An order-independent transparency demo framework, including optimizations and benchmark support. With published strategies for fast sorting, this demo shows performance improvements over the basic per-pixel linked lists method of up to a factor of 5 or more.
NOTE: I haven't had time to prettify everything (I won't argue with people calling the code a mess). There are makefiles for gcc/Fedora 20 and solution/project files tested with vs2013/Windows 7.
A little description of OIT and this code is here: http://www.heuristic42.com/5/opengl/oit/
This code, and the LFB depndency, is the source code for Efficient Layered Fragment Buffer Techniques (well, updated source), Backwards Memory Allocation and Improved OIT and Fast Sorting for Exact OIT of Complex Scenes found here.
This project requires the following two to be in the same directory. I.e. /path/oit/, /path/lfb/ and /path/pyarlib.
Methods in the layered fragment buffer (LFB) repository are also called Per-Pixel Linked Lists (PPLLs), the A-Buffer (at a stretch IMO), the Dynamic Fragment Buffer (DFB). The S "sparsity", L "multi-layered framebuffer condensation", D "deque" buffers are also quite closely related. See http://en.wikipedia.org/wiki/Order-independent_transparency.
Models/meshes are not in the repository. Most can be found here: http://goanna.cs.rmit.edu.au/~pknowles/models.html.
The scenes are stored in xml files, and model search paths are specified in config.cfg.
Currently obj, ctm and 3ds are the main supported model formats.
- ` - toggle HUD
- space - toggle camera lock to origin
- w/a/s/d - fly
- mouse drag - left/middle/right to rotate/pan/zoom
- 1-6 - load scene
- b - execute benchmark (must provide test xml file as argument)
- h - find maximum depth complexity
- z - rotate 360 degrees, recording times
- q - enable scene edit mode (very simple maya-like e/r/t move/rotate/scale)