diff options
author | Kenneth Heafield <github@kheafield.com> | 2020-03-08 22:22:24 +0300 |
---|---|---|
committer | Kenneth Heafield <github@kheafield.com> | 2020-03-08 22:22:24 +0300 |
commit | 261a5fbcf7558fc3c2ac22b33fe0c2930d440fc3 (patch) | |
tree | f912e6a3cec9691e509d3e63eb3672ae6a357d2c /multiply.h | |
parent | 1323f245adff16b351b61310625b743b22e90059 (diff) |
Change to INTGEMM_COMPILER_SUPPORTS_AVX512BW and update test
Diffstat (limited to 'multiply.h')
-rw-r--r-- | multiply.h | 4 |
1 files changed, 2 insertions, 2 deletions
@@ -36,7 +36,7 @@ INTGEMM_AVX2 static inline __m256i PermuteSummer(__m256i pack0123, __m256i pack4 return _mm256_add_epi32(rev, blended); } -#ifdef INTGEMM_COMPILER_SUPPORTS_AVX512 +#ifdef INTGEMM_COMPILER_SUPPORTS_AVX512BW /* Only INTGEMM_AVX512F is necessary but due to GCC 5.4 bug we have to set INTGEMM_AVX512BW */ INTGEMM_AVX512BW static inline __m256i PermuteSummer(__m512i pack0123, __m512i pack4567) { // Form [0th 128-bit register of pack0123, 0st 128-bit register of pack4567, 2nd 128-bit register of pack0123, 2nd 128-bit register of pack4567] @@ -104,7 +104,7 @@ target inline Register Pack0123(Register sum0, Register sum1, Register sum2, Reg INTGEMM_PACK0123(INTGEMM_SSE2, __m128i) INTGEMM_PACK0123(INTGEMM_AVX2, __m256i) -#ifdef INTGEMM_COMPILER_SUPPORTS_AVX512 +#ifdef INTGEMM_COMPILER_SUPPORTS_AVX512BW /* Only INTGEMM_AVX512F is necessary but due to GCC 5.4 bug we have to set INTGEMM_AVX512BW */ INTGEMM_PACK0123(INTGEMM_AVX512BW, __m512i) #endif |