Dynamic Ray Stream Traversal implemented in Embree 2.0.

Paper: http://fileadmin.cs.lth.se/graphics/research/papers/2014/drst

Software and hardware requirements

Requires Microsoft Visual Studio 2013, Intel C++ Compiler XE 14.0 and Windows 7 64-bit.

Requires a CPU with AVX2 support and has only been tested on a mobile Haswell chip with 128 MB L4 cache. It is likely that the optimal ray stream size is different on a processor with a different cache hierarchy.

Getting started

You can export the code using the following svn-command:

svn export http://drst.googlecode.com/svn/trunk/embree-2.0

The project requires that Intel C++ Compiler XE 14.0 is installed into C:\Program Files (x86)\Intel\Composer XE 2013 SP1. If your path differs you need to change the VC++ Directories in each project.

Set startup-project to "renderer" and working directory to $(SolutionDir).

For best results, run Embree without the Visual Studio debugger, 16 spp and bvh4.triangle8 acceleration structure. Example command-line arguments:

-c ../crown/crown.ecs -spp 16 -accel bvh4.triangle8

where crown.ecs is the scene to be rendered (assumes that the crown has been stored in that location, look for it at the Embree project's home page).

Switching between single-ray and dynamic ray stream traversal

You can switch single-ray traversal on and off by changing the bool-variable streamTraversal in integratorrenderer.cpp in device_singleray.