Enum PlanStep 

pub enum PlanStep {
    KernelLaunch {
        nodes: Vec<NodeId>,
        function_name: String,
        config: KernelConfig,
        stream: StreamId,
    },
    Memcpy {
        node: NodeId,
        dir: MemcpyDir,
        size_bytes: usize,
        stream: StreamId,
    },
    Memset {
        node: NodeId,
        size_bytes: usize,
        value: u8,
        stream: StreamId,
    },
    EventRecord {
        event_id: usize,
        stream: StreamId,
    },
    EventWait {
        event_id: usize,
        stream: StreamId,
    },
    HostCallback {
        node: NodeId,
        label: String,
        stream: StreamId,
    },
    Barrier {
        node: NodeId,
        stream: StreamId,
    },
}

A single step in the execution plan.

Variants

KernelLaunch

Launch a (possibly fused) kernel.

Fields

nodes: Vec<NodeId>

Node(s) contributing to this launch (singleton if not fused).

function_name: String

Function name to look up in the loaded PTX module.

config: KernelConfig

Launch configuration.

stream: StreamId

Stream to submit on.

Memcpy

Transfer memory between host and device (or device to device).

Fields

node: NodeId

dir: MemcpyDir

size_bytes: usize

stream: StreamId

Memset

Fill a device buffer with a byte pattern.

Fields

node: NodeId

size_bytes: usize

value: u8

stream: StreamId

EventRecord

Record a CUDA event on stream to signal completion of preceding work.

Fields

event_id: usize

Synthetic event ID (for matching with EventWait).

stream: StreamId

EventWait

Wait for a previously recorded event before continuing on stream.

Fields

event_id: usize

stream: StreamId

HostCallback

Host-side callback inserted into the stream.

Fields

node: NodeId

label: String

stream: StreamId

Barrier

No-op barrier (no device work, used for explicit synchronisation).

Fields

node: NodeId

stream: StreamId

Implementations

impl PlanStep

pub fn stream(&self) -> StreamId

Returns the stream this step executes on.

pub fn tag(&self) -> &'static str

Returns a short tag string for display.

pub fn is_compute(&self) -> bool

Returns true if this step represents GPU compute work.

Trait Implementations

impl Clone for PlanStep

fn clone(&self) -> PlanStep

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for PlanStep

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

Formats the value using the given formatter.

impl Display for PlanStep

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

Formats the value using the given formatter.

impl PartialEq for PlanStep

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

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for PlanStep