Enum DeviceMsg

Source

pub enum DeviceMsg {
Show 30 variants
    Alloc(Box<dyn AllocDispatch>),
    CopyToHost(Box<dyn CopyToHostDispatch>),
    CopyFromHost(Box<dyn CopyFromHostDispatch>),
    Allocate {
        len: usize,
        reply: Sender<Result<GpuRef<f32>, GpuError>>,
    },
    AllocateF32 {
        len: usize,
        reply: Sender<Result<GpuRef<f32>, GpuError>>,
    },
    AllocateF64 {
        len: usize,
        reply: Sender<Result<GpuRef<f64>, GpuError>>,
    },
    AllocateI8 {
        len: usize,
        reply: Sender<Result<GpuRef<i8>, GpuError>>,
    },
    AllocateI32 {
        len: usize,
        reply: Sender<Result<GpuRef<i32>, GpuError>>,
    },
    AllocateI64 {
        len: usize,
        reply: Sender<Result<GpuRef<i64>, GpuError>>,
    },
    AllocateU8 {
        len: usize,
        reply: Sender<Result<GpuRef<u8>, GpuError>>,
    },
    AllocateU32 {
        len: usize,
        reply: Sender<Result<GpuRef<u32>, GpuError>>,
    },
    AllocateU64 {
        len: usize,
        reply: Sender<Result<GpuRef<u64>, GpuError>>,
    },
    CopyToHostF32 {
        src: GpuRef<f32>,
        dst: HostBuf<f32>,
        reply: Sender<Result<HostBuf<f32>, GpuError>>,
    },
    CopyFromHostF32 {
        src: HostBuf<f32>,
        dst: GpuRef<f32>,
        reply: Sender<Result<HostBuf<f32>, GpuError>>,
    },
    CopyToHostF64 {
        src: GpuRef<f64>,
        dst: HostBuf<f64>,
        reply: Sender<Result<HostBuf<f64>, GpuError>>,
    },
    CopyFromHostF64 {
        src: HostBuf<f64>,
        dst: GpuRef<f64>,
        reply: Sender<Result<HostBuf<f64>, GpuError>>,
    },
    CopyToHostI32 {
        src: GpuRef<i32>,
        dst: HostBuf<i32>,
        reply: Sender<Result<HostBuf<i32>, GpuError>>,
    },
    CopyFromHostI32 {
        src: HostBuf<i32>,
        dst: GpuRef<i32>,
        reply: Sender<Result<HostBuf<i32>, GpuError>>,
    },
    CopyToHostU32 {
        src: GpuRef<u32>,
        dst: HostBuf<u32>,
        reply: Sender<Result<HostBuf<u32>, GpuError>>,
    },
    CopyFromHostU32 {
        src: HostBuf<u32>,
        dst: GpuRef<u32>,
        reply: Sender<Result<HostBuf<u32>, GpuError>>,
    },
    CopyToHostU8 {
        src: GpuRef<u8>,
        dst: HostBuf<u8>,
        reply: Sender<Result<HostBuf<u8>, GpuError>>,
    },
    CopyFromHostU8 {
        src: HostBuf<u8>,
        dst: GpuRef<u8>,
        reply: Sender<Result<HostBuf<u8>, GpuError>>,
    },
    Sgemm(Box<SgemmRequest>),
    SnapshotContext {
        reply: Sender<Option<Arc<CudaContext>>>,
    },
    SnapshotStream {
        reply: Sender<Option<Arc<CudaStream>>>,
    },
    SnapshotChildren {
        reply: Sender<Option<KernelChildren>>,
    },
    WatchGeneration {
        reply: Sender<Receiver<u64>>,
    },
    Stats {
        reply: Sender<DeviceLoad>,
    },
    ContextReady {
        children: KernelChildren,
    },
    ContextLost,
}

Expand description

Public messages sent to a DeviceActor.

Phase 0.4 — the formerly-21 dtype-enumerated Allocate* / CopyToHost* / CopyFromHost* variants collapse into 3 boxed dispatchers:

DeviceMsg::Alloc — typed allocation
DeviceMsg::CopyToHost — D2H async copy
DeviceMsg::CopyFromHost — H2D async copy

Each carries a Box<dyn …Dispatch> whose concrete payload is an AllocReq<T> / CopyToHostReq<T> / CopyFromHostReq<T> for some T: CudaDtype. GpuRef<T> keeps its static dtype on both ends — the box is purely a uniform mailbox surface.

The legacy Allocate* / CopyToHost* / CopyFromHost* variants remain as #[deprecated] aliases. Existing call sites compile and run unchanged; the handler arm constructs the equivalent Box<dyn …Dispatch> and forwards through the new path.

Variants§

§

Alloc(Box<dyn AllocDispatch>)

Phase 0.4 generic alloc. Construct via DeviceMsg::alloc::<T> or Box::new(AllocReq::<T> { … }) directly.

§

CopyToHost(Box<dyn CopyToHostDispatch>)

Phase 0.4 generic D2H copy.

§