Performance Analysis and Optimization of the OP2 Framework on Many-Core Architectures

Authors:
M.B. Giles;G.R. Mudalige;Z. Sharif;G. Markall;P.H.J. Kelly
Affiliations:
-;-;-;-;-
Venue:
The Computer Journal
Year:
2012

Citing 0
Cited 7

Predictive modeling and analysis of OP2 on distributed memory GPU clusters

Proceedings of the second international workshop on Performance modeling, benchmarking and simulation of high performance computing systems
Mesh independent loop fusion for unstructured mesh applications

Proceedings of the 9th conference on Computing Frontiers
Using domain-specific languages and access-execute descriptors to expand the parallel code synthesis design space: keynote talk

Proceedings of the 1st ACM SIGPLAN workshop on Functional high-performance computing
Predictive modeling and analysis of OP2 on distributed memory GPU clusters

ACM SIGMETRICS Performance Evaluation Review
Abstractions to separate concerns in semi-regular grids

Proceedings of the 27th international ACM conference on International conference on supercomputing
Vectorizing Unstructured Mesh Computations for Many-core Architectures

Proceedings of Programming Models and Applications on Multicores and Manycores
Design and initial performance of a high-level unstructured mesh framework on heterogeneous parallel systems

Parallel Computing

Quantified Score

Hi-index	0.00

Visualization

Abstract

This paper presents a benchmarking, performance analysis and optimization study of the OP2 ‘active’ library, which provides an abstraction framework for the parallel execution of unstructured mesh applications. OP2 aims to decouple the scientific specification of the application from its parallel implementation, and thereby achieve code longevity and near-optimal performance through re-targeting the application to execute on different multi-core/many-core hardware. Runtime performance results are presented for a representative unstructured mesh application on a variety of many-core processor systems, including traditional X86 architectures from Intel (Xeon based on the older Penryn and current Nehalem micro-architectures) and GPU offerings from NVIDIA (GTX260, Tesla C2050). Our analysis demonstrates the contrasting performance between the use of CPU (OpenMP) and GPU (CUDA) parallel implementations for the solution of an industrial-sized unstructured mesh consisting of about 1.5脗聽million edges. Results show the significance of choosing the correct partition and thread-block configuration, the factors limiting the GPU performance and insights into optimizations for improved performance.