OpenCL FFTs

I’m trying AMD’s clAmdFft and it works fine on AMD’s GPUs and Intel’s CPUs, but it fails with my NVIDIA 9800GT card.

I tried the AMD and the NVIDIA’s SDK.

http://devgurus.amd.com/message/1280952#1280952
http://devgurus.amd.com/message/1286564#1286564

It seems that NVIDIA has a problem in its runtime library. Can I expect a fix for this? Should we expect an OpenCL FFT library from NVIDIA?

Try the Apple OpenCL FFT Implementation.
http://developer.apple.com/library/mac/#samplecode/OpenCL_FFT/Introduction/Intro.html

I’ve heard good things, I’ve ran it a few times myself but have not benchmarked it.

I’m also interested in hearing if there are any others that work, and are written in OpenCL.

The Apple’s FFT doesn’t work with CPUs, and I had problems (artifacts) with AMD graphics cards.

I heard that Apple’s OpenCL FFT was made for NV G80 GPUs, maybe that’s why it doesn’t works well wigh ATI/AMD.

Are there any opencl fft for newer nvidia cards and/or amd cards? The fft is a too important library to be missing.