The ABINIT Users Mailing List ( CLOSED )

[Alain Jacques <ajs.jacques@gmail.com>]

I love good news :-) Don't trash your binaries too fast. Maybe they are
not so bad ...

dp 5.1.3 comes with Kohn Sham .kss files produced on a big endian
architecture that's different than yours. Your just compiled DP expect
unformatted binary files in little endian order suitable for your x86
processor. A tweak to read big endian binary files on an x86 CPU is
explained on
http://gcc.gnu.org/onlinedocs/gfortran/GFORTRAN_005fCONVERT_005fUNIT.html
So a first try consists of defining the GFORTRAN_CONVERT_UNIT
environment variable before running DP tests: if you're working in bash,
do an export GFORTRAN_CONVERT_UNIT=big_endian and remake the tests. It
may work. If it still complains, a possible explanation is that the .kss
files were produced with a different record marker length as the one
your compiler is defaulting ; there is no standard for this and it
varies between compilers brand and versions.
For all these difficulties, Abinit authors promote NetCDF data files and
a standard way to write them ETSF-IO ... they automatically care for
endianess and record lengths. Unfortunately not all abinitio
applications adhere to this standard ...
If you compiled DP with NetCDF support, it is able to read its own
NetCDF .nc files. So a second try consists of running the tests with the
.nc files in place of the corresponding .kss data. You have to modify
the perl test script or to run them by hand. For example, in the test
directory, edit t1.in to change verbose=3 to verbose=2; run the first
test with ../src/dp-5.1.3 -k si8H2k-etsf.nc -i t1.in > dp.log. Does it
work now? Why verbose=2 - because my DP binaries produced with
gfortran-4.1.2 segfault with verbose=3. Try ... your mileage may vary.

Assuming you have compiled Abinit and DP with the same compiler and the
same word length (32 bit or 64 bit), they should be able to exchange
binary .kss files (...).

Alain


文沈 wrote:
> Hi, Alain,
> I have one good news and one bad news here.
> good news is that following your steps, I passed make,but the bad news
> is when it comes make tests,
> it gives errors
>
> Test 1: rpa calculation 1q->0 tamm-dancoff approx ............cat:
> outlf.mdf: No such file or directory
> cat: outlf.mdf: No such file or directory
> Illegal division by zero at ./testpl line 82.
> make[1]: *** [all] Error 9
> make[1]: Leaving directory `/home/wenshen/Desktop/dp-5.1.3/test'
> make: *** [tests] Error 2
>
> So I don't think it make properly
>