[GEM-dev] vertex_array

james tittle tigital at mac.com
Tue Aug 16 18:16:20 CEST 2005


hmmm,

...welcome back IOhannes!  Hope ya had a good break:  I'm still  
looking for one myself :-\

On Aug 16, 2005, at 10:09 AM, zmoelnig at iem.at wrote:

> Quoting "B. Bogart" <ben at ekran.org>:
>
>
>> Hey again.
>>
>> Well it seems my fresh vertex_array checkout is not helping with
>> compiling on OSX. The build system clearly does not match the new
>> autoconf one. Anyhow:
>>
>
> yes the new autoconf (in Gem/src instead of Gem/src/Gnu) has no  
> notion of the
> vertex_array.
> in theory you just need to add "Vertex" to the GEM_TARGETS in  
> Make.config (after
> configure), but i have no way to test this right now.
>
> so i gues it is really time to make the vertex_array stuff ready  
> for the main
> branch (even if there are different types for color/vertex/ 
> texcoord/...; this
> could be changed later on if need is at hand; or do costy  
> conversions between
> those arrays at the user's own choice)

...yeh, ben, you've opened an unfinished project, so proceed very  
carefully!  When last we left (as I recall), IOhannes, chris, and I  
were bickering about IOhannes' latest change to the vertex_array  
branch:  he had made all "arrays" equal at 4 floats:  the idea being  
that we can then plug anything into anything and let someone else  
sort it out later...

...chris and I didn't (and I presume still don't) like this, firstly  
because there's no such thing as a 4 element normal, texcoord, color,  
etc....as a matter of fact, only vertices can have 4 elements...so,  
you're put in the position of potentially doubling the amount of data  
being used, potentially for some undefined flexibility...

...the whole matter was left there when IOhannes said "show me some  
numbers" about how big a slowdown we're talking about, and I never  
got around to actually doing that...

...I guess until the benchmarks are done and we can all fall on one  
side or the other, I'd like to point out that if we use vertex buffer  
objects (or any kind of buffer objects), and we do want to use them,  
it should be noted that you can't map a buffer for 4 elements to  
two...maybe can't is too harsh a word:  let's say that it's  
experimental at best, and expects alot from the gl driver, especially  
considering we're dealing with multiple platforms, gpus, etc...

ok?
jamie




More information about the GEM-dev mailing list