happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Relax requirement of checks in async CUDA copies from backend and buffer type to just buffer type, to avoid linking issues

This commit is contained in:
aendk 2025-12-19 11:30:03 +01:00
parent 9171aca782
commit 1233fdda5f
1 changed files with 5 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
ggml/src/ggml-cuda/ggml-cuda.cu
View File

@ -59,7 +59,6 @@
#include "ggml-cuda/cumsum.cuh"
#include "ggml-cuda/fill.cuh"
#include "ggml.h"
#include "ggml-cpu.h"
#include <algorithm>
#include <array>
@ -2792,13 +2791,13 @@ static bool ggml_backend_cuda_cpy_tensor_async(ggml_backend_t backend_src, ggml_
ggml_backend_buffer_t buf_dst = dst->view_src ? dst->view_src->buffer : dst->buffer;
//enables async copies from CPU to CUDA, instead of only CUDA-to-CUDA
bool copy_from_cpu = ggml_backend_is_cpu(backend_src) && ggml_backend_buffer_is_host(src->buffer);
bool copy_from_host = ggml_backend_buffer_is_host(src->buffer);
if (!(copy_from_cpu || ggml_backend_is_cuda(backend_src)) || !ggml_backend_is_cuda(backend_dst)) {
if (!(copy_from_host || ggml_backend_is_cuda(backend_src)) || !ggml_backend_is_cuda(backend_dst)) {
return false;
}
if (!(copy_from_cpu || ggml_backend_buffer_is_cuda(src->buffer)) || !ggml_backend_buffer_is_cuda(dst->buffer)) {
if (!(copy_from_host || ggml_backend_buffer_is_cuda(src->buffer)) || !ggml_backend_buffer_is_cuda(dst->buffer)) {
return false;
}
@ -2809,14 +2808,14 @@ 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_cpu && (cuda_ctx_src->device != buf_ctx_src->device || cuda_ctx_dst->device != buf_ctx_dst->device)) {
if (!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
return false;
}
if (copy_from_cpu) {
if (copy_from_host) {
if (!cuda_ctx_dst->stream()) {
return false;
}