12 September 2012

233. Compiling netlib's lapack and blas on Debian Testing (Wheezy)

In addition to specific BLAS/LAPACK libs such as ACML, MKL, and ATLAS netlib provides (what I understand to be) reference versions of BLAS and LAPACK. Presumably these are slower than optimised versions of blas/lapack, but it doesn't hurt being familiar with them.

Here's how to compile those versions.



BLAS
sudo mkdir /opt/netlib
sudo chown $USER /opt/netlib
mkdir /opt/netlib/blas/lib -p
wget http://www.netlib.org/blas/blas.tgz
tar xvf blas.tgz
cd BLAS/

Edit make.inc
OPTS = -O3 -shared -m64 -march=native -fPIC


make all
gfortran -shared -Wl,-soname,libnetblas.so -o libblas.so.1.0.1 *.o -lc
ln -s libblas.so.1.0.1 libnetblas.so
cp lib*blas* /opt/netlib/blas/lib


To see whether everything linked ok:
 ldd libnetblas.so 
        linux-vdso.so.1 =>  (0x00007ffff1bc6000)
        libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00002b42ec030000)
        libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00002b42ec3b8000)
        libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00002b42ec6ce000)
        libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00002b42ec950000)
        libquadmath.so.0 => /usr/lib/x86_64-linux-gnu/libquadmath.so.0 (0x00002b42ecb67000)
        /lib64/ld-linux-x86-64.so.2 (0x00002b42ebaf3000)




LAPACK
(inspired by this and this)
mkdir -p /opt/netlib/lapack
sudo apt-get install cmake-curses-gui
cd ~/tmp
wget http://www.netlib.org/lapack/lapack-3.4.1.tgz
tar xvf lapack-3.4.1.tgz
cd lapack-3.4.1/
mkdir build
cd build
ccmake ../

Hit 'c' to generate a configuration. Navigate with arrow keys and hit enter to change values. Change to the values in red:
 
 BUILD_COMPLEX                   *ON
 BUILD_COMPLEX16                 *ON
 BUILD_DOUBLE                    *ON
 BUILD_SHARED_LIBS               *ON
 BUILD_SINGLE                    *ON
 BUILD_STATIC_LIBS               *ON
 BUILD_TESTING                   *ON
 CMAKE_BUILD_TYPE                *     
 CMAKE_INSTALL_PREFIX            */opt/netlib/lapack
 LAPACKE                         *OFF
 LAPACKE_WITH_TMG                *OFF
 USE_OPTIMIZED_BLAS              *ON
 USE_XBLAS                       *OFF

Then hit 'c' which might give you (change the values in red) -- I got some errors about ACML/eula here, but don't worry about that.

NOTE: this will only work if you already have blas installed in a standard location. If you don't get the BLAS_FOUND etc. then you should hit 'c' again and then 'g'. Next edit your CMakeCache.txt and paste the variables (without line numbers) you find below this section, then do ccmake ../ again and make sure everything looks ok, and generate using 'g'.

 BLAS_FOUND                       TRUE
 BLAS_GENERIC_FOUND               ON
 BLAS_GENERIC_blas_LIBRARY        /opt/netlib/blas/lib/libnetblas.so
 BLAS_LIBRARIES                   /opt/netlib/blas/lib/libnetblas.so
 BLAS_LINKER_FLAGS
 BUILD_COMPLEX                   *ON
 BUILD_COMPLEX16                 *ON
 BUILD_DOUBLE                    *ON
 BUILD_SHARED_LIBS               *OFF
 BUILD_SINGLE                    *ON
 BUILD_STATIC_LIBS               *ON
 BUILD_TESTING                   *ON
 CMAKE_BUILD_TYPE                *     
 CMAKE_INSTALL_PREFIX            */usr/local 
 LAPACKE                         *OFF
 LAPACKE_WITH_TMG                *OFF
 USE_OPTIMIZED_BLAS              *ON
 USE_XBLAS                       *OFF
The hit 'c' again. If there were no issues, hit 'g' which writes the configuration and exits.


make
[100%] Building Fortran object TESTING/EIG/CMakeFiles/xeigtstz.dir/__/__/INSTALL/dsecnd_INT_ETIME.f.o
Linking Fortran executable ../../bin/xeigtstz
[100%] Built target xeigtstz

make install
Install the project...
-- Install configuration: ""
-- Installing: /opt/netlib/lapack/lib/pkgconfig/lapack.pc
-- Installing: /opt/netlib/lapack/lib/cmake/lapack-3.4.1/lapack-config.cmake
-- Installing: /opt/netlib/lapack/lib/cmake/lapack-3.4.1/lapack-config-version.cmake
-- Installing: /opt/netlib/lapack/lib/cmake/lapack-3.4.1/lapack-targets.cmake
-- Installing: /opt/netlib/lapack/lib/cmake/lapack-3.4.1/lapack-targets-noconfig.cmake
-- Installing: /opt/netlib/lapack/lib/liblapack.so
-- Removed runtime path from "/opt/netlib/lapack/lib/liblapack.so"
-- Installing: /opt/netlib/lapack/lib/libtmglib.so
-- Removed runtime path from "/opt/netlib/lapack/lib/libtmglib.so"


tree /opt/netlib/ -d
/opt/netlib/
|-- blas
|   `-- lib
`-- lapack
    `-- lib
        |-- cmake
        |   `-- lapack-3.4.1
        `-- pkgconfig

7 directories

CMakeCache.txt variables:
 16 
 17 BLAS_FOUND:STRING=TRUE
 18 
 19 //Whether not the GENERIC library was found and is usable
 20 BLAS_GENERIC_FOUND:BOOL=TRUE
 21 
 22 //Path to a library.
 23 BLAS_GENERIC_blas_LIBRARY:FILEPATH=/opt/netlib/blas/lib/libnetblas.so
 24 
 25 BLAS_LIBRARIES:PATH=/opt/netlib/blas/lib/libnetblas.so
 26 


Testing the libraries:
I built gromacs against the new libs to make sure they 'worked'

sudo mkdir /opt/gromacs
sudo chown ${USER} /opt/gromacs
cd ~/tmp
wget ftp://ftp.gromacs.org/pub/gromacs/gromacs-4.5.5.tar.gz
tar xvf gromacs-4.5.5.tar.gz
cd gromacs-4.5.5/
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/netlib/blas/lib:/opt/netlib/lapack/lib
export LDFLAGS="-l:/opt/netlib/blas/lib/libnetblas.so -l:/opt/netlib/lapack/lib/liblapack.so"
./configure --disable-mpi --enable-float --with-external-blas --with-external-lapack --program-suffix=_netlib --prefix=/opt/gromacs/gromacs-4.5.5
make
make install

Check that it linked ok:
ldd /opt/gromacs/gromacs-4.5.5/bin/grompp_netlib
        linux-vdso.so.1 =>  (0x00007fffb83f2000)
        libgmxpreprocess.so.6 => /opt/gromacs/gromacs-4.5.5/lib/libgmxpreprocess.so.6 (0x00002b6411cfa000)
        libmd.so.6 => /opt/gromacs/gromacs-4.5.5/lib/libmd.so.6 (0x00002b6411fcd000)
        libfftw3f.so.3 => /usr/lib/x86_64-linux-gnu/libfftw3f.so.3 (0x00002b64123ad000)
        libxml2.so.2 => /usr/lib/x86_64-linux-gnu/libxml2.so.2 (0x00002b64127b0000)
        libgmx.so.6 => /opt/gromacs/gromacs-4.5.5/lib/libgmx.so.6 (0x00002b6412b10000)
        libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00002b6412fe5000)
        libnetblas.so => /opt/netlib/blas/lib/libnetblas.so (0x00002b64131e9000)
        liblapack.so => /opt/netlib/lapack/lib/liblapack.so (0x00002b64134cc000)
        libnsl.so.1 => /lib/x86_64-linux-gnu/libnsl.so.1 (0x00002b6413ece000)
        libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00002b64140e6000)
        libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00002b6414369000)
        libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00002b6414585000)
        libz.so.1 => /lib/x86_64-linux-gnu/libz.so.1 (0x00002b641490c000)
        liblzma.so.5 => /lib/x86_64-linux-gnu/liblzma.so.5 (0x00002b6414b24000)
        /lib64/ld-linux-x86-64.so.2 (0x00002b6411ad8000)
        libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00002b6414d47000)
        libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00002b641505d000)
        libquadmath.so.0 => /usr/lib/x86_64-linux-gnu/libquadmath.so.0 (0x00002b6415274000)

Here are some input files (it's not a 'real' md run -- I just needed something small and quick to run):
step1.top:
#include "/opt/gromacs/gromacs-4.5.5/share/gromacs/top/ffoplsaa.itp"
#include "/opt/gromacs/gromacs-4.5.5/share/gromacs/top/oplsaa.ff/tip4p.itp"

[system]
test 

[molecules]

step1.mdp:

integrator = md
define      = -DFLEXIBLE
emtol      = 1000.0
emstep     = 0.001
nsteps     = 5000
nstlist    = 1
ns_type    = grid 
rlist      = 0.9
coulombtype= PME  
rcoulomb   = 0.9  
rvdw       = 1.0  
pbc        =  xyz

genbox_netlib -o step1.gro -cs /opt/gromacs/gromacs-4.5.5/share/gromacs/top/tip4p.gro -box 4x4x4 -p step1.top

grompp_netlib -f step1.mdp -po step2.mdp -p step1.top -pp step2.top -c step1.gro -o step2.tpr

mdrun_netlib -v -s step2.tpr -o step3.trr -x step3.xtc -cpo step3.cpt -c step3.gro -e step3.edr -g step3.log

On my old AMD II X3 I got about 7.7 GFLOPS with Openblas and 7.8 GFLOPS with the above libs. Note that the run is shorter than a minute so it's pretty useless for benchmarking. However, there's no obvious MAJOR penalty.


If you don't have cmake:
cp INSTALL/make.inc.gfortran make.inc

Edit make.inc
 15 FORTRAN  = gfortran
 16 OPTS     = -O2 -fPIC -m64
 17 DRVOPTS  = $(OPTS)
 18 NOOPT    = -O0 -fPIC -m64
 19 LOADER   = gfortran
 20 LOADOPTS =
Edit Makefile
 11 #lib: lapacklib tmglib
 12 lib: blaslib variants lapacklib tmglib
Run make
make
-->  Tests passed: 13176


   -->   LAPACK TESTING SUMMARY  <--
  Processing LAPACK Testing output found in the TESTING direcory
SUMMARY              nb test run  numerical error    other error  
================    =========== ================= ================  
REAL              1077227  0 (0.000%) 0 (0.000%) 
DOUBLE PRECISION 1078039  0 (0.000%) 0 (0.000%) 
COMPLEX           522814  0 (0.000%) 0 (0.000%) 
COMPLEX16          552410  0 (0.000%) 0 (0.000%) 

--> ALL PRECISIONS 3230490  0 (0.000%) 0 (0.000%) 




Older version:
In the oldest version of this post I did the blas compilation by hand:

gfortran -O2 -fPIC -m64 -march=native -funroll-all-loops -c *.f

To build a static library:
ar rvs libblas.a *.o

To build a shared/dynamic library:
gfortran -shared -Wl,-soname,libnetblas.so -o libblas.so.1.0.1 *.o -lc 


ldd libblas.so.1.0.1
        linux-vdso.so.1 =>  (0x00007fff301af000)
        libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00002aeeac390000)
        libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00002aeeac718000)
        libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00002aeeaca2e000)
        libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00002aeeaccb0000)
        libquadmath.so.0 => /usr/lib/x86_64-linux-gnu/libquadmath.so.0 (0x00002aeeacec7000)
        /lib64/ld-linux-x86-64.so.2 (0x00002aeeabedd000)

Either way:
cp libblas* /opt/netlib/blas/lib

To test:
wget http://www.netlib.org/blas/sblat1
mv sblat1 sblat1.f

And EITHER
gfortran sblat1.f -l:libblas.a

OR
ln -s libblas.so.1.0.1 libnetblas.so
gfortran sblat1.f -l:libnetblas.so

THEN
./a.out
 
Real BLAS Test Program Results
Test of subprogram number  1             SDOT 
                                    ----- PASS -----

 Test of subprogram number  2            SAXPY 
                                    ----- PASS -----

 Test of subprogram number  3            SROTG 
                                    ----- PASS -----

 Test of subprogram number  4             SROT 
                                    ----- PASS -----

 Test of subprogram number  5            SCOPY 
                                    ----- PASS -----

 Test of subprogram number  6            SSWAP 
                                    ----- PASS -----

 Test of subprogram number  7            SNRM2 
                                    ----- PASS -----

 Test of subprogram number  8            SASUM 
                                    ----- PASS -----

 Test of subprogram number  9            SSCAL 
                                    ----- PASS -----

 Test of subprogram number 10            ISAMAX
                                    ----- PASS -----

No comments:

Post a Comment