On Mon, May 6, 2013 at 4:00 PM, Shaohao Chen <shchen at www.phys.lsu.edu>wrote:
> Dear managers and users, > > I got a problem when using PETSc to make parallel codes. Could you please > help me out? > > Basically, I need to do is to update the values of some blocks of a large > matrix (not the whole matrix), > and I need to do it many times in a big loop. These blocks could be > assembled to the same or > different processors. If using "MatSetValues", it would spend much time > for the data transfer between > Are you guessing, or have you measured this? > different processors. I expect "MatSubMatrixUpdate" could do the job > better. Am I correct? But I can > Almost certainly not. It cannot do less communication than MatSetValues(), its just easier sometimes. Matt > not find an example of using "MatSubMatrixUpdate" on the website. It is > said that the > "MatSubMatrixUpdate" is only in a "developer" level and the users should > use some other functions to > replace it. What other functions should I use? > > Thank you for your attention! > > -- > Shaohao Chen > Department of Physics & Astronomy, > Louisiana State University, > Baton Rouge, LA > -- 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/20130506/e4bcbcd6/attachment.html>
