There were two independent bugs exposed through the additional (correct) checks for minus zero. One on ARM and one on ia32/x64. r8878 includes the original fix and should fix all these as well.
Den 9. aug. 2011 11.10 skrev Florian Schneider <[email protected]>: > I identified repeated deoptimization on a Math.floor operation as the > performance problem. We deoptimize on a zero input, but should not if we > know that the input can't be negative 0, but just +0. > > > > Den 4. aug. 2011 18.17 skrev Kevin Millikin <[email protected]>: > > Thanks for the report. I will revert while we investigate. >> >> >> On Thu, Aug 4, 2011 at 3:59 PM, M Capewell >> <[email protected]>wrote: >> >>> This patch (committed as r8804) is preventing the Kraken benchmark >>> suite completing on ARM. Backing it out seems to fix things on the >>> head. >>> >>> -- >>> v8-dev mailing list >>> [email protected] >>> http://groups.google.com/group/v8-dev >>> >> >> -- >> v8-dev mailing list >> [email protected] >> http://groups.google.com/group/v8-dev >> > > -- v8-dev mailing list [email protected] http://groups.google.com/group/v8-dev
