Selecting Mathematical Method for Systolic Processing
Authors: M. P. Bekakos, I. Ž . Milovanović, T. I. Tokić, Ć . B. Dolićanin, E.I.Milovanović
Keywords: systolic arrays, matrix multiplication
Abstract:
The most important aspect in the design of a systolic arrays is the mapping of the algorithm to the processor array. However, not all algorithms can be systolized. Only highly regular algorithms with the structure of nested loops are suitable for systolic implementation. Therefore, one has to choose an appropriate mathematical method that can be systolized. In this paper we analyze the problem of choosing the most suitable mathematicam method for systolic implementation. We illustrate this problem on the example of matrix multiplication.
References:
[1] K. T. JOHNSON, A. R. HURSON, B. SHIRAZI, General purpose systolic arrays, IEEE Computer
11 (1993), 20-31.
[2] S. Y. KUNG, it VLSI array processors, Prentice Hall, New Jersey, 1988.
[3] H. T. KUNG, Why systolic architectures?, Computer, 15 (1982), 37-46.
[4] S. G. SEDUKHIN, it The designing and analysis of systolic algorithms and structures, Programming,
2 (1990), 20-40. (in Russian)
[5] V. V. VOEVODIN, Mathematical models and methods in parallel processing, Nauka,
Moscow, 1986. (in Russian)
[6] I. Z. MILENTIJEVIć , I. Ž. MILOVANOVIć, E. I. MILOVANOVIć , M. K. STOJčEV, The
design of optimal planar systolic arrays for matrix multiplication, Comput. Math. Appl., 33,
6, (1997), 17-35.
[7] I. Ž . MILOVANOVIć, T. I. TOKIć , E. I. MILOVANOVIć , M. K. STOJčEV, Determining the
number of processing elements in systolic arrays, Facta Universitatis, Ser. Math. Inform.,
Vol 15, 1 (2000), 123-132.
[8] J. M. ORTEGA, Introduction to parallel and vector solution on linear systems, Plenum Press,
1988.
[9] I. Ž . MILOVANOVIć, E. I. MILOVANOVIć , I. Z. MILENTIJEVIć , M. K. STOJčEV, Designing
of processor-time optimal systolic arrays for band matrix-vector multiplication, Comput.
Math. Appl., Vol 32, 2 (1996), 21-31.