LargeScale Polygon Rendering

1
Large-Scale Polygon Rendering
2
Solutions

• Decimation
• Visibility Culling
• Parallel Rendering
• Others

3
Surface Decimation
Use a smaller number of triangles to represent
the same surface -gt preserve the topology
Removing vertices Contracting edges Etc.
4
Visibility Culling
Do not display polygons that are not visible to
the eyes
1 million triangles
188,000 triangles
5
Parallel Rendering
6
3D Rendering Pipeline
Geometry Processing
Rasterization
7
Parallel Rendering
sequential
G R
display
8
Parallel Rendering (2)

• Geometry processing parallelized by assigning
  each processor a subset of primitives (polygons)
• Rasterization parallelized by assigning each
  processor a portion of pixel calculations

9
Parallel Rendering (3)
What is rendering anyway? Calculate the effect
of each primitive on each pixel During
transformation, a primitive can go anywhere
on the screen Rendering can be seen as sorting
primitives to the screen
10
Parallel Rendering (4)
Rendering as a sorting process
11
Sorting primitives
Where to sort? Sorting can pretty much happy
anywhere Where/When to sort will affect the
structure of the parallel rendering system
G R
G R
G R
G R
G R
G R
display
12
Sort-First Parallel Rendering
13
Sort-First Parallel Rendering
Distribute the primitives early in the rendering
pipeline

• Subdivide the screen
• Pre-transform the primitives
• Distribute the primitives
• Each processor render its own
• primitives
• No communication needed
• afterwards

P2
P1
P3
P4
P1 P2 P3
P4
14
Sort-Middle Parallel Rendering
15
Sort-Middle Parallel Rendering
Distribute the primitives in the middle of
pipeline
P1 P2 P3 P4
1. Arbitrary assignment 2. Geometry processing
3. Sorting
P1 P2 P3 P4
Rasterization
16
Sort-Last Parallel Rendering
17
Sort-Last Parallel Rendering
Distribute the primitives in the end of pipeline
P1 P2 P3 P4
1. Arbitrary assignment 2. Geometry processing
3. Rasterization
18
Sort-Last Parallel Rendering
P1 P2 P3 P4
Sort images with z Compositing/z-buffer
P1 P2 P3 P4
19
Uniprocessor Cost
20
Sort First Overhead
21
Sort first pros and cons

• Pros
• Low communication cost
• Processors implement the entire pipeline
• Cons
• Susceptible to load imbalance
• Need special handling for frame to frame coherence

22
Sort Middle Overhead
23
Sort middle pros and cons

• Pros
• General and straightforward
• Cons
• High communication cost if tessellation ratio is
  high
• Susceptible to load imbalance between
  rasterization

24
Sort Last Overhead
25
Sort last Pros and Cons

• Pros
• Independent rendering of a full pipeline up to
  pixel merging
• Less Prone to load imbalance
• Cons
• Pixel traffic can be extremely high, particularly
  for oversampling

26
Comparison
27
Comparisons

• Pre-transformation and bucketization overhead may
  not be significant
• SL-sparse has little processing overhead beyond
  pixel merging
• SL-full requires much more effort for merging
  pixels
• Sort first communication cost can be small if
  frame to frame coherence is utilized
• Simple and large primitives favor sort first and
  middle. Small and complex primitive favor sort
  last sparse

28
Parallel Volume Rendering

• Object space subdivision

29
Parallel Volume Rendering

• Image Compositing