https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114832
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |ASSIGNED
Assignee|unassigned at gcc dot gnu.org |rguenth at gcc dot
gnu.org
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
With the early break code we now allow multiple blocks in the loop body but
guess this only happens for blocks with exits. In this case we have a simple
forwarder which seems to confuse the code. Usually if-conversion gets rid of
those so -fno-tree-loop-if-convert is critical to trigger this.
If we want to fix this on the release branch I think we should reject loops
with this kind of CFG again. In stage1 we should fix the real issue.
I think the fix is just
diff --git a/gcc/tree-vect-loop-manip.cc b/gcc/tree-vect-loop-manip.cc
index 8d9b533d50f..43c7881c640 100644
--- a/gcc/tree-vect-loop-manip.cc
+++ b/gcc/tree-vect-loop-manip.cc
@@ -1523,7 +1523,7 @@ slpeel_tree_duplicate_loop_to_edge_cfg (class loop *loop,
edge loop_exit,
exit_dest = exit->dest;
was_imm_dom = (get_immediate_dominator (CDI_DOMINATORS,
- exit_dest) == loop->header ?
+ exit_dest) == exit->src ?
true : false);
/* Also copy the pre-header, this avoids jumping through hoops to
