Dear Hong, thanks for your quick answer.
Yes, we replace it with superlu's ilu with drop tolerance. To use it, configure petsc with '--download-superlu --download-parmetis' Then run your code with option -pc_type ilu -pc_factor_mat_solver_package superlu -mat_superlu_ilu_droptol<>
see available options with '-help |grep superlu'
Ah, ok. Is there a possibility to hardwire that in the source code? We run different integration tests with different solver/pc combinations with one single executable call (currently without any options).
Suggest to use petsc-dev for such function, because we have fixed several bugs in superlu interface.
Ok, thanks. Will this functionality be available also in the next release (p7) and when is this expected (approximately)? Thanks and best regards Tobias
we recently switched from 3.0.0-p11 to 3.1-p6 and are now facing a minor problem in some of our test cases: We use seqaij matrix format and GMRES with ILU dt.
Our code contained a statement using PCFactorSetUseDropTolerance() which does not exist any longer. So we renamed the function call to PCFactorSetDropTolerance (same signature) which has no online docu but found by google ;-) Obviously, this function has not the same functionality as our tests fail.
In 3.0.0-p11, the docu of the "Summary of Sparse Linear Solvers Available from PETSc" showed a line containing: ILU dt: ILU dt seqaij Sparsekit (table survey)
This is not included in the docu of 3.1-p6 (http://www.mcs.anl.gov/petsc/petsc-as/documentation/linearsolvertable.html ) any more. Does that mean that the ILU dt version is not supported by the (default) petsc? I also checked the changelog but did not find anything there; sorry if I missed sth.
Any help or infos will be highly appreciated ;-)
Thanks and best regards Tobias