[Python-Dev] progress: compiling python2.5 under msys (specifically but not exclusively under wine) with msvcr80

Roumen Petrov bugtrack at roumenpetrov.info
Sun Jan 25 22:05:31 CET 2009


Roumen Petrov wrote:
> Cesare Di Mauro wrote:
>> Have you made some benchmarks like pystone?

Its seems to me version 2.6.1 is not optimized build so I 
remove(uninstall) it.

I repeat the pystone tests with an optimized GCC(mingw32) build.

- python-trunk-GCC(mingw32, local, native, optimized)
--  shell=cmd.exe
35453,3; 35700,4; 35747,3; 35615,5; 35632,3; 35661,8; 35547,1
   average=35622,5
   deviation=98,0
-- shell=bash.exe(msys)
36002,1; 35884,4; 35961,7; 35859,5; 35997,3; 36062,9; 35747,1
   average=35930,7
   deviation=107,2

- python-2.6-MSVC
--  shell=cmd.exe
35891,3; 35827,9; 35791,3; 35901,7; 35876,5; 36081,1; 36149,2
   average=35931,3
   deviation=132,7
-- shell=bash.exe(msys)
35532,9; 35621,1; 35526,8; 35639,4; 35671,2; 35702,4; 35633,0;
   average=35618,1
   deviation=66,1

I don't have idea why performance of official python 2.6 goes down(see 
previous results below). It is same PC. Every tested program load own files.

The result show unexpected behaviour:
- the MSVC build is faster by ~0.9% if it is run under cmd.exe then msys 
bash;
- the GCC build is faster by ~0.9% if it is run under msys bash.

Otherwise results lock similar but note that builds use different source 
base and in this case we may can't compare.


The old results:
> There is result from pystone test test run an old PC (NT 5.1):
> - 2.6(official build):
>   42194,6; 42302,4; 41990,8; 42658,0; 42660,6; 42770,1
>   average=42429,4
>   deviation=311,6
> - 2.6.1(official build):
>   35612,1; 35778,8; 35666,7; 35697,9; 35514,9; 35654,0
>   average=35654,1
>   deviation=88,1
> - trunk(my mingw based build):
>   35256,7; 35272,5; 35247,2; 35270,7; 35225,6; 35233,5
>   average=35251,0
>   deviation=19,2
> 
> There is problem with python performance between 2.6 and 2.6.1 ~ 19% :(.
> Also the test for GCC-mingw is not with same source base.
> 
> Roumen

Roumen


More information about the Python-Dev mailing list