It turns out our GCC has quite an interesting bug:

      typeof(1.0fi) != float _Complex
      typeof((float _Complex)1.0fi) != float _Complex
      typeof((float _Complex)1.0i) == float _Complex

In other words: if casting to an equal size, GCC seems to take a
shortcut. By casting down from a double to a float, GCC doesn't take
this shortcut, yielding the proper type.