aboutsummaryrefslogtreecommitdiff
Commit message (Collapse)AuthorAge
* CarpetAdaptiveRegrid: Improve CCTK_Warn replacementErik Schnetter2012-09-11
|
* CarpetLib: add preliminary support for higher order restrictionRoland Haas2012-09-11
|
* CarpetInterp: "m" -> "mglevel" when accessing ↵Christian Reisswig2012-09-11
| | | | groupdata.AT(gi).activetimelevels.AT(mglevel).AT(rl) to find out how many active timelevels are available. When using "m", I get a std::vector out of range.
* CarpetInterp: Don't use CCTK_ActiveTimeLevels to get number of active ↵Christian Reisswig2012-09-11
| | | | | | timelevels. Use groupdata.AT(gi).activetimelevels.AT(m).AT(rl) instead. This rempoves a problem in CarpetInterp's error check.
* CarpetReduce: increase warning level about possible deadlock toRoland Haas2012-09-11
| | | | | WARN_DEBUG from WARN_PICKY since there might still be thorns around that do so
* CarpetEvolutionMask: add relative error tolerance of 1e-12Roland Haas2012-09-11
|
* CarpetIOScalar: add test cases for all combinations ofRoland Haas2012-09-11
| | | | one_file_per_group and all_reductions_in_one_file
* CarpetIOScalar: correct column labelsRoland Haas2012-09-11
| | | | when both one_file_per_group and all_reductions_in_one_file are active
* CarpetReduce: output calling routine when warning about reductionsRoland Haas2012-09-11
| | | | in local mode
* CarpetLib: Fill overlap zones via prolongation during regrid.Christian Reisswig2012-09-11
|
* CarpetLib: disable debug outputRoland Haas2012-09-11
|
* Remove all #warning statementsErik Schnetter2012-09-11
|
* CarpetLib: Correct determining overlap zonesErik Schnetter2012-09-11
|
* CarpetLib: Correct determining buffer and overlap zonesErik Schnetter2012-09-11
|
* LoopControl: Explain why downcast is safeErik Schnetter2012-09-11
|
* CarpetLib: Disable OpenMP collapse statementsErik Schnetter2012-09-11
| | | | These do not work on all systems, and we don't autoconf them yet.
* CarpetEvolutionMask: use ML_BSSN instead of BSSN_MoL in testsuitesRoland Haas2012-09-11
|
* CarpetEvolutionMask: use dd.buffer_widths instead of buffer_width parameterRoland Haas2012-09-11
|
* CarpetLib: Remove superfluous memoryof() definitionsErik Schnetter2012-09-11
| | | | | Remove definitions for memoryof(T*) sinfc there is already memoryof(T const*). xlC does not like having both.
* Carpet: Run overlap zone tests on 2 processesErik Schnetter2012-09-11
|
* Carpet: Add test cases for overlap zonesErik Schnetter2012-09-11
|
* Carpet…: Rename buffer2 to overlap zones; correct implementation in dhErik Schnetter2012-09-11
| | | | | | Rename buffer2 zones to overlap zones. Complete implementing overlap zones in the dh class.
* CarpetLib: Remove warnings for non-aligned grid variablesErik Schnetter2012-09-11
|
* Carpet: Correct setting core affinityErik Schnetter2012-09-11
| | | | Correct upper loop bound
* CarpetRegrid2: Take buffer2 zones into accountErik Schnetter2012-09-11
|
* Carpet: Correct condition for using buffer2 zonesErik Schnetter2012-09-11
|
* CarpetLib: Use vectors instead of lists to store bboxset bboxesErik Schnetter2012-09-11
|
* Carpet, CarpetLib: Complete patch to support buffer2 zonesErik Schnetter2012-09-11
|
* Carpet: Correct error in thread id handlingErik Schnetter2012-09-11
|
* CarpetLib: Check whether grid functions are alignedErik Schnetter2012-09-11
| | | | Check whether grid functions are aligned to multiples of a cache line.
* Add a timer around the scheduled functions barrierIan Hinder2012-09-11
|
* Add parameter sync_barriers to insert an MPI barrier before every syncIan Hinder2012-09-11
| | | | This causes each process to wait for all the others to reach this point in the execution. This will happen anyway (for most of the processes) during the sync, and adding this barrier means that this time is not included in the sync timer. Also add a new timer to time this barrier.
* CarpetIOF5: Add example for outputting grid arraysErik Schnetter2012-09-11
|
* CarpetIOF5: Add time step information to grid arraysErik Schnetter2012-09-11
|
* Carpet: Remove debug outputErik Schnetter2012-09-11
|
* Carpet: Set CPU affinity correctly for multiple MPI processes per hostErik Schnetter2012-09-11
|
* CarpetIOHDF5: use version 1.6 of the HDF5 API within hdf5_recombinerknarf2012-09-11
| | | | | | | There are still systems with only version 1.6 installed, or broken installs of version 1.8 (Debian system packages at least up to squeeze). These systems benefit from a compiling hdf5_recombiner, while version 1.8 is not really required to be used here.
* Output XML timer tree at end of runIan Hinder2012-09-11
|
* Carpet: get thread number from CarpetLib's dist objectRoland Haas2012-09-11
|
* CarpetLib: store OpenMP thread number in dist objectRoland Haas2012-09-11
|
* CarpetRegrid2: Ensure that the refinement hierarchy is symmetry about the ↵Erik Schnetter2012-09-11
| | | | | | origin, if desired Introduce a new parameter that tests whether the refinement hierarchy is symmetric about the origin.
* CarpetLib: Improve comment of dh data structuresErik Schnetter2012-09-11
|
* Carpet: Time OutputGH operationsErik Schnetter2012-09-11
|
* Carpet: Do not count ghost zones when calculating speedErik Schnetter2012-09-11
| | | | When calculating the speed of a simulation, ignore ghost zones. Use two counters, one including the outer boundaries, the other excluding it.
* CarpetIOASCII: Declare function prototypeErik Schnetter2012-09-11
|
* Make timer tree output precision and threshold configurable with parametersIan Hinder2012-09-11
|
* CarpetLib: output regions when not-contained error is detected inRoland Haas2012-09-11
| | | | restrict_3d_cc_rf2 and restrict_3d_vc_rf2
* Carpet: make OutputGridCoordinates fully multipatch awareRoland Haas2012-09-11
| | | | | | | This is achieved by using domain information in domainspecs the way OutputGrids does. The key functional change was to use Multipatch_ConvertFromPhysicalBoundary (used to fill domainspecs) rather than CoordBase::ConvertFromPhysicalBoundary.
* Backed out changeset 74ddd545dbb1Roland Haas2012-09-11
| | | | | I give up. I cannot get this right today. The previous patch would have allowed a thread to switch sockets which is undesirable.
* Carpet: let thread run on any core [n0,n0+num_threads) instead of binding toRoland Haas2012-09-11
| | | | | | | single core This is the behaviour of eg. Kraken's aprun --cc numa_node. Avoids using omp_get_thread_num().
generated by cgit v1.2.3 (git 2.39.1) at 2024-06-29 16:29:00 +0000