%0 = "tf.Const"() {value = dense<42> : tensor<1xi64>} : () -> tensor<1xi64>
  return
}
func.func @init2() attributes {tf_saved_model.exported_names = ["init2"]} {
  %0 = "tf.Const"() {value = dense<43> : tensor<1xi64>} : () -> tensor<1xi64>
  return
}
func.func @init3() attributes {tf_saved_model.exported_names = ["init3"]} {
  %0 = "tf.Const"() {value = dense<44> : tensor<1xi64>} : () -> tensor<1xi64>
  return
}

// CHECK-LABEL: func @transposeConv2D
func.func @transposeConv2D(%input: tensor<1x32x32x3xf32>, %filter: tensor<1x1x3x8xf32>) -> tensor<1x7x7x8xf32> {

  // CHECK: %[[ARG_PERM:.*]] = "tf.Const"() <{value = dense<[0, 3, 1, 2]> : tensor<4xi64>}>
  // CHECK: %[[ARG_TRANSPOSE:[0-9]*]] = "tf.Transpose"(%arg0, %[[ARG_PERM]])

  // CHECK: %[[CONV2D:[0-9]*]] = "tf.Conv2D"(%[[ARG_TRANSPOSE]], %arg1)
  // CHECK-SAME: data_format = "NCHW"

  func.return %2, %3 : tensor<*xi64>, tensor<*xi64>
}

// -----

// CHECK-LABEL: @not_eliminate_multinomial
func.func @not_eliminate_multinomial(%0: tensor<*xf32>, %1: tensor<*xi32>) -> (tensor<*xi64>, tensor<*xi64>) {
  // CHECK-NEXT: tf.Multinomial
  // CHECK-SAME: seed = 0
  // CHECK-NEXT: tf.Multinomial
  // CHECK-SAME: seed = 1
  // CHECK-NEXT: tf.Multinomial

  func.return %cst : tensor<1x750xf32>

  // CHECK-DAG: %0 = mhlo.constant dense<7.680000e+02> : tensor<f32>
  // CHECK: %1 = "mhlo.broadcast_in_dim"(%0) <{broadcast_dimensions = dense<> : tensor<0xi64>}> : (tensor<f32>) -> tensor<1x750xf32>
  // CHECK: return %1 : tensor<1x750xf32>
}

// CHECK-LABEL: @unfold_splat_constant_integer
func.func @unfold_splat_constant_integer() -> tensor<1x750xi32> {

module {
  func.func @main() -> tensor<1xi64> {
  %0 = "tfl.custom"() {custom_code = "stablehlo.constant", custom_option = #tfl<const_bytes : "0x76616C75650001020109010101062C022401">} : () -> tensor<1xi64>
  func.return %0 : tensor<1xi64>
  }
}

// CHECK:  module {
// CHECK-NEXT:    func @main() -> tensor<1xi64> {
// CHECK-NEXT:    %[[c0:.+]] = stablehlo.constant dense<2> : tensor<1xi64>
// CHECK-NEXT:    return %[[c0]] : tensor<1xi64>

  // CHECK-DAG: %[[SCALE_BCAST:.+]] = "mhlo.broadcast_in_dim"(%[[SCALE]]) <{broadcast_dimensions = dense<1> : tensor<1xi64>}> : (tensor<256xf32>) -> tensor<4x256xf32>
  // CHECK-DAG: %[[OFFSET_BCAST:.+]] = "mhlo.broadcast_in_dim"(%[[OFFSET]]) <{broadcast_dimensions = dense<1> : tensor<1xi64>}> : (tensor<256xf32>) -> tensor<4x256xf32>

    %4 = "tf.AddV2"(%3, %1) {device = ""} : (tensor<?xi64>, tensor<i64>) -> tensor<?xi64>
    %5 = "tf.LookupTableFindV2"(%0, %arg0, %cst) {device = ""} : (tensor<!tf_type.resource>, tensor<?x!tf_type.string>, tensor<i64>) -> tensor<*xi64>
    return %5 : tensor<*xi64>
  }

// Check that HashTable op is lifted.
// CHECK: func.func private @serving_default

// CHECK: %[[XLA_CALL_MODULE:.+]] = "tf.XlaCallModule"(%[[CUSTOM_AGGREGATOR_0]], %[[CST]])
// CHECK-SAME: _entry_function = @composite_dot_general_fn_1, _original_entry_function = "composite_dot_general_fn_1"

// -----

func.func @foldDivF64() -> tensor<2xf64> {
  %0 = stablehlo.constant dense<[2.0, 3.0]> : tensor<2xf64>
  %1 = stablehlo.constant dense<[4.0, 6.0]> : tensor<2xf64>
  %2 = stablehlo.divide %0, %1 : tensor<2xf64>
  return %2 : tensor<2xf64>
}

// CHECK-LABEL: foldDivF64
// CHECK: stablehlo.constant dense<5.000000e-01> : tensor<2xf64>

// -----

func.func @foldDivI32() -> tensor<2xi32> {

      %0 = mhlo.constant dense<1.000000e+00> : tensor<f64>
      %1 = mhlo.convert %0 : (tensor<f64>) -> tensor<i64>
      %2 = mhlo.reshape %1 : (tensor<i64>) -> tensor<1xi64>
      %3 = "mhlo.dynamic_iota"(%2) {iota_dimension = 0 : i64} : (tensor<1xi64>) -> tensor<?xi32>
      %4 = mhlo.multiply %3, %3 : tensor<?xi32>
      return %4 : tensor<?xi32>
    }
  })";
  CellReader<int64_t> legal_error(kFailedLegalizationStreamz);