A Stream Compiler (ASC)

for Computing with Field Programmable Gate Arrays (FPGAs)






Summary:

FPGAs enable the design of accelerators for microprocessors. The last decade has shown that we can accelerate certain parts of applications by up to three orders of magnitude. The two main scientific challenges of acceleration with FPGAs are:

  1. FPGAs are ~5x slower than microprocessors in terms of clock frequency, and
  2. there is no standard programming methodology for FPGA such as it exists for microprocessors.
The first challenge is traditionally attacked by using parallelism. In the case of FPGAs we have the additional option to explore parallelism and pipelining on the word (instruction) level but also on the bit-level. Parallelism on the bit-level, or BLP, is exploited by optimizing the the number representation of data values inside the FPGA.

The second challenge is an active area of current research in industry and academia. The solution provided in the ASC project is based on a strong reliance on object-oriented features of C++ to build a tightly integrated module generation and compiler system. ASC can achieve orders of magnitude speedup for a substantial set of applications or application domains such as signal processing (e.g. for wireless, or voice), compression, encryption, image and video processing, graphics, networking, VLSI CAD and scientific computing.

ballTalks: Show and Tell (Nov. 2001) [pdf, www]

Stanford, Computer Systems Colloquium (May 2nd, 2001)[pdf, www,Stanfordwebcast]

Princeton (Dec. 2000) [pdf, www]


ballStudent Projects:

             Benjamin Tai, M.Sc. Thesis, "A Python Interface for Hardware Scriping with ASC," [in progress]

                 Altaf Gaffar, PhD Thesis, "Floating Point Precision Analysis using Automatic Differentiation," [in progress]

                 Umang Garg, M.Sc. Thesis, "Numerical Computing in the Logarithmic Domain," [report]

                 Henry Styles, Summer 2001, "Accelerating Capacitance Extraction with ASC," [report, slides]

                 Zhining Huang, Summer 2001, "Accelerating Graph Algorithms," [report, slides]

                 Nicols Boullis, Summer 2000, "Elementary Function Evaluation on FPGAs," [report]

                 Heiko Huebert, Dipl.Work 2000, "Scheduling Dataflow Graphs for StReAm"
 

ASC Users Manual: preliminary version (March 2002) [pdf]

ballRelevant Publications:

     Oskar Mencer, Nicolas Boullis, Wayne Luk, Henry Styles,"Parameterized Function Evaluation for FPGAs,"
Field-Programmable Logic (FPL), Belfast, Ireland, September 2001.

    Oskar Mencer, Marco Platzner, Martin Morf, Michael J. Flynn, "Object-oriented Domain-Specific Compilers for Programming FPGAs,"
IEEE Transactions on VLSI, special issue on Reconfigurable Computing, March 2001.

     Oskar Mencer, Heiko Huebert, Martin Morf, Michael J. Flynn, "StReAm: Object-Oriented Programming of Stream Architectures using PAM-Blox",
Plenary Session, Field-Programmable Logic (FPL), Villach, Austria, September 2000.

     Oskar Mencer, Luc Semeria, Martin Morf, Jean-Marc Delosme, "Application of Reconfigurable CORDIC Architectures,"
     The Journal of VLSI Signal Processing, special issue on Reconfigurable Computing, Kluwer, March 2000.

     Oskar Mencer, Martin Morf, Michael J. Flynn, "PAM-Blox: High Performance FPGA Design for Adaptive Computing,"
     IEEE Symposium on FPGAs for Custom Computing Machines (FCCM), Napa Valley, April 1998.






Last update : 27/11/01