[Bug target/18665] [3.4/4.0 Regression] -ftrapv borks up simpleinteger arithmetic

[Paul Schlie <schlie at comcast dot net>]

> ------- Additional Comments From rth at gcc dot gnu dot org  2004-12-06
> Look at the routines we implement in libgcc and ask again.  We never did
> implement sub-word routines.

Which is an interesting point, but wouldn't it be more ideal to define
default built-in's for each of a target's defined unique sized types
independently of it's word size?

(Thereby enabling operations to operate on the smallest sized operands
necessary to satisfy it's destination precision requirements without having
to otherwise unnecessarily promote smaller operations to their larger
defined counterparts if advantageous to do so?)