Suggestions for performance upgrades to the Makaqu codebase

[JasonX]

[leileilol]

[frag.machine]

[mankrip]

[mh]

Well how many people actually run a 486 as their regular machine in 2010? OK, I know if you still have one and you use it for some retro-kick fetish ( ) you'll want good performance on it, but seriously...

Optimizing for 486 performance is the wrong thing to be doing right now. Get it running reasonably well on a machine that managed to outlive the Brontosaurus first, then worry about 486s if you think there's return on investment in it for you.

With the software renderer I've found that a major bottleneck lies in writing pixels to the screen buffer. I don't know if Makaqu has done it, but getting rid of the SciTech MGL, writing a native DirectDraw (depending on your OS) driver, and unrolling some of the loops that write to screen can give major benefits.

[JasonX]

I'm not really targeting old PC's, guys, i'm planning to get a clean and fast codebase so i can port it to interesting hardware, such as the GP2X. I don't want to mess with GL support for those ports, my idea is to have a codebase as unified as possible, with mostly IFDEFs for different platforms.

[mankrip]

[leileilol]

[frag.machine]

[mh]

[JasonX]

[Baker]

Well ...

MH's software interpolation code is blindingly fast. And easy to implement. Uses no asm. Totally portable. IF you are getting ok performance with WinQuake, you would think that you would also get ok performance with MH's software interpolation.




Makaqu has a ton of awesome stuff like skybox support, some great QC extensions like scale, but if the only thing that you are really after is interpolation ...

[p.s. qbism added decals and some other stuff to his modified Makaqu engine. Those could be factors as well. I seemed to get very satisfactory performance with qbism's engine in Flash.]

[mankrip]

[JasonX]

[mankrip]