cuda.core._memory._managed_buffer.AccessedBySetProxy#

class cuda.core._memory._managed_buffer.AccessedBySetProxy( buf: ManagedBuffer, )#

Live driver-backed view of set_accessed_by advice for a managed buffer.

Reads (__contains__, __iter__, len(...)) call cuMemRangeGetAttribute; writes (add, discard) call cuMemAdvise. There is no in-memory mirror, so the view always reflects the current driver state.

Note

The driver returns integer device ordinals (-1 for host); host NUMA distinctions applied via Host(numa_id=...) collapse to a generic Host() when iterating this set.

Methods

__init__( buf: ManagedBuffer, )#

add( location: Device | Host, ) → None#: Apply set_accessed_by advice for location.

clear()#: This is slow (creates N new iterators!) but effective.

discard( location: Device | Host, ) → None#

Apply unset_accessed_by advice for location.

Per the MutableSet contract, discard is a no-op for elements not in the set. set_accessed_by only accepts Device and the generic Host() — NUMA-aware host variants (Host(numa_id=...), Host.numa_current()) can never enter the set, so discarding them is silently ignored rather than forwarded to the driver.

isdisjoint(other)#: Return True if two sets have a null intersection.

pop()#: Return the popped value. Raise KeyError if empty.

remove(value)#: Remove an element. If not a member, raise a KeyError.