diff options
| author | H.J. Lu <hjl.tools@gmail.com> | 2021-03-05 07:20:28 -0800 |
|---|---|---|
| committer | H.J. Lu <hjl.tools@gmail.com> | 2021-03-29 07:40:17 -0700 |
| commit | 91264fe3577fe887b4860923fa6142b5274c8965 (patch) | |
| tree | 172c0193a4e0a6b0e695fd8e0df51f0d746dd817 /sysdeps/x86_64/multiarch/ifunc-memcmp.h | |
| parent | 1b968b6b9b3aac702ac2f133e0dd16cfdbb415ee (diff) | |
| download | glibc-91264fe3577fe887b4860923fa6142b5274c8965.tar.gz glibc-91264fe3577fe887b4860923fa6142b5274c8965.tar.xz glibc-91264fe3577fe887b4860923fa6142b5274c8965.zip | |
x86-64: Add memcmp family functions with 256-bit EVEX
Update ifunc-memcmp.h to select the function optimized with 256-bit EVEX instructions using YMM16-YMM31 registers to avoid RTM abort with usable AVX512VL, AVX512BW and MOVBE since VZEROUPPER isn't needed at function exit.
Diffstat (limited to 'sysdeps/x86_64/multiarch/ifunc-memcmp.h')
| -rw-r--r-- | sysdeps/x86_64/multiarch/ifunc-memcmp.h | 13 |
1 files changed, 10 insertions, 3 deletions
diff --git a/sysdeps/x86_64/multiarch/ifunc-memcmp.h b/sysdeps/x86_64/multiarch/ifunc-memcmp.h index d5df541ec4..5ac41a19b8 100644 --- a/sysdeps/x86_64/multiarch/ifunc-memcmp.h +++ b/sysdeps/x86_64/multiarch/ifunc-memcmp.h @@ -23,17 +23,24 @@ extern __typeof (REDIRECT_NAME) OPTIMIZE (sse2) attribute_hidden; extern __typeof (REDIRECT_NAME) OPTIMIZE (ssse3) attribute_hidden; extern __typeof (REDIRECT_NAME) OPTIMIZE (sse4_1) attribute_hidden; extern __typeof (REDIRECT_NAME) OPTIMIZE (avx2_movbe) attribute_hidden; +extern __typeof (REDIRECT_NAME) OPTIMIZE (evex_movbe) attribute_hidden; static inline void * IFUNC_SELECTOR (void) { const struct cpu_features* cpu_features = __get_cpu_features (); - if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER) - && CPU_FEATURE_USABLE_P (cpu_features, AVX2) + if (CPU_FEATURE_USABLE_P (cpu_features, AVX2) && CPU_FEATURE_USABLE_P (cpu_features, MOVBE) && CPU_FEATURES_ARCH_P (cpu_features, AVX_Fast_Unaligned_Load)) - return OPTIMIZE (avx2_movbe); + { + if (CPU_FEATURE_USABLE_P (cpu_features, AVX512VL) + && CPU_FEATURE_USABLE_P (cpu_features, AVX512BW)) + return OPTIMIZE (evex_movbe); + + if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER)) + return OPTIMIZE (avx2_movbe); + } if (CPU_FEATURE_USABLE_P (cpu_features, SSE4_1)) return OPTIMIZE (sse4_1); |