A Tensor Algebra Compiler library interface and runtime

Abstract

Tensor algebra is a powerful tool for computing on multidimensional data and has applications in many fields. The number of possible tensor operations is infinite, so it is impossible to manually implement all of them to operate on different tensor dimensions. The Tensor Algebra Compiler (taco) introduced a compiler approach to automatically generate kernels for any compound tensor algebra operation on any input tensor formats. In this thesis, we present a new API for the taco library. The API removes the need to call compiler methods with the introduction of a delayed execution framework. Additionally, the API introduces multiple important tensor algebra features previously unavailable in taco. Finally, we propose extensions to taco's code generation algorithm to automatically generate tensor API methods for any tensor format. The proposed API makes taco code cleaner, shorter and more elegant. Furthermore, the extensions to its code generation algorithm make the API scalable to new formats and operations.