Matrix-Vector Data Structure Tutorial: The multiplication of a matrix and a vector is a common operation in applications such as 3D graphics games. Multi-threading the code will help performance, both on DP Xeon processor-based systems and hyperthread technology systems. This tutorial explains you how to speed up matrix-vector multiplication with SSE instructions, threading, and data re-structuring. Learn more about hyperthread technology in this thread data structure tutorial.

---

by Kiefer Kuah, Intel Corp.

Speeding up matrix-vector multiplication with SSE instructions, threading, and data restructuring

Single-instruction multiple-data instructions, threading, and restructuring data are three common optimization methods. The performance impact that these methods can make on matrix-vector multiplication was investigated here. The different implementations were tested on three hardware configurations and compared to the version written in C. All three methods resulted in measurable gains. As expected, while it resulted in gains on the Hyper-Threading–technology–enabled system and the dual-processor Intel Xeon chip system, the multithreaded version did not exhibit performance benefits on the uniprocessor system.

Introduction
The multiplication of a matrix and a vector is a common operation in applications such as 3D graphics games. We investigated a few ways to write the code for this operation and assess the performance of each version on a 2.8-GHz uniprocessor Hyper-Threading–technology–enabled Intel Pentium 4 processor system as well as on a 2.4-GHz Pentium 4 Xeon system.