[PROJ] Vector/SIMD acceleration

[Even Rouault]

> https://eigen.tuxfamily.org/dox/group__CoeffwiseMathFunctions.html

Oh ok, the Array type, not the Vector type. That's a bit better.

But playing with that, I can see for example that vectorized sin() is only 
implemented for SSE2/AVX using the code from http://gruntthepeon.free.fr/
ssemath/ for single precision, not double that we need for PROJ. So by default 
Eigen with double type fallbacks to serialized evaluation of sin() for each 
member of the array.

We could probably provide an implementation using Sleef, but from what I see 
that would require to modify Eigen headers themselves.

-- 
Spatialys - Geospatial professional services
http://www.spatialys.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/proj/attachments/20200417/7ba8e086/attachment.html>