func.func @multiply(%arg0: tensor<1xf32>, %arg1: tensor<1xf32>) -> tensor<1xf32> {
  %0 = "vhlo.multiply_v1"(%arg0, %arg1) : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>
  func.return %0 : tensor<1xf32>
}

// CHECK:func.func private @multiply(%arg0: tensor<1xf32>, %arg1: tensor<1xf32>) -> tensor<1xf32> {
// CHECK: %0 = "vhlo.multiply_v1"(%arg0, %arg1) : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>
// CHECK: return %0 : tensor<1xf32>
// CHECK:}

    %2 = "tf.MatMul"(%arg2, %arg1) {} : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>
    %3 = "tf.AddV2" (%arg1, %arg1)  : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>
    %4 = "tf.Identity"(%1) {device = ""} : (tensor<i32>) -> tensor<i32>
    %5 = "tf.Identity"(%3) {device = ""} : (tensor<*xf32>) -> tensor<*xf32>
    %6 = "tf.Identity"(%2) {device = ""} : (tensor<*xf32>) -> tensor<*xf32>

// CHECK-LABEL: QuantizeCustomOp
// CustomOp-LABEL: QuantizeCustomOp
func.func @QuantizeCustomOp(%arg0: tensor<1x1x1x1xf32>) -> (tensor<*xf32>, tensor<*xf32>, tensor<*xf32>) attributes {tf.entry_function = {inputs = "input", outputs = "custom_op"}} {
  %0 = "quantfork.stats"(%arg0) {layerStats = dense<[0.000000e+00, 2.550000e+02]> : tensor<2xf32>} : (tensor<1x1x1x1xf32>) -> tensor<1x1x1x1xf32>
  %w_1 = arith.constant dense<127.0> : tensor<4096x1x1x1xf32>

  func.return %0 : tensor<*xf32>
}

// CHECK-NOT: "tfl.reshape"
// CHECK: "tfl.pack"

// -----

func.func @squaredDifference(%arg0: tensor<4xf32>, %arg1: tensor<4xf32>) -> tensor<4xf32> {
  %0 = "tfl.squared_difference"(%arg0, %arg1) : (tensor<4xf32>, tensor<4xf32>) -> tensor<4xf32>
  func.return %0 : tensor<4xf32>
}

    %0 = "tf.Sub"(%cst_0, %cst_1) : (tensor<8xf32>, tensor<8xf32>) -> tensor<8xf32>
    return %0 : tensor<8xf32>
  }
// CHECK: @serving_default
// CHECK-DAG: %[[VAR_HANDLE_2:.*]] = "tf.VarHandleOp"() {{.*shared_name = "const_0".*}} : () -> tensor<!tf_type.resource<tensor<8xf32>>>
// CHECK-DAG: %[[READ_VAR_0:.*]] = "tf.ReadVariableOp"(%[[VAR_HANDLE_2]]) : (tensor<!tf_type.resource<tensor<8xf32>>>) -> tensor<8xf32>

func.func @RemoveRedundantPack(%arg0: tensor<2x5xf32>) -> (tensor<2x5xf32>, tensor<5xf32>) {
  %0:2 = "tfl.unpack"(%arg0) {axis = 0 : i32, num = 2 : i32} : (tensor<2x5xf32>) -> (tensor<5xf32>, tensor<5xf32>)
  %1 = "tfl.pack"(%0#0, %0#1) {axis = 0 : i32, values_count = 2 : i32} : (tensor<5xf32>, tensor<5xf32>) -> (tensor<2x5xf32>)
  func.return %1, %0#0: tensor<2x5xf32>, tensor<5xf32>
  // CHECK: %[[UNPACK:.*]]:2 = "tfl.unpack"

  %1 = "tf.TPUCompileMlirAndExecute"(%0) {metadata = "metadata", mlir_module = "mlir_module", operandSegmentSizes = array<i32: 1, 0>, producer_name = "producer_name"} : (tensor<f32>) -> tensor<i32>
  func.return %1 : tensor<i32>
}

// -----

func.func @executeop_input_same_execute_op(%arg0: tensor<i32>, %arg1: tensor<2xf32>) -> (tensor<i32>) {

  %7 = "tfl.add"(%2, %1) {fused_activation_function = "NONE"} : (tensor<4xf32>, tensor<  f32>) -> tensor<4xf32>
  %8 = "tfl.add"(%2, %3) {fused_activation_function = "NONE"} : (tensor<4xf32>, tensor<4xf32>) -> tensor<4xf32>
  %9 = "tfl.add"(%2, %3) {fused_activation_function = "SIGN_BIT"} : (tensor<4xf32>, tensor<4xf32>) -> tensor<4xf32>

  %b = arith.constant dense<-1.23697901> : tensor<64xf32>
  %conv = "tfl.conv_2d"(%arg0, %w, %b) {dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32, fused_activation_function = "NONE", padding = "SAME", stride_h = 2 : i32, stride_w = 2 : i32} : (tensor<1x224x224x3xf32>, tensor<64x3x3x3xf32>, tensor<64xf32>) -> tensor<1x112x112x64xf32>
  func.return %conv : tensor<1x112x112x64xf32>

module {
  func.func @simpleTest(%arg0: tensor<1xf32>, %arg1: tensor<1xf32>, %arg2: tensor<1xf32>, %arg3: tensor<1xf32>) -> tensor<2x1xf32> {
    %0 = func.call @func_0_GPU_FLOAT(%arg0, %arg1, %arg2) {tac.interface_name = "func_0"} : (tensor<1xf32>, tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>
    %1 = func.call @func_1_GPU_FLOAT(%arg0, %arg3) {tac.interface_name = "func_1"} : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>