[MITgcm-support] Adjoint tutorial problems

Patrick Heimbach heimbach at MIT.EDU
Fri Feb 3 23:20:49 EST 2012


Great, good job!

We have two ways of how to write adjoint-related checkpoints:
1. (default) merge many state variables into single array
   then write that array (and vice versa) causing large record lengths;
2. instead, using below flag writes (reads) every state variable into (from) separate file.

Reason for these alternatives is to be able (to some extent)
to adapt I/O performance to target platform when running (I/O-heavy) 
parallel adjoint, i.e. write few files (large record lengths)
vs. writing many files (small record lengths).
Not clear though why it's a problem for such a small config. as yours.
Perhaps there are ways (compiler options) to "tune" ifort?

But for the time being you should be good to go
(at least until you hit performance issues ;)

Cheers
p.

On Feb 3, 2012, at 10:46 PM, Marcello Magaldi wrote:

> Thanks Matt!
> 
> It looks like
> 
> #define AUTODIFF_USE_OLDSTORE_3D
> 
> makes it go thru.
> 
> Do you guys have any idea why is that?
> Do you think that it is worthy a better investigation?
> 
> Thanks again,
> M.
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support

---
Patrick Heimbach | heimbach at mit.edu | http://www.mit.edu/~heimbach
MIT | EAPS 54-1420 | 77 Massachusetts Ave | Cambridge MA 02139 USA
FON +1-617-253-5259 | FAX +1-617-253-4464 | SKYPE patrick.heimbach





More information about the MITgcm-support mailing list