On Sun, Sep 4, 2011 at 10:13 PM, John <[email protected]> wrote: > Hi Jean, > I am still getting familiar with GPGPU and OpenCl, I have a quick question : > the python binding you use will generate the OpenCl kernels? or it just > provides APIs to execute the pre written OpenCl kernels?
Hi John, The pyopencl package only provides an API to execute OpenCL kernels. The kernels can be written directly in C99. They can also be generated from Python using codepy (using codepy.cgen.opencl) but IMO it's simpler and more efficient to write them directly in C99. > Do you have the kernels written in C for each type of parallel operations > that you support? Working on it... Like I said in the initial post to this thread, the current Python component implementations are simple independent functions which are easy to translate to C, C++, Java etc. I'm translating them to C99 with some algorithm tweaks to leverage parallelism as much as possible. I've made good progress during this long weekend but it's probably going to take me 1 or 2 more weekends to get something running end to end. I'll update this mailing list as I progress. -- Jean-Sebastien
