A Predictive Modelling based Approach to Portable Parallel Compilation for Heterogeneous Multi-cores

Lead Research Organisation: University of Edinburgh
Department Name: Sch of Informatics

Abstract

Modern computers at their heart consist of multiple processingelements. These multi-core processors have the capability ofdelivering high performance with reduced energy consumption, but arehighly challenging to program. As the number of cores is relativelysmall in number at present, operating systems can make good use ofthem. In the near future, however, the number of cores will rise andvary in type and capability. Currently, this means that programmerswill soon have to think in parallel and work out how to partitiondifferent parts of their programs to run on different types ofcores. Each time this program is run on a new platfrom or the currentone is upgraded, this task will have to be repeated. Due to the sheercomplexity of this process, as hardware increases in size andcomplexity, software will not be able to utilise its potential, leadingto software stagnation.This project aims to prevent this software stagnation by investigatingnew techniques to automatically learn how to utilise new multi-coreplatforms. Using ideas and techniques first developed in artificialintelligence, we will develop a system that automatically learns howto adapt software to work on new platforms. It uses statisticalmachine learning to determine what type of cores to use to give thebest performance and also predicts when software is out-of date.If successful it will be of significant benefit to academics workingin the area, UK industry and in the long term applicationsprogrammers.

Planned Impact

How to effectively utilise multi-cores is the number one priority for academic systems researchers worldwide. It is also one of the main concerns of processor manufacturers and IP vendors. For this reason ARM has provided a letter of support. The demand for ever more cores is likely to fall if no-one can actually achieve any noticable performance improvement. If this project succeeds it will have significant impact on these groups. In addition, application programmers will be able to achieve stable scalable performance and plan their software evolution. This in turn will have significant impact on the economy, given the size of the computing systems market. The following academic groups will benefit from this research - Parallelising Compilers - Runtime Systems - Language Design - Computer Architecture - Machine Learning for Systems Industrial groups - Embedded Systems Providers - Compiler and Tools Vendors - Application Developers
 
Description It is possible to exploit GPUS effectively for general
computing
Exploitation Route Further research projects, technolgy transfer
Sectors Digital/Communication/Information Technologies (including Software)

URL http://groups.inf.ed.ac.uk/PreMapp/
 
Description Ciompiler technology used by ARM
First Year Of Impact 2017
Sector Digital/Communication/Information Technologies (including Software)
Impact Types Economic

 
Description EPSRC Programme Grant
Amount 4,135,048 ج.م.‏ (EGP)
Funding ID EP/K008730/1 
Organisation Engineering and Physical Sciences Research Council (EPSRC) 
Sector Public
Country United Kingdom
Start 03/2013 
End 02/2016
 
Description Pervasive Parallelism CDT
Amount £3,892,290 (GBP)
Funding ID EP/L01503X/1 
Organisation Engineering and Physical Sciences Research Council (EPSRC) 
Sector Public
Country United Kingdom
Start 03/2014 
End 09/2022