[Bug target/100777] New: MMA builtin usage ICEs when used in a #pragma omp parallel and using -fopenmp

[bergner at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100777

            Bug ID: 100777
           Summary: MMA builtin usage ICEs when used in a #pragma omp
                    parallel and using -fopenmp
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: bergner at gcc dot gnu.org
  Target Milestone: ---

bergner@pike:~/$ cat bug.i 
void
foo (__vector_quad *dst)
{
#pragma omp parallel
  {
    __builtin_mma_xxsetaccz (dst);
  }
}

bergner@pike:~/$ gcc -S -O1 -mcpu=power10 -fopenmp bug.i 
during GIMPLE pass: ssa
bug.i: In function ‘foo’:
bug.i:8:1: internal compiler error: Segmentation fault
    8 | }
      | ^
0x10b89f63 crash_signal
        /home/bergner/gcc/gcc-fsf-mainline-base/gcc/toplev.c:327
0x10ea5a00 verify_ssa(bool, bool)
        /home/bergner/gcc/gcc-fsf-mainline-base/gcc/tree-ssa.c:1070
...

This fails on trunk back to GCC10 (ie, first MMA support).