[pypy-dev] cppyy performance numbers (was: ype object 'BlackholeInterpreter' has no attribute 'bhimpl_direct_ptradd')

Carl Friedrich Bolz cfbolz at gmx.de
Fri Nov 26 16:31:44 CET 2010


Hi Wim, hi all,

On 11/25/2010 01:32 AM, wlavrijsen at lbl.gov wrote:
> so I'm revising my numbers after finding out that I was using a debug
> version
> of ROOT/Reflex ...
>
> PyROOT: 48.6
> PyCintex: 50.2
> pypy-c: 5.5
> C++: 0.05

I did some benchmarks with the new fast path that I just added, it gives 
the following numbers:

pypy-c without fast path: 5.9 (my laptop seems a bit slower than yours)
pypy-c with fast path: 0.22
C++: 0.05

So Anto's and my optimization helped a lot, we got another factor of 25 
faster. A bit more than a factor of 4 slower than C++, and it should 
work for many (non-static) method and constructor calls.

>
> Note that although PyROOT speeds up only marginally in opt mode, pypy-c has
> a huge speedup. For PyROOT, this is b/c most of the time is spent in
> CPython
> through API calls and that was already in opt mode. For pypy-c, this is
> telling me quite clearly that I've got to work some on capi.py. ;)
>
> Of course, the factor 6 that I saw originally (v.s. 2 during the sprint) is
> now a factor 10 almost ...

Are you sure that the units are the same? Because if yes, it is a factor 
of 100, not 10.

Cheers,

Carl Friedrich

P.S.: For the people missing context, see here:
http://morepypy.blogspot.com/2010/07/cern-sprint-report-wrapping-c-libraries.html



More information about the Pypy-dev mailing list