Just as a side-note, and although this does not seem feasible on the RPi
(at least yet), anybody generally interested in audio DSP on GPUs may want
to have a look at:
Savoija, Lauri, Vesa Valimaki, and Julius O. Smith. "Audio Signal
Processing Using Graphics Processing Units." *Journal of the Audio
Engineering Society* 59.1-2 (2011): 3-19.
In my experience, doing scientific computations by tricking a GPU into
thinking that it's processing graphics (e.g. by using OpenGL), is quite
painful. So in the absence of an official framework like OpenCL or CUDA,
which provides a relatively simple means of abusing a GPU, chasing this one
on the RPi would be a nightmare...
Cheers,
Jon.