[RC5] finishing up an RC5 core using MMX

Jason Stratos Papadopoulos jasonp at Glue.umd.edu
Sun Jul 26 17:29:59 EDT 1998


Hello. <Deep breath> On the off chance that an RC5 core using MMX is 
still an open problem, I'm almost finished writing one, and would be 
happy to make it available upon completion, along with documentation.

The finished version will require approximately 470 cycles to test a 
single key, for a net keyrate of about 430 kkeys/sec on a 200MHz Pentium
MMX. This comes out to about 1.57 times the average rate on the RC5
benchmark page for the same machine.

The code as written tests four keys at a time, and uses MMX instructions
almost entirely. It's very complicated, and pairs almost perfectly 
(more than 399 mips on a 200MHz machine). I'm more than a little proud
of it.

A few questions though:
   Can an MMX core run up the keyrate on the PII and 6x86? These
   machines already seem to test far more keys than the MHz suggest,
   meaning an MMX core would only speed up the Pentium.

   How much of a speedup can distribute.net expect from this? I ask
   mainly to inflate my ego :)

   Finally, are there specific calling conventions required? Remi
   Guyomarch's driver programs aren't clear on the matter. As a related
   issue, what should I do to make the final product public without
   violating ITAR?

Lemme know what to do...

jasonp
--
To unsubscribe, send 'unsubscribe rc5' to majordomo at lists.distributed.net
rc5-digest subscribers replace rc5 with rc5-digest



More information about the rc5 mailing list