Hi,
This patch is to fix a latent vectorizer bug uncovered by previous one.
Function optimize_mask_stores
creates if-then structure in CFG in order to revert if-conversion
transformation. It adds newly created
then_bb to outer loop when vectorizing two levels loop nest and masked_store is
in the inner loop.
Without this patch, gcc.dg/tree-ssa/pr71077.c would be broken. Bootstrap and
test on AArch64, is it OK?
2017-03-27 Bin Cheng <bin.ch...@arm.com>
* tree-vect-loop.c (optimize_mask_stores): Add bb to the right
loop.
diff --git a/gcc/tree-vect-loop.c b/gcc/tree-vect-loop.c
index 6bbf816..af874e7 100644
--- a/gcc/tree-vect-loop.c
+++ b/gcc/tree-vect-loop.c
@@ -7229,6 +7229,7 @@ optimize_mask_stores (struct loop *loop)
unsigned nbbs = loop->num_nodes;
unsigned i;
basic_block bb;
+ struct loop *bb_loop;
gimple_stmt_iterator gsi;
gimple *stmt;
auto_vec<gimple *> worklist;
@@ -7267,11 +7268,16 @@ optimize_mask_stores (struct loop *loop)
last = worklist.pop ();
mask = gimple_call_arg (last, 2);
bb = gimple_bb (last);
- /* Create new bb. */
+ /* Create then_bb and if-then structure in CFG, then_bb belongs to
+ the same loop as if_bb. It could be different to LOOP when two
+ level loop-nest is vectorized and mask_store belongs to the inner
+ one. */
e = split_block (bb, last);
+ bb_loop = bb->loop_father;
+ gcc_assert (loop == bb_loop || flow_loop_nested_p (loop, bb_loop));
join_bb = e->dest;
store_bb = create_empty_bb (bb);
- add_bb_to_loop (store_bb, loop);
+ add_bb_to_loop (store_bb, bb_loop);
e->flags = EDGE_TRUE_VALUE;
efalse = make_edge (bb, store_bb, EDGE_FALSE_VALUE);
/* Put STORE_BB to likely part. */