Struct DeviceProfile 

pub struct DeviceProfile {

    pub backend: &'static str,
    pub supports_subgroup_ops: bool,
    pub supports_indirect_dispatch: bool,
    pub supports_distributed_collectives: bool,
    pub supports_specialization_constants: bool,
    pub supports_f16: bool,
    pub supports_bf16: bool,
    pub supports_trap_propagation: bool,
    pub supports_tensor_cores: bool,
    pub has_mul_high: bool,
    pub has_dual_issue_fp32_int32: bool,
    pub has_subgroup_shuffle: bool,
    pub has_shared_memory: bool,
    pub max_native_int_width: u32,
    pub max_workgroup_size: [u32; 3],
    pub max_invocations_per_workgroup: u32,
    pub max_shared_memory_bytes: u32,
    pub max_storage_buffer_binding_size: u64,
    pub subgroup_size: u32,
    pub compute_units: u32,
    pub regs_per_thread_max: u32,
    pub l1_cache_bytes: u32,
    pub l2_cache_bytes: u32,
    pub mem_bw_gbps: u32,
    pub timing_quality: DeviceTimingQuality,
    pub supports_device_timestamps: bool,
    pub supports_hardware_counters: bool,
    pub ideal_unroll_depth: u32,
    pub ideal_vector_pack_bits: u32,
    pub ideal_workgroup_tile: [u32; 3],
    pub shared_memory_bank_count: u32,
    pub shared_memory_bank_width_bytes: u32,
}

Device capability snapshot used across driver-shared planning.

Fields

backend: &'static str

Stable backend identifier.

supports_subgroup_ops: bool

The device and lowering path support subgroup intrinsics.

supports_indirect_dispatch: bool

The backend supports indirect dispatch.

supports_distributed_collectives: bool

The backend lowers distributed collective communication nodes.

supports_specialization_constants: bool

The backend supports compile-time specialization constants.

supports_f16: bool

The backend lowers binary16 natively.

supports_bf16: bool

The backend lowers bfloat16 natively.

supports_trap_propagation: bool

The backend preserves explicit trap propagation.

supports_tensor_cores: bool

The backend lowers matrix-engine operations for supported shapes.

has_mul_high: bool

Native unsigned multiply-high is available to lowering strategies.

has_dual_issue_fp32_int32: bool

Integer and float pipelines can issue concurrently.

has_subgroup_shuffle: bool

Subgroup shuffle-like communication is available.

has_shared_memory: bool

Explicit workgroup/shared memory is available.

max_native_int_width: u32

Maximum native integer width in bits.

max_workgroup_size: [u32; 3]

Maximum workgroup dimensions.

max_invocations_per_workgroup: u32

Maximum invocations in one workgroup.

max_shared_memory_bytes: u32

Shared memory per workgroup in bytes.

max_storage_buffer_binding_size: u64

Maximum single storage-buffer binding in bytes.

subgroup_size: u32

Native subgroup size, or 0 when unknown.

compute_units: u32

Physical compute-unit count, or 0 when unknown.

regs_per_thread_max: u32

Maximum registers per thread, or 0 when unknown.

l1_cache_bytes: u32

L1 cache size in bytes, or 0 when unknown.

l2_cache_bytes: u32

L2 cache size in bytes, or 0 when unknown.

mem_bw_gbps: u32

Peak memory bandwidth in GB/s, or 0 when unknown.

timing_quality: DeviceTimingQuality

Timing-data quality exposed by this backend/device.

supports_device_timestamps: bool

Device timestamp queries/events are available for dispatch timing.

supports_hardware_counters: bool

Hardware counter sampling is available for benchmark telemetry.

ideal_unroll_depth: u32

Device-profile preferred unroll depth, or 0 when unknown.

ideal_vector_pack_bits: u32

Device-profile preferred vector pack width in bits, or 0 when unknown.

ideal_workgroup_tile: [u32; 3]

Device-profile preferred workgroup tile, or [0, 0, 0] when unknown.

shared_memory_bank_count: u32

Shared-memory bank count, or 0 when unknown.

shared_memory_bank_width_bytes: u32

Shared-memory bank width in bytes, or 0 when unknown.

Implementations

impl DeviceProfile

pub const fn conservative(backend: &'static str) -> Self

Conservative profile for a backend that has not probed a device.

pub fn from_backend(backend: &dyn VyreBackend) -> Self

Build a profile from the stable backend trait capability methods.

pub const fn validation_capabilities(self) -> BackendCapabilities

Validation capability projection.

pub const fn adapter_caps(self) -> AdapterCaps

Optimizer capability projection.

pub const fn strategy_capabilities(self) -> BackendCapabilities

Strategy capability projection.

Trait Implementations

impl Clone for DeviceProfile

fn clone(&self) -> DeviceProfile

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for DeviceProfile

impl Debug for DeviceProfile

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

Formats the value using the given formatter.

impl Default for DeviceProfile

fn default() -> Self

Returns the “default value” for a type.

impl Eq for DeviceProfile

impl From<DeviceProfile> for AdapterCaps

fn from(profile: DeviceProfile) -> Self

Converts to this type from the input type.

impl From<DeviceProfile> for BackendCapabilities

fn from(profile: DeviceProfile) -> Self

Converts to this type from the input type.

impl PartialEq for DeviceProfile

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

Tests for self and other values to be equal, and is used by ==.

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

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for DeviceProfile