vasp 4.6.26 mpi on Athron64 with ifce 8.1 dosen't work

Message

atogo · #1 Post by **atogo** » Sat Apr 30, 2005 3:14 am

Hello,

I compile vasp 4.6.26 mpi with ifce 8.1 (x86-64) and mpich 1.2.6 on Athron64.
When running it, I got error message like this,

------
LDA part: xc-table for Ceperly-Alder, Vosko type interpolation para-ferro
POSCAR, INCAR and KPOINTS ok, starting setup
WARNING: wrap around errors must be expected
FFT: planning ... 7
reading WAVECAR
entering main loop
N E dE d eps ncg rms rms(c)

/usr/local/calc/mpich/bin/mpirun.ch_p4: line 243: 2241 Segmentation fault /usr/local/src/vasp.4.6.26-mpi-gamma-ifc81-mkl72/vasp -p4pg
-----

I have no idea how to compile in this environment.
I tried two types of makefiles which are modified slightly from makefile.linux_ifc_P4 and makefile.linux_ifc_ath.

I could run vasp 4.6.21 mpi by using slightly modified makefile.linux_ifc_P4.

Would anyone give me any hint ?

Thank you,

#2 Post by **admin** » Wed Jun 01, 2005 8:36 am

concerning the compilation, there are no significant differences between
vasp.4.6.21 and vasp.4.6.26 --> if you can compile and run vasp.4.6.21 sucessfully on your specific architecture, vasp.4.6.26 should run as well.

dajiang · #3 Post by **dajiang** » Fri Jun 03, 2005 9:06 pm

I found similar problems with both serial and parallel versions, on both P4 and AMD64 machines. Just sent a bug report. One thing I noted is that the newest intel ifort8.1 compiler (l_fc_pc_8.1_028) doesn't seem to have the segmention fault problem, while the 64 bit (l_fce_pc_8.1_028) still does. Also the segmention fault only occurs with larger supercells.

#4 Post by **admin** » Thu Jun 16, 2005 9:37 am

the segmentation fault seems to be due to a compiler bug of rel. 8 of the intel
Fortran compiler, which affects the memory handling in hamil.F.
-- only vasp-releases >= 4.6.25, compiled with rel.8 of the Intel compiler are
affected,
-- ALL code releases are compiled correctly with ifc7 and the compiler70-version of mpif77
-- all older code releases are compiled correctly with rel.8 as well

Prof. Kresse has implemented a work-around in hamil.F for this case, which he will put onto our ftp server within the next few days.

atogo · #5 Post by **atogo** » Fri Jul 29, 2005 1:48 am

Hello,

I tried to compile 4.6.27 by IFC9.0 and it works.
(Though it couldn't work with fftw...)
Thank you very much.