I created a common implementation, allowing OpenCL and HIP to be supported with the same code. Another benefit is that the OpenCL platform will now also compute the Pytorch model with CUDA or HIP if it's available, instead of always doing it on the CPU.
I created a common implementation, allowing OpenCL and HIP to be supported with the same code. Another benefit is that the OpenCL platform will now also compute the Pytorch model with CUDA or HIP if it's available, instead of always doing it on the CPU.