[patch tree-optimization]: Improve handling of conditional-branches on targets with high branch costs

Richard Guenther richard.guenther@gmail.com
Mon Oct 17 15:06:00 GMT 2011


On Mon, Oct 17, 2011 at 3:36 PM, Kai Tietz <ktietz70@googlemail.com> wrote:
> Sure,
>
> Is simplier and also handles (A T[-IF] (B T-IF C) -> (A T B) T-IF C
> case, which can happen by framing in conditions.
>
> @@ -8380,13 +8400,65 @@ fold_truth_andor (location_t loc, enum t
>      lhs is another similar operation, try to merge its rhs with our
>      rhs.  Then try to merge our lhs and rhs.  */
>   if (TREE_CODE (arg0) == code
> -      && 0 != (tem = fold_truthop (loc, code, type,
> -                                  TREE_OPERAND (arg0, 1), arg1)))
> +      && 0 != (tem = fold_truth_andor_1 (loc, code, type,
> +                                        TREE_OPERAND (arg0, 1), arg1)))
>     return fold_build2_loc (loc, code, type, TREE_OPERAND (arg0, 0), tem);
>
> -  if ((tem = fold_truthop (loc, code, type, arg0, arg1)) != 0)
> +  if ((tem = fold_truth_andor_1 (loc, code, type, arg0, arg1)) != 0)
>     return tem;
>
> +  if ((BRANCH_COST (optimize_function_for_speed_p (cfun),
> +                   false) >= 2)
> +      && LOGICAL_OP_NON_SHORT_CIRCUIT)
> +    {
> +      enum tree_code ncode, icode;
> +
> +      ncode = (code == TRUTH_ANDIF_EXPR || code == TRUTH_AND_EXPR)
> +             ? TRUTH_AND_EXPR : TRUTH_OR_EXPR;
> +      icode = ncode == TRUTH_AND_EXPR ? TRUTH_ANDIF_EXPR : TRUTH_ORIF_EXPR;
> +
> +      /* Transform ((A AND-IF B) AND[-IF] C) into (A AND-IF (B AND C)),
> +        or ((A OR-IF B) OR[-IF] C) into (A OR-IF (B OR C))
> +        We don't want to pack more than two leafs to a non-IF AND/OR
> +        expression.
> +        If tree-code of left-hand operand isn't an AND/OR-IF code and not
> +        equal to IF-CODE, then we don't want to add right-hand operand.
> +        If the inner right-hand side of left-hand operand has
> +        side-effects, or isn't simple, then we can't add to it,
> +        as otherwise we might destroy if-sequence.  */
> +      if (TREE_CODE (arg0) == icode
> +         && simple_operand_p_2 (arg1)
> +         /* Needed for sequence points to handle trappings, and
> +            side-effects.  */
> +         && simple_operand_p_2 (TREE_OPERAND (arg0, 1)))
> +       {
> +         tem = fold_build2_loc (loc, ncode, type, TREE_OPERAND (arg0, 1),
> +                                arg1);
> +         return fold_build2_loc (loc, icode, type, TREE_OPERAND (arg0, 0),
> +                                 tem);
> +       }
> +       /* Same as abouve but for (A AND[-IF] (B AND-IF C)) -> ((A AND B) AND-IF C),
> +          or (A OR[-IF] (B OR-IF C) -> ((A OR B) OR-IF C).  */
> +      else if (TREE_CODE (arg1) == icode
> +         && simple_operand_p_2 (arg0)
> +         /* Needed for sequence points to handle trappings, and
> +            side-effects.  */
> +         && simple_operand_p_2 (TREE_OPERAND (arg1, 0)))
> +       {
> +         tem = fold_build2_loc (loc, ncode, type,
> +                                arg0, TREE_OPERAND (arg1, 0));
> +         return fold_build2_loc (loc, icode, type, tem,
> +                                 TREE_OPERAND (arg1, 1));
> +       }
> +      /* Transform (A AND-IF B) into (A AND B), or (A OR-IF B)
> +        into (A OR B).
> +        For sequence point consistancy, we need to check for trapping,
> +        and side-effects.  */
> +      else if (code == icode && simple_operand_p_2 (arg0)
> +               && simple_operand_p_2 (arg1))
> +       return fold_build2_loc (loc, ncode, type, arg0, arg1);
> +    }
> +
>   return NULL_TREE;
>  }

Ok with the rest of the changes I requested.

Richard.

> Regards,
> Kai
>



More information about the Gcc-patches mailing list