Page 1 of 1

Vasp on 64-bit Intels using ifort/icc

Posted: Wed Nov 28, 2007 1:39 pm
by plesset
I have been trying to get VASP to work on 64-bit Intel processors, using mpich2-1.0.6p1 and ifort/icc_10.1.008 and the larest MKL_10.0.011 as well. The mpich compiles without problems and works as supposed. VASP compiles as well, but running the standard benchmark test results in:

BRMIX: very serious problems
the old and the new charge density differ
old charge density: 597.39141 new 597.70047

The serial version works fine and without errors. But I also get the same error if I compile a serial version and use FFTW (which I compile myself and passes all tests) instead of the default FFT.

I have been using:
setenv FC ifort
setenv F90 ifort
setenv F77 ifort
setenv CC icc
setenv CXX icc
setenv CPP 'icc -E'

setenv FFLAGS '-w -O3 -static'
setenv F77FLAGS '-w -O3 -static'
setenv CFLAGS '-w -O3 -static'
setenv CXXLAGS '-w -O3 -static'

For compiling vasp I added in "-FR -lowercase -w -mP2OPT_hlo_loadpair=FALSE -auto -i4" to the Makefile

Has anyone here experience compiling VASP on 64-bit Intel processors using ifort/icc?

Thanks!

Vasp on 64-bit Intels using ifort/icc

Posted: Thu Nov 29, 2007 1:05 pm
by MYoshiya
Hi.
I have the exactly same problem as that you encounter as I reported in the other thread entitled 'optimization failure?'. Decreasing optimization level allows you to bypass the problem.

Vasp on 64-bit Intels using ifort/icc

Posted: Fri Nov 30, 2007 9:13 am
by admin
please simply use the standard makefile.linux_ifc_P4 (and replace ifc by ifort). please also mind that in MPICH, all integers are default-C standard (32bit), therefore the compiler option -i4 should be avoided (conflict when passing inegers from/to MPI).

Vasp on 64-bit Intels using ifort/icc

Posted: Fri Nov 30, 2007 4:07 pm
by michaeleric.williams
I am using a very similar system to you and am having problems compiling. Could you tell me what your BLAS and LAPACK lines in your makefile are?

Thanks