| #
82dfd14a
|
| 04-Jun-2014 |
Vijay Mahadevan <vijay.m@gmail.com> |
Fixing a nasty memory corruption bug in non-zero allocation and set zero entries. And other miscellaneous cleanups.
|
| #
f90c3b0e
|
| 22-Jan-2014 |
Vijay Mahadevan <vijay.m@gmail.com> |
Updating DMMoab interface to the latest PETSc changes related to PetscNew. Removing occurrences of stringstream and replacing with PetscSNPrintf for manipulating IO options.
|
| #
b8ecf6d3
|
| 22-Nov-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Adding more documentation to all main source files and making sure the documentation is generated according to PETSc standards.
|
| #
f6829af0
|
| 21-Nov-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Documentation updates for all the DMMoab implementation sources and re-arranging routines to place public API first and static/private routines next.
|
| #
351b8a77
|
| 02-Oct-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Making sure that the interface is const correct when dealing with MOAB objects on the getter methods.
|
| #
addae81c
|
| 12-Sep-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Rename nfields->numFields and fields->fieldNames to be consistent with PetscSection. Need to make a move eventually to this structure to handle the internal field and point DOF.
|
| #
6465f021
|
| 11-Sep-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Fixes to make sure the MOAB based tag_iterate functions will work correctly even when local entities are non-contiguous. No more restrictions for either the MOAB or native vector in DMMoab now.
|
| #
6d9eb265
|
| 20-Nov-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
A very important fix for the ghosted vector updates. The parallel bugs in incorrect residual evaluations in ex2.cxx were two-fold.
1) The ghost updates were incorrect. 2) The VecScatter context was
A very important fix for the ghosted vector updates. The parallel bugs in incorrect residual evaluations in ex2.cxx were two-fold.
1) The ghost updates were incorrect. 2) The VecScatter context was wrongly formed. Perhaps a VecCopy might suffice for local Global-Local transfers. Optimize later. 3) The BC evaluation had a bug due to the way we were accessing the bit vector with EntityHandle static casts. This could be a problem in evaluating field DOF data also. Currently not an issue but need to check if this can be optimized. 4) ex2.cxx runs correctly in serial and parallel. Preliminary profiling against ts/examples/tutorials/ex25.cxx shows that the FEM version against FD version is about 2-3 times slower in TSFunctionEval. This needs to be profiled further and corrected. 5) Now support both native PETSc and native MOAB-tag based Vec depending on whether mesh entities or contiguous.
Parallel bugs fixed. Optimizations upcoming to improve speedup.
show more ...
|
| #
bb8f3634
|
| 07-Sep-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Adding the first draft capability to allow both MOAB and native Vec with DMMoab. Also setting the necessary Mapping data so that SetLocal and SetBlockedLocal can be used.
|
| #
a2eadca2
|
| 24-Aug-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Some bug fixes for the ghosted exchange of tag data during Get/Restore Array for DMMoabVec
|
| #
eb9d2429
|
| 20-Nov-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Adding an extra API functionality to use the local DOF to assemble vectors and matrices.
- This should generally be faster than using VecSetValues. Use DMMoabGetLocalFieldDofs for these operations.
Adding an extra API functionality to use the local DOF to assemble vectors and matrices.
- This should generally be faster than using VecSetValues. Use DMMoabGetLocalFieldDofs for these operations. - Should work with MatSetValuesLocal, VecSetValuesLocal also (untested).
show more ...
|
| #
ac5d1045
|
| 13-Aug-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Fixed a possible bug when restoring array.
|
| #
e23c60eb
|
| 20-Nov-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Fix the NNZ computation by removing the old methods and implemented a more rigorous traversal to find nnz and onz. Other minor cleanups in DMMoab - remove prints, debug statements etc.
|
| #
fc418013
|
| 20-Nov-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Bug fixes for deciphering leading dimension/DMCreateLocalVector_Moab and data output feature additions.
9cb894f: Adding some options for writing MOAB output files. 1e30170: Fixed a bug with leading
Bug fixes for deciphering leading dimension/DMCreateLocalVector_Moab and data output feature additions.
9cb894f: Adding some options for writing MOAB output files. 1e30170: Fixed a bug with leading dimension specification. MOAB->set_dimension should not be used. Miscellaneous fixes for DOF of fields: making global DOF index list as part of DM private data. 70830f3: Remove redundant use of gsindices and reuse global DOF list from DM. Fixed a bug in DMCreateLocalVector_Moab.
show more ...
|
| #
9088682f
|
| 20-Nov-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Private interface changes in DMMoab.
d64c012: Adding storage for holding the boundary vertices and faces for easier traversal during FEM assembly calculations. Could move it somewhere else if need b
Private interface changes in DMMoab.
d64c012: Adding storage for holding the boundary vertices and faces for easier traversal during FEM assembly calculations. Could move it somewhere else if need be. d7dd518: Adding more public interface methods to Output the mesh+solution fields, to set a Vec for a field handled by DMMoab and accessors for getting the boundary entities. bdfc3ff: Fix a bug in checking whether a tag is valid.
show more ...
|
| #
efd17f3e
|
| 18-Jun-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Add MOAB specific Vector array get/restore methods for both write and read-only operations to ease the use of MoabVec in FEM based computations. The arrays typically have a ghosted representation if
Add MOAB specific Vector array get/restore methods for both write and read-only operations to ease the use of MoabVec in FEM based computations. The arrays typically have a ghosted representation if the MOAB mesh is ghosted itself, else represents a truly distributed array.
show more ...
|
| #
db66d124
|
| 20-Nov-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Add important ltog and gtol routines apart from documentation and cleanup of code.
|
| #
032b8ab6
|
| 20-Nov-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Several essential feature additions and internal modifications for DMMoab.
1) Petscify the DMMoab implementation and separate out the Mat and Vec handling parts to separate files. 2) Private data st
Several essential feature additions and internal modifications for DMMoab.
1) Petscify the DMMoab implementation and separate out the Mat and Vec handling parts to separate files. 2) Private data structure definitions moved to petsc-private. 3) Other feature additions include clean ways to perform local-global and global-local scatters. 4) Now there is a separate SetUp routine that handles consistency checks and initialization all necessary data for DMMoab. 5) The Matrix creation from DMMoab can handle AIJ and BAIJ type matrix creations based on how the user defined a tag. This assumes that the primary solution is defined as a single tag in MOAB and hence block size is set in DMMoab globally. This should change in order to allow more flexibility to create block matrices for single solution fields or combinations of them. Similar changes for Vec can be provided based on tag structure. Perhaps taking in a MoabVec input can resolve this to maintain consistency.
show more ...
|
| #
20e9df35
|
| 22-Nov-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Adding more documentation to all main source files and making sure the documentation is generated according to PETSc standards.
|
| #
7f25a9c9
|
| 21-Nov-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Documentation updates for all the DMMoab implementation sources and re-arranging routines to place public API first and static/private routines next.
|
| #
21c76061
|
| 02-Oct-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Making sure that the interface is const correct when dealing with MOAB objects on the getter methods.
|
| #
5ff14e8b
|
| 12-Sep-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Rename nfields->numFields and fields->fieldNames to be consistent with PetscSection. Need to make a move eventually to this structure to handle the internal field and point DOF.
|
| #
7756a0bb
|
| 11-Sep-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Fixes to make sure the MOAB based tag_iterate functions will work correctly even when local entities are non-contiguous. No more restrictions for either the MOAB or native vector in DMMoab now.
|
| #
f2497645
|
| 20-Nov-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
A very important fix for the ghosted vector updates. The parallel bugs in incorrect residual evaluations in ex2.cxx were two-fold.
1) The ghost updates were incorrect. 2) The VecScatter context was
A very important fix for the ghosted vector updates. The parallel bugs in incorrect residual evaluations in ex2.cxx were two-fold.
1) The ghost updates were incorrect. 2) The VecScatter context was wrongly formed. Perhaps a VecCopy might suffice for local Global-Local transfers. Optimize later. 3) The BC evaluation had a bug due to the way we were accessing the bit vector with EntityHandle static casts. This could be a problem in evaluating field DOF data also. Currently not an issue but need to check if this can be optimized. 4) ex2.cxx runs correctly in serial and parallel. Preliminary profiling against ts/examples/tutorials/ex25.cxx shows that the FEM version against FD version is about 2-3 times slower in TSFunctionEval. This needs to be profiled further and corrected. 5) Now support both native PETSc and native MOAB-tag based Vec depending on whether mesh entities or contiguous.
Parallel bugs fixed. Optimizations upcoming to improve speedup.
show more ...
|
| #
31c6faee
|
| 07-Sep-2013 |
Vijay Mahadevan <vijay.m@gmail.com> |
Adding the first draft capability to allow both MOAB and native Vec with DMMoab. Also setting the necessary Mapping data so that SetLocal and SetBlockedLocal can be used.
|