🚀 go-pugleaf

In article <7343@brl-smoke.ARPA> gwyn@brl.arpa (Doug Gwyn (VLD/VMB) <gwyn>) writes:
>In article <770@entropy.ms.washington.edu> martin@entropy.ms.washington.edu (Don Martin) writes:
>>A book that is highly recomded in spite of the poor 
>>implemetation of DES ( 1 bit per integer ).
>
>Why do you consider that a poor implementation decision?
>It's the way I would have done it.  Packed bits would be
>too slow.

It's more fun to keep them in 48-bit words (or 32 if that's all you've
got) so that you can do operations on them a word at a time.  For the
permutations, for example, you can precompute the result of the permutation
applied to each byte in each position, and OR them together.  The XOR
operations happen immediately, of course.  Hauling 6- or 12-bit chunks
out to stuff through S-boxes (or doubled S-boxes if you have the luxury
of lots of memory) is simplified if you're using the 48-bit size, but
can be handled with a few shifts and logical operations without much
trouble even with 32-bit words.

Somebody (Dan Hoey?) posted some useful hacks a few months ago, including
an absolutely wonderful way to make the initial and final permutations
almost painless using a modification of a HAKMEM algorithm by Rick
Gumpertz and himself.  These used very fancy shifting and logical ops
to do the IP and FP in much less time than the table-lookup stuff that
I described briefly above (which still would need to be used for the
combined S-P operation).

	Jim Gillogly
-- 
	Jim Gillogly
	{hplabs, ihnp4}!sdcrdcf!randvax!jim
	jim@rand-unix.arpa    [HASA: U (Spam) division]