Multiply(context.get(), second_combined_value.get(),
               second_negative_one.get(), status.get()));
  ASSERT_EQ(TF_GetCode(status.get()), TF_OK) << TF_Message(status.get());

  // Un-pack the parallel tensor to verify that the operation was
  // successful. The resulting structure should be:
  //   second_device{first_device{1. * 3., 2. * 3.}, 3. * 3.}.
  std::array<TensorHandlePtr, 2> second_components;

1 !important;\n  flex-shrink: 1 !important;\n}\n\n.justify-content-start {\n  -ms-flex-pack: start !important;\n  justify-content: flex-start !important;\n}\n\n.justify-content-end {\n  -ms-flex-pack: end !important;\n  justify-content: flex-end !important;\n}\n\n.justify-content-center {\n  -ms-flex-pack: center !important;\n  justify-content: center !important;\n}\n\n.justify-content-between {\n  -ms-flex-pack: justify !important;\n  justify-content: space-between !important;\n}\n\n.justify-content-around...

1 !important;\n  flex-shrink: 1 !important;\n}\n\n.justify-content-start {\n  -ms-flex-pack: start !important;\n  justify-content: flex-start !important;\n}\n\n.justify-content-end {\n  -ms-flex-pack: end !important;\n  justify-content: flex-end !important;\n}\n\n.justify-content-center {\n  -ms-flex-pack: center !important;\n  justify-content: center !important;\n}\n\n.justify-content-between {\n  -ms-flex-pack: justify !important;\n  justify-content: space-between !important;\n}\n\n.justify-content-around...

      // One of the inputs we're trying to pack is on a custom device. We'll let
      // the first custom device we see handle all of the packing.
      auto* custom_device_handle =
          tensorflow::down_cast<tensorflow::CustomDeviceTensorHandle*>(
              unwrapped_handle);
      tensorflow::ImmediateExecutionTensorHandle* result;
      status->status = custom_device_handle->device()->Pack(

                   &FloatDeallocator, nullptr),
      TF_DeleteTensor);
  return TensorHandlePtr(TFE_NewTensorHandle(tensor.get(), status));
}

// Helper to un-pack `num_replicas` TFE_TensorHandles from one parallel handle.
template <std::size_t num_replicas>
void ExtractPerDeviceValues(
    TFE_Context* context, TFE_TensorHandle* input,

    components.reserve(underlying_devices_.size());
    for (int j = 0; j < underlying_devices_.size(); ++j) {
      components.push_back(std::move(per_device_output_tensors[j][i]));
    }
    if (expected_output_shapes[i].IsFullyDefined()) {
      per_device_outputs.push_back(ParallelTensor::FromTensorHandles(
          *this, std::move(components),
          absl::Span<const int64_t>(expected_output_shapes[i].dim_sizes()),

  }

  // We use power-of-2 tables, and this is the highest int that's a power of 2
  static final int MAX_TABLE_SIZE = Ints.MAX_POWER_OF_TWO;

  // Represents how tightly we can pack things, as a maximum.
  private static final double DESIRED_LOAD_FACTOR = 0.7;

  // If the set has this many elements, it will "max out" the table size

  // custom device.
  //
  // Many devices will want to simply return an "unimplemented" status
  // here. This is the default behavior if `pack` is null when passed to
  // TFE_RegisterCustomDevice.
  TFE_TensorHandle* (*pack)(TFE_Context* context, TFE_TensorHandle** handles,
                            int num_handles, TF_Status* s,
                            void* device_info) = nullptr;

container .toast-warning {\n  background-color: #ffc107;\n}\n\n.pace {\n  z-index: 1048;\n}\n\n.pace .pace-progress {\n  z-index: 1049;\n}\n\n.pace .pace-activity {\n  z-index: 1050;\n}\n\n.pace-primary .pace .pace-progress {\n  background: #007bff;\n}\n\n.pace-barber-shop-primary .pace {\n  background: #ffffff;\n}\n\n.pace-barber-shop-primary .pace .pace-progress {\n  background: #007bff;\n}\n\n.pace-barber-shop-primary .pace .pace-activity {\n  background-image: linear-gradient(45deg, rgba(255,...

  // Add a sync point to make sure that variables have been initialized
  // before the function execution starts.
  TFE_ContextAsyncWait(ctx, status);
  EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  // Pack 3 variable handles into one TFE_TensorHandle.
  // When remote is false, function device is placed on task0. Handle types are
  // REMOTE, REMOTE, LOCAL on task0. When remote is true, function device is