Pytorch is not compiled with NCCL support

I rebuild pytorch again. The overall steps are the same as the “Build from Source” part of this thread: PyTorch for Jetson , except for USE_SYSTEM_NCCL=0 & USE_NCCL=1 &USE_DISTRIBUTED=1.
The little test still gives :

distributed is available True
nccl is available False
/home/jane/.local/lib/python3.8/site-packages/torch/cuda/nccl.py:15: UserWarning: PyTorch is not compiled with NCCL support
  warnings.warn('PyTorch is not compiled with NCCL support')
cuda nccl is available False
distributed is initialized True
distributed is initialized True