|
|
"Alex" <alex@xxxxxxxxxxxxx> writes:
> Hello,
>
> Phil Carmody wrote:
> > "Alex" <alex@xxxxxxxxxxxxx> writes:
> > > Current sieving by software is looking for factors for the
> > > remaining 9 k values for n < 50,000,000 and p is approaching
> > > the limit of the software siever at 2^50.
> >
> > The only reason I claimed that the limit of my sieve (NbeGon)
> > was p=2^52 was that my C implementation was portable across
> > all platforms and that many (all but x86) don't have floating
> > point precision greater than that. Paul's sieve, written in
> > x86 asm, and NbeGon run on x86 do not have such a 2^52 limit.
> > (Having a 2^63 one instead.)
>
> This is proth_sieve by Paul Jobling and Mikael Klasson. From what
> I've read the cmov version (x86 assembly for Athlons) has a limit
> of 2^50. That's what SoB is using at the moment.
I can imagine a P4/SSE2 version might have that limitation, but
there's no need for an Athlon version to, particularly if you're
using assembly language routines. That's what the 80-bit FPU is
designed for.
> Thanks for the tip on floating point, it had never even crossed
> my mind, although I would guess you're using FFT multiplcation and
> getting the modulus operation for free.
Nope, only use FFTs for bignums. Not a bignum in sight here.
> My siever will never be quick but it's a good learning experience.
You can't beat learning by doing.
Phil
--
If a religion is defined to be a system of ideas that contains unprovable
statements, then Godel taught us that mathematics is not only a religion, it
is the only religion that can prove itself to be one. -- John Barrow
|
|