SOFA plugin wrapping the Eigen linear solvers into SOFA components
This linear solver component is based on Eigen::ConjugateGradient (https://eigen.tuxfamily.org/dox/classEigen_1_1ConjugateGradient.html).
Its equivalent in SOFA is CGLinearSolver (without preconditioner), or ShewchukPCGLinearSolver (with preconditioner).
The available templates are:
- EigenSparseMatrixd
- CompressedRowSparseMatrixMat3x3d
- CompressedRowSparseMatrixd
In XML, the declaration of this component looks like:
<EigenConjugateGradient template="EigenSparseMatrixd" iterations="1000" tolerance="1e-9" preconditioner="diagonal"/>A scene example is available.
This linear solver component is based on Eigen::SimplicialLDLT (https://eigen.tuxfamily.org/dox/classEigen_1_1SimplicialLDLT.html).
Its equivalent in SOFA is SparseLDLSolver.
The available templates are:
- EigenSparseMatrixd
- CompressedRowSparseMatrixMat3x3d
- CompressedRowSparseMatrixd
In XML, the declaration of this component looks like:
<EigenSimplicialLDLT template="CompressedRowSparseMatrixMat3x3d"/>Scene examples are available:
EigenConjugateGradient and EigenSimplicialLDLT can have the choice of a template based on an Eigen sparse matrix, or a SOFA type CompressedRowSparseMatrix.
Both templates have different approaches and can have an impact on the performances.
If possible, the use of CompressedRowSparseMatrixMat3x3d is preferable as it brings significant speed up of the matrix assembly.
The following measures are given on a indicative basis.
Both solvers have been templated with CompressedRowSparseMatrixMat3x3d.
The measurements were made without GUI, running 1000 time steps.
| Measure | SparseLDLSolver | EigenSimplicialLDLT |
|---|---|---|
| total (s) | 3.79887 | 2.45837 |
| total (FPS) | 263.236 | 406.774 |
| MBKBuild (average ms) | 0.53 | 0.52 |
| MBKSolve (average ms) | 3.07 | 1.74 |
The scene is available in FEMBAR_EigenSimplicialLDLT.scn.
Both solvers have been templated with CompressedRowSparseMatrixMat3x3d.
The measurements were made without GUI, running 1000 time steps.
| Measure | SparseLDLSolver | EigenSimplicialLDLT |
|---|---|---|
| total (s) | 13.3605 | 10.5815 |
| total (FPS) | 74.8476 | 94.5048 |
| MBKBuild (average ms) | 5.79 | 5.84 |
| MBKSolve (average ms) | 7.02 | 4.16 |
The scene is available in Cylinder_EigenSimplicialLDLT.scn
Both solvers have been templated with CompressedRowSparseMatrixMat3x3d.
The measurements were made without GUI, running 200 time steps.
| Measure | SparseLDLSolver | EigenSimplicialLDLT |
|---|---|---|
| total (s) | 3.9661 | 3.30239 |
| total (FPS) | 50.4274 | 60.5622 |
| MBKBuild (average ms) | 9.57 | 9.07 |
| MBKSolve (average ms) | 7.75 | 4.96 |
The scene is available in DiamondRobot.py3scn.scn
As any SOFA plugin, follow the build procedure on sofa-framework.org/community/doc/plugins/build-a-plugin-from-sources/