ASSERT_TRUE(out != nullptr);
      EXPECT_EQ(TF_INT32, TF_TensorType(out));
      EXPECT_EQ(0, TF_NumDims(out));  // scalar
      ASSERT_EQ(sizeof(int32_t), TF_TensorByteSize(out));
      int32_t* output_contents = static_cast<int32_t*>(TF_TensorData(out));
      EXPECT_EQ(expected_results[i], *output_contents);
    }
  }

  void CompareInt32Tensor(const std::vector<int32_t>& expected, TF_Tensor* t) {

bool GetGraphDef(TF_Graph* graph, tensorflow::GraphDef* graph_def) {
  TF_Status* s = TF_NewStatus();
  TF_Buffer* buffer = TF_NewBuffer();
  TF_GraphToGraphDef(graph, buffer, s);
  bool ret = TF_GetCode(s) == TF_OK;
  EXPECT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
  if (ret) ret = graph_def->ParseFromArray(buffer->data, buffer->length);
  TF_DeleteBuffer(buffer);
  TF_DeleteStatus(s);
  return ret;
}

    const auto tensor_interface =
        tensorflow::down_cast<const tensorflow::TensorInterface*>(src->tensor);

    if (dst->dims() != 0) {
      return InvalidArgument(
          "Malformed TF_RESOURCE tensor: expected a scalar, got a tensor with "
          "shape ",
          dst->shape().DebugString());
    }
    *dst = tensorflow::Tensor(tensorflow::DT_RESOURCE, dst->shape());

  EXPECT_FALSE(executed);
  ASSERT_FALSE(TF_GetCode(status.get()) == TF_OK);

  // Custom devices can refuse to do type-based dispatch (as hcustom1 is
  // configured to do)
  matmul.reset(MatMulOp(context.get(), hcustom1.get(), hcpu.get()));
  num_retvals = 1;
  executed = false;
  TFE_Execute(matmul.get(), &retval, &num_retvals, status.get());
  EXPECT_FALSE(executed);

  bool valid = true;
  int64_t expected_stride = 1;
  for (int i = ndim - 1; i >= 0; --i) {
    // Empty tensors are always compact regardless of strides.
    if (shape_arr[i] == 0) return true;
    // Note that dimensions with size=1 can have any stride.
    if (shape_arr[i] != 1 && stride_arr[i] != expected_stride) {
      valid = false;
    }
    expected_stride *= shape_arr[i];
  }
  return valid;
}

  status_ =
      RunModel(IdentityNGradModel, immediate_execution_ctx_.get(),
               {x1.get(), x2.get()}, absl::MakeSpan(outputs), UseFunction());
  ASSERT_EQ(errors::OK, status_.code()) << status_.message();
  EXPECT_EQ(outputs[0], nullptr);
  ASSERT_NO_FATAL_FAILURE(CheckTensorValue(outputs[1], {1.0f}, /*dims*/ {},
                                           /*abs_error*/ 0));
  outputs[1]->Unref();
}

#ifdef PLATFORM_GOOGLE

If the conversion is successful, but the generated model is wrong, then state what is wrong:

- Model produces wrong results and/or has lesser accuracy.
- Model produces correct results, but it is slower than expected.

### 4. (optional) RNN conversion support
If converting TF RNN to TFLite fused RNN ops, please prefix [RNN] in the title.

### 5. (optional) Any other info / logs

    TF_DeleteDir(dirpath.c_str(), s);
    ASSERT_TF_OK(s) << "TF_DeleteDir failed for " << dirpath << ": "
                    << TF_Message(s);

    TF_DeleteStatus(s);
    break;
  }

  ASSERT_TRUE(found) << "expected at least one temp dir";

  TF_StringStreamDone(tempdirs);
}

TEST(TestEnv, TestTimeFunctions) {
  ASSERT_GE(TF_NowSeconds(), 946684800);  // Midnight Jan 1, 2000
  ASSERT_GE(TF_NowMicros(), 946684800 * 1e6);

    `TFCI` env param pointing to a list of `env` files.
2.  `utilities/setup.sh`, initialized by all top-level scripts, reads and sets
    values from those `TFCI` paths.
    -   `set -a` / `set -o allexport` exports the variables from `env` files so
        all scripts can use them.
    -   `utilities/setup_docker.sh` creates a container called `tf` with all
        `TFCI_` variables shared to it.

// the device is not usable. In case of a bad status, `device.delete_device` is
// still called on `device_info` (i.e. the caller does not retain ownership).
//
// 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,