From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gcc-bugs-return-88472-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org>
Received: (qmail 22435 invoked by alias); 9 Mar 2004 17:17:04 -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 22421 invoked by uid 48); 9 Mar 2004 17:17:03 -0000
Date: Tue, 09 Mar 2004 17:17:00 -0000
From: "stl at caltech dot edu" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Message-ID: <20040309171658.14504.stl@caltech.edu>
Reply-To: gcc-bugzilla@gcc.gnu.org
Subject: [Bug optimization/14504] New: Missed Bit Twiddling Optimization
X-Bugzilla-Reason: CC
X-SW-Source: 2004-03/txt/msg01129.txt.bz2
List-Id: <gcc-bugs.sourceware.org>

g++ 3.3.2 on i686-pc-linux-gnu with -s -O3 -fomit-frame-pointer compiles:

unsigned long cond_mask(bool flag, unsigned long mask, unsigned long target) {
    return flag ? target | mask : target & ~mask;
}

into:

cmpb    $0, 4(%esp)
movl    8(%esp), %eax
movl    12(%esp), %edx
je      .L2
orl     %edx, %eax
ret
.L2:
notl    %eax
andl    %edx, %eax
ret

This appears to be a straightforward translation of the code into assembly. 
With the same options, this:

unsigned long cond_mask(bool flag, unsigned long mask, unsigned long target) {
    return (mask | target ^ 0xFFFFFFFFUL + flag) ^ 0xFFFFFFFFUL + flag;
}

is compiled into:

movzbl  4(%esp), %edx
movl    12(%esp), %eax
movl    8(%esp), %ecx
decl    %edx
xorl    %edx, %eax
orl     %ecx, %eax
xorl    %edx, %eax
ret

Again this appears to be a straightforward translation of the code into 
assembly (instead of flag + 0xFFFFFFFFUL, it uses static_cast<unsigned long>
(flag) - 1, which is the same thing).

However, the rewritten code lacks a branch yet does the exact same thing.

g++ ought to be able to perform this reasonably simple transformation on its 
own - if, indeed, the transformation is desirable (which I think it is).

NB:
I do not know assembly, so I may have deleted important information from g++'s 
output. I don't think I have, however.

This case was suggested by ryanm@microsoft.com, who said "this is something I 
will never expect a compiler to be able to optimize for me.  ^_^". I wrote the 
transformed code; perhaps there is a cleverer way to transform it which I am 
not aware of.

-- 
           Summary: Missed Bit Twiddling Optimization
           Product: gcc
           Version: 3.3.2
            Status: UNCONFIRMED
          Severity: enhancement
          Priority: P2
         Component: optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: stl at caltech dot edu
                CC: gcc-bugs at gcc dot gnu dot org,ryanm at microsoft dot
                    com
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: i686-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14504