Hi Mikael,

thanks a lot for your comments!

>> +      actual_arglist->expr = gfc_copy_expr (e);
>> +      actual_arglist->next = gfc_get_actual_arglist ();
> Another one is needed.  I get a segmentation fault with SUM.

Fixed by using gfc_build_intrisic_call.  I have also put SUM
into the test case.


>> +  if (dim != NULL)
>> +    {
> Minor, but I think you can assume dim != NULL.  Same for mask.

Fixed.

>> +  /* We changed things from under the expression walker.  Walking the
>> +     old tree would mess up things, so let's not do that.  */
>> +  return 1;
> I think this prevents any further reduction optimization. The following
> variant of your test case doesn't avoid the temporary:

You're right; I also could not come up with a test case where this
didn't work.

I have put this

>      do i=1,3
>         if (any([abs(a(i,1) - b(i,1)) > acc,  &
>              (j==i+1,j=3,8)])) cycle
>         if (any([abs(a(i,2) - b(i,2)) > acc, &
>              abs(a(i,3) - b(i,3)) > acc, lo(i,:)])) cycle
>         c = c + i
>      end do

into the test case.

Updated test case and patch attached.

OK for trunk?

	Thomas

2013-01-13  Thomas Koenig  <tkoenig@gcc.gnu.org>

         PR fortran/55806
         * frontend-passes.c (optimize_reduction):  New function,
         including prototype.
         (callback_reduction):  Likewise.
         (gfc_run_passes):  Also run optimize_reduction.
         (copy_walk_reduction_arg):  New function.
         (dummy_code_callback):  New function.

2013-01-13  Thomas Koenig  <tkoenig@gcc.gnu.org>

         PR fortran/55806
         * gfortran.dg/array_constructor_40.f90:  New test.