[Gmsh] Meshing is slow because Netgen is too verbose
cgeuzaine at ulg.ac.be
Thu Aug 2 12:16:30 CEST 2012
On 02 Aug 2012, at 10:00, Serban Georgescu <serbange at 09.alumni.u-tokyo.ac.jp> wrote:
> Hi Christophe,
> Thanks for the quick reply.
> I have tried "-v 3" but that removes almost all meshing output, including information that is quite useful to me (e.g. time taken in 2D and 3D meshing, time for optimization, resulting number of elements). So I was looking for a way that removes that extra output and leaves just the "important" information. Basically, I would like to get something like it was for Gmsh 2.5, when Netgen was not that verbose.
> Do you know a way to do this?
I've removed the "testout" in svn; this should match what we did in 2.5.
Thanks for the feedback!
> Thanks again!
> On 2 Aug 2012, at 07:20, Christophe Geuzaine <cgeuzaine at ulg.ac.be> wrote:
>> Hi Serban - Use "-v 3" ?
>> On 01 Aug 2012, at 14:39, Serban Georgescu <serbange at 09.alumni.u-tokyo.ac.jp> wrote:
>>> Hi guys,
>>> I am have been using Gmsh for an year or so with very good results (and I am very grateful to the developers for the wonderful work).
>>> Recently, I switched to version 2.6 (from 2.5) and I found that for some of the models the meshing time is much increased.
>>> The main reason for this is the large amount of output that Netgen generates.
>>> Basically, Netgen outputs loads of the following Info:
>>> Info : Freezone:
>>> Info : (0, 0, 0)
>>> Info : (1.2, -5.5e-17, -3.1e-17)
>>> Info : (-0.45, 1, 2.9e-18)
>>> Info : (-2.3, 1.2, -0.31)
>>> Info : (-0.54, 0.74, -0.1)
>>> Info : (-0.96, 0.74, -0.1)
>>> Info : (-0.38, 0.38, -0.1)
>>> Info : Mapping found!!: Rule Tetrahedron Vis a Vis Point with 3 edges (1)
>>> Info : 3 1 2 7
>>> Info : 3 = 3
>>> Here are some results for one of the models:
>>> Gmsh 2.5 -> ~ 6 sec
>>> Gmsh 2.6 with output redirected to file à ~ 11 sec
>>> Gmsh 2.6 with output redirected to /dev/nullà ~ 8 sec
>>> Gmsh 2.6 with manually editingcontrib/Netgen/nglib_gmsh.cpp by commenting line 75 (//testout = new ostream(new mystreambuf());) and un-commenting line 66 (testout = new ofstream ("test.out");) à ~ 5 sec
>>> So basically, in this case, the extra verbosity from netgen makes the mesher take twice as much.
>>> I have other cases where the increase is much larger.
>>> My question (and wish) is: would it be possible to connect this output to the Gmsh verbosity argument (-v)?
>>> I would be quite happy to (at least try to) add this functionality myself should somebody tell me how could I get access the verbosity value from contrib/Netgen/nglib_gmsh.cpp .
>>> Thanks in advance,
>>> gmsh mailing list
>>> gmsh at geuz.org
>> Prof. Christophe Geuzaine
>> University of Liege, Electrical Engineering and Computer Science
Prof. Christophe Geuzaine
University of Liege, Electrical Engineering and Computer Science
More information about the gmsh