Takagi Factorization on GPU using CUDA

Takagi factorization or symmetric singular value decomposition is a special form of SVD applicable to symmetric complex matrices. The computation takes advantage of symmetry to reduce computation and storage requirements. The Jacobi method with chess tournament ordering was used to perform the computation in parallel on a GPU using the CUDA programming model. We were able to achieve speedups of over 11x and 7x over CPU serial and Pthreads implementations, respectively, for matrix sizes greater than 512 × 512.

[1]  W. Gragg,et al.  Singular value decompositions of complex symmetric matrices , 1988 .

[2]  Jie Cheng,et al.  Programming Massively Parallel Processors. A Hands-on Approach , 2010, Scalable Comput. Pract. Exp..

[3]  T. Hahn.  Routines for the diagonalization of complex matrices , 2006 .

[4]  Xiaohong Wang,et al.  A Parallel Jacobi Method for the Takagi Factorization , 2002, PDPTA.

[5]  James Demmel,et al.  Jacobi's Method is More Accurate than QR , 1989, SIAM J. Matrix Anal. Appl..

[6]  Wei Xu,et al.  A twisted factorization method for symmetric SVD of a complex symmetric tridiagonal matrix , 2009, Numer. Linear Algebra Appl..

[7]  P. J. Narayanan,et al.  Singular value decomposition on GPU using CUDA , 2009, 2009 IEEE International Symposium on Parallel & Distributed Processing.