[petsc-users] my code runs too slow

[abhyshr at mcs.anl.gov]

If you want to use the PETSc to compute the jacobian via finite differencing then here's the sequence of calls 






ierr = SNESComputeDefaultJacobian(ts_snes,CV_Y,&J,&J,&flag);CHKERRQ(ierr); 
ierr = MatGetColoring(J,MATCOLORINGSL,&iscoloring);CHKERRQ(ierr); 
ierr = MatFDColoringCreate(J,iscoloring,&matfdcoloring);CHKERRQ(ierr); 
ierr = ISColoringDestroy(iscoloring);CHKERRQ(ierr); ...you missed this 
ierr = MatFDColoringSetFunction(matfdcoloring,(PetscErrorCode (*)(void))f,(void*)&appctx);CHKERRQ(ierr); 
ierr = MatFDColoringSetFromOptions(matfdcoloring);CHKERRQ(ierr); 
ierr = TSSetRHSJacobian(ts,J,J,TSDefaultComputeJacobianColor,matfdcoloring); 




if you are using your own jacobian evaluation then use SNESSetJacobian followed by SNESComputeJacobian instead of SNESComputeDefaultJacobian. 


Shri 

----- "Xuan YU" <xxy113 at psu.edu> wrote: 
> ierr = MatCreateSeqAIJ(PETSC_COMM_SELF,N,N,10,PETSC_NULL,&J);CHKERRQ(ierr); 

ierr = MatAssemblyBegin(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 
ierr = MatAssemblyEnd(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 

ierr = SNESComputeJacobian(ts_snes,CV_Y,&J,&J,&flag);CHKERRQ(ierr); 
ierr = MatGetColoring(J,MATCOLORINGSL,&iscoloring);CHKERRQ(ierr); 
ierr = MatFDColoringCreate(J,iscoloring,&matfdcoloring);CHKERRQ(ierr); 
ierr = MatFDColoringSetFunction(matfdcoloring,(PetscErrorCode (*)(void))f,(void*)&appctx);CHKERRQ(ierr); 
ierr = MatFDColoringSetFromOptions(matfdcoloring);CHKERRQ(ierr); 
ierr = TSSetRHSJacobian(ts,J,J,TSDefaultComputeJacobianColor,matfdcoloring); 


> 
These are the Jacobian related codes. 

> 

> 

> 

> 

On Jul 7, 2010, at 1:51 PM, Satish Balay wrote: 





total: nonzeros=1830 
> 

mallocs used during MatSetValues calls =1830 
> 
> Looks like you are zero-ing out the non-zero structure - before 
> assembling the matrix. 
> 
> Are you calling MatZeroRows() or MatZeroEntries() or something else - 
> before assembling the matrix? 
> 
> Satish 
> 
> On Wed, 7 Jul 2010, Xuan YU wrote: 
> 
> 

I made a change: ierr = 
> 

MatCreateSeqAIJ(PETSC_COMM_SELF,N,N,5,PETSC_NULL,&J);CHKERRQ(ierr); 
> 


> 

Time of the code did not change much, and got the info: 
> 

Matrix Object: 
> 

type=seqaij, rows=1830, cols=1830 
> 

total: nonzeros=1830, allocated nonzeros=36600 
> 

total number of mallocs used during MatSetValues calls =1830 
> 

not using I-node routines 
> 


> 


> 


> 

On Jul 7, 2010, at 12:51 PM, Satish Balay wrote: 
> 


> 





total: nonzeros=1830, allocated nonzeros=29280 
> 





total number of mallocs used during MatSetValues calls =1830 
> 




> 



There is something wrong with your preallocation or matrix 
> 



assembly. You should see zero mallocs for efficient assembly. 
> 




> 



http://www.mcs.anl.gov/petsc/petsc-as/documentation/faq.html#efficient-assembly 
> 




> 



satish 
> 




> 




> 



On Wed, 7 Jul 2010, Xuan YU wrote: 
> 




> 





Hi, 
> 






> 





I finite difference Jacobian approximation for my TS model. The size of 
> 





the 
> 





vector is 1830. I got the following info with(-ts_view): 
> 






> 





type: beuler 
> 





maximum steps=50 
> 





maximum time=50 
> 





total number of nonlinear solver iterations=647 
> 





total number of linear solver iterations=647 
> 





SNES Object: 
> 





type: ls 
> 





line search variant: SNESLineSearchCubic 
> 





alpha=0.0001, maxstep=1e+08, minlambda=1e-12 
> 





maximum iterations=50, maximum function evaluations=10000 
> 





tolerances: relative=1e-08, absolute=1e-50, solution=1e-08 
> 





total number of linear solver iterations=50 
> 





total number of function evaluations=51 
> 





KSP Object: 
> 





type: gmres 
> 





GMRES: restart=30, using Classical (unmodified) Gram-Schmidt 
> 





Orthogonalization with no iterative refinement 
> 





GMRES: happy breakdown tolerance 1e-30 
> 





maximum iterations=10000, initial guess is zero 
> 





tolerances: relative=1e-05, absolute=1e-50, divergence=10000 
> 





left preconditioning 
> 





using PRECONDITIONED norm type for convergence test 
> 





PC Object: 
> 





type: ilu 
> 





ILU: out-of-place factorization 
> 





0 levels of fill 
> 





tolerance for zero pivot 1e-12 
> 





using diagonal shift to prevent zero pivot 
> 





matrix ordering: natural 
> 





factor fill ratio given 1, needed 1 
> 





Factored matrix follows: 
> 





Matrix Object: 
> 





type=seqaij, rows=1830, cols=1830 
> 





package used to perform factorization: petsc 
> 





total: nonzeros=1830, allocated nonzeros=1830 
> 





total number of mallocs used during MatSetValues calls =0 
> 





not using I-node routines 
> 





linear system matrix = precond matrix: 
> 





Matrix Object: 
> 





type=seqaij, rows=1830, cols=1830 
> 





total: nonzeros=1830, allocated nonzeros=29280 
> 





total number of mallocs used during MatSetValues calls =1830 
> 





not using I-node routines 
> 






> 






> 





50 output time step takes me 11.877s. So I guess there is something not 
> 





appropriate with my Jacobian Matrix. Could you please tell me how to speed 
> 





up 
> 





my code? 
> 






> 





Thanks! 
> 






> 





Xuan YU 
> 





xxy113 at psu.edu 
> 






> 






> 






> 






> 




> 




> 


> 

Xuan YU (俞烜) 
> 

xxy113 at psu.edu 
> 


> 


> 


> 


> 
> 

> 
> 
Xuan YU ( 俞烜 ) 
xxy113 at psu.edu 

> 


>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20100707/31079bb5/attachment.htm>