CGAL users discussion list

[Sylvain Pion <>]

Maarten Moesen wrote:
> Dear CGAL Programmers,
>
> I'm currently using CGAL 3.2 and I'm very happy with it. But I noticed
> that my CGAL processes when they're running use increasing amounts of
> memory, and do not release their memory although I do delete() etc.
> properly.
>
> I wrote the following little test program:
[...]
> The output for the inexact constructions kernel is OK for me:
>
> Start of program: Virtual process size: 21598208 Resident process size:
> 9842688
>
> doubles allocated: Virtual process size: 37605376 Resident process size:
> 25890816
> doubles destroyed: Virtual process size: 21602304 Resident process size:
> 9887744
>
> Points allocated: Virtual process size: 69603328 Resident process size:
> 57888768
> Points destroyed: Virtual process size: 21602304 Resident process size:
> 9887744
>
> But all other kernels yield something like: (here Cartesian<double>)
>
> Start of program: Virtual process size: 21594112 Resident process size:
> 9842688
>
> doubles allocated: Virtual process size: 37601280 Resident process size:
> 25890816
> doubles destroyed: Virtual process size: 21598208 Resident process size:
> 9887744
>
> Points allocated: Virtual process size: 93536256 Resident process size:
> 81891328
> Points destroyed: Virtual process size: 85532672 Resident process size:
> 73887744
>
> How come that this last value is so high? What am I doing wrong? 
> I would expect that it goes back to normal as with the doubles and the
> inexact constructions kernel. (This is the reason why my processes grow
> so large... )
>
> BTW: I'm using CGAL 3.2 and g++ 4.0.3 on a linux system 
> Program compiled with -O2 and -DNDEBUG
> The Memory sizer seems to work fine, it gives about the same as top does
> and the memory-widgets on my pc.

This is to be expected, unfortunately.

We have something in the works (Lazy_kernel) which will significantly
reduce the memory usage of the exact constructions kernel, but it's not
ready for prime time yet.

There will always be a difference with the pure double kernel though,
for exact constructions.


Here is what I get with the stuff to come :

Start of program: Virtual process size: 13541376 Resident process size: 
4448256

doubles allocated: Virtual process size: 29548544 Resident process size: 
20516864
doubles destroyed: Virtual process size: 13545472 Resident process size: 
4513792

Points allocated: Virtual process size: 181452800 Resident process size: 
172515328
Points destroyed: Virtual process size: 13565952 Resident process size: 
4644864


Compared to the current EPECK :

Start of program: Virtual process size: 13537280 Resident process size: 
4452352

doubles allocated: Virtual process size: 29544448 Resident process size: 
20512768
doubles destroyed: Virtual process size: 13541376 Resident process size: 
4509696

Points allocated: Virtual process size: 253480960 Resident process size: 
244510720
Points destroyed: Virtual process size: 205479936 Resident process size: 
196509696


--
Sylvain