[Trilinos-Users] problem when trying to use HIPS

[Riccardo Rossi]

Dear Mike,
             actually i haven't done any wrapping ... as i understand HIPS
interface is included with trilinos ifpack since release 10.10

for example a quick googling gives this within your doxygen documentation

http://trilinos.sandia.gov/packages/docs/r10.10/packages/ifpack/browser/doc/html/test_2HIPS_2cxx__main_8cpp.html


i am attempting  using it exactly as one of the ifpack preconditioners...
the failure seems to happen as soon as the trilinos calls the hips lib,
however the hips appears to be correctly compiled since it passes its own
tests.


if you have a look at the file i attach i am using the aztec factory to
construct the preconditioner as if it was a standard ILU or ILUT one (which
do work)


i don't think any of the tests checks for correctness of this, but i may
well be wrong. Do you have any clue on who may have contributed the
interface?

thanks in advance
Riccardo


On Fri, Aug 30, 2013 at 5:54 PM, Heroux, Mike <MHeroux at csbsju.edu> wrote:

> Riccardo,
>
> I assume you have wrapped HIPS as an Epetra_Operator and are calling the
> SetPrecOperator method in AztecOO, correct?
>
>  One thing that comes to mind, which is easily overlooked, is that AztecOO
> expects your Epetra_Operator implementation to support the situation where
> the input and output vectors are the same array.  In other words, the
> ApplyInverse method of your Epetra_Operator must handle the case where the
> X and Y arguments are the same object.  This is briefly mentioned here:
>
>
> http://trilinos.sandia.gov/packages/docs/r11.4/packages/epetra/doc/html/classEpetra__Operator.html#aac0eb3b0061a6e67a5666194e744c343
>
> This is an artifact of the underlying Aztec.  If HIPS doesn't support this
> mode, you can create your own temporary Y vector at the beginning of
> ApplyInverse, and copy the result back to the real Y vector at the end of
> ApplyInverse.
>
> Beyond this issue, the only advice I can give is to make sure you are
> checking error codes from the return arguments.  Sometimes you get
> information back from one of them.
>
> If you keep having trouble, please send more details.
>
> Mike
>
> From: Riccardo Rossi <rrossi at cimne.upc.edu<mailto:rrossi at cimne.upc.edu>>
> Date: Wednesday, August 28, 2013 2:33 PM
> To: "trilinos-users at software.sandia.gov<mailto:
> trilinos-users at software.sandia.gov>" <trilinos-users at software.sandia.gov
> <mailto:trilinos-users at software.sandia.gov>>
> Subject: [Trilinos-Users] problem when trying to use HIPS
>
> Dear List,
>
>        i am getting a segfault with no apparent reason when trying to use
> HIPS
> through AztecOO
>
> i attach the makefile.inc used in compiling hips
>
> the configure i use for the trilinos
>
> the file that i use to attempt using the solver...
>
> my system is ubuntu 12.04 and i use OpenMPI
>
> i use our own python wrappers to trilinos, the settings are given as
>
> aztec_parameters = ParameterList()
> aztec_parameters.set("AZ_solver", "AZ_gmres")
> aztec_parameters.set("AZ_kspace", 100)
> aztec_parameters.set("AZ_output", 1)
>
> preconditioner_type = "HIPS"
> preconditioner_parameters = ParameterList()
> preconditioner_parameters.set("hips: id",0);
> #preconditioner_parameters.set("hips: strategy",1) ---> note that i set
> this in the C++ code!!
>
> overlap_level = 0
> nit_max = 300
> tol = 1e-6
>
> solver.structure_linear_solver = AztecSolver(aztec_parameters,
> preconditioner_type, preconditioner_parameters, tol, nit_max, overlap_level)
> #so
>
> any hint on what i may be doing wrong will be very welcome (i am convinced
> i miss a fundamental parameter but the failure provides no info at all...)
>
> thanks in advance
>
> Riccardo
>
> --
>
> Dr. Riccardo Rossi, Civil Engineer
>
> Member of Kratos Team
>
> International Center for Numerical Methods in Engineering - CIMNE
> Campus Norte, Edificio C1
>
> c/ Gran Capitán s/n
>
> 08034 Barcelona, España
>
> Tel:        (+34) 93 401 56 96
>
> Fax:       (+34) 93.401.6517
>
> web:       www.cimne.com<http://www.cimne.com/>
>



-- 

Dr. Riccardo Rossi, Civil Engineer

Member of Kratos Team

International Center for Numerical Methods in Engineering - CIMNE
Campus Norte, Edificio C1

c/ Gran Capitán s/n

08034 Barcelona, España

Tel:        (+34) 93 401 56 96

Fax:       (+34) 93.401.6517
web:       www.cimne.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://software.sandia.gov/pipermail/trilinos-users/attachments/20130830/6022424b/attachment.html 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: aztec_solver.h
Type: text/x-chdr
Size: 9676 bytes
Desc: not available
Url : https://software.sandia.gov/pipermail/trilinos-users/attachments/20130830/6022424b/attachment.bin