Rakudo spectest time and parrot GCs

Patrick R. Michaud pmichaud at pobox.com
Thu May 5 12:15:11 UTC 2011


On Thu, May 05, 2011 at 08:10:18AM +1000, Vasily Chekalkin wrote:
> Can you try parrot 3.3 with attached patch?

The patch seems to work wonders on my desktop, generally a 19%
or better improvement over 2011.01:

Dell Dimension 9150, 4GB RAM, 64-bit, Kubuntu 11.04 x86_64
Version                    T1     T2     T3     T4    Fastest   vs 2011.01
--------------------------------------------------------------------------
rakbench-01 (core.pm):
  2011.01/3.0 (ms2)       212    212    219    215      212       100.0%
  2011.04/3.3 (gms)       291    292    291    291      291       135.3%
  2011.04/3.3-p1 (gms)    174    174    176    175      174        80.9%

rakbench-02 (sin.t):
  2011.01/3.0 (ms2)      38.0   38.8   38.1   38.9     38.0       100.0%
  2011.04/3.3 (gms)      47.6   47.2   47.2   47.2     47.2       124.2%
  2011.04/3.3-p1 (gms)   40.9   41.0   40.9   40.9     40.9       107.6

rakbench-03 (atan2.t):
  2011.01/3.0 (ms2)      37.2   37.4   37.2   37.1     37.1       100.0%
  2011.04/3.3 (gms)      33.6   33.6   33.5   33.3     33.3        89.8%
  2011.04/3.3-p1 (gms)   30.0   29.9   30.4   30.4     29.9        80.6%

rakbench-04 (rx.t):
  2011.01/3.0 (ms2)     155.7  152.7  154.3  153.7    152.7       100.0%
  2011.04/3.3 (gms)     123.1  122.6  122.6  122.3    122.3        80.1%
  2011.04/3.3-p1 (gms)   96.4   97.2   96.1   96.6     96.1        62.9%

I'm not sure why we aren't seeing a big improvement in the sin.t test.
I'm running a spectest on the 2011.04/3.3-p1 install now, just to verify 
that the patch doesn't break something else.  I may also do a fresh
round of tests, just as confirmation.

> My main suspition that we have quite different memory layouts between
> _compiled_ parrot.

I suspect you're correct; that would explain a lot of what I've been
seeing in these benchmark results.

Pm


More information about the parrot-dev mailing list