|
Title: Optimal ideas Post by: fromhell on June 24, 2011, 06:31:54 PM brainfart time
The Pandora was recently released and shipped, and Quake3 is super crappy on it. That has got me thinking, "HOW MAKE OA FASTER!" - Use less and less textures per model (merging kyonshi to one TGA instead of like 5 would speed up rendering quite a bit in theory for example) - Fake texture atlas for marks achieved by tcMod scale (up to 4 marks on one texture file saving some texture switching) - Throw in ADPCM wav loading. There IS ACTUALLY DECODING code in the q3 source, GPLED, but it's completely unused. When it was implemented in 1999 it was seen as some cheap MP3 alternative (several 90s games used ADPCM for music for example. Even Command & Conquer (all, up to Red Alert 2) had some form of ADPCM). The advantage? Loading times, plus you're not going to notice 16-bit clarity through a handheld anyhow. You could say "WHY NOT OGG!1!!" well, look. Ogg's good for music but terrible for memory and loading when it comes to decoding to raw WAVs. - MDR, IQM or some sort of skeletal model format that works that I can export to with ease. This saves memory and doesn't take much more CPU than MD3 does. Of course for OA3 i'll always try to keep optimising in mind the first time around. Title: Re: Optimal ideas Post by: Peter Silie on June 26, 2011, 03:35:25 AM +1 ;)
Title: Re: Optimal ideas Post by: fromhell on July 20, 2011, 05:44:01 AM I wonder if chaining two tcMod Scale commands upon another will have CPU impact from coordinate generation
like i'd have some sort of 8 picture atlas in one texutre, where i'd select just by scaling i'll have to test this after I experiment with benchmarks of an atlas-based decal shaders vs the unique texture decals before I go any further with the atlas crazy ideas |