%5 = "quantfork.qcast"(%4) {volatile} : (tensor<1x3xf32>) -> tensor<1x3x!quant.uniform<i8:f32, 0.13170163023705575:-1>>
    %6 = "quantfork.dcast"(%5) : (tensor<1x3x!quant.uniform<i8:f32, 0.13170163023705575:-1>>) -> tensor<1x3xf32>
    %7 = stablehlo.reshape %6 : (tensor<1x3xf32>) -> tensor<3x1xf32>

func.func @main(%arg0: tensor<1x4xf32>) -> tensor<1x3xf32> {
  %0 = stablehlo.constant dense<1.0> : tensor<4x3xf32>
  %1 = stablehlo.dot_general %arg0, %0, contracting_dims = [1] x [0], precision = [DEFAULT, DEFAULT] : (tensor<1x4xf32>, tensor<4x3xf32>) -> tensor<1x3xf32>
  return %1 : tensor<1x3xf32>
}
// CHECK: @main(%[[ARG_0:.+]]: tensor<1x4xf32>) -> tensor<1x3xf32>

    return %1 : tensor<1x3xf32>
  }

  func.func private @composite_dot_general_fn(%arg0: tensor<1x2xf32>, %arg1: tensor<2x3xf32>) -> tensor<1x3xf32> attributes {_from_xla_call_module} {
    %0 = stablehlo.dot_general %arg0, %arg1, contracting_dims = [1] x [0] : (tensor<1x2xf32>, tensor<2x3xf32>) -> tensor<1x3xf32>
    return %0 : tensor<1x3xf32>
  }
}

}

// 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>

  %1 = "tf.Transpose"(%0, %cst_0): (tensor<1x2xf32>, tensor<2xi32>) -> tensor<2x1xf32>
  %2 = "tf.Transpose"(%0, %cst_0): (tensor<1x2xf32>, tensor<2xi32>) -> tensor<2x1xf32>
  func.return %1, %2 : tensor<2x1xf32>, tensor<2x1xf32>

// CHECK:  %cst = arith.constant

  %1 = "tf.Transpose"(%0, %cst_0): (tensor<1x2xf32>, tensor<2xi32>) -> tensor<2x1xf32>
  %2 = "tf.Transpose"(%0, %cst_0): (tensor<1x2xf32>, tensor<2xi32>) -> tensor<2x1xf32>
  func.return %1, %2 : tensor<2x1xf32>, tensor<2x1xf32>

// CHECK:  %cst = arith.constant

    %3 = stablehlo.concatenate %2, %1, dim = 0 : (tensor<1x3xf32>, tensor<1x3xf32>) -> tensor<2x3xf32>
    return %3 : tensor<2x3xf32>
  }
  func.func private @composite_dot_general_fn_1(%arg0: tensor<1x2xf32>, %arg1: tensor<2x3xf32>) -> tensor<1x3xf32> attributes {_from_xla_call_module, tf_quant.composite_function} {

  %8 = "tfl.concatenation"(%2, %0) {axis = -1 : i32, fused_activation_function = "NONE"} : (tensor<1x1xf32>, tensor<1x1xf32>) -> tensor<1x2xf32>
  %9 = "quantfork.stats"(%8) {layerStats = dense<[-0.488159984, 0.189515018]> : tensor<2xf32>} : (tensor<1x2xf32>) -> tensor<1x2xf32>
  %10 = "tfl.concatenation"(%9, %7) {axis = -1 : i32, fused_activation_function = "NONE"} : (tensor<1x2xf32>, tensor<1x2xf32>) -> tensor<1x4xf32>

        return %2 : tensor<?x2xf32>
      }
      func.func private @composite_fn_1(%arg0: tensor<?x2xf32>, %arg1: tensor<2x2xf32>, %arg2: tensor<2xf32>) -> tensor<?x2xf32> attributes {_from_xla_call_module, tf_quant.composite_function} {
        return %arg0 : tensor<?x2xf32>
      }
    }
  )mlir";

func.func @RemoveRedundantPack(%arg0: tensor<2x5xf32>) -> (tensor<2x5xf32>, tensor<5xf32>) {
  %0:2 = "tfl.unpack"(%arg0) {axis = 0 : i32, num = 2 : i32} : (tensor<2x5xf32>) -> (tensor<5xf32>, tensor<5xf32>)
  %1 = "tfl.pack"(%0#0, %0#1) {axis = 0 : i32, values_count = 2 : i32} : (tensor<5xf32>, tensor<5xf32>) -> (tensor<2x5xf32>)
  func.return %1, %0#0: tensor<2x5xf32>, tensor<5xf32>
  // CHECK: %[[UNPACK:.*]]:2 = "tfl.unpack"