%1 = "tf.ExpandDims"(%0, %cst_0) : (tensor<4x?x?xf32>, tensor<i32>) -> tensor<4x?x?x1xf32>
  %2 = "tf.Conv2D"(%1, %arg1) {padding = "VALID", strides = [1, 1, 1, 1]} : (tensor<4x?x?x1xf32>, tensor<5x5x1x1xf32>) -> tensor<4x?x?x1xf32>
  %3 = "tf.Squeeze"(%2) {squeeze_dims = [3]} : (tensor<4x?x?x1xf32>) -> tensor<4x?x?xf32>

    %0 = "tf.Sub"(%cst_0, %cst_1) : (tensor<8xf32>, tensor<8xf32>) -> tensor<8xf32>
    return %0 : tensor<8xf32>
  }
// CHECK: @serving_default
// CHECK-DAG: %[[VAR_HANDLE_2:.*]] = "tf.VarHandleOp"() {{.*shared_name = "const_0".*}} : () -> tensor<!tf_type.resource<tensor<8xf32>>>
// CHECK-DAG: %[[READ_VAR_0:.*]] = "tf.ReadVariableOp"(%[[VAR_HANDLE_2]]) : (tensor<!tf_type.resource<tensor<8xf32>>>) -> tensor<8xf32>

// CHECK-NEXT:  %[[quant:.*]] = "tfl.quantize"(%[[split]]#0) <{qtype = tensor<2x!quant.uniform<u8:f32, 1.000000e+00>>}> {volatile} : (tensor<2xf32>) -> tensor<2x!quant.uniform<u8:f32, 1.000000e+00>>
// CHECK-NEXT:  return %[[quant]] : tensor<2x!quant.uniform<u8:f32, 1.000000e+00>>

// CHECK:         }
func.func @tfl_wrapped_jax_random_uniform(%arg0: tensor<2xui32>) -> tuple<tensor<1x2xf32>> {
  // This is a fake jax random uniform body.
  %0 = stablehlo.constant dense<0.0> : tensor<2xf32>
  %1 = "stablehlo.reshape"(%0) : (tensor<2xf32>) -> tensor<1x2xf32>
  %2 = "stablehlo.tuple"(%1) : (tensor<1x2xf32>) -> tuple<tensor<1x2xf32>>
  func.return %2 : tuple<tensor<1x2xf32>>

  %7 = "tf.BiasAdd"(%6, %cst) {data_format = "NHWC", device = ""} : (tensor<*xf32>, tensor<2xf32>) -> tensor<*xf32>
  func.return %2, %5, %7 : tensor<*xf32>, tensor<*xf32>, tensor<*xf32>

// CHECK: %[[CONST_0:.*]] = "tf.Const"() <{value = dense<0.000000e+00> : tensor<2xf32>}> : () -> tensor<2xf32>
// CHECK: %[[PARTITIONEDCALL_0:.*]] = "tf.PartitionedCall"(%arg0, %arg1, %[[CONST_0]])

  func.return %0 : tensor<256x32x32x16xf32>
}

func.func @testConv2DDynamicShape(tensor<?x32x32x3xf32>, tensor<16x3x3x3xf32>, tensor<16xf32>) -> tensor<?x32x32x16xf32> {
^bb0(%arg0: tensor<?x32x32x3xf32>, %arg1: tensor<16x3x3x3xf32>, %arg2: tensor<16xf32>):
  // CHECK: _arithmetic_count = -1 : i64

        tensor<1x1x5xf32>,
        tensor<2x5xf32>, tensor<2x5xf32>, tensor<2x5xf32>, tensor<2x5xf32>,
        tensor<2x4xf32>, tensor<2x4xf32>, tensor<2x4xf32>, tensor<2x4xf32>,
        tensor<2xf32>, tensor<2xf32>, tensor<2xf32>,
        tensor<2xf32>, tensor<2xf32>, tensor<2xf32>, tensor<2xf32>,
        tensor<4x2xf32>, tensor<4xf32>,
        tensor<1x4xf32>, tensor<1x2xf32>,

    %4 = "quantfork.stats"(%2) {layerStats = dense<[-2795.36523, 4609.57373]> : tensor<2xf32>} : (tensor<*xf32>) -> tensor<*xf32>
    %5 = "tf.BiasAdd"(%4, %3) {data_format = "NHWC", device = ""} : (tensor<*xf32>, tensor<2xf32>) -> tensor<*xf32>
    %6 = "quantfork.stats"(%5) {layerStats = dense<[-2788.31055, 4616.62842]> : tensor<2xf32>} : (tensor<*xf32>) -> tensor<*xf32>

    // expected-remark@+1 {{Cost: 10}}
    %44 = "tf.ConcatV2"(%36, %37, %38, %39, %40, %41, %42, %43, %10) {device = "/job:localhost/replica:0/task:0/device:CPU:0"} : (tensor<?x1xf32>, tensor<?x1xf32>, tensor<?x1xf32>, tensor<?x1xf32>, tensor<?x1xf32>, tensor<?x1xf32>, tensor<?x1xf32>, tensor<?x1xf32>, tensor<i32>) -> tensor<?x8xf32>
    // expected-remark@+1 {{Cost: 1}}

    %cst_0 = "tf.Const"() {value = dense<[-1, 10]> : tensor<2xi32>} : () -> tensor<2xi32>
    %1 = "tf.MatMul"(%arg0, %arg1) {
      transpose_a = false, transpose_b = false
    } : (tensor<1x10xf32>, tensor<10x10xf32>) -> tensor<1x10xf32>  loc(fused["MatMul:", "test_opt_out"])