%1 = "tf.BiasAdd"(%0, %cst_0) {data_format = "NHWC"} : (tensor<1x3x2x2xf32>, tensor<2xf32>) -> tensor<1x3x2x2xf32>
  %2 = "tf.Mul"(%0, %cst_1) : (tensor<1x3x2x2xf32>, tensor<2xf32>) -> tensor<1x3x2x2xf32>
  func.return %1, %2 : tensor<1x3x2x2xf32>, tensor<1x3x2x2xf32>
}
// CHECK: func @not_fuse_conv2d_with_bias_and_mul

  %0 = "tf.Conv2D"(%output, %cst) <{data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 2, 2, 1], use_cudnn_on_gpu = true}> {attr_map = "0:strides,1:use_cudnn_on_gpu,2:padding,3:explicit_paddings,4:dilations", device = ""} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<1x2x2x2xf32>

    %2 = stablehlo.convolution(%1, %0)
      dim_numbers = [b, 0, 1, f]x[0, 1, i, o]->[b, 0, 1, f],
      window = {
        stride = [1, 1], pad = [[0, 0], [1, 1]],
        lhs_dilate = [1, 1],
        rhs_dilate = [1, 1]
      }
      {
        batch_group_count = 1 : i64,
        feature_group_count = 1 : i64
      } : (tensor<1x3x2x3xf32>, tensor<2x3x3x2xf32>)

func.func @broadcast_to_to_reshape(%arg0: tensor<4x4x4xf32>, %arg1 : tensor<4xi32>) -> tensor<1x4x4x4xf32> {
  %0 = "tfl.broadcast_to"(%arg0, %arg1) : (tensor<4x4x4xf32>, tensor<4xi32>) -> tensor<1x4x4x4xf32>
  // CHECK: "tfl.reshape"
  // CHECK-SAME: (tensor<4x4x4xf32>, tensor<4xi32>) -> tensor<1x4x4x4xf32>
  func.return %0 : tensor<1x4x4x4xf32>
}

// Converts tfl.broadcast_to to tfl.reshape if input and output have the same

  %0 = "tf.Conv2D"(%output, %cst) <{data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 2, 2, 1], use_cudnn_on_gpu = true}> {attr_map = "0:strides,1:use_cudnn_on_gpu,2:padding,3:explicit_paddings,4:dilations", device = ""} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<1x2x2x2xf32>

  %1 = mhlo.multiply %0, %cst1 : tensor<1x1x2x4xf32>
  // CHECK:      return %[[RES]] : tensor<1x1x2x4xf32>
  func.return %1 : tensor<1x1x2x4xf32>
}

// CHECK-LABEL: @foldBroadcastInDimBeforeMulOp_bcast_dim_2D_float
func.func @foldBroadcastInDimBeforeMulOp_bcast_dim_2D_float() -> (tensor<1x2x2x3xf32>) {

func.func private @conv(%input: tensor<1x3x4x3xf32> {tf._user_specified_name = "input_tensor"}) -> tensor<*xf32> attributes {tf._construction_context = "kEagerRuntime", tf._input_shapes = [#tf_type.shape<1x3x4x3>]} {
  %weight = arith.constant dense_resource<__elided__> : tensor<2x3x3x2xf32>
  %bias = arith.constant dense<[7.11401462, 7.05456924]> : tensor<2xf32>

  %9 = "tfl.quantize"(%8) {qtype = tensor<1x3x1x1x!quant.uniform<i8:f32, 1.0:2>>, volatile} : (tensor<1x3x1x1xf32>) -> tensor<1x3x1x1x!quant.uniform<i8:f32, 1.0:2>>
  %10 = "tfl.dequantize"(%9) : (tensor<1x3x1x1x!quant.uniform<i8:f32, 1.0:2>>) -> tensor<1x3x1x1xf32>

// CHECK: %0 = "tfl.transpose"(%arg0, %cst_0) : (tensor<2x3x4x1xf32>, tensor<4xi32>) -> tensor<1x2x3x4xf32>
// CHECK: %1 = tfl.add(%0, %cst) <{fused_activation_function = "NONE"}> : (tensor<1x2x3x4xf32>, tensor<5x2x3x4xf32>) -> tensor<5x2x3x4xf32>

// -----

// CHECK-LABEL: doubleTposeOneBroadcastInput
func.func @doubleTposeOneBroadcastInput(%arg0: tensor<2x3x4x1xf32>, %arg1: tensor<2x3x4x5xf32>) -> tensor<5x2x3x4xf32> {

// -----

// CHECK-LABEL: transpose_simple_4d
func.func @transpose_simple_4d() -> tensor<5x2x3x4xf32> {
  %0 = stablehlo.constant dense<1.000000e+0> : tensor<2x3x4x5xf32>
  %1 = stablehlo.transpose %0, dims = [3, 0, 1, 2] : (tensor<2x3x4x5xf32>) -> tensor<5x2x3x4xf32>
  return %1 : tensor<5x2x3x4xf32>
}
// CHECK-DAG: %[[CONST_0:.+]] = stablehlo.constant dense<1.000000e+00> : tensor<5x2x3x4xf32>
// CHECK-NOT: transpose