Subject: Re: [ublas] Numeric traits for ublas bounded_vector and matrix
From: Rutger ter Borg (rutger_at_[hidden])
Date: 2010-02-23 08:06:14
Jesse Perla wrote:
> Ignoring my glaring misuse of ublas....
> While all of the stride information is useful, I still think it is
> necessary to have information on the sizes of the static allocations used
> by the vector/matrix
> This is the main reason I am using these metafunctions. My classes have
> huge numbers of small vectors/matrices which would likely kill performance
> prevent loop unrolling if I didn't have information on the static sizes.
> In the end, I also do a runtime assertion that they have the correct
Cool stuff. What I think would be really cool, is a BLAS-backend that does
compile-time unrolling / that is specialized for static-sized matrices and
vectors. Then you can write identical code for the dynamic/static-sized
cases, but still have the performance benefit.