On Wed, 5 Oct 2011 01:42:58 +0500, Ziyad Saeed <[email protected]> wrote: > AMD released Aparapi project that allows Java developers to use GPU and does > it in a wonderful way. > > First you don't right C code at all. It converts the java code at runtime to > be able to run on the GPU. PyOpenCL does that as well via meta programming. > check
PyOpenCL does not convert Python code. It can execute some types of numpy code unchanged. Aside from that it > Second it is smart enough to adapt itself from a single code whether > to execute on a OpenCL device or not. You never write two separate > codes. If an OpenCL device is detected it uses it, if not then uses > java thread pool. Essentially taking all the headache away. It > automatically covers all your bases with a single code base. I'd consider this a non-issue since very capable (and free) OpenCL implementations for the CPU exist that do not impose any hardware requirements at all. In addition, not requiring an OpenCL implementation and still offering all of PyOpenCL's current features would amount to implementing OpenCL--which is clearly outside the scope of PyOpenCL. The project you describe might be realizable *using* PyOpenCL, but it is not PyOpenCL itself. > Can PyOpenCL adapt itself if OpenCL device isn't detected. No, by design. Hope this clarifies things. :) Andreas
pgpGSVvMgka7c.pgp
Description: PGP signature
_______________________________________________ PyOpenCL mailing list [email protected] http://lists.tiker.net/listinfo/pyopencl
