Welcome to the new version of CaltechAUTHORS. Login is currently restricted to library staff. If you notice any issues, please email coda@library.caltech.edu
Published January 1988 | public
Book Section - Chapter

Portable programming within a message-passing model: the FFT as an example

Abstract

This paper describes a portable programming environment for MIMD concurrent processors based on an object-oriented, message-passing paradigm. The basis of this environment is the Virtual Machine Loosely Synchronous Communication System (VMLSCS) which is designed to be used for loosely synchronous problems. VMLSCS is structured to make efficient use of hierarchical memory, and permits communication and calculation to be overlapped on certain concurrent processors. As an example, the use of VMLSCS in performing both one-dimensional and multi-dimensional fast Fourier transforms (FFTs) on concurrent multiprocessors is described. It is shown that all necessary interprocessor communication can be performed by a single routine, vm_index. Thus the construction of a portable concurrent FFT rests on the implementation of vm_index on the target machines. In the multi-dimensional algorithm a strip decomposition is applied to each of the directions in turn so that each of the FFTs performed in a particular direction are done in one processor. This allows fast sequential one-dimensional FFTs to be exploited. The implementation of vm_index on both homogeneous and inhomogeneous hypercubes, and shared memory multiprocessors is discussed.

Additional Information

© 1988 ACM. The support of the Department of Energy, under grant number DE-FG03-85ER25009, is gratefully acknowledged. I would also like to acknowledge helpful comments by Ed Felten. John Salmon and Stuart Anderson contributed to my understanding of parallel FFT algorithms. Finally, I would like to thank Geoffrey Fox for reading a earlier draft and making valuable suggestions.

Additional details

Created:
August 19, 2023
Modified:
October 24, 2023