[OpenMP] 'omp unroll partial' loop generates too many iterations

[mikerice1969]

In the following test the loop executes one more time than expected:

void unroll_partial_remainder(int *A)
{
    #pragma omp parallel for
    #pragma omp unroll partial(4)
    for (int i = 0; i < 12; ++i)
        A[i] = i;
}

#include <stdio.h>
#include <assert.h>

int main() {

  int A[24];
  for(int i = 0; i<24; i++){ A[i]=0; }

  // write 12 elements 0 thru 11.
  unroll_partial_remainder(A);

  // Off by one?
  assert(A[12] == 0);

  return 0;
}

$ clang -fopenmp -fopenmp-version=51 u.c && ./a.out
a.out: u.c:21: int main(): Assertion `A[12] == 0' failed.
Aborted (core dumped)

$ clang --version
clang version 15.0.0 (git@github.com:llvm/llvm-project ef87865b98fa25af1d2c045bab1268b2a1503374)
Target: x86_64-unknown-linux-gnu

[llvmbot]

@llvm/issue-subscribers-openmp

[llvmbot]

@llvm/issue-subscribers-clang-frontend

[Meinersbur]

Thanks for the report. I fixed it in acec070 after verifying a variety of other combinations are working as expected. I already did that at the patch stage, I don't know how this slipped through.

/cherry-pick acec070

[llvmbot]

Failed to cherry-pick: acec070

https://github.com/llvm/llvm-project/actions/runs/2360595720

Please manually backport the fix and push it to your github fork. Once this is done, please add a comment like this:

/branch <user>/<repo>/<branch>

[Meinersbur]

/branch meinersbur/llvm-project/unroll-off-by-one