uis

I guess in Linux it would be so much easier. KMS for modswitching and DRI for 3d stuff.

Nyuzi GPGPU had demo of Quake

But why not just do branch delay? Or some sort of prepared branches like on ELBRUS where you can write to predicate register many instrucions ago and take conditional branch at same cost of non-conditional or maybe even prefetch intstructions into second copy of pipeline before stage where you do btanching(for example fetching if you do it on decode stage). Or some sort of explicit branch slots.

Anyway, I registered here just to reply to you. I kinda want to talk to you more because you do what is interesting for me too.