Parent Directory
| 
Revision Log
| Links to HEAD: | (view) (annotate) |
| Sticky Revision: |
ripley now supports paso solvers again and returns an appropriate matrix type id. Changed the getSystemMatrixTypeId() method to take a full SolverBuddy instance and made some other simplifications.
Fast-forward to latest trunk to be able to use both Paso or cusp.
Fast forward to current trunk rev 4947. All tests that don't require a direct solver pass without MPI.
first proof of concept - self-contained ripley system matrix with diagonal storage, CG solver, no preconditioner, no MPI, no blocks, quite a few quick'n'dirty hacks. Solves poisson faster than paso :-)
A branch to have fun with diagonal storage for ripley.
This commit is brought to you by the number 4934 and the tool "meld". Merge of partially complete split world code from branch.
"Some" SystemMatrix clean up.....
SystemMatrixPattern shptr
Coupler/Connector shared ptrs.
paso::Coupler and paso::Connector.
paso::SystemMatrixPattern
adding support for reading binary grids from gzip compressed binary files, ripley brick only so far
adding skeleton of fast Lame assemblers adjusted automatic ripley domain subdivision ( should solve issue #94 ) more tab->space conversions
Remove randomFill python method from ripley domains. All random data objects (for all domain types) should be generated using esys.escript.RandomData() The only filtered random we have is gaussian on ripley but it is triggered by passing the tuple as the last arg of RandomData(). While the interface is a bit more complicated (in that you always need to pass in shape and functionspace) it does mean we have a common interface for all domains. Removed randomFill from DataExpanded. The reasoning behind this is to force domains to call the util function themselves and enforce whatever consistancy requirements they have. Added version of blocktools to deal with 2D case in Ripley. Use blocktools for the 2D transfers [This was cleaner than modifying the previous implementation to deal with variable shaped points]. Note that under MPI, ripley can not generate random data (even unfiltered) if any of its per rank dimensions is <4 elements on any side. Unit tests for these calls are in but some extra checks still needed.
I changed some files. Updated copyright notices, added GeoComp.
Fixed a spelling error and missing virtual (not sure that one matters). Added _untested_ 3D gaussian smoothed random data
Fixes necessary to get a clang/macports build working on 10.9
added dirac point interpolation to ripley (and fixed dirac points in Brick) added fast wave assembler for Brick and minor correction to fast wave assembler for Rectangle
Eliminated all const_cast<Data*> hacks in ripley and finley now that Data.getSampleDataRO returns a const pointer.
Added dirac support to ripley, added interface for custom assemblers for ripleydomains (also added the custom assembler for 2D VTI waves), changed synthetic_VTI example to use the new, faster custom assembler
Implemented reverse reading of grid data from netcdf files. Fixes #44.
Step 1 for #31: wrap parameters in an object.
Added support for more input data types in ER Mapper files. Tests will follow soon.
templated writeBinaryGrid and a bit of cleanup.
added ripley method that returns grid parameters as python tuple.
Preparations for Voxet (GOCAD) binary grid output + code clean up. Changed node id's order to remove code duplication in weipa. More work to do...
Implemented claimed support for mixing data with different resolutions.
Round 1 of copyright fixes
Implemented netCDF grid reader analogous to binary grid reader in ripley.
First pass of updating copyright notices
Implemented binary grid reader in ripley which is now used in ER Mapper source. Interface/details likely to change but need to get this running now.
Added heuristics for number of domain subdivisions to ripley so ripley domains can now be created without specifying the values manually (but a warning will be printed).
Added declspecs for all ripley classes.
Merged ripley rectangular domain into trunk.
Since setX() is not supported the constructors now take location and extent. Python wrapper allows construction the old way with l0,l1 or with tuples (x0,x1),(y0,y1).
3D tests pass, yeah :-)
Preparations for PDE boundary element assembly. Removed some code that slipped in before.
Checkpoint. Getting weipa to like ripley again.
Fixed integration now that elements overlap + code cleanup.
PDE assemblage with reduced order done.
Can solve full order systems of PDEs in 2D and 3D now.
-implemented addPDEToRHS() and setToSize() -added a few missing calls to requireWrite() -added assemblePDESystem() to Rectangle but haven't tested yet
Implemented ownSample() for DOF and Elements.
Fixed interpolation from DOF to nodes and moved common code to the base class.
Made element IDs globally unique so ownSample can work (to be implemented). Fixed couple block in 3D so PDEs seem to work now with MPI :-)
checkpoint. 3D single PDEs almost working
Took over Node->DOF interpolation to Brick and parallelised some loops.
Checkpoint - reinstated DOF FS and tried to fix couple blocks...
In Ripley Solution==ContinuousFunction and ReducedSolution==ReducedCF now. Removed a test in escript that relied on failure when trying to tag Data on Solution/ReducedSolution.
Implemented (Face)Elements->Reduced(Face)Elements interpolation and started separating DOF and nodes. Also, implemented operator==() so that a==b for different domain objects a and b which are in the same state.
All "util on ripley" tests pass now :-) -added support for node/element/face tags -implemented setToNormal() -added a bunch of omp nowait
Integrals in 2D and 3D for rectangle & brick elements & face elements in regular and reduced orders.
Interpolation & Gradient for reduced elements & faces.
Interpolation on faces.
Gradient and interpolation on elements for Brick domain.
Gradient for Rectangle elements.
[x] Rectangle... [x] Brick node id's and weipa compatibility
Next iteration
This form allows you to request diffs between any two revisions of this file. For each of the two "sides" of the diff, enter a numeric revision.
| ViewVC Help | |
| Powered by ViewVC 1.1.26 |