Thanks Maris,

[BTW: your mail was sent only to me, so I'm forwarding to the list.]

On 04/08/17 16:04, Maris Nartiss wrote:
1. It is not a dependency, as OpenMP code should run just fine also
when compiled without OpenMP. Those are just comments in code thus
backwards safe.

Right. I now see that all code is conditionalized on _OPENMP being set and I suppose that happens when you use --with-openmp as a compiler switch ?

2. It is not as easy as it seems, as every place, where
parallelisation is introduced, needs to be examined carefully to not
introduce hard to track issues when common variables are changed and
to not make even worse performance by making CPU cache dirty when it
isn't. Of course these problems affect only binaries compiled with
OpenMP. Thus OpenMP support in GRASS GIS could be marked as "use at
own risk".


This thus means that merging the OpenMP version into the trunk version will not have any consequences for the "normal" user. However, does it mean that maintenance of the modules will become more complex ?

I see that in v.surf.rst the OpenMP stuff has already been integrated. It's only been 7 months, but up to now noone has complained, so I think we could integrate openmp into the two other modules as well (r.sun + r.sim.water).

Moritz
_______________________________________________
grass-dev mailing list
[email protected]
https://lists.osgeo.org/mailman/listinfo/grass-dev

Reply via email to