|
 |
00db10 |
commit 83d776f979342f923b5c3d2a5b43afab841c6086
|
|
|
00db10 |
Author: Andrew Senkevich <andrew.senkevich@intel.com>
|
|
|
00db10 |
Date: Sat Dec 19 02:47:28 2015 +0300
|
|
|
00db10 |
|
|
|
00db10 |
Added memset optimized with AVX512 for KNL hardware.
|
|
|
00db10 |
|
|
|
00db10 |
It shows improvement up to 28% over AVX2 memset (performance results
|
|
|
00db10 |
attached at <https://sourceware.org/ml/libc-alpha/2015-12/msg00052.html>).
|
|
|
00db10 |
|
|
|
00db10 |
* sysdeps/x86_64/multiarch/memset-avx512-no-vzeroupper.S: New file.
|
|
|
00db10 |
* sysdeps/x86_64/multiarch/Makefile (sysdep_routines): Added new file.
|
|
|
00db10 |
* sysdeps/x86_64/multiarch/ifunc-impl-list.c: Added new tests.
|
|
|
00db10 |
* sysdeps/x86_64/multiarch/memset.S: Added new IFUNC branch.
|
|
|
00db10 |
* sysdeps/x86_64/multiarch/memset_chk.S: Likewise.
|
|
|
00db10 |
* sysdeps/x86/cpu-features.h (bit_Prefer_No_VZEROUPPER,
|
|
|
00db10 |
index_Prefer_No_VZEROUPPER): New.
|
|
|
00db10 |
* sysdeps/x86/cpu-features.c (init_cpu_features): Set the
|
|
|
00db10 |
Prefer_No_VZEROUPPER for Knights Landing.
|
|
|
00db10 |
|
|
|
00db10 |
Index: glibc-2.17-c758a686/sysdeps/x86/cpu-features.c
|
|
|
00db10 |
===================================================================
|
|
|
00db10 |
--- glibc-2.17-c758a686.orig/sysdeps/x86/cpu-features.c
|
|
|
00db10 |
+++ glibc-2.17-c758a686/sysdeps/x86/cpu-features.c
|
|
|
00db10 |
@@ -76,6 +76,8 @@ init_cpu_features (struct cpu_features *
|
|
|
00db10 |
|
|
|
00db10 |
case 0x57:
|
|
|
00db10 |
/* Knights Landing. Enable Silvermont optimizations. */
|
|
|
00db10 |
+ cpu_features->feature[index_Prefer_No_VZEROUPPER]
|
|
|
00db10 |
+ |= bit_Prefer_No_VZEROUPPER;
|
|
|
00db10 |
|
|
|
00db10 |
case 0x37:
|
|
|
00db10 |
case 0x4a:
|
|
|
00db10 |
Index: glibc-2.17-c758a686/sysdeps/x86/cpu-features.h
|
|
|
00db10 |
===================================================================
|
|
|
00db10 |
--- glibc-2.17-c758a686.orig/sysdeps/x86/cpu-features.h
|
|
|
00db10 |
+++ glibc-2.17-c758a686/sysdeps/x86/cpu-features.h
|
|
|
00db10 |
@@ -33,6 +33,7 @@
|
|
|
00db10 |
#define bit_AVX512F_Usable (1 << 12)
|
|
|
00db10 |
#define bit_AVX512DQ_Usable (1 << 13)
|
|
|
00db10 |
#define bit_Prefer_MAP_32BIT_EXEC (1 << 16)
|
|
|
00db10 |
+#define bit_Prefer_No_VZEROUPPER (1 << 17)
|
|
|
00db10 |
|
|
|
00db10 |
/* CPUID Feature flags. */
|
|
|
00db10 |
|
|
|
00db10 |
@@ -93,6 +94,7 @@
|
|
|
00db10 |
# define index_AVX512F_Usable FEATURE_INDEX_1*FEATURE_SIZE
|
|
|
00db10 |
# define index_AVX512DQ_Usable FEATURE_INDEX_1*FEATURE_SIZE
|
|
|
00db10 |
# define index_Prefer_MAP_32BIT_EXEC FEATURE_INDEX_1*FEATURE_SIZE
|
|
|
00db10 |
+# define index_Prefer_No_VZEROUPPER FEATURE_INDEX_1*FEATURE_SIZE
|
|
|
00db10 |
|
|
|
00db10 |
# if defined (_LIBC) && !IS_IN (nonlib)
|
|
|
00db10 |
# ifdef __x86_64__
|
|
|
00db10 |
@@ -270,6 +272,7 @@ extern const struct cpu_features *__get_
|
|
|
00db10 |
# define index_AVX512F_Usable FEATURE_INDEX_1
|
|
|
00db10 |
# define index_AVX512DQ_Usable FEATURE_INDEX_1
|
|
|
00db10 |
# define index_Prefer_MAP_32BIT_EXEC FEATURE_INDEX_1
|
|
|
00db10 |
+# define index_Prefer_No_VZEROUPPER FEATURE_INDEX_1
|
|
|
00db10 |
|
|
|
00db10 |
#endif /* !__ASSEMBLER__ */
|
|
|
00db10 |
|
|
|
00db10 |
Index: glibc-2.17-c758a686/sysdeps/x86_64/multiarch/Makefile
|
|
|
00db10 |
===================================================================
|
|
|
00db10 |
--- glibc-2.17-c758a686.orig/sysdeps/x86_64/multiarch/Makefile
|
|
|
00db10 |
+++ glibc-2.17-c758a686/sysdeps/x86_64/multiarch/Makefile
|
|
|
00db10 |
@@ -16,7 +16,8 @@ sysdep_routines += strncat-c stpncpy-c s
|
|
|
00db10 |
strcat-sse2-unaligned strncat-sse2-unaligned \
|
|
|
00db10 |
strcat-ssse3 strncat-ssse3 strlen-sse2-pminub \
|
|
|
00db10 |
strnlen-sse2-no-bsf strrchr-sse2-no-bsf strchr-sse2-no-bsf \
|
|
|
00db10 |
- memcmp-ssse3 strstr-sse2-unaligned
|
|
|
00db10 |
+ memcmp-ssse3 strstr-sse2-unaligned \
|
|
|
00db10 |
+ memset-avx512-no-vzeroupper
|
|
|
00db10 |
ifeq (yes,$(config-cflags-sse4))
|
|
|
00db10 |
sysdep_routines += strcspn-c strpbrk-c strspn-c strstr-c strcasestr-c varshift
|
|
|
00db10 |
CFLAGS-varshift.c += -msse4
|
|
|
00db10 |
Index: glibc-2.17-c758a686/sysdeps/x86_64/multiarch/ifunc-impl-list.c
|
|
|
00db10 |
===================================================================
|
|
|
00db10 |
--- glibc-2.17-c758a686.orig/sysdeps/x86_64/multiarch/ifunc-impl-list.c
|
|
|
00db10 |
+++ glibc-2.17-c758a686/sysdeps/x86_64/multiarch/ifunc-impl-list.c
|
|
|
00db10 |
@@ -20,6 +20,7 @@
|
|
|
00db10 |
#include <string.h>
|
|
|
00db10 |
#include <wchar.h>
|
|
|
00db10 |
#include <ifunc-impl-list.h>
|
|
|
00db10 |
+#include <sysdep.h>
|
|
|
00db10 |
#include "init-arch.h"
|
|
|
00db10 |
|
|
|
00db10 |
/* Maximum number of IFUNC implementations. */
|
|
|
00db10 |
@@ -66,12 +67,24 @@ __libc_ifunc_impl_list (const char *name
|
|
|
00db10 |
IFUNC_IMPL (i, name, __memset_chk,
|
|
|
00db10 |
IFUNC_IMPL_ADD (array, i, __memset_chk, 1, __memset_chk_sse2)
|
|
|
00db10 |
IFUNC_IMPL_ADD (array, i, __memset_chk, 1,
|
|
|
00db10 |
- __memset_chk_x86_64))
|
|
|
00db10 |
+ __memset_chk_x86_64)
|
|
|
00db10 |
+#ifdef HAVE_AVX512_ASM_SUPPORT
|
|
|
00db10 |
+ IFUNC_IMPL_ADD (array, i, __memset_chk,
|
|
|
00db10 |
+ HAS_ARCH_FEATURE (AVX512F_Usable),
|
|
|
00db10 |
+ __memset_chk_avx512_no_vzeroupper)
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ )
|
|
|
00db10 |
|
|
|
00db10 |
/* Support sysdeps/x86_64/multiarch/memset.S. */
|
|
|
00db10 |
IFUNC_IMPL (i, name, memset,
|
|
|
00db10 |
IFUNC_IMPL_ADD (array, i, memset, 1, __memset_sse2)
|
|
|
00db10 |
- IFUNC_IMPL_ADD (array, i, memset, 1, __memset_x86_64))
|
|
|
00db10 |
+ IFUNC_IMPL_ADD (array, i, memset, 1, __memset_x86_64)
|
|
|
00db10 |
+#ifdef HAVE_AVX512_ASM_SUPPORT
|
|
|
00db10 |
+ IFUNC_IMPL_ADD (array, i, memset,
|
|
|
00db10 |
+ HAS_ARCH_FEATURE (AVX512F_Usable),
|
|
|
00db10 |
+ __memset_avx512_no_vzeroupper)
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+ )
|
|
|
00db10 |
|
|
|
00db10 |
/* Support sysdeps/x86_64/multiarch/rawmemchr.S. */
|
|
|
00db10 |
IFUNC_IMPL (i, name, rawmemchr,
|
|
|
00db10 |
Index: glibc-2.17-c758a686/sysdeps/x86_64/multiarch/memset-avx512-no-vzeroupper.S
|
|
|
00db10 |
===================================================================
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ glibc-2.17-c758a686/sysdeps/x86_64/multiarch/memset-avx512-no-vzeroupper.S
|
|
|
00db10 |
@@ -0,0 +1,194 @@
|
|
|
00db10 |
+/* memset optimized with AVX512 for KNL hardware.
|
|
|
00db10 |
+ Copyright (C) 2015 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include <sysdep.h>
|
|
|
00db10 |
+
|
|
|
00db10 |
+#if defined HAVE_AVX512_ASM_SUPPORT && IS_IN (libc)
|
|
|
00db10 |
+
|
|
|
00db10 |
+#include "asm-syntax.h"
|
|
|
00db10 |
+#ifndef MEMSET
|
|
|
00db10 |
+# define MEMSET __memset_avx512_no_vzeroupper
|
|
|
00db10 |
+# define MEMSET_CHK __memset_chk_avx512_no_vzeroupper
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+
|
|
|
00db10 |
+ .section .text,"ax",@progbits
|
|
|
00db10 |
+#if defined PIC
|
|
|
00db10 |
+ENTRY (MEMSET_CHK)
|
|
|
00db10 |
+ cmpq %rdx, %rcx
|
|
|
00db10 |
+ jb HIDDEN_JUMPTARGET (__chk_fail)
|
|
|
00db10 |
+END (MEMSET_CHK)
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
+
|
|
|
00db10 |
+ENTRY (MEMSET)
|
|
|
00db10 |
+ vpxor %xmm0, %xmm0, %xmm0
|
|
|
00db10 |
+ vmovd %esi, %xmm1
|
|
|
00db10 |
+ lea (%rdi, %rdx), %rsi
|
|
|
00db10 |
+ mov %rdi, %rax
|
|
|
00db10 |
+ vpshufb %xmm0, %xmm1, %xmm0
|
|
|
00db10 |
+ cmp $16, %rdx
|
|
|
00db10 |
+ jb L(less_16bytes)
|
|
|
00db10 |
+ cmp $512, %rdx
|
|
|
00db10 |
+ vbroadcastss %xmm0, %zmm2
|
|
|
00db10 |
+ ja L(512bytesormore)
|
|
|
00db10 |
+ cmp $256, %rdx
|
|
|
00db10 |
+ jb L(less_256bytes)
|
|
|
00db10 |
+ vmovups %zmm2, (%rdi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x40(%rdi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x80(%rdi)
|
|
|
00db10 |
+ vmovups %zmm2, 0xC0(%rdi)
|
|
|
00db10 |
+ vmovups %zmm2, -0x100(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, -0xC0(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, -0x80(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, -0x40(%rsi)
|
|
|
00db10 |
+ ret
|
|
|
00db10 |
+
|
|
|
00db10 |
+L(less_256bytes):
|
|
|
00db10 |
+ cmp $128, %dl
|
|
|
00db10 |
+ jb L(less_128bytes)
|
|
|
00db10 |
+ vmovups %zmm2, (%rdi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x40(%rdi)
|
|
|
00db10 |
+ vmovups %zmm2, -0x80(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, -0x40(%rsi)
|
|
|
00db10 |
+ ret
|
|
|
00db10 |
+
|
|
|
00db10 |
+L(less_128bytes):
|
|
|
00db10 |
+ cmp $64, %dl
|
|
|
00db10 |
+ jb L(less_64bytes)
|
|
|
00db10 |
+ vmovups %zmm2, (%rdi)
|
|
|
00db10 |
+ vmovups %zmm2, -0x40(%rsi)
|
|
|
00db10 |
+ ret
|
|
|
00db10 |
+
|
|
|
00db10 |
+L(less_64bytes):
|
|
|
00db10 |
+ cmp $32, %dl
|
|
|
00db10 |
+ jb L(less_32bytes)
|
|
|
00db10 |
+ vmovdqu %ymm2, (%rdi)
|
|
|
00db10 |
+ vmovdqu %ymm2, -0x20(%rsi)
|
|
|
00db10 |
+ ret
|
|
|
00db10 |
+
|
|
|
00db10 |
+L(less_32bytes):
|
|
|
00db10 |
+ vmovdqu %xmm0, (%rdi)
|
|
|
00db10 |
+ vmovdqu %xmm0, -0x10(%rsi)
|
|
|
00db10 |
+ ret
|
|
|
00db10 |
+
|
|
|
00db10 |
+L(less_16bytes):
|
|
|
00db10 |
+ cmp $8, %dl
|
|
|
00db10 |
+ jb L(less_8bytes)
|
|
|
00db10 |
+ vmovq %xmm0, (%rdi)
|
|
|
00db10 |
+ vmovq %xmm0, -0x08(%rsi)
|
|
|
00db10 |
+ ret
|
|
|
00db10 |
+
|
|
|
00db10 |
+L(less_8bytes):
|
|
|
00db10 |
+ vmovd %xmm0, %ecx
|
|
|
00db10 |
+ cmp $4, %dl
|
|
|
00db10 |
+ jb L(less_4bytes)
|
|
|
00db10 |
+ mov %ecx, (%rdi)
|
|
|
00db10 |
+ mov %ecx, -0x04(%rsi)
|
|
|
00db10 |
+ ret
|
|
|
00db10 |
+
|
|
|
00db10 |
+L(less_4bytes):
|
|
|
00db10 |
+ cmp $2, %dl
|
|
|
00db10 |
+ jb L(less_2bytes)
|
|
|
00db10 |
+ mov %cx, (%rdi)
|
|
|
00db10 |
+ mov %cx, -0x02(%rsi)
|
|
|
00db10 |
+ ret
|
|
|
00db10 |
+
|
|
|
00db10 |
+L(less_2bytes):
|
|
|
00db10 |
+ cmp $1, %dl
|
|
|
00db10 |
+ jb L(less_1bytes)
|
|
|
00db10 |
+ mov %cl, (%rdi)
|
|
|
00db10 |
+L(less_1bytes):
|
|
|
00db10 |
+ ret
|
|
|
00db10 |
+
|
|
|
00db10 |
+L(512bytesormore):
|
|
|
00db10 |
+ mov __x86_64_shared_cache_size_half(%rip), %rcx
|
|
|
00db10 |
+ cmp %rcx, %rdx
|
|
|
00db10 |
+ ja L(preloop_large)
|
|
|
00db10 |
+ cmp $1024, %rdx
|
|
|
00db10 |
+ ja L(1024bytesormore)
|
|
|
00db10 |
+
|
|
|
00db10 |
+ vmovups %zmm2, (%rdi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x40(%rdi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x80(%rdi)
|
|
|
00db10 |
+ vmovups %zmm2, 0xC0(%rdi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x100(%rdi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x140(%rdi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x180(%rdi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x1C0(%rdi)
|
|
|
00db10 |
+ vmovups %zmm2, -0x200(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, -0x1C0(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, -0x180(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, -0x140(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, -0x100(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, -0xC0(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, -0x80(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, -0x40(%rsi)
|
|
|
00db10 |
+ ret
|
|
|
00db10 |
+
|
|
|
00db10 |
+/* Align on 64 and loop with aligned stores. */
|
|
|
00db10 |
+L(1024bytesormore):
|
|
|
00db10 |
+ sub $0x100, %rsi
|
|
|
00db10 |
+ vmovups %zmm2, (%rax)
|
|
|
00db10 |
+ and $-0x40, %rdi
|
|
|
00db10 |
+ add $0x40, %rdi
|
|
|
00db10 |
+
|
|
|
00db10 |
+L(gobble_256bytes_loop):
|
|
|
00db10 |
+ vmovaps %zmm2, (%rdi)
|
|
|
00db10 |
+ vmovaps %zmm2, 0x40(%rdi)
|
|
|
00db10 |
+ vmovaps %zmm2, 0x80(%rdi)
|
|
|
00db10 |
+ vmovaps %zmm2, 0xC0(%rdi)
|
|
|
00db10 |
+ add $0x100, %rdi
|
|
|
00db10 |
+ cmp %rsi, %rdi
|
|
|
00db10 |
+ jb L(gobble_256bytes_loop)
|
|
|
00db10 |
+ vmovups %zmm2, (%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x40(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x80(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, 0xC0(%rsi)
|
|
|
00db10 |
+ ret
|
|
|
00db10 |
+
|
|
|
00db10 |
+/* Align on 128 and loop with non-temporal stores. */
|
|
|
00db10 |
+L(preloop_large):
|
|
|
00db10 |
+ and $-0x80, %rdi
|
|
|
00db10 |
+ add $0x80, %rdi
|
|
|
00db10 |
+ vmovups %zmm2, (%rax)
|
|
|
00db10 |
+ vmovups %zmm2, 0x40(%rax)
|
|
|
00db10 |
+ sub $0x200, %rsi
|
|
|
00db10 |
+
|
|
|
00db10 |
+L(gobble_512bytes_nt_loop):
|
|
|
00db10 |
+ vmovntdq %zmm2, (%rdi)
|
|
|
00db10 |
+ vmovntdq %zmm2, 0x40(%rdi)
|
|
|
00db10 |
+ vmovntdq %zmm2, 0x80(%rdi)
|
|
|
00db10 |
+ vmovntdq %zmm2, 0xC0(%rdi)
|
|
|
00db10 |
+ vmovntdq %zmm2, 0x100(%rdi)
|
|
|
00db10 |
+ vmovntdq %zmm2, 0x140(%rdi)
|
|
|
00db10 |
+ vmovntdq %zmm2, 0x180(%rdi)
|
|
|
00db10 |
+ vmovntdq %zmm2, 0x1C0(%rdi)
|
|
|
00db10 |
+ add $0x200, %rdi
|
|
|
00db10 |
+ cmp %rsi, %rdi
|
|
|
00db10 |
+ jb L(gobble_512bytes_nt_loop)
|
|
|
00db10 |
+ sfence
|
|
|
00db10 |
+ vmovups %zmm2, (%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x40(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x80(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, 0xC0(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x100(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x140(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x180(%rsi)
|
|
|
00db10 |
+ vmovups %zmm2, 0x1C0(%rsi)
|
|
|
00db10 |
+ ret
|
|
|
00db10 |
+END (MEMSET)
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
Index: glibc-2.17-c758a686/sysdeps/x86_64/multiarch/memset.S
|
|
|
00db10 |
===================================================================
|
|
|
00db10 |
--- glibc-2.17-c758a686.orig/sysdeps/x86_64/multiarch/memset.S
|
|
|
00db10 |
+++ glibc-2.17-c758a686/sysdeps/x86_64/multiarch/memset.S
|
|
|
00db10 |
@@ -29,6 +29,13 @@ ENTRY(memset)
|
|
|
00db10 |
HAS_ARCH_FEATURE (Prefer_SSE_for_memop)
|
|
|
00db10 |
jz 2f
|
|
|
00db10 |
leaq __memset_sse2(%rip), %rax
|
|
|
00db10 |
+#ifdef HAVE_AVX512_ASM_SUPPORT
|
|
|
00db10 |
+ HAS_ARCH_FEATURE (AVX512F_Usable)
|
|
|
00db10 |
+ jz 2f
|
|
|
00db10 |
+ HAS_ARCH_FEATURE (Prefer_No_VZEROUPPER)
|
|
|
00db10 |
+ jz 2f
|
|
|
00db10 |
+ leaq __memset_avx512_no_vzeroupper(%rip), %rax
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
2: ret
|
|
|
00db10 |
END(memset)
|
|
|
00db10 |
|
|
|
00db10 |
Index: glibc-2.17-c758a686/sysdeps/x86_64/multiarch/memset_chk.S
|
|
|
00db10 |
===================================================================
|
|
|
00db10 |
--- glibc-2.17-c758a686.orig/sysdeps/x86_64/multiarch/memset_chk.S
|
|
|
00db10 |
+++ glibc-2.17-c758a686/sysdeps/x86_64/multiarch/memset_chk.S
|
|
|
00db10 |
@@ -30,6 +30,13 @@ ENTRY(__memset_chk)
|
|
|
00db10 |
HAS_ARCH_FEATURE (Prefer_SSE_for_memop)
|
|
|
00db10 |
jz 2f
|
|
|
00db10 |
leaq __memset_chk_sse2(%rip), %rax
|
|
|
00db10 |
+#ifdef HAVE_AVX512_ASM_SUPPORT
|
|
|
00db10 |
+ HAS_ARCH_FEATURE (AVX512F_Usable)
|
|
|
00db10 |
+ jz 2f
|
|
|
00db10 |
+ HAS_ARCH_FEATURE (Prefer_No_VZEROUPPER)
|
|
|
00db10 |
+ jz 2f
|
|
|
00db10 |
+ leaq __memset_chk_avx512_no_vzeroupper(%rip), %rax
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
2: ret
|
|
|
00db10 |
END(__memset_chk)
|
|
|
00db10 |
|