I believe that literals could be sorted by alignment requirement instead of
length, this way minimizing the needed padding bytes. That is, multiples of 8
first, then multiples of 4, then multiples of 2, then odd lenghts, that need to
be aligned on even addresses because of relative addressing, then "true" odds,
that dont have this special requirement.
> It does seem like it could be useful if literals inherited the
> worst-case alignment requirements imposed by the instruction operand(s)
> that reference them. But, doing so could introduce some new issues.
> Literals are currently sorted by length. Aligning short literals on a
> halfword, fullword, doubleword or quadword boundary might require the
> assembler to insert a lot of "filler" bytes in the literal pool. Would
> that be acceptable?