[flang-commits] [flang] [llvm] [flang] Support static-unit array slices in FIR LoopVersioning (PR #222723)

[via flang-commits]

Ritanya-B-Bharadwaj wrote:

Is this pointer-array case covered by this PR, or is support planned in a follow-up?
```
module pointer_stride_repro
  use iso_fortran_env, only: real64
  implicit none
contains
  subroutine accumulate(dst, x, y, z)
    complex(real64), pointer, intent(inout) :: dst(:)
    complex(real64), pointer, intent(in) :: x(:), y(:), z(:)

    dst(:) = dst(:) + x(:) * y(:) * z(:)
  end subroutine
end module

flang -O3 -ffast-math -fstack-arrays -fversion-loops-for-stride -Rpass=loop-vectorize -S -emit-llvm repro.f90 -o repro.ll
```

On main, this reports vectorization, but the vector path requires byte strides of 1 . Contiguous elements have a byte stride of 16 , so they take the scalar fallback.

https://github.com/llvm/llvm-project/pull/222723