Hi Ben & co,
On 15/02/16 09:34, Kenneth Hoste wrote:
HI Ben,
On 14/02/16 22:11, Ben Roberts wrote:
Since an EB-mediated installation of the new PGI compilers is on my
to do list, I could have a crack at it. Is there an existing
framework module that it would be ideally suited for? A quick look in
the “tools” directory didn’t suggest an obvious fit.
Thanks for looking into this, let us know if you need any help.
Yeah, there's no ideal fit, but maybe just throw it into the
'filetools' module?
Depending on how much stuff we're talking, you can also add a new
module 'easybuild.tools.licensing' or something like that (filetools
is getting rather big).
I was annoyed enough by this this morning to dive in myself.
framework PR: https://github.com/hpcugent/easybuild-framework/pull/1633
updated IntelBase easyblock:
https://github.com/hpcugent/easybuild-easyblocks/pull/839
I think Damian is going to update
https://github.com/hpcugent/easybuild-easyblocks/pull/837 to also us the
find_flexlm_licenses function in the PGI easyblock.
regards,
Kenneth
regards,
Kenneth
Cheers,
Ben
On 13/02/2016, at 11:46 PM, Kenneth Hoste <[email protected]>
wrote:
Hi Joachim,
On 11/02/16 12:14, Joachim Hein wrote:
Hi,
We just build a PGI compiler using PGI-15.10-GCC-4.9.3-2.25.eb and
had some fun with the license handling. We use flexlm and set an
env-variable (LM_LICENSE_FILE). The icc easyconfigs can handle
this, the PGI config does not. In the end we created a dummy
licences file (EasyBuild seems only to check for the existence)
and modified the module file for what we need. So we can now
continue.
Using an
export [email protected]
type statement did not make things build. It does for the icc.
When comparing the easy-configs (PGI and ICC) I didn’t notice the
obvious difference, so that should be in the easyblock. I am
wondering whether some with better expertise on those could have a
look and “upgrade” the PGI.
There's indeed a bit of magic going on for the Intel tools, that
could be useful in other places as well (in the easyblock for PGI,
but also in other places).
The IntelBase generic easyblock, which serves as a base for the
different easyblocks for the Intel compilers/libraries/tools, has
the magic in configure_step [1].
We could look into fleshing this out into a function to framework,
such that the IntelBase and PGI (and other) easyblocks can easily
call out to it.
Things are a bit entangled in IntelBase.configure_step right now
(some things in there are really specific to the Intel tools), so
this is not exactly trivial, but definitely doable.
And now the million dollar question: who's up for it? :-)
regards,
Kenneth
[1]
https://github.com/hpcugent/easybuild-easyblocks/blob/7fe31aa39dd2196298e1f2e328271c9f1d766baa/easybuild/easyblocks/generic/intelbase.py#L204
