I wonder if this is a result of the Q32 macro not containing the "u" suffix and defaulting to a signed int.

Of course the result of the calculation is the same, but that old level of gcc may prefer subtracting when adding negative constants.