On Mar 15, 2009, at 5:42 AM, Daniel Dunbar wrote: > These artifacts look like floating point inaccuracies to me. Another way to view these would be to use clang at -O0. If they improve, we can then bugpoint them down to the optimization pass that did it?