Lines of Code
[root]/CCA/Components/LoadBalancers
| Author | Changes | Lines of Code | Lines per Change |
|---|---|---|---|
| Totals | 105 (100.0%) | 1140 (100.0%) | 10.8 |
| allen | 55 (52.4%) | 581 (51.0%) | 10.5 |
| ahumphrey | 18 (17.1%) | 517 (45.4%) | 28.7 |
| jas | 29 (27.6%) | 26 (2.3%) | 0.8 |
| dav | 3 (2.9%) | 16 (1.4%) | 5.3 |
standardized the error output for the factories
0 lines of code changed in 2 files:
Beginnings MALLOC_TRACE support removal, e.g. MALLOC_TRACE_TAG_SCOPE and other macros.
This library, though once useful, no longer works with Uintah and is not maintained, plus inextricably tied to SCI_MALLOC, which ultimately needs to go away.
0 lines of code changed in 4 files:
added a SolverCommon class, made OnTheFlyAnalysis a first class UintahParallelComponent, added acess methods for component controll in UintahParallelComponent
20 lines of code changed in 4 files:
Cleanup of the Scheduler to use ports for all components
74 lines of code changed in 8 files:
more cleanup of the recompiling
81 lines of code changed in 12 files:
additional changes related to the Application Common
46 lines of code changed in 7 files:
update to move to a application common class and remove the shared state
123 lines of code changed in 6 files:
added processor node identification to the proc group, also cleaned up method names
186 lines of code changed in 13 files:
Fix some of the comparison between signed and unsigned integer expressions warnings.
5 lines of code changed in 3 files:
Both temporal and spatial scheduling working correctly for Radiometer.
Effectively, spatial tasks are scheduled "level->eachSpatialPatch()" as opposed to "level->eachPatch". Working to generalize this within the infratructure. For now, schedule spatial tasks independently (independent task for each patch the spatial task runs on), as opposed to scheduling with a spatial PatchSubset. This is so the load balancer can distribute patches more effectively (as opposed to lumping all spatial tasks on a particular node), and so that MPI dependencies can be generated and met correctly.
64 lines of code changed in 3 files:
Reintegration of incite_dev branch. Improvements/Features/Optimizations include (but not limited too):
1.) Temporal scheduling, e.g. support for multiple primary task graphs (not w-cycle)
2.) Significantly improved task graph compilation times for multi-level RMCRT at large core/patch counts
3.) GPU super patch support
4.) Highly optimized RMCRT GPU kernels
5.) Significant reduction in infrastructure memory footprint
ARCHES: methane_rmcrt test is broken due to issues with trunk code. Good luck Todd, let us know if you need another set of eyes.
WASATCH: RMCRT-Burns-Christon UDA comparison fails as the trunk input file specifies an unused variable to store. Branch code has fixed this.
-"compare_uda -ignoreVariable "VRFlux" rmcrt-burns-christon.uda.trunk rmcrt-burns-christon.uda.branch" compares exactly.
-just need to remove <save label="VRFlux"/> from original input and regenerate GS for this test.
Please let us know if anyone sees any residual issues with this commit. We are standing by to deal with any fallout.
145 lines of code changed in 2 files:
* Significant cleanup for RuntimeStats usage. Correspondingly, this cleans up the execute() methods for all schedulers.
* Consolidate Dout objects in RuntimeStats class - MPIStats, WaitTimes, and coopted ExecTimes SCI_DEBUG env vars now all work well. Note that MPIStats makes use of the individual timers for each MPI call in UintahMPI.h. This is the most accurate, per-MPI-call info we can get, e.g. Count, Time and number of messages for several message size categories.
* runTask() methods in MPIScheduler and UnifiedScheduler now feed the loadblancer contribution and exectimes map the same values (task execution time from beginning of doit() to end of done() ), computed by the DetailedTasks themselves.
* Removed more usage of usingMPI() condition. This method will ultimately be removed as it is meaningless now, because Uintah is always "usingMPI", meaning for even 1 MPI proc runs, Uintah inits the MPI runtime.
* Updated environmentalFlags.txt to reflect new offerings from RuntimeStats class.
1 lines of code changed in 1 file:
Update copyright.
21 lines of code changed in 26 files:
Removed Time.h and replaced with Timers.hpp
49 lines of code changed in 2 files:
Backing out committ 56299 for now. The optimizations made in this commit have triggered the DWDatabase race condition Brad and I discussed at a high level recently.
157 lines of code changed in 4 files:
Separate decl/impl of DetailedTask and DetailedTasks classes.
This drastically reduces confusion when trying to study any of this code (e.g. MPI engine and TG compile stuff).
Also removed significant amount of dead and/or redundant code.
More work to conform to what seems to be forming into a more widely adpoted coding standard.
150 lines of code changed in 4 files:
add min/max bounds to the in-situ vars, fixed handle checks
2 lines of code changed in 1 file:
Continuation of commit r56249 - getting into the Core and the UCF now.
Mostly cosmetic changes - removing the actual hack (see Handle.h below) to
allow for better type safety using c++11 nullptr.
These changes have passed the local RT on Baja.
- The biggest update in this commit (the update to remove the c++11 warning hack) is that you will
no longer be able to compare a ProblemSpecP to 0 (you will need to compare to 'nullptr').
- White space, coding standards, indentation, remove tabs.
- Line up some variables to make them easier to read.
- Use of 80+ columns of text (we decided to allow this 10 years ago) and it makes reading longer lines of code on the screen easier.
- Use {} for if/else clauses even if they are only one line long.
- Please put "else" and "else if" on a new line. The added white space makes following the branching much easier.
M CCA/Components/Arches/BoundaryConditions/BoundaryFunctorHelper.cc
M CCA/Components/LoadBalancers/DynamicLoadBalancer.cc
M CCA/Components/LoadBalancers/LoadBalancerCommon.cc
M CCA/Components/LoadBalancers/ParticleLoadBalancer.cc
M CCA/Components/Schedulers/DynamicMPIScheduler.cc
M CCA/Components/Schedulers/MPIScheduler.cc
M CCA/Components/Schedulers/UnifiedScheduler.cc
M CCA/Ports/PIDXOutputContext.cc
M Core/GeometryPiece/GeometryPieceFactory.cc
M Core/Grid/BoundaryConditions/BCDataArray.cc
M Core/Grid/BoundaryConditions/BCDataArray.h
- Pass const strings as references instead of copying.
M Core/Grid/BoundaryConditions/BCUtils.cc
M Core/Grid/BoundaryConditions/BoundCondReader.cc
M Core/Grid/Level.h
M Core/Grid/Level.cc
- Pass const params as const.
- Removed m_patches variables as it isn't used anywhere.
- Very few real changes in this file, use diff -w.
M Core/Grid/Patch.h
M Core/Grid/Patch.cc
- Place .h one line functions on one line as it makes the viewing of the Class spec easier
(and this is the main purpose of the .h file.)
- Line up vars/ for easier reading.
M Core/Grid/Task.cc
M Core/Grid/Variables/ComputeSet.h
- Place .h one line functions on one line as it makes the viewing of the Class spec easier
(and this is the main purpose of the .h file.)
M Core/Grid/Variables/ParticleVariable.h
M Core/Grid/Variables/ParticleVariable_special.cc
M Core/Grid/Variables/Variable.cc
- Fix end comment on readPIDX()
M Core/Util/Handle.h
- Here is the hack that is being removed... The "==" operator was overloaded in order
to allow for comparison of pointers to 0. When this is removed, then (previous to these
commits) you would be spammed by warnings about ambiguous "==" operator. By removing this
operator and forcing the comparison to nullptr (as of c++11), type safety is increased.
16 lines of code changed in 3 files: