%1 = "tfl.add"(%cst_2, %cst_1) {fused_activation_function = "NONE"} : (tensor<2x2x2xf32>, tensor<  2x2xf32>) -> tensor<2x2x2xf32>
  %2 = "tfl.add"(%cst_0, %cst_2) {fused_activation_function = "NONE"} : (tensor<    2xf32>, tensor<2x2x2xf32>) -> tensor<2x2x2xf32>

  func.return %0, %1, %2 : tensor<2x2xf32>, tensor<2x2x2xf32>, tensor<2x2x2xf32>

  %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.return %6 : tensor<1x401408xf32>
}

// CHECK-LABEL: quantized_constant
func.func @quantized_constant(%arg0: tensor<1x2xf32>) -> tensor<2x2xf32> {
  %1 = "tfl.quantize"(%arg0) {qtype = tensor<1x2x!quant.uniform<u8:f32, 1.0>>, volatile} : (tensor<1x2xf32>) -> tensor<1x2x!quant.uniform<u8:f32, 1.0>>

// CHECK-LABEL: fuseTileWithBinaryOp
func.func @fuseTileWithBinaryOp(%arg0: tensor<1x1xf32>) -> tensor<1x2xf32> {
  %cst = arith.constant dense<[[1,2]]> : tensor<1x2xi32>
  %cst1 = arith.constant dense<[[3.0, 4.0]]> : tensor<1x2xf32>
  %0 = "tfl.sqrt"(%arg0) : (tensor<1x1xf32>) -> tensor<1x1xf32>
  %1 = "tfl.tile"(%0, %cst) : (tensor<1x1xf32>, tensor<1x2xi32>) -> tensor<1x2xf32>

  // CHECK-SAME: %[[ARG0:.*]]: tensor<3x2xf32>
  // CHECK-SAME: %[[ARG1:.*]]: tensor<1x2xf32>
  // CHECK-SAME: %[[ARG2:.*]]: tensor<2x5xf32>
  func.func private @concatenate_and_composite(%arg0: tensor<3x2xf32>, %arg1: tensor<1x2xf32>, %arg2: tensor<2x5xf32>) -> tensor<4x5xf32> {
    // CHECK: %[[Q1:.*]] = "quantfork.qcast"(%[[ARG0]]) {volatile} : (tensor<3x2xf32>) -> tensor<3x2x!quant.uniform<i8:f32, 5.000000e-03:-1>>

    %7 = call @uniform_quantize_0(%arg0, %0, %1) : (tensor<1x2xf32>, tensor<1x1xf32>, tensor<1x1xi8>) -> tensor<1x2xi32>
    %8 = stablehlo.convert %7 : (tensor<1x2xi32>) -> tensor<1x2xf32>
    %9 = stablehlo.convert %2 : (tensor<2x3xi8>) -> tensor<2x3xf32>
    %10 = stablehlo.dot_general %8, %9, contracting_dims = [1] x [0] : (tensor<1x2xf32>, tensor<2x3xf32>) -> tensor<1x3xf32>
    %11 = stablehlo.convert %3 : (tensor<1x3xi32>) -> tensor<1x3xf32>

        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>,
        none, none, none, none) -> tensor<*xf32>

      return %0 : tensor<2x2xf32>
    }
    func.func private @composite_fn_1(%arg0: tensor<2x2xf32>, %arg1: tensor<2x2xf32>) -> tensor<2x2xf32> attributes {tf_quant.composite_function} {

}
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:

  // CHECK-DAG: %[[EXP:.*]] = "tf.Exp"(%[[SHIFTED]]) : (tensor<2x3xf32>) -> tensor<2x3xf32>
  // CHECK-DAG: %[[SUM:.*]] = "tf.Sum"(%[[EXP]], %[[AXIS]]) <{keep_dims = true}> : (tensor<2x3xf32>, tensor<1xi64>) -> tensor<2x1xf32>
  // CHECK-DAG: %[[RESULT:.*]] = "tf.Div"(%[[EXP]], %[[SUM]]) : (tensor<2x3xf32>, tensor<2x1xf32>) -> tensor<2x3xf32>
  // CHECK: return %[[RESULT]]