So I was reading "The Official OpenGL Guide" and in a section where they taught material lighting, they suddenly used the "flat" qualifier for an input variable in the fragment shader.
I google'd the matter and all I came up with was "flat shading" and "smooth shading" and the differences between them, which I cannot understand how it is related to a simple "MatIndex" variable.
here's the example code from the book:
struct MaterialProperties {
vec3 emission;
vec3 ambient;
vec3 diffuse;
vec3 specular;
float shininess;
};
// a set of materials to select between, per shader invocation
const int NumMaterials = 14;
uniform MaterialProperties Material[NumMaterials];
flat in int MatIndex; // input material index from vertex shader
What is this all about?
In the general case, there is not a 1:1 mapping between a vertex and a fragment. By default, the associated data per vertex is interpolated across the primitive to generate the corresponding associated data per fragment, which is what smooth shading does.
Using the flat keyword, no interpolation is done, so every fragment generated during the rasterization of that particular primitive will get the same data. Since primitives are usually defined by more than one vertex, this means that the data from only one vertex is used in that case. This is called the provoking vertex in OpenGL.
Also note that integer types are never interpolated. You must declare them as flat in any case.
In your specific example, the code means that each primitive can only have one material, which is defined by the material ID of the provoking vertex of each primitive.
It's part of how the attribute is interpolated for the fragment shader, the default is a perspective correct interpolation.
in the GLSL spec section 4.5
A variable qualified as flat will not be interpolated. Instead, it
will have the same value for every fragment within a triangle. This
value will come from a single provoking vertex, as described by the
OpenGL Graphics System Specification. A variable may be qualified as
flat can also be qualified as centroid or sample, which will mean the same thing as qualifying it only as flat
Integral attributes need to be flat.
You can find the table of which vertex is the provoking vertex in the openGL spec table 13.2 in section 13.4.
Related
If we pass a varying from any geometry stage (vertex, geometry or tess shader) to fragment shader, we always loose some information. Basically, we loose it in two ways:
By interpolation: smooth, noperspective or centroid - does not matter. If we passed 3 floats (one per vertex) in geometry stage, we will get only one mixed float in fragment stage.
By discarding. When doing flat interpolation, hardware discards all values except one from provoking vertex.
Why does OpenGL not allow functionality like this:
Vertex shader:
// nointerp is an interpolation qualifier I would like to have
// along with smooth or flat.
nointerp out float val;
main()
{
val = whatever;
}
Fragment shader:
nointerp in float val[3];
// val[0] might contain the value from provoking vertex,
// and the rest of val[] elements contain values from vertices in winding order.
main()
{
// some code
}
In GLSL 330 I need to make integer indexing tricks or divide by barycentric coordinates in fragment shader, if I want values from all vertices.
Is it hard to implement in hardware, or is it not widely requested by shader coders? Or am I not aware of it?
Is it hard to implement in hardware, or is it not widely requested by shader coders?
It is usually just not needed in the typical shading algorithms. So traditionally, there has been the automatic (more or less) interpolation for each fragment. It is probably not too hard to implement in current gen hardware, because at least modern desktop GPUs typically use "pull-model interpolation" (see Fabian Giesen's blog article) anyway, meaning the actual interpolation is done in the shader already, the fixed-function hw just provides the interpolation coefficients. But this is hidden from you by the driver.
Or am I not aware of it?
Well, in unextended GL, there is currently (GL 4.6) no such feature. However, there are two related GL extensions:
GL_AMD_shader_explicit_vertex_parameter
GL_NV_fragment_shader_barycentric
which basically provide the features you are asking for.
Most GLSL shaders are using a attribute for the color in the vertex shader, which will be forwarded as varying to the fragment shader. Like this:
attribute vec4 position;
attribute vec4 color;
uniform mat4 mvp;
varying vec4 destinationColor;
void main(){
destinationColor = color;
gl_Position = mvp * position;
};
Setting the color can be done with glVertexAtribPointer() to pass one color per vertex or with glVertexAttrib4fv() to pass a global color for all vertexes. I try to understand the difference to the predefined variable gl_Color in the vertex shader (if there is any difference at all). i.e.
attribute vec4 position;
uniform mat4 mvp;
varying vec4 destinationColor;
void main(){
destinationColor = gl_Color;
gl_Position = mvp * position;
};
and using glColorPointer() to pass one color per vertex or glColor4fv() to use a global color for all vertexes. To me the second shader looks better (= more efficient?), because it uses less attributes. But all tutorials & online resources are using the first approach - so I wonder if I missed anything or if there is no difference at all.
What is better practice when writing GLSL shaders?
To me the second shader looks better (= more efficient?), because it uses less attributes.
It does not use fewer attributes. It just uses fewer explicit attribute declarations. All of the work needed to get that color value to OpenGL is still there. It's still being done. The hardware is still fetching data from a buffer object or getting it from the glColor context value or whatever.
You just don't see it in your shader's text. But just because you don't see it doesn't mean that it happens for free.
User-defined attributes are preferred for the following reasons:
User-defined attributes make it clear how many resources your shaders are using. If you want to know how many attributes you need to provide to a shader, just look at the global declarations. But with predefined attributes, you can't do this; you have to scan through the entire vertex shader for any gl_* names that name a predefined attribute.
User-defined attributes can do more things. If you want to pass integer values as integers to the vertex shader, you must use a user-defined attribute. If you need to pass a double-precision float to the vertex shader, again, a predefined attribute cannot help you.
Predefined attributes were removed from core OpenGL contexts. OSX, for example, does not allow the compatibility profile. You can still use OpenGL 2.1, but if you want to use any OpenGL version 3.2 or greater on OSX, you cannot use removed functionality. And the built-in vertex attributes were removed in OpenGL 3.1.
Predefined attributes were never a part of OpenGL ES 2.0+. So if you want to write shaders that can work in OpenGL ES, you again cannot use them.
So basically, there's no reason to use them these days.
if I remember correctly gl_Color is deprecated remnant from the old style API without VAO/VBO using glBegin() ... glEnd(). If you go to core profile there is no gl_Color anymore ... so I assume you use old OpenGL version or compatibility profile.
If you try to use gl_Color in core profile (for example 4.00) you got:
0(35) : error C7616: global variable gl_Color is removed after version 140
Which means gl_Color was removed from GLSL 1.4
It is not entirely matter of performance but the change in graphic rendering SW architecture or hierarchy of the GL calls if you want.
The vertex shader is expected to output vertices positions in clip space:
Vertex shaders, as the name implies, operate on vertices.
Specifically, each invocation of a vertex shader operates on a single
vertex. These shaders must output, among any other user-defined
outputs, a clip-space position for that vertex. (source: Learning Modern 3D Graphics Programming, by Jason L. McKesson)
It has a built-in variable named gl_Position for that.
Similarly, the fragment shader is expected to output colors:
A fragment shader is used to compute the output color(s) of a
fragment. [...] After the fragment shader executes, the fragment
output color is written to the output image. (source: Learning
Modern 3D Graphics Programming, by Jason L. McKesson)
but there is no gl_Color built-in variable defined for that as stated here: opengl44-quick-reference-card.pdf
Why that (apparent) inconsistency in the OpenGL API?
That is because the OpenGL pipeline uses gl_Position for several tasks. The manual says: "The value written to gl_Position will be used by primitive assembly, clipping, culling and other fixed functionality operations, if present, that operate on primitives after vertex processing has occurred."
In contrast, the pipeline logic does not depend on the final pixel color.
The accepted answer does not adequately explain the real situation:
gl_Color was already used once-upon-a-time, but it was always defined as an input value.
In compatibility GLSL, gl_Color is the color vertex pointer in vertex shaders and it takes on the value of gl_FrontColor or gl_BackColor depending on which side of the polygon you are shading in a fragment shader.
However, none of this behavior exists in newer versions of GLSL. You must supply your own vertex attributes, your own varyings and you pick between colors using the value of gl_FrontFacing. I actually explained this in more detail in a different question related to OpenGL ES 2.0, but the basic principle is the same.
In fact, since gl_Color was already used as an input variable this is why the output of a fragment shader is called gl_FragColor instead. You cannot have a variable serve both as an input and an output in the same stage. The existence of an inout storage qualifier may lead you to believe otherwise, but that is for function parameters only.
I'm having some problem understanding one line in the most basic (flat) shader example while reading OpenGL SuperBible.
In chapter 6, Listing 6.4 and 6.5 it introduces the following two very basic shaders.
6.4 Vertex Shader:
// Flat Shader
// Vertex Shader
// Richard S. Wright Jr.
// OpenGL SuperBible
#version 130
// Transformation Matrix
uniform mat4 mvpMatrix;
// Incoming per vertex
in vec4 vVertex;
void main(void)
{
// This is pretty much it, transform the geometry
gl_Position = mvpMatrix * vVertex;
}
6.5 Fragment Shader:
// Flat Shader
// Fragment Shader
// Richard S. Wright Jr.
// OpenGL SuperBible
#version 130
// Make geometry solid
uniform vec4 vColorValue;
// Output fragment color
out vec4 vFragColor;
void main(void)
{
gl_FragColor = vColorValue;
}
My confusion is that it says vFragColor in the out declaration while saying gl_FragColor in main().
On the other hand, in code from the website, it has been corrected to 'vFragColor = vColorValue;' in the main loop.
What my question is that other then being a typo in the book, what is the rule for naming out values of shaders? Do they have to follow specific names?
On OpenGL.org I've found that gl_Position is required for the out of the vertex shader. Is there any such thing for the fragment shader? Or it is just that if there is only one output, then it will be the color in the buffer?
What happens when there is more then one out of a fragment shader? How does the GLSL compiler know which one to use in the buffer?
As stated in the GLSL specification for version 1.3, the use of gl_FragColor in the fragment shader is deprecated. Instead, you should use a user defined output variable like the
vFragColor variable described in your fragment shader. As you said, it's a typo.
What is the rule for naming out values of shaders?
The variable name can be anything you like, unless it collides with any existing names.
What happens when there is more then one out of a fragment shader? How does the GLSL compiler know which one to use in the buffer?
When there is more than one out in the fragment shader, you should assign slots to the fragment shader outputs by calling BindFragDataLocation. You can then say which slots will render to which render target by calling DrawBuffers.
The specification states that if you have one output variable in the fragment shader defined, it will be assigned to index 0 and output 0. For more information, I recommend you take a look at it yourself.
gl_FragColor was the original output variable in early versions of GLSL. This was the color of the fragment that was to be drawn.
Your initial confusion is justified, as there's no reason to declare that out variable and then write to glFragColor.
In later versions it became customizable, such that you could give arbitrary names to your output variables. You can map these arbitrary outputs to specific buffers with the command glBindFragDataLocation.
I'm not 100% positive, but I believe if you don't call this function before linking, then your output variables will be randomly assigned to buffers. If you only have one output, then it should always be assigned to buffer 0.
I'm trying to write a shader using cg (for ogre3d). I can't seem to parse a working shader that I'd like to use as a starting point for my own code.
Here's the declaration for the shader:
void main
(
float2 iTexCoord0 : TEXCOORD0,
out float4 oColor : COLOR,
uniform sampler2D covMap1,
uniform sampler2D covMap2,
uniform sampler2D splat1,
uniform sampler2D splat2,
uniform sampler2D splat3,
uniform sampler2D splat4,
uniform sampler2D splat5,
uniform sampler2D splat6,
uniform float splatScaleX,
uniform float splatScaleZ
)
{...}
My questions:
iTexCoord0 is obviously an input parameter. Why is it not declared uniform?
(oColor is obviously an output parameter. No question)
covMap1 - splat6 are textures. Are these parameters or something loaded into the graphics card memory (like globals)? The ogre definition for the shader program doesn't list them as parameters.
Are splatScaleX and splatScaleZ also parameters? The ogre definition for the shader program also doesn't list these as parameters.
Does the order of declaration mean anything when sending values from an external program?
I'd like to pass in an array of floats (the height map). I assume that would be
uniform float splatScaleZ,
uniform float heightmap[1024]
)
{...}
If I don't pass one of the parameters will the shader just not be executed (and my object will be invisible because it has no texture)?
Is there a better way to debug these than just hit/miss and guess?
iTexCoord0 is obviously an input parameter. Why is it not declared uniform?
Uniforms are not the same things as input parameters. iTexCoord is a varying input, which is to say that for every vertex, it can have a unique value. This is set with commands like glVertexAttribPointer. Things like vertex coordinates, normals, texcoords, vertex colors, are examples of what you might use a varying input for.
Uniforms are the other hand are intended to be static for an entire draw call, or potentially for the entire frame or life of the program. They are set with glUniform* commands. A uniform might be something like the modelview matrix for an object, or the position of the sun for a lighting calculation. They don't change very often.
[edit] These specific commands I think actually work with GLSL, but the theory should be the same for CG. Lookup a cg specific tutorial to figure out the exact commands to set varyings and uniforms.
covMap1 - splat6 are textures. Are these parameters or something loaded into the graphics card memory (like globals)? The ogre definition for the shader program doesn't list them as parameters.
My CG is a little rusty, but if its the same as GLSL then the sampler2d is a uniform that takes an index that represent which sampler you want to sample from. When you do something like glActiveTexture(GL_TEXTURE3), glBindTexture(n), then you set the sampler uniform to "3" you can sample from that texture.
Does the order of declaration mean anything when sending values from an external program?
No, the variables are referred to in the external program by their string variable names.
If I don't pass one of the parameters will the shader just not be executed (and my object will be invisible because it has no texture)?
Unknown. They will likely have "some" initial value, though whether that makes any sense and will generate anything visible is hard to guess.
Is there a better way to debug these than just hit/miss and guess?
http://http.developer.nvidia.com/CgTutorial/cg_tutorial_appendix_b.html
See section B.3.8