cc1: error: unrecognized command line option "-ipo"

[Andreas Schreiber]

Hey guys

I found today that I cannot compile my PC runs any more. The compiler
crashes with:
> make[2]: Entering directory
> `/afs/ipp-garching.mpg.de/home/a/aschreib/tmp/test/src'
> if [ "" ]; then \
> rm -f cparam.inc; \
> rm -f cparam_pencils.inc; \
> ln -s cparam.inc cparam.inc; \
> ln -s cparam_pencils.inc cparam_pencils.inc; \
> fi
> sed -e 's/REAL/DOUBLE_PRECISION/g' mpicomm.f90 > mpicomm_double.f90
> mpicc -DFUNDERSC=1 -ipo -DDOUBLE_PRECISION -o syscalls_ansi.o -c
> syscalls_ansi.c
> mpif90 -implicitnone -ipo -O3 -mcmodel=large -shared-intel -r8 -std95
> -o cparam.o -c cparam.f90
> mpif90 -implicitnone -ipo -O3 -mcmodel=large -shared-intel -r8 -o
> fftpack.o -c fftpack.f90
> scripts/mkdummyinc -d noinitial_condition.f90 -s
> noinitial_condition.f90 > initial_condition_dummies.inc
> ln -s ghostfold_mpicomm.f90 ghostfold_mpicomm_double.f90
> mpicc -DFUNDERSC=1 -ipo -DDOUBLE_PRECISION -o nogsl.o -c nogsl.c
> cc1: error: unrecognized command line option "-ipo"
> make[2]: *** [syscalls_ansi.o] Error 1
> make[2]: *** Waiting for unfinished jobs....
> gfortran: unrecognized option '-shared-intel'
> gfortran: unrecognized option '-r8'
> gfortran: unrecognized option '-shared-intel'
> gfortran: unrecognized option '-r8'
> cc1: error: unrecognized command line option "-ipo"
> make[2]: *** [nogsl.o] Error 1
> f951: error: unrecognized command line option "-std95"
> make[2]: *** [cparam.o] Error 1
> make[2]: Leaving directory
> `/afs/ipp-garching.mpg.de/home/a/aschreib/tmp/test/src'
> make[1]: *** [code] Error 2
> make[1]: Leaving directory
> `/afs/ipp-garching.mpg.de/home/a/aschreib/tmp/test/src'
> make: *** [default] Error 2
> Make failed: <>

With an older pencil-code Version it compiles fine! So what changed lately?

best wishes
Andreas

[Philippe Bourdin]

Hello Andreas,

seems your config tries to use the Intel_MPI setup together with GNU
compilers. Can you maybe check, if you have correctly installed/loaded
the Intel compilers or alternatively switch to the GNU-GCC_MPI setup?
And which .conf file are you using exactly? (Not visible in the output.)

Thanks and best greetings,

Philippe.

[Andreas Schreiber]

Hi Philippe

all I do is "module load intel" as before (as said, old pc version does
its job) and use pc_build, after pc_build --cleanup.

Im not really sure whether I know how to switch to a GNU-GCC_MPI setup.
The .conf file is 'pencil-code/config/hosts/RZG/theo02.bc.rzg.mpg.de.conf'

Does someone changed an enviroment variable so that wrong compilers are
loaded in my theo02 case?

best wishes
Andreas

whole compile log:
> No such file: </etc/version>
> Found config file
> </afs/ipp-garching.mpg.de/home/a/aschreib/pencil-code/config/hosts/RZG/theo02.bc.rzg.mpg.de.conf>
> Running make '-j' 'FFLAGS_DOUBLE=-r8'
> 'CFLAGS_DOUBLE=-DDOUBLE_PRECISION' 'LD_MPI=' 'CFLAGS_FFTW3='
> 'FFLAGS_FFTW3=' 'LD_FFTW3=' 'CFLAGS_FFTW2=' 'FFLAGS_FFTW2='
> 'LD_FFTW2=' 'FC=mpif90' 'F77=mpif77' 'FFLAGS=-implicitnone -ipo -O3
> -mcmodel=large -shared-intel' 'FSTD_95=-std95' 'FSTD_2003=-std03'
> 'CC=mpicc' 'CFLAGS=-DFUNDERSC=1 -ipo' 'AR=xiar'
> precision-check: No such file src/.current-precision -- creating new
> one at ./src/scripts/precision-check line 36.

> gfortran: unrecognized option '-shared-intel'
> gfortran: unrecognized option '-r8'
> gfortran: unrecognized option '-shared-intel'
> gfortran: unrecognized option '-r8'

> f951: error: unrecognized command line option "-std95"

> cc1: error: unrecognized command line option "-ipo"

> make[2]: *** [cparam.o] Error 1

> make[2]: *** Waiting for unfinished jobs....
> make[2]: *** [syscalls_ansi.o] Error 1

> cc1: error: unrecognized command line option "-ipo"
> make[2]: *** [nogsl.o] Error 1

> make[1]: *** [code] Error 2

> make: *** [default] Error 2
> Make failed: <>

======================================================================
Andreas Schreiber Voice: +49 6221 528 335
Max-Planck-Institut
für Astronomie mailto:andreass...@mpia.de
Königstuhl 17
D-69117 Heidelberg, Germany
======================================================================

[Andreas Schreiber]

Somehow the conf files in my old PC dir and the newst rep are different!?!?

old:
> #
> # THEO cluster at Rechenzentrum Garching, with Intel compiler.
> #
> # One must have
> # module load intel
> # and
> # setenv I_MPI_ROOT /afs/@cell/common/soft/intel/impi/4.1.0
> # in .cshrc or .bashrc.
> #
> %include compilers/ifort
>
> %section environment
> I_MPI_ROOT = /afs/@cell/common/soft/intel/impi/4.1.0
> %endsection environment
>
> %section Makefile
> FC = /afs/@cell/common/soft/intel/impi/4.1.0/bin64/mpiifort
> FFLAGS = -O3 -I/afs/@cell/common/soft/intel/impi/4.1.0/include64
> -mcmodel=large -shared-intel
> %endsection Makefile
>
> %section runtime
> nprocpernode = 1
> local_disc = 0
> one_local_disc = 0
> mpirun = /afs/@cell/common/soft/intel/impi/4.1.0/bin64/mpirun
> nodelist=`cat $TMPDIR/machines`
> SSH='ssh -x'
> SCRATCH_DIR=/var/tmp/$USER
> %endsection runtime

new:
> # THEO cluster at Rechenzentrum Garching, with Intel compiler.
> #
> # One must execute first in the shell:
> # $ module load intel
>
> %include compilers/Intel_MPI
>
> %section Makefile
> FFLAGS += -O3 -mcmodel=large -shared-intel
> %endsection Makefile
>
> %section runtime
> nprocpernode = 1
> local_disc = 0
> one_local_disc = 0
> mpiexec = mpirun
> %endsection runtime

>


======================================================================
Andreas Schreiber Voice: +49 6221 528 335
Max-Planck-Institut
für Astronomie mailto:andreass...@mpia.de
Königstuhl 17
D-69117 Heidelberg, Germany
======================================================================

[Andreas Schreiber]

also the config/compilers/ifort is missing?
I remember someone doning a compiler cleanup. should I thus change my
config files for theo02? to what?

[Philippe Bourdin]

Hello Andreas,

yes I have cleaned this up a bit. Seems RZG has changed their setup for
mpif90, which before should have been ifort, but now is gfortran.
I will try a small change, please update and retry...

[Andreas Schreiber]

Thank you very much! But error now changed to:

> ...
> mpiifort -implicitnone -ipo -O3 -mcmodel=large -shared-intel -r8
> -std95 -o nolsode_for_chemistry.o -c nolsode_for_chemistry.f90
> mpiifort -implicitnone -ipo -O3 -mcmodel=large -shared-intel -r8
> -std95 -o noNSCBC.o -c noNSCBC.f90
> mpiifort -implicitnone -ipo -O3 -mcmodel=large -shared-intel -r8
> -std95 -o noentropy.o -c noentropy.f90
> mpiifort -implicitnone -ipo -O3 -mcmodel=large -shared-intel -r8
> -std95 -o shock_highorder.o -c shock_highorder.f90
> mpiifort -implicitnone -ipo -O3 -mcmodel=large -shared-intel -r8
> -std95 -o hydro.o -c hydro.f90
> xiar r magnetic.a nomagnetic.o magnetic/nomeanfield.o
> magnetic/nomeanfield_demfdt.o
> make[2]: xiar: Command not found
> make[2]: *** [magnetic.a] Error 127

> make[2]: *** Waiting for unfinished jobs....

> make[2]: Leaving directory
> `/afs/ipp-garching.mpg.de/home/a/aschreib/tmp/test/src'

> make[1]: *** [code] Error 2

> make[1]: Leaving directory
> `/afs/ipp-garching.mpg.de/home/a/aschreib/tmp/test/src'

> make: *** [default] Error 2
> Make failed: <>

======================================================================
Andreas Schreiber Voice: +49 6221 528 335
Max-Planck-Institut
für Astronomie mailto:andreass...@mpia.de
Königstuhl 17
D-69117 Heidelberg, Germany
======================================================================

[Philippe Bourdin]

Hello Andreas,

> Thank you very much! But error now changed to:

> > mpiifort -implicitnone -ipo -O3 -mcmodel=large -shared-intel -r8
> > -std95 -o nolsode_for_chemistry.o -c nolsode_for_chemistry.f90

Ok, ifort/icc now works fine.

> > xiar r magnetic.a nomagnetic.o magnetic/nomeanfield.o
> > magnetic/nomeanfield_demfdt.o
> > make[2]: xiar: Command not found

Really strange! Because 'xiar' is part of the Intel compiler package!
Can you please ask the RZG technical staff, what the Intel "ar" is now
called in their setup?

Alternatively, which version of the Intel package do you have?
Can you please send me the output of:
$ module load intel
$ mpiifort --version
$ mpiicc --version
$ xiar --version
$ ar --version

One more alternative, can you try to compile with only standard options:
$ make cleann
$ pc_setupsrc
$ pc_build -f Intel_MPI-mpiifort

(The -f option overrides your host.conf file and uses the given one.)

[Andreas Schreiber]

shall I just reset everything to the old but still working setup?

>>> xiar r magnetic.a nomagnetic.o magnetic/nomeanfield.o
>>> magnetic/nomeanfield_demfdt.o
>>> make[2]: xiar: Command not found
> Really strange! Because 'xiar' is part of the Intel compiler package!
> Can you please ask the RZG technical staff, what the Intel "ar" is now
> called in their setup?
>
> Alternatively, which version of the Intel package do you have?
> Can you please send me the output of:
> $ module load intel
> $ mpiifort --version
> $ mpiicc --version
> $ xiar --version
> $ ar --version

aschreib@theo02:~/tmp/test> module load intel
aschreib@theo02:~/tmp/test> mpiifort --version
ifort (IFORT) 12.1.6 20120928
Copyright (C) 1985-2012 Intel Corporation. All rights reserved.

aschreib@theo02:~/tmp/test> mpiicc --version
icc (ICC) 12.1.6 20120928
Copyright (C) 1985-2012 Intel Corporation. All rights reserved.

aschreib@theo02:~/tmp/test> xiar --version
-bash: xiar: command not found
aschreib@theo02:~/tmp/test> ar --version
GNU ar (GNU Binutils; SUSE Linux Enterprise 11) 2.23.1
Copyright 2012 Free Software Foundation, Inc.
This program is free software; you may redistribute it under the terms of
the GNU General Public License version 3 or (at your option) any later
version.
This program has absolutely no warranty.

> One more alternative, can you try to compile with only standard options:
> $ make cleann
> $ pc_setupsrc
> $ pc_build -f Intel_MPI-mpiifort

Same error msg: xiar: command not found

Could xiar be a part of an other package? Lets wait what RZG Support
tells me...

best
Andy

[Philippe Bourdin]

Hello Andreas,

> shall I just reset everything to the old but still working setup?

Better not, because we need a working solution that other (future) users
can also use on RZG clusters.

> aschreib@theo02:~/tmp/test> mpiifort --version
> ifort (IFORT) 12.1.6 20120928

> aschreib@theo02:~/tmp/test> mpiicc --version
> icc (ICC) 12.1.6 20120928

> aschreib@theo02:~/tmp/test> xiar --version
> -bash: xiar: command not found
> aschreib@theo02:~/tmp/test> ar --version
> GNU ar (GNU Binutils; SUSE Linux Enterprise 11) 2.23.1

This looks like a mis-configuration to me...
The Intel-"ar" must also be somewhere, it is part of the Intel compiler:
"xiar (archive tool for IPO objects) should be present in the same
directory as ifort and icc, a mandatory part of the standard install."
source: https://software.intel.com/de-de/forums/topic/269744

> Lets wait what RZG Support tells me...

Yes, I fear we have to wait for that answer...

If you urgently need to do something, inbetween you may use:
$ pc_build -f GNU-GCC_MPI
until we have the answer, which might hopefully be already tomorrow.

[Andreas Schreiber]

solved:

> Lieber Herr Schreiber,
>
> auf dem Theo-Cluster gibt es bei der Default-Version des Intel-Compilers
> tatsaechlich kein "xiar". Aber versuchen Sie doch mal ein
>
> "module swap intel/12.1 intel/14.0"
>
> Schvne Gr|_e / Sch=F6ne Gr=FC=DFe / Schoene Gruesse
> Werner Nagel

======================================================================
Andreas Schreiber Voice: +49 6221 528 335
Max-Planck-Institut
für Astronomie mailto:andreass...@mpia.de
Königstuhl 17
D-69117 Heidelberg, Germany
======================================================================