c/7527: gcc 2.95.3 and gcc 3.0.4 generate incorrect code

c/7527: gcc 2.95.3 and gcc 3.0.4 generate incorrect code

[Thomas D. Dean]

>Number:         7527
>Category:       c
>Synopsis:       gcc 2.95.3 and gcc 3.0.4 generate incorrect code
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Wed Aug 07 10:26:00 PDT 2002
>Closed-Date:
>Last-Modified:
>Originator:     Thomas D. Dean
>Release:        3.0.4 m68hc1x-20020616
>Organization:
>Environment:
System: FreeBSD asus 4.6-STABLE FreeBSD 4.6-STABLE #7: Mon Jul 22 14:18:44 PDT 2002 root@asus:/usr/src/sys/compile/ASUS i386


	
host: i386-unknown-freebsd4.6
build: i386-unknown-freebsd4.6
target: m6811-unknown-none
configured with: ./configure --target=m6811-elf --program-prefix=m6811-elf- --enable-languages=c,c++
>Description:
	
Compiling the simple 10 statement c-source, below, causes incorrect
code to be generated.  The 40 lines of comments in the code explain
details.

The statement 'if(!_io_ports[0x59]&0x80) {...}' is compiled incorrectly.

The problem was first noticed with 

# m6811-elf-gcc -v
Reading specs from /usr/local/lib/gcc-lib/m6811-elf/3.0.4/specs
Configured with: ./configure --target=m6811-elf --program-prefix=m6811-elf- \
                 --enable-languages=c,c++
Thread model: single
gcc version 3.0.4 m68hc1x-20020616

and reproduced with 

# gcc -v
Using builtin specs.
gcc version 2.95.3 20010315 (release) [FreeBSD]

tomdean

/*
 * example to generate incorrect code.
 *
 * compiled with `gcc -Wall -o f f.c
 */

volatile unsigned char *_io_ports = (unsigned char *)0x7500;

int main() {
  /*
   * produces correct code
   */
  if(!(_io_ports[0x39]&0x80)) {
	_io_ports[0x29] = 0xff;
  }
  /*
   * produces incorrect code - accesses _io_ports[0x59], but,
   * the if() is ignored.  There is no value testing or conditional
   * branch in the elf file.  The above code is the preferred syntax,
   * but, the below code should do something with the value.
   *
   * at 8048485 in the output of objdump, below, _io_ports[0x59] is
   * accessed as 0x59(%eax).  The next line, 8048488, is the start of
   * 'return 0'.  There should be some form of conditional between.
   *
   */
  if(!_io_ports[0x59]&0x80) {
	_io_ports[0x49] = 0;
  }

  return 0;
}

/*
 * objdump -d f  shows
 * ...
 *
 * 0804846c <main>:
 *  804846c:       55                      push   %ebp
 *  804846d:       89 e5                   mov    %esp,%ebp
 *  804846f:       8b 15 d8 94 04 08       mov    0x80494d8,%edx
 *  8048475:       8a 42 39                mov    0x39(%edx),%al
 *  8048478:       84 c0                   test   %al,%al
 *  804847a:       7c 04                   jl     8048480 <main+0x14>
 *  804847c:       c6 42 29 ff             movb   $0xff,0x29(%edx)
 *  8048480:       a1 d8 94 04 08          mov    0x80494d8,%eax
 *  8048485:       8a 40 59                mov    0x59(%eax),%al
 *  8048488:       31 c0                   xor    %eax,%eax
 *  804848a:       c9                      leave  
 *  804848b:       c3                      ret    
 *
 * ...
 */

>How-To-Repeat:
	
Compile the source code above
>Fix:
	
The first conditional, 'if(!(_io_ports[0x39]&0x80))' generates correct code.
The 2nd conditional,   'if(!_io_ports[0x39]&0x80)' generates incorrect code.
>Release-Note:
>Audit-Trail:
>Unformatted:
 medium
 wrong-code

Re: c/7527: gcc 2.95.3 and gcc 3.0.4 generate incorrect code

[Stian]

The following reply was made to PR c/7527; it has been noted by GNATS.

From: Stian <stian@knowledge-network.no>
To: gcc-gnats@gcc.gnu.org
Cc: gcc-prs@gcc.gnu.org, gcc-bugs@gcc.gnu.org, nobody@gcc.gnu.org, 
	tomdean@asus.tddhome
Subject: Re: c/7527: gcc 2.95.3 and gcc 3.0.4 generate incorrect code
Date: 08 Aug 2002 12:08:55 +0200

 The second conditional makes more sense if you use () at the right places
 
 if(!(_io_ports[0x39]&0x80))
 
 
 If you don't add them, your statement is the same as
 
 if((!_io_ports[0x39])&0x80)
 
 That always come out false, then ! makes the result of the io-read either 0 or 1.
 
 
 
 Stian
 
 
 

Re: c/7527: gcc 2.95.3 and gcc 3.0.4 generate incorrect code

[rth]

Synopsis: gcc 2.95.3 and gcc 3.0.4 generate incorrect code

State-Changed-From-To: open->closed
State-Changed-By: rth
State-Changed-When: Wed Aug  7 11:16:27 2002
State-Changed-Why:
    Not a bug.  You are confused about operator precedence.

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=7527