Skip to content

Commit 86c90e3

Browse files
committed
metal : disable concurrency optimization
1 parent f015b26 commit 86c90e3

File tree

1 file changed

+3
-3
lines changed

1 file changed

+3
-3
lines changed

llama.cpp

+3-3
Original file line numberDiff line numberDiff line change
@@ -6605,8 +6605,8 @@ struct llama_context * llama_new_context_with_model(
66056605
llama_free(ctx);
66066606
return NULL;
66076607
}
6608-
ggml_metal_graph_find_concurrency(ctx->ctx_metal, gf, false);
6609-
ggml_allocr_set_parse_seq(ctx->alloc, ggml_metal_get_concur_list(ctx->ctx_metal), ggml_metal_if_optimized(ctx->ctx_metal));
6608+
//ggml_metal_graph_find_concurrency(ctx->ctx_metal, gf, false);
6609+
//ggml_allocr_set_parse_seq(ctx->alloc, ggml_metal_get_concur_list(ctx->ctx_metal), ggml_metal_if_optimized(ctx->ctx_metal));
66106610
}
66116611
#endif
66126612
// measure memory requirements for the graph
@@ -6621,7 +6621,7 @@ struct llama_context * llama_new_context_with_model(
66216621
ctx->alloc = ggml_allocr_new(ctx->buf_alloc.data, ctx->buf_alloc.size, tensor_alignment);
66226622
#ifdef GGML_USE_METAL
66236623
if (ctx->ctx_metal) {
6624-
ggml_allocr_set_parse_seq(ctx->alloc, ggml_metal_get_concur_list(ctx->ctx_metal), ggml_metal_if_optimized(ctx->ctx_metal));
6624+
//ggml_allocr_set_parse_seq(ctx->alloc, ggml_metal_get_concur_list(ctx->ctx_metal), ggml_metal_if_optimized(ctx->ctx_metal));
66256625
}
66266626
#endif
66276627
#ifdef GGML_USE_CUBLAS

0 commit comments

Comments
 (0)