But running CUDA code has remained largely locked to Nvidia hardware. Now, an open-source project is working to break that barrier.