----- Original Message ----- From: "Barry Smith" <[email protected]> To: "PETSc users list" <petsc-users at mcs.anl.gov> Sent: Saturday, January 01, 2011 2:12 AM Subject: Re: [petsc-users] pastix solver break at pastix_checkMatrix
Sorry. Yes there was another bug. What caused both of these problems is that Pastix requires a symmetric nonzero structure and the interface made some assumptions that the PETSc matrix had a symmetric nonzero structure which is usually true, hence it did not crash for most matrices people use. I've attached another copy of pastix.c follow the same procedure again. Barry Thank you, Barry. The new version of pastix.c works. And the previous patch of memory allocation seems useless, you can reset it back. Further more, valgrind reported serious memory leak. However, I think it is nothing to do with the petsc interface. Maybe update pastix to the latest version can solve this problem.
