https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107160
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Assignee|linkw at gcc dot gnu.org |rguenth at gcc dot
gnu.org
Status|NEW |ASSIGNED
--- Comment #15 from Richard Biener <rguenth at gcc dot gnu.org> ---
I think that either vect_find_reusable_accumulator should punt on this or
vect_create_epilog_for_reduction shouldn't register this accumulator.
The caller of vect_find_reusable_accumulator checks for vec_num == 1
(in the epilog) but we register for vec_num == 2. For SLP reductions
we "fail" to reduce to a single accumulator.
int vec_num = vec_oprnds0.length ();
gcc_assert (vec_num == 1 || slp_node);
The following fixes this:
diff --git a/gcc/tree-vect-loop.cc b/gcc/tree-vect-loop.cc
index 1996ecfee7a..b1442a93581 100644
--- a/gcc/tree-vect-loop.cc
+++ b/gcc/tree-vect-loop.cc
@@ -6232,7 +6232,8 @@ vect_create_epilog_for_reduction (loop_vec_info
loop_vinfo,
}
/* Record this operation if it could be reused by the epilogue loop. */
- if (STMT_VINFO_REDUC_TYPE (reduc_info) == TREE_CODE_REDUCTION)
+ if (STMT_VINFO_REDUC_TYPE (reduc_info) == TREE_CODE_REDUCTION
+ && vec_num == 1)
loop_vinfo->reusable_accumulators.put (scalar_results[0],
{ orig_reduc_input, reduc_info });
