Matrix multiplication on LUCAS

Multiplication of two N by N matrices involves N³ multiplications of elements. The task allows a large amount of parallelism to be utilized, indicating that it can be efficiently executed on a parallel computer. This paper describes how matrix multiplication is performed on LUCAS, an SIMD type parallel processor with bit-serial processing elements. The interconnection network is of Perfect Shuffle/Exchange type. The case of study is when the number of processing elements is between N² and N³. The algorithm presented can be applied to any computer with the same interconnection structure. Formulas showing how the execution time depends on data length and matrix size are presented together with measured values from execution on LUCAS.