// CHECK-NEXT:      %9 = "tfl.unsorted_segment_prod"(%4, %6, %7) : (tensor<3xi32>, tensor<3xi32>, tensor<i32>) -> tensor<1xi32>
// CHECK-NEXT:      %10 = "tfl.pseudo_const"() <{value = dense<4> : tensor<1xi32>}> : () -> tensor<1xi32>
// CHECK-NEXT:      %11 = "tfl.concatenation"(%10, %9, %8) <{axis = 0 : i32, fused_activation_function = "NONE"}> : (tensor<1xi32>, tensor<1xi32>, tensor<1xi32>) -> tensor<3xi32>

  %outputs_50 =  "tf.Mul"(%outputs_30, %outputs_48) {device = ""} : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>
  %outputs_52 =  "tf.Reshape"(%outputs_50, %outputs_0) {device = ""} : (tensor<*xf32>, tensor<2xi32>) -> tensor<*xf32>
  %outputs_54 =  "tf.MatMul"(%outputs_40, %outputs_52) {device = "", transpose_a = false, transpose_b = false} : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>

    %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>

  %b = arith.constant dense<0.0> : tensor<16xf32>
  %conv_3d = "tfl.conv_3d"(%arg0, %w, %b) {dilation_d_factor = 1 : i32, dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32, fused_activation_function = "NONE", padding = "SAME", stride_d = 1 : i32, stride_h = 1 : i32, stride_w = 1 : i32} : (tensor<1x32x32x32x8xf32>, tensor<1x1x1x8x16xf32>, tensor<16xf32>) -> tensor<1x32x32x32x16xf32>
  func.return %conv_3d : tensor<1x32x32x32x16xf32>

}

// CHECK-LABEL: QuantizeStridedSlice
func.func @QuantizeStridedSlice(tensor<12x2x2x5x!quant.uniform<u8:f32, 0.1>>, tensor<1xi32>, tensor<1xi32>, tensor<1xi32>) -> tensor<1x2x2x5xf32> {
^bb0(%arg0: tensor<12x2x2x5x!quant.uniform<u8:f32, 0.1>>, %arg1: tensor<1xi32>, %arg2: tensor<1xi32>, %arg3: tensor<1xi32>):
  %0 = "tfl.dequantize"(%arg0) : (tensor<12x2x2x5x!quant.uniform<u8:f32, 0.1>>) -> tensor<12x2x2x5xf32>

      %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>

func.func @main(%arg0: tensor<4xf32>) -> tensor<4xf32> {
  %0 = "tfl.pseudo_const" () {value = dense<1.0> : tensor<4xf32>} : () -> tensor<4xf32> loc("Const")
  %1 = "tfl.squared_difference"(%arg0, %0) {fused_activation_function = "NONE"} : (tensor<4xf32>, tensor<4xf32>) -> tensor<4xf32> loc("squared_difference")
  %2 = "tfl.mul"(%0, %1) {fused_activation_function = "NONE"} : (tensor<4xf32>, tensor<4xf32>) -> tensor<4xf32> loc("mul")

  // 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>

func.func @sharding_custom_call_removed(%arg0: tensor<3xf32>) -> tensor<3xf32> {
  %1 = stablehlo.custom_call @Sharding(%arg0) {mhlo.sharding = ""} : (tensor<3xf32>) -> tensor<3xf32>
  return %1 : tensor<3xf32>
}
// CHECK-NOT: custom_call

// -----

// Tests that a custom_call that is not @Sharding is not removed.

// CHECK-LABEL: custom_call_not_removed
func.func @custom_call_not_removed(%arg0: tensor<3xf32>) -> tensor<3xf32> {

  // Confirm that attributes that cannot be stored in the flatbuffer options
  // for a given operator are dropped silently.
  %1 = "tfl.squared_difference"(%arg0, %0) {fused_activation_function = "NONE"} : (tensor<4xf32>, tensor<4xf32>) -> tensor<4xf32> loc("squared_difference")
  %2 = "tfl.mul"(%arg0, %1) {fused_activation_function = "NONE"} : (tensor<4xf32>, tensor<4xf32>) -> tensor<4xf32> loc("mul")