[Bug] Fix ddp bug when pytorch>=1.7 (#769)

* fix ddp bug when pytorch>=1.7

* fix ddp bug when pytorch>=1.7

* support pat

* fix docs

* use print_log instead of get_logger
Co-authored-by: nbei <631557085@qq.com>

mmcv/parallel/distributed.py

@@ -3,6 +3,7 @@ import torch
 from torch.nn.parallel.distributed import (DistributedDataParallel,
                                            _find_tensors)
 
+from mmcv import print_log
 from mmcv.utils import TORCH_VERSION
 from .scatter_gather import scatter_kwargs
 
@@ -28,6 +29,15 @@ class MMDistributedDataParallel(DistributedDataParallel):
         ``self.module.forward()`` with ``self.module.train_step()``.
         It is compatible with PyTorch 1.1 - 1.5.
         """
+
+        # In PyTorch >= 1.7, ``reducer._rebuild_buckets()`` is moved from the
+        # end of backward to the beginning of forward.
+        if (TORCH_VERSION >= '1.7' and 'parrots'
+                not in TORCH_VERSION) and self.reducer._rebuild_buckets():
+            print_log(
+                'Reducer buckets have been rebuilt in this iteration.',
+                logger='mmcv')
+
         if getattr(self, 'require_forward_param_sync', True):
             self._sync_params()
         if self.device_ids:
@@ -60,6 +70,14 @@ class MMDistributedDataParallel(DistributedDataParallel):
         ``self.module.forward()`` with ``self.module.val_step()``.
         It is compatible with PyTorch 1.1 - 1.5.
         """
+        # In PyTorch >= 1.7, ``reducer._rebuild_buckets()`` is moved from the
+        # end of backward to the beginning of forward.
+        if (TORCH_VERSION >= '1.7' and 'parrots'
+                not in TORCH_VERSION) and self.reducer._rebuild_buckets():
+            print_log(
+                'Reducer buckets have been rebuilt in this iteration.',
+                logger='mmcv')
+
         if getattr(self, 'require_forward_param_sync', True):
             self._sync_params()
         if self.device_ids: