You and me both… But I think I’m getting it.
I think your post basically said the same thing as my last one did.
And yes, currently (before implementing this newest idea) I am doing just that, a ‘solid strip of triangles’ - but I am looking for a better way by injecting indexing into the mix. Hence the point of the whole thread. Now that we got my non-Indexed half figured out, now we are getting to the good part.
aye!
i too got to this point a couple times and i still didn’t try that last thing that would solve the whole indexing problem - at a high cost tho. i guess the post transform cache and other magic things GPU’s do these days would not snap in.
we could just grab the mesh, as it is defined in the .obj format and store everything in SSBO’s and source not the vertices but more the “faces”.
if you think about it, once your mesh becomes really complex, the “unwrapped” vbo will look more and more like a waste of memory. normals alone, using smoothing groups or per face materials.
so if attributes are more “per face” than “per vertex” we could use a geo-shader which sources the “face” data, which would be just a pack of indices into other data buffers. emit everything in flight. least memory usage, most processing.
this sample show how to source vertex data from a SSBO :
https://github.com/g-truc/ogl-samples/blob/master/tests/gl-430-draw-without-vertex-attrib.cpp
they use the [icode]gl_VertexID[/icode] to query the data.
but the idea is open to any data format you can imagine.
Aaaaand you lost me. I am sorry, I don’t know Squat about shaders yet. I have basic pass through ones for testing, but I can’t even get data To them yet, so still working on it. And Geo-shaders are just a faint silhouette on the horizon right now.