3DMark06, Cascaded Shadow Maps (CSM)

Started by
34 comments, last by wolf 18 years, 2 months ago
Hey zedzeek,
where does the light come from on your figure ... does it go into the same direction as the view direction?
If the light is coming from a different direction, this should look different? Maybe like this:



well ... actually the third frustum is wrong in that figure ... but overall it shows all the problems.

- Wolf

P.S: I am still trying to optimize the whole light frustum calculation ...

[Edited by - wolf on February 16, 2006 1:22:44 PM]
Advertisement
Hi eq,
I will get MSAA for the render target on the hardware platform I work ... so this is not a problem for me. Isn't that possible on PC graphics cards as well?

- Wolf
Quote:Original post by Guoshima
I see, but like you mentioned. How do you take care of tall object casting shadows into the scene like this, but not visible in your frustum. That's what I had this big region behind me.


Sorry but I fail to see what the problem is. The light view uses an orthographic projection matrix. I choose this matrix such that the frustum is fully visible. This defines the x and y parameters for the projections. Then I compute the z parameters in such a way that all shadow casters in the scene (not just the current frustum) are included (i.e. the minimum and maximumg z-value as seen from the light for all objects disregarding their frustum). Thus the projection matrix projects all shadow casters that can project a shadow in the current frustum into the unit cube. What's wrong with this idea?
true .. I was just a little bit mixed up. I first though about dividing the currently culled meshes from the view frustum into different regions but this of course doesn't work. You have to perform your culling again for each region you want to render, and then it works fine. Sorry for the misunderstanding.
Quote:Original post by wolf
Hi eq,
I will get MSAA for the render target on the hardware platform I work ... so this is not a problem for me. Isn't that possible on PC graphics cards as well?
- Wolf


The render target is MSAA, but if you need to use THAT target as a texture it has to be "downsampled" to a non AA'd surface.
If you can get away with the lighting without using the "shadow mask" as a texture (i.e might be able to use destination alpha as the shadow mask), then there's no problem.


<<<
The render target is MSAA, but if you need to use THAT target as a texture it has to be "downsampled" to a non AA'd surface.
If you can get away with the lighting without using the "shadow mask" as a texture (i.e might be able to use destination alpha as the shadow mask), then there's no problem.
<<<
Oh I see.

As long as I do not want to blur the a screenspace black-and-white texture, I might just render everything four times into the framebuffer ... this might be expensive, because I have to use all those expensive shaders, but I would save the in-between step.

Let's see how it comes out ...

- Wolfgang

This topic is closed to new replies.

Advertisement