http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50014
Ira Rosen <irar at il dot ibm.com> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |ASSIGNED
CC| |irar at il dot ibm.com
AssignedTo|unassigned at gcc dot |irar at gcc dot gnu.org
|gnu.org |
--- Comment #2 from Ira Rosen <irar at il dot ibm.com> 2011-08-08 12:27:44 UTC
---
Somehow when getting a copy of a vector operand in reduction we use a dummy
def_type, which causes that we try to make a copy of a constant instead of just
using it.
I am going to test this patch:
Index: tree-vect-loop.c
===================================================================
--- tree-vect-loop.c (revision 177423)
+++ tree-vect-loop.c (working copy)
@@ -4310,7 +4310,7 @@ vectorizable_reduction (gimple stmt, gim
VEC (tree, heap) *vec_oprnds0 = NULL, *vec_oprnds1 = NULL, *vect_defs =
NULL;
VEC (gimple, heap) *phis = NULL;
int vec_num;
- tree def0, def1, tem;
+ tree def0, def1, tem, op0, op1 = NULL_TREE;
/* In case of reduction chain we switch to the first stmt in the chain, but
we don't update STMT_INFO, since only the last stmt is marked as
reduction
@@ -4767,8 +4767,6 @@ vectorizable_reduction (gimple stmt, gim
/* Handle uses. */
if (j == 0)
{
- tree op0, op1 = NULL_TREE;
-
op0 = ops[!reduc_index];
if (op_type == ternary_op)
{
@@ -4798,11 +4796,19 @@ vectorizable_reduction (gimple stmt, gim
{
if (!slp_node)
{
- enum vect_def_type dt = vect_unknown_def_type; /* Dummy */
- loop_vec_def0 = vect_get_vec_def_for_stmt_copy (dt,
loop_vec_def0);
+ enum vect_def_type dt;
+ gimple dummy_stmt;
+ tree dummy;
+
+ vect_is_simple_use (ops[!reduc_index], loop_vinfo, NULL,
+ &dummy_stmt, &dummy, &dt);
+ loop_vec_def0 = vect_get_vec_def_for_stmt_copy (dt,
+ loop_vec_def0);
VEC_replace (tree, vec_oprnds0, 0, loop_vec_def0);
if (op_type == ternary_op)
{
+ vect_is_simple_use (op1, loop_vinfo, NULL, &dummy_stmt,
+ &dummy, &dt);
loop_vec_def1 = vect_get_vec_def_for_stmt_copy (dt,
loop_vec_def1);
VEC_replace (tree, vec_oprnds1, 0, loop_vec_def1);
