[EdDAzevedo]

Motivation

This PR fixes a race condition in SYEVJ GPU kernel. The problem showed up on gfx11 architecture.
The solution is to add calls to __syncthreads() to avoid race condition hazards in accessing arrays cosine_res[] and sines_diag[] in LDS shared memory.

This will address

JIRA ROCM-31219
https://amd-hub.atlassian.net/browse/ROCM-31219

Technical Details

The problem was initially identified by github user zpengmei in PR 9135 (now closed)

#9135

The issue was initially thought to be related to "[AMDGPU] Miscompiled post-barrier LDS reload in a large-workgroup HIP kernel on gfx11 (silent wrong results) #207883)

llvm/llvm-project#207883

Anshil Gandhi (angandhi@amd.com) has identified the source of the problem of race condition where synchronizations are required.

This PR adds the calls to __synchthreads() in the GPU kernel for SYEVJ.

Test Plan

Pass google tests for all routines that use the Jacobi method on all architectures, including gfx11.

Test Result

All tests passed.

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/TheRock/blob/main/GOVERNANCE.md#pull-requests.

[Copilot]

🟡 Changes recommended

Restore reference-result validation and add regression coverage for the affected gfx11 sizes.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Fixes a gfx11 SYEVJ GPU-kernel race by adding synchronization around LDS accesses. The change also removes validation checks from host LAPACK reference helpers.

Changes:

• Adds synchronization barriers for shared-memory reads and writes.
• Removes info == 0 assertions from reference routines.
• Requires regression coverage for affected matrix sizes.

File summaries

File	Summary
projects/rocsolver/library/src/lapack/roclapack_syevj_heevj.hpp	Adds barriers protecting SYEVJ/HEEVJ LDS accesses.
projects/rocsolver/clients/common/misc/lapack_host_reference.cpp	Removes host-reference LAPACK status checks.

Review details

Suppressed comments (4)

projects/rocsolver/clients/common/misc/lapack_host_reference.cpp:8651

• Removing these assertions from the sytf2/sytrf reference wrappers makes factorization failures (including singular pivots) invisible at the wrapper boundary. The associated tests can then compare results without an explicit reference-status check, weakening failure diagnostics; please preserve the check or return/handle info explicitly.

    ssytf2_(&uploC, &n, A, &lda, ipiv, info);

projects/rocsolver/clients/common/misc/lapack_host_reference.cpp:6331

• This removes the debug check that the sytrs2 reference solve succeeded. Without replacing it with explicit info handling, a failed CPU reference can be consumed as if it were valid and obscure the real test failure; please preserve the check or propagate the status to the caller.

    ssytrs2_(&uploC, &n, &nrhs, A, &lda, ipiv, B, &ldb, &(work[0]), &info);

projects/rocsolver/clients/common/misc/lapack_host_reference.cpp:8702

• This removes the debug-time check for sytrf's returned info while the wrapper still writes the factorization outputs. If the CPU factorization fails, the test can proceed with an invalid reference without any diagnostic; preserve validation or propagate and handle the status explicitly.

    ssytrf_(&uploC, &n, A, &lda, ipiv, work, &lwork, info);

projects/rocsolver/library/src/lapack/roclapack_syevj_heevj.hpp:178

• The check-in SYEVJ/HEEVJ parameterization still samples n=40 and n=67, so it does not exercise the gfx11 small-kernel window around n=48–58 that this synchronization change is intended to fix. Please add regression cases for the affected sizes (or otherwise add a focused test) so this barrier ordering is covered in CI.

    __syncthreads();

• Files reviewed: 2/2 changed files
• Comments generated: 1
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[therock-pr-bot]

✅ All Checks Passed — Ready for Review

Check	Status	Details
📝 PR Description	✅ Pass	—
⛔ Forbidden Files	✅ Pass	—
🧪 Unit Test	✅ Pass	—
🔎 pre-commit	✅ Pass	—
🚫 Draft PR	🔜 To Be Enabled	—
🚩 Feature Flag	🔜 To Be Enabled	—
📊 Code Coverage	🔜 To Be Enabled	—
🤖 therock-pr-bot	✅ Pass	—

🎉 All checks passed! This PR is ready for review.

📖 Need help? See the Policy FAQ for details on every check and how to fix failures.

🙋 Wish to Override Policy?

[therock-pr-bot]

🎉 All checks passed! This PR is ready for review.

[Copilot]

🟡 Changes recommended

The added size cases do not reliably reproduce the data-dependent, intermittent race.

Get a fresh assessment by requesting another Copilot review.

Review details

• Files reviewed: 4/4 changed files
• Comments generated: 1
• Review effort level: Balanced

[assistant-librarian]

TheRock Submodule Bump Activity

Newest first

• 2026-10-19 — Included in TheRock via ROCm/TheRock#9348.