sdgoij/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-05-13 20:44:09 +00:00
Code Issues Packages Projects Releases Wiki Activity

Minor cleanup

Browse Source
This commit is contained in:
aendk
2026-01-09 17:07:19 +01:00
committed by Georgi Gerganov
parent d776354dc9
commit b7376c3ed7
2 changed files with 2 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
ggml/src/ggml-backend.cpp
View File

@@ -769,12 +769,6 @@ static bool ggml_backend_implicitly_synced(ggml_backend_t current_backend) {
* For these backends, we can skip costly explicit synchronizations during compute split scheduling.
*/
static bool disable_scheduler_sync_opt = (getenv("GGML_SCHED_DISABLE_SYNC_OPT") != nullptr);
if (disable_scheduler_sync_opt) {
return false;
}
// To not change any APIs or change what ggml-base links to, we can only detect backends by string matching
auto backend_name = ggml_backend_name(current_backend);
if (strncmp(backend_name, "CUDA", 4) == 0) {

6
ggml/src/ggml-cuda/ggml-cuda.cu
View File

@@ -2768,7 +2768,8 @@ static bool ggml_backend_cuda_cpy_tensor_async(ggml_backend_t backend_src, ggml_
ggml_backend_cuda_buffer_context * buf_ctx_src = (ggml_backend_cuda_buffer_context *) buf_src->context;
ggml_backend_cuda_buffer_context * buf_ctx_dst = (ggml_backend_cuda_buffer_context *) buf_dst->context;
if (!copy_from_host && (cuda_ctx_src->device != buf_ctx_src->device || cuda_ctx_dst->device != buf_ctx_dst->device)) {
if ((copy_from_host && cuda_ctx_dst->device != buf_ctx_dst->device) ||
!copy_from_host && (cuda_ctx_src->device != buf_ctx_src->device || cuda_ctx_dst->device != buf_ctx_dst->device)) {
#ifndef NDEBUG
GGML_LOG_DEBUG("%s: backend and buffer devices do not match\n", __func__);
#endif // NDEBUG
@@ -2776,9 +2777,6 @@ static bool ggml_backend_cuda_cpy_tensor_async(ggml_backend_t backend_src, ggml_
}
if (copy_from_host) {
if (!cuda_ctx_dst->stream()) {
return false;
}
CUDA_CHECK(cudaMemcpyAsync(dst->data, src->data, ggml_nbytes(dst), cudaMemcpyHostToDevice, cuda_ctx_dst->stream()));
} else if (backend_src != backend_dst) {
// copy on src stream