rubinius?
rogerdpack
Good news:
it runs some things about as fast as 1.9 does, for me.
Bad news:
it blew up (size wise) with larger arrays.
http://groups.google.com/group/rubinius-dev/browse_thread/thread/c6a04503c3accb63
it also is still like 3x as slow as crystalizer.
http://groups.google.com/group/rubinius-dev/browse_thread/thread/de4472f7b4ec7778
Seems there's room for improvement all around us. Now if I could just
try out MacRuby...
-r
rogerdpack
> it blew up (size wise) with larger arrays.
>
> it also is still like 3x as slow as crystalizer.
Good news--seems I was running rubinius incorrectly. If you run it
"against methods only" it actually runs, on one test almost twice as
fast as crystalizer.
http://groups.google.com/group/rubinius-dev/browse_thread/thread/de4472f7b4ec7778
however on a different test it runs a tidge slower than crystalizer.
Also note that this is with an unoptimized crystalizer. If I add
inline math crystalizer is 3.5, crystalizer unoptimized 5.3, rubinius
is 2.6s, so they're pretty close. I bet a "super crystalizer" could
still beat it, but not by much, and since most of the bottlenecks are
probably in the stdlib at that point...working on the stdlib (or C
extensions) is probably more important, anyway :)
Another bench:
And on RBS' benchmarks/micro-benchmarks/bm_gc_array:
1.9.1: 58
rubinius: 26s
jruby straight: 18.3
jruby --server --fast: 12.5
I'm starting to like jruby and rubinius more and more.
Also, I wonder if Rubinius' GC isn't as good as jruby's?
-r