ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-03-11 06:20:09 +00:00
Code Issues Packages Projects Releases Wiki Activity

cuda: set device to src device before p2p copy

Browse Source
This commit is contained in:
Kawrakow
2025-12-17 11:43:36 +00:00
parent d1dd45b4b9
commit 64908da772
1 changed files with 1 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
ggml/src/ggml-cuda.cu
View File

@@ -3451,6 +3451,7 @@ GGML_CALL static bool ggml_backend_cuda_cpy_tensor_async(ggml_backend_t backend_
needs_f16_f32_copy = true;
} else {
ggml_cuda_set_device(cuda_ctx_src->device);
CUDA_CHECK(cudaMemcpyPeerAsync(dst->data, cuda_ctx_dst->device, src->data, cuda_ctx_src->device, ggml_nbytes(dst), cuda_ctx_src->stream()));
}
#endif
@@ -3458,10 +3459,8 @@ GGML_CALL static bool ggml_backend_cuda_cpy_tensor_async(ggml_backend_t backend_
// record event on src stream after the copy
if (!cuda_ctx_src->copy_event) {
ggml_cuda_set_device(cuda_ctx_src->device);
CUDA_CHECK(cudaEventCreateWithFlags(&cuda_ctx_src->copy_event, cudaEventDisableTiming));
}
CUDA_CHECK(cudaEventRecord(cuda_ctx_src->copy_event, cuda_ctx_src->stream()));
// wait on dst stream for the copy to complete