CGAL users discussion list

[Gärtner Bernd <>]

Hi Florian,
 
one minute sounds way too slow for 50 constraints, indeed. 
One obvious thing is -O2 -DNDEBUG (the latter is important to disable
some quite costly runtime checks). With 50 constraints, what is the
number of variables? Also, if you send me a program, I might suggest
some runtime improvements. For example, input type double is
preferrable over int since only the former triggers floating-point
filtering by default.
 
Best,
Bernd Gaertner.

________________________________

Von: Florian Thiel 
[mailto:]
Gesendet: Do 21.06.2007 17:43
An: 

Betreff: [cgal-discuss] Speed of qp solver / using simulation together with qp



Hi!

We're using CGAL in a university project for a simulation of falling
cuboids (in 2D, so the simulation uses rectangles only). This involves
quadratic optimization. For every point (we've got 9 per cuboid) a
bunch of constraints are generated. The optimization generally works
but is very slow. For two cubes (about 50 constraints) every step (the
optimization calculates the acceleration for every point) takes about
one minute to compute. We'd need about 20 cubes and  at least 1,000
steps for our simulations, so this is clearly way to slow. Are there
any common programming traps which make qp solving so slow or is this
the speed you should be expecting in a situation like ours? What are
the common optimizations one could use?

Another thought: At the moment we're including all the constraints in
the qp formula (like deformation and collision). Would it be possible
to let CGAL do these things? Can we create predicates which evaluate
to true for each simulation step (i.e. considering only results in qp
for which all predicates hold)?

Thanks in advance,
Florian
--
You are currently subscribed to cgal-discuss.
To unsubscribe or access the archives, go to
https://lists-sop.inria.fr/wws/info/cgal-discuss