vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

minor

Browse source
This commit is contained in:
jon-chuang 2023-04-30 20:56:46 +08:00
parent fb469ed972
commit 470cc4c5d1
1 changed files with 2 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
ggml.c
View file

@ -497,7 +497,7 @@ static inline int hsum_i32_4(const __m128i a) {
} }
// AVX routine provided by GH user jon-chuang // AVX routine provided by GH user jon-chuang
#if (__AVX2__ || __AVX512F__) && FMA #if (__AVX2__ || __AVX512F__) && __FMA__
// Given A = K X M, B = K X N, compute one row of C = A^TB // Given A = K X M, B = K X N, compute one row of C = A^TB
void ggml_mul_row_f32_tall_skinny(const float * A, const float * B, float * C, int M, int N, int K) { void ggml_mul_row_f32_tall_skinny(const float * A, const float * B, float * C, int M, int N, int K) {
alignas(32) float res_vec[8]; alignas(32) float res_vec[8];
@ -8123,7 +8123,7 @@ static void ggml_compute_forward_mul_mat_f32(
assert(ne2 == ne02); assert(ne2 == ne02);
assert(ne3 == ne03); assert(ne3 == ne03);
#if defined(__AVX2__) || defined(__AVX__) #if (__AVX512F__ || __AVX2__ || __AVX__) && __FMA__
if ((ggml_cpu_has_avx2() && ne00 <= 48) || ne00 <= 32) { if ((ggml_cpu_has_avx2() && ne00 <= 48) || ne00 <= 32) {
// Handle tall and skinny matrices // Handle tall and skinny matrices
// TODO(jon-chuang): Also check that we only handle 2D matrices? // TODO(jon-chuang): Also check that we only handle 2D matrices?