Confirmed that pytests pass

python/cugraph-dgl/cugraph_dgl/dataloading/dataloader.py

@@ -211,8 +211,12 @@ def __iter__(self):
             self._sampling_output_dir, "epoch_" + str(self.epoch_number)
         )
         if isinstance(self.cugraph_dgl_dataset, HomogenousBulkSamplerDataset):
+            deduplicate_sources = True
+            prior_sources_behavior = "carryover"
             renumber = True
         else:
+            deduplicate_sources = False
+            prior_sources_behavior = None
             renumber = False
 
         bs = BulkSampler(
@@ -223,6 +227,8 @@ def __iter__(self):
             seeds_per_call=self._seeds_per_call,
             fanout_vals=self.graph_sampler._reversed_fanout_vals,
             with_replacement=self.graph_sampler.replace,
+            deduplicate_sources=deduplicate_sources,
+            prior_sources_behavior=prior_sources_behavior,
             renumber=renumber,
         )
         if self.shuffle:

python/cugraph-dgl/cugraph_dgl/dataloading/utils/sampling_helpers.py

@@ -168,10 +168,14 @@ def _create_homogeneous_sampled_graphs_from_tensors_perhop(tensors_batch_d, edge
     if edge_dir == "out":
         raise ValueError("Outwards edges not supported yet")
     graph_per_hop_ls = []
+    seednodes = None
     for hop_id, tensor_per_hop_d in tensors_batch_d.items():
         if hop_id != "map":
             block = _create_homogeneous_dgl_block_from_tensor_d(
-                tensor_per_hop_d, tensors_batch_d["map"]
+                tensor_per_hop_d, tensors_batch_d["map"], seednodes
+            )
+            seednodes = torch.concat(
+                [tensor_per_hop_d["sources"], tensor_per_hop_d["destinations"]]
             )
             graph_per_hop_ls.append(block)
 
@@ -184,10 +188,28 @@ def _create_homogeneous_sampled_graphs_from_tensors_perhop(tensors_batch_d, edge
     return input_nodes, output_nodes, graph_per_hop_ls
 
 
-def _create_homogeneous_dgl_block_from_tensor_d(tensor_d, renumber_map):
+def _create_homogeneous_dgl_block_from_tensor_d(tensor_d, renumber_map, seednodes=None):
     rs = tensor_d["sources"]
     rd = tensor_d["destinations"]
-    block = dgl.create_block((rs, rd))
+
+    max_src_nodes = rs.max()
+    max_dst_nodes = rd.max()
+    if seednodes is not None:
+        # If we have isolated vertices
+        # sources can be missing from seednodes
+        # so we add them
+        # to ensure all the blocks are
+        # linedup correctly
+        max_dst_nodes = max(max_dst_nodes, seednodes.max())
+
+    data_dict = {("_N", "_E", "_N"): (rs, rd)}
+    num_src_nodes = {"_N": max_src_nodes.item() + 1}
+    num_dst_nodes = {"_N": max_dst_nodes.item() + 1}
+    print("num_src_nodes", num_src_nodes)
+    print("num_dst_nodes", num_dst_nodes)
+    block = dgl.create_block(
+        data_dict=data_dict, num_src_nodes=num_src_nodes, num_dst_nodes=num_dst_nodes
+    )
     if "edge_id" in tensor_d:
         block.edata[dgl.EID] = tensor_d["edge_id"]
     block.srcdata[dgl.NID] = renumber_map[block.srcnodes()]

python/cugraph-dgl/examples/dataset_from_disk_cudf.ipynb

@@ -135,7 +135,7 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "7.32 s ± 778 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
+      "7.08 s ± 596 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
      ]
     }
    ],
@@ -217,7 +217,7 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "7.39 s ± 524 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
+      "7.34 s ± 353 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
      ]
     }
    ],