Use PredIteratorCache in LCSSA, which gives a 37% overall speedup on the testcase...