Rank-Polymorphism for Shape-Guided Blocking

Explore a 36-minute video presentation from the FHPNC 2023 conference that delves into rank-polymorphism for shape-guided blocking in array programming. Learn how researchers from Heriot-Watt University and Radboud University Nijmegen demonstrate the power of rank-polymorphic array languages in simplifying the specification of generically blocked algorithms. Discover how this approach enables the creation of blocked versions of numerical algorithms on matrices or tensors, improving performance through better cache locality. Examine the proposed dependently-typed framework for rank-polymorphic arrays and its application to matrix multiplication. Gain insights into the benefits of this method, including guaranteed lack of out-of-bound indexing and proof of result consistency across blocked versions. Observe the practical implementation in the SaC array language, showcasing both concise code and impressive performance improvements compared to established libraries like OpenBLAS and Intel's MKL on multi-core systems.