Re: [Torbjorn Granlund: GNU Multiple Precision Arithmetic Library]

[markh@wimsey.bc.ca (Mark C. Henderson)]

> Well, I've done some comparison between gmp 1.3 and RSAREF (as distributed
> with RIPEM 1.07) on a RS6000/320
> 
> The key generation speeds turned out to be almost exactly the same using
> the "aixgcc/xlc mixed" version of RSAREF/RIPEM and gmp compiled entirely
> with IBM's xlc compiler.
> 
> (for those who haven't poked around in RIPEM, I'll say that this RS6000
> version of RIPEM uses the "long long" data type to get 32 bit by 32 bit
> multiplications with 64 bit results quickly).
> 
> So, it wouldn't surprise me if some clever person could optimize
> gmp for the RS6000 and end up with something faster than what's
> provided in RIPEM 1.07.

In fact, after reading some code, I noticed the optimisations are
in there, if one compiles with gcc (a few bytes of RS6k assembler
makes a significant difference!). Anyway, the particular benchmark
I'm playing with (key generation) is now about 20% faster with gmp
than with the routines provided with RIPEM.

Mark

-- 
Mark Henderson
markh@wimsey.bc.ca
RIPEM key available by key server/finger/E-mail
  MD5OfPublicKey: F1F5F0C3984CBEAF3889ADAFA2437433