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
When obtaining the device for a pointer, we currently rely on an attribute which only became available with CUDA 9.2 ; but we could, in fact, obtain the context handle, then obtain its device ID using a second API call. Let's do that to support CUDA 9.1 and earlier.
The text was updated successfully, but these errors were encountered:
* Can now get the device for a pointer in CUDA before 9.2 - by getting the context, then getting the device for the context
* Added `memory::device_of()` and `memory::context_of()`
* Added `memory::detail_` functions for returning the context handle or device id of a pointer, without having to remember the pointer attributes mechanism
* Can now get the device for a pointer in CUDA before 9.2 - by getting the context, then getting the device for the context
* Added `memory::device_of()` and `memory::context_of()`
* Added `memory::detail_` functions for returning the context handle or device id of a pointer, without having to remember the pointer attributes mechanism
* Can now get the device for a pointer in CUDA before 9.2 - by getting the context, then getting the device for the context
* Added `memory::device_of()` and `memory::context_of()`
* Added `memory::detail_` functions for returning the context handle or device id of a pointer, without having to remember the pointer attributes mechanism
When obtaining the device for a pointer, we currently rely on an attribute which only became available with CUDA 9.2 ; but we could, in fact, obtain the context handle, then obtain its device ID using a second API call. Let's do that to support CUDA 9.1 and earlier.
The text was updated successfully, but these errors were encountered: