public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [PATCH] Fix PR83255
@ 2017-12-04 14:01 Richard Biener
  0 siblings, 0 replies; only message in thread
From: Richard Biener @ 2017-12-04 14:01 UTC (permalink / raw)
  To: gcc-patches


Been sitting in my tree (fixes SPEC 2k6 miscompares).

Bootstrapped and tested on x86-64-unknown-linux-gnu, applied.

Richard.

2017-12-04  Richard Biener  <rguenther@suse.de>

	PR tree-optimization/83255
	* graphite-isl-ast-to-gimple.c (translate_isl_ast_node_for):
	Re-add zero-iteration check.

	* gcc.dg/graphite/pr83255.c: New testcase.

Index: gcc/graphite-isl-ast-to-gimple.c
===================================================================
--- gcc/graphite-isl-ast-to-gimple.c	(revision 255375)
+++ gcc/graphite-isl-ast-to-gimple.c	(working copy)
@@ -720,6 +720,32 @@ translate_isl_ast_node_for (loop_p conte
     ub = integer_zero_node;
 
   edge last_e = single_succ_edge (split_edge (next_e));
+
+  /* Compensate for the fact that we emit a do { } while loop from
+     a for ISL AST.
+     ???  We often miss constraints on niter because the SESE region
+     doesn't cover loop header copies.  Ideally we'd add constraints
+     for all relevant dominating conditions.  */
+  if (TREE_CODE (lb) == INTEGER_CST && TREE_CODE (ub) == INTEGER_CST
+      && tree_int_cst_compare (lb, ub) <= 0)
+    ;
+  else
+    {
+      tree one = build_one_cst (POINTER_TYPE_P (type) ? sizetype : type);
+      /* Adding +1 and using LT_EXPR helps with loop latches that have a
+	 loop iteration count of "PARAMETER - 1".  For PARAMETER == 0 this
+	 becomes 2^k-1 due to integer overflow, and the condition lb <= ub
+	 is true, even if we do not want this.  However lb < ub + 1 is false,
+	 as expected.  */
+      tree ub_one = fold_build2 (POINTER_TYPE_P (type)
+				 ? POINTER_PLUS_EXPR : PLUS_EXPR,
+				 type, ub, one);
+      create_empty_if_region_on_edge (next_e,
+				      fold_build2 (LT_EXPR, boolean_type_node,
+						   lb, ub_one));
+      next_e = get_true_edge_from_guard_bb (next_e->dest);
+    }
+
   translate_isl_ast_for_loop (context_loop, node, next_e,
 			      type, lb, ub, ip);
   return last_e;
Index: gcc/testsuite/gcc.dg/graphite/pr83255.c
===================================================================
--- gcc/testsuite/gcc.dg/graphite/pr83255.c	(nonexistent)
+++ gcc/testsuite/gcc.dg/graphite/pr83255.c	(working copy)
@@ -0,0 +1,28 @@
+/* { dg-do run } */
+/* { dg-options "-O -floop-nest-optimize -fdump-tree-graphite-details" } */
+
+int rx, in;
+
+int
+main (void)
+{
+  const int tj = 3;
+  int as[tj];
+  static int l4;
+
+  while (l4 < 1)
+    {
+      for (rx = 0; rx < tj; ++rx)
+	{
+	  for (in = 0; in < tj; ++in)
+	    as[in] = 1;
+	  as[rx] = 0;
+	}
+      ++l4;
+    }
+
+  if (as[tj - 1] != 0)
+    __builtin_abort ();
+}
+
+/* { dg-final { scan-tree-dump "loop nest optimized" "graphite" } } */

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

only message in thread, other threads:[~2017-12-04 14:01 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-12-04 14:01 [PATCH] Fix PR83255 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).