Struct GpuBackendDeviceProfile 

pub struct GpuBackendDeviceProfile { /* private fields */ }

Pure per-device construction profile validated before owner acquisition.

Encapsulates device-local resource policies and configuration defaults, including capacity, reserve, query limit, active-attempt count, source work, installation bounds, capabilities, and telemetry. Retains no memory ownership; the deployment owner binds a profile with explicit GpuMemoryOwnership via GpuBackendDeviceConfig::new.

Implementations

impl GpuBackendDeviceProfile

pub fn new(device_ordinal: usize) -> Self

Starts a profile with automatic capacity discovery and two active attempts. Callers may override any device-local bound below.

pub fn with_memory_limit(self, memory_limit: GpuDeviceMemoryLimit) -> Self

Sets the admission capacity policy. Validation and device discovery resolve the policy before managed capacity is installed.

pub fn with_device_capabilities( self, device_capabilities: DeviceCapabilities, ) -> Self

Freezes the native execution capabilities available on this device.

The default accepts every capability with a modeled native contract. Supplying an exact profile is useful for heterogeneous deployments and is consumed by admission before any waiter is queued.

pub fn with_backend_reserve_bytes(self, backend_reserve_bytes: u64) -> Self

Sets bytes reserved outside managed query and cache capacity. Backend construction rejects a reserve that leaves no managed capacity.

pub fn with_cache_cap_bytes(self, cache_cap_bytes: u64) -> Self

Sets the aggregate accounted resident-cache ceiling for this device.

pub fn cache_cap_bytes(&self) -> u64

Returns the accounted resident-cache ceiling in bytes. The default is zero.

pub fn with_default_query_limit_bytes( self, default_query_limit_bytes: u64, ) -> Self

Sets the immutable default memory limit for each query on this device. When unset, it resolves to the query share divided by the maximum active-attempt count, rounded down to whole bytes.

pub fn with_source_work_per_attempt( self, source_work_per_attempt: usize, ) -> Self

Sets source work reserved per active attempt.

This is part of the immutable device envelope; execution/session source scheduling controls must not change controller capacity.

pub fn with_max_active_attempts(self, max_active_attempts: usize) -> Self

Sets the device admission concurrency bound. Zero is rejected by backend validation; existing explicit installation bounds remain in force.

pub fn with_derived_installation_bounds( self, max_active_attempts: usize, ) -> Self

Sets the active-attempt limit and clears explicit installation bounds so construction derives them from this profile.

Prefer this over spelling with_max_active_attempts plus with_installation_bounds separately: the pair carries the same number twice with no type-level guarantee they agree. Explicit bounds remain available through with_installation_bounds for deployments that deliberately overprovision stream or worker capacity.

pub fn with_installation_bounds( self, installation_bounds: GpuBackendDeviceInstallationBounds, ) -> Self

Sets explicit stream, worker, and execution-lane bounds.

Most callers should prefer Self::with_derived_installation_bounds; use this only to deliberately overprovision beyond the derived A_d * (1 + S_d) stream formula.

pub fn with_telemetry(self, install_telemetry: bool) -> Self

Enables or disables installation of diagnostic device telemetry. Telemetry does not participate in query admission accounting.

pub fn device_ordinal(&self) -> usize

Returns the CUDA device ordinal selected by this declaration.

pub fn device_capabilities(&self) -> &DeviceCapabilities

Immutable native execution capabilities selected for this device.

pub fn memory_limit(&self) -> GpuDeviceMemoryLimit

Returns the unresolved admission capacity policy, resolved at device discovery.

pub fn backend_reserve_bytes(&self) -> u64

Returns bytes reserved outside managed query and cache capacity. The default is zero.

pub fn default_query_limit_bytes(&self) -> Option<u64>

Returns the explicit default query grant in bytes. None requests floor((managed_capacity_bytes - cache_cap_bytes) / max_active_attempts) after device capacity resolution.

pub fn source_work_per_attempt(&self) -> usize

Returns auxiliary source-work slots reserved for each active attempt. The default is one.

pub fn max_active_attempts(&self) -> usize

Returns the maximum simultaneously admitted attempts on this device. The default is two.

pub fn installation_bounds(&self) -> Option<GpuBackendDeviceInstallationBounds>

Returns the explicit installation bounds. None requests the values returned by Self::derived_installation_bounds.

pub fn derived_installation_bounds( &self, ) -> Result<GpuBackendDeviceInstallationBounds>

The installation bounds a profile without explicit ones receives: A_d * (1 + S_d) stream slots, A_d execution lanes, and one worker per active attempt plus the worker reserved for context setup and release cleanup. Explicit bounds must cover the stream slots and workers of this derivation.

pub fn install_telemetry(&self) -> bool

Returns whether device telemetry is installed. New profiles enable it.

Trait Implementations

impl Clone for GpuBackendDeviceProfile

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for GpuBackendDeviceProfile

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

Formats the value using the given formatter.

impl Eq for GpuBackendDeviceProfile

impl PartialEq for GpuBackendDeviceProfile

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

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for GpuBackendDeviceProfile