AMD Releases Open-Source R600/700 3D Code (Phoronix)

Well what do you expect? You want AMD to risk losing Windows certification for its Windows drivers, thereby hamstringing the single largest (by a huge margin) target audience of its consumer graphics hardware?

Yeah, great idea. Then AMD's graphics business tanks and has to be shut down, and then AMD users end up stuck the same place Matrox users used to be. Remember the old Matrox cards, once the paradigm of Open Source graphics hardware for consumer enthusiasts? How years after it was no longer even remotely top-of-the-line Linux users were still buying up the G200, G400, G450, and so on at ridiculous markups? Until eventually even the Open Source support dried up because the cards were so ancient as to be unusable for modern desktop needs?

You want AMD to risk going there just so that you could get faster Linux driver releases?

"I highly doubt it—they had complete HW, IRQ, and register specs available, and I presume that HW and SW development were done in parallel."

No, actually, they didn't. Read up on what the actual "sanitization" process was -- much of it was taking the docs created by the HW team and fixing all of the great deal of errors, and making sure working usable code could be produced from those specs without needing the ability to call up the HW engineering team and ask questions the specs don't answer.

Additionally, the Windows drivers were developed years ago. The Linux drivers being developed today should make it much easier to support new hardware at or before release in the future. Windows got a head start, as the driver engineering team started those drivers years before the hardware hit the streets. Linux has to play catch up. Be mad at ATI for not supporting Linux 5 years ago if you want, but don't get upset because AMD's recent decision to support Linux can't magically produce many man-decades worth of work in a couple of months.

Yes, AMD could have thrown more resources at it. But let's be perfectly honest with ourselves -- a niche market is not worth a huge investment. It isn't. No matter how much we might want to believe Linux users matter more than other people, the fact is that we don't. That a desktop OS that gets maybe 1-2% of the desktop market share gets _ANY_ full-time engineers from AMD is amazing all on its own.

Going forward, AMD is letting open-source developers talk to the chip designers like the Windows driver others talk to them, and generally not bothering to try to write complete documents for public (or even internal or NDA) consumption.

To illustrate this, consider the AVR ATmega8. This is a reasonably straightforward microcontroller with a smallish number of built-in peripherals. The documentation is a 308-page PDF which has gone through 20 revisions after it was written, over the course of 6 years. The chip was intended from initial design time to be fully explained to external users. One of the things that's really great about this manufacturer, Atmel, and this line in particular, is that their documentation is accurate and informative; most companies' datasheets have inaccuracies (using unusual features doesn't actually work the way the datasheet says it does) and gaps (documentation for register A assumes that register B has the default value, and for register B assumes that register A has the default value, and nobody knows what happens if you change both). And this is for companies that provide datasheets in order that people will design devices that use the parts and then buy a lot of them. Writing extensive and reasonably accurate hardware specs for a graphics chipset is just not an efficient use of a huge amount of time.

1. For the last >year we have been releasing Catalyst Linux driver support simultaneously with Windows support. The long delays between initial launch and Linux were in 2007 and earlier.

2. Getting 6xx/7xx 3d information took about 10 months from start to finish. Most of that was spent learning to program the chip and distilling knowledge from >20M lines of source code and hundreds of engineers into something small and useful to the open source community. Before the 6xx/7xx 3d effort we were working on 2d/3d information for earlier GPU generations and on display/modesetting support for both older and newer GPUs (remember the project was launched in Sep 2007).

3. It took us about 16 months to largely catch up on 6 years of hardware development (the last round of open source community support ended around 2002). The delay depends on the generation; you could say it took us 72 months to get r300 information out, but again the project only started 15 months ago. The delay for open source 7xx info was obviously a lot less -- a few weeks from launch to initial display/modesetting support and a few months to release of 3d engine info.

But true, for newer cards one needs radeonhd driver, which still lacks many features.