[Mlir-commits] [mlir] [OpenMP] Enable simd in non-reduction composite constructs (PR #146097)

[Kajetan Puchalski]

mrkajetanp wrote:

I ran both gfortran and Fujitsu test suites (alongside the LLVM one) to make sure this does not introduce regressions in those. For a number of test Fortran programs using composite constructs with simd, I manually inspected the IR and the final binary to make sure that with this change, vectorisable instructions inside these constructs turn into vectorised instructions even with `-fno-vectorize`, where without this change they do not.

https://github.com/llvm/llvm-project/pull/146097