public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [PATCH] Fix PR91293
@ 2019-07-31 13:16 Richard Biener
  0 siblings, 0 replies; only message in thread
From: Richard Biener @ 2019-07-31 13:16 UTC (permalink / raw)
  To: gcc-patches


The following fixes wrong-code when SLP ends up swapping operands
of a reduction stmt, confusing code-generation later.  Fixed by
not doing such swapping.

Bootstrapped on x86_64-unknown-linux-gnu, testing in progress.

Richard.

2019-07-31  Richard Biener  <rguenther@suse.de>

	PR tree-optimization/91293
	* tree-vect-slp.c (vect_build_slp_tree_2): Do not swap operands
	of reduction stmts.

	* gcc.dg/vect/pr91293-1.c: New testcase.
	* gcc.dg/vect/pr91293-2.c: Likewise.
	* gcc.dg/vect/pr91293-3.c: Likewise.

Index: gcc/tree-vect-slp.c
===================================================================
--- gcc/tree-vect-slp.c	(revision 273930)
+++ gcc/tree-vect-slp.c	(working copy)
@@ -1296,6 +1296,9 @@ vect_build_slp_tree_2 (vec_info *vinfo,
 	  && nops == 2
 	  && oprnds_info[1]->first_dt == vect_internal_def
 	  && is_gimple_assign (stmt_info->stmt)
+	  /* Swapping operands for reductions breaks assumptions later on.  */
+	  && STMT_VINFO_DEF_TYPE (stmt_info) != vect_reduction_def
+	  && STMT_VINFO_DEF_TYPE (stmt_info) != vect_double_reduction_def
 	  /* Do so only if the number of not successful permutes was nor more
 	     than a cut-ff as re-trying the recursive match on
 	     possibly each level of the tree would expose exponential
Index: gcc/testsuite/gcc.dg/vect/pr91293-1.c
===================================================================
--- gcc/testsuite/gcc.dg/vect/pr91293-1.c	(nonexistent)
+++ gcc/testsuite/gcc.dg/vect/pr91293-1.c	(working copy)
@@ -0,0 +1,19 @@
+/* { dg-do run } */
+/* { dg-additional-options "-msse4.1" { target { sse4_runtime } } } */
+
+long long a;
+unsigned b, c;
+int d = 62;
+void e(long long *f, int p2) { *f = p2; }
+int main()
+{
+  for (int g = 2; g <= d; g++)
+    {
+      b += g + 4;
+      c += 5 - g;
+    }
+  e(&a, b);
+  if (a != 2196)
+    __builtin_abort ();
+  return 0;
+}
Index: gcc/testsuite/gcc.dg/vect/pr91293-2.c
===================================================================
--- gcc/testsuite/gcc.dg/vect/pr91293-2.c	(nonexistent)
+++ gcc/testsuite/gcc.dg/vect/pr91293-2.c	(working copy)
@@ -0,0 +1,19 @@
+/* { dg-do run } */
+/* { dg-additional-options "-msse4.1" { target { sse4_runtime } } } */
+
+long long a;
+unsigned b, c;
+int d = 62;
+void e(long long *f, int p2) { *f = p2; }
+int main()
+{
+  for (int g = 2; g <= d; g++)
+    {
+      c += 5 - g;
+      b += g + 4;
+    }
+  e(&a, b);
+  if (a != 2196)
+    __builtin_abort ();
+  return 0;
+}
Index: gcc/testsuite/gcc.dg/vect/pr91293-3.c
===================================================================
--- gcc/testsuite/gcc.dg/vect/pr91293-3.c	(nonexistent)
+++ gcc/testsuite/gcc.dg/vect/pr91293-3.c	(working copy)
@@ -0,0 +1,20 @@
+/* { dg-do run } */
+/* { dg-additional-options "-msse4.1" { target { sse4_runtime } } } */
+
+long long a;
+unsigned b, c;
+int d = 62;
+void e(long long *f, int p2) { *f = p2; }
+int xx = 5, yy = 4;
+int main()
+{
+  for (int g = 2; g <= d; g++)
+    {
+      c += xx - g;
+      b += yy + g;
+    }
+  e(&a, b);
+  if (a != 2196)
+    __builtin_abort ();
+  return 0;
+}

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

only message in thread, other threads:[~2019-07-31 13:16 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-07-31 13:16 [PATCH] Fix PR91293 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).