[Trilinos-Users] Using my own matrix format with

Baker, Christopher G. bakercg at ornl.gov
Mon Feb 13 05:22:34 MST 2012


The Trilinos linear algebra libraries all provide an interface by which a user may implement their own linear operator for use in iterative solvers. For example, someone who wants to implement their own sparse matrix operator instead of using the Epetra_CrsMatrix class (one of Epetra's compressed sparse matrix classes) would implement their functionality into the base class, Epetra_Operator. This requires that you have implemented the local operations in addition to all of the communication necessary to support those.


-----Original Message-----
From: Karol Gruchany [karol.gruchany at gmail.com<mailto:karol.gruchany at gmail.com>]
Sent: Monday, February 13, 2012 03:36 AM Eastern Standard Time
To: trilinos-users at software.sandia.gov
Subject: [Trilinos-Users] Using my own matrix format with


My friend and I are going to perform a lot of high performance computations on sparse matrices. Due to specifity of the matrices (the majority of them are steady-state matrices), there is possibility to store them in very compressed form. My friend wrote a class for storing all of the matrix in compressed form in memory and performing some simple operations on the matrix. AFAIK, most of popular iterative solvers use very simple operations (like multiplication matrix by vector) and here is my questions: is there a possibility to implement our class with elementary operators overloaded (or just simple methods) to some of Trilinos project package and use it by iterative solvers? Or maybe instead of bringing a whole class, just provide a methods or functions which does low-level operations?�

Best regards,


More information about the Trilinos-Users mailing list