Module nccl 

2026-09-26: Raw NCCL bindings used by metrale-comm, plus the CUDA driver stream and event calls it needs. The function declarations follow the installed nccl.h (2.29.7); NcclConfig does not (see its doc).

The extern functions are unchecked: a caller passes a communicator from ncclCommInitRank, device pointers valid for the element count on that communicator’s device, and a stream of that device. The wrappers below (create_stream to stream_ready) check only the status code.

Owner: metrale-gpu-sys. Invariants: none beyond the types.

Structs

NcclConfig

2026-09-26: Meant for ncclCommInitRankConfig, which nothing in the workspace calls. It does not match ncclConfig_t in the installed nccl.h (2.29.7): that struct starts with magic NCCL_API_MAGIC (0xcafebeef), holds netName as a pointer, and has more fields.

NcclUniqueId

2026-09-26: ncclUniqueId: 128 bytes, passed by value.

Enums

NcclDataType

2026-09-26: ncclDataType_t, the values up to ncclBfloat16 (9) as in nccl.h.

NcclRedOp

2026-09-26: ncclRedOp_t’s built-in operations, values as in nccl.h.

NcclResult

2026-09-26: ncclResult_t, values 0 to 7 as in nccl.h.

Functions

check_nccl

2026-09-26: Ok for Success; otherwise an error naming context, the text of ncclGetErrorString and the code.

create_event

create_stream

destroy_event

destroy_stream

ncclAllGather^⚠

ncclAllReduce^⚠

ncclBroadcast^⚠

ncclCommAbort^⚠

ncclCommDeregister^⚠

ncclCommDestroy^⚠

ncclCommGetAsyncError^⚠

ncclCommInitRank^⚠

ncclCommInitRankConfig^⚠

ncclCommRegister^⚠

ncclGetErrorString^⚠

ncclGetUniqueId^⚠

ncclGroupEnd^⚠

ncclGroupStart^⚠

ncclMemAlloc^⚠

ncclMemFree^⚠

ncclRecv^⚠

ncclReduceScatter^⚠

ncclSend^⚠

nccl_mem_alloc^⚠

2026-09-26: ncclMemAlloc: size bytes of device memory from NCCL’s allocator; nccl.h notes the allocation may be larger than requested.

nccl_mem_free^⚠

2026-09-26: ncclMemFree for a pointer from nccl_mem_alloc.

record_event

stream_ready

2026-09-26: cuStreamQuery, without waiting: true when the stream’s work is done, false on CUDA_ERROR_NOT_READY (600), an error otherwise.

stream_wait_event

sync_stream

Type Aliases

NcclComm

2026-09-26: ncclComm_t, an opaque communicator handle.