Setup of matrix for instance shader

问题

I want to draw instanced cubes.

I can call GL.DrawArraysInstanced(PrimitiveType.Triangles, 0, 36, 2); successfully.

My problem is that all the cubes are drawn at the same position and same rotation. How can i change that individually for every cube?

To create different positions and so on, i need a matrix for each cube, right? I created this:

Matrix4[] Matrices = new Matrix4[]{
  Matrix4.Identity, //do nothing
  Matrix4.Identity * Matrix4.CreateTranslation(1,0,0) //move a little bit
};

GL.BindBuffer(BufferTarget.ArrayBuffer, matrixBuffer);
GL.BufferData(BufferTarget.ArrayBuffer, (IntPtr)(sizeof(float) * 16 * Matrices.Length), Matrices, BufferUsageHint.StaticDraw);

This should create a buffer where i can store my matrices. matrixBuffer is the pointer to my buffer. Im not sure if the size is correct, i took float * 4 (for Vector4) * 4 (for 4 vectors) * array-size.

Draw Loop:

GL.BindBuffer(BufferTarget.ArrayBuffer, matrixBuffer);
GL.VertexAttribPointer(3, 4, VertexAttribPointerType.Float, false, 0, 0);
//GL.VertexAttribDivisor(3, ?);
GL.EnableVertexAttribArray(3);

GL.DrawArraysInstanced(PrimitiveType.Triangles, 0, 36, 2);

Any higher number than 4 in VertexAttribPointer(..., 4, VertexattribPointerType.Float, ...); causes a crash. I tought i have to set that value to 16?

Im not sure if i need a VertexAttribDivisor, probably i need this every 36th vertex so i call GL.VertexAttribDivisor(3, 36);? But when i do that, i can't see any cube.

My vertex shader:

#version 330 core
layout(location = 0) in vec3 position;
layout(location = 1) in vec4 color;
layout(location = 2) in vec2 texCoord;
layout(location = 3) in mat4 instanceMatrix;

uniform mat4 projMatrix;

out vec4 vColor;
out vec2 texCoords[];

void main(){
  gl_Position = instanceMatrix * projMatrix * vec4(position, 1.0);
  //gl_Position = projMatrix * vec4(position, 1.0);
  texCoords[0] = texCoord;
  vColor = color;
}

So my questions:

What is wrong with my code?
Why can i set the size-parameter of VertexAttribPointer only to 4?
What is the correct value for VertexAttribDivisor?

Edit:

Based on the answer of Andon M. Coleman i made this changes:

GL.BindBuffer(BufferTarget.UniformBuffer, matrixBuffer);
GL.BufferData(BufferTarget.UniformBuffer, (IntPtr)(sizeof(float) * 16), IntPtr.Zero, BufferUsageHint.DynamicDraw);
//Bind Buffer to Binding Point
GL.BindBufferBase(BufferRangeTarget.UniformBuffer, matrixUniform, matrixBuffer);

matrixUniform = GL.GetUniformBlockIndex(shaderProgram, "instanceMatrix");
//Bind Uniform Block to Binding Point
GL.UniformBlockBinding(shaderProgram, matrixUniform, 0);

GL.BufferSubData(BufferTarget.UniformBuffer, IntPtr.Zero, (IntPtr)(sizeof(float) * 16 * Matrices.Length), Matrices);

And shader:

#version 330 core
layout(location = 0) in vec4 position; //gets vec3, fills w with 1.0
layout(location = 1) in vec4 color;
layout(location = 2) in vec2 texCoord;

uniform mat4 projMatrix;
uniform UniformBlock
{ mat4 instanceMatrix[]; };

out vec4 vColor;
out vec2 texCoords[];

void main(){
  gl_Position = projMatrix * instanceMatrix[0] * position;
  texCoords[0] = texCoord;
  vColor = color;
}

回答1:

You have discovered the hard way that vertex attribute locations are always 4-component.

The only way to make a 4x4 matrix a per-vertex attribute is if you concede that mat4 is 4x as large as vec4.

Consider the declaration of your `mat4` vertex attribute:

layout(location = 3) in mat4 instanceMatrix;

You might naturally think that location 3 stores 16 floating-point values, but you would be wrong. Locations in GLSL are always 4-component. Thus, mat4 instanceMatrix actually occupies 4 different locations.

This is essentially how `instanceMatrix` actually works:

layout(location = 3) in vec4 instanceMatrix_Column0;
layout(location = 4) in vec4 instanceMatrix_Column1;
layout(location = 5) in vec4 instanceMatrix_Column2;
layout(location = 6) in vec4 instanceMatrix_Column3;

Fortunately, you do not have to write your shader that way, it is perfectly valid to have a mat4 vertex attribute.

However, you do have to write your C# code to behave that way:

GL.BindBuffer(BufferTarget.ArrayBuffer, matrixBuffer);
GL.VertexAttribPointer(3, 4, VertexAttribPointerType.Float, false, 64,  0); // c0
GL.VertexAttribPointer(4, 4, VertexAttribPointerType.Float, false, 64, 16); // c1
GL.VertexAttribPointer(5, 4, VertexAttribPointerType.Float, false, 64, 32); // c2
GL.VertexAttribPointer(6, 4, VertexAttribPointerType.Float, false, 64, 48); // c3

Likewise, you must setup your vertex attribute divisor for all 4 locations:

GL.VertexAttribDivisor (3, 1);
GL.VertexAttribDivisor (4, 1);
GL.VertexAttribDivisor (5, 1);
GL.VertexAttribDivisor (6, 1);

Incidentally, because vertex attributes are always 4-component, you can actually declare:

layout(location = 0) in vec4 position;

And stop writing ugly code like this:

gl_Position = instanceMatrix * projMatrix * vec4(position, 1.0);

This is because missing components in a vertex attribute are automatically expanded by OpenGL.

(0.0, 0.0, 0.0, 1.0)

If you declare a vertex attribute as vec4 in the GLSL shader, but only supply data for XYZ, then W is automatically assigned a value of 1.0.

In actuality, you do not want to store your matrices per-vertex. That is a waste of multiple vertex attribute locations. What you may consider is an array of uniforms, or better yet a uniform buffer. You can index this array using the Vertex Shader pre-declared variable: gl_InstanceID. That is really the most sensible way to approach this, because you may find yourself using more properties per-instance than you have vertex attribute locations (mininum 16 in GL 3.3, only a few GPUs actually support more than 16).

Keep in mind that there is a limit to the number of vec4 uniforms a vertex shader can use in a single invocation, and that a mat4 counts as 4x the size of a vec4. Using a uniform buffer will allow you to draw many more instances than a plain old array of uniforms would.

来源：https://stackoverflow.com/questions/23629776/setup-of-matrix-for-instance-shader

标签

opengl

matrix

opentk