[Bug middle-end/77366] New: Rev. 2ac4967 prevents loop unrolling for s390

[rdapp at linux dot vnet.ibm.com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77366

            Bug ID: 77366
           Summary: Rev. 2ac4967 prevents loop unrolling for s390
           Product: gcc
           Version: 7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: rdapp at linux dot vnet.ibm.com
                CC: krebbel at gcc dot gnu.org
  Target Milestone: ---
              Host: s390
            Target: s390

Since 2ac4967f49a70e1bd0bb28a142324f527dac3743 the following loop is not being
unrolled anymore (-O3 -funroll-loops):

void
foo(unsigned int size, unsigned int *state)
{
   unsigned int i;

   for(i = 0; i < size; i++)
    {
       if(*state & 1)
         {
           *state ^= 1;
         }
    }
}

The revision adds an additional check that allows the split-paths pass to
duplicate a basic block (returned NULL before the patch) which in turn causes
check_simple_exit() in loop-iv.c to not find a proper niter_desc. This seems
due to

if (!dominated_by_p (CDI_DOMINATORS, loop->latch, exit_bb))

failing, i.e. check_simple_exit() doesn't see that the exit condition is
checked every iteration.