View Full Version : Shadow mapping with Detonator 52.16 - Broken?

10-27-2003, 02:22 AM
Has anyone had any problems with shadow mapping with the new detonator release? I just noticed that it's broken in my app and I've tried the "Hardware Shadow Maps (cg)" effect in the Effects browser and I get the same results.

Basically everything comes out black (even my skybox - which is odd - it might be that something is not turning off as it should). Sometimes the very edges of the objects are partially visible.

10-27-2003, 02:36 AM
Hmmm. I've run some other demos and they appear to work OK (Humus' shadows that rock, Ikka's Engdemo)

10-27-2003, 05:57 AM
I too get a similar problem. In my case I think that previously when a pixel did not appear in the shadow map it would be assigned as NOT IN shadow. It now looks like pixels that do not appear in the shadow map are being assigned as IN shadow.

10-27-2003, 09:17 AM
The shadow map demos in the effects browser work for me. I have a GeForce FX 5600 Ultra.


10-27-2003, 02:53 PM
In my case it looks like the z-values might be being distorted in some way. That would explain why the edges of objects are lit and the rest is in shadow (I don't have anything outside the shadow area so I don't know what is happening).

I have a 5900 Ultra - what do you have MattS?

10-27-2003, 07:08 PM
Originally posted by rgpc:
Hmmm. I've run some other demos and they appear to work OK (Humus' shadows that rock, Ikka's Engdemo)

The difference just dawned on me, these two demos (AFAIK) use FP's whereas I do not...

10-29-2003, 07:59 AM
I have a 5800 FX. nVidia were giving them away at a conference. After plugging it in I found out why. I recently got an ATI 9800 (which is lovely btw).

Anyway I get the same effect on this new card that I saw on the 5800. Unfortuately due to recent hardware crashes/changes and software changes I cannot be sure what has happened. Either the nVidia drivers have changed some default value or (more likely) I have changed something in my code which results in the different results. I will have a play with some old and new builds on a machine with older drivers and report back.

Oh yeah meant to mention originally but I think the "shadows that rock" demo does not use hardware shadow mapping but instead does it in a fragment program since it is using a cube map to do shadow mapping for a point light. Oh just read you last post, you noticed this.


10-29-2003, 08:43 AM
OK, I've done some testing.

The problem I was experiencing was indeed my fault. Apologies if this gave you false hope. In my case it was due to border colour being set differently.

Once I sorted that out shadow mapping appears to work as expected using both render to depth texture and copy to depth texture. It also works in both fixed function and fragment programs. In my case the fragment program assumes the texture behaves as a shadow map since nVidia's drivers do this.


10-29-2003, 03:36 PM
Yes I remember receiving an invitation to the nVidia conference where they were giving away the fx's (thought it might have been a 5200 though - didn't say in the e-mail). Main problem was the conference was on the other side of the planet to me...

As for the shadow mapping problem, I have reported it to nVidia as a possible bug, haven't had any response though (flagged it as low priority - I prefer the stencil shadows anyhow).