MathGroup Archive 2010

[Date Index] [Thread Index] [Author Index]

Search the Archive

Re: Mathematica 8: first impressions

  • To: mathgroup at smc.vnet.net
  • Subject: [mg114241] Re: Mathematica 8: first impressions
  • From: Daniel Lichtblau <danl at wolfram.com>
  • Date: Sun, 28 Nov 2010 06:54:35 -0500 (EST)

----- Original Message -----
> From: "David Bailey" <dave at removedbailey.co.uk>
> To: mathgroup at smc.vnet.net
> Sent: Saturday, November 27, 2010 2:36:21 AM
> Subject: [mg114208] Re: Mathematica 8: first impressions
> On 26/11/10 10:28, Daniel Lichtblau wrote:
> 
> >
> > Actually I would like to see native support for 64 bit integers e.g.
> > in packed arrays. It will take more memory in the case of small
> > integers, but also improve performance for integer linear and
> > polynomial algebra. I think the improvement would be worth the added
> > memory consumption. Granted, for some purposes this might just be
> > added memory use for no gain.
> >
> > I'd say the same about quad precision for hardware floating point
> > arithmetic, except it does not appear to be looming on the near term
> > horizon, in terms of general vendor support.
> >
> > Daniel Lichtblau
> > Wolfram Research
> >
> 
> Could you support packed arrays of 64-bit integers as a special case
> without doubling the size of all machine integers. If you doubled the
> size of all machine integers, I imagine that could impact badly on a
> lot
> of user software.
> 
> David Bailey
> 
> http://www.dbaileyconsultancy.co.uk

Offhand I cannot imagine why. The main memory users are likely to be packed arrays. Other integers will not take up much room.

In other places where native integer size matters e.g. denoting Length, iterator variable ranges, etc., we want to exploit the potential of larger sizes. People who want to restrict against that will need to do so explicitly in their code. Relying on current machine-specific limits to enforce program-specific size bounds is always a bad idea, and I'd not want to cripple all usage just to accomodate what I'd regard as bad programming.

I do not claim this to be the reason you might want 32 bit machine integers, but it is the only place other than packed arrays where I see the distinction between 32 and 64 bits as likely to have an impact.

Daniel Lichtblau
Wolfram Research


  • Prev by Date: Re: Documentation of Eigenvalues[], Eigensystem[]
  • Next by Date: How to short-circuit match failure?
  • Previous by thread: Re: Mathematica 8: first impressions
  • Next by thread: FindRoot to solve large number of polynomials -- how to speed it up?