/trunk/src Developers: qymeng

Activity by Clock Time

Activity in Directories

Most Recent Commits

qymeng 2014-05-22 13:01 Rev.: 51864

When a reduction variable being computed by multiple tasks,
schedule the reduction task after the very last task that computes this variable.

27 lines of code changed in 5 files:

• CCA/Components/MPM: AMRMPM.cc (new)
• CCA/Components/Schedulers: SchedulerCommon.cc (+11 -1), SchedulerCommon.h (+3), TaskGraph.cc (+12)

qymeng 2014-05-16 10:16 Rev.: 51845

remove reduction task map

0 lines of code changed in 3 files:

• CCA/Components/Schedulers: SchedulerCommon.cc (-2), SchedulerCommon.h (-3), TaskGraph.cc (-12)

qymeng 2014-05-14 12:36 Rev.: 51836

restrict to use only one thread for the first timestep
We should treat restart and init timestep separately in the future,
so that init timestep can use multi-thread while restart timestep use only one thread.

6 lines of code changed in 2 files:

• CCA/Components/MPM: AMRMPM.cc (+1)
• CCA/Components/Schedulers: UnifiedScheduler.cc (+5 -2)

qymeng 2014-05-01 10:39 Rev.: 51796

fix hash map namspace issue

6 lines of code changed in 3 files:

• CCA/Components/Schedulers: DWDatabase.h (+1 -5)
• include: sci_hash_map.h (new)

qymeng 2014-05-01 10:08 Rev.: 51795

Removed many of datawarehouse locks by using atomic operations

401 lines of code changed in 19 files:

• CCA/Components/ICE: impAMRICE.cc (+3), impICE.cc (+2)
• CCA/Components/MPM: ImpMPM.cc (+2)
• CCA/Components/Schedulers: DWDatabase.h (+234 -113), DetailedTasks.cc (+34 -1), DetailedTasks.h (+10), OnDemandDataWarehouse.cc (+73 -167), OnDemandDataWarehouse.h (+9 -3), SchedulerCommon.cc (+19 -5), SchedulerCommon.h (+4), TaskGraph.cc (+2 -1), UnifiedScheduler.cc (+1 -1)
• CCA/Components/SimulationController: AMRSimulationController.cc (+3 -1)
• CCA/Ports: DataWarehouse.h (+1), Scheduler.h (+2)
• Core/Grid: UnknownVariable.cc (+2 -2)

qymeng 2014-03-04 17:32 Rev.: 51591

add computesWithScratchGhost for Wasatch

37 lines of code changed in 2 files:

• Core/Grid: Task.cc (+26), Task.h (+11 -4)

qymeng 2014-02-07 17:21 Rev.: 51538

better scheduling copydata timestep

32 lines of code changed in 1 file:

• CCA/Components/Schedulers: SchedulerCommon.cc (+32 -24)

qymeng 2014-02-07 16:07 Rev.: 51537

parallelly compute refineset

32 lines of code changed in 1 file:

• CCA/Components/Schedulers: SchedulerCommon.cc (+32 -6)

qymeng 2014-01-30 11:12 Rev.: 51518

fixed a mismatch of task computes and dw put.
task computes a reduction var into null(global) level, however puts it into level 0

1 lines of code changed in 1 file:

• CCA/Components/Arches: Arches.cc (+1 -1)

qymeng 2014-01-21 15:30 Rev.: 51483

fix memory leak of postD2H and postH2D copies.

4 lines of code changed in 1 file:

• CCA/Components/Schedulers: UnifiedScheduler.cc (+4)

qymeng 2014-01-06 13:52 Rev.: 51430

Loop through only neigbours instead of all processors when assign resources/nodes to ONCE_PER_PROC task.
This change speeded up the assign resource method by 6900X on Mira with 256K cores.

6 lines of code changed in 1 file:

• CCA/Components/LoadBalancers: LoadBalancerCommon.cc (+6 -2)

qymeng 2014-01-05 23:18 Rev.: 51426

When more than one tasks compute a same reduction variable into same level,
only schedule one reduction task after the very last task that computes this reduction variable.

28 lines of code changed in 3 files:

• CCA/Components/Schedulers: SchedulerCommon.cc (+11 -6), SchedulerCommon.h (+3), TaskGraph.cc (+14 -2)

qymeng 2014-01-05 16:51 Rev.: 51425

Turn off topologicalSort() of task graph compiling.
Move reduction task generation part of topo sort to ScheduerCommon.
This greatly speeded up the taskgraph compling on very large scale.

43 lines of code changed in 2 files:

• CCA/Components/Schedulers: SchedulerCommon.cc (+39), TaskGraph.cc (+4 -6)