[MITgcm-devel] [MITgcm-cvs] MITgcm/tools/build_options CVS Commit

[Patrick Heimbach]

Haven't followed this thread thoroughly, but sounds reasonable to me
(sorry for having triggered the optfile flurry).
-p.



On Apr 11, 2010, at 10:07 AM, Menemenlis, Dimitris (3248) wrote:

> David and other Pleaides users, as checked in the Pleiades option  
> file only works on 1216 out of 7104 Pleiades nodes.
> In a first step, I suggest that we check in 2 separate option files:
> (i) a generic option file with SSE4.1 that supports all processors,  
> i.e., including the 5888 Harpertown nodes, and
> (ii) a _nehalem option file with SSE4.2 that is specific to the  
> newer 1216 Nehalem processors.
> A third option file can be added, if needed, when the Westmere  
> processors become available.
> In a second step, if someone has the time to run some proper timing  
> tests, we could implement Constantino's suggestions:
>
>> My suggestions:
>> a) Separate optfiles (per queue). It explodes the number of  
>> optfiles but solves
>> the problem for good.
>> b) The "generic" Pleides optfile uses the -ax instead of the -x  
>> flag. That flag
>> builds a generic binary with optional code paths depending on the  
>> actual
>> processor used. So I would have a -axSSSE3,SSE4.1 (and in the  
>> future ,SSE4.2)
>>
>> See the way they do it at
>> http://astromatic.iap.fr/wsvn/public/filedetails.php?repname=public 
>> +software.swarp&path=%2Ftrunk%2Facx_prog_cc_optim.m4&
>> The -ax flag is not as fast as the -x flag as it involves runtime  
>> processor ID
>> checks but it will allow the same binary to run everywhere on  
>> pleiades.
>
> Let me know if this is acceptable.  Dimitris
>
>
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel

---
Patrick Heimbach | heimbach at mit.edu | http://www.mit.edu/~heimbach
MIT | EAPS 54-1518 | 77 Massachusetts Ave | Cambridge MA 02139 USA
FON +1-617-253-5259 | FAX +1-617-253-4464 | SKYPE patrick.heimbach