| 摘要 |
The general purpose of the present invention is to provide a optimized procedure for performing matrix by matrix multiplication in a parallel architecture computer, either digital or optical, that is generally capable of handling two-dimensionally structured image data sets. The present invention provides a process for performing matrix by matrix multiplication in a parallel computer architecture having first and second means for representing spatially distributed data values obtainable from corresponding first and second matrices of values, said matrices having therein defined rows and columns of said values and said first and second data representing means having defined therein rows and columns of data representation positions, said architecture further having means for accumulating spatially distributed resultant values, the process comprising the steps of: (a) representing the respective data values of a first column of values of said first matrix in the corresponding rows of said first data representing means; (b) representing the respective data values of a first row of values of said second matrix in the corresponding columns of said second data representing means; (c) multiplying each of the values represented by said first data representing means with the spatially corresponding values represented by said second data representing means such that a matrix of resultant values is obtained, each resultant value being the product of the value of its corresponding row of said first column of said first matrix of values and the value of its corresponding column of said first row of said second matrix of values; (d) accumulating the respective resultant values as a matrix of cummulative resultant values; and (e) repeating steps (a)-(d) above for each successive column of said first matrix and correspondingly successive row of said second matrix.
|
