On 20/01/14 04:34, Andreas Kloeckner wrote: > OIC. That said, I'd suspect that the perf gain from the page-locked > transfer is likely higher than from the persistent MPI request, but I > might of course be wrong.
Just to further confuse the issue I was looking through the NVIDIA OpenCL best practices guide [1] where this topic is discussed (p9). Nowhere in their recommended solution do they mention um-mapping the buffer. It therefore seems as if what you proposed is -- in practice -- the best solution. Regards, Freddie. [1] <http://www.nvidia.com/content/cudazone/CUDABrowser/downloads/papers/NVIDIA_OpenCL_BestPracticesGuide.pdf>
signature.asc
Description: OpenPGP digital signature
_______________________________________________ PyOpenCL mailing list [email protected] http://lists.tiker.net/listinfo/pyopencl
