%13 = "tfl.square"(%6) : (tensor<f32>) -> tensor<f32>

  func.return %7, %8, %9, %10, %11, %12, %13 : tensor<f32>, tensor<f32>, tensor<f32>, tensor<f32>, tensor<f32>, tensor<f32>, tensor<f32>
}

// CHECK-LABEL: @max_with_neg_f32_max_val
// CHECK-SAME: (%[[ARG0:.+]]: tensor<f32>)
func.func @max_with_neg_f32_max_val(%arg0 : tensor<f32>) -> (tensor<f32>, tensor<f32>) {

// CHECK-NEXT: [[res:%.+]] = "quantfork.dcast"([[conv]]) : (tensor<*x!quant.uniform<i8:f32, 0.023529411764705882:-128>>) -> tensor<*xf32>

// CHECK-LABEL: testDotToDotGeneralVectorVector
func.func @testDotToDotGeneralVectorVector(%arg0: tensor<3072xf32>, %arg1: tensor<3072xf32>) -> tensor<f32> {
  %0 = "mhlo.dot"(%arg0, %arg1) : (tensor<3072xf32>, tensor<3072xf32>) -> tensor<f32>
  func.return %0 : tensor<f32>

// CHECK:      %[[RES:.*]] = "mhlo.dot_general"(%arg0, %arg1) <{
// CHECK-SAME:   dot_dimension_numbers = #mhlo.dot<
// CHECK-SAME:     lhs_contracting_dimensions = [0],

  %5 = "tfl.add"(%4, %arg3) {tac.device = "CPU", tac.inference_type = "QUANTIZED_INT8", fused_activation_function = "NONE"} : (tensor<1x384x128x!quant.uniform<i8:f32, 0.3:3>>, tensor<128x!quant.uniform<i8:f32, 0.2:-4>>) -> tensor<1x384x128x!quant.uniform<i8:f32, 0.3:-3>>
  func.return %5 : tensor<1x384x128x!quant.uniform<i8:f32, 0.3:-3>>

// CHECK:         %[[ADD:.*]] = f32[] add(f32[] %[[DIV]], f32[]
// CHECK:         %[[CLAMP:.*]] = f32[] clamp(f32[] %[[MIN:.*]], f32[] %[[ADD]],
// CHECK:         %[[ROUND:.*]] = f32[] round-nearest-even(f32[] %[[CLAMP]])
// CHECK:         %[[CONVERT_0:.*]] = u8[] convert(f32[] %[[ROUND]])
// CHECK:         %[[CONVERT_1:.*]] = s32[] convert(u8[] %[[CONVERT_0]])
// CHECK:         %[[SUB:.*]] = s32[] subtract(s32[] %[[CONVERT_1]],

    func @outside_compilation() -> tensor<f32> {
      %0 = "tf_device.cluster"() ( {
        %1 = "tf.Const"() {_xla_outside_compilation = "0", value = dense<1.0> : tensor<f32>} : () -> (tensor<f32>)
        %2 = "tf.Identity"(%1) {_xla_outside_compilation = "0"} : (tensor<f32>) -> (tensor<f32>)
        %3 = "tf.AddV2"(%1, %2) : (tensor<f32>, tensor<f32>) -> (tensor<f32>)
        tf_device.return %3 : tensor<f32>

  %0:3 = "tf.CaseRegion"(%index) ({
      %mul = "tf.Mul"(%arg0, %arg1) : (tensor<f32>, tensor<f32>) -> tensor<f32>
      "tf.Yield"(%arg1, %mul, %arg2) : (tensor<f32>, tensor<f32>, tensor<!tf_type.resource>) -> ()
    }, {
      %sub = "tf.Sub"(%arg0, %arg1) : (tensor<f32>, tensor<f32>) -> tensor<f32>
      "tf.Yield"(%arg1, %sub, %arg2) : (tensor<f32>, tensor<f32>, tensor<!tf_type.resource>) -> ()
    }, {

  %0 = "tf.XlaAllReduce"(%input, %group_assignment) {reduce_op = "Add", mode = "CrossReplica"} : (tuple<tensor<f32>, tensor<f32>>, tensor<2x1xi32>) -> tuple<tensor<f32>, tensor<f32>>
  func.return %0 : tuple<tensor<f32>, tensor<f32>> 
}


// -----

// CHECK: module attributes

func.func private @f(%arg0: f32, %arg1: f32) -> (f32, f32) {
  %0 = "some_op"(%arg0) : (f32) -> f32
  return %0, %arg1 : f32, f32
}

// CHECK-LABEL: @removes_second_passthrough_arg
func.func @removes_second_passthrough_arg(%arg0: f32, %arg1: f32) -> (f32, f32) {
  // CHECK: %0 = call @f(%arg0)
  %0, %1 = call @f(%arg0, %arg1) : (f32, f32) -> (f32, f32)
  // CHECK: return %0, %arg1
  return %0, %1 : f32, f32
}

// -----

    // CHECK-DAG: %[[CST_1:.*]] = "tf.Const"() <{value = dense<3> : tensor<i32>}> : () -> tensor<i32>
    // CHECK-NEXT: %[[UQ:.*]] = "tf.UniformQuantize"(%arg0, %cst, %cst_0) <{quantization_axis = -1 : i64, quantization_max_val = 127 : i64, quantization_min_val = -128 : i64}> : (tensor<1xf32>, tensor<f32>, tensor<i32>) -> tensor<1x!tf_type.qint8>