spectrometerHBH opened a new pull request, #20154: URL: https://github.com/apache/tvm/pull/20154
Two independent fixes found while running the TIRx kernel registry correctness suite on sm100. **Version-gate the CUDA 12.8 tensor-map enums.** `CU_TENSOR_MAP_SWIZZLE_128B_ATOM_32B` and the other CUDA 12.8 additions are enumerators, not macros, so every `#ifdef` guard around them in `cuda_device_api.cc` was always false: the swizzle/dtype variants they gate were unconditionally rejected, and requesting a `128B_ATOM_32B` swizzle failed host validation with `Unsupported swizzle enum value: 4` even on CUDA 13.2. The guards are now `CUDA_VERSION >= 12080` checks, and the dead guards around the base-enum `TFLOAT32`/`FLOAT32_FTZ` members (present wherever the tensor-map API exists) are dropped. **Harden the registry-test GPU locking.** The per-device flock release ran after `torch.cuda.empty_cache()` in the same `finally`; when a kernel fault poisons the CUDA context, `empty_cache()` raises, and pytest's saved traceback keeps the frame (and the lock file descriptors) alive — the held flock never releases, and the same worker deadlocks against itself when its next test reopens the lock file. The release now sits in a nested `finally`. Also skip `sm100_fp8_fp4_mega_moe` in this suite: its dedicated multi-process scheduler validates physical-device assignments that reject any process already owning a CUDA context, which every process in this suite has by the time the free-memory probe has run. Verified on sm100: the two `gdn_cp_prefill_sm100` configs that requested the `128B_ATOM_32B` swizzle now pass, the MegaMoE configs skip, and the previously deadlocking registry suite completes. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
