[mvapich-discuss] xFLAGS and RPMs

[Damian Alvarez]

Dear mailing list,

I think we've found a "feature" inhered from MPICH2, which might lead to undesired behaviour.

MPICH2 makes a distinction between xFLAGS, MPICH2LIB_xFLAGS, MPICH2_MAKE_xFLAGS, and MPICH2_MPIX_FLAGS. This can be seen in section 3 of their README (http://www.mpich.org/static/downloads/3.2/mpich-3.2-README.txt) which has been there for a long time.

It seems like MVAPICH2 inhered that behaviour. I am unsure if this is in the MVAPICH2 documentation. I couldn't find it but I didn't look too deeply.

The issue is the following: using xFLAGS (like CFLAGS, CXXFLAGS, etc) is a fairly common way to compile software. Due to this "feature", the flags used to compile the compiler/runtime will be transparently and [maybe] inadvertently added to the mpixxx wrappers. That means that applications compiled with MPI will pass the same set of flags used to compile MPI itself, and any additional flag that the user might provide. That can be conflictive and can lead to incompatible compiler flags and/or underperforming options.

The distinction between CFLAGS and MPICH2LIB_xFLAGS, etc,  has been mentioned in the mailing list a couple of times before, but I didn't find any discussion where this issue is highlighted. That's why I am writing to the mailing list. Maybe it is worth a discussion and be mentioned in the documentation in a more prominent position.

I guess that due to this obscure "feature", the RPMs provided for MVAPICH2-GDR have something that I guess was not intended. The MPI wrappers pass this set of flags to the underlying compiler: -O2 -g -pipe -Wall -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector --param=ssp-buffer-size=4 -m64 -mtune=generic. I doubt flags like -O2 -g -mtune=generic are supposed to be passed to the compiler regardless of what the user specifies.

Am I totally off, or this set of flags shouldn't be included in the wrappers shipped on the RPMs?

Damian

--
Dr. Damian Alvarez
Juelich Supercomputing Centre (JSC)
Institute for Advanced Simulation
Forschungszentrum Juelich GmbH
52425 Juelich, Germany
Phone: +49 2461 61 9336
WWW: http://www.fz-juelich.de/jsc/
JSC is member of the
Gauss Centre for Supercomputing



------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------
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
------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20160118/457e1b4c/attachment.html>