Christian Trott
Christian Trott
Another advantage is that no implementation of the customization point is needed for the common case of `container_type::data()` returning `element_type*` .
My thinking always was that the vast majority of accesses happen via mdspan not mdarray. You use mdarray to create the original data, but all your functions actually take mdspan....
Does the original OpenMP offload code also use managed memory? Or do you explicitly move memory via data clauses? Generally I would NOT expect any OpenMP offload code to be...
Retest this please.
Was the post part maybe something the resilience plugin needed? @nmm0
We are still waiting on confirmation that this works at all properly, which may require CUDA 12.4 and Drivers 550
hm interesting problem. If we allow `create_mirror_view` on const data to not return a new view, we may need to allow `deep_copy(const_a, const_a)`: i.e. allow deep_copy between two views of...
That said I am not sure if the current behavior is bad or not, we should think this through carefully.
@pzehner apologies lol there may be some unfortunate legacy behavior we need to change to get stuff being fully consistent ...
Not sure what the cases are, but I guess the thing this would now "wrongly" complain about is two layout stride views with offset 1 between them and stride 2?...