On Sat, Sep 12, 2009 at 3:21 AM, John Rose <john.r...@sun.com> wrote: > I just pushed that fix. I also pushed (what I hope are) the fixes for > the last GC bugs. Please let me know what happens.
Updated numbers for Surinx, new build as of a few minutes ago: ~/projects/surinx ➔ jruby --server -J-XX:+UnlockExperimentalVMOptions -J-XX:+EnableInvokeDynamic bin/surinx examples/fib.sx 0.091 0.068 0.066 0.067 0.067 0.066 0.066 0.065 0.069 0.066 I don't have Remi's backport handy, but it's obviously very close. JRuby numbers are nearly as fast as non-indy now too: ~/projects/jruby ➔ jruby --server bench/bench_fib_recursive.rb 100 0.473000 0.000000 0.473000 ( 0.421000) 0.241000 0.000000 0.241000 ( 0.241000) 0.239000 0.000000 0.239000 ( 0.239000) 0.242000 0.000000 0.242000 ( 0.242000) 0.239000 0.000000 0.239000 ( 0.239000) 0.236000 0.000000 0.236000 ( 0.236000) 0.236000 0.000000 0.236000 ( 0.236000) ~/projects/jruby ➔ jruby --server -J-XX:+UnlockExperimentalVMOptions -J-XX:+EnableInvokeDynamic -J-XX:InlineSmallCode=1500 -J-XX:MaxInlineSize=150 bench/bench_fib_recursive.rb 100 0.444000 0.000000 0.444000 ( 0.378000) 0.236000 0.000000 0.236000 ( 0.236000) 0.231000 0.000000 0.231000 ( 0.231000) 0.234000 0.000000 0.234000 ( 0.234000) 0.233000 0.000000 0.233000 ( 0.233000) 0.235000 0.000000 0.235000 ( 0.235000) 0.235000 0.000000 0.235000 ( 0.236000) If I remove some special-case code for math operations, indy actually improves, beating stock JRuby: ~/projects/jruby ➔ jruby --server -J-XX:+UnlockExperimentalVMOptions -J-XX:+EnableInvokeDynamic -J-XX:InlineSmallCode=1500 -J-XX:MaxInlineSize=150 bench/bench_fib_recursive.rb 100 0.429000 0.000000 0.429000 ( 0.365000) 0.237000 0.000000 0.237000 ( 0.237000) 0.231000 0.000000 0.231000 ( 0.230000) 0.231000 0.000000 0.231000 ( 0.230000) 0.229000 0.000000 0.229000 ( 0.230000) 0.229000 0.000000 0.229000 ( 0.229000) 0.227000 0.000000 0.227000 ( 0.227000) 0.226000 0.000000 0.226000 ( 0.227000) It's time for that blog post I've been promising about "how to" and "why it's awesome". - Charlie _______________________________________________ mlvm-dev mailing list mlvm-dev@openjdk.java.net http://mail.openjdk.java.net/mailman/listinfo/mlvm-dev