Hello, I have been reading about the topic of VBOs and how they will benefit my rendering. But I don't understand how you can use the typical transformation matrices to transform data that is held in a VBO.

For example, let's say I have a VBO storing all the geometry of my scene, including a cube that rotates/moves independently from the rest of the geometry.

I want to render the geometry held in the VBO but I also want to apply a transformation to the cube that is inside.

What's a good approach to do that? Do I apply the transformation to the vertices and then put the vertex information of the cube with the transformation applied, updating the VBO in every single frame?

I'd want to be able to apply the transformation with a vertex shader, but I don't see the way to do that.

thanks.

I think you misunderstood what VBO is. A single VBO shouldn't store the entire scene data (unless you have a pretty simple scene).

Correct usage of VBO would be:

1. At application load time, create and initialize a VBO per object.
2. At you render loop - the VBOs are already initialized, just bind them, set the correct render state and draw. No need to reload the vertex data.

There is usually no sense in putting ALL geometry in ONE vertex buffer and draw that entirely in ONE call.

1. You can use several vertex buffers. You will at least put static geometry into another vertex buffer than dynamic geometry (e.g. skinned meshes), because the former one needs to be pushed once into GPU memory and may remain there for many of the subsequent frames, while the latter one can be used only once before becoming obsolete. See the usage parameter to glBufferData and companions. However, in reality you will use a finer distinction though.

2. You can use the same vertex buffer for several draw calls. Due to the count parameter of glDraw* routines, you can define a subset of a vertex buffer to be drawn in a particular call. This allows you to alter vertex independent parameters, e.g. the MODELVIEW matrix, for different parts of the same vertex buffer.

3. You can use several MODELVIEW matrices in one call. If you have enough uniform variables available, or are pushing MODELVIEW matrices by misusing a floating point texture, (or somewhat else), and give an index along with each vertex in the vertex buffer to specify what MODELVIEW matrix to use, it is possible to deal with several distinct MODELVIEW matrices in one draw call.

4. You can, of course, also transform by CPU.

My advice would be to use another draw call with its own MODELVIEW matrix.

EDIT: totally misunderstood your question! As the others above have stated, you wouldn't put all geometry into a VBO. You would have a VBO per object or per surface or something. Then you would be able to draw those objects individually with separate uniforms.

In most cases your VBO will be initialized once with vertex data that's in "model space" and then you'll leave the VBO alone. When you want to transform the object into world space, you'll do that in the vertex shader. You can send your transformation matrices to the vertex shader with uniform buffers.

#version 410

in vec4 position;
in vec3 color;

out vec4 out_position;

uniform Transform {
    mat4 model_view_projection;
};

void main()
{
    out_position = position * model_view_projection;
}

In that code model_view_projection is a member of the Transform uniform buffer. On the CPU side you would make a buffer that contained your model-view-projection matrix, and then bind that buffer to the location of Transform before drawing.

This way you never have to update your VBO, and all transformations are done efficiently on the GPU.

EDIT: totally misunderstood your question! As the others above have stated, you wouldn't put all geometry into a VBO. You would have a VBO per object or per surface or something. Then you would be able to draw those objects individually with separate uniforms.

In most cases your VBO will be initialized once with vertex data that's in "model space" and then you'll leave the VBO alone. When you want to transform the object into world space, you'll do that in the vertex shader. You can send your transformation matrices to the vertex shader with uniform buffers.

#version 410

in vec4 position;
in vec3 color;

out vec4 out_position;

uniform Transform {
    mat4 model_view_projection;
};

void main()
{
    out_position = position * model_view_projection;
}

In that code model_view_projection is a member of the Transform uniform buffer. On the CPU side you would make a buffer that contained your model-view-projection matrix, and then bind that buffer to the location of Transform before drawing.

This way you never have to update your VBO, and all transformations are done efficiently on the GPU.

What If I want to update the geometry? (add/remove vertices to the VBO)

And If I have, let's say, lots of small sprites (textured quads/triangles), a single VBO for each of them doesn't look like a good idea...