Fast Dynamic Lighting

[mh]

Continuing the derail...

Yes, draw the gun model as the first thing in your frame; on modern hardware you'll get early Z rejection on other scene items which gives a small but useful perf boost. On older hardware (where the Z test runs after the pixel shader) it should still help a small bit.

[taniwha]

Getting a little back on topic...

For various reasons (including chasing a bug for Spirit), I got nvidia's drivers working on my system again, and noticed QF is significantly (~2.5x) faster on nvidia than nouveau, and the GL (fix function) renderer is faster than the GLSL renderer on nv, but the other way on nouveau. Getting curious, I generated some graphs while running on nv (too much hassle to swap to nouveau just for timing).

[edit]Just to make it explicit: this is timedemo demo3.

Click on the images for full-size/lossless.

• GL renderer (~570fps)
  
• GLSL renderer (~515fps)
  
• GLSL renderer, no particles (~540fps)
  
• GLSL renderer, no dynamic lights (~860fps) [edit 2] vertical scale is 0.1ms/large tick, it seems.
  
• GLSL renderer, no particles or dynamic lights (~900fps)
  

Vertical scale is 1ms/large tick, horizontal scale is 100 frames/tick.
It seems I still have a lot of work to do on dlights (particles seem to be ok).

For the curious, I generated the graphs using QF's demo_timeframes feature (written by Rhamph years ago). It dumps the frame ms to timeframes.txt in the user's quake root. After that, timeframes.py in QF's tools/misc directory to generate timeframes.ppm. I then converted to png/jpg for the web.

I don't know why the thumbnails got random widths: the full size images are all the same width (different heights, though)