-gideon
> On Mar 10, 2015, at 10:03 AM, Jed Brown <[email protected]> wrote:
>
> Gideon Simpson <[email protected]> writes:
>
>> I had two questions on time stepping routines:
>>
>> 1. What’s the proper usage of TSMonitorSolutionBinary? As near as I can
>> tell from source, I need to call:
>>
>> PetscViewerBinaryOpen(PETSC_COMM_WORLD, "sim.bin",FILE_MODE_WRITE, &viewer);
>>
>> TSMonitorSet(ts, TSMonitorSolutionBinary, viewer,(PetscErrorCode
>> (*)(void**))PetscViewerDestroy);
>>
>> I’m guessing it’s necessary to include the PetscViewerDestroy command?
>
> If you created a viewer, you need to destroy it.
But the destruction of the viewer is handled by the TSMonitor routine, and not
done manually, with a PetscViewerDestroy routine?
>
>> Also, is there a reason why I’m not passing a pointer to the viewer,
>> but rather just the view itself?
>
> All PETSc objects are pointers.
>
> ierr = PetscOptionsString("-ts_monitor_solution_binary","Save each solution
> to a binary
> file","TSMonitorSolutionBinary",0,monfilename,PETSC_MAX_PATH_LEN,&flg);CHKERRQ(ierr);
> if (flg) {
> PetscViewer ctx;
> if (monfilename[0]) {
> ierr =
> PetscViewerBinaryOpen(PetscObjectComm((PetscObject)ts),monfilename,FILE_MODE_WRITE,&ctx);CHKERRQ(ierr);
> ierr = TSMonitorSet(ts,TSMonitorSolutionBinary,ctx,(PetscErrorCode
> (*)(void**))PetscViewerDestroy);CHKERRQ(ierr);
> } else {
> ctx = PETSC_VIEWER_BINARY_(PetscObjectComm((PetscObject)ts));
> ierr = TSMonitorSet(ts,TSMonitorSolutionBinary,ctx,(PetscErrorCode
> (*)(void**))NULL);CHKERRQ(ierr);
> }
> }
What I find confusing about that, though, is that you call
PetscViewerBinaryOpen with &ctx, a pointer to the viewer object, and this makes
sense, because the command is:
PetscErrorCode PetscViewerBinaryOpen(MPI_Comm comm,const char
name[],PetscFileMode type,PetscViewer *binv)
but for TSMonitorSet and TSMonitorSolutionBinary, it’s written as:
PetscErrorCode TSMonitorSet(TS ts,PetscErrorCode
(*monitor)(TS,PetscInt,PetscReal,Vec,void*),void *mctx,PetscErrorCode
(*mdestroy)(void**))
PetscErrorCode TSMonitorSolutionBinary(TS ts,PetscInt step,PetscReal ptime,Vec
u,void *viewer)
which makes it look like i should be passing an &ctx, and not the ct. itself.
But this is a minor point. One works, and one doesn't
>
>> 2. Is there a built in way to track the time steps that the TS solver
>> takes? Assuming it’s using adaptivity, this could be variable. I imagine I
>> could use a TSMonitor routine to print it to the screen, but is there some
>> way to get it into a file?
>
> Do you want output from -ts_monitor or -ts_adapt_monitor? The former
> takes a filename argument and the latter probably should (instead of
> just beeing a boolean).
I was more interested in -ts_monitor. I suppose I just need to do string
processing on that to get the actual times.
