ver4a/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

cuda : fix device sync on buffer clear (#14033)

Browse source
This commit is contained in:
Diego Devesa 2025-06-09 07:36:26 -07:00 • committed by GitHub
parent 201b31dc2e
commit 8f47e25f56
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 2 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

5
ggml/src/ggml-cuda/ggml-cuda.cu
View file

@ -615,9 +615,8 @@ static void ggml_backend_cuda_buffer_clear(ggml_backend_buffer_t buffer, uint8_t
ggml_backend_cuda_buffer_context * ctx = (ggml_backend_cuda_buffer_context *)buffer->context; ggml_backend_cuda_buffer_context * ctx = (ggml_backend_cuda_buffer_context *)buffer->context;
ggml_cuda_set_device(ctx->device); ggml_cuda_set_device(ctx->device);
CUDA_CHECK(cudaDeviceSynchronize()); CUDA_CHECK(cudaMemsetAsync(ctx->dev_ptr, value, buffer->size, cudaStreamPerThread));
CUDA_CHECK(cudaMemset(ctx->dev_ptr, value, buffer->size)); CUDA_CHECK(cudaStreamSynchronize(cudaStreamPerThread));
CUDA_CHECK(cudaDeviceSynchronize());
} }
static const ggml_backend_buffer_i ggml_backend_cuda_buffer_interface = { static const ggml_backend_buffer_i ggml_backend_cuda_buffer_interface = {