There's a team at UC-Berkeley which is working on compiling C++ to a combination of CPU/GPU/FPGA code right now, which might change things. I think you need a special architecture to take full advantage of it though.