[MITgcm-devel] new computer

[Jean-Michel Campin]

Hi Martin,

I think that we have similar issues regarding some changes in
ifort options with version of intel compiler, in
 linux_amd64_ifort (older version) and linux_amd64_ifort11 (used
also with v.12).
But I don't think it would be a good idea to have an other
optfile linux_ia64_ifort11 and to leave linux_ia64_ifort as it is
(and no longer tested). So I would vote for updating the
present linux_ia64_ifort to work with new compiler (v.12 and
likely will work also with v.11).

There might be few edits that I would suggest:
1) don't need  $MCMODEL in NOOPTFLAGS since NOOPTFLAGS is
 replacing FOPTIM (and not FFLAGS) when compiling files from 
 NOOPTFILES
2) commented line: #MAKEDEPEND=cyrus_makedepend
 would change it to #MAKEDEPEND=tools_xmakedepend
 (this is the new default tested in few exp. such as in 
  lab_sea/build/genmake_local)
3) I would lower the optimisation level in CFLAGS
 since the C source code we have is not crutial to optimise
 (in fact, I sometime even put a "-O0" !)
I can do those edits after you check-in your changes.

Cheers,
Jean-Michel

On Wed, Jan 11, 2012 at 03:39:52PM +0100, Martin Losch wrote:
> Hi Jean-Michel,
> 
> since mitgcm-devel is down: I am now able to run testreport, even with the proper global_ocean.90x40x15 on our SGI UV100 with ifort 12.1.0
> 
> Unfortunately I had to deviate from the linux_ia64_ifort option file
> 1. removed -static because I do not have static mpi-libraries
> 2. removed deprecated flag -w95
> 3. replaced -mp by -fp-model precise -fp-model except, not sure if this is the approriate combination for ieee conformant floating point arithmetic (I just don't understand "man ifort")
> 
> see attached file.
> 
> Do you still think that this file is generic enough to check in as a new linux_ia64_ifort?
> 
> Martin
> 
> 
> On Jan 9, 2012, at 6:27 PM, Jean-Michel Campin wrote:
> 
> > Hi Martin,
> > 
> > I think it's a good idea to add the kind of "generic" MPI_INC_DIR etc ...
> > in this optfile. 
> > And you can safely remove (or may be, comment out) the -static option, 
> > since it's not the standard way (I only see it in this particular optfile: 
> > linux_ia64_ifort+mpi_altix_meinesz, and only commented out in 1 other 
> > amd64_ifort optfile) and, although this optfile was used for non-mpi 
> > built on columbia, the columbia mpi optfile does not have the -static
> > option.
> > 
> > Waiting for new testreport results !
> > Cheers,
> > Jean-Michel
> > 
> > On Mon, Jan 09, 2012 at 03:21:56PM +0100, Martin Losch wrote:
> >> Hi Jean-Michel,
> >> 
> >> we have a new platform that I could include into our testing scheme, it's an "SGI UV100" with Linux and ifort.
> >> 
> >> The closest build-options files is linux_ia64_ifort, but rather than adding a new build_options file, I'd like to use this generic files. I added some generic lines about MPI_INC_DIR (just as in e.g. linux_amd_gfortran), that I could check in. All works fine except for the ifort flag "-static":
> >> The mpi libraries on this sgi are only available as dynamic libraries, so that I cannot use the flag "-static". Do you have a good idea how I can still use this opt file or will I have to make a specific one for this machine?
> >> 
> >> 
> >> Martin
> >> 
> >> 
>