2009-04-15 21:41:47

AMD uvolnilo specifikace - na vlastní bulvy

Přímé svědectví od mého spolupracovníka, který byl u toho když se uvolňovali specifikace ke grafickým kartám. Původně jsem to chtěl přeložit, ale vy jste inteligenti, tak si to vychutnejte v originále. Přetisknuto se svolením Adama.

Adam Jackson wrote:

I know, I was there.

So the story goes something like this.

I was at the X Developers' Summit in Cambridge this week. On Monday a couple of guys from AMD get up and talk about what their plans are. They're literally everything we asked for: active endorsement of open source drivers, non-NDA access to documentation, engineering resources within AMD working on the open drivers, and a support backchannel for the X developers so we can ask questions about the things the docs don't cover and we can't manage to figure out. They've also been working with Novell to develop a reference driver, which should be available within the next week or so.

None of which came as news to me. We've been working with AMD on this for the better part of a year now. I've been in fairly constant communication with our partner management about what our requirements are, what we'd like to see in an ideal universe, all that good stuff. Mad props to Russell Doty for doing the communication work on this.

About two months ago on a conference call, AMD committed to doing the open source thing and doing it right, which was probably one of the best days of the year for me.

Wednesday rolls around at XDS, and the SuSE guys get up to demo their progress with the driver. Still a few rough edges to file down but doing pretty good for only six weeks' worth of work. As the talk is winding down Egbert says "... and the docs will be available publically really soon". At which point John Bridgman of ATI says "Approved for public release about 45 seconds ago. Lemme burn a CD."

Said CD gets handed to Dave Airlie, who promptly uploads them. Within the first five hours, we'd pushed about 71000 copies - not a huge number until you realize that's nearly a terabyte. And those docs are just output setup for _two_ ASICs, nothing about 2D or 3D accel at all yet. These chips are phenomenally complicated and every single document requires explicit signoff to release, so we're in for a serious deluge of data.

The avivo driver hackers in the room start digging through the docs to convert register names of the form AVIVO_DC_TMDS1_MYSTERY3 into something more meaningful. Someone commits, Phoronix (an enthusiast news site and forum) posts a story saying "AMD docs already help AVIVO driver effort", and include a link in the story directly to the commit in gitweb. Which just happens to be the same machine as serves the docs.

Suddenly we're not pushing data anymore.

Lessons learned:

  1. AMD is absolutely serious about this.
  2. While git is a phenomenally scalable revision control system, gitweb scales only slightly better than a Commodore 64.
  3. Despite having Gig-E backbone-attached hosting in Boston (x.org) and Portland (freedesktop.org), we're probably going to need to serve future documentation drops with bittorrent.

Posted by Miroslav Suchý | Permanent link
comments powered by Disqus