The result of projMat * viewMat * modelMat * vertPos should result in a screen-space position... right?

Question

And given that, shouldn't all position values that end up being rendered be between the values -1 and 1?

I tried passing said position value as the "color" value from my vertex shader to my fragment shader to see what would happen, and expected a gradient the full way across the screen (at least, where geometry exists).

I would EXPECT the top-right corner of the screen to have color value rgb(1.0,1.0,?.?) (? because the z value might vary), and that would gradiate towards (0.0,0.0,?.?) at the very center of the screen (and anything to the bottom left would have 0'd r and g components because their value would be negative).

But instead, what I'm getting looks like the gradiation is happening in a smaller scale towards the center of my screen (attached): 1, meaning the blue component is always maxed out"> (

Why is this? This makes it look like the geometry position resulting from the composition of my matrices and position is a value between -10ish and 10ish...?

Any ideas what I might be missing?

edit- don't worry about the funky geometry. if it helps, what's being rendered is a quad behind ~100 unit triangles randomly rotated about the origin. debugging stuff.

Answer 1

You are confusing several things here.

projMat * viewMat * modelMat * vPos will transform vPos into clip space, not screen space. However, using your coordinates as colors, you don't even want screen space (which are the pixel coordinates relative to the output window, and accessible via gl_FragmentPosition in the fragment shader), but you want [-1,1] normalized device coordinates. You get to NDC by dividing by the w component of your clipspace value.

From the images you get I can guess that projMat is a projective transform, since in the orthogonal case, clip.w will typically be 1 for all vertices (not necessarily, but very likely), and the results would look more like you would expect. You can see from your image that x and y (r and g) are zero in the center. However, those values are not limited to the interval [-1,1], but your shader output values are clamped to [0,1] so you don't see much of a gradient. Your z coord is >= 1 anywhere, so it does not change at all in the image.

If you would use the NDC coords as colors, you would indeed see a red gradient in the right half, and a green gradient in the upper half, but the other areas where the value is below 0 will still be clamped to zero. You would also get some more info in the blue channel (although it might be possible that NDC z is <= 0 for your whole scene), but you should be aware of the nonlinear z distortions introducted by the divide.