ACCELERATION OF THE MATRIX MULTIPLICATION OF THIN-FILLED MATRICES BY USING COLUMN FOLDING AND COMPRESSION

摘要

The disclosed embodiments relate to accelerating sparse matrix multiplication (SMM) using column convolution and compression. In one example, a processor in response to an SMM instruction with fields for specifying locations of a first, a second and an output matrix, the second matrix being a sparse matrix, uses execution circuitry to package the second matrix by replacing one or more zero-valued elements by non-zero elements yet to be processed, each of the replaced elements further comprising a field for identifying its logical position within the second matrix, and the execution circuitry is further configured for each non-zero element on row M and column K of the specified one first matrix to produce a product of the element and each corresponding non-zero element on row K, column N of the packed second matrix and each product generated with a previous value of a corresponding element on row M and column N of the specified starting mat accumulate rix.