The singular value decomposition (SVD) has numerous applications, including signal processing, data compression, principal component analysis (PCA), pattern recognition, and so on. Many applications involve large-size data matrices; however, the SVD is computationally intensive. This paper presents a high-performance SVD^algorithm for large matrices. Its high performance is achieved by utilizing multicore architecture for parallelism and exploiting locality to reduce the traffic between fast memory (caches) and slow memory (disk).
展开▼
