64/multiarch/ifunc-impl-list.c
(__libc_ifunc_impl_list): Add __wmemset_sse2_unaligned,
__wmemset_avx2_unaligned, __wmemset_avx512_unaligned,
__wmemset_chk_sse2_unaligned, __wmemset_chk_avx2_unaligned
and __wmemset_chk_avx512_unaligned.
* sysdeps/x86_64/multiarch/memset-avx2-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S
(VDUP_TO_VEC0_AND_SET_RETURN): Renamed to ...
(MEMSET_VDUP_TO_VEC0_AND_SET_RETURN): This.
(WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN): New.
(WMEMSET_SYMBOL): Likewise.
* sysdeps/x86_64/multiarch/memset-vec-unaligned-erms.S: Updated.
(WMEMSET_CHK_SYMBOL): New.
(WMEMSET_CHK_SYMBOL (__wmemset_chk, unaligned)): Likewise.
(WMEMSET_SYMBOL (__wmemset, unaligned)): Likewise.
* sysdeps/x86_64/multiarch/memset.S (WMEMSET_SYMBOL): New.
(libc_hidden_builtin_def): Also define __GI_wmemset and
__GI___wmemset.
(weak_alias): New.
* sysdeps/x86_64/multiarch/wmemset.c: New file.
* sysdeps/x86_64/multiarch/wmemset.h: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk-nonshared.S: Likewise.
* sysdeps/x86_64/multiarch/wmemset_chk.c: Likewise.
* sysdeps/x86_64/wmemset.c: Likewise.
* sysdeps/x86_64/wmemset_chk.c: Likewise.
2017-06-05 18:09:48 +00:00
|
|
|
__wmemset_chk_avx512_unaligned))
|
2012-10-10 14:49:12 +00:00
|
|
|
#endif
|
|
|
|
|
|
2022-06-10 16:13:29 +00:00
|
|
|
return 0;
|
2012-10-10 14:49:12 +00:00
|
|
|
}
|
