search

arvindm95 / unladen-swallow

Automatically exported from code.google.com/p/unladen-swallow
Other
0 stars 0 forks source link

2% regression in slowpickle benchmark #53

Closed GoogleCodeExporter closed 8 years ago

GoogleCodeExporter commented 8 years ago 
The slowpickle benchmark is showing a 2% performance regression against our
2009Q1 release with -L2. Other benchmarks are showing an improvement of 5-10%.

2009Q1 vs trunk@r610 (-L2; gcc 4.3.1; 64-bit)
perf.py options: --args=,-L2 -r

slowpickle:
Min: 0.736 -> 0.754: 2.36% slower
Avg: 0.746 -> 0.756: 1.28% slower
Significant (t=-9.445019, a=0.95)

Original issue reported on code.google.com by collinw on 8 Jun 2009 at 9:41

GoogleCodeExporter commented 8 years ago 
Interesting:

2009Q1 vs trunk@r669 (-j always -O2; gcc 4.0.1; 32-bit)
perf.py options: --args ",-j always -O2"

slowpickle:
Min: 0.736 -> 0.625: 17.71% faster
Avg: 0.739 -> 0.627: 17.83% faster
Significant (t=386.726331, a=0.95)
Stddev: 0.002 -> 0.002: 41.70% smaller

I need to see if this is a difference between 32- and 64-bit codegen or if the 
original 
observations were measurement error, or what.

Original comment by collinw on 24 Jun 2009 at 11:23

GoogleCodeExporter commented 8 years ago 
I can't reproduce this in the 64-bit build.

Original comment by collinw on 30 Jun 2009 at 1:25