MFFT: A package for two- and three-dimensional vectorized discrete fourier transforms

✍ Scribed by A. Nobile; V. Roberto

• Publisher: Elsevier Science
• Year: 1986
• Tongue: English
• Weight: 1018 KB
• Volume: 42
• Category: Article
• ISSN: 0010-4655
• DOI: 10.1016/0010-4655(86)90039-1

✦ Synopsis

Title ofprogram: MFFT Nature ofphysical problem Any problem requiring Fourier transforms. Catalogue number: AALL Program obtainable from: CPC Program Library, Queen's Uni-Method of solution versity of Belfast, N. Ireland (see application form in this The Discrete Fourier Transform (DFT) is calculated using the Fast Fourier Transform (FFT [1]) algorithms implemented in issue)

place in a vectorized form. In particular the radix-2, radix-3,

Computer: CRAY X-MP 12; Installation: CINECA, Bologna, radix-5 and mixed-radix algorithms are implemented in two Italy distinct versions: decimation-in-time (or Cooley-Tukey) and decimation-in-frequency (or Gentleman-Sande). Operating system: COS 1.14 Programming language: FORTRAN 77

Restrictions on the complexity of the program

In the data matrix to be transformed each dimension has to be High speed storage required: The code requires roughly 27500 a product of integer powers of 2, 3, 5. In real transforms the words when all the routines are loaded. The complex NL x NM first dimension in the input data has to be an even number. transform requires roughly 4 * (NL + NM) + max(NL,NM) words for tables; a real transform requires additional 2 * (NL + NM) words; for a 3-dimensional NL X NMX NN transform, Typical running times the tables require normally 4* (NL -~-NM + NN) + Top performances are reached for input data lengths which are max(NL,NM,NN) words, but they can grow up to about powers of 2. The CPU time required by a complex DFT 4*NL*NM+ 9 (NL+NM+NN) words, if special treatment 64*64*64 is roughly 0.2 s, by a real DFT 512*512 is 0.12 s.

for small NL is required. This estimate does not include the If powers of 3 and 5 are present, then the typical running time memory required for the data to be transformed is increased with up to 60%.