} : (tensor<?x100352xf32>, tensor<100352x10xf32>) -> tensor<?x10xf32>
    %3 = "tf.Identity"(%2) {device = ""} : (tensor<?x10xf32>) -> tensor<?x10xf32>
    return %3 : tensor<?x10xf32>
  }

  // CHECK-LABEL: func.func private @composite_dot_general_fn_1
  func.func private @composite_dot_general_fn_1(%arg0: tensor<?x100352xf32>, %arg1: tensor<100352x10xf32>) -> tensor<?x10xf32> {
    // CHECK-NOT: tf.CustomAggregator

}

// -----

// CHECK-LABEL: func @testBatchMatMulV2NoBatchDimension
func.func @testBatchMatMulV2NoBatchDimension(%lhs: tensor<5x10xf32>, %rhs: tensor<10x10xf32>) -> (tensor<5x10xf32>) {
  %0 = "tf.BatchMatMulV2"(%lhs, %rhs) : (tensor<5x10xf32>, tensor<10x10xf32>) -> tensor<5x10xf32>
  func.return %0 : tensor<5x10xf32>
}

// -----

// -----

module {
  func.func @same_scale_ptq_test(%arg0: tensor<*xf32>) -> tensor<*xf32> {
    %cst = arith.constant dense<[-1, 144]> : tensor<2xi32>
    %cst_1 = arith.constant dense<1.0> : tensor<144x10xf32>
    %cst_2 = arith.constant dense<0.1> : tensor<10xf32>
    %0 = "quantfork.stats"(%arg0) {
      layerStats = dense<[-1.28, 1.28]> : tensor<2xf32>
    } : (tensor<*xf32>) -> tensor<*xf32>
    %1 = "tf.MaxPool"(%0) {

  %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: testFoldEnsureShapeOp
func.func @testFoldEnsureShapeOp(%arg0: tensor<10x20xf32>) -> (tensor<10x20xf32>, tensor<10x20xf32>, tensor<20x10xf32>) {
  %0 = "tf.EnsureShape"(%arg0) {shape = #tf_type.shape<10x20>} : (tensor<10x20xf32>) -> tensor<10x20xf32>
  %1 = "tf.EnsureShape"(%arg0) {shape = #tf_type.shape<?x20>} : (tensor<10x20xf32>) -> tensor<10x20xf32>
  // Failing case which should not be folded.

  %2 = "tf.Add"(%1, %1) {
    _xla_compile_device_type = "TPU", _replication_info = "cluster",
    device = "/task:0/device:TPU:0", dtype = f32
  } : (tensor<1x80xf32>, tensor<1x80xf32>) -> tensor<1x80xf32>
  %3 = "tf.ResourceGatherNd"(%arg0, %0) {
    Tindices = i32
  } : (tensor<*x!tf_type.resource<tensor<80xf32>>>, tensor<i32>) -> tensor<1x80xf32>

    ```mlir
      %0 = "tf.Const"() {value = dense<[[42.0]]> : tensor<1x1xf32>} : () -> tensor<1x1xf32>
      %1 = "tf.Const"() {device = "", value = dense<[[42.0]]> : tensor<1x1xf32>} : () -> tensor<1x1xf32>
      %2 = "tf.Const"() {device = "baz", value = dense<[[42.0]]> : tensor<1x1xf32>} : () -> tensor<1x1xf32>
    ```

    then running this pass with 'default-device=foobar', we get:

    ```mlir

  %0 = "mhlo.reshape"(%arg0) : (tensor<1x1x512xf32>) -> tensor<1x512xf32>
  %1 = "mhlo.dot"(%0, %arg1) : (tensor<1x512xf32>, tensor<512x13x!quant.uniform<i8:f32, 0.00285>>) -> tensor<1x13xf32>
  %2 = "mhlo.reshape"(%1) : (tensor<1x13xf32>) -> tensor<1x1x13xf32>
  func.return %2 : tensor<1x1x13xf32>

// CHECK:      %[[RES:.*]] = "mhlo.dot_general"(%arg0, %arg1) <{
// CHECK-SAME:   dot_dimension_numbers = #mhlo.dot<

    // CHECK-NEXT: return %[[DIV]] : tensor<4x10xf32>
    %0 = "tf.Softsign"(%arg0) : (tensor<4x10xf32>) -> tensor<4x10xf32>
    func.return %0 : tensor<4x10xf32>
}

// -----

// CHECK-LABEL: func @softsign_grad
func.func @softsign_grad(%arg0: tensor<4x10xf32>, %arg1: tensor<4x10xf32>) -> tensor<4x10xf32> {

    // CHECK-NEXT: %[[ONE:.*]] = mhlo.constant dense<1.000000e+00> : tensor<f32>

  // CHECK-LABEL: @xla_svd
  func.func @xla_svd(%arg0: tensor<1x1xf32>) -> (tensor<1xf32>, tensor<1x1xf32>, tensor<1x1xf32>) {
    // CHECK-NOT: XlaSvd
    %s, %u, %v = "tf.XlaSvd"(%arg0) {max_iter = 1, epsilon = 1.0E-09 : f32, precision_config = ""} : (tensor<1x1xf32>) -> (tensor<1xf32>, tensor<1x1xf32>, tensor<1x1xf32>)
    func.return %s, %u, %v : tensor<1xf32>, tensor<1x1xf32>, tensor<1x1xf32>
  }

  func.func @identity(%arg0: f32) -> f32 {