TAMM: Tensor Algebra for Many-body Methods

What is it about?

Tensor contraction operations represent a significant fraction of the computing time in computational chemistry calculations. Cutting this computational cost is difficult—computer scientists would need to optimize the high-performance computing constructs the operations use, while scientific application developers focus on the algorithmic requirements for the calculations. Our paper outlines a new framework, Tensor Algebra for Many-body Methods (TAMM), that features a modular structure to support the sustainable development of new electronic structure methods. This structure allows the framework to incorporate new algorithmic advances and optimizations to support different hardware architectures in an independent fashion. Thus, TAMM can take advantage of the various accelerators, types of memory configurations, and topologies of current and emerging computing systems without affecting the tensor algebra formulations. Additionally, TAMM provides a tensor algebra interface that allows users input tensor operations in a mathematical Einstein notation. TAMM’s performance and productivity gains compared to other tensor algebra frameworks are also presented in the paper.

Featured Image

Photo by Joshua Sortino on Unsplash

Photo by Joshua Sortino on Unsplash

Why is it important?

Emerging exascale computing architectures present an exciting opportunity for computational chemistry, but coding challenges hinder this advancement. New formulations must be able to accurately capture the chemical environment while also being scalable and portable. TAMM provides a sustainable and scalable framework for tensor algebra calculations through its modular design. This interface allows scientific application developers and high-performance computing developers to separately address method development and optimization concerns.

Perspectives