Re: AES in LibTomCrypt

From: Tom St Denis (tomstdenis_at_iahu.ca)
Date: 06/12/03 

Date: Thu, 12 Jun 2003 19:42:45 GMT

John E. Hadstate wrote:
> "Tom St Denis" <tomstdenis@iahu.ca> wrote in message
> news:Hj0Ga.65019$G_.5528@news02.bloor.is.net.cable.rogers.com...
>
>>And on x86 platforms it doesn't matter that much. For example, my LTC
>>timing [new code] does 2^21 loops, AES was clocked at 453 cycles [Athlon
>>XP, niced] or for a total of about 950,009,856 cycles. A RDTSC takes 11
>>cycles on the Athlon [and 86 on the P4 iirc...] which means it
>>contributes a huge whopping sum of zero to the overall cycle count since
>>it is performed once at the end.
>>
>
>
> Tom,
>
> When you say "AES was clocked at 453 cycles...", did you estimate that from
> timing, or did you use some software to process the object code and count
> the cycles?
>
> If you estimated from timing, exactly what did you do to get that figure?

My high-res timer [will be part of LTC v0.86] basically does this

     y1 = 0;
     t_start();
     while (y1 < (1<<22)) {
       DO256;
       y1 += 256;
     }
     t1 = t_read();

Where t1/y1 is the cycle time per call [I recently changed it to 2^22 to
get slightly better resolution].

The fact that the encrypt takes the majority of time the skew from
overhead goes to nill after 2^22 iterations.

So when I say AES takes 453 cycles I mean that the cycle count was
roughly 453 * 2^22 for the inner loop.

Tom

Relevant Pages

  • Two Points
    ... and Eigen do not involve any essential role for temperature cycles. ... In fact, Tom, I am not aware of any published proposals for the origin ... that do involve a diurnal temperature cycle. ... Stability over function. ...
    (sci.bio.evolution)
  • Re: Connecting Intel 386EX to external device
    ... embedded systems. ... > The rd# of processor is connected to rd# of external device. ... There are many possible reasons for timing ... The first silicon version had very nasty bus cycle timing problems. ...
    (comp.programming)
  • Re: Two Points
    ... But both you and Tom should ... > be aware that driving a chemical cycle by an environmental cycle is ... Antoine Muller's proposal involves a temperature cycle, ... Your idea is still strange if you think that an environment ...
    (sci.bio.evolution)
  • Re: Ignition modules
    ... A bit more experimentation, this time with Wico A body, electric drill and ... timing strobe:- ... The device only fires on one half cycle just after the ...
    (uk.rec.engines.stationary)
  • Re: AAA, etc.
    ... timing those two routines, it should certainly run for both. ... Problem is that you can reproduce the circumstances for the code I ran, but I cannot reproduce the circumstances in which you timed the code, inside of the next 20-30 years:D ... Cycle counting is rarly interessting. ... It was _much_ slower than expanding other kinds of virtual memory.. ...
    (alt.lang.asm)