On 14 March 2012 11:48, Richard Guenther <richard.guent...@gmail.com> wrote:
> On Wed, Mar 14, 2012 at 11:39 AM, Frederic Riss <frederic.r...@gmail.com>
> wrote:
>> Doing that at expansion time looks like papering over some other issue
>> though.
>
> Can you check where the COND_EXPR is introduced? That place should
> be fixed to properly fold.
Quite logically it's the if-conversion that generates the COND_EXPRs.
The following patch tries to fold if-converted stmts right after their
creation; it fixes my test-case but hasn't undergone any other
validation:
diff --git a/gcc/tree-if-conv.c b/gcc/tree-if-conv.c
index ca9503f..59fc162 100644
--- a/gcc/tree-if-conv.c
+++ b/gcc/tree-if-conv.c
@@ -1278,6 +1278,7 @@ predicate_scalar_phi (gimple phi, tree cond,
gimple new_stmt;
basic_block bb;
tree rhs, res, arg, scev;
+ gimple_stmt_iterator new_gsi;
gcc_assert (gimple_code (phi) == GIMPLE_PHI
&& gimple_phi_num_args (phi) == 2);
@@ -1322,6 +1323,11 @@ predicate_scalar_phi (gimple phi, tree cond,
new_stmt = gimple_build_assign (res, rhs);
SSA_NAME_DEF_STMT (gimple_phi_result (phi)) = new_stmt;
gsi_insert_before (gsi, new_stmt, GSI_SAME_STMT);
+ if (arg != rhs) {
+ /* COND_EXPR might fold to {MIN,MAX}_EXPR. */
+ new_gsi = gsi_for_stmt (new_stmt);
+ fold_stmt (&new_gsi);
+ }
update_stmt (new_stmt);
if (dump_file && (dump_flags & TDF_DETAILS))
