From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gcc-bugs-return-198573-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org>
Received: (qmail 6623 invoked by alias); 2 Oct 2006 13:56:14 -0000
Received: (qmail 6557 invoked by uid 48); 2 Oct 2006 13:56:04 -0000
Date: Mon, 02 Oct 2006 13:56:00 -0000
Message-ID: <20061002135604.6556.qmail@sourceware.org>
X-Bugzilla-Reason: CC
References: <bug-27986-5748@http.gcc.gnu.org/bugzilla/>
Subject: [Bug middle-end/27986] [4.0/4.1/4.2 Regression] jump to middle of loop on entry with using old version of an variable
In-Reply-To: <bug-27986-5748@http.gcc.gnu.org/bugzilla/>
Reply-To: gcc-bugzilla@gcc.gnu.org
To: gcc-bugs@gcc.gnu.org
From: "amacleod at redhat dot com" <gcc-bugzilla@gcc.gnu.org>
Mailing-List: contact gcc-bugs-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
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
X-SW-Source: 2006-10/txt/msg00134.txt.bz2
List-Id: <gcc-bugs.sourceware.org>



------- Comment #4 from amacleod at redhat dot com  2006-10-02 13:56 -------
This is not something out of ssa can resolve on its own.

given this sequence:
  # s_2 = PHI <s_5(0), s_9(1)>;
  # d_1 = PHI <d_6(0), d_10(1)>;
<L0>:;
  D.1287_8 = MEM[base: d_1];
  s_9 = s_2 + D.1287_8;          <<<---  s_2 and s_9 have different values
  d_10 = d_1 + 4B;
  if (s_9 < g_11) goto <L0>; else goto <L1>;

  # s_7 = PHI <s_2(1)>;          <<<--- original value of s_2 used here
<L1>:;
  *v_3 = s_7;

When s_9 is assigned s_2 + D.1287_8  , they have different values, so when s_2
is used in the PHI assigned to s_7, there is a stretch over which s_9 and s_2
are both live, and contain different values.

out of ssa cannot assign them to the same variable.

you are going to have a copy in the loop no matter what you do... 

right now we get:
<bb 2>:
  s = *(v + 4B);

<L0>:;
  s.31 = s + MEM[base: d]{*d};
  d = d + 4B;
  if (s.31 < g) goto <L6>; else goto <L1>;

<L6>:;
  s = s.31;
  goto <bb 3> (<L0>);

<L1>:;
  *v = s;
  return;


the other alternative would be:
<L0>:;
  s.31 = s
  s = s + MEM[base: d]{*d};
  d = d + 4B;
  if (s < g) goto <L6>; else goto <L1>;

<L6>:;
  goto <bb 3> (<L0>);

<L1>:;
  *v = s.31;


Is that any better? It looks pretty much the same to me.  It might help on a 2
address machine where you need r1 = r1 + exp, but that about it.


so Im not sure I understand what you want out of ssa to do with the code here.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27986