Skip to main content
GameDev.net gamedev.net
🔒 Locked

GLSL access depth buffer from shader

Started by jay8ee Feb 18, 2010 at 6:12 PM 16 replies 28.6k views
Original Post
jay8ee
jay8ee
Hey guys, I want to read the depth buffer of my FBO in one of my shaders so I can run some DOF equation. I am having trouble reading the buffer however. I setup the depth texture with GL_DEPTH_STENCIL_EXT so I have the stencil value to play with too, but thats another issue for another day. 

glBindTexture(GL_TEXTURE_2D, textures[depth]);
glTexImage2D(GL_TEXTURE_2D, 0, GL_DEPTH24_STENCIL8_EXT, wWidth, wHeight, 0, GL_DEPTH_STENCIL_EXT, GL_UNSIGNED_INT_24_8_EXT, 0);
glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST );
glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST );
glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE );
glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE );

then..

glFramebufferTexture2DEXT(GL_FRAMEBUFFER_EXT, GL_DEPTH_ATTACHMENT_EXT, GL_TEXTURE_2D, textures[depth], 0);
I don't how I access just the depth component of the texture from a shader (if it is possible at all?). Can I pass the the depth texture as a sampler? If anyone can help, then thanks.
coordz
coordz
For sure you can just bind the depth as a sampler. Just make sure you set your depth comparison to NONE.
jay8ee
jay8ee
Huh? I don't understand.. If I set the texture as a sampler, how do i read of the first 24bits and leave out the stencil value?
idinev
idinev
The texture2D() automatically gives you values in the [0;1] range. The value is (z/w+1)/2
Hodgman
Hodgman
Quote:
Original post by idinev
The texture2D() automatically gives you values in the [0;1] range. The value is (z/w+1)/2
To expand on that answer, you just sample the red component of the depth buffer, and you get the full 24 bit value as a float.
float depth = texture2D( depthBuf, texCoords ).r;
jay8ee
jay8ee
Quote:
Original post by Hodgman
Quote:
Original post by idinev
The texture2D() automatically gives you values in the [0;1] range. The value is (z/w+1)/2
To expand on that answer, you just sample the red component of the depth buffer, and you get the full 24 bit value as a float.
float depth = texture2D( depthBuf, texCoords ).r;


Awesome, I'm just on my way to work so i'll give this a blast later. If this is the case as you say, can you access the stencil value with .b?
jay8ee
jay8ee
float depth = texture2D(depthImg, gl_TexCoord[0].xy).r;

this doesn't work after all. Always seems to equate to 1 :S. I can't work it out. Here is what I got to try and just output the depth as colour...

uniform sampler2D depthImg;void main(){	float depth = texture2D(depthImg, gl_TexCoord[0].xy).r;	gl_FragColor = vec4(depth, depth, depth, 1.0);}


and in the c++ code...

glActiveTexture(GL_TEXTURE0);glBindTexture(GL_TEXTURE_2D, textures[depth]);shaderDepthColour.UseProgram();RenderFullscreenQuad();


seems like I am only getting a white screen though :(


idinev
idinev
Yes, it "seems"

Try this in your shader, to convert to linear:
	float LinearizeDepth(float zoverw){		float n = 1.0; // camera z near		float f = 20000.0; // camera z far		return (2.0 * n) / (f + n - zoverw * (f - n));	}uniform sampler2D depthImg;void main(){	float depth = texture2D(depthImg, gl_TexCoord[0].xy).r;	depth = LinearizeDepth(depth)*77;	gl_FragColor = vec4(depth, depth, depth, 1.0);}


jay8ee
jay8ee
Quote:
Original post by idinev
Yes, it "seems"

Try this in your shader, to convert to linear:
*** Source Snippet Removed ***


This doesn't work either... is it potentially because I am using the wrong format for my texture?

glTexImage2D(GL_TEXTURE_2D, 0, GL_DEPTH24_STENCIL8_EXT, wWidth, wHeight, 0, GL_DEPTH_STENCIL_EXT, GL_UNSIGNED_INT_24_8_EXT, 0);

perhaps when using float data type in the shader the value is getting clamped because it is a 24 uint in the texture?

idinev
idinev
Hopefully you simply didn't change the "n" and "f" vars to match your persp-proj values. Also toy with the "77" value .

Or check if really the depth==1.0 :
float depth = texture2D(depthImg, gl_TexCoord[0].xy).r;depth = depth==1.0 ? 0 : 1; gl_FragColor = vec4(depth, depth, depth, 1.0);


The other stuff is: glTexParami(gl_texture_compare_mode) - has to be gl_none, otherwise the sampler will be doing shadowmap comparison (which will result in the exact problem you currently describe).
jay8ee
jay8ee
still no joy on this :/ seems like I'm doing something really wrong, or I am missing something so stupid... If I write the depth value from the shader to just the red channel, then the screen is infact red. So depth is clamping to 1.0.

I'll break it all down, maybe someone can see my school boy error.

Vertex shader:
// shaderDepthColour.vsvoid main(){	gl_TexCoord[0].xy = gl_MultiTexCoord0.xy;	gl_Position = gl_Vertex;}


Fragment Shader:
// shaderDepthColour.fs
float LinearizeDepth(float zoverw){	float n = 1.0; //	float f = 20000.0;		return(2.0 * n) / (f + n - zoverw * (f - n));	}uniform sampler2D depthImg;void main(){	float depth = texture2D(depthImg, gl_TexCoord[0].xy).r;	depth = LinearizeDepth(depth)*77;	gl_FragColor = vec4(depth, depth, depth, 1.0);}


CPP code:
////////////buffer setup////enum TypesTexure{ colour, normal, depth, texTypeCount, };glGenTextures(texTypeCount, textures);// colour buffer textureglBindTexture(GL_TEXTURE_2D, textures[colour]);glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA32F_ARB, wWidth, wHeight, 0, GL_RGBA, GL_FLOAT, 0);glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);// normal buffer textureglBindTexture(GL_TEXTURE_2D, textures[normal]);glTexImage2D(GL_TEXTURE_2D, 0, INTERNAL_FORMAT, wWidth, wHeight, 0, GL_RGBA, GL_FLOAT, 0);glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);// depth bufferglBindTexture(GL_TEXTURE_2D, textures[depth]);glTexImage2D(GL_TEXTURE_2D, 0, GL_DEPTH24_STENCIL8_EXT, wWidth, wHeight, 0, GL_DEPTH_STENCIL_EXT, GL_UNSIGNED_INT_24_8_EXT, 0);glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_COMPARE_MODE, GL_NONE);glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST );glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST );glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE );glTexParameteri( GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE );// framebufferglGenFramebuffersEXT(1, &frameBuffer);glBindFramebufferEXT(GL_FRAMEBUFFER_EXT, frameBuffer);glFramebufferTexture2DEXT(GL_FRAMEBUFFER_EXT, GL_COLOR_ATTACHMENT0_EXT, GL_TEXTURE_2D, textures[colour], 0);glFramebufferTexture2DEXT(GL_FRAMEBUFFER_EXT, GL_COLOR_ATTACHMENT1_EXT, GL_TEXTURE_2D, textures[normal], 0);glFramebufferTexture2DEXT(GL_FRAMEBUFFER_EXT, GL_DEPTH_ATTACHMENT_EXT, GL_TEXTURE_2D, textures[depth], 0);glFramebufferTexture2DEXT(GL_FRAMEBUFFER_EXT, GL_STENCIL_ATTACHMENT_EXT, GL_TEXTURE_2D, textures[depth], 0);////////// render pass 1 (write textures)////glBindFramebufferEXT(GL_FRAMEBUFFER_EXT, frameBuffer);const GLenum buffersToUse[] = { GL_COLOR_ATTACHMENT0_EXT, GL_COLOR_ATTACHMENT1_EXT };glDrawBuffers(2, buffersToUse);glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);shaderColour.UseProgram();glEnable(GL_CULL_FACE);glEnable(GL_DEPTH_TEST);/// render as normalglDisable(GL_CULL_FACE);glDisable(GL_DEPTH_TEST);glDrawBuffer(GL_COLOR_ATTACHMENT0_EXT);glBindFramebufferEXT(GL_FRAMEBUFFER_EXT, 0);////////// render pass 2 draw depth////glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);glActiveTexture(GL_TEXTURE0);shaderDepthColour.UseProgram();glBindTexture(GL_TEXTURE_2D, textures[depth]);RenderFullscreenQuad();


I can render the normal colour just fine, but no depth :/

Sorry for wall of text
idinev
idinev
But did you try the "n, f" and depth==1.0 suggestions?

You _must_ set both "n" and "f" to the zNear and zFar values you passed in the gluPerspective() .

Separately, try the "depth= depth==1.0 ? 0:1;" approach, to have the gpu check whether the values are constant 1.0 , or it's just the RGBA8 precision tricking you into thinking it's all pure white. (in most scenes, most pixels will have values around [0.9996;0.9999999] )
jay8ee
jay8ee
Quote:
Original post by idinev
But did you try the "n, f" and depth==1.0 suggestions?

You _must_ set both "n" and "f" to the zNear and zFar values you passed in the gluPerspective() .

Separately, try the "depth= depth==1.0 ? 0:1;" approach, to have the gpu check whether the values are constant 1.0 , or it's just the RGBA8 precision tricking you into thinking it's all pure white. (in most scenes, most pixels will have values around [0.9996;0.9999999] )


I set the values to my gluperspective (0.01 and 100.0) and I also tried the "depth= depth==1.0 ? 0:1;" approach, that still just results in a white screen, (or red if i send it to just the alpha channel). Really don't know what the hell is going on. The scene when rendered as geom colour or normal colour or indeed fully shaded looks like it should in terms of depth, so there must be actual values in the buffer, right?
idinev
idinev
^^"

"depth= depth==1.0 ? 0:1;" -> if this results in white, then depth!=1.0

Also, again:

float LinearizeDepth(float zoverw){	float n = 0.01; // You had to change this	float f = 100.0; // and this		return(2.0 * n) / (f + n - zoverw * (f - n));	}
jay8ee
jay8ee
Quote:
Original post by idinev
"depth= depth==1.0 ? 0:1;" -> if this results in white, then depth!=1.0


oh of course xD. But then if I run it through the LinearizeDepth func it still give white screen :S


idinev
idinev
Did you vary that "77" value I noted about?
jay8ee
jay8ee
Omg you are an absolute hero..

I did mess with the '77' value, but not enough :P. If i put in in the range of 1 - to 5, the I can actually see the depth! anything above 6 and its all white.

Thanks for sticking with this and helping me even though i'm such a dunce xD.
idinev
idinev
You're welcome :)

Topic Locked

This topic has been locked by a moderator. New replies are not allowed.

Sign in to reply to this topic.