The ABINIT Users Mailing List ( CLOSED )

[Yunkai Zhou <zhou@sp.msi.umn.edu>]

Hi,
I am trying to get familiar with abinit.
But during the lib building process, I already found
some features of abinit not so appealing.
(please consider all my criticisms not as offense
but as my goodwill to wish that abinit will become
better and better).

The first is that I don't see (from my brief look at installation
guide) how to assign other libs to $(LIBS).
For example, I set the variable

LIBS= -L$(HOME)/lib/fftw-2.1.5/libs/ibm/lib  -lfftw -lrfftw \
      -L$(HOME)/lib/LAPACK/lib/ -llapack  -lessl -lm

But I got error:
--
ld: 0706-006 Cannot find or open library file: -l lapack
        ld:open(): A file or directory in the path name does not exist.
make: 1254-004 The error code from the last command is 255.
--

I think there is not reason to insist users to use the attached
LAPACK FFT etc (LAPACK v3.0 is available and pretty easy to install),
newer version of FFTW is available too.  What's more, the user does
not need to rebuild libs that are already available.  Vender optimized
libs would possibly be better choices.


The 2nd is about speed. I found it difficult to understand why you
have to produce a temp *_ccp.f file, then compile it, then remove it.
On some remote file systems, this is really slow!! It took several
hours to compile and build allseq (I am using pretty fast IBM regatta).
Not because of the compile time is long, but because creating the temp
files on remote disk is slow. Even after I move abinit to the local
disk (usually in some /scratch dirs), the temp files creation speed
improves, but it is still a bit slow to build abinit.
I have not looked into source files carefully (I guess the _ccp.f
is to put the included files in the source file?? But cannot compilers
handle the include directly? Modern compilers are developing so fast).

Thanks,

--
Yunkai Zhou