Title :
Matrix multiplication on the connection machine
Author :
Johnsson, S. Lennart ; Harris, Tim ; Mathur, Kapil K.
Author_Institution :
Thinking Machines Corp., 245 First Street, Cambridge, MA and Department of Computer Science, Yale University, New Haven, CT
Abstract :
A data parallel implementation of the multiplication of matrices of arbitrary shapes and sizes is presented. A systolic algorithm based on a rectangular processor layout is used by the implementation. All processors contain submatrices of the same size for a given operand. Matrix-vector multiplication is used as a primitive for local matrix-matrix multiplication in the Connection Machine system CM-2 implementation. The peak performance of the local matrix-matrix multiplication is in excess of 20 Gflops s-1. The overall algorithm including all required data motion has a peak performance of 5.8 Gflops s-1.
Keywords :
Algorithm design and analysis; Bandwidth; Parallel algorithms; Parallel architectures; Permission; Pipelines; Registers; Scientific computing; Shape; Software libraries;
Conference_Titel :
Supercomputing, 1989. Supercomputing '89. Proceedings of the 1989 ACM/IEEE Conference on
Conference_Location :
Reno, NV, United States
Print_ISBN :
0-89791-341-8
DOI :
10.1145/76263.76298
