Christian Trott
Christian Trott
As Mark said: we clarified that in the last revision. Regarding the iterate over the whole array as a single dimension vector, we explicitly decided against that. You can still...
It is certainly possible that someone could propose such a convenience thing, but it gets fairly iffy fairly fast in terms of corner cases which would lead to subtle bugs....
Considering where the design process is and the time limit for getting into C++23 we would need to do this via a national body comment.
Note this only became a problem due to the attempt to undo multiple nested scalings, if we wouldn't try to undo it you are back to the () case. Furthermore,...
Adding the unlikely does make it slower in cases where you trigger that path (i.e. run this thing in a for loop instead a parallel for).
But generally speaking folks do many more copies of views inside a kernel than outside.
Note to us: we need a performance test for this as well. Both inside kernel and outside.
We will assign this to someone.
Hi, thanks for the change, we don't have a super active maintainer right now for this. Anyway: if you wanted to check for Kokkos Version we could get rid of...
Again it depends. If you are still stuck on Kokkos 3.7 I think this is fine (and it will work with 4.0 and likely 4.1). But if you are anyway...