Re: [eigen] Indexes: why signed instead of unsigned?

[ Thread Index | Date Index | More Archives ]

I've started doing these changes. I think that we only need to change
runtime indices: for the template parameter indices, int should be
good enough. That limits the amount of changes needed. See my current
diff, in case you want to pick it up.

Andrea: these changes will make it possible (even easy) to later add
the ability for Sparse matrices to have a configurable Index type; but
i'm not going to do this change myself as I have enough to do :)


2010/5/12 Andrea Arteaga <yo.eres@xxxxxxxxx>:
> Just a word from my recent experience about sparse matrices: I think it
> would be important to let the user decide the type of the indices. For dense
> matrices it doesn't matter if the indices are signed or unsigned, 32 or 64
> bit, since the data structure does not contain data of this type (except for
> two numbers for rows and cols); but sparse matrices have one array for data,
> one for row indices and one for "column indices" (I'm talking about CRS
> storage), i.e. two arrays are involved in this decision.
> I'm writing a library for interchanging data between C++ programs and
> Matlab, and I'm providing support for Eigen and Boost::ublas. This kind of
> problems are common in my work: having a template parameter to decide the
> type of the indices would be very useful.
> Andrea Arteaga

Attachment: diff
Description: Binary data

Mail converted by MHonArc 2.6.19+