%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>

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:}

        none, none, none,
        tensor<3xf32>, tensor<3xf32>, tensor<3xf32>, tensor<3xf32>,
        none, none,
        tensor<1x3xf32>, tensor<1x3xf32>,
        none, none, none, none) -> tensor<1x2x3xf32>
  %17 = "quantfork.stats"(%16) {layerStats = dense<[-0.1, 0.1]> : tensor<2xf32>} : (tensor<1x2x3xf32>) -> tensor<1x2x3xf32>
  func.return %17 : tensor<1x2x3xf32>

    %0 = "tfl.pseudo_const"() {value = dense<0.962260901> : tensor<1xf32>} : () -> tensor<1xf32>
    %1 = func.call @func_0_GPU_FLOAT(%arg0, %0) {tac.device = "GPU", tac.inference_type = "FLOAT", tac.interface_name = "func_0"} : (tensor<1x200x200x200xf32>, tensor<1xf32>) -> tensor<1x200x200x200xf32>
    %2 = "tfl.pseudo_const"() {value = dense<0.895973444> : tensor<1xf32>} : () -> tensor<1xf32>

    %cst = "tf.Const"() <{value = dense<true> : tensor<i1>}> {device = ""} : () -> tensor<i1>
    %cst_0 = "tf.Const"() <{value = dense<[0.117216609, 0.933735609, 0.0728900209]> : tensor<3xf32>}> {device = ""} : () -> tensor<3xf32>

  // CHECK-SAME:      <[-1.000000e+00, 1.000000e+00]> : tensor<2xf32>}>
  // CHECK-SAME:      : (tensor<40x37xf32>) -> tensor<40x37xf32>
  // CHECK-NEXT:    %[[cst:.*]] = "tfl.pseudo_const"() <{value = dense<
  // CHECK-SAME:      1.000000e+00> : tensor<40xf32>}> : () -> tensor<40xf32>
  // CHECK-NEXT:    %[[fc:.*]]:2 = "tfl.fully_connected"(%[[stat]], %arg1,

  %perm = arith.constant dense<[3, 0, 1, 2]> : tensor<4xi32>
  %0 = "tfl.transpose"(%arg0, %perm) : (tensor<2x3x4x5xf32>, tensor<4xi32>) -> tensor<5x2x3x4xf32>
  %perm1 = arith.constant dense<[3, 0, 1, 2]> : tensor<4xi32>
  %1 = "tfl.transpose"(%arg1, %perm1) : (tensor<2x3x4x5xf32>, tensor<4xi32>) -> tensor<5x2x3x4xf32>
  %2 = tfl.add %0, %1 { fused_activation_function = "NONE" } : tensor<5x2x3x4xf32>

      %outputs_9, %control_10 = tf_executor.island wraps "tf.Mul"(%outputs_7, %outputs) {T = "tfdtype$DT_FLOAT", _tpu_replicate = "cluster", device = "", name = "mul"} : (tensor<0xf32>, tensor<f32>) -> tensor<0xf32>
      %outputs_11, %control_12 = tf_executor.island wraps "tf.Placeholder"() {device = "", dtype = "tfdtype$DT_FLOAT", name = "y", shape = "tfshape$dim { }"} : () -> tensor<0xf32>

  %cst_1 = "tf.Const"() {value = dense<[0.3, 0.4, 0.3, 0.4, 0.3, 0.4, 0.3, 0.4]> : tensor<8xf32>} : () -> tensor<8xf32>

  // CHECK-DAG: %[[CST_0:.+]] = mhlo.constant dense<[1.000000e-01, 2.000000e-01]> : tensor<2xf32>
  // CHECK-DAG: %[[CST_1:.+]] = mhlo.constant dense<[3.000000e-01, 4.000000e-01]> : tensor<2xf32>
  // CHECK: %[[CST_BCAST:.+]] = "mhlo.broadcast_in_dim"(%[[CST_0]]) <{broadcast_dimensions = dense<3> : tensor<1xi64>}> : (tensor<2xf32>) -> tensor<1x1x3x2xf32>