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

done

Browse source
This commit is contained in:
jon-chuang 2023-04-26 23:03:54 +08:00
parent 8ead56c03a
commit 8cead20746
1 changed files with 2 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

5
ggml.c
View file

@ -494,7 +494,7 @@ void ggml_mul_row_f32_tall_skinny(const float * A, const float * B, float * C, i
} }
// Store the result in the corresponding row of C // Store the result in the corresponding row of C
_mm256_store_ps(&res_vec, c_vec); _mm256_store_ps((float *) &res_vec, c_vec);
for (int k = 0; k < 8; ++k) { for (int k = 0; k < 8; ++k) {
C[j+k] = res_vec[k]; C[j+k] = res_vec[k];
@ -7700,7 +7700,7 @@ static void ggml_compute_forward_mul_mat_f32(
const int64_t ne02 = src0->ne[2]; const int64_t ne02 = src0->ne[2];
const int64_t ne03 = src0->ne[3]; const int64_t ne03 = src0->ne[3];
#if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS) || defined(GGML_USE_CUBLAS) || defined(__AVX2__) || defined(__AVX__) #if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS) || defined(GGML_USE_CUBLAS)
const int64_t ne10 = src1->ne[0]; const int64_t ne10 = src1->ne[0];
#endif #endif
const int64_t ne11 = src1->ne[1]; const int64_t ne11 = src1->ne[1];
@ -7758,7 +7758,6 @@ static void ggml_compute_forward_mul_mat_f32(
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?
assert(ne00 == ne10);
if (params->type == GGML_TASK_INIT || params->type == GGML_TASK_FINALIZE) { if (params->type == GGML_TASK_INIT || params->type == GGML_TASK_FINALIZE) {
return; return;
} }