TFX: A System for Deriving Multi-Core Applications

The physical and economic limits of processor advances are driving the industry toward multi-core architectures. In a short time, they will find their way into systems ranging from high end servers down to embedded systems. Parallel programming in one form or another will become much more common, because it will become competitively necessary. At present, however, there is inadequate tool support for developing applications that take advantage of the underlying parallel architecture. High-level abstractions and parallelizing compilers rarely yield good processor utilization. Thus developers typically resort to programming at a low-level of abstraction where the complexities of the target architecture are exposed to the application developer. This is time consuming, error-prone, and hard to test: in short, uneconomical. We propose the development of TFX to make a radical improvement in the productivity of developers, while still yielding high-performance and high processor utilization. TFX will be based on a novel, two-pronged approach to developing multi-core applications. Recognizing the benefits of abstraction in improving productivity, TFX will support user-directed refinements from a high level language. Recognizing the importance of specialization, TFX will integrate platform-specific optimization templates for achieving high performance and for making the most efficient use of the resources provided by the architecture.