public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [PATCH] tree-optimization/107569 - avoid wrecking earlier folding in FRE/PRE
@ 2023-03-23 16:21 Richard Biener
  0 siblings, 0 replies; only message in thread
From: Richard Biener @ 2023-03-23 16:21 UTC (permalink / raw)
  To: gcc-patches; +Cc: Jakub Jelinek

The following avoids picking up dead code left over from folding
during FRE/PRE, effectively undoing propagations.

Bootstrapped and tested on x86_64-unknown-linux-gnu, pushed.

	PR tree-optimization/107569
	* tree-ssa-sccvn.cc (eliminate_dom_walker::eliminate_stmt):
	Do not push SSA names with zero uses as available leader.
	(process_bb): Likewise.

	* g++.dg/opt/pr107569.C: New testcase.
---
 gcc/testsuite/g++.dg/opt/pr107569.C | 29 +++++++++++++++++++++++++++++
 gcc/tree-ssa-sccvn.cc               | 17 +++++++++++------
 2 files changed, 40 insertions(+), 6 deletions(-)
 create mode 100644 gcc/testsuite/g++.dg/opt/pr107569.C

diff --git a/gcc/testsuite/g++.dg/opt/pr107569.C b/gcc/testsuite/g++.dg/opt/pr107569.C
new file mode 100644
index 00000000000..e03941c7862
--- /dev/null
+++ b/gcc/testsuite/g++.dg/opt/pr107569.C
@@ -0,0 +1,29 @@
+// { dg-do compile }
+// { dg-require-effective-target c++20 }
+// { dg-options "-O2 -fdump-tree-evrp -fdump-tree-vrp1" }
+
+namespace std
+{
+  constexpr bool isfinite (float x) { return __builtin_isfinite (x); }
+  constexpr bool isfinite (double x) { return __builtin_isfinite (x); }
+  constexpr bool isfinite (long double x) { return __builtin_isfinite (x); }
+}
+
+bool
+foo (double x)
+{
+  if (!std::isfinite (x))
+    __builtin_unreachable ();
+
+  return std::isfinite (x);
+}
+
+bool
+bar (double x)
+{
+  [[assume (std::isfinite (x))]];
+  return std::isfinite (x);
+}
+
+/* { dg-final { scan-tree-dump "return 1;" "evrp" } } */
+/* { dg-final { scan-tree-dump-times "return 1;" 2 "vrp1" } } */
diff --git a/gcc/tree-ssa-sccvn.cc b/gcc/tree-ssa-sccvn.cc
index d5b081a309f..6b8d38b270c 100644
--- a/gcc/tree-ssa-sccvn.cc
+++ b/gcc/tree-ssa-sccvn.cc
@@ -7197,10 +7197,14 @@ eliminate_dom_walker::eliminate_stmt (basic_block b, gimple_stmt_iterator *gsi)
     }
 
   /* Make new values available - for fully redundant LHS we
-     continue with the next stmt above and skip this.  */
-  def_operand_p defp;
-  FOR_EACH_SSA_DEF_OPERAND (defp, stmt, iter, SSA_OP_DEF)
-    eliminate_push_avail (b, DEF_FROM_PTR (defp));
+     continue with the next stmt above and skip this.
+     But avoid picking up dead defs.  */
+  tree def;
+  FOR_EACH_SSA_TREE_OPERAND (def, stmt, iter, SSA_OP_DEF)
+    if (! has_zero_uses (def)
+	|| (inserted_exprs
+	    && bitmap_bit_p (inserted_exprs, SSA_NAME_VERSION (def))))
+      eliminate_push_avail (b, def);
 }
 
 /* Perform elimination for the basic-block B during the domwalk.  */
@@ -8046,9 +8050,10 @@ process_bb (rpo_elim &avail, basic_block bb,
 	avail.eliminate_stmt (bb, &gsi);
       else
 	/* If not eliminating, make all not already available defs
-	   available.  */
+	   available.  But avoid picking up dead defs.  */
 	FOR_EACH_SSA_TREE_OPERAND (op, gsi_stmt (gsi), i, SSA_OP_DEF)
-	  if (! avail.eliminate_avail (bb, op))
+	  if (! has_zero_uses (op)
+	      && ! avail.eliminate_avail (bb, op))
 	    avail.eliminate_push_avail (bb, op);
     }
 
-- 
2.35.3

^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2023-03-23 16:21 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-03-23 16:21 [PATCH] tree-optimization/107569 - avoid wrecking earlier folding in FRE/PRE 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).