[quake3] Re: quake3 MacOS hackery, continued...

[Ryan C. Gordon]

> It's your guys project, so if you guys want to dump SMP support,  
> that's your decision.  But Mac users have grown accustomed to the  
> support being their so if it's not their anymore, well, you know Mac  
> users...

I'm not sure I do know Mac users in the way you're insinuating, but no
matter.

For what it's worth, I can duplicate the results on the Mac...a timedemo
of demos/FOUR.DM_68, with no SMP, ran at 218fps on a dual G5/2.0GHz with
an NV40...with r_smp 1, it got 288. This is with the SDL code and some
Mac-specifics to make it SMP-friendly.

Granted, I'm way past the default value of com_maxfps here either way,
but that is a significant boost.

I was misunderstanding the code as something that renders in one thread
and does game logic in the other, but that's not the case; it looks like
it renders in both, so you're issuing GL commands on one CPU while the
GL is processing the other CPU's frame.

Frankly, I think it's a miracle that this works at all, and the fact
that it does is probably a testament to how much engineering Apple was
willing to throw at this specific problem at the time to improve benchmarks.

On linux, the SMP code is a HUGE performance _loss_...same timedemo,
dual Opteron 240 system with a Quadro FX 1100, yields 206fps with r_smp
1, and 248fps _without_ it.

So I propose we add a few lines to sdl_glimp.c to allow Mac OS X to use
SMP, and leave it disabled everywhere else. I think that should make
everyone happy.

--ryan.