AJITPar: Adaptive Just-In-Time Parallelisation

Abstract

A key element of the multicore software crisis is a lack of abstraction: most parallel code mixes coordination and computation, and assumptions about the architecture are often hard-coded, tying the code to a specific architecture. For problems with regular parallelism, portable parallel performance may still be achieved by static compilation techniques that specialise the code for the given architecture. However, the more common case of irregular parallelism can't be tackled statically; instead dynamic specialisation and dynamic scheduling are required.

AJITPar (Adaptive Just-In-Time Parallelisation) aims to achieve portable parallel performance by combining dynamic trace-based just-in-time compilation of a high-level parallel functional language with dynamic demand-driven scheduling of parallelism.

AJITPar is funded by EPSRC (EP/L000687/1), from October 2013 to May 2017.

Research Team

Phil Trinder (Principal Investigator)
Patrick Maier (Researcher Co-Investigator & Technical Lead)
Magnus Morton (PhD Student)

Visitors

SICSA Distinguished Visiting Fellow Ryan R. Newton, June 2014

SICSA Distinguished Visiting Fellow Sam Tobin-Hochstadt, Jan/Feb 2017

Publications

Patrick Maier, Magnus Morton, Phil Trinder. JIT Costing Adaptive Skeletons for Performance Portability. In Proceedings of FHPC 2016, ACM. DOI 10.1145/2975991.2975995

Magnus Morton, Patrick Maier, Phil Trinder. JIT-Based Cost Analysis for Dynamic Program Transformations. ENTCS 330:5-25 (2016) DOI 10.1016/j.entcs.2016.12.012

Patrick Maier, Magnus Morton, Phil Trinder. Towards an Adaptive Skeleton Framework for Performance Portability. Technical Report TR-2016-001, University of Glasgow. 2016. PDF

Magnus Morton, Patrick Maier, Phil Trinder. Costing JIT Traces. Technical Report TR-2015-001, University of Glasgow. 2015. PDF

Presentations

Patrick Maier. JIT Costing Adaptive Skeletons for Performance Portability. FHPC 2016, 22 September 2016, Nara, Japan.
Patrick Maier. JIT Cost Analysis for Adaptive Skeletons. SPLS, 22 June 2016, Heriot-Watt, UK.
Magnus Morton. JIT-Based Cost Analysis for Dynamic Program Transformations. RAC 2016, 2 April 2016, Eindhoven, The Netherlands.
Project Review. 01 Oct 2015, Microsoft Research, Cambridge, UK.
Patrick Maier. Towards an Adaptive Framework for Performance Portability. IFL 2015, 16 Sep 2015, Koblenz, Germany. slides (PDF)
Magnus Morton. Costing JIT Traces for Adaptive Parallelism. CEFP 2015, 8 July 2015, Budapest, Hungary.
Magnus Morton. Costing JIT Traces for Adaptive Parallelism. SPLS, 15 June 2015, St Andrews, UK.
Magnus Morton. Costing and Transforming JIT Traces for Adaptive Parallelism. TFP 2015, 3 June 2015, Sophia Antipolis, France.

Related Projects

AJITPar builds on HPC-GAP (High Performance Computational Algebra and Discrete Mathematics), which was funded by EPSRC (EP/G05553X/1) from 2009 to 2013.
Glasgow is a partner in CoDiMa (Computational Discrete Mathematics), a CCP funded by EPSRC (EP/M022641/1) from 2015 to 2020.
Glasgow leads MaRIONet (Manycore Research Innovation and Opportunities Network), funded by EPSRC (EP/P006434/1) from 2016 to 2019.
AJITPar is related to the PhD project of Blair Archibald.
- In 2017, Blair won a SSI (Software Sustainability Institute) fellowship.