%one = "glue.constant"() { value = 1: i32 } : () -> i32
  %done = "glue.compare" (%one, %one) { predicate = #glue<"compare LTE"> } : (i32, i32) -> i1
  %2 = mhlo.constant dense<[[1.1]]> : tensor<1x1xf32>
  %3 = mhlo.multiply %2, %2 : tensor<1x1xf32>
  %cst = "tf.Const"() {value = dense<0.0> : tensor<f32>} : () -> tensor<f32>
  %0 = "tf.AddV2"(%arg0, %cst) {device = "/device:CPU:0"} : (tensor<f32>, tensor<f32>) -> tensor<f32>

}

// CHECK-LABEL: QuantizeConcat
func.func @QuantizeConcat(tensor<1x2xf32>, tensor<1x2xf32>) -> tensor<2x2x!quant.uniform<u8:f32, 1.000000e-01:128>> {
^bb0(%arg0: tensor<1x2xf32>, %arg1: tensor<1x2xf32>):
  %0 = "tfl.concatenation"(%arg0, %arg1) {axis = 0 : i32, fused_activation_function = "NONE"} : (tensor<1x2xf32>, tensor<1x2xf32>) -> tensor<2x2xf32>

  // Use other output
  %3:6 = "tf.FusedBatchNormV3"( %2#0, %arg1, %arg2, %arg3, %arg4) {T = "tfdtype$DT_FLOAT", U = "tfdtype$DT_FLOAT", data_format = "NHWC", epsilon = 0.001 : f32, is_training = false} : (tensor<8x8x8x8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>) -> (tensor<8x8x8x8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>, tensor<8xf32>)

}
func.func @_func(%arg0: tensor<2x4xf32>, %arg1: tensor<4x2xf32>) -> tensor<2x2xf32> {
  %0 = "tf.MatMul"(%arg0, %arg1) {_XlaSharding = "\08\03\1A\02\02\01\22\02\00\01"} : (tensor<2x4xf32>, tensor<4x2xf32>) -> tensor<2x2xf32>
  %1 = "tf.Identity"(%0) : (tensor<2x2xf32>) -> tensor<2x2xf32>
  return %1 : tensor<2x2xf32>
}

// -----
// The following op sharding is used in the following test case:

    %9 = "tfl.fully_connected"(%8, %cst_10, %cst_0) {fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<4x5xf32>, tensor<8x5xf32>, tensor<8xf32>) -> tensor<4x8xf32>
    %10:4 = "tfl.split"(%cst_5, %9) {num_splits = 4 : i32} : (tensor<i32>, tensor<4x8xf32>) -> (tensor<4x2xf32>, tensor<4x2xf32>, tensor<4x2xf32>, tensor<4x2xf32>)

func.func @main(%arg0: tensor<5x7xf32>) -> tensor<5x7xf32> {
  func.return %arg0: tensor<5x7xf32>
// CHECK-LABEL: main
// CHECK: return %arg0 : tensor<5x7xf32>
}

// - transpose
//
func.func @transpose_2d(%arg0: tensor<2x3xf32>) -> tensor<3x2xf32> {
  %0 = "mhlo.transpose"(%arg0) <{permutation = dense<[1, 0]> : tensor<2xi64>}> : (tensor<2x3xf32>) -> tensor<3x2xf32>
  func.return %0 : tensor<3x2xf32>

  %2 = "tf.Cast"(%1) {Truncate = false} : (tensor<1x2xbf16>) -> tensor<1x2xf32>
  %3 = "tf.IdentityN"(%2) {device = ""} : (tensor<1x2xf32>) -> tensor<1x2xf32>
  return %3 : tensor<1x2xf32>
}

// CHECK: func @cast_bf16_matmul_to_fp32
// CHECK-DAG: %[[cst:.*]] = "tf.Const"() <{value = dense<{{.*}}> : tensor<10x2xf32>}> : () -> tensor<10x2xf32>
// CHECK: %[[matmul:.*]] = "tf.MatMul"(%arg0, %[[cst]])

// -----

func.func @testAddReluPack(%arg0: tensor<1xf32>, %arg1: tensor<1xf32>) {
   // CHECK: tac.device = "GPU", tac.inference_type = "FLOAT"
  %0 = "tfl.add"(%arg0, %arg1) {fused_activation_function = "RELU6"} : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>
   // CHECK: tac.device = "GPU", tac.inference_type = "FLOAT"
  %1 = "tfl.add"(%arg0, %0) {fused_activation_function = "RELU"} : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>

// CHECK:           %[[VAL_4:.*]]:2 = call @func_0_GPU_FLOAT(%[[VAL_0]], %[[VAL_1]], %[[VAL_2]], %[[VAL_3]]) {tac.device = "GPU", tac.inference_type = "FLOAT", tac.interface_name = "func_0"} : (tensor<1xf32>, tensor<1xf32>, tensor<1xf32>, tensor<1xf32>) -> (tensor<1xf32>, tensor<1xf32>)

func.func @add_dense_dense_float_mixfng_1_n() -> tensor<2x2xf32> {
  %cst_0 = arith.constant dense<[[1.5, -2.5]]> : tensor<1x2xf32>
  %cst_1 = arith.constant dense<[[-3.], [4.]]> : tensor<2x1xf32>

  %0 = "tfl.add"(%cst_0, %cst_1) {fused_activation_function = "NONE"} : (tensor<1x2xf32>, tensor<2x1xf32>) -> tensor<2x2xf32>

  func.return %0 : tensor<2x2xf32>