|
|
@@ -55,7 +55,11 @@ cudaStreamSynchronize(starpu_cuda_get_local_stream());
|
|
|
Unfortunately, some CUDA libraries do not have stream variants of
|
|
|
kernels. That will lower the potential for overlapping.
|
|
|
|
|
|
-StarPU already does appropriate calls for the CUBLAS library.
|
|
|
+Calling starpu_cublas_init() makes StarPU already do appropriate calls for the
|
|
|
+CUBLAS library. Some libraries like Magma may however change the current stream,
|
|
|
+one then has to call cublasSetKernelStream(starpu_cuda_get_local_stream()); at
|
|
|
+the beginning of the codelet to make sure that CUBLAS is really using the proper
|
|
|
+stream.
|
|
|
|
|
|
If the kernel can be made to only use this local stream or other self-allocated
|
|
|
streams, i.e. the whole kernel submission can be made asynchronous, then
|
