Mark Hoemmen
Mark Hoemmen
https://youtu.be/ok5l26Fs0LI?t=57
@ajpowelsnl Hi! Not sure how y'all want to do that, but cppreference does it this way: https://en.cppreference.com/w/cpp/container/span/deduction_guides
I think @alanw0 may want a fix for this too.
@tawiesn @ibaned If it helps, Tpetra has a (currently not parallel) `sort2` function, that sorts one array and applies the resulting permutation to an array of the same length. Tpetra...
@tawiesn If you come up with a good interface, let me know :-)
@gmackey Thanks! Just curious -- how does thrust::sort work when called on device? Does it work like a thread team collective, is it sequential, or does it launch a new...
@tawiesn See the following (now closed) issue for sorting of short arrays: https://github.com/trilinos/Trilinos/issues/887
Tpetra already has all this code -- you just need to stick KOKKOS_FUNCTION in front of stuff (I wouldn't do KOKKOS_INLINE_FUNCTION since it probably wouldn't make sense to inline general...
@gmackey Thanks! A team sort would be nice :-) I've prototyped some sort functions for arrays of fixed lengths (2, 3, 4, 8) in Trilinos/packages/tpetra/core/src/Tpetra_Details_shortSort.hpp. They use sort networks to...
@brian-kelley Have you considered using [CUB](https://docs.nvidia.com/cuda/cub/index.html) for the CUDA back-end? CUB should always come with your CUDA installation.