Hi guys, I've got a software rendering pipeline programmed, but I have a few weird things happening, (and I think they're related) 1) z-buffering doesn't work - stuff gets drawn over older stuff 2) after multiplying by modelview and projection matrices, my z and w values are REALLY close, always 3) after the homogeneous divide (division by w), my z values (and w values) are all around 1, if not exactly 1 4) I think this also messes stuff up when doing perspective-correct interpolation Does that sound wrong to you guys?

Thanks :) Yeah, I made my own classes, though I've tested my rotation, translation, ModelView, PerspectiveProjection, and MVP matrices against OpenGL's. They turned out the same for the same input. Since I multiplied matrices together in the process, my class's operator* should be working too.
I'll try working out the transformations on paper and see how that goes.

Update: Multiplying the MVP matrix times vertices is giving me the same results as my program. Weird...

[Edited by - meteorstorm42 on October 29, 2009 10:04:58 PM]

Like you have done already make a little test rig in OpenGL and use the debugger to interrogate values.

glGetFloatv() will allow you to get different matrices from OpenGL and see what values they have. Whilst very lazy I find this a lot easier than writing out the math myself ;-)


When you say new stuff gets drawn over the old stuff on your framebuffer are you clearing it each time you call your draw method? Looking at double buffering at this point may also be helpful.

Quote:Original post by Jason Z
So if your matrix classes work as intended (or at least the same as OpenGL) then the next area to work through is to see if your culling code is working properly - are the objects that you render completely on screen and within the view frustum? If so, you could look into the parameters you use to create your projection matrix and ensure that there are reasonable values in there too.

If all that works out, then what you should see after performing a WVP transformation is the w-component will contain the view space depth of the vertex. Check if this is being calculated correctly (which it probably is since you already tested it...), and if so then you can do the w-divide and move on to rasterization.

Can you describe how you do your rasterization and z-interpolation? This may be the source of your errors too - how you store them and so on. If you post screen shots it may also be helpful to diagnose what is happening.

I'm doing this for the projection matrix:
setProjectionMatrix( fovy = 45.0, aspectRatio = 1.0, zNear = 0.1, zFar = 600 );
It gives me the same matrix as glutPerspective()

Yup, all triangles that even straddle the viewable volume are culled (keeping this part simple for now...)

By WVP, did you mean MVP? I'll try doing a few more matrix multiplies in octave, see if my program's doing it correctly...

I'm using a line-walking algorithm for rasterizing. Individual triangles are drawn fine (aside from the textures not being perspective-correct), but it looks like deeper triangles are going over closer ones.
My z-buffer starts out with float_max and then I check if the incoming fragment's z is less than where I'm trying to draw it.

I just started trying to implement perspective-correct interpolation, so I'm unsure how that part of it is working, but as I was debugging it, I noticed the equal z and w values (coming into the rasterizer).