SPARSE MATRIX BY VECTOR MULTIPLICATION

摘要

The invention involves pre-processing the matrix according to an encoding scheme whereby the non-zero data (in any numerical format), blocking information, the row an column offset indices within a block are represented by state machine control words which are combined in a single data stream. Thus, a single vector may be used to store all of the matrix information required to compute a sparse matrix by vector multiplication. Therefore, the system can be used effectively with a single memory channel. Also, it can be used in parallel with multiple independent memory channels. This method of matrix-by-vector multiplication achieves allows very high FPU utilization to be achieved for low bandwidth matrices such as those from finite element calculations. Also, it allows local memory buffers to be simple, and so there is no need for a complex cache architecture.