Trait rust_cuda::lend::LendToCudaAsync

source · 
pub trait LendToCudaAsync: RustToCudaAsync {
    // Required methods
    fn lend_to_cuda_async<'stream, O, E: From<CudaError>, F: FnOnce(Async<'_, 'stream, HostAndDeviceConstRef<'_, DeviceAccessible<<Self as RustToCuda>::CudaRepresentation>>, NoCompletion>) -> Result<O, E>>(
        &self,
        stream: Stream<'stream>,
        inner: F,
    ) -> Result<O, E>
       where Self: Sync;
    fn lend_to_cuda_mut_async<'a, 'stream, O, E: From<CudaError>, F: for<'b> FnOnce(Async<'b, 'stream, HostAndDeviceMutRef<'_, DeviceAccessible<<Self as RustToCuda>::CudaRepresentation>>, NoCompletion>) -> Result<O, E>, T: 'a>(
        this: BoxRefMut<'a, T, Self>,
        stream: Stream<'stream>,
        inner: F,
    ) -> Result<(Async<'a, 'stream, BoxRefMut<'a, T, Self>, CompletionFnMut<'a, Self>>, O), E>
       where Self: Sync + SafeMutableAliasing;
    fn move_to_cuda_async<'stream, O, E: From<CudaError>, F: for<'a> FnOnce(Async<'a, 'stream, HostAndDeviceOwned<'_, DeviceAccessible<<Self as RustToCuda>::CudaRepresentation>>, NoCompletion>) -> Result<O, E>>(
        self,
        stream: Stream<'stream>,
        inner: F,
    ) -> Result<O, E>
       where Self: Send + RustToCuda<CudaRepresentation: StackOnly, CudaAllocation: EmptyCudaAlloc>;
}
Available on crate feature host only.

Required Methods§

source

fn lend_to_cuda_async<'stream, O, E: From<CudaError>, F: FnOnce(Async<'_, 'stream, HostAndDeviceConstRef<'_, DeviceAccessible<<Self as RustToCuda>::CudaRepresentation>>, NoCompletion>) -> Result<O, E>>( &self, stream: Stream<'stream>, inner: F, ) -> Result<O, E>
where Self: Sync,

Lends an immutable copy of &self to CUDA:

  • code in the CUDA kernel can only access &self through the DeviceConstRef inside the closure
  • after the closure, &self will not have changed, i.e. interior mutability is not handled by this method

Since the HostAndDeviceConstRef is wrapped in an Async with NoCompletion, this Async can be safely dropped or forgotten without changing any behaviour. Therefore, this Async does not need to be returned from the inner closure.

§Errors

Returns a CudaError iff an error occurs inside CUDA

source

fn lend_to_cuda_mut_async<'a, 'stream, O, E: From<CudaError>, F: for<'b> FnOnce(Async<'b, 'stream, HostAndDeviceMutRef<'_, DeviceAccessible<<Self as RustToCuda>::CudaRepresentation>>, NoCompletion>) -> Result<O, E>, T: 'a>( this: BoxRefMut<'a, T, Self>, stream: Stream<'stream>, inner: F, ) -> Result<(Async<'a, 'stream, BoxRefMut<'a, T, Self>, CompletionFnMut<'a, Self>>, O), E>
where Self: Sync + SafeMutableAliasing,

Lends a mutable borrow of &mut self to CUDA iff Self is SafeMutableAliasing:

  • code in the CUDA kernel can only access &mut self through the DeviceMutRef inside the closure
  • after the closure, &mut self will reflect the changes from the kernel execution
§Errors

Returns a rustacuda::errors::CudaError iff an error occurs inside CUDA

source

fn move_to_cuda_async<'stream, O, E: From<CudaError>, F: for<'a> FnOnce(Async<'a, 'stream, HostAndDeviceOwned<'_, DeviceAccessible<<Self as RustToCuda>::CudaRepresentation>>, NoCompletion>) -> Result<O, E>>( self, stream: Stream<'stream>, inner: F, ) -> Result<O, E>
where Self: Send + RustToCuda<CudaRepresentation: StackOnly, CudaAllocation: EmptyCudaAlloc>,

Moves self to CUDA iff self is StackOnly.

Since the HostAndDeviceOwned is wrapped in an Async with NoCompletion, this Async can be safely dropped or forgotten without changing any behaviour. Therefore, this Async does not need to be returned from the inner closure.

§Errors

Returns a CudaError iff an error occurs inside CUDA

Object Safety§

This trait is not object safe.

Implementors§