Benchmarking: Gromacs-4.6

Requirements

cmake 2.8 (for build)
CUDA 5.0

Download

wget ftp://ftp.gromacs.org/pub/gromacs/gromacs-4.6.tar.gz

Build

module load cmake/2.8.10.2
module load cuda/5.0

cmake ../gromacs-4.6 -DGMX_GPU=ON -DGMX_BUILD_OWN_FFTW=ON \
     -DCUDA_TOOLKIT_ROOT_DIR=/shared/apps/cuda/cuda-5.0 \
     -DLIB_INSTALL_DIR=/shared/apps/cuda/gromacs/4.6/lib \
     -DCMAKE_INSTALL_PREFIX=/shared/apps/cuda/gromacs/4.6

make -j 24

Modulefiles

File location: /shared/modulefiles
File name: gpu/gromacs/4.6

#%Module 1.0
#
prereq  cuda/5.0
system "source /shared/apps/cuda/gromacs/4.6/bin/GMXRC.bash"
prepend-path  PATH      /shared/apps/cuda/gromacs/4.6/bin
prepend-path  LD_LIBRARY_PATH /shared/apps/cuda/gromacs/4.6/lib

Run

module load gpu/gromacs/4.6
mdrun -s <input file>

Build Errors

Intel Compilers

Error

/home/boston/gromacs-4.6.3/include/gmx_math_x86_avx_128_fma_double.h(674): error: a value of type "int" cannot be assigned to an entity of type "__m128d"
      PolyCP0  = _mm_macc_pd(PolyCP1, w, PolyCP0);
               ^

compilation aborted for /home/boston/gromacs-4.6.3/src/gmxlib/bondfree.c (code 4)
make[2]: *** [src/gmxlib/CMakeFiles/gmx.dir/bondfree.c.o] Error 4
make[1]: *** [src/gmxlib/CMakeFiles/gmx.dir/all] Error 2
make: *** [all] Error 2

Fix

You can fix the error by updating the CMakeCache.txt after you cmake Gromacs.

Update GMX_CPU_ACCELERATION:STRING=AVX_128_FMA to

//Accelerated CPU kernels. Pick one of: None, SSE2, SSE4.1, AVX_128_FMA, // AVX_256, IBM_QPX, Sparc64_HPC_ACE GMX_CPU_ACCELERATION:STRING=None

After which you can proceed with $ make $ make install

It’s probably because Intel compilers do not understand the intrinsic types of AMD (AVX_128_FMA as identified by Gromacs) – meaning: the Intel compiler cannot produce optimized code for AMD.