Dodgy arithmetic in fixed1616ToReal()

Whilst examining the XCB code, I noticed the following:

static inline qreal fixed1616ToReal(FP1616 val)
{
    return (qreal(val >> 16)) + (val & 0xFFFF) / (qreal)0xFFFF;
}

Whilst this is an improvement over an earlier version that used 0xFF, I think this is wrong unless FP1616 has extremely unconventional semantics (such that e.g. 1.FFFF == 2.0000). I think a more plausible definition is simply

static inline qreal fixed1616ToReal(FP1616 val)
{
    return qreal(val) / 0x1000;
}