%3 = "tf.AddN"(%1, %arg0, %1) : (tensor<2xf32>, tensor<2xf32> , tensor<2xf32>) -> tensor<2xf32>
  %4 = "tf.AddN"(%1, %1) : (tensor<2xf32>, tensor<2xf32>) -> tensor<2xf32>
  %5 = "tf.AddN"(%arg0, %1, %0) : (tensor<2xf32>, tensor<2xf32>, tensor<2xf32>) -> tensor<2xf32>
  func.return %2, %3, %4, %5: tensor<2xf32>, tensor<2xf32>, tensor<2xf32>, tensor<2xf32>
}

// CHECK-LABEL: func @addNWithZerosInt

  %0 = "tfl.unidirectional_sequence_lstm"(%arg0,...

  func.return %0 : tensor<?x?xf32>

// CHECK-LABEL: dynamicReshape
// CHECK-NEXT:  %[[reshape:.*]] = "tfl.reshape"(%arg0, %arg1) : (tensor<*xf32>, tensor<2xi32>) -> tensor<?x?xf32>
// CHECK-NEXT:  return %[[reshape]] : tensor<?x?xf32>
}

func.func @dynamicReshapeI64(%arg0: tensor<*xf32>, %arg1: tensor<2xi64>) -> tensor<?x?xf32> {
  %0 = "tf.Reshape"(%arg0, %arg1) : (tensor<*xf32>, tensor<2xi64>) -> tensor<?x?xf32>

// CHECK:           }) : (tensor<1x?xf32>, tensor<1x0xf32>, tensor<1x0xf32>, tensor<1x0xf32>, tensor<1x0xf32>, tensor<1x3xf32>, tensor<1x3xf32>, tensor<1x3xf32>, tensor<1x3xf32>, none, none, none, tensor<1xf32>, tensor<1xf32>, tensor<1xf32>, tensor<1xf32>, tensor<3x1xf32>, tensor<3xf32>, tensor<1x3xf32>, tensor<1x1xf32>, tensor<1xf32>, tensor<1xf32>, tensor<1xf32>, tensor<1xf32>) -> tensor<1x3xf32>

          %1 = call @then_branch_func(%arg1) : (tensor<*xf32>) -> tensor<*xf32>
          "tf.Yield"(%1) : (tensor<*xf32>) -> ()
        },  {
          %1 = call @else_branch_func(%arg1) : (tensor<*xf32>) -> tensor<*xf32>
          "tf.Yield"(%1) : (tensor<*xf32>) -> ()
        }) {is_stateless = false} : (tensor<i1>) -> tensor<*xf32>
    ```

    will be transformed into this functional operation

// `tfl.reshape`.

func.func @dynamic_reshape_float(%arg0: tensor<?x3xf32>, %arg1: tensor<2xi32>) -> tensor<?x?xf32> {
  %0 = "stablehlo.dynamic_reshape"(%arg0, %arg1) : (tensor<?x3xf32>, tensor<2xi32>) -> tensor<?x?xf32>
  return %0 : tensor<?x?xf32>
}

// CHECK-LABEL: func @dynamic_reshape_float
// CHECK: stablehlo.dynamic_reshape
// CHECK-NOT: tfl.reshape

// -----

  // The first Operand is assumed to be a TensorType around a variant with a
  // single subtype (e.g. tensor<!tf_type.variant<tensor<2xi32>>>). We will
  // copy this type to the first result, and copy the singular variant subtype
  // to the second result (tensor<2xi32>).
  DCOMMENT_OP(op, "Inferring shape for TensorListPopBackOp.");

  auto src_list_handle_t =

//   %res = "tfl.concatenation"(%unpack#0, %unpack#1, %unpack#2)
//        {axis = -1 : i32, fused_activation_function = "NONE"}
//
// can be optimized to
//
//   %cst = arith.constant dense<[1, 6]> : tensor<2xi32>
//   %res = "tfl.reshape"(%input, %cst)
struct FuseUnpackAndConcatToReshape
    : public OpRewritePattern<TFL::ConcatenationOp> {
  using OpRewritePattern::OpRewritePattern;