Why do pca(dataset), svd(cov(dataset)) provide different principal components but same eigen values?

In Matlab there are two functions pca() and princomp() which give exactly same results for finding principal components.

[u s v]=pca(dataset) or [u1 s1 v1]=princomp(dataset).    I found u=u1.

However, if I use [a b c]=svd(cov(dataset)). [ cov() gives covarience of dataset after mean normalization]

It is found that u != a , why???? However, all three provide same eigen values.