Is ovr_GetFovTextureSize special?

I pointed out this to people at #pypy, they told me it's because the cpython bunded libffi library had issue with returning structs on windows. It's been fixed but suppressed because cpython has fixed the problem in ctypes instead of libffi.

They told that the problem is described in rpython/translator/c/src/libffi_msvc/prep_cif.c:119 just mentioning because you tend to find this kind of posts in google without enough information in them.

At 0.8.0 version of oculus rift SDK, the answer to FovPort -> Projection can be found from the LibOVR/Src/OVR_StereoProjection.cpp

I use ovrMatrix4f_Projection(). You give it the eye fov structure and the desired near and far planes and it returns a perspective matrix.
The matrix is a little different to normal because the rift screens and lenses are asymmetric, the DK2 lenses aren't centred on the screen horizontally (slightly more fov to the sides of the head than the nose) and the CV1 has more downwards fov than upwards. So the view frustum is skewed.

For

"cheery" wrote:
I guess I could work around the function entirely if I knew what it did exactly.

ovr_GetFovTextureSize works out what resolution texture you need so that the pixels in the centre of the view are the specified size relative to the native display panel resolution.
The barrel distortion that the sdk does makes pixels in the centre bulge out bigger. To combat this, you need to render at higher than 1920x1080, so when they bulge (effectively becoming lower res) they become 1920x1080.
Under standard setting with 1:1 pixel scale the texture resolution returned by ovr_GetFovTextureSize on a DK2 is 1182x1464 (that's for one eye). That's what you need to render at to get 1080p on screen after the distortion step.

You can pick any values for the texture size though, if you don't mind lower quality. No idea what CV1 will return (we know it's panel resolution: 1080x1200 each eye, but we don't know the distortion values).

I have it rendering on DK1, but would like if it worked on CV1. I guess I'll go and try to fix the issue on pypy.

So far, I've got render on both left and right eye. I also receive tracking data. There is still something weird though. I wonder if I have filled the LayerFov and ViewScaleDesc correct. On the left eye I only see the right corner of the left-side image, and on the right eye I only see the left corner of the right-side image. It is very heavily distorted.

Here's what I have so far in code: http://pastebin.com/PeWpFi7C

I hope I also had the equations to calculate the eye poses and respective matrices to position the models. But now even static posing doesn't seem to converge. I'll try it again tmr.

On line 408, I had mixed up .x and .y coordinates when I copied over the perspective matrix. That caused this kind of issue with rendering:



Now I got it solved and the image converges. Once I fix the FFI issue I have, lever will be officially ready for Oculus!

Newbie question: What is the difference between pixelsPerDisplayPixel for ovr_GetFovTextureSize and MSAA?

Sorry I'm borrowing this topic since I'm a new member :)