AMBER Archive (2007)

Subject: Re: AMBER: Blue Gene, compile Amber9?

From: Robert Duke (rduke_at_email.unc.edu)
Date: Tue Apr 10 2007 - 10:42:11 CDT

Hi Mike -
I don't have access to a current BG/L machine; the instructions for config were done for a machine maintained by ibm about a year ago (by me). This looks like it is due to a change in function layout in the default system libraries; something is being sucked in by the system that references something else in the system that is not being sucked in. So you have to either go to doc or the system people to see how you are supposed to reference system libraries for this particular installation/system release. This is always the fun stuff - has nothing to do with the applications software, but with the system. Good luck; maybe Ross can help you some since he does have access to a BG/L machine (note that although you found a link to this stuff at SDSC, it is actually my stuff from dev/test that I did at the ibm BG/L center in Rochester. Carlos Sosa may also know the answer to this one since he supports these machines at ibm (cpsosa_at_msi.umn.edu).
Regards - Bob Duke
  ----- Original Message -----
  From: Mike Hanby
  To: amber_at_scripps.edu
  Sent: Tuesday, April 10, 2007 11:13 AM
  Subject: RE: AMBER: Blue Gene, compile Amber9?

  I'm attempting to use the PMEMD config.h from the scripps web site:

  http://amber.scripps.edu/config.bgl.html

   

  The compilation starts complaining on line 213 of the attached output about "undefined refrence to `__ctype_b'

   

  1501-510 Compilation successful for file erfcfun.f90.

  blrts_xlf90 -o pmemd gbl_constants.o gbl_datatypes.o state_info.o file_io_dat.o parallel_dat.o mdin_ctrl_dat.o mdin_ewald_dat.o prmtop_dat.o inpcrd_dat.o dynamics_dat.o img.o parallel.o pme_direct.o pme_recip.o pme_fft.o fft1d.o bspline.o pme_force.o pbc.o nb_pairlist.o cit.o dynamics.o bonds.o angles.o dihedrals.o runmd.o loadbal.o shake.o runmin.o constraints.o axis_optimize.o gb_ene.o veclib.o gb_force.o timers.o pmemd_lib.o runfiles.o file_io.o bintraj.o pmemd_clib.o pmemd.o random.o degcnt.o erfcfun.o nmr_calls.o nmr_lib.o get_cmdline.o master_setup.o alltasks_setup.o pme_setup.o ene_frc_splines.o nextprmtop_section.o -L/bgl/BlueLight/ppcfloor/bglsys/lib -lmpich.rts -lmsglayer.rts -lrts.rts -ldevices.rts -lm

  /opt/ibmcmp/xlf/bg/10.1/blrts_lib/libxlf90.a(beginio.o): In function `_xlfBeginIO':

  /build/neptune/101/xlf.101.bgl-prod-opt-bgl/xlf/rte2/src/io/shr/beginio.c:(.text+0x23e): undefined reference to `__ctype_b'

  /build/neptune/101/xlf.101.bgl-prod-opt-bgl/xlf/rte2/src/io/shr/beginio.c:(.text+0x103a): undefined reference to `__ctype_b'

   

  I've attached the "make install 2>&1 |tee pmemd-make.log" output and my config.h file

   

  Does anyone have any suggestions?

   

------------------------------------------------------------------------------

  From: owner-amber_at_scripps.edu [mailto:owner-amber_at_scripps.edu] On Behalf Of Mike Hanby
  Sent: Tuesday, April 10, 2007 09:04
  To: amber_at_scripps.edu
  Subject: RE: AMBER: Blue Gene, compile Amber9?

   

  Thanks for the feedback Ross and Carlos, I'll compile just the PMEMD component then. And thanks for the link to the SDSC page :-)

   

------------------------------------------------------------------------------

  From: owner-amber_at_scripps.edu [mailto:owner-amber_at_scripps.edu] On Behalf Of Ross Walker
  Sent: Friday, April 06, 2007 16:24
  To: amber_at_scripps.edu
  Subject: RE: AMBER: Blue Gene, compile Amber9?

   

  Dear Mike,

   

  There is little point trying to run the rest of Amber on Blue Gene. The scaling of sander is not sufficient to be efficient on Blue Gene and the memory requirements will likely stop you running any decent size systems.

   

  PMEMD scales much better than sander and requires less memory. The memory usage also goes down with the cpu count making it more suitable for Blue Gene. Note PMEMD 9 supports both PME simulations and implicit solvent GB simulations.

   

  I attach the config.h I used to build pmemd 9 on SDSC's Blue Gene. Try the one on the Amber website first though since this may have been optimized by Bob Duke and give better performance. Benchmarks and job submission scripts are here: http://coffee.sdsc.edu/rcw/amber_sdsc/

   

  Note: Blue Gene is not a machine that was actively targetted by the Amber developers during the Amber 9 development and so the code has not had any specific optimizations for Blue Gene. Amber 10's PMEMD may contain some Blue Gene tweaks that will improve the performance but this won't be released until next year. Bob Duke may be able to comment more.

   

  All the best

  Ross

  /\
  \/
  |\oss Walker

  | HPC Consultant and Staff Scientist |
  | San Diego Supercomputer Center |
  | Tel: +1 858 822 0854 | EMail:- ross_at_rosswalker.co.uk |
  | http://www.rosswalker.co.uk | PGP Key available on request |

  Note: Electronic Mail is not secure, has no guarantee of delivery, may not be read every day, and should not be used for urgent or sensitive issues.

   

     

----------------------------------------------------------------------------

    From: owner-amber_at_scripps.edu [mailto:owner-amber_at_scripps.edu] On Behalf Of Mike Hanby
    Sent: Friday, April 06, 2007 10:23
    To: amber_at_scripps.edu
    Subject: AMBER: Blue Gene, compile Amber9?

    Howdy,

     

    Has anyone compiled Amber9 for use on an IBM Blue Gene/L (bgl) system that wouldn't mind sharing the steps they used to cross compile the code using the blrts_xl compilers?

     

    I found one section on the Amber site that provides a config.h file for compiling pmemd but nothing to do with the rest of Amber.

     

    Thanks, Mike

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu