auto ReluGradModel = BuildGradModel(ReluModel, registry_);

  float X_vals[] = {1.0f, 2.0f, 3.0f, -5.0f, -4.0f, -3.0f, 2.0f, 10.0f, -1.0f};
  int64_t X_dims[] = {3, 3};
  AbstractTensorHandlePtr X;
  {
    AbstractTensorHandle* X_raw;
    status_ = TestTensorHandleWithDims<float, TF_FLOAT>(
        immediate_execution_ctx_.get(), X_vals, X_dims, 2, &X_raw);

TEST(CAPI, TestFromProto) {
  Tensor t_cc(DT_FLOAT, TensorShape({2, 3}));
  t_cc.flat<float>().setConstant(1.0);
  tensorflow::TensorProto t_proto;
  t_cc.AsProtoField(&t_proto);

  TF_Buffer* t_buffer = TF_NewBuffer();
  TF_CHECK_OK(MessageToBuffer(t_proto, t_buffer));

  const int num_bytes = 6 * sizeof(float);
  float* values =
      reinterpret_cast<float*>(tensorflow::cpu_allocator()->AllocateRaw(

    status_ = TestScalarTensorHandle<float, TF_FLOAT>(
        immediate_execution_ctx_.get(), 1.0f, &x1_raw);
    ASSERT_EQ(errors::OK, status_.code()) << status_.message();
    x1.reset(x1_raw);
  }

  AbstractTensorHandlePtr x2;
  {
    AbstractTensorHandle* x2_raw = nullptr;
    status_ = TestScalarTensorHandle<float, TF_FLOAT>(
        immediate_execution_ctx_.get(), 1.0f, &x2_raw);

    ctx.reset(ctx_raw);
  }

  AbstractTensorHandlePtr x;
  {
    AbstractTensorHandle* x_raw = nullptr;
    float data[] = {0., 0., 0., 0., 0., 0., 0., 0};
    int64_t dim_sizes[] = {2, 4};
    Status s = TestTensorHandleWithDims<float, TF_FLOAT>(ctx.get(), data,
                                                         dim_sizes, 2, &x_raw);
    ASSERT_EQ(errors::OK, s.code()) << s.message();

#include "tensorflow/core/util/port.h"

using tensorflow::string;
using tensorflow::tstring;

TFE_TensorHandle* TestScalarTensorHandle(TFE_Context* ctx, float value) {
  float data[] = {value};
  TF_Status* status = TF_NewStatus();
  TF_Tensor* t = TFE_AllocateHostTensor(ctx, TF_FLOAT, nullptr, 0, status);
  memcpy(TF_TensorData(t), &data[0], TF_TensorByteSize(t));
  TFE_TensorHandle* th = TFE_NewTensorHandleFromTensor(ctx, t, status);

    ASSERT_EQ(num_elem_numerical, num_elem_analytical);

    float* dnumerical = new float[num_elem_numerical]{0};
    memcpy(&dnumerical[0], TF_TensorData(numerical_tensor),
           TF_TensorByteSize(numerical_tensor));
    float* danalytical = new float[num_elem_analytical]{0};
    memcpy(&danalytical[0], TF_TensorData(analytical_tensor),
           TF_TensorByteSize(analytical_tensor));

          return absl::OkStatus();
        case 32:
          *tf_dtype = TF_DataType::TF_FLOAT;
          return absl::OkStatus();
        case 64:
          *tf_dtype = TF_DataType::TF_DOUBLE;
          return absl::OkStatus();
        default:
          return tensorflow::errors::InvalidArgument("Unsupported Float bits: ",
                                                     dtype.bits);
      }
      break;

    ASSERT_EQ(errors::OK, s.code()) << s.message();
    ctx.reset(ctx_raw);
  }

  AbstractTensorHandlePtr x;
  {
    AbstractTensorHandle* x_raw = nullptr;
    Status s = TestScalarTensorHandle<float, TF_FLOAT>(ctx.get(), 1.0f, &x_raw);
    ASSERT_EQ(errors::OK, s.code()) << s.message();
    x.reset(x_raw);
  }

  // Pseudo-code:
  //
  // tape.watch(x)
  // y = exp(x)
  // outputs = tape.gradient(y, x)

}
Status SetAttrFloatList(AbstractOperation* op_, const char* attr_name,
                        const float* values, int num_values,
                        ForwardOperation* forward_op_) {
  forward_op_->attrs.Set(attr_name,
                         gtl::ArraySlice<const float>(values, num_values));
  return op_->SetAttrFloatList(attr_name, values, num_values);
}

        "//tensorflow/c/experimental/ops:nn_ops",
        "//tensorflow/core:test",
        "//tensorflow/core:test_main",
        "//tensorflow/core/platform:tensor_float_32_utils",
    ] + if_libtpu(
        if_false = ["//tensorflow/compiler/mlir/tensorflow/c:mlir_c_api_registration"],
        if_true = [],
    ),
)

tf_cuda_cc_test(
    name = "math_grad_test",