- Keywords: programming languages, program analysis, quantitative analysis, design patterns
- Project type: Individual or small group
- Requirements: Functional programming skills
This project seeks to better understand the prevalent programming patterns in computational science to help inform future language designs targeted at aiding scientific computing.
There is currently a lack of quantitative data on programming idioms, approaches, and needs in science. Many studies suggesting key idioms are high-level and anecdotal (e.g., [1]). Other studies are too low-level (at the syntactic- level, e.g., [2, 3, 5]) rather than taking into account language semantics. The aim of this project is to develop a programming pattern analysis tool for Fortran that can be used to analyse whole code-base and extract metrics about the programming patterns employed.
This project will be build on CamFort, a research infrastructure for analysing and transforming Fortran code base (see some details in [4]). Fortran remains a popular language for scientific computing, partly due to the legacy of long-lived models written in Fortran that continue to be useful. There is a large amount of scientific computing code in Fortran available for analysis and I have a corpus of models (including the Met Office UM, which is about a million lines of code in length) available as a target for the tool.
In the first instance, the analysis tool will examine the use of iteration over arrays in terms of their data dependencies (e.g., are the loops potentially data parallel), side effects (e.g., how easy is it to make safe loop transformations), and data access patterns (both reading and writing). The project will require some refinement and exploration of what can be analysed and tracked, and what this data might be used for.
Potential results and idioms might be, for example, that X% of programs have a gather pattern (read a number elements from an array and write to a single location at a time), with a finite data access stencil, and a shared portion of data under Y kb of expected storage. If X% is significant, then this implies we should focus future language design efforts on supporting such computations as primitive.
There is already a simple tool for doing some array analysis here: https://github.com/camfort/array-analyse with results published recently [6].
[1] K. Asanovic, R. Bodik, et al. The Parallel Computing Laboratory at U.C. Berkeley: A re- search agenda based on the Berkeley view. Technical Report UCB/EECS-2008-23, University of California, Mar 2008.
[2] Knuth, Donald E. "An empirical study of FORTRAN programs." Software: Practice and experience 1.2 (1971): 105-133.
[3] M. M Mendez, J. Overbey, and F. G. Tinetti. Legacy fortran software: applying syntactic metrics to global climate models. In XVIII Congreso Argentino de Ciencias de la Computacion, 2012.
[4] http://github.com/camfort
[5] R.-G. Urma and A. Mycroft. Programming language evolution via source code query languages. In 4th workshop on Evaluation and usability of programming languages and tools, pages 35-38. ACM, 2012.
[6] D. Orchard, M. Contrastin, M. Danish, A. Rice, Verifying Spatial Properties of Array Computations, Proceedings of the ACM on Programming Languages, OOPSLA(1), 2017