On 13/04/2017 16:53, Markus Geimer wrote:
Maik,
As far as I can see in my build logs, this is nothing that is induced
by EasyBuild. It seems as if the configure scripts of various math
libraries needed to build GCC (i.e., GMP, isl, etc.) set these flags
automatically based on the host architecture. To fix this behavior,
you need to figure out how to convince the configure scripts to use
a "conservative" setting and then modify the GCC easyblock accordingly.
This is indeed not done by EasyBuild, since for installations done with
the 'dummy' toolchain the build environment (e.g. $CFLAGS etc.) is not
set up by EasyBuild, hence --optarch has no effect whatsoever.
By default, EasyBuild performs the 3-stage bootstrap procedure for GCC,
which results in a native compiler.
What you want to do is basically build a cross-compiler...
regards,
Kenneth
Markus
On 04/13/2017 03:57 PM, Maik Schmidt wrote:
Nope, wasn't that. Even when explictly setting --optarch=GENERIC on the
eb call, it uses whatever the host architecture is in the second and
third build stages, e.g. "-march=haswell -mtune=haswell"
Am 13.04.2017 um 11:53 schrieb Maik Schmidt:
Okay, I think it might be due to the fact that GCC/GCCcore are
actually built with the dummy toolchain and so my compiler-specific
optarch that only includes "Intel" and "GCC" is not applied there. Can
I also add "dummy" to the optarch as another toolchain?
Best,
Maik
Am 13.04.2017 um 10:13 schrieb Maik Schmidt:
I have noticed that when doing a GCC/GCCcore build, it always uses
the HOST architecture for -march and -mtune starting in compile stage
2, even though I have set something else via optarch. This is a real
problem for heterogenous clusters if you do not make sure that the
build host always is one with the oldest architecture, because then
it leads to the error:
"internal compiler error: Illegal instruction"
when the resulting GCC is used on an older node.
Is this a known issue, or is there a way to make the GCC build
process respect what EasyBuild sets as the target architecture, even
during the later compile stages?
Thank you kindly in advance.
--
Dr. Markus Geimer
Juelich Supercomputing Centre
Institute for Advanced Simulation
Forschungszentrum Juelich GmbH
52425 Juelich, Germany
Phone: +49-2461-61-1773
Fax: +49-2461-61-6656
E-Mail: [email protected]
WWW: http://www.fz-juelich.de/jsc
------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------
Forschungszentrum Juelich GmbH
52425 Juelich
Sitz der Gesellschaft: Juelich
Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498
Vorsitzender des Aufsichtsrats: MinDir Dr. Karl Eugen Huthmacher
Geschaeftsfuehrung: Prof. Dr.-Ing. Wolfgang Marquardt (Vorsitzender),
Karsten Beneke (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt,
Prof. Dr. Sebastian M. Schmidt
------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------
