Enhancing the Performance of Multigrid Smoothers in Simultaneous Multithreading Architectures
Carlos García (Universidad Complutense de Madrid)Manuel Prieto (Universidad Complutense de Madrid)
Javier Setoain (Universidad Complutense de Madrid)
Francisco Tirado (Universidad Complutense de Madrid)
Abstract:
We have addressed in this paper the implementation of red-black multigrid smoothers on high-end microprocessors. Most of the previous work about this topic has been focused on cache memory issues due to its tremendous impact on performance.
In this paper, we have extended these studies taking Simultaneous Multithreading (SMT)into account.
With the introduction of SMT, new possibilities arise, which makes highly advisable a revision of the different alternatives.
A new strategy is proposed that focused on inter-thread sharing to tolerate the increasing penalties caused by memory accesses. Performance results on an IBM's Power5 based system reveal that our alternative scheme can compete with and even improve sophisticated schemes based on tailored loop fusion and tiling transformations aimed at improving temporal locality.
Keywords:
Numerical Methods (PDE), Tuning, Multigrid smoothers,
