Revisions of dolfin
Matthias Mailänder (Mailaender)
accepted
request 281978
from
Kshitij Kulshreshtha (polyconvex)
(revision 7)
- update to version 1.5.0 * DG demos working is parallel * Simplify re-use of LU factorisations * CMake 3 compatibility * Make underlying SLEPc object accessible * Full support for linear algebra backends with 64-bit integers * Add smoothed aggregation AMG elasticity demo * Add support for slepc4py * Some self-assignment fixes in mesh data structures * Deprecated GenericDofMap::geometric_dimension() * Experimental support for ghosted meshes (overlapping region in parallel) * Significant memory reduction in dofmap storage * Re-write dofmap construction with significant performance and scaling improvements in parallel * Switch to local (process-wise) indexing for dof indices * Support local (process-wise) indexing in linear algerbra backends * Added support for PETSc 3.5, require version >= 3.3 * Exposed DofMap::tabulate_local_to_global_dofs, MeshEntity::sharing_processes in Python * Added GenericDofmap::local_dimension("all"|"owned"|"unowned") * Added access to SLEPc or slepc4py EPS object of SLEPcEigenSolver (requires slepc4py version >= 3.5.1) * LinearOperator can now be accessed using petsc4py * Add interface (PETScTAOSolver) for the PETSc nonlinear (bound-constrained) optimisation solver (TAO) * Add GenericMatrix::nnz() function to return number of nonzero entries in matrix (fixes #110) * Add smoothed aggregation algerbraic multigrid demo for elasticity * Add argument 'function' to project, to store the result into a preallocated function * Remove CGAL dependency and mesh generation, now provided by mshr * Python 2.7 required * Add experimental Python 3 support. Need swig version 3.0.3 or later * Move to py.test, speed up unit tests and make tests more robust in parallel * Repeated initialization of PETScMatrix is now an error * MPI interface change: num_processes -> size, process_number -> rank * Add optional argument project(..., function=f), to avoid superfluous allocation * Remove excessive printing of points during extrapolation * Clean up DG demos by dropping restrictions of Constants: c('+') -> c * Fix systemassembler warning when a and L both provide the same subdomain data. * Require mesh instead of cell argument to FacetArea, FacetNormal, CellSize, CellVolume, SpatialCoordinate, Circumradius, MinFacetEdgeLength, MaxFacetEdgeLength * Remove argument reset_sparsity to assemble() * Simplify assemble() and Form() signature: remove arguments mesh, coefficients, function_spaces, common_cell. These are now all found by inspecting the UFL form * Speed up assembly of forms with multiple integrals depending on different functions, e.g. f*dx(1) + g*dx(2). * Handle accessing of GenericVectors using numpy arrays in python layer instead of in hard-to-maintain C++ layer * Add support for mpi groups in jit-compilation * Make access to HDFAttributes more dict like * Add 1st and 2nd order Rush Larsen schemes for the PointIntegralSolver * Add vertex assembler for PointIntegrals * Add support for assembly of custom_integral * Add support for multimesh assembly, function spaces, dofmaps and functions * Fix to Cell-Point collision detection to prevent Points inside the mesh from falling between Cells due to rounding errors * Enable reordering of cells and vertices in parallel via SCOTCH and the Giibs-Poole-Stockmeyer algorithm * Efficiency improvements in dof assignment in parallel, working on HPC up to 24000 cores * Introduction of PlazaRefinement methods based on refinement of the Mesh skeleton, giving better quality refinement in 3D in parallel * Basic support for 'ghost cells' allowing integration over interior facets in parallel
Kshitij Kulshreshtha (polyconvex)
committed
(revision 6)
Try to get this to compile on obs again. Too much memory is required for parallel compilation. It was suggested that serial make might work with less memory and succeed in compilation.
Dirk Stoecker (dstoecker)
accepted
request 215446
from
Kshitij Kulshreshtha (polyconvex)
(revision 5)
- backport vtk-6 support from upstream, otherwise no visualisation: * dolfin-1.3-vtk-6-enable.patch - corrected installation path due to failing compilation of demo files
Matthias Mailänder (Mailaender)
accepted
request 213749
from
Kshitij Kulshreshtha (polyconvex)
(revision 4)
update to latest version. unfortunately openSUSE_Factory build does not succeed due to faulty build of boost library in Factory (missing zlib interface). I'm not sure how to fix that.
Dmitry Roshchin (Dmitry_R)
accepted
request 181319
from
Kshitij Kulshreshtha (polyconvex)
(revision 3)
This updates dolfin to version 1.2.0 and activates some more interesting features. The package cgal from the graphics project is required for building. A _constraints file was added because otherwise the compiler quits in OBS with an out of memory error. - Activate features by using optional modules CGAL for geometry manipulation HDF5 format support in parallel SLEPc support for eigenvalues - Cleanup spec file - update to version 1.2.0 * Fixes bug where child/parent hierarchy in Python were destroyed * Add utility script dolfin-get-demos * MeshFunctions in python now support iterable protocol * Expose ufc::dofmap::tabulate_entity_dofs to GenericDofMap interface * Expose ufc::dofmap::num_entity_dofs to GenericDofMap interface * Allow setting of row dof coordinates in preconditioners (only works with PETSc backed for now) * Expose more PETSc/ML parameters * Improve speed to tabulating coordinates in some DofMap functions * Feature: Add support for passing a Constant as default value to a CompiledExpression * Fix bug in dimension check for 1-D ALE * Remove some redundant graph code * Improvements in speed of parallel dual grapj builder * Fix bug in XMDF output for cell-based Functions * Fixes for latest version of clang compiler * LocalSolver class added to efficiently solve cell-wise problems * New implementation of periodic boundary conditions. Now incorporated into the dofmap * Optional arguments to assemblers removed * SymmetricAssembler removed * Domains for assemblers can now only be attached to forms * SubMesh can now be constructed without a CellFunction argument, if the MeshDomain contains marked celldomains. * MeshDomains are propagated to a SubMesh during construction * Simplify generation of a MeshFunction from MeshDomains: No need to call mesh_function with mesh * Rename dolfin-config.cmake to DOLFINConfig.cmake
Corot Sebastien (scorot)
accepted
request 160680
from
Corot Sebastien (scorot)
(revision 2)
- update to version 1.1.0 * Changelog to long see ChangeLog file for a full list
Corot Sebastien (scorot)
accepted
request 144707
from
Corot Sebastien (scorot)
(revision 1)
An open source CFD software
Displaying revisions 21 - 27 of 27