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/strnlen/wcslen/wcsnlen optimized with AVX2.
|
2022-01-01 18:54:23 +00:00
|
|
|
Copyright (C) 2017-2022 Free Software Foundation, Inc.
|
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
|
|
|
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/>. */
|
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
|
|
|
|
2022-07-13 23:32:59 +00:00
|
|
|
#include <isa-level.h>
|
|
|
|
|
|
|
|
#if ISA_SHOULD_BUILD (3)
|
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
|
|
|
|
|
|
|
# include <sysdep.h>
|
|
|
|
|
|
|
|
# ifndef STRLEN
|
|
|
|
# define STRLEN __strlen_avx2
|
|
|
|
# endif
|
|
|
|
|
|
|
|
# ifdef USE_AS_WCSLEN
|
|
|
|
# define VPCMPEQ vpcmpeqd
|
|
|
|
# define VPMINU vpminud
|
2021-04-19 23:36:07 +00:00
|
|
|
# define CHAR_SIZE 4
|
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
|
|
|
# else
|
|
|
|
# define VPCMPEQ vpcmpeqb
|
|
|
|
# define VPMINU vpminub
|
2021-04-19 23:36:07 +00:00
|
|
|
# define CHAR_SIZE 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
|
|
|
# endif
|
|
|
|
|
|
|
|
# ifndef VZEROUPPER
|
|
|
|
# define VZEROUPPER vzeroupper
|
|
|
|
# endif
|
|
|
|
|
2021-03-05 15:26:42 +00:00
|
|
|
# ifndef SECTION
|
|
|
|
# define SECTION(p) p##.avx
|
|
|
|
# endif
|
|
|
|
|
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
|
|
|
# define VEC_SIZE 32
|
2021-04-19 23:36:07 +00:00
|
|
|
# define PAGE_SIZE 4096
|
2021-06-23 05:56:29 +00:00
|
|
|
# define CHAR_PER_VEC (VEC_SIZE / CHAR_SIZE)
|
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
|
|
|
|
2021-03-05 15:26:42 +00:00
|
|
|
.section SECTION(.text),"ax",@progbits
|
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
|
|
|
ENTRY (STRLEN)
|
|
|
|
# ifdef USE_AS_STRNLEN
|
2021-04-19 23:36:07 +00:00
|
|
|
/* Check zero length. */
|
2021-06-23 05:56:29 +00:00
|
|
|
# ifdef __ILP32__
|
|
|
|
/* Clear upper bits. */
|
|
|
|
and %RSI_LP, %RSI_LP
|
|
|
|
# else
|
2019-01-21 19:36:36 +00:00
|
|
|
test %RSI_LP, %RSI_LP
|
2021-06-23 05:56:29 +00:00
|
|
|
# endif
|
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
|
|
|
jz L(zero)
|
2021-04-19 23:36:07 +00:00
|
|
|
/* Store max len in R8_LP before adjusting if using WCSLEN. */
|
|
|
|
mov %RSI_LP, %R8_LP
|
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
|
|
|
# endif
|
2021-04-19 23:36:07 +00:00
|
|
|
movl %edi, %eax
|
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
|
|
|
movq %rdi, %rdx
|
|
|
|
vpxor %xmm0, %xmm0, %xmm0
|
2021-04-19 23:36:07 +00:00
|
|
|
/* Clear high bits from edi. Only keeping bits relevant to page
|
|
|
|
cross check. */
|
|
|
|
andl $(PAGE_SIZE - 1), %eax
|
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
|
|
|
/* Check if we may cross page boundary with one vector load. */
|
2021-04-19 23:36:07 +00:00
|
|
|
cmpl $(PAGE_SIZE - VEC_SIZE), %eax
|
|
|
|
ja L(cross_page_boundary)
|
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
|
|
|
|
|
|
|
/* Check the first VEC_SIZE bytes. */
|
2021-04-19 23:36:07 +00:00
|
|
|
VPCMPEQ (%rdi), %ymm0, %ymm1
|
2021-06-23 05:56:29 +00:00
|
|
|
vpmovmskb %ymm1, %eax
|
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
|
|
|
# ifdef USE_AS_STRNLEN
|
2021-04-19 23:36:07 +00:00
|
|
|
/* If length < VEC_SIZE handle special. */
|
2021-06-23 05:56:29 +00:00
|
|
|
cmpq $CHAR_PER_VEC, %rsi
|
2021-04-19 23:36:07 +00:00
|
|
|
jbe L(first_vec_x0)
|
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
|
|
|
# endif
|
2021-04-19 23:36:07 +00:00
|
|
|
/* If empty continue to aligned_more. Otherwise return bit
|
|
|
|
position of first match. */
|
|
|
|
testl %eax, %eax
|
|
|
|
jz L(aligned_more)
|
|
|
|
tzcntl %eax, %eax
|
|
|
|
# ifdef USE_AS_WCSLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* NB: Divide bytes by 4 to get wchar_t count. */
|
2021-04-19 23:36:07 +00:00
|
|
|
shrl $2, %eax
|
|
|
|
# endif
|
|
|
|
VZEROUPPER_RETURN
|
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
|
|
|
|
|
|
|
# ifdef USE_AS_STRNLEN
|
2021-04-19 23:36:07 +00:00
|
|
|
L(zero):
|
|
|
|
xorl %eax, %eax
|
|
|
|
ret
|
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
|
|
|
|
2021-04-19 23:36:07 +00:00
|
|
|
.p2align 4
|
|
|
|
L(first_vec_x0):
|
|
|
|
/* Set bit for max len so that tzcnt will return min of max len
|
|
|
|
and position of first match. */
|
2021-06-23 05:56:29 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
|
|
|
/* NB: Multiply length by 4 to get byte count. */
|
|
|
|
sall $2, %esi
|
|
|
|
# endif
|
2021-04-19 23:36:07 +00:00
|
|
|
btsq %rsi, %rax
|
|
|
|
tzcntl %eax, %eax
|
|
|
|
# ifdef USE_AS_WCSLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* NB: Divide bytes by 4 to get wchar_t count. */
|
2021-04-19 23:36:07 +00:00
|
|
|
shrl $2, %eax
|
|
|
|
# endif
|
|
|
|
VZEROUPPER_RETURN
|
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
|
|
|
# endif
|
|
|
|
|
|
|
|
.p2align 4
|
2021-04-19 23:36:07 +00:00
|
|
|
L(first_vec_x1):
|
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
|
|
|
tzcntl %eax, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
/* Safe to use 32 bit instructions as these are only called for
|
|
|
|
size = [1, 159]. */
|
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
|
|
|
# ifdef USE_AS_STRNLEN
|
2021-04-19 23:36:07 +00:00
|
|
|
/* Use ecx which was computed earlier to compute correct value.
|
|
|
|
*/
|
2021-06-23 05:56:29 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
|
|
|
leal -(VEC_SIZE * 4 + 1)(%rax, %rcx, 4), %eax
|
|
|
|
# else
|
2021-04-19 23:36:07 +00:00
|
|
|
subl $(VEC_SIZE * 4 + 1), %ecx
|
|
|
|
addl %ecx, %eax
|
2021-06-23 05:56:29 +00:00
|
|
|
# endif
|
2021-04-19 23:36:07 +00:00
|
|
|
# else
|
|
|
|
subl %edx, %edi
|
|
|
|
incl %edi
|
|
|
|
addl %edi, %eax
|
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
|
|
|
# endif
|
|
|
|
# ifdef USE_AS_WCSLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* NB: Divide bytes by 4 to get wchar_t count. */
|
2021-04-19 23:36:07 +00:00
|
|
|
shrl $2, %eax
|
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
|
|
|
# endif
|
2021-04-19 23:36:07 +00:00
|
|
|
VZEROUPPER_RETURN
|
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
|
|
|
|
|
|
|
.p2align 4
|
2021-04-19 23:36:07 +00:00
|
|
|
L(first_vec_x2):
|
|
|
|
tzcntl %eax, %eax
|
|
|
|
/* Safe to use 32 bit instructions as these are only called for
|
|
|
|
size = [1, 159]. */
|
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
|
|
|
# ifdef USE_AS_STRNLEN
|
2021-04-19 23:36:07 +00:00
|
|
|
/* Use ecx which was computed earlier to compute correct value.
|
|
|
|
*/
|
2021-06-23 05:56:29 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
|
|
|
leal -(VEC_SIZE * 3 + 1)(%rax, %rcx, 4), %eax
|
|
|
|
# else
|
2021-04-19 23:36:07 +00:00
|
|
|
subl $(VEC_SIZE * 3 + 1), %ecx
|
|
|
|
addl %ecx, %eax
|
2021-06-23 05:56:29 +00:00
|
|
|
# endif
|
2021-04-19 23:36:07 +00:00
|
|
|
# else
|
|
|
|
subl %edx, %edi
|
|
|
|
addl $(VEC_SIZE + 1), %edi
|
|
|
|
addl %edi, %eax
|
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
|
|
|
# endif
|
2021-04-19 23:36:07 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* NB: Divide bytes by 4 to get wchar_t count. */
|
2021-04-19 23:36:07 +00:00
|
|
|
shrl $2, %eax
|
|
|
|
# endif
|
|
|
|
VZEROUPPER_RETURN
|
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
|
|
|
|
2021-04-19 23:36:07 +00:00
|
|
|
.p2align 4
|
|
|
|
L(first_vec_x3):
|
|
|
|
tzcntl %eax, %eax
|
|
|
|
/* Safe to use 32 bit instructions as these are only called for
|
|
|
|
size = [1, 159]. */
|
|
|
|
# ifdef USE_AS_STRNLEN
|
|
|
|
/* Use ecx which was computed earlier to compute correct value.
|
|
|
|
*/
|
2021-06-23 05:56:29 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
|
|
|
leal -(VEC_SIZE * 2 + 1)(%rax, %rcx, 4), %eax
|
|
|
|
# else
|
2021-04-19 23:36:07 +00:00
|
|
|
subl $(VEC_SIZE * 2 + 1), %ecx
|
|
|
|
addl %ecx, %eax
|
2021-06-23 05:56:29 +00:00
|
|
|
# endif
|
2021-04-19 23:36:07 +00:00
|
|
|
# else
|
|
|
|
subl %edx, %edi
|
|
|
|
addl $(VEC_SIZE * 2 + 1), %edi
|
|
|
|
addl %edi, %eax
|
|
|
|
# endif
|
|
|
|
# ifdef USE_AS_WCSLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* NB: Divide bytes by 4 to get wchar_t count. */
|
2021-04-19 23:36:07 +00:00
|
|
|
shrl $2, %eax
|
|
|
|
# endif
|
|
|
|
VZEROUPPER_RETURN
|
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
|
|
|
|
2021-04-19 23:36:07 +00:00
|
|
|
.p2align 4
|
|
|
|
L(first_vec_x4):
|
|
|
|
tzcntl %eax, %eax
|
|
|
|
/* Safe to use 32 bit instructions as these are only called for
|
|
|
|
size = [1, 159]. */
|
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
|
|
|
# ifdef USE_AS_STRNLEN
|
2021-04-19 23:36:07 +00:00
|
|
|
/* Use ecx which was computed earlier to compute correct value.
|
|
|
|
*/
|
2021-06-23 05:56:29 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
|
|
|
leal -(VEC_SIZE * 1 + 1)(%rax, %rcx, 4), %eax
|
|
|
|
# else
|
2021-04-19 23:36:07 +00:00
|
|
|
subl $(VEC_SIZE + 1), %ecx
|
|
|
|
addl %ecx, %eax
|
2021-06-23 05:56:29 +00:00
|
|
|
# endif
|
2021-04-19 23:36:07 +00:00
|
|
|
# else
|
|
|
|
subl %edx, %edi
|
|
|
|
addl $(VEC_SIZE * 3 + 1), %edi
|
|
|
|
addl %edi, %eax
|
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
|
|
|
# endif
|
2021-04-19 23:36:07 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* NB: Divide bytes by 4 to get wchar_t count. */
|
2021-04-19 23:36:07 +00:00
|
|
|
shrl $2, %eax
|
|
|
|
# endif
|
|
|
|
VZEROUPPER_RETURN
|
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
|
|
|
|
2021-04-19 23:36:07 +00:00
|
|
|
.p2align 5
|
|
|
|
L(aligned_more):
|
|
|
|
/* Align data to VEC_SIZE - 1. This is the same number of
|
|
|
|
instructions as using andq with -VEC_SIZE but saves 4 bytes of
|
|
|
|
code on the x4 check. */
|
|
|
|
orq $(VEC_SIZE - 1), %rdi
|
|
|
|
L(cross_page_continue):
|
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
|
|
|
/* Check the first 4 * VEC_SIZE. Only one VEC_SIZE at a time
|
|
|
|
since data is only aligned to VEC_SIZE. */
|
2021-04-19 23:36:07 +00:00
|
|
|
# ifdef USE_AS_STRNLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* + 1 because rdi is aligned to VEC_SIZE - 1. + CHAR_SIZE
|
|
|
|
because it simplies the logic in last_4x_vec_or_less. */
|
2021-04-19 23:36:07 +00:00
|
|
|
leaq (VEC_SIZE * 4 + CHAR_SIZE + 1)(%rdi), %rcx
|
|
|
|
subq %rdx, %rcx
|
2021-06-23 05:56:29 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
|
|
|
/* NB: Divide bytes by 4 to get the wchar_t count. */
|
|
|
|
sarl $2, %ecx
|
|
|
|
# endif
|
2021-04-19 23:36:07 +00:00
|
|
|
# endif
|
|
|
|
/* Load first VEC regardless. */
|
|
|
|
VPCMPEQ 1(%rdi), %ymm0, %ymm1
|
|
|
|
# ifdef USE_AS_STRNLEN
|
|
|
|
/* Adjust length. If near end handle specially. */
|
|
|
|
subq %rcx, %rsi
|
|
|
|
jb L(last_4x_vec_or_less)
|
|
|
|
# endif
|
2021-06-23 05:56:29 +00:00
|
|
|
vpmovmskb %ymm1, %eax
|
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
|
|
|
testl %eax, %eax
|
|
|
|
jnz L(first_vec_x1)
|
|
|
|
|
2021-04-19 23:36:07 +00:00
|
|
|
VPCMPEQ (VEC_SIZE + 1)(%rdi), %ymm0, %ymm1
|
2021-06-23 05:56:29 +00:00
|
|
|
vpmovmskb %ymm1, %eax
|
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
|
|
|
testl %eax, %eax
|
|
|
|
jnz L(first_vec_x2)
|
|
|
|
|
2021-04-19 23:36:07 +00:00
|
|
|
VPCMPEQ (VEC_SIZE * 2 + 1)(%rdi), %ymm0, %ymm1
|
2021-06-23 05:56:29 +00:00
|
|
|
vpmovmskb %ymm1, %eax
|
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
|
|
|
testl %eax, %eax
|
|
|
|
jnz L(first_vec_x3)
|
|
|
|
|
2021-04-19 23:36:07 +00:00
|
|
|
VPCMPEQ (VEC_SIZE * 3 + 1)(%rdi), %ymm0, %ymm1
|
2021-06-23 05:56:29 +00:00
|
|
|
vpmovmskb %ymm1, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
testl %eax, %eax
|
|
|
|
jnz L(first_vec_x4)
|
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
|
|
|
|
2021-04-19 23:36:07 +00:00
|
|
|
/* Align data to VEC_SIZE * 4 - 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
|
|
|
# ifdef USE_AS_STRNLEN
|
2021-04-19 23:36:07 +00:00
|
|
|
/* Before adjusting length check if at last VEC_SIZE * 4. */
|
2021-06-23 05:56:29 +00:00
|
|
|
cmpq $(CHAR_PER_VEC * 4 - 1), %rsi
|
2021-04-19 23:36:07 +00:00
|
|
|
jbe L(last_4x_vec_or_less_load)
|
|
|
|
incq %rdi
|
|
|
|
movl %edi, %ecx
|
|
|
|
orq $(VEC_SIZE * 4 - 1), %rdi
|
|
|
|
andl $(VEC_SIZE * 4 - 1), %ecx
|
2021-06-23 05:56:29 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
|
|
|
/* NB: Divide bytes by 4 to get the wchar_t count. */
|
|
|
|
sarl $2, %ecx
|
|
|
|
# endif
|
2021-04-19 23:36:07 +00:00
|
|
|
/* Readjust length. */
|
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
|
|
|
addq %rcx, %rsi
|
2021-04-19 23:36:07 +00:00
|
|
|
# else
|
|
|
|
incq %rdi
|
|
|
|
orq $(VEC_SIZE * 4 - 1), %rdi
|
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
|
|
|
# endif
|
2021-04-19 23:36:07 +00:00
|
|
|
/* Compare 4 * VEC at a time forward. */
|
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
|
|
|
.p2align 4
|
|
|
|
L(loop_4x_vec):
|
2021-04-19 23:36:07 +00:00
|
|
|
# ifdef USE_AS_STRNLEN
|
|
|
|
/* Break if at end of length. */
|
2021-06-23 05:56:29 +00:00
|
|
|
subq $(CHAR_PER_VEC * 4), %rsi
|
2021-04-19 23:36:07 +00:00
|
|
|
jb L(last_4x_vec_or_less_cmpeq)
|
|
|
|
# endif
|
2021-06-23 05:56:29 +00:00
|
|
|
/* Save some code size by microfusing VPMINU with the load.
|
|
|
|
Since the matches in ymm2/ymm4 can only be returned if there
|
|
|
|
where no matches in ymm1/ymm3 respectively there is no issue
|
|
|
|
with overlap. */
|
2021-04-19 23:36:07 +00:00
|
|
|
vmovdqa 1(%rdi), %ymm1
|
|
|
|
VPMINU (VEC_SIZE + 1)(%rdi), %ymm1, %ymm2
|
|
|
|
vmovdqa (VEC_SIZE * 2 + 1)(%rdi), %ymm3
|
|
|
|
VPMINU (VEC_SIZE * 3 + 1)(%rdi), %ymm3, %ymm4
|
|
|
|
|
|
|
|
VPMINU %ymm2, %ymm4, %ymm5
|
|
|
|
VPCMPEQ %ymm5, %ymm0, %ymm5
|
2021-06-23 05:56:29 +00:00
|
|
|
vpmovmskb %ymm5, %ecx
|
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
|
|
|
|
2021-04-19 23:36:07 +00:00
|
|
|
subq $-(VEC_SIZE * 4), %rdi
|
|
|
|
testl %ecx, %ecx
|
|
|
|
jz L(loop_4x_vec)
|
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
|
|
|
|
|
|
|
|
2021-04-19 23:36:07 +00:00
|
|
|
VPCMPEQ %ymm1, %ymm0, %ymm1
|
2021-06-23 05:56:29 +00:00
|
|
|
vpmovmskb %ymm1, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
subq %rdx, %rdi
|
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
|
|
|
testl %eax, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
jnz L(last_vec_return_x0)
|
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
|
|
|
|
2021-04-19 23:36:07 +00:00
|
|
|
VPCMPEQ %ymm2, %ymm0, %ymm2
|
2021-06-23 05:56:29 +00:00
|
|
|
vpmovmskb %ymm2, %eax
|
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
|
|
|
testl %eax, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
jnz L(last_vec_return_x1)
|
|
|
|
|
|
|
|
/* Combine last 2 VEC. */
|
|
|
|
VPCMPEQ %ymm3, %ymm0, %ymm3
|
2021-06-23 05:56:29 +00:00
|
|
|
vpmovmskb %ymm3, %eax
|
|
|
|
/* rcx has combined result from all 4 VEC. It will only be used
|
|
|
|
if the first 3 other VEC all did not contain a match. */
|
2021-04-19 23:36:07 +00:00
|
|
|
salq $32, %rcx
|
|
|
|
orq %rcx, %rax
|
|
|
|
tzcntq %rax, %rax
|
|
|
|
subq $(VEC_SIZE * 2 - 1), %rdi
|
|
|
|
addq %rdi, %rax
|
|
|
|
# ifdef USE_AS_WCSLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* NB: Divide bytes by 4 to get wchar_t count. */
|
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
|
|
|
shrq $2, %rax
|
2021-04-19 23:36:07 +00:00
|
|
|
# endif
|
2021-03-05 15:26:42 +00:00
|
|
|
VZEROUPPER_RETURN
|
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
|
|
|
|
2021-04-19 23:36:07 +00:00
|
|
|
|
|
|
|
# ifdef USE_AS_STRNLEN
|
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
|
|
|
.p2align 4
|
2021-04-19 23:36:07 +00:00
|
|
|
L(last_4x_vec_or_less_load):
|
2021-06-23 05:56:29 +00:00
|
|
|
/* Depending on entry adjust rdi / prepare first VEC in ymm1.
|
|
|
|
*/
|
2021-04-19 23:36:07 +00:00
|
|
|
subq $-(VEC_SIZE * 4), %rdi
|
|
|
|
L(last_4x_vec_or_less_cmpeq):
|
|
|
|
VPCMPEQ 1(%rdi), %ymm0, %ymm1
|
|
|
|
L(last_4x_vec_or_less):
|
2021-06-23 05:56:29 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
|
|
|
/* NB: Multiply length by 4 to get byte count. */
|
|
|
|
sall $2, %esi
|
|
|
|
# endif
|
|
|
|
vpmovmskb %ymm1, %eax
|
|
|
|
/* If remaining length > VEC_SIZE * 2. This works if esi is off
|
|
|
|
by VEC_SIZE * 4. */
|
2021-04-19 23:36:07 +00:00
|
|
|
testl $(VEC_SIZE * 2), %esi
|
|
|
|
jnz L(last_4x_vec)
|
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
|
|
|
|
2021-04-19 23:36:07 +00:00
|
|
|
/* length may have been negative or positive by an offset of
|
|
|
|
VEC_SIZE * 4 depending on where this was called from. This fixes
|
|
|
|
that. */
|
|
|
|
andl $(VEC_SIZE * 4 - 1), %esi
|
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
|
|
|
testl %eax, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
jnz L(last_vec_x1_check)
|
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
|
|
|
|
2021-04-19 23:36:07 +00:00
|
|
|
subl $VEC_SIZE, %esi
|
|
|
|
jb L(max)
|
|
|
|
|
|
|
|
VPCMPEQ (VEC_SIZE + 1)(%rdi), %ymm0, %ymm1
|
2021-06-23 05:56:29 +00:00
|
|
|
vpmovmskb %ymm1, %eax
|
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
|
|
|
tzcntl %eax, %eax
|
|
|
|
/* Check the end of data. */
|
2021-04-19 23:36:07 +00:00
|
|
|
cmpl %eax, %esi
|
|
|
|
jb L(max)
|
|
|
|
subq %rdx, %rdi
|
|
|
|
addl $(VEC_SIZE + 1), %eax
|
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
|
|
|
addq %rdi, %rax
|
|
|
|
# ifdef USE_AS_WCSLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* NB: Divide bytes by 4 to get wchar_t count. */
|
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
|
|
|
shrq $2, %rax
|
|
|
|
# endif
|
2021-03-05 15:26:42 +00:00
|
|
|
VZEROUPPER_RETURN
|
2021-04-19 23:36:07 +00:00
|
|
|
# endif
|
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
|
|
|
|
|
|
|
.p2align 4
|
2021-04-19 23:36:07 +00:00
|
|
|
L(last_vec_return_x0):
|
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
|
|
|
tzcntl %eax, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
subq $(VEC_SIZE * 4 - 1), %rdi
|
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
|
|
|
addq %rdi, %rax
|
2021-04-19 23:36:07 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* NB: Divide bytes by 4 to get wchar_t count. */
|
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
|
|
|
shrq $2, %rax
|
2021-04-19 23:36:07 +00:00
|
|
|
# endif
|
2021-03-05 15:26:42 +00:00
|
|
|
VZEROUPPER_RETURN
|
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
|
|
|
|
|
|
|
.p2align 4
|
2021-04-19 23:36:07 +00:00
|
|
|
L(last_vec_return_x1):
|
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
|
|
|
tzcntl %eax, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
subq $(VEC_SIZE * 3 - 1), %rdi
|
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
|
|
|
addq %rdi, %rax
|
2021-04-19 23:36:07 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* NB: Divide bytes by 4 to get wchar_t count. */
|
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
|
|
|
shrq $2, %rax
|
2021-04-19 23:36:07 +00:00
|
|
|
# endif
|
2021-03-05 15:26:42 +00:00
|
|
|
VZEROUPPER_RETURN
|
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
|
|
|
|
2021-04-19 23:36:07 +00:00
|
|
|
# ifdef USE_AS_STRNLEN
|
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
|
|
|
.p2align 4
|
2021-04-19 23:36:07 +00:00
|
|
|
L(last_vec_x1_check):
|
|
|
|
|
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
|
|
|
tzcntl %eax, %eax
|
|
|
|
/* Check the end of data. */
|
2021-04-19 23:36:07 +00:00
|
|
|
cmpl %eax, %esi
|
|
|
|
jb L(max)
|
|
|
|
subq %rdx, %rdi
|
|
|
|
incl %eax
|
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
|
|
|
addq %rdi, %rax
|
|
|
|
# ifdef USE_AS_WCSLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* NB: Divide bytes by 4 to get wchar_t count. */
|
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
|
|
|
shrq $2, %rax
|
|
|
|
# endif
|
2021-03-05 15:26:42 +00:00
|
|
|
VZEROUPPER_RETURN
|
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
|
|
|
|
|
|
|
L(max):
|
|
|
|
movq %r8, %rax
|
2021-04-19 23:36:07 +00:00
|
|
|
VZEROUPPER_RETURN
|
|
|
|
|
|
|
|
.p2align 4
|
|
|
|
L(last_4x_vec):
|
|
|
|
/* Test first 2x VEC normally. */
|
|
|
|
testl %eax, %eax
|
|
|
|
jnz L(last_vec_x1)
|
|
|
|
|
|
|
|
VPCMPEQ (VEC_SIZE + 1)(%rdi), %ymm0, %ymm1
|
2021-06-23 05:56:29 +00:00
|
|
|
vpmovmskb %ymm1, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
testl %eax, %eax
|
|
|
|
jnz L(last_vec_x2)
|
|
|
|
|
|
|
|
/* Normalize length. */
|
|
|
|
andl $(VEC_SIZE * 4 - 1), %esi
|
|
|
|
VPCMPEQ (VEC_SIZE * 2 + 1)(%rdi), %ymm0, %ymm1
|
2021-06-23 05:56:29 +00:00
|
|
|
vpmovmskb %ymm1, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
testl %eax, %eax
|
|
|
|
jnz L(last_vec_x3)
|
|
|
|
|
|
|
|
subl $(VEC_SIZE * 3), %esi
|
|
|
|
jb L(max)
|
|
|
|
|
|
|
|
VPCMPEQ (VEC_SIZE * 3 + 1)(%rdi), %ymm0, %ymm1
|
2021-06-23 05:56:29 +00:00
|
|
|
vpmovmskb %ymm1, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
tzcntl %eax, %eax
|
|
|
|
/* Check the end of data. */
|
|
|
|
cmpl %eax, %esi
|
|
|
|
jb L(max)
|
|
|
|
subq %rdx, %rdi
|
|
|
|
addl $(VEC_SIZE * 3 + 1), %eax
|
|
|
|
addq %rdi, %rax
|
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
|
|
|
# ifdef USE_AS_WCSLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* NB: Divide bytes by 4 to get wchar_t count. */
|
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
|
|
|
shrq $2, %rax
|
|
|
|
# endif
|
2021-03-05 15:26:42 +00:00
|
|
|
VZEROUPPER_RETURN
|
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
|
|
|
|
|
|
|
|
|
|
|
.p2align 4
|
2021-04-19 23:36:07 +00:00
|
|
|
L(last_vec_x1):
|
|
|
|
/* essentially duplicates of first_vec_x1 but use 64 bit
|
|
|
|
instructions. */
|
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
|
|
|
tzcntl %eax, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
subq %rdx, %rdi
|
|
|
|
incl %eax
|
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
|
|
|
addq %rdi, %rax
|
2021-04-19 23:36:07 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* NB: Divide bytes by 4 to get wchar_t count. */
|
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
|
|
|
shrq $2, %rax
|
2021-04-19 23:36:07 +00:00
|
|
|
# endif
|
2021-03-05 15:26:42 +00:00
|
|
|
VZEROUPPER_RETURN
|
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
|
|
|
|
|
|
|
.p2align 4
|
2021-04-19 23:36:07 +00:00
|
|
|
L(last_vec_x2):
|
|
|
|
/* essentially duplicates of first_vec_x1 but use 64 bit
|
|
|
|
instructions. */
|
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
|
|
|
tzcntl %eax, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
subq %rdx, %rdi
|
|
|
|
addl $(VEC_SIZE + 1), %eax
|
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
|
|
|
addq %rdi, %rax
|
2021-04-19 23:36:07 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* NB: Divide bytes by 4 to get wchar_t count. */
|
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
|
|
|
shrq $2, %rax
|
2021-04-19 23:36:07 +00:00
|
|
|
# endif
|
2021-03-05 15:26:42 +00:00
|
|
|
VZEROUPPER_RETURN
|
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
|
|
|
|
|
|
|
.p2align 4
|
2021-04-19 23:36:07 +00:00
|
|
|
L(last_vec_x3):
|
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
|
|
|
tzcntl %eax, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
subl $(VEC_SIZE * 2), %esi
|
|
|
|
/* Check the end of data. */
|
|
|
|
cmpl %eax, %esi
|
|
|
|
jb L(max_end)
|
|
|
|
subq %rdx, %rdi
|
|
|
|
addl $(VEC_SIZE * 2 + 1), %eax
|
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
|
|
|
addq %rdi, %rax
|
2021-04-19 23:36:07 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* NB: Divide bytes by 4 to get wchar_t count. */
|
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
|
|
|
shrq $2, %rax
|
2021-04-19 23:36:07 +00:00
|
|
|
# endif
|
|
|
|
VZEROUPPER_RETURN
|
|
|
|
L(max_end):
|
|
|
|
movq %r8, %rax
|
2021-03-05 15:26:42 +00:00
|
|
|
VZEROUPPER_RETURN
|
2021-04-19 23:36:07 +00:00
|
|
|
# endif
|
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
|
|
|
|
2021-06-23 05:56:29 +00:00
|
|
|
/* Cold case for crossing page with first load. */
|
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
|
|
|
.p2align 4
|
2021-04-19 23:36:07 +00:00
|
|
|
L(cross_page_boundary):
|
|
|
|
/* Align data to VEC_SIZE - 1. */
|
|
|
|
orq $(VEC_SIZE - 1), %rdi
|
|
|
|
VPCMPEQ -(VEC_SIZE - 1)(%rdi), %ymm0, %ymm1
|
2021-06-23 05:56:29 +00:00
|
|
|
vpmovmskb %ymm1, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
/* Remove the leading bytes. sarxl only uses bits [5:0] of COUNT
|
|
|
|
so no need to manually mod rdx. */
|
|
|
|
sarxl %edx, %eax, %eax
|
|
|
|
# ifdef USE_AS_STRNLEN
|
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
|
|
|
testl %eax, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
jnz L(cross_page_less_vec)
|
|
|
|
leaq 1(%rdi), %rcx
|
|
|
|
subq %rdx, %rcx
|
2021-06-23 05:56:29 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
|
|
|
/* NB: Divide bytes by 4 to get wchar_t count. */
|
|
|
|
shrl $2, %ecx
|
|
|
|
# endif
|
2021-04-19 23:36:07 +00:00
|
|
|
/* Check length. */
|
|
|
|
cmpq %rsi, %rcx
|
|
|
|
jb L(cross_page_continue)
|
|
|
|
movq %r8, %rax
|
|
|
|
# else
|
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
|
|
|
testl %eax, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
jz L(cross_page_continue)
|
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
|
|
|
tzcntl %eax, %eax
|
2021-04-19 23:36:07 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
2021-06-23 05:56:29 +00:00
|
|
|
/* NB: Divide length by 4 to get wchar_t count. */
|
2021-04-19 23:36:07 +00:00
|
|
|
shrl $2, %eax
|
|
|
|
# endif
|
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
|
|
|
# endif
|
2021-04-19 23:36:07 +00:00
|
|
|
L(return_vzeroupper):
|
|
|
|
ZERO_UPPER_VEC_REGISTERS_RETURN
|
|
|
|
|
|
|
|
# ifdef USE_AS_STRNLEN
|
|
|
|
.p2align 4
|
|
|
|
L(cross_page_less_vec):
|
|
|
|
tzcntl %eax, %eax
|
2021-06-23 05:56:29 +00:00
|
|
|
# ifdef USE_AS_WCSLEN
|
|
|
|
/* NB: Multiply length by 4 to get byte count. */
|
|
|
|
sall $2, %esi
|
|
|
|
# endif
|
2021-04-19 23:36:07 +00:00
|
|
|
cmpq %rax, %rsi
|
|
|
|
cmovb %esi, %eax
|
|
|
|
# ifdef USE_AS_WCSLEN
|
|
|
|
shrl $2, %eax
|
|
|
|
# endif
|
2021-03-05 15:26:42 +00:00
|
|
|
VZEROUPPER_RETURN
|
2021-04-19 23:36:07 +00:00
|
|
|
# endif
|
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
|
|
|
|
|
|
|
END (STRLEN)
|
|
|
|
#endif
|