[Pw_forum] cp.x and lemieux TRU64 Alpha

[aaron at nemo.physics.ncsu.edu]

Hi,

The problem I'd like to solve is the following:

Our cp.x build on the Pittsburg Supercomputing Center alpha cluster 
Lemieux is failing.  The code runs perfectly while we minimize the 
electronic wavefunction.  When we minimize the ionic structure the run 
fails with the following error:

______________standard out error message_________________________
forrtl: error (72): floating overflow
forrtl: error (76): IOT trap signal
_________________________________________________________________

In the pipline output the message is as follows:

*****************************************************************
<>
 subroutine rhoofr: total integrated electronic density
 in g-space = 50.000000     in r-space = 50.000000


                total energy =      -75.04974 a.u.
              kinetic energy =       39.40916 a.u.
        electrostatic energy =      -56.31996 a.u.
                         esr =        4.15664 a.u.
                       eself =       61.30307 a.u.
      pseudopotential energy =      -48.85354 a.u.
  n-l pseudopotential energy =       12.79418 a.u.
 exchange-correlation energy =      -22.07958 a.u.
           average potential =       -0.10767 a.u.

exception system:  exiting due to multiple internal errors:
        handler returned invalid disposition
        exception dispatch or unwind stuck in infinite loop
*****************************************************************

We have isolated the failure to the subroutine ortho.  It performs roughly 
five iterations and the error occurs.

Is any forum member running CPV/cp.x on the psc machines or similar 
architecture?   

I'm attaching the make.sys for inspection. We have reduced the
optimizations with no change in the result.  Also we used both old and new 
methods of configure.

Thanks in advance for any advice you can provide,

Aaron
-------------- next part --------------
OSHOME=/usr/users/0/georgea/AiMDF90
#
# System-dependent definitions for HP/COMPAQ alpha parallel machines
# (contributed by Guido Roma). Edit according to your needs
#

# Use fft routines from the dxml or cxml library:

# CPPFLAGS = -D__ALPHA -D__PARA -D__MPI -DDXML -I$(OSHOME)/include

# Use precompiled fftw library (version <= 2.1.5, NOT v.3!)
#
# In this case, specify also how to load the fftw library (FFTW_LIB)
# and the path to the fftw.h include file (FFTW_INC_DIR). Example:
# FFTW_LIB=-L/usr/local/src/fftw-2.1.3/fftw/.libs -lfftw
# FFTW_INC_DIR=/usr/local/src/fftw-2.1.3/fftw
# CPPFLAGS = -D__ALPHA -D__PARA  -D__MPI -D__FFTW \
#  -I$(OSHOME)/include -I$(FFTW_INC_DIR)

# Use the local copy of fftw
#
CPPFLAGS = -D__ALPHA -D__PARA -D__MPI -D__FFTW -D__USE_INTERNAL_FFTW \
  -I$(OSHOME)/include -I./

#
# Fortran compiler:
#
F90 = f90
F77 = f90
CC = cc

#
# Fortran options:
#
FFLAGS = -O -fpe0 -real_size 64 -align dcommons -align records
# Flags for FPMD
# FFLAGS = -double_size 64 -fpconstant -real_size 64 -fast \
#        -math_library fast -tune ev68

#
F77FLAGS = $(FFLAGS)
F90FLAGS = $(FFLAGS) -free -cpp $(CPPFLAGS)

CCFLAGS = -O $(CPPFLAGS)
#
# This is needed to tell the compiler where modules are
#
MODULEFLAG= -I$(OSHOME)/Modules -I$(OSHOME)/PW -I$(OSHOME)/PH

#
# Libraries (mpi + cxml + fftw) :
#
LIBS = -lfmpi -lpmpi -lmpi -lelan -lcxml $(FFTW_LIB)
#
# Loader:
# 
LD=$(F90)
LDFLAGS = $(OSHOME)/flib/ptools.a $(OSHOME)/flib/flib.a $(OSHOME)/clib/clib.a $(LIBS)

AR = ar
ARFLAGS = rv