[CUDA][P2P] Check device capability in requires_cuda_p2p_access (py…

…torch#134523)

Tests seem to fail on e.g., Volta without this given the compile time meacros used e.g., in https://github.com/pytorch/pytorch/blob/79b7fff18820e4f62a02534a961d5930040e3475/torch/csrc/distributed/c10d/intra_node_comm.cu#L487

Pull Request resolved: pytorch#134523
Approved by: https://github.com/yifuwang, https://github.com/Skylion007

test/distributed/test_symmetric_memory.py

@@ -28,7 +28,9 @@
 
 def requires_cuda_p2p_access():
     cuda_p2p_access_available = (
-        torch.cuda.is_available() and torch.cuda.device_count() >= 2
+        torch.cuda.is_available()
+        and torch.cuda.get_device_capability() >= (8, 0)
+        and torch.cuda.device_count() >= 2
     )
     num_devices = torch.cuda.device_count()
     for i in range(num_devices - 1):