[petsc-users] passing information to TSIFunction

[Sanjay Kharche]

Hi 

Thanks for that. I have since resolved my issue for purpose. I will however implement this suggestion by Jed as it seems more efficient and more consistent with the Petsc paradigm.

cheers
Sanjay


________________________________________
From: Jed Brown [jed at jedbrown.org]
Sent: 09 February 2015 04:19
To: Sanjay Kharche; petsc-users at mcs.anl.gov
Subject: Re: [petsc-users] passing information to TSIFunction

Sanjay Kharche <Sanjay.Kharche at manchester.ac.uk> writes:

> Hi
>
> I need to pass a 2D array of ints to user defined functions, especially RHS. Ideally, this 2D array is dynamically created at run time to make my application general. Yesterday's discussion is below.
>
> I did this in the application context:
>
> /* User-defined data structures and routines           */
> /* AppCtx: used by FormIFunction() */
> typedef struct {
>   DM        da; // DM instance in which u, r are placed.
>   PetscInt  geometry[usr_MY][usr_MX]; // This is static, so the whole thing is visible to everybody who gets the context. This is my working solution as of now.
>   Vec        geom; // I duplicate u for this in calling function. VecDuplicate( u , &user.geom). I cannot pass this to RHS function, I cannot access values in geom in the called function.
>   PetscInt **geomet; // I calloc this in the calling function. I cannot access the data in RHS function
>   int **anotherGeom; // just int.
> } AppCtx;
>
> This static geometry 2D array can be seen in all functions that receive the application context. This is a working solution to my problem, although not ideal. The ideal solution would be if I can pass and receive something like geom or geomet which are dynamically created after the 2D DA is created in the calling function.

Why not create a DMDA (serial and redundant if you prefer) for the
geometry, store the geometry in a Vec, and use DMDAVecGetArray() to
access it?

I like to compose the geometry with the state DM and use DMCoarsenHooks
if using rediscretized geometric multigrid (thus keeping the application
context resolution-independent), but if you're putting
resolution-dependent stuff in the context, you can just add the dmgeom
and vector.

> I am working through the manual and the examples, but some indication of how to solve this specific issue will be great.
>
> cheers
> Sanjay
>
>
>
>
> ________________________________
> From: Matthew Knepley [knepley at gmail.com]
> Sent: 04 February 2015 21:03
> To: Sanjay Kharche
> Cc: petsc-users at mcs.anl.gov
> Subject: Re: [petsc-users] passing information to TSIFunction
>
> On Wed, Feb 4, 2015 at 2:59 PM, Sanjay Kharche <Sanjay.Kharche at manchester.ac.uk<mailto:Sanjay.Kharche at manchester.ac.uk>> wrote:
>
> Hi
>
> I started with the ex15.c example from ts. Now I would like to pass a 2D int array I call data2d to the FormIFunction which constructs the udot - RHS. FormIFunction is used in Petsc's TSSetIFunction. My data2d is determined at run time in the initialisation on each rank. data2d is the same size as the solution array and the residual array.
>
> I tried adding a Vec to FormIFunction, but Petsc's TSIFunction (    TSSetIFunction(ts,r,FormIFunction,&user);  ) expects a set number & type of arguments to FormIFunction. I tried passing data2d as a regular int pointer as well as a Vec. As a Vec, I tried to access the data2d in a similar way as the solution vector, which caused the serial and parallel execution to produce errors.
>
> 1) This is auxiliary data which must come in through the context argument. Many many example use a context
>
> 2) You should read the chapter on DAs in the manual. It describes the data layout. In order for your code to
>     work in parallel I suggest you use a Vec and cast to int when you need the value.
>
>  Thanks,
>
>     Matt
>
> Any ideas on how I can get an array of ints to FormIFunction?
>
> thanks
> Sanjay
>
>
> The function declaration:
> // petsc functions.
> extern PetscInt FormIFunction(TS,PetscReal,Vec,Vec,Vec,void*, Vec); // last Vec is supposed to be my data2D, which is a duplicate of the u.
>
> I duplicate as follows:
>    DMDACreate2d(PETSC_COMM_WORLD, DM_BOUNDARY_NONE, DM_BOUNDARY_NONE,DMDA_STENCIL_STAR,usr_MX,usr_MY,PETSC_DECIDE,PETSC_DECIDE,1,1,NULL,NULL,&da);
>    user.da = da;
>    DMCreateGlobalVector(da,&u);
>    VecDuplicate(u,&r);
>    VecDuplicate(u,&Data2D);  // so my assumption is that data2D is part of da, but I cannot see/set its type anywhere
>
> The warnings/notes at build time:
>
>> make sk2d
> /home/sanjay/petsc/linux-gnu-c-debug/bin/mpicc -o sk2d.o -c -fPIC -Wall -Wwrite-strings -Wno-strict-aliasing -Wno-unknown-pragmas -g3 -O0   -I/home/sanjay/petsc/include -I/home/sanjay/petsc/linux-gnu-c-debug/include    `pwd`/sk2d.c
> /home/sanjay/petscProgs/Work/twod/sk2d.c: In function ‘main’:
> /home/sanjay/petscProgs/Work/twod/sk2d.c:228:4: warning: passing argument 3 of ‘TSSetIFunction’ from incompatible pointer type [enabled by default]
> /home/sanjay/petsc/include/petscts.h:261:29: note: expected ‘TSIFunction’ but argument is of type ‘PetscInt (*)(struct _p_TS *, PetscReal,  struct _p_Vec *, struct _p_Vec *, struct _p_Vec *, void *, struct _p_Vec *)’
>
>
>
>
> --
> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead.
> -- Norbert Wiener