public inbox for gcc@gcc.gnu.org
 help / color / mirror / Atom feed
* Problems with loop unroller and gcse in 3.3?
@ 2003-09-19 13:37 Richard Guenther
  2003-09-19 13:55 ` Richard Guenther
  0 siblings, 1 reply; 2+ messages in thread
From: Richard Guenther @ 2003-09-19 13:37 UTC (permalink / raw)
  To: gcc

Hi!

Are there any known issues with the loop unroller and gcse in 3.3 from
g++-3.3 (GCC) 3.3.2 20030908 (prerelease)?

I have a POOMA C++ testcase that segfaults with -O2 -funroll-loops, but
doesnt with -O2 -funroll-loops -fno-gcse (also removing -funroll-loops
fixes the testcase). The segfault looks like

Program received signal SIGSEGV, Segmentation fault.
0x40182cb3 in chunk_free () from /lib/libc.so.6
(gdb) bt
#0  0x40182cb3 in chunk_free () from /lib/libc.so.6
#1  0x40182c53 in free () from /lib/libc.so.6
#2  0x4006a353 in operator delete(void*) (ptr=0xc0560000)
    at /home/rguenth/src/gcc/gcc3.3/libstdc++-v3/libsupc++/del_op.cc:39
#3  0x0804f397 in ~Engine (this=0xbfffeb94)
    at /home/rguenth/src/pooma-bugfixes/r2/src/Engine/BrickEngine.cpp:145
#4  0x0804af13 in main (argc=1, argv=0xbfffec94)
    at /home/rguenth/src/pooma-bugfixes/r2/src/Domain/DomainBase.h:185
#5  0x4012b9ed in __libc_start_main () from /lib/libc.so.6

exceptions are disabled.

Any ideas? (I'm trying to produce a standalone testcase and check more
recent 3.3 CVS)

Richard.

--
Richard Guenther <richard dot guenther at uni-tuebingen dot de>
WWW: http://www.tat.physik.uni-tuebingen.de/~rguenth/

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: Problems with loop unroller and gcse in 3.3?
  2003-09-19 13:37 Problems with loop unroller and gcse in 3.3? Richard Guenther
@ 2003-09-19 13:55 ` Richard Guenther
  0 siblings, 0 replies; 2+ messages in thread
From: Richard Guenther @ 2003-09-19 13:55 UTC (permalink / raw)
  To: gcc

On Fri, 19 Sep 2003, Richard Guenther wrote:

> Hi!
>
> Are there any known issues with the loop unroller and gcse in 3.3 from
> g++-3.3 (GCC) 3.3.2 20030908 (prerelease)?

I just checked gcc 3.4 from
gcc-3.4 (GCC) 3.4 20030908 (experimental)

and the problem is not visible with neither -funroll-loops nor
-fold-unroll-loops. GCC 3.2.3 also doesnt show this problem, so its a
regression.

> I have a POOMA C++ testcase that segfaults with -O2 -funroll-loops, but
> doesnt with -O2 -funroll-loops -fno-gcse (also removing -funroll-loops
> fixes the testcase). The segfault looks like
>
> Program received signal SIGSEGV, Segmentation fault.
> 0x40182cb3 in chunk_free () from /lib/libc.so.6
> (gdb) bt
> #0  0x40182cb3 in chunk_free () from /lib/libc.so.6
> #1  0x40182c53 in free () from /lib/libc.so.6
> #2  0x4006a353 in operator delete(void*) (ptr=0xc0560000)
>     at /home/rguenth/src/gcc/gcc3.3/libstdc++-v3/libsupc++/del_op.cc:39
> #3  0x0804f397 in ~Engine (this=0xbfffeb94)
>     at /home/rguenth/src/pooma-bugfixes/r2/src/Engine/BrickEngine.cpp:145
> #4  0x0804af13 in main (argc=1, argv=0xbfffec94)
>     at /home/rguenth/src/pooma-bugfixes/r2/src/Domain/DomainBase.h:185
> #5  0x4012b9ed in __libc_start_main () from /lib/libc.so.6
>
> exceptions are disabled.
>
> Any ideas? (I'm trying to produce a standalone testcase and check more
> recent 3.3 CVS)
>
> Richard.
>
> --
> Richard Guenther <richard dot guenther at uni-tuebingen dot de>
> WWW: http://www.tat.physik.uni-tuebingen.de/~rguenth/
>
>

--
Richard Guenther <richard dot guenther at uni-tuebingen dot de>
WWW: http://www.tat.physik.uni-tuebingen.de/~rguenth/

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2003-09-19 13:37 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2003-09-19 13:37 Problems with loop unroller and gcse in 3.3? Richard Guenther
2003-09-19 13:55 ` Richard Guenther

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).