I need to compute svd of a matrix with m<n. I was trying to use “cusolverDnSgesvd” but it only supports matrices with m>=n.
I found one solution to use “cusolverDnSgesvdj” (using Jacobi method) which has no such restriction but is only available on CUDA 9.0 .
Is there any way to compute this svd on CUDA 8.0 or do I need to take transpose twice to get my work done?
Thanks in advance!