Parallel Path Tracer Computer Systems Lab Presentation Stuart
Parallel Path Tracer Computer Systems Lab Presentation Stuart Maier
Abstract Computer generation of highly realistic images has been a difficult problem. Although there algorithms that can generate images that look essentially real, they take large amounts of time to render. This project explores ways of distributing that onto multiple computers, in order to speed up the process.
Graphics Background Many different algorithms Most unrealistic Real-life based algorithms developed Ray tracing Path tracing Spectral rendering
Ray Tracing Image from Wikipedia. Rays shoot through screen Rays seek light sources Phong shading
Path Tracing Rays don't seek light Rays can: Find light Be absorbed Be reflected Be transmitted Probabilistic
Spectral Rendering Complete color space RGB part illuminated Image from Wikipedia. Many different colors can create the same RGB value
Parallel Background BOINC Multiple independent clients No interclient communication Clients barely talk with server Splitting data Pixels are divided up among clients
Design Written in C with MPI and Open. GL libraries Currently implements a parallel path tracer with Cornell Box colors Six components Main Renderer Object Spectrum Vector XML
Main Runs the entire show Talks with MPI and Open. GL Starts up the renderer Passes off to the Renderer component
Renderer Holds information about the scene Renders the scene Performs the high-level ray calculations
Object Information about objects Calculates ray-object intersections
Spectrum Contains the Cornell Box colors Can retrieve these colors Converts spectral colors to RGB ones for use in the renderer
Vector Generic library Creates 3 -D vectors and processes them Used for spatial and color calculations
XML First component called after Main Reads in the scene file Passes off information to Renderer
Results
Analysis Path Tracer works in parallel Many basic parts work More realistic Global Illumination Cornell Box colors Transition to spectral rendering soon
- Slides: 16