Not sure how much use it will be, but AMD's GPUOpen initiative recently released a tool[1] that translates native CUDA code to portable C++ that can be recompil

Not sure how much use it will be, but AMD's GPUOpen initiative recently released a tool[1] that translates native CUDA code to portable C++ that can be recompiled into either CUDA or OpenCL. I haven't played with it at all so I can't verify how well it works. [1] https://github.com/GPUOpen-ProfessionalCompute-Tools/HIP https://github.com/GPUOpen-ProfessionalCompute-Tools/HIP