139. compiling nwchem with custom ATLAS on debian

I'm not yet sure, but there seems to be something weird with this build. Use the openblas build here preferentially:  http://verahill.blogspot.com.au/2012/05/nwchem-with-openblas.html

i.e. do not use this build without testing it against builds without external blas/atlas and against other libraries.

I am not an expert

The atlas build is pretty much like what was described here:
http://verahill.blogspot.com.au/2012/05/compile-atlas-blas-on-debian-testing.html

---

Build atlas
Check you cpu governor:

cat /sys/devices/system/cpu/cpu1/cpufreq/scaling_governor 

ondemand

Next, set it to performance for all cores (example for three-core AMD)
sudo cpufreq-set -g performance
sudo cp /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor /sys/devices/system/cpu/cpu1/cpufreq/scaling_governor
sudo cp /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor /sys/devices/system/cpu/cpu2/cpufreq/scaling_governor

Prepare the install directory:
sudo mkdir /opt/ATLAS
sudo chown ${USER} /opt/ATLAS

Prepare the compile directory
mkdir ~/tmp
cd ~/tmp
 wget http://downloads.sourceforge.net/project/math-atlas/Developer%20%28unstable%29/3.9.72/atlas3.9.72.tar.bz2
tar xvf atlas3.9.72.tar.bz2
cd ATLAS/
mkdir build/
vim Make.top 
change -V on line 6 to -v

cd build
.././configure --prefix=/opt/ATLAS

There's any level of detail in what you can pass to configure. -A AMD64 would be one option. Look at the bottom for more info.

make
make install
cp lib/lib* /opt/ATLAS/lib

Once it's all done, set the cpu governor back if you so desire.
sudo cpufreq-set -g ondemand
sudo cp /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor /sys/devices/system/cpu/cpu1/cpufreq/scaling_governor
sudo cp /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor /sys/devices/system/cpu/cpu2/cpufreq/scaling_governor

---

nwchem

For python support you need to edit src/config/makefile.h (see http://verahill.blogspot.com.au/2012/04/adding-python-support-to-nwchem-under.html)

export LARGE_FILES=TRUE

export TCGRSH=/usr/bin/ssh

export NWCHEM_TOP=/opt/nwchem/nwchem-6.0

export NWCHEM_TARGET=LINUX64

export NWCHEM_MODULES="all python"

export PYTHONHOME=/usr

export PYTHONVERSION=2.7

export BLASOPT="-L/opt/ATLAS/lib -latlas -lblas -llapack"

export USE_MPI=y

export USE_MPIF=y

export MPI_LOC=/usr/lib/openmpi/lib

export MPI_INCLUDE=/usr/lib/openmpi/include

export LIBRARY_PATH=$LIBRARY_PATH:/usr/lib/openmpi/lib

export LIBMPI="-lmpi -lopen-rte -lopen-pal -ldl -lmpi_f77 -lpthread"

cd $NWCHEM_TOP/src

make clean

make nwchem_config

make FC=gfortran

Note: don't use HAS_BLAS=y. It gave me no end of grief.

Also, IF you binary executes without a hitch at this point it means that you're libs are not being loaded.  You need to put

export LD_LIBRARY_PATH=/opt/ATLAS/lib:$LD_LIBRARY_PATH

in your ~/.bashrc

Building takes a while, but should go ok. Make sure you have python2.7-dev installed, and if you're using a different python version, make sure to change the options above.

---

Here's also an example ~/.nwchemrc file for no particular reason at all -- if you've got ecce installed you might want to make sure yours is correct.

nwchem_basis_library /opt/nwchem/nwchem-6.0/src/basis/libraries/
nwchem_nwpw_library /opt/nwchem/nwchem-6.0/src/nwps/libraryps/
ffield amber
amber_1 /opt/nwchem/nwchem-6.0/src/data/amber_s/
amber_2 /opt/nwchem/nwchem-6.0/src/data/amber_x/
spce /opt/nwchem/nwchem-6.0/src/data/solvents/spce.rst
charmm_s /opt/nwchem/nwchem-6.0/src/data/charmm_s/
charmm_x /opt/nwchem/nwchem-6.0/src/data/charmm_x/

---

You can get a list over compile options for ATLAS using
make xprint_enums ; ./xprint_enums

after you've done .././configure

See http://sourceforge.net/tracker/index.php?func=detail&aid=3021404&group_id=23725&atid=379483

138. Compile ATLAS blas on debian testing

I first tried the latest stable and gave up due to "ERROR 639 DURING CACHE EDGE DETECTION!!" errors. Hence, I used unstable.

Start here

mkdir /opt/ATLAS
chown ${USER}  /opt/ATLAS

mkdir ~/tmp
cd ~/tmp
 wget http://downloads.sourceforge.net/project/math-atlas/Developer%20%28unstable%29/3.9.72/atlas3.9.72.tar.bz2

tar xvf atlas3.9.72.tar.bz2

cd ATLAS/


Edit ATLAS/Make.top 
change the V on line 6 to lowercase i.e. from

- $(ICC) -V 2>&1  >> bin/INSTALL_LOG/ERROR.LOG

to

- $(ICC) -v 2>&1  >> bin/INSTALL_LOG/ERROR.LOG

mkdir build/
cd build/
sudo apt-get install cpufrequtils

Since I don't like computers overheating I normally throttle my cpu. To unset throttling on a AMD64 (phenom II) six core system, before compile:
/usr/bin/cpufreq-selector -g performance
sudo cp /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor /sys/devices/system/cpu/cpu1/cpufreq/scaling_governor

sudo cp /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor /sys/devices/system/cpu/cpu2/cpufreq/scaling_governor

sudo cp /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor /sys/devices/system/cpu/cpu3/cpufreq/scaling_governor

sudo cp /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor /sys/devices/system/cpu/cpu4/cpufreq/scaling_governor

sudo cp /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor /sys/devices/system/cpu/cpu5/cpufreq/scaling_governor

Alternatively, you can disable throttle checking using the -Si cputhrchk 0 configure switch. If you don't, and throttling is on, configure will fail.

OK, continuing with the build, in your build/ directory:

.././configure --prefix=/opt/ATLAS -A AMD64

Before you continue, do
make xprint_enums ; ./xprint_enums
and study the output -- you might find a better ARCH fit than e.g. AMD64.  If so, run the configure command above again. There is a lot of other options which you can configure as well. Obviously, nothing prevents you from experimenting and re-compiling if something goes wrong. You can also run without any -A switch at all.

Continuing:
make
make install
cp lib/lib* /opt/ATLAS/lib

Your libs should be in /opt/ATLAS/lib -- when you compiled something to include the new libs, use
"-L/opt/ATLAS/lib"

137. Setting up Gaussian g09 on debian -- precompiled binaries

Most people would use a precompiled and pre-installed copy of gaussian on their local computational grid. If you do, however, purchase your own copy, or work at an institution with a site license, then you can install gaussian on your local beowulf cluster (or over-powered desktop).

And it's easy. I've presumed that your username is verahill, and your group is verahill.

You need csh

sudo apt-get install csh

Next set up a destination directory

mkdir /opt/gaussian
sudo chown verahill:verahill /opt/gaussian

copy the gaussian binaries to /opt/gaussian/g09 so that you have

/opt/gaussian/g09
|-- basis
|-- bsd
`-- tests
    |-- com
    |-- ia64
    `-- newz

Gaussian wants almost everything to be executable
chmod +x /opt/gaussian/g09/*
chmod +x /opt/gaussian/g09/bsd/*
cd /opt/gaussian/g09
csh bsd/install

Next, edit your ~/.bashrc and add (anywhere)

export g09root=/opt/gaussian
export GAUSS_SCRDIR=/scratch
. /opt/gaussian/g09/bsd/g09.profile
export PATH=$PATH:/opt/gaussian/g09/bsd:/opt/gaussian/g09/local

Source your bashrc to make changes take effect immediately:

source ~/.bashrc

You need to edit the bashrc of anyone wanting to use gaussian

And you're done!