On Mon, 2007-04-09 at 16:57 -0700, Chris Lattner wrote: > On Apr 9, 2007, at 4:47 PM, Reid Spencer wrote: > > > On Mon, 2007-04-09 at 18:40 -0500, Devang Patel wrote: > >> > >> Changes in directory llvm/test/Transforms/LoopRotate: > >> > >> PhiRename-1.ll updated: 1.1 -> 1.2 > >> --- > >> Log message: > >> > >> Add check for opt crash. > > > > This idiom is becoming more and more prevalent. While it works, its > > also > > doubling the runtime of each test its used in. I'm wondering if we > > could > > invest in a "notgrep" script (since we're invoking "not" anyway) that > > would fail on empty input. Is this sufficient to avoid having to do > > things twice? > > I'm not worried about it. If you're worried about testing time, I'd > suggest changing tests like CodeGen/ARM/vfp.ll to use %t to only > invoke llc once :)
Well, those need to be done anyway. I'm not overly concerned about it but in the last year time to run the regressions has gone from about 45 seconds to > 3 minutes. Obviously, a lot of that is due to nearly 1000 new test cases, but I can imagine the number will only grow in the future. When we get to 5000 and its taking 15 minutes, perhaps then it becomes important? > > -Chris > _______________________________________________ llvm-commits mailing list llvm-commits@cs.uiuc.edu http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits