I wake up with a clear vision, but sadly my laptop card doesn't do displacement mapping nor
glVertexAttribDivisor so I can't test it out; I'm left sharing here:
With geomipmapping, the grid at any factor is transposable - if you pass in an offset - say as a uniform - you can reuse the same vertex and index array again and again.
If you also pass in the offset into the heightmap as a uniform, the vertex shader can do displacement mapping. If the displacement map is mipmapped, you get the advantages of trilinear filtering for distant maps.
And, if the scenery is closer, rather than exposing that the you have a world made out of quads, you can use your transposable grid vertex array and indices to do vertex-shader interpolation (fancy splines) to do super-smooth infinite zoom?
So I have some questions:
- does it work? In theory, in practice?
- does anyone do it? Does this technique have a name? Papers, demos, anything I can look at?
glVertexAttribDivisormean that you can have a single
glMultiDrawElementsEXTor similar approach to draw all your terrain tiles in one call rather than setting up the uniforms and emitting each tile? Would this offer any noticeable gains?
- does a heightmap that is GL_LUMINANCE take just one byte per pixel(=vertex)? (On mainstream cards, obviously. Does storage vary in practice?) Does going to the effort of reusing the same vertices and indices mean that you can basically fill the GPU RAM with heightmap and not a lot else, giving you either bigger landscapes or more detailed landscapes/meshes for the same bang?
- is mipmapping the displacement map going to work? On future cards? Is it going to introduce unsurmountable inaccuracies if it is enabled?