[overture] Re: parallel ins with PETSc

  • From: Bill Henshaw <henshaw@xxxxxxxx>
  • To: overture@xxxxxxxxxxxxx
  • Date: Mon, 07 Jan 2008 19:47:25 -0800

Hi Joel:

  You should NOT copy PETScSolver.C nor buildEquationSolvers.C. from the 
Overture directory.
The cg make system will automatically find these files and compile them for you.
All you should need to do is set
   usePETSc   := on
in cg/ins/Makefile. When this flag is set the above files will be compiled in 
the correct way.

Sorry for the confusion.

...Bill


Joel Guerrero wrote:
Hi Bill,

PETScSolver.C should be also recompile by cg???. So, according to your message, I should change the cg makefile in a way that buildEquationSolvers.C is recompile during cg compilation?. Because so far I'm just copying these two files (PETScSolver.C and buildEquationSolvers.C ) from the overture installation dir.

HAve a nice day,

Joel

----- Original Message ----
From: Bill Henshaw <henshaw@xxxxxxxx>
To: overture@xxxxxxxxxxxxx
Sent: Tuesday, January 8, 2008 3:05:39 PM
Subject: [overture] Re: parallel ins with PETSc


Hi Joel,

  Your error message is different now since Oges cannot find PETScNew
 (which
is the parallel PETSc solver)

 > Oges::buildEquationSolvers:ERROR: solver PETScNew is not currently
 available

Check the file Overture/oges/buildEquationSolvers.C to see where this
 error
occurs. You may need to recompile this file. A new version of this file
 should be
compiled by cg with the compile flag  -DOVERTURE_USE_PETSC, and this
 new version over-rides
the version that is in the Overture library.

...Bill



Joel Guerrero wrote:
Hi,

Sorry for bringing again this subject but I still have problems and
 this is driving me nuts.  With the serial version of cg ins I don't have
 problems linking it with petsc, but with the parallel version I still
 have the same problem that is:
initialize the solution...
 interpolateAndApplyBoundaryConditions (start) steps=-1 t=0: number
 of array ID's has increased to 684
Cgins::project: project the initial conditions <<<<
 $$$$$$$$$$$$$$$ Cgins: updateToMatchGrid(CompositeGrid & cg)
 $$$$$$$$$$$$
Oges::buildEquationSolvers:ERROR: solver PETScNew is not currently
 available
 You may have to copy and edit the file
 Overture/Oges/buildEquationSolvers.C
 and a file like PETScEquationSolver.C (if you are trying to use
 PETSc)
 and then link the files to your application in order to get a
 non-standard solver
 See the Oges documentation for further details
error
Overture::abort: I am now going to purposely abort so that you can
 get a traceback from a debugger
7:31775] [14]
 ../../../../cg_parallel.v22/ins/bin/cgins(_ZN3MPI3Win4FreeEv+0xdd) [0x8059151]
[linux-p8i7:31775] *** End of error message ***
mpirun noticed that job rank 0 with PID 31775 on node linux-p8i7
exited on signal 6 (Aborted).
1 additional process aborted (not shown)


the compilation process goes fine, all the necessary files are
 present and also all the required libraries.   I also checked the shared
 library dependencies and the executable is proper linked to the petsc
 libraries.  At this point I dont know what could be the problem, do I need
 to change something in the source code in order to have the parallel
 petsc solver ????

Regards,

joel

----- Original Message ----
From: Bill Henshaw <henshaw@xxxxxxxx>
To: overture@xxxxxxxxxxxxx
Sent: Sunday, January 6, 2008 6:16:25 PM
Subject: [overture] Re: parallel ins with PETSc


Joel:

 > Oges::buildEquationSolvers:ERROR: solver PETSc is not currently
 available

   It looks like you are trying to use the serial "PETSc" solver
 instead of the
parallel version. Use the "PETScNew" option when specifying the
 solver
 to use
or "choose best iterative solver" should also work.

    pressure solver options
     * PETScNew
     choose best iterative solver

You could also run with gdb and set a break in OgesParameters::set(
 OptionEnum option, int value, real rvalue )
to make sure that the solver is set to PETScNew at some point.

Many cmd files in ins/cmd run in parallel -- grep for "mpirun" in
 *.cmd
 for exmaples

...Bill

Joel Guerrero wrote:
Hi,

The PETSc option is switch on in the makefile.  Also the
 buildEquationSolvers.o and PETScSolver.o are proper linked.  The
 compilation
 process goes fine but when I try to use parallel ins I always get
 the same
 error.  Besides the file cicp.cmd, is there another parallel
 example??
cheers,

jg





 
____________________________________________________________________________________
Looking for last minute shopping deals? Find them fast with Yahoo! Search.
  http://tools.search.yahoo.com/newsearch/category.php?category=shopping






      
____________________________________________________________________________________
Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it now. http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ

Other related posts: