Top Page > Browsing

Tips for installation of OpenMX Date: 2007/09/25 17:12 Name: T.Ozaki Dear All, The following is tips for installation of OpenMX. Most problems in installation of OpenMX come from compilation of LAPACK and BLAS and its linking. Thus, I shall show tips for installation of OpenMX on several platforms. (1) Intel Pentium 4 and Xeon (32 bit) A simple way is to use highly optimized ATLAS library (libatlas_p4.a) provided by Dr. Axel Kohlmeyer. http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/cpmd-linux.html#atlas Then, you may link these libraries in makefile of OpenMX, CC = mpicc -O3 -I/usr/local/include LIB = -L/usr/local/lib -lfftw3 -latlas_p4 -static Another way is to use CLAPACK http://www.netlib.org/clapack/ The compilation of CLAPACK is made by 1. get clapack.tgz from netlib.org 2. tar zxvf clapack.tgz 3. cd CLAPACK/ 4. cp -f INSTALL/make.inc.LINUX ./make.inc (You can change the compiler and compile options in the make.inc.) 5. make f2clib 6. make blaslib 7. cd SRC 8. make (lapack_LINUX.a is made) 9. cd ../TESTING/MATGEN/ 10. make 11. cd ../../ 12. If you are a super user, do cp lapack_LINUX.a /usr/local/lib/liblapack.a cp blas_LINUX.a /usr/local/lib/libblas.a cp F2CLIBS/libF77.a /usr/local/lib/libg2c.a cp F2CLIBS/libI77.a /usr/local/lib/libI77.a cp clapack.h /usr/include cp F2CLIBS/f2c.h /usr/include /sbin/ldconfig Then, you may link these libraries in makefile of OpenMX, CC = mpicc -O3 -I/usr/local/include LIB = -L/usr/local/lib -lfftw3 -llapack -lblas -lg2c -static or CC = mpicc -Dbraswrap -O3 -I/usr/local/include LIB = -L/usr/local/lib -lfftw3 -llapack -lblas -lg2c -lI77-static 12' if you are not a super user, do mkdir ~/lib mkdir ~/include cp lapack_LINUX.a ~/lib/liblapack.a cp blas_LINUX.a ~/lib/libblas.a cp F2CLIBS/libF77.a ~/lib/libg2c.a cp F2CLIBS/libI77.a ~/lib/libI77.a cp clapack.h ~/include cp F2CLIBS/f2c.h ~/include Then, you may link these libraries in makefile of OpenMX, CC = mpicc -O3 -I/usr/local/include -I$(HOME)/include LIB = -L$(HOME)/lib -lfftw3 -llapack -lblas -lg2c -static or CC = mpicc -Dbraswrap -O3 -I/usr/local/include -I$(HOME)/include LIB = -L$(HOME)/lib -lfftw3 -llapack -lblas -lg2c -lI77 -static (2) Intel Pentium D and Xeon (EM64T) A simple way is to use highly optimized ATLAS library (libatlas_x86_64.a) provided by Dr. Axel Kohlmeyer. http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/cpmd-linux.html#atlas Then, you may link these libraries in makefile of OpenMX, CC = mpicc -O3 -I/usr/local/include -I$(HOME)/include LIB = -L$(HOME)/lib -lfftw3 -latlas_x86_64 -static (3) AMD Opteron A simple way is to use ACML library (acml-3-1-0-gnu-64bit.tgz) provided by AMD. http://developer.amd.com/acml.aspx Then, you may link these libraries in makefile of OpenMX, CC = mpicc -fast -I$(HOME)/include LIB = -L$(HOME)/lib -lfftw3 -lacml -L/opt/gcc33/lib64 -lg2c (4) Intel Itanium A simple way is to use MKL library. Then, you may link these libraries in makefile of OpenMX, CC = mpicc -O3 -I$(HOME)/include -I/opt/intel/mkl/include -Wl,--allow-multiple-definition LIB = -L$(HOME)/lib -lfftw3 -L/opt/intel/mkl/lib/64 -lmkl_lapack -lmkl_ip -lguide -L/usr/lib -lpthread -static ------------------------------------------ As for Intel icc compilear Version 9.0 There is a specific problem for the intel icc compiler Version 9.0. When "-O3" is used for icc Version 9.0 as compile option, the icc compiler often optimizes a code incorrectly, and as a result the executable file behaves in a unexpected mannar, although I am not sure whether this is a bug of Version 9.0 or not. (Note: I did not see the problem in Versions 7.0 and 8.0) The solution is to use another option instead of "-O3". I checked the effect of the compiler option for Version 9.0 and obtained the following result. "-O1" works well "-O2" works well "-O3" no good for bulk systems "-Os" works but does not work for some of input files "-fast" may work (but I am not sure) A better choice is to use "-O2" for Intel icc Version 9.0. ------------------------------------------ As for Portland PGCC compiler Version 7.0 There is a specific problem for the portland pgcc compiler Version 7.0. When "-fast" or more is used for icc Version 7.0 as compile option, the pgcc compiler often optimizes a code incorrectly, and as a result the executable file behaves in a unexpected mannar, although I am not sure whether this is a bug of Version 7.0 or not. (Note: I did not see the problem in Versions 5.1 and 6.1) The solution is to use another option instead of "-O3". I checked the effect of the compiler option for Version 9.0 and obtained the following result. "-O1" works well "-O2" works well "-O3" works well "-fast" no good "-fastsse" no good A better choice is to use "-O3" for Porland pgcc Version 7.0.