[RC5] Re: A whole bunch of things

gindrup at okway.okstate.edu gindrup at okway.okstate.edu
Fri Nov 21 15:27:01 EST 1997


     PPCs have parallel execution in the FP as well as the integer unit, so 
     both pipelines could be used to crack keys.  Some care would have to 
     be taken to minimize the number of stalls.
            -- Eric Gindrup ! gindrup at Okway.okstate.edu


______________________________ Reply Separator _________________________________
Subject: RE: [RC5] Re: A whole bunch of things 
Author:  <rc5 at llamas.net > at SMTP
Date:    1997/11/21 15:14


there is no advantage on a PPC's to work in float.  Alpha's might get an 
advantage since they're optimized for float.
     
Kevin
     
> -----Original Message-----
> From:	Jared McIntyre [SMTP:jmcintyre at DEPAUW.EDU] 
> Sent:	Friday, November 21, 1997 2:25 PM
> To:	rc5-digest at llamas.net
> Subject:	[RC5] Re: A whole bunch of things 
> 
> So what chips would be legit to try this on (especially the ones we 
> could
> parrallel process with). And what is the chance that we could get 
> someone
> to implement this. We could really do a lot more keys by crunching 
> both
> ways.  My PPC is itching to do some floating point work. 
> 
> jared
> 
> >     I posted a method for doing the all important rotates in FP. 
> The
> >     modulo additions are trivial.  The only real trouble you'd have 
> is on
> >     architectures with only one "full size" FP register or some 
> (popular)
> >     idiotic architecture that uses a FPU *stack*.  Then you'd have 
> to
> >     continually be loading and reloading "A" and "B" into the FPU 
> >     register.
> >
> >     On (popular) idiotic hardware, I'd expect this to be no better 
> than
> >     the rotateless clients.  However, this has the advantage of 
> being
> >     parallelizable with the normal client on the same hardware. 
> Some
> >     other platforms also allow parallelized FP execution. 
> >            -- Eric Gindrup ! gindrup at Okway.okstate.edu
> ---------------------------------------------------------------------- 
> -------
> >     Nope.  Get an IEEE doubbe (80 bit mantissa).  Put two adjacent 
> copies
> >     of the argument in the most significant bytes (setting the other 
> two
> >     bytes to 0).  Call the byte position collections occupied by the 
> two
> >     copies "high-order" and "low-order" indicating the relative 
> >     significance of their bits.  Set the exponent to represent 
> 2^-(32-k).
> >     Add 1.  This slides the bits we originally inserted by 32-k bits 
> to
> >     the right.  The bits in the position called "low-order" are now 
> the
> >     original argument shifted k bits to the left.
> >     If you want right shift, set the exponent to represent 2^-k 
> instead.
> >            -- Eric Gindrup ! gindrup at okway.okstate.edu 
> >
> 
> 
> 
> ___________________Schwartz____________jmcintyre at depauw.edu___ 
> 
> """"""
> O    O    from Allen Ginsberg's - Refrain 
>   \/
>  ....     -   nobody cares when a man goes mad 
>   ..          he is sorry, God is glad
>               shadow changes into bone 
>               shadow changes into bone 
> 
> 
> --
> To unsubcribe, send 'unsubscribe rc5' to majordomo at llamas.net 
> rc5-digest subscribers replace rc5 with rc5-digest
--
To unsubcribe, send 'unsubscribe rc5' to majordomo at llamas.net 
rc5-digest subscribers replace rc5 with rc5-digest
     


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



More information about the rc5 mailing list