TEST(TFCompileTest, Add) {
  AddComp add;
  EXPECT_EQ(add.arg0_data(), add.arg_data(0));
  EXPECT_EQ(add.arg1_data(), add.arg_data(1));

  add.arg0() = 1;
  add.arg1() = 2;
  EXPECT_TRUE(add.Run());
  EXPECT_EQ(add.error_msg(), "");
  EXPECT_EQ(add.result0(), 3);
  EXPECT_EQ(add.result0_data()[0], 3);
  EXPECT_EQ(add.result0_data(), add.results()[0]);

  add.arg0_data()[0] = 123;
  add.arg1_data()[0] = 456;
  EXPECT_TRUE(add.Run());

  void set_arg0_data(const void* data) {
    set_arg_data(0, data);
  }
  float* arg0_data() {
    return static_cast<float*>(arg_data(0));
  }
  float& arg0(size_t dim0, size_t dim1) {
    return (*static_cast<float(*)[1][2]>(
        arg_data(0)))[dim0][dim1];
  }
  const float* arg0_data() const {
    return static_cast<const float*>(arg_data(0));
  }
  const float& arg0(size_t dim0, size_t dim1) const {

  }
  {{TYPE}}* arg{{NAME}}_data() {
    return static_cast<{{TYPE}}*>(arg_data({{I}}));
  }
  {{TYPE}}& arg{{NAME}}({{DIM_VARS}}) {
    return (*static_cast<{{TYPE}}(*){{DIM_SIZES}}>(
        arg_data({{I}}))){{INDICES}};
  }
  const {{TYPE}}* arg{{NAME}}_data() const {
    return static_cast<const {{TYPE}}*>(arg_data({{I}}));
  }
  const {{TYPE}}& arg{{NAME}}({{DIM_VARS}}) const {

// clang-format on

namespace tensorflow {
namespace tfcompile {
namespace {

void zero_buffers(XlaCompiledCpuFunction* computation) {
  for (int i = 0; i < computation->num_args(); ++i) {
    memset(computation->arg_data(i), 0, computation->arg_size(i));
  }
}

// Trivial test that runs the generated function to ensure it doesn't crash.
TEST(TEST_NAME, NoCrash) {
  Eigen::ThreadPool pool(port::MaxParallelism());

// clang-format on

namespace tensorflow {
namespace tfcompile {
namespace {

void zero_buffers(XlaCompiledCpuFunction* computation) {
  for (int i = 0; i < computation->num_args(); ++i) {
    memset(computation->arg_data(i), 0, computation->arg_size(i));
  }
}

// Trivial test that runs the generated function to ensure it doesn't crash.
TEST(TEST_NAME, NoCrash) {
  Eigen::ThreadPool pool(port::MaxParallelism());

  };

  StreamExecutor* executor = GetExecutor(0);
  TF_ASSERT_OK_AND_ASSIGN(auto stream, executor->CreateStream());
  size_t size = sizeof(int);
  int src_data = 34;
  int dst_data = 2;
  DeviceMemoryBase device_src(&src_data, size);
  TF_ASSERT_OK(stream->Memcpy(&dst_data, device_src, size));
  ASSERT_EQ(dst_data, 34);
}

TEST_F(StreamExecutorTest, MemcpyFromHost) {

// CHECK-LABEL: @tf__my_requantize
tfr.func @tf__my_requantize(%input: !tfr.tensor) -> !tfr.tensor {
  %raw_data = tfr.quant_raw_data(%input) : (!tfr.tensor) -> !tfr.tensor
  %scale, %zp = tfr.quant_qparam(%input) : (!tfr.tensor) -> (!tfr.tensor, !tfr.tensor)
  %result = tfr.call @tf__requantize(%raw_data, %scale, %zp) : (!tfr.tensor, !tfr.tensor, !tfr.tensor) -> !tfr.tensor
  tfr.return %result : !tfr.tensor
}

 */
class HpackRoundTripTest : HpackDecodeTestBase() {
  internal class StoriesTestProvider : SimpleProvider() {
    override fun arguments(): List<Any> = createStories(RAW_DATA)
  }

  private val bytesOut = Buffer()
  private val hpackWriter = Hpack.Writer(out = bytesOut)

  @ParameterizedTest
  @ArgumentsSource(StoriesTestProvider::class)
  fun testRoundTrip(story: Story) {

Eigen::half APFloatToEigenHalf(const APFloat& val) {
  uint16_t raw_data = val.bitcastToAPInt().getZExtValue();
  return Eigen::numext::bit_cast<Eigen::half>(raw_data);
}

APFloat EigenHalfToAPFloat(const Eigen::half& val) {
  uint16_t raw_data = Eigen::numext::bit_cast<uint16_t>(val);
  return APFloat(APFloat::IEEEhalf(), APInt(16, raw_data));
}

void PopulateEncodingParams(const std::vector<int>& block_size,

// CHECK:  %[[RSQRT:.*]] = "tf.Rsqrt"(%[[VARIANCE]])
// CHECK:  %[[MUL1:.*]] = "tf.Mul"(%[[ARG_T:.*]], %[[RSQRT]])
// CHECK:  %[[MUL2:.*]] = "tf.Mul"(%[[ARG_M:.*]], %[[RSQRT]])
// CHECK:  %[[SUB:.*]] = "tf.Sub"(%[[ARG_BETA:.*]], %[[MUL2]])
// CHECK:  %[[RESULT:.*]] = "tf.Add"(%[[MUL1]], %[[SUB]])
// CHECK:  return %[[RESULT]]
}

func.func @batchNormWithGlobalNormalizationWithScaleAfterNormalization(