Excuse me, but nouveau/nv50,nvc0 have been supporting MSAA for a year now (http://cgit.freedesktop.org/mesa/mes...2820b00a9ce667).
Michael, this is wrong:
It runs entirely on the GPU.MLAA in Mesa though is done on the CPU
Of course Mesa as a whole may be cpu-limited, but please don't confuse the two
- read the docs to pick a suitable opcode, if it exists
- check the right file on how the other opcodes around it are implemented, copy-paste a suitable function / add an argument to the handler
- bind it to that function in the tables for the hw
And the result:
I don't think it was just a simple oversight -- IIRC there was some nuance with earlier GL versions where the extension had to be supported but only 1x support was required (which meant that real MSAA support was *not* really needed at that point).
More likely a missed nuance in a 200-page specification...
Just thinking about it fills me with feelings and emotions that no man should be encumbered with in a professional or academic setting.
BTW, there's no support for MSAA pre-Gen 6 because only Sandy Bridge onwards have hardware for it.