A New VIA OpenChrome Gallium3D Driver Is Under Development

[curaga]

Reply to exhibit B: There was a working VIA 3d driver since 2003, but nobody cared enough, so it was thrown out of mesa 3 years ago.

It didn't work in 2008 when I tried to use it. So from my POV it never reached "working" status even for limited GL 1.2, not even talking about the chips' full abilities.

Surely you can do better than that, especially to the man who pioneered modesetting, and who did so on VIA hardware, and who REs ARM GPUs these days.

Oh, well aware who you are It could be argued that a 2d driver does not need to go the hard way like you did, but using the bios tables is acceptable (not sure if Via had such).

[libv]

It didn't work in 2008 when I tried to use it. So from my POV it never reached "working" status even for limited GL 1.2, not even talking about the chips' full abilities.



Oh, well aware who you are It could be argued that a 2d driver does not need to go the hard way like you did, but using the bios tables is acceptable (not sure if Via had such).

Of course. VIA had register dumps taken straight out of the BIOS in their code. This is why and where structured display driver development started. Openchrome forked away in the middle of me doing that work though, as i would not accept adding a VBE based codepath back in. This was two steps back from what i was working towards. As far as i know, the multiple codepaths plague the openchrome driver still today. It will be a decade since that fork soon.

As for PLLs, use the radeon driver for evidence that it is not that easy, and doing it properly is a lot of work (which only one man so far has done).

Doing modesetting right is _very_ hard. The combinations are endless, the different blocks, and there are usually many, of which all are usually different, rarely work together under all conditions. The amount of hardware needed to support all configurations out there is also vast. And it's not as if you can emulate one of the steps in there.

In comparison, anything 3D is pretty well contained and very overseeable. You always take bits from one piece of memory, have some combination of hardware in between, and throw the result to another bit of memory, so everything can be nicely quantified and measured. With the right _software_ infrastructure, you can test the hell out of it, and you only need like 1-2 devices per chip generation, not every board with a different display layout. Everything has actually become a lot easier with shaders, as you can always throw in a slightly different shader program, and verify that it is not the commandstream that's messed up.

Everything is easy in 3d land, at least much easier than in display land.

Sadly, everyone just confuses the blinginess with difficulty.

[lepetit]

the documentation of via is good or not enougt ?

http://www.freedesktop.org/wiki/Openchrome/Development/