Section: New Results

Minimizing the synchronization overhead of X10 programs

The CAMUS team has for long focused on compiling, optimizing, and parallelizing sequential programs. The project described in this section is somewhat unusual in this context, in that it targets programs written in an explicitly parallel language, and applies polyhedral modeling techniques to reschedule computations, effectively introducing parallel-to-parallel program transformations. This work has been done in collaboration with the Inria COMPSYS team at ENS Lyon, and first results will be presented at the Compiler Construction conference (CC'14) in April 2014.

The need to leverage the computing power of multi-core processors (and distributed computers) has lead to the design of explicitly parallel programming languages. Such languages often employ a fork/join model, and include syntax to launch and synchronize tasks (also called activities) with well-defined semantics. This brings parallel constructions under the control of the compiler, and introduces new optimization opportunities. Our work has focused on the various synchronization primitives available to the programmer, and more specifically on how one type of synchronization can be replaced with another for specific classes of programs, the goal being to minimize the synchronization overhead. We have demonstrated significant speedups on programs written using the X10 programming language, and have obtained similar results on equivalent Habanero-Java programs.

More specifically, our proposed optimization works by eliminating the use of clocks in X10 programs whose activities can be characterized with a polyhedral time-domain. The X10 language basically has two activity synchronization primitives: one is the explicit use of “clocks” (synchronization barriers) during activity execution, the other is the implicit use of activity containers that synchronize only on the end of activities. Under reasonable conditions on the patterns of activity creation and control, we have shown that long-running activities using clocks can be replaced by short-lived activities synchronized only on the end of their containers, and that this transformation provides a significant gain at run time. This work has two main contributions. First, it extends a known transformation framework to the case where the original program is already parallel. Second, it shows that the polyhedral model has applications far beyond its current use in data dependence and memory locality analyzes. This work also opens up new research directions. First, it turns out that our transformation is far more general than the use we currently make of it, and therefore that it provides a solid basis for other optimizations of parallel programs. Second, the polyhedral model we have developed provides an immediate cost model for synchronization primitives, which is not used in our current work, but may provide sound heuristics to adapt the optimization phase to the characteristics of specific run time components. We plan to explore these aspects in the near future.

This work has been done in collaboration with Paul Feautrier, member of the COMPSYS Inria team, in ENS Lyon. The CAMUS team has invited Paul Featrier for one week in June 2013 in Strasbourg. We are currently seeking funding to organize more frequent stays at either Lyon or Strasbourg.

This work has been invited for presentation at the LCPC workshop held in Lyon in July 2013 (http://labexcompilation.ens-lyon.fr/cpc2013 ). An extended version of this work has been accepted for publication at the Compiler Construction conference, to be held in April 2014.