Re: [eigen] compilation failed for not aligned tiny matrix

[ Thread Index | Date Index | More lists.tuxfamily.org/eigen Archives ]


2009/10/6 Gael Guennebaud <gael.guennebaud@xxxxxxxxx>:
>
> Hi,
>
> ok, here the bug is that computeInverse() expect a PlainMatrixType which in
> your case is equivalent to TinyRMatrix but without the DontAlign option. Yes
> that's a bug in Eigen.
>
> But now, I'm wondering why computeInverse is not a template function ? Think
> about submatrices, or, e.g., to perform: m.transpose().computeInverse(inv);
> - here m.transpose() is row major and so computeInverse will expect a row
> major matrix.

Yes, all these functions that take a ptr/ref to result should be
template in the result type, you're perfectly right.

>
> Moreover I'm wondering whether computeInverse() is still useful since we can
> make inv = a.inverse() works in place just like compute inverse.

Indeed, but that part (Inverse.h) is the trickiest of the current
solve API refactoring (btw, sorry for being slow) for reasons i'll
explain, so it's the part that i'll do last. The problems is that
these API changes go together with a reformulation of the problem of
"checking for existence of a solution" and the case of inverting tiny
matrices is special here.

Benoit



Mail converted by MHonArc 2.6.19+ http://listengine.tuxfamily.org/