On Tue, 19 Jul 2011, Michael Matz wrote:
> Hi,
>
> On Tue, 19 Jul 2011, Richard Guenther wrote:
>
> > *************** forward_propagate_comparison (gimple stm
> > *** 1164,1170 ****
> > }
> > /* We can propagate the condition into a statement that
> > computes the logical negation of the comparison result. */
> > ! else if (gimple_assign_rhs_code (use_stmt) == TRUTH_NOT_EXPR)
> > {
> > tree type = TREE_TYPE (gimple_assign_rhs1 (stmt));
> > bool nans = HONOR_NANS (TYPE_MODE (type));
> > --- 1165,1174 ----
> > }
> > /* We can propagate the condition into a statement that
> > computes the logical negation of the comparison result. */
> > ! else if ((gimple_assign_rhs_code (use_stmt) == BIT_NOT_EXPR
> > ! && TYPE_PRECISION (TREE_TYPE (lhs)) == 1)
> > ! || (gimple_assign_rhs_code (use_stmt) == BIT_XOR_EXPR
> > ! && integer_onep (gimple_assign_rhs2 (use_stmt))))
>
> In XOR case you want either precision==1 or integer_all_onesp (the fixed
> variant), otherwise it's no inversion.
As discussed offline that's not what we want. This check is to handle
non-1-precision BOOLEAN_TYPEs (which still have always 1 for true and 0
for false). Consider that the above is always in the context of
a comparison producing the value that is xored with 1.
Richard.
