ASSERT_EQ(errors::OK, status_.code()) << status_.message();
    y.reset(y_raw);
  }

  // TODO(srbs): Rename ops::Add to ops::AddV2 and AddRegister to
  // AddV2Registerer.
  status_ = registry_.Register("AddV2", AddRegisterer);
  ASSERT_EQ(errors::OK, status_.code()) << status_.message();

  ASSERT_NO_FATAL_FAILURE(CompareNumericalAndAutodiffGradients(

                                           TF_Status* s);

// Create an operation suitable to use with the provided context. The operation
// requires its type (e.g. "AddV2") to be set independently.
TF_AbstractOp* TF_NewAbstractOp(TF_ExecutionContext* ctx);
void TF_DeleteAbstractOp(TF_AbstractOp*);

// TODO(srbs): Add APIs for specifying attrs etc.
// `op_type` must outlive `op`.

#include "tensorflow/c/experimental/ops/array_ops.h"
#include "tensorflow/c/experimental/ops/math_ops.h"
#include "tensorflow/c/experimental/ops/nn_ops.h"

using std::vector;
using tensorflow::ops::AddV2;
using tensorflow::ops::Div;
using tensorflow::ops::DivNoNan;
using tensorflow::ops::MatMul;
using tensorflow::ops::Mul;
using tensorflow::ops::Neg;
using tensorflow::ops::OnesLike;
using tensorflow::ops::SqrtGrad;

  TF_DeleteStatus(status);

  return var_handle;
}

TFE_Op* AddOp(TFE_Context* ctx, TFE_TensorHandle* a, TFE_TensorHandle* b) {
  TF_Status* status = TF_NewStatus();

  TFE_Op* op = TFE_NewOp(ctx, "AddV2", status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_OpAddInput(op, a, status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_OpAddInput(op, b, status);

*   Enable JIT-compiled i64-indexed kernels on GPU for large tensors with more than 2**32 elements.
    *   Unary GPU kernels: Abs, Atanh, Acos, Acosh, Asin, Asinh, Atan, Cos, Cosh, Sin, Sinh, Tan, Tanh.
    *   Binary GPU kernels: AddV2, Sub, Div, DivNoNan, Mul, MulNoNan, FloorDiv, Equal, NotEqual, Greater, GreaterEqual, LessEqual, Less.

* `tf.lite`
    * Add experimental supports conversion of models that may be larger than 2GB before buffer deduplication