Raj,

I will most likely be in a similar situation soon.  Were you able to
solve this and get a generic build that works on both Intel and AMD?
I didn't see a corresponding PR yet...

Markus


On 12/01/2017 04:33 PM, Raj Ayyampalayam wrote:
> I managed to get it working on both platforms by adding --enable-fat
> option to the stage2_stuff GMP configure options in the EB_GCC
> easyblock. I will look at the ISL build setting.
> 
> Thanks!
> -Raj
> 
> On Fri, Dec 1, 2017 at 3:27 AM Markus Geimer <m.gei...@fz-juelich.de
> <mailto:m.gei...@fz-juelich.de>> wrote:
> 
>     All,
> 
>     > I believe that it's not GCC itself but the auxiliary infrastructure
>     > components (GMP, MPFR, etc.) that auto-optimize for the build
>     > architecture.  Though I have to recheck the build logs to confirm
>     > this.
> 
>     I've just checked the build logs of GCCcore 6.3.0 and can confirm
>     that at least the GMP and ISL builds (for 'stage2_stuff') use
>     platform-dependent '-march' flags...
> 
>     Markus
> 
> 
>     > If this is the case, we would need to find a way to convince their
>     > configure scripts to build generic code, and then adjust the EasyBlock
>     > accordingly.
>     >
>     > Markus
>     >
>     >
>     > On 11/30/2017 05:40 PM, Alvarez, Damian wrote:
>     >> I do not know why you are experiencing this, but I don’t think
>     --optarch
>     >> plays any role here. --optarch=GENERIC will look for the appropriate
>     >> value for the compiler used, but GCCcore is built on top of
>     dummy, which
>     >> doesn’t have any optarch defined. So EB shouldn’t add any
>     architecture
>     >> specific bits to GCCcore itself. Packages on top of GCCcore (like
>     >> binutils) are a different matter.
>     >>
>     >>
>     >>
>     >> So I’d look more into the details of building GCC per se, not in the
>     >> details of EB.
>     >>
>     >>
>     >>
>     >> Just my 2 cents
>     >>
>     >>
>     >>
>     >> Damian.
>     >>
>     >>
>     >>
>     >> *From: *<easybuild-requ...@lists.ugent.be
>     <mailto:easybuild-requ...@lists.ugent.be>> on behalf of Raj Ayyampalayam
>     >> <ans...@gmail.com <mailto:ans...@gmail.com>>
>     >> *Reply-To: *"easybuild@lists.ugent.be
>     <mailto:easybuild@lists.ugent.be>" <easybuild@lists.ugent.be
>     <mailto:easybuild@lists.ugent.be>>
>     >> *Date: *Tuesday 28 November 2017 at 21:20
>     >> *To: *"easybuild@lists.ugent.be
>     <mailto:easybuild@lists.ugent.be>" <easybuild@lists.ugent.be
>     <mailto:easybuild@lists.ugent.be>>
>     >> *Subject: *[easybuild] Building GCCcore that will run on AMD and
>     Intel CPU
>     >>
>     >>
>     >>
>     >> Hello,
>     >>
>     >>
>     >>
>     >> I am trying to build a the software stack for our new cluster using
>     >> easybuild. The cluster has older AMD Opterons and new Intel Broadwell
>     >> based machines. We have been building applications on the AMD
>     machines
>     >> and recently realized that they would not run on Intel machines.
>     >>
>     >>
>     >>
>     >> I read the easybuild docs and tried to use the --optarch=GENERIC
>     option
>     >> to build the entire application stack including the toolchains. I am
>     >> having issues with getting GCCcore to work on both platforms. If we
>     >> build GCCcore on AMD it does not work on Intel platforms and vice
>     versa.
>     >>
>     >>
>     >>
>     >> It seems that the problem lies in the stage2 part of the build. In
>     >> stage1, the build system type is identified as
>     x86_64-unknown-linux-gnu
>     >> and for stage2 it is identified as broadwell-pc-linux-gnu.
>     >>
>     >>
>     >>
>     >> Am I missing some configuration options or is there a trick to
>     building
>     >> GCC for multiple platforms?
>     >>
>     >>
>     >>
>     >> Thanks,
>     >>
>     >> Raj
>     >>
> 
>     --
>     Dr. Markus Geimer
>     Juelich Supercomputing Centre
>     Institute for Advanced Simulation
>     Forschungszentrum Juelich GmbH
>     52425 Juelich, Germany
> 
>     Phone:  +49-2461-61-1773 <tel:+49%202461%20611773>
>     Fax:    +49-2461-61-6656 <tel:+49%202461%20616656>
>     E-Mail: m.gei...@fz-juelich.de <mailto:m.gei...@fz-juelich.de>
>     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
>     
> ------------------------------------------------------------------------------------------------
>     
> ------------------------------------------------------------------------------------------------
> 

-- 
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: m.gei...@fz-juelich.de
WWW:    http://www.fz-juelich.de/jsc

Reply via email to