[CPMD-list] essl errors with lapack and essl on IBM_JS20.
Joseph Hargitai
joseph.hargitai at nyu.edu
Fri Dec 8 17:43:48 CET 2006
Yesterday with the help of Axel we made the lapack essl version work by
adding dsygv to the -qextname list. However, as Axel pointed out - the
-qextname list should only really have -printmemsize - the rest should
be picked up without an issue. (which is not our experience)
For my understanding:
JS20 compiled Lapack with essl should compile without a long -qextname
list vs std Lapack - which we are using in this case?
As far as Goto - we have it - just to make sure - use standard Lapack
with Goto, not JS20?
j
-------------- next part --------------
Yes -
it looks like you are "merging" too much -
I suggest to compile lapack on you machine and then link with -lapack
-lessl ( as inthe IBM-JS20-ESSL-MPI conf)
if you want to use atlas from Marenostrum - please try to build
appropriate lapack merging the atlas lapack subset and then use/modify
the IBM-JS20-ATLAS-MPI conf.
Best performance is obtained using goto ( if you have it) + std lapack +
ESSL as in IBM-JS20-GOTO-MPI .
Best Regards,
Alessandro
Alessandro CURIONI, PhD
Research Staff Member
Computational Biochemistry and Material Science group
IBM Research Division - Zurich Research Laboratory
Saumerstrasse 4
8003 Rueschlikon - Switzerland
e-mail: cur at zurich.ibm.com
www: www.zurich.ibm.com
Tel: +41-1-7248633
Fax: +41-1-7248958
Joseph Hargitai <joseph.hargitai at nyu.edu>
Sent by: cpmd-list-bounces at cpmd.org
12/08/2006 05:02 PM
To
cpmd-list at cpmd.org
cc
Subject
[CPMD-list] essl errors with lapack and essl on IBM_JS20.
Dear All:
We are testing some cpmd runs on our IBM JS20 cluster. (PPC970, Redhat
4, Myrinet) The compile is
with lapack essl and mpi. Some runs work okay - others have this error,
which, unless we are mistaken can be traced back to essl
errors/misinterpretations.
We compiled with the following flags (had to include the qextname -
otherwise we had issues compiling. Maybe there is a better way)
We also tried, based on an old posting - llapack lessl llapack - link
ordering - same errors and both -DFFT_ESSL and -DFFT_DEFAULT
The lapack is standard, we also used other compiles with Atlas specially
compiled for JS20s - from Mare Nostrum. We merged the lapack libs with
the standard /usr/lib64 (maybe this is the issue - we would need a JS20
compiled Lapack)
(We are okay compiling on Sunfire 8 dual core cpus with 48GB memory -
using Pathscale and ACML.)
Compile script followed by the runtime error:
#INFO#
#INFO# IBM-JS20-MPI-64bit-essl
#INFO# you need lapack essl and goto 64 bit libraries
#INFO#
#INFO# A. Curioni
#INFO#
IRAT=2
FFLAGS=' -q64 -qmaxmem=32768 -qtune=ppc970 -qarch=ppc970
-qextname=dgelss:dsbevx:dstev:dsyev:dsyevx:dsysv:dtrtrs:printmemsize:zhbevx:zheev:zheevx:zhegv
'
FFLAGS_GROMOS=' -qarch=ppc970 -qdpc'
LFLAGS='-q64 -L/usr/local/lib -llapack -lessl \
-qarch=ppc970 '
CFLAGS=' -q64 -qarch=ppc970 '
CPP=' /opt/ibmcmp/xlf/9.1/exe/cpp -P'
CPPFLAGS=' -D__IBM -DLAPACK -DFFT_ESSL -D__PWRLinux -DMALLOC8
-DPOINTER8 -DPARALLEL=PARALLEL -DMP_LIBRARY=__MPI '
NOOPT_FLAG=' -O0 '
NOOPT_OBJS=' control.o memory.o freem.o'
AR='/usr/bin/ar ruv'
RANLIB='/usr/bin/ranlib'
if [ $debug ]; then
CC='xlc_r -c -g -C -qflttrap'
FC='/usr/local/mpich/1.2.6..0.94/mx/ppc64/smp/ibmcmp64/ssh/bin/mpif77 -c
-g -C -qflttrap'
LD='/usr/local/mpich/1.2.6..0.94/mx/ppc64/smp/ibmcmp64/ssh/bin/mpif77 -g
-C -qflttrap'
else
CC='xlc_r -c -O3 -qstrict'
FC='/usr/local/mpich/1.2.6..0.94/mx/ppc64/smp/ibmcmp64/ssh/bin/mpif77 -c
-O3 -qstrict'
LD='/usr/local/mpich/1.2.6..0.94/mx/ppc64/smp/ibmcmp64/ssh/bin/mpif77
-O3 -qstrict'
fi
-----------
Error running the code:
****************************************************************
* ATOM MASS RAGGIO NLCC PSEUDOPOTENTIAL *
* C 12.0112 1.2000 NO GOEDECKER S NONLOCAL *
* P NONLOCAL *
* D LOCAL *
* H 1.0080 1.2000 NO S LOCAL *
* Si 28.0860 1.2000 NO GOEDECKER S NONLOCAL *
* P NONLOCAL *
* D LOCAL *
****************************************************************
PARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARA
NCPU NGW NHG PLANES GXRAYS HXRAYS ORBITALS Z-PLANES
0 4570 36483 5 233 929 8 1
1 4569 36527 5 234 930 8 1
2 4567 36529 5 234 930 8 1
3 4567 36531 5 234 930 8 1
4 4567 36539 5 234 930 8 1
5 4569 36547 5 234 930 8 1
6 4569 36533 5 234 930 9 1
7 4569 36533 5 234 930 8 1
8 4569 36527 5 234 930 8 1
9 4569 36533 5 234 930 8 1
10 4567 36537 5 234 930 8 1
11 4567 36535 5 234 930 8 1
12 4565 36543 5 234 930 8 1
13 4565 36541 5 234 930 8 1
14 4565 36537 5 234 930 8 1
15 4569 36531 5 234 930 8 1
16 4570 36537 5 232 930 8 1
17 4570 36558 5 232 932 8 1
18 4572 36550 5 232 932 9 1
19 4572 36526 5 232 932 8 1
20 4574 36538 5 232 932 8 1
21 4572 36538 5 232 932 8 1
22 4568 36544 5 232 932 8 1
23 4566 36538 5 232 932 8 1
G=0 COMPONENT ON PROCESSOR : 0
PARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARA
*** LOADPA| SIZE OF THE PROGRAM IS 11164/ 95108 kBYTES ***
OPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPEN
NUMBER OF CPUS PER TASK 1
OPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPEN
*** RGGEN| SIZE OF THE PROGRAM IS 12296/ 96240 kBYTES ***
************************** SUPERCELL ***************************
SYMMETRY: ORTHORHOMBIC
LATTICE CONSTANT(a.u.): 29.23690
CELL DIMENSION: 29.2369 1.0000 2.0533 .0000 .0000 .0000
VOLUME(OMEGA IN BOHR^3): 51315.70009
LATTICE VECTOR A1(BOHR): 29.2369 .0000 .0000
LATTICE VECTOR A2(BOHR): .0000 29.2369 .0000
LATTICE VECTOR A3(BOHR): .0000 .0000 60.0327
RECIP. LAT. VEC. B1(2Pi/BOHR): .0342 .0000 .0000
RECIP. LAT. VEC. B2(2Pi/BOHR): .0000 .0342 .0000
RECIP. LAT. VEC. B3(2Pi/BOHR): .0000 .0000 .0167
REAL SPACE MESH: 120 120 252
WAVEFUNCTION CUTOFF(RYDBERG): 40.00000
DENSITY CUTOFF(RYDBERG): (DUAL= 4.00) 160.00000
NUMBER OF PLANE WAVES FOR WAVEFUNCTION CUTOFF: 109647
NUMBER OF PLANE WAVES FOR DENSITY CUTOFF: 876835
****************************************************************
*** RINFORCE| SIZE OF THE PROGRAM IS 16856/ 101200 kBYTES ***
*** FFTPRP| SIZE OF THE PROGRAM IS 20940/ 104392 kBYTES ***
GENERATE ATOMIC BASIS SET
C SLATER ORBITALS
2S ALPHA= 1.6083 OCCUPATION= 2.00
2P ALPHA= 1.5679 OCCUPATION= 2.00
H SLATER ORBITALS
1S ALPHA= 1.0000 OCCUPATION= 1.00
Si SLATER ORBITALS
3S ALPHA= 1.6344 OCCUPATION= 2.00
3P ALPHA= 1.4284 OCCUPATION= 2.00
INITIALIZATION TIME: 2.22 SECONDS
*** WFOPTS| SIZE OF THE PROGRAM IS 26228/ 168252 kBYTES ***
*** PHFAC| SIZE OF THE PROGRAM IS 26492/ 197268 kBYTES ***
*** ATOMWF| SIZE OF THE PROGRAM IS 37528/ 227624 kBYTES ***
ATRHO| CHARGE(R-SPACE): 388.000000 (G-SPACE): 388.000000
DSYGV : 2538-2015
The number of elements (ARG NO. 11) in a work array must be greater than
or equal to (2124173152).
DSYGV : 2538-2005
The size of the leading dimension (ARG NO. 5) of an array must be greater
than zero.
DSYGV : 2538-2006
The number of rows (ARG NO. 9) of a matrix must be less than or equal
to the size of the leading dimension (ARG NO. 5) of its array.
DSYGV : 2538-2006
The number of rows (ARG NO. 9) of a matrix must be less than or equal
to the size of the leading dimension (ARG NO. 8) of its array.
DSYGV : 2538-2099
End of input argument error reporting. For more information, refer to
Engineering and Scientific Subroutine Library Guide and Reference
(SA22-7904).
DSYGV : 2538-2604
Execution terminating due to error count for error number 2099.
DSYGV : 2538-2605
Message summary: 2005 - 1
DSYGV : 2538-2605
Message summary: 2006 - 2
DSYGV : 2538-2605
Message summary: 2015 - 1
DSYGV : 2538-2605
Message summary: 2099 - 1
best,
joseph hargitai
j
_______________________________________________
CPMD-list mailing list
CPMD-list at cpmd.org
http://cpmd.org/mailman/listinfo/cpmd-list
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://cpmd.org/pipermail/cpmd-list/attachments/20061208/1711aa99/attachment.html
-------------- next part --------------
_______________________________________________
CPMD-list mailing list
CPMD-list at cpmd.org
http://cpmd.org/mailman/listinfo/cpmd-list
More information about the CPMD-list
mailing list