• improved GPU support for all platforms
• bumped CMake requirement to 3.19
  • the AMD HIP/ROCm backend requires CMake 3.21
• improved CLI and MATLAB interfaces
• numerous performance improvements and some bug-fixes

• new module: Tasmanian Optimization
  • implements Gradient Descent and Particle Swarm methods
• added differentiation method for all sparse grid types
• improved libEnsemble-Tasmanian integration
  • now includes the dynamic adaptive construction
• miscellaneous bug fixes and performance improvements

• added new Addon methods for constructing quadrature for an arbitrary weight function (not necessarily positive)
• stabilized the HIP/ROCm and oneAPI/DPC++ backends
• complete the Fortran 2003 interface
• numerous API tweaks, bug fixes, and performance improvements

• added DPC++ backend using oneMKL
• stabilized the HIP backend
• added the MPI addon methods to Fortran 2003
• some performance enhancements and bug fixes

• added HIP/ROCm backend to the linear algebra methods
  • including HIP kernels, ROCm linear algebra, and MAGMA-HIP
• CMake now builds only shared or static libraries
• numerous performance enhancements and bug fixes

• expanded the usage of advanced accelerated linear algebra from the LAPACK, cuSovlver and MAGMA libraries
• added a method for constructing a grid from a set of unstructured data
• the CUDA and MAGAM acceleration modes can be enabled through the Python PIP installer
• added swig generated Fortran 2003 interface using Fortran classes
• numerous performance enhancements and bug fixes

• added a python pip-installer for Tasmanain 7.0
• fixed several build-system bugs, most notably concerning GNU Make and MPI
• Note: there is no point in updating an existing working installation of version 7.0

• included Addon module with templates for automated parallel surrogate construction (MPI and Threads)
• updated the C++ API to fully utilize C++2011 features, containers, lambdas, etc.
• reworked the DREAM API, now includes both C++ and Python modules
• all acceleration modes now work with all grid types
• all grid types include refinement procedures (previously Fourier refinement was missing)
• numerous performance enhancements and bug-fixes

• Tasmanian can use UTK MAGMA as back-end for GPU Accelerated Linear Algebra
• New type of grid: Fourier grids use trigonometric basis and preserve periodicity
• Added CUDA kernels for Sequence and Fourier grids
• In addition to arrays, the external API accepts std::vector<type> with basic error checking
• Numerous bug-fixes and performance enhancements, see the changelog for details

Tasmanian version 5.1

• added custom CUDA kernels for evaluations for LocalPolynomial grids
• CUDA accelerated evaluations can be used from MATLAB
• added binary file format for read/write
• added Fortran 90/95 interface (very experimental)
• numerous bug fixes and performance enhancements