| --- published/pipeline/nvrtc_tce.py | |
| +++ portable/pipeline/nvrtc_tce.py | |
| use_cubin = _os.environ.get("NVRTC_CUBIN", "1") == "1" | |
| if use_cubin: | |
| arch = arch.replace(b"compute_", b"sm_") | |
| - opts = [b"--gpu-architecture=" + arch, b"-I/usr/local/cuda/include", | |
| + opts = [b"--gpu-architecture=" + arch, b"-I/tmp/kernelbench-h100-attempt2/deps/nvidia/cuda_runtime/include", b"-I/tmp/kernelbench-h100-attempt2/deps/nvidia/cuda_cccl/include", | |
| b"--std=c++17", b"-default-device"] | |
| arr = (ctypes.c_char_p * len(opts))(*opts) | |
| rc = nvrtc.nvrtcCompileProgram(prog, len(opts), arr) | |