v3.0.0

Full Changelog: v2.0.0...v3.0.0

v2.0.0

v2.0.0 | 2022.11.09

• Refactored the full structure to a more object oriented focused design, reducing redundancy and increasing reusability:
  • Tile representation abstraction for both dense and compressed tiles
  • Templated implementation for different base datatypes
  • Revisted high-level API
  • Generic resolution for different tile combinations
• Provide scripts for generating and building the library based on CMake
• Provide a test suite based on the Catch2 framework for unit testing
• Provide high-level examples showcasing generic GEMM, and a matrix-matrix multiplication driver based on the tile low-rank GEMM that is useful for benchmarking, along with a benchmarking script
• Support single and double precision arithmetics
• Support the following backends:
  • MKL/OpenBlas for x86 CPU platforms
  • CUDA for NVIDIA GPUs

C++

v1.0.0 | 2021.05.01

• Modern C++ design
  • Templates for precision to minimize codebase
    • Instantiation for float, double, complex, and complex-double
  • Exceptions to avoid silently ignoring errors
  • Use standard containers. e.g., std::vector
• Use BLAS++ and LAPACK++ as abstraction layers to provide C++ APIs for the
  Basic Linear Algebra Subroutines (BLAS) and LAPACK (Linear Algebra PACKage).
• Support flexible and non-uniform tile sizes
• Support all combinations of GEMM
  • ddd, ddc, dcd, dcc, cdd, cdc, ccd, and ccc; where c is compressed tile
    and d is dense tile
• Provide a complete build script generator based on CMake
• Provide a comprehensive testing suite based on TestSweeper framework