func.func @valid_unranked_inputs_on_reshape(%arg0: tensor<3x4xi32>, %arg1: tensor<*xi32>) -> tensor<3x4xi32> {
  // CHECK: "tfl.reshape"(%arg0, %arg1)
  %0 = "tfl.reshape"(%arg0, %arg1) : (tensor<3x4xi32>, tensor<*xi32>) -> tensor<3x4xi32>
  func.return %0 : tensor<3x4xi32>
}

// -----

// CHECK-LABEL: valid_one_dynamic_dim_on_reshape
func.func @valid_one_dynamic_dim_on_reshape(%arg0: tensor<3x4xi32>) -> tensor<1x3x4xi32> {

// `tfl.slice` when stride is 1.

func.func @slice(
    %arg0: tensor<3x4x!quant.uniform<i8:f32, 2.000000e+00:-1>>
  ) -> tensor<2x2x!quant.uniform<i8:f32, 2.000000e+00:-1>> {
  %0 = "stablehlo.slice"(%arg0) {
    start_indices = array<i64: 1, 2>,
    limit_indices = array<i64: 3, 4>,
    strides = array<i64: 1, 1>
  } : (
    tensor<3x4x!quant.uniform<i8:f32, 2.000000e+00:-1>>

}

func.func @where(%arg0: tensor<3x5xi1>) -> tensor<?x2xi64> {
  %0 = "tf.Where"(%arg0) : (tensor<3x5xi1>) -> tensor<?x2xi64>
  func.return %0 : tensor<?x2xi64>
  // CHECK-LABEL: where
  // CHECK: "tfl.where"(%arg0) : (tensor<3x5xi1>) -> tensor<?x2xi64>
}

func.func @floor_mod(%arg0: tensor<5xf32>, %arg1: tensor<5xf32>) -> tensor<5xf32> {

  func.return %0 : tensor<4x2x!tf_type.string>
}

// -----

// CHECK-LABEL: func @testBitcast
func.func @testBitcast(%arg0: tensor<3x4xui16>) -> tensor<3x4x!tf_type.quint16> {
  %0 = "tf.Bitcast"(%arg0) : (tensor<3x4xui16>) -> tensor<3x4x!tf_type.quint16>
  func.return %0 : tensor<3x4x!tf_type.quint16>
}

// -----

// CHECK-LABEL: func @testBitcast_v2
func.func @testBitcast_v2(%arg0: tensor<3x2xi16>) -> tensor<3xi32> {

}

// Returns a RankedTensorType which is similar to `input_type` but replaces the
// dimension size of `dim` with `dim_size`.  For example,
// `SubstituteRankedTensorTypeDimSize(tensor<3x4xi32>, 1, 2)` returns
// `tensor<3x2xi32>`.
static RankedTensorType SubstituteRankedTensorTypeDimSize(
    RankedTensorType input_type, int64_t dim, int64_t dim_size) {
  auto shape = input_type.getShape().vec();