Kick-off Meeting

Kick-off Meeting

Thanks to everyone who was involved in the adaptive compilation cluster. The kick-off meeting was a huge success and we had many great presentations.

This page shows the projects each participant wants to work on. You can also download their presentations. I'll continue to update as I receive the files and information from the speakers. In addition, even if you didn't speak or were unable to attend the meeting, if you want to collaborate on a specific project, send me an email and I'll add you to this page.

Here are the details so far of people seeking collaborations:

• Lee Howes (Imperial) We are interested in developing active library technology. As an example of this see our LCPC paper, describing taking stencil based image processing loops and actively optimising combinations of these components to produce substantial performance improvements on multi-core architectures. The challenge here is to implement the fusion of such loops within gcc and we like to see this as a challenge problem, and a demonstrator for active library technology. There are many questions remaining about how a library writer should present optimisation validity and policy information to the compiler to make this happen. Collaboration in these areas would be of interest.
• Jonas Maebe (Ghent) I am mainly interested assisting people with using/expanding Diablo (to perform all sorts of static binary rewriting and analyses) to help them perform their research. Also, I'm interested in work involving machine learning.
• Grigori Fursin (INRIA) I plan to work with
  • Ghent University (Lieven et al) on characterizing/cluster program behavior in presence of multiple datasets;
  • Imperial college on run-time adaptation for irregular programs with multiple datasets
  • ST Micro on split compilation, i.e. speeding up run-time JIT optimizations
  • Edinburgh on soft errors
  • Possibly with ARM and Edinburgh on using predictive modelling to build realistic simulator power models
• Pertti Kellomaki (Tampere) We are interested in compilation for a parameterizable architecture, where the number of FUs and the interconnects can vary widely. As I mentioned in the meeting, we are using LLVM, and we are interested in sharing experiences with using it.
• Hans Vandierendonck (Ghent) We'd like to collaborate on extracting thread-level parallelism and on run-time support (eg. virtual machines) for extracting this parallelism in a speculative way.