func.func @reshape_removeIdentity(tensor<4x4x4xf32>) -> tensor<4x4x4xf32> {
^bb0(%arg0: tensor<4x4x4xf32>) :
  %cst = arith.constant dense<[4, 4, 4]> : tensor<3xi32>
  %0 = "tfl.reshape"(%arg0, %cst) : (tensor<4x4x4xf32>, tensor<3xi32>) -> tensor<4x4x4xf32>
  func.return %0 : tensor<4x4x4xf32>

// CHECK-LABEL: func @reshape_removeIdentity
// CHECK:  return %arg0 : tensor<4x4x4xf32>
}

static const char *const module_with_add =
    R"(module {
func.func @main(%arg0: tensor<3x4x5xf32>, %arg1: tensor<3x4x5xf32>) -> tensor<3x4x5xf32> {
  %0 = "tf.AddV2"(%arg0, %arg1) : (tensor<3x4x5xf32>, tensor<3x4x5xf32>) -> tensor<3x4x5xf32>
  func.return %0 : tensor<3x4x5xf32>
}
}
)";

static const char *const module_with_sub =
    R"(module {

  func.return %2 : tensor<4x4x4xf32>
}

static const char *const module_with_add =
    R"(module {
func.func @main(%arg0: tensor<3x4x5xf32>, %arg1: tensor<3x4x5xf32>) -> tensor<3x4x5xf32> {
  %0 = "tf.AddV2"(%arg0, %arg1) : (tensor<3x4x5xf32>, tensor<3x4x5xf32>) -> tensor<3x4x5xf32>
  func.return %0 : tensor<3x4x5xf32>
}
}
)";

// Test pass filter.
TEST(DataDumperLoggerConfig, TestPassFilter) {
  mlir::DialectRegistry mlir_registry;

func.func @lower_einsum(%arg0: tensor<3x4x5xf32>, %arg1: tensor<3x5x6xf32>) -> tensor<3x4x6xf32> {
  %0 = "tf.Einsum"(%arg0, %arg1) {T = "tfdtype$DT_FLOAT", equation = "ijk,ikm->ijm"}: (tensor<3x4x5xf32>, tensor<3x5x6xf32>) -> tensor<3x4x6xf32>
  func.return %0 : tensor<3x4x6xf32>
}
// CHECK-LABEL: lower_einsum
// CHECK: "tf.BatchMatMulV2"(%arg0, %arg1) <{adj_x = false, adj_y = false}> : (tensor<3x4x5xf32>, tensor<3x5x6xf32>) -> tensor<3x4x6xf32>

// CHECK-NEXT: return %[[ES]] : tensor<3x2x3xf32>
}

// CHECK-LABEL: my_leaky_relu
func.func @my_leaky_relu(%arg0: tensor<2x3xf32>) -> tensor<3x2x3xf32> {
  %0 = "tf.MyLeakyRelu"(%arg0) {alpha=3.0 : f32} : (tensor<2x3xf32>) -> tensor<3x2x3xf32>
  func.return %0 : tensor<3x2x3xf32>

func.func @xla_dot_v2(%arg0: tensor<?x2x3xf32>, %arg1: tensor<3x4x5xf32>) -> (tensor<?x2x4x5xf32>) {
  %0 = "tf.XlaDotV2"(%arg0, %arg1) {device = "", dimension_numbers = "\0A\01\02\12\01\00", precision_config = ""} : (tensor<?x2x3xf32>, tensor<3x4x5xf32>) -> tensor<?x2x4x5xf32>
  func.return %0 : tensor<?x2x4x5xf32>
}

// CHECK: func @xla_dot_v2

func.func @main(tensor<4x4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4xf32>, tensor<4xf32>, tensor<4xf32>, tensor<4xf32>, tensor<4xf32>, tensor<4xf32>, tensor<4xf32>, tensor<4x4xf32>, tensor<4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>, tensor<4x4xf32>) -> tensor<4x4x4xf32> {
// CHECK: {
// CHECK-NEXT:   version: 3,

func.func @partitioned_input_rank_mismatch(%arg0: tensor<!tf_type.resource<tensor<1x4x4xf32>>>) {
  // expected-error @+1 {{rank}}
  %0 = "tf.TPUPartitionedInputV2"(%arg0) {_XlaSharding = "\08\03\1A\05\04\01\01\01\02\22\08\00\01\02\03\04\05\06\070\01", partition_dims = [4, 1, 1, 2], is_packed = true} : (tensor<!tf_type.resource<tensor<1x4x4xf32>>>) -> tensor<!tf_type.resource<tensor<4x4x4xf32>>>

      return %0 : tensor<1x1x4xf32>
    }
  )mlir";

  const OwningOpRef<ModuleOp> module_op =
      ParseModuleOpString(kXlaCallModuleOpWithQuantizationMethodAttr);
  ASSERT_TRUE(module_op);