|
|
1feee8 |
commit e570b865b53f33453d97160791a7d97e38bcc6e8
|
|
|
1feee8 |
Author: Aurelien Jarno <aurelien@aurel32.net>
|
|
|
1feee8 |
Date: Mon Oct 3 23:16:46 2022 +0200
|
|
|
1feee8 |
|
|
|
1feee8 |
x86-64: Require BMI1/BMI2 for AVX2 strrchr and wcsrchr implementations
|
|
|
1feee8 |
|
|
|
1feee8 |
The AVX2 strrchr and wcsrchr implementation uses the 'blsmsk'
|
|
|
1feee8 |
instruction which belongs to the BMI1 CPU feature and the 'shrx'
|
|
|
1feee8 |
instruction, which belongs to the BMI2 CPU feature.
|
|
|
1feee8 |
|
|
|
1feee8 |
Fixes: df7e295d18ff ("x86: Optimize {str|wcs}rchr-avx2")
|
|
|
1feee8 |
Partially resolves: BZ #29611
|
|
|
1feee8 |
|
|
|
1feee8 |
Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
|
|
|
1feee8 |
(cherry picked from commit 7e8283170c5d6805b609a040801d819e362a6292)
|
|
|
1feee8 |
|
|
|
1feee8 |
diff --git a/sysdeps/x86_64/multiarch/ifunc-avx2.h b/sysdeps/x86_64/multiarch/ifunc-avx2.h
|
|
|
1feee8 |
index 52bd00ea5cab6b22..877f007dd6e38fe8 100644
|
|
|
1feee8 |
--- a/sysdeps/x86_64/multiarch/ifunc-avx2.h
|
|
|
1feee8 |
+++ b/sysdeps/x86_64/multiarch/ifunc-avx2.h
|
|
|
1feee8 |
@@ -30,6 +30,7 @@ IFUNC_SELECTOR (void)
|
|
|
1feee8 |
const struct cpu_features* cpu_features = __get_cpu_features ();
|
|
|
1feee8 |
|
|
|
1feee8 |
if (CPU_FEATURE_USABLE_P (cpu_features, AVX2)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE_P (cpu_features, BMI1)
|
|
|
1feee8 |
&& CPU_FEATURE_USABLE_P (cpu_features, BMI2)
|
|
|
1feee8 |
&& CPU_FEATURE_USABLE_P (cpu_features, LZCNT)
|
|
|
1feee8 |
&& CPU_FEATURES_ARCH_P (cpu_features, AVX_Fast_Unaligned_Load))
|
|
|
1feee8 |
diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
|
|
|
1feee8 |
index d1fc1e75d6706413..84f9e73e2b7df816 100644
|
|
|
1feee8 |
--- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
|
|
|
1feee8 |
+++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
|
|
|
1feee8 |
@@ -498,15 +498,21 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
1feee8 |
/* Support sysdeps/x86_64/multiarch/strrchr.c. */
|
|
|
1feee8 |
IFUNC_IMPL (i, name, strrchr,
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strrchr,
|
|
|
1feee8 |
- CPU_FEATURE_USABLE (AVX2),
|
|
|
1feee8 |
+ (CPU_FEATURE_USABLE (AVX2)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI1)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)),
|
|
|
1feee8 |
__strrchr_avx2)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strrchr,
|
|
|
1feee8 |
(CPU_FEATURE_USABLE (AVX2)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI1)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)
|
|
|
1feee8 |
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
1feee8 |
__strrchr_avx2_rtm)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strrchr,
|
|
|
1feee8 |
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
1feee8 |
- && CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (AVX512BW)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI1)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)),
|
|
|
1feee8 |
__strrchr_evex)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strrchr, 1, __strrchr_sse2))
|
|
|
1feee8 |
|
|
|
1feee8 |
@@ -687,15 +693,20 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
1feee8 |
/* Support sysdeps/x86_64/multiarch/wcsrchr.c. */
|
|
|
1feee8 |
IFUNC_IMPL (i, name, wcsrchr,
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, wcsrchr,
|
|
|
1feee8 |
- CPU_FEATURE_USABLE (AVX2),
|
|
|
1feee8 |
+ (CPU_FEATURE_USABLE (AVX2)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI1)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)),
|
|
|
1feee8 |
__wcsrchr_avx2)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, wcsrchr,
|
|
|
1feee8 |
(CPU_FEATURE_USABLE (AVX2)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI1)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)
|
|
|
1feee8 |
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
1feee8 |
__wcsrchr_avx2_rtm)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, wcsrchr,
|
|
|
1feee8 |
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
1feee8 |
&& CPU_FEATURE_USABLE (AVX512BW)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI1)
|
|
|
1feee8 |
&& CPU_FEATURE_USABLE (BMI2)),
|
|
|
1feee8 |
__wcsrchr_evex)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, wcsrchr, 1, __wcsrchr_sse2))
|