[root]/Packages/Uintah/CCA/Components/LoadBalancers
Author | Changes | Lines of Code | Lines per Change |
---|---|---|---|
Totals | 70 (100.0%) | 1379 (100.0%) | 19.7 |
worthen | 37 (52.9%) | 819 (59.4%) | 22.1 |
luitjens | 19 (27.1%) | 527 (38.2%) | 27.7 |
mjc | 3 (4.3%) | 11 (0.8%) | 3.6 |
dav | 4 (5.7%) | 11 (0.8%) | 2.7 |
dgroulx | 4 (5.7%) | 8 (0.6%) | 2.0 |
jas | 1 (1.4%) | 2 (0.1%) | 2.0 |
harman | 2 (2.9%) | 1 (0.1%) | 0.5 |
Reverted load balancing algorithm to an earlier algorithm. This
algorithim is more robust. Also this should fix a DLB/MPM bug.
0 lines of code changed in 4 files:
Revamped load balancer logic in an attempt to achieve decent load
balance.
118 lines of code changed in 2 files:
tweaked the load balancer logic slightly to improve load balance
2 lines of code changed in 1 file:
Changed the default targetImbalance parameter to 5%
1 lines of code changed in 1 file:
Modified the SFC load balancing algorithm to provide a better load
balance.
66 lines of code changed in 2 files:
Add Per-Component LB stats for Regridder, LoadBalancer, TG Execution, TG mpi wait, TG compilation, available with SCI_DEBUG=ComponentTimings:+
3 lines of code changed in 1 file:
SFC.h
Added more sorting algorithms for space-filling curve generation and added
optimizations to the current methods.
22 lines of code changed in 1 file:
Fix small DLB bug resulting from last commit and remove debug print from AMRSC
1 lines of code changed in 1 file:
Remove concept of time-refinement from Grid, and store whether the sim is lockstep in SharedState. Have components compute delt based on the level it's on and have the SimulationController worry about the inter-level constraints. (Also consolidate SimCnt::doInitialTimestepRegridding into doInitialTimestep and doRegridding)
3 lines of code changed in 1 file:
Use fabs instead of abs.
2 lines of code changed in 1 file:
Modified the SFC generation to use a different style of indexing. In
addition new merging methods have been implemented but need more testing
so are disabled by default.
22 lines of code changed in 2 files:
Overhaul the DynamicLoadBalancer
State no longer maintained internally, called by SimController
Should now work with any combination of restart, regrid, nthproc,
timestep restarts, etc.
Schedulers:
Modify the way Particle Subsets determine whether they need to send
the number of particles to receiving processor
Don't reserve cell-refinement-ratio of cells for each var, just the
number of intra-level ghost cells required
MPMICE:
Move a cout to a cout_doing
Regridder:
Use Regridder vars on matl 0
128 lines of code changed in 6 files:
Replace Scheduler::noScrubVar and copyDataVar with overrideDefaultBehavior. Enable restarts to regrid where the old uda left off rather than waiting a timestep first
24 lines of code changed in 2 files:
Fix hopefully last of 3d amr bugs
3 lines of code changed in 1 file:
cleaned up output
1 lines of code changed in 2 files:
Changed float to double. Floating point error was causing the dynamic
load balancer to assign patches to non-existant processors. This only
occured when there were lots of patches.
11 lines of code changed in 2 files:
Force time-refinement to be based on cell refinement. This is probably temporary until we get the arbitrary-timestep per level working.
Also provides a "lockstep" knob for AMR that will in essence force TRR To be 1
Most of the changes are just moving adjustDelt from SimulationState back to Level.
1 lines of code changed in 1 file:
Improve the relationship between Load Balancing, Scrubbing, Multi-Taskgraphing, and refluxing
- Provide a Conditional message-send, which enables the TG to only compile once instead
of twice for a load-balance (and allows multi-tg to run following a lb)
- Have the LB threshold be based on the theoretical loadBalance%
- Deprecate nonCopyDataVars
17 lines of code changed in 1 file:
AMR Scrubbing
12 lines of code changed in 1 file:
Get neighborhood to (hopefully) be correct for large 3D 4+level problems. Also prune down calls to selectPatches when there are no ghost cells, and change the reflux iterators when the coarse and fine patches share the same boundary
30 lines of code changed in 1 file:
(15 more)