On Mon, May 5, 2014 at 8:22 PM, huaibao zhang <[email protected]>wrote:
> > Matt, > > THanks for the answer. > I think my question is why have to do assembly? > In the piece of my code, 2 processors are inserting the dada to a public > vector soon_n. > This is explained very well in the book Using MPI: http://www.mcs.anl.gov/research/projects/mpi/usingmpi/ If process 0 inserts a value for process 1, then somehow process 1 must be told. That happens in VecAssembly(). Matt > Paul > > > > On May 5, 2014, at 9:18 PM, Matthew Knepley <[email protected]> wrote: > > On Mon, May 5, 2014 at 8:14 PM, huaibao zhang <[email protected]>wrote: > >> >> Hello, >> >> I looked up the manual, but still felt quite confused about why have to >> do assembly. Does it have to do with parallelization? Since all of the >> processors are loading the data at the same time, they need to a pause >> before one can use the whole vector? >> > > If one process sets a value owned by another process, it has to tell it. > > Matt > > >> See a piece of code: >> >> for (int c=0;c<grid[gid].cellCount;++c) { >> row=grid[gid].myOffset+c; >> value=p; >> >> VecSetValues(soln_n,1,&row,&value,INSERT_VALUES); >> } >> VecAssemblyBegin(soln_n); VecAssemblyEnd(soln_n); >> >> >> Thanks, >> Paul >> > > > > -- > 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 > > > -- 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
