Well, I'm all for a low level approach. But going a little higher allows the GPU to handle a lot of things. Or do you propose to incorporate something like CUDA/OpenCL into WASM?