// 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.

  %3 = "tf.Add"(%2, %arg1) : (tensor<?x8x10xf32>, tensor<8x10xf32>) -> tensor<?x8x10xf32>
  %4 = "tf.Add"(%2, %arg2) : (tensor<?x8x10xf32>, tensor<8x10xf32>) -> tensor<?x8x10xf32>
  %5 = "tf.Add"(%arg1, %arg2) : (tensor<8x10xf32>, tensor<8x10xf32>) -> tensor<8x10xf32>

  %0 = arith.constant dense<-1> : tensor<1xi32>
  %1 = "tfl.reduce_max"(%arg0, %0) {keep_dims = true} : (tensor<10x10xf32>, tensor<1xi32>) -> tensor<10x1xf32>
  %2 = tfl.sub(%arg0, %1) {fused_activation_function = "NONE"} : (tensor<10x10xf32>, tensor<10x1xf32>) -> tensor<10x10xf32>
  %3 = "tfl.exp"(%2) : (tensor<10x10xf32>) -> tensor<10x10xf32>

module attributes {tf.versions = {producer = 179 : i32}} {
  func.func @main(%arg0: tensor<10x19xf32>, %arg1: tensor<19x10xf32> {mhlo.is_same_data_across_replicas = true}) -> tensor<10x19xf32> {
    %0 = "tf.Shape"(%arg0) : (tensor<10x19xf32>) -> tensor<2xi64>
    %1 = "tf.Reshape"(%arg1, %0) : (tensor<19x10xf32>, tensor<2xi64>) -> tensor<10x19xf32>
    func.return %1 : tensor<10x19xf32>
  }
}

// -----

func.func @pack_CPU(%arg0: tensor<100xf32>, %arg1: tensor<100xf32>) -> tensor<2x100xf32> attributes {tac.device = "CPU", tac.interface_name = "func_2"} {
  // CHECK: tac.cost = 1.000000e+02
  %0 = "tfl.pack"(%arg0, %arg1) {axis = 0 : i32, tac.device = "CPU", values_count = 2 : i32} : (tensor<100xf32>, tensor<100xf32>) -> tensor<2x100xf32>
  func.return %0 : tensor<2x100xf32>
}

module attributes {tf.versions = {producer = 179 : i32}} {

  // CHECK-LABEL: func @main
  func.func @main(%arg0: tensor<10x19xf32>, %arg1: tensor<19x10xf32> {mhlo.is_same_data_across_replicas = true}) -> tensor<?xi64> {
    %0 = "tf.Shape"(%arg0) : (tensor<10x19xf32>) -> tensor<2xi64>
    %1 = "tf.Reshape"(%arg1, %0) : (tensor<19x10xf32>, tensor<2xi64>) -> tensor<?x?xf32>

    // CHECK: %[[RESULT:.*]] = mhlo.constant dense<[10, 19]>

  } : (tensor<9x10x!tf_type.qint8>, tensor<f32>, tensor<i32>) -> tensor<9x10xf32>
  %0 = "tf.MatMul"(%input, %filter_new) {
  } : (tensor<8x9xf32>, tensor<9x10xf32>) -> tensor<8x10xf32>
  return %0 : tensor<8x10xf32>
})mlir";
  constexpr absl::string_view kProgram = R"mlir(
func.func @main(%input: tensor<8x9xf32>, %filter: tensor<9x10xi8>) -> tensor<8x10xf32> {
  %filter_scale = "tf.Const"() { value = dense<0.0235> : tensor<f32> } : ()

  %1 = "tf.MatMul"(%arg0, %arg0) {device = "", transpose_a = false, transpose_b = false} : (tensor<10x10xi32>, tensor<10x10xi32>) -> tensor<10x10xi32>
  %2 = "tf.PreventGradient"(%0) : (tensor<10x10xi32>) -> tensor<10x10xi32>
  %3 = "tf.PreventGradient"(%1) : (tensor<10x10xi32>) -> tensor<10x10xi32>
  %4 = "tf.AddV2"(%2, %3) {device = ""} : (tensor<10x10xi32>, tensor<10x10xi32>) -> tensor<10x10xi32>

// -----

func.func @handles_mhlo_regions() {
  "glue.someop"() : () -> ()
  %4 = mhlo.constant dense<0.0> : tensor<200x10xf32>
  %13 = mhlo.constant dense<-0.0> : tensor<f32>
  %29 = mhlo.reduce(%4 init: %13) applies mhlo.add across dimensions = [1] : (tensor<200x10xf32>, tensor<f32>) -> tensor<200xf32>
  return
}

// CHECK: func @handles_mhlo_regions
// CHECK: call [[name:@[^(]*]](
// CHECK: func [[name]]()

func.func @testSoftmax(%arg0: tensor<10x10xf32>) -> tensor<10x10xf32> {
  // CHECK: _arithmetic_count = 6400 : i64
  %0 = "tfl.softmax"(%arg0) {beta = 1.000000e+00 : f32} : (tensor<10x10xf32>) -> tensor<10x10xf32>
  func.return %0 : tensor<10x10xf32>
}

func.func @testTanh(%arg0: tensor<10x10xf32>) -> tensor<10x10xf32> {
  // CHECK: _arithmetic_count = 6400 : i64
  %0 = "tfl.tanh"(%arg0) : (tensor<10x10xf32>) -> tensor<10x10xf32>