GENERAL CONCLUSIONS


  • Assembled a comprehensive collection of FFT algorithms available from our website at www.isip.msstate.edu

  • Benchmarked the performance on DEC Alpha, Pentium Pro, and UltraSparc and SunSparc 20 processors

  • We have shown the tradeoff between memory usage and computation speed

  • Compiler optimizations have significant effect on performance of algorithms - effect speed by up to 15%

  • Bilateral algorithms like FHT are preferable

  • Computer architecture has significant effect on performance

  • Arithmetic complexity plays a vital role only on weak floating point processors