[Patch, Fortran] Use -huge()-1 as most negative integer for maxval/maxloc (PR30512)

[Tobias Burnus]

Thomas Koenig wrote:
> This is OK, with a few minor nits:
>   
I'm not sure I understand the nits correctly.

> On Thu, Feb 01, 2007 at 09:59:27AM +0100, Tobias Burnus wrote:
>   
>> ===================================================================
>> --- gcc/fortran/trans-intrinsic.c	(Revision 121416)
>> +++ gcc/fortran/trans-intrinsic.c	(Arbeitskopie)
>> @@ -1984,11 +1984,16 @@
>>        gcc_unreachable ();
>>      }
>>  
>> -  /* Most negative(+HUGE) for maxval, most negative (-HUGE) for minval.  */
>> +  /* Most negative(-HUGE) for maxloc, most positiv (+HUGE) for minloc.  */
>>     
> This should read
>   
>> -  /* Most negative(-HUGE) for maxval, most positive (-HUGE) for minval.  */
>> +  /* Most negative(-HUGE) for maxval, most positive (+HUGE) for minval.  */
>>     

I think the way it is in my patch is commentwise correct. In my tree it
currently (with patch) reads as:


gfc_conv_intrinsic_minmaxloc (gfc_se * se, gfc_expr * expr, int op)
[...]
    case BT_REAL:
      tmp = gfc_conv_mpfr_to_tree (gfc_real_kinds[n].huge,
expr->ts.kind); [...]
    case BT_INTEGER:
      tmp = gfc_conv_mpz_to_tree (gfc_integer_kinds[n].huge, expr->ts.kind);
[...]
  /* Most negative(-HUGE) for maxloc, most positiv (+HUGE) for minloc.  */
  if (op == GT_EXPR)
    tmp = fold_build1 (NEGATE_EXPR, TREE_TYPE (tmp), tmp);
  gfc_add_modify_expr (&se->pre, limit, tmp);

  /* Most negative for BT_INTEGER is -HUGE-1.  */
  if (op == GT_EXPR && expr->ts.type == BT_INTEGER)
    tmp = build2 (MINUS_EXPR, TREE_TYPE (tmp), tmp,
                  build_int_cst (type, 1));


gfc_conv_intrinsic_minmaxval (gfc_se * se, gfc_expr * expr, int op)
[...]
  /* Most negative(-HUGE) for maxval, most positive (+HUGE) for minval.  */
  if (op == GT_EXPR)
    tmp = fold_build1 (NEGATE_EXPR, TREE_TYPE (tmp), tmp);

  /* Most negative for BT_INTEGER is -HUGE-1.  */
  if (op == GT_EXPR && expr->ts.type == BT_INTEGER)
    tmp = build2 (MINUS_EXPR, TREE_TYPE (tmp), tmp,
                  build_int_cst (type, 1));


> and this
> /* Most negative(-HUGE-1) for maxval, most positive (+HUGE) for minval.  */
>   

While I agree that one could improve the wording, I think using
"(-HUGE-1)" is confusing for BT_REAL.
How about something like the following?


  /* Most negative for maxloc, most positiv (+HUGE) for minloc.
     Most negative is -HUGE for BT_REAL and -HUGE-1 for BT_INTEGER.  */
  if (op == GT_EXPR)
    tmp = fold_build1 (NEGATE_EXPR, TREE_TYPE (tmp), tmp);
  gfc_add_modify_expr (&se->pre, limit, tmp);

  if (op == GT_EXPR && expr->ts.type == BT_INTEGER)
    tmp = build2 (MINUS_EXPR, TREE_TYPE (tmp), tmp,
                  build_int_cst (type, 1));


Tobias