Indeed, you want to get clinfo to report more that 0 platforms/devices.
You need an OpenCL implementation even for the CPUs. I seem to remember that Intel offers one for download (for Intel CPUs) but I don't have a link. OTOH I don't expect OpenCL on CPUs to be actually useful other than for trying out/debugging.
For an Nvidia GPU, it is my impression that the CUDA *toolkit* brings the OpenCL driver. (I don't have experience with Nouveau)
For an AMD GPU, one may use either ROCm, or amdgpu-pro, which both have different OpenCL compilers.
So the first step is to get clinfo to recognize something.
Afterwards when starting programming, you may have a look at clwrap.h in gpuowl source code for some utils like compiling a kernel etc.
|