[MITgcm-support] vectorization

[Chris Hill]

As far as I know the code vectorizes just as well as any other code. All the
key dynamics loops are written in a vector friendly form. Every CPU out
there these days does what is called streaming. Loops that stream well also
vectorize well.

As far as I understand from Patrick the code actually runs reasonably OK on
the NEC if you set the compiler options right. Detlef had someone test the
code who set the compiler and CPP options all wrong. As is usual in this
area Detlef has never actually has said anything to me or Patrick directly,
nor has he provided any information on the runs nor has anyone ever set me
up an account. 

Chris
> -----Original Message-----
> From: mitgcm-support-bounces at mitgcm.org 
> [mailto:mitgcm-support-bounces at mitgcm.org] On Behalf Of Martin Losch
> Sent: Friday, May 07, 2004 9:37 AM
> To: mitgcm-support at mitgcm.org
> Subject: Re: [MITgcm-support] vectorization
> 
> Hi,
> 
> I don't need information specifically on the NEC, anything 
> "vector" is fine. I am supposed to give a talk about these 
> things and I would like to have an answer for this type of 
> question: "yes, but on vector machines ...", because I just 
> heard through the grapevine that Detlef is not too happy 
> about MITgcm and NEC, but I don't know the reason.
> 
> Martin
> 
> 
> On Friday, May 7, 2004, at 03:28 PM, Chris Hill wrote:
> 
> > Martin,
> >
> >  Patrick has information on NEC performance I think. So far only 
> > Detlef has been running on these and he doesn’t tell us 
> what he finds. 
> > Is someone can can me an account I'd be happy to take a look.
> >
> >  The electricity to building 54 went out for a second time 
> this week.
> > http://mitgcm.org should be back up shortly.
> >
> > Chris
> >
> >> -----Original Message-----
> >> From: mitgcm-support-bounces at mitgcm.org 
> >> [mailto:mitgcm-support-bounces at mitgcm.org] On Behalf Of 
> Martin Losch
> >> Sent: Friday, May 07, 2004 9:15 AM
> >> To: MITgcm Support
> >> Subject: [MITgcm-support] vectorization
> >>
> >> Hi,
> >> are there any presentable numbers about the performance of the 
> >> MITgcm-code on vector machines, such as the NEC in Hamburg?
> >> Can't find anything on the web, probably because 
> http://mitgcm.org/ 
> >> does not work? (error 404)
> >>
> >> Martin
> >>
> >> Martin Losch // mailto:mlosch at awi-bremerhaven.de 
> >> Alfred-Wegener-Institut für Polar- und Meeresforschung Postfach 
> >> 120161, 27515 Bremerhaven, Germany
> >> Tel./Fax: ++49(471)4831-1872
> >> http://www.awi-bremerhaven.de/People/show?mlosch
> >>
> >> _______________________________________________
> >> MITgcm-support mailing list
> >> MITgcm-support at mitgcm.org
> >> http://dev.mitgcm.org/mailman/listinfo/mitgcm-support
> >>
> >
> >
> > _______________________________________________
> > MITgcm-support mailing list
> > MITgcm-support at mitgcm.org
> > http://dev.mitgcm.org/mailman/listinfo/mitgcm-support
> 
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://dev.mitgcm.org/mailman/listinfo/mitgcm-support
>