tensorflow::TensorHandleFromInterface(tensorflow::unwrap(h));
  if (handle->Type() != TensorHandle::LOCAL) {
    status->status = tensorflow::errors::InvalidArgument(
        "DLPack doesn't support ", handle->TypeString(), " tensor");
    return nullptr;
  }
  const tensorflow::Tensor* tensor;
  status->status = handle->Tensor(&tensor);
  if (!status->status.ok()) {
    return nullptr;
  }
  return tensor;

  for (int i = 0; i < *num_retvals; i++) {
    // TODO(srbs): Manage refcount of ForwardOperation's inputs/outputs.
    forward_op_->outputs.push_back(retvals[i]);
  }
  // TODO(b/166669239): This is needed to support AttrBuilder::Get for string
  // attributes. Number type attrs and DataType attrs work fine without this.
  // Consider getting rid of this and making the behavior between number types
  // and string consistent.

 public:
  Status Compute(AbstractContext* ctx,
                 absl::Span<AbstractTensorHandle* const> grad_outputs,
                 absl::Span<AbstractTensorHandle*> grad_inputs) override {
    // TODO(b/161805092): Support broadcasting.

    DCHECK(grad_outputs[0]);
    grad_inputs[0] = grad_outputs[0];
    grad_inputs[1] = grad_outputs[0];

    grad_inputs[0]->Ref();
    grad_inputs[1]->Ref();
    return absl::OkStatus();
  }

//
// This API is highly experimental, and in particular is expected to change when
// it starts supporting operations with attributes and when tf.function support
// is added.
TF_CAPI_EXPORT extern void TFE_RegisterCustomDevice(TFE_Context* ctx,
                                                    TFE_CustomDevice device,

        "//tensorflow/core/platform:tstring",
        "@com_google_absl//absl/status",
        "@local_xla//xla/tsl/c:tsl_status",
    ] + select({
        "//tensorflow:with_xla_support": [
            "//tensorflow/compiler/jit",
            "//tensorflow/compiler/tf2xla:xla_compiler",
        ],
        "//conditions:default": [],
    }) + if_tensorrt([

                                          int num_handles, TF_Status* status,
                                          void* device_info) {
  TF_SetStatus(status, TF_UNIMPLEMENTED,
               "This custom device does not support packing tensors.");
  return nullptr;
}
}  // namespace

extern "C" {

bool TFE_IsCustomDevice(TFE_Context* ctx, const char* device_name) {

#include "tensorflow/c/tf_file_statistics.h"
#include "tensorflow/c/tf_status.h"

// Implementation of a filesystem for POSIX environments.
// This filesystem will support `file://` and empty (local) URI schemes.

static void* plugin_memory_allocate(size_t size) { return calloc(1, size); }
static void plugin_memory_free(void* ptr) { free(ptr); }

            "//tensorflow/core/lib/llvm_rtti",
            "@com_google_absl//absl/types:variant",
        ],
    }) + select({
        "//tensorflow:with_xla_support": [
            "//tensorflow/compiler/jit",
            "//tensorflow/compiler/jit:xla_device",
            "//tensorflow/compiler/tf2xla:xla_compiler",
        ],
        "//conditions:default": [],
    }) + [

    for the required packages. Alternatively, use the said
    [tensorflow/build Docker images](https://hub.docker.com/r/tensorflow/build)
    (`tensorflow/tensorflow:devel` and `tensorflow/tensorflow:devel-gpu` are no
    longer supported for) development. Use TF SIG Build Dockerfiles in
    development to avoid installing the packages directly on your system (in
    which case remember to change the directory from `/root` to `/tensorflow`

    return {};
  }
  const auto& fdef_lib = gdef.library();
  if (fdef_lib.gradient_size() > 0) {
    status->status = tensorflow::errors::Internal(
        "GradientDef is not supported in reading Dataset related functions: ",
        text_proto);
    return {};
  }
  std::vector<UniqueFuncPtr> ret;
  for (const FunctionDef& fdef : fdef_lib.function()) {