Re: [eigen] Memory Map a SparseMatrix

["Gael Guennebaud" <gael.guennebaud@xxxxxxxxx>]

Hi,

indeed, it is currently not possible to map external data as a sparse
object but that's on the TODO list.

Unlike the Map class for dense matrix, here I think I'll simply reuse
SparseMatrix and add a flag to prevent deletion of the data, etc...

Cheers,
Gael.

On Mon, Jan 12, 2009 at 11:01 AM, Jens Mueller <jens.k.mueller@xxxxxx> wrote:
> Hi,
>
> I'm using the current SVN revision 909800 trying to use of the started
> sparse matrix implementation in Eigen.
> Because I will reuse the data of some sparse matrices very often, I want
> to store it inside a file. I already have some C code that uses memory
> mapped files. I came across the Map function in SparseArray. I suppose
> there should also be a Map function passing a pointer to the outer
> indices and a SparseArray for the values with its appropriate indices
> and the needed sizes.
> The trouble is you can't pass pointers that weren't allocated using
> new[] because SparseArray/SparseMatrix uses delete[]. Suppose I pass a
> pointer not allocated using new[] I run into troubles. I suppose this is
> also the case for normal matrices.
> In my case I'm using Boost.Interprocess to do like this:
> // mmap file
> file_mapping m_file(filename.c_str(), read_write);
> mapped_region region(m_file, read_write);
>
> and now I like to use the Map function with pointers pointing as needed
> to memory mapped region like this:
> MappedMatrix m = MappedMatrix::Map(region.get_address(), ...)
>
> I'm very unsure whether this is a proper way to achieve storing and
> reading matrices to/from files. Maybe other ways to do this are more
> appropriate, e.g. object serialization. My approach is to store the data
> in a memory mapped file.
> How do you elegantly store an Eigen matrix in a file? Maybe I'm misusing
> the Map function for this purpose.
>
> Regards,
> Jens
>
>
>