Generation of control and data flow graphs from scheduled and pipelined assembly code

  • Authors:

  • David C. Zaretsky;Gaurav Mittal;Robert Dick;Prith Banerjee

  • Affiliations:

  • Department of Electrical Engineering and Computer Science, Northwestern University, Evanston, IL;Department of Electrical Engineering and Computer Science, Northwestern University, Evanston, IL;Department of Electrical Engineering and Computer Science, Northwestern University, Evanston, IL;College of Engineering, University of Illinois at Chicago, Chicago, IL

  • Venue:
  • LCPC'05 Proceedings of the 18th international conference on Languages and Compilers for Parallel Computing
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

High-level synthesis tools generally convert abstract designs described in a high-level language into a control and data flow graph (CDFG), which is then optimized and mapped to hardware. However, there has been little work on generating CDFGs from highly pipelined software binaries, which complicate the problem of determining data flow propagation and dependencies. This paper presents a methodology for generating CDFGs from highly pipelined and scheduled assembly code that correctly represents the data dependencies and propagation of data through the program control flow. This process consists of three stages: generating a control flow graph, linearizing the assembly code, and generating the data flow graph. The proposed methodology was implemented in the FREEDOM compiler and tested on 8 highly pipelined software binaries. Results indicate that data dependencies were correctly identified in the designs, allowing the compiler to perform complex optimizations to reduce clock cycles.