}

// CHECK-LABEL: prepareAdd
func.func @prepareAdd(%arg0: tensor<2x2xf32>) -> tensor<2x2xf32> {
  %cst = arith.constant dense<[[0.0, 1.0], [2.0, 255.0]]> : tensor<2x2xf32>
  %add = "tfl.add"(%arg0, %cst) {fused_activation_function="NONE"} : (tensor<2x2xf32>, tensor<2x2xf32>) -> tensor<2x2xf32>
  func.return %add : tensor<2x2xf32>

tensor<1024xf32>, tensor<32x1024xf32>, tensor<1024x1xf32>, tensor<1024xf32>, tensor<1024x1xf32>, tensor<i64>, tensor<i32>, tensor<32x1024xf32>, tensor<i64>, tensor<i64>, tensor<1024xf32>, tensor<32x1024xf32>, tensor<1024x1xf32>, tensor<1024xf32>, tensor<1024x1xf32>, tensor<i64>, tensor<i32>, tensor<32x1024xf32>, tensor<i64>, tensor<i64>, tensor<1024xf32>, tensor<32x1024xf32>, tensor<1024x1xf32>, tensor<1024xf32>, tensor<1024x1xf32>, tensor<i64>, tensor<i32>, tensor<32x1024xf32>, tensor<i64>, tensor<i64>,...

    has_token_input_output = false, module = "", platforms = [],
    version = 5 : i64
  } : (tensor<1x2xf32>, tensor<2x3xf32>) -> tensor<1x3xf32>
  return %0 : tensor<1x3xf32>
}

// CHECK-LABEL: func.func @qdq_for_dot_general_weight_empty
// CHECK-SAME: (%[[ARG_0:.+]]: tensor<1x2xf32>) -> tensor<1x3xf32>
// CHECK: %[[CST:.+]] = "tf.Const"() <{value = dense<3.000000e-01> : tensor<2x3xf32>}> : () -> tensor<2x3xf32>

// CHECK: return

  func.func private @composite_dot_general_fn(%arg0: tensor<1x4xf32>, %arg1: tensor<4x3xf32>) -> tensor<1x3xf32> attributes {_from_xla_call_module} {
      %0 = stablehlo.dot_general %arg0, %arg1, contracting_dims = [1] x [0] : (tensor<1x4xf32>, tensor<4x3xf32>) -> tensor<1x3xf32>
      return %0 : tensor<1x3xf32>
  }
}

// -----

  func.func private @uniform_quantize_0(%arg0: tensor<1x2xf32>, %arg1: tensor<1x1xf32>, %arg2: tensor<1x1xi8>) -> tensor<1x2xi32> {
    %0 = stablehlo.convert %arg0 : (tensor<1x2xf32>) -> tensor<1x2xi32>
    return %0 : tensor<1x2xi32>
  }
// CHECK: @uniform_quantize_0
  func.func private @uniform_quantize_1(%arg0: tensor<1x3xf32>, %arg1: tensor<1x1xf32>, %arg2: tensor<1x1xi8>) -> tensor<1x3xi8> {

// CHECK:           %[[VAL_17:.*]] = tfl.mul(%[[VAL_15]], %[[VAL_16]]) <{fused_activation_function = "NONE"}> {tac.device = "GPU", tac.inference_type = "FLOAT"} : (tensor<1x384x128xf32>, tensor<128xf32>) -> tensor<1x384x128xf32>

      time_major = false} : (
        tensor<1x2x3xf32>,
        tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>,
        tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>, tensor<1x1xf32>,
        none, none, none,
        tensor<3xf32>, tensor<3xf32>, tensor<3xf32>, tensor<3xf32>,
        none, none,
        tensor<1x3xf32>, tensor<1x3xf32>,
        none, none, none, none) -> tensor<1x2x3xf32>

module {
    func.func @main(%arg0: tensor<8x128xf32>, %arg1: tensor<f32>) -> tensor<11x131xf32> {
      %0 = "tfl.custom"(%arg0, %arg1) {custom_code = "stablehlo.pad", custom_option = #tfl<const_bytes : "0x656467655F70616464696E675F6869676800020203656467655F70616464696E675F6C6F7700020100696E746572696F725F70616464696E6700020000033E2A17030103311E0B2C2C2C062401">} : (tensor<8x128xf32>, tensor<f32>) -> tensor<11x131xf32>

  %1 = "tf.ReadVariableOp"(%0) {device = "/CPU:0"} : (tensor<!tf_type.resource<tensor<1x1xf32>>>) -> tensor<1x1xf32>
  %2:2 = "tf.Split"(%cst, %arg0) {device = "/CPU:0"} : (tensor<i32>,  tensor<1x4xf32>) -> (tensor<1x2xf32>, tensor<1x2xf32>)
  %3:2 = "tf.Split"(%cst, %2#0) {device = "/CPU:0"} : (tensor<i32>,  tensor<1x2xf32>) -> (tensor<1x1xf32>, tensor<1x1xf32>)

  %2:2 = "tf.IteratorGetNext"(%arg0) {device = "/device:CPU:0"} : (tensor<*x!tf_type.resource>) -> (tensor<128xf32>, tensor<128xf32>)
  %3:2 = "tf.IteratorGetNext"(%arg1) {device = "/device:CPU:0"} : (tensor<*x!tf_type.resource>) -> (tensor<128xf32>, tensor<128xf32>)
  // CHECK: "tf.TPUCompileSucceededAssert"(%[[COMPILE]]#0)
  "tf_device.launch"() ({