[Bug tree-optimization/17742] New: C Optimization error with -O1 and -O2 on i686

[Bug tree-optimization/17742] New: C Optimization error with -O1 and -O2 on i686

[m dot calderbank at iname dot com]

The following program compiles incorrectly on i686 using the latest CVS code and
optimization level -O1 or -O2. But -O3 is OK, as is -fno-ivopts.
I am aware that bug #17474 concerns -fno-ivopts but lack the knowledge to
determine if this is a related problem.

Compiled with gcc -O1 -save-temps test.c:

--- Begin test.c ---
#define advance(f) do { ++pos; if ((f) && pos >= argc) return(2); } while (0)
int pos;
int argc; char **argv;
int term();

int
main(int margc, char **margv)
{
  argc = margc; argv = margv;
  pos = 1;
  return term();
}

int
term()
{
  int value;
  if (pos >= argc) return (2);
  if (argv[pos][0] == '!' && argv[pos][1] == '\0')
    {
      value = 0;
      while (pos < argc && argv[pos][0] == '!' && argv[pos][1] == '\0')
        {
          advance (1);
          value = 1 - value;
        }
      return (value ? !term() : term());
    }
  return (0);
}
--- End test.c ---

Output of gcc -v:

Reading specs from
/home/mark/tarballs/gcc-cvs/gcc-build/usr/lib/gcc/i686-pc-linux-gnu/4.0.0/specs
Configured with: ../gcc/configure
--prefix=/home/mark/tarballs/gcc-cvs/gcc-build/usr
--libexecdir=/home/mark/tarballs/gcc-cvs/gcc-build/usr/lib --enable-shared
--enable-languages=c
Thread model: posix
gcc version 4.0.0 20040929 (experimental)

The file test.c is cut down from the coreutils program of the same name, which
also suffers from this bug. When supplied with three or more arguments of which
at least the first two are exclamation marks, the program enters an infinite
loop. e.g.

./a.out ! ! foo

I've included a portion of my assembly output below to point out what I think is
going wrong. Line (1) earmarks register %ecx for the 'value' variable. Line (2)
is supposed to be performing ++pos, but in fact performs "pos = original-pos +
value + 1". Lines (3) flip "value" between 0 and 1. Of course this breaks after
two iterations, and the loop never exits.

        movl    $0, %ecx               ; (1)
        leal    0(,%eax,4), %ebx
.L13:
        leal    1(%edi,%ecx), %eax     ; (2)
        movl    %eax, pos
        cmpl    %eax, -16(%ebp)
        jg      .L14
        movl    $2, %eax
        jmp     .L8
.L14:
        movl    $1, %eax               ; (3)
        subl    %ecx, %eax             ; (3)
        movl    %eax, %edx             ; (3)
        movl    %eax, %ecx             ; (3)
        leal    (%ebx,%eax,4), %eax
        movl    (%eax,%esi), %eax
        cmpb    $33, (%eax)
        jne     .L16
        cmpb    $0, 1(%eax)
        je      .L13

-- 
           Summary: C Optimization error with -O1 and -O2 on i686
           Product: gcc
           Version: 4.0.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: tree-optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: m dot calderbank at iname dot com
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17742

[Bug tree-optimization/17742] [4.0 Regression] C Optimization error with -O1 on i686

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-09-29 22:46 -------
It does look like -fno-ivopts fixes the problem and it does not look related to PR 17474 at all.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rakdver at gcc dot gnu dot
                   |                            |org
           Keywords|                            |wrong-code
            Summary|C Optimization error with - |[4.0 Regression] C
                   |O1 and -O2 on i686          |Optimization error with -O1
                   |                            |on i686
   Target Milestone|---                         |4.0.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17742

[Bug tree-optimization/17742] [4.0 Regression] C Optimization error with -O1 on i686

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-09-30 21:46 -------
Hmm, this fails for me with -O0 on powerpc-apple-darwin.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17742

[Bug tree-optimization/17742] [4.0 Regression] C Optimization error with -O1 on i686

[m dot calderbank at iname dot com]

------- Additional Comments From m dot calderbank at iname dot com  2004-10-01 00:21 -------
Actually -O3 is not OK as I thought at first: it produces wrong return codes 
rather than going infinite.
Inspection of assembly and test.c.00.expand shows that the problem is the same: 
it's treating pos and value as equivalent.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17742

[Bug tree-optimization/17742] [4.0 Regression] C Optimization error with -O1 on i686

[reichelt at gcc dot gnu dot org]

------- Additional Comments From reichelt at gcc dot gnu dot org  2004-11-08 23:20 -------
Here's a reduced testcase. It goes into an infinite loop when compiled
with optimization (-O, -O2, -O3, -Os).

=====================
int i=0;

int main()
{
    int j=0;
    while (++i<3)
        j = 1-j;
    return j;
}
=====================


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |reichelt at gcc dot gnu dot
                   |                            |org
           Severity|normal                      |critical
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
           Keywords|                            |monitored
   Last reconfirmed|0000-00-00 00:00:00         |2004-11-08 23:20:37
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17742

[Bug tree-optimization/17742] [4.0 Regression] C Optimization error with -O1 on i686

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-11-08 23:25 -------
This is most likely the same as PR 17926.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
OtherBugsDependingO|                            |17926
              nThis|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17742

[Bug tree-optimization/17742] [4.0 Regression] C Optimization error with -O1 on i686

[reichelt at gcc dot gnu dot org]

------- Additional Comments From reichelt at gcc dot gnu dot org  2004-11-08 23:38 -------
Zdenek, the bug really appears with your ivopts patch
http://gcc.gnu.org/ml/gcc-cvs/2004-09/msg00234.html


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17742

[Bug tree-optimization/17742] [4.0 Regression] C Optimization error with -O1 on i686

[rakdver at gcc dot gnu dot org]

------- Additional Comments From rakdver at gcc dot gnu dot org  2004-11-09 00:15 -------
This is a problem due to scev. It claims that j is an induction variable with
step 1. Ummm....

Patch in testing.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17742

[Bug tree-optimization/17742] [4.0 Regression] C Optimization error with -O1 on i686

[rakdver at gcc dot gnu dot org]

------- Additional Comments From rakdver at gcc dot gnu dot org  2004-11-09 11:46 -------
Patch:

http://gcc.gnu.org/ml/gcc-patches/2004-11/msg00716.html

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |patch


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17742

[Bug tree-optimization/17742] [4.0 Regression] C Optimization error with -O1 on i686

[cvs-commit at gcc dot gnu dot org]

------- Additional Comments From cvs-commit at gcc dot gnu dot org  2004-11-11 20:12 -------
Subject: Bug 17742

CVSROOT:	/cvs/gcc
Module name:	gcc
Changes by:	rakdver@gcc.gnu.org	2004-11-11 20:12:36

Modified files:
	gcc            : ChangeLog tree-scalar-evolution.c 

Log message:
	PR tree-optimization/17742
	* tree-scalar-evolution.c (follow_ssa_edge_in_rhs): Handle
	MINUS_EXPR correctly.

Patches:
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/ChangeLog.diff?cvsroot=gcc&r1=2.6279&r2=2.6280
http://gcc.gnu.org/cgi-bin/cvsweb.cgi/gcc/gcc/tree-scalar-evolution.c.diff?cvsroot=gcc&r1=2.11&r2=2.12



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17742

[Bug tree-optimization/17742] [4.0 Regression] C Optimization error with -O1 on i686

[rakdver at gcc dot gnu dot org]

------- Additional Comments From rakdver at gcc dot gnu dot org  2004-11-11 20:14 -------
Fixed.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17742