Lines of Code
[root]/CCA/Components/LoadBalancers
| Author | Changes | Lines of Code | Lines per Change |
|---|---|---|---|
| Totals | 112 (100.0%) | 1714 (100.0%) | 15.3 |
| luitjens | 34 (30.4%) | 1649 (96.2%) | 48.5 |
| jas | 78 (69.6%) | 65 (3.8%) | 0.8 |
Add the executing processor to its neighorhood. Previously if a processor had no patches on all levels it would not be in its own neighborhood. This in turn caused once per proc tasks to cause a deadlock.
0 lines of code changed in 2 files:
Added some MallocTrace profiling.
Enabled parallel SFC generation. This was disabled when we ran into an allgatherv bug on Kraken which has now been fixed.
0 lines of code changed in 2 files:
Fix doxygen errors.
1 lines of code changed in 2 files:
Don't call MPI functions when only using a single processor.
0 lines of code changed in 2 files:
only output constant values if ProfileStats:+ is on.
1 lines of code changed in 1 file:
Minor changes to the statistics output
0 lines of code changed in 2 files:
Updated load balancers to be able to easily utilize what was
previously experimental code. Also added some more defaults
so that the user doesn't have to specify many parameters for a
load balancer.
44 lines of code changed in 8 files:
Modified PLB threshold criteria
1 lines of code changed in 1 file:
Adjusted where the weighting functions are applied.
3 lines of code changed in 1 file:
Added the ability to weight particles or cells when partitioning the patches into particle or cell patches.
9 lines of code changed in 3 files:
Initial commit for a particle specific load balancer.
This load balancer is attempting to address performance issues in the DLB associated with particles.
In particular, the DLB balances on a single cost metric which is a function of the number of particles
and cells. This load balancer seperates the metric and load balances in two phases. During the first
phase any patches that have particles are assigned to processors and in the second phase all remaining patches
are assigned to processors. During the first phase the load balancer attemptes to equalize the number of particles
on each processor. The second phase then load balances the remaining processors while attempting to equalize
the number of cells on each processor, this is done while mainting the assignments from the first phase.
Theoretically this will help avoid synchronization problems associated with having varying numbers of particles
on each processor.
1423 lines of code changed in 6 files:
Moved some classes from DLB.h to LBC.h so they can be more readily reused. Also fixed the indenting of LBC.h
115 lines of code changed in 3 files:
Update copyright dates.
1 lines of code changed in 1 file:
Update copyright dates
21 lines of code changed in 33 files:
Undoing commit 45403
21 lines of code changed in 21 files:
Update copyright date.
21 lines of code changed in 21 files:
Undid part of my last commit as it was messing with other lb options.
1 lines of code changed in 1 file:
When profiling only collect particles when profile data does not exist.
4 lines of code changed in 1 file:
Added some debugging code that is turned off
48 lines of code changed in 1 file: