Daniel Arndt

Even better! Let me update accordingly then.

It appears that this feature isn't available with the compiler drops on the Argonne tests beds yet. So we'll have to wait some. > Implementations can (and [do](https://github.com/intel/llvm/blob/c1c918482cff323ec26f13767a0b6ab6d844a2b6/sycl/include/sycl/handler.hpp#L367-L379)) implement parallel_for(range)...

> We cannot directly template with policies as they are functions that take the iteration space in argument and we need to set it internally (the length of the view)....

> I wanted to talk about `Kokkos::Experimental::copy` of (std) algorithms. Ah, that would create a dependency from `core` to `algorithms` which might be problematic. > The call for *VectorRange is...

@tcclevenger Can you please have a look?

> I also can't remember, but is deep-copying between devices possible yet? See 49a7ea6

> Is this expected that `deep_copy()` cannot correctly copy to non-default device without passing an exespace? Anything that doesn't have an execution space instance is assumed to be executed on...

Broken in https://github.com/kokkos/kokkos/commit/4d8629f26c should be `cudaFuncSetAttribute` instead of `cudaFuncSetAttributes`.

A fix with test should look somewhat like ```C++ diff --git a/core/src/Cuda/Kokkos_Cuda_Instance.hpp b/core/src/Cuda/Kokkos_Cuda_Instance.hpp index db764fb1b..660ef308b 100644 --- a/core/src/Cuda/Kokkos_Cuda_Instance.hpp +++ b/core/src/Cuda/Kokkos_Cuda_Instance.hpp @@ -449,10 +449,10 @@ class CudaInternal { } template -...

> If create_mirror_view is called without an execution space, the mirrored view is duplicated, and its elements are not constant. Otherwise, if it is called with an execution space, the...