some windows benchmarks

[rogerdpack]

Here's a probably unscientific hack at comparing mingw, VS 2008,
jruby--on windows.

>ruby_19\bin\ruby -v
ruby 1.9.0 (2008-11-22 revision 20317) [i386-mingw32]

>gcc -v
gcc version 3.4.5 (mingw-vista special r3)

and the VC2008 downloaded is version:

>ruby-1.9.1p0-mswin32_90\bin\ruby -v
ruby 1.9.1p0 (2009-01-30 revision 21907) [i386-mswin32_90]

and >jruby -v
jruby 1.1.6 (ruby 1.8.6 patchlevel 114) (2008-12-17 rev 8388) [x86-
java]

ran each test 3 times, took the fastest. Results seemed reasonably
steady.

Note that I used jruby without any command line parameters, and that
the KRI's are without a patched GC, so some improvement is possible.
Note also that the gcc version is a little old, so mingw could
possibly improve there.

0.00 means it failed

ming vc9 jruby
0.12 0.18 3.00 Startup
0.02 0.02 0.04 core-features/bm_app_factorial.rb
0.06 0.08 0.14 core-features/bm_app_factorial.rb
0.47 0.44 0.00 core-features/bm_app_factorial.rb
0.00 0.00 0.01 core-features/bm_app_fib.rb
1.08 1.27 1.89 core-features/bm_app_fib.rb
12.2 15.5 21.3 core-features/bm_app_fib.rb
20.7 23.9 4.41 core-features/bm_app_raise.rb
0.06 0.09 0.10 core-features/bm_app_tak.rb
0.25 0.34 0.36 core-features/bm_app_tak.rb
0.81 0.97 1.19 core-features/bm_app_tak.rb
2.70 3.33 4.22 core-features/bm_app_tarai.rb
3.30 4.06 5.09 core-features/bm_app_tarai.rb
3.97 4.88 6.04 core-features/bm_app_tarai.rb
3.09 4.05 8.13 core-features/bm_loop_times.rb
totals
48.87 59.14 55.94
totals without bm_app_raise [i.e. everything except speed of raising
an exception]
28.03 35.06 48.52

Sorry it's not a comparison with a newer 1.9 version for the mingw
version, but I'd imagine it's pretty close to 1.9.1
Thanks!

-=r

[Charles Oliver Nutter]

rogerdpack wrote:
> Note that I used jruby without any command line parameters, and that
> the KRI's are without a patched GC, so some improvement is possible.
> Note also that the gcc version is a little old, so mingw could
> possibly improve there.

You probably want --server in JRuby. It should be quite a bit better.
And hopefully this is with a Java 1.6/6 version or better.

- Charlie

[Charles Oliver Nutter]

Also, if you're interested in trying a nightly, there's a --fast option
that turns on a few sometimes-unsafe optimizations. For the benchmarks
you listed, it will be another noticeable improvement.

- Charlie