SVD problem_ Is it a bug?

A couple of days ago I tested the SVD operator on my text dataset with 23000 features on a relatively high performance machine. After 10 hours, the algorithm was finished and I got about 2000 features in the new reduced space (SVD's output). But all of the values in the are zero!!!!! It should be mentioned that I tested the SVD operator on iris data set and it worked fine.

Any one knows what is wrong here? Is it possible that I have encountered a bug in SVD implementation for sparse matrix?

cheers.