[MITgcm-support] output.txt empty

[Martin Losch]

Hi Nicolas,

inlining: the sx compiler only looks for routines to inline in the  
same file unless you give it the names of files to search for  
routines. Unfortunately, when you give it a filename and this file  
does not exist, then the compilation aborts with an error. Therefore,  
in FEXTRAFLAGS, there are only routines that are linked from model/ 
src, that is, that are always linked into the build-directory, so it  
should be safe to use this option without any modification. I noticed  
that the comment describing this flag is no longer up to date,  
because FEXTRAFLAGS does not affect genmake2. If you want to inline  
subroutines from some of the packages (currently I cannot thing of an  
example candidate, but there are definitely some), then you'd have to  
include them yourself.

I am also quite sure that only the last few lines in the opt files  
are AWI-specific and you should be able to use it for cross compiling  
"as is" (unless you need netcdf).

I also noticed that in some cases the gather/scatter_2d routines  
break during vectorization (give segmentation faults). I don't think  
that this has been fixed yet (in the compiler!). If you have this  
problem, let me know and I'll give you the work-around (which I have  
not checked in, because it's a work around, but maybe I should, what  
do the code-czars think?).

Martin

On 9 Jun 2008, at 17:13, Nicolas Grisouard wrote:

> Martin Losch wrote:
>>
>> To improve the performance on your SX8, there is sx8 template  
>> optfile:
>> tools/build_options/SUPER-UX_SX-8_sxf90+mpi_awi
>> that may be a good starting point for the SX-compiler on an SX8.  
>> Some of the paths are specific to "our" machine at the AWI, but in  
>> general you should be able to use this file.
>>
>> Martin
>
> Vielen Dank für die Hilfe! Indeed, I might have to adapt a few  
> things (and more important, to know what to adapt). I'll have a  
> chat with the IDRIS assistance to know what to do and if there are  
> some differences indeed, I'll come up with a new optfile and  
> propose it.
>
> By the way, I'm not familiar with inlining (actually, today was the  
> first day I heard about it). It looks like we use the code in  
> different configurations so from what I have understood, I don't  
> think your FEXTRAFLAGS would suit to me, would they ? Is there an  
> easy way to set this line fro scratch ?
>
> Noch Mal vielen Dank,
>
> Nicolas.
>
> -- 
> Nicolas GRISOUARD
> PhD Student - ERES division
> Laboratoire des Ecoulements Geophysiques et Industriels
> BP 53
> 38041 Grenoble cedex 9 France
> tel : +33 (0)476 825 037 - fax : +33 (0)476 825 271
>
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support