Skip to content

changes necessary for compilation with llvm flang - #3500

Open
gbarter wants to merge 1 commit into
OpenFAST:devfrom
gbarter:llvmflang23
Open

gbarter wants to merge 1 commit into
OpenFAST:devfrom
gbarter:llvmflang23

Conversation

@gbarter

@gbarter gbarter commented Oct 24, 2026

Copy link
Copy Markdown

Feature or improvement description

(use of Copilot to) Enable compilation with full LLVM toolchain, with the changes here needed for compilation with LLVM Flang. Compilation with the LLVM suite should hopefully make future conda packaging easier, give a more consistent cross-platform approach, and improve the Fortan-C/C++ interfaces. Care was taken to not impact the compilation with Classic Flang or ROCm Flang as used in some of the HPC systems with Kynema.

In addition to touching a few cmake-files, there were numerous string-copy calls that pass through other compilers fine but not flang. Also, at least one instance of a cot(x) function needing to be flipped to 1/tan(x). Finally, LLVM Flang does not also ship with openmp, so one or two of the use OMP_LIB calls was moved inside an #ifdef

I expect some scrutiny and iteration during code review.

Related issue, if one exists

Impacted areas of the software

The r-test suite was run locally on my Mac with errors just in the numerical precision

Additional supporting information

Generative AI usage

Assisted by: GitHub Copilot support@github.com

Test results, if applicable

Might need guidance from the OF team here on how best to accommodate an LLVM Flang option in the test suite.

  • r-test branch merging required

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant