ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-30 11:21:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

Make split mode graph work with vision enabled (#1392)

Browse Source
This commit is contained in:
Kawrakow
2026-03-10 06:56:39 +01:00
committed by GitHub
parent 666ea0e983
commit 14492bfdd2
1 changed files with 2 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -512,8 +512,9 @@ ggml_backend_cuda_context::ggml_backend_cuda_context(int device) :
auto info = const_cast<ggml_cuda_device_info*>(&ggml_cuda_info()); auto info = const_cast<ggml_cuda_device_info*>(&ggml_cuda_info());
if (info->all_ctx[device]) { if (info->all_ctx[device]) {
GGML_CUDA_LOG_WARN("%s: a context for device %d already exists?\n", __func__, device); GGML_CUDA_LOG_WARN("%s: a context for device %d already exists?\n", __func__, device);
} else{
info->all_ctx[device] = this;
} }
info->all_ctx[device] = this;
} }
ggml_backend_cuda_context::~ggml_backend_cuda_context() { ggml_backend_cuda_context::~ggml_backend_cuda_context() {
@@ -527,9 +528,6 @@ ggml_backend_cuda_context::~ggml_backend_cuda_context() {
std::unique_lock<std::mutex> lock(ggml_cuda_lock); std::unique_lock<std::mutex> lock(ggml_cuda_lock);
ggml_cuda_lock_cv.wait(lock, []{ return ggml_cuda_lock_counter == 0; }); ggml_cuda_lock_cv.wait(lock, []{ return ggml_cuda_lock_counter == 0; });
auto info = const_cast<ggml_cuda_device_info*>(&ggml_cuda_info());
info->all_ctx[this->device] = nullptr;
if (copy_event != nullptr) { if (copy_event != nullptr) {
CUDA_CHECK(cudaEventDestroy(copy_event)); CUDA_CHECK(cudaEventDestroy(copy_event));
} }