[FEA]: Run tests through `compute-sanitizer` in CI

alliepiper commented 5 months ago

To check for memory safety, the tests should be run through compute-sanitizer regularly in CI. Due to the expense, this might be best as a nightly/weekly check.

Work is ongoing in #1879.

### Related Tasks
- [ ] https://github.com/NVIDIA/cccl/issues/1911
- [ ] https://github.com/NVIDIA/cccl/issues/1017
- [ ] https://github.com/NVIDIA/cccl/issues/1887
- [ ] https://github.com/NVIDIA/cccl/issues/1888
- [ ] https://github.com/NVIDIA/cccl/issues/1889
- [ ] #1890
- [ ] https://github.com/NVIDIA/cccl/issues/1891
- [ ] #1892
- [ ] #1893
- [ ] #1894
- [ ] #1895
- [ ] #1896
- [ ] #1897
- [ ] #1899
- [ ] https://github.com/NVIDIA/cccl/issues/1917
- [ ] #1898
- [ ] #1902
- [ ] #1903
- [ ] #1906
- [ ] #1907
- [ ] #1909
- [ ] #1910
- [ ] https://github.com/NVIDIA/cccl/issues/1790

jrhemstad commented 5 months ago

What I've envisioned with adding nightly/weekly testing is to start by expanding the matrix.yaml file to add top-level nightly: and weekly: matrices alongside the existing pull_request: matrix.

lilohuang commented 3 months ago

Some of them might be harmless. However, it would be great to find a way to appease the compute-sanitizer so that CUDA programmers can trust the NVIDIA compute-sanitizer to identify any potential bugs while using the thrust/cub library. Thank you so much. 👍

fbusato commented 1 month ago

@alliepiper FYI I opened another issue for host-side sanitizers https://github.com/NVIDIA/cccl/issues/2241

NVIDIA / cccl

[FEA]: Run tests through `compute-sanitizer` in CI #1618