[Trilinos-Users] creating_epetra_import_assertion_fail

Blahos, Jiri j.blahos at imperial.ac.uk
Wed Jan 22 11:06:28 MST 2020


Hello everyone,

I am getting an error in one of the assertions when constructing an instance of epetra_import. It happens for a specific mesh (I have a finite element code) and specific number of mpis.
The failing assertion is:

assert(curr_LID!=-1); // Internal error

in file Epetra_BasicDirectory.cpp, line 644. I am using Trilinos with tag 12.18.1 on the github repo.

The assertion fails when I use 48 mpis on a cluster I am using.

The line in my code that fails:

mMatrixImport = new Epetra_Import(*mStandardDofMapSemilocal, *mStandardDofMap);

where mMatrixImport is Epetra_Import* and both mStandardDofMapSemilocal and mStandardDofMap are type Epetra_Map*.

I have printed both maps into files on all ranks, but couldn't find anything unusual in them.
The files are attached (map1 is mStandardDofMapSemilocal, map2 is mStandardDofMap).
The mStandardDofMapSemilocal map is as the name suggests overlapped, which means some of the indices are local to more than one mpi rank.

Just wondering if anyone has any suggestions about what could be wrong with the maps or the import construction.

Best regars,
Jiří Blahoš

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://trilinos.org/pipermail/trilinos-users_trilinos.org/attachments/20200122/c259eb8a/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: map_files.zip
Type: application/x-zip-compressed
Size: 410084 bytes
Desc: map_files.zip
URL: <http://trilinos.org/pipermail/trilinos-users_trilinos.org/attachments/20200122/c259eb8a/attachment-0001.bin>


More information about the Trilinos-Users mailing list