Add support for bfloat16.
Add ncclAvg reduction operation.
Improve performance for aggregated operations.
Improve performance for tree.
Improve network error reporting.
Add NCCL_NET parameter to force a specific network.
Add NCCL_IB_QPS_PER_CONNECTION parameter to split IB traffic onto multiple queue pairs.
Fix topology detection error in WSL2.
Fix proxy memory elements affinity (improve alltoall performance).
Fix graph search on cubemesh topologies.
Fix hang in cubemesh during NVB connections.
Este commit está contenido en:
Ke Wen
2021-07-08 14:12:04 -07:00
padre 3fec2fa5ee
commit 7e51592129
Se han modificado 52 ficheros con 3496 adiciones y 2469 borrados
+1
Ver fichero
@@ -16,6 +16,7 @@ ncclResult_t bootstrapInit(ncclUniqueId* id, int rank, int nranks, void** commSt
ncclResult_t bootstrapAllGather(void* commState, void* allData, int size);
ncclResult_t bootstrapSend(void* commState, int peer, int tag, void* data, int size);
ncclResult_t bootstrapRecv(void* commState, int peer, int tag, void* data, int size);
ncclResult_t bootstrapBarrier(void* commState, int *ranks, int tag, int rank, int nranks);
ncclResult_t bootstrapRemAlloc(size_t size, int rank, void* commState, int* id, cudaIpcMemHandle_t* ipc, void** ptr);
ncclResult_t bootstrapRemFree(int id, int rank, void* commState);
ncclResult_t bootstrapClose(void* commState);