TFE_DeleteExecutor(executor);
  TFE_ContextRemoveFunction(ctx, "AddVariablesFunction", status);
  TFE_DeleteContext(ctx);

  TF_DeleteStatus(status);

  // TODO(b/136478427): Figure out how to correctly shut the server down.
  worker_server1.release();
  worker_server2.release();
}

TEST(CAPI, TestLocalFunctionWithPackedInput) {
  TestFunctionWithPackedInput(/*remote=*/false);
}

index 185402f..bbdeae0 100644
--- a/sysdeps/aarch64/dl-machine.h
+++ b/sysdeps/aarch64/dl-machine.h
@@ -49,23 +49,11 @@ elf_machine_load_address (void)
   /* To figure out the load address we use the definition that for any symbol:
      dynamic_addr(symbol) = static_addr(symbol) + load_addr
 
-     The choice of symbol is arbitrary. The static address we obtain

  TFE_Executor* executor = TFE_ContextGetExecutorForThread(ctx);
  TFE_ExecutorWaitForAllPendingNodes(executor, status);
  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  // TODO(b/136478427): Figure out how to correctly shut the server down.
  worker_server.release();

  // Update the server def with a new set of names (worker instead of
  // localhost).

#ifdef __cplusplus
} /* end extern "C" */
#endif

#ifdef __cplusplus
// A workaround to ease conversion to and from numpy objects and
// TFE_TensorHandle's.
//
// TODO(ashankar): Figure out an alternative scheme that precludes the need for
// these API-boundary breaking methods.
namespace tensorflow {
class Tensor;
}  // namespace tensorflow

  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_DeleteExecutor(executor);
  TFE_DeleteContext(ctx);

  TF_DeleteStatus(status);

  // TODO(b/136478427): Figure out how to correctly shut the server down.
  worker_server.release();
}

TEST(CAPI, RemoteExecute) { TestRemoteExecute(false); }
TEST(CAPI, RemoteExecuteAsync) { TestRemoteExecute(true); }

        !DeviceNameUtils::IsSameAddressSpace(
            underlying_devices_[component_index], underlying_devices_[0])) {
      // Device names are from different address spaces, or we can't figure out
      // whether they are, so we'll fully-qualify everything.
      return underlying_devices_;
    }
  }
  std::vector<std::string> local_names;
  local_names.reserve(underlying_devices_.size());

  Status Lookup(const ForwardOperation& op,
                std::unique_ptr<GradientFunction>* gradient_function) const;

 private:
  absl::flat_hash_map<string, GradientFunctionFactory> registry_;
};

// TODO(srbs): Figure out if we can avoid declaring this in the public header.
// Wrapper for a tensor output of an operation executing under a tape.
//
// `GetID` returns a unique id for the wrapped tensor which is used to maintain

  ASSERT_EQ(errors::OK, status_.code()) << status_.message();

  ASSERT_NO_FATAL_FAILURE(CompareNumericalAndAutodiffGradients(
      ExpModel, BuildGradModel(ExpModel, registry_),
      immediate_execution_ctx_.get(), {x.get()}, UseFunction()));
}

TEST_P(CppGradients, TestMatMulGrad) {
  // TODO(vnvo2409): Figure out why `gradient_checker` does not work very
  // well with `MatMul` and remove `TestMatMul*` in

        immediate_execution_ctx_.get(), X_vals, X_dims, 2, &X_raw);
    ASSERT_EQ(errors::OK, status_.code()) << status_.message();
    X.reset(X_raw);
  }

  ASSERT_NO_FATAL_FAILURE(CompareNumericalAndAutodiffGradients(
      ReluModel, ReluGradModel, immediate_execution_ctx_.get(), {X.get()},
      UseFunction()));

  // Mathematically, Relu isn't differentiable at `0`. So `gradient_checker`

    ::tensorflow::Status _s(__VA_ARGS__);                         \
    if (!TF_PREDICT_TRUE(_s.ok())) {                              \
      ::tensorflow::Set_TF_Status_from_Status(C_STATUS, _s);      \
      TF_OpKernelContext_Failure(CTX, C_STATUS);                  \
      TF_DeleteStatus(C_STATUS);                                  \
      return;                                                     \