[gmx-users] Configuration bug preventing GROMACS 5.0 build with Intel compiler?

Mark Abraham mark.j.abraham at gmail.com
Wed Sep 3 22:59:56 CEST 2014 

Hi,

I'm not aware of an intrinsic problem with GROMACS and icc 14 - that icc
version + CUDA 5.5 + OpenMPI compiles and passes tests (with a million
warnings presumably from the wrong CUDA version for icc, which normally
seems to be fine in practice). IntelMPI is not often tested with GROMACS,
so I would experiment with taking it and CUDA out of the mix and observing
how your mileage varies. Frankly, gcc will do just as well or slightly
better...

You should be able to hack out the erroneous detection by using cmake
-Disfinite_compile_ok=0
etc. but I would guess that the fault here is in some wrapper compiler not
propagating the inner failure correctly - all CMake can do is take the
return value from the compiler.

Mark


On Wed, Sep 3, 2014 at 7:00 PM, Paul Caheny <Paul.Caheny at uk.fujitsu.com>
wrote:

> Hi All,
>
> I'm attempting to compile GROMACS 5.0 (as downloaded from the tarball
> available here: http://www.gromacs.org/Downloads) with Intel Compiler
> version 14 and Intel MPI version 5 without success. I'm also compiling the
> CUDA functionality (with CUDA 6.5) but CUDA/MPI functionality doesn't seem
> to be related to the compilation problem I'm seeing. I'm using CMAKE 3.0.1
> which I compiled myself from source with the same Intel 14 compiler.
>
> Here is the version info:
>
> icc --version
> icc (ICC) 14.0.3 20140422
> Copyright (C) 1985-2014 Intel Corporation.  All rights reserved.
>
> icpc --version
> icpc (ICC) 14.0.3 20140422
> Copyright (C) 1985-2014 Intel Corporation.  All rights reserved.
>
> mpirun --version
> Intel(R) MPI Library for Linux* OS, Version 5.0 Build 20140605
> Copyright (C) 2003-2014, Intel Corporation. All rights reserved.
>
> I'm compiling like so:
>
> module load compiler/intel/14.0/3.174
> module load mpi/intel/5.0/0.028
>
> tar -zxvf gromacs-5.0.tar.gz
> cd gromacs-5.0
> mkdir build
> cd build
>
> ../../../CMAKE/cmake-3.0.1_intel14.0/bin/cmake .. \
> -DCMAKE_C_COMPILER=mpiicc -DCMAKE_CXX_COMPILER=mpiicpc \
> -DGMX_MPI=on -DGMX_GPU=on
> -DCUDA_TOOLKIT_ROOT_DIR=/app/libraries/cuda/6.5/sb \
> -DCUDA_HOST_COMPILER=/app/compilers/intel/
> 14.0.3.174/composer_xe_2013_sp1.3.174/bin/intel64/icpc \
> -DCUDA_VERBOSE_BUILD=ON -DCUDA_VER=6.5
> -DCUDA_INSTALL_PATH=/app/libraries/cuda/6.5/sb \
> -DCMAKE_INSTALL_PREFIX=../../../G2/install \
> -DGMX_FFT_LIBRARY=mkl \
> -DCMAKE_EXPORT_COMPILE_COMMANDS=ON -DCMAKE_VERBOSE_MAKEFILE=ON
> make
> make install
>
> During install I see the following:
>
>   file INSTALL cannot find
>   "/home/paul.caheny/ISV/FY14Q2/GROMACS/G2/gromacs-5.0/build/bin/gmx_mpi".
>
> Going backwards through the compilation I first see multiple instances of
> undefined references to _finite and _aligned_free like this one during link:
>
> /app/libraries/impi/5.0.0.028/intel64/bin/mpiicpc    -mavx   -w3 -wd111
> -wd177 -wd181 -wd193 -wd271 -wd304 -wd383 -wd424 -wd444 -wd522 -wd593
> -wd869 -wd981 -wd1418 -wd1419 -wd1572 -wd1599 -wd2259 -wd2415 -wd2547
> -wd2557 -wd3280 -wd3346 -wd1782   -ip -funroll-all-loops -alias-const
> -ansi-alias   -O3 -DNDEBUG  -mkl=sequential
> CMakeFiles/template.dir/template.cpp.o  -o ../../bin/template -rdynamic
> ../../lib/libgromacs_mpi.so.0.0.0 -openmp ../../lib/libgpu_utils.a
> ../../lib/libnbnxn_cuda.a ../../lib/libcuda_tools.a
> /app/libraries/cuda/6.5/sb/lib64/libcudart.so -ldl -lrt -lm -lz
> -Wl,-rpath,/home/paul.caheny/ISV/FY14Q2/GROMACS/G2/gromacs-5.0/build/lib:/app/libraries/cuda/6.5/sb/lib64
> ../../lib/libgromacs_mpi.so.0.0.0: undefined reference to `_finite'
> ../../lib/libgromacs_mpi.so.0.0.0: undefined reference to `_aligned_free'
>
> ...and earlier during compilation:
>
> /home/paul.caheny/ISV/FY14Q2/GROMACS/G2/gromacs-5.0/src/gromacs/math/utilities.c(731):
> warning #266: function "_finite" declared implicitly
>       returnval = _finite(x);
>
> /home/paul.caheny/ISV/FY14Q2/GROMACS/G2/gromacs-5.0/src/gromacs/utility/smalloc.c(359):
> warning #266: function "_aligned_free" declared implicitly
>           _aligned_free(free);
>
> This seems to be related to the CMAKE configuration successfully finding
> functions that don't exist on my platform, in particular from:
>
> -- Checking for isfinite
> -- Performing Test isfinite_compile_ok
> -- Performing Test isfinite_compile_ok - Success
> -- Checking for isfinite - yes
> -- Checking for _isfinite
> -- Performing Test _isfinite_compile_ok
> -- Performing Test _isfinite_compile_ok - Success
> -- Checking for _isfinite - yes
> -- Checking for _finite
> -- Performing Test _finite_compile_ok
> -- Performing Test _finite_compile_ok - Success
> -- Checking for _finite - yes
>
> -- Looking for _aligned_malloc
> -- Looking for _aligned_malloc - found
>
> According to build/CMakeFiles/CMakeOutput.log and CMakeError.log all the
> above checks pass, yet in CMakeOutput.log all except the first check
> (isfinite) are logged as having undefined references, yet the check still
> reports success. This is the just the output for _finite_compile_ok but the
> _isfinite_compile_ok and _aligned_malloc follow a similar pattern:
>
> Performing C SOURCE FILE Test _finite_compile_ok succeded with the
> following output:
> Change Dir:
> /home/paul.caheny/ISV/FY14Q2/GROMACS/G2/gromacs-5.0/build/CMakeFiles/CMakeTmp
>
> Run Build Command:/usr/bin/gmake "cmTryCompileExec3947929838/fast"
> /usr/bin/gmake -f CMakeFiles/cmTryCompileExec3947929838.dir/build.make
> CMakeFiles/cmTryCompileExec3947929838.dir/build
> gmake[1]: Entering directory
> `/home/paul.caheny/ISV/FY14Q2/GROMACS/G2/gromacs-5.0/build/CMakeFiles/CMakeTmp'
> /home/paul.caheny/ISV/FY14Q2/GROMACS/CMAKE/cmake-2.8.12.2_intel14.0/bin/cmake
> -E cmake_progress_report
> /home/paul.caheny/ISV/FY14Q2/GROMACS/G2/gromacs-5.0/build/CMakeFiles/CMakeTmp/CMakeFiles
> 1
> Building C object CMakeFiles/cmTryCompileExec3947929838.dir/src.c.o
> /app/libraries/impi/5.0.0.028/intel64/bin/mpiicc   -std=gnu99 -w3 -wd111
> -wd177 -wd181 -wd193 -wd271 -wd304 -wd383 -wd424 -wd444 -wd522 -wd593
> -wd869 -wd981 -wd1418 -wd1419 -wd1572 -wd1599 -wd2259 -wd2415 -wd2547
> -wd2557 -wd3280 -wd3346   -D_finite_compile_ok
> -I/home/paul.caheny/ISV/FY14Q2/GROMACS/G2/gromacs-5.0/build/CMakeFiles/CMakeTmp/float.h
>   -o CMakeFiles/cmTryCompileExec3947929838.dir/src.c.o   -c
> /home/paul.caheny/ISV/FY14Q2/GROMACS/G2/gromacs-5.0/build/CMakeFiles/CMakeTmp/src.c
> /home/paul.caheny/ISV/FY14Q2/GROMACS/G2/gromacs-5.0/build/CMakeFiles/CMakeTmp/src.c(4):
> warning #266: function "_finite" declared implicitly
>     _finite(f);
>     ^
>
> /home/paul.caheny/ISV/FY14Q2/GROMACS/G2/gromacs-5.0/build/CMakeFiles/CMakeTmp/src.c(4):
> warning #592: variable "f" is used before its value is set
>     _finite(f);
>             ^
>
> Linking C executable cmTryCompileExec3947929838
> /home/paul.caheny/ISV/FY14Q2/GROMACS/CMAKE/cmake-2.8.12.2_intel14.0/bin/cmake
> -E cmake_link_script CMakeFiles/cmTryCompileExec3947929838.dir/link.txt
> --verbose=1
> /app/libraries/impi/5.0.0.028/intel64/bin/mpiicc  -std=gnu99 -w3 -wd111
> -wd177 -wd181 -wd193 -wd271 -wd304 -wd383 -wd424 -wd444 -wd522 -wd593
> -wd869 -wd981 -wd1418 -wd1419 -wd1572 -wd1599 -wd2259 -wd2415 -wd2547
> -wd2557 -wd3280 -wd3346   -D_finite_compile_ok
> CMakeFiles/cmTryCompileExec3947929838.dir/src.c.o  -o
> cmTryCompileExec3947929838 -rdynamic
> CMakeFiles/cmTryCompileExec3947929838.dir/src.c.o: In function `main':
> /home/paul.caheny/ISV/FY14Q2/GROMACS/G2/gromacs-5.0/build/CMakeFiles/CMakeTmp/src.c:(.text+0x40):
> undefined reference to `_finite'
> gmake[1]: Leaving directory
> `/home/paul.caheny/ISV/FY14Q2/GROMACS/G2/gromacs-5.0/build/CMakeFiles/CMakeTmp'
>
> Source file was:
> #include <float.h>
> int main(void) {
>   float f;
>   _finite(f);
> }
>
> I've also tried with CMAKE 2.8, same result. Is this something anyone else
> is seeing with Intel compilers? Version 14 is apparently the only Intel
> compiler supported by CUDA 6.5. I also tried with Intel Compiler version
> 12.1 + CUDA 5.5, I saw the same apparently erroneous behaviour during the
> CMAKE configuration but in this case the build ended even earlier for what
> I think is an unrelated compilation problem:
>
> /home/paul.caheny/ISV/FY14Q2/GROMACS/G5_INTEL12.1_CUDA5.5/gromacs-5.0/src/external/boost/boost/config/compiler/intel.hpp(40):
> warning #47: incompatible redefinition of macro "BOOST_COMPILER" (declared
> at line 11 of
> "/home/paul.caheny/ISV/FY14Q2/GROMACS/G5_INTEL12.1_CUDA5.5/gromacs-5.0/src/external/boost/boost/config/compiler/nvcc.hpp")
>   #define BOOST_COMPILER "Intel C++ version "
> BOOST_STRINGIZE(BOOST_INTEL_CXX_VERSION)
>           ^
>
> /home/paul.caheny/ISV/FY14Q2/GROMACS/G5_INTEL12.1_CUDA5.5/gromacs-5.0/src/external/boost/boost/config/suffix.hpp(34):
> warning: #pragma GCC system_header cannot be used in the primary source file
>
> /home/paul.caheny/ISV/FY14Q2/GROMACS/G5_INTEL12.1_CUDA5.5/gromacs-5.0/src/external/boost/boost/config/suffix.hpp(496):
> error: identifier "__int128" is undefined
>
> /home/paul.caheny/ISV/FY14Q2/GROMACS/G5_INTEL12.1_CUDA5.5/gromacs-5.0/src/external/boost/boost/config/suffix.hpp(497):
> error: expected a ";"
>
> 2 errors detected in the compilation of
> "/tmp/tmpxft_0000797f_00000000-12_nbnxn_cuda_data_mgmt.compute_35.cpp1
>
>
> I've tried with GCC 4.8.0/OpenMPI 1.6.4 and it compiles ok and as
> expected. Any ideas whether I'm doing something wrong in the build or
> otherwise whether there's a good workaround for the Intel Compiler 14?
>
>
> Best Regards,
>
> Paul.
>
>
>
> ---
>
> Paul Caheny
> Researcher
> Technical Computing Research
> Fujitsu Laboratories of Europe Ltd (FLE)
> phone: +44 (0) 20 8606 4529
> e-mail: Paul.Caheny at uk.fujitsu.com<mailto:Paul.Caheny at uk.fujitsu.com>
> web: http://www.fujitsu.com/emea/about/fle/<
> http://www.fujitsu.com/emea/about/fle>
>
>
> ______________________________________________________________________
>
>  Fujitsu Laboratories of Europe Limited
>  Hayes Park Central, Hayes End Road, Hayes, Middlesex, UB4 8FE
>  Registered No. 4153469
>
>  This e-mail and any attachments are for the sole use of addressee(s) and
>  may contain information which is privileged and confidential. Unauthorised
>  use or copying for disclosure is strictly prohibited. The fact that this
>  e-mail has been scanned by Trendmicro Interscan does not guarantee that
>  it has not been intercepted or amended nor that it is virus-free.
> --
> Gromacs Users mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> send a mail to gmx-users-request at gromacs.org.
>

More information about the gromacs.org_gmx-users mailing list