From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gcc-bugs-return-158851-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org>
Received: (qmail 15677 invoked by alias); 19 Sep 2005 18:52:21 -0000
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
Received: (qmail 15600 invoked by uid 48); 19 Sep 2005 18:52:14 -0000
Date: Mon, 19 Sep 2005 18:52:00 -0000
From: "rguenth at gcc dot gnu dot org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Message-ID: <20050919185212.23970.rguenth@gcc.gnu.org>
Reply-To: gcc-bugzilla@gcc.gnu.org
Subject: [Bug tree-optimization/23970] New: loop-invariant-motion is not doing it's work
X-Bugzilla-Reason: CC
X-SW-Source: 2005-09/txt/msg02445.txt.bz2
List-Id: <gcc-bugs.sourceware.org>

void Ekin(double *e, int *stridee,
          double *vx, int *stridevx,
          double *vy, int *stridevy,
          double *vz, int *stridevz,
          int *sz)
{
  int i1 = sz[0];
  int j1 = sz[1];
  int k1 = sz[2];
  int i, j, k;
  for (k=0; k<k1; ++k)
   for (j=0; j<j1; ++j)
    for (i=0; i<i1; ++i)
     {
       e[i + j * stridee[1] + k * stridee[2]]
        = 0.128 * (
            ((vx[i + j * stridevx[1] + k * stridevx[2]]
              + vx[i+1 + j * stridevx[1] + k * stridevx[2]])
             * (vx[i + j * stridevx[1] + k * stridevx[2]]
                + vx[i+1 + j * stridevx[1] + k * stridevx[2]]))
          + ((vy[i + j * stridevy[1] + k * stridevy[2]]
              + vy[i + (j+1) * stridevy[1] + k * stridevy[2]])
             * (vy[i + j * stridevy[1] + k * stridevy[2]]
                + vy[i + (j+1) * stridevy[1] + k * stridevy[2]]))
          + ((vz[i + j * stridevz[1] + k * stridevz[2]]
              + vz[i + j * stridevz[1] + (k+1) * stridevz[2]])
             * (vz[i + j * stridevz[1] + k * stridevz[2]]
                + vz[i + j * stridevz[1] + (k+1) * stridevz[2]])));

     }
}

lim moves all the j*stridev?[1] and k*stridev?[2] to the j-loop but
does not move the k*stridev?[2] to the k-loop.  This results in the
following asm.

Ekin:
        pushl   %ebp
        movl    %esp, %ebp
        pushl   %edi
        pushl   %esi
        pushl   %ebx
        subl    $44, %esp
        movl    40(%ebp), %eax
        movl    $0, -24(%ebp)
        movl    (%eax), %edx
        movl    4(%eax), %ecx
        movl    8(%eax), %eax
        movl    %edx, -40(%ebp)
        movl    %ecx, -36(%ebp)
        testl   %eax, %eax
        movl    %eax, -32(%ebp)
        jle     .L13
.L4:
        movl    -24(%ebp), %edx
        movl    -36(%ebp), %eax
        incl    %edx
        testl   %eax, %eax
        movl    %edx, -48(%ebp)
        jle     .L7
        movl    -40(%ebp), %ecx
        movl    $0, -16(%ebp)
        testl   %ecx, %ecx
        jle     .L17
.L10:
        movl    28(%ebp), %eax
        movl    -24(%ebp), %ebx
        movl    -16(%ebp), %edi
        movl    $0, -56(%ebp)
        fldl    .LC0
        movl    8(%eax), %edx
        movl    4(%eax), %ecx
        incl    %edi
        movl    %edi, -28(%ebp)
        movl    -16(%ebp), %edi
        imull   %edx, %ebx
        movl    36(%ebp), %edx
        movl    4(%edx), %eax
        movl    8(%edx), %esi
        movl    12(%ebp), %edx
        imull   %eax, %edi
        movl    -16(%ebp), %eax
        movl    %edi, -44(%ebp)
        movl    4(%edx), %edi
        movl    -24(%ebp), %edx
        imull   %edi, %eax
        movl    12(%ebp), %edi
        imull   8(%edi), %edx
        addl    %edx, %eax
        movl    -16(%ebp), %edx
        sall    $3, %eax
        movl    %eax, -20(%ebp)
        movl    20(%ebp), %eax
        movl    4(%eax), %edi
        movl    -24(%ebp), %eax
        imull   %edi, %edx
        movl    20(%ebp), %edi
        imull   8(%edi), %eax
        addl    %eax, %edx
        movl    -16(%ebp), %eax
        imull   %ecx, %eax
        addl    %ebx, %eax
        leal    0(,%eax,8), %edi
        movl    -28(%ebp), %eax
        imull   %eax, %ecx
        movl    -24(%ebp), %eax
        addl    %ecx, %ebx
        movl    -44(%ebp), %ecx
        imull   %esi, %eax
        sall    $3, %ebx
        addl    %ecx, %eax
        movl    -48(%ebp), %ecx
        sall    $3, %eax
        movl    %eax, -52(%ebp)
        imull   %ecx, %esi
        movl    -44(%ebp), %ecx
        addl    %esi, %ecx
        leal    0(,%ecx,8), %esi
        .p2align 4,,15
.L5:
        movl    16(%ebp), %eax
        movl    24(%ebp), %ecx
        incl    -56(%ebp)
        fldl    (%eax,%edx,8)
        faddl   8(%eax,%edx,8)
        incl    %edx
        movl    -52(%ebp), %eax
        addl    $8, -52(%ebp)
        fldl    (%edi,%ecx)
        addl    $8, %edi
        faddl   (%ecx,%ebx)
        addl    $8, %ebx
        movl    32(%ebp), %ecx
        fldl    (%eax,%ecx)
        faddl   (%ecx,%esi)
        fxch    %st(2)
        addl    $8, %esi
        movl    -20(%ebp), %eax
        movl    8(%ebp), %ecx
        fmul    %st(0), %st
        fxch    %st(1)
        fmul    %st(0), %st
        faddp   %st, %st(1)
        fxch    %st(1)
        fmul    %st(0), %st
        faddp   %st, %st(1)
        fmul    %st(1), %st
        fstpl   (%eax,%ecx)
        addl    $8, %eax
        movl    %eax, -20(%ebp)
        movl    -56(%ebp), %eax
        cmpl    %eax, -40(%ebp)
        jne     .L5
        fstp    %st(0)
        movl    -28(%ebp), %edx
        cmpl    %edx, -36(%ebp)
        jle     .L7
.L18:
        movl    -40(%ebp), %ecx
        movl    %edx, -16(%ebp)
        testl   %ecx, %ecx
        jg      .L10
.L17:
        movl    -16(%ebp), %ecx
        incl    %ecx
        movl    %ecx, -28(%ebp)
        movl    -28(%ebp), %edx
        cmpl    %edx, -36(%ebp)
        jg      .L18
.L7:
        movl    -48(%ebp), %eax
        cmpl    %eax, -32(%ebp)
        movl    %eax, -24(%ebp)
        jne     .L4
.L13:
        addl    $44, %esp
        popl    %ebx
        popl    %esi
        popl    %edi
        popl    %ebp
        ret

-- 
           Summary: loop-invariant-motion is not doing it's work
           Product: gcc
           Version: 4.1.0
            Status: UNCONFIRMED
          Keywords: missed-optimization
          Severity: normal
          Priority: P2
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: rguenth at gcc dot gnu dot org
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23970