Some shadow map optimizations for Wicked Engine, now 16 viewports can be fully utilized for multiple lights' shadow draw calls with instancing. 256 dynamic pointlight shadows in Sponza:

Now even multiple light types can be fitted into one draw: - directional: multiple cascades - point: up to 6 frustums - spot/rect: 1 frustum Previously the viewport instancing only happened for cascades and point lights, separately. Now all of them are batched.

Viewports/cameras are created for visible light frustums while they can fit into the 16 available viewports (graphics API limit). Once 16 viewports are full, the scene is rendered once into all viewports. Instancing is used to select SV_ViewportArrayIndex by VS as before.

About 5ms CPU time for shadow render in this scene (depends heavily on culling)

Nice :) If I understand it , instead of geometry shader you are doing multiple instances and selecting the frustum for eg in Pointlights?

Yes, the vertex shader gets the camera/viewport index from the instancing data.

Thanks for the clarification. Love your work on wicked engine.

*inspects closer* *gets dabbledazed by disco lights* Yeah that seems wicked enough.

Does the resolution of these frustums need to match?

No, resolution is dynamic for each light, depends on its screen size.