what camera does the built-in shader variables reference during a custom ShadowCaster pass

Hi!

I'am currently doing a custom vert shader for my materials that snaps each object to a grid. That part is done and working, however when I try to do the same thing with a custom ShadowCaster pass, I get these artifacts:

They do not appear when I use the "Legacy Shaders/VertexLit/SHADOWCASTER" pass. However, then, the shadow is not at the correct place. (duh!)

IF the built-in shader variables reference the same things (ie camera etc) as my main pass does, I see no reason for these artifacts to occur, given that the vertices should be in the same position on both passes no matter the camera.

So, what differs from the ShadowCaster pass compared to my main pass? Variable wise ofc. Or does that pass only run once per draw call to populate the declared shadowmap? Ignoring how many cameras are in the scene, since the shadowmap is in its own shadowspace (maybe in reference to the light object)? Idk im just spitballing tbh.

I use:
unity_ObjectToWorld
unity_WorldToObject
UNITY_MATRIX_VP
UnityObjectToClipPos()

And render using multiple orthographic cameras.

 

unity_ObjectToWorld

and

unity_WorldToObject

will be exactly the same. The

UNITY_MATRIX_VP

, and by extension the

UnityObjectToClipPos()

which uses it, will be completely different. This is because the VP matrix is for what the currently rendered view is. When rendering the shadow maps the VP matrix is from the “view” of the light. For directional lights this is a view projection matrix (or several if using cascades) that isn’t exposed anywhere to c#.

If you’re snapping to a world space grid, there shouldn’t be a problem. If you’re snapping to a screen space grid you’ll likely want to use the

unity_CameraProjection

and

unity_CameraInvProjection

matrices, along with

unity_WorldToCamera

and

unity_CameraToWorld

to get consistent view or clip space positions between the camera rendering and shadow map rendering.