Re: mpi opteron myrinet

From: Gengbin Zheng (gzheng_at_ks.uiuc.edu)
Date: Fri Dec 31 2004 - 00:08:20 CST

You can just copy the conv-mach-gm2.h and conv-mach-gm2.sh (I assume you
have gm version 2) to charm/src/arch/mpi-linux-amd64 direcotry and then
compile with "gm2" option.
I have checked in these two files to cvs, and you will be able to
download nighly build source tomorrow with this change.

Gengbin

Damon Smith wrote:

>All the charm and namd object files are compiled:
>ELF 64-bit LSB relocatable, AMD x86-64, version 1 (SYSV), not stripped
>
>I suppose I should try the opteron arch for the new version of namd.
>Charm++ doesn't have a myrinet gm option for mpi-linux-amd64 though, so
>I don't think it's going to work properly until that happens.
>
>Thanks again
>
>On Thu, 2004-12-30 at 22:50 -0600, Gengbin Zheng wrote:
>
>
>>Hi,
>>
>> Most of these gcc problem with templates and gcc3.4 have been fixed in
>>the latest version. For build on opteron, you will have to use "opteron"
>>option because it sets up several configurations for compiling on
>>opteron, NAMD won't run even you can compile the binary.
>>
>>I am wondering if there is any additional error messages before the
>>messages you patesed below, I mean the error message given by the actual
>>gcc compiler? Could it because of not compiling object file in 64 bits,
>>use command 'file foo.o' to verify. If this is the reason, use -m64
>>compiler flag to generate 64 bit obj files.
>>
>> New version of NAMD has Linux-amd64-MPI.arch for compiling with
>>Opteron and MPI.
>>The content of this file is:
>>
>>NAMD_ARCH = Linux-amd64
>>CHARMARCH = mpi-linux-opteron
>>
>>CXX = g++ -DSOCKLEN_T=socklen_t -DNO_STRSTREAM_H
>>CXXOPTS = -O3 -m64 -fexpensive-optimizations -ffast-math
>>CC = gcc
>>COPTS = -O3 -m64 -fexpensive-optimizations -ffast-math
>>
>>You need to change CHARMARCH to your charm version compiled.
>>
>>Gengbin
>>
>>Damon Smith wrote:
>>
>>
>>
>>>Hi all,
>>>
>>>Sorry for this really long email, but as the subject suggests, I had a
>>>go at compiling namd 2.5 on opteron with mpich 1.2.6 and myrinet, using
>>>gcc 3.4. I haven't found good docs on this, I've only seen docs for
>>>net-linux with amd64 and mpi, but no myrinet.
>>>
>>>A lot of the C++ code in charm++ and namd won't compile because of
>>>gcc3.4's changes to the way templates work, but it's a simple fix, I can
>>>post a patch if anyone is interested. The details on the gcc problem
>>>are here:
>>>http://www.dis.com/gnu/gcc/Name-lookup.html
>>>(it won't break the code for old compilers either)
>>>
>>>
>>>Once I got past that, and a lot more mucking around, it's hard to work
>>>out what to do. Charm++ won't compile with:
>>>mpi-linux opteron gm
>>>
>>>It gets the following error:
>>>Fatal Error by charmc in
>>>directory /home/san02/damon/tmp/charm-5.8/mpi-linux-gm-opteron/tmp
>>> Command mpicc -rdynamic -o ../lib_so/libconv-core.so -shared
>>>-L/home/san02/damon/tmp/charm-5.8/mpi-linux-gm-opteron/lib_so
>>>convcore.o conv-conds.o queueing.o msgmgr.o cpm.o cpthreads.o futures.o
>>>cldb.o topology.o random.o debug-conv.o generate.o edgelist.o conv-ccs.o
>>>ccs-builtins.o traceCore.o traceCoreCommon.o converseProjections.o
>>>machineProjections.o quiescence.o isomalloc.o global-nop.o returned
>>>error code 1
>>>charmc exiting...
>>>make[2]: *** [../lib/libconv-core.a] Error 1
>>>
>>>but it will compile with just
>>>mpi-linux gm
>>>
>>>and then namd doesn't appear to have an option for amd64 or opteron with
>>>mpi and myrinet. So I modified the i686-gm-mpi to use -march=opteron,
>>>and it compiles ok, but gets a linker error:
>>>
>>>collect2: ld returned 1 exit status
>>>Fatal Error by charmc in
>>>directory /home/san02/damon/tmp/NAMD_2.5_Source/Linux-i686-MPI
>>> Command mpiCC -rdynamic -O3 -march=opteron -ffast-math -static
>>>-L/home/san02/damon/tmp/fftw-3.0.1/lib
>>>-L/home/san02/damon/fftw-linux/lib
>>>-L/home/san02/damon/tmp/plugins//LINUXAMD64
>>>-L/home/san02/damon/tmp/plugins//LINUXAMD64/molfile
>>>-L/home/san02/damon/tmp/plugins/compile/LINUXAMD64
>>>-L/home/san02/damon/tmp/plugins/compile/LINUXAMD64/molfile
>>>-I/home/san02/damon/tmp/charm-5.8/mpi-linux-gm/include -o namd2
>>>-L/home/san02/damon/tmp/charm-5.8/mpi-linux-gm/bin/../lib
>>>-I/home/san02/damon/tmp/charm-5.8/mpi-linux-gm/bin/../include /home/san02/damon/tmp/charm-5.8/mpi-linux-gm/bin/../lib/libldb-rand.o obj/buildinfo.o obj/common.o obj/dcdlib.o obj/erf.o obj/main.o
>>>[..snip..]
>>>obj/pub3dfft.o obj/vmdsock.o obj/parm.o obj/imd.o moduleinit.o
>>>-lmoduleNeighborLB
>>>-lmodulecommlib /home/san02/damon/tmp/charm-5.8/mpi-linux-gm/bin/../lib/libmemory-default.o /home/san02/damon/tmp/charm-5.8/mpi-linux-gm/bin/../lib/libthreads-default.o -lck -lconv-cplus-y -lconv-core -lconv-util -lm -lckqt -lpmpich -lgm -lpthread -ldl -lz -lsrfftw -lsfftw -lmolfile_plugin -lm -lmoduleNeighborLB -lmodulecommlib returned error code 1
>>>charmc exiting...
>>>
>>>So that's where I give up. Those linker errors tell me nothing useful.
>>>Has anyone, anywhere got namd to compile with gcc, on opteron, using mpi
>>>and myrinet? If so how??
>>>
>>>I maybe should try getting the latest version of charm from CVS, as it
>>>appears to have changes to the amd64 code as of 6 days ago.
>>>
>>>Thanks,
>>>
>>>Damon
>>>
>>>
>>>
>>>

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:38:06 CST