warp.set_peer_access_enabled

warp.set_peer_access_enabled(target_device, peer_device, enable)

Enable or disable direct access from peer_device to the memory of target_device.

Enabling peer access can improve the speed of peer-to-peer memory transfers, but can have a negative impact on memory consumption and allocation performance.

This applies to memory allocated using default CUDA allocators. For memory allocated using CUDA pooled allocators, use set_mempool_access_enabled().

Parameters:

• target_device (Device | str | None)

• peer_device (Device | str | None)

• enable (bool)

Return type:

None