https://hgpu.org/?p=29156
Fast Truncated SVD of Sparse and Dense Matrices on Graphics Processors