I have read that one way to perform non-inferiority (= one-sided) tests is to build two-sided tests with the alpha risk twice the wanted value. Eg : if you want to test non-inferiority with 5% alpha risk, test equality with 10% alpha risk.
One way to do this with SAS using covariates to stratify analysis is through GLM (fixed effects only) or MIXED (fixed and random effects) procedures.
I am not sure if this way of performing one-sided tests is really exact, or just a workaround with close results to the true method (or something totally insane, by the way, but I'm sure I've read some people doing so).
Olivier