Fix 1D multi-rank MPI_GATHERV bug in post-process silo output#1138
Open
sbryngelson wants to merge 2 commits intoMFlowCode:masterfrom
Open
Fix 1D multi-rank MPI_GATHERV bug in post-process silo output#1138sbryngelson wants to merge 2 commits intoMFlowCode:masterfrom
sbryngelson wants to merge 2 commits intoMFlowCode:masterfrom
Conversation
Contributor
|
CodeAnt AI is reviewing your PR. Thanks for using CodeAnt! 🎉We're free for open-source projects. if you're enjoying it, help us grow by sharing. Share on X · |
Contributor
PR Reviewer Guide 🔍Here are some key observations to aid the review process:
|
Contributor
PR Code Suggestions ✨No code suggestions found for the PR. |
Contributor
📝 WalkthroughWalkthroughAdds 1D-specific MPI gather handling to spatial/data extent collection in Changes
Sequence Diagram(s)sequenceDiagram
rect rgba(200,200,255,0.5)
participant Rank_i as Non-root Rank
end
rect rgba(180,255,200,0.5)
participant Root as Root Rank (proc_rank==0)
end
rect rgba(255,220,180,0.5)
participant MPI as MPI Library
end
Note over Rank_i,Root: 1D path (n==0 && grid_geometry!=3) flow
Rank_i->>MPI: MPI_GATHER(ext_scalar -> ext_temp[rank_index])
MPI->>Root: deliver ext_temp array
Root->>Root: assemble spatial_extents/data_extents from ext_temp per rank
Note over Rank_i,Root: Multi-D path (n>0 or grid_geometry==3) flow
Rank_i->>MPI: MPI_GATHERV(local_extent, recvcounts, displs)
MPI->>Root: deliver gathered arrays
Root->>Root: assign spatial_extents/data_extents from gathered arrays
Estimated code review effort🎯 4 (Complex) | ⏱️ ~45 minutes Poem
🚥 Pre-merge checks | ✅ 3 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (3 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing touches
🧪 Generate unit tests (beta)
⚔️ Resolve merge conflicts (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
codeant-ai
bot
added
the
size:XS
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Pull request overview
This PR updates a comment in the post-processing data output module (m_data_output) to slightly simplify the description of the m_global_parameters import.
Changes:
- Updated the inline comment on
use m_global_parametersfrom “Global parameters for the code” to “Global parameters”.
Contributor
Nitpicks 🔍
|
Contributor
|
CodeAnt AI finished reviewing your PR. |
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #1138 +/- ##
==========================================
- Coverage 44.03% 44.02% -0.01%
==========================================
Files 70 70
Lines 20649 20658 +9
Branches 2053 2059 +6
==========================================
+ Hits 9093 9095 +2
- Misses 10368 10373 +5
- Partials 1188 1190 +2 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
sbryngelson
changed the title
The 1D paths in s_mpi_gather_spatial_extents and s_mpi_gather_data_extents reused recvcounts/displs arrays sized for grid defragmentation (m+1 per rank), but each rank only sends 1 scalar value. This sendcount/recvcounts mismatch is undefined behavior per the MPI standard and caused nondeterministic crashes with Intel MPI, preventing silo files from being written. Replace MPI_GATHERV with MPI_GATHER + temp buffer for the 1D case. Multi-D paths and 1D defragmentation functions are unchanged. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
![qodo-code-review[bot]](https://avatars.githubusercontent.com/in/484649?s=60&v=4){kind=small}
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.
Summary
s_mpi_gather_spatial_extentsands_mpi_gather_data_extentsinsrc/post_process/m_mpi_proxy.fppfor 1D multi-rank runsrecvcounts/displsarrays are sized for grid defragmentation (m+1per rank) but were being reused for scalar extent gathers where each rank sends exactly 1 element — a sendcount/recvcounts mismatch that is undefined behavior per the MPI standard4*displsinstead of2*displs, causing buffer overflows on the root processMPI_GATHERVwithMPI_GATHER+ temp buffer for the 1D case, avoiding the mismatched shared arrays entirelyBug details
The
1D -> 2 MPI Rankstest (UUID0FCCE9F1) was failing nondeterministically on the Intel no-debug CI job (Github (ubuntu, mpi, no-debug, true)) with:This was a latent bug exposed when the Intel oneAPI packages were unpinned (from
intel-oneapi-mpi-2021.7.1to latest). Newer Intel MPI versions are stricter about theMPI_GATHERVsendcount/recvcounts mismatch, causing crashes that prevent silo files from being written. The bug was masked with GCC/OpenMPI (which tolerates the mismatch) and in debug mode (which changes stack/heap layout).Test plan
1D -> 2 MPI Rankstest passes on Intel no-debug CISummary by CodeRabbit
Refactor
Chores