25.08.2013, 02:28, "John Regehr" <[email protected]>: >> I think great addition to it would be binary search version of >> remove-unused-function. > > Definitely. This should be easy given the new binary search + clang > pass code. There probably are a few other passes that will work well in > binary search mode. > > Regarding pass_lines::0, I think we just have to try and see. One > tricky thing that Yang has already run into is that every now and then > pass_lines get really lucky and removes 50% or 75% of the test case on > the first try, but it fails to get lucky if you make any changes to the > test case before running pass_lines::0.
Even when it does not get lucky to remove 50%, it usually is lucky enough to remove ~10% quickly (at high granularity values). I think it may even be beneficial to run all stages of pass_lines (0, 1, 2, 10) one time with limit on granularity, it should take much less time than full run and have very good reduction size for time. -- Regards, Konstantin
