| Commit message (Collapse) | Author | Age |
| |
|
| |
|
|
|
| |
This give more fine-grained control, and is consistent with CaKernel.
|
| |
|
| |
|
|
|
| |
This was supposed to have been committed before.
|
|
|
| |
A calculation has a name, stored in its Name entry. This is what is used when generating the code files, and is what users can use to do relative scheduling (before, after, etc). Sometimes (e.g. for CaKernel), the routine scheduled with the Cactus scheduler in schedule.ccl needs to have a different name. This is what is returned by GetCalculationScheduleName. We now always schedule such functions "<schedulename> as <name>" where <name> is the real name of the calculation.
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
| |
Otherwise the data never gets copied back to the CPU from the device
|
| |
|
| |
|
|\
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Conflicts:
Auxiliary/Cactus/KrancNumericalTools/GenericFD/src/MathematicaCompat.h
Examples/kranc.th
Tools/CodeGen/Kranc.m
Tools/CodeGen/KrancTensor.m
Tools/CodeGen/Schedule.m
Tools/CodeGen/Thorn.m
|
| | |
|
| |
| |
| | |
This option forces Kranc to put the calculation in a group and select and schedule boundary conditions directly after the calculation is called.
|
| | |
|
| |
| |
| |
| |
| |
| |
| | |
- CodeGenC.m still contains Fortran variants. These might be moved into CodeGenFortran.m.
- Several other modules essentially need to pull in all three packages due to newly broken abstraction barriers. This can be fixed incrementally.
- Indent and format all three files consistently
|
| |
| |
| |
| |
| |
| | |
Schedule boundary conditions in MoL_PseudoEvolutionBoundaries instead
of at postrestrict and at postrestrictinitial. This ensures these
boundary conditions are applied after regridding as well.
|
|/
|
| |
These can be set to the names of functions or groups to perform relative scheduling.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Before this commit, Kranc required the user to use the Schedule key in
a calculation to give a Cactus schedule specification. Scheduling
analysis functions which compute only on the grid interior (e.g. those
that take derivatives) was problematic because boundary conditions
were not applied (e.g. symmetries and physical), leading to either
poison or undefined values on the grid. Calculations scheduled in
ANALYSIS could not have prolongation in time with mesh refinement.
This commit introduces a new way to schedule functions which is
completely automatic. If the Schedule key is omitted, the calculation
is scheduled in MoL_PseudoEvolution which means it will be called in
all necessary places. Each such calculation will have its own
schedule group, in which boundary conditions are also selected and
applied (if necessary). Boundary conditions are also applied
in POSTRESTRICT and POSTRESTRICTINITIAL.
|
|
|
|
|
|
|
|
|
|
|
| |
Merge was correct after all.
This reverts commit 04dfdcd32eec91978ba050a472aae2a6e7091c05.
Conflicts:
Tools/CodeGen/CalculationFunction.m
Tools/CodeGen/Thorn.m
|
|
|
|
|
| |
This reverts commit 815307e624fdb8c0ee2eefc644c9bbd8244ad7f2, reversing
changes made to 8e006230f24be02831d390eaad9b90b7a7c77d2c.
|
|
|
|
| |
functions into Interface.m.
|
|
Also move NonevolvedTimelevels to KrancGroups.m
|