Opteron installation

Message

danmajchrzak · #1 Post by **danmajchrzak** » Wed Jun 08, 2005 4:36 pm

We've been running the 32 bit version on an Athlon and Xeon based clusters. We have recently aquired a small opteron cluster. I've compiled the latest code using the PGI compilers and (path modified) pgi opteron make file. At this point I've only compiled the serial version. When I try to run it on the benchmark files I get the following output:

./vasp
vasp.4.6.26 15Dec04 complex
POSCAR found : 1 types and 8 ions
Error reading item 'LREAL_COMPAT' from file INCAR.
WARNING: mass on POTCAR and INCAR are incompatible
typ 1 Mass 63.55000000000000
63.54600000000000
LDA part: xc-table for Ceperly-Alder, standard interpolation
internal error: XML_INCAR called with a vector, use XML_INCAR_V instead
I_CONSTRAINED_M
FORTRAN STOP

Has anyone seen this behavior? Have a solution?

#2 Post by **admin** » Mon Jun 13, 2005 7:32 am

These errors do not occur on any of our machines (even the HP-UX box),
most probably, the subroutine READER (reader.R) was not compiled correctly.

ad 1) (ion's mass): Did you read it in explicitely (POMASS-tag in INCAR)? if so, it has to match the mass read from POTCAR to 1e-03.
The Cu-mass is 63.546 in all POTCARS.

2) XML_INCAR:
I fear you will have to do some debugging to figure out the reason for the error.
The relevant lines are located in constrmag.F (lines 50 ff in vasp.4.6.26), please insert
the
WRITE(*,*) N
line as indicated by !test:

I_CONSTRAINED_M=0
CALL RDATAB(LOPEN,'INCAR',IU5,'I_CONSTRAINED_M','=','#',';','I', &
& I_CONSTRAINED_M,RDUM,CDUM,LDUM,CHARAC,N,1,IERR)
IF (((IERR/=0).AND.(IERR/=3)).OR. &
& ((IERR==0).AND.(N=0) &
WRITE(IU0,*)'Error reading item ''I_CONSTRAINED_M'' from file INCAR.'
GOTO 150
ENDIF
!test
WRITE(*,*) N
!test
CALL XML_INCAR('I_CONSTRAINED_M','I',I_CONSTRAINED_M,RDUM,CDUM,LDUM,CHARAC,N)

The problematic call is XML_INCAR. Insert the indicated lines, recompile and
run the code. N should be 1 or 0. If this is not the case the compiler has
incorrectly compiled RDATAB (vasp.4.lib/rdatab.X). If it is 0 or 1, it is most likely the module
xml.F that was not correctly compiled. Try to decrease the
optimisation for the problematic routines in the makefile and recompile again.

pavel · #3 Post by **pavel** » Thu Oct 19, 2006 11:48 am

I've got exactly the same error after compilation on Opteron cluster with pgf90.

Some remarks to the reply of head admin:
to ad 1): The error occurs during running benchmark, not by reading a user written file. The same warning is reproduced in the VASP Guide (p.17 The Installation of Vasp). Certainly, it is not a reason for crash.

to ad 2) After insertion of the proposed debug print in constrmag.F I've got N=4598385715892453376, which suggests that RDATAB was compiled incorrectly.

Decreasing the level of optimization for RDATAB contained in drdatab.F (vasp.4.lib) does not help.

This topic was posted nearly two years ago and, probably, a solution has been already found. Dear danmajchrzak, I would appreciate very much if you could report your solution.

Apropo, the error "Error reading item 'LREAL_COMPAT' from file INCAR." was issued earlier by subroutine LREAL_COMPAT_MODE. What is the reason for this error?
<span class='smallblacktext'>[ Edited Thu Oct 19 2006, 01:55PM ]</span>

pavel · #4 Post by **pavel** » Thu Oct 19, 2006 1:40 pm

Solution found: both vasp and vasp.lib should be compiled with the same length of integer (compiler option -i8)!

Makefile for vasp.4.lib (makefile_linux_pg) does not have this explicit option and was compiled by default with -i4. This was the reason for the error "Error reading item 'LREAL_COMPAT' from file INCAR." and "internal error: XML_INCAR called with a vector, use XML_INCAR_V instead".

Compilation of both vasp.lib and vasp with option -i4 resulted in "Segmentation fault", while option -i8 (makefile_linux_pgi_opt) works well.