public inbox for gcc-prs@sourceware.org
help / color / mirror / Atom feed
* Re: optimization/3387: GCC 3.0 optimisation -O2 -fPIC produces incorrect code
@ 2001-09-09 10:44 sirl
0 siblings, 0 replies; 4+ messages in thread
From: sirl @ 2001-09-09 10:44 UTC (permalink / raw)
To: gcc-bugs, gcc-prs, nobody, veillard, wbrack
Synopsis: GCC 3.0 optimisation -O2 -fPIC produces incorrect code
State-Changed-From-To: open->feedback
State-Changed-By: sirl
State-Changed-When: Sun Sep 9 10:44:30 2001
State-Changed-Why:
With current gcc-3_0-branch this testcase seems to produce correct code, see below. Can you please verify that the current gcc-3_0-branch works for you in the original context too?
.file "xml2bug.c"
.text
.align 16
.globl bug
.type bug,@function
bug:
pushl %ebp
movl %esp, %ebp
pushl %edi
pushl %esi
pushl %ebx
subl $20, %esp
call .LPR0
addl $_GLOBAL_OFFSET_TABLE_, %ebx
movl 8(%ebp), %edi
leal 12(%ebp), %esi
pushl %esi
pushl %edi
call f1@PLT
addl $16, %esp
testl %eax, %eax
je .L3
subl $8, %esp
pushl %esi
pushl %edi
call f2@PLT
addl $16, %esp
testl %eax, %eax
je .L3
.L1:
leal -12(%ebp), %esp
popl %ebx
popl %esi
popl %edi
popl %ebp
ret
.p2align 4,,7
.L3:
subl $8, %esp
pushl %esi
pushl %edi
call f2@PLT
xorl %eax, %eax
jmp .L1
.Lfe1:
.size bug,.Lfe1-bug
.LPR0:
movl (%esp), %ebx
ret
.ident "GCC: (GNU) 3.0.2 20010909 (prerelease)"
http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view&pr=3387&database=gcc
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: optimization/3387: GCC 3.0 optimisation -O2 -fPIC produces incorrect code
@ 2001-09-12 14:24 sirl
0 siblings, 0 replies; 4+ messages in thread
From: sirl @ 2001-09-12 14:24 UTC (permalink / raw)
To: gcc-bugs, gcc-prs, nobody, veillard, wbrack
Synopsis: GCC 3.0 optimisation -O2 -fPIC produces incorrect code
State-Changed-From-To: feedback->closed
State-Changed-By: sirl
State-Changed-When: Wed Sep 12 14:24:28 2001
State-Changed-Why:
Fix backported to branch, will be in 3.0.2.
http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view&pr=3387&database=gcc
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: optimization/3387: GCC 3.0 optimisation -O2 -fPIC produces incorrect code
@ 2001-06-22 19:56 William M. Brack
0 siblings, 0 replies; 4+ messages in thread
From: William M. Brack @ 2001-06-22 19:56 UTC (permalink / raw)
To: nobody; +Cc: gcc-prs
The following reply was made to PR optimization/3387; it has been noted by GNATS.
From: "William M. Brack" <wbrack@mmm.com.hk>
To: <gcc-gnats@gcc.gnu.org>
Cc:
Subject: Re: optimization/3387: GCC 3.0 optimisation -O2 -fPIC produces incorrect code
Date: Sat, 23 Jun 2001 10:50:59 +0800
This is a multi-part message in MIME format.
------=_NextPart_000_0046_01C0FBD2.637B41E0
Content-Type: text/plain;
charset="iso-8859-1"
Content-Transfer-Encoding: 7bit
Looking at the bug database for this report, it appears that the attached
file was somehow lost during the web submission. If true, please add it to
the PR for me; if false, please ignore this (and excuse my ignorance).
W. Brack
------=_NextPart_000_0046_01C0FBD2.637B41E0
Content-Type: application/octet-stream;
name="bug1.c"
Content-Transfer-Encoding: quoted-printable
Content-Disposition: attachment;
filename="bug1.c"
/*=0D
The following program generates incorrect code (judged by reading the =
generated=0D
assembler source) when compiled with flags -O2 and -fPIC. It seems to =
compile=0D
correctly with flags -O1 and -fPIC, as well as with flags -O2 only.=0D
=0D
The critical problem is that the routine "f1" modifies the referenced =
param (str),=0D
so the restoration of the initial value is required ( "str =3D tmp;" ). =
However,=0D
the generated code deletes that statement.=0D
=0D
Please note: This fragment is completely artificial - it is a VERY =
reduced excerpt=0D
from a reasonably complex program in which the problem caused complete =
failure.=0D
Any criticism about about "poor coding" will be taken as a personal =
insult ;>)=0D
*/=0D
=0D
int f1(int uri, char **str);=0D
int f2(int uri, char **str);=0D
=0D
int bug(int uri, char *str) {=0D
int ret;=0D
char *tmp =3D str; /* Save the initial value of param str */=0D
=0D
ret =3D f1(uri, &str); /* This routine modifies the value of str =
*/=0D
if (ret !=3D 0) {=0D
str =3D tmp; /* Restore the initial value of str */=0D
ret =3D f2(uri, &str);=0D
}=0D
if (ret !=3D 0)=0D
return(ret);=0D
=0D
ret =3D f2(uri, &str);=0D
=0D
return(0);=0D
}=0D
------=_NextPart_000_0046_01C0FBD2.637B41E0--
^ permalink raw reply [flat|nested] 4+ messages in thread
* optimization/3387: GCC 3.0 optimisation -O2 -fPIC produces incorrect code
@ 2001-06-22 17:56 wbrack
0 siblings, 0 replies; 4+ messages in thread
From: wbrack @ 2001-06-22 17:56 UTC (permalink / raw)
To: gcc-gnats; +Cc: veillard
>Number: 3387
>Category: optimization
>Synopsis: GCC 3.0 optimisation -O2 -fPIC produces incorrect code
>Confidential: no
>Severity: serious
>Priority: medium
>Responsible: unassigned
>State: open
>Class: wrong-code
>Submitter-Id: net
>Arrival-Date: Fri Jun 22 17:56:00 PDT 2001
>Closed-Date:
>Last-Modified:
>Originator: wbrack@mmm.com.hk
>Release: GCC-3.0
>Organization:
>Environment:
i686-pc-linux-gnu
>Description:
Under optimisation, one necessary assignment statement (judged by reading the assembler output) is deleted. Further comment in attached file, checked with web interface to codesourcery as demonstrating the problem with (latest) compiler.
>How-To-Repeat:
compile the attached (small) routine
>Fix:
>Release-Note:
>Audit-Trail:
>Unformatted:
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2001-09-12 14:24 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2001-09-09 10:44 optimization/3387: GCC 3.0 optimisation -O2 -fPIC produces incorrect code sirl
-- strict thread matches above, loose matches on Subject: below --
2001-09-12 14:24 sirl
2001-06-22 19:56 William M. Brack
2001-06-22 17:56 wbrack
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).