Vertex attributes are passed in view space instead of model space to the vertex shader

Relating to this topic : http://forum.processing.org/two/discussion/7484/vertex-attribute-bug-#Item_5

I just set a color if the vertex X position is > 0 or not. So if the vertex is really in world space, not in view, if i move the camera the color would stay at the same position on the box. Here, the color move on the box when i move the camera not the world(so the view matrix ?)

Is-it a normal behaviour ?

PDE

import peasy.*;

PShader cubemapShader; PeasyCam cam;

void setup() { size(800, 640, P3D); frameRate(60);

// Load cubemap shader. cubemapShader = loadShader("cubemapfrag.glsl", "cubemapvert.glsl"); cam = new PeasyCam(this, 0, 0, 0, 180); cam.setYawRotationMode(); }

void draw() { background(0); shader(cubemapShader); box(50); resetShader(); }

Vertex


uniform mat4 transform;

attribute vec4 position;

varying vec4 pos;

void main() {

pos = position;

gl_Position = transform * position;

}

Fragment


varying vec4 pos;

void main() { if(pos.x > 0.0){ gl_FragColor = vec4(1,0,0,1); } else{ gl_FragColor = vec4(0,0,1,1); } }

Thanks for your report. What you observe is correct, right now vertex positions are premultiplied by modelview matrix before passing them to vertex shader. We will discuss if there is a way to make it easier for people to get model coordinates.

Right now you can convert them back using inverted modelview matrix like this:

import peasy.*;

PShader cubemapShader;
PeasyCam cam;

void setup() {
  size(800, 640, P3D);
  frameRate(60);

  // Load cubemap shader.
  cubemapShader = loadShader("cubemapfrag.glsl", "cubemapvert.glsl");
  cam = new PeasyCam(this, 0, 0, 0, 180);
  cam.setYawRotationMode();
}

void draw() {
  background(0);
  shader(cubemapShader);
  cubemapShader.set("modelviewInv", ((PGraphicsOpenGL) g).modelviewInv);

  box(50);
  resetShader();
}

// vertex shader
uniform mat4 transform;

attribute vec4 position;

varying vec4 pos;

uniform mat4 modelviewInv;

void main() {
  pos = position * modelviewInv;

  gl_Position = transform * position;
}

// fragment shader
varying vec4 pos;

void main() { 
  if (pos.x > 0.0) {
    gl_FragColor = vec4(1.0,0.0,0.0,1.0);
  } else {
    gl_FragColor = vec4(0.0,0.0,1.0,1.0);
  }
}

If you need more help, please post on processing forum.

We will discuss if there is a way to make it easier for people to get model coordinates.

Please also discuss a way for easy world space coordinates. Is this even possible, currently?

Just for information, the matrix you get by the PGraphicsOpenGL object seams not really right.

For example the cameraInv and the camera matrix are the same of the modelview / modelviewInv, according usual GLSL, the cameraInv and camera may be the same of the view / viewInv matrix.

More strange, when i get the modelview from the PGraphicsOpenGL, and pass this matrix to the shader via Shader.set, this matrix is not the same passed automatically by the processing.

So lot of strange things ...

@P0ulp

Hi guys, by default the vertices are indeed sent to the vertex shader already multiplied by the modelview matrix (and so the modelview matrix in the vertex shader is set to the identity to avoid applying the transformation twice, so this might be one reason for the discrepancies you have observed).

The reason for this seemingly puzzling approach is to optimize performance when drawing multiple shapes: since it is most efficient to minimize the number of buffer copies between CPU and GPU memory, the renderer batches as many shapes together as possible to draw them in a single call. However, if each shape had a different transformation, then it wouldn't be possible to group them because they would have different uniform matrices.

You can disable the batching with the following hint: hint(DISABLE_OPTIMIZED_STROKE)

@P0ulp regarding the relationship between the modelview and the camera matrices in Processing is the following:

modelview = camera * [all other geometry transformations applied after camera()]

The camera matrix is really not used for anything by the OpenGL renderer (by the exception of the modelXYZ functions), it more of a legacy from the renderers in 1.x

I'm happy to discuss the matrix handling further if you thing it needs to be changed/improved/clarified.

We will add a section in the wiki explaining the workarounds discussed in this thread.

After reading this thread, I really need some help to understand the tutorial on Shaders in the official Processing page. https://processing.org/tutorials/pshader/
If the modelview matrix in vertex shader is the identity, this means that transform * vertex = projection * vertex ?
Even though, in some tutorials with the position in eye coordinates, expressions like "ecPosition = modelview * position" are used, does this equal to "ecPosition = position" ?
I am not sure what the meaning of modelview matrix is in the vertex shader in Processing.

Please use the forum for coding questions.

Navigation Menu

Repository navigation

Vertex attributes are passed in view space instead of model space to the vertex shader #2904

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Vertex attributes are passed in view space instead of model space to the vertex shader #2904

Description

Activity

JakubValtar commented on Oct 21, 2014

AmnonOwed commented on Oct 21, 2014

P0ulp commented on Oct 23, 2014

codeanticode commented on Nov 3, 2014

codeanticode commented on Jul 16, 2015

technosalon commented on Sep 3, 2017

benfry commented on Sep 4, 2017

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions