On Apr 28, 2011, at 12:33 PM, Christian Thalinger wrote: > On Apr 28, 2011, at 12:16 PM, Christian Thalinger wrote: >> On Apr 27, 2011, at 5:54 AM, Charles Oliver Nutter wrote: >>> I prepared this for someone else, but I thought folks here might be >>> interested in it too. >>> >>> This gist contains hotspot x86 (32-bit) assembly output for JRuby's >>> dynopt mode and invokedynamic (on a couple-week-old OS X OpenJDK >>> build). I haven't spent a lot of time investigating. >> >> I took a look at it. I used 64-bit x86 since the code is a bit smaller than >> with 32-bit. >> >> The code is almost identical but three things popped into my eye (the output >> is from PrintOptoAssembly): >> >> 1. The obvious one: the method handle call site guard: >> >> 1a4 B32: # B160 B33 <- B31 B149 B123 Freq: 0.499969 >> 1a4 movq R10, byte[int:>=0]<ciObject ident=770 PERM address=0xe99088> >> * # ptr >> 1ae movq R10, [R10 + #1576 (32-bit)] # ptr >> 1b5 movq R11, [R10 + #32 (8-bit)] # ptr >> 1b9 movq R8, java/lang/invoke/AdapterMethodHandle:exact * # ptr >> 1c3 cmpq R11, R8 # ptr >> 1c6 jne,u B160 P=0.000000 C=-1.000000 > > I did an experiment and hacked a version of HotSpot that does not emit the > call site guard. Since the fib benchmark is monomorphic this works and shows > that we just need to get rid of the guard to get on a par with dynopt > performance. The numbers are 32-bit x86.
I have now a patch that makes the command line switch tweaking superfluous and the default performance looks pretty good (see below, 32-bit x86). Charlie, what benchmark could I use for more real world application numbers? bench_string_ops.rb sounds and looks promising. With bench_string_ops.rb I see that dynopt performance isn't always better than "normal" (whatever normal is). I guess that is expected? -- Christian $ bin/jruby.sh --server -Xcompile.dynopt=false bench/bench_fib_recursive.rb 10 35 1.090000 0.000000 1.090000 ( 1.059000) 1.012000 0.000000 1.012000 ( 1.012000) 0.977000 0.000000 0.977000 ( 0.977000) 0.979000 0.000000 0.979000 ( 0.979000) 0.979000 0.000000 0.979000 ( 0.979000) 0.979000 0.000000 0.979000 ( 0.979000) 0.979000 0.000000 0.979000 ( 0.979000) 0.979000 0.000000 0.979000 ( 0.979000) 0.978000 0.000000 0.978000 ( 0.978000) 0.987000 0.000000 0.987000 ( 0.987000) $ bin/jruby.sh --server -Xcompile.dynopt=true bench/bench_fib_recursive.rb 10 35 1.038000 0.000000 1.038000 ( 1.005000) 0.763000 0.000000 0.763000 ( 0.763000) 0.763000 0.000000 0.763000 ( 0.763000) 0.763000 0.000000 0.763000 ( 0.763000) 0.763000 0.000000 0.763000 ( 0.763000) 0.765000 0.000000 0.765000 ( 0.765000) 0.764000 0.000000 0.764000 ( 0.764000) 0.763000 0.000000 0.763000 ( 0.763000) 0.772000 0.000000 0.772000 ( 0.772000) 0.762000 0.000000 0.762000 ( 0.762000) $ bin/jruby.sh --server -Xcompile.invokedynamic=true bench/bench_fib_recursive.rb 10 35 0.912000 0.000000 0.912000 ( 0.879000) 0.787000 0.000000 0.787000 ( 0.787000) 0.786000 0.000000 0.786000 ( 0.786000) 0.789000 0.000000 0.789000 ( 0.789000) 0.789000 0.000000 0.789000 ( 0.789000) 0.791000 0.000000 0.791000 ( 0.791000) 0.788000 0.000000 0.788000 ( 0.788000) 0.788000 0.000000 0.788000 ( 0.788000) 0.787000 0.000000 0.787000 ( 0.787000) 0.802000 0.000000 0.802000 ( 0.802000) _______________________________________________ mlvm-dev mailing list mlvm-dev@openjdk.java.net http://mail.openjdk.java.net/mailman/listinfo/mlvm-dev
