Value of Shadowmeshes

[Springheel]

It's easy to make a shadowmesh for an animated mesh, but it's not easy to make a *good* one. You have to make sure the shadowmesh doesn't stick out anywhere (otherwise it casts ugly shadows onto the visible mesh) both at rest AND while in motion.

 

I was getting a bit drained working on it, so I decided to confirm that it really was important.

 

I ran the following test in a cube room with two shadow-casting lights and 10 AI in motion.

 

With the shadowmesh on: 15 FPS.

 

With the shadowmesh turned off and the full mesh casting shadows: 9-10 FPS.

 

So yes, in case anyone else was wondering, the shadowmeshes are important.

[Tels]

It's easy to make a shadowmesh for an animated mesh, but it's not easy to make a *good* one. You have to make sure the shadowmesh doesn't stick out anywhere (otherwise it casts ugly shadows onto the visible mesh) both at rest AND while in motion.

 

I was getting a bit drained working on it, so I decided to confirm that it really was important.

 

I ran the following test in a cube room with two shadow-casting lights and 10 AI in motion.

 

With the shadowmesh on: 15 FPS.

 

With the shadowmesh turned off and the full mesh casting shadows: 9-10 FPS.

 

So yes, in case anyone else was wondering, the shadowmeshes are important.

 

The shadowmesh should become even more important with complex environment, because every shadow casting polygon will be computed against every shadow-receiving polygon. Could you do a quick test (with maybe some complex models or just some duplicated patches/brushes) to see what the effect really is?

[OrbWeaver]

The shadowmesh should become even more important with complex environment, because every shadow casting polygon will be computed against every shadow-receiving polygon.

 

Shadow volumes are computed once for each light, the number of "receiving" polygons doesn't make any difference to the calculation.

 

A complex environment will of course generate more shadows, and be more expensive to render in itself, but shouldn't have any effect on the shadows which are falling on it.

[Tels]

Shadow volumes are computed once for each light, the number of "receiving" polygons doesn't make any difference to the calculation.

 

A complex environment will of course generate more shadows, and be more expensive to render in itself, but shouldn't have any effect on the shadows which are falling on it.

 

But what if a shadow polygon falls on 1 vs. 3 polygons. Doesn't this make the shadow-casting process more complicated?

[OrbWeaver]

But what if a shadow polygon falls on 1 vs. 3 polygons. Doesn't this make the shadow-casting process more complicated?

 

No, because the algorithm does not involve calculating intersections between shadow volumes and rendered geometry; it is per-pixel, not per-vertex or per-poly.

 

Wikipedia has a reasonably good description of stencil shadows. Note how the shadow effect arises from rendering the shadow volume itself into the stencil buffer, and then using this buffer to test against the depth values of the final rendered scene to determine whether a given pixel is in shadow or not.

[Tels]

No, because the algorithm does not involve calculating intersections between shadow volumes and rendered geometry; it is per-pixel, not per-vertex or per-poly.

 

Wikipedia has a reasonably good description of stencil shadows. Note how the shadow effect arises from rendering the shadow volume itself into the stencil buffer, and then using this buffer to test against the depth values of the final rendered scene to determine whether a given pixel is in shadow or not.

 

Ah thanx. So the amount of work in this technique (which is used in D3) is a per-light pass, and thus depends only on the shadow polygons and the screen resolution, not what is in shadow or not. (If I understood that correctly)

[OrbWeaver]

Ah thanx. So the amount of work in this technique (which is used in D3) is a per-light pass, and thus depends only on the shadow polygons and the screen resolution, not what is in shadow or not. (If I understood that correctly)

 

Yes, exactly. That is why light count and shadow-mesh complexity are so much more important than the polycount of rendered objects themselves.