Thanks for the data, the numbers look fine. Best regards, Vladimir Ivanov PS: as I told you privately, some Octane benchmarks are noisy and 5 forks aren't enough to get stable results. On 7/16/15 4:42 PM, Michael Haupt wrote:
Hi Paul, Vladimir,
the data exhibit some decreases and some increases. Please comment.
Best,
Michael
Original JDK: Benchmark Mode Samples Score Score error Units o.m.b.s.o.Box2DBench.test avgt 25 68.136 5.515 ms/op o.m.b.s.o.CodeLoadBench.test avgt 25 3827.190 59.335 ms/op o.m.b.s.o.CryptoBench.test avgt 25 9.987 0.084 ms/op o.m.b.s.o.DeltaBlueBench.test avgt 25 939.489 8.072 us/op o.m.b.s.o.EarleyBoyerBench.test avgt 25 21.667 0.369 ms/op o.m.b.s.o.GbemuBench.test avgt 25 69.289 1.381 ms/op o.m.b.s.o.MandreelBench.test avgt 25 1168.844 22.119 ms/op o.m.b.s.o.NavierStokesBench.test avgt 25 7.316 0.019 ms/op o.m.b.s.o.PdfJSBench.test avgt 25 108.611 9.979 ms/op o.m.b.s.o.RaytraceBench.test avgt 25 7002.523 52.355 us/op o.m.b.s.o.RegexpBench.test avgt 25 99.227 2.064 ms/op o.m.b.s.o.RichardsBench.test avgt 25 310.353 1.122 us/op o.m.b.s.o.TypescriptBench.test avgt 25 5.093 0.092 s/op
Modified JDK: Benchmark Mode Samples Score Score error Units o.m.b.s.o.Box2DBench.test avgt 25 57.906 1.937 ms/op o.m.b.s.o.CodeLoadBench.test avgt 25 3789.170 38.922 ms/op o.m.b.s.o.CryptoBench.test avgt 25 9.854 0.120 ms/op o.m.b.s.o.DeltaBlueBench.test avgt 25 937.746 3.167 us/op o.m.b.s.o.EarleyBoyerBench.test avgt 25 20.890 0.428 ms/op o.m.b.s.o.GbemuBench.test avgt 25 68.949 0.421 ms/op o.m.b.s.o.MandreelBench.test avgt 25 1208.157 18.836 ms/op o.m.b.s.o.NavierStokesBench.test avgt 25 6.774 0.017 ms/op o.m.b.s.o.PdfJSBench.test avgt 25 111.189 11.428 ms/op o.m.b.s.o.RaytraceBench.test avgt 25 6905.455 137.892 us/op o.m.b.s.o.RegexpBench.test avgt 25 96.781 1.628 ms/op o.m.b.s.o.RichardsBench.test avgt 25 310.157 1.956 us/op o.m.b.s.o.TypescriptBench.test avgt 25 4.991 0.054 s/op
Am 15.07.2015 um 13:58 schrieb Michael Haupt <michael.haupt@oracle.com>:
Hi Paul, Vladimir,
thanks - I'll be back with perf numbers ASAP.
Best,
Michael
Am 15.07.2015 um 12:51 schrieb Paul Sandoz <paul.sandoz@oracle.com>:
On Jul 15, 2015, at 10:23 AM, Michael Haupt <michael.haupt@oracle.com> wrote:
Dear all,
please review and sponsor this change. RFE: https://bugs.openjdk.java.net/browse/JDK-8062543 Webrev: http://cr.openjdk.java.net/~mhaupt/8062543/webrev.00/
The change removes the no longer needed MethodHandleImpl.castReference method. Its uses are replaced with Class.cast.
Look good. I agree with Vladimir on checking for performance regressions (i suspect/hope! there will not be any).
Paul.
Tested with: * jtreg (java.lang.invoke, java.util.stream) * JPRT (hotspot and jdk_lang test sets) * UTE (vm.quick-pcl.testlist, vm.defmeth.testlist)
Thanks,
Michael