%4:4 = "tf.CustomAggregator"(%3) {calibration_method = 1 : i32, id = "1", max_percentile = 0.000000e+00 : f32, min_percentile = 0.000000e+00 : f32, num_bins = 0 : i32} : (tensor<3x11xf32>) -> (tensor<3x11xf32>, tensor<f32>, tensor<f32>, tensor<*xi64>)
    %5 = stablehlo.add %4#0, %1 : tensor<3x11xf32>
    %6 = stablehlo.multiply %5, %1 : tensor<3x11xf32>

func.func @ensureBiasForConv2d(%arg0: tensor<128x32x32x3xf32>, %arg1: tensor<32x1x1x3xf32>) -> tensor<128x32x32x32xf32> {
  %cst = "tfl.no_value"() {value = unit} : () -> none
  %0 = "tfl.conv_2d"(%arg0, %arg1, %cst) {dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32, fused_activation_function = "NONE", padding = "VALID", stride_h = 1 : i32, stride_w = 1 : i32} : (tensor<128x32x32x3xf32>, tensor<32x1x1x3xf32>, none) -> tensor<128x32x32x32xf32>

  // attribute `_quantization_method` set to `"no_quantization {}"`.
  constexpr absl::string_view kXlaCallModuleOpWithQuantizationMethodAttr =
      R"mlir(
    func.func @main(%arg0: tensor<1x1x3xf32>, %arg1: tensor<3x4xf32>) -> tensor<1x1x4xf32> {

// CHECK: return %[[identity]] : tensor<1x2x2x6xf32>

func.func @cast_bf16_batch_matmul_v2_to_fp32(%arg0: tensor<1x1x10xf32>) -> (tensor<1x1x2xf32>) {
  %cst = "tf.Const"() {device = "", value = dense<1.000000e+01> : tensor<10x2xbf16>} : () -> tensor<10x2xbf16>
  %0 = "tf.Cast"(%arg0) {Truncate = false, device = ""} : (tensor<1x1x10xf32>) -> tensor<1x1x10xbf16>

// CHECK-NEXT:    return %2 : tensor<3x2xf32>
}

func.func @transpose_3d(%arg0: tensor<1x2x3xf32>) -> tensor<3x2x1xf32> {
  %0 = "mhlo.transpose"(%arg0) <{permutation = dense<[2, 1, 0]> : tensor<3xi64>}> : (tensor<1x2x3xf32>) -> tensor<3x2x1xf32>
  func.return %0 : tensor<3x2x1xf32>

// CHECK-LABEL:   transpose_3d
// CHECK-NEXT:    %0 = "tfl.pseudo_const"() <{value = dense<[2, 1, 0]> : tensor<3xi64>}> : () -> tensor<3xi64>

//CHECK:func.func private @slice(%arg0: tensor<160x20x1xf32>) -> tensor<1x1x1xf32> {
//CHECK-NEXT: %0 = "vhlo.slice_v1"(%arg0) <{limit_indices = #vhlo.tensor_v1<dense<0> : tensor<3xi64>>, start_indices = #vhlo.tensor_v1<dense<1> : tensor<3xi64>>, strides = #vhlo.tensor_v1<dense<1> : tensor<3xi64>>}> : (tensor<160x20x1xf32>) -> tensor<1x1x1xf32> 
//CHECK-NEXT: return %0 : tensor<1x1x1xf32>
//CHECK-NEXT:}

  %0 = "mhlo.broadcast_in_dim"(%arg0) <{broadcast_dimensions = dense<[1, 2, 3]> : tensor<3xi64>, name = "broadcast.0"}> : (tensor<8x1x16xf32>) -> tensor<3x8x8x16xf32>
  func.return %0 : tensor<3x8x8x16xf32>
}

// CHECK-LABEL:   func @broadcast_in_dim_general_case(
// CHECK-SAME:                                        %[[VAL_0:.*]]: tensor<3x1x16xf32>) -> tensor<3x8x8x16xf32> {

func.func @EliminateReduceOpsFloat(%arg: tensor<1x2x1x3xf32>, %arg_scalar: tensor<f32>, %arg_unknown: tensor<?xf32>, %axis_unknown: tensor<?xi32>) -> (tensor<f32>, tensor<f32>, tensor<2x1x3xf32>, tensor<1x2x1x3xf32>, tensor<1x1x1x3xf32>, tensor<1x1x3xf32>, tensor<1x2x3xf32>, tensor<1x2x1x3xf32>, tensor<1x2x1xf32>, tensor<1x2x1x1xf32>, tensor<?xf32>) {
  %axis_0 = arith.constant dense<0> : tensor<1xi32>

func.func @transpose_3d_int32(%arg0: tensor<1x2x3xf32>) -> tensor<3x2x1xf32> {
  %permutation = "tf.Const"() {value = dense<[2, 1, 0]> : tensor<3xi32>} : () -> (tensor<3xi32>)
  // CHECK: "mhlo.transpose"
  %0 = "tf.Transpose"(%arg0, %permutation) : (tensor<1x2x3xf32>, tensor<3xi32>) -> tensor<3x2x1xf32>
  func.return %0 : tensor<3x2x1xf32>
}

// -----

// CHECK-LABEL: @transpose_3d

    %elem_shape = "tf.Const"() {value = dense<[-1, 1]> : tensor<2xi32>} : () -> tensor<2xi32>
    %size = "tf.Const"() {value = dense<10> : tensor<i32>} : () -> tensor<i32>
    %tensors = "tf._SomeOp"() : () -> tensor<3x16x1xf32>
    %indices = "tf.Const"() {value = dense<[2, 5, 9]> : tensor<3xi32>} : () -> tensor<3xi32>
    // CHECK: TensorListReserve
    // CHECK-SAME: tensor<!tf_type.variant<tensor<16x1xf32>>>