Subject: Re: [geometry] SIMD optimizations
From: Adam Wulkiewicz (adam.wulkiewicz_at_[hidden])
Date: 2016-01-28 18:23:25
Hi Vladimir,
Vladimir Voznesensky wrote:
> Every node of rtree concludes information about several geometrical
> objects or boxes bounding such objects. Is it possible to zip
> (transpose) such information into SIMD vectors (using libsimdpp, for
> example) to use fast SIMD instructions without messing up too many
> stuff in boost::geometry?
How would you like to zip/transpose it exactly? Instead of having a
container of e.g. points, to have D containers of coordinates, where D
is the dimension?
And what would you like to do with them? Would they still be used as
coordinates?
Would you like to change only the nodes contents, provide some
lower-level algorithms e.g. testing for spatial overlap or would you
like to use entirely different balancing and querying algorithms?
Regards,
Adam