The ABINIT Users Mailing List ( CLOSED )

["Kevin Jorissen" <kevin.jorissen@ua.ac.be>]

Dear abinit users,

I am trying to install abinit v4.0 on a cluster of xeon dual proc. machines
using the Intel Fortran Compiler 7.0-64.  Recent e-mails on the forum report
optimization problems compiling abinit v3,4,3 on an Itanium using the
7.0.64, and mention upgrading to 7.0.79 as a remedy.
I tried anyway.

I use the makefile_macros /Intel_P6/m_m.IFC_v7trial.  I change the path for
ifc compiler, cpp, and perl.  I keep the optimization flags -tpp7 and -xW in
FFLAGS* since they are correct for Intel Xeon processors (I would like to
remark at this point that, contrary to what was implied in one of the
e-mails on the forum, they are NOT correct for Itanium machines.  There, one
should use either tpp1 or tpp2, but not tpp7.  I'm not sure about the xW).

This is the first part of my makefile_macros (I'm definitely interested in
parallel calcul, too, but first things first - sequential, that is ).
####################################################################
# For the sequential version
# Machine type
MACHINE=P6
# Fortran optimized compilation
FC=/usr/bin/ifc
#Michel wrote to use  -O3 -xW -tpp7 , except for drive, iovars, and psp
#For v6 it was -FR -O3 -w -cl,../Machine_dept_files/Intel_P6/my_work.pcl
FFLAGS=-FR -O0 -xW -tpp7
FFLAGS_Src_2psp =-FR -O0
FFLAGS_Src_3iovars =-FR -O0
FFLAGS_Src_9drive =-FR -O0
FFLAGS_LIBS= -O3
# C preprocessor, used to preprocess the fortran source.
CPP=/usr/bin/cpp
CPP_FLAGS=-P -traditional -D__IFC
# C optimized compilation.
CC=cc
CFLAGS=-O
# Location of perl .
PERL=/usr/bin/perl
# List of machine-dependent routines
MACHINE_DEP_C_SEQ_SUBS_LIST=etime.o
####################################################################

No compilation errors (a million warnings and comments about obsolete
features etc. :)).  make test1 and make tests are OK.
In the next tests (Test_v1,2,3,4) I get quite lengthy fldiff.reports,
containing
*the harmless features described in the installation notes
*different numbers of comments and warnings, presumably equally harmless
*numerical differences.  most of these look pretty innocent at first sight
(e.g. ETOT differs slightly, but converges in the same number of iterations
to the same value), e.g. (from Test-V1)
 Case_8
17
<  Starting date : Mon 17 Mar 2003.
>  Starting date : Sat 22 Feb 2003.
171
<  frms,max,avg= 1.7837379E+01 3.0895246E+01   3.404E-06  1.584E-06
3.858E-06 e/A
>  frms,max,avg= 1.7837379E+01 3.0895246E+01   3.404E-06  1.584E-06
3.857E-06 e/A
289
<  Delivered    3 WARNINGs and   1 COMMENTs to log file.
>  Delivered    1 WARNINGs and   1 COMMENTs to log file.
290
< +Overall time at end (sec) : cpu=         15.1  wall=         15.1
> +Overall time at end (sec) : cpu=         14.1  wall=         28.4

but there are tons of them.  Also, maybe less innocent : (Test_v3)
864
<  RCP: (index,theta,phi)  17  0.15708587E+01  0.59666371E+00
>  RCP: (index,theta,phi)  17  0.15708587E+01  0.97413262E+00
867
<  RCP: (index,theta,phi)  18  0.97413262E+00 -0.75460054E-04
>  RCP: (index,theta,phi)  18  0.59667247E+00 -0.11940436E-03
870
<  RCP: (index,theta,phi)  19  0.15708634E+01  0.97412386E+00
>  RCP: (index,theta,phi)  19  0.15707919E+01  0.25447975E+01
873
<  RCP: (index,theta,phi)  20  0.59666371E+00 -0.11109318E-03
>  RCP: (index,theta,phi)  20  0.97400121E+00  0.31415873E+01
885
<  RCP: (index,theta,phi)  24  0.59666371E+00  0.15709074E+01
>  RCP: (index,theta,phi)  24  0.59667247E+00  0.15709157E+01

I tried with the global FFLAGS set either to O0 (no optim.) and O3 (full
optimization).  In both cases I get this kind of output (examples from O0).

Would you consider such an installation to be succesful, or should I
reinstall?  And if so, what do you recommend - newer compiler, other
options?

Thank you for your assistance,

Kevin Jorissen
University of Antwerp
kevin.jorissen@ua.ac.be