// CHECK:           return %[[VAL_2]] : tensor<1x4xi8>
// CHECK:         }
func.func @bitwise_or_broadcast_chlo(%arg0: tensor<1xi8>, %arg1: tensor<1x4xi8>) -> tensor<1x4xi8> {
  %0 = "chlo.broadcast_or"(%arg0, %arg1) {broadcast_dimensions = array<i64: 1>} : (tensor<1xi8>, tensor<1x4xi8>) -> tensor<1x4xi8>
  func.return %0 : tensor<1x4xi8>
}

// CHECK-LABEL:   func @bitwise_or_dynamic(

    %0 = stablehlo.convert %arg0 : (tensor<1x2xf32>) -> tensor<1x2xi8>
    return %0 : tensor<1x2xi8>
  }
// CHECK: @uniform_quantize_0
  func.func private @uniform_quantize_1(%arg0: tensor<1x3xf32>, %arg1: tensor<1x1xf32>, %arg2: tensor<1x1xi8>) -> tensor<1x3xi8> {
    %0 = stablehlo.convert %arg0 : (tensor<1x3xf32>) -> tensor<1x3xi8>
    return %0 : tensor<1x3xi8>
  }

    %6 = "tf.Identity"(%5) {device = ""} : (tensor<2x3xi8>) -> tensor<2x3xi8>
    %7 = "tf.Cast"(%6) {Truncate = false, device = ""} : (tensor<2x3xi8>) -> tensor<2x3xi32>
    %8 = "tf.Sub"(%7, %cst_4) {device = ""} : (tensor<2x3xi32>, tensor<i32>) -> tensor<2x3xi32>
    %9 = "tf.Identity"(%cst_2) {device = ""} : (tensor<3x4xi8>) -> tensor<3x4xi8>

  // CHECK: %[[BROADCAST:.*]] = "mhlo.broadcast_in_dim"(%arg0) <{broadcast_dimensions = dense<1> : tensor<1xi64>}> : (tensor<4xui8>) -> tensor<2x4xui8>
  // CHECK:  mhlo.shift_right_logical %[[BROADCAST]], %arg1 : tensor<2x4xui8>
  %0 = "tf.RightShift"(%arg0, %arg1) : (tensor<4xui8>, tensor<2x4xui8>) -> tensor<2x4xui8>
  func.return %0 : tensor<2x4xui8>
}

    %1 = stablehlo.uniform_quantize %arg0 : (tensor<1x4xf32>) -> tensor<1x4x!quant.uniform<i8:f32, 6.000000e-03:-128>>
    // CHECK: call @quantized_dot_general_relu_fn
    // CHECK-SAME: -> tensor<1x3xf32>

  // CHECK: return %[[ADD]] : tensor<14xf32>
}

func.func @scatter_nd_updates_bool(%arg0: tensor<1x24xi1>, %arg1: tensor<1x2x2xi32>, %arg2: tensor<1x2xi1>) -> tensor<1x24xi1> {
  %0 = "tf.TensorScatterUpdate"(%arg0, %arg1, %arg2) : (tensor<1x24xi1>, tensor<1x2x2xi32>, tensor<1x2xi1>) -> tensor<1x24xi1>
  func.return %0 : tensor<1x24xi1>

// CHECK-LABEL: scatter_nd_updates_bool(

  // CHECK: %[[WHILE_RESULT:.*]]:2 = mhlo.while(%[[ARG0:.*]] = %[[INT0]], %[[ARG1:.*]] = %[[INT1]])
  // CHECK-SAME: tensor<2x2xi8>, tensor<2x2xi8>

  // CHECK: cond

  // CHECK: do
  // CHECK: mhlo.return %[[ARG0]], %[[ARG1]] : tensor<2x?xi8>, tensor<?x2xi8>

  %2:2 = "tf.WhileRegion"(%0, %1) ({
  ^bb0(%carg0: tensor<2x2x!tf_type.qint8>, %carg1: tensor<2x2x!tf_type.qint8>):

  %cst = "tfl.pseudo_qconst"() {qtype = tensor<1x2x!quant.uniform<u8:f32, 1.0>>, value = dense<-76> : tensor<1x2xi8>} : () -> tensor<1x2x!quant.uniform<u8:f32, 1.0>>

  // CHECK: return %[[RESULT]] : tensor<1x4xi32>
  %sizes = "tf.Const"() {value = dense<[1, 4]> : tensor<2xi64>} : () -> (tensor<2xi64>)
  %0 = "tf.Slice"(%arg0, %arg1, %sizes) : (tensor<3x4xi32>, tensor<2xi64>, tensor<2xi64>) -> tensor<1x4xi32>
  func.return %0 : tensor<1x4xi32>
}

// CHECK-LABEL: slice_variable_start_negsize

}

TEST_F(ConvertTfQuantToMhloIntTest, UniformRequantizeToValidGraph) {
  constexpr absl::string_view kProgram = R"mlir(
func.func @main(%input: tensor<10xi8>) -> tensor<10xi8> {
  %input_scale = "tf.Const"() { value = dense<0.2235> : tensor<f32> } : () -> tensor<f32>
  %input_zp = "tf.Const"() { value = dense<-2> : tensor<i32> } : () -> tensor<i32>