Using State-of-the-Art Sparse Matrix Optimizations for Accelerating the Performance of Multiphysics Simulations

摘要

Multiphysics simulations are at the core of modern Computer Aided Engineering (CAE) allowing the analysis of multiple, simultaneously acting physical phenomena. These simulations often rely on Finite Element Methods (FEM) and the solution of large linear systems which, in turn, end up in multiple calls of the costly Sparse Matrix-Vector Multiplication (SpM×V) kernel. The major-and mostly inherent-performance problem of the this kernel is its very low flop:byte ratio, meaning that the algorithm must retrieve a significant amount of data from the memory hierarchy in order to perform a useful operation. In modern hardware, where the processor speed has far overwhelmed that of the memory subsystem, this characteristic becomes an overkill.