From: Gengbin Zheng (gzheng_at_ks.uiuc.edu)
Date: Tue Aug 10 2004 - 16:54:08 CDT
Hi Randy,
There is a NAMD wiki page about Myrinet at:
http://www.ks.uiuc.edu/Research/namd/wiki/index.cgi?NamdOnMyrinet
NAMD/Charm++ on old GM ver 1 used to work smoothly. Starting from GM 2,
there is some incompatibility problem due to the introduction of pthread
in GM. Charm++ has to switch to a pthread implementation (when use "gm2"
in build command) of user level thread which seems to may or may not
work depending on how GM is configured.
Normally, just compile Charm++ with or without "gm2" build option, one
of which should work.
You may also want to fix the include and library search path. There are
many ways to do this,
You can put everything in build (as in Shantenu's email below ) for example:
./build charm++ mpi-linux-ia64 ecc gm --no-shared -DCMK_OPTIMIZE=1
-I/usr/local/mpich/mpich-gm-1.2.5..10-intel-r2/include
-L/usr/local/mpich/mpich-gm-1.2.5..10- intel-r2/lib/
-L/opt/gm/lib/ -lgm -lpthread -lmpich -lpmpich
-lpthread
to something a little simplified like:
./build charm++ mpi-linux-ia64 ecc gm --no-shared
--basedir /usr/local/mpich/mpich-gm-1.2.5..10-intel-r2
--libdir /opt/gm/lib/
-DCMK_OPTIMIZE=1 -lgm -lpthread -lmpich -lpmpich -lpthread
Or you can modify all compilation flags in:
charm/src/arch/mpi-linux-ia64,
specifically file "conv-mach.sh" (for default build) and "cc-ecc.sh"
(when specifying ecc in build)
Gengbin
Shantenu Jha wrote:
>>Does anyone know if NAMD runs on Itanium2 Linux clusters?
>>
>>
>
>Yes.
>
>I've pasted some notes and the build commands used to compile NAMD
>(with the assistance of Matt Harvey, UCL) on the NCSA teragrid Itanium
>cluster below, but please note, before using the pasted build commands
>locally, that in all probability they'll need some tweaking to deal
>with your specific platform.
>
>[I'd have posted these on the NamdWiki as well, but it isn't accepting
>edits, which maybe related to recent web server problems at
>ks.uiuc.edu].
>
>
>
>>I've looked at the FAQ, which almost implies that NCSA has ported
>>NAMD to their Itanium cluster. But there are no details. Has that
>>happened? Who might I contact to learn more about how it was done?
>>
>>
>
>The NCSA Itanium cluster is part of the teragrid and has a prescribed
>software environment, called the common-teragrid-software stack
>(CTSS). There probably isn't much point going into the details of
>CTSS, as the default environment (accessed via "softenv") is
>documented else where, but it is useful to mention that the default
>compiler is the intel 8.0.
>
>I personally had lots of difficulty getting the compiler to use the
>right version of mpich and maybe something to watch for. Also, the
>teragrid IA64 cluster has myrinet interconnects and I had to
>explicitly load the GM libraries (myricom's addon to mpich) , though I
>got away with GM and not GM2.
>
>I used charm 5.8 eventually, although later I was able to get the
>default charm++ version that ships with NAMD 2.5 working as well.
>
>To build charm++, on the teragrid IA64 use "mpi-linux-ia64 ecc gm"
>platform option. The full command used was:
>
> ./build charm++ mpi-linux-ia64 ecc gm --no-shared -DCMK_OPTIMIZE=1
>-I/usr/local/mpich/mpich-gm-1.2.5..10-intel-r2/include
>-L/usr/local/mpich/mpich-gm-1.2.5..10- intel-r2/lib/
>-L/opt/gm/lib/ -lgm -lpthread -lmpich -lpmpich
>-lpthread
>
>There is at least one other subtelty: either tweak the Makefiles to
>reflect a problem with directory naming; easier to make a softlink to
> mpi-linux-ia64-gm -> mpi-linux-ia64-gm-ecc/
>
>Regarding the simplearrayhello validation program (in pgms) , although
>charm++ compiled, the validation program still required the following:
>
>make OPTS="-L/usr/local/mpich/mpich-gm-1.2.5..10-intel-r2/lib/ -L/opt/gm/lib/ -lmpich
> -lpmpich -lgm -lpthread"
>
>With charm++ built and the validation programs running, you're almost there.
>
>For NAMD I used the Linux-ia64-MPI-GM-ecc option to config (the ecc stands
>for the Intel 64bit compiler, GM for Myrinet specific stuff). I didn't try
>the net-linux version, as we'll be using it on 100's of px , rather than
>10's.
>
>The Makearch file in the directory Linux-ia64-MPI-GM-ecc is,
>
>include .rootdir/Make.charm
>include .rootdir/arch/Linux-ia64-MPI-GM-ecc.arch
>CHARM = $(CHARMBASE)/$(CHARMARCH)
>NAMD_PLATFORM = $(NAMD_ARCH)-MPI-GM
>include .rootdir/arch/$(NAMD_ARCH).base
>include .rootdir/arch/$(NAMD_ARCH).tcl
>include .rootdir/arch/$(NAMD_ARCH).fftw
>include .rootdir/arch/$(NAMD_ARCH).plugins
>
>and you can get the required 64 bit tcl,fftw and plugins from the
>ks.uiuc.edu website (as mentioned in the notes.txt file).
>
>Finally I had to add the following for the generated Makefile:
>
>MPICHLIB=-L/usr/local/mpich/mpich-gm-1.2.5..10-intel-r2/lib/
>-L/opt/gm/lib/ -lmpich -lpmpich -lgm -lpthread
>
>and introduce a $(MPICHLIB) in the dependency list of namd2 target.
>
>Don't know if that helps, but good luck!
>
>Yes, I too had hell getting all this working, but thanks to great
>support locally (Matt Harvey), we was able to get there.. I'm sure the
>charm++/NAMD developers know this, but currently the barrier to
>deployment of source code is almost prohibitively high.
>
>Shantenu
>
>
>
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:38:49 CST