[geos-devel] JTS/GEOS performance - smart-ptr vs ref counting

[Norman Vine]

Martin Davis writes:
>
> Norman Vine
> > 
> > Martin Davis writes:
> > > 
> > > 
> > > That was pretty much my point - smart pointers alone don't produce 
> > > memory efficiency, they just make the code safer (and hopefully 
> > > easier).
> > 
> > Well they are a little faster then any garbage collection 
> > routine could ever be but you are right 'speed issues' aren't 
> > why they are used.
> > 
> > Has anyone profiled the code recently,  without profiling 
> > statistics optimization discussions are really just guesswork.
> 
> Something to note here is that the original profiling numbers given by
> strk were based on the buffer algorithm.  Buffer is by far the most
> complex algorithm in JTS, and inherently has to create lots of new
> coordinate sequences (amongst other things).  There may be no way to
> remove this requirement, in which case the only way to improve the speed
> is to increase the speed of memory allocation.

One thing that would probably speed things a bit is to inline the
trivial constructors and destructors :-)

It might be worth looking into special memory managers but AFAICT 
most of the 'coordinates' are actually stored in STL vectors and these 
are generally about as efficient as you can get at minimizing allocation
costs.

Norman