A (very basic) OpenCL-to-CUDA compiler/runtime API. Requires access to the OpenCL code at compile-time, and the generated CUDA must be linked into the final program.
Supports only a very limited subset of OpenCL excluding e.g. textures but including fine-grained shared virtual memory.
NOTE: This project is NOT actively maintained.