[Gradient Compression] Directly let world_size = process_group.size()

Address the comment on #49417 (comment)

Differential Revision: [D25673997](https://our.internmc.facebook.com/intern/diff/D25673997/)

ghstack-source-id: 119021459
Pull Request resolved: #49715

torch/distributed/algorithms/ddp_comm_hooks/default_hooks.py

@@ -18,9 +18,7 @@ def allreduce_hook(
         >>> ddp_model.register_comm_hook(process_group, allreduce_hook)
     """
     group_to_use = process_group if process_group is not None else dist.group.WORLD
-    world_size = (
-        process_group.size() if process_group is not None else dist.get_world_size()
-    )
+    world_size = process_group.size()
 
     tensor = bucket.get_tensors()[0]
     fut = dist.all_reduce(tensor, group=group_to_use, async_op=True).get_future()
@@ -46,9 +44,7 @@ def fp16_compress_hook(
         >>> ddp_model.register_comm_hook(process_group, fp16_compress_hook)
     """
     group_to_use = process_group if process_group is not None else dist.group.WORLD
-    world_size = (
-        process_group.size() if process_group is not None else dist.get_world_size()
-    )
+    world_size = process_group.size()
 
     compressed_tensor = bucket.get_tensors()[0].to(torch.float16)
 
@@ -100,9 +96,7 @@ def _allgather_then_aggregate_hook(
     """
     group_to_use = process_group if process_group is not None else dist.group.WORLD
     rank = process_group.rank() if process_group is not None else dist.get_rank()
-    world_size = (
-        process_group.size() if process_group is not None else dist.get_world_size()
-    )
+    world_size = process_group.size()
 
     tensor = bucket.get_tensors()[0]
     fut = dist.all_gather(

torch/distributed/algorithms/ddp_comm_hooks/powerSGD_hook.py

@@ -126,9 +126,7 @@ def powerSGD_hook(
     """
     process_group = state.process_group
     group_to_use = process_group if process_group is not None else dist.group.WORLD
-    world_size = (
-        process_group.size() if process_group is not None else dist.get_world_size()
-    )
+    world_size = group_to_use.size()
 
     # The input tensor is a flattened 1D tensor.
     input_tensor = bucket.get_tensors()[0]
@@ -363,9 +361,7 @@ def batched_powerSGD_hook(
     """
     process_group = state.process_group
     group_to_use = process_group if process_group is not None else dist.group.WORLD
-    world_size = (
-        process_group.size() if process_group is not None else dist.get_world_size()
-    )
+    world_size = process_group.size()
 
     # The input tensor is a flattened 1D tensor.
     input_tensor = bucket.get_tensors()[0]

torch/distributed/algorithms/ddp_comm_hooks/quantization_hooks.py

@@ -63,9 +63,7 @@ def quantization_pertensor_hook(
     """
     group_to_use = process_group if process_group is not None else dist.group.WORLD
     rank = process_group.rank() if process_group is not None else dist.get_rank()
-    world_size = (
-        process_group.size() if process_group is not None else dist.get_world_size()
-    )
+    world_size = group_to_use.size()
 
     tensor = bucket.get_tensors()[0]
 
@@ -144,9 +142,7 @@ def quantization_perchannel_hook(
     """
     group_to_use = process_group if process_group is not None else dist.group.WORLD
     rank = process_group.rank() if process_group is not None else dist.get_rank()
-    world_size = (
-        process_group.size() if process_group is not None else dist.get_world_size()
-    )
+    world_size = process_group.size()
 
     tensor = bucket.get_tensors()[0]