Suggest Deprecating Current JIT System

Nick Glencross nick.glencross at gmail.com
Wed Sep 2 13:48:28 UTC 2009


Hi all,

On Wed, Sep 2, 2009 at 1:56 PM, Andy Dougherty <doughera at lafayette.edu> wrote:
>
> Here are a few random observations:
>
> Current JIT:
>    Can anyone provide reasonable benchmarks for whether this actually
>    provides any significant performance boost?  (Obviously, this has
>    to be done by someone on a platform where it currently works.)


I think the best JIT success story has been the PIR MD5 library because:

  * It only uses simple instructions which pretty much all JIT
  * Does lots of looping so that it gets lots of payback from the JIT'ing

It's not representative of real code I know, but shows the current JIT
code shining. Gives at least a 10-fold increase, even over the fast
core.

(I'm not suggesting sticking with the current JIT implementation)

Cheers,

Nick


nickg at jenny:~/src/parrot$ time ./parrot examples/library/md5sum.pir
~/Desktop/mono-139174.tar
2d5b5571ac09f6f82e09a54635a56a11	/home/nickg/Desktop/mono-139174.tar

real	2m41.270s
user	2m39.822s
sys	0m1.128s


nickg at jenny:~/src/parrot$ time ./parrot --runcore=fast
examples/library/md5sum.pir ~/Desktop/mono-139174.tar
2d5b5571ac09f6f82e09a54635a56a11	/home/nickg/Desktop/mono-139174.tar

real	1m47.448s
user	1m46.111s
sys	0m1.084s


nickg at jenny:~/src/parrot$ time ./parrot --runcore=jit
examples/library/md5sum.pir ~/Desktop/mono-139174.tar
2d5b5571ac09f6f82e09a54635a56a11	/home/nickg/Desktop/mono-139174.tar

real	0m10.231s
user	0m8.993s
sys	0m1.204s


More information about the parrot-dev mailing list