* [PATCH] x86-64: Optimize strlen/strnlen/wcslen/wcsnlen with AVX2
@ 2017-06-01 18:13 H.J. Lu
2017-06-02 19:48 ` H.J. Lu
0 siblings, 1 reply; 3+ messages in thread
From: H.J. Lu @ 2017-06-01 18:13 UTC (permalink / raw)
To: GNU C Library
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.
Any comments?
H.J.
---
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
strlen-avx2, strnlen-avx2 and 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, __wcsnlen_avx2 and __wcsnlen_sse2.
* sysdeps/x86_64/multiarch/strlen-avx2.S: New file.
* sysdeps/x86_64/multiarch/strlen.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/strnlen.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcslen.S: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen-avx2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen.S: Likewise.
---
sysdeps/x86_64/multiarch/Makefile | 4 +-
sysdeps/x86_64/multiarch/ifunc-impl-list.c | 28 ++
sysdeps/x86_64/multiarch/strlen-avx2.S | 394 +++++++++++++++++++++++++++++
sysdeps/x86_64/multiarch/strlen.S | 64 +++++
sysdeps/x86_64/multiarch/strnlen-avx2.S | 4 +
sysdeps/x86_64/multiarch/strnlen.S | 65 +++++
sysdeps/x86_64/multiarch/wcslen-avx2.S | 4 +
sysdeps/x86_64/multiarch/wcslen.S | 55 ++++
sysdeps/x86_64/multiarch/wcsnlen-avx2.S | 5 +
sysdeps/x86_64/multiarch/wcsnlen.S | 55 ++++
10 files changed, 677 insertions(+), 1 deletion(-)
create mode 100644 sysdeps/x86_64/multiarch/strlen-avx2.S
create mode 100644 sysdeps/x86_64/multiarch/strlen.S
create mode 100644 sysdeps/x86_64/multiarch/strnlen-avx2.S
create mode 100644 sysdeps/x86_64/multiarch/strnlen.S
create mode 100644 sysdeps/x86_64/multiarch/wcslen-avx2.S
create mode 100644 sysdeps/x86_64/multiarch/wcslen.S
create mode 100644 sysdeps/x86_64/multiarch/wcsnlen-avx2.S
create mode 100644 sysdeps/x86_64/multiarch/wcsnlen.S
diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile
index 48aba0f..f33f21a 100644
--- a/sysdeps/x86_64/multiarch/Makefile
+++ b/sysdeps/x86_64/multiarch/Makefile
@@ -13,6 +13,7 @@ sysdep_routines += strncat-c stpncpy-c strncpy-c strcmp-ssse3 \
memcpy-ssse3-back \
memmove-ssse3-back \
memmove-avx512-no-vzeroupper strcasecmp_l-ssse3 \
+ strlen-avx2 strnlen-avx2 \
strncase_l-ssse3 strcat-ssse3 strncat-ssse3\
strcpy-ssse3 strncpy-ssse3 stpcpy-ssse3 stpncpy-ssse3 \
strcpy-sse2-unaligned strncpy-sse2-unaligned \
@@ -35,5 +36,6 @@ ifeq ($(subdir),wcsmbs)
sysdep_routines += wmemcmp-sse4 wmemcmp-ssse3 wmemcmp-c \
wmemchr-avx2 \
wmemcmp-avx2 \
- wcscpy-ssse3 wcscpy-c
+ wcscpy-ssse3 wcscpy-c \
+ wcslen-avx2 wcsnlen-avx2
endif
diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
index ae09241..c2b07b3 100644
--- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
@@ -165,6 +165,20 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
__rawmemchr_avx2)
IFUNC_IMPL_ADD (array, i, rawmemchr, 1, __rawmemchr_sse2))
+ /* Support sysdeps/x86_64/multiarch/strlen.S. */
+ IFUNC_IMPL (i, name, strlen,
+ IFUNC_IMPL_ADD (array, i, strlen,
+ HAS_ARCH_FEATURE (AVX2_Usable),
+ __strlen_avx2)
+ IFUNC_IMPL_ADD (array, i, strlen, 1, __strlen_sse2))
+
+ /* Support sysdeps/x86_64/multiarch/strnlen.S. */
+ IFUNC_IMPL (i, name, strnlen,
+ IFUNC_IMPL_ADD (array, i, strnlen,
+ HAS_ARCH_FEATURE (AVX2_Usable),
+ __strnlen_avx2)
+ IFUNC_IMPL_ADD (array, i, strnlen, 1, __strnlen_sse2))
+
/* Support sysdeps/x86_64/multiarch/stpncpy.S. */
IFUNC_IMPL (i, name, stpncpy,
IFUNC_IMPL_ADD (array, i, stpncpy, HAS_CPU_FEATURE (SSSE3),
@@ -309,6 +323,20 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
__wcscpy_ssse3)
IFUNC_IMPL_ADD (array, i, wcscpy, 1, __wcscpy_sse2))
+ /* Support sysdeps/x86_64/multiarch/wcslen.S. */
+ IFUNC_IMPL (i, name, wcslen,
+ IFUNC_IMPL_ADD (array, i, wcslen,
+ HAS_ARCH_FEATURE (AVX2_Usable),
+ __wcslen_avx2)
+ IFUNC_IMPL_ADD (array, i, wcslen, 1, __wcslen_sse2))
+
+ /* Support sysdeps/x86_64/multiarch/wcsnlen.S. */
+ IFUNC_IMPL (i, name, wcsnlen,
+ IFUNC_IMPL_ADD (array, i, wcsnlen,
+ HAS_ARCH_FEATURE (AVX2_Usable),
+ __wcsnlen_avx2)
+ IFUNC_IMPL_ADD (array, i, wcsnlen, 1, __wcsnlen_sse2))
+
/* Support sysdeps/x86_64/multiarch/wmemchr.S. */
IFUNC_IMPL (i, name, wmemchr,
IFUNC_IMPL_ADD (array, i, wmemchr,
diff --git a/sysdeps/x86_64/multiarch/strlen-avx2.S b/sysdeps/x86_64/multiarch/strlen-avx2.S
new file mode 100644
index 0000000..1dc823a
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strlen-avx2.S
@@ -0,0 +1,394 @@
+/* strlen/strnlen/wcslen/wcsnlen optimized with AVX2.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#if IS_IN (libc)
+
+# include <sysdep.h>
+
+# ifndef STRLEN
+# define STRLEN __strlen_avx2
+# endif
+
+# ifdef USE_AS_WCSLEN
+# define VPCMPEQ vpcmpeqd
+# define VPMINU vpminud
+# else
+# define VPCMPEQ vpcmpeqb
+# define VPMINU vpminub
+# endif
+
+# ifndef VZEROUPPER
+# define VZEROUPPER vzeroupper
+# endif
+
+# define VEC_SIZE 32
+
+ .section .text.avx,"ax",@progbits
+ENTRY (STRLEN)
+# ifdef USE_AS_STRNLEN
+ /* Check for zero length. */
+ testq %rsi, %rsi
+ jz L(zero)
+# ifdef USE_AS_WCSLEN
+ shl $2, %rsi
+# endif
+ movq %rsi, %r8
+# endif
+ movl %edi, %ecx
+ movq %rdi, %rdx
+ vpxor %xmm0, %xmm0, %xmm0
+
+ /* Check if we may cross page boundary with one vector load. */
+ andl $(2 * VEC_SIZE - 1), %ecx
+ cmpl $VEC_SIZE, %ecx
+ ja L(cros_page_boundary)
+
+ /* Check the first VEC_SIZE bytes. */
+ VPCMPEQ (%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+
+# ifdef USE_AS_STRNLEN
+ jnz L(first_vec_x0_check)
+ /* Adjust length and check the end of data. */
+ subq $VEC_SIZE, %rsi
+ jbe L(max)
+# else
+ jnz L(first_vec_x0)
+# endif
+
+ /* Align data for aligned loads in the loop. */
+ addq $VEC_SIZE, %rdi
+ andl $(VEC_SIZE - 1), %ecx
+ andq $-VEC_SIZE, %rdi
+
+# ifdef USE_AS_STRNLEN
+ /* Adjust length. */
+ addq %rcx, %rsi
+
+ subq $(VEC_SIZE * 4), %rsi
+ jbe L(last_4x_vec_or_less)
+# endif
+ jmp L(more_4x_vec)
+
+ .p2align 4
+L(cros_page_boundary):
+ andl $(VEC_SIZE - 1), %ecx
+ andq $-VEC_SIZE, %rdi
+ VPCMPEQ (%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ /* Remove the leading bytes. */
+ sarl %cl, %eax
+ testl %eax, %eax
+ jz L(aligned_more)
+ tzcntl %eax, %eax
+# ifdef USE_AS_STRNLEN
+ /* Check the end of data. */
+ cmpq %rax, %rsi
+ jbe L(max)
+# endif
+ addq %rdi, %rax
+ addq %rcx, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(aligned_more):
+# ifdef USE_AS_STRNLEN
+ /* "rcx" is less than VEC_SIZE. Calculate "rdx + rcx - VEC_SIZE"
+ with "rdx - (VEC_SIZE - rcx)" instead of "(rdx + rcx) - VEC_SIZE"
+ to void possible addition overflow. */
+ negq %rcx
+ addq $VEC_SIZE, %rcx
+
+ /* Check the end of data. */
+ subq %rcx, %rsi
+ jbe L(max)
+# endif
+
+ addq $VEC_SIZE, %rdi
+
+# ifdef USE_AS_STRNLEN
+ subq $(VEC_SIZE * 4), %rsi
+ jbe L(last_4x_vec_or_less)
+# endif
+
+L(more_4x_vec):
+ /* Check the first 4 * VEC_SIZE. Only one VEC_SIZE at a time
+ since data is only aligned to VEC_SIZE. */
+ VPCMPEQ (%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x0)
+
+ VPCMPEQ VEC_SIZE(%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x1)
+
+ VPCMPEQ (VEC_SIZE * 2)(%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x2)
+
+ VPCMPEQ (VEC_SIZE * 3)(%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x3)
+
+ addq $(VEC_SIZE * 4), %rdi
+
+# ifdef USE_AS_STRNLEN
+ subq $(VEC_SIZE * 4), %rsi
+ jbe L(last_4x_vec_or_less)
+# endif
+
+ /* Align data to 4 * VEC_SIZE. */
+ movq %rdi, %rcx
+ andl $(4 * VEC_SIZE - 1), %ecx
+ andq $-(4 * VEC_SIZE), %rdi
+
+# ifdef USE_AS_STRNLEN
+ /* Adjust length. */
+ addq %rcx, %rsi
+# endif
+
+ .p2align 4
+L(loop_4x_vec):
+ /* Compare 4 * VEC at a time forward. */
+ vmovdqa (%rdi), %ymm1
+ vmovdqa VEC_SIZE(%rdi), %ymm2
+ vmovdqa (VEC_SIZE * 2)(%rdi), %ymm3
+ vmovdqa (VEC_SIZE * 3)(%rdi), %ymm4
+ VPMINU %ymm1, %ymm2, %ymm5
+ VPMINU %ymm3, %ymm4, %ymm6
+ VPMINU %ymm5, %ymm6, %ymm5
+
+ VPCMPEQ %ymm5, %ymm0, %ymm5
+ vpmovmskb %ymm5, %eax
+ testl %eax, %eax
+ jnz L(4x_vec_end)
+
+ addq $(VEC_SIZE * 4), %rdi
+
+# ifndef USE_AS_STRNLEN
+ jmp L(loop_4x_vec)
+# else
+ subq $(VEC_SIZE * 4), %rsi
+ ja L(loop_4x_vec)
+
+L(last_4x_vec_or_less):
+ /* Less than 4 * VEC and aligned to VEC_SIZE. */
+ addl $(VEC_SIZE * 2), %esi
+ jle L(last_2x_vec)
+
+ VPCMPEQ (%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x0)
+
+ VPCMPEQ VEC_SIZE(%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x1)
+
+ VPCMPEQ (VEC_SIZE * 2)(%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+
+ jnz L(first_vec_x2_check)
+ subl $VEC_SIZE, %esi
+ jle L(max)
+
+ VPCMPEQ (VEC_SIZE * 3)(%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+
+ jnz L(first_vec_x3_check)
+ movq %r8, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(last_2x_vec):
+ addl $(VEC_SIZE * 2), %esi
+ VPCMPEQ (%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+
+ jnz L(first_vec_x0_check)
+ subl $VEC_SIZE, %esi
+ jle L(max)
+
+ VPCMPEQ VEC_SIZE(%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x1_check)
+ movq %r8, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(first_vec_x0_check):
+ tzcntl %eax, %eax
+ /* Check the end of data. */
+ cmpq %rax, %rsi
+ jbe L(max)
+ addq %rdi, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(first_vec_x1_check):
+ tzcntl %eax, %eax
+ /* Check the end of data. */
+ cmpq %rax, %rsi
+ jbe L(max)
+ addq $VEC_SIZE, %rax
+ addq %rdi, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(first_vec_x2_check):
+ tzcntl %eax, %eax
+ /* Check the end of data. */
+ cmpq %rax, %rsi
+ jbe L(max)
+ addq $(VEC_SIZE * 2), %rax
+ addq %rdi, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(first_vec_x3_check):
+ tzcntl %eax, %eax
+ /* Check the end of data. */
+ cmpq %rax, %rsi
+ jbe L(max)
+ addq $(VEC_SIZE * 3), %rax
+ addq %rdi, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(max):
+ movq %r8, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(zero):
+ xorl %eax, %eax
+ ret
+# endif
+
+ .p2align 4
+L(first_vec_x0):
+ tzcntl %eax, %eax
+ addq %rdi, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(first_vec_x1):
+ tzcntl %eax, %eax
+ addq $VEC_SIZE, %rax
+ addq %rdi, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(first_vec_x2):
+ tzcntl %eax, %eax
+ addq $(VEC_SIZE * 2), %rax
+ addq %rdi, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(4x_vec_end):
+ VPCMPEQ %ymm1, %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x0)
+ VPCMPEQ %ymm2, %ymm0, %ymm2
+ vpmovmskb %ymm2, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x1)
+ VPCMPEQ %ymm3, %ymm0, %ymm3
+ vpmovmskb %ymm3, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x2)
+ VPCMPEQ %ymm4, %ymm0, %ymm4
+ vpmovmskb %ymm4, %eax
+ testl %eax, %eax
+L(first_vec_x3):
+ tzcntl %eax, %eax
+ addq $(VEC_SIZE * 3), %rax
+ addq %rdi, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+END (STRLEN)
+#endif
diff --git a/sysdeps/x86_64/multiarch/strlen.S b/sysdeps/x86_64/multiarch/strlen.S
new file mode 100644
index 0000000..2847440
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strlen.S
@@ -0,0 +1,64 @@
+/* Multiple versions of strlen
+ All versions must be listed in ifunc-impl-list.c.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include <sysdep.h>
+#include <init-arch.h>
+
+/* Define multiple versions only for the definition in libc. */
+#if IS_IN (libc)
+ .text
+ENTRY(strlen)
+ .type strlen, @gnu_indirect_function
+ LOAD_RTLD_GLOBAL_RO_RDX
+ HAS_ARCH_FEATURE (Prefer_No_VZEROUPPER)
+ jnz 1f
+ HAS_ARCH_FEATURE (AVX2_Usable)
+ jz 1f
+ HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load)
+ jz 1f
+ leaq __strlen_avx2(%rip), %rax
+ ret
+
+1: leaq __strlen_sse2(%rip), %rax
+ ret
+END(strlen)
+
+# undef ENTRY
+# define ENTRY(name) \
+ .type __strlen_sse2, @function; \
+ .p2align 4; \
+ .globl __strlen_sse2; \
+ .hidden __strlen_sse2; \
+ __strlen_sse2: cfi_startproc; \
+ CALL_MCOUNT
+# undef END
+# define END(name) \
+ cfi_endproc; .size __strlen_sse2, .-__strlen_sse2
+
+# ifdef SHARED
+# undef libc_hidden_builtin_def
+/* It doesn't make sense to send libc-internal strlen calls through a PLT.
+ The speedup we get from using AVX2 instructions is likely eaten away
+ by the indirect call in the PLT. */
+# define libc_hidden_builtin_def(name) \
+ .globl __GI_strlen; __GI_strlen = __strlen_sse2
+# endif
+#endif
+
+#include "../strlen.S"
diff --git a/sysdeps/x86_64/multiarch/strnlen-avx2.S b/sysdeps/x86_64/multiarch/strnlen-avx2.S
new file mode 100644
index 0000000..c4062b2
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strnlen-avx2.S
@@ -0,0 +1,4 @@
+#define STRLEN __strnlen_avx2
+#define USE_AS_STRNLEN 1
+
+#include "strlen-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/strnlen.S b/sysdeps/x86_64/multiarch/strnlen.S
new file mode 100644
index 0000000..0c2289a
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strnlen.S
@@ -0,0 +1,65 @@
+/* Multiple versions of strnlen
+ All versions must be listed in ifunc-impl-list.c.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include <sysdep.h>
+#include <init-arch.h>
+
+/* Define multiple versions only for the definition in libc. */
+#if IS_IN (libc)
+ .text
+ENTRY(__strnlen)
+ .type __strnlen, @gnu_indirect_function
+ LOAD_RTLD_GLOBAL_RO_RDX
+ HAS_ARCH_FEATURE (Prefer_No_VZEROUPPER)
+ jnz 1f
+ HAS_ARCH_FEATURE (AVX2_Usable)
+ jz 1f
+ HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load)
+ jz 1f
+ leaq __strnlen_avx2(%rip), %rax
+ ret
+
+1: leaq __strnlen_sse2(%rip), %rax
+ ret
+END(__strnlen)
+
+# undef ENTRY
+# define ENTRY(name) \
+ .type __strnlen_sse2, @function; \
+ .p2align 4; \
+ .globl __strnlen_sse2; \
+ .hidden __strnlen_sse2; \
+ __strnlen_sse2: cfi_startproc; \
+ CALL_MCOUNT
+# undef END
+# define END(name) \
+ cfi_endproc; .size __strnlen_sse2, .-__strnlen_sse2
+
+# ifdef SHARED
+/* It doesn't make sense to send libc-internal strnlen calls through a PLT.
+ The speedup we get from using AVX2 instructions is likely eaten away
+ by the indirect call in the PLT. */
+# undef libc_hidden_def
+# define libc_hidden_def(name) \
+ .globl __GI_strnlen; __GI_strnlen = __strnlen_sse2; \
+ .globl __GI___strnlen; __GI___strnlen = __strnlen_sse2
+# endif
+#endif
+
+#include "../strnlen.S"
diff --git a/sysdeps/x86_64/multiarch/wcslen-avx2.S b/sysdeps/x86_64/multiarch/wcslen-avx2.S
new file mode 100644
index 0000000..c9224f1
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcslen-avx2.S
@@ -0,0 +1,4 @@
+#define STRLEN __wcslen_avx2
+#define USE_AS_WCSLEN 1
+
+#include "strlen-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcslen.S b/sysdeps/x86_64/multiarch/wcslen.S
new file mode 100644
index 0000000..04369b6
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcslen.S
@@ -0,0 +1,55 @@
+/* Multiple versions of wcslen
+ All versions must be listed in ifunc-impl-list.c.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include <sysdep.h>
+#include <init-arch.h>
+
+/* Define multiple versions only for the definition in libc. */
+#if IS_IN (libc)
+ .text
+ENTRY(__wcslen)
+ .type __wcslen, @gnu_indirect_function
+ LOAD_RTLD_GLOBAL_RO_RDX
+ HAS_ARCH_FEATURE (Prefer_No_VZEROUPPER)
+ jnz 1f
+ HAS_ARCH_FEATURE (AVX2_Usable)
+ jz 1f
+ HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load)
+ jz 1f
+ leaq __wcslen_avx2(%rip), %rax
+ ret
+
+1: leaq __wcslen_sse2(%rip), %rax
+ ret
+END(__wcslen)
+
+# undef ENTRY
+# define ENTRY(name) \
+ .type __wcslen_sse2, @function; \
+ .p2align 4; \
+ .globl __wcslen_sse2; \
+ .hidden __wcslen_sse2; \
+ __wcslen_sse2: cfi_startproc; \
+ CALL_MCOUNT
+# undef END
+# define END(name) \
+ cfi_endproc; .size __wcslen_sse2, .-__wcslen_sse2
+#endif
+
+#include "../wcslen.S"
diff --git a/sysdeps/x86_64/multiarch/wcsnlen-avx2.S b/sysdeps/x86_64/multiarch/wcsnlen-avx2.S
new file mode 100644
index 0000000..fac8354
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcsnlen-avx2.S
@@ -0,0 +1,5 @@
+#define STRLEN __wcsnlen_avx2
+#define USE_AS_WCSLEN 1
+#define USE_AS_STRNLEN 1
+
+#include "strlen-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcsnlen.S b/sysdeps/x86_64/multiarch/wcsnlen.S
new file mode 100644
index 0000000..0893bea
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcsnlen.S
@@ -0,0 +1,55 @@
+/* Multiple versions of wcsnlen
+ All versions must be listed in ifunc-impl-list.c.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include <sysdep.h>
+#include <init-arch.h>
+
+/* Define multiple versions only for the definition in libc. */
+#if IS_IN (libc)
+ .text
+ENTRY(__wcsnlen)
+ .type __wcsnlen, @gnu_indirect_function
+ LOAD_RTLD_GLOBAL_RO_RDX
+ HAS_ARCH_FEATURE (Prefer_No_VZEROUPPER)
+ jnz 1f
+ HAS_ARCH_FEATURE (AVX2_Usable)
+ jz 1f
+ HAS_ARCH_FEATURE (AVX_Fast_Unaligned_Load)
+ jz 1f
+ leaq __wcsnlen_avx2(%rip), %rax
+ ret
+
+1: leaq __wcsnlen_sse2(%rip), %rax
+ ret
+END(__wcsnlen)
+
+# undef ENTRY
+# define ENTRY(name) \
+ .type __wcsnlen_sse2, @function; \
+ .p2align 4; \
+ .globl __wcsnlen_sse2; \
+ .hidden __wcsnlen_sse2; \
+ __wcsnlen_sse2: cfi_startproc; \
+ CALL_MCOUNT
+# undef END
+# define END(name) \
+ cfi_endproc; .size __wcsnlen_sse2, .-__wcsnlen_sse2
+#endif
+
+#include "../wcsnlen.S"
--
2.9.4
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] x86-64: Optimize strlen/strnlen/wcslen/wcsnlen with AVX2
2017-06-01 18:13 [PATCH] x86-64: Optimize strlen/strnlen/wcslen/wcsnlen with AVX2 H.J. Lu
@ 2017-06-02 19:48 ` H.J. Lu
2017-06-09 11:32 ` H.J. Lu
0 siblings, 1 reply; 3+ messages in thread
From: H.J. Lu @ 2017-06-02 19:48 UTC (permalink / raw)
To: GNU C Library
[-- Attachment #1: Type: text/plain, Size: 1541 bytes --]
On Thu, Jun 1, 2017 at 11:12 AM, H.J. Lu <hongjiu.lu@intel.com> wrote:
> 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.
>
> Any comments?
>
> H.J.
> ---
> * sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
> strlen-avx2, strnlen-avx2 and 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, __wcsnlen_avx2 and __wcsnlen_sse2.
> * sysdeps/x86_64/multiarch/strlen-avx2.S: New file.
> * sysdeps/x86_64/multiarch/strlen.S: Likewise.
> * sysdeps/x86_64/multiarch/strnlen-avx2.S: Likewise.
> * sysdeps/x86_64/multiarch/strnlen.S: Likewise.
> * sysdeps/x86_64/multiarch/wcslen-avx2.S: Likewise.
> * sysdeps/x86_64/multiarch/wcslen.S: Likewise.
> * sysdeps/x86_64/multiarch/wcsnlen-avx2.S: Likewise.
> * sysdeps/x86_64/multiarch/wcsnlen.S: Likewise.
Updated patch with IFUNC selector in C.
--
H.J.
[-- Attachment #2: 0005-x86-64-Optimize-strlen-strnlen-wcslen-wcsnlen-with-A.patch --]
[-- Type: text/x-patch, Size: 26364 bytes --]
From eca10ba617eea905abfaa7576f6185c1f3f47fa5 Mon Sep 17 00:00:00 2001
From: "H.J. Lu" <hjl.tools@gmail.com>
Date: Fri, 19 May 2017 12:19:42 -0700
Subject: [PATCH 5/8] 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, wcsnlen-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, __wcsnlen_avx2 and __wcsnlen_sse2.
* 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-sse2.S: Likewise.
* sysdeps/x86_64/multiarch/wcsnlen.c: Likewise.
---
sysdeps/x86_64/multiarch/Makefile | 4 +-
sysdeps/x86_64/multiarch/ifunc-impl-list.c | 28 ++
sysdeps/x86_64/multiarch/strlen-avx2.S | 394 +++++++++++++++++++++++++++++
sysdeps/x86_64/multiarch/strlen-sse2.S | 32 +++
sysdeps/x86_64/multiarch/strlen.c | 30 +++
sysdeps/x86_64/multiarch/strnlen-avx2.S | 4 +
sysdeps/x86_64/multiarch/strnlen-sse2.S | 36 +++
sysdeps/x86_64/multiarch/strnlen.c | 33 +++
sysdeps/x86_64/multiarch/wcslen-avx2.S | 4 +
sysdeps/x86_64/multiarch/wcslen-sse2.S | 26 ++
sysdeps/x86_64/multiarch/wcslen.c | 31 +++
sysdeps/x86_64/multiarch/wcsnlen-avx2.S | 5 +
sysdeps/x86_64/multiarch/wcsnlen-sse2.S | 26 ++
sysdeps/x86_64/multiarch/wcsnlen.c | 31 +++
14 files changed, 683 insertions(+), 1 deletion(-)
create mode 100644 sysdeps/x86_64/multiarch/strlen-avx2.S
create mode 100644 sysdeps/x86_64/multiarch/strlen-sse2.S
create mode 100644 sysdeps/x86_64/multiarch/strlen.c
create mode 100644 sysdeps/x86_64/multiarch/strnlen-avx2.S
create mode 100644 sysdeps/x86_64/multiarch/strnlen-sse2.S
create mode 100644 sysdeps/x86_64/multiarch/strnlen.c
create mode 100644 sysdeps/x86_64/multiarch/wcslen-avx2.S
create mode 100644 sysdeps/x86_64/multiarch/wcslen-sse2.S
create mode 100644 sysdeps/x86_64/multiarch/wcslen.c
create mode 100644 sysdeps/x86_64/multiarch/wcsnlen-avx2.S
create mode 100644 sysdeps/x86_64/multiarch/wcsnlen-sse2.S
create mode 100644 sysdeps/x86_64/multiarch/wcsnlen.c
diff --git a/sysdeps/x86_64/multiarch/Makefile b/sysdeps/x86_64/multiarch/Makefile
index eb42b19..915b44f 100644
--- a/sysdeps/x86_64/multiarch/Makefile
+++ b/sysdeps/x86_64/multiarch/Makefile
@@ -13,6 +13,7 @@ sysdep_routines += strncat-c stpncpy-c strncpy-c strcmp-ssse3 \
memcpy-ssse3-back \
memmove-ssse3-back \
memmove-avx512-no-vzeroupper strcasecmp_l-ssse3 \
+ strlen-sse2 strnlen-sse2 strlen-avx2 strnlen-avx2 \
strncase_l-ssse3 strcat-ssse3 strncat-ssse3\
strcpy-ssse3 strncpy-ssse3 stpcpy-ssse3 stpncpy-ssse3 \
strcpy-sse2-unaligned strncpy-sse2-unaligned \
@@ -35,7 +36,8 @@ ifeq ($(subdir),wcsmbs)
sysdep_routines += wmemcmp-sse4 wmemcmp-ssse3 wmemcmp-c \
wmemchr-sse2 wmemchr-avx2 \
wmemcmp-avx2-movbe \
- wcscpy-ssse3 wcscpy-c
+ wcscpy-ssse3 wcscpy-c \
+ wcslen-sse2 wcsnlen-sse2 wcslen-avx2 wcsnlen-avx2
endif
ifeq ($(subdir),debug)
diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
index f60535b..f139efc 100644
--- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
@@ -166,6 +166,20 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
__rawmemchr_avx2)
IFUNC_IMPL_ADD (array, i, rawmemchr, 1, __rawmemchr_sse2))
+ /* Support sysdeps/x86_64/multiarch/strlen.S. */
+ IFUNC_IMPL (i, name, strlen,
+ IFUNC_IMPL_ADD (array, i, strlen,
+ HAS_ARCH_FEATURE (AVX2_Usable),
+ __strlen_avx2)
+ IFUNC_IMPL_ADD (array, i, strlen, 1, __strlen_sse2))
+
+ /* Support sysdeps/x86_64/multiarch/strnlen.S. */
+ IFUNC_IMPL (i, name, strnlen,
+ IFUNC_IMPL_ADD (array, i, strnlen,
+ HAS_ARCH_FEATURE (AVX2_Usable),
+ __strnlen_avx2)
+ IFUNC_IMPL_ADD (array, i, strnlen, 1, __strnlen_sse2))
+
/* Support sysdeps/x86_64/multiarch/stpncpy.S. */
IFUNC_IMPL (i, name, stpncpy,
IFUNC_IMPL_ADD (array, i, stpncpy, HAS_CPU_FEATURE (SSSE3),
@@ -310,6 +324,20 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
__wcscpy_ssse3)
IFUNC_IMPL_ADD (array, i, wcscpy, 1, __wcscpy_sse2))
+ /* Support sysdeps/x86_64/multiarch/wcslen.S. */
+ IFUNC_IMPL (i, name, wcslen,
+ IFUNC_IMPL_ADD (array, i, wcslen,
+ HAS_ARCH_FEATURE (AVX2_Usable),
+ __wcslen_avx2)
+ IFUNC_IMPL_ADD (array, i, wcslen, 1, __wcslen_sse2))
+
+ /* Support sysdeps/x86_64/multiarch/wcsnlen.S. */
+ IFUNC_IMPL (i, name, wcsnlen,
+ IFUNC_IMPL_ADD (array, i, wcsnlen,
+ HAS_ARCH_FEATURE (AVX2_Usable),
+ __wcsnlen_avx2)
+ IFUNC_IMPL_ADD (array, i, wcsnlen, 1, __wcsnlen_sse2))
+
/* Support sysdeps/x86_64/multiarch/wmemchr.S. */
IFUNC_IMPL (i, name, wmemchr,
IFUNC_IMPL_ADD (array, i, wmemchr,
diff --git a/sysdeps/x86_64/multiarch/strlen-avx2.S b/sysdeps/x86_64/multiarch/strlen-avx2.S
new file mode 100644
index 0000000..1dc823a
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strlen-avx2.S
@@ -0,0 +1,394 @@
+/* strlen/strnlen/wcslen/wcsnlen optimized with AVX2.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#if IS_IN (libc)
+
+# include <sysdep.h>
+
+# ifndef STRLEN
+# define STRLEN __strlen_avx2
+# endif
+
+# ifdef USE_AS_WCSLEN
+# define VPCMPEQ vpcmpeqd
+# define VPMINU vpminud
+# else
+# define VPCMPEQ vpcmpeqb
+# define VPMINU vpminub
+# endif
+
+# ifndef VZEROUPPER
+# define VZEROUPPER vzeroupper
+# endif
+
+# define VEC_SIZE 32
+
+ .section .text.avx,"ax",@progbits
+ENTRY (STRLEN)
+# ifdef USE_AS_STRNLEN
+ /* Check for zero length. */
+ testq %rsi, %rsi
+ jz L(zero)
+# ifdef USE_AS_WCSLEN
+ shl $2, %rsi
+# endif
+ movq %rsi, %r8
+# endif
+ movl %edi, %ecx
+ movq %rdi, %rdx
+ vpxor %xmm0, %xmm0, %xmm0
+
+ /* Check if we may cross page boundary with one vector load. */
+ andl $(2 * VEC_SIZE - 1), %ecx
+ cmpl $VEC_SIZE, %ecx
+ ja L(cros_page_boundary)
+
+ /* Check the first VEC_SIZE bytes. */
+ VPCMPEQ (%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+
+# ifdef USE_AS_STRNLEN
+ jnz L(first_vec_x0_check)
+ /* Adjust length and check the end of data. */
+ subq $VEC_SIZE, %rsi
+ jbe L(max)
+# else
+ jnz L(first_vec_x0)
+# endif
+
+ /* Align data for aligned loads in the loop. */
+ addq $VEC_SIZE, %rdi
+ andl $(VEC_SIZE - 1), %ecx
+ andq $-VEC_SIZE, %rdi
+
+# ifdef USE_AS_STRNLEN
+ /* Adjust length. */
+ addq %rcx, %rsi
+
+ subq $(VEC_SIZE * 4), %rsi
+ jbe L(last_4x_vec_or_less)
+# endif
+ jmp L(more_4x_vec)
+
+ .p2align 4
+L(cros_page_boundary):
+ andl $(VEC_SIZE - 1), %ecx
+ andq $-VEC_SIZE, %rdi
+ VPCMPEQ (%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ /* Remove the leading bytes. */
+ sarl %cl, %eax
+ testl %eax, %eax
+ jz L(aligned_more)
+ tzcntl %eax, %eax
+# ifdef USE_AS_STRNLEN
+ /* Check the end of data. */
+ cmpq %rax, %rsi
+ jbe L(max)
+# endif
+ addq %rdi, %rax
+ addq %rcx, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(aligned_more):
+# ifdef USE_AS_STRNLEN
+ /* "rcx" is less than VEC_SIZE. Calculate "rdx + rcx - VEC_SIZE"
+ with "rdx - (VEC_SIZE - rcx)" instead of "(rdx + rcx) - VEC_SIZE"
+ to void possible addition overflow. */
+ negq %rcx
+ addq $VEC_SIZE, %rcx
+
+ /* Check the end of data. */
+ subq %rcx, %rsi
+ jbe L(max)
+# endif
+
+ addq $VEC_SIZE, %rdi
+
+# ifdef USE_AS_STRNLEN
+ subq $(VEC_SIZE * 4), %rsi
+ jbe L(last_4x_vec_or_less)
+# endif
+
+L(more_4x_vec):
+ /* Check the first 4 * VEC_SIZE. Only one VEC_SIZE at a time
+ since data is only aligned to VEC_SIZE. */
+ VPCMPEQ (%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x0)
+
+ VPCMPEQ VEC_SIZE(%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x1)
+
+ VPCMPEQ (VEC_SIZE * 2)(%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x2)
+
+ VPCMPEQ (VEC_SIZE * 3)(%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x3)
+
+ addq $(VEC_SIZE * 4), %rdi
+
+# ifdef USE_AS_STRNLEN
+ subq $(VEC_SIZE * 4), %rsi
+ jbe L(last_4x_vec_or_less)
+# endif
+
+ /* Align data to 4 * VEC_SIZE. */
+ movq %rdi, %rcx
+ andl $(4 * VEC_SIZE - 1), %ecx
+ andq $-(4 * VEC_SIZE), %rdi
+
+# ifdef USE_AS_STRNLEN
+ /* Adjust length. */
+ addq %rcx, %rsi
+# endif
+
+ .p2align 4
+L(loop_4x_vec):
+ /* Compare 4 * VEC at a time forward. */
+ vmovdqa (%rdi), %ymm1
+ vmovdqa VEC_SIZE(%rdi), %ymm2
+ vmovdqa (VEC_SIZE * 2)(%rdi), %ymm3
+ vmovdqa (VEC_SIZE * 3)(%rdi), %ymm4
+ VPMINU %ymm1, %ymm2, %ymm5
+ VPMINU %ymm3, %ymm4, %ymm6
+ VPMINU %ymm5, %ymm6, %ymm5
+
+ VPCMPEQ %ymm5, %ymm0, %ymm5
+ vpmovmskb %ymm5, %eax
+ testl %eax, %eax
+ jnz L(4x_vec_end)
+
+ addq $(VEC_SIZE * 4), %rdi
+
+# ifndef USE_AS_STRNLEN
+ jmp L(loop_4x_vec)
+# else
+ subq $(VEC_SIZE * 4), %rsi
+ ja L(loop_4x_vec)
+
+L(last_4x_vec_or_less):
+ /* Less than 4 * VEC and aligned to VEC_SIZE. */
+ addl $(VEC_SIZE * 2), %esi
+ jle L(last_2x_vec)
+
+ VPCMPEQ (%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x0)
+
+ VPCMPEQ VEC_SIZE(%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x1)
+
+ VPCMPEQ (VEC_SIZE * 2)(%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+
+ jnz L(first_vec_x2_check)
+ subl $VEC_SIZE, %esi
+ jle L(max)
+
+ VPCMPEQ (VEC_SIZE * 3)(%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+
+ jnz L(first_vec_x3_check)
+ movq %r8, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(last_2x_vec):
+ addl $(VEC_SIZE * 2), %esi
+ VPCMPEQ (%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+
+ jnz L(first_vec_x0_check)
+ subl $VEC_SIZE, %esi
+ jle L(max)
+
+ VPCMPEQ VEC_SIZE(%rdi), %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x1_check)
+ movq %r8, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(first_vec_x0_check):
+ tzcntl %eax, %eax
+ /* Check the end of data. */
+ cmpq %rax, %rsi
+ jbe L(max)
+ addq %rdi, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(first_vec_x1_check):
+ tzcntl %eax, %eax
+ /* Check the end of data. */
+ cmpq %rax, %rsi
+ jbe L(max)
+ addq $VEC_SIZE, %rax
+ addq %rdi, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(first_vec_x2_check):
+ tzcntl %eax, %eax
+ /* Check the end of data. */
+ cmpq %rax, %rsi
+ jbe L(max)
+ addq $(VEC_SIZE * 2), %rax
+ addq %rdi, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(first_vec_x3_check):
+ tzcntl %eax, %eax
+ /* Check the end of data. */
+ cmpq %rax, %rsi
+ jbe L(max)
+ addq $(VEC_SIZE * 3), %rax
+ addq %rdi, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(max):
+ movq %r8, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(zero):
+ xorl %eax, %eax
+ ret
+# endif
+
+ .p2align 4
+L(first_vec_x0):
+ tzcntl %eax, %eax
+ addq %rdi, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(first_vec_x1):
+ tzcntl %eax, %eax
+ addq $VEC_SIZE, %rax
+ addq %rdi, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(first_vec_x2):
+ tzcntl %eax, %eax
+ addq $(VEC_SIZE * 2), %rax
+ addq %rdi, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+ .p2align 4
+L(4x_vec_end):
+ VPCMPEQ %ymm1, %ymm0, %ymm1
+ vpmovmskb %ymm1, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x0)
+ VPCMPEQ %ymm2, %ymm0, %ymm2
+ vpmovmskb %ymm2, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x1)
+ VPCMPEQ %ymm3, %ymm0, %ymm3
+ vpmovmskb %ymm3, %eax
+ testl %eax, %eax
+ jnz L(first_vec_x2)
+ VPCMPEQ %ymm4, %ymm0, %ymm4
+ vpmovmskb %ymm4, %eax
+ testl %eax, %eax
+L(first_vec_x3):
+ tzcntl %eax, %eax
+ addq $(VEC_SIZE * 3), %rax
+ addq %rdi, %rax
+ subq %rdx, %rax
+# ifdef USE_AS_WCSLEN
+ shrq $2, %rax
+# endif
+ VZEROUPPER
+ ret
+
+END (STRLEN)
+#endif
diff --git a/sysdeps/x86_64/multiarch/strlen-sse2.S b/sysdeps/x86_64/multiarch/strlen-sse2.S
new file mode 100644
index 0000000..d0c2991
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strlen-sse2.S
@@ -0,0 +1,32 @@
+/* strlen optimized with SSE2.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#if IS_IN (libc)
+# define strlen __strlen_sse2
+
+# ifdef SHARED
+# undef libc_hidden_builtin_def
+/* It doesn't make sense to send libc-internal strlen calls through a PLT.
+ The speedup we get from using AVX2 instructions is likely eaten away
+ by the indirect call in the PLT. */
+# define libc_hidden_builtin_def(name) \
+ .globl __GI_strlen; __GI_strlen = __strlen_sse2
+# endif
+#endif
+
+#include "../strlen.S"
diff --git a/sysdeps/x86_64/multiarch/strlen.c b/sysdeps/x86_64/multiarch/strlen.c
new file mode 100644
index 0000000..8384035
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strlen.c
@@ -0,0 +1,30 @@
+/* Multiple versions of strlen.
+ All versions must be listed in ifunc-impl-list.c.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+/* Define multiple versions only for the definition in libc. */
+#if IS_IN (libc)
+# define strlen __redirect_strlen
+# include <string.h>
+# undef strlen
+
+# define SYMBOL_NAME strlen
+# include "ifunc-sse2-avx2.h"
+
+libc_ifunc_redirected (__redirect_strlen, strlen, IFUNC_SELECTOR ());
+#endif
diff --git a/sysdeps/x86_64/multiarch/strnlen-avx2.S b/sysdeps/x86_64/multiarch/strnlen-avx2.S
new file mode 100644
index 0000000..c4062b2
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strnlen-avx2.S
@@ -0,0 +1,4 @@
+#define STRLEN __strnlen_avx2
+#define USE_AS_STRNLEN 1
+
+#include "strlen-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/strnlen-sse2.S b/sysdeps/x86_64/multiarch/strnlen-sse2.S
new file mode 100644
index 0000000..7db8821
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strnlen-sse2.S
@@ -0,0 +1,36 @@
+/* strnlen optimized with SSE2.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#if IS_IN (libc)
+# define __strnlen __strnlen_sse2
+
+# ifdef SHARED
+/* It doesn't make sense to send libc-internal strnlen calls through a PLT.
+ The speedup we get from using AVX2 instructions is likely eaten away
+ by the indirect call in the PLT. */
+# undef libc_hidden_def
+# define libc_hidden_def(name) \
+ .globl __GI_strnlen; __GI_strnlen = __strnlen_sse2; \
+ .globl __GI___strnlen; __GI___strnlen = __strnlen_sse2
+# endif
+
+# undef weak_alias
+# define weak_alias(__strnlen, strnlen)
+#endif
+
+#include "../strnlen.S"
diff --git a/sysdeps/x86_64/multiarch/strnlen.c b/sysdeps/x86_64/multiarch/strnlen.c
new file mode 100644
index 0000000..bb09b83
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/strnlen.c
@@ -0,0 +1,33 @@
+/* Multiple versions of strnlen.
+ All versions must be listed in ifunc-impl-list.c.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+/* Define multiple versions only for the definition in libc. */
+#if IS_IN (libc)
+# define strnlen __redirect_strnlen
+# define __strnlen __redirect___strnlen
+# include <string.h>
+# undef __strnlen
+# undef strnlen
+
+# define SYMBOL_NAME strnlen
+# include "ifunc-sse2-avx2.h"
+
+libc_ifunc_redirected (__redirect_strnlen, __strnlen, IFUNC_SELECTOR ());
+weak_alias (__strnlen, strnlen);
+#endif
diff --git a/sysdeps/x86_64/multiarch/wcslen-avx2.S b/sysdeps/x86_64/multiarch/wcslen-avx2.S
new file mode 100644
index 0000000..c9224f1
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcslen-avx2.S
@@ -0,0 +1,4 @@
+#define STRLEN __wcslen_avx2
+#define USE_AS_WCSLEN 1
+
+#include "strlen-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcslen-sse2.S b/sysdeps/x86_64/multiarch/wcslen-sse2.S
new file mode 100644
index 0000000..e7b24ea
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcslen-sse2.S
@@ -0,0 +1,26 @@
+/* wcslen optimized with SSE2.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#if IS_IN (libc)
+# define __wcslen __wcslen_sse2
+
+# undef weak_alias
+# define weak_alias(__wcslen, wcslen)
+#endif
+
+#include "../wcslen.S"
diff --git a/sysdeps/x86_64/multiarch/wcslen.c b/sysdeps/x86_64/multiarch/wcslen.c
new file mode 100644
index 0000000..3f8ad87
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcslen.c
@@ -0,0 +1,31 @@
+/* Multiple versions of wcslen.
+ All versions must be listed in ifunc-impl-list.c.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+/* Define multiple versions only for the definition in libc. */
+#if IS_IN (libc)
+# define __wcslen __redirect_wcslen
+# include <wchar.h>
+# undef __wcslen
+
+# define SYMBOL_NAME wcslen
+# include "ifunc-sse2-avx2.h"
+
+libc_ifunc_redirected (__redirect_wcslen, __wcslen, IFUNC_SELECTOR ());
+weak_alias (__wcslen, wcslen);
+#endif
diff --git a/sysdeps/x86_64/multiarch/wcsnlen-avx2.S b/sysdeps/x86_64/multiarch/wcsnlen-avx2.S
new file mode 100644
index 0000000..fac8354
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcsnlen-avx2.S
@@ -0,0 +1,5 @@
+#define STRLEN __wcsnlen_avx2
+#define USE_AS_WCSLEN 1
+#define USE_AS_STRNLEN 1
+
+#include "strlen-avx2.S"
diff --git a/sysdeps/x86_64/multiarch/wcsnlen-sse2.S b/sysdeps/x86_64/multiarch/wcsnlen-sse2.S
new file mode 100644
index 0000000..846466b
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcsnlen-sse2.S
@@ -0,0 +1,26 @@
+/* wcsnlen optimized with SSE2.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#if IS_IN (libc)
+# define __wcsnlen __wcsnlen_sse2
+
+# undef weak_alias
+# define weak_alias(__wcsnlen, wcsnlen)
+#endif
+
+#include "../wcsnlen.S"
diff --git a/sysdeps/x86_64/multiarch/wcsnlen.c b/sysdeps/x86_64/multiarch/wcsnlen.c
new file mode 100644
index 0000000..35541b1
--- /dev/null
+++ b/sysdeps/x86_64/multiarch/wcsnlen.c
@@ -0,0 +1,31 @@
+/* Multiple versions of wcsnlen.
+ All versions must be listed in ifunc-impl-list.c.
+ Copyright (C) 2017 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+/* Define multiple versions only for the definition in libc. */
+#if IS_IN (libc)
+# define __wcsnlen __redirect_wcsnlen
+# include <wchar.h>
+# undef __wcsnlen
+
+# define SYMBOL_NAME wcsnlen
+# include "ifunc-sse2-avx2.h"
+
+libc_ifunc_redirected (__redirect_wcsnlen, __wcsnlen, IFUNC_SELECTOR ());
+weak_alias (__wcsnlen, wcsnlen);
+#endif
--
2.9.4
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH] x86-64: Optimize strlen/strnlen/wcslen/wcsnlen with AVX2
2017-06-02 19:48 ` H.J. Lu
@ 2017-06-09 11:32 ` H.J. Lu
0 siblings, 0 replies; 3+ messages in thread
From: H.J. Lu @ 2017-06-09 11:32 UTC (permalink / raw)
To: GNU C Library
On Fri, Jun 2, 2017 at 12:48 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
> On Thu, Jun 1, 2017 at 11:12 AM, H.J. Lu <hongjiu.lu@intel.com> wrote:
>> 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.
>>
>> Any comments?
>>
>> H.J.
>> ---
>> * sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Add
>> strlen-avx2, strnlen-avx2 and 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, __wcsnlen_avx2 and __wcsnlen_sse2.
>> * sysdeps/x86_64/multiarch/strlen-avx2.S: New file.
>> * sysdeps/x86_64/multiarch/strlen.S: Likewise.
>> * sysdeps/x86_64/multiarch/strnlen-avx2.S: Likewise.
>> * sysdeps/x86_64/multiarch/strnlen.S: Likewise.
>> * sysdeps/x86_64/multiarch/wcslen-avx2.S: Likewise.
>> * sysdeps/x86_64/multiarch/wcslen.S: Likewise.
>> * sysdeps/x86_64/multiarch/wcsnlen-avx2.S: Likewise.
>> * sysdeps/x86_64/multiarch/wcsnlen.S: Likewise.
>
> Updated patch with IFUNC selector in C.
>
>
I will check it in.
--
H.J.
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2017-06-09 11:32 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-06-01 18:13 [PATCH] x86-64: Optimize strlen/strnlen/wcslen/wcsnlen with AVX2 H.J. Lu
2017-06-02 19:48 ` H.J. Lu
2017-06-09 11:32 ` H.J. Lu
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).