Forum Discussion
Oculus SDK and Stereoscopic Rendering Design
I'm pretty new to stereoscopic rendering but I'm wondering about something trying to implement the SDK.
Since we need to render to a bigger render target than the backbuffer anyways, why is Oculus using a full-sized render target and only renders to half of it for each eye?
I dont get the idea behind it. Why dont you just render to a [width*renderscale/2, height*renderscale] rendertarget for each pass?
For me it seems like a lot of unnecessary overhead (viewport adjustments, etc..) and wasted graphics memory (half of the bigger render target is always unused).
Since we need to render to a bigger render target than the backbuffer anyways, why is Oculus using a full-sized render target and only renders to half of it for each eye?
I dont get the idea behind it. Why dont you just render to a [width*renderscale/2, height*renderscale] rendertarget for each pass?
For me it seems like a lot of unnecessary overhead (viewport adjustments, etc..) and wasted graphics memory (half of the bigger render target is always unused).
2 Replies
- I'm not entirely sure what you mean. Are you asking,
"Why are the Rift's stereoscopic buffers side-by-side instead of interleaved frame-by-frame?"
or are you asking
"Why do the Rift SDK samples render to one large 1280x800 rendertarget instead of to two smaller 640x800 rendertargets?"
or is your question about something that I'm missing? - Almost the second question: "Why do the Rift SDK samples render to one large 1280x800 * ~1.7 rendertarget instead of to one smaller 640x800 * ~1.7 rendertarget?"
After the first eye is rendered, you can just clear the one-eye rendertarget and redraw on it (the final backbuffer is a different rendertarget).
I guess this is not really an issue, just preference. In the meantime, I adapted the Oculus shader to work on a single-eye texture instead of the two-in-one texture.
I just think the whole setup is a little complicated and could be done in a more simple way, so that it's easier to understand.
