Through the computing method and parallel computer null network of the alliance linear equation by the memory

摘要

PROBLEM TO BE SOLVED: To solve simultaneous linear equations at high speed by using a comparatively small memory while efficiently performing transfer between processors. ;SOLUTION: Blocks arranged at processors are cyclically selected and numbered (1) and these numbered blocks are cyclically processed in the order of numbers so that LU decomposition is performed (2). In the case of matrix product calculation in the LU decomposition, data as the calculation object of matrix product are divided and transferred to the respective processors and by simultaneously performing the calculation of the matrix product at each processor to the divided data and parallel transfer, transfer time is shortened. Next, each processor performs formed substitution concerning the LU decomposed result and transfers the result to the adjacent processor (3). Besides, backward substitution is similarly performed as well. When transferring the result of forward/backward substitution to the other processor, the excessive data of one bit are transferred at least in addition to the transfer data and from these one-bit data, the reception of data is confirmed at the respective processors.;COPYRIGHT: (C)1999,JPO