Texture Sampling

From OpenGL Wiki
Revision as of 04:53, 10 August 2009 by V-man (talk | contribs)
Jump to: navigation, search

After Compiling Your Shader

In order to setup texture sampling, after you have compiled your shader, you need to get the location for the samplers.
For getting a shader's uniform's location, the shader doesn't need to be bound when you call glGetUniformLocation.
Let's assume you have a sampler called MyDiffuseTexture, MyEnvironmentMap, MyGlossMap in your FS

 uniform sampler2D MyDiffuseTexture;
 uniform samplerCube MyEnvironmentMap;
 uniform sampler2D MyGlossMap;

In your code, get the locations

 the_location0 = glGetUniformLocation(ProgramObject, "MyDiffuseTexture");
 the_location1 = glGetUniformLocation(ProgramObject, "MyEnvironmentMap");
 the_location2 = glGetUniformLocation(ProgramObject, "MyGlossMap");

Keep in mind that "locations" never change. In other words, you only need to call glGetUniformLocation once.
By default, each of those uniform values is 0. If you use the shader as is, an error will be raised. You would need to call glGetError(). The state of the shader is also consider invalid. This is how you can check it :

 int isValid;
 glValidateProgram(ProgramObject);
 glGetProgramiv(ProgramObject, GL_VALIDATE_STATUS, &isValid);

So, to set up those uniforms, bind the shader and call glUniform1i since they are considered as integers

 glUseProgram(ProgramObject);
 //Bind to tex unit 0
 glUniform1i(the_location0, 0);
 //Bind to tex unit 1
 glUniform1i(the_location1, 1);
 //Bind to tex unit 2
 glUniform1i(the_location2, 2);

Notice how glUniform1i doesn't take ProgramObject as a parameter, unlike glGetUniformLocation. This means, you must bind the shader before calling glUniform1i. Binding is done with glUseProgram.
The other thing to keep in mind is that the uniforms belong to the shader. This means that if you bind another shader, then bind this shader, you don't need to call glUniform1i to set things up again. The values are remembered.

Do not call glUniform1i(the_location0, tex[0]). You can't bind a texture that way.
To bind your texture, use glBindTexture

 glActivateTexture(GL_TEXTURE0);
 glBindTexture(GL_TEXTURE_2D, tex[0]);
 glActivateTexture(GL_TEXTURE1);
 glBindTexture(GL_TEXTURE_CUBE_MAP, tex[1]); //Because of uniform samplerCube MyEnvironmentMap
 glActivateTexture(GL_TEXTURE2);
 glBindTexture(GL_TEXTURE_2D, tex[2]);

Also, we use glUniform1i(the_location0, 0), glUniform1i(the_location1, 1), glUniform1i(the_location2, 2) but nothing is forcing us to use those samplers. You can even try glUniform1i(the_location0, 0), glUniform1i(the_location1, 1), glUniform1i(the_location2, 14) but make sure you bind the texture to the correct sampler with glActivateTexture(GL_TEXTURE14) and glBindTexture(GL_TEXTURE_2D, tex[2]).

Just be careful with the sampler number. Call

 glGetIntegerv(GL_MAX_TEXTURE_IMAGE_UNITS, &MaxTextureImageUnits);

to know how many samplers your GPU supports in the fragment shader. A GPU like the Gf 8800 supports 32 samplers while previous generation GPUs support 16 samplers.

In The GLSL Shader

So what happens when you sample the texture?
When you sample it, you get a normalized value if the texture is some fixed point format. By fixed point, we mean integer formats. Integer formats are BYTE, UNSIGNED_BYTE, SHORT, UNSIGNED_SHORT, INTEGER, UNSIGNED_INTEGER.
These would be formats like GL_RGBA8, GL_LUMINANCE8, GL_ALPHA8, GL_LUMINANCE16 and the many other formats listed in the GL spec.

 vec4 texel0 = texture2D(MyDiffuseTexture, TexCoord0);
 vec4 texel1 = textureCube(MyEnvironmentMap, TexCoord1);
 vec4 texel2 = texture2D(MyGlossMap, TexCoord2);

The values returned will be from 0.0 to 1.0 for each RGBA component. In other words, values are normalized, 0.0 to 1.0.
The GPU reads the integer format and converts them automatically.
If for some reason, you don't want it to do a conversion, see http://www.opengl.org/wiki/index.php/GL_EXT_texture_integer
That extension is available on Geforce 8 and up.

If the texture is a floating point format such as GL_RGBA32F, then since this is a 32 bit float, then no conversion takes place.
If the texel has a value like -489.5, then that's what you get.
If you are using GL_RGBA16F, then the 16 bit float gets upconverted (cast) to 32 bit float.
If you don't want it to convert, do the following

 half4 texel0 = texture2D(My16BitFloatTexture, TexCoord0);

That works on nVidia. You must make sure not to define any version number in your shader else nVidia drivers will consider half4 not defined.
half, half2, half3, half4 are not defined in the standard of GL 2.1

 #version 110

Don't put the above at the top of your shader if your shader doesn't follow the standard!