cuda.core.utils.advise#

cuda.core.utils.advise(targets, advice, location=None, *, options=None)#

Apply managed-memory advice to one or more allocation ranges.

Parameters:

  • targets (Buffer | Sequence[Buffer]) – One or more managed allocations to advise.

  • advice (str | CUmem_advise) – Managed-memory advice. String aliases ("set_read_mostly", "unset_read_mostly", "set_preferred_location", "unset_preferred_location", "set_accessed_by", "unset_accessed_by") and CUmem_advise enum values are accepted.

  • location (Location | Device | int | Sequence[…]) – Target location(s). Required for advice values that consult a location; ignored (may be None) for set_read_mostly, unset_read_mostly, and unset_preferred_location. A sequence must match len(targets).

  • options (None) – Reserved for future per-call flags. Must be None.

On this page