Vertex Post-Processing

OpenGL Rendering Pipeline

Rendering Pipeline Overview

---

• Vertex Specification
  • Vertex Rendering
  • Primitive
• Vertex Processing
  • Vertex Shader
  • Tessellation
  • Geometry Shader
• Vertex Post-Processing
  • Transform Feedback
• Primitive Assembly
  • Face Culling
• Rasterization
• Fragment Shader
• Per-Sample Processing
  • Scissor Test
  • Stencil Test
  • Depth Test
  • Blending
  • Logical Operation
  • Write Mask

Vertex Post-Processing is the stage in the OpenGL Rendering Pipeline where the vertex outputs of the Vertex Processing undergo a variety of operations. Many of these are setup for Primitive Assembly and Rasterization stages.

Transform Feedback

Main article: Transform Feedback

Transform feedback is a way of recording values output from the Vertex Processing stage into Buffer Objects.

Clipping

Primitives generated by previous stages are collected and then clipped to the view volume. Each vertex has a clip-space position (the gl_Position output of the last Vertex Processing stage). The viewing volume for a vertex is defined by:

${\begin{aligned}-w_{c}&\leq x_{c}&\leq w_{c}\\-w_{c}&\leq y_{c}&\leq w_{c}\\-w_{c}&\leq z_{c}&\leq w_{c}\end{aligned}}$

This volume can be modified by depth clamping as well as the addition of user-defined clip-planes. The total volume that primitives are clipped to, including user-defined clip planes, is the clipping volume.

The way primitives are clipped to this clipping volume depends on the basic Primitive type:

Points

Points are not really "clipped". If a point is in any way outside of the clipping volume, then the primitive is discarded (ie: not rendered). Points can be bigger than one pixel, but the clipping remains; if the center of the point (the actual gl_Position value) is outside of the clipping range, it is discarded. Yes, this means that Point Sprites will disappear when the center moves off-screen.

Platform Issue (NVIDIA): These cards will not clip points "properly". That is, they will do what people generally want (only discard the point if it is fully outside the volume), rather than what the OpenGL specification requires. Be advised that other hardware does what OpenGL asks.

Lines

If the line is entirely outside of the volume, it is discarded. If the line is partially outside of the volume, then it is clipped; new vertex coordinates are computed for one or both vertices, as appropriate. The end-point of such a clipped vertex is on the boundary of the clipping volume.

Triangles

A triangle is clipped to the viewing volume by generating appropriate triangles who's vertices are on the boundary of the clipping volume. This may generate more than 1 triangle, as appropriate. If a triangle is entirely outside of the viewing volume, it is culled.

When primitives are clipped, new per-vertex outputs must be generated for them. These are generated via linear interpolation (in clip-space) of the output values. Flat-shaded outputs don't get this treatment.

Depth clamping

The clipping behavior against the Z position of a vertex (ie: $-w_{c}\leq z_{c}\leq w_{c}$) can be turned off by activating depth clamping. This is done with glEnable(GL_DEPTH_CLAMP). This will cause the clip-space Z to remain unclipped by the front and rear viewing volume.

Note: With perspective projections, you still get clipping with the sides of the viewing volume. Depth clamping turns a frustum into a pyramid, and pyramids still end at the tip. So objects that go behind the camera are still clipped. What is unclipped are objects between the projection near-plane and the camera.

The Z value computations will proceed as normal through the pipeline. After computing the window-space position, the resulting Z value will be clamped to the glDepthRange.

User-defined clipping

Uses

User-defined clipping can be used for many things. For example, in GUI elements, it is often desired to ensure that objects inside a GUI window cannot be drawn outside of that window. There are several tools one could use to ensure this, including changing employing Scissor Testing.

However, most of these solutions have one important downside: the parameters can only be changed between rendering calls. Which means, if you have a large batch of work that you want rendered, where parts of that batch are in different windows, you must do break it up into multiple rendering calls. This decreases performance.

User-defined clipping can help. These parameters can be per-vertex values, sampled from vertex arrays. The Vertex Shader, or other final processing stage, will be able to use them to set the gl_ClipDistance outputs.

The key is to set the clipping parameters based on the requested clipping region and the world-space position of the vertices.

There are some downsides to this approach.

First, it requires an additional per-vertex attribute, who's values will typically be shared among many vertices. And this attribute would likely have to be a full 2D clipping region: an XY position and width+height. Storing such positions would probably require 16-bits per position, as 8 bits would not provide enough resolution for most displays. So that's a total of 8 bytes added to each vertex.

It is possible to mitigate this by using a UBO to store an array of clipping regions, with each vertex simply specifying an integer index into that array.

Second, when user-defined clipping is used together with Multisampling, GUI elements may be drawn outside of the clipping region. User-defined clipping happens at the vertex level, not the fragment level. So it remains possible for multisampled Rasterization to generate Fragments outside of the pixel area assigned by the user-defined clipping region.

Perspective divide

The clip-space positions returned from the clipping stage are transformed into normalized device coordinates (NDC) via this equation:

${\begin{pmatrix}x_{ndc}\\y_{ndc}\\z_{ndc}\end{pmatrix}}={\begin{pmatrix}{\tfrac {x_{c}}{w_{c}}}\\{\tfrac {y_{c}}{w_{c}}}\\{\tfrac {z_{c}}{w_{c}}}\end{pmatrix}}$

Viewport transform

The viewport transform defines the transformation of vertex positions from NDC space to window space. These are the coordinates that are rasterized to the output image.

The viewport is defined by a number of viewport parameters. These parameters are set by these functions:

void glViewport(GLint x​, GLint y​, GLsizei width​, GLsizei height​);

void glDepthRange(GLdouble nearVal​, GLdouble farVal​);

void glDepthRangef(GLfloat nearVal​, GLfloat farVal​);

The second two functions set the same parameters, the near and far values of the depth range.

Given the viewport parameters, we compute the window-space coordinates via these equations:

${\begin{pmatrix}x_{w}\\y_{w}\\z_{w}\end{pmatrix}}={\begin{pmatrix}{\begin{aligned}{\tfrac {width}{2}}x_{ndc}&+x+{\tfrac {width}{2}}\\{\tfrac {height}{2}}y_{ndc}&+y+{\tfrac {height}{2}}\\{\tfrac {farVal-nearVal}{2}}z_{ndc}&+{\tfrac {farVal+nearVal}{2}}\end{aligned}}\end{pmatrix}}$

Where x​, y​, width​, height​, nearVal​, and farVal​ are the viewport parameters.

Viewport array

Viewport Arrays

Core in version		4.6
Core since version		4.1
Core ARB extension	ARB_viewport_array

Multiple viewports can be used in OpenGL. The specific viewport for a particular primitive can be set by the Geometry Shader. If the GS does not specify a viewport, then viewport number 0 is selected. The computation works as above, except where it says "the viewport parameters", it means "the viewport parameters for the primitive's viewport index".

There are sets of viewports, indexed on the half-open range [0, GL_MAX_VIEWPORTS). Each index has its own depth range and viewport coordinates. The previously defined functions will only set the value for viewport index 0.

To set the viewport parameters for a particular index, use this pair of functions:

void glViewportIndexedf(GLuint index​, GLfloat x​, GLfloat y​, GLfloat w​, GLfloat h​)

void glViewportIndexedfv(GLuint index​, const GLfloat *v​)

void glDepthRangeIndexed(GLuint index​, GLdouble nearVal​, GLdouble farVal​)

The index​ is the viewport index to set the parameters for. glViewportIndexedfv takes an array of 4 floats, in the same orders as the parameters for glViewportIndexedf.

Multiple viewport indices can be set with a single function, via these APIs:

void glViewportArrayv(GLuint first​, GLsizei count​, const GLfloat *v​)

void glDepthRangeArrayv(GLuint first​, GLsizei count​, const GLdouble *v​)

The first​ index is the first viewport index to set. count​ is the number of viewport indices to be set by the function. v​ is an array of viewport values, which contains count​ * 4 or 2 values, depending on the function being called. The values for a single viewport index are in the same order as the arguments in the regular function calls.