Where to find authoritative MP Results

Richard Outerbridge

unread,

Dec 12, 2009, 9:42:16 PM12/12/09

to

Could anyone please recommend an authoritative source implementation of
(provably-correct) multiple-precision floating-point arithmetic
within 4 ulps (roughly ulp = bits) in the least-significant-digit over
multiplication, division and modulo for any arbitrary precision greater
than 30 decimal digits?

Noob

unread,

Dec 14, 2009, 7:48:52 AM12/14/09

to

Richard Outerbridge wrote:

I was not aware that multi-precision floating-point was a requirement
in cryptography.

Wolfgang Ehrhardt

unread,

Dec 14, 2009, 12:23:08 PM12/14/09

to

Although you did not mention your programming language, I think you
should have look at MPFR from http://www.mpfr.org/.

<quote>The MPFR library is a C library for multiple-precision
floating-point computations with correct rounding.</quote>
--
Do not use the e-mail address from the header! You can get a
valid address from http://home.netsurf.de/wolfgang.ehrhardt
(Free open source Crypto, CRC/Hash, MPArith for Pascal/Delphi)

Richard Outerbridge

unread,

Feb 6, 2010, 4:32:57 AM2/6/10

to

In article <hg5c65$pe8$1...@aioe.org>, Noob <ro...@127.0.0.1> wrote:

> I was not aware that multi-precision floating-point was a requirement
> in cryptography.

In a running-key (pseudo-OTP) multiplicative implementation one might
want to rely on identically-computed least-significant computed bits
from floating-point operations.

Expensive to compute, but would it be at least as expensive to search
for the more significant truncated bits to achieve key recovery?

outer

unruh

unread,

Feb 6, 2010, 1:21:00 PM2/6/10

to

On 2010-02-06, Richard Outerbridge <ou...@interlog.com> wrote:
> In article <hg5c65$pe8$1...@aioe.org>, Noob <ro...@127.0.0.1> wrote:
>
>> I was not aware that multi-precision floating-point was a requirement
>> in cryptography.
>
> In a running-key (pseudo-OTP) multiplicative implementation one might
> want to rely on identically-computed least-significant computed bits
> from floating-point operations.

One would not want to do that unless one did not care that one's
recipient could not read the messages. The truncation and roundoff error
could well vary from machine to machine or processor to processor.
A useful crypto system must be deterministic and the recipient must
obtain the same results as the sender.

Richard Outerbridge

unread,

Feb 8, 2010, 2:32:17 AM2/8/10

to

In article <slrnhmrcoc...@wormhole.physics.ubc.ca>,
unruh <un...@wormhole.physics.ubc.ca> wrote:

> On 2010-02-06, Richard Outerbridge <ou...@interlog.com> wrote:
> > In article <hg5c65$pe8$1...@aioe.org>, Noob <ro...@127.0.0.1> wrote:
> >
> >> I was not aware that multi-precision floating-point was a requirement
> >> in cryptography.
> >
> > In a running-key (pseudo-OTP) multiplicative implementation one might
> > want to rely on identically-computed least-significant computed bits
> > from floating-point operations.
>
> One would not want to do that unless one did not care that one's
> recipient could not read the messages. The truncation and roundoff error
> could well vary from machine to machine or processor to processor.
> A useful crypto system must be deterministic and the recipient must
> obtain the same results as the sender.

Obviously. That's true of all mechanisms, from the Jeffersonian to the
Enigma, and to DEA and AES implementations. So my question remains:

Mok-Kong Shen

unread,

Feb 8, 2010, 11:21:41 AM2/8/10

to

If there are incompatibilites due to hardware and software/compiler,
some appropriate trials in the basic operations should reveal these,
I suppose. Due to the fall of costs, arrangement to have communication
partners to have compatible hard- software is fortunately no longer
a severe bottle neck now as compared to decades before in my humble
view.

M. K. Shen