diff options
| author | Kenneth Heafield <github@kheafield.com> | 2018-06-15 12:37:11 +0300 |
|---|---|---|
| committer | Kenneth Heafield <github@kheafield.com> | 2018-06-15 12:37:11 +0300 |
| commit | 5753664e707b8d956136d8f0f8b74395edf47940 (patch) | |
| tree | 224a796f037ef01b0738702a5b9de056949380b8 /avx2_gemm.h | |
| parent | 5451b34b3a315b954897636442b4f61e70f8dd6d (diff) | |
8-way Reduce, contrastive version with better memory
Diffstat (limited to 'avx2_gemm.h')
| -rw-r--r-- | avx2_gemm.h | 1 |
1 files changed, 1 insertions, 0 deletions
diff --git a/avx2_gemm.h b/avx2_gemm.h index 26a0cee..a3f959e 100644 --- a/avx2_gemm.h +++ b/avx2_gemm.h @@ -15,6 +15,7 @@ void Quantize8(const float *input, int8_t *output, float quant_mult, std::size_t // C should be the usual 4-byte alignment. void MatrixMult16(const __m256i *A, const __m256i *B, float *C, float unquant_mult, int num_A_rows, int num_B_rows, int width); void MatrixMult8(const __m256i *A, const __m256i *B, float *C, float unquant_mult, int num_A_rows, int num_B_rows, int width); +void MatrixMult8Contrast(const __m256i *A, const __m256i *B, float *C, float unquant_mult, int num_A_rows, int num_B_rows, int width); } // namespace AVX2 #endif // __AVX2__ |