BoundaryConditions
(31 files, 6226 lines)
PatchBVH
(9 files, 821 lines)
Variables
(91 files, 13344 lines)
[root]/Core/Grid
BoundaryConditions
(31 files, 6226 lines)
PatchBVH
(9 files, 821 lines)
Variables
(91 files, 13344 lines)
| Author | Changes | Lines of Code | Lines per Change |
|---|---|---|---|
| Totals | 204 (100.0%) | 2515 (100.0%) | 12.3 |
| allen | 41 (20.1%) | 1171 (46.6%) | 28.5 |
| mcconnell | 6 (2.9%) | 400 (15.9%) | 66.6 |
| ahumphrey | 27 (13.2%) | 366 (14.6%) | 13.5 |
| guilkey | 1 (0.5%) | 340 (13.5%) | 340.0 |
| harman | 17 (8.3%) | 106 (4.2%) | 6.2 |
| jas | 100 (49.0%) | 61 (2.4%) | 0.6 |
| jholmen | 5 (2.5%) | 45 (1.8%) | 9.0 |
| dav | 5 (2.5%) | 26 (1.0%) | 5.2 |
| jthornoc | 2 (1.0%) | 0 (0.0%) | 0.0 |
printTaskLevels()
- Can now handle tasks with patches on multiple levels.
0 lines of code changed in 2 files:
Generalize the concept of a task graph having tasks with "distal" halo requirements, meaning the number, order, and appearance of TGs with distal requires is arbitrary. This is a much more robust way to determine distal neighborhood creation, etc. Prior to this, the code was quite brittle, depending on the distal TG being last in the order.
14 lines of code changed in 4 files:
Declare "i" in "for" loop rather than outside to get rid of several
warnings.
0 lines of code changed in 2 files:
* Brace initializers for POD and pointer_type class members.
* cbegin() and/or cend() for const_iterators, especially for methods that return a const_iterator, e.g., Level::const_patch_iterator Level::patchesBegin() const.
* Use prefix ++ operator (e.g., ++iter) for more complicated types, avoiding tmp creation.
* Prefer using const when possible.
* Use auto keyword, leaving correct type deduction to the compiler.
12 lines of code changed in 4 files:
Update and revive DW task access checking and RunningTaskInfo code.
Fixed a few other typos and formatting in these areas while under the hood.
25 lines of code changed in 2 files:
Infrastructure restructuring and cleanup while starting last thesis-related task graph work.
Big things that I will always change when I see them.... and I saw much of this recently:
1.) ALWAYS use {} for 1-line conditionals.
if (<predicate>) {
// do stuff
}
2.) Use C++ static _cast over C-style casts, static_cast<int>(myvar) is checked at compile time, C casts are not and can fail at runtime.
3.) use "nullptr" over "0 or NULL" in the context of pointers and their assignment. Using NULL or 0 can be ambigous, e.g., compiler converting to non-pointer type 'int' from NULL (nullptr can't be assigned to an integral type). nullptr is a keyword and usage/intent is clear. From the standard: "The pointer literal is the keyword nullptr. It is an rvalue of type std::nullptr_t.".
4.) Things like Dout/Debugstreams that are specific to a compilation unit should be placed into an unnamed namespace (which is a utility to make an identifier translation unit local). The effect is almost identical to decalring something static.
5.) Use meanigful names, a bit lengthy is OK if it helps clarity. Remember, the compiler doesn't care about your names (values just go into registers), but the people maintaining your code do.. Code is meant to be read and maintained.
6.) Whitespace is our friend, but 80 column lines went away long ago.....
140 lines of code changed in 2 files:
Reworking the fastCpdiInterpolator to revert to the regular cpdiInterpolator
if particles span more than two grid cells in any direction. Some additional
modifications might be forthcoming.
340 lines of code changed in 1 file:
planeAverage:
- fixed memory leaks
- Creat a custom MPI_communicator. In multilevel/processor problems there's no guarentee that every rank
will own patches. Thanks Dav and Alan.
- Return immediately if a task has an empty patchSubset.
DbgOutput:
- printTask(): allow for empty patchSubsets.
41 lines of code changed in 1 file:
planeAverage.*
- switched from debugStream to Dout so the output doesn't become garbled when running in parallel.
- Decreased number of significant figures in output from 15 to 14. This hides the numerical fuzz
that's picked up when running in parallel. This allows for direct comparisons beweeen serial output
- Added clone virtual function allowing for deep copies of vector < vector <shared_ptr> > d_allLevels_planarVars
- computePlanarSums(): removed flow control logic d_progressVar. This task needs to be computed on all patches.
- Added resetProgressVar() task. It's trival but has to execute after the other tasks.
*** single and 4 core test problem are giving the same results for 2 level advect problem.
DbgOutput.*
- Added another implementation of printTask()
25 lines of code changed in 3 files:
Replaced tabs with spaces.
20 lines of code changed in 6 files:
display()
- revert commit 58886. Onceperproc patch sets can contain patches from several levels causing
getLevel() to puke.
2 lines of code changed in 1 file:
display()
- output level index for oncePerProc tasks
1 lines of code changed in 1 file:
deprecated the SimulationTime class and moved it to be fully contained in the ApplicationCommon
0 lines of code changed in 4 files:
changed the delta T reduction to be last and modified the var name to match the xml name
15 lines of code changed in 2 files:
removal of shared state and introduction of a material manager
74 lines of code changed in 13 files:
initial commit for the new material manager aka old shared state
494 lines of code changed in 3 files:
added default values
13 lines of code changed in 1 file:
Level::getTotalCellsInRegion()
- Take into account different variable types (CC, NC, SFC(*)) when computing the patch extents.
14 lines of code changed in 2 files:
Changed exception message to be more descriptive.
3 lines of code changed in 1 file:
This commit brings the PIDX data I/O branch (named: pidx_xml) back into the trunk.
All PIDX data I/O is, by default, turned off. To use PIDX you have to configure
against the PIDX library (either installing manually and pointing at it, or using
the --with-pidx flag (not 100% certain that this works at the moment)), and you must
set <DataArchiver type="PIDX"> (as opposed to the default which is type="UDA") in
the .ups input file.
Other caveats:
- Particles are not (currently?) supported in PIDX so any simulation that has a
particle data type will fail (most will print an error that PIDX does not
support them, but it seems some just die).
- Simulations that have overlapping patches (this may just be AMR type
sims) (eg: multibox_rmcrt_coal_1L) can save (slightly different) results.
In the multibox_rmcrt_coal_1L, we saw (only) a handful of differences like this:
((A 1 vs a 0 in a couple of locations) Also note that the UDA itself compared
successfully, but the checkpoint underneath it did not...)
DIFFERENCE [int 36, 5, 0] Time: 0.0003106658 Level: 0 Patch1: 0 Patch2: 0 Material: 0 Variable: areaFractionFX
multibox_RMCRT_coal_1L.uda/checkpoints goldStandard/.../multibox_RMCRT_coal_1L.uda/checkpoints
1 0
- "#if HAVE_PIDX" is used inconsistently through a number of files. Some PIDX stuff is #if'd out, while other code
is left in and compiles, it is just not used.
- There are a lot of cleanups that can and should take place wrt the PIDX integration.
While the code passes all the local RT tests (with the exception of particle tests),
I don't believe it has been run at large scale (though it should work) on a BGQ or Quartz@llnl.
Summary of changes: (Note, see pidx_xml branch commits if you want more information.)
M CCA/Ports/ApplicationInterface.h
M CCA/Components/Application/ApplicationCommon.h
- Added the ability to record (and retrieve) the number for the last regrid time step.
M CCA/Components/DataArchiver/DataArchiver.h
M CCA/Components/DataArchiver/DataArchiver.cc
- When saving as PIDX, only create/output timestep.xml on the initial (I/O) time step that occurs after the regrid.
- Fixes that allow PIDX to save correctly (and efficiently) in parallel. If you are ever trying
to trace this back, read the comments from the commit logs for the "pidx_xml" branch.
- Remove createPIDX_dirs() as it was commented out and Sid tells me that PIDX creats its own dirs and we don't need this.
- Lot of white space fixes.
M CCA/Components/Examples/RMCRT_Test.cc
M CCA/Components/PostProcessUda/PostProcess.cc
- The DataArchive (DA) queryVariables() function now also returns the number of materials for each variable.
- "else" and "else if" should be on a separate line to make it easier to read.
M CCA/Components/Models/Radiation/RMCRT/RayGPU.cc
- Differentiate the output between the GPU and non-GPU reports so the viewer knows explicitly whether the GPU code was run.
M CCA/Components/Schedulers/OnDemandDataWarehouse.h
M CCA/Components/Schedulers/OnDemandDataWarehouse.cc
- Cosmetics: Line up variables, use more than 80 chars per line, white space.
Add in comments denoting default param values.
M CCA/Components/Schedulers/SchedulerCommon.cc
- Cosmetics.
M CCA/Components/SimulationController/AMRSimulationController.cc
- Cosmetics: Use more white space, 80+ chars per line for better readability, etc.
M CCA/Ports/PIDXOutputContext.h
- const'ify some params.
- checkReturnCode() is a static function so mark it so.
M CCA/Ports/PIDXOutputContext.cc
- White spacee. Code alignment for readability.
M Core/DataArchive/DataArchive.h
M Core/DataArchive/DataArchive.cc
- Add in PIDX functions. Note, queryPIDXSerial() is used (currently) by standalone codes (eg: puda).
Its purpose is to wrap the setupQueryPIDX()/queryPIDX() calls that have been added to
allow PIDX to truly run in parallel. At some point it would be good to update puda/compare_uda/etc
to run in parallel...
- See previous comments from PIDX branch for more details on the PIDX communicators that are necessary
to transfer data as some data does not live on all cores and thus can't use the golbal comm.
- Rename a few functions (queryAndSetParticlePositionName()) to be more accurate as to what they do.
- There are still a number of hacks in this code that need to be addressed (eg: var_materials map).
M Core/Geometry/IntVector.h
- Mostly cosmetics...
White space for readability.
Place 1-line functions in .h on one line to make it easier to see all provided functions.
M Core/Grid/Grid.cc
- White space, variable alignment for readability.
M Core/Grid/Level.h
M Core/Grid/Level.cc
- White space, coding standard (return type on its own line).
- Fix logic error in computeVariableExtents() that did not handle periodic boundaries correctly.
This fix was suggested as probably the correct way to do it by Todd... seems to pass tests and did fix
the issue (PIDX saving a particular something or other) that led me to it.
M Core/Grid/Patch.cc
- Place "else" on its own line for readability.
M Core/Grid/Variables/GridVariable.h
- Added emitPIDX().
- Opening "{" for functions in .h files goes on same line.
M Core/Grid/Variables/GridVariableBase.h
M Core/Grid/Variables/GridVariableBase.cc
- Organize #includes.
- White space, line up vars, etc for better readability.
- Return type on its own line.
M Core/Grid/Variables/ParticleData.h
- #include ParticleSubset.h for particleIndex.
M Core/Grid/Variables/ParticleSubset.h
- Cosmetic white space.
M Core/Grid/Variables/ParticleVariable.h
- White space, variable line up, etc.
- Added emitPIDX() function for particles. Note, particles do not fully work yet (on PIDX side).
And thus, this function hasn't really been tested.
M Core/Grid/Variables/ParticleVariableBase.h
M Core/Grid/Variables/ParticleVariableBase.cc
M Core/Grid/Variables/PerPatchBase.h
M Core/Grid/Variables/PerPatchBase.cc
M Core/Grid/Variables/ReductionVariableBase.h
M Core/Grid/Variables/ReductionVariableBase.cc
- Cosmetic white space. Coding standards. Etc.
M Core/Grid/Variables/Variable.h
M Core/Grid/Variables/Variable.cc
- Organize #includes.
- Use "const" on params that are const.
- Cosmetic white space, etc.
M Core/Grid/Variables/VarnameMatlPatch.h
- Pass non-changing strings by const reference (instead of copying them).
- White space for readability.
M Core/ProblemSpec/ProblemSpec.cc
- Line up vars, white space.
M R_Tester/helpers/compare_dat_files.pl
- Strip "/" from end of uda names.
- Clean up output to make it more readable (Todd I hope this doesn't break some of your scripts...)
M R_Tester/helpers/compare_dats
- White space.
M R_Tester/helpers/runSusTests.py
- Fix typo.
M StandAlone/compare_uda.cc
- White space.
- Line up vars.
- Coding standard.
- Line up case statement for better readability.
- 0 => nullptr
- query() now returns if the var was found (because PIDX has a list of all variables the simulation starts with
and the vars may not actually exist in the UDA.
- var => var_name (as it is actually the name, and not the var).
- Use 80+ columns.
M StandAlone/partvarRange.cco
M StandAlone/selectpart.cc
M StandAlone/tools/compute_Lnorm_udas.cc
M StandAlone/tools/dumpfields/dumpfields.cc
M StandAlone/tools/extractors/extractF.cc
M StandAlone/tools/extractors/extractS.cc
M StandAlone/tools/extractors/extractV.cc
M StandAlone/tools/extractors/faceextract.cc
M StandAlone/tools/extractors/lineextract.cc
M StandAlone/tools/extractors/partextract.cc
M StandAlone/tools/extractors/particle2tiff.cc
M StandAlone/tools/extractors/timeextract.cc
M StandAlone/tools/puda/AA_MMS.cc
M StandAlone/tools/puda/ER_MMS.cc
M StandAlone/tools/puda/GV_MMS.cc
M StandAlone/tools/puda/ICE_momentum.cc
M StandAlone/tools/puda/PIC.cc
M StandAlone/tools/puda/POL.cc
M StandAlone/tools/puda/asci.cc
M StandAlone/tools/puda/jacquie.cc
M StandAlone/tools/puda/jim1.cc
M StandAlone/tools/puda/jim2.cc
M StandAlone/tools/puda/jim3.cc
M StandAlone/tools/puda/pressure.cc
M StandAlone/tools/puda/puda.cc
M VisIt/interfaces/archiveInterface.cc
- queryVariables() now returns number of materials.
M StandAlone/tools/puda/varsummary.h
M StandAlone/tools/puda/varsummary.cc
- Swat some evil endls.
M build_scripts/build_pidx.sh
- FYI... Not sure if changing PIDX_TAG to "master" actually works...
M buildbot_try.sh
- Added a number of comments about use of this script.
- Fixed some typos.
- Spread command to multiple lines for readability.
26 lines of code changed in 5 files:
Remove unused TaskGraph data structure. This was a remnant from the old topological sort, which has since been removed along with supporting code.
3 lines of code changed in 1 file:
rolled back changes to r58119
317 lines of code changed in 2 files:
Update copyright date.
2 lines of code changed in 2 files:
cleanup of depencendies to applications
81 lines of code changed in 2 files:
moved Todd'd print to here
25 lines of code changed in 2 files:
For infrastructure: Update existing DebugStreams and Dout objects to be constructed with a desription matching that found in environmentalFlags.txt. EnvironmentalFlags,txt itself also updated.
Components can individually go in and do this same thing now, ultimately deprecating and eventually removing environmentalFlags.txt (hopefully).
2 lines of code changed in 3 files:
Make sure to set SimulationState MaterialSet objects to nullptr in SimulationState::clearMaterials(). This was causing the problem with the Switcher and IMPM, etc. Valgrind gives the failing Switcher tests a clean bill of health now.
A few other minor, cosmetic changes.
8 lines of code changed in 1 file:
Remove redundant code in SimulationState::clearMaterials.
With the switcher component, this was incorrectly decrementing the ref count on material sets, which is likely the reason ComputeSubsets were being deleted prematurely and subsequently casusing double delete issues when switching occurred. This problem would not manifest in anything but the switcher, as clearing materials only happened implicitly by the infrastructure via SimulationState DTOR. Switcher was calling it explicitly in Switcher::needRecompile().
Some other cleanup, e.g 0->nullptr, braces, white space, etc.
125 lines of code changed in 3 files:
cleanup of the namespace for debug streams and removed using namespace std
110 lines of code changed in 6 files:
Cleanup, update debug output, default initialization (in header files, e.g. m_ptr{nullptr}), etc
33 lines of code changed in 4 files:
minor remaning of debug stream names
3 lines of code changed in 2 files:
addition debug stream naming
1 lines of code changed in 1 file:
restructured teh Dout and DebugStream to register the streams to avoid conflicts
4 lines of code changed in 2 files:
added code to remove warning for opt builds
3 lines of code changed in 1 file:
Remove the machinery for shared DetailedTasks object. (1) this no longer makes sense in a multi-task graph environement, as teh whole point of multile primary graphs is that the DetailedTasks set will differ, (2) turns out this original "enhancement" and all supporting code was not even being excercised (for many years now), and (3) when actually enabled, any benifit was negligible relative to the overall complexity it introduced, even for the shared scrub table.
Many other refactorings, cleanups and other cosmetic changes, all as part of the next wave of task graph optimizations.
4 lines of code changed in 3 files:
* Added TaskType Hypre.
* Updated TaskType logic to support Task::Hypre.
* Updated Kokkos-OpenMP Scheduler logic to exit OpenMP thread partitions when a Hypre task is selected.
* Cleaned up whitespace and formatting.
This commit allows Hypre tasks to be differentiated from OncePerProc tasks. This differentiation is used within the Kokkos-OpenMP Scheduler to make all OpenMP threads within an MPI process available to a Hypre task as opposed to only those within the OpenMP thread partition that selected the Hypre task.
39 lines of code changed in 3 files:
* Updated KokkosOpenMPScheduler.cc to use Kokkos::OpenMP::partition_master() for Kokkos-based builds.
* Added -npartitions and -nthreadsperpartition run-time options.
* Set the Kokkos-OpenMP Scheduler as the default scheduler for Kokkos-based builds.
* Added bulletproofing within SchedulerFactory.cc.
* Updated ProcessorGroup.h to use m_comm for -nthreads 1 or -npartitions 1.
* Added MasterLock.h to help avoid mixing synchronization primitives.
* Replaced use of #include <mutex> with #include <Core/Parallel/MasterLock.h>.
* Replaced use of std::mutex with Uintah::MasterLock.
* Cleaned up whitespace.
This commit extends the Kokkos-OpenMP Scheduler to support parallel execution of tasks within an MPI process.
Scheduler logic is based upon that of the Unified Scheduler, where individual threads are used as task runners.
In the context of the Kokkos-OpenMP Scheduler, individual partitions are used as task runners.
Individual partitions execute Kokkos-based tasks in parallel across the threads belonging to that partition and existing serial tasks serially.
Note, this scheduler requires MPI_THREAD_MULTIPLE support to run with multiple partitions per MPI process.
6 lines of code changed in 2 files:
added missing delete calls
48 lines of code changed in 1 file:
fixed copy paste error in check
2 lines of code changed in 1 file:
Update copyright date.
61 lines of code changed in 100 files:
(1 more)