Cuda's future

I’m just thinking about how Cuda will evolve.
Does it have any way to became a cross-platform tool to enable stream processing powered software on both GPUs, CPUs (MMX, SSE, etc…) and mixes of them (IBM’s Cell) ?
Something like Rapidmind but more open and free, wouldn’t it be cool? What do you think about it?

I think OpenMP is more like something you’re looking for. CUDA is meant as a GPU language, not as an general model for any kind of parallel system.

OpenMP can’t run on GPUs. I’m looking for something like Rapidmin but open source…