On Nov 9, 2010, at 11:56 AM, Rongliang Chen wrote: > Yes, I know. So this is why I need to use MAT_FINAL_ASSEMBLY instead of > MAT_FLUSH_ASSEMBLY in the function MatAssemblyBegin/End when switching > between ADD_VALUES and INSERT_VALUES in MatSetValues(). > But when I use the finally assembled matrix to multiply a vector using > MatMult. I find that the output of MatMult is wrong if I use > multiprocessors. So I doubt that the problem is come from > "MAT_FINAL_ASSEMBLY instead of MAT_FLUSH_ASSEMBLY", but I am not sure. Thanks.
Well the problem is either in your code or in PETSc. Send to petsc-maint at mcs.anl.gov a code that demonstrates the problem and we'll look into the problem. Barry > > Regards, > > Rongliang > > On Tue, Nov 9, 2010 at 10:31 AM, Barry Smith <bsmith at mcs.anl.gov> wrote: > > On Nov 9, 2010, at 11:27 AM, Rongliang Chen wrote: > > > Message: 2 > > Date: Mon, 8 Nov 2010 19:57:16 -0600 > > From: Barry Smith <bsmith at mcs.anl.gov> > > Subject: Re: [petsc-users] Can I use MAT_FINAL_ASSEMBLY instead of > > MAT_FLUSH_ASSEMBLY? > > To: PETSc users list <petsc-users at mcs.anl.gov> > > Message-ID: <52AFD8FB-17C1-4B51-97E9-C77112B69F3C at mcs.anl.gov> > > Content-Type: text/plain; charset=us-ascii > > > > > > On Nov 8, 2010, at 7:49 PM, Rongliang Chen wrote: > > > > > Hi everyone, > > > > > > Can I use MAT_FINAL_ASSEMBLY instead of MAT_FLUSH_ASSEMBLY in the > > > function MatAssemblyBegin/End when switching between ADD_VALUES and > > > INSERT_VALUES in MatSetValues()? Thank you! > > > > > > Regards, > > > > > > Rongliang > > > > Yes, but generally there is no reason to. It is best to use the flush until > > your final assembly because then it doesn't do any work that needs to be > > redone or prematurely free memory that may be needed later. > > > > Why do you want to do that? > > > > Barry > > > > > > > > Hi Barry, > > > > Thank you for your reply. I find that when I use MAT_FINAL_ASSEMBLY instead > > of MAT_FLUSH_ASSEMBLY in the function MatAssemblyBegin/End, it will > > cause problem. I do not know why. > > "cause problem" is not very helpful. What does that mean? Send all output > to petsc-maint at mcs.anl.gov if it doesn't work. > > > > The reason that I want to use it like this is that I first set the matrix > > with ADD_VALUES and then I need to zero some rows of the matrix using > > MatZeroRowsLocalIS(), at last I still need to set some elements of the > > matrix with INSERT_VALUES. Between MatSetValues() and MatZeroRowsLocalIS() > > I need to use MAT_FINAL_ASSEMBLY. > > The order is: > > ADD_VALUES--->MatZeroRowsLocalIS()--->INSERT_VALUES. > > You need to use FINAL_ASSEMLY before the zero rows calls. Flush is not > enough in that case. > > Barry > > > > > Thanks! > > > > Regards, > > > > Rongliang > >
