this article shows an approach to extend the HAT programming model with an API for explicit tensor-core programming. Furthermore, it shows how to make this approach generic to be able to process computations expressed with the proposed HAT tensor core API on accelerators without explicit tensor instructions.
