%1 = "tf.Add"(%0, %arg5) : (tensor<8x8x40xf32>, tensor<40xf32>) -> tensor<8x8x40xf32>
  %2 = "tf.BatchMatMulV2"(%1, %arg4) {adj_x = false, adj_y = true} : (tensor<8x8x40xf32>, tensor<10x40xf32>) -> tensor<8x8x10xf32>
  %3 = "tf.Add"(%2, %arg1) : (tensor<8x8x10xf32>, tensor<8x10xf32>) -> tensor<8x8x10xf32>

func.func @invalidSelect(%arg0: tensor<1x8xi1>, %arg1: tensor<1x8x8xi32>, %arg2: tensor<1x8x8xi32>) -> tensor<1x8x8xi32> {
  // expected-error @+1 {{requires that pred is a scalar OR has the same rank as t and e OR is a vector}}
  %0 = "tf.Select"(%arg0, %arg1, %arg2) : (tensor<1x8xi1>, tensor<1x8x8xi32>, tensor<1x8x8xi32>) -> tensor<1x8x8xi32>
  func.return %0: tensor<1x8x8xi32>
}

// -----

  func.return %0#0 : tensor<8x8x8x8xf32>
}

// -----

// CHECK-LABEL: fusedBatchNormGrad_Training

  func.return %0#0, %1#0 : tensor<8x8x8x8xf32>, tensor<8x8x8x8xf32>
}
// CHECK: func @customAggregator
// CHECK-NEXT: %[[stats:.*]] = "quantfork.stats"(%arg0) <{layerStats = dense<[-1.000000e-01, 2.000000e-01]> : tensor<2xf32>}> : (tensor<8x8x8x8xf32>) -> tensor<8x8x8x8xf32>
// CHECK-NEXT: return %[[stats]], %arg0

  %gather = "tf.TensorListGather"(%tl, %arg1, %elem_shape) : (tensor<!tf_type.variant<tensor<8x9xf32>>>, tensor<3xi32>, tensor<2xi32>) -> tensor<3x8x9xf32>
  // CHECK-NEXT: return %[[GATHER]] : tensor<3x8x9xf32>
  func.return %gather: tensor<3x8x9xf32>
}

// -----

// Test scatter into existing tensor list.

module {
    func.func @main(%arg0: tensor<8x128xf32>, %arg1: tensor<f32>) -> tensor<11x131xf32> {
      %0 = "tfl.custom"(%arg0, %arg1) {custom_code = "stablehlo.pad", custom_option = #tfl<const_bytes : "0x656467655F70616464696E675F6869676800020203656467655F70616464696E675F6C6F7700020100696E746572696F725F70616464696E6700020000033E2A17030103311E0B2C2C2C062401">} : (tensor<8x128xf32>, tensor<f32>) -> tensor<11x131xf32>

  %relu = "tf.Relu"(%add) : (tensor<8x8x8x8xf32>) -> tensor<8x8x8x8xf32>
  func.return %relu : tensor<8x8x8x8xf32>
}

// CHECK-LABEL: func @FusedBatchNormAddReluTwoUses
func.func @FusedBatchNormAddReluTwoUses(%arg0: tensor<8x8x8x8xf32>, %arg1: tensor<8xf32>, %arg2: tensor<8xf32>, %arg3: tensor<8xf32>, %arg4: tensor<8xf32>) -> (tensor<8x8x8x8xf32>, tensor<8x8x8x8xf32>) {

func.func @fakeQuantArgs(%arg0: tensor<8x8x8x8xf32>) -> tensor<8x8x8x8xf32> {
  %0 = "tf.FakeQuantWithMinMaxArgs"(%arg0) {
    min = -0.1 : f32, max = 0.2 : f32, num_bits = 8
  } : (tensor<8x8x8x8xf32>) -> tensor<8x8x8x8xf32>
  func.return %0 : tensor<8x8x8x8xf32>
}
// CHECK: func @fakeQuantArgs
// CHECK-NEXT: %[[q:.*]] = "quantfork.qcast"(%arg0) : (tensor<8x8x8x8xf32>) -> tensor<8x8x8x8x!quant.uniform<i8:f32, 0.0011764706057660721:-43>>

// CHECK-NEXT: return %[[dq]] : tensor<2x2xf32>
}

// CHECK-LABEL: prepareStatistics
func.func @prepareStatistics(%arg0: tensor<8x4x3xf32>) -> tensor<8x4x3xf32> {
  %0 = "quantfork.stats"(%arg0) {
    layerStats = dense<[-1.0, 1.0]> : tensor<2xf32>
  } : (tensor<8x4x3xf32>) -> tensor<8x4x3xf32>
  %1 = "quantfork.stats"(%0) {
    layerStats = dense<[-1.0, 1.0]> : tensor<2xf32>,
    axisStats = dense<[
      [-1.0, 1.0],

      tf_device.return %2 : tensor<1x8x2xf32>
    // CHECK: () -> tensor<1x8x2xf32>
    }) {device = "/device:CPU:0"} : () -> tensor<*xf32>
    // CHECK: "tf.Cast"(%{{.*}}) <{Truncate = false}> : (tensor<1x8x2xf32>) -> tensor<*xf32>
    // CHECK: (tensor<i32>, tensor<1x8x2xf32>) -> (tensor<1x8x1xf32>, tensor<1x8x1xf32>)