Repository navigation
CI: validate OpenBLAS and select a compatible kernel for ASE tests - #8348
Open
Stardust0831 wants to merge 2 commits into
Open
Stardust0831 wants to merge 2 commits into
Stardust0831 wants to merge 2 commits into
Conversation
Collaborator
|
That adds quite a few complexity. Would it be possible to simply choose the generic kernel for GitHub CI? Btw I'm surprised that such fault can be related to system OpenBLAS... |
This branch has not been deployed
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The GNU CI image's OpenBLAS 0.3.20 Cooperlake kernel returns incorrect DSYMM results. In the Gamma-only Si calculator and restart tests, this corrupts the density and aborts the fourth SCF iteration with a charge integral of
-17.6846andCan't find even an electron!.Check the OpenBLAS library linked to ABACUS before running the ASE tests. Preserve a working kernel selection; when the regression probe fails, select and verify an alternative supported by the CPU.
Linked Issue
Related: #7794.
What's changed?
interfaces/ASE_interface/tests/that resolves ABACUS's OpenBLAS dependency and compares 30 small DSYMM cases against scalar matrix products.1e-12before exportingOPENBLAS_CORETYPEthroughGITHUB_ENV.This changes the ASE CI environment. It does not modify the numerical engine, solver, scientific assertions, or reference data.
Unit Tests and/or Case Tests for my changes
Validation used upstream
eba09c937c73, built in the failing GNU image (sha256:b195875e7f3262d26eb4e07960afc0e7a193861699cd281eb08f65562880b17d), on an AMD Ryzen 9 9950X3D. Runtime tests usedOMP_NUM_THREADS=1,OPENBLAS_NUM_THREADS=1, andMKL_NUM_THREADS=1.Failure isolation: selecting Cooperlake reproduced both original Si failures in three repetitions, including the original CI's first three energies and fourth-step density diagnostics. Replacing only DSYMM with its scalar definition made both tests pass while other BLAS operations retained Cooperlake. The standalone DSYMM maximum error was
2.786for Cooperlake, versus at most4.44e-16for the passing kernels.Selector results:
Commands run inside the image, with
OPENBLAS_CORETYPE=CooperlakeandGITHUB_ENV=/evidence/selector-relocation/cooperlake.envset for the workflow step:The selector also preserved the healthy default without writing an override. The workflow working directory and subprocess invocation were exercised in the original image; the helper and policy-test contents match those used for the complete numerical validation.
With exactly the emitted environment applied to the following processes (commands from the repository root):
Staged governance and whitespace checks passed:
Validation limits: the AVX2 feature mask was exercised on the same physical AVX-512-capable host, not on separate AVX2-only hardware. Other architectures have policy-test coverage only. The probe covers this DSYMM regression, not every BLAS operation or shape. The original hosted logs do not record the selected OpenBLAS kernel, so its Cooperlake selection is inferred from the host and exact reproduced numerical signature. GitHub-hosted validation of the selector remains pending; a particular newer OpenBLAS release has not been established as the dependency fix.
Governance Notes
source/changes. The Python helpers are invoked directly by the workflow and need no CMake linkage.Reminder
AGENTS.mdanddocs/developers_guide/agent_governance.md.source/changes.