Add back torch load_library inplace update op (#1581)

Summary:
Pull Request resolved: #1581

Load inplace update dependencies after BC issue is resolved.

Reviewed By: brad-mengchi

Differential Revision: D43026352

fbshipit-source-id: b365535ca5896e88eaac95bab72afe6a7fb34733

Author: jianyuh

fbgemm_gpu/bench/split_table_batched_embeddings_benchmark.py

@@ -2513,7 +2513,7 @@ def emb_inplace_update(  # noqa C901
     logging.info(
         f"Emb inplace update (including H2D for metadata): "
         f"T: {T}, D: {D}, E: {E}, N: {N}, "
-        f"BW: {read_write_bytes / time_per_iter / 1.0e9: .2f} GB/s, "  # noqa: B950
+        f"BW: {read_write_bytes / time_per_iter / 1.0e9:.2f} GB/s, "  # noqa: B950
         f"T: {time_per_iter * 1.0e6:.0f}us"
     )
 
@@ -2566,7 +2566,7 @@ def emb_inplace_update(  # noqa C901
     logging.info(
         f"Emb inplace update (pure device update op): "
         f"T: {T}, D: {D}, E: {E}, N: {N}, "
-        f"BW: {read_write_bytes / time_per_iter / 1.0e9: .2f} GB/s, "  # noqa: B950
+        f"BW: {read_write_bytes / time_per_iter / 1.0e9:.2f} GB/s, "  # noqa: B950
         f"T: {time_per_iter * 1.0e6:.0f}us"
     )
 

fbgemm_gpu/codegen/split_embedding_codegen_lookup_invoker.template

@@ -18,13 +18,8 @@ torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:sparse_ops_cpu")
 torch.ops.load_library(
     "//deeplearning/fbgemm/fbgemm_gpu:split_table_batched_embeddings"
 )
-# try:
-#     torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:embedding_inplace_update")
-#     torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:embedding_inplace_update_cpu")
-# except OSError:
-#     # Keep for BC: will be deprecated soon.
-#     torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu/fb:embedding_inplace_update")
-#     torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu/fb:embedding_inplace_update_cpu")
+torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:embedding_inplace_update")
+torch.ops.load_library("//deeplearning/fbgemm/fbgemm_gpu:embedding_inplace_update_cpu")
 
 {% else %}
 #import os

fbgemm_gpu/test/split_table_batched_embeddings_test.py

@@ -4905,9 +4905,8 @@ def test_embedding_inplace_update(
         )
 
         weights_ty_list = [weights_ty] * T
-        # if open_source:
-        #     test_internal = False
-        test_internal = False
+        if open_source:
+            test_internal = False
 
         # create two embedding bag op with random weights
         locations = [location] * T