Directory Core/Parallel/

Lines of Code

Developers

Most Recent Commits

allen 2017-12-13 18:22 Rev.: 57440

added a SolverCommon class, made OnTheFlyAnalysis a first class UintahParallelComponent, added acess methods for component controll in UintahParallelComponent

3 lines of code changed in 2 files:

• Core/Parallel: UintahParallelComponent.h (new)

ahumphrey 2017-11-29 13:55 Rev.: 57370

Guard the DataArchiver DebugStream (check if dbg.active()). This is causing issues when not active on mira. DebugStreams need to go away in favor of much simpler Dout, as DebugStreams unnecessarily inherit from the standard library (std::ostream) and are unduley compicated for basic debug output.

Also cleaned up poor whitespace, linebreaks and indentation, other formatting. Seems like someone went through the infrastructure with a faulty code formatter previously...

31 lines of code changed in 1 file:

• Core/Parallel: UintahParallelComponent.h (+31 -26)

allen 2017-11-22 21:12 Rev.: 57334

additional changes related to the Application Common

3 lines of code changed in 1 file:

• Core/Parallel: UintahParallelComponent.h (+3 -1)

allen 2017-10-25 12:48 Rev.: 57176

added processor node identification to the proc group, also cleaned up method names

96 lines of code changed in 6 files:

• Core/Parallel: Parallel.cc (+3 -3), ProcessorGroup.cc (new), ProcessorGroup.h (+18 -4)

ahumphrey 2017-07-13 12:37 Rev.: 56781

Working with SFC and loadblancer functionality, re-enable SFC test.
Other minor cleanup, fix some comments.

1 lines of code changed in 2 files:

• Core/Parallel: Parallel.h (+1 -2)

ahumphrey 2017-07-10 15:33 Rev.: 56777

Simplfy unnecessarily complicated proc0cout macro.
Just print from proc0/masterThread (T0), always the same regardless of number of threads or scheduler type.

2 lines of code changed in 1 file:

• Core/Parallel: Parallel.h (+2 -4)

ahumphrey 2017-06-06 14:15 Rev.: 56609

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.

2 lines of code changed in 1 file:

• Core/Parallel: Parallel.h (+2 -1)

ahumphrey 2017-04-25 22:40 Rev.: 56480

* 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.

4 lines of code changed in 2 files:

• Core/Parallel: UintahMPI.h (+4 -4)

jas 2017-04-20 13:18 Rev.: 56452

Update copyright.

16 lines of code changed in 26 files:

• Core/Parallel: BufferInfo.cc (new), BufferInfo.h (+1 -1), CrowdMonitor.hpp (new), PackBufferInfo.cc (+1 -1), PackBufferInfo.h (new), Parallel.cc (+1 -1), Parallel.h (+1 -1), ParallelContext.h (+1 -1), ProcessorGroup.cc (+1 -1), ProcessorGroup.h (+1 -1), UintahMPI.h (+1 -1), UintahParallelComponent.cc (+1 -1), UintahParallelComponent.h (+1 -1), UintahParallelPort.cc (new), UintahParallelPort.h (+1 -1), sub.mk (+1 -1)

allen 2017-04-18 21:09 Rev.: 56445

removed needless incude of time.h

0 lines of code changed in 1 file:

• Core/Parallel: Parallel.cc (-6)

allen 2017-04-13 16:24 Rev.: 56421

Removed Time.h and replaced with Timers.hpp

15 lines of code changed in 1 file:

• Core/Parallel: Parallel.cc (+15 -2)

ahumphrey 2017-02-28 17:43 Rev.: 56270

Continuing with infrastrucutre cleanup and refactoring:
-----------------------------------------------
- White space, coding standards, indentation, remove tabs.
- Line up some variables to make them easier to read.
- Use of 80+ columns of text - this was somewhat archaic
- Use {} for if/else clauses even if they are only one line long.
- Put "else" and "else if" on a new line. The added white space makes following the branching much easier.
- remove instances of "using namespace::std" when found.
- nullptr checks


Class Member Naming Conventions:
-----------------------------------------------
m_var_name - class member
s_var_name - static
g_var_name - global
t_var_name - thread local

229 lines of code changed in 4 files:

• Core/Parallel: BufferInfo.cc (+81 -72), BufferInfo.h (+50 -36), PackBufferInfo.cc (+58 -53), PackBufferInfo.h (+40 -29)