add intel/2014b toolchain definition, which includes GCC/4.8.3 as a dependency#948
Conversation
There was a problem hiding this comment.
this is a conservative choice, maybe using the latest imkl v11.1.3.174 is also possible, I'm checking on that with Intel support
There was a problem hiding this comment.
reply from Intel support:
Intel compilers v13.1.x are officially supported by MKL 11.1.x. So yes, it is safe to use the latest MKL release with icc/ifort 2013.5.192.
However, when MKL 11.2 comes out later this year (Q4) and if you decide to use it then we'd strongly recommend you to update your compilers to v14.x. This is mainly because of OpenMP compatibility requirement. MKL 11.2 may have issues with OpenMP runtimes provided by compilers earlier than v14.
So, I'll update this PR to use the latest imkl version.
|
Jenkins: test this please |
1 similar comment
|
Jenkins: test this please |
|
We could maybe all agree on an environment variable to set to point to the intel license file? and if this is set don't try to do a default path. |
There was a problem hiding this comment.
os.path.expanduser("~") is prefered over os.getenv('HOME')
But actually I would prefer using something like an INTELLICENSEFILE environment variable...
There was a problem hiding this comment.
On Wed, Jul 2, 2014 at 4:58 PM, Jens Timmerman [email protected]
wrote:
os.path.expanduser("~") is prefered over os.getenv('HOME')
Can you elaborate?
I have seen the opposite effect (situations where $HOME works, while ~
does not; no proof at the moment)
There was a problem hiding this comment.
the ~ approach is supposed to work on Windows too... I haven't seen that approach fail yet (It's used in a couple of places in EB).
There was a problem hiding this comment.
In that case, I'd agree... let's have it as such and when it breaks we fix
it!
On 2 Jul 2014 17:03, "Kenneth Hoste" [email protected] wrote:
In easybuild/easyconfigs/i/imkl/imkl-11.0.5.192-GCC-4.8.3.eb:
+versionsuffix = '-%s-%s' % (gcc, gccver)
+
+# deps for interface build
+dependencies = [
- ('icc', compver, versionsuffix),
- ('ifort', compver, versionsuffix),
- ('impi', '4.1.3.049', versionsuffix),
+]
+dontcreateinstalldir = 'True'
+
+interfaces = True
+
+# license file
+import os
+license_file = os.path.join(os.getenv('HOME'), "licenses", "intel", "license.lic")the ~ approach is supposed to work on Windows too... I haven't seen that
approach fail yet (It's used in a couple of places in EB).—
Reply to this email directly or view it on GitHub
https://github.com/hpcugent/easybuild-easyconfigs/pull/948/files#r14461765
.
There was a problem hiding this comment.
@fgeorgatos python will expand ~ to $HOME if it is set, otherwise it will look the home dir up in /etc/passwd and also do other tricks on windows. So it is prefered.
There was a problem hiding this comment.
what we really need is a constant provided by the framework that points to the home directory, rather than importing os in an easyconfig file...
So let's fix it that way rather than switching to expanduser in here. Anyone up for that? @JensTimmerman?
|
@JensTimmerman: that's already covered in the |
|
Good enought for me then. |
|
Seems fine. |
|
Test report by @boegel |
|
Merging this in after extensive testing; I've rebuilt all |
add intel/2014b toolchain definition, which includes GCC/4.8.3 as a dependency
depends on easybuilders/easybuild-easyblocks#429 to ensure that FFTW wrappers libraries for Intel MKL are installed correctly
@gjbex, @hajgato, @JensTimmerman: please review?