On Mon, Jun 27, 2022 at 10:23 PM Hannu Krosing <hannuk@google.com> wrote:
>
> > Another thought: for non-x86 platforms, the SIMD nodes degenerate to
> > "simple loop", and looping over up to 32 elements is not great
> > (although possibly okay). We could do binary search, but that has bad
> > branch prediction.
>
> I am not sure that for relevant non-x86 platforms SIMD / vector
> instructions would not be used (though it would be a good idea to
> verify)
By that logic, we can also dispense with intrinsics on x86 because the
compiler will autovectorize there too (if I understand your claim
correctly). I'm not quite convinced of that in this case.
> I would definitely test before assuming binary search is better.
I wasn't very clear in my language, but I did reject binary search as
having bad branch prediction.
--
John Naylor
EDB: http://www.enterprisedb.com