Re: [eigen] Large matrix problem

[Manoj Rajagopalan <rmanoj@xxxxxxxxx>]

I've seen a similar problem in my machine with is running a 32-bit OS. Unless 
you have a 64-bit OS, new[] /malloc() won't be able to allocate so much 
memory. You are requesting 4+eps GB of RAM but 32-bit OS can only address 4GB 
and some of that will be allocated to OS activities so you can only request 
an effective 3GB or so at most.

cheers!
Manoj



On Thursday 01 July 2010 02:19:05 am Jian Yang wrote:
> Hi everyone,
>
> I'm trying to create a 14000 x 14000*22 matrix using the following code
>
> MatrixXf a=MatrixXf::Zero(14000, 14000*22);
>
> The program stopped with the error: "Segmentation fault".
>
> I have 64G RAM in my machine. I think I have enough memory to create
> such matrix.
>
> Anyone knows what's going on?
>
> Cheers
>
> Jian
>
> -----Original Message-----
> From: Listengine [mailto:listengine@xxxxxxxxxxxxxxxxx] On Behalf Of
> Benoit Jacob
> Sent: Sunday, 27 June 2010 4:05 AM
> To: eigen@xxxxxxxxxxxxxxxxxxx
> Subject: Re: [eigen] [patch] LDLt decomposition with rank-deficient
> matrices
>
> Hey Ben,
> Thanks for the patches and sorry that I don't have time to look at
> them today: we are currently in an incredible rush to get beta1 out
> the door, and that means a lot of documentation work. I think that
> your stuff is OK to apply post-beta1 anyway. Just wanted to let you
> know why we might have long response times these days.
> Benoit
>
> 2010/6/26 Ben Goodrich <bgokgm@xxxxxxxxxxxxxx>:
> > Hi,
> >
> > On Fri, Jun 25, 2010 at 1:50 AM, Ben Goodrich <bgokgm@xxxxxxxxxxxxxx>
>
> wrote:
> >>>> 5) I copied-and-pasted a block inside /test/cholesky.cpp and
>
> exercised
>
> >>>> the pivot=false option. It seems to work when you do ./check.sh
> >>>> cholesky. I did some other tests locally with singular matrices,
>
> but
>
> >>>> /test/cholesky.cpp does not seem to have any tests with singular
> >>>> matrices, so maybe some should be added?
> >>>
> >>> why not.
> >>
> >> I have not added the singular tests yet, but I can do that soon.
> >
> > This patch does so. It seems to work okay with both the Pivoting and
> > NoPivoting options from the previous patch.
> >
> >>> You should also make the solve function skips the transpositions
>
> when
>
> >>> no pivoting has been computed.
> >>
> >> I have not done this yet either. What did you decide about making a
> >> new class versus putting a flag in the class definition?
> >
> > Gael?
> >
> > Thanks,
> > Ben