[ET Trac] [Einstein Toolkit] #928: openmp parallelization within Exact broken

Einstein Toolkit trac-noreply at einsteintoolkit.org
Mon Oct 8 18:49:05 CDT 2012


#928: openmp parallelization within Exact broken
------------------------------------+---------------------------------------
  Reporter:  knarf                  |       Owner:                     
      Type:  defect                 |      Status:  confirmed          
  Priority:  minor                  |   Milestone:  ET_2012_11         
 Component:  EinsteinToolkit thorn  |     Version:  development version
Resolution:                         |    Keywords:                     
------------------------------------+---------------------------------------

Comment (by eschnett):

 A single region implies a barrier, but this barrier can be omitted with a
 nowait clause. If there was a barrier, things would break, since the
 barrier requires that each thread processes the exact same number of grid
 points, which we cannot guarantee.

 In many cases, saved variables can be omitted, especially if they only
 provide a shorter name for parameter that has a much longer name. However,
 in many other cases saved variables really speed up things, e.g. if they
 calculate large matrices for boosts or evaluate Bessel functions.

 In this case, the simplest solution is to use threadprivate variables.

-- 
Ticket URL: <https://trac.einsteintoolkit.org/ticket/928#comment:14>
Einstein Toolkit <http://einsteintoolkit.org>
The Einstein Toolkit


More information about the Trac mailing list