}

// -----

func.func @testTFLiteDetectionPostProcess(%arg0: tensor<1x64x64x32xf32>, %arg1: tensor<1x64x64x32xf32>, %arg2: tensor<1x64x64x32xf32>) -> (tensor<1x32x32x32xf32>, tensor<1x32x32x32xf32>) {
  %0, %1, %2, %3 = "tfl.custom_tf"(%arg0, %arg1, %arg2) ({
  ^bb0(%a0: tensor<1x64x64x32xf32>, %a1: tensor<1x64x64x32xf32>, %a2: tensor<1x64x64x32xf32>):

      // CHECK-OPT: %{{.*}} = stablehlo.custom_call @tf.ResizeBilinear(%arg0, %cst) {align_corners = false, device = "", half_pixel_centers = true} : (tensor<1x32x32x128xf32>, tensor<2xi32>) -> tensor<1x64x64x128xf32>
      %1 = "tf.ResizeBilinear"(%arg0, %0) {
        align_corners = false, device = "", half_pixel_centers = true

  func.return %0 : tensor<1x64x84x32xf32>

  %0 = "tfl.custom"(%arg0, %arg1, %arg2) {custom_code = "Convolution2DTransposeBias", custom_option = #tfl<const_bytes : "0x010000000200000002000000">} : (tensor<32x4x4x128xf32>, tensor<1x32x42x128xf32>, tensor<4xi32>) -> tensor<1x64x84x32xf32>
  func.return %0 : tensor<1x64x84x32xf32>
}
// CHECK-LABEL: main

  %0 = "tfl.custom"(%arg0, %arg1, %arg2) {custom_code = "Convolution2DTransposeBias", custom_option = #tfl<const_bytes : "0x010000000200000002000000">} : (tensor<32x4x4x128xf32>, tensor<1x32x42x128xf32>, tensor<4xi32>) -> tensor<1x64x84x32xf32>
  func.return %0 : tensor<1x64x84x32xf32>
}
// CHECK-LABEL: main

module {
func.func @main(%arg0: tensor<21x32x32x128xf32>, %arg1: tensor<2xi32>) -> tensor<1x64x64x128xf32> {
  %0 = "tf.ResizeBilinear"(%arg0, %arg1) {align_corners = false, device = "", half_pixel_centers = true} : (tensor<21x32x32x128xf32>, tensor<2xi32>) -> tensor<1x64x64x128xf32>
  func.return %0 : tensor<1x64x64x128xf32>
}

  %27 = mhlo.subtract %26, %19 : tensor<1x64x32x32xf32>
  %28 = "mhlo.broadcast_in_dim"(%1) <{broadcast_dimensions = dense<3> : tensor<1xi64>}> : (tensor<32xf32>) -> tensor<1x64x32x32xf32>
  %29 = mhlo.multiply %27, %28 : tensor<1x64x32x32xf32>
  %30 = mhlo.add %19, %29 : tensor<1x64x32x32xf32>

  %cst = "tfl.no_value"() {value = unit} : () -> none
  // CHECK: _arithmetic_count = 176160768 : i64
  %0 = "tfl.transpose_conv"(%arg0, %arg1, %arg2, %cst) {padding = "SAME", stride_h = 2 : i32, stride_w = 2 : i32, fused_activation_function = "NONE"} : (tensor<4xi32>, tensor<32x4x4x128xf32>, tensor<1x32x42x128xf32>, none) -> tensor<1x64x84x32xf32>
  func.return %0 : tensor<1x64x84x32xf32>
}

  %1 = "tfl.add"(%0, %cst) {fused_activation_function = "NONE"} : (tensor<1x64x84x32xf32>, tensor<32xf32>) -> tensor<1x64x84x32xf32>
  func.return %1 : tensor<1x64x84x32xf32>

  // CHECK-DAG: %[[SHAPE:.*]] = arith.constant dense<[1, 64, 84, 32]> : tensor<4xi32>

  %cst_1 = arith.constant dense<0> : tensor<2x2xi32>
  %0 = "tf.SpaceToBatchND"(%arg0, %cst, %cst_1) : (tensor<1x128x128x3xf32>, tensor<2xi32>, tensor<2x2xi32>) -> tensor<4x64x64x3xf32>
  %1 = "tf.Conv2D"(%0, %arg1) {padding = "VALID", strides = [1, 1, 1, 1]} : (tensor<4x64x64x3xf32>, tensor<5x5x3x8xf32>) -> tensor<4x60x60x8xf32>
  %2 = "tf.BatchToSpaceND"(%1, %cst, %cst_1) : (tensor<4x60x60x8xf32>, tensor<2xi32>, tensor<2x2xi32>) -> tensor<1x120x120x8xf32>