public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r13-6942] tree-optimization/109342 - wrong code with edge equivalences in VN
@ 2023-03-30 11:02 Richard Biener
0 siblings, 0 replies; only message in thread
From: Richard Biener @ 2023-03-30 11:02 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:1d0ba4467dd9cad11eb9ff547442e3ce6292b892
commit r13-6942-g1d0ba4467dd9cad11eb9ff547442e3ce6292b892
Author: Richard Biener <rguenther@suse.de>
Date: Thu Mar 30 11:32:19 2023 +0200
tree-optimization/109342 - wrong code with edge equivalences in VN
The following testcase shows a problem in how we query valitity for
equivalences on edges when the edge is a backedge and thus refering
to a block thats later in the iteration order we use for VN. That
causes the dominated_by_p_w_unex helper to look at edge executable
state that's not yet computed and thus still at optimistic not
executable state.
The following makes sure to use a plain dominance check in these cases.
PR tree-optimization/109342
* tree-ssa-sccvn.cc (vn_nary_op_get_predicated_value): New
overload for edge. When that edge is a backedge use
dominated_by_p directly.
* g++.dg/torture/pr109342.C: New testcase.
Diff:
---
gcc/testsuite/g++.dg/torture/pr109342.C | 33 +++++++++++++++++++++++++++++
gcc/tree-ssa-sccvn.cc | 37 ++++++++++++++++++++++++---------
2 files changed, 60 insertions(+), 10 deletions(-)
diff --git a/gcc/testsuite/g++.dg/torture/pr109342.C b/gcc/testsuite/g++.dg/torture/pr109342.C
new file mode 100644
index 00000000000..2cdfa764faf
--- /dev/null
+++ b/gcc/testsuite/g++.dg/torture/pr109342.C
@@ -0,0 +1,33 @@
+// { dg-do run }
+
+unsigned long int var_4 = 42;
+unsigned long int var_14 = 10;
+unsigned char var_16 = 1;
+unsigned short var_18 = 0;
+short var_75;
+
+inline const int &foo(const int &b, const int &c)
+{
+ return b < c ? c : b;
+}
+inline unsigned long &bar(unsigned long &b, unsigned long &c)
+{
+ return !c ? c : b;
+}
+
+void __attribute__((noipa))
+test(unsigned long var_4, unsigned long var_14,
+ unsigned char var_16, unsigned short var_18)
+{
+ for (bool h = 0; h < (bool)foo(var_16 ? -7 : 4, var_4 ? var_4 : var_18);
+ h = 2)
+ var_75 = bar(var_4, var_14);
+}
+
+int main()
+{
+ test(var_4, var_14, var_16, var_18);
+ if (var_75 != 42)
+ __builtin_abort();
+ return 0;
+}
diff --git a/gcc/tree-ssa-sccvn.cc b/gcc/tree-ssa-sccvn.cc
index 6b8d38b270c..99609538f54 100644
--- a/gcc/tree-ssa-sccvn.cc
+++ b/gcc/tree-ssa-sccvn.cc
@@ -4583,22 +4583,39 @@ static bool
dominated_by_p_w_unex (basic_block bb1, basic_block bb2, bool);
static tree
-vn_nary_op_get_predicated_value (vn_nary_op_t vno, basic_block bb)
+vn_nary_op_get_predicated_value (vn_nary_op_t vno, basic_block bb,
+ edge e = NULL)
{
if (! vno->predicated_values)
return vno->u.result;
for (vn_pval *val = vno->u.values; val; val = val->next)
for (unsigned i = 0; i < val->n; ++i)
- /* Do not handle backedge executability optimistically since
- when figuring out whether to iterate we do not consider
- changed predication. */
- if (dominated_by_p_w_unex
- (bb, BASIC_BLOCK_FOR_FN (cfun, val->valid_dominated_by_p[i]),
- false))
- return val->result;
+ {
+ basic_block cand
+ = BASIC_BLOCK_FOR_FN (cfun, val->valid_dominated_by_p[i]);
+ /* Do not handle backedge executability optimistically since
+ when figuring out whether to iterate we do not consider
+ changed predication.
+ When asking for predicated values on an edge avoid looking
+ at edge executability for edges forward in our iteration
+ as well. */
+ if (e && (e->flags & EDGE_DFS_BACK))
+ {
+ if (dominated_by_p (CDI_DOMINATORS, bb, cand))
+ return val->result;
+ }
+ else if (dominated_by_p_w_unex (bb, cand, false))
+ return val->result;
+ }
return NULL_TREE;
}
+static tree
+vn_nary_op_get_predicated_value (vn_nary_op_t vno, edge e)
+{
+ return vn_nary_op_get_predicated_value (vno, e->src, e);
+}
+
/* Insert the rhs of STMT into the current hash table with a value number of
RESULT. */
@@ -5928,7 +5945,7 @@ visit_phi (gimple *phi, bool *inserted, bool backedges_varying_p)
ops, &vnresult);
if (! val && vnresult && vnresult->predicated_values)
{
- val = vn_nary_op_get_predicated_value (vnresult, e->src);
+ val = vn_nary_op_get_predicated_value (vnresult, e);
if (val && integer_truep (val)
&& !(sameval_e && (sameval_e->flags & EDGE_DFS_BACK)))
{
@@ -5947,7 +5964,7 @@ visit_phi (gimple *phi, bool *inserted, bool backedges_varying_p)
we can change sameval to def. */
if (EDGE_COUNT (bb->preds) == 2
&& (val = vn_nary_op_get_predicated_value
- (vnresult, EDGE_PRED (bb, 0)->src))
+ (vnresult, EDGE_PRED (bb, 0)))
&& integer_truep (val)
&& !(e->flags & EDGE_DFS_BACK))
{
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2023-03-30 11:02 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-03-30 11:02 [gcc r13-6942] tree-optimization/109342 - wrong code with edge equivalences in VN Richard Biener
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).