[Flang][OpenMP] Erase trip count for generic kernels

[DominikAdamski]

We determine the type of the kernel based on the MLIR code, which changes during the lowering phase. Some kernels, such as those with multiple workshare loops, are initially classified as SPMD kernels, but are later recognized as generic kernels during PFT lowering. In such cases, we need to identify the change in type and clear the trip count if it was previously set.

[skatrak]

Thank you Dominik, I can see that without this change the trip count would be incorrectly set if there are multiple OpenMP loop nests inside of the target region. I just have a minimal suggestion to help readability.

I'm just wondering if we should think about this case as well (possibly as a separate patch):

!$omp target teams
  !$omp distribute parallel do
  do i=1, n
    ! ...
  end do
  !$omp end distribute parallel do

  x = 5
  ! ...
!$omp end target teams

My feeling is that it will lower the loop and set the trip count because at that point it would be detected as an SPMD loop, but afterwards it would keep adding code to the omp.teams region, so it would break that assumption. If I'm not wrong, maybe a more resilient solution would involve looking at the PFT rather than the half-built MLIR to check these things.

[DominikAdamski]

If I'm not wrong, maybe a more resilient solution would involve looking at the PFT rather than the half-built MLIR to check these things.

I think it's a good idea to look at PFT. @kparzysz Is OpenMP PFT stable enough that we can use it to determine the type of kernel?