[Trilinos-Users] OpenMp on Epetra

Heroux, Mike MHeroux at csbsju.edu
Thu Apr 11 10:24:57 MDT 2013

This information is correct.  In addition, you can add
-DEPETRA_HAVE_OMP_NONASSOCIATIVE to the C++ flags list if you want to
enable threaded execution of dot products and norms.  By default these are
not enabled due to bitwise variations in answers and the general minimal
role parallelization of these kernels has on performance.


On 4/11/13 10:48 AM, "Ramsey, James J CIV (US)"
<james.j.ramsey14.civ at mail.mil> wrote:

>I found the description that told Epetra can support OpenMp from
>trilinos11 version, in somewhere in the web;;
>The following sentence is what I found in the web.
>OpenMP support for Epetra. Epetra is the original and still dominant
>collection of data classes for Trilinos. Epetra can be compiled with
>OpenMP support enabled. Turning this option on will enable
> threaded computation of sparse matrix multiplication, and all vector and
>multivector computations. It includes support for NUMA data placement.
>Sparse triangular solves are presently NOT threaded.
>But I couldn¹t find any example code and I had no idea how to set it.
>Look at http://trilinos.sandia.gov/Trilinos11CMakeQuickstart.txt
>I think that all you need to do is add "-D
>Trilinos_ENABLE_OpenMP:BOOL=ON" to your CMake command line when
>configuring, and Epetra will use OpenMP behind the scenes.
>Trilinos-Users mailing list
>Trilinos-Users at software.sandia.gov

More information about the Trilinos-Users mailing list