[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: randn benchmarks
From: |
David Bateman |
Subject: |
Re: randn benchmarks |
Date: |
Sat, 24 Jan 2004 10:39:59 +0100 |
User-agent: |
Mutt/1.4.1i |
I don't want more precision, just that the cycle time of the random
integer generator in the zigurat code is 32 bit and thus repeats in
2^32. I see now that only this has to be replaced in the code to
get a longer repetition time. In fact we could even use the Mersenne
Twister code here...
D.
According to Paul Kienzle <address@hidden> (on 01/23/04):
>
> On Jan 23, 2004, at 7:36 AM, David Bateman wrote:
> >... In any case, the ziggurat code in
> >this paper is for 2^32 bit seeds, which is a bit short for my liking.
> >I'd prefer to target a 64 bit seed as does matlab, so a possible
> >solution is just to rewrite the code from the paper for a 64 bit
> >seed..
>
> Do you want to generate 53 bits of precision in rand() or are you
> happy with 32 bit? It will of course slow down the code by almost
> a factor of two.
>
> Paul Kienzle
> address@hidden
--
David Bateman address@hidden
Motorola CRM +33 1 69 35 48 04 (Ph)
Parc Les Algorithmes, Commune de St Aubin +33 1 69 35 77 01 (Fax)
91193 Gif-Sur-Yvette FRANCE
The information contained in this communication has been classified as:
[x] General Business Information
[ ] Motorola Internal Use Only
[ ] Motorola Confidential Proprietary
- Re: benchmarks - sort, (continued)
- benchmarks, John W. Eaton, 2004/01/06
- packages (was: benchmarks), John W. Eaton, 2004/01/06
- randn benchmarks, David Bateman, 2004/01/22
- Re: randn benchmarks, Paul Kienzle, 2004/01/23
- Re: randn benchmarks, Dirk Eddelbuettel, 2004/01/23
- Re: randn benchmarks, Dmitri A. Sergatskov, 2004/01/23
- Re: randn benchmarks, Dirk Eddelbuettel, 2004/01/24
- Re: randn benchmarks, Paul Kienzle, 2004/01/24
- Re: randn benchmarks, Paul Kienzle, 2004/01/24
- Re: randn benchmarks, Dmitri A. Sergatskov, 2004/01/24
- Re: randn benchmarks, David Bateman, 2004/01/25
eigenvalues 3 times speedup patch [Was: benchmarks], David Bateman, 2004/01/23