Jump to content
  • Advertisement
Sign in to follow this  
Vanderry

OpenGL Shadow mapping with GLSL

This topic is 1199 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

Hey guys!

 

I'm having some serious trouble implementing shadow mapping in my OpenGL/OpenGL ES application. There are many sources and discussions out there, but they often cover different versions of the API or omit some details. The way I'm doing it is by first constructing an orthographic projection matrix based on a light source. The "frustum" is adjusted to contain a terrain mesh.

 

Vector3 offset = {1.0f, -1.0f, 1.0f}; // Light direction
offset.Normalize();
offset *= 30.0f;
Vector3 center = {15.0f, 0.0f, 15.0f};
Vector3 eye = center - offset;
Vector3 up = {0.0f, 1.0f, 0.0f};
Matrix4 viewMatrix = LookAt(eye, center, up);

Matrix4 projectionMatrix = Ortho(-30.0f, 30.0f, -30.0f, 30.0f, 0.0f, 60.0f); // Depth range is 60.0

Matrix4 shadowMatrix = viewMatrix * projectionMatrix;

 

Then I am drawing all the models that populate the world to the shadow map, and projecting it onto the terrain. It looks great for things that are positioned above ground. The problem emerges when they are not, in which case they are projected just the same. Naturally they should appear in the shadow map, so I am assuming that it is the depth comparison that fails. The shadow map is being rendered directly to a texture (no render buffer involved), with settings GL_RGBA and GL_UNSIGNED_BYTE. The attachment is done through GL_COLOR_ATTACHMENT0, and for the color value I have tried to use a linearized and normalized gl_FragCoord.z, but now I am more inclined to use the same manually projected vertex z that is used in the other render step. The shaders look roughly like this:

 

// Shadow vertex shader
attribute vec4 a_Position;
uniform mat4 u_ShadowMatrix;
uniform mat4 u_WorldMatrix;
varying vec4 v_ShadowCoordinate;
void main()
{
  v_ShadowCoordinate = u_ShadowMatrix * u_WorldMatrix * a_Position;
  gl_Position = v_ShadowCoordinate;
}
 
// Shadow fragment shader
varying vec4 v_ShadowCoordinate;
void main()
{
  gl_FragColor = vec4(v_ShadowCoordinate.z / 60.0); // Without packing
}
 
// Render vertex shader
uniform mat4 u_WorldMatrix;
uniform mat4 u_ViewProjectionMatrix;
uniform mat4 u_ShadowMatrix;
attribute vec4 a_Position;
attribute vec2 a_TextureCoordinate;
varying vec2 v_TextureCoordinate;
varying vec4 v_ShadowCoordinate;
void main()
{
  // I have a suspicion this might mess up the z, but it's vital for the projection
  mat4 shadowBiasMatrix = mat4(
    0.5f, 0.0f, 0.0f, 0.0f,
    0.0f, 0.5f, 0.0f, 0.0f,
    0.0f, 0.0f, 0.5f, 0.0f,
    0.5f, 0.5f, 0.5f, 1.0f);
 
  v_TextureCoordinate = a_TextureCoordinate;
  v_ShadowCoordinate = shadowBiasMatrix * u_ShadowMatrix * u_WorldMatrix * a_Position;
  gl_Position = u_ViewProjectionMatrix * u_WorldMatrix * a_Position;
}
 
// Render fragment shader
uniform sampler2D u_Texture;
uniform sampler2D u_ShadowMap;
varying vec2 v_TextureCoordinate;
varying vec4 v_ShadowCoordinate;
void main()
{
  vec4 textureColor = texture2D(u_Texture, v_TextureCoordinate);
  vec4 shadowMapColor = texture2D(u_ShadowMap, v_ShadowCoordinate.xy);
  if ((shadowMapColor.x * 60.0) < v_ShadowCoordinate.z) // Same projection as in the shadow shader, right?
    shadowMapColor = vec4(vec3(0.5), 1.0);
  else
    shadowMapColor = vec4(1.0); // No shadow
  
  gl_FragColor = textureColor * shadowMapColor;
}
 
I have tried different ways of computing the shadow map, with more elaborate packing methods. Ultimately I am assuming that OpenGL clamps the values between 0.0 and 1.0, or rather 0 and 255 as in the case of my texture format. There has to be a more obvious problem with my implementation. I can mention that changes to the render state such as dither, depth test and face culling doesn't make a noticeable difference. Can anyone think of a possible flaw? Any tips are greatly appreciated.
 
- David
Edited by Vanderry

Share this post


Link to post
Share on other sites
Advertisement

I guess I am getting better results from using gl_FragCoord.z in the shadow fragment shader, in which case the problem is how to depth test the fragments. I did at one point pass the target texture into the shader, to write the minimum of the old and new z but that didn't work so well in OpenGL ES...

Share this post


Link to post
Share on other sites

The shadow map is being rendered directly to a texture (no render buffer involved), with settings GL_RGBA and GL_UNSIGNED_BYTE.

That would be the first problem. Render shadows to float textures.
 

gl_FragColor = vec4(v_ShadowCoordinate.z / 60.0); // Without packing

You didn’t divide by .w. 60.0 is an arbitrary number—there is no reason to expect that you will get correct results this way.
gl_FragColor = vec4(v_ShadowCoordinate.z / v_ShadowCoordinate.w);
gl_FragColor = vec4(gl_FragCoord.z); // Alternative.
 

 

mat4 shadowBiasMatrix = mat4(
    0.5f, 0.0f, 0.0f, 0.0f,
    0.0f, 0.5f, 0.0f, 0.0f,
    0.0f, 0.0f, 0.5f, 0.0f,
    0.5f, 0.5f, 0.5f, 1.0f);

Shouldn’t this be column-major?
 

v_TextureCoordinate = a_TextureCoordinate;
  v_ShadowCoordinate = shadowBiasMatrix * u_ShadowMatrix * u_WorldMatrix * a_Position;

This should all be moved to the CPU, I’ve never heard of the shadow coordinate being set in the vertex shader (I would have to spend more time than I am willing to think about whether or not it is possible), below this you have magic numbers again—let’s just stop there.


This section needs to be rewritten.
There is plenty of source code here: http://codeflow.org/entries/2013/feb/15/soft-shadow-mapping/#hard-shadow-mapping
And here: http://www.opengl-tutorial.org/intermediate-tutorials/tutorial-16-shadow-mapping/


L. Spiro Edited by L. Spiro

Share this post


Link to post
Share on other sites


Vanderry, on 09 Mar 2015 - 8:32 PM, said:
The shadow map is being rendered directly to a texture (no render buffer involved), with settings GL_RGBA and GL_UNSIGNED_BYTE.
That would be the first problem. Render shadows to float textures.

 

Right you are. That takes care of all the packing business.

 


Vanderry, on 09 Mar 2015 - 8:32 PM, said:
gl_FragColor = vec4(v_ShadowCoordinate.z / 60.0); // Without packing
You didn’t divide by .w. 60.0 is an arbitrary number—there is no reason to expect that you will get correct results this way.
gl_FragColor = vec4(v_ShadowCoordinate.z / v_ShadowCoordinate.w);
gl_FragColor = vec4(gl_FragCoord.z); // Alternative.

 

I could see how this would be appropriate when using the depth value calculated by OpenGL, to linearize it if my understanding is correct. I have been assuming that a manual transformation wouldn't suffer the same "skewing". After all, if I performed the same calculation in C++, then the results should be linear, and w would remain 1.0. The 60.0, or view depth, was a desperate attempt to contain the result within the 0.0-1.0 range. I would use a uniform in practice. With the many opportunities for error here, please don't hesitate to correct me.

 


Vanderry, on 09 Mar 2015 - 8:32 PM, said:
mat4 shadowBiasMatrix = mat4(
    0.5f, 0.0f, 0.0f, 0.0f,
    0.0f, 0.5f, 0.0f, 0.0f,
    0.0f, 0.0f, 0.5f, 0.0f,
    0.5f, 0.5f, 0.5f, 1.0f);
Shouldn’t this be column-major?

 

My math support classes are assuming row-majority. It would have been a good catch if this wasn't the case.

 


Vanderry, on 09 Mar 2015 - 8:32 PM, said:
v_TextureCoordinate = a_TextureCoordinate;
  v_ShadowCoordinate = shadowBiasMatrix * u_ShadowMatrix * u_WorldMatrix * a_Position;
This should all be moved to the CPU, I’ve never heard of the shadow coordinate being set in the vertex shader (I would have to spend more time than I am willing to think about whether or not it is possible), below this you have magic numbers again—let’s just stop there.

 

Isn't the purpose of varying variables to let OpenGL generate fragment-adjusted values across faces? It certainly seems to work, but I may just be doing a whole lot of overcalulation if my assumption is wrong.

 

Thank you very much for taking the time to read this mess :)

Share this post


Link to post
Share on other sites
Since it is already back from the dead…

Isn't the purpose of varying variables to let OpenGL generate fragment-adjusted values across faces?

That’s not what I meant. These matrices (included in my previous quote) should all be concatenated on the CPU side and only the final matrix sent to the shader. There are only very rare cases where it is justified to combine matrices in a shader.


L. Spiro

Share this post


Link to post
Share on other sites
Sign in to follow this  

  • Advertisement
×

Important Information

By using GameDev.net, you agree to our community Guidelines, Terms of Use, and Privacy Policy.

We are the game development community.

Whether you are an indie, hobbyist, AAA developer, or just trying to learn, GameDev.net is the place for you to learn, share, and connect with the games industry. Learn more About Us or sign up!

Sign me up!