AuroraMiddleware/glibc
1
0
Fork 0
mirror of https://sourceware.org/git/glibc.git synced 2024-11-22 13:00:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

Skip SSE4.2 versions on Intel Silvermont

Browse Source 
SSE2/SSSE3 versions are faster than SSE4.2 versions on Intel Silvermont.
This commit is contained in:
Liubov Dmitrieva 2013-06-28 15:28:50 -07:00 committed by H.J. Lu
parent 89cd956937
commit 6308fd9a46
6 changed files with 51 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
ChangeLog
View File

@ -1,3 +1,17 @@
2013-06-28 Liubov Dmitrieva <liubov.dmitrieva@intel.com>
* sysdeps/x86_64/multiarch/init-arch.c (__init_cpu_features): Set
bit_Slow_SSE4_2 and bit_Prefer_PMINUB_for_stringop for Intel
Silvermont.
* sysdeps/x86_64/multiarch/init-arch.h (bit_Slow_SSE4_2): New
macro.
(index_Slow_SSE4_2): Likewise.
(index_Prefer_PMINUB_for_stringop): Likewise.
* sysdeps/x86_64/multiarch/strchr.S: Skip SSE4.2 version if
bit_Slow_SSE4_2 is set.
* sysdeps/x86_64/multiarch/strcmp.S: Likewise.
* sysdeps/x86_64/multiarch/strrchr.S: Likewise.
2013-06-28 Ryan S. Arnold <rsa@linux.vnet.ibm.com> 2013-06-28 Ryan S. Arnold <rsa@linux.vnet.ibm.com>
* sysdeps/powerpc/Makefile: Add comment about generating an offset to * sysdeps/powerpc/Makefile: Add comment about generating an offset to

10
sysdeps/x86_64/multiarch/init-arch.c
View File

@ -81,8 +81,16 @@ __init_cpu_features (void)
case 0x37: case 0x37:
/* Unaligned load versions are faster than SSSE3 /* Unaligned load versions are faster than SSSE3
on Silvermont. */ on Silvermont. */
#if index_Fast_Unaligned_Load != index_Prefer_PMINUB_for_stringop
# error index_Fast_Unaligned_Load != index_Prefer_PMINUB_for_stringop
#endif
#if index_Fast_Unaligned_Load != index_Slow_SSE4_2
# error index_Fast_Unaligned_Load != index_Slow_SSE4_2
#endif
__cpu_features.feature[index_Fast_Unaligned_Load] __cpu_features.feature[index_Fast_Unaligned_Load]
|= bit_Fast_Unaligned_Load; |= (bit_Fast_Unaligned_Load
| bit_Prefer_PMINUB_for_stringop
| bit_Slow_SSE4_2);
break; break;
default: default:

4
sysdeps/x86_64/multiarch/init-arch.h
View File

@ -23,6 +23,7 @@
#define bit_AVX_Usable (1 << 6) #define bit_AVX_Usable (1 << 6)
#define bit_FMA_Usable (1 << 7) #define bit_FMA_Usable (1 << 7)
#define bit_FMA4_Usable (1 << 8) #define bit_FMA4_Usable (1 << 8)
#define bit_Slow_SSE4_2 (1 << 9)
/* CPUID Feature flags. */ /* CPUID Feature flags. */
@ -62,6 +63,7 @@
# define index_AVX_Usable FEATURE_INDEX_1*FEATURE_SIZE # define index_AVX_Usable FEATURE_INDEX_1*FEATURE_SIZE
# define index_FMA_Usable FEATURE_INDEX_1*FEATURE_SIZE # define index_FMA_Usable FEATURE_INDEX_1*FEATURE_SIZE
# define index_FMA4_Usable FEATURE_INDEX_1*FEATURE_SIZE # define index_FMA4_Usable FEATURE_INDEX_1*FEATURE_SIZE
# define index_Slow_SSE4_2 FEATURE_INDEX_1*FEATURE_SIZE
#else /* __ASSEMBLER__ */ #else /* __ASSEMBLER__ */
@ -156,9 +158,11 @@ extern const struct cpu_features *__get_cpu_features (void)
# define index_Fast_Copy_Backward FEATURE_INDEX_1 # define index_Fast_Copy_Backward FEATURE_INDEX_1
# define index_Slow_BSF FEATURE_INDEX_1 # define index_Slow_BSF FEATURE_INDEX_1
# define index_Fast_Unaligned_Load FEATURE_INDEX_1 # define index_Fast_Unaligned_Load FEATURE_INDEX_1
# define index_Prefer_PMINUB_for_stringop FEATURE_INDEX_1
# define index_AVX_Usable FEATURE_INDEX_1 # define index_AVX_Usable FEATURE_INDEX_1
# define index_FMA_Usable FEATURE_INDEX_1 # define index_FMA_Usable FEATURE_INDEX_1
# define index_FMA4_Usable FEATURE_INDEX_1 # define index_FMA4_Usable FEATURE_INDEX_1
# define index_Slow_SSE4_2 FEATURE_INDEX_1
# define HAS_ARCH_FEATURE(name) \ # define HAS_ARCH_FEATURE(name) \
((__get_cpu_features ()->feature[index_##name] & (bit_##name)) != 0) ((__get_cpu_features ()->feature[index_##name] & (bit_##name)) != 0)

2
sysdeps/x86_64/multiarch/strchr.S
View File

@ -29,6 +29,8 @@ ENTRY(strchr)
jne 1f jne 1f
call __init_cpu_features call __init_cpu_features
1: leaq __strchr_sse2(%rip), %rax 1: leaq __strchr_sse2(%rip), %rax
testl $bit_Slow_SSE4_2, __cpu_features+CPUID_OFFSET+index_Slow_SSE4_2(%rip)
jnz 2f
testl $bit_SSE4_2, __cpu_features+CPUID_OFFSET+index_SSE4_2(%rip) testl $bit_SSE4_2, __cpu_features+CPUID_OFFSET+index_SSE4_2(%rip)
jz 2f jz 2f
leaq __strchr_sse42(%rip), %rax leaq __strchr_sse42(%rip), %rax

34
sysdeps/x86_64/multiarch/strcmp.S
View File

@ -88,14 +88,16 @@ ENTRY(STRCMP)
jne 1f jne 1f
call __init_cpu_features call __init_cpu_features
1: 1:
testl $bit_Slow_SSE4_2, __cpu_features+CPUID_OFFSET+index_Slow_SSE4_2(%rip)
jnz 2f
leaq STRCMP_SSE42(%rip), %rax leaq STRCMP_SSE42(%rip), %rax
testl $bit_SSE4_2, __cpu_features+CPUID_OFFSET+index_SSE4_2(%rip) testl $bit_SSE4_2, __cpu_features+CPUID_OFFSET+index_SSE4_2(%rip)
jnz 2f jnz 3f
leaq STRCMP_SSSE3(%rip), %rax 2: leaq STRCMP_SSSE3(%rip), %rax
testl $bit_SSSE3, __cpu_features+CPUID_OFFSET+index_SSSE3(%rip) testl $bit_SSSE3, __cpu_features+CPUID_OFFSET+index_SSSE3(%rip)
jnz 2f jnz 3f
leaq STRCMP_SSE2(%rip), %rax leaq STRCMP_SSE2(%rip), %rax
2: ret 3: ret
END(STRCMP) END(STRCMP)
# ifdef USE_AS_STRCASECMP_L # ifdef USE_AS_STRCASECMP_L
@ -109,16 +111,18 @@ ENTRY(__strcasecmp)
# ifdef HAVE_AVX_SUPPORT # ifdef HAVE_AVX_SUPPORT
leaq __strcasecmp_avx(%rip), %rax leaq __strcasecmp_avx(%rip), %rax
testl $bit_AVX_Usable, __cpu_features+FEATURE_OFFSET+index_AVX_Usable(%rip) testl $bit_AVX_Usable, __cpu_features+FEATURE_OFFSET+index_AVX_Usable(%rip)
jnz 2f jnz 3f
# endif # endif
testl $bit_Slow_SSE4_2, __cpu_features+CPUID_OFFSET+index_Slow_SSE4_2(%rip)
jnz 2f
leaq __strcasecmp_sse42(%rip), %rax leaq __strcasecmp_sse42(%rip), %rax
testl $bit_SSE4_2, __cpu_features+CPUID_OFFSET+index_SSE4_2(%rip) testl $bit_SSE4_2, __cpu_features+CPUID_OFFSET+index_SSE4_2(%rip)
jnz 2f jnz 3f
leaq __strcasecmp_ssse3(%rip), %rax 2: leaq __strcasecmp_ssse3(%rip), %rax
testl $bit_SSSE3, __cpu_features+CPUID_OFFSET+index_SSSE3(%rip) testl $bit_SSSE3, __cpu_features+CPUID_OFFSET+index_SSSE3(%rip)
jnz 2f jnz 3f
leaq __strcasecmp_sse2(%rip), %rax leaq __strcasecmp_sse2(%rip), %rax
2: ret 3: ret
END(__strcasecmp) END(__strcasecmp)
weak_alias (__strcasecmp, strcasecmp) weak_alias (__strcasecmp, strcasecmp)
# endif # endif
@ -133,16 +137,18 @@ ENTRY(__strncasecmp)
# ifdef HAVE_AVX_SUPPORT # ifdef HAVE_AVX_SUPPORT
leaq __strncasecmp_avx(%rip), %rax leaq __strncasecmp_avx(%rip), %rax
testl $bit_AVX_Usable, __cpu_features+FEATURE_OFFSET+index_AVX_Usable(%rip) testl $bit_AVX_Usable, __cpu_features+FEATURE_OFFSET+index_AVX_Usable(%rip)
jnz 2f jnz 3f
# endif # endif
testl $bit_Slow_SSE4_2, __cpu_features+CPUID_OFFSET+index_Slow_SSE4_2(%rip)
jnz 2f
leaq __strncasecmp_sse42(%rip), %rax leaq __strncasecmp_sse42(%rip), %rax
testl $bit_SSE4_2, __cpu_features+CPUID_OFFSET+index_SSE4_2(%rip) testl $bit_SSE4_2, __cpu_features+CPUID_OFFSET+index_SSE4_2(%rip)
jnz 2f jnz 3f
leaq __strncasecmp_ssse3(%rip), %rax 2: leaq __strncasecmp_ssse3(%rip), %rax
testl $bit_SSSE3, __cpu_features+CPUID_OFFSET+index_SSSE3(%rip) testl $bit_SSSE3, __cpu_features+CPUID_OFFSET+index_SSSE3(%rip)
jnz 2f jnz 3f
leaq __strncasecmp_sse2(%rip), %rax leaq __strncasecmp_sse2(%rip), %rax
2: ret 3: ret
END(__strncasecmp) END(__strncasecmp)
weak_alias (__strncasecmp, strncasecmp) weak_alias (__strncasecmp, strncasecmp)
# endif # endif

2
sysdeps/x86_64/multiarch/strrchr.S
View File

@ -32,6 +32,8 @@ ENTRY(strrchr)
jne 1f jne 1f
call __init_cpu_features call __init_cpu_features
1: leaq __strrchr_sse2(%rip), %rax 1: leaq __strrchr_sse2(%rip), %rax
testl $bit_Slow_SSE4_2, __cpu_features+CPUID_OFFSET+index_Slow_SSE4_2(%rip)
jnz 2f
testl $bit_SSE4_2, __cpu_features+CPUID_OFFSET+index_SSE4_2(%rip) testl $bit_SSE4_2, __cpu_features+CPUID_OFFSET+index_SSE4_2(%rip)
jz 2f jz 2f
leaq __strrchr_sse42(%rip), %rax leaq __strrchr_sse42(%rip), %rax