[RC5] Clocks per Key on Pentium vs PPC604

RÈmi Guyomarch rguyom at mail.dotcom.fr
Wed Jan 7 02:00:59 EST 1998


Mary Conner wrote:
> 
> I'm not a PPC expert, but I do know that the rotl instruction on a PPC
> is always 1 cycle long, whereas the rotl instruction on x86's is 2
> cycles when rotating by the cl register, 1 cycle otherwise.  A
> substantial number of the rotates are by register, so there's some
> of your cycle savings right there.

rotl cycle count vary from one x86 implementation to another.
The Pentium spend 1 cycle executing rotl with an immediate count, and 4
cycles executing rotl with a register as the rotation count. Even worst,
this instruction use both pipelines and cannot pair with anything else.
The situtation is better on the 6x86, where rotl is always pairable and
lasts 1 or 2 cycles.
On the K5, rotl execute in 1 cycle, whatever rotation count is
(immediate or register) but cannot pair with another rotl or shift
operation.
On the P6, it seems to execute in 1 cycle and pair with anything.

> I know there is an rc5-coders list for those doing overall coding
> on clients, but would it be possible to set up a publically available
> list for those of us hacking on the publically released cores?  This
> stuff is probably pretty dry and boring for the majority of the
> rc5 mailing list.

This thread is by far the most on-topic thing I've seen on this list for
the last few days...

-- 
RÈmi		Don't waste your computer's time. Distribute it!
			http://www.distributed.net/
	    RC5 cores source code : http://wwwperso.hol.fr/~guyom001/


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



More information about the rc5 mailing list