集体通信库(MCCL)文档
- MCCL 概览
- 设置
- 使用 MCCL
- MCCL API
- 通信器创建和管理函数
- mcclGetLastError
- mcclGetErrorString
- mcclGetVersion
- mcclGetUniqueId
- mcclCommInitRank
- mcclCommInitAll
- mcclCommInitRankConfig
- mcclCommInitRankScalable
- mcclCommSplit
- mcclCommFinalize
- mcclCommDestroy
- mcclCommAbort
- mcclCommGetAsyncError
- mcclCommCount
- mcclCommCuDevice
- mcclCommUserRank
- mcclCommRegister
- mcclCommDeregister
- mcclMemAlloc
- mcclMemFree
- 集体通信函数
- 组调用
- 点对点通信函数
- 类型
- 用户定义的约简操作符
- 通信器创建和管理函数
- 从 MCCL 1 迁移到 MCCL 2
- 示例
- MCCL 和 MPI
- 环境变量
- 系统配置
- MCCL_SOCKET_IFNAME
- MCCL_SOCKET_FAMILY
- MCCL_SOCKET_RETRY_CNT
- MCCL_SOCKET_RETRY_SLEEP_MSEC
- MCCL_SOCKET_NTHREADS
- MCCL_NSOCKS_PERTHREAD
- MCCL_CROSS_NIC
- MCCL_IB_HCA
- MCCL_IB_TIMEOUT
- MCCL_IB_RETRY_CNT
- MCCL_IB_GID_INDEX
- MCCL_IB_ADDR_FAMILY
- MCCL_IB_ADDR_RANGE
- MCCL_IB_ROCE_VERSION_NUM
- MCCL_IB_SL
- MCCL_IB_TC
- MCCL_IB_FIFO_TC
- MCCL_IB_RETURN_ASYNC_EVENTS
- MCCL_OOB_NET_ENABLE
- MCCL_OOB_NET_IFNAME
- MCCL_UID_STAGGER_THRESHOLD
- MCCL_UID_STAGGER_RATE
- MCCL_NET
- MCCL_NET_PLUGIN
- MCCL_TUNER_PLUGIN
- MCCL_PROFILER_PLUGIN
- MCCL_IGNORE_CPU_AFFINITY
- MCCL_CONF_FILE
- MCCL_DEBUG
- MCCL_DEBUG_FILE
- MCCL_DEBUG_SUBSYS
- MCCL_DEBUG_TIMESTAMP_FORMAT
- MCCL_DEBUG_TIMESTAMP_LEVELS
- MCCL_COLLNET_ENABLE
- MCCL_COLLNET_NODE_THRESHOLD
- MCCL_TOPO_FILE
- MCCL_TOPO_DUMP_FILE
- MCCL_SET_THREAD_NAME
- 调试
- MCCL_P2P_DISABLE
- MCCL_P2P_LEVEL
- MCCL_P2P_DIRECT_DISABLE
- MCCL_SHM_DISABLE
- MCCL_BUFFSIZE
- MCCL_NTHREADS
- MCCL_MAX_NCHANNELS
- MCCL_MIN_NCHANNELS
- MCCL_CHECKS_DISABLE
- MCCL_CHECK_POINTERS
- MCCL_LAUNCH_MODE
- MCCL_IB_DISABLE
- MCCL_IB_AR_THRESHOLD
- MCCL_IB_QPS_PER_CONNECTION
- MCCL_IB_SPLIT_DATA_ON_QPS
- MCCL_IB_MUSA_SUPPORT
- MCCL_IB_PCI_RELAXED_ORDERING
- MCCL_IB_ADAPTIVE_ROUTING
- MCCL_IB_ECE_ENABLE
- MCCL_MEM_SYNC_DOMAIN
- MCCL_CUMEM_ENABLE
- MCCL_CUMEM_HOST_ENABLE
- MCCL_NET_GDR_LEVEL (原 MCCL_IB_GDR_LEVEL)
- MCCL_NET_GDR_C2C
- MCCL_NET_GDR_READ
- MCCL_NET_SHARED_BUFFERS
- MCCL_NET_SHARED_COMMS
- MCCL_SINGLE_RING_THRESHOLD
- MCCL_LL_THRESHOLD
- MCCL_TREE_THRESHOLD
- MCCL_ALGO
- MCCL_PROTO
- MCCL_NVB_DISABLE
- MCCL_PXN_DISABLE
- MCCL_P2P_PXN_LEVEL
- MCCL_RUNTIME_CONNECT
- MCCL_GRAPH_REGISTER
- MCCL_LOCAL_REGISTER
- MCCL_LEGACY_MUSA_REGISTER
- MCCL_SET_STACK_SIZE
- MCCL_GRAPH_MIXING_SUPPORT
- MCCL_DMABUF_ENABLE
- MCCL_P2P_NET_CHUNKSIZE
- MCCL_P2P_LL_THRESHOLD
- MCCL_ALLOC_P2P_NET_LL_BUFFERS
- MCCL_COMM_BLOCKING
- MCCL_CGA_CLUSTER_SIZE
- MCCL_MAX_CTAS
- MCCL_MIN_CTAS
- MCCL_NVLS_ENABLE
- MCCL_IB_MERGE_NICS
- MCCL_MNNVL_ENABLE
- MCCL_RAS_ENABLE
- MCCL_RAS_ADDR
- MCCL_RAS_TIMEOUT_FACTOR
- MCCL_LAUNCH_ORDER_IMPLICIT
- MCCL_LAUNCH_RACE_FATAL
- 系统配置
- 故障排除

