rL356570 introduced a test which only passes with the default openmp library, libomp, and fails with other openmp libraries, such as libgomp. Explicitly choose libomp.
Details
Diff Detail
- Repository
- rC Clang
Event Timeline
Interesting.
What happens if libomp is not being built?
What happens if libomp is not present?
Or more generally, why does it even matter whether there is runtime or not, this only does paring+sema, no codegen/execution.
If the compiler is built without OpenMP support -fopenmp is converted to -fopenmp=libgomp frontend option.
I'm far from an omp expert (only running across this because some tests failed), but I would guess both those cases to break.
But this is broken anyway when libomp isn't being used. So the set of broken cases is going from (people that don't use libomp) to (people that don't have libomp) which I think is a much smaller subset.
Or more generally, why does it even matter whether there is runtime or not, this only does paring+sema, no codegen/execution.
I'm very surprised myself. Hence my question "has something gone wrong" on D59214. Hopefully this fix/hack can be reverted once that's figured out.