RE: [Algorithms] Vertex data entering the pipeline

[Shawn H. <sha...@cl...>]

Jon Watte writes:
>> Could vertex and normal buffers have seperate indices, so 
>> less memory is used up by duplicated normals, and (hopefully) 
>> they are only transformed once.
> 
> That's how 3dsmax and other modelers store the data internally. 
> However, every time this subject gets mentioned, some hardware 
> vendors will jump in and claim that their research shows this 
> would actually be slower. I tend to agree -- when we tried 
> storing models like this, the size of the file often went UP 
> because of the added index sizes

Gamecube, anyone?

It uses separate indices for positions, normals, UV's, and whatever
else you like, and you can usually pack meshes down to a lot less
than on any other platform (which is a good thing given how little
RAM the thing has :-)

I don't see this being anything of a speed boost for rendering,
though, as it makes access far more random and less cache-friendly.
Hardware likes nice simple linear access to big chunks of
contiguous data, which is pretty much mutually exclusive with
doing compression through indexing.


-- 
Shawn Hargreaves
Climax Brighton