Re: Boost Digest, Vol 7346, Issue 1

From: Alfredo Correa (alfredo.correa_at_[hidden])
Date: 2024-09-20 20:48:44

Hi Seth,

On Fri, Sep 20, 2024 at 5:00 AM <boost-request_at_[hidden]> wrote:

>
> Date: Thu, 19 Sep 2024 16:28:28 +0200
> From: Seth <bugs_at_[hidden]>
>
> On Thu, Sep 19, 2024, at 1:13 PM, Artyom Beilis via Boost wrote:
> > 1. You haven't mentioned even once OpenCV
> > [...]
> > 4. If you do implement many high performance algorithms this library by
> any
> > means shouldn't be header only.
> >
> > What is not clear to me is why should I use one over some existing
> solution
> > like OpenCV?
>
> All of the points raised are actually a single one, and seem to
> misunderstand the scope of the library? I suppose it should be seen as a
> successor to Boost MultiArray, leveraging modern language features and
> side-stepping some of the pitfalls.
>

yes, exactly.
You seem to understand exactly what I am talking about. (Also a big fan of
your Stackoverflow answers).
I just gave a long answer to Artyom's message in the list.

> That's not at all useless, and interop examples would be very valuable.
> However, not all code needs vectorization (beyond what the compiler will
> automatically do). I'm more interested in the relation to
> https://en.cppreference.com/w/cpp/container/mdspan - which seems to fill
> the same space: building block for flexibly indexed contiguous storage.
>

>From this section:
https://gitlab.com/correaa/boost-multi#substitutability-with-standard-vector-and-span

"Multi-dimensinal arrays can interoperate with C++23's non-owning mdspan.
Preliminarily, Multi's subarrays (arrays) can be converted (viewed as)
mdspan."
https://godbolt.org/z/aWW3vzfPj

Also see
https://gitlab.com/correaa/boost-multi#appendix-comparison-to-other-array-libraries-mdspan-boostmultiarray-etc

Thank you,
Alfredo