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

[Hong Zhang]

David Ferreira wrote:
> Hi,
> I looked into the linux_ia64_ifort+mpi_ice_nas optfile and compared it 
> to the one I used
> for testreport and they are very similar, mine being a bit more 
> standard. The only significant
> difference is in the optimization :
> FOPTIM='-O2 -align -ip -xSSE4.2' in mine
> against
> FOPTIM='-O3'     in linux_ia64_ifort+mpi_ice_nas
>
> The -O2 is because I didn't see much a difference in speed compared to 
> -O3 and
> it is safer. The -xSSE4.2 is specific to the normal_N queue and should 
> be replaced by
> -xSSE4.1 for the normal queue (and it does speed up things in my case).
>
> I can replace linux_ia64_ifort+mpi_ice_nas by my version, or we can 
> work something out ?
> What do people think?
> david
>
Hi David
thanks for initiating the fixing.
Notice each cpu has 1GB ram (for nomal queue) and program could easily 
reach that threshold.
So Patrick's check-in with two sets of FFLAGS/CFLAGS could be added.
(http://mitgcm.org/viewvc/MITgcm/MITgcm/tools/build_options/linux_x86_64_ifort_pleiades_n_nas?revision=1.1&view=markup&sortby=date)
hong