EFFICIENT MATRIX MULTIPLICATION USING HARDWARE INTRINSICS AND PARALLELISM WITH C#

Abstract

In this paper we improve the efficiency of the simple matrix-multiplicationalgorithm using parallelism and hardware instrinsics with C# and .Net Task ParallelLibrary. We demonstrate how additional processing can result the overall shorterexecution times. We present an algorithm in C# that achieves up to five times performanceimprovement on midrange hardware, using single-instruction-multiple-datainstructions and unsafe code foo direct access to memory where other techniques arenot possible in the .Net managed environment.