Re: [eigen] Initial implementation of tensor support

[ Thread Index | Date Index | More Archives ]

Hi there,

Sorry for replying only so late, but I didn't notice your
email before I went on vacation.

A very cool contribution. Maybe I will stop using Boost.MultiArray
combined with Eigen Maps for my next project...

Boost's multi array feature is very useful indeed, but at least
when I last tried it, the performance was just not good enough
for any serious numerical work. And using Eigen Maps as an
overlay is not an option for my purposes.

I know eigen targets C++98, but do the eigen developers allow C++11
code in unsupported/?  I hope so, as C++11 seems truly necessary to
get support for arbitrary-rank tensors.

Is there a scientific need for dimensionality beyond the maximum
standardized Fortran array dimension?

Well, 7 does seem to be large enough for the vast majority of
cases, but if you can do it properly in the first place, why
not? Also, even if it's possible to implement this with fixed
index objects in C++98, note that this will result in a huge
amount of code duplication. And my fancy symmetry stuff could
not be implemented at compile time in that case, unless you
want to add specializations for all possible symmetries...


Mail converted by MHonArc 2.6.19+