Hello all,
I would like to process some data, first with OpenCL and then use CUFFT to do an FFT. Is there any way of converting a cl_mem object (buffer or image) to a cuda memory object without doing it myself on the CPU side?
I know of no direct way to do this, and have never tried it, but it may be possible to create an OpenGL buffer object that is first mapped to OpenCL and then released and re-mapped using CUDA calls.
I know of no direct way to do this, and have never tried it, but it may be possible to create an OpenGL buffer object that is first mapped to OpenCL and then released and re-mapped using CUDA calls.
You can run a simple kernel that saves the buffer’s pointer address into a buffer (make sure you know whether its 32 or 64-bit) and then read that back. I’m pretty sure the addresses are the same in CUDA and OpenCL if you’re using NVidia’s OpenCL driver. Might be a bit slow if you have to do it a lot.
I would be interesting to know how this turned out. LFdS, did you get any OpenCL/CUDA interoperability? Have anyone else tried this?