From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gcc-bugs-return-411848-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org>
Received: (qmail 8380 invoked by alias); 10 Jan 2013 17:47:12 -0000
Received: (qmail 8255 invoked by uid 48); 10 Jan 2013 17:46:42 -0000
From: "aldyh at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug target/55565] [4.8 regression] FAIL: gcc.target/powerpc/ppc-mov-1.c scan-assembler-not fmr [0-9]+,[0-9]+
Date: Thu, 10 Jan 2013 17:47:00 -0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: changed
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: target
X-Bugzilla-Keywords:
X-Bugzilla-Severity: normal
X-Bugzilla-Who: aldyh at gcc dot gnu.org
X-Bugzilla-Status: NEW
X-Bugzilla-Priority: P1
X-Bugzilla-Assigned-To: aldyh at gcc dot gnu.org
X-Bugzilla-Target-Milestone: 4.8.0
X-Bugzilla-Changed-Fields:
Message-ID: <bug-55565-4-ywetSNyXja@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-55565-4@http.gcc.gnu.org/bugzilla/>
References: <bug-55565-4@http.gcc.gnu.org/bugzilla/>
X-Bugzilla-URL: http://gcc.gnu.org/bugzilla/
Auto-Submitted: auto-generated
Content-Type: text/plain; charset="UTF-8"
MIME-Version: 1.0
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
X-SW-Source: 2013-01/txt/msg00967.txt.bz2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55565

--- Comment #4 from Aldy Hernandez <aldyh at gcc dot gnu.org> 2013-01-10 17:46:39 UTC ---
I compared the code generated by trunk with the generated code in rev 190339
which broke the test.  The trunk code is more optimal than when the test
"passed", so I suggest either removing the dg-final in the test, or modifying
the regex in the dg-final.

In the testcase we have something functionally equivalent to (where x is the
original function argument passed in f1):

  if (huge + x > 0.0) {
    if ((int)x > 0)
      return 0.0;
    ...
  } else
    return x;

When the test passed, we used to generate:

.L17:
        ld %r8,.LC0@toc(%r2)
        ld %r10,.LC2@toc(%r2)
        lfd %f13,0(%r8)
        lfd %f0,0(%r10)
        fadd %f1,%f1,%f13
        fcmpu %cr7,%f1,%f0
        bng %cr7,.L4           /* if (!(huge + x > 0.0)) return x; */
        ...
        ...

.L4:                           /* return x; */
        std %r9,-8(%r1)
        ori 2,2,0
        lfd %f1,-8(%r1)
        blr

Notice that since we perform the fadd onto %f1, we need to reload %f1 from
memory.

On trunk though, we keep x/%f1 and 0.0/%f0 around so we can return them
directly.  We can return x/%f1 directly in one case, or copy 0.0/%f0 to the
return register (%f1) in the another case.

.L18:
        ld %r8,.LC6@toc(%r2)
        ld %r9,.LC1@toc(%r2)
        lfd %f13,0(%r8)
        lfd %f0,0(%r9)
        fadd %f13,%f1,%f13
        fcmpu %cr7,%f13,%f0
        bnglr %cr7            /* if (!(huge + x > 0.0)) return x; */
        cmpdi %cr7,%r10,0
        fmr %f1,%f0
        bgelr %cr7            /* if (huge+x>0.0 && (int)x >0) return 0.0 */

Bottom line, on trunk we avoid a branch and memory load/stores.

Testing a patch.