Abstract
User-level components of applications can be made performance-aware by annotating them with performance model and other metadata. We present a component model and a composition framework for the automatically optimized composition of applications for modern GPU-based systems from such components, which may expose multiple implementation variants. The framework targets the composition problem in an integrated manner, with the ability to do global performance-aware composition across multiple invocations. We demonstrate several key features of our framework relating to performance-aware composition including implementation selection, both with performance characteristics being known (or learned) beforehand as well as cases when they are learned at runtime. We also demonstrate hybrid execution capabilities of our framework on real applications. “Furthermore, we present a bulk composition technique that can make better composition decisions by considering information about upcoming calls along with data flow information extracted from the source program by static analysis. The bulk composition improves over the traditional greedy performance aware policy that only considers the current call for optimization.”
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
In this article, we use the terms implementation, implementation variant and variant interchangably.
We consider a recursive function as a special scenario to avoid combinatorial explosion of the solution space.
We have not encountered any such scenario yet in any application that we have ported to our framework.
By default, the system generates composition code for our own GCF runtime library. The user can set the -starpu switch to generate code for the StarPU runtime system.
Registering a variable to the runtime system creates a unique data handle (with information about size, memory address etc.) for that data in the runtime system which can be used for controlling its state and data transfers.
A point represents a single execution with certain performance relevant properties.
References
Augonnet C et al (2009) Automatic calibration of performance models on heterogeneous multicore architectures. In: Euro-Par Workshops (HPPC 2009), LNCS, vol 6043
Hong S, Kim H (2009) An analytical model for a GPU architecture with memory-level and thread-level parallelism awareness. In: Proceedings of the Annual International Symposium on Computer Architecture (ISCA)
Karcher T, Pankratius V (2011) Run-time automatic performance tuning for multicore applications. In: Euro-Par, LNCS vol 6852
Ansel J et al (2009) PetaBricks: a language and compiler for algorithmic choice. In: Proceedings conference on Programming Language Design and Implementation (PLDI)
Linderman MD et al (2008) Merge: a programming model for heterogeneous multi-core systems. In: Proceedings of the International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS)
Wernsing JR, Stitt G (2010) Elastic computing: a framework for transparent, portable, and adaptive multi-core heterogeneous computing. In: Proceedings of the Conference on Languages, Compilers, and Tools for Embedded Systems (LCTES)
Gregg C, Hazelwood K (2011) Where is the data? Why you cannot debate CPU vs. GPU performance without the answer. In: International Symposium on Performance Analysis of Systems and Software (ISPASS)
Quinlan D, Liao C (2011) The ROSE source-to-source compiler infrastructure. In: Cetus users and compiler infrastructure workshop, USA
Augonnet C et al (2011) StarPU: a unified platform for task scheduling on heterogeneous multicore architectures. Concurr Comput Pract Exper 23:187–198
Feautrier P (1991) Dataflow analysis of array and scalar references. Intl J Parallel Program 20(1)
Kicherer M et al (2011) Cost-aware function migration in heterogeneous systems. In: Proceedings of the Conference on High Performance and Embedded Architectures and Compilers (HiPEAC)
Li L, Dastgeer U, Kessler C (2013) Adaptive off-line tuning for optimized composition of components for heterogeneous many-core systems. In: Seventh Int. Workshop on Automatic Performance Tuning (iWAPT-2012), Proc. VECPAR-2012 Conference
Che S et al (2009) Rodinia: a benchmark suite for heterogeneous computing. In: IEEE International Symposium on Workload Characterization (IISWC)
Topcuoglu H et al (2002) Performance-effective and low-complexity task scheduling for heterogeneous computing. IEEE Trans Par Dist Syst 13(3)
Korch M, Rauber, T (2006) Optimizing locality and scalability of embedded Runge-Kutta solvers using block-based pipelining. J Parallel Distrib Comput 66(3)
Kicherer M, Nowak F, Buchty R, Karl W (2012) Seamlessly portable applications:Managing the diversity of modern heterogeneous systems. ACM Trans Archit Code Optim 8(4):42(1–42:20)
Dastgeer U, Li L, Kessler C (2012) The PEPPHER composition tool: Performance-aware dynamic composition of applications for GPU-based systems. In: MuCoCoS, SC12
Lee S, Vetter JS (2012) Early evaluation of directive-based gpu programming models for productive exascale computing. In: Conference for high performance computing, networking, storage and analysis
Reyes R, Sande F (2011) Automatic code generation for GPUs in llc. J Supercomput 58:349–356
Kessler CW, Löwe W (2012) Optimized composition of performance-aware parallel components. Concurr Comput Pract Exper 24(5):481–498
Ericsson M (2008) Composition and optimization. Växjö University Press, Kalmar
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Dastgeer, U., Kessler, C. Performance-aware composition framework for GPU-based systems. J Supercomput 71, 4646–4662 (2015). https://doi.org/10.1007/s11227-014-1105-1
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11227-014-1105-1