--------------------
Change log
--------------------
FEMT-beta36
October 7, 2015
- New functions Solid::GetElementSize and Solid::GetTotalSize, return the area or volume of all the geometry.
FEMT-beta35
September 22, 2015
- CheckElements now returns unconnected groups of elements.
FEMT-beta34
September 18, 2015
- FEMSolver and FEMSolver.Schur use CheckNodes and CheckElements
- New functions Mesh::CheckNodes and Mesh::Check elements to find unconnected items.
- Solid problem type can handle plain-stress and plain-strain problems.
- New function Mesh::GetAdjacencies(Matrix& adjacency) to obtain adjacencies by facet.
- New macro FacetNode to obtain the (local) nodes of a facet.
- Change in Formula class, now all variables defined are stored in the class. Also now it uses the operator () to evaluate.
- New funcion MPI::Test.
FEMT-beta33
November 29, 2012, 01:17:37 a.m.
- New functions TransposeStructure() in CSRMatrix.h to obtain the transpose of a sparse matrix but only its structure.
- Some improvements in SparseApproximateInverse memory usage.
- SparseApproximateInverse renamed to SparseApproximateInverseSymmetric.
- New function SparseApproximateInverseUnsymmetric to calculate preconditioners for non symmetric matrices.
- BiconjugateGradientSparseApproximateInverse now uses SparseApproximateInverseUnsymmetric to generate the preconditioner. This one is a lot faster and uses less memory.
- new operator now accumulate Memory::current_usage using an atomic operation. This solve invalid memory leak messages.
- Update to Makefiles.
FEMT-beta32
October 28, 2012, 02:29:43 a.m.
- Improved adjacency calculation for dual graphs. Removed ElementAdjacency function from Partition.cpp. New function Mesh::GetAdjacencies.
- Makefiles for Visual Studio's nmake. This allows building FEMT projects regardless Visual Studio version.
- MatSover did not saved result file.
- New function Mesh::ElementColoring to separate a elements in a mesh in groups without contact. Useful when parallelizing assembling of stiffness matrix.
- New GiD problemtype: Coloring. To test coloring algorithms.
- New routine for filling incomplete LDL' preconditioner using Munksgaard's algorithm to keep incomplete LDL' SPD. Thanks to Dr. Maria Luisa Sandoval Solis for her advice on this solver.
- New solver EqnSolver.Schur. It is similar to FEMSolver.Schur but for systems of equations.
- Renamed FEMSolver.MPI to FEMSolver.Schur.
- Renamed Heat.MPI to Heat.Schur.
- Renamed Solid.MPI to Solid.Schur.
- Changed Partitioning.bas to support 3D surface meshes.
FEMT-beta31
June 20, 2012, 05:38:14 p.m.
- Added support for building with Open-MPI for Windows (OMPI_IMPORTS define and use of libmpi.lib).
- Error in GiD problemtypes with function input controls, GiD inserts #WIDTH# command into preprocess files (regression to beta29).
- Cleanup of unused functions in SchurComplement.h.
- Regression to METIS 4.0.1. Version 4.0.3 has memory allocation problems for large problems.
- In Solver.MPI, Heat.MPI and FEMSolver.MPI, in generation of systems of equations, chaged solver_parameters.threads to solver_parameters.substructuring_threads.
FEMT-beta30
May 29, 2012, 04:31:57 a.m.
- In Solid and Heat problem types, function input controls have a larger width.
- FEMSolver, FEMSolver.MPI and EqnSolver pipe names no longer need prefix ('\\pipe\.\' or '/tmp/').
- Bug in FEMSolver, FEMSolver.MPI and EqnSolver when only three parameters are given, it was reported as a invalid command line.
FEMT-beta29
May 27, 2012, 01:19:09 a.m.
- Geometry template files now use nelems() function to count elements.
- New class String to handle simple strings.
- New fuction File::CreateTemp creates a temporal file, returns file name.
- File::NamedPipe now returns the pipe name.
- FEMSolver.MPI running (needs testing with larger problems).
- RunFEMSolver.MPI.cmd and RunFEMSolver.MPI.sh are updated to handle Shur parameters.
- Solid.MPI and Heat.MPI do not print solver parameters (these are not used).
- "Read me.txt" replaced with "ReadMe.html"
FEMT-beta28
May 10, 2012, 07:28:49 a.m.
- Some directory movements: examples -> gid/examples, problemtypes -> gid/problemtypes, source/problems, source/GiDProblemTypes.
- New macros Throw and ReThrow to encapsulate DebugPosition and throw.
- Assert.cpp removed, its functionality is now in Debug.cpp.
- Iterative solvers stop iterations if a NaN appears.
- Corrected syntax error in ElectricPotential (strengt -> strength).
- New improved List and Set containers. Memory allocation is done by blocks, fewer calls to new and delete.
- VERSION macro is now defined in BuildFlags.mk (GCC or Intel) or as a shared project property (Visual C++).
FEMT-beta27
April 18, 2012, 04:31:28 p.m.
- Ooops! in LUSolver::LUSolver, missing section of code, it was accidentally removed in beta-23.
FEMT-beta26
April 13, 2012, 09:07:37 p.m.
- Added to ShapeFuncions.h (needed when compiling with gcc 4.7.0)
- New examples: EqnSolverExample.py, FEMSolverExample.py.
FEMT-beta25
- Changed FFLAGS for Fortran examples (-mcmodel=large fails to build in Mac OS).
- Incorrect sign when calculing fluxes in Heat problem.
FEMT-beta24
- License changed to GNU LESSER GENERAL PUBLIC LICENSE Version 3.
- Now EqnSolver and FEMSolver can read solver parameters from a file.
FEMT-beta23
March 21, 2012, 12:29:11 p.m.
- Typo error in ElectricPotential problem file template.
- Rewrite new and delete operators in Memory.cpp, malloc_size (Mac OS X), _msize (Windows) and _malloc_usable_size (glibc) are used to retrive the amount of memory allocated.
- Change EqnSolver to reorder b vector for direct methods before calling solver. It is needed when solver is called multiple times.
- New method for distribute and multiply sparse matrices in SchurComplementMaster and SchurComplementSlave classes.
- New examples: EqnSolverExample.for, EqnSolverExample.cpp, FEMSolverExample.for, FEMSolverExample.cpp.
FEMT-beta22
February 26, 2012, 04:50:11 a.m.
- In Partitioning problem type, elimination of partition_method_name.
- Flags to avoid subnormal-float-point problems in build/gcc/Buildflags.mk: -ffast-math -msse2 -mfpmath=sse.
- Correction of Makefiles to build correctly Partition.cpp
- Exception class no longer receives a message.
- All exceptions are thrown as static member values.
- New macro DebugPossition to show messages prior to throw exceptions.
FEMT-beta21
February 20, 2012, 05:53:14 a.m.
- SchurComplement.h, Partition.h and Partition.cpp are now in the Substructuring directory.
- Remove DomainDecomposition directory.
- Partitioning now only partitions using StructurePartitioning method.
- StructurePartitioning always reorders systems of equatiosn for Cholesky factorizations.
- METIS updated to 4.0.3.
- Fixed error on thickness in Solid::FillF for normal forces in 2D problems.
- Solid, Heat and ElectricPotential problem types modified to support substructuring, prior domain decomposition methods have been eliminated.
FEMT-beta20
February 18, 2012, 06:32:04 a.m.
- Now the SchurComplementMaster::MultiplyKBB works with non-blocking communication. It improves speed a lot!
- PBMSave function in PBMFile was corrected, saved image was shifted by one column and one row.
- Restructuration of SchurComplementMaster and SchurComplementSlave to apply fixed conditions internally.
- Now Schur complement method works correctly with Neumann conditions.
- CSRMatrix/CSCMatrix modified to allow rows/columns with zero entries.
FEMT-beta19
February 13, 2012, 19:08:07 p.m.
- Complete redesign of SchurComplementMaster SchurComplementSlave classes. A new solver was implemented to solve (KBB - sum_i bar_KBBi)*uB = (fB - sum_i bar_FBi), now uses a distributed version of conjugate gradient with Jacobi preconditioner. This saves huge memory in master process.
FEMT-beta18
February 12, 2012, 10:04:58 a.m.
- Compensation for fixed conditions now is done before calling SchurComplementMaster::SendSystemOfEquations. This reduces memory consuption on slave processes.
- Reorganization of SchurComplementSlave, new methods ReceiveMatrix and ReceiveVector, Initialize now is simplified.
- New DenseConjugateGradient and DenseConjugateGradientJacobi solvers.
- SchurComplementSlave now always uses sparse CholeskySolver.
- Solver parameters now defines the dense solver used to calculate the Schur complement. Available solvers are: DenseCholesky, DenseConjugateGradient and DenseConjugateGradientJacobi.
FEMT-beta17
February 11, 2012, 03:35:29 a.m.
- New containers for triangular matrices, LowerTriangularMatrix and UpperTriangularMatrix.
- Cholesky solver for full matrices.
- New solvers for triangular matrices LowerTriangularSystem and UpperTriangularSystem.
- buid/gcc and buid/icc Makefiles modified to read CPPFLAGS from BuildFlags.mk.
- New class DenseSolver, similar to Solver but for dense matrices.
- New class DenseCholeskySolver, simular to CholeskySolver but for dense matrices.
- New method Assembler::AssembleAe for dense matrices.
- StructurePartitioning method was created to partition a domain in substructures without overlapping.
- Shur complement method is implemented in DomainDecomposition/SchurComplement.h.
- Heat.MPI now uses Shur complement method.
- Solid.MPI now uses Shur complement method.
FEMT-beta16
January 25, 2012, 05:59:55 a.m.
- Bug in CSRMatrix.h file, function Convert(const CSRMatrix& R, CSCMatrix& C). Invalid number of columns for C.
- Small type cast correction in MatFile::Store for full matrices.
- For Visual Studio, new configurations "Debug (no MPI)" and "Release (no MPI)" to be able to build without MPI.
FEMT-beta15
November 18, 2011, 02:23:49 a.m.
- Solid and Heat problems now allow functions up to 65535 characters long.
- The Partitioning problemtype had wrongly defined the executable name.
- Partitioning executable did not run in Windows, missing include .
- ElectricPotential, critical sections removed due to inconsistency operation. Now problem data is duplicated.
FEMT-beta14
November 02, 2011, 01:49:01 a.m.
- EqnSolver tool project is renamed to MatSolver.
- New tool project created with the name EqnSolver, it uses named pipes (like FEMSolver) to read and solve a system of equations.
- Bug in Vector::Sort, failed to sort with types different to integer.
- Bug in File::NamedPipe, if unlink fails because the file does not exist an exception is thrown. Now the result of unlink is not verified.
- FEMSolver.MPI is changed to use named pipes.
October 21, 2011, 05:21:25 a.m.
FEMT-beta13
- Makefiles modified to detect if MPI is pressent, if so, build projects that uses it.
- In ElectricPotential problems electrodes can be made of several segments.
October 19, 2011, 10:05:47 a.m.
FEMT-beta12
- ElectricPotential problems now gives results of sensitivity on nodes (to plot surfaces).
- Sensitivity area factor corrected.
October 14, 2011, 03:55:40 p.m.
FEMT-beta11
- Now all solver related functions has a last parameter threads, to allow fine control of the number of threads assigned to each routine.
- For ElectricPotential problem, sensitivity analysis is calculated in parallel.
- Assembler::AssembleRow remade it did not work for reordered matrices.
- In the tool FEMSolver, the command command_get_row_indexes now returns the correct indexes for a row.
- FEMSolver now works with reordered matrices.
October 12, 2011, 10:20:20 a.m.
FEMT-beta10
- New function ShapeFunctions::ElementDetJ to calculate element Jacobian determinant without calculating shape functions.
- ElectricPotential now calculates capacitances and sensitivity analysis.
October 10, 2011, 02:52:07 p.m.
FEMT-beta9
- FEMSolver now works with named pipes.
- New tool FEMSolverExample to test FEMSolver
- New tool EqnSolver, it reads a system of equations from a MAT-File 4 and solves it.
- New routines for reordering before Choleksy factorization: FindingAnOrdering, Reorder (for matrices) and Reorder (for vectors).
September 23, 2011, 07:54:20 p.m.
FEMT-beta8
- New problem type to calculate Electric Potentials.
- New Float class that stores properties for float and double types.
- List class have new AppendFirst and AppendLast to add a ListItem created alone.
- New function Mesh::GetFacetNodes.
- For all problems, now problem files are named ".problem.dat".
FEMT-beta7
- Ops fixed in CSRMatrix::Resize.
- File class now support creation of named pipes.
- MatFile now saves sparse matrices ordering entries first by column and second by row (this is required by MatLab but not by Octave).
July 07, 2011, 11:51:27 p.m.
FEMT-beta6
- Bug on SchwarzAlternating and ReactionAlternating methods MPI object must be passed by reference.
- In SchwarzAlternatingSalve and ReactionAlternatingSlave solver initialization is before other tasks to improve speed.
- FEMSolver.MPI, in main.mpi.cpp casted offsets using (off_t) in order to work with huge files.
July 03, 2011, 09:01:07 p.m.
FEMT-beta5
- New domain decomposition method: ReactionAlternating.
- New partition method: InterleavedPartitioning.
- Now it is posible to select between SchwarzAlternating and ReactionAlternating methods.
- UpdateFactor is now a property of the domain decomposition tab.
June 12, 2011, 10:12:00 p.m.
FEMT-beta4
- In domain decomposition problems, the code to mark local links as fixed conditions now is inside DomainDecomposition::SchwarzAlternatingSlave.
- Formula::ParseNumber bug with integer followed by a '+' or '-' signs.
- File::ReadLine added.
- Heat::Heat and Solid::Solid now read user functions using File::ReadLine.
- Move Solver/DomainDecomposition.h to DomainDecomposition/DomainDecomposition.h
- Split Schwarz alternating routines from DomainDecomposition to a sub class DomainDecomposition/SchwarzAlternating.h.
June 04, 2011, 05:32:03 a.m.
FEMT-beta3
- New Solid problem example of a gelatin dynamics with self weight and two forces applied.
- For Heat and Solid problems, non fixed conditions are stored first on problem file.
- For Heat and Solid problems, fixed conditions are imposed with an 'or' operation.
May 31, 2011, 05:02:57 p.m.
FEMT-beta2
- New problemtype called Partitioning, it only partitions a geometry given a number of partitions and overlapping.
- Schwarz alternating method now uses a update factor to mix old and new solutions for each sub-domain.
- FEMSolver.MPI now can save individual meshes (nodes coordinates must be in input file).
May 28, 2011, 01:04:07 p.m.
FEMT-beta1
- New class Math/Formula to parse and evaluate formulas entered as text.
- Now problem types (Solid and Heat) support definition of user functions to allow boundary conditions to depend of position or time.
- FormatFloat change to allow selectio of type of float (fixed, exponential or automatic).
May 24, 2011, 01:57:34 p.m.
FEMT-alpha11
- Revert changes to File.h to support off_t.
- DomainDecompoisition::SchwarzAlternatingMaster now reports the number of iterations.
- FEMSolver.MPI tested.
May 23, 2011, 05:31:52 a.m.
FEMT-alpha10
- Heat problems now save flux results on gaussian points.
- New option to calculate/save temperature and flux results.
- Solid problems now save stain and stress results on gaussian points.
- New option to calculate/save displacemenent, stain and stress results.
May 20, 2011, 08:43:07 a.m.
FEMT-alpha9
- A version of FEMSolver that implements domain decomposition with MPI is created.
- FEMSolver and FEMSolver.MPI read solver parameters from a file.
- Fixed weird behaviour with Memory::current_usage and OpenMP atomic operation, changed to a critical section.
- new and delete operators now allocate and free memory inside critical sections as an extra protection.
May 17, 2011, 01:45:10 p.m.
FEMT-alpha8
- Bug in Solver::Solver generating b_adjust.
May 14, 2011, 12:14:03 a.m.
FEMT-alpha7
- Small bug, Cholesky and LU solvers did not show advance.
- New tool FEMSolver, it reads a geometry and elemental matrices then assembles the corresponding system of equations and solves it. Tested with CALSEF.
May 06, 2011, 06:30:43 a.m.
FEMT-alpha6
- FiniteElement class is now splited to several classes to have a more flexible architecture. These classes are: Assembler, Geometry, ShapeFunctions.
- Paritioning and reordering routines where moved to the class Mesh.
- PrintInfo routines where moved to each class constructor.
- Merged DomainDecompositionParameters into SolverParameters.
- Memory class has two new static members peak_usage and current_usage.
- new and delete operators now update Memory::peak_usage and Memory::current_usage these funcions are thread safe.
- Programs now reports peak memory usage.
- Programs now reports memory leaks.
- Big Opps! Solver class destructor is now declared as virtual (tssss).
- Reorganized DomainDecomposition class (old FiniteElement class is not needed).
- Removed Solver::SetFixed, instead set b_i = u_i where i is a fixed condition.
- Modified Solver::CompensateFixed, now have two methods to adjust fixed conditions, by matrix (needed for domain decomposition) or by vector (simpler and faster).
- Solid, Solid.MPI, Heat and Heat.MPI programs where reorganized to use Assembler, Geometry and ShapeFunctions classes.
- Log messages where reorganized. log_level=1 shows information messages, log_level=2 shows iteration steps.
- Now in MPI programs slaves does not show log messages.
Apr 25, 2011, 04:35:29 a.m.
FEMT-alpha5
- Dynamic structural mechanics problem is solved using finite difference method (alpha method) in time.
- New example of a 2D bar moving.
- Solid::PrintInfo now prints dynamic and mass forces parameters.
Apr 19, 2011, 09:40:52 p.m.
FEMT-alpha4
- SolverParameters::PrintInfo now only prints only parameters used.
- Thermal diffusivity formula was wrong.
- Now dynamic heat diffusion problem is solved using finite difference method in time.
- Added Temperature_Volume condition in Heat-1.bas.
Apr 14, 2011, 04:34:07 a.m.
FEMT-alpha3
- Fixed oops in SymbolicCholeskyDecomposition parallelizing L generation.
- Changed File::GoTo from int to long to be able to handle huge files.
- Temperature conditions in Heat problem now have a fixed field (preparation for dynamic problems).
- Modified Solid and Heat in preparation to handle dynamic problems.
- FiniteElement::AssembleX and FiniteElement::AssembleFixed are merged into a single function.
- FiniteElement::FillX and FiniteElement::FillFixed are merged into a single function.
Apr 01, 2011, 06:03:08 p.m.
FEMT-alpha2
- Correction to Makefiles (mpic++ instead of g++ for MPI projects).
- Added an option to save the system of equations in MatLab 4 format (checkbox in problem types).
- Solver::Calculate does not call PreCalculate and PostCalculate, these funcions where replaced by CompensateFixed, CheckSolution and SetFixed.
Mar 12, 2011, 01:44:10 p.m.
FEMT-alpha1
- Project renamed from ElementCrunching to FEMT (Finite Element Method Tools).
- SchwarzAlternatingMaster modified, now partitioning is done outside this function.
- FiniteElement::SaveMesh now is usable.
- Add an option to save mesh.
- Conditions for Heat problems now use Temperature and HeatFlow instead of Dirichlet and Neumann.
- Now when a problem file is saved, it reports material names.
- Save multiple works in domain decomposition.
Feb 27, 2011, 04:21:40 a.m.
ElementCrunching-1.4.0
- Fixed facet integration for normal conditions on Heat and Solid problems.
- Problems projects are divided in single process and multiple processes (MPI).
- New problem Dynamic (in construction).
Nov 27, 2010, 05:39:16 a.m.
ElementCrunching-1.3.9
- New Diff program to compare numerically two post.res files from the same problem.
- New functions PrintInfo in SolverParameters, DomainDecompositionParameters and Finite element classes to display data.
Nov 20, 2010, 01:02:34 p.m.
ElementCrunching-1.3.8
- Oops in SchwarzAlternatingSlave. Calling solver.MemoryUsed() after deleting Solver object.
Nov 20, 2010, 04:44:11 a.m.
ElementCrunching-1.3.7
- Minor fix in METIS/proto.h (it gives a warning in Intel compiler, but if it is fixed METIS fails to work).
Nov 19, 2010, 04:32:31 p.m.
ElementCrunching-1.3.6
- Modifications in CSRMatrix and CSCMatrix to work with Intel compiler (it is not necesary without -O3, broken compiler?)
- Makefiles updated
Nov 19, 2010, 04:27:50 a.m.
ElementCrunching-1.3.5
- Improved SymbolicCholeskyFactorization (like 50% less time).
- Remake of Transpose and Convert functions for sparse matrices (now are faster without using more memory).
- New functions SortByIndexes in CSRMatrix, CSCMatrix and CSVector.
- Functions SortIndexes and SortByIndexes in CSRMatrix and CSCMatrix run in parallel.
Nov 18, 2010, 04:50:14 p.m.
ElementCrunching-1.3.4
- Corrected a bug in FactorizedSparseApproximate inverse.
- Convert functions for sparse matrices modified (now them use more memory but are a lot faster).
- Vector, Matrix, CSVector, CSRMatrix and CSCMatrix now have a MemoryUsed function.
- new and delete operators modified. Now memory usage is calculate unsing MemoryUsed functions on containers.
Nov 16, 2010, 02:00:40 p.m.
ElementCrunching-1.3.3
- Improved Factorized sparse approximate inverse by avoiding memory allocation inside threads.
- Transpose functions for sparse matrices modified (now them use more memory but are a lot faster).
Nov 15, 2010, 02:05:45 a.m.
ElementCrunching-1.3.2
- Routines to save/load files with MatrixMarket format (untested).
- Factorized sparse approximate inverse [Chow01]
- Conjugate gradient using FSAI.
- Problem types modified to support preconditiner_threshold parameter.
- SolverParameters class now reads preconditiner_threshold parameter.
Nov 10, 2010, 00:35:20 a.m.
ElementCrunching-1.3.1
- New class CSVector to handle sparse vectors.
- CSCMatrix tested.
- Functions to store/retrive CSCMatrix in MatLab format files.
- Functions to save CSCMatrix to CSV and PBM files.
- Functions to convert from CSCMatrix to CSRMatrix and vice versa.
- Sparse approximate inverse preconditioner (via MR iteration)
- Incomplete factorization preconditioners work better without reordering [Benz02].
Oct 23, 2010, 04:34:02 a.m.
ElementCrunching-1.3.0
- Heat problem now supports source conditions.
- Initial implementation of the eparse approximate inverse preconditioner added (not completed).
Oct 2, 2010, 03:15:20 a.m.
ElementCrunching-1.2.5
- Improved Cholesky, Cholesky2 and LU factorizations algorithms (like 10% faster).
- Now Chokesy and Cholesky2 could run in parallel.
- Bug fixed on MatFile Retrive routines.
- CSCMatrix class added (untested).
Sep 7, 2010, 01:24:45 a.m.
ElementCrunching-1.2.4
- Added new CodeLite project files.
- Updated Clean batch files for all build systems.
- Corrected a bug in gaussian quadrature of 5 points, added quadratures up to 7 points.
- Updated CSVFile routines to use the Format class.
Sep 3, 2010, 04:08:48 a.m.
ElementCrunching-1.2.3
- Added makefiles to use Intel C++.
- Reorganization of Interger.h.
- Corrected Young module in solid materials (by a scale of 1000).
Sep 3, 2010, 12:00:37 a.m.
ElementCrunching-1.2.2
- In SchwarzAlternatingMaster splited a big critical section into several small pieces to improve domain decomposition in parallel.
- Added Exception messages in FiniteElement member functions.
- Corrected a comment in Solid-1.bas
Sep 2, 2010, 02:08:28 a.m.
ElementCrunching-1.2.1
- Removed the use of global N and dN storage in Solid and Heat. This was not thread safe!
Sep 2, 2010, 12:09:44 a.m.
ElementCrunching-1.2.0
- Corrected use of mass forces for Solid.
- Added gravity property for Solid
- Corrected neumann conditions of Head problem.
- Added a critical section in SchwarzAlternatingMaster when sending system of equations to slaves.
- Simplification of Shape functions code (now there is a single function to calculate N and dN).
Aug 31, 2010, 02:14:07 p.m.
ElementCrunching-1.1.0
- Corrected the use of thickness for 2D Solid.
- Added mass forces to Solid (2D and 3D).
Aug 31, 2010, 02:14:07 p.m.
ElementCrunching-1.0.1
- Reinserted threads inside MPI data i/o. Problem persist in GNU/Linux when using more than 1 thread in a multicore CPU. (May be it is a problem with OpenMPI).
- Updated test script.
Aug 31, 2010, 09:22:16 a.m.
ElementCrunching-1.0.0
- Program arguments modified, now only is needed.
- Checked examples for Heat and Solid in 2 and 3 dimensions.
- Run scripts modified.
- Correct bug on differences statistics.
- Remove threads inside MPI data i/o.
- Remove -lcompat from Solid and Heat Makefiles.
Aug 30, 2010, 11:53:50 a.m.
ElementCrunching-rc3
- Heat code recreated.
- Heat problem_type remade.
- Time functions created.
- Log routines remade, now each log line shows a timestamp.
- Log messages modified.
Aug 29, 2010, 11:55:40 a.m.
ElementCrunching-rc2
- Corrected bug with Neumman conditions.
- Folder reorganization, now build folder is outside source.
- Created Arc2D and Arc3D problems as example for the Solid program.
- Now problem executables are stored in problemtypes/.gid/.
- Update Makefiles for GCC.
Aug 28, 2010, 11:55:40 a.m.
ElementCrunching-rc1
- Reordering routines for direct solvers.
- Partition routine moved to FiniteElement class.
- Reordering routine moved to FiniteElement class.
Aug 24, 2010, 03:07:14 a.m.
ElementCrunching-beta6
- Routines for 3D solids fixed.
- Need to add and check permutation routines for direct solvers.
Aug 22, 2010, 05:07:05 p.m.
ElementCrunching-beta5
- Solid for 3D geometries.
- Integration rules for 2D and 3D problems.
- Checked to work with conjugate gradient (not working with direct solvers).
- Need to add and check permutation routines for direct solvers.
Aug 17, 2010, 01:43:45 p.m.
ElementCrunching-beta4
- Corrected Makefiles.
- Corrected CodeLite project files.
- Addition of degrees_of_freedom to geometry file, it simplifies FiniteElement class.
Aug 15, 2010, 00:04:19 a.m.
ElementCrunching-beta3
- Avoid sending zero-length messages among slaves.
- Check for valid solution.
Aug 14, 2010, 08:26:11 a.m.
ElementCrunching-beta2
- Better separation of data files (now: geometry, problem, solver, domain_decomposition).
- Definition of number of threads for solver and for domain decompositioning.
Aug 13, 2010, 06:05:16 p.m.
ElementCrunching-beta1
- Parition working for elements with mode nodes than vertex.
- Schwartz alternating method working.
- Solid problem in 2D type working with triangular elements.
- MPI routines arranged in a class.
- Memory allocation routines modified to collect information about of current and maximum memory allocated.
- Log routines improved.
- CodeLite project added.
Jul 19, 2010, 04:19:05 a.m.
ElementCrunching-alpha22
- Assembling routines for Solid problem.
- Integration points and weights for triangles and quadrilaterals.
Jul 17, 2010, 01:31:55 a.m.
ElementCrunching-alpha21
- Shape functions for: tetrahedron (4 and 10 nodes), hexahedron (8, 20 and 27 nodes).
Jul 14, 2010, 07:58:44 a.m.
ElementCrunching-alpha20
- Shape functions for: triangles (3 and 6 nodes), quadrilateral (4, 8 and 9 nodes), tetrahedron (4 nodes).
Jul 12, 2010, 05:52:14 a.m.
ElementCrunching-alpha19
- Creation of FiniteElement class, it has routines to load greometry (mesh, nodes), calculate shape functions (currently only triangles of 3 and 6 nodes)
Jul 10, 2010, 05:44:34 a.m.
ElementCrunching-alpha18
- New solver Cholesky2 (LDL')
- File class redesign. New Format class
- New DiagonalSystem routine
- Heat problemtype definition
- Basic MPI routines wrappers
- Log routines
- New OrderedList container
Jun 28, 2010, 02:51:17 a.m.
ElementCrunching-alpha17
- Fix bug in FillLUDecomposition function (size_i not checked)
- Performance improvement in FillLUDecomposition and FillCholeskyDecomposition
Jun 18, 2010, 01:05:25 a.m.
ElementCrunching-alpha16
- Allow OpenMP parallelization in FillLUDecomposition function
- New solver ConjugateGradientIncompleteLUSolver
- Insertion of __restrict directive in all solvers
Jun 17, 2010, 03:38:09 a.m.
ElementCrunching-alpha15
- Corrected bug in CSRMatrix, function CopyRow
- Major improvement of FillLUDecomposition (making use of the known symetric structure to calculate U' instead of U)
Jun 12, 2010, 04:20:09 a.m.
ElementCrunching-alpha14
- Implementation of MinimumDegree reordering algorithm to improve fill-in in Cholesky decomposition
- Insertion of METIS routines for reordering and partitioning.
- Geometry load routines checked.
Jun 8, 2010, 04:12:07 a.m.
ElementCrunching-alpha13
- Corrected bugs in CSRMatrix: Fill routine and in the constructor called with a connectivity matrix.
- Change Fill(T* data) to Copy(T* data) to avoid ambiguity in call.
Jun 6, 2010, 03:09:44 a.m.
ElementCrunching-alpha12
- Solvers for symetric matrices tested (Cholesky, ConjugateGradient, ConjugateGradientJacobi, ConjugateGradientIncompleteCholesky).
- Solvers for non-symetric matrices tested (LU, BiconjugateGradient, BiconjugateGradientJacobi, BiconjugateGradientIncompleteLU).
- SymbolicCholeskyDecomposition now accepts a k parameter to have an imcomplete Cholesky decomposition.
- DataFile, Nodes, Mesh and Geometry clases reincorporated to project (untested).
- Corrected bug in CSRMatrix in Resize routine.
Jun 5, 2010, 05:11:34 a.m.
ElementCrunching-alpha11
- Corrected Set and List classes for exception handling.
- CholeskyFillIn now works with a single data array.
- Routines for writing PBM files.
- Implementation of LU solver for matrices with symetric structure and non-sumetric data.
- Update of ConjugateGradient and ConjugateGradientJacobi solvers.
- Correction of MatFile routines to handle data type better.
Jun 4, 2010, 07:01:45 a.m.
ElementCrunching-alpha10
- CSRMatrix, Matrix and Vector containers restructurated to improve speed.
- Routines for loading/saving Matlab .mat4 file data for CSRMatrix, Matrix and Vector containers.
- Remake of routines for writting CSV files.
- Now new[] doesn't throw an exception, this behaviour is passed to containers. It makes constructor initializations easier.
Jun 1, 2010, 05:22:34 a.m.
ElementCrunching-alpha9
- Makefile updated
- Compilation errors using GCC fixed. Template base class functions must be called using this->foo();
May 31, 2010, 12:04:05 a.m.
ElementCrunching-alpha8
- Solvers BiconjugateGradient BiconjugateGradientJacobi and BiconjugateGradientIncompleteLU tested with non-symetic matrices. Stop criteria changed to norm(r) < tolerance
May 30, 2010, 07:29:25 p.m.
ElementCrunching-alpha7
- New solvers BiconjugateGradientJacobi and BiconjugateGradientIncompleteLU (pending to test them with a non-symetic matrices)
- IncompleteLU factorization (with k = 0)
- FillLUDecomposition function
May 29, 2010, 04:40:39 a.m.
ElementCrunching-alpha6
- New FillIn class used to create a bitmap of a CSRMatrix
- ConjugateGradientIncompleteCholesky solver is working
- New solver BiconjugateGradient (pending to test it with a non-symetic matrix)
- Traspose function for a CSRMatrix
May 28, 2010, 07:34:07 a.m.
ElementCrunching-alpha5
- Creation fo Solver base class
- Inclution of solvers: Cholesky, UpperTriangular, LowerTriangular, ConjugateGradient, ConjugateGradientJacobi, ConjugateGradientIncompleteCholesky (not working)
- Routines for loading and saving vectors, matrices and matrices shape
- Function to test float and double values for +inf, -int and NaN.
May 26, 2010, 04:44:50 a.m.
ElementCrunching-alpha4
- Revision and modification of Matrix, Vector, List, Set, CSRMatrix classes
May 24, 2010, 06:43:28 a.m.
ElementCrunching-alpha3
- Project restructuration by directories
- Template file modified to suppor normal conditions
- Change to use a class for exception handling
- Makefile for GCC
May 22, 2010, 03:26:10 a.m.
ElementCrunching-alpha2
- Multiple ploblems with multiple solvers, materials and boundary conditions
- Routines for reading Nodes, Meshes
May 19, 2010, 04:04:07 a.m.
ElementCrunching-alpha1
- Improvement of problem type: handling different kinds of materials, count number of meshes
- Basic routines for parsing and loading geometry data
- Routines for exeption handling, asserts, debug messages, vectors, matrices
May 12, 2010, 08:01:42 a.m.
ElementCrunching-alpha0
- Creation of generic problemtype for GiD. It accepts multiple meshes with different element types at the same time(linear, triangle, quadrilateral, tetrahedra, and hexahedra)