Uploaded image for project: 'JDK'
  1. JDK
  2. JDK-8051725

Improve expansion of Conv2B nodes in the middle-end

XMLWordPrintable

    • Icon: Enhancement Enhancement
    • Resolution: Fixed
    • Icon: P5 P5
    • 21
    • 9, 10
    • hotspot
    • b25
    • x86

      Originally found here:
      http://stackoverflow.com/a/24899011/2613885

      The benchmark from http://cr.openjdk.java.net/~shade/8051725/ yields:

      o.o.IfSetneBench.test0 2.286 0.011 ns/op
      o.o.IfSetneBench.test1 1.010 0.003 ns/op
      o.o.IfSetneBench.test1neg 1.010 0.003 ns/op
      o.o.IfSetneBench.test42 1.011 0.005 ns/op
      o.o.IfSetneBench.test42neg 1.008 0.002 ns/op

      Notice test0 is slower than any other variant. This is because it indeed uses the weird SETNE optimization:
       http://cr.openjdk.java.net/~shade/8051725/test0.perfasm

      ...while other tests are doing the plain cmp:
       http://cr.openjdk.java.net/~shade/8051725/test1.perfasm

            jkarthikeyan Jasmine Karthikeyan
            shade Aleksey Shipilev
            Votes:
            0 Vote for this issue
            Watchers:
            8 Start watching this issue

              Created:
              Updated:
              Resolved: