| 304006b3 | 14-Aug-2015 |
Vijay Mahadevan <vijay.m@gmail.com> |
Add new API functions to support mesh copies, view and modifications and
creation of sub-DM objects
* Adding new operations (DMView and DMClone) for DMMoab.
* Adding several public API functions to
Add new API functions to support mesh copies, view and modifications and
creation of sub-DM objects
* Adding new operations (DMView and DMClone) for DMMoab.
* Adding several public API functions to support mesh modifications
and creation of sub-DM objects
* Store the local/global min/max ID space in DMMoab object
show more ...
|
| 755f3dfb | 11-May-2015 |
Vijay Mahadevan <vijay.m@gmail.com> |
Enhancements to DMMoab driven through the 2-D and 3-D example use-cases.
* Updates to the FEM examples to verify convergence and use of GMG
preconditioners with arbitrary levels.
++ TODO: Handle
Enhancements to DMMoab driven through the 2-D and 3-D example use-cases.
* Updates to the FEM examples to verify convergence and use of GMG
preconditioners with arbitrary levels.
++ TODO: Handle mixed meshes in the future.
* Updates to the parallel structured grid mesh-generation which was failing
for non-divisible parallel square meshes.
* Fix a bug in boundary element computation through skinner.
- This was tricky to find and the ksp/tutorials/ex35.cxx helped when
the convergence order was not as expected. After verifying the bases,
the issue was found to be the B.C treatment which missed certain
elements that only had a node on the boundary and not any edges/faces,
which meant that Skinner did not acknowledge these as boundary elements.
* Some profiling monitors are now available during mesh generation.
* Added a new API to get the mesh hierarchy level.
* Fix the nnz computation routine for interpolation operators.
The allocation is still suboptimal in parallel.
show more ...
|
| 49d66b22 | 03-Mar-2015 |
Vijay Mahadevan <vijay.m@gmail.com> |
Several key bug fixes to get the boundary data and UMR working correctly
for both the 2-D and 3-D scenarios.
* Do not use the skinner to get boundary data if we are on a refined level.
We should u
Several key bug fixes to get the boundary data and UMR working correctly
for both the 2-D and 3-D scenarios.
* Do not use the skinner to get boundary data if we are on a refined level.
We should use it only for coarsest levels and then just decipher based
on coarse level entity parent-child property inheritance.
* Updates to the ex35 and ex36 examples to make them verifiable through MMS
and modify tests to use the new API changes.
+ ex35 and ex36 can now compute the L2 discretization error wrt exact solution.
+ Update examples and tests to respect the new API change to accept ghost layers when
loading a mesh.
+ Added an exact solution for problem 2 (MMS) and fix forcing function
so that we can verify the solution convergence
* Several bug fixes and minor enhancements for RefineMesh with edges/faces
enabled. Also tested the computation of all the boundary vertices/faces/edges
to impose BC data for the operators.
* Fix a bug in the numbering and decomposition of HEX->TETS during mesh generation.
Now the numberings are consistent and nested refinement works cleanly for arbitrary levels.
Tested with impls/moab/tests/ex3 for dim=[1,2,3] with simplex=[on,off].
* Added code to exchange ghost layers for all hierarchically generated mesh levels
* Fix issues with boundary marker caching and querying
* Fixes for parallel ghosting for refined entities. Added flexibility to specify ghost
rings for both in-memory mesh and loaded from file.
* Bugfix for read options (mem corruption)
* Update the NNZ allocation an creation of interpolation matrices in parallel.
show more ...
|
| d0d83648 | 28-Feb-2015 |
Vijay Mahadevan <vijay.m@gmail.com> |
Update online mesh generation and convert ex36 to a 3-D example.
* Mesh generated through hierarchy is creating artifacts in the
solution (even when solved with ILU, strangely). Need verification.
*
Update online mesh generation and convert ex36 to a 3-D example.
* Mesh generated through hierarchy is creating artifacts in the
solution (even when solved with ILU, strangely). Need verification.
* Also the solution on tets have many clustered data on the boundary.
This was caused due to inconsistent usage of skinner + DoF setup.
* Now, we do not need intermediate entities (edges/faces) when
creating higher dimensional element unless user requests it.
show more ...
|
| ce27a4ee | 02-Mar-2015 |
Vijay Mahadevan <vijay.m@gmail.com> |
The box mesh generation routine was ad-hoc. Complete revamp of the
implementation along with good verification of the generated mesh. The
GenLargeMesh example in MOAB has been useful here but in the
The box mesh generation routine was ad-hoc. Complete revamp of the
implementation along with good verification of the generated mesh. The
GenLargeMesh example in MOAB has been useful here but in the future,
need to expose this as an API in MOAB so that we can just make use of
that. This replication of code is stupid.
* Fix parallel generation and ghost layer creation
* Adding options to control refinement etc from command line.
show more ...
|
| 941e0cff | 28-Feb-2015 |
Vijay Mahadevan <vijay.m@gmail.com> |
Updating the memory allocation for interpolation matrices to reduce number of unnecessary nonzero entries.
Potentially, this operation needs to be discretization-aware. Link to
PetscFe etc in the fu
Updating the memory allocation for interpolation matrices to reduce number of unnecessary nonzero entries.
Potentially, this operation needs to be discretization-aware. Link to
PetscFe etc in the future can be helpful here.
show more ...
|