Subject: Re: [boost] [Review Request] Multiprecision Arithmetic Library
From: Paul A. Bristow (pbristow_at_[hidden])
Date: 2012-04-06 13:58:32
> -----Original Message-----
> From: boost-bounces_at_[hidden] [mailto:boost-bounces_at_[hidden]] On Behalf Of
> Sent: Friday, April 06, 2012 11:57 AM
> To: boost_at_[hidden]
> Subject: Re: [boost] [Review Request] Multiprecision Arithmetic Library
> >> But let's face it, my O(N^2) is very slowwww for digits.
> >> In your opinion, should I take a crack at this if I find a spare afternoon?
> > Maybe. My understanding is that FFT doesn't become viable until the digit count grows truely
> huge? Would the Karatsuba algorithm be viable sooner and help more users?
> > My gut feeling is that the current implementation is perfectly fine up to a few hundred (and
> thousand?) decimal places, which will probably take care of most use cases.... well my use cases
> In my previous unpublished work (mp_cpp, also using base-10^8), I tuned to the following:
> static const INT32 mp_elem_karatsuba_min = static_cast< INT32>(40); // Approx. 280 digits.
> static const INT32 mp_elem_fft_min = static_cast< INT32>(129); // Approx. 900 digits.
> I've got a recursive Karatsuba template available in my catalog already.
> Maybe we should try it out. The FFT mentioned above was FFTW (a sadly non-BPL wonder of mankind)
> a vanilla FFT would be, maybe, 2 or 3 times slower.
> > So maybe examples are a higher priority for now?
> Yes, you are right. I just don't want to get into trouble with the community.
> Everyone wants to compute a million digits of pi, and they might get mad at us if boost can't do
> But you're right below. We need to stop and get a work of high quality out there to the
> improving it later, like in the fall of 2012.
Agreed - it needs user exposure.
> In fact, as I told you and Paul Bristow
> (both of whom have helped so much in this project), I'm actually booked solid through summer.
I fully understand that. But I'd be willing work up the handful of examples that you mentioned if
that would help (I think it would - I think novices need much more help). Novice views on whether
they find the package useful to them would a (small part) factor in a review.
--- Paul A. Bristow, Prizet Farmhouse, Kendal LA8 8AB UK +44 1539 561830 07714330204 pbristow_at_[hidden]
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk