Stencil computation optimization and auto-tuning on state-of-the-art multicore architectures

Abstract

Understanding the most efficient design and utilization of emerging multicore systems is one of the most challenging questions faced by the mainstream and scientific computing industries in several decades. Our work explores multicore stencil (nearest-neighbor) computations-a class of algorithms at the heart of many structured grid codes, including PDE solvers. We develop a number of effective optimization strategies, and build an auto-tuning environment that searches over our optimizations and their parameters to minimize runtime, while maximizing performance portability. To evaluate the effectiveness of these strategies we explore the broadest set of multicore architectures in the current HPC literature, including the Intel Clovertown, AMD Barcelona, Sun Victoria Falls, IBM QS22 PowerXCell 8i, and NVIDIA GTX280. Overall, our auto-tuning optimization methodology results in the fastest multicore stencil performance to date. Finally, we present several key insights into the architectural tradeoffs of emerging multicore designs and their implications on scientific algorithm development. *. Node power under a computational load can differ dramatically from both idle power and from the manufacturer's peak power specifications.

Machine learning for predictive auto-tuning with boosted regression trees

“…This hand-tuning rarely generalizes well to new hardware generations or different input domains, is prone to error, results in unmaintainable code, and does not even guarantee optimal performance. One of the reasons is that GPU kernels can yield staggeringly large optimization spaces [Datta et al, 2008]. The problem is further compounded by the fact that these spaces can be highly discontinuous [Ryoo et al, 2008], difficult to explore, and optimal performance is often realized at the edge of "performance cliffs" induced by hard device-specific constraints (e.g.…”

“…Two major auto-tuning approaches have emerged in the extensive literature covering the subject (see surveys in e.g. [Vuduc et al, 2001, Williams, 2008, Datta et al, 2008, Cavazos, 2008, Li et al, 2009, Park et al, 2011): analytical model-driven optimization and empirical optimization [Yotov et al, 2003].…”

A scalable parallel algorithm for dynamic range-limited n -tuple computation in many-body molecular dynamics simulation

“…By looping over all cells, it generates the entire S (n) . This is analogous to stencil computation for solving partial differential equations, in which a stencil (equivalent to computation pattern in our framework) defines local computation rules for each grid point and its neighbor grid points and the stencil is applied to all grid points in a lattice to solve the problem [27,28].…”

Data Layout Transformation for Stencil Computations on Short-Vector SIMD Architectures

“…A number of works have addressed optimizations of stencil computations on emerging multicore platforms [7], [16], [17], [6], [27], [26], [11], [37], [10], [4], [9], [40], [38], [41], [8], [39]. In addition, other transformations such as tiling of stencil computations for multicore architectures have been addressed in [43], [25], [21], [34].…”