* Added native HIP backend for targeting AMD GPUs. * Rewritten CUDA backend which no longer depends on PyCUDA. * Updated defaults for CUDA backend to improve out of the box performance.
1.10.0
New Features:
- Added support for writing specific regions of the domain to disk. - Fixed statistics collection for dual time stepping integrators.
1.9.0
New Features:
- Improved strong scaling. - Added support for Gmsh v4.1. - Fixed performance issue with OpenCL backend.
1.8.5
New Features:
- Added support for CLBlast. - Added support for Gmsh v4. - Added the optimal Runge-Kutta schemes of Vermeire et al. for pseudo time-stepping.