cuda.core.utils.CopyOptions#

class cuda.core.utils.CopyOptions(
src_access_order: MemcpySrcAccessOrder | str = 'stream',
src_location_hint: Device | Host | None = None,
dst_location_hint: Device | Host | None = None,
overlap_mode: MemcpyOverlapMode | str = 'default',
)#

Attribute bundle for a single copy within a batched memcpy.

Parameters:

  • src_access_order (MemcpySrcAccessOrder or str) – Hint describing how the source will be accessed. Default is "stream" (stream-ordered access).

  • src_location_hint (cuda.core.Device | cuda.core.Host | None) – Hint for the source memory location. Honored only for managed memory on devices with concurrent managed access and for system-allocated pageable memory on devices with pageable memory access; ignored for all other memory types. Does not prefetch memory and does not set persistent memory advice. None means no hint.

  • dst_location_hint (cuda.core.Device | cuda.core.Host | None) – Hint for the destination memory location. Same semantics and restrictions as src_location_hint. None means no hint.

  • overlap_mode (MemcpyOverlapMode or str) – Hint requesting that the copy overlap with concurrent compute work. This is advisory; it has an effect only on devices that support it. Default is "default".

On this page