Abstract
In this research paper we present an implementation of a singular value decomposition algorithm designed specifically for the graphics processing unit. It consists of two parts: orthogonal matrix decomposition and matrix diagonalization. Presented an implementation of bidiagonalization algorithm where we calculate the main bidiagonal matrix and two orthogonal multipliers using a series of House- holder transformations, as well as diagonalization algorithm with the help of Givens rotation matrices. Bothe these parts are implemented in jCUDA environment. Experiments have been conducted, the results of which have been thoroughly investigated on the matter of time consumption and calculations error. We’ve also compared our implementation with alternatives both on central and graphic processors.
Publisher
National Academy of Sciences of Ukraine (Co. LTD Ukrinformnauka) (Publications)
Reference10 articles.
1. Malashonok H. I., Sukharskyi S. S. A GPU-based orthogonal matrix factoriza- tion algorithm that produces a two-diago- nal shape. NaUKMA. Computer Science. 2021. retrieved from http://nrpcomp. ukma.edu.ua/article/view/246581.
2. Malashonok H. I., Semylitko М.Y. Parallel SVD algorithm for a three-diagonal matrix on a video card using the Nvidia CUDA architecture. NaUKMA. Computer Sci- ence. 2021. retrieved from http://nrpcomp. ukma.edu.ua/article/view/246582.
3. Malashonok H. I., Savchenko S. O., "Matrychni alhorytmy rozbyttia mnozhyn dlia rekomendatsiinykh system". NaUKMA, 2019.
4. S. Lahabar and P. J. Narayanan, «Singular valuedecompositiononGPUusingCUDA,» 2009 IEEE International Symposium on Parallel & Distributed Processing, 2009, pp. 1-10, retrieved from http://doi. org/10.1109/IPDPS.2009.5161058.
5. Persson, "Householder Reflectors and Givens Rotations", MIT 18.335J / 6.337J Introduction to Numerical Methods. retrieved from https://math.dartmouth. edu/~m116w17/Householder.pdf.