Presentation
P06 - Code-Generation of Highly Efficient Finite Element Operations Using the MLIR Compiler Infrastructure
Presenter
DescriptionThe immense financial and environmental cost of high performance computing (HPC) infrastructure demands highly efficient and hardware specific software. In modern exascale hardware, the development of efficient kernels requires addressing both hardware heterogeneity and the memory bandwidth bottleneck. A popular approach to overcoming the latter is by maximizing the floating point operations (FLOPS) per byte of memory, namely the arithmetic intensity, of the numerical scheme. In computational fluid dynamics (CFD) higher order numerical schemes such has the spectral/hp element method (SEM) have been found to have a desirably high and tunable arithmetic intensity, as the higher order shape functions allow each element to contribute with more resolved approximations of the dynamics being simulated. In this work we present the initial stages of the development of NektarIR, a compiler for the Nektar++ spectral/hp element framework. Built using the MLIR compiler infrastructure, NektarIR aims to facilitate the generation of highly efficient, hardware specific, implementations of finite element evaluations for Nektar++.
TimeMonday, June 1610:20 - 10:50 CEST
LocationCampussaal - Plenary Room
Session Chair
Event Type
PASC Poster
