[Dune] UG grid problem: ID overflow DDD_HdrConstructor

[Eike Mueller]

Hi Oliver and Markus,

thank you very much for your replies. As my email says, I already had a look at FS #250, and followed the instructions there, 
but this does not help.

I attach an example code, which reads the grid from the attached .dgf file, refines it, attaches data to the grid and writes it 
out as a .vtu file.

The executable is called as

testsphericalgridgenerator <gridfile> <refcount>

where <gridfile> is the .dgf file to read from and <refcount> is the number of refinenement steps.

The example file contains six elements, so when running on 6 preocessors, each should end up with one element and refine it 
<refcount> times. I ran it through an interactive session on a HECToR node, but it segfaults if I run on 1 or 6 cores. So far I 
haven't been able to find out exactly where it crashes.

The code also works perfectly fine (both sequentially and in parallel) on my desktop, where I just use the default settings for 
the GID in the UG and Dune installation (i.e. on my desktop I did not add the -DDDD_GID_T=long and --with-ddd-maxprocbits=20 
flags at compile time).

Thanks,

Eike

Oliver Sander wrote:
>   Hi Eike,
> sorry for the late reply.  AFAIK replacing the id type is UG has hardly 
> been tested,
> and I am not surprised that people find bugs.  Can you please send me a test
> program and I promise to have a look.
> best,
> Oliver
> 
> Am 23.11.2012 18:53, schrieb Eike Mueller:
>> Dear dune-list,
>>
>> I'm having trouble with using the parallel DUNE grid, and I keep 
>> getting the
>>
>> DDD [000] FATAL 02221: global ID overflow DDD_HdrConstructor
>>
>> error. My grid is not very large, I use 6*2^5 = 6144 elements on 6 
>> processors. I had a look at Flyspray Task #250 [1], and followed the 
>> instructions there, in particular I set -DDDD_GID_T=long and use 20 
>> bits for the processor part (--with-ddd-maxprocbits=20). I recompiled 
>> both UG and Dune (version 2.2) with these flags, and also recompiled 
>> my own code with them. I checked that long is really 8 bytes (=64 
>> byte) on the machine I'm using, but I still get the error.
>> I generate the grid using the UG grid factory by reading from a .dgf 
>> file with the attached classes.
>> If I make the grid smaller, the problem eventually disappears but I 
>> get a segmentation fault.
>> I should also say that on a different machine, where I did not change 
>> the maxprocbit and GID_T, it all works fine for small grids.
>>
>> Is there anything I might have overlooked? For example, do I have to 
>> be careful with the datatypes I use when constructing the grid with 
>> the factory?
>>
>> Thanks a lot,
>>
>> Eike
>>
>> [1] 
>> http://www.dune-project.org/flyspray/index.php?do=details&task_id=250&string=UG&project=1&type[0]=&sev[0]=&pri[0]=&due[0]=&reported[0]=&cat[0]=&status[0]=open&percent[0]=&opened=&dev=&closed=&duedatefrom=&duedateto=&changedfrom=&changedto=&openedfrom=&openedto=&closedfrom=&closedto=
>>
>>
>>
>>
>>
>> _______________________________________________
>> Dune mailing list
>> Dune at dune-project.org
>> http://lists.dune-project.org/mailman/listinfo/dune
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Dune mailing list
> Dune at dune-project.org
> http://lists.dune-project.org/mailman/listinfo/dune


-- 
Dr Eike Mueller
Research Officer

Department of Mathematical Sciences
University of Bath
Bath BA2 7AY, United Kingdom

+44 1225 38 5633
e.mueller at bath.ac.uk
http://people.bath.ac.uk/em459/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: sphericalshell.tgz
Type: application/x-compressed-tar
Size: 4748 bytes
Desc: not available
URL: <https://lists.dune-project.org/pipermail/dune/attachments/20121126/0d2168d0/attachment.bin>