Build Status


This package provides an implementation of the FameSVD algorithm via the BLAS and LAPACK routines syrk, syevr and gemm, The provided method is faster than the SVD algorithm used in the Julia standard library and as shown in the paper faster than the Krylov-Method and Randomized-PCA.

alt text

Please note that column size was kept contstant at 1000 and the machine used had 16GB DDR4 RAM and an Intel i7-8565U CPU running at 4.6GHz.


The package provides the function fsvd which returns an LinearAlgebra.SVD object.

S = FameSVD.fsvd(A)


Xiaocan Li, Shuo Wang and Yinghao Cai: "FameSVD: Fast and Memory-efficient Singular Value Decomposition"; arXiv:1906.12085v1