René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"[BUG]: cuda.core tests fail with torch cu126 installed (NVRTC 12.6 options, torch CAI v2)","articleBody":"### Is this a duplicate?\n\n- [x] I confirmed there appear to be no [duplicate issues](https://github.com/NVIDIA/cuda-python/issues) for this bug and that I agree to the [Code of Conduct](CODE_OF_CONDUCT.md)\n\n### Type of Bug\n\nSomething else\n\n### Component\n\ncuda.core\n\n### Describe the bug\n\nWith cuda-core 1.2.2 `[cu12]` and torch 2.14.1+cu126 in the same environment, `cuda_core/tests` has 10 failures that come from the tests, not from cuda.core. The same parametrizations are unchanged on `main`.\n\n1. `test_program.py::test_cpp_program_with_various_options`: `ProgramOptions(no_cache=True)`, `ProgramOptions(frandom_seed=\"12345\")` and `ProgramOptions(ofast_compile=\"min\")` fail with `NVRTC_ERROR_INVALID_OPTION`. torch cu126 pins `nvidia-cuda-nvrtc-cu12==12.6.85`, and these options need NVRTC 12.9. Compiling an empty kernel with each option through `cuda.bindings.nvrtc`:\n\n   | Option | NVRTC 12.6.85 | 12.8.93 | 12.9.86 |\n   | --- | --- | --- | --- |\n   | `--no-cache` | unrecognized option | unrecognized option | OK |\n   | `--frandom-seed=12345` | unrecognized option | unrecognized option | OK |\n   | `--Ofast-compile=min` | invalid argument `min` | invalid argument `min` | OK |\n   | `--pch` | unrecognized option | OK | OK |\n\n   The PCH parametrizations in the same list are already skipped below NVRTC 12.8; these three have no version gate.\n\n2. `test_utils.py::TestViewCudaArrayInterfaceGPU::test_cuda_array_interface_gpu[torch-*]` (7 cases) fail with `BufferError: only CUDA Array Interface v3 or above is supported`. torch still exports `__cuda_array_interface__` with `\"version\": 2`, so `StridedMemoryView.from_cuda_array_interface` rejects every torch sample that `_gpu_array_samples()` adds. This test is listed in #2077 because CI has no array library installed for it; with torch installed it does run and these samples fail.\n\n### How to Reproduce\n\n1. `pip install \"cuda-core[cu12]==1.2.2\" torch --index-url ...` (torch 2.14.1+cu126), plus the test dependencies.\n2. Run `pytest cuda_core/tests/test_program.py cuda_core/tests/test_utils.py` from the `cuda-core-v1.2.2` sources on a CUDA GPU.\n\n### Expected behavior\n\nThe three option cases skip on NVRTC older than 12.9, like the PCH cases skip below 12.8. The CAI test either leaves out torch samples or expects the `BufferError` for CAI v2 producers.\n\n### Operating System\n\nUbuntu 24.04, x86_64\n\n### nvidia-smi output\n\nTesla V100-SXM2-16GB, driver 580 (CUDA 13.0). The failures are not GPU-specific.\n","author":{"url":"https://github.com/0z5a","@type":"Person","name":"0z5a"},"datePublished":"2026-11-02T19:42:26Z","url":"https://github.com/NVIDIA/cuda-python/issues/3333"}