CGAL users discussion list

[Stephen Wong <>]

Hi Peter,

The description in the File I/O section of the User Manual is excellent, and 
I should have read it carefully before.  It restates what you have in this 
thread.  I agree with you when you mention that that section in the manual 
can be overlooked.  Perhaps a separate heading with "Kernel Selection" can 
provide more insight to the user.

Yes, I have used NDEBUG and O2.

Stephen

-----Original Message-----
From: Peter Hachenberger 
[mailto:]
Sent: February-22-08 11:39 AM
To: 

Subject: RE: [cgal-discuss] [3d_boolean_ops] Most Efficient Rational Kernel

Hi Stephen,

I introduced these new items, because it was clear that filtered kernels
are inefficient with the old ones. The new ones allow efficient
filtering. Since I mostly wanted to speed up Cartesian kernels and
therefore allow handling of double coordinates, I write about it at a
seemingly strange place of the documentation. It's in the user manual
under File I/O.

BTW, do you compile with -DNDEBUG and -O2? That usually gives a big
efficiency boost.

Peter

On Fri, 2008-02-22 at 11:19 -0800, Stephen Wong wrote:
> Hi Peter,
>
> I didn't use SNC_indexed_items before.  I am noticing about 30-40% 
> improvement in computation time (30% faster) after using 
> Exact_constructions_exact_predicates and SNC_indexed_items!
>
> What does the indexed items do?  It's not mentioned in the documentation - 
> in fact, the current documentation suggests that you should stay with the 
> default, SNC_items.
>
> Thanks for your excellent analysis.
>
> Stephen
>
> -----Original Message-----
> From: Peter Hachenberger 
> [mailto:]
> Sent: February-22-08 4:14 AM
> To: 
> 
> Subject: Re: [cgal-discuss] [3d_boolean_ops] Most Efficient Rational Kernel
>
> Hi Stephen,
>
> I performed some tests. Assuming you have gmp but no leda, the
> Exact_predicate_exact_constructions_kernel is up to double as fast as
> all the others. Don't forget to use the SNC_indexed_items with that
> kernel. Also all the other kernels (besides the extended kernels) profit
> from the SNC_indexed_items. The second fastest seems to be
> Simple_cartesian<Gmpq>. The homogeneous kernels were in general slower
> than the Cartesians. That would be different with LEDA.
> Homogeneous<leda_integer> is very fast.
>
> Peter
>
> On Mon, 2008-02-18 at 11:38 -0800, Stephen Wong wrote:
> > Hi there,
> >
> >
> >
> > I am wondering what is the most computationally-efficient kernel that
> > is allowable in Nef Polyhedra in 3D (Boolean operations), in the set
> > of rational numbers Q.  Right now I am using Cartesian<Gmpq>, a
> > variant of Exact_predicate_exact_constructions_kernel.  I am wondering
> > if there is a Kernel (than I have stated) that performs more
> > efficiently.  If not, is there a more efficient number type than Gmpq
> > that is compatible with 3D Nef Polyhedra?
> >
> >
> >
> > On the same topic of discussion, if I were to use kernels representing
> > Z, which is the most efficient kernel?
> >
> >
> >
> > Thanks,
> >
> >
> >
> > Stephen
> >
> >
> --
> You are currently subscribed to cgal-discuss.
> To unsubscribe or access the archives, go to
> https://lists-sop.inria.fr/wws/info/cgal-discuss
>
--
You are currently subscribed to cgal-discuss.
To unsubscribe or access the archives, go to
https://lists-sop.inria.fr/wws/info/cgal-discuss