<div style="line-height:1.7;color:#000000;font-size:14px;font-family:arial"><DIV>Hello,</DIV>
<DIV>    I have tried AMG, but there are some problems. I use the command:<BR>    mpiexec -n 4 ./ex4f -ksp_type gmres -pc_type gamg -pc_gamg_agg_nsmooths 1 -ksp_gmres_restart 170 -ksp_rtol 1.0e-15 -ksp_converged_reason.<BR>    The matrix has a size of 30000. However, compared with -pc_type asm,<BR>the amg need more time:asm needs 4.9s, amg needs 13.7s. I did several tests<BR>and got the same conclusion. When it begins, the screen shows the information: <BR>[0]PCSetData_AGG bs=1 MM=7601. I do not know the meaning. And if there is some<BR>parameters that affect the performance of AMG?<BR>    Besides, I want to confirm a conception. In my view, AMG itself can be a solver<BR>like gmres. It can also be used as a preconditioner like jacobi and is used by combining<BR>with other solver. Is it right? If it is right, how use AMG solver?<BR>    My codes are attached.<BR>    Thanks.<BR>  
                                           Jim</DIV>
<DIV><BR>-----------codes------------------------------------------<BR>      call MatCreate(PETSC_COMM_WORLD,A,ierr)<BR>      call MatSetSizes(A,PETSC_DECIDE,PETSC_DECIDE,m,n,ierr)<BR>      call MatSetType(A, MATMPIAIJ,ierr)<BR>      call MatSetFromOptions(A,ierr)<BR>!premalloc<BR>      !find the max non-zero numbers of all rows<BR>       maxnonzero=0<BR>       do 19,II=1,m<BR>        !no-zero numbers of this row<BR>        maxnonzeroII=NROWIN(II+1)-NROWIN(II)<BR>        if (maxnonzeroII>maxnonzero) then<BR>        maxnonzero=maxnonzeroII<BR>        endif<BR>  19  continue<BR>     &nbs
 p; call MatMPIAIJSetPreallocation(A,maxnonzero,PETSC_NULL_INTEGER,<BR>     &  maxnonzero,PETSC_NULL_INTEGER,ierr)</DIV>
<DIV>      call MatGetOwnershipRange(A,Istart,Iend,ierr)<BR>!set values per row<BR>       do 10,II=Istart+1,Iend<BR>        !no-zero numbers of this row<BR>        rowNum=NROWIN(II+1)-NROWIN(II)<BR>        <BR>        allocate(nColPerRow(rowNum))<BR>        allocate(valuePerRow(rowNum))<BR>        <BR>        kValStart=NROWIN(II)+1-1<BR>        kValEnd=NROWIN(II)+rowNum-1<BR>        <BR>        !column index<BR>        nColPerRow=NNZIJ(kValStart:kValEnd)-1<BR>        valuePerRow=VALUE(kValStart:kValEnd)<BR>&nb
 sp;       nRow=II-1<BR>        call MatSetValues(A,ione,nRow,rowNum,nColPerRow,valuePerRow,<BR>     &       INSERT_VALUES,ierr)<BR>       deallocate(nColPerRow)<BR>       deallocate(valuePerRow)<BR>  10  continue<BR>      call MatAssemblyBegin(A,MAT_FINAL_ASSEMBLY,ierr)<BR>      call MatAssemblyEnd(A,MAT_FINAL_ASSEMBLY,ierr)<BR>      call VecCreateMPI(PETSC_COMM_WORLD,PETSC_DECIDE,m,b,ierr)<BR>      call VecSetFromOptions(b,ierr)<BR>      call VecDuplicate(b,u,ierr)<BR>      call VecDuplicate(b,x,ierr)<BR>      call KSPCreate(PETSC_COMM_WORLD,ksp,ierr)<BR>      call KSPSetOperators(ksp,A,A,DIF
 FERENT_NONZERO_PATTERN,ierr)<BR>      call PetscOptionsHasName(PETSC_NULL_CHARACTER,'-my_ksp_monitor',  &<BR>     &                    flg,ierr)<BR>      if (flg) then<BR>        call KSPMonitorSet(ksp,MyKSPMonitor,PETSC_NULL_OBJECT,          &<BR>     &                     PETSC_NULL_FUNCTION,ierr)<BR>      endif<BR>      call KSPSetFromOptions(ksp,ierr)<BR>      call PetscOptionsHasName(PETSC_NULL_CHARACTER,                    
 &<BR>     &     '-my_ksp_convergence',flg,ierr)<BR>      if (flg) then<BR>        call KSPSetConvergenceTest(ksp,MyKSPConverged,                  &<BR>     &          PETSC_NULL_OBJECT,PETSC_NULL_FUNCTION,ierr)<BR>      endif<BR>      !Assing values to 'b'<BR>      bTemp=Iend-Istart<BR>      ioneb=1<BR>      <BR>      do 12,II=Istart,Iend-1<BR>            voneb=F(II+1)<BR>         call VecSetValues(b,ioneb,II,voneb,INSERT_VALUES,ierr)<BR>  12    continue
 </DIV>
<DIV>      call VecAssemblyBegin(b,ierr)<BR>      call VecAssemblyEnd(b,ierr)<BR>      call KSPSolve(ksp,b,x,ierr)<BR>------codes-------------------------------------------------</DIV>
<DIV> </DIV>
<DIV><BR><BR><BR><BR><BR><BR> </DIV>
<DIV></DIV>
<DIV id="divNeteaseMailCard"></DIV>
<DIV><BR></DIV>>At 2012-11-01 22:00:28,"Jed Brown" <jedbrown@mcs.anl.gov> wrote:<BR>
<BLOCKQUOTE style="BORDER-LEFT: #ccc 1px solid; MARGIN: 0px 0px 0px 0.8ex; PADDING-LEFT: 1ex" id="isReplyContent">
<DIV class="gmail_extra">>Yes, it's faster to understand this error message than to have "mysteriously slow performance".</DIV>
<DIV class="gmail_extra"><BR></DIV>
<DIV class="gmail_extra">><EM>* Preallocation routines now automatically set MAT_NEW_NONZERO_ALLOCATION_ERR, if you intentionally preallocate less than necessary then </EM>><EM>use MatSetOption(mat,MAT_NEW_NONZERO_ALLOCATION_ERR,PETSC_FALSE) to disable the error generation.<BR></EM><BR>><A href="http://www.mcs.anl.gov/petsc/documentation/changes/33.html">http://www.mcs.anl.gov/petsc/documentation/changes/33.html</A><BR><BR>
<DIV class="gmail_quote">>On Thu, Nov 1, 2012 at 8:57 AM, w_ang_temp <SPAN dir="ltr"><<A href="mailto:w_ang_temp@163.com" target="_blank">w_ang_temp@163.com</A>></SPAN> wrote:<BR>
<BLOCKQUOTE style="BORDER-LEFT: rgb(204,204,204) 1px solid; MARGIN: 0px 0px 0px 0.8ex; PADDING-LEFT: 1ex" class="gmail_quote">
<DIV>>Do you mean that the two versions have a different in this point? If I use the new version, I have to</DIV>
<DIV>>make some modifications on my codes?</DIV></BLOCKQUOTE></DIV><BR></DIV></BLOCKQUOTE></div><br><br><span title="neteasefooter"><span id="netease_mail_footer"></span></span>