// CHECK: return %[[v4]] : tensor<?x36x32xf32>
}

func.func @einsum_with_runtime_shape2(%arg0 : tensor<?x?x8x64xf32>, %arg1 : tensor<8x8x64xf32>) -> tensor<?x?x8xf32> {
  %0 = "tf.Einsum"(%arg0, %arg1) {device = "", equation = "ABNH,DNH->ABD"} : (tensor<?x?x8x64xf32>, tensor<8x8x64xf32>) -> tensor<?x?x8xf32>
  func.return %0 : tensor<?x?x8xf32>
// CHECK-LABEL: einsum_with_runtime_shape2

func.func @check_propagation_for_output_sharding_from_tf_matmul(%arg0: tensor<2x4xf32>, %arg1: tensor<4x2xf32>) -> (tensor<1x2xf32>, tensor<1x2xf32>) {
  %0 = "tf_device.cluster_func"(%arg0, %arg1) {func = @_func, use_spmd_for_xla_partitioning = true, use_tpu = true, num_cores_per_replica = 2 : i64} : (tensor<2x4xf32>, tensor<4x2xf32>) -> tensor<2x2xf32>

func.func @testPackedTPUPartitionedInputV2(tensor<2x4xf32>, tensor<2x4xf32>) -> tensor<4x4xf32> {
^bb0(%arg0: tensor<2x4xf32>, %arg1: tensor<2x4xf32>):
  // expected-error @+1 {{expected 1 inputs, got 2}}
  %0 = "tf.TPUPartitionedInputV2"(%arg0, %arg1) {partition_dims = [2, 1], is_packed = true} : (tensor<2x4xf32>, tensor<2x4xf32>) -> tensor<4x4xf32>
  func.return %0 : tensor<4x4xf32>
}

// -----

    %13 = "tf.Mul"(%12, %cst_0) {device = ""} : (tensor<2x4xf32>, tensor<f32>) -> tensor<2x4xf32>
    %14 = "tf.Relu"(%13) {device = ""} : (tensor<2x4xf32>) -> tensor<2x4xf32>
    %15 = "tf.Minimum"(%14, %cst) {device = ""} : (tensor<2x4xf32>, tensor<f32>) -> tensor<2x4xf32>
    %16 = "tf.Identity"(%15) {device = ""} : (tensor<2x4xf32>) -> tensor<2x4xf32>

  // CHECK:  return
}

func.func @tile(tensor<2x3xf32>, tensor<2xi32>) -> tensor<2x6xf32> {
^bb0(%arg0: tensor<2x3xf32>, %arg1: tensor<2xi32>):
  %cst = "tf.Const"() { value = dense<[1, 2]> : tensor<2xi32> } : () -> tensor<2xi32>
  %0 = "tf.Tile"(%arg0, %cst) : (tensor<2x3xf32>, tensor<2xi32>) -> tensor<2x6xf32>
  func.return %0 : tensor<2x6xf32>

  // CHECK-LABEL: tile

// -----

// CHECK-LABEL: func @einsum
func.func @einsum(%arg0: tensor<2x3xf32>, %arg1: tensor<3x4xf32>) -> tensor<2x4xf32> {
  // CHECK:  mhlo.einsum
  %0 = "tf.Einsum"(%arg0, %arg1) {equation = "ab,bc->ac"} : (tensor<2x3xf32>, tensor<3x4xf32>) -> tensor<2x4xf32>
  func.return %0: tensor<2x4xf32>
}

// -----

// CHECK-LABEL: func @unary_einsum

  %30 = "tf.GatherV2"(%9, %29, %5) {batch_dims = 0 : i64, device = ""} : (tensor<32x1xf32>, tensor<2x16xi32>, tensor<i32>) -> tensor<2x16x1xf32>
  %31 = "tf.AddV2"(%19, %28) {device = ""} : (tensor<2x16xi32>, tensor<2x16xi32>) -> tensor<2x16xi32>
  %32 = "tf.GatherV2"(%9, %31, %5) {batch_dims = 0 : i64, device = ""} : (tensor<32x1xf32>, tensor<2x16xi32>, tensor<i32>) -> tensor<2x16x1xf32>

%0 = "mhlo.dot_general"(%arg0, %arg1) {
  dot_dimension_numbers = #mhlo.dot<
    lhs_batching_dimensions = [0, 1],
    rhs_batching_dimensions = [0, 1],
    lhs_contracting_dimensions = [3],
    rhs_contracting_dimensions = [3]
  >} : (tensor<2x?x2x3xf32>, tensor<2x?x4x3xf32>) -> tensor<2x?x2x4xf32>
func.return %0 : tensor<2x?x2x4xf32>

  %4 = "tfl.dequantize"(%3) : (tensor<32x12x!quant.uniform<u8<1:255>:f32, 0.021826678373682216:151>>) -> tensor<32x12xf32>
  %5 = "tfl.fully_connected"(%2, %4, %cst) {fused_activation_function = "NONE", keep_num_dims = false, weights_format = "DEFAULT"} : (tensor<1x224x224x3xf32>, tensor<32x12xf32>, tensor<32xf32>) -> tensor<1x112x112x32xf32>

    %8 = "tf.ConcatV2"(%2, %6, %cst_1) : (tensor<2x2xf32>, tensor<2x2xf32>, tensor<i32>) -> tensor<2x4xf32>
    return %8 : tensor<2x4xf32>
  }
  func.func private @composite_matmul_fn_2(%arg0: tensor<2x2xf32>, %arg1: tensor<2x2xf32>) -> tensor<2x2xf32> attributes {tf_quant.composite_function} {