Weird behavior for log_summary
Hi, I am checking the performances of my code via -log_summary, but the output is incomplete (please see attached) file. I configured petsc with the following options: if __name__ == '__main__': import sys import os sys.path.insert(0, os.path.abspath('config')) import configure configure_options = [ '--with-batch=1 ', '--known-mpi-shared=0 ', '--known-mpi-shared-libraries=0', '--known-memcmp-ok ', '--with-blas-lapack-lib=-L/opt/acml/5.3.1/gfortran64/lib -lacml', '--COPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--FOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--CXXOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--with-x=0 ', '--with-debugging=0', '--with-clib-autodetect=0 ', '--with-cxxlib-autodetect=0 ', '--with-fortranlib-autodetect=0 ', '--with-shared-libraries=0 ', '--with-mpi-compilers=1 ', '--with-cc=cc ', '--with-cxx=CC ', '--with-fc=ftn ', # '--with-64-bit-indices', '--download-hypre=1', '--download-blacs=1 ', '--download-scalapack=1 ', '--download-superlu_dist=1 ', '--download-metis=1 ', '--download-parmetis=1 ', ] configure.petsc_configure(configure_options) Any idea about this issue? Thanks, Michele
[NID 04001] 2015-06-04 19:07:24 Apid 25022256: initiated application termination Application 25022256 exit signals: Killed Application 25022256 resources: utime ~271s, stime ~15107s, Rss ~188536, inblocks ~5078831, outblocks ~12517984 Usually this kind of message indicates that either the OS or the batch system killed the process for some reason: often because it ran out of time or maybe memory. Can you run in batch with a request for more time? Do smaller jobs run through ok? If utime means user time and stime means system time then this is very bad, the system time is HUGE relative to the user time. Barry
On Jun 5, 2015, at 9:22 PM, Michele Rosso <[email protected]> wrote:
Hi,
I am checking the performances of my code via -log_summary, but the output is incomplete (please see attached) file. I configured petsc with the following options:
if __name__ == '__main__': import sys import os sys.path.insert(0, os.path.abspath('config')) import configure configure_options = [ '--with-batch=1 ', '--known-mpi-shared=0 ', '--known-mpi-shared-libraries=0', '--known-memcmp-ok ', '--with-blas-lapack-lib=-L/opt/acml/5.3.1/gfortran64/lib -lacml', '--COPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--FOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--CXXOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--with-x=0 ', '--with-debugging=0', '--with-clib-autodetect=0 ', '--with-cxxlib-autodetect=0 ', '--with-fortranlib-autodetect=0 ', '--with-shared-libraries=0 ', '--with-mpi-compilers=1 ', '--with-cc=cc ', '--with-cxx=CC ', '--with-fc=ftn ', # '--with-64-bit-indices', '--download-hypre=1', '--download-blacs=1 ', '--download-scalapack=1 ', '--download-superlu_dist=1 ', '--download-metis=1 ', '--download-parmetis=1 ', ] configure.petsc_configure(configure_options)
Any idea about this issue? Thanks,
Michele
<log_summary>
Hi Barry, I run a small test case like you suggested: this results in no error, but the problem with log_summary still persists. Please find attached the output of log_summary. Thanks, Michele On Fri, 2015-06-05 at 21:34 -0500, Barry Smith wrote:
[NID 04001] 2015-06-04 19:07:24 Apid 25022256: initiated application termination Application 25022256 exit signals: Killed Application 25022256 resources: utime ~271s, stime ~15107s, Rss ~188536, inblocks ~5078831, outblocks ~12517984
Usually this kind of message indicates that either the OS or the batch system killed the process for some reason: often because it ran out of time or maybe memory.
Can you run in batch with a request for more time? Do smaller jobs run through ok?
If utime means user time and stime means system time then this is very bad, the system time is HUGE relative to the user time.
Barry
On Jun 5, 2015, at 9:22 PM, Michele Rosso <[email protected]> wrote:
Hi,
I am checking the performances of my code via -log_summary, but the output is incomplete (please see attached) file. I configured petsc with the following options:
if __name__ == '__main__': import sys import os sys.path.insert(0, os.path.abspath('config')) import configure configure_options = [ '--with-batch=1 ', '--known-mpi-shared=0 ', '--known-mpi-shared-libraries=0', '--known-memcmp-ok ', '--with-blas-lapack-lib=-L/opt/acml/5.3.1/gfortran64/lib -lacml', '--COPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--FOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--CXXOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--with-x=0 ', '--with-debugging=0', '--with-clib-autodetect=0 ', '--with-cxxlib-autodetect=0 ', '--with-fortranlib-autodetect=0 ', '--with-shared-libraries=0 ', '--with-mpi-compilers=1 ', '--with-cc=cc ', '--with-cxx=CC ', '--with-fc=ftn ', # '--with-64-bit-indices', '--download-hypre=1', '--download-blacs=1 ', '--download-scalapack=1 ', '--download-superlu_dist=1 ', '--download-metis=1 ', '--download-parmetis=1 ', ] configure.petsc_configure(configure_options)
Any idea about this issue? Thanks,
Michele
<log_summary>
On Jun 8, 2015, at 12:16 PM, Michele Rosso <[email protected]> wrote:
Hi Barry,
I run a small test case like you suggested: this results in no error, but the problem with log_summary still persists. Please find attached the output of log_summary.
I cannot explain why nothing gets listed in the log summary. It appears to have used the command line options and hence a Krylov solver. I am dumbfounded why no events are listed in the log_summary. Barry
Thanks, Michele
On Fri, 2015-06-05 at 21:34 -0500, Barry Smith wrote:
[NID 04001] 2015-06-04 19:07:24 Apid 25022256: initiated application termination Application 25022256 exit signals: Killed Application 25022256 resources: utime ~271s, stime ~15107s, Rss ~188536, inblocks ~5078831, outblocks ~12517984
Usually this kind of message indicates that either the OS or the batch system killed the process for some reason: often because it ran out of time or maybe memory.
Can you run in batch with a request for more time? Do smaller jobs run through ok?
If utime means user time and stime means system time then this is very bad, the system time is HUGE relative to the user time.
Barry
On Jun 5, 2015, at 9:22 PM, Michele Rosso <[email protected]> wrote:
Hi,
I am checking the performances of my code via -log_summary, but the output is incomplete (please see attached) file. I configured petsc with the following options:
if __name__ == '__main__': import sys import os sys.path.insert(0, os.path.abspath('config')) import configure configure_options = [ '--with-batch=1 ', '--known-mpi-shared=0 ', '--known-mpi-shared-libraries=0', '--known-memcmp-ok ', '--with-blas-lapack-lib=-L/opt/acml/5.3.1/gfortran64/lib -lacml', '--COPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--FOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--CXXOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--with-x=0 ', '--with-debugging=0', '--with-clib-autodetect=0 ', '--with-cxxlib-autodetect=0 ', '--with-fortranlib-autodetect=0 ', '--with-shared-libraries=0 ', '--with-mpi-compilers=1 ', '--with-cc=cc ', '--with-cxx=CC ', '--with-fc=ftn ', # '--with-64-bit-indices', '--download-hypre=1', '--download-blacs=1 ', '--download-scalapack=1 ', '--download-superlu_dist=1 ', '--download-metis=1 ', '--download-parmetis=1 ', ] configure.petsc_configure(configure_options)
Any idea about this issue? Thanks,
Michele
<log_summary>
<outfile.txt>
Is there any external software I should link at compile time to allow PETSc to provide log infos? Also, is there any define in petscconf.h that controls profiling? Michele On Mon, 2015-06-08 at 12:48 -0500, Barry Smith wrote:
On Jun 8, 2015, at 12:16 PM, Michele Rosso <[email protected]> wrote:
Hi Barry,
I run a small test case like you suggested: this results in no error, but the problem with log_summary still persists. Please find attached the output of log_summary.
I cannot explain why nothing gets listed in the log summary. It appears to have used the command line options and hence a Krylov solver. I am dumbfounded why no events are listed in the log_summary.
Barry
Thanks, Michele
On Fri, 2015-06-05 at 21:34 -0500, Barry Smith wrote:
[NID 04001] 2015-06-04 19:07:24 Apid 25022256: initiated application termination Application 25022256 exit signals: Killed Application 25022256 resources: utime ~271s, stime ~15107s, Rss ~188536, inblocks ~5078831, outblocks ~12517984
Usually this kind of message indicates that either the OS or the batch system killed the process for some reason: often because it ran out of time or maybe memory.
Can you run in batch with a request for more time? Do smaller jobs run through ok?
If utime means user time and stime means system time then this is very bad, the system time is HUGE relative to the user time.
Barry
On Jun 5, 2015, at 9:22 PM, Michele Rosso <[email protected]> wrote:
Hi,
I am checking the performances of my code via -log_summary, but the output is incomplete (please see attached) file. I configured petsc with the following options:
if __name__ == '__main__': import sys import os sys.path.insert(0, os.path.abspath('config')) import configure configure_options = [ '--with-batch=1 ', '--known-mpi-shared=0 ', '--known-mpi-shared-libraries=0', '--known-memcmp-ok ', '--with-blas-lapack-lib=-L/opt/acml/5.3.1/gfortran64/lib -lacml', '--COPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--FOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--CXXOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--with-x=0 ', '--with-debugging=0', '--with-clib-autodetect=0 ', '--with-cxxlib-autodetect=0 ', '--with-fortranlib-autodetect=0 ', '--with-shared-libraries=0 ', '--with-mpi-compilers=1 ', '--with-cc=cc ', '--with-cxx=CC ', '--with-fc=ftn ', # '--with-64-bit-indices', '--download-hypre=1', '--download-blacs=1 ', '--download-scalapack=1 ', '--download-superlu_dist=1 ', '--download-metis=1 ', '--download-parmetis=1 ', ] configure.petsc_configure(configure_options)
Any idea about this issue? Thanks,
Michele
<log_summary>
<outfile.txt>
Michele Rosso <[email protected]> writes:
Is there any external software I should link at compile time to allow PETSc to provide log infos? Also, is there any define in petscconf.h that controls profiling?
There is PETSC_USE_LOG, but that should always be on, in which case everything is run-time rather than compile-time. When you run a PETSc example, does the log_summary correctly contain events? Is it possible that your code is turning off events somewhere?
Jed, In the petscconf.h I have #ifndef PETSC_USE_LOG #define PETSC_USE_LOG 1 #endif so I guess that is not the problem. I run ex50: I attached the output. It does prints the summary. The I guess there is something wrong with my code. I call mpi_init before petsc_initialize and then I finalize everything with petsc_finalize. I do not believe I added any other command that could turn off the log_summary output. Thanks, Michele On Mon, 2015-06-08 at 12:02 -0600, Jed Brown wrote:
Michele Rosso <[email protected]> writes:
Is there any external software I should link at compile time to allow PETSc to provide log infos? Also, is there any define in petscconf.h that controls profiling?
There is PETSC_USE_LOG, but that should always be on, in which case everything is run-time rather than compile-time. When you run a PETSc example, does the log_summary correctly contain events? Is it possible that your code is turning off events somewhere?
Michele Rosso <[email protected]> writes:
Jed,
In the petscconf.h I have
#ifndef PETSC_USE_LOG #define PETSC_USE_LOG 1 #endif
so I guess that is not the problem. I run ex50: I attached the output. It does prints the summary. The I guess there is something wrong with my code. I call mpi_init before petsc_initialize and then I finalize everything with petsc_finalize.
Uh, there is no petsc_finalize (it's spelled PetscFinalize), but remember that it doesn't finalize MPI unless PetscInitialize called MPI_Init.
I do not believe I added any other command that could turn off the log_summary output.
Then incrementally figure out what is different between your code and the example. Or use a debugger to track down the difference.
Ok, I will go step by step and see what is wrong. Just to make sure: if I initialize with mpi_init and then call petscinitialize and finally petscfinalize without calling mpi_finalize, petsc should still print the output of log_summary, correct? Michele On Mon, 2015-06-08 at 12:54 -0600, Jed Brown wrote:
Michele Rosso <[email protected]> writes:
Jed,
In the petscconf.h I have
#ifndef PETSC_USE_LOG #define PETSC_USE_LOG 1 #endif
so I guess that is not the problem. I run ex50: I attached the output. It does prints the summary. The I guess there is something wrong with my code. I call mpi_init before petsc_initialize and then I finalize everything with petsc_finalize.
Uh, there is no petsc_finalize (it's spelled PetscFinalize), but remember that it doesn't finalize MPI unless PetscInitialize called MPI_Init.
I do not believe I added any other command that could turn off the log_summary output.
Then incrementally figure out what is different between your code and the example. Or use a debugger to track down the difference.
Michele Rosso <[email protected]> writes:
Ok, I will go step by step and see what is wrong. Just to make sure: if I initialize with mpi_init and then call petscinitialize and finally petscfinalize without calling mpi_finalize, petsc should still print the output of log_summary, correct?
The log_summary printing comes first, but you still need to call MPI_Finalize to clean up MPI correctly.
On Jun 8, 2015, at 12:53 PM, Michele Rosso <[email protected]> wrote:
Is there any external software I should link at compile time to allow PETSc to provide log infos?
No
Also, is there any define in petscconf.h that controls profiling?
Yes, but it would stop anything from being printed, not just the events. Barry
Michele
On Mon, 2015-06-08 at 12:48 -0500, Barry Smith wrote:
On Jun 8, 2015, at 12:16 PM, Michele Rosso <[email protected]> wrote:
Hi Barry,
I run a small test case like you suggested: this results in no error, but the problem with log_summary still persists. Please find attached the output of log_summary.
I cannot explain why nothing gets listed in the log summary. It appears to have used the command line options and hence a Krylov solver. I am dumbfounded why no events are listed in the log_summary.
Barry
Thanks, Michele
On Fri, 2015-06-05 at 21:34 -0500, Barry Smith wrote:
[NID 04001] 2015-06-04 19:07:24 Apid 25022256: initiated application termination Application 25022256 exit signals: Killed Application 25022256 resources: utime ~271s, stime ~15107s, Rss ~188536, inblocks ~5078831, outblocks ~12517984
Usually this kind of message indicates that either the OS or the batch system killed the process for some reason: often because it ran out of time or maybe memory.
Can you run in batch with a request for more time? Do smaller jobs run through ok?
If utime means user time and stime means system time then this is very bad, the system time is HUGE relative to the user time.
Barry
On Jun 5, 2015, at 9:22 PM, Michele Rosso <[email protected]> wrote:
Hi,
I am checking the performances of my code via -log_summary, but the output is incomplete (please see attached) file. I configured petsc with the following options:
if __name__ == '__main__': import sys import os sys.path.insert(0, os.path.abspath('config')) import configure configure_options = [ '--with-batch=1 ', '--known-mpi-shared=0 ', '--known-mpi-shared-libraries=0', '--known-memcmp-ok ', '--with-blas-lapack-lib=-L/opt/acml/5.3.1/gfortran64/lib -lacml', '--COPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--FOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--CXXOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--with-x=0 ', '--with-debugging=0', '--with-clib-autodetect=0 ', '--with-cxxlib-autodetect=0 ', '--with-fortranlib-autodetect=0 ', '--with-shared-libraries=0 ', '--with-mpi-compilers=1 ', '--with-cc=cc ', '--with-cxx=CC ', '--with-fc=ftn ', # '--with-64-bit-indices', '--download-hypre=1', '--download-blacs=1 ', '--download-scalapack=1 ', '--download-superlu_dist=1 ', '--download-metis=1 ', '--download-parmetis=1 ', ] configure.petsc_configure(configure_options)
Any idea about this issue? Thanks,
Michele
<log_summary>
<outfile.txt>
Hi Barry, I accidentally found out what triggers the problem with log_summary. If I specify log_summary from command line, it works as expected. If instead I open and load an options file with -log_summary in it, the output is incomplete. I am actually concerned about using options files now because this weird behavior could affect the correct loading of other options as well. Michele On Fri, 2015-06-05 at 21:34 -0500, Barry Smith wrote:
[NID 04001] 2015-06-04 19:07:24 Apid 25022256: initiated application termination Application 25022256 exit signals: Killed Application 25022256 resources: utime ~271s, stime ~15107s, Rss ~188536, inblocks ~5078831, outblocks ~12517984
Usually this kind of message indicates that either the OS or the batch system killed the process for some reason: often because it ran out of time or maybe memory.
Can you run in batch with a request for more time? Do smaller jobs run through ok?
If utime means user time and stime means system time then this is very bad, the system time is HUGE relative to the user time.
Barry
On Jun 5, 2015, at 9:22 PM, Michele Rosso <[email protected]> wrote:
Hi,
I am checking the performances of my code via -log_summary, but the output is incomplete (please see attached) file. I configured petsc with the following options:
if __name__ == '__main__': import sys import os sys.path.insert(0, os.path.abspath('config')) import configure configure_options = [ '--with-batch=1 ', '--known-mpi-shared=0 ', '--known-mpi-shared-libraries=0', '--known-memcmp-ok ', '--with-blas-lapack-lib=-L/opt/acml/5.3.1/gfortran64/lib -lacml', '--COPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--FOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--CXXOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--with-x=0 ', '--with-debugging=0', '--with-clib-autodetect=0 ', '--with-cxxlib-autodetect=0 ', '--with-fortranlib-autodetect=0 ', '--with-shared-libraries=0 ', '--with-mpi-compilers=1 ', '--with-cc=cc ', '--with-cxx=CC ', '--with-fc=ftn ', # '--with-64-bit-indices', '--download-hypre=1', '--download-blacs=1 ', '--download-scalapack=1 ', '--download-superlu_dist=1 ', '--download-metis=1 ', '--download-parmetis=1 ', ] configure.petsc_configure(configure_options)
Any idea about this issue? Thanks,
Michele
<log_summary>
On Wed, Jul 1, 2015 at 3:52 PM, Michele Rosso <[email protected]> wrote:
Hi Barry,
I accidentally found out what triggers the problem with log_summary. If I specify log_summary from command line, it works as expected. If instead I open and load an options file with -log_summary in it, the output is incomplete.
Same experience here. Many stuffs right before "Memory usage is given in bytes:" do not show if '-log_summary' is specified in input file. Ling
I am actually concerned about using options files now because this weird behavior could affect the correct loading of other options as well.
Michele
On Fri, 2015-06-05 at 21:34 -0500, Barry Smith wrote:
[NID 04001] 2015-06-04 19:07:24 Apid 25022256: initiated application termination Application 25022256 exit signals: Killed Application 25022256 resources: utime ~271s, stime ~15107s, Rss ~188536, inblocks ~5078831, outblocks ~12517984
Usually this kind of message indicates that either the OS or the batch system killed the process for some reason: often because it ran out of time or maybe memory.
Can you run in batch with a request for more time? Do smaller jobs run through ok?
If utime means user time and stime means system time then this is very bad, the system time is HUGE relative to the user time.
Barry
On Jun 5, 2015, at 9:22 PM, Michele Rosso <[email protected]> wrote:> > Hi,> > I am checking the performances of my code via -log_summary, but the output is incomplete (please see attached) file.> I configured petsc with the following options:> > if __name__ == '__main__':> import sys> import os> sys.path.insert(0, os.path.abspath('config'))> import configure> configure_options = [> '--with-batch=1 ',> '--known-mpi-shared=0 ',> '--known-mpi-shared-libraries=0',> '--known-memcmp-ok ',> '--with-blas-lapack-lib=-L/opt/acml/5.3.1/gfortran64/lib -lacml',> '--COPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ',> '--FOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ',> '--CXXOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ',> '--with-x=0 ',> '--with-debugging=0',> '--with-clib-autodetect=0 ',> '--with-cxxlib-autodetect=0 ',> '--with-fortranlib-autodetect=0 ',> '--with-shared-libraries=0 ',> '--with-mpi-compilers=1 ',> '--with-cc=cc ',> '--with-cxx=CC ',> '--with-fc=ftn ',> # '--with-64-bit-indices',> '--download-hypre=1',> '--download-blacs=1 ',> '--download-scalapack=1 ',> '--download-superlu_dist=1 ',> '--download-metis=1 ',> '--download-parmetis=1 ',> ]> configure.petsc_configure(configure_options)> > Any idea about this issue?> Thanks,> > Michele> > > > > <log_summary>
Ahh yes, I see. I have added the following text to PetscOptionsInsertFile() to clarify Usually, instead of using this command, one should list the file name in the call to PetscInitialize(), this insures that certain options such as -log_summary or -malloc_debug are processed properly. This routine only sets options into the options database that will be processed by later calls to XXXSetFromOptions() it should not be used for options listed under PetscInitialize().
On Jul 1, 2015, at 4:52 PM, Michele Rosso <[email protected]> wrote:
Hi Barry,
I accidentally found out what triggers the problem with log_summary. If I specify log_summary from command line, it works as expected. If instead I open and load an options file with -log_summary in it, the output is incomplete. I am actually concerned about using options files now because this weird behavior could affect the correct loading of other options as well.
Michele
On Fri, 2015-06-05 at 21:34 -0500, Barry Smith wrote:
[NID 04001] 2015-06-04 19:07:24 Apid 25022256: initiated application termination Application 25022256 exit signals: Killed Application 25022256 resources: utime ~271s, stime ~15107s, Rss ~188536, inblocks ~5078831, outblocks ~12517984
Usually this kind of message indicates that either the OS or the batch system killed the process for some reason: often because it ran out of time or maybe memory.
Can you run in batch with a request for more time? Do smaller jobs run through ok?
If utime means user time and stime means system time then this is very bad, the system time is HUGE relative to the user time.
Barry
On Jun 5, 2015, at 9:22 PM, Michele Rosso <[email protected]> wrote:
Hi,
I am checking the performances of my code via -log_summary, but the output is incomplete (please see attached) file. I configured petsc with the following options:
if __name__ == '__main__': import sys import os sys.path.insert(0, os.path.abspath('config')) import configure configure_options = [ '--with-batch=1 ', '--known-mpi-shared=0 ', '--known-mpi-shared-libraries=0', '--known-memcmp-ok ', '--with-blas-lapack-lib=-L/opt/acml/5.3.1/gfortran64/lib -lacml', '--COPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--FOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--CXXOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--with-x=0 ', '--with-debugging=0', '--with-clib-autodetect=0 ', '--with-cxxlib-autodetect=0 ', '--with-fortranlib-autodetect=0 ', '--with-shared-libraries=0 ', '--with-mpi-compilers=1 ', '--with-cc=cc ', '--with-cxx=CC ', '--with-fc=ftn ', # '--with-64-bit-indices', '--download-hypre=1', '--download-blacs=1 ', '--download-scalapack=1 ', '--download-superlu_dist=1 ', '--download-metis=1 ', '--download-parmetis=1 ', ] configure.petsc_configure(configure_options)
Any idea about this issue? Thanks,
Michele
<log_summary>
Barry, thank you for the clarification! Best, Michele On Wed, 2015-07-01 at 20:02 -0500, Barry Smith wrote:
Ahh yes, I see. I have added the following text to PetscOptionsInsertFile() to clarify
Usually, instead of using this command, one should list the file name in the call to PetscInitialize(), this insures that certain options such as -log_summary or -malloc_debug are processed properly. This routine only sets options into the options database that will be processed by later calls to XXXSetFromOptions() it should not be used for options listed under PetscInitialize().
On Jul 1, 2015, at 4:52 PM, Michele Rosso <[email protected]> wrote:
Hi Barry,
I accidentally found out what triggers the problem with log_summary. If I specify log_summary from command line, it works as expected. If instead I open and load an options file with -log_summary in it, the output is incomplete. I am actually concerned about using options files now because this weird behavior could affect the correct loading of other options as well.
Michele
On Fri, 2015-06-05 at 21:34 -0500, Barry Smith wrote:
[NID 04001] 2015-06-04 19:07:24 Apid 25022256: initiated application termination Application 25022256 exit signals: Killed Application 25022256 resources: utime ~271s, stime ~15107s, Rss ~188536, inblocks ~5078831, outblocks ~12517984
Usually this kind of message indicates that either the OS or the batch system killed the process for some reason: often because it ran out of time or maybe memory.
Can you run in batch with a request for more time? Do smaller jobs run through ok?
If utime means user time and stime means system time then this is very bad, the system time is HUGE relative to the user time.
Barry
On Jun 5, 2015, at 9:22 PM, Michele Rosso <[email protected]> wrote:
Hi,
I am checking the performances of my code via -log_summary, but the output is incomplete (please see attached) file. I configured petsc with the following options:
if __name__ == '__main__': import sys import os sys.path.insert(0, os.path.abspath('config')) import configure configure_options = [ '--with-batch=1 ', '--known-mpi-shared=0 ', '--known-mpi-shared-libraries=0', '--known-memcmp-ok ', '--with-blas-lapack-lib=-L/opt/acml/5.3.1/gfortran64/lib -lacml', '--COPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--FOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--CXXOPTFLAGS=-march=bdver1 -fopenmp -O3 -ffast-math -fPIC ', '--with-x=0 ', '--with-debugging=0', '--with-clib-autodetect=0 ', '--with-cxxlib-autodetect=0 ', '--with-fortranlib-autodetect=0 ', '--with-shared-libraries=0 ', '--with-mpi-compilers=1 ', '--with-cc=cc ', '--with-cxx=CC ', '--with-fc=ftn ', # '--with-64-bit-indices', '--download-hypre=1', '--download-blacs=1 ', '--download-scalapack=1 ', '--download-superlu_dist=1 ', '--download-metis=1 ', '--download-parmetis=1 ', ] configure.petsc_configure(configure_options)
Any idea about this issue? Thanks,
Michele
<log_summary>
participants (4)
-
Barry Smith -
Jed Brown -
Michele Rosso -
Zou (Non-US), Ling