Skip to main content

Crate baracuda_cufile_sys

Crate baracuda_cufile_sys 

Source
Expand description

Raw FFI + dynamic loader for NVIDIA cuFile (GPUDirect Storage).

Linux-only: cuFile calls ibverbs + NVIDIA’s GDS kernel driver and has no Windows/macOS analogue. On non-Linux platforms the loader returns LoaderError::UnsupportedPlatform.

Modules§

CUfileOpcode
Opcode selector for CUfileIOParams_t.

Structs§

CUfileDescr_t
CUfileDescr_t — struct describing a file (Unix fd or Win32 handle) for registration. cuFile’s own layout is an outer tagged union; we model only the Unix-fd path since this API is Linux-only anyway.
CUfileError_t
CUfileError_t — status plus a CUDA error code (for I/O that dispatched through the GPU).
CUfileIOEvents_t
CUfileIOEvents_t — status for a single batched-I/O entry.
CUfileIOParams_t
CUfileIOParams_t — 1 entry in a batched-IO request.
CUfileOpError
cuFile operation status (a.k.a. CUfileOpError).
Cufile

Functions§

cufile

Type Aliases§

CUfileBatchHandle_t
Opaque batch-handle (maps to CUfileBatchHandle_t).
CUfileHandle_t
CUfileHandle_t — opaque handle returned by cuFileHandleRegister.
PFN_cuFileBatchIOCancel
PFN_cuFileBatchIODestroy
PFN_cuFileBatchIOGetStatus
PFN_cuFileBatchIOSetUp
PFN_cuFileBatchIOSubmit
PFN_cuFileBufDeregister
PFN_cuFileBufRegister
PFN_cuFileDriverClose
PFN_cuFileDriverGetProperties
PFN_cuFileDriverOpen
PFN_cuFileDriverSetMaxCacheSize
PFN_cuFileDriverSetMaxDirectIOSize
PFN_cuFileDriverSetMaxPinnedMemSize
PFN_cuFileDriverSetPollMode
PFN_cuFileGetVersion
PFN_cuFileHandleDeregister
PFN_cuFileHandleRegister
PFN_cuFileOpStatusError
PFN_cuFileRead
PFN_cuFileReadAsync
PFN_cuFileStreamDeregister
PFN_cuFileStreamRegister
PFN_cuFileUseCount
PFN_cuFileWrite
PFN_cuFileWriteAsync