//   purposes and doesn't affect logic. module_name - What the input module name
//   is for debugging help.
//
// Output: Modifies the input module in place with clustered operations.
//   status - Whether the transformation to cluster the input MLIR module was
//   successful.
tensorflow::Status RunFunctionTf2xlaClusteringBridge(
    mlir::ModuleOp module, bool is_supported_by_replicated_brige,

      cluster.constraints = std::move(member.constraints);

    // Add operation to the cluster.
    if (auto op = member.source.dyn_cast<Operation *>())
      cluster.operations.emplace_back(op);
  }

  llvm::SmallVector<Cluster> clusters;
  for (auto &kv : root_clusters) {
    Cluster &cluster = kv.getSecond();
    // Skip degenerate clusters formed by a single basic block argument.

Clustered nodes: 2178
Unclustered nodes: 446
Number of clusters: 1

unclustered size 446
 AssignAddVariableOp 2
 Const 2
 DivNoNan 1
 Identity 1
 NoOp 1
 ReadVariableOp 2
 VarHandleOp 435
 _Retval 2
cluster 0 size 2178
 Add 17
 AddN 72
 ArgMax 1
 AssignAddVariableOp 1
 AssignSubVariableOp 106
 BiasAdd 1
 BiasAddGrad 1
 Cast 3
 Const 357
 Conv2D 53
 Conv2DBackpropFilter 53
 Conv2DBackpropInput 52

Clustered nodes: 2725
Unclustered nodes: 606
Number of clusters: 2

unclustered size 606
 AssignAddVariableOp 1
 Const 108
 DivNoNan 1
 Identity 2
 Merge 53
 PlaceholderWithDefault 1
 ReadVariableOp 2
 Switch 1
 _Arg 435
 _Retval 2
cluster 0 size 1910
 Add 16
 AddN 71
 ArgMax 1
 AssignAddVariableOp 1
 BiasAdd 1
 BiasAddGrad 1
 Cast 115
 Const 407
 Conv2D 53
 Conv2DBackpropFilter 53

llvm::SmallVector<Cluster> FindClustersInTheBlock(
    Block* block, const ClusteringPolicySet& policies,
    std::function<bool(Operation* op)> filter = {});

// Creates a `tf_device.cluster` operation from the clustered operations.
tf_device::ClusterOp CreateClusterOp(Cluster& cluster, StringAttr policy = {});

// -------------------------------------------------------------------------- //

// ```
//
// Assuming `tf.OpA` and `tf.OpC` are clustered together, the region will be
// inserted right after `tf.OpC`. The live-out `%0`, however, is used by
// `tf.OpB`, which won't dominate the region. This function reorders all users
// of the cluster op to be placed after the cluster op itself so that SSA
// dominance is preserved after cluster op creation.
void ReorderOpResultUses(mlir::Operation* cluster);

}  // namespace mlir::TF

      TF_RET_CHECK(absl::SimpleAtoi(*maybe_cluster, &cluster));
      clustered_nodes++;
    }
    clusters[cluster][n->type_string()]++;
    cluster_size[cluster]++;
  }

  string result =
      absl::StrCat("Clustered nodes: ", clustered_nodes,
                   "\nUnclustered nodes: ", cluster_size[kNoCluster],
                   "\nNumber of clusters: ", clusters.size() - 1, "\n\n");

// file name suggests, these are graphdefs that have been dumped right before
// the mark_for_compilation pass.  There should be one
// before_mark_for_compilation_<N>.pbtxt for every TF graph that was
// auto-clustered, out of which usually only one is the "main" graph that's
// running training/inference.
//
// Copy the pbtxt for that "main" graph to tensorflow/compiler/jit/tests/

    // We don't want to decluster F in a graph like
    //
    //   Input -> OP -> Shape -> F -> Reshape
    //
    // Doing so will break up the cluster.  Even if we were okay with breaking
    // up the cluster we will at least have to relabel the two clusters to have
    // different cluster names.
    //
    // We may want to revisit this in the future: we may have cases where OP is

  auto clusters = GetClusters(*graph);
  EXPECT_EQ(4, clusters.size());
  EXPECT_EQ(clusters["B"], clusters["C"]);
  EXPECT_EQ(clusters["E"], clusters["F"]);
  EXPECT_NE(clusters["B"], clusters["E"]);
  EXPECT_TRUE(clusters.find("A") == clusters.cend());
  EXPECT_TRUE(clusters.find("D") == clusters.cend());
}

TEST(XlaCompilationTest, UncompilableCycles) {