[Bf-committers] profiling subsurf

Brecht Van Lommel brechtvanlommel at pandora.be
Sat Nov 25 17:44:59 CET 2006


Hey,

Ton Roosendaal wrote:
> I've done some profiling tests, and found out that subsurf is slowest in 
> generating UV maps and in subdividing vertexgroups. That's both features 
> we could make optional... and probably the same for Sticky and other 
> custom layers. This is usually not needed for 3d editing.

If I remember correctly, Ben said in the last sunday meeting he would 
look into a system that would only preserve the layers that are needed.

> Problem is that sometimes you do require vertexgroups, like for particle 
> (hair) systems.
> Nevertheless, also for hair you want to get interactive editing in a 3D 
> window.
> 
> Derivedmesh already has flags to indicate 'render quality', so probably 
> we can use this flag for triggering subdivision of custom layers (apart 
> from vertex locations).
> To facilitate modifiers to keep working on subsurf levels, you then can 
> just copy the layer info (from nearest vertex or so).

Just copying might be slow as well, since it does a MEM_malloc for each 
vertex. But it would be faster of course.

Maybe the subsurf modifier could allow 0 levels, so no subdivision at 
all for e.g. hair with vertex groups. Since the particles already follow 
the smoothed vertex normals, there doesn't seem to be much difference 
with or without subsurf. But that might not be true in cases more 
complicated than I tested..

Brecht.



More information about the Bf-committers mailing list