From NWChem
Viewed 366 times, With a total of 4 Posts
|
|
|
Clicked A Few Times
Threads 3
Posts 8
|
|
| 9:33:02 PM - Wed, Jul 20th 2011 |
|
Does anyone have info on building version 6 on an SGI Altix ICE (Dual quad Intel Nehalems, Infiniband interconnect, ifort with SGI or Intel MPI)?
I've tried with ARMCI_NETWORK = MPI-SPAWN, OPENIB, and none at all - always the same error.
setenv NWCHEM_TOP $WORKDIR/nwchem-6.0
setenv NWCHEM_MODULES "all"
setenv USE_MPI y
setenv LIBMPI "-lmpi"
setenv LARGE_FILES TRUE
setenv ARMCI_NETWORK MPI-SPAWN
setenv NWCHEM_TARGET LINUX64
setenv USE_64TO32 y
setenv HAS_BLAS yes
setenv BLASOPT "-lmkl_intel_lp64 -lmkl_sequential -lmkl_core"
alias gmake "make FC=ifort FFLAGS=-O2"
> mpiexec_mpt -n 1 /usr/local/usp/nwchem/6.0/bin/nwchem qmmm_grad0.nw
MA internal error: MAi_inform_base: invalid datatype: 4337916969969
MA internal fatal error: MA_sizeof: unable to set sizes of FORTRAN datatypes
MPI: MPI_COMM_WORLD rank 0 has terminated without calling MPI_Finalize()
MPI: aborting job
Any guidance is appreciated.
Thanks, Chris
|
|
|
-
Bert NWChemDeveloper, bureaucrat, sysop
|
|
|
Gets Around
Threads 1
Posts 173
|
|
| 9:50:30 PM - Wed, Jul 20th 2011 |
|
Chris,
If you do not specify a network, you need to use the parallel command. Actually, for now (as this is network independent I think) you can compile without ARMCI_NETWORK set and run nwchem from the command line (without mpiexec).
Why are you choosing to set USE_64TO32 y ? This option is only necessary when your platform is 64 bit and the optimized BLASS libraries have 32 bit integers. Did you do a "make 64_TO_32" before building?
Looks like the ICE clusters have Mellanox IB cards. See the INSTALL file under the heading "BUILDING AND RUNNING NWCHEM ON INFINIBAND CLUSTERS WITH OPENIB" for environment variables being set.
Bert
Quote:Orleans704 Jul 20th 9:33 pmDoes anyone have info on building version 6 on an SGI Altix ICE (Dual quad Intel Nehalems, Infiniband interconnect, ifort with SGI or Intel MPI)?
I've tried with ARMCI_NETWORK = MPI-SPAWN, OPENIB, and none at all - always the same error.
setenv NWCHEM_TOP $WORKDIR/nwchem-6.0
setenv NWCHEM_MODULES "all"
setenv USE_MPI y
setenv LIBMPI "-lmpi"
setenv LARGE_FILES TRUE
setenv ARMCI_NETWORK MPI-SPAWN
setenv NWCHEM_TARGET LINUX64
setenv USE_64TO32 y
setenv HAS_BLAS yes
setenv BLASOPT "-lmkl_intel_lp64 -lmkl_sequential -lmkl_core"
alias gmake "make FC=ifort FFLAGS=-O2"
> mpiexec_mpt -n 1 /usr/local/usp/nwchem/6.0/bin/nwchem qmmm_grad0.nw
MA internal error: MAi_inform_base: invalid datatype: 4337916969969
MA internal fatal error: MA_sizeof: unable to set sizes of FORTRAN datatypes
MPI: MPI_COMM_WORLD rank 0 has terminated without calling MPI_Finalize()
MPI: aborting job
Any guidance is appreciated.
Thanks, Chris |
|
|
|
|
|
Clicked A Few Times
Threads 3
Posts 8
|
|
| 10:26:24 PM - Wed, Jul 20th 2011 |
|
Thanks Bert. When I unset ARMCI_NETWORK, I could not run with ./nwchem. mpirun was required.
All I see in the doc about OPENIB is to set paths to IB and MPI. On our ICE all that is handled in the module for SGI MPT. All I have to do to com pile a MPI program on this machine is ifort foo.f -lmpi
The 64 to 32 bit thing is a carryover from our Cray, which was in the instructions from the vendor. I don't know if Intel MKL's BLAS is optimzed for 64 bits.
Thanks, Chris
|
|
|
|
|
|
Clicked A Few Times
Threads 3
Posts 8
|
|
| 1:29:30 PM - Fri, Jul 22nd 2011 |
|
MA errors
|
Anyone familiar with these errors?
MA internal error: MAi_inform_base: invalid datatype: 4337916969969
MA internal fatal error: MA_sizeof: unable to set sizes of FORTRAN datatypes
MPI: MPI_COMM_WORLD rank 0 has terminated without calling MPI_Finalize()
MPI: aborting job
Thanks, Chris
|
|
|
-
Bert NWChemDeveloper, bureaucrat, sysop
|
|
|
Gets Around
Threads 1
Posts 173
|
|
| 12:10:47 AM - Wed, Aug 3rd 2011 |
|
To compile a basic single processor version of NWChem, without using the IB network or MPI, unset the MPI variables and the ARMCI_NETWORK. This should generate a single processor version.
Your errors are related to i4/i8 issues. The USE_64TO32 could be a culprit. Did you get a clean source from the NWChem web site?
Bert
Quote:Orleans704 Jul 22nd 1:29 pmAnyone familiar with these errors?
MA internal error: MAi_inform_base: invalid datatype: 4337916969969
MA internal fatal error: MA_sizeof: unable to set sizes of FORTRAN datatypes
MPI: MPI_COMM_WORLD rank 0 has terminated without calling MPI_Finalize()
MPI: aborting job
Thanks, Chris |
|
|
AWC's:
2.5.10 MediaWiki - Stand Alone Forum Extension
Forum theme style by: AWC
