http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55110
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-11-26
10:35:44 UTC ---
I'd say either we should nuke the bogus
gcc_assert (STMT_VINFO_RELATED_STMT (orig_stmt_info) == stmt);
assertion, which predates the introduction of
STMT_VINFO_PATTERN_DEF_STMT (and later STMT_VINFO_PATTERN_DEF_SEQ) - both in
4.7 cycle, or just return false; if it isn't equal.
We won't successfully vectorize this anyway, orig_code is still TRUNC_DIV_EXPR
and that is not a handled reduction (and even if we'd handled it as the signed
MULT_HIGHPART_EXPR reduction, that isn't handled either).
So, I'd propose this (if we wanted to assert something, it would need to
walk the STMT_VINFO_PATTERN_DEF_SEQ sequence and checking whether stmt isn't in
there too):
2012-11-26 Jakub Jelinek <[email protected]>
PR tree-optimization/55110
* tree-vect-loop.c (vectorizable_reduction): Don't assert
that STMT_VINFO_RELATED_STMT of orig_stmt is stmt.
* gcc.dg/pr55110.c: New test.
--- gcc/tree-vect-loop.c.jj 2012-11-21 16:00:12.000000000 +0100
+++ gcc/tree-vect-loop.c 2012-11-26 11:24:42.903995009 +0100
@@ -4624,7 +4624,6 @@ vectorizable_reduction (gimple stmt, gim
if (orig_stmt)
{
orig_stmt_info = vinfo_for_stmt (orig_stmt);
- gcc_assert (STMT_VINFO_RELATED_STMT (orig_stmt_info) == stmt);
gcc_assert (STMT_VINFO_IN_PATTERN_P (orig_stmt_info));
gcc_assert (!STMT_VINFO_IN_PATTERN_P (stmt_info));
}
--- gcc/testsuite/gcc.dg/pr55110.c.jj 2012-11-26 11:31:32.008587313 +0100
+++ gcc/testsuite/gcc.dg/pr55110.c 2012-11-26 11:31:09.000000000 +0100
@@ -0,0 +1,13 @@
+/* PR tree-optimization/55110 */
+/* { dg-do compile } */
+/* { dg-options "-O1 -ftree-vectorize" } */
+
+int
+foo (int x)
+{
+ int a, b;
+ for (b = 0; b < 8; b++)
+ for (a = 0; a < 2; a++)
+ x /= 3;
+ return x;
+}
