[petsc-dev] MatGetLocalSubMatrix, MatLocalReference implementation

[Barry Smith]

On Nov 20, 2010, at 5:36 PM, Jed Brown wrote:

> On Sun, Nov 21, 2010 at 00:01, Barry Smith <bsmith at mcs.anl.gov> wrote:
> Just pick what you want, but if you do use "local" then it should match the use of "local" in other parts of PETSc (for example a local Vec is a ghosted "part" of a global Vec). For example, if you suddenly starting calling unghosted sequential Vec parts of a global Vec it would be confusing. So just use your judgement, I can't tell what you are asking half the time so don't always expect a coherent response.
> 
> Okay, this thing I need denotes a subset of the usual "local Vec".  You introduced "global IS" to denote subsets of the global Vec so I think my calling it "local IS" is entirely consistent.  I tend to think of the relations in the sense of
> 
>   (global Vec, global IS) induces, via the "local" operation, (local Vec, local IS)
> 
> rather than
> 
>   (global Vec, local Vec) induces, via the "get IS" operation, (global IS, local IS)
> 
> although this diagram commutes with a suitable ISLocalToGlobalMapping.
> 
> I would like to rename your DMCompositeGetGlobalIndices to DMCompositeGetISLocalToGlobalMappings (and make it return ISLocalToGlobalMappings, obviously).  Since the function was never called (never declared in a public header), I can't imagine this will inconvenience anyone.

   Fine.


> 
> BTW, that function currently uses a PFIDENTITY hack which is really gross if the user gets crazy and uses single precision, I'll make it use integer arithmetic.
> 
> Jed