A Generic and Scalable Pipeline for GPU Tetrahedral Grid Rendering

1
A Generic and Scalable Pipelinefor GPU
Tetrahedral Grid Rendering

• Joachim Georgii, Rüdiger Westermann

2
Motivation

• Rendering tetrahedral grids?
• Frequently used in simulations
• Visualization of internal properties
• Interactive Rendering of
• Large data
• Deformable models
• Topologically changing models

3
Motivation

• Rendering tetrahedral grids?
• Frequently used in simulations
• Visualization of internal properties
• Interactive Rendering of
• Large data
• Deformable models
• Topologically changing models

Deformable models
Topological changes
4
Yet another approach?

• Interactive Rendering
• Previous approaches applicable?
• Cell Projection (Shirley 90)
• Powersort (Cignoni 95)
• MPVO (Williams 92)
• k-Buffer (Callahan 05)
• Raycasting (Weiler 03)
• Resampling (Weiler 01, Westermann 01)

Deformable models
Topological changes
? ?
? ?
? ?
? ?
? ?
5
Yet another resampler?

• Explore new graphics hardware features
• Minimize bus bandwidth
• Advantages
• Per-element calculations performed only once
• Shared vertex and attribute arrays
• No additional memory
• Raysampling in barycentric coordinates
• ? minimal arithmetic and memory access operations

6
Method Overview

• Element Assembly Primitive Construction
• Build a tetrahedral element on the GPU
• Fragment Stage
• Tetrahedra sampling

7
Shell Partitioning

• Tetrahedra partitioning and sampling

8
Element Assembly

• A single vertex is rendered per tetrahedron
• 4 references into vertex texture
• Get vertex coordinates (4 x texfetches)
• Transform to eye-coordinates

Vertex Texture
1 Vertex
9
Primitive Construction

• Traverse rays in barycentric coordinates
• needed anyway to interpolate attributes
• Determine local barycentric coordinates (LBC)
• Render triangle strip (backfaces are culled)
• 6 vertices with
• barycentric ray direction, length (interpolated
  by rasterizer)
• barycentric eye point, IDs

10
Fragment Stage
Eye
Screen
Shell
11
Fragment Stage cont.

• Discard samples outside the tetrahedron
• min bi lt 0, output 0
• Interpolate attributes (scalars, texcoords)
• Optimization
• MRTs
• Up to 16 samples per fragment/shell? Less access
  operations to scalars/texcoord texture
• ? Only (cheap) incremental update on bi
• ? Partitioning gets better

12
Direct volume rendering
13
Is there more?

• No need to only sample equidistantly along view
  rays
• We can also
• Sample at iso-values ? Iso-surface rendering
• Sample at front-/backfaces? Cell Projection

14
Iso-Surface Rendering

• Sample tetrahedron at
• Depth test
• instead of blending stage
• Gradients
• Lighting

15
Cell Projection

• Determine ray exit point
• Idea
• Point on face ? one barycentric coordinate 0
• Determine 4 candidate values zout
• Choose smallest value ? exit point
• Get attenuation from length of ray segment
• Get scalars on entry and exit point

16
Implementation

• Element Assembly ? vertex shader
• Fragment Stage ? fragment shader
• Primitive Construction ? geometry shader
• Whats a geometry shader?
• Specified in Direct3D 10
• Available next week
• Therefore 2 passes, render-to-vertex-buffer

17
Results

• Scalar
• Bluntfin
• (190k tets)
• 400 slices
• 9 fps
• 3D texture coordinates
• Visible male head
• (3800k tets)
• 512x512x302 texture
• 600 slices
• 1 fps

• DEMO

18
Conclusion

• We have proposed a rendering pipeline for
  tetrahedral grids that is
• Generic
• Direct volume rendering
• Iso-surface rendering
• Cell projection
• Scalable
• Large datasets
• Graphics hardware development
• Designed for upcoming graphics hardware
• Geometry shader

19
Future Work

• Verification on future Direct3D 10 GPUs
• Early ray termination
• ClearView

20
Thanks for your attention