Function TileGrad

tensorflow/cc/gradients/array_grad.cc:582–612

REGISTER_GRADIENT_OP("BroadcastTo", BroadcastToGrad);

Status TileGrad(const Scope& scope, const Operation& op,
                const std::vector<Output>& grad_inputs,
                std::vector<Output>* grad_outputs) {
  if (op.num_inputs() != 2) {
    return errors::InvalidArgument("Tile requires 2 inputs");
  }
  if (grad_inputs.size() != 1) {
    return errors::InvalidArgument("Tile grad requires 1 grad input");
  }

  Shape::Attrs shape_attrs;
  shape_attrs.out_type_ = op.input_type(1);
  auto input_shape = Shape(scope, op.input(0), shape_attrs);
  // We interleave multiples and input_shape to get split_shape,
  // reshape grad to split_shape, and reduce along all even
  // dimensions (the tiled dimensions) to get the result
  // with shape input_shape.  For example
  //   input_shape = [20, 30, 40]
  //   multiples = [2, 3, 4]
  //   split_shape = [2, 20, 3, 30, 4, 40]
  //   axes = [0, 2, 4]
  auto stack = Stack(scope, {op.input(1), input_shape.output});
  auto perm = Range(scope, Sub(scope, Rank(scope, stack), 1), -1, -1);
  auto split_shape = Reshape(scope, Transpose(scope, stack, perm), {-1});
  auto axes = Range(scope, Const(scope, 0), Size(scope, split_shape.output), 2);
  auto input_grad = ReduceSum(
      scope, Reshape(scope, grad_inputs[0], split_shape.output), axes.output);
  grad_outputs->push_back(input_grad.output);
  grad_outputs->push_back(NoGradient());
  return scope.status();
}
REGISTER_GRADIENT_OP("Tile", TileGrad);

// Create a constant of the provided d_type;