Using domain-specific languages and access-execute descriptors to expand the parallel code synthesis design space: keynote talk
Pages 1 - 2
Abstract
This talk is about the following idea: can we simultaneously raise the level at which programmers can reason about code, and also provide the compiler with a model of the computation that enables it to generate faster code than you could reasonably write by hand?
We have been working with three large computational fluid dynamics frameworks [3, 5, 8], and I will present some of our experience in building compiler tools at various levels of abstraction. Our primary goal is to build tools that automatically synthesise the best possible implementation. By getting the abstraction right, we can capture design choices far beyond what a conventional compiler can do.
I will illustrate this with examples involving low-level parallel code generation (eg for GPUs) [1, 4], high-level cross-cutting almost-algorithmic choices (such as whether to actually build a global sparse system matrix) [6, 7], and semantic properties (enabling massive common subexpression elimination in finite-element assembly). I will also show some of the power of a generative approach in supporting free navigation of the alternatives, such as refining either the mesh or the polynomial order in a finite-element fluid dynamics application [2].
What is the right code to generate, for a given hardware platform? How does this change as problem parameters change? The key, we believe, is to start with the right representation of the problem, and to build tools that can automate the combination of code generation alternatives.
References
[1]
C. Bertolli, A. Betts, P. H. J. Kelly, G. R. Mudalige, and M. B. Giles. Mesh independent loop fusion for unstructured mesh applications. In J. Feo, P. Faraboschi, and O. Villa, editors, Conf. Computing Frontiers, pages 43--52. ACM, 2012. ISBN 978-1-4503-1215-8.
[2]
C. D. Cantwell, S. J. Sherwin, R. M. Kirby, and P. H. J. Kelly. From h to p efficiently: selecting the optimal spectral/hp discretisation in three dimensions. Math. Mod. Nat. Phenom., 6 (3): 84--96, 2011.
[3]
D. DR, W. CR, and K. SC. Fluidity: A fully unstructured anisotropic adaptive mesh computational modeling framework for geodynamics. Geochem Geophy Geosy, 12, June 2011.
[4]
M. B. Giles, G. R. Mudalige, Z. Sharif, G. R. Markall, and P. H. J. Kelly. Performance analysis and optimization of the op2 framework on many-core architectures. Comput. J., 55 (2): 168--180, 2012.
[5]
A. Logg, K.-A. Mardal, and G. N. Wells, editors. Automated Solution of Differential Equations by the Finite Element Method, volume 84 of Lecture Notes in Computational Science and Engineering. Springer, 2012. 10.1007/978-3-642-23099-8. URL http://dx.doi.org/10.1007/978-3-642-23099-8.
[6]
G. Markall, A. Slemmer, D. Ham, P. Kelly, C. Cantwell, and S. Sherwin. Finite element assembly strategies on multi-core and many-core architectures. International Journal for Numerical Methods in Fluids, January 2012.
[7]
G. R. Markall, D. A. Ham, and P. H. J. Kelly. Towards generating optimised finite element solvers for gpus from high-level specifications. Procedia CS, 1 (1): 1815--1823, 2010.
[8]
P. Moinier, J.-D. Muller, and M. Giles. Edge-based multigrid and preconditioning for hybrid grids. AIAA Journal, 40 (10): 1954--1960, 2002.
Recommendations
SystemCoDesigner—an automatic ESL synthesis approach by design space exploration and behavioral synthesis for streaming applications
With increasing design complexity, the gap from ESL (Electronic System Level) design to RTL synthesis becomes more and more crucial to many industrial projects. Although several behavioral synthesis tools exist to automatically generate synthesizable ...
Bootstrapping domain-specific meta-languages in language workbenches
GPCE 2016: Proceedings of the 2016 ACM SIGPLAN International Conference on Generative Programming: Concepts and ExperiencesIt is common practice to bootstrap compilers of programming languages. By using the compiled language to implement the compiler, compiler developers can code in their own high-level language and gain a large-scale test case. In this paper, we ...
An Optimal Parallel Nonoverlapping Domain Decomposition Iterative Procedure
A nonoverlapping domain decomposition iterative procedure is developed and analyzed for second order elliptic problems in $\mathbb R^N$. Its convergence is proved. The method is based on a Robin-type consistency condition with two parameters, called a ...
Comments
Information & Contributors
Information
Published In
September 2012
110 pages
Copyright © 2012 Author.
Sponsors
Publisher
Association for Computing Machinery
New York, NY, United States
Publication History
Published: 15 September 2012
Check for updates
Author Tags
Qualifiers
- Abstract
Conference
ICFP'12
Sponsor:
ICFP'12: ACM SIGPLAN International Conference on Functional Programming
September 15, 2012
Copenhagen, Denmark
Acceptance Rates
Overall Acceptance Rate 18 of 25 submissions, 72%
Upcoming Conference
ICFP '25
- Sponsor:
- sigplan
Contributors
Other Metrics
Bibliometrics & Citations
Bibliometrics
Article Metrics
- 0Total Citations
- 95Total Downloads
- Downloads (Last 12 months)1
- Downloads (Last 6 weeks)0
Reflects downloads up to 09 Feb 2025
Other Metrics
Citations
View Options
Login options
Check if you have access through your login credentials or your institution to get full access on this article.
Sign in