Compiling on mike2 fails b/c C compiler cannot compile code

[Former user]

Trying to compile the future ET_Mayer on Mike2 (HPC CCT LSU) I get failures during the CST stage:

checking whether the C compiler (/home/compilers/Intel/parallel_studio_xe_2018.0/compilers_and_libraries_2018/linux/bin/intel64/icc -g -xHOST -align -std=gnu99 -U__STRICT_ANSI__ -ansi_alias -gcc-name=/usr/local/compilers/gcc/4.9.2/bin/gcc -Wl,-rpath,/usr/local/compilers/gcc/4.9.2/lib64) works... no
configure: error: installation or configuration problem: C compiler cannot create executables (see configs/sim/config-data/config.log for details).

and config.log contains:

This file contains any messages produced by compilers while
running configure, to aid debugging if configure makes a mistake.

configure:912: checking host system type
configure:938: checking for mawk
configure:938: checking for gawk
configure:1007: checking whether make sets ${MAKE}
configure:1072: checking whether the C compiler (/home/compilers/Intel/parallel_studio_xe_2018.0/compilers_and_libraries_2018/linux/bin/intel64/icc -g -xHOST -align -std=gnu99 -U__STRICT_ANSI__ -ansi_alias -gcc-name=/usr/local/compilers/gcc/4.9.2/bin/gcc -Wl,-rpath,/usr/local/compilers/gcc/4.9.2/lib64) works
configure:1089: /home/compilers/Intel/parallel_studio_xe_2018.0/compilers_and_libraries_2018/linux/bin/intel64/icc -o conftest -g -xHOST -align -std=gnu99 -U__STRICT_ANSI__ -ansi_alias -gcc-name=/usr/local/compilers/gcc/4.9.2/bin/gcc -DMPICH_IGNORE_CXX_SEEK -Wl,-rpath,/usr/local/compilers/gcc/4.9.2/lib64 conftest.c  1>&5
/tmp/icc5Wh1JP.o: In function `main':
conftest.c:(.text+0x1a): undefined reference to `__intel_new_feature_proc_init'
configure: failed program was:

#line 1084 "configure"
#include "confdefs.h"

main(){return(0);} int PilotMain(){return(0);}

[Roland Haas]

Log files from failed attempt.

[Roland Haas]

http://einsteintoolkit.org/testsuite_results/index.php lists shelob as a supported (and tested) machine for ET_Proca and my recollection is that mike and shelob are the same system by now. Is this correct?

[Peter Diener]

I can not reproduce this even with an almost empty environment (disabling loading the modules I would normally load). My build starts normally and thorns are being compiled as I write this.

@Roland Haas What does your environment look like?

‌

[Roland Haas]

@Peter Diener It should not be anything special since the first thing I did was rm -rf * in my $HOME and I really have not done anything on mike for a long time. All the dotfiles look sane to me.

My .soft file reads just @default and both .bashrc and .bash_profile look like vanilla versions to me. I am not quite sure how to make softenv show me a list of loaded modules. I could provide the output of env but that may be too much. I have made my $HOME on mike2 readable if you would like to poke around.

I have tried both running locally on mike as well as using simfactory’s remote option:

simfactory/bin/sim --remote mike build

and both give the same result.
The only non-standard setting for mike in simfactory that I have is setting an allocation.

[Steven R. Brandt]

@Roland Haas We switched from softenv to modules some time ago. I believe the defaults changed at that time. See section 3.2.0 to upgrade: http://www.hpc.lsu.edu/docs/guides.php?system=SuperMike2

[Peter Diener]

@Roland Haas Oh, I see. They switched from softenv to modules a while back. In section 3.2.0 of the user guide (http://www.hpc.lsu.edu/docs/guides.php?system=SuperMike2) it is explained how to switch over. Can you try that and see if that helps?

[Roland Haas]

Let me try this. Currently when I type module it reports -bash: module: command not found….

Alright, running the command indicated in the help page seems to have fixed the issue, or at least simfactory is now compiling. I will close this ticket once it is done compiling.

So this seems to have been a case of bit-rot.

[Peter Diener]

Later on during compilation there is an issue with MPI, that I’m looking into now.

[Peter Diener]

@Roland Haas I got compilation and linking to work by specifying modules to load in the mike.ini machine file even with an environment with no modules loaded. I also tested that the testsuite runs. Do you wanna test if this works for you or should we just close the ticket?

[Roland Haas]

Fixed in git hash 486e8e6 "Make things work with an empty environment." of simfactory2.

[Roland Haas]

@Peter Diener Fine with me, I will re-open the ticket should anything fails once more.