[petsc-users] pastix solver break at pastix_checkMatrix

[Gong Ding]

----- Original Message ----- 
From: "Barry Smith" <bsmith at mcs.anl.gov>
To: "PETSc users list" <petsc-users at mcs.anl.gov>
Sent: Thursday, December 30, 2010 10:55 PM
Subject: Re: [petsc-users] pastix solver break at pastix_checkMatrix



  Please run program with -ksp_view_binary and send us the file binaryoutput to petsc-maint at mcs.anl.gov so we can reproduce the problem.


   Barry


Ok, I use pastix as linear solver of SNES.  
(The coding style of pastix seems much better than mumps. I'd like to try if I can speed it by GPU-based BLAS.) 

The settings are listed as follows:
ierr = KSPSetType (ksp, (char*) KSPPREONLY); assert(!ierr);
ierr = PCSetType  (pc, (char*) PCLU); assert(!ierr);
ierr = PCFactorSetMatSolverPackage (pc, MAT_SOLVER_PASTIX); assert(!ierr);
ierr = PCFactorSetReuseFill(pc, PETSC_TRUE);assert(!ierr);
ierr = PCFactorSetReuseOrdering(pc, PETSC_TRUE); assert(!ierr);
ierr = PCFactorSetColumnPivot(pc, 1.0); genius_assert(!ierr);
//ierr = PCFactorReorderForNonzeroDiagonal(pc, 1e-20); assert(!ierr);<-- Caught signal number 11 SEGV error will occure when diag value < 1e-20
ierr = PCFactorSetShiftType(pc,MAT_SHIFT_NONZERO);assert(!ierr);

The attached matrix is dumped  at the end of jacobian assemble.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: break.mat
Type: application/octet-stream
Size: 208768 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20101231/ddf534cc/attachment-0001.obj>