const Edge* e;
    TF_RETURN_IF_ERROR(dst->input_edge(edges[i].dst_input, &e));
    Node* src = e->src();
    int src_output = e->src_output(), dst_input = e->dst_input();
    g->RemoveEdge(e);

    // Find or create placeholder node.
    string new_name =
        absl::StrCat(src->name(), "_oc_to_oc_placeholder_", src_output);
    auto placeholder_index = std::make_pair(src->name(), src_output);

  // attribute of the MLIR op. The assigned device might be remote, not
  // available during compilation or compilation only device for on demand
  // execution which may create a recursion if used for constant folding.
  constexpr char kHostCpu[] = "/job:localhost/replica:0/task:0/CPU:0";
  TFE_OpSetDevice(op, kHostCpu, status);
  RETURN_FAILURE_IF_ERROR(status);

}

}  // namespace

Status MapFunctionSignaturesFromTFSavedModelMLIR(
    mlir::ModuleOp module,
    llvm::function_ref<void(const TFRTSavedModelSignatureInfo&)> map_fn) {
  // Create bound inputs for each functions.
  mlir::SymbolTable symbol_table(module);
  tensorflow::Status status = absl::OkStatus();
  module.walk([&symbol_table, map_fn, &status](mlir::func::FuncOp func) {

// not revisited in future calls to ScanForValue, so callers must take
// care to order their calls.
//
// Useful for merging multiple sorted lists in O(n) time.
class SinglePassSearch {
 public:
  // Creates a SinglePassSearch object that can be used to search in `values`.
  // Does not take ownership of `values`. `values` must outlive this.
  // `values` must be sorted.
  explicit SinglePassSearch(absl::Span<int const> values)

        func.emitError() << "Failed to duplicate values for the stateful op\n";
        return signalPassFailure();
      }
    }
  }
}

}  // namespace

/// Creates an instance of the TensorFlow Lite dialect SplitMergedOperands
/// pass.
std::unique_ptr<OperationPass<func::FuncOp>> CreateSplitMergedOperandsPass() {
  return std::make_unique<SplitMergedOperandsPass>();
}

// the cluster.
//
// Declustering f will increase the number of compilations only if f is a
// one-to-many "function" i.e. isn't a function at all.  RNG is one possible
// example, depending on how we look at it.  But we never create clusters where
// such f's would be marked as must-be-constant.
//
// We assume here that the extra repeated (repeated compared to a clustered f
// where it will always be constant folded) host-side computation of f does not

  }

  // Rewrite return if there are variable writes.
  const int return_operands_size = return_operands.size();
  if (return_operands_size > num_results_before) {
    builder.create<func::ReturnOp>(return_op.getLoc(), return_operands);
    return_op.erase();
  }

  // Update function argument and result types with new resource subtypes.

  AddShapeLegalizationPasses(pm);
  pm.addPass(createReplaceStablehloOpsInMainFunctionWithXlaCallModuleOpsPass());
  // ReplaceStablehloOpsInMainFunctionWithXlaCallModuleOpsPass may create
  // duplicate constants. Add canonicalizer to deduplicate.
  pm.addNestedPass<func::FuncOp>(mlir::createCanonicalizerPass());
  pm.addPass(TF::CreateXlaCallModuleSerializationPass());
}

    Value value = effect.getValue();
    found_any_effect = true;

    // We only collect value-based side effects here for which we can use
    // resource alias analysis. Other side effects are treated as op-based
    // side effects.
    if (!ShouldUseResourceAliasAnalysis(effect)) continue;
    if (value) processed_values.insert(value);

    TypeID type_id = effect.getResource()->getResourceID();

#include "tensorflow/core/platform/status.h"
#include "tensorflow/core/public/session_options.h"
#include "tsl/platform/statusor.h"

#define DEBUG_TYPE "run-tf-graph-optimization"

namespace tensorflow {
namespace {
// Creates a pass to convert MLIR to Graph, run user-specified Graph
// Optimization Passes and convert back to MLIR.
// Constraints: This pass expects that all operations in the MLIR module either