http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51269
--- Comment #1 from Ira Rosen <irar at il dot ibm.com> 2011-11-23 10:46:00 UTC --- Created attachment 25895 --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25895 patch Right, this looks like a bug. The attached (not really tested) patch tries to propagate prologue_after_cost_adjust. Unfortunately, the patch will not help with your problem. This param has no effect on peeling. I think you need a param min-niters-for-peeling to control the minimum of scalar loop iterations to apply peeling for alignment (in case that all the unaligned accesses are supported). Could you please open a new PR for this?