[Mlir-commits] [mlir] [Flang][OpenMP][MLIR] Initial array section mapping MLIR -> LLVM-IR lowering utilising omp.bounds (PR #68689)

[llvmlistbot at llvm.org]

agozillon wrote:

The gfortran regressions should be fixed by: https://github.com/llvm/llvm-project/commit/68c384676cf92289d823576f915ed296d209354f which re-introduces the very basic implicit capture we have currently that I a little too hastily removed in this PR. It appears while we don't use it in any of our tests, we do depend on the behavior in the regression test suite to some extent. So until the IFA series of patches lands, we must continue to keep this little bit of behavior to satisfy the regression tests, when the IFA patches from @TIFitis land this segment can likely be removed (can be done in the IFA patches themselves, or I can do so after the fact).  

Unfortunately, that means at the moment we occasionally (such as when using do loops) capture constants that we don't handle very well yet. Hopefully the IFA patch fixes this when it lands, but if it doesn't it's something I'm happy to look into as it's resolved downstream, so should just need a little nudge somewhere in the right direction. But in any case, I had to change some tests to utilise do while's with explicit arguments. 

https://github.com/llvm/llvm-project/pull/68689