Here you will find a short guideline about how to compile in 'Dalma'
Before starting any compilation, user have to make some decisions first, like
1.- decide the compiler to use
2.- select the requiered dependencies
3.- if it is a parallel application, decide which MPI implementation to use
4.- choose the best compilation flags
5.- compile the code
For the compiler, we have available GCC compiler suite version 4.9.3 and INTEL compiler version 16.1 Both are easily available through the modules system:
[hpc@~]$ module load gcc
Loading module 'gcc/4.9.3'
[hpc@~]$ module load intel
Loading module 'intel/16.1'
Dependencies are application specific and users should know what they need. They just could inquire the modules system to check if the required dependency is available either as 'default' or inside 'all'
To see if a specific dependecy is available, users may need execute module avail command. For instance, to inquire the modules if your dependency (let's say fftw3) is available as default, you may need to
[hpc@~]$ module purge
[hpc@~]$ module avail fftw3
------------------------------------ /share/apps/NYUAD/modules/SOFTWARE -----------------------------------------
fftw3/3.3.4
[hpc@~]$ module show fftw3
-------------------------------------------------------------------
/share/apps/NYUAD/modules/SOFTWARE/fftw3/3.3.4:
module-whatis 'fftw3' HERE GOES THE DESCRIPTION
setenv FFTW3_ROOT /share/apps/NYUAD/fftw3/avx2/3.3.4
setenv FFTW3_INC /share/apps/NYUAD/fftw3/avx2/3.3.4/include
setenv FFTW3_LIB /share/apps/NYUAD/fftw3/avx2/3.3.4/lib
setenv FFTW3_BIN /share/apps/NYUAD/fftw3/avx2/3.3.4/bin
setenv FFTW3_MAN /share/apps/NYUAD/fftw3/avx2/3.3.4/share/man
prepend-path CPATH /share/apps/NYUAD/fftw3/avx2/3.3.4/include
prepend-path CPLUS_INCLUDE_PATH /share/apps/NYUAD/fftw3/avx2/3.3.4/include
prepend-path C_INCLUDE_PATH /share/apps/NYUAD/fftw3/avx2/3.3.4/include
prepend-path OBJC_INCLUDE_PATH /share/apps/NYUAD/fftw3/avx2/3.3.4/include
prepend-path LD_LIBRARY64_PATH /share/apps/NYUAD/fftw3/avx2/3.3.4/lib
prepend-path LD_LIBRARY_PATH /share/apps/NYUAD/fftw3/avx2/3.3.4/lib
prepend-path LIBRARY_PATH /share/apps/NYUAD/fftw3/avx2/3.3.4/lib
prepend-path PKG_CONFIG_PATH /share/apps/NYUAD/fftw3/avx2/3.3.4/lib/pkgconfig
prepend-path PATH /share/apps/NYUAD/fftw3/avx2/3.3.4/bin
prepend-path MANPATH /share/apps/NYUAD/fftw3/avx2/3.3.4/share/man
-------------------------------------------------------------------
If the version that is offered as 'default' doesn't match your needs, you ay want to look into 'all' and load the specific need
[hpc@~]$ module load all
Loading module 'all'
[hpc@~]$ module avail fftw3
--------------------------------------------- /share/apps/NYUAD/modules/ALL -----------------------------------------------
fftw3/avx/3.3.4 fftw3/gcc_4.9.3/mvapich2_2.2b/sse4.2/3.3.4 fftw3/intel_16.1/mvapich2_2.2b/avx/3.3.4
fftw3/avx2/3.3.4 fftw3/gcc_4.9.3/openmpi_1.10.2/avx/3.3.4 fftw3/intel_16.1/mvapich2_2.2b/avx2/3.3.4
fftw3/gcc_4.9.3/mvapich2_2.2b/avx/3.3.4 fftw3/gcc_4.9.3/openmpi_1.10.2/avx2/3.3.4 fftw3/intel_16.1/mvapich2_2.2b/sse4.2/3.3.4
fftw3/gcc_4.9.3/mvapich2_2.2b/avx2/3.3.4 fftw3/gcc_4.9.3/openmpi_1.10.2/sse4.2/3.3.4 fftw3/intel_16.1/openmpi_1.10.2/avx/3.3.4
There are two MPI implementations available: OpenMPI, version 1.10.2 and MVAPICH2, version 2.2b
By default, both MPI implementations are available for the GCC compiler suite. So if you load the module openmpi or mvapich2 outside 'all', the versions that will be loaded are linked to GCC.
[hpc@~]$ module load openmpi
Loading module 'openmpi/1.10.2'
the latter will load openmpi to be used with GCC compiler
In case you need the ones for INTEL compiler, then you will have to load 'all' and then select the correct module
[hpc@~]$ module load all
Loading module 'all'
[hpc@~]$ module avail openmpi
---------------------------------------------- /share/apps/NYUAD/modules/ALL ----------------------------------------------
openmpi/gcc_4.9.3/1.10.2 openmpi/intel_16.1/1.10.2
[hpc@~]$ module load openmpi/intel_16.1/1.10.2
Loading module 'openmpi/intel_16.1/1.10.2'
The recommended compiler flags to use with the GCC compiler suite are
-O3 -mavx2 -mfma -march=haswell -funroll-loops
For intel compiler, the flags become
-O3 -mieee-fp -xCORE-AVX2
Those options are valid for the processor architecture used in 'Dalma'. For a different machine you will have to use a different set of flags.
NYUAD HPC Apps Team:
--------------------
- Benoit Marchand
- Guowei He
- Jorge Naranjo
Please refer to the online documentation available here