public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [PATCH] Fix PR81790
@ 2017-08-15  9:28 Richard Biener
  0 siblings, 0 replies; only message in thread
From: Richard Biener @ 2017-08-15  9:28 UTC (permalink / raw)
  To: gcc-patches


This PR uncovers a missed-optimization with looking up CONSTRUCTORs
resulting from match-and-simplify simplification.  Those appear in
"GENERIC" form while VN internally uses exploded form.

Thus make vn_lookup_simplify_result mediate between the two -- with
the additional complication of vn_nary_simplify getting here
(eventually) with non-simplified exploded form already.

Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk.

Richard.

2017-08-15  Richard Biener  <rguenther@suse.de>

	PR tree-optimization/81790
	* tree-ssa-sccvn.c (vn_lookup_simplify_result): Handle both
	CONSTRUCTORs from simplifying and VN.

	* gcc.dg/torture/pr81790.c: New testcase.

Index: gcc/tree-ssa-sccvn.c
===================================================================
--- gcc/tree-ssa-sccvn.c	(revision 251086)
+++ gcc/tree-ssa-sccvn.c	(working copy)
@@ -1646,13 +1646,25 @@ static unsigned mprts_hook_cnt;
 /* Hook for maybe_push_res_to_seq, lookup the expression in the VN tables.  */
 
 static tree
-vn_lookup_simplify_result (code_helper rcode, tree type, tree *ops)
+vn_lookup_simplify_result (code_helper rcode, tree type, tree *ops_)
 {
   if (!rcode.is_tree_code ())
     return NULL_TREE;
+  tree *ops = ops_;
+  unsigned int length = TREE_CODE_LENGTH ((tree_code) rcode);
+  if (rcode == CONSTRUCTOR
+      /* ???  We're arriving here with SCCVNs view, decomposed CONSTRUCTOR
+         and GIMPLEs / match-and-simplifies, CONSTRUCTOR as GENERIC tree.  */
+      && TREE_CODE (ops_[0]) == CONSTRUCTOR)
+    {
+      length = CONSTRUCTOR_NELTS (ops_[0]);
+      ops = XALLOCAVEC (tree, length);
+      for (unsigned i = 0; i < length; ++i)
+	ops[i] = CONSTRUCTOR_ELT (ops_[0], i)->value;
+    }
   vn_nary_op_t vnresult = NULL;
-  tree res = vn_nary_op_lookup_pieces (TREE_CODE_LENGTH ((tree_code) rcode),
-				       (tree_code) rcode, type, ops, &vnresult);
+  tree res = vn_nary_op_lookup_pieces (length, (tree_code) rcode,
+				       type, ops, &vnresult);
   /* We can end up endlessly recursing simplifications if the lookup above
      presents us with a def-use chain that mirrors the original simplification.
      See PR80887 for an example.  Limit successful lookup artificially
Index: gcc/testsuite/gcc.dg/torture/pr81790.c
===================================================================
--- gcc/testsuite/gcc.dg/torture/pr81790.c	(revision 0)
+++ gcc/testsuite/gcc.dg/torture/pr81790.c	(working copy)
@@ -0,0 +1,28 @@
+/* { dg-do compile } */
+/* { dg-additional-options "--param sccvn-max-scc-size=10" } */
+
+typedef int a __attribute__ ((__vector_size__ (16)));
+typedef struct
+{
+  a b;
+} c;
+
+int d, e;
+
+void foo (c *ptr);
+
+void bar ()
+{
+  double b = 1842.9028;
+  c g, h;
+  if (d)
+    b = 77.7998;
+  for (; e;)
+    {
+      g.b = g.b = g.b + g.b;
+      h.b = (a){b};
+      h.b = h.b + h.b;
+    }
+  foo (&g);
+  foo (&h);
+}

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

only message in thread, other threads:[~2017-08-15  7:01 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-08-15  9:28 [PATCH] Fix PR81790 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).