[rc5] Why are Alpha's so much slower then wintel boxe
gindrup at okway.okstate.edu
Mon Nov 3 09:27:01 EST 1997
This is certainly slower on Intels. Can you ensure that it is slower
on non-rotating platforms.
An, you approximately remember correctly. RC5-32/12/8 is using 32-bit
words and therefore is performing 32-bit rotates. This still ensures
one valid copy starting at the 33rd least significant bit.
Also, on rotating architectures with separate FP pipes, this would
allow an extra key or two per millisecond...
-- Eric Gindrup ! gindrup at Okway.okstate.edu
______________________________ Reply Separator _________________________________
Subject: Re: [rc5] Why are Alpha's so much slower then wintel boxe
Author: <rc5 at llamas.net > at SMTP
Date: 10/31/97 4:50 PM
At 06:25 PM 10/31/1997 -0600, Eric Gindrup wrote:
> Oops. Drop the last term from that long sum and add that the most
> significant byte is now useless.
> Also, I got a bit fuzzy about x. It would have been better to say:
> "Let abcdefgh be some 8-bit... and let x = [abcdefgh, 8-3-1]" I was
> thinking I had said this in the rest of the message. (*sigh*)
Your solution's slower than what we've got now, unfortunately. (Plus, since
we're rotating larger numbers (32 or 64 bits, IIRC) you may still run into
FP truncation/rounding issues) Faking a left rotate takes an AND, an OR, a
left shift, and a right one. (Not necessarily in that order :) You can do
it with a right rotate, too, but I don't think there's one of those around
in the Alphas either.
----------------------------------------"it's like this"-------------------
Dan Sugalski even samurai
sugalskd at osshe.edu have teddy bears
and even the teddy bears
To unsubscribe, send email to majordomo at llamas.net with 'unsubscribe rc5' in the
To unsubscribe, send email to majordomo at llamas.net with 'unsubscribe rc5' in the body.
More information about the rc5