nvpro-samples / gl_occlusion_culling Goto Github PK

In the scene vertex shader the matrices are accessed by the index and the byte offset.
texelFetch(tex,idx*4 + 0)
but in the raster shader, the matrices are indexed with additional offsets. As far as I can tell the matrix buffer is constructed the same. What does the following mean in the raster vert shader?
#ifndef MATRIX_WORLD
#define MATRIX_WORLD 0
#endif

#ifndef MATRIX_WORLD_IT
#define MATRIX_WORLD_IT 1
#endif

#ifndef MATRICES
#define MATRICES 2
#endif

And the index is constructed int matindex = (matrixIndex * MATRICES + MATRIX_WORLD_IT) * 4;

I don't understand how this works when it looks like the same matrix buffer is used for both shaders?
I appreciate any insights.

Cheers

Simon

As we know, The geometry shader is not that fast as a computer shader. I have an idea, if we calculate the three edges every instance in computer shader, then it can gain more performance. I plan to implement this by Vulkan. So could this algorithm be used in that way?

Another question is that what's the difference between these 3 sides to just drawing bounding box sides for early fragment test.

Thank you.