Milepost GCC Crash

[Nik]

Hi
I was compiling some files using Milepost GCC yesterday, but I had
trouble compiling one particular file. Now Yuriy mentioned that this
might happen on some combinations of flags but right now I was just
saving the compiler passes. I am giving the error here and I have
uploaded the c file in the files section. Please tell me whether this
has anything to do with Milepost, because the file is easily compiled
when I use GCC 4.4.0. Thanks.

/media/Misc/Linux/Mozilla/Milepost/install/bin/gcc -o vorbis_res0.o -c
-I../../../dist/system_wrappers -include /home/nik/192srcpgo/config/
gcc_hidden.h -DOSTYPE=\"Linux2.6\" -DOSARCH=Linux -I/home/nik/
192srcpgo/media/libvorbis/lib -I. -I../../../dist/include -I../../../
dist/include/nsprpub -I/home/nik/192srcpgo/ff-opt/dist/include/nspr -
I/home/nik/192srcpgo/ff-opt/dist/include/nss -fPIC -Wall -W -
Wno-unused -Wpointer-arith -Wcast-align -W -Wno-long-long -pedantic -
fno-strict-aliasing -pthread -pipe -DNDEBUG -DTRIMMED -Os -freorder-
blocks -fno-reorder-functions -include ../../../mozilla-config.h -
DMOZILLA_CLIENT -Wp,-MD,.deps/vorbis_res0.pp /home/nik/192srcpgo/media/
libvorbis/lib/vorbis_res0.c
ICI: loading plugin ...
/home/nik/192srcpgo/media/libvorbis/lib/vorbis_res0.c: In function ‘T.
72’:
/home/nik/192srcpgo/media/libvorbis/lib/vorbis_res0.c:525: internal
compiler error: Segmentation fault
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.
make[3]: *** [vorbis_res0.o] Error 1
make[3]: Leaving directory `/home/nik/192srcpgo/ff-opt/media/libvorbis/
lib'

[Grigori Fursin]

Thanks for the info, Nik.
Can you please attach the source file so that we could check it?..
I think I changed the settings of this group to let users
attach files ...
Cheers,
Grigori

--
You received this message because you are subscribed to the Google Groups "ctuning-discussions"
group.
To post to this group, send email to ctuning-d...@googlegroups.com.
To unsubscribe from this group, send email to ctuning-discuss...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/ctuning-discussions?hl=en.

[Nik]

Yea, I've already posted it in the files section. But I havn't posted
any of the header files included. It will take me a long time to find
all of them. If you require them, then please tell me and I'll try to
put them all together. Thanks :).

[Joern Rennecke]

Quoting Nik <nikhi...@gmail.com>:

Much more useful would be if you posted preprocessed source, the exact
options passed to the failed cc1 run, and the exact version of the
compiler.

For the first two, re-do the failed compilation with -v --save-temps,
then attach the generated*.i file, and cut & paste the compiler output
before the crash.

[Nik]

Hi
I have done as you suggested and you can find both vorbis_res0.i and
vorbis_res0_compiler_options.txt attached. I am using gcc 4.4.0. The
file compiles normally, but the moment I put ICI_USE=1, the
compilation crashes. Please take a look. Thanks for your help..

[Nik]

[Joern Rennecke]

Quoting Nik <nikhi...@gmail.com>:

> Hi
> I have done as you suggested and you can find both vorbis_res0.i and
> vorbis_res0_compiler_options.txt attached.

Actually, your email doesn't have attachments.

[Nik]

They are uploaded in the files of the group... Isn't that where you
want them to be?

[Nik]

I went through the file being compiled, trying to reduce it and found
that the error is limited to just one function "_01forward". As far as
I could see, the error comes on line 4043 of vorbis_res0.i or line 612
of vorbis_res0.c. Without this function, milepost easily compiles the
file.
Hope this helps.

[Yuri Kashnikoff]

Hi Nik,

I will take a look tomorrow at it and will try to assist you with this problem.
I am sorry for delay, just too many things to do :)

Thanks!

[Grigori Fursin]

Thanks a lot, Nik - this should be easier to check.
My colleagues and I have to finish some urgent things the coming
week, but I identified the problem with the cTuning optimization
prediction web service and should fix it during the week starting
from 8th of February ... We will keep in touch about that ...
Take care,
Grigori

-----Original Message-----
From: ctuning-d...@googlegroups.com [mailto:ctuning-d...@googlegroups.com] On Behalf
Of Nik

--

[Yuri Kashnikoff]

Hi Nik,

Sorry for the long delay. We have found a couple of problems and are
finishing them. So, please wait a while, hopefully everything would be
finished next week.

Thanks!

On Sun, Jan 31, 2010 at 11:53 PM, Grigori Fursin
<grigori...@inria.fr> wrote:
> Thanks a lot, Nik - this should be easier to check.
> My colleagues and I have to finish some urgent things the coming
> week, but I identified the problem with the cTuning optimization
> prediction web service and should fix it during the week starting
> from 8th of February ... We will keep in touch about that ...
> Take care,

> Grigori
>
> -----Original Message-----
> From: ctuning-d...@googlegroups.com [mailto:ctuning-d...@googlegroups.com] On Behalf
> Of Nik

> Sent: Friday, January 29, 2010 3:38 PM
> To: ctuning-discussions
> Subject: Re: Milepost GCC Crash
>

[Nik]

Hey
Thanks a lot for working on this. I understand about the delay. I'll
be waiting. Thanks again.

[Nik]

Hi
Are there any updates on the crash issue and the web service? If so
please let me know. Thanks.

[Grigori Fursin]

Hi Nik,

After many sleepless nights this week ;), we managed
to find most of the issues with the prediction services
(and actually I rewrote it completely to better predict
execution time, code size and compilation time). I plan
to update both web service and MILEPOST GCC this week
and will give you a patch before releasing the tool.

The bug is not fixed yet since we spent most of the time
on data analysis/verification this week and also fixed
a few more bugs in CCC framework and benchmarks (that
we should also release shortly). We will try to check/fix
the bug this week too so will communicate with you about that ...

Hope to send you software soon since I am very interested
to test it on some more realistic libraries/programs ...

Will keep in touch,
Grigori (and Yuriy and Abdul)

-----Original Message-----
From: ctuning-d...@googlegroups.com [mailto:ctuning-d...@googlegroups.com] On Behalf
Of Nik
Sent: Saturday, February 13, 2010 6:53 PM
To: ctuning-discussions
Subject: Re: Milepost GCC Crash

--

[Nik]

Hi

Thanks a lot for working on this. I look forward to working on the new
patch. Thanks.

Regards
Nik

[Abdul Wahid Memon]

Hello Nik

I have tried to compile the file with ICI and it is generating the passes as it should be generating. Milepost compiler seems to be working fine.

Regards

Abdul

[Grigori Fursin]

Hi Abdul,

Did you use MILEPOST GCC 4.4.0 that is available for download at SVN?...

Did you try to extract features?.. I think this is when it may crash?..

Thanks,

Grigori

 

 

From: ctuning-d...@googlegroups.com [mailto:ctuning-d...@googlegroups.com] On Behalf Of Abdul Wahid Memon
Sent: Monday, February 15, 2010 2:47 PM
To: ctuning-d...@googlegroups.com
Subject: Re: Milepost GCC Crash

 

Hello Nik

[Abdul Wahid Memon]

Hello Sir

Yes I used the latest one (at SVN) and the feature extraction process is working correctly.

Regards

Abdul

On Mon, Feb 15, 2010 at 2:49 PM, Grigori Fursin <grigori...@inria.fr> wrote:

Hi Abdul,

Did you use MILEPOST GCC 4.4.0 that is available for download at SVN?...

Did you try to extract features?.. I think this is when it may crash?..

Thanks,

Grigori

 

 

From: ctuning-d...@googlegroups.com [mailto:ctuning-d...@googlegroups.com] On Behalf Of Abdul Wahid Memon
Sent: Monday, February 15, 2010 2:47 PM
To: ctuning-d...@googlegroups.com

Subject: Re: Milepost GCC Crash

 

Hello Nik

[Grigori Fursin]

Actually, we just found out that Abdul has been using CCC version to extract features,

not milepost-gcc wrapper. Since I am cleaning it up at the moment, we will recheck it

later ...

Cheers,

[Nik]

Yea, I am not using CCC.. Maybe the problem is in Milepost GCC wrapper
as you said. Thanks for the work here. Do let me know if it is
something I am doing wrong, which I need to correct.. For the record,
I am using 4.4.0...