cuda : fix device sync on buffer clear (llama/14033)

This commit is contained in:
Diego Devesa
2025-06-09 07:36:26 -07:00
committed by Georgi Gerganov
parent 26282282fa
commit e16a84cd95

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_cuda_set_device(ctx->device);
CUDA_CHECK(cudaDeviceSynchronize());
CUDA_CHECK(cudaMemset(ctx->dev_ptr, value, buffer->size));
CUDA_CHECK(cudaDeviceSynchronize());
CUDA_CHECK(cudaMemsetAsync(ctx->dev_ptr, value, buffer->size, cudaStreamPerThread));
CUDA_CHECK(cudaStreamSynchronize(cudaStreamPerThread));
}
static const ggml_backend_buffer_i ggml_backend_cuda_buffer_interface = {