Load balancing changes answer

[rpwagner]

-DENABLE_LOAD_BALANCE causes a change in the result. This was found by comparing the results of running an [example cosmology simulation]. [Changeset 2284] is the first placed where a vanilla build will show a difference in the result.

Unless someone has an argument for why load balancing should modify the results, this is a bug.

[jbordner]

I'll look into this for awhile, but may throw it back at Brian if I get bored. Grep shows the only difference is calling / not calling CommunicationLoadBalanceGrids?().

[jbordner]

The problem has been narrowed down to gravity. It works if gravity is turned off (grids are still load balanced and the results are bitwise identical), and the first change in any field occurs across the call to PrepareDensityField?().

[jbordner]

The first change seems to occur in the following function:

   EvolveLevel()
      PrepareDensityField()
         PrepareGravitatingMassField()
            CopyOverlappingParticleMassFields()
               grid::CheckForOverlap()
                  grid::AddOverlappingParticleMassField()

[jbordner]

David Collins found and fixed the bug! See SVN revision 2715 http://mngrid.ucsd.edu/projects/enzo/changeset/2715