From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gcc-prs-return-16773-listarch-gcc-prs=gcc.gnu.org@gcc.gnu.org>
Received: (qmail 25255 invoked by alias); 3 Apr 2002 18:06:02 -0000
Mailing-List: contact gcc-prs-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
List-Archive: <http://gcc.gnu.org/ml/gcc-prs/>
List-Post: <mailto:gcc-prs@gcc.gnu.org>
List-Help: <mailto:gcc-prs-help@gcc.gnu.org>
Sender: gcc-prs-owner@gcc.gnu.org
Received: (qmail 25224 invoked by uid 71); 3 Apr 2002 18:06:01 -0000
Date: Wed, 03 Apr 2002 10:06:00 -0000
Message-ID: <20020403180601.25223.qmail@sources.redhat.com>
To: nobody@gcc.gnu.org
Cc: gcc-prs@gcc.gnu.org,
From: Dan Nicolaescu <dann@godzilla.ICS.UCI.EDU>
Subject: Re: optimization/5738: GCSE missed optimization
Reply-To: Dan Nicolaescu <dann@godzilla.ICS.UCI.EDU>
X-SW-Source: 2002-04/txt/msg00237.txt.bz2
List-Id: <gcc-prs.sourceware.org>

The following reply was made to PR optimization/5738; it has been noted by GNATS.

From: Dan Nicolaescu <dann@godzilla.ICS.UCI.EDU>
To: Daniel Berlin <dan@dberlin.org>
Cc: rth@gcc.gnu.org, gcc-bugs@gcc.gnu.org, gcc-prs@gcc.gnu.org, 
    nobody@gcc.gnu.org, gcc-gnats@gcc.gnu.org
Subject: Re: optimization/5738: GCSE missed optimization
Date: Wed, 03 Apr 2002 10:01:16 -0800

 Daniel Berlin <dan@dberlin.org> writes:
 
   > On 3 Apr 2002 rth@gcc.gnu.org wrote:
   > 
   > > Synopsis: GCSE missed optimization
   > > 
   > > State-Changed-From-To: open->closed
   > > State-Changed-By: rth
   > > State-Changed-When: Wed Apr  3 02:25:09 2002
   > > State-Changed-Why:
   > >     That's not how partial redundancy elimination (PRE) works.
   > >     The object with PRE is to minimize the number of evaluations
   > >     of an expression *along a path*. 
   > 
   > Please don't close this PR, it's correct.
 
 I want to add one more data point to this: the cfg-branch does a
 little better on this testcase, it finds some of the common code on
 the 2 branches of the if, but not all of it. 
 
 struct foo { unsigned short * p;};
 
 unsigned short
 foo (struct foo *s, unsigned long *coord, _Bool delta)
 {
   unsigned short change;
   if (delta) {
     change = *((s->p)++);
     coord +=  change;
     return change;
   } else {
     change = *((s->p)++);
     coord +=  change; 
     coord += *((s)->p++) <<  8;
     return change;
   }
 }
 
 the generated SPARC assembly for "foo" is: 
 
 CVS HEAD gcc -O2                   cfg-branch gcc -O2
 foo:                               foo:
         !#PROLOGUE# 0                      !#PROLOGUE# 0
         !#PROLOGUE# 1                      save    %sp, -112, %sp  
         andcc   %o2, 0xff, %g0             !#PROLOGUE# 1
         be      .LL2                       andcc   %i2, 0xff, %g0
         mov     %o0, %o3                   be      .LL2
         ld      [%o0], %o0                 mov     %i0, %i3
         lduh    [%o0], %o2                 ld      [%i0], %i0
         add     %o0, 2, %o0                b       .LL4
         sll     %o2, 16, %o1               add     %i0, 2, %i2
         st      %o0, [%o3]         .LL2:
         b       .LL1                       ld      [%i0], %i0
         srl     %o1, 16, %o0               add     %i0, 4, %i2
 .LL2:                              .LL4:
         ld      [%o0], %o0                 lduh    [%i0], %i1
         lduh    [%o0], %o2                 st      %i2, [%i3]
         add     %o0, 4, %o1                sll     %i1, 16, %i1
         st      %o1, [%o3]                 srl     %i1, 16, %i0
         mov     %o2, %o0                   ret
 .LL1:                                      restore
         retl                       
         nop                        
 
 
 the function "foo" above should be roughly simplified to:
 
 unsigned short
 bar (struct foo *s, unsigned long *coord, _Bool delta)
 {
   unsigned short change;
   change = *((s->p)++);
   coord +=  change;
   if (delta)
     ;
   else 
     coord += *((s)->p++) <<  8;
   return change;
 }
 
 bar:
         !#PROLOGUE# 0
         !#PROLOGUE# 1
         ld      [%o0], %o1
         mov     %o0, %o3
         lduh    [%o1], %o0
         andcc   %o2, 0xff, %g0
         add     %o1, 2, %o1
         sll     %o0, 16, %o0
         st      %o1, [%o3]
         srl     %o0, 16, %o0
         bne     .LL6
         add     %o1, 2, %o2
         st      %o2, [%o3]
 .LL6:
         retl
         nop