Re: [eigen] Ideas for better documenting coeffwise math functions

[ Thread Index | Date Index | More Archives ]

Hi Gael,

I think it's a good format, but it needs some information about accuracy. Moreover, there should be some mention of the fact that versions for scalar, AVX and SSE differ. Indeed, sometimes they even use different algorithms and it would be nice to have some indication about relative speed and accuracy between them.


On Wed, Jun 1, 2016 at 2:51 PM, Gael Guennebaud <gael.guennebaud@xxxxxxxxx> wrote:
Hi list,

as we are supporting more and more coefficient-wise math functions, we need a good way to structure and document them. To this end, I've started to sketch a table summarising the essential information about each of them:

- available API ( ;foo(a) ; a.cwiseFoo())
- what it does in a few words/equation
- what is used under the hood, i.e., which STL function (with hyperlinks to cppreference)?, built-in implementation? (and if so, which scalar types are supported?) Which function has to be provided for custom scalar type?
- SIMD status: supported instruction set, supported scalar types, accuracy (i.e., disabled in non-fast math mode).

You can see a start here:

I've included a very few functions just to see how it looks like, and to get feedbacks before making it complete.

 - Would you find such a table useful?
 - Does it provides too many or too few information?
 - Is the information clearly presented?
 - Any general comments on this aspect?

And most importantly, feel free to volunteer to complete this task ;)


Mail converted by MHonArc 2.6.19+