Struct GpuStream 

pub struct GpuStream<'a> { /* private fields */ }

Handle to an orchestrated pipeline. Couples a compiled megakernel to its submission + completion infrastructure.

Implementations

impl<'a> GpuStream<'a>

pub fn new() -> Self

Create a pipeline handle with no io_uring stream attached.

Examples

use vyre_runtime::GpuStream;

let stream = GpuStream::new();

assert!(!stream.is_shutdown_requested());

pub fn with_uring(self, stream: AsyncUringStream<'a>) -> Self

Attach an io_uring stream for GPU-visible reads. Linux-only.

Use uring::NvmeGpuIngestDriver::new_gpudirect when the caller requires the native NVMe → BAR1 path instead of registered mapped reads.

pub fn poll(&mut self) -> Result<u32, PipelineError>

Reap completions and bump the megakernel tail pointer.

Errors

Propagates any uring syscall error from the underlying ring.

pub fn request_shutdown(&mut self)

Request graceful shutdown of the pipeline.

pub fn is_shutdown_requested(&self) -> bool

Whether shutdown has been requested.

pub unsafe fn wait_for_observable( host_visible_addr: *const u32, current: u32, timeout_ns: u64, ) -> Result<(), PipelineError>

Block until the megakernel writes a new value into the observable word. Uses futex_waitv on Linux 5.16+.

Errors

• PipelineError::NotLinux on non-Linux hosts.
• PipelineError::IoUringSyscall on futex errors.

Safety

host_visible_addr must be host-mapped and outlive this call.

Trait Implementations

impl Default for GpuStream<'_>

fn default() -> Self

Returns the “default value” for a type.