Skip to content

cmake: replace -fopenmp-target-fast with -O3 for LLVMFlang AMD GPU builds#1450

Merged
sbryngelson merged 1 commit into
MFlowCode:masterfrom
sbryngelson:amd-fixes
May 20, 2026
Merged

cmake: replace -fopenmp-target-fast with -O3 for LLVMFlang AMD GPU builds#1450
sbryngelson merged 1 commit into
MFlowCode:masterfrom
sbryngelson:amd-fixes

Conversation

@sbryngelson
Copy link
Copy Markdown
Member

@sbryngelson sbryngelson commented May 19, 2026

Summary

Replaces -fopenmp-target-fast with -O3 in the LLVMFlang OpenMP target offload compile flags.

-fopenmp-target-fast is present in the minimal reproducer for the array constructor corruption bug reported in #1449. Replacing it with explicit -O3 preserves optimization level while avoiding the flag combination that triggers silent wrong values in GPU_PARALLEL_LOOP regions on AMD MI250X (gfx90a).

Closes #1449

Test plan

  • Build with ./mfc.sh build --gpu mp -j 8 on Frontier using AMD flang
  • Run IBM test cases (./mfc.sh test --only IBM -j 8) and confirm ghost point corrections are correct
  • Run full test suite on Frontier to check for regressions

@MFlowCode MFlowCode deleted a comment from qodo-code-review Bot May 19, 2026
@codecov
Copy link
Copy Markdown

codecov Bot commented May 20, 2026

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 61.31%. Comparing base (8e34431) to head (84a1bff).

Additional details and impacted files
@@           Coverage Diff           @@
##           master    #1450   +/-   ##
=======================================
  Coverage   61.31%   61.31%           
=======================================
  Files          72       72           
  Lines       19771    19771           
  Branches     2852     2852           
=======================================
  Hits        12123    12123           
  Misses       5699     5699           
  Partials     1949     1949           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

@sbryngelson sbryngelson merged commit ab146e9 into MFlowCode:master May 20, 2026
88 checks passed
@sbryngelson sbryngelson deleted the amd-fixes branch May 20, 2026 03:02
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Development

Successfully merging this pull request may close these issues.

amdflang: array constructor assignment to private arrays in GPU_PARALLEL_LOOP produces wrong values (m_ibm.fpp, m_collisions.fpp)

1 participant