Skip to Content.
Sympa Menu

forum - Compile ABINIT successfully of PC-Linux parallel version

forum@abinit.org

Subject: The ABINIT Users Mailing List ( CLOSED )

List archive

Compile ABINIT successfully of PC-Linux parallel version


Chronological Thread 
  • From: "Zhang Mike" <mz24cn@hotmail.com>
  • To: forum@abinit.org
  • Subject: Compile ABINIT successfully of PC-Linux parallel version
  • Date: Sat, 28 Dec 2002 23:31:24 +0800

Because the available parallel version of ABINIT for PC-Linux cluster is very old, last week I tried compiling ABINIT based on Lam-MPI interface and PGI-F90 compiler on a 8 node plus a server node PC-Linux beowulf cluster. I compiled and RUN the sequential version and the parallel version 3.3.4 all SUCCESSFULLY. I would like to upload the binary files of PC-Linux beowulf cluster parallel version if required. You also can make the parallel version by yourself. Only should you notice the makefile_macro must be modified to fit the path setup of your cluster.

However, I encountered several "odd" problems and I will describe them in details.

I found when launching "make", it firstly invoked the "cpp" such as:
/lib/cpp -C -DP6 -P -DMPI respfn.f
As we know, macro P6 and MPI are required when building the parallel version on PC. Cpp generated respfn_cpp.f which would be compiled by pgf90. However, I found "cpp" also paid attention to the comments in source files which began with a '!' or '!!'. The majority of source files are compiled successfully except tddft.f, initsph.f, mkffnl.f. For example, the error is
/lib/cpp -C -DP6 -P tddft.f > tddft_cpp.f
hf77 -Mextend -Mfree -O2 -c tddft_cpp.f
PGF90-S-0034-Syntax error at or near , (tddft_cpp.f: 55)
The 55th line of tddft.f is !! systems: Theory, Computational Methods, and Functionals", by M.E. Casida,
After compiling by cpp, it changed into (in tddft_cpp.f)
!! systems: Theory, Computational Methods, and Functionals"
, by M.E. Casida,
I don't know why there exists the carriage return character in the line after cpp's compilation. The phrase ", by M.E. Casida," inevitably incurred a compilation error. I tried deleting the "-C" switch on cpp command line but it was no use. So I simply got rid of the 54-55th line of tddft.f to remove the error.
The compilation errors in initsph.f and in mkffnl.f are similar as that in tddft.f. I changed the 20th line of initsph.f,
!! pawsph% nphi - " " " " " " phi " (there are odd " characters)
into
!! pawsph% nphi - " " " " " " phi "" (there are even " characters)
and changed the 10th, 12th, 81st, 92nd lines of mkffnl.f,
!! ... FFNL's ...
into !! ... FFNL''s ...
All was Okay and compilation was accomplished.

Then I run the built parallel version as follows, (enter Test_in directory)
mpirun -np 8 ../abinip <test1.files >&test1.log
All results are same as those of suquential version, but it seems the abinip is not faster than abinis :(
I am also tring to build abinip by Intel Fortran Compiler. I've found the binaries compiled by IFC run faster than that by PGI on PC. Unfortunately, I have downloaded IFC_for_Linux several times but all are corrupted (it is very big, up to ac. 80MB). Let me try again :)
Good continuation,
Mike Zhang

--------------------------------------------------------
Where there is a will,there is a way.
_\|/_
(@ @)
+-------------------ooO-(_)-Ooo------------------------+
| Zhang Hua |
| M.S. Candidate |
| The 11th Lab of Institute of Plasma Physics, |
| Chinese Academy of Science |
+------------------------------------------------------+
| Addr : P.O. BOX 1126, Hefei Anhui, PRChina |
| P.C. : 230031 |
| Tel : +86-551-5591377(O),5592922(H), |
| Fax : +86-551-5591310 |
| Mobile: +86-13965034376 |
| Email : xw23@163.com(private) |
| mz24cn@hotmail.com(academic) |
|_________________________Oooo.________________________|
.oooO ( )
( ) ) /
\ ( (_/
\_)





_________________________________________________________________
与联机的朋友进行交流,请使用 MSN Messenger: http://messenger.msn.com/cn



Archive powered by MHonArc 2.6.16.

Top of Page