Hi, https://github.com/NVIDIA/cuda-tile/blob/8a775693b18303d6c696be6ffd06dadad1b32a8e/cmake/IncludeLLVM.cmake#L26-L27 It would be more convenient to enable CACHE for these two macros. Regards, Peter Han