o i@svddlZddlZddlmZddlmZmZddlZgdZ dZ ddZ ddZ d d Z d d Zd eejeejfddfddZde ddfddZdde ddfddd eejdeeejeejfdededeeejjdeeejddfddZ d!d eejdeddfddZ  d"d eejdeejddfddZe ddfd eejdeejdeddfdd ZdS)#N)Sequence)OptionalUnion) all_reducereduce broadcast all_gatherreduce_scattercCsxttjdstjddddSt}|D]%}|jrdS|s#dS|js)dS| }||vr4dS| |qdS)N_nccl_all_reducez)PyTorch is not compiled with NCCL support stacklevelFT) hasattrtorch_Cwarningswarnset is_sparse is_contiguousis_cuda get_deviceadd)tensorsdevicestensordevicerQ/sda-disk/www/egybert/egybert_env/lib/python3.10/site-packages/torch/cuda/nccl.py is_availables  rcCsTtj}|d?}|d?d@}|d@}tjd}|dkr$|||fS||||fS)a Returns the version of the NCCL. This function returns a tuple containing the major, minor, and patch version numbers of the NCCL. The suffix is also included in the tuple if a version suffix exists. Returns: tuple: The version information of the NCCL. izutf-8)rr _nccl_version_nccl_version_suffixdecode)vermajorminorpatchsuffixrrrversion$s    r+cCs tjSN)rr_nccl_unique_idrrrr unique_id9s r.cCstj|||Sr,)rr_nccl_init_rank) num_ranksuidrankrrr init_rank=sr3inputsreturncCs&t|tjjr t|tjrtddS)Nz(Inputs should be a collection of tensors) isinstance collectionsabc ContainerrTensor TypeError)r4rrr_check_sequence_typeAs r<cCs4t||dur |}t|tj|||||dSr,)r<rrr r4outputsopstreamscommsrrrrHs r)r>outputrootr?r@r>cCst||dur|durtdtjdtdd||}n$t|tjs7t|tj j r7tjdtdd||}n |dur?||n|}tj ||||||dS)Nz'output' and 'outputs' can not be both specified. 'outputs' is deprecated in favor of 'output', taking in a single output tensor. The signature of reduce is: reduce(inputs, output=None, root=0, op=SUM, streams=None, comms=None).z`nccl.reduce` with an output tensor list is deprecated. Please specify a single output tensor with argument 'output' instead instead.r r z\nccl.reduce with an output tensor list is deprecated. Please specify a single output tensor.) r< ValueErrorrr FutureWarningr6rr:r7r8rr _nccl_reduce)r4rBrCr?r@rAr>_outputrrrrRs.   rcCst|tj||||dSr,)r<rr_nccl_broadcast)r4rCr@rArrrr}srcCs&t|t|tj||||dSr,)r<rr_nccl_all_gather)r4r>r@rArrrrsrcCs(t|t|tj|||||dSr,)r<rr_nccl_reduce_scatterr=rrrr sr )rNN)NN)r7rcollections.abcrtypingrr torch.cudar__all__SUMrr+r.r3r:r<rintcudaStreamrrrr rrrrs "   ,