An Efficient FFT Algorithm for Superscalar and VLIW Processor Architectures
✍ Scribed by Chris Basoglu; Woobin Lee; Yongmin Kim
- Publisher
- Elsevier Science
- Year
- 1997
- Tongue
- English
- Weight
- 480 KB
- Volume
- 3
- Category
- Article
- ISSN
- 1077-2014
No coin nor oath required. For personal study only.
✦ Synopsis
Real-Time Imaging 3, 441-453 (1997) applications, including digital video, images, graphics and audio. These new processors employ instruction-level parallelism, which includes the superscalar and very long instruction word (VLIW) computer architectures. Intel Pentium Pro, Hewlett-Packard PA-RISC 8000, Sun Microsystems UltraSPARC, and Texas Instruments TMS320C80 are all recently developed superscalar/VLIW processors designed to meet today's new computing environments. However, little research has been done on optimally mapping the FFT algorithm onto such architectures to improve its execution time.
A previous study has shown that some optimization can be achieved in VLIW processors [8]. By extending this idea, we developed a more efficient 1D and 2D FFT algorithm which could run on most general-purpose superscalar/VLIW microprocessors in real-time or near real-time. The new algorithm speeds up the computation by utilizing the superscalarl/LIW microprocessor's ability to perform multiple operations concurrently. We implemented this algorithm on a high-end, commercially available programmable processor. We demonstrated the algorithm's applicability to such processors and achieved the reduction in execution time necessary to be useful in real-time applications. This programmable approach via the new FFT algorithm provides the flexibility and adaptability to new applications and changing requirements, which is a definite advantage over specialized hardware-based solutions.
In this paper we briefly review the conventional FFT algorithm and previous computational approaches. We then discuss our new FFT algorithm in detail and describe its implementation. Finally, the performance of our algorithm is presented and compared to that of previous approaches.