[petsc-users] MatFDColorCreate takes really big portion of the total time

[Bao Kai]

Hi, all,

I use the SNES in petsc-3.2 to solve my problem. The problem is a 3-D
finite difference problem with structured grid.

I use MatFDColorCreate to generate the Jacobian matrix. I just found
that when the size of problem is big, MatFDColorCreate takes really
long time.  The following results is the summary with size of the mesh
to be 1000^3. 90% of the time is costed in MatFDColorCreate.

    237 MatGetOrdering         1 1.0 1.3502e-03 1.1 0.00e+00 0.0
0.0e+00 0.0e+00 2.0e+00  0  0  0  0  0   0  0  0  0  0     0
    238 MatZeroEntries        39 1.0 9.2822e-02 1.2 0.00e+00 0.0
0.0e+00 0.0e+00 0.0e+00  0  0  0  0  0   0  0  0  0  0     0
    239 MatFDColorCreate      10 1.0 3.2863e+03 1.0 0.00e+00 0.0
1.6e+07 5.0e+02 6.8e+02 90  0  0  0  3  90  0  0  0  3     0
    240 MatFDColorApply       20 1.0 2.5288e+01 1.0 3.54e+07 1.1
4.6e+08 2.0e+03 8.0e+01  1  0  5  5  0   1  0  5  5  0 42708
    241 MatFDColorFunc       560 1.0 9.5386e+00 1.3 0.00e+00 0.0
4.5e+08 2.0e+03 0.0e+00  0  0  5  5  0   0  0  5  5  0     0

And the following the code I use.

    262        call DMGetMatrix(solv%da, MATAIJ, solv%jac,ierr)
    263        call
DMGetColoring(solv%da,IS_COLORING_GLOBAL,MATAIJ,iscoloring,ierr)
    264        call MatFDColoringCreate(solv%jac,iscoloring,matfdcoloring,ierr)
    265        call
MatFDColoringSetFunction(matfdcoloring,FormFunction,equ,ierr)
    266        call MatFDColoringSetFromOptions(matfdcoloring,ierr)
    267        call SNESSetJacobian(solv%snes, solv%jac,
solv%jac,SNESDefaultComputeJacobianColor, matfdcoloring, ierr)
    268        call ISColoringDestroy(iscoloring,ierr)

I am wondering if there is anything I can do to improve this problem.

Thank you very much.

Best Regards,
Kai