Enum TransferPath 

pub enum TransferPath {
    SameDevice,
    Peer,
    SharedMemory,
    DeviceCopy,
    HostStaging,
}

How bytes can move from one device to another. Backends advertise the paths they support as the source; Router consumes this to pick the cheapest available path between two specific devices.

Bandwidth + latency estimates land in A5 alongside Router integration; for A4 the variants exist as discriminators only.

Variants

SameDevice

No transfer needed (same device instance).

Peer

Direct peer-to-peer (CUDA P2P, NVLink, AMD Infinity Fabric, GPUDirect). Requires explicit enablement on both ends.

SharedMemory

Zero-copy via shared memory mapping (UMA, ResizableBAR, dma-buf). Common on integrated GPUs and Apple Silicon.

DeviceCopy

Bulk transfer engine (cudaMemcpy, vkCmdCopyBuffer + staging buffer). The default for cross-device GPU↔GPU on standard hardware.

HostStaging

Through CPU as intermediary. Universal fallback — every backend supports this. Highest latency but always available.

Trait Implementations

impl Clone for TransferPath

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Copy for TransferPath

impl Debug for TransferPath

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Eq for TransferPath

impl Hash for TransferPath

fn hash<__H: Hasher>(&self, state: &mut __H)

Feeds this value into the given Hasher.

fn hash_slice<H>(data: &[Self], state: &mut H)

where H: Hasher, Self: Sized,

Feeds a slice of this type into the given Hasher.

impl PartialEq for TransferPath

fn eq(&self, other: &Self) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for TransferPath