Hi,
On Mon, 12 Apr 2004, Guanglei Cui wrote:
> I've been trying to compile amber8 for SCALI with
>
> configure -static -scali -athlon pgf90
>
> $SCALI_HOME has been set properly. I found I had to remove -lscasci
> -lscartl from the config.h to pass the compilation. However, the job
> didn't start. In PBS output,
>
> --- mpimon --- Aborting run after timeout while waiting for process(es)
> to start
>
> and sander stopped at
>
> ---------------------------------------------------
> | Local SIZE OF NONBOND LIST = 2092754
> | TOTAL SIZE OF NONBOND LIST = 2092754
>
> Has anyone had similar experience with SCALI?
There are a couple of problems:
1.
The scali section of configure has been mangled;
the patch below fixes this (for example -DMPI was not included in fppflags).
2.
Apparently the combination of static and scali was never tried.
I have just made several unsuccessful attempts to execute a
statically linked scali sander. Details are below; I give up.
----------------------------------
1.
********>Bugfix 2:
Author: Scott Brozell
Date: 04/12/2004
Programs: sander, sander.LES
Description: the configure option -scali does not produce MPI executables.
Fix: apply the following patch to amber8/src/configure
------------------------------------------------------------------------------
*** configure 12 Mar 2004 00:19:15 -0000 1.224
--- configure 12 Apr 2004 22:19:36 -0000
***************
*** 282,287 ****
--- 282,289 ----
scali_root='/opt/scali'
if [ -z "$SCALI_HOME" ] ; then
if [ -d $scali_root ]; then
+ echo "Warning: the SCALI_HOME environment variable is not defined !"+ SCALI_HOME="$scali_root"
echo "Caution: Setting SCALI_HOME to $scali_root (guessed)"
else
PAR="SCALI"
***************
*** 289,296 ****
EXAMPLE="/usr/local/scali"
par_error
fi
fi
! if [ ! -d $SCALI_HOME/lib ] ||[ ! -d $SCALI_HOME/include ]
then echo " missing dirs"; missingfiles ; fi
if [ ! -f $SCALI_HOME/include/mpif.h ] ;
then echo " missing include"; missingfiles ; fi
--- 291,301 ----
EXAMPLE="/usr/local/scali"
par_error
fi
+ else
+ echo "SCALI_HOME is set to $SCALI_HOME"
fi
!
! if [ ! -d $SCALI_HOME/lib ] || [ ! -d $SCALI_HOME/include ]
then echo " missing dirs"; missingfiles ; fi
if [ ! -f $SCALI_HOME/include/mpif.h ] ;
then echo " missing include"; missingfiles ; fi
***************
*** 300,308 ****
then echo " missing libs"; missingfiles; fi
done
! echo " SCALI_HOME is set to $SCALI_HOME"
! loadlib="$loadlib -lm -L$SCALI_HOME/lib -lfmpi -lmpi -lscasci -lscirtl"
! fppflags="-I$SCALI_HOME/include $fppflags"
;;
none) # Nothing to do
--- 305,321 ----
then echo " missing libs"; missingfiles; fi
done
! if [ $static = "yes" ]; then
! # scali/doc/ScaMPI/FAQ item 20 (Version: 1.20 Date: Oct 15th 2001):
! #loadlib="$loadlib -L$SCALI_HOME/lib -lfmpi -lmpi -Wl,-Bdynamic -lpthread -ldl -Wl,-Bstatic"
! # builds but does not execute.
! # Neither does static linking of -lpthread -ldl.
! echo "Error: static linking with SCALI is not supported !"
! exit 1
! else
! loadlib="$loadlib -L$SCALI_HOME/lib -lfmpi -lmpi"
! fi
! fppflags="-I$SCALI_HOME/include $fppflags -DMPI"
;;
none) # Nothing to do
------------------------------------------------------------------------------
Temporary workarounds: use a different MPI implementation.
----------------------------------
2.
Verbose details on static and scali attempts:
setenv DO_PARALLEL "mpirun -npn 1 -np 1"
Following scali/doc/ScaMPI/FAQ item 20 (Version: 1.20 Date: Oct 15th 2001):
loadlib="$loadlib -L$SCALI_HOME/lib -lfmpi -lmpi -Wl,-Bdynamic -lpthread -ldl -Wl,-Bstatic"
But
/opt/scali/bin/mpimon -stdin all ../../exe/sander -O -i in.md -c crd.md.23 -o cytosine.out -- gremlin44 1
sander-0(mpi:18098.gremlin.scripps.edu): relocation error: sander-0(mpi:18098.gremlin.scripps.edu): undefined symbol: dl_SdbReadPciConfig
--- mpimon --- Aborting run after process-0 terminated abnormally Childprocess 17077 exited with exitcode 127 ---
Since libpthread.a and libdl.a exist:
LOADLIB= -L/opt/scali/lib -lfmpi -lmpi -lpthread -ldl -llapack -lblas
But
/opt/scali/bin/mpimon -stdin all ../../exe/sander -O -i in.md -c crd.md.23 -o cytosine.out -- gremlin44 1
ERROR: Failed to dynamically load semaphore symbols: /opt/scali/lib/: cannot read file data: Is a directory
--- mpimon --- Aborting run after process-0 terminated abnormally Childprocess 17067 exited with exitcode 1 ---
Apr 12 14:55:40: (mpimon.gremlin.scripps.edu)(6151) Error: Timed out while waiting for all Ready in MonitorInputSocketHander
Scott
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Tue Apr 13 2004 - 00:53:00 PDT
