🚀 go-pugleaf

In article <1992Dec30.152808.25311@netcom.com> tm@netcom.com (Toshiyasu Morita) writes:
>I'm looking for a fast min() and max() algorithm, preferably without branching.
>
>Anyone know of any?
>

I think this is the standard r1=min(r1,r2) trick, assuming two registers
holding unsigned quantities on an 80x86:

    sub   r2,r1   ; set carry if r1>r2, leave r2-r1 in r2 
    sbb   rs,rs   ; scratch register = 0ffffh if we want r2 
    and   rs,r2   ; zero difference if we want r1
    add   r1,rs   ; replace r1 value with r2 value if r1>r2

This trashes r2, and requires a scratch register, but it screams
if you have a >386 which bites when you branch.  There is an
analog of this for max.