[Tests] Guard TIRX CUDA tests by compute capability#19699
[Tests] Guard TIRX CUDA tests by compute capability#19699spectrometerHBH wants to merge 9 commits into
Conversation
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request decorates various CUDA-related tests with @tvm.testing.requires_cuda_compute_version to ensure they only run on compatible hardware (SM 90 or SM 100). A review comment correctly identifies that test_tma_cache_policy_operand_codegen is decorated with version 9, but actually requires version 10 because it uses a Blackwell-specific (SM 100) intrinsic.
Important
The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.
| assert "((unsigned long long)(&(A_map)))" in src | ||
|
|
||
|
|
||
| @tvm.testing.requires_cuda_compute_version(9) |
There was a problem hiding this comment.
The test test_tma_cache_policy_operand_codegen uses T.cuda.sm100_tma_2sm_mbarrier_addr (on line 349), which is a Blackwell (SM 100) specific intrinsic. Guarding it with compute version 9 (Hopper) will cause compilation or runtime failures on Hopper GPUs. It should be guarded by compute version 10 instead.
| @tvm.testing.requires_cuda_compute_version(9) | |
| @tvm.testing.requires_cuda_compute_version(10) |
PR apache#19699 (draft, being split by topic) carried a few optional dependency guards alongside its CUDA compute-capability work. Move them here where the optional-dependency gating lives: - support/test_popen_pool.py: skip the module without cloudpickle (converted from a pytestmark skipif to the module-level pytest.importorskip used throughout this PR; ruff's E402 exemption covers importorskip but not pytestmark assignments) - s_tir/meta_schedule/test_meta_schedule_builder.py: skip test_meta_schedule_missing_build_func without cloudpickle - s_tir/meta_schedule/test_meta_schedule_cost_model.py: requires_xgboost marker on the five XGBModel tests - codegen/test_target_codegen_blob.py: skip test_cuda_multi_lib without cloudpickle
1 participant
No description provided.