Hi, I'm developer of Alea.cuBase (http://www.quantalea.net), which is new CUDA tool in Microsoft new F# language. We developed a kernel DSL using F#, and generate LLVM and finally compiled into PTX by nvidia backend.

So now I'm seeking a way to make it possible to run PTX on other platforms, such as AMD, Intel Phi, etc. OpenCL is one way, but it need lot of work. I wonder if gpuocelot would help. In the backend, we use NVIDIA Driver API call, using .NET P/Invoke to call the nvcuda.dll (installed by driver) directly, so we bypass any stuff related to C language, even we don't need nvcc toolkit.

But I saw gpuocelot uses CUDA Runtime API. So my question are: 1) is it possible to provide a dll that simply replace the nvcuda.dll but with the same CUDA Driver API interface? 2) Will it support Intel Phi?

Regards, Xiang.