avoid permuting inverse indices when not necessary (#1807)

Summary:
Pull Request resolved: #1807

if inverse indices and embedding names are in the same order + no duplicates we can avoid the index select to permute the inverse indices tensor

Reviewed By: zainhuda

Differential Revision: D55039535

fbshipit-source-id: c00a193e5a8f34914cd67c627cb88d79e02cfdc9

Author: joshuadeng

torchrec/distributed/embeddingbag.py

@@ -311,7 +311,7 @@ def __init__(
         self,
         awaitables: List[Awaitable[torch.Tensor]],
         inverse_indices: Tuple[List[str], torch.Tensor],
-        inverse_indices_permute_indices: torch.Tensor,
+        inverse_indices_permute_indices: Optional[torch.Tensor],
         batch_size_per_feature_pre_a2a: List[int],
         uncombined_embedding_dims: List[int],
         embedding_names: List[str],
@@ -331,9 +331,11 @@ def __init__(
     def _wait_impl(self) -> KeyedTensor:
         embeddings = [w.wait() for w in self._awaitables]
         batch_size = self._inverse_indices[1].numel() // len(self._inverse_indices[0])
-        indices = torch.index_select(
-            self._inverse_indices[1], 0, self._inverse_indices_permute_indices
-        )
+        permute_indices = self._inverse_indices_permute_indices
+        if permute_indices is not None:
+            indices = torch.index_select(self._inverse_indices[1], 0, permute_indices)
+        else:
+            indices = self._inverse_indices[1]
         reindex_output = torch.ops.fbgemm.batch_index_select_dim0(
             inputs=embeddings[0] if len(embeddings) == 1 else torch.cat(embeddings),
             indices=indices.view(-1),
@@ -768,25 +770,25 @@ def _create_inverse_indices_permute_indices(
             index_per_name[name.split("@")[0]]
             for name in self._uncombined_embedding_names
         ]
-        self._inverse_indices_permute_indices = _pin_and_move(
-            torch.tensor(permute_indices),
-            inverse_indices[1].device,
-        )
+        if len(permute_indices) != len(index_per_name) or permute_indices != sorted(
+            permute_indices
+        ):
+            self._inverse_indices_permute_indices = _pin_and_move(
+                torch.tensor(permute_indices),
+                inverse_indices[1].device,
+            )
 
     # pyre-ignore [14]
     def input_dist(
         self, ctx: EmbeddingBagCollectionContext, features: KeyedJaggedTensor
     ) -> Awaitable[Awaitable[KJTList]]:
+        ctx.variable_batch_per_feature = features.variable_stride_per_key()
+        ctx.inverse_indices = features.inverse_indices_or_none()
         if self._has_uninitialized_input_dist:
             self._create_input_dist(features.keys())
             self._has_uninitialized_input_dist = False
-        ctx.variable_batch_per_feature = features.variable_stride_per_key()
-        ctx.inverse_indices = features.inverse_indices_or_none()
-        if (
-            ctx.variable_batch_per_feature
-            and self._inverse_indices_permute_indices is None
-        ):
-            self._create_inverse_indices_permute_indices(ctx.inverse_indices)
+            if ctx.variable_batch_per_feature:
+                self._create_inverse_indices_permute_indices(ctx.inverse_indices)
         with torch.no_grad():
             if self._has_features_permute:
                 features = features.permute(