|
 |
1feee8 |
commit 414fc856ff4e011e62b88a21d30294637a152dc7
|
|
|
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 str(n)casecmp implementations
|
|
|
1feee8 |
|
|
|
1feee8 |
The AVX2 str(n)casecmp implementations use 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 10f79d3670b036925da63dc532b122d27ce65ff8)
|
|
|
1feee8 |
|
|
|
1feee8 |
diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
|
|
|
1feee8 |
index 8d649e263eb24b8a..ca64b34c146a76f9 100644
|
|
|
1feee8 |
--- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
|
|
|
1feee8 |
+++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
|
|
|
1feee8 |
@@ -386,13 +386,16 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
1feee8 |
IFUNC_IMPL (i, name, strcasecmp,
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strcasecmp,
|
|
|
1feee8 |
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
1feee8 |
- && CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (AVX512BW)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)),
|
|
|
1feee8 |
__strcasecmp_evex)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strcasecmp,
|
|
|
1feee8 |
- CPU_FEATURE_USABLE (AVX2),
|
|
|
1feee8 |
+ (CPU_FEATURE_USABLE (AVX2)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)),
|
|
|
1feee8 |
__strcasecmp_avx2)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strcasecmp,
|
|
|
1feee8 |
(CPU_FEATURE_USABLE (AVX2)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)
|
|
|
1feee8 |
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
1feee8 |
__strcasecmp_avx2_rtm)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strcasecmp,
|
|
|
1feee8 |
@@ -407,13 +410,16 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
1feee8 |
IFUNC_IMPL (i, name, strcasecmp_l,
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strcasecmp,
|
|
|
1feee8 |
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
1feee8 |
- && CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (AVX512BW)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)),
|
|
|
1feee8 |
__strcasecmp_l_evex)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strcasecmp,
|
|
|
1feee8 |
- CPU_FEATURE_USABLE (AVX2),
|
|
|
1feee8 |
+ (CPU_FEATURE_USABLE (AVX2)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)),
|
|
|
1feee8 |
__strcasecmp_l_avx2)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strcasecmp,
|
|
|
1feee8 |
(CPU_FEATURE_USABLE (AVX2)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)
|
|
|
1feee8 |
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
1feee8 |
__strcasecmp_l_avx2_rtm)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strcasecmp_l,
|
|
|
1feee8 |
@@ -542,13 +548,16 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
1feee8 |
IFUNC_IMPL (i, name, strncasecmp,
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strncasecmp,
|
|
|
1feee8 |
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
1feee8 |
- && CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (AVX512BW)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)),
|
|
|
1feee8 |
__strncasecmp_evex)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strncasecmp,
|
|
|
1feee8 |
- CPU_FEATURE_USABLE (AVX2),
|
|
|
1feee8 |
+ (CPU_FEATURE_USABLE (AVX2)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)),
|
|
|
1feee8 |
__strncasecmp_avx2)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strncasecmp,
|
|
|
1feee8 |
(CPU_FEATURE_USABLE (AVX2)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)
|
|
|
1feee8 |
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
1feee8 |
__strncasecmp_avx2_rtm)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strncasecmp,
|
|
|
1feee8 |
@@ -564,13 +573,16 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
|
|
|
1feee8 |
IFUNC_IMPL (i, name, strncasecmp_l,
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strncasecmp,
|
|
|
1feee8 |
(CPU_FEATURE_USABLE (AVX512VL)
|
|
|
1feee8 |
- && CPU_FEATURE_USABLE (AVX512BW)),
|
|
|
1feee8 |
+ & CPU_FEATURE_USABLE (AVX512BW)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)),
|
|
|
1feee8 |
__strncasecmp_l_evex)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strncasecmp,
|
|
|
1feee8 |
- CPU_FEATURE_USABLE (AVX2),
|
|
|
1feee8 |
+ (CPU_FEATURE_USABLE (AVX2)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)),
|
|
|
1feee8 |
__strncasecmp_l_avx2)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strncasecmp,
|
|
|
1feee8 |
(CPU_FEATURE_USABLE (AVX2)
|
|
|
1feee8 |
+ && CPU_FEATURE_USABLE (BMI2)
|
|
|
1feee8 |
&& CPU_FEATURE_USABLE (RTM)),
|
|
|
1feee8 |
__strncasecmp_l_avx2_rtm)
|
|
|
1feee8 |
IFUNC_IMPL_ADD (array, i, strncasecmp_l,
|
|
|
1feee8 |
diff --git a/sysdeps/x86_64/multiarch/ifunc-strcasecmp.h b/sysdeps/x86_64/multiarch/ifunc-strcasecmp.h
|
|
|
1feee8 |
index 40819caf5ab10337..e61d6e9497bce9d9 100644
|
|
|
1feee8 |
--- a/sysdeps/x86_64/multiarch/ifunc-strcasecmp.h
|
|
|
1feee8 |
+++ b/sysdeps/x86_64/multiarch/ifunc-strcasecmp.h
|
|
|
1feee8 |
@@ -32,6 +32,7 @@ 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, BMI2)
|
|
|
1feee8 |
&& CPU_FEATURES_ARCH_P (cpu_features, AVX_Fast_Unaligned_Load))
|
|
|
1feee8 |
{
|
|
|
1feee8 |
if (CPU_FEATURE_USABLE_P (cpu_features, AVX512VL)
|