Lines of Code
[root]/CCA/Components/Schedulers
| Author | Changes | Lines of Code | Lines per Change |
|---|---|---|---|
| Totals | 386 (100.0%) | 11083 (100.0%) | 28.7 |
| ahumphrey | 197 (51.0%) | 8981 (81.0%) | 45.5 |
| harman | 54 (14.0%) | 882 (8.0%) | 16.3 |
| dav | 20 (5.2%) | 689 (6.2%) | 34.4 |
| jas | 91 (23.6%) | 293 (2.6%) | 3.2 |
| guilkey | 3 (0.8%) | 83 (0.7%) | 27.6 |
| knolla | 2 (0.5%) | 73 (0.7%) | 36.5 |
| jholmen | 8 (2.1%) | 73 (0.7%) | 9.1 |
| jsutherland | 7 (1.8%) | 7 (0.1%) | 1.0 |
| sidharth | 4 (1.0%) | 2 (0.0%) | 0.5 |
If
SCI_DEBUG ExecTimes:+
is set output the exectimes files.
0 lines of code changed in 4 files:
Make sure and distinguish between ExecTimes, WaitTimes and MPIScheduler_TimingsOut SCI_DEBUG flags. This was missed when this code, common to all schedulers was extracted into a method (outputTimingStats) and put into MPIScheduler.
0 lines of code changed in 2 files:
Some cleanup and work toward refining and better understanding MPI timings and stats.
Pull ancient VAMPIR weeds.
125 lines of code changed in 7 files:
Unified scheduler can run multi-threaded on intial timestep. This speeds up the 0th timestep considerably, especially with large problems and particle creation, etc.
4 lines of code changed in 1 file:
Consolidate ExecTimes, WaitTimes, and MPI Timing code into parent (MPI) scheduler. This eliminates miles of redundnat and error-prone code. Now it only has to be correct in one place.
162 lines of code changed in 7 files:
PIDX support if configured with --with-pidx option : works only for CCtype grids
2 lines of code changed in 4 files:
Make the SCI_DEBUG flag "DynamicMPI_TimingsOut" actually output timings files. This is needed for the Sandia frameworks tech report.
Also do some various formatting, cleanup and add some comments.
51 lines of code changed in 3 files:
Remove antiquated TAU support.
15 lines of code changed in 13 files:
Added method:
bool isRestartInitTimestep()
A component can query the scheduler and determine if it is the initial timestep after
a restart.
AMRSimulationController: run()
- moved
bool first = true;
if (d_restarting) {
d_scheduler->setRestartInitTimestep(first);
}
upstream, before doInitialTimestep().
3 lines of code changed in 2 files:
Revert "- Get the framework to compile when C++11 is turned on."
We can remove this once we switch over to C++11 entirely. But not quite yet...
3 lines of code changed in 4 files:
- Get the framework to compile when C++11 is turned on.
- Remove sci_algorithm.h. This must have been from prehistoric times.
4 lines of code changed in 3 files:
print()
- if the "domain" is not defined then don't print it.
0 lines of code changed in 2 files:
added PIDX code to DataArchiver
73 lines of code changed in 2 files:
level: added operator<<
DWDatabase: print()
- outputing patch or level information depending on domainType.
1 lines of code changed in 1 file:
print()
- more descriptive output.
0 Name: abskg domain ptr: 0x1b664e0 ID: 2 matl:0
0 Name: sigmaT4 domain ptr: 0x1b665a0 ID: 3 matl:0
0 Name: abskg domain ptr: 0x1b665a0 ID: 3 matl:0
0 Name: cellType domain ptr: 0x1b664e0 ID: 2 matl:0
0 Name: sigmaT4 domain ptr: 0x1b66480 ID: 1 matl:0
0 Name: cellType domain ptr: 0x1b665a0 ID: 3 matl:0
2 lines of code changed in 1 file:
* Added output to exectimes.x.x files to indicate which timesteps that TaskExecTimes have been accumulated over.
* Updated execsummary output to indicate which timesteps that TaskExecTimes have been accumulated over.
29 lines of code changed in 4 files:
Revert commit 53533: For once per proc task, turn off all but one thread.
1 lines of code changed in 1 file:
Accurately report specs on all available on-node GPUs.
5 lines of code changed in 1 file:
Comment out multi-GPU code that's not working on local SCI cluster. Will fix this soon.
7 lines of code changed in 1 file:
Remove ProblemSetupException thrown when trying to use > 1 available GPU per node. We correctly support multiple GPUs.
5 lines of code changed in 1 file:
getItem() & getLevelItem()
- You must find the patchID or levelIndex before searching the variable DB for the label. In simulations
with a large number of patches the previous method would find the label for the wrong patch and
crash.
This bug was tricky!
0 lines of code changed in 4 files:
Revert c++-11 namespace changes.
93 lines of code changed in 14 files:
Explicit namespace usage as part of the transition to c++-11.
92 lines of code changed in 10 files:
Remove the commented out <omp.h> since we are not using any omp calls.
0 lines of code changed in 1 file:
Talked to John and omp.h (OpenMP) is not needed in this file. Removing it to fix Mac build.
1 lines of code changed in 1 file:
For once per proc task, turn off all but one thread.
59 lines of code changed in 1 file:
Throw an exception if the user is trying to use multiple GPUs. The simulation chugs along
as if it's working when really the some of the GPU kernels are being skipped. This was
observed on a machine with two different GPUs and may not hold when the GPUs are the same make/model.
10 lines of code changed in 1 file:
Changes required to use c++-11.
6 lines of code changed in 3 files:
removed print statement that clutters the mind.
0 lines of code changed in 1 file:
gpu_stats formatting and now outputting the nCells. It's much easier to follow.
22 lines of code changed in 1 file:
postH2DCopies()
- simplified logic for deciding if a variable is a "levelItem."
- Moved H2dRequiresLock up stream, needed to avoid issues when multiple threads
query existsLevelDB() simultaneously. Kudos to Dav for helping on this!
preallocateDeviceMemory() & postD2HCopies()
- gpu_stats: Added the number cells in the x,y,z dirs.
137 lines of code changed in 1 file:
allocateAndPut()
- added comments
4 lines of code changed in 1 file:
Now outputting MPI rank and threadID in taskdbg statements. You can now
grep <rank>.<ID> so watch what each thread is doing.
The getRankThread() function should be moved upstream so others can use it.
10 lines of code changed in 1 file:
Initialize the cudaStream to NULL.
<< operator:
- if a task is using the cuda stream then output the stream address.
7 lines of code changed in 1 file:
Cleanup TaskDBG output in threaded schedulers, reporting in ThreadedMPI and Unified schedulers has not been consistent. Also restrict thread-specific debugstream output.
16 lines of code changed in 2 files:
Prepend all gpustats output with mpirank.threadID information. For example:
0.0 Post H2D copy of REQUIRES ( abskg), L-1, patch: 1, size (bytes) = 157216 from 0x168eb90 to 0x1304a40000, using stream 0x7f44d80008c0
0.1 Post H2D copy of REQUIRES ( abskg), L-1, patch: 2, size (bytes) = 157216 from 0x16eeb10 to 0x1304a66800, using stream 0x1bc5880
Useful when debugging thread issues.
44 lines of code changed in 2 files:
postH2DCopies()
- output cuda stream pointer in DebugStream gpuStats
- removed repetitive calls to dw->getGPUDW()
30 lines of code changed in 1 file:
Updated method names. Should have been in previous commit.
3 lines of code changed in 1 file:
added
existsLevelDB() // checks for variable in levelDB
removeLevelDB() // flags variable for removal
clear()
- added loop over levelDB items and remove them.
89 lines of code changed in 2 files:
Throw an exception if the user tries to use the -gpu option without the Unified scheduler.
7 lines of code changed in 1 file:
(82 more)