Check out http://humus.name/index.php?page=3D&ID=81

Be carefull thought, all of his code uses column major matrix' for directx.

Couldn't you follow a tutorial which outlines the concept in DirectX9, then apply what you've learnt to DirectX11?

That's what I did. Works a treat. Also, updating all the code to work with DirectX11 ensured I knew exactly what was going on, rather than simply copy and pasting. You should find areas that can be optimized also, as most tutorials are designed to be understood rather than written for performance.

If you understand the concept of deferred shading, you should be able to implement it using just the msdn.

Well, the code difference between directx 10 and 11 is very very small. The only difference is you have function calls with an 11, instead of a 10. If you get the directx 10 version working, then just convert it to 11. Should take like.... 30 minutes to do the conversion. But, do not convert before you even get the directx 10 version working --thats asking for compounded problems.

Quote:Original post by simotix
If there is something that shows a step by guide for DirectX10, I would of course follow that, but if there is, could someone please tell me what improvements DirectX11 has given in the context of Deferred Rendering over DirectX10?

Off the top of my head...

1. You can create "read-only" depth stencil views. This is useful if you sample the depth buffer directly for reconstructing position. In D3D10, you couldn't have a depth stencil buffer bound for depth testing AND sample it as a shader resource view simultaneously. Now you can do it if you create a read-only view, and make sure that depth/stencil writes are disabled.

2. This is actually a 10.1 feature, but still useful in 11: you can run pixel shaders at per-sample frequency. Handy for enabling MSAA with a light prepass/deferred lighting setup, since you need to calculate and store lighting at sample resolution.

3. You can take SV_Coverage as an input for the pixel shader. This lets you avoid shading subsamples where the depth/stencil or coverage test failed when MSAA is enabled. Otherwise you would have to run the shader at per-sample frequency. SV_Coverage also gives you a more direct way of detecting edge pixels, rather than using the centroid sampling trick demonstrated in one Humus's samples.

4. Another 10.1 feature: you can sample MSAA depth buffers.

5. You can use compute shaders to perform your lighting on a fixed grid. This is the fastest way to shade many (hundreds) of lights. See this sample.