Module uring 

Linux io_uring integration. Compiled out on macOS / Windows. Linux io_uring scaffolding for NVMe → GPU-visible memory streaming. All items here are gated on cfg(target_os = "linux") and compiled out on every other platform.

Public surface:

• IoUringState - raw syscall + mmap wrapper for the SQ/CQ rings.
• GpuMappedBuffer - typed wrapper around a GPU-visible memory region: either a registered host-visible mapping or a BAR1 peer-memory allocation.
• AsyncUringStream - the submission glue: pushes reads into the SQ and advances an atomic tail pointer the megakernel observes.
• NvmeGpuIngestDriver - publishes completed slots into the megakernel io_queue; new_gpudirect requires the native NVMe → BAR1 path.

Re-exports

pub use driver::CompletedIngest;

pub use driver::NativeReadPath;

pub use driver::NvmeGpuIngestDriver;

pub use driver::NvmeGpuIngestTelemetry;

pub use gpudirect::encode_nvme_read_sqe;

pub use gpudirect::GpuDirectCapability;

pub use gpudirect::NVME_CMD_READ;

pub use io_loop::MegakernelIoLoop;

pub use io_loop::RegisteredIoDestination;

pub use pump::UringMegakernelPump;

pub use ring::IoUringState;

pub use stream::AsyncUringStream;

pub use stream::GpuMappedBuffer;

pub use stream::Iovec;

Modules

driver

End-to-end ingest driver: file/NVMe -> io_uring -> mapped slot -> io_queue.

gpudirect

GPUDirect Storage capability probe + passthrough helper.

io_loop

Autonomous IO loop for persistent megakernel.

pump

File-read → megakernel ring-slot pump. Linux-only.

ring

Raw io_uring orchestrator and syscall wrappers.

stream

AsyncUringStream - drives io_uring reads into GPU-visible memory and advances the megakernel tail pointer on each completion.