Bypassing auto hardware shadow maps
Author
I'm trying to allow my texture alpha values to show 'holes' in my shadow map (ie, so that tree textures can become shadow mapped). It seems that the NVIDIA hardware (and newer versions of the ATI) have hardware shadow mapping built into the shading language, so that when a tex2Dproj is performed, shadow map calculates are automatically computed for you.
I've tried stepping around it, (ie enabling alpha testing before rendering the objects) and for some reason, i can't seem to alter the shadow map image.
Has anyone else ran into this problem? Or atleast offer some advice?
thanks
FROG!
This has nothing to do with tex2Dproj, nothing. It's all about how you genereate the shadow map. Enabling alpha test when rendering to the shadow map should do the trick.
Sunray, simply enabling alpha testing wont work if you are using floating point textures.
ElFrogo, what you'll need to do is use a ARGB texture and store the floating point value (the depth) in the RGB channels and the alpha of the texture in the alpha channel.
Take a look at this thread here where I explain how to pack a float into 24bits:
http://www.gamedev.net/community/forums/topic.asp?topic_id=322318
Your shadow mapping pixel shader then becomes:
Then enable alpha testing and you should be good to go. This will work with any ps 2.0 card, not just cards with hardware shadow maps.
ElFrogo, what you'll need to do is use a ARGB texture and store the floating point value (the depth) in the RGB channels and the alpha of the texture in the alpha channel.
Take a look at this thread here where I explain how to pack a float into 24bits:
http://www.gamedev.net/community/forums/topic.asp?topic_id=322318
Your shadow mapping pixel shader then becomes:
PS_OUTPUT ShadowMapPS(VS_OUTPUT vs){ PS_OUTPUT o; // // Sample the texture // float4 textureSample = tex2D(TextureSampler, vs.textureCoords); // // Pack the depth value into 24 bits (RGB) and place the texture alpha into (A) // o.colour = float4(PackDepth24(vs.depth), textureSample.a); return o;}Then enable alpha testing and you should be good to go. This will work with any ps 2.0 card, not just cards with hardware shadow maps.
Author
hey buttza, thanks. but won't that limit my precision to 24 bits?
Also, won't a simple clip(x) work? or is texkill still too performance heavy on sm3.0 cards??
EDIT: Let me rephrase that: When rendering an object into my r32 texture, I called clip(-1) on objects with an alpha value <.2f; However, all the pixels were drawn. I modified the algo, to call clip(-1) on ALL pixels drawn, and still, they all came through in the shadow map. I'm starting to think that my understanding on how exactially the pixel data gets into the 'shadow map.' If we're doing a texkill on every pixel, shouldn't nothing be rendered!?
FROG!
[Edited by - ElFrogo on June 1, 2005 9:59:14 AM]
Also, won't a simple clip(x) work? or is texkill still too performance heavy on sm3.0 cards??
EDIT: Let me rephrase that: When rendering an object into my r32 texture, I called clip(-1) on objects with an alpha value <.2f; However, all the pixels were drawn. I modified the algo, to call clip(-1) on ALL pixels drawn, and still, they all came through in the shadow map. I'm starting to think that my understanding on how exactially the pixel data gets into the 'shadow map.' If we're doing a texkill on every pixel, shouldn't nothing be rendered!?
FROG!
[Edited by - ElFrogo on June 1, 2005 9:59:14 AM]
ElFrogo, yes it will limit you to 24bits, but in my experience it's enough. 16bit shadow maps should be enough in small areas or for small lights.
Thanks for the suggestion on using clip, i'd never thought of using that. I tried it and it is certainly fast enough with no apparent difference between my way and using clip. I'll use clip in my next game though.
I simply used:
clip(-0.4f + textureSample.a);
And it worked fine, I'm not sure why you're having problems. I'm using sm2.0 btw.
Thanks for the suggestion on using clip, i'd never thought of using that. I tried it and it is certainly fast enough with no apparent difference between my way and using clip. I'll use clip in my next game though.
I simply used:
clip(-0.4f + textureSample.a);
And it worked fine, I'm not sure why you're having problems. I'm using sm2.0 btw.
Author
Hey buttza, thanks for the help. Turns out it was a shader pipeline issue.
Clip works very well, i agree.
Thanks for the input
FROG!
Clip works very well, i agree.
Thanks for the input
FROG!
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement