[petsc-users] A very strange bug about using multiple KSPs

[Sun, Hui]

Thank you, Matthew! It does solve the problem.

Best,
Hui


________________________________
From: Matthew Knepley [knepley at gmail.com]
Sent: Tuesday, March 03, 2015 8:26 AM
To: Sun, Hui
Cc: Dave May; petsc-users at mcs.anl.gov
Subject: Re: [petsc-users] A very strange bug about using multiple KSPs

On Tue, Mar 3, 2015 at 10:15 AM, Sun, Hui <hus003 at ucsd.edu<mailto:hus003 at ucsd.edu>> wrote:
Dave, thank you for your suggestions. I don't have a debugger, and I think without debugger we can still figure out if a function is called or not.

Putting the print commend before or after PetscFunctionBeginUser doesn't really matter. I just tried to put it after PetscFunctionBeginUser, but something doesn't print just doesn't print.

The ComputeMatrix function is stored in the DM, or more precisely in the DMKSP object. Since both KSPs have the
same DM, they have the same function.

I believe you can DMClone(da, &da2), and then this will work. Please tell me if it does not.

  Thanks,

    Matt

There are other ways to see that computeMatrix1 is not called. For example, I just change b1 to a vector of 1's, and b2 to a vector of 0.5's, and ComputeMatrix1 to give diagonal matrix with entries 1, and  ComputeMatrix2 to give diagonal matrix with entries 2.

Then I output the matrices and the vectors into a file, and visualize them in MATLAB. It does show me that the ksp1 and ksp2 both use ComputeMatrix2; while the ksp1 uses the rhs b1, and the ksp2 uses the rhs b2. This is not expected because I've set ksp1 to use ComputeMatrix1 before:


    ierr = KSPSetComputeOperators(ksp1,ComputeMatrix1,NULL);CHKERRQ(ierr);

    ierr = KSPSetComputeOperators(ksp2,ComputeMatrix2,NULL);CHKERRQ(ierr);


Now if I comment out the lines


    ierr = KSPSetComputeOperators(ksp2,ComputeMatrix2,NULL);CHKERRQ(ierr);

and


    ierr = KSPSolve(ksp2,b2,x2);CHKERRQ(ierr);


Then ksp1 start to use ComputeMatrix1, as can be visualized from both the output file and the terminal print. So it shows me that there are some crosstalk between ksp1 and ksp2. If I set

    ierr = KSPSetComputeOperators(ksp1,ComputeMatrix1,NULL);CHKERRQ(ierr);

ksp1 uses ComputeMatrix1. But if I set


    ierr = KSPSetComputeOperators(ksp1,ComputeMatrix1,NULL);CHKERRQ(ierr);

    ierr = KSPSetComputeOperators(ksp2,ComputeMatrix2,NULL);CHKERRQ(ierr);

ksp1 uses ComputeMatrix2.



________________________________
From: Dave May [dave.mayhem23 at gmail.com<mailto:dave.mayhem23 at gmail.com>]
Sent: Tuesday, March 03, 2015 1:11 AM
To: Sun, Hui
Cc: petsc-users at mcs.anl.gov<mailto:petsc-users at mcs.anl.gov>
Subject: Re: [petsc-users] A very strange bug about using multiple KSPs

Try moving your print statement in ComputeMatrix2 to be after PetscFunctionBeginUser;

You should really use a debugger to see if these functions are being called or not - don't use print statements.

On Tuesday, 3 March 2015, Sun, Hui <hus003 at ucsd.edu<mailto:hus003 at ucsd.edu>> wrote:
I set up 2 KSPs through KSPSetComputeOperators, with separate compute matrix functions.



    ierr = KSPSetComputeOperators(ksp1,ComputeMatrix1,NULL);CHKERRQ(ierr);

    ierr = KSPSetComputeOperators(ksp2,ComputeMatrix2,NULL);CHKERRQ(ierr);

Within each computeMatrix function, I define the matrix, and I also print something to the screen. So


PetscErrorCode ComputeMatrix1(KSP ksp, Mat J,Mat jac, void *ctx)

{

    ......

    ierr    = PetscPrintf(PETSC_COMM_WORLD,"1\n");CHKERRQ(ierr);

    ......

}


PetscErrorCode ComputeMatrix2(KSP ksp, Mat J,Mat jac, void *ctx)

{

    ......

    ierr    = PetscPrintf(PETSC_COMM_WORLD,"2\n");CHKERRQ(ierr);

    ......

}


Now if I compile and execute the code, it only print 2, completely ignoring 1, which is not expected, since I have setup ksp1 to use ComputeMatrix1.

Does anyone know what might be wrong? Thank you. I have attached the short code here in this email as well.

Hui



--
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
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20150303/c9641834/attachment.html>