[gmx-users] Re: Alpha ev67 cluster

Nuno R. L. Ferreira nunolf at ci.uc.pt
Tue Nov 4 13:38:00 CET 2003

> On Mon, 3 Nov 2003, Nuno R. L. Ferreira wrote:
> > Trying for the first time to compile gmx314 in a cluster!
> What is the OS of your cluster? Last time I checked Digital Unix/Tru64 was
> faster than linux on the same machine. But that was more than a year
> ago...

COMPAQ XP 1000 (processor 21264A at 667 MHz, EV67)
RedHat 7.2 (2.4.9-12 kernel)
gcc version 2.96 20000731 (Red Hat Linux 7.1 2.96-87)
Compaq Fortran Compiler X1.1.1-1684-46B5P
MPICH --> MPIVERSION="1.2.3 of : 2002/01/23 21:03:59"

> > >From what I read, some of you have similiar clusters.
> > Two quick questions.
> > 1) Do you compile fftw213 and gmx with the same F and C compilers?
> > 2) Do you compile with gnu compilers or do you use ccc and fort? No
> > problems?
> In the past ccc en fort were faster, but maybe gcc has improved. The only
> way to find out is to benchmark I guess.

Yes , you're right. But, I had some problems with compilation, that's why
I'm starting from scratch, before anoying you with my problems.
But, here they are.
When I compile fftw213 with CC=ccc, it gives me an error :
(configure with
CC=ccc, --prefix, --exec-prefix, --enable-type-prefix, --enable-mpi, --enabl
Error from make.log

mpicc -O3 -o test_sched test_sched.o .libs/libsfftw_mpi.a -lm
../fftw/.libs/libsfftw.a -lm -lm
../fftw/.libs/libsfftw.a(putils.o): In function `fftw_make_node_rgeneric':
putils.o(.text+0x3bc): undefined reference to `_OtsDivide32'
../fftw/.libs/libsfftw.a(putils.o): In function `fftw_complete_twiddle':
putils.o(.text+0x868): undefined reference to `_OtsDivide32'
putils.o(.text+0x86c): undefined reference to `_OtsDivide32'
../fftw/.libs/libsfftw.a(putils.o)(.text+0x86c): more undefined references
to `_OtsDivide32' follow
../fftw/.libs/libsfftw.a(putils.o): In function `fftw_factor':
putils.o(.text+0xe70): undefined reference to `_OtsRemainder32'
../fftw/.libs/libsfftw.a(putils.o)(.text+0xeb0): more undefined references
to `_OtsRemainder32' follow
../fftw/.libs/libsfftw.a(twiddle.o)(.text+0x144): undefined reference to
../fftw/.libs/libsfftw.a(twiddle.o)(.text+0x22c): more undefined references
to `__cpml_sincos' follow
../fftw/.libs/libsfftw.a(twiddle.o)(.text+0x328): undefined reference to
twiddle.o(.text+0x90c): undefined reference to `__cpml_sincos'
../fftw/.libs/libsfftw.a(twiddle.o)(.text+0x90c): more undefined references
to `__cpml_sincos' follow
collect2: ld returned 1 exit status
make[1]: *** [test_sched] Error 1
make[1]: Leaving directory `/home/biomd/gmx_test/fftw-2.1.3/mpi'
make: *** [all-recursive] Error 1

After that I compiled with gcc, and all went just fine (no problems with
make check also). Also compiled statically gmx314 with gcc, and apparently
also compiled with no problems. After that, I have run some benchmarks (with
PME and cut-off). Curiously, some of them did not write to the output files.
The dppc bench ran OK on more than 1 CPU.
Tried also some runs with my system (protein + solvent) and as before,
grompp works fine without any errors, mdrun starts with no problems, but
then just doesn't write anything to outfiles.

Part of my mdrun log, at the moment it stops writing. The same appens for
e.g. with villin bench.

Initializing LINear Constraint Solver
  number of constraints is 12927
  average number of constraints coupled to one constraint is 2.0

Any insight from you gmx's will be appretiated.


More information about the gromacs.org_gmx-users mailing list