Ray Tracing with Dynamic Spatial Decomposition using MPI - PowerPoint PPT Presentation

1 / 18
About This Presentation
Title:

Ray Tracing with Dynamic Spatial Decomposition using MPI

Description:

Bump-mapping to produce realistic surfaces. Exploit Parallelism (using MPI) ... Bump Mapping. Motion Blur. Parallel Implementation. Master - Worker Paradigm. Renderer ... – PowerPoint PPT presentation

Number of Views:31
Avg rating:3.0/5.0
Slides: 19
Provided by: ccGa
Category:

less

Transcript and Presenter's Notes

Title: Ray Tracing with Dynamic Spatial Decomposition using MPI


1
Ray Tracing with Dynamic Spatial Decomposition
using MPI
  • Rafal Zboinski

2
Ray Tracing
  • Basic Techniques
  • Phong Shading Model
  • Diffuse, Specular
  • Reflection
  • Refraction
  • Beers law Percentage of light transmitted T
    e-(edc)
  • (e absorbity, d distance, c
    concentration/density)
  • Advanced Ray Tracing Techniques
  • Shoot out multiple rays to produce more realistic
    effects
  • Soft Shadows using spherical area light sources
  • Motion Blur to produce realistic
    animation/movies
  • Texture Mapping
  • Bi-linear interpolation to smooth textures
  • Bump-mapping to produce realistic surfaces
  • Exploit Parallelism (using MPI)
  • Embarrassingly Parallel
  • Dynamically decompose scene into chunks of
    optimal size to render movies on lots of CPUs

3
Basic Ray Tracer
4
Textures
5
Soft Shadows
6
Bump Mapping
7
Motion Blur
8
Parallel Implementation
9
Master - Worker Paradigm
  • Renderer
  • Basic Renderer for a one CPU implementation (No
    MPI)
  • Handles both ray tracing and 2D output
  • RendererMaster
  • Only handles 2D and manages workers
  • Doesnt touch ray tracing
  • RendererWorker
  • Only handles ray tracing computation doesnt
    know about windows, X, etc.

Renderer
RendererMaster
RendererWorker
10
Algorithm Overview
Processing (hard to compute)
t 1
Current Frame
Ahead of the game
t 2
Overflow Frame
11
Problems
  • We dont have to wait on CPUs to finish but..
  • Very small grid sizes with lots of CPUs require
    many overflow buffers
  • approx buffers Processors/(Grids per frame)
  • Only used 2 buffers
  • Even worse when there is a complex data/slow
    nodes

Computationally Complex Nodes
Slow nodes
T 3
T 4
T 1
12
Problem Pictures
13
Grid Size and Effect on Performance
  • Very fine grid causes increased bandwidth
    requirements
  • Goal Adapt grid size to optimize performance
  • Single Processor Times
  • 8x8 Grid 11.8302 sec/frame
  • 16x16 Grid 14.693 sec/frame

14
8x8 Grid, 512x512 - Rohan
15
16x16 Grid, 512x512 - Rohan
16
Future Work
  • Use more overflow buffers to prevent overwriting
    but..
  • Attempt to load-balance per frame to prevent
    skipping ahead

17
Movies and Questions
18
Movies and Questions
Write a Comment
User Comments (0)
About PowerShow.com