Jean-Matthieu Etancelin <[email protected]> writes: > Yes, thats perfect ! When the Buffer is mapped, the async copy is really fast. > So it comes an other question: > - What are the limitations of using the Buffer in kernels while it is mapped, > is it necessary to release the buffer before ?
I wouldn't advise using the buffer with ALLOC_HOST_PTR as storage to back a computation--that'll be pretty slow. The runtime will either copy-in and copy-out behind your back or employ some type of memory mapping over PCIe. Neither of these will match the bandwidth and latency of native GPU memory. So I would use two buffers--one on the host with ALLOC_HOST_PTR and one without any flags (that will wind up sitting on the GPU). Andreas
pgpk9qq4otNwo.pgp
Description: PGP signature
_______________________________________________ PyOpenCL mailing list [email protected] http://lists.tiker.net/listinfo/pyopencl
