module attributes {tf.versions = {bad_consumers = [], min_consumer = 12 : i32, producer = 1219 : i32}, tf_saved_model.semantics} {

  func.func @add(%arg0: tensor<?xf32> {tf_saved_model.index_path = ["input1"]}, %arg1: tensor<?xf32> {tf_saved_model.index_path = ["input2"]}) -> (tensor<?xf32> {tf_saved_model.index_path = ["result"]}) attributes {tf.entry_function = {control_outputs = "", inputs = "input1:0,input2:0", outputs = "result:0"}, tf_saved_model.exported_names = ["add"]} {

TEST_F(ConvertAssetArgsTest, ConvertsSingleAssetArg) {
  OwningOpRef<ModuleOp> module_op = ParseModuleOpString(R"mlir(
    module {
      "tf_saved_model.asset"() {filename = "assets/file_0.txt", sym_name = "__tf_saved_model_asset0"} : () -> ()
      func.func @main(%arg_0: tensor<!tf_type.string> {tf_saved_model.bound_input = @__tf_saved_model_asset0}) -> () attributes {tf.entry_function = {inputs = "arg_0:0", outputs = ""}} {
        return
      }
    }

# CHECK-SAME: producer

# CHECK: "tf_saved_model.global_tensor"() <{is_mutable, sym_name = "[[VAR:[a-zA-Z_0-9]+]]", type = tensor<1x3xf32>, value = {{.*}} : tensor<1x3xf32>}> : () -> ()

# CHECK:      func {{@[a-zA-Z_0-9]+}}(
# CHECK-SAME:   [[ARG0:%.*]]: tensor<3x1xf32> {tf_saved_model.index_path = ["x"]},
# CHECK-SAME:   [[ARG1:%.*]]: tensor<!tf_type.resource<tensor<1x3xf32>>> {tf_saved_model.bound_input = @[[VAR]]})

from tensorflow.compiler.mlir.tensorflow.tests.tf_saved_model import common


class TestModule(tf.Module):

  def __init__(self):
    super(TestModule, self).__init__()
    # CHECK: "tf_saved_model.global_tensor"() <{is_mutable, {{.*}} type = tensor<*xf32>, value = dense<0.000000e+00> : tensor<1xf32>}> {tf_saved_model.exported_names = ["v0"]} : () -> ()

module attributes {tf.versions = {bad_consumers = [], min_consumer = 12 : i32, producer = 975 : i32}, tf_saved_model.semantics}  {
  func.func @serving_default(%arg0: tensor<1x20x20x28xf32> {tf_saved_model.index_path = ["a"]}) -> (tensor<1x40x40x28xf32> {tf_saved_model.index_path = ["b"]}) attributes {tf.entry_function = {control_outputs = "", inputs = "c:0", outputs = "d:0"}, tf_saved_model.exported_names = ["serving_default"]} {
      %c = stablehlo.constant dense<40> : tensor<2xi32>

// CHECK-NOT: tf_saved_model.semantics
// CHECK: tfrt.cost_threshold = 1024
// CHECK-SAME: tfrt.merge_inter_dependent_streams = true
module attributes {tf_saved_model.semantics} {

// CHECK-NOT: "tf_saved_model.session_initializer"
"tf_saved_model.session_initializer"() { initializers = [@func_init] } : () -> ()

// CHECK-LABEL: _tfrt_resource_init

  func.func @serving_default(%arg0: tensor<1024x1024xf32> {tf_saved_model.index_path = ["input_tensor"]}) -> (tensor<1024x3xf32> {tf_saved_model.index_path = ["output1"]}, tensor<1024x3xf32> {tf_saved_model.index_path = ["output2"]}) attributes {tf.entry_function = {control_outputs = "", inputs = "serving_default_input_tensor:0", outputs = "PartitionedCall:0"}, tf_saved_model.exported_names = ["serving_default"]} {

    (i.e. `obj[0]["x"]`).

    The function arg attribute
    `tf_saved_model.bound_input = @some_symbol_name` references a
    `tf_saved_model.global_tensor` that is bound to that function argument
    and implicitly passed in when the function is invoked from a user-level API.
    TODO(silvasean): Consider replacing tf_saved_model.bound_input arg attrs
    with "get_global @some_global_tensor" in the function body.
  }];

#include "mlir/Support/LLVM.h"  // from @llvm-project
#include "tensorflow/compiler/mlir/tensorflow/ir/tf_ops.h"
#include "tensorflow/compiler/mlir/tensorflow/ir/tf_saved_model.h"

namespace mlir {
namespace tf_saved_model {
namespace {

#define GEN_PASS_DEF_DEDUPBOUNDINPUTBINDINGPASS
#include "tensorflow/compiler/mlir/tensorflow/transforms/tf_savedmodel_passes.h.inc"

class DedupBoundInputBindingPass