From NWChem
Viewed 2763 times, With a total of 4 Posts
|
|
|
Just Got Here
Threads 1
Posts 4
|
|
| 7:59:28 AM PDT - Thu, Apr 18th 2013 |
|
Hi,
I have tried to build nwchem for threaded execution.
I used the following settings:
- !/bin/bash
...
...
...
export USE_OPENMP=y
export BLASOPT="$MKLLIB/libmkl_intel.a $MKLLIB/libmkl_intel_thread.a $MKLLIB/libmkl_core.a -lpthread -lguide"
export LAPACKOPT="$MKLLIB/libmkl_intel.a $MKLLIB/libmkl_intel_thread.a $MKLLIB/libmkl_core.a -lpthread -lguide"
...
...
...
I have also added "-openmp" in corresponding lines of CFLAGS, FFLAGS, LDFLAGS in the makefile.h.
The compilation is done without any error.
When I try to run a sample case "h3tr1.nw" using:
"OMP_NUM_THREADS=8 nwchem h3tr1.nw"
the run is successful, but "nproc" in out put remains 1.
I do not want to run using "mpi" in a single box.
How can I fix the problem?
Thanks.
Mahmoud
|
|
|
-
Edoapra Forum:Admin, Forum:Mod, bureaucrat, sysop
|
|
|
Forum Vet
Threads 9
Posts 1472
|
|
| 9:25:38 AM PDT - Thu, Apr 18th 2013 |
|
Expected behavior
|
Mahmoud
The nproc printed at the beginning of the NWChem output correctly reports the number of MPI processes spawned by your job.
OMP_NUM_THREADS=8 will simply have 8 threads that will execute each MKL (=BLAS) call used in your NWChem run.
Cheers, Edo
|
|
|
|
|
|
Just Got Here
Threads 1
Posts 4
|
|
| 1:07:22 PM PDT - Thu, Apr 18th 2013 |
|
Edo
You are right. When I compiled using mpi, nproc=8 was correctly shown in output, but the wall time was too high.
So, I preferred to use the openmp. In my run, the threading was not activated because the wall time remains the same as in the case when using the command "nwchem h3tr1.nw". So, I am not sure if I have built in a correct manner.
Could you please comment the way I compiled for openmp.
Thanks, Mahmoud
|
|
|
-
Edoapra Forum:Admin, Forum:Mod, bureaucrat, sysop
|
|
|
Forum Vet
Threads 9
Posts 1472
|
|
| 5:46:28 PM PDT - Fri, Apr 19th 2013 |
|
Mahmoud
I think that you might have compiled correctly for the Openmp parallelization in MKL.
However, the problem might simply be that the input you are using is not make BLAS call that use matrix dimensions large enough to be efficiently spread among the MKL threads.
Edo
|
|
|
|
|
|
Just Got Here
Threads 1
Posts 4
|
|
| 8:21:53 AM PDT - Sun, Apr 21st 2013 |
|
Edo
Thank you very much. I managed to build nwchem suitable to my needs.
Cheers,
Mahmoud
|
|
|
AWC's:
2.5.10 MediaWiki - Stand Alone Forum Extension
Forum theme style by: AWC
