Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

In this Discussion

[Solved] Unusual transform between vertex and fragment shader

edited September 2018 in Shaders Posts: 21

When investigating co-ordinate systems used by the app I noticed variables interpolated between the vertex and fragment shader have an unusual implicit conversion.

Let w and h denote the built-in Codea variables WIDTH and HEIGHT, respectively.

gl_Position is (-w, -h) at the bottom left corner, (w, h) at the top right corner, and (0, 0) in the centre. I expected this to be (-1.0, -1.0), (1.0, 1.0), and (0.0, 0.0) (a.k.a. normalised device co-ordinates) respectively, but it's not a concern - it's quite sensible.

The thing I find troubling is that there is an implicit conversion from these 'semi-window' co-ordinates to normalised device co-ordinates between the vertex and fragment stages on variables that are not gl_Position. The implication is that, for example, if I want to interpolate a 100% red vertex in the fragment shader, I have to set the color variable to (w, 0, 0) and, if I want to pass a 100% green vertex to the fragment shader, I have to set the color variable to (0, h, 0) so that the conversion maps the color values to unit range. This is totally surprising and unintuitive!

The following program demonstrates:


function setup()
    parameter.number("x", 1.0, 1024.0)
    m = mesh()
    m.shader = shader("Project:Foo")
    m:addRect(0.5 * WIDTH, 0.5 * HEIGHT, WIDTH, HEIGHT)
    print(WIDTH, HEIGHT)
    local vertices = m:buffer("position")
    for i = 1, 4 do

function draw()
    background(40, 40, 50)
    m.shader.uColorScale = x
    m.shader.uDimensions = vec2(WIDTH, HEIGHT)

Vertex shader

attribute vec4 position;

uniform vec2 uDimensions;
uniform highp float uColorScale;

varying lowp vec4 vColor;

void main()
    vColor = uColorScale * vec4(position.x / uDimensions.x, position.y / uDimensions.y, 0.0, 1.0);
    gl_Position = position;

Fragment shader

precision highp float;

uniform vec2 uDimensions;
varying lowp vec4 vColor;

void main()
    if (vColor.x > 1.0 || vColor.y > 1.0) {
        gl_FragColor = vec4(1.0, 1.0, 1.0, 1.0);
    } else {
        gl_FragColor = vColor;

See attached screenshots for output.

I'd like to know whether there is a way around this that doesn't involve scaling by the window size.

image.png 168.3K
image.png 68.8K


  • Update: It turns out the output of gl_Position should in fact be normalised device co-ordinates and not screen co-ordinates. It seems during investigation I have managed to thoroughly confuse myself! :blush:

  • One source of confusion comes from the default setting for projectionMatrix() which is equivalent to ortho(-WIDTH, WIDTH, -HEIGHT, HEIGHT) with a max z value of 10.0.

    Another is that the square in the shader lab follows this default projection convention but with different dimensions. The square itself is defined in terms of this co-ordinate system which makes the live output a bit useless if you want to work in a different co-ordinate system. This can be worked around although it would be better if one could configure the shader lab, i.e. set the input vertices and size of live preview area.

    In the regular editor, one can switch to NDC with ortho(-1.0, 1.0, -1.0, 1.0), bearing in mind the max z value is still 10.0.

Sign In or Register to comment.