vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

cuda : fix split buffer free

Browse source
This commit is contained in:
slaren 2024-01-10 21:31:28 +01:00
parent 74066f8c41
commit c522c112b3
1 changed files with 6 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
ggml-cuda.cu
View file

@ -10092,10 +10092,14 @@ struct ggml_backend_cuda_split_buffer_context {
for (ggml_tensor_extra_gpu * extra : tensor_extras) {
for (int id = 0; id < g_device_count; ++id) {
for (int64_t is = 0; is < MAX_STREAMS; ++is) {
if (extra->events[id][is] != nullptr) {
CUDA_CHECK(cudaEventDestroy(extra->events[id][is]));
}
}
if (extra->data_device[id] != nullptr) {
CUDA_CHECK(cudaFree(extra->data_device[id]));
}
}
delete extra;
}
}