Re: OGR-27 OpenCL core write attempt
Jeff Lawson <[email protected]> Wed, 15 Aug 2012 10:15:01 -0500
| Newsgroups | gmane.comp.distributed-net.hardware |
|---|---|
| Message-ID | <CAAEc9z-RzGAXy6MJVMhQz5j4H=GhMPNy=AqGV1RBwfq8dqawbw@mail.gmail.com> |
It is exciting that you are also looking at OpenCL for OGR, as we are very close to internally making available OpenCL cores for RC5-72. Unfortunately, I don't think it will be possible for you to eliminate the choose data tables since they are intrinsic to the OGR algorithm and the parameters that we have selected for the project. What performance are you currently seeing on your OpenCL core relative to the CPU core for OGR? On Tue, Aug 14, 2012 at 12:17 PM, Konrad Rzepecki <[email protected]> wrote: > Hi > > I try to write OpenCL OGR-27 core (see attachment), but I have encounter > problem with choose() macro. This is in fact some big table generated in way I > don't understood. > > So I have question. > Can I replace it some way, or make some static pregenerated table for ALL > working packets? Note that memory access is costly on GPU, so little more > counting can be preferable over memory read. > > > PS. I'm new to OGR & OpenCL coding. > > -- > Konrad Rzepecki (Hannibal) > _______________________________________________ > Hardware mailing list > Hardware-Ra3b/QYEcJ3d140v2zMXi0fjHoOT/h/[email protected] > http://lists.distributed.net/mailman/listinfo/hardware > _______________________________________________ Hardware mailing list Hardware-Ra3b/QYEcJ3d140v2zMXi0fjHoOT/h/[email protected] http://lists.distributed.net/mailman/listinfo/hardware