New Raytracing Patch II

[Stephan Reiter]

I just finished a new version that produces better bytecode. The code is
also better to read because I abstracted away most of the emit-calls into
classes à la Float, Float2, etc.

I also gave SIMD-code generation a shot, but it resulted in no speedup at
all and made the code difficult to be followed. So there's only the scalar
shading pipeline included ...

Patch is available from here:
http://www.file-upload.net/download-760882/rtpatch1270.zip.html

Comments and test reports are appreciated! :-)
Stephan

2008/3/30, Stephan Reiter <stephan.reiter at gmail.com>:
>
>  So, I have been working the past few weeks to get rid of some
> shortcomings of the older raytracing patch and present to you a new patch
> with the following features:
>
> Rapdio 0.7:
> - Restructured code, respects C++ RAII idiom and separates scene
> management from the acceleration structure through templates (next version
> will feature BIH and Grid as replacements for the BVH, BIH should yield
> better results in the ioquake3 environment).
> - Rewritten kd-tree builder, does not depend on openmp to utilize all
> cores (although openmp allows speeding up some parts of the process)
>
> rt-ioquake3:
> - Runtime code generation for shaders using the LLVM: this is blazingly
> fast and does not require GCC to be installed anymore!
> - SIMD code is not generated yet, expect lower performance due to use of
> scalar code only.
> - Support for reflections in Elite Force has been removed. We should think
> about adding new keywords to the Q3 shader language for raytraced effects.
>
> Some notes regarding compilation:
> - Code is based on revision 1270.
> - Rapido is not built with the ioquake3 makefile - build it manually in
> code/rapido
> - Get the latest version of the LLVM from their Subversion repository, svn
> co http://llvm.org/svn/llvm-project/llvm/trunk llvm
> - Expect a crash under Mac OS X: I know there is a bug in the kd-tree
> builder, but I haven't been able to locate until now. Maybe someone else is
> more lucky and could help, which I would appreciate A LOT!
> - Lightmaps are not sampled yet.
>
> Stephan
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://icculus.org/pipermail/quake3/attachments/20080401/d5067d2d/attachment.htm>