// CHECK: "tfl.unidirectional_sequence_lstm"(%arg0, %arg1, %arg2, %arg3, %arg4, %arg5, %arg6, %arg7, %arg8, %arg9, %arg10, %arg11, %arg12, %arg13, %arg14, %arg15, %arg16, %arg17, %arg18, %arg19, %arg20, %arg21, %arg22, %arg23) <{fused_activation_function = "NONE", time_major = false}>...

// CHECK: %[[NONE:.+]] = "tfl.no_value"() <{value}> : () -> none
// CHECK: "tfl.lstm"(%arg0, %arg1, %arg2, %arg3, %arg4, %arg5, %arg6, %arg7, %arg8, %[[NONE]], %[[NONE]], %[[NONE]], %arg9, %arg10, %arg11, %arg12, %arg13, %[[NONE]], %arg19, %arg20, %arg15, %arg16, %arg17, %arg18)
}

func.func @keepCustomFlexOps(%arg0: tensor<1x10xf32>) -> tensor<1x10xf32> {

// CHECK-LABEL: testLiftDotConcatLHSAndRHS
func.func @testLiftDotConcatLHSAndRHS(%arg0: tensor<1x72x128xf32>, %arg1: tensor<1x128x72xf32>, %arg2: tensor<1x72x128xf32>, %arg3: tensor<1x128x72xf32>, %arg4: tensor<1x72x128xf32>, %arg5: tensor<1x128x72xf32>, %arg6: tensor<1x72x128xf32>, %arg7: tensor<1x128x72xf32>) -> tensor<4x72x72xf32> {
  %0 = "mhlo.dot_general"(%arg0, %arg1) {
    dot_dimension_numbers = #mhlo.dot<
      lhs_batching_dimensions = [0],

  "tf.AssignVariableOp"(%arg0, %arg2) : (tensor<*x!tf_type.resource<tensor<32xf32>>>, tensor<32xf32>) -> ()
  // CHECK-NEXT: "tf.AssignVariableOp"(%[[ARG_5]], %[[ARG_6]])
  "tf.AssignVariableOp"(%arg5, %arg6) : (tensor<*x!tf_type.resource<tensor<2xf32>>>, tensor<2xf32>) -> ()
  %read1 = "tf.ReadVariableOp"(%arg1) : (tensor<*x!tf_type.resource<tensor<64xf32>>>) -> tensor<64xf32>

  // CHECK: %0 = "tf.PartitionedCall"(%cst, %arg0) <{config = "", config_proto = "", executor_type = "", f = @"sort_map/while_cond/TfMlrtAsyncWhilePredicate"}> : (tensor<i32>, tensor<i32>) -> tensor<i1>

//CHECK-NEXT: return %0 : tensor<i1>
//CHECK-NEXT:}

func.func @dynamic_update_slice(%arg0: tensor<4x4xi64>, %arg1: tensor<2x3xi64>, %arg2: tensor<i64>, %arg3: tensor<i64>) -> tensor<4x4xi64> {

    %cst = "tfl.no_value"() {value = unit} : () -> none

// -----

// CHECK-LABEL: call_op
func.func @call_op(%arg0: !tfr.tensor<T>, %arg1: !tfr.tensor_list<TL>, %arg2: i32) -> !tfr.tensor<K> {
  %0 = tfr.call @Foo(%arg0, %arg1, %arg2) : (!tfr.tensor<T>, !tfr.tensor_list<TL>, i32) -> !tfr.tensor<K>
  func.return %0 : !tfr.tensor<K>
}

// -----

// CHECK-LABEL: call_op_arg_attr(%arg0: i32) -> !tfr.tensor<K>
func.func @call_op_arg_attr(%arg0: i32) -> !tfr.tensor<K> {

    // CHECK: %[[OUT:.*]], %[[MEAN:.*]], %[[VAR:.*]] = "mhlo.batch_norm_training"({{.*}}, %arg1, %arg2) <{epsilon = 1.000000e-03 : f32, feature_index = 3 : i64}> : (tensor<8x8x8x8xf32>, tensor<8xf32>, tensor<8xf32>) -> (tensor<8x8x8x8xf32>, tensor<8xf32>, tensor<8xf32>)

  %4:2 = "mhlo.reduce"(%arg0, %3, %0, %1) ({
  ^bb0(%arg1: tensor<f32>, %arg2: tensor<i32>, %arg3: tensor<f32>, %arg4: tensor<i32>):
    %7 = "mhlo.compare"(%arg1, %arg3) {comparison_direction = #mhlo<comparison_direction GT>} : (tensor<f32>, tensor<f32>) -> tensor<i1>
    %8 = "mhlo.compare"(%arg1, %arg1) {comparison_direction = #mhlo<comparison_direction NE>} : (tensor<f32>, tensor<f32>) -> tensor<i1>