ggerganov commited on
Commit
6c4692f
·
unverified ·
1 Parent(s): 2078d85

ggml : when using BLAS start only 1 CPU thread

Browse files
Files changed (1) hide show
  1. ggml.c +1 -0
ggml.c CHANGED
@@ -7257,6 +7257,7 @@ void ggml_graph_compute(struct ggml_context * ctx, struct ggml_cgraph * cgraph)
7257
  node->src1->type == GGML_TYPE_F32) {
7258
  #if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
7259
  if (ggml_compute_forward_mul_mat_use_blas(node->src0, node->src1, node)) {
 
7260
  cur = sizeof(float)*(node->src0->ne[0]*node->src0->ne[1]);
7261
  } else {
7262
  cur = sizeof(ggml_fp16_t)*ggml_nelements(node->src1);
 
7257
  node->src1->type == GGML_TYPE_F32) {
7258
  #if defined(GGML_USE_ACCELERATE) || defined(GGML_USE_OPENBLAS)
7259
  if (ggml_compute_forward_mul_mat_use_blas(node->src0, node->src1, node)) {
7260
+ node->n_tasks = 1;
7261
  cur = sizeof(float)*(node->src0->ne[0]*node->src0->ne[1]);
7262
  } else {
7263
  cur = sizeof(ggml_fp16_t)*ggml_nelements(node->src1);