You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy constructor and copy assignment - should be present.
Move constructor and assignment - I think these are implicitly deleted due to the user declared destructor.
In the current implementation of oneMKL, the implicit copy assignment / constructors are deleted due to the unique_ptr to the backend-specific descriptor implementation class.
For copying, it isn't clear what the behavour should be:
If constructor was copied, except for the backend-specific implementation class, and then recommitted, copying would become a very expensive operation.
Throwing on copying an already committed constructor at runtime would be annoying.
I don't see any downsides to move constructors / assignment operators.
What should the behaviour of the copy constructor be, and should there be move constructors?
The text was updated successfully, but these errors were encountered:
hjabird
added a commit
to hjabird/oneAPI-spec
that referenced
this issue
Jul 7, 2023
* Relates to uxlfoundation#487
* Adds move constructor/assignment to descriptor class
* Otherwise implicitly deleted.
* Add explicit copy constructor/assignment to descriptor class
* Otherwise ambigiuous
* Notes that copy operation is by value.
The oneMKL descriptor class has its constructor / destructor / copy / move interface defined as follows:
Consequently one might expect:
In the current implementation of oneMKL, the implicit copy assignment / constructors are deleted due to the unique_ptr to the backend-specific descriptor implementation class.
For copying, it isn't clear what the behavour should be:
I don't see any downsides to move constructors / assignment operators.
What should the behaviour of the copy constructor be, and should there be move constructors?
The text was updated successfully, but these errors were encountered: