Struct cuda_oxide::stream::Stream

pub struct Stream<'a> { /* fields omitted */ }

A stream of asynchronous operations operating in a Context

Implementations

impl<'a> Stream<'a>

pub fn new(_handle: &Rc<Handle<'a>>) -> CudaResult<Self>

Creates a new stream for a handle

pub fn sync(&mut self) -> CudaResult<()>

Drives all pending tasks on the stream to completion

pub fn is_synced(&self) -> CudaResult<bool>

Returns Ok(true) if the stream has finished processing all queued tasks.

pub fn wait_32<'b>(
    &'b mut self,
    addr: &'b DevicePtr<'a>,
    value: u32,
    mode: WaitValueMode,
    flush: bool
) -> CudaResult<()>

Wait for a 4-byte value in a specific location to compare to value by mode.

pub fn wait_64<'b>(
    &mut self,
    addr: &'b DevicePtr<'a>,
    value: u64,
    mode: WaitValueMode,
    flush: bool
) -> CudaResult<()>

Wait for a 8-byte value in a specific location to compare to value by mode.

pub fn write_32<'b>(
    &'b mut self,
    addr: &'b DevicePtr<'a>,
    value: u32,
    no_memory_barrier: bool
) -> CudaResult<()>

Writes a 4-byte value to device memory asynchronously

pub fn write_64<'b>(
    &'b mut self,
    addr: &'b DevicePtr<'a>,
    value: u64,
    no_memory_barrier: bool
) -> CudaResult<()>

Writes a 8-byte value to device memory asynchronously

pub fn callback<F: FnOnce() + Send + Sync>(
    &mut self,
    callback: F
) -> CudaResult<()>

Calls a callback closure function callback once all prior tasks in the Stream have been driven to completion. Note that it is a memory leak to drop the stream before this callback is called. The callback is not guaranteed to be called if the stream errors out. Also note that it is erroneous in libcuda to make any calls to libcuda from this callback. The callback is called from a CUDA internal thread, however this is an implementation detail of libcuda and not guaranteed.

pub unsafe fn launch<'b, D1: Into<Dim3>, D2: Into<Dim3>, K: KernelParameters>(
    &mut self,
    f: &Function<'a, 'b>,
    grid_dim: D1,
    block_dim: D2,
    shared_mem_size: u32,
    parameters: K
) -> CudaResult<()>

Launch a CUDA kernel on this Stream with the given grid_dim grid dimensions, block_dim block dimensions, shared_mem_size allocated shared memory pool, and parameters kernel parameters. It is undefined behavior to pass in parameters that do not conform to the passes CUDA kernel. If the argument count is wrong, CUDA will generally throw an error. If your parameters is accurate to the kernel definition, then this function is otherwise safe.

Trait Implementations

impl<'a> Drop for Stream<'a>

fn drop(&mut self)

Executes the destructor for this type.