I posted a thread couple days ago on the Max forum about the use of GPU as an audio processor. No reply so far, so it seems that the question of how to implement audio shaders is not to priority, at least not now. I am curious what are the prospects of using the GPU in more complex tasks inside the max/msp environment.
With the right implementation you could use the GPU for arbitrary calculation tasks of whatever data you want, provided they're in the right format and the operations are supported. So that could include audio samples too, like as a 2D matrix (maybe 4096 x 4096 samples or something), or higher dimensions, depending on what's most efficient.
Pretty interesting stuff with a TON of potential to extend processing power beyond the CPU... freeing it up for other things.