- C++ template wrappers for BLAS implementations using CPU, CUDA or ROCm frameworks
- support for real and complex, single and double-precision
- included support for matrix multiply and triangular solve methods using sparse matrices
- added implementation for several iterative Krylov solvers (CPU and GPU versions)
- support for some LAPACK methods (CPU only)
- support for extended registers (sse3, avx, avx512) using templates wrapping CPU intrinsics