Difference between revisions of "Project Plan"

End date: 31 August 2013
Work Package Leader: Embecosm
Contributing Partners: Embecosm

• Definition of compiler and hardware independent interface for machine learning compiler.
• Selection of a set of software characteristics to be exploited during the optimisation selection process.
• Identify target for first implementation, GCC or LLVM.

• Identify target for first implementation, GCC or LLVM.
• Determine degree of integration with specific compilers.
• Identify machine learning interface.
• Identify feature selection methodology.
• Iterative refinement on 2-4.

• General computing infrastructure

• None

• Work Package 2

End date: 30 Nov 2014
Work Package Leader: Embecosm
Contributing Partners: Embecosm

• Implementation of prototype framework with one compiler, identified in Work Package 1
• Implementation of prototype framework with other compiler.

• Write code for use with first compiler.
• Write documentation for use with first compiler.
• Implement regression tests for use with first compiler.
• Extend support for use with second compiler (code, documentation, regression).
• Iterative refinement on 1-4.

• General computing infrastructure

• Work Package 1

• None

End date: 31 July 2013
Work Package Leader: University of Bristol
Contributing Partners: University of Bristol

• Implementation of Hardware Measurement

• Design board, reusing existing expertise
• Board Implementation
• Board testing

• PCB Manufacturing (outsourced)

• None

• Work Package 4 (to instrument boards)

End date: 30 September 2013
Work Package Leader: University of Bristol
Contributing Partners: University of Bristol, Embecosm

• Creation of a set of applications for training.
• Creation of a set of applications for initial testing of trained systems.
• Development of larger case studies for testing purposes.
• Selection of target embedded systems for testing.

• Select suitable test and training applications from existing benchmark suites.
• Choice of case studies from wider community.
• Selection of embedded systems representative of industrial/commercial applications in consultation with community.
• Integration of embedded systems with hardware test platform (Work Package 3)

• Selection of embedded systems
• Community engagement platforms

• None

• None

End date: 31 August 2014
Work Package Leader: University of Bristol
Contributing Partners: University of Bristol, Embecosm

• Understand current machine learning techniques and decide if relevant
• Select approach(es) for incorporating into framework
• Refinement in the light of ongoing project development and experience

• Review existing uses including MILEPOST, directed learning, abductive learning (Literature)
• Whole team working days to bring together theory with implementers to select approach and specify API
• Decision on choice of training approach, e.g. FFD, random, etc.
• Iterative review during second year of programme, inc. potential for reordering

• General computing infrastructure

• None

• None

End date: 31 August 2013
Work Package Leader: University of Bristol
Contributing Partners: University of Bristol, Embecosm

• Theoretical analysis of why energy optimisations work.

• Look at all existing work with energy measurement of relevant systems to look for data to guide implementers of optimisation passes to implement those specific for energy minimisation.

• General computing infrastructure

• None

• None

End date: 31 May 2014
Work Package Leader: Embecosm
Contributing Partners: Embecosm

• Set up of training infrastructure and demonstration with tests using existing optimisations
• Refine/repeat for use with new optimisations

• Set up infrastructure for existing optimisations
• Train infrastructure with small set with existing optimisations
• Test with small set with existing optimisations
• Repeat above with new optimisations from Work Package 8

• High performance workstation
• Embedded systems with hardware energy measuring

• Work Package 2 (Compiler Infrastructure)
• Work Package 4 (Test hardware, test applications)
• Work Package 8 (New Optimisations)

• None

End date: 31 May 2014
Work Package Leader: Embecosm
Contributing Partners: Embecosm

• Design and implement optimisation passes in the GCC and LLVM compilers.

• Design new optimisation passes in light of theory from Work Package 6
• Implement new optimisation passes in GCC
• Reimplement optimisation passes in LLVM

• General computing infrastructure

• None

• None

End date: 30 Nov 2014
Work Package Leader: University of Bristol
Contributing Partners: University of Bristol, Embecosm

• Full evaluation of infrastructure using full training sets, full test sets and large case studies for both with and without our new optimisations.

• Train infrastructure with existing optimisations
• Evaluate with smaller tests with existing optimisations
• Evaluate with case studies with existing optimisations
• Repeat above with new optimisations
• Write paper detailing findings
• Review and refine paper

• Instrumented embedded systems
• HPC Facilities

• None

• None

End date: 30 Nov 2014
Work Package Leader: University of Bristol, Embecosm
Contributing Partners: University of Bristol, Embecosm

• Business case development inc. market analysis
• Engagement with relevant communities
• Engagement with potential customers
• Academic and business publications

• Develop business case by engagement of all stakeholders
• Ongoing review of business case throughout project, leading to updated exploitation plan
• Engagement with the technical community through participation in workshops (including EACO, NMI, etc.), conferences (including GNU Tools Cauldron, LLVM Developer Conference, etc.), presentations, training events, new media using the skills of AB Open
• Engagement with potential customers
• Publication of papers as described in various Work Packages.

• General computing infrastructure, including website and social media (http://mageec.org)

• None

• None