[CI] Optimize unit tests including adding stochasticity by hughperkins · Pull Request #709 · Genesis-Embodied-AI/quadrants

Skip the slowest tests by default

Hugh Perkins (deskai7)

committed

Trim n parametrize lists to {smallest, 12}

Hugh Perkins (deskai7)

committed

Mark more single-test outliers as slow (round 2)

Hugh Perkins (deskai7)

committed

Slow-mark long-tail test_tile16 outliers (round 3)

Hugh Perkins (deskai7)

committed

Revert function-level @pytest.mark.slow decorators

Hugh Perkins (deskai7)

committed

[Demo] cholesky_blocked: take N / N_ENVS / WARMUP / ITERS via argparse

Hugh Perkins (deskai7)

committed

[Test] test_tile16_cholesky_blocked_demo: invoke demo in smoke-mode

Hugh Perkins (deskai7)

committed

[Test] test_matmul_chain_qipc_sizes: parametrize on matrix shapes

Hugh Perkins (deskai7)

committed

[Test] test_gdar_mpm: parametrize on particles_side / n_grid / num_steps

Hugh Perkins (deskai7)

committed

[Test] test_device_{reduce,exclusive_scan}: fuse {add,min,max} into one op-parametrized test

Hugh Perkins (deskai7)

committed

[Style] black: reformat test_tile16_cholesky_blocked_demo cmd list + run_tests help string

Hugh Perkins (deskai7)

committed

[Test] test_subgroup_full_matches_tiled: fuse 20 thin subgroup-op wrappers into 2 op-parametrized tests

Hugh Perkins (deskai7)

committed

[Test] test_block_reduce{,_all}: fuse {add,min,max} into op-parametrized tests

Hugh Perkins (deskai7)

committed

[Test] test_block_inclusive: fuse {add,min,max} into one op-parametrized test

Hugh Perkins (deskai7)

committed

[Test] test_block_exclusive_minmax: fuse {min,max} into one op-parametrized test

Hugh Perkins (deskai7)

committed

[Test] @pytest.mark.sample: per-test stochastic parametrize subsampling

Hugh Perkins (deskai7)

committed

[Test] test_tile16: bump @pytest.mark.sample(n=4) -> n=6

Hugh Perkins (deskai7)

committed

[Style] Reflow @sample comment blocks to 120c (from AI default ~110c)

Hugh Perkins (deskai7)

committed

[BugFix] @sample: propagate seed to xdist workers via workerinput

Hugh Perkins (deskai7)

committed

[BugFix] @sample: propagate seed to xdist workers via QD_SAMPLE_SEED env var

Hugh Perkins (deskai7)

committed

[BugFix] @sample: pick seed in run_tests.py, propagate via --sample-seed argv

Hugh Perkins (deskai7)

committed

[DebugTemp] @sample: log seed/argv/workerinput in each pytest_collection_modifyitems call

Hugh Perkins (deskai7)

and

cursoragent

committed

[BugFix] @sample: sample from sorted(group) so xdist workers see identical order

Hugh Perkins (deskai7)

committed

[DebugTemp2] re-add sample debug log

Hugh Perkins (deskai7)

and

cursoragent

committed

[BugFix] @sample: derive per-test RNG via sha256, not tuple-seeding of random.Random

Hugh Perkins (deskai7)

committed

[Style] Reflow CI-flagged 80c-wrapped comments to 120c

Hugh Perkins (deskai7)

committed

[Doc] contributing.md: shorten testing bullet per PR review

Hugh Perkins (deskai7)

committed

[Doc] unit_testing.md: apply PR review feedback

Hugh Perkins (deskai7)

committed

Merge branch 'main' into hp/mark-slow-tests

hughperkins

authored

[Doc] address second round of PR review on unit_testing + contributing

Hugh Perkins (deskai7)

committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[CI] Optimize unit tests including adding stochasticity#709

[CI] Optimize unit tests including adding stochasticity#709
hughperkins wants to merge 30 commits into
mainfrom
hp/mark-slow-tests

Commits on May 19, 2026

Skip the slowest tests by default

Trim n parametrize lists to {smallest, 12}

Mark more single-test outliers as slow (round 2)

Slow-mark long-tail test_tile16 outliers (round 3)

Revert function-level @pytest.mark.slow decorators

[Demo] cholesky_blocked: take N / N_ENVS / WARMUP / ITERS via argparse

[Test] test_tile16_cholesky_blocked_demo: invoke demo in smoke-mode

[Test] test_matmul_chain_qipc_sizes: parametrize on matrix shapes

[Test] test_gdar_mpm: parametrize on particles_side / n_grid / num_steps

[Test] test_device_{reduce,exclusive_scan}: fuse {add,min,max} into one op-parametrized test

[Style] black: reformat test_tile16_cholesky_blocked_demo cmd list + run_tests help string

[Test] test_subgroup_full_matches_tiled: fuse 20 thin subgroup-op wrappers into 2 op-parametrized tests

[Test] test_block_reduce{,_all}: fuse {add,min,max} into op-parametrized tests

[Test] test_block_inclusive: fuse {add,min,max} into one op-parametrized test

[Test] test_block_exclusive_minmax: fuse {min,max} into one op-parametrized test

[Test] @pytest.mark.sample: per-test stochastic parametrize subsampling

[Test] test_tile16: bump @pytest.mark.sample(n=4) -> n=6

[Style] Reflow @sample comment blocks to 120c (from AI default ~110c)

[BugFix] @sample: propagate seed to xdist workers via workerinput

[BugFix] @sample: propagate seed to xdist workers via QD_SAMPLE_SEED env var

[BugFix] @sample: pick seed in run_tests.py, propagate via --sample-seed argv

[DebugTemp] @sample: log seed/argv/workerinput in each pytest_collection_modifyitems call

[BugFix] @sample: sample from sorted(group) so xdist workers see identical order

[DebugTemp2] re-add sample debug log

[BugFix] @sample: derive per-test RNG via sha256, not tuple-seeding of random.Random

[Style] Reflow CI-flagged 80c-wrapped comments to 120c

[Doc] contributing.md: shorten testing bullet per PR review

[Doc] unit_testing.md: apply PR review feedback

Merge branch 'main' into hp/mark-slow-tests

[Doc] address second round of PR review on unit_testing + contributing