AP-931 Streaming SIMD Extensions - LU Decomposition
This application note describes LU Decomposition of matrices with arbitrary dimensions using Intel?s Streaming SIMD Extensions.
The performance of the code, which uses the Streaming SIMD Extensions for LU Decomposition, is approximately 2.6x times faster (for 15 x 15 matrices) than a generic C code implementation (See section 5.1). With increasing matrix dimension, the performance ratio (for 30 x 30 ? 3.5, 40 x 40 ? 4.0) increases as well. These measurements are based on tests run on a 450MHz Pentium® III processor.