qiurui
/
glibc
mirror of git://sourceware.org/git/glibc.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

x86: Remove mem{move|cpy}-ssse3-back 

With SSE2, SSE4.1, AVX2, and EVEX versions very few targets prefer
SSSE3. As a result it is no longer worth it to keep the SSSE3
versions given the code size cost.
Reviewed-by: H.J. Lu <hjl.tools@gmail.com>
This commit is contained in:
Noah Goldstein 2022-04-14 11:47:38 -05:00
parent 41bfe224e5
commit d85916e30a
5 changed files with 6 additions and 3212 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
sysdeps/x86_64/multiarch/Makefile
View File

@ -17,7 +17,6 @@ sysdep_routines += \
memcmpeq-evex \ memcmpeq-evex \
memcmpeq-sse2 \ memcmpeq-sse2 \
memcpy-ssse3 \ memcpy-ssse3 \
memcpy-ssse3-back \
memmove-avx-unaligned-erms \ memmove-avx-unaligned-erms \
memmove-avx-unaligned-erms-rtm \ memmove-avx-unaligned-erms-rtm \
memmove-avx512-no-vzeroupper \ memmove-avx512-no-vzeroupper \
@ -25,7 +24,6 @@ sysdep_routines += \
memmove-evex-unaligned-erms \ memmove-evex-unaligned-erms \
memmove-sse2-unaligned-erms \ memmove-sse2-unaligned-erms \
memmove-ssse3 \ memmove-ssse3 \
memmove-ssse3-back \
memrchr-avx2 \ memrchr-avx2 \
memrchr-avx2-rtm \ memrchr-avx2-rtm \
memrchr-evex \ memrchr-evex \

15
sysdeps/x86_64/multiarch/ifunc-impl-list.c
View File

@ -132,9 +132,6 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
IFUNC_IMPL_ADD (array, i, __memmove_chk, IFUNC_IMPL_ADD (array, i, __memmove_chk,
CPU_FEATURE_USABLE (AVX512VL), CPU_FEATURE_USABLE (AVX512VL),
__memmove_chk_evex_unaligned_erms) __memmove_chk_evex_unaligned_erms)
IFUNC_IMPL_ADD (array, i, __memmove_chk,
CPU_FEATURE_USABLE (SSSE3),
__memmove_chk_ssse3_back)
IFUNC_IMPL_ADD (array, i, __memmove_chk, IFUNC_IMPL_ADD (array, i, __memmove_chk,
CPU_FEATURE_USABLE (SSSE3), CPU_FEATURE_USABLE (SSSE3),
__memmove_chk_ssse3) __memmove_chk_ssse3)
@ -177,8 +174,6 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
IFUNC_IMPL_ADD (array, i, memmove, IFUNC_IMPL_ADD (array, i, memmove,
CPU_FEATURE_USABLE (AVX512VL), CPU_FEATURE_USABLE (AVX512VL),
__memmove_avx512_unaligned_erms) __memmove_avx512_unaligned_erms)
IFUNC_IMPL_ADD (array, i, memmove, CPU_FEATURE_USABLE (SSSE3),
__memmove_ssse3_back)
IFUNC_IMPL_ADD (array, i, memmove, CPU_FEATURE_USABLE (SSSE3), IFUNC_IMPL_ADD (array, i, memmove, CPU_FEATURE_USABLE (SSSE3),
__memmove_ssse3) __memmove_ssse3)
IFUNC_IMPL_ADD (array, i, memmove, 1, __memmove_erms) IFUNC_IMPL_ADD (array, i, memmove, 1, __memmove_erms)
@ -872,9 +867,6 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
IFUNC_IMPL_ADD (array, i, __memcpy_chk, IFUNC_IMPL_ADD (array, i, __memcpy_chk,
CPU_FEATURE_USABLE (AVX512VL), CPU_FEATURE_USABLE (AVX512VL),
__memcpy_chk_evex_unaligned_erms) __memcpy_chk_evex_unaligned_erms)
IFUNC_IMPL_ADD (array, i, __memcpy_chk,
CPU_FEATURE_USABLE (SSSE3),
__memcpy_chk_ssse3_back)
IFUNC_IMPL_ADD (array, i, __memcpy_chk, IFUNC_IMPL_ADD (array, i, __memcpy_chk,
CPU_FEATURE_USABLE (SSSE3), CPU_FEATURE_USABLE (SSSE3),
__memcpy_chk_ssse3) __memcpy_chk_ssse3)
@ -908,8 +900,6 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
IFUNC_IMPL_ADD (array, i, memcpy, IFUNC_IMPL_ADD (array, i, memcpy,
CPU_FEATURE_USABLE (AVX512VL), CPU_FEATURE_USABLE (AVX512VL),
__memcpy_evex_unaligned_erms) __memcpy_evex_unaligned_erms)
IFUNC_IMPL_ADD (array, i, memcpy, CPU_FEATURE_USABLE (SSSE3),
__memcpy_ssse3_back)
IFUNC_IMPL_ADD (array, i, memcpy, CPU_FEATURE_USABLE (SSSE3), IFUNC_IMPL_ADD (array, i, memcpy, CPU_FEATURE_USABLE (SSSE3),
__memcpy_ssse3) __memcpy_ssse3)
IFUNC_IMPL_ADD (array, i, memcpy, IFUNC_IMPL_ADD (array, i, memcpy,
@ -958,9 +948,6 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
IFUNC_IMPL_ADD (array, i, __mempcpy_chk, IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
CPU_FEATURE_USABLE (AVX512VL), CPU_FEATURE_USABLE (AVX512VL),
__mempcpy_chk_evex_unaligned_erms) __mempcpy_chk_evex_unaligned_erms)
IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
CPU_FEATURE_USABLE (SSSE3),
__mempcpy_chk_ssse3_back)
IFUNC_IMPL_ADD (array, i, __mempcpy_chk, IFUNC_IMPL_ADD (array, i, __mempcpy_chk,
CPU_FEATURE_USABLE (SSSE3), CPU_FEATURE_USABLE (SSSE3),
__mempcpy_chk_ssse3) __mempcpy_chk_ssse3)
@ -1003,8 +990,6 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
IFUNC_IMPL_ADD (array, i, mempcpy, IFUNC_IMPL_ADD (array, i, mempcpy,
CPU_FEATURE_USABLE (AVX512VL), CPU_FEATURE_USABLE (AVX512VL),
__mempcpy_evex_unaligned_erms) __mempcpy_evex_unaligned_erms)
IFUNC_IMPL_ADD (array, i, mempcpy, CPU_FEATURE_USABLE (SSSE3),
__mempcpy_ssse3_back)
IFUNC_IMPL_ADD (array, i, mempcpy, CPU_FEATURE_USABLE (SSSE3), IFUNC_IMPL_ADD (array, i, mempcpy, CPU_FEATURE_USABLE (SSSE3),
__mempcpy_ssse3) __mempcpy_ssse3)
IFUNC_IMPL_ADD (array, i, mempcpy, 1, IFUNC_IMPL_ADD (array, i, mempcpy, 1,

14
sysdeps/x86_64/multiarch/ifunc-memmove.h
View File

@ -25,7 +25,6 @@ extern __typeof (REDIRECT_NAME) OPTIMIZE (sse2_unaligned)
extern __typeof (REDIRECT_NAME) OPTIMIZE (sse2_unaligned_erms) extern __typeof (REDIRECT_NAME) OPTIMIZE (sse2_unaligned_erms)
attribute_hidden; attribute_hidden;
extern __typeof (REDIRECT_NAME) OPTIMIZE (ssse3) attribute_hidden; extern __typeof (REDIRECT_NAME) OPTIMIZE (ssse3) attribute_hidden;
extern __typeof (REDIRECT_NAME) OPTIMIZE (ssse3_back) attribute_hidden;
extern __typeof (REDIRECT_NAME) OPTIMIZE (avx_unaligned) attribute_hidden; extern __typeof (REDIRECT_NAME) OPTIMIZE (avx_unaligned) attribute_hidden;
extern __typeof (REDIRECT_NAME) OPTIMIZE (avx_unaligned_erms) extern __typeof (REDIRECT_NAME) OPTIMIZE (avx_unaligned_erms)
attribute_hidden; attribute_hidden;
@ -94,17 +93,14 @@ IFUNC_SELECTOR (void)
} }
} }
if (!CPU_FEATURE_USABLE_P (cpu_features, SSSE3) if (CPU_FEATURE_USABLE_P (cpu_features, SSSE3)
|| CPU_FEATURES_ARCH_P (cpu_features, Fast_Unaligned_Copy)) && !CPU_FEATURES_ARCH_P (cpu_features, Fast_Unaligned_Copy))
{ {
return OPTIMIZE (ssse3);
}
if (CPU_FEATURE_USABLE_P (cpu_features, ERMS)) if (CPU_FEATURE_USABLE_P (cpu_features, ERMS))
return OPTIMIZE (sse2_unaligned_erms); return OPTIMIZE (sse2_unaligned_erms);
return OPTIMIZE (sse2_unaligned); return OPTIMIZE (sse2_unaligned);
} }
if (CPU_FEATURES_ARCH_P (cpu_features, Fast_Copy_Backward))
return OPTIMIZE (ssse3_back);
return OPTIMIZE (ssse3);
}

3181
sysdeps/x86_64/multiarch/memcpy-ssse3-back.S
View File

File diff suppressed because it is too large Load Diff

4
sysdeps/x86_64/multiarch/memmove-ssse3-back.S
View File

@ -1,4 +0,0 @@
#define USE_AS_MEMMOVE
#define MEMCPY __memmove_ssse3_back
#define MEMCPY_CHK __memmove_chk_ssse3_back
#include "memcpy-ssse3-back.S"