Derivation of cegui projection matrix

[fedyakin]

Hello,
I've been trying to reverse engineer the values in the projection matrix created in OgreRenderTarget::updateMatrix and failing. There are a few things that have me stumped so far.
1) The location of non-zero values in the projection matrix don't match those in my (very limited) understanding of perspective and ortho projection matrices.
2) There are two magic numbers that I can't figure out the meaning behind.
Below is my best guess at what is in the matrix

Code: Select all

k1 = 3.732050808
k2 = 0.267949192431123
(k1*w)/h,      0,      0,            -h/(2*k2)
0,             -k1,    0,             h/(2*k2)
0,             0,      -1.666666,     0
0,             0,       1,            h/(2*k2)


The reason I am trying to figure this out, is I believe the projection matrix might be slightly wrong. I have been testing our engine using nvidia's stereo 3d and all but the interface work perfectly. My best theory is that the projection matrix is resulting in incorrect w values. In case it matters this is using cegui 0.7.2 and "CEGUI::OgreRenderer - Official OGRE based 2nd generation renderer module".

Can anyone shed some light on how the transformation matrix was derived?

ps: The -1.66666 is the result of a fair bit of math involving the near and far distances, but all the terms cancel out and the result is -5/3

[CrazyEddie]

Off the top of my head, I can't actually remember. I'll try to remember to look at it and report back when I have some information

CE

[fedyakin]

I have made a little progress on getting things working in stereo 3d. There is a note in nvidia's best practices paper (http://developer.download.nvidia.com/whitepapers/2010/NVIDIA%203D%20Vision%20Automatic.pdf) that states that scissor clipping is not performed per eye. I happened to notice that the geometry buffer enables scissor clipping, so I commented out that line. Once I did that, some of the interface renders correctly in stereo 3d.

I'm still investigating the remaining broken windows, but perhaps blaming the projection matrix was hasty. Thanks for looking at it, and I hope you didn't waste to much time.

// I think that the 3.732050808 in the projection matrix might be the cot (a/2) with a=30 degrees.