From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gcc-bugs-return-435219-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org>
Received: (qmail 30186 invoked by alias); 20 Nov 2013 12:08:26 -0000
Mailing-List: contact gcc-bugs-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
List-Id: <gcc-bugs.gcc.gnu.org>
List-Archive: <http://gcc.gnu.org/ml/gcc-bugs/>
List-Post: <mailto:gcc-bugs@gcc.gnu.org>
List-Help: <mailto:gcc-bugs-help@gcc.gnu.org>
Sender: gcc-bugs-owner@gcc.gnu.org
Received: (qmail 30151 invoked by uid 48); 20 Nov 2013 12:08:23 -0000
From: "rguenth at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug tree-optimization/59058] [4.7/4.8/4.9 Regression] wrong code at -O3 on x86_64-linux-gnu (affecting gcc 4.6 to trunk)
Date: Wed, 20 Nov 2013 12:08:00 -0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: changed
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: tree-optimization
X-Bugzilla-Version: 4.9.0
X-Bugzilla-Keywords: wrong-code
X-Bugzilla-Severity: normal
X-Bugzilla-Who: rguenth at gcc dot gnu.org
X-Bugzilla-Status: ASSIGNED
X-Bugzilla-Priority: P2
X-Bugzilla-Assigned-To: rguenth at gcc dot gnu.org
X-Bugzilla-Target-Milestone: 4.7.4
X-Bugzilla-Flags:
X-Bugzilla-Changed-Fields:
Message-ID: <bug-59058-4-UoNbOZ4f3Q@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-59058-4@http.gcc.gnu.org/bugzilla/>
References: <bug-59058-4@http.gcc.gnu.org/bugzilla/>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit
X-Bugzilla-URL: http://gcc.gnu.org/bugzilla/
Auto-Submitted: auto-generated
MIME-Version: 1.0
X-SW-Source: 2013-11/txt/msg01996.txt.bz2

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59058

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
Unfortunately

@@ -2930,11 +2931,31 @@ number_of_exit_cond_executions (struct l
   if (chrec_contains_undetermined (ret))
     return ret;

-  ret = chrec_fold_plus (type, ret, build_int_cst (type, 1));
-  if (TREE_CODE (ret) == INTEGER_CST
-      && TREE_OVERFLOW (ret))
+  /* Handle constants without widening.  */
+  if (TREE_CODE (ret) == INTEGER_CST)
+    {
+      double_int adj = tree_to_double_int (ret) + double_int_one;
+      if (double_int_fits_to_tree_p (type, adj))
+       return double_int_to_tree (type, adj);
+    }
+
+  /* For the remaining case widen to an unsigned type from a
+     signed one or to one with at least one bit more precision
+     but not larger than using word_mode.  */
+  if (!TYPE_UNSIGNED (type))
+    type = unsigned_type_for (type);
+  else if (GET_MODE_CLASS (TYPE_MODE (type)) == MODE_INT
+          && TYPE_PRECISION (type) < GET_MODE_PRECISION (word_mode))
+    {
+      enum machine_mode mode;
+      mode = smallest_mode_for_size (TYPE_PRECISION (type) + 1, MODE_INT);
+      type = build_nonstandard_integer_type (GET_MODE_PRECISION (mode), 1);
+    }
+  else
     return chrec_dont_know;

+  ret = chrec_convert (type, ret, NULL);
+  ret = chrec_fold_plus (type, ret, build_int_cst (type, 1));
   return ret;
 }

breaks loops which use size_t as induction variable type.  But they
are possibly miscompiled anyway.  Not sure if we can rely on
the availability of arithmetics on modes > word_mode, can we?