Struct ReduceRequest 

pub struct ReduceRequest<T: NcclReduceSupported> {
    pub send: GpuRef<T>,
    pub recv: Option<GpuRef<T>>,
    pub op: ReduceOp,
    pub root: i32,
    pub reply: Sender<Result<(), GpuError>>,
}

Reduce: each rank contributes send; the result lands in recv on root. On non-root ranks, recv may be None.

Fields

send: GpuRef<T>

recv: Option<GpuRef<T>>

op: ReduceOp

root: i32

reply: Sender<Result<(), GpuError>>

Trait Implementations

impl<T: NcclReduceSupported> CollectiveDispatch for ReduceRequest<T>

fn dtype_kind(&self) -> DispatchDType

fn device_id(&self) -> Option<u32>

fn dispatch(self: Box<Self>, ctx: &CollectiveDispatchCtx<'_>)