}

// -----

func.func @testMaxUnpooling2D(%arg0: tensor<1x8x8x128xf32>, %arg1: tensor<1x8x8x128xf32>) -> tensor<1x8x8x128xf32> {
  // custom op for "tfl.max_unpooling_2d"(%arg0, %arg1) {filter_h = 2 : i32, filter_w = 2 : i32, padding = "SAME", stride_h = 2 : i32, stride_w = 2 : i32} : (tensor<1x8x8x128xf32>, tensor<1x8x8x128xf32>) -> (tensor<1x8x8x128xf32>)

  // CHECK: %[[VAL_1:.*]] = "tf.Identity"(%[[VAL_0]]) : (tensor<*xf32>) -> tensor<*xf32>
  // CHECK: return %[[VAL_1]]
  %0 = "tf.Conv2D"(%arg2, %arg1) {data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 1, 1], use_cudnn_on_gpu = true} : (tensor<8x32x32x3xf32>, tensor<1x1x3x128xf32>) -> tensor<*xf32>

  // CHECK-NEXT: [[RESULT:%.*]] = "tf.BiasAdd"([[SQUEEZE]], [[BIAS]]) : (tensor<1x128x128xf32>, tensor<128xf32>) -> tensor<1x128x128xf32>
  // CHECK-NEXT: return [[RESULT]] : tensor<1x128x128xf32>
}

func.func @testDilatedDepthWiseConvWithExpandSqueeze1(%arg0: tensor<1x128x128xf32>, %arg1: tensor<5x5x1x1xf32>, %arg2: tensor<128xf32>) -> tensor<1x128x128xf32> {

// CHECK:           [[VAL_7:%.*]] = "tfl.slice"([[VAL_1]], [[VAL_3]], [[VAL_5]]) : (tensor<1x8x8x1024xf32>, tensor<4xi32>, tensor<4xi32>) -> tensor<1x8x8x256xf32>
// CHECK:           [[VAL_8:%.*]] = "tfl.slice"([[VAL_1]], [[VAL_4]], [[VAL_5]]) : (tensor<1x8x8x1024xf32>, tensor<4xi32>, tensor<4xi32>) -> tensor<1x8x8x256xf32>

  %7 = "tfl.add"(%1, %6) {tac.device = "GPU", tac.inference_type = "FLOAT", fused_activation_function = "NONE"} : (tensor<1x128x128xf32>, tensor<1x128x128xf32>) -> tensor<1x128x128xf32>
  func.return %7 : tensor<1x128x128xf32>
}

// CHECK:   func @norm1(%[[VAL_0:.*]]: tensor<1x128x128xf32>) -> tensor<1x128x128xf32> {

func.func private @func_20_GPU_FLOAT(%arg0: tensor<128x128xf32>, %arg1: tensor<3xi32>) -> tensor<1x128x128xf32> attributes {tac.device = "GPU", tac.inference_type = "FLOAT", tac.interface_name = "func_20"} {
  %0 = "tfl.reshape"(%arg0, %arg1) {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<128x128xf32>, tensor<3xi32>) -> tensor<1x128x128xf32>
  func.return %0 : tensor<1x128x128xf32>
}

// CHECK:           %[[VAL_2:.*]] = "tf.Conv2D"(%[[VAL_0]], %[[VAL_1]]) <{data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 1, 1], use_cudnn_on_gpu = true}> : (tensor<1x8x8x207xf32>, tensor<3x3x207x16xf32>) -> tensor<1x8x8x16xf32>
// CHECK:           return %[[VAL_2]] : tensor<1x8x8x16xf32>
// CHECK:         }

  %894 = "tfl.reshape"(%arg0, %cst_10) : (tensor<128xf32>, tensor<4xi32>) -> tensor<1x1x1x128xf32>
  %895 = "tfl.mul"(%894, %arg1) {fused_activation_function = "NONE"} : (tensor<1x1x1x128xf32>, tensor<1x512x512x128xf32>) -> tensor<1x512x512x128xf32>
  %896 = "tfl.mul"(%arg2, %894) {fused_activation_function = "NONE"} : (tensor<1x512x512x128xf32>, tensor<1x1x1x128xf32>) -> tensor<1x512x512x128xf32>