Andrey Prokopenko
Andrey Prokopenko
Here are the results on Frontier (`attach` meaning pair of value and index, `iota` meaning storing just integers and relying on indexable getter) for the benchmark (code is [here](https://github.com/aprokop/ArborX/blob/243f8820571092d1391e49bcf596b8081fb23d43/benchmarks/develop/develop.cpp)): ```...
> If they need better performance, they use the advanced API and they can expect substantial improvement (not a 5% difference). I think it's really is case specific. It would...
> My worry is that someone does a performance comparison between ArborX with the simple interface and another library. Yeah, that's my worry too. It's a tradeoff between making the...
@dalg24 What are your thoughts on this PR? Particularly, do you think it's a reach to provide second argument as `int` instead of making a user to spell out `Iota`?
TODO: make a new construct a free function (outside of indexes)
I simplified the Caliper code a bit. The specific options can be passed through `CALI_CONFIG`.
> Not sure it is a good idea What do you see as the difference between `Kokkos::View` and `std::vector` specializations?
> What would we want to do about host device annotations on the access functions? Not use them. Everything will be `HostSpace`, so we won't need annotations.