November 2011 - petsc-dev - mailman3.cels.anl.gov

no manual pages for DM subclasses
by Barry Smith 14 Nov '11

14 Nov '11

None of these beasties have manual pages. #define DMDA "da" #define DMADDA "adda" #define DMCOMPOSITE "composite" #define DMSLICED "sliced" #define DMMESH "mesh" #define DMCARTESIAN "cartesian" #define DMIGA "iga" #define DMREDUNDANT "redundant"

1 0

Re: [petsc-dev] DMCartesian vs DMMesh?
by Matthew Knepley 14 Nov '11

14 Nov '11

On Mon, Nov 14, 2011 at 7:21 PM, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > > On Nov 14, 2011, at 1:08 PM, Matthew Knepley wrote: > > > On Mon, Nov 14, 2011 at 6:18 PM, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > > > > > > > > Also since Mesh is a generic word but DMMesh is built on Sieve should > DMMesh be renamed to DMSieve? > > > > I don't like that, because no one will understand what we are talking > about. DMMesh makes sense. I might be > > alright with DMUnstructuredMesh, but that is overkill. > > > > Matt > > DMMesh is kind of like having a PCPRECONDITIONER I hear the criticism. I don't hear the alternative. What other word do people use for an unstructured mesh? Matt > > Barry > > > > > > > Barry > > > > -- > > 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

1 0

Re: [petsc-dev] DMCartesian vs DMMesh?
by Barry Smith 14 Nov '11

14 Nov '11

On Nov 14, 2011, at 1:08 PM, Matthew Knepley wrote: > On Mon, Nov 14, 2011 at 6:18 PM, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > > > > Also since Mesh is a generic word but DMMesh is built on Sieve should DMMesh be renamed to DMSieve? > > I don't like that, because no one will understand what we are talking about. DMMesh makes sense. I might be > alright with DMUnstructuredMesh, but that is overkill. > > Matt DMMesh is kind of like having a PCPRECONDITIONER Barry > > > Barry > > -- > 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

1 0

Re: [petsc-dev] DMCartesian vs DMMesh?
by Matthew Knepley 14 Nov '11

14 Nov '11

On Mon, Nov 14, 2011 at 6:18 PM, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > > What is the relationship between DMCartesian and DMMesh? > Somewhat, but only in the implementation, which is not that important. > Should DMCartesian be removed? > I will get rid of it, when I have the replacement done. What I want here is a Cartesian topology that allows arbitrary dof layout. DMCartesian almost does this in any dimension, but cannot do edges and faces correctly since I could not figure out the gray code for this. Note, that this is not trivial with DMDA, because 1) The communication code would have to be fixed 2) It is not easy (ugly) to construct an FEM iteration 3) It also does nto know about edges and faces > Also since Mesh is a generic word but DMMesh is built on Sieve should > DMMesh be renamed to DMSieve? I don't like that, because no one will understand what we are talking about. DMMesh makes sense. I might be alright with DMUnstructuredMesh, but that is overkill. Matt > > Barry > -- 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

1 0

DMCartesian vs DMMesh?
by Barry Smith 14 Nov '11

14 Nov '11

What is the relationship between DMCartesian and DMMesh? Should DMCartesian be removed? Also since Mesh is a generic word but DMMesh is built on Sieve should DMMesh be renamed to DMSieve? Barry

1 0

Different SNES initial residual semantics?
by Jed Brown 13 Nov '11

13 Nov '11

The initial residuals for snes ex28 have changed (much larger now), but all subsequent convergence history is identical. I don't think anything I was messing with would have caused this.

1 0

Re: [petsc-dev] developer hint
by Jed Brown 13 Nov '11

13 Nov '11

On Sun, Nov 13, 2011 at 16:53, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > But DMDA does not? So it could be that DMSliced is just too much of the > original design and needs updating as DMDA was modernised Yeah, this was a simple fix. Pushed.

1 0

Re: [petsc-dev] removing DMRefine/CoarsenHierarchy?
by Barry Smith 13 Nov '11

13 Nov '11

On Nov 13, 2011, at 8:57 PM, Jed Brown wrote: > On Sun, Nov 13, 2011 at 18:58, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > Does anyone object if I remove DMRefine/CoarsenHierarchy() and just have DMRefine/Coarsen() take a number of levels argument? > > So: remove DMRefine/Coarsen and then rename the Hierarchy versions? Sounds fine to me, you can just pass one level. > > The comm argument would be removed from the calling sequence, doesn't work anyways. > > The need to use the same communicator on coarse grids is pretty limiting, in my opinion. It may be tricky to implement, but I'm not sure we should remove it from the interface. It can be passed in some other way. Barry

1 0

Re: [petsc-dev] removing DMRefine/CoarsenHierarchy?
by Jed Brown 13 Nov '11

13 Nov '11

On Sun, Nov 13, 2011 at 18:58, Barry Smith <bsmith(a)mcs.anl.gov> wrote: > Does anyone object if I remove DMRefine/CoarsenHierarchy() and just have > DMRefine/Coarsen() take a number of levels argument? > So: remove DMRefine/Coarsen and then rename the Hierarchy versions? Sounds fine to me, you can just pass one level. > The comm argument would be removed from the calling sequence, doesn't work > anyways. > The need to use the same communicator on coarse grids is pretty limiting, in my opinion. It may be tricky to implement, but I'm not sure we should remove it from the interface.

1 0

removing DMRefine/CoarsenHierarchy?
by Barry Smith 13 Nov '11

13 Nov '11

Does anyone object if I remove DMRefine/CoarsenHierarchy() and just have DMRefine/Coarsen() take a number of levels argument? Barry The comm argument would be removed from the calling sequence, doesn't work anyways.

1 0