Hi
I've had a look at the SVD code, and if I understand it correctly, the routine takes a matrix resident on the host and returns a result on the host. This seems to suggest that the GPU is used to accelerate certain parts of the algorithm, but that a lot of the work is done on the host.
Are there any plans to make a version of SVD which takes a matrix resident on the GPU and returns a result resident on the GPU?
Regards
Jacques
