Fix composed-graph cudaErrorMisalignedAddress at 4D production shapes and CI OOM on lcov install#269
Merged
Conversation
…tion shapes — root-cause planner/executor bug, fix, add regression tests
The bare 'apt-get install -y lcov' pulls 78 recommended packages (fontconfig, fonts-dejavu, libgd, etc.), exhausting runner memory. The runner receives SIGKILL during unpack before any test runs. Adding --no-install-recommends limits the install to lcov and its hard dependencies only.
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #269 +/- ##
==========================================
+ Coverage 85.70% 85.78% +0.08%
==========================================
Files 129 129
Lines 6457 6495 +38
==========================================
+ Hits 5534 5572 +38
Misses 923 923 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
chhwang
changed the title
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fix composed-graph cudaErrorMisalignedAddress at 4D shapes; fix CI OOM
Two bugs in the planner/executor caused
cudaErrorMisalignedAddressorsilent data corruption when composing ops on 4D tensors where H > W
(e.g.,
(1,4,128,32)).Bug 1 — Cast
_InShape::W == 1branch (cast.h:23): scalar pathprocessed 1 element while
NelemPerThread = 2, skipping every otherH element. Triggered at shapes like
(1,4,128,1).Bug 2 — Default tile
(1, 64)for Cast and RoPE: at(1,4,128,32)(H=128, W=32), the tile spans 2× W, forcing H-consecutive access that
misaligns addresses.
Fix: remove the dead
W==1scalar branch incast.h; adddefault_configoverrides inops_cast.cppandops_rope.cppselectingtile
(32,2)when H > W and W ≥ 2. Add 23 regression tests intest_composed_graph_shapes.pycovering rope, composed rmsnorm, andcomposed silu·gate at the exact production shapes that crashed.
CI fix: add
--no-install-recommendstoapt-get install lcovinut.yml. The bare install pulled 78 recommended packages (fontconfig,fonts-dejavu, etc.), causing OOM/SIGKILL on the self-hosted runner.
Files changed:
ark/include/kernels/cast.h— remove dead W==1 branchark/ops/ops_cast.{cpp,hpp}— add default_config for H > Wark/ops/ops_rope.{cpp,hpp}— add default_config for H > Wpython/unittest/ops/test_composed_graph_shapes.py— 23 regression tests.github/workflows/ut.yml— --no-install-recommends for lcov