While working on PR57359 I noticed that a change during GCC 9
development broke a pointer equality check in ref_always_accessed.
The following corrects this - the way LIM works it is enough
to check for store vs. load.
The issue leads to more conditional SMs than necessary.
Bootstrap / regtest running on x86_64-unknown-linux-gnu.
Richard.
This fixes a regression when canonicalizing refs for LIM PR84362.
This possibly unshares and rewrites the refs in the internal data
and thus pointer equality no longer works in ref_always_accessed
computation.
2020-04-28 Richard Biener <rguent...@suse.de>
* tree-ssa-loop-im.c (ref_always_accessed::operator ()):
Just check whether the stmt stores.
---
gcc/tree-ssa-loop-im.c | 9 +++++----
1 file changed, 5 insertions(+), 4 deletions(-)
diff --git a/gcc/tree-ssa-loop-im.c b/gcc/tree-ssa-loop-im.c
index 2bb47e9d6d2..471b620f2c4 100644
--- a/gcc/tree-ssa-loop-im.c
+++ b/gcc/tree-ssa-loop-im.c
@@ -2407,20 +2407,21 @@ ref_always_accessed::operator () (mem_ref_loc *loc)
{
class loop *must_exec;
- if (!get_lim_data (loc->stmt))
+ struct lim_aux_data *lim_data = get_lim_data (loc->stmt);
+ if (!lim_data)
return false;
/* If we require an always executed store make sure the statement
- stores to the reference. */
+ is a store. */
if (stored_p)
{
tree lhs = gimple_get_lhs (loc->stmt);
if (!lhs
- || lhs != *loc->ref)
+ || !(DECL_P (lhs) || REFERENCE_CLASS_P (lhs)))
return false;
}
- must_exec = get_lim_data (loc->stmt)->always_executed_in;
+ must_exec = lim_data->always_executed_in;
if (!must_exec)
return false;
--
2.13.7
