Catching up...
Tried hacking in the most disgusting way the icc and ifort ebs...
Replacing the sources var with the parallel_studio_xe_2015_update3.tgz
tarball works, sorta, at least for the icc installation, tried the
same for the ifort and got out of disk space on the VM where I was
trying this...then I realized the truth in Kenneth words at the start
of the thread "it's likely you'll end up with an installation that's a
lot fatter than it needs to". I guess if I start with enough space it
should work...but will end up with an inordinate amount of duplicate
and useless stuff.
From looking at the silent.cfg file within the tarball it looks like
one can tell install.sh to only install specific components:
# list of components to install, valid values are: {ALL, DEFAULTS, anythingpat}
COMPONENTS=DEFAULTS
Fiddling with intelbase.py, etc is beyond me...can you guys think of
an "easy" way to hack the proper "COMPONENTS" into the silent.cfg via
the eb file (if one could find them) to make this work?
Cheers,
C.
--
Dr. Cristián G. Sánchez
UNIVERSIDAD NACIONAL DE CORDOBA
Facultad de Ciencias Químicas
Departamento de Matemática y Física
Haya de la Torre y Medina Allende
EDIFICIO INTEGRADOR - SUBSUELO
CIUDAD UNIVERSITARIA
X5000HUA CORDOBA-ARGENTINA
Phone: +54-351-5353850 (Ext. 55020)
[email protected]
--
On Fri, Sep 18, 2015 at 11:29 AM, Jack Perdue <[email protected]> wrote:
> On 09/18/2015 05:43 AM, Kenneth Hoste wrote:
>>
>> On 18/09/15 12:31, Fokko Masselink wrote:
>>>
>>> Take iccifort-2015.1.133.eb and iccifort-2015.1.133-GCC-4.9.2.eb for
>>> example:
>>>
>>> iccifort-2015.1.133.eb has these:
>>> dependencies = [
>>> ('icc', version),
>>> ('ifort', version),
>>> ]
>>>
>>>
>>> and iccifort-2015.1.133-GCC-4.9.2.eb has these:
>>> dependencies = [
>>> ('icc', version, versionsuffix),
>>> ('ifort', version, versionsuffix),
>>> ]
>>>
>>> If you look at icc-2015.1.133-GCC-4.9.2.eb
>>> it installs icc and has GCC as dependency.
>>>
>>> though you could also change it, so that icc-2015.1.133-GCC-4.9.2.eb
>>> doesn't
>>> install the icc, but just has 2 dependencies: icc-2015.1.133.eb and
>>> GCC-4.9.2.eb.
>>>
>>> That way you de-duplicate the installation of icc. Right?
>>>
>>> Any reason not to do it like that?
>>
>> Yes, there is a reason.
>>
>> During the installation of icc, I'm not sure what the Intel installer
>> doesn't perform magic to figure out which GCC version is used, to then
>> hardcode that in one way or another in the installation.
>>
>> In that sense, using icc as a dependency next to a GCC dependency may
>> yield a different installation of icc as it would be if the EB-provided GCC
>> is available.
>>
>> This only matters w.r.t. the easyconfigs themselves: typically, you would
>> only install one of the icc easyconfigs, not both (with/without GCC).
>>
>> Just go for the one where a GCC is included underneath icc, and just skip
>> the one that relies on the system GCC?
>>
> I've been meaning to look at icc after my experiences with IBM's xlc
> to see if we could do something similar to what I do on Power/BG. e.g.
>
> http://www.siliconslick.com/easybuild/ebfiles_repo_cleaned/curie/xlc/xlc-13.1.0.1.484.eb
>
> Basically, install the vendor's compiler and then come up with
> different configurations (overriden via an environment variable)
> for the different GCCs used underneath. That avoids a lot
> of duplication.
>
> jack
>
>
