DCT/DST on the GPU

Hi,

There are some smoothing and filtering operations that I need to perform on large arrays that I would like to run on GPUs, but they need DCT and DST operations that don't seem to be available for gpuArray types. I know all flavors of each transform can be mimicked with FFTs, but doubling the size of my arrays to fit complex types limits the size of the data I can work with, and a deconvolution operation that needs to happen can't be done with convn without the N^2 runtime killing the overall performance.

I can probably work around the lack of a gpuArray-compatible type-I DCT, but there doesn't seem to be any gpu-compatible DST routine at all. I suppose this has to do with the lack of support for these transforms in cuFFT, but is there any possible workaround for this? Perhaps a different library out there could do the job?

Thanks,

Rodrigo