[pooma-dev] Assigning to internal guards

[John H. Hall]

>
> Of course, in multiple dimensions you generally don't want internal 
> guards on anything that you're not going to stencil, so in practice it 
> isn't clear how often this practice will really pay off. (Internal 
> guards can consume a lot of memory in 3D.)
>
>         Jim

Jim, et. al.:
In practice, this (calculating internal/external guard cells instead of 
communicating) would be one of the biggest optimizations possible. 
Especially when you introduce the concept of dependent and independent 
fields. Usually only the independent fields need to communicate to fill 
their guard cells. The dependent fields can just calculate their guards 
and move on as pretty as you please.

Because of the limitations on scalar coding right now, once you start 
using guard cells, pretty much every field ends up requiring them also 
(this has to do with alignment issues and all patches starting their 
coordinate indexing at (0,0,0)). The devil is of course in the details, 
but, you can't generally get away with mixing and matching fields with 
and without guard cells right now except in the simple case of all SIMD 
syntax (which of course is what you were referring to).

I've got the GCC work back online. Mark's contract is extended through 
next year. Hopefully, we can soon get back to thinking about these 
other exciting issues.

Regards,
John

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: text/enriched
Size: 1396 bytes
Desc: not available
URL: <http://sourcerytools.com/pipermail/pooma-dev/attachments/20020925/4ee1c8eb/attachment.bin>