Multi-Core Software

Future-Proof Data Parallel Algorithms and Software on Intel® Multi-Core Architecture

CONCLUSION

Future-proofing algorithms for multi-core architectures is an important way to continue to reap the performance benefits of Moore's Law scaling. Data parallel programming models offer a promising abstraction to use for forward-scaling, but it is often limited by too-narrowly defined types and operators. This severely limits the scope of applicability for such models. In Ct, we are attempting to build a system that delivers a more general data parallel (indeed, a deterministic task parallel) model while providing the essential framework for forward-scaling.

A serious challenge is acknowledging the realities of modern software development methods and assuring compatibility with legacy code and programming methodologies. We believe that using the Ct Dynamic Engine's particular flavor of adaptive compilation and run-time is the most effective way to extract chains of performance code without seriously compromising the language design or the large software investment by developers. More radical language redesigns are likely to appear at some point, but we view this more incremental approach as a flexible and highly productive way to leverage multi-core architectures while developing the basic parallel algorithms and design methods. In fact, we expect that future languages will almost certainly encompass some form (if not the exact form) of the ideas in Ct.

Back to Top