[autovect] [patch] Make store motion use alias oracle
Revital1 Eres
ERES@il.ibm.com
Thu Mar 15 09:12:00 GMT 2007
gcc-patches-owner@gcc.gnu.org wrote on 14/03/2007 16:26:57:
> Revital1 Eres <ERES@il.ibm.com> writes:
>
> > +/* Return true if STMT may have certain side effects
> > + and thus ifcvt can not be applied. Otherwise return false. */
> > +static bool
> > +may_have_side_effect_p (tree stmt)
>
> Can you give this function a different name? It does not seem to test
> for whether the statement may have side effects: e.g., statements do
> not have side effects merely because they are at the end of a basic
> block. The function seems to test for some other property and I think
> it would be clearer if it had some other name.
>
In that case maybe it will be more appropriate to cancel
the function and insert all those constrains to the caller
if_convertible_gimple_modify_stmt_p ()
Revital
Index: tree-if-conv.c
===================================================================
--- tree-if-conv.c (revision 122880)
+++ tree-if-conv.c (working copy)
@@ -346,17 +346,28 @@
if_convertible_gimple_modify_stmt_p (struct loop *loop, basic_block bb,
tree m_expr)
{
+ tree lhs, rhs;
+
if (dump_file && (dump_flags & TDF_DETAILS))
{
fprintf (dump_file, "-------------------------\n");
print_generic_stmt (dump_file, m_expr, TDF_SLIM);
}
- /* Be conservative and do not handle immovable expressions. */
- if (movement_possibility (m_expr) == MOVE_IMPOSSIBLE)
+ if (TREE_CODE (stmt) != GIMPLE_MODIFY_STMT)
+ return false;
+
+ lhs = GIMPLE_STMT_OPERAND (stmt, 0);
+ rhs = GIMPLE_STMT_OPERAND (stmt, 1);
+
+ /* Some of these constrains might be too conservative. */
+ if (stmt_ends_bb_p (stmt) || stmt_ann (stmt)->has_volatile_ops
+ || (TREE_CODE (lhs) == SSA_NAME
+ && SSA_NAME_OCCURS_IN_ABNORMAL_PHI (lhs))
+ || TREE_SIDE_EFFECTS (rhs))
{
if (dump_file && (dump_flags & TDF_DETAILS))
- fprintf (dump_file, "stmt is movable. Don't take risk\n");
+ fprintf (dump_file, "stmt not suitable for ifcvt\n");
return false;
}
> Thanks.
>
> Ian
More information about the Gcc-patches
mailing list