Re: [pooma-dev] status

[Scott Haney]

On Tuesday, June 5, 2001, at 02:40 AM, Allan Stokes wrote:

> I began authoring a test document in DocBook describing an approach to
> optimal uniform partitioning which caught my fancy while I was reviewing
> code in src/Partition.  This will be a two or three page exploration of
> writing simple equations, psuedo-code, code formatting, etc. in a 
> DocBook
> context.
>
> This week I will document the Pooma domain concept, and perhaps 
> contribute
> some documentation to the NewField div stencil which Jeffrey and 
> Stephen are
> presently working on.  I suggested that I begin by comparing Jeffrey's 
> Pooma
> implementation to similar constructs from the Blitz library, which 
> factors
> the code quite differently.

Hi Allan,

This sounds like a very good plan.

When you're looking at domains, could you also try to construct a plan 
for simplifying the current implementation, which makes domains more 
heavyweight and harder to optimize than they should be, and improving 
the physical design.

Comparing with Blitz is always a good thing since Todd has great ideas. 
However, support for parallelism; first-class fields with geometry, 
centering, and boundary conditions; and arbitrary data representation 
provide some additional constraints that drive our design. We'd need to 
have some *really* good reasons to start changing things very much at 
this point. BTW, there is another interesting package that uses 
sophisticated stencil objects from LANL called Pdeum. The developer's 
name is Martin Staley at address@hidden

Scott

P.S. Most of the Proximation folks have been working away on an internal 
project, which is supposed to be done by mid-June. Therefore, there is 
not a lot of POOMA progress to report. Also, I'm going on vacation June 
7. I'll be back June 18th. Stephen is going on vacation June 11 and will 
be back the same day as me. We're both going to get some POOMA work in 
before we leave.