Lines of Code
[root]/CCA/Components/LoadBalancers
| Author | Changes | Lines of Code | Lines per Change |
|---|---|---|---|
| Totals | 120 (100.0%) | 1052 (100.0%) | 8.7 |
| ahumphrey | 56 (46.7%) | 883 (83.9%) | 15.7 |
| allen | 14 (11.7%) | 79 (7.5%) | 5.6 |
| dav | 13 (10.8%) | 62 (5.9%) | 4.7 |
| jas | 35 (29.2%) | 26 (2.5%) | 0.7 |
| harman | 2 (1.7%) | 2 (0.2%) | 1.0 |
fixed boolean chaeck and added modified var mapper for saving vars in the timestep.xml file
3 lines of code changed in 6 files:
The main changes in this commit only effect a simulation when <outputNthProc> is
set (> 1) in the .ups file.
This commit allows for PIDX output on every process for a time step
output, while using every Nth process (if so specified) for UDA
checkpoint output. Eventually the need for this Nth Proc changing
hack (and Nth proc output itself) should go away when PIDX fully takes
over I/O.
Note, when using "Nth Process Output", both a checkpoint and a time step save
CANNOT occur on the same time step. Therefore the time step output is postponed
for one time step.
On a slightly different note, we use the term proc badly... we should use process,
or perhaps better yet, rank... as proc is usually taken to mean "processor", but
in MPI's case, means process - and the term MPI uses for process is rank.
D CCA/Ports/LoadBalancer.cc
D CCA/Ports/LoadBalancer.h
A + CCA/Ports/LoadBalancerPort.cc
A + CCA/Ports/LoadBalancerPort.h
- Renamed from LoadBalancer to LoadBalancerPort to avoid confusion with the LoadBalancer itself.
- Added "setNthProc()" so that the SimulationController can change the value as needed.
M CCA/Components/DataArchiver/DataArchiver.h
M CCA/Components/DataArchiver/DataArchiver.cc
- Added postponeNextOutputTimestep(). This pushes off the output of a time step by one
time step. (This occurs if using Nth Process Output and a checkpoint and a time step
where to occur on the same time step.)
- Added savingAsPIDX() boolean query function so that the SimulationController
can avoid doing all of this hack when we are not using PIDX.
- Removed "getCurrentTimestep()" and "getCurrentTime()", as, per the documentation on them,
they should not be used anyway; you should get this info from the shared state. Note,
the DataArchiver name (getCurrentTime()) is different from the shared state name
(getElapsedTime()) which was also confusing.
- Pass const string parameters by reference.
- White space cleanup / alignments.
- If the input file has no <save> labels, but specifies an output interval, throw an exception.
- If the 'logname' is not available when running a simulation, put 'unknown' instead of ignoring.
Should probably find a better fix then 'unknown', but that will have to wait for another day.
- Swat some evil endls. :)
- Don't calculate 'currsecs' on every process, just on rank 0.
- Removed all calls to usingMPI() as Alan deprecated it in a recent commit.
- procOnLevel is (I hope) a boolean array... so declare and use it as such.
- Don't name a variable with a single letter. n -> var_cnt
- Did I mention white space cleanups... Please for the sake of older eyes...
M CCA/Components/ICE/ICE.cc
M CCA/Components/Models/FluidsBased/AdiabaticTable.cc
M CCA/Components/Models/FluidsBased/NonAdiabaticTable.cc
M CCA/Components/Models/FluidsBased/NonAdiabaticTable.h
M CCA/Components/Models/FluidsBased/SimpleRxn.cc
M CCA/Components/Models/FluidsBased/SimpleRxn.h
M CCA/Components/Models/Radiation/RMCRT/Radiometer.cc
M CCA/Components/OnTheFlyAnalysis/MinMax.cc
M CCA/Components/OnTheFlyAnalysis/1stLawThermo.cc
M CCA/Components/OnTheFlyAnalysis/containerExtract.cc
M CCA/Components/OnTheFlyAnalysis/lineExtract.cc
M CCA/Components/OnTheFlyAnalysis/momentumAnalysis.cc
M CCA/Components/OnTheFlyAnalysis/particleExtract.cc
M CCA/Components/OnTheFlyAnalysis/planeExtract.cc
M CCA/Components/OnTheFlyAnalysis/statistics.cc
- Use shared state getElapsedTime()
- Naming conventions: sharedState -> d_sharedState (to match other class vars in file).
- White space / alignments / etc.
M CCA/Components/LoadBalancers/CostProfiler.cc
- White space separation of #includes between CCA and Core
M CCA/Components/Arches/PressureSolverV2.cc
M CCA/Components/Arches/Radiation/RadPetscSolver.cc
M CCA/Components/Examples/Poisson2.cc
M CCA/Components/Examples/DOSweep.cc
M CCA/Components/Examples/RMCRT_Test.cc
M CCA/Components/Examples/RegridderTest.cc
M CCA/Components/Examples/SolverTest1.cc
M CCA/Components/FVM/ElectrostaticSolve.cc
M CCA/Components/ICE/impAMRICE.cc
M CCA/Components/ICE/impICE.cc
M CCA/Components/LoadBalancers/CostProfiler.h
M CCA/Components/LoadBalancers/DynamicLoadBalancer.cc
M CCA/Components/LoadBalancers/ParticleLoadBalancer.cc
M CCA/Components/MPM/AMRMPM.cc
M CCA/Components/MPM/ImpMPM.cc
M CCA/Components/MPM/SerialMPM.cc
M CCA/Components/Parent/Switcher.cc
M CCA/Components/ReduceUda/UdaReducer.h
M CCA/Components/ReduceUda/UdaReducer.cc
M CCA/Components/Regridder/RegridderCommon.h
M CCA/Components/Regridder/RegridderCommon.cc
M CCA/Components/Regridder/SingleLevelRegridder.cc
- LoadBalancer -> LoadBalancerPort
- White space cleanup / alignments.
M CCA/Components/LoadBalancers/LoadBalancerCommon.h
M CCA/Components/LoadBalancers/LoadBalancerCommon.cc
- LoadBalancer -> LoadBalancerPort.
- Add setNthRank().
- getOutputProc() -> getOutputRank().
- Whitespace / alignment.
- Fix comment to indicate getOutputProc is returning an MPI rank.
M CCA/Components/LoadBalancers/ProfileDriver.h
M CCA/Components/LoadBalancers/ProfileDriver.cc
M CCA/Components/Regridder/TiledRegridder.cc
M CCA/Components/Schedulers/MPIScheduler.cc
M CCA/Components/Schedulers/SchedulerCommon.h
M CCA/Components/Schedulers/SchedulerCommon.cc
- Cosmetics: Whitespace / alignment / coding standards / alphabetize #includes / etc.
- LoadBalancer -> LoadBalancerPort
M CCA/Components/Schedulers/OnDemandDataWarehouse.h
M CCA/Components/Schedulers/OnDemandDataWarehouse.cc
M CCA/Components/Schedulers/Relocate.h
M CCA/Components/Schedulers/Relocate.cc
- Name class variables according to naming convention. (eg: aborted -> d_aborted).
- Be consistent with naming convention of class variables in same file.
- LoadBalancer -> LoadBalancerPort.
M CCA/Components/Schedulers/TaskGraph.h
M CCA/Components/Schedulers/TaskGraph.cc
- LoadBalancer -> LoadBalancerPort.
- White space. Name variable "subset" more accurately: "handling_rank". I don't know
why it was named "subset" to begin with as it doesn't appear to be a "set" to me...
M CCA/Components/Schedulers/UnifiedScheduler.cc
- White space.
M CCA/Components/SimulationController/AMRSimulationController.cc
- There are a lot of cosmetic cleanups in this file, but the real/main change is:
- Add in hack to handle changing from "nth process output" (for checkpoints) to
"1 process output" (for time steps) when running with PIDX (and nth process is > 1).
- Declare variables (start) where they are used, not at the top of a function.
- Move the terminal output of the 0th timestep out of the main simulation loop.
- White space / alignment / etc.
- LoadBalancer -> LoadBalancerPort.
M CCA/Components/SimulationController/SimulationController.h
M CCA/Components/SimulationController/SimulationController.cc
M CCA/Components/Solvers/AMR/AMRSolver.cc
M CCA/Components/Solvers/AMR/HypreDriver.cc
M CCA/Components/Solvers/AMR/HypreDriverStruct.cc
M CCA/Components/Solvers/CGSolver.cc
M CCA/Components/Solvers/HypreSolver.cc
M CCA/Components/Wasatch/Expressions/Coordinate.cc
M CCA/Components/Wasatch/Expressions/PoissonExpression.cc
M CCA/Components/Wasatch/Expressions/Pressure.cc
M CCA/Components/Wasatch/Expressions/RadiationSource.cc
M CCA/Components/Wasatch/Wasatch.cc
M CCA/Ports/Scheduler.h
M CCA/Ports/sub.mk
M Core/DataArchive/DataArchive.h
M Core/DataArchive/DataArchive.cc
M Core/Grid/Grid.h
M Core/Grid/Grid.cc
M Core/Grid/Level.h
M Core/Grid/Level.cc
M StandAlone/tools/uda2vis/uda2vis.cc
- LoadBalancer -> LoadBalancerPort.
- Cosmetic white space, alpha, etc.
M CCA/Ports/Output.h
- Removed getCurrentTimestep() and getCurrentTime() as the user should get
them from the shared state.
- Added postponeNextOutputTimestep().
- Added savingAsPIDX().
62 lines of code changed in 13 files:
Remove SingleProcessor Scheduler/LoadBalancer.
Most of the work here has been in removing the need for Parallel::determineIfRunningUnderMPI(), as running Uintah with MPI is now an invariant, even with only a single process. We ALWAYS run Uintah with MPI.
The last simple step will be to remove usage of Parallel::usingMPI() (which now simply returns true), and also do away with the "-mpi" command line option. Right now sus has been modified to silently ignore "-mpi" and once the nightly RT scripts have been modified, we can deprecate usage.
Note that the following examples of a single process run are synonymous and all use the MPI scheduler with 1 rank:
./sus input.ups
./sus -mpi input.ups
mpirun -np 1 ./sus input.ups
mpirun -np 1 ./sus -mpi input.ups
75 lines of code changed in 8 files:
added additional vars and streams to the in-situ
62 lines of code changed in 3 files:
backed out sim state changes
5 lines of code changed in 1 file:
cleanup of the simulation state - most everything is now private
5 lines of code changed in 1 file:
Significant refactoring and clenaup in schedulers and loadbalancers prior to assesment of the required changes to the MPI engine to support Kokkos views. Also moving away from all mutex protected debug output, in favor of the light-weight, printf-based Dout class and DOUT macros. We now get fully coherent debug output, regardless of the proc/thread counts. Also moving to standardized naming conventions for class, static, thread-local and global variables.
167 lines of code changed in 10 files:
Introduction of the Lockfree Pool data structure and the new CommunicationList. Using this now to replace CommRecMPI and its use of the problematic MPI_Testsome() and MPIWaitsome() calls. We now store individual requests in a lock/wait/contention free Pool and call MPI_Test() and MPI_Wait on individual MPI_Requests.
This fixes the MPI_Buffer memory leak seen in the threaded scheduler, in which multiple threads think they will recieve a message, allocate a buffer and then only one thread does the actual recieve and calls the after-communcation handler to clean up the buffer. This memory leak was most pronounced at large scale with RMCRT due to the global halo requirement.
Also backing out the support for non-uniform ghost cells across AMR levels for now, until the issue of some required messages not being generated from coarse radiation mesh to some ranks. This has to do within partial dependencybatches being created due to an incomplete processor neighborhood list across levels.
MISC:
* cleaned up some old TAU remnants in doc directory and build system
* refactor and cleanup in MPIScheduler
* removed unused source code, including the old ThreadedScheduler (only one threaded scheduler now - Unified)
* cleaned up non-existent entreis in environmentalFlags.txt
2 lines of code changed in 1 file:
Fix failing AMR-ICE tests.
9 lines of code changed in 1 file:
Allow ghost cell requirements to vary across levels for tasks scheduled on the fine level, e.g. RMCRT_DO. This reduces the LoadBalancer::createNeighborhood complexity for RMCRT_DO significantly, e.g. from 45 minutes to 15 seconds with 64^3 patches on the fine level.
NOTE: This is a unique case and the solution was a trade off: a simple design for the edge-case far outweighed the complexity a generalized solution involved. This was effectively the addition of one integer in Task and SchedulerCommon and some simple logic elsewhere.
Also fixed some logic on sched_rayTrace_dataOnion to more clearly cycle through course levels for requriements.
82 lines of code changed in 1 file:
Significant refactoring of Level and LoadBalancerCommon prior to adding in per-level max_ghost_cells machinery.
386 lines of code changed in 4 files:
Redirect all MPI calls through the lightweight wrapper (header only)
This allows for standardized error checking and easy collection of runtime stats. All MPI funtions can be called from the wrapper by replacing:
MPI_ with Uintah::MPI::
e.g.
Uintah::MPI::Isend(...)
Uintah::MPI::Reduce(...)
Also enables MPI3 wrappers when MPI3 is available and protects against these when not. Will want MPI3 for non-blocking collectives. MPI3 is availalbe on Mira but not Titan yet.
* Note src/scripts/wrap_mpi_calls.sh has been added to the src tree. This has the sed foo to do this replacement src tree-wide.
* Have tested this with OpenMPI, MPICH, IntelMPI and also built on Titan and Mira.
30 lines of code changed in 6 files:
Revert -r55443, until folks can get compilers upgraded and buildbot MPI is updated.
30 lines of code changed in 5 files:
Redirect all MPI calls through the lightweight wrapper (header only)
This allows for standardized error checking and easy collection of runtime stats. All MPI funtions can be called from the wrapper by replacing:
MPI_ with Uintah::MPI::
e.g.
Uintah::MPI::Isend(...)
Uintah::MPI::Reduce(...)
Also enables MPI3 wrappers when MPI3 is available and protects against these when not. Will want MPI3 for non-blocking collectives. MPI3 is availalbe on Mira but not Titan yet.
* Note src/scripts/wrap_mpi_calls.sh has been added to the src tree. This has the sed foo to do this replacement src tree-wide.
* Have tested this with OpenMPI, MPICH, IntelMPI and also built on Titan and Mira.
30 lines of code changed in 5 files:
Revert r55429.
Still having an MPICH/OpenMPI compatibility problem.
30 lines of code changed in 5 files:
Redirect all MPI calls through the lightweight wrapper (header only)
This allows for standardized error checking and easy collection of runtime stats. All MPI funtions can be called from the wrapper by replacing:
MPI_ with Uintah::MPI::
e.g.
Uintah::MPI::Isend(...)
Uintah::MPI::Reduce(...)
Also enables MPI3 wrappers when MPI3 is available and protects against these when not. Will want MPI3 for non-blocking collectives. MPI3 is availalbe on Mira but not Titan yet.
* Note src/scripts/wrap_mpi_calls.sh has been added to the src tree. This has the sed foo to do this replacement src tree-wide.
30 lines of code changed in 5 files:
Removal of src/Core/Thread and related refactoring throughout the code-base.
This is the first step in a series of infrastrucutre overhauls to modernize Uintah. Though this all passes local RT (both CPU and GPU tests), I expect some fallout we haven't considered and will be standing by to deal with any issues. Once the dust settles, we will move to replacing Core/Malloc with jemalloc.
* We are now using the standard library for all multi-threading needs within the infrastructure, e.g. std::atomic, std::thread, std::mutex, etc.
* The Unified Scheduler is now the only multi-threaded scheduler, e.g. ThreadedMPIScheduler no longer exists (though the source will soon be placed into an attic).
* Threads spawned by the Unified Scheduler are detached by default (not joinable), allowing for easy, clean and independent execution. There are no longer ConditionVariables used to signal worker threads, just a simple enum for thread-state.
* What was Core/Thread/Time.* is now Core/Util/Time.* - a next step will be to migrate all internal timers, etc to use std::chrono.
* NOTE: Though much cleanup has occurred with this commit, there is still significant cleanup and formatting to be done. The scope of this commit neccessitates a more incremental approach.
12 lines of code changed in 5 files:
Move non-conflicting classes that were in SCIRun namespace to Uintah namespace.
3 lines of code changed in 7 files:
global: generalization
- d_timingStats -> d_runTimeStats
This infomap will hold more than just times.
SimulationController::printSimulationStats()
- output the units for each of the runTimeStats
DataArchiver::outputVariables()
- save timing info when outputting PIDX formatted files.
2 lines of code changed in 2 files:
fixed the pointer passing
2 lines of code changed in 1 file:
moved the timing stat over to infomapper
2 lines of code changed in 2 files:
Update copyright date to 2016.
23 lines of code changed in 28 files: