CUDA-compatible replacement layer for non-NVIDIA GPUs that runs unmodified CUDA applications with near-native performance as a general GPU compute runtime library.