Finite element assembly strategies on multi‐core and many‐core architectures

Abstract

SUMMARY We demonstrate that radically differing implementations of finite element methods (FEMs) are needed on multi‐core (CPU) and many‐core (GPU) architectures, if their respective performance potential is to be realised. Our numerical investigations using a finite element advection–diffusion solver show that increased performance on each architecture can only be achieved by committing to specific and diverse algorithmic choices that cut across the high‐level structure of the implementation. Making these commitments to achieve high performance for a single architecture leads to a loss of performance portability. Data structures that include redundant data but enable coalesced memory accesses are faster on many‐core architectures, whereas redundancy‐free data structures that are accessed indirectly are faster on multi‐core architectures. The Addto algorithm for global assembly is optimal on multi‐core architectures, whereas the Local Matrix Approach is optimal on many‐core architectures despite requiring more computation than the Addto algorithm. These results demonstrate the value in making the correct choice of algorithm and data structure when implementing FEMs, spectral element methods and low‐order discontinuous Galerkin methods on modern high‐performance architectures. Copyright © 2012 John Wiley & Sons, Ltd.

A stabilised nodal spectral element method for fully nonlinear water waves

“…In this work we focus exclusively on the first two requirements and we pave the way for the fulfilment of the last one, which is being addressed in ongoing work. The use of modern many-core hardware is attractive for acceleration of the high-order spectral element framework and for enabling practical computations of realistic engineering problems [49,50].…”

Performance Analysis of a High-Level Abstractions-Based Hydrocode on Future Computing Systems

“…Although well established finite element methods could be supported by such a declarative abstraction, it lacks the flexibility offered by frameworks such as OP2 for developing new applications/algorithms. Currently, a runtime code generation, compilation and execution framework that is based on Python, called PyOP2 [27], and a larger framework that supports finite element application development called Firedrake [1,21] is being developed at Imperial College London.…”

Rapid development and adjoining of transient finite element models

“…As the mathematical structure of the PDE is available for analysis, important equation-specific optimisations may be performed that low level compilers cannot automate, or that would be too laborious to implement by hand [7,8]. UFL is remarkably compact: a finite element discretisation that might take thousands or tens of thousands of lines of FORTRAN or C++ code to implement can be cleanly expressed in just a handful of lines of UFL.…”

“…* denotes the Hermitian transpose. If the adjoint model solution λ is known then one can compute the derivative with respect to many parameters, simultaneously, via (8).…”

“…In particular, if the forward model and functional are differentiable, then the total derivative of the functional with respect to model parameters should be computed exactly (except for precision related errors) by equation (8).…”