[Bug fortran/107424] [13 Regression] ICE in gfc_trans_omp_do, at fortran/trans-openmp.cc:5397 - and wrong code - with non-rectangular loops

["burnus at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107424

--- Comment #5 from Tobias Burnus <burnus at gcc dot gnu.org> ---
For completeness, the C testcase to previous comment (comment 3) is as follows.
Contrary to Fortran, I put a 'lastprivate' to all variables – to match the
current
patch (→ see (A) in comment 3).

  int n,m,p, i,j,k;
  n = 11; m = 23; p = 27;
  #pragma omp simd collapse(3) lastprivate(k) lastprivate(i) lastprivate(j)
  for (i = 1; i <= n; i++)
    for (j = 1; j <= m; j++)
      for (k = j - 41; k <= p; k++)
        if (k < 1 - 41 || k > p)
          __builtin_printf ("%d %d %d\n", i, j, k);
  if (i != n + 1) __builtin_abort ();
  if (j != m + 1) __builtin_abort ();
  if (k != p + 1) __builtin_abort ();

The code works fine. The original dump is also the same, except for the '* 1':
  #pragma omp simd lastprivate(j) lastprivate(i) lastprivate(k) collapse(3)
     ...
      for (k = j * 1 + -41; k <= p; k++ )

but in the gimple dump, the result is:
    #pragma omp simd lastprivate(j) lastprivate(i) lastprivate(k) collapse(3) \
                     private(k.2) private(j.1) private(i.0)
    for (i.0 = 1; i.0 <= n; i.0 = i.0 + 1)
      for (j.1 = 1; j.1 <= m; j.1 = j.1 + 1)
        for (k.2 = j.1 * 1 + -41; k.2 <= p; k.2 = k.2 + 1)

where 'j' has been replaced by 'j.1' and not been hosted before the loop.