High-level Synthesis of Multi-process Behavioral Descriptions

  • Authors:
  • Weidong Wang;Niraj K. Jha;Anand Raghunathan;Sujit Dey

  • Affiliations:
  • -;-;-;-

  • Venue:
  • VLSID '03 Proceedings of the 16th International Conference on VLSI Design
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper presents a new high-level synthesis methodology togenerate optimized implementations for multi-process behavioraldescriptions. The concurrent communicating processes specificationparadigm is widely used in digital circuit and system design,and is employed in all popular hardware description languages.It has been shown that interprocess communication andsynchronization can result in complex timing inter-dependencies,which significantly affect the performance of a multi-process system.However, previous research on high-level synthesis typicallytakes a one-process-at-a-time approach, and the effects of inter-processcommunication and synchronization are ignored whenperforming tasks such as scheduling, resource sharing, etc.In this paper, we demonstrate that state-of-the-art high-levelsynthesis tools can generate significantly sub-optimal implementationsfor behaviors that contain concurrent communicating processes.We present an analysis of how inter-process communicationimpacts high-level synthesis steps, and describe a newmethodology to adapt existing high-level synthesis tools to optimizemulti-process descriptions. Our methodology is basedon executing multi-process performance analysis and process-by-processscheduling in an iterative manner. The results of performanceanalysis are used to identify critical and near-critical operations,and to judiciously partition the global resource budgetinto constraints for each process. The process-level constraintsare used to drive scheduling for individual processes, so as tospeed up the overall system critical path. We present algorithmsfor key steps in the proposed methodology.We have performed extensive experiments in the context of acommercial high-level design flow to evaluate the proposed techniques.The results clearly demonstrate the utility of our techniquesin synthesizing implementations with superior area, performance,and energy consumption. For example, up to 40.0%performance improvement (average of 35.6%) was achieved withlittle or no area overheads (average of 4.8%). In effect, the proposedtechniques lead to a shift of the entire area-delay trade-offcurve for a design to include superior designs that were hithertoinfeasible. Our techniques also simultaneously result in upto 50.0% (average of 33.5%) improvement in energy and up to69.0% (average of 58.3%) in the energy-delay product.