|
 |
1feee8 |
commit b9cbb8dd48b545f3b36b5d411481dc0bd118ee94
|
|
|
1feee8 |
Author: Aurelien Jarno <aurelien@aurel32.net>
|
|
|
1feee8 |
Date: Mon Oct 3 23:16:46 2022 +0200
|
|
|
1feee8 |
|
|
|
1feee8 |
x86-64: Require BMI2 for AVX2 strncmp implementation
|
|
|
1feee8 |
|
|
|
1feee8 |
The AVX2 strncmp implementations uses the 'bzhi' instruction, which
|
|
|
1feee8 |
belongs to the BMI2 CPU feature.
|
|
|
1feee8 |
|
|
|
1feee8 |
NB: It also uses the 'tzcnt' BMI1 instruction, but it is executed as BSF
|
|
|
1feee8 |
as BSF if the CPU doesn't support TZCNT, and produces the same result
|
|
|
1feee8 |
for non-zero input.
|
|
|
1feee8 |
|
|
|
1feee8 |
Partially fixes: b77b06e0e296 ("x86: Optimize strcmp-avx2.S")
|
|
|
1feee8 |
Partially resolves: BZ #29611
|
|
|
1feee8 |
|
|
|
1feee8 |
Reviewed-by: Noah Goldstein <goldstein.w.n@gmail.com>
|
|
|
1feee8 |
(cherry picked from commit fc7de1d9b99ae1676bc626ddca422d7abee0eb48)
|
|
|
1feee8 |
|
|
|
1feee8 |
diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
|
|
|
1feee8 |
index 70931f15985334af..34d5f6efe5421014 100644
|
|
|
1feee8 |
--- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
|
|
|
1feee8 |
+++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
|
|
|
1feee8 |
@@ -1022,15 +1022,18 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
1feee8 |
/* Support sysdeps/x86_64/multiarch/strncmp.c. */
|
|
|
1feee8 |
IFUNC_IMPL (i, name, strncmp,
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strncmp,
|
|
|
1feee8 |
- CPU_FEATURE_USABLE (AVX2),
|
|
|
1feee8 |
+ (CPU_FEATURE_USABLE (AVX2)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)),
|
|
|
1feee8 |
__strncmp_avx2)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strncmp,
|
|
|
1feee8 |
(CPU_FEATURE_USABLE (AVX2)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)
|
|
|
1feee8 |
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
1feee8 |
__strncmp_avx2_rtm)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strncmp,
|
|
|
1feee8 |
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
1feee8 |
- && CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (AVX512BW)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)),
|
|
|
1feee8 |
__strncmp_evex)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strncmp, CPU_FEATURE_USABLE (SSE4_2),
|
|
|
1feee8 |
__strncmp_sse42)
|
|
|
1feee8 |
diff --git a/sysdeps/x86_64/multiarch/strncmp.c b/sysdeps/x86_64/multiarch/strncmp.c
|
|
|
1feee8 |
index f94a421784bfe923..7632d7b2dd4447aa 100644
|
|
|
1feee8 |
--- a/sysdeps/x86_64/multiarch/strncmp.c
|
|
|
1feee8 |
+++ b/sysdeps/x86_64/multiarch/strncmp.c
|
|
|
1feee8 |
@@ -39,11 +39,11 @@ 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, BMI2)
|
|
|
1feee8 |
&& CPU_FEATURES_ARCH_P (cpu_features, AVX_Fast_Unaligned_Load))
|
|
|
1feee8 |
{
|
|
|
1feee8 |
if (CPU_FEATURE_USABLE_P (cpu_features, AVX512VL)
|
|
|
1feee8 |
- && CPU_FEATURE_USABLE_P (cpu_features, AVX512BW)
|
|
|
1feee8 |
- && CPU_FEATURE_USABLE_P (cpu_features, BMI2))
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE_P (cpu_features, AVX512BW))
|
|
|
1feee8 |
return OPTIMIZE (evex);
|
|
|
1feee8 |
|
|
|
1feee8 |
if (CPU_FEATURE_USABLE_P (cpu_features, RTM))
|