386 optimized bitblit code

[Kevin Ryde]

Brian Hurt <bhurt at spnz.org> writes:
>
> Attached is a first cut at 386-optimized code.

Unless I've missed something, what you're doing looks like either
mpn_lshift or mpn_rshift, according to the overlap direction.  See
that code for what it takes to go fast on x86s.  (In particular mmx is
usually what you want for the mmx chips, shrdl is sub-optimal on say
athlon, and very slow on pentium4.)