Tasmanian

--------------

* added support for Apple silicon M chips
* python required some work-arounds and bug-fixes, C++ was good before

* added more multicore cpu support
* parallelized curved tensor selection and large beta, e.g., ipcurved
* parallelized setting surplus refinement
* compatibility with gcc parallel STL algorithms

* implemented a new algorithm for global sparse Kronecker
* significant speedup when loading needed values

* dropped support for defunct python 2
* wasn't tested before and it was partially or fully broken already

--------------

* added option to automatically download/configure/build MAGMA
* enabled with -DTasmanian_MAGMA_DOWNLOAD=ON
* improved error checking for the CLI interface
* affects mostly the internal structure
* improved MATLAB interface
* better internal filenames for the grid files
* possible to interface with an external grid file
* added bindings for the asynchronous (dynamic) construction
* updated GPU support
* OneAPI support for 2023
* CUDA 12 support
* required CMake 3.19 and 3.21 when using HIP
* switched to native search for CUDA mathlibs
* switched to HIP as a CMake language
* switched to DPC++ using clang++ with automatically added -fsycl flag

--------------

* new module: Tasmanian Optimization
* implements Gradient Descent and Particle Swarm methods
* included examples and documentation
* credit goes to William Kong
* added differentiation method for all sparse grid types
* improved libEnsemble-Tasmanian integration
* now includes the dynamic adaptive construction
* migrated the tasmanian.ornl.gov page content to github
* miscellaneous bug fixes and performance improvements

--------------

* added new method for constructing quadrature rules with negative weight functions
* the weight function needs to be bounded from below
* the methods are added to the Addons module under the Exotic Quadrature name
* credit goes to William Kong
* stabilized the DPC++ backend and improved the oneMKL compatibility
* cleaned the build system and improved the way the dependencies are handled
* stabilized the Fortrain 2003 interface to cover all Sparse Grid functionality
* GPU and MPI methods are now fully supported through Fortran 2003
* added Fortran 2003 methods for complex numbers for Fourier grids
* the old Fortran 90/95 interface is now deprecated
* improved the grid compression methods
* now can remove points by keeping a specific number of points/basis-functions
* moved the post-install testing to CMake
* simplifies the spack testing for the E4S compatibility
* improved the libEnsemble-Tasmanian integration, see the documentation
* improved the logic for accepting external handles and queues for all GPUs
* deprecated the simple install script, use CMake directly

--------------

* added DPC++ backend using oneMKL
* added both kernels and calls to BLAS/LAPACK methods
* the kernels for the global grids is missing atomics (currently very slow)
* updated Rocm interface
* using the Rocm OpenMP implementation in libiomp5.so and libomp.so
* added github CI covering a MacOSX build and basic Ubuntu build
* credit goes to Viktor Reshniak
* extended the Fortrain 2003 methods
* added functions, e.g., returnPoints() as members to TasmanianSparseGrid
* added MPI methods to the Fortain interface
* updates to the GNU Make build system
* removed Fortran from the GNU Make build system, use CMake
* the added complexity to the SWIG build system is too much to maintain
* GNU Make now defaults to Python 3, removed the make python3 target

--------------

* added AMD HIP/ROCm backend
* all CUDA capabilities have an AMD HIP/ROCm equivalent
* this includes MAGMA-HIP capabilities
* capabilities are accessed with accel_gpu_cuda and accel_gpu_cublas
* removed the XSDK names for options
* the XSDK requirement was alleviated due to spack
* CMake now builds only one type of library shared or static
* matlab interface works well with shared libs thanks to rpaths
* magma search is simpler looking for either shared or static libs
* ::shared and ::static targets are deprecated