TBLIS is a library and framework for performing tensor operations,
especially tensor contraction, using native algorithms. In this
context, native means that TBLIS does not translate tensor operations
into matrix operations using the BLAS libraries. Instead, TBLIS uses
the philosophy (and certain low-level kernels) of the BLIS framework
to implement tensor contraction with essentially zero overhead
compared to matrix multiplication. TBLIS also aims to offer a
convenient C++ interface for tensors and tensor operations, as well
as a stable C interface with high portability.


  Optional dependence hwloc, autodetect.