Assembly using the “novec” mode does not crash anymore.
Small patch release with the following fixes:
- A proper error message is now given if Exafmm is not available.
- Numba now does not produce an error message in single-precision.
- Fixed the issue that evaluation points had to be handed over as the correct data type to Bempp.
- VTK Export now does not produce errors
- The Maxwell dieelectric notebook was updated for Bempp-cl 0.2
This release contains a number of major improvements:
- In addition to OpenCL all operators are now also implemented in Numba, allowing Bempp to be run on systems without OpenCL stack. However, OpenCL is preferrable due to its better performance.
- Internally, the exact implementation of operators has been fully abstracted out so that it is straight forward to write other compute backends (e.g. Cuda, Sycl, etc.)
- All integral operators and potentials can now be assembled via FMM via interfaces to the Exafmm-t library (https://github.com/exafmm/exafmm-t).
- Decorators for grid function callables have been worked over so that it is easy to disable Just-In-Time compilation. Alternatively, a vectorised of grid function callables is now possible.
- The FEniCS interface now allows to obtain trace matrices for Nedelec function spaces for Maxwell FEM/BEM problems.
- Bempp-cl now has automatic deployment with generation of Docker images and a fully automated testing infrastructure.
- Many smaller bug fixes and performance improvements.
- Export now works with meshio version 4
- P1 spaces are correctly generated for open domains if
- First release of Bempp-cl