AMD Publishes Open-Source HD 7000, Trinity Code

[bridgman]

But let's be charitable and say that, through some miracle, AMD managed to cut the time between the kernel push and the mesa push by a whopping 60%. 222 * 0.40 = 88 days. 88 days from March 20, 2012 works out to be June 16, 2012. Even I'm optimistic enough to say that, indeed, an initial push of Gallium3d on June 16, 2012 would probably be enough to stabilize the driver in June-July and get it into Ubuntu 12.10 and turned on by default by some miracle.

But the question is, will AMD be able to cut their time to release by that much? Or will it sit behind closed doors for a period closer to the track record of 222 days for Evergreen? Every day that tick-tick-ticks by is slimming the chances that the G3D driver will land in distros being developed in 2012. And without the G3D bits, you have absolutely nothing.

Mesa push happened today (16 days ?). Is that soon enough ?

The code is in a public branch on Alex's freedesktop.org repo until Tom comes back from vacation next week and we work out where the llvm bits should go -- a big chunk of the 114 KSLOC is LLVM back-end, including both common-across-all-AMD-GPU bits and SI-specific bits.

[archibald]

Right, let's see if I can cover all the responses:

• Hurrah! Congratulations AMD!
• I'm really glad about this - I'm going to order a new card.
• The release->driver support time is coming down very quickly
• I already own a 7XXX - I can't believe it took this long!
• Does it have UVD support?
• Why don't you work on UVD?
• AMD should just open up Catalyst
• No open source EVERYTHING, right NOW means AMD don't even support linux - go to Nvidia!

Have I missed anything out? *grin*

Incidentally, I'm all of the first 3.

[curaga]

I guess the silent majority is like me, that is, "I'll consider a SI/Trinity when the FOSS support is at least as good as it is now for Evergreen".

[archibald]

Damn, I knew I'd forget something!

Being a BSD user I'm a bit (!) behind the times when it comes to graphics cards, so I'd have to buy a non-cayman Northern Islands card - these being the most recent cards supported by UMS and the highest-spec cards that are in my price range.

[bridgman]

• Does it have UVD support?
• Why don't you work on UVD?

Have I missed anything out? *grin*.

You forgot "Why are you wasting your time working on UVD ?"

[AlbertP]

I guess the silent majority is like me, that is, "I'll consider a SI/Trinity when the FOSS support is at least as good as it is now for Evergreen".

Could also be, "I won't consider an SI/Trinity until other forum users have ran it successfully with FOSS"

[allquixotic]

Mesa push happened today (16 days ?). Is that soon enough ?

The code is in a public branch on Alex's freedesktop.org repo until Tom comes back from vacation next week and we work out where the llvm bits should go -- a big chunk of the 114 KSLOC is LLVM back-end, including both common-across-all-AMD-GPU bits and SI-specific bits.

YES, that is most definitely soon enough. Thank you.

You know what they say about pessimists, right? No matter what happens you'll be pleasantly surprised.

I had a feeling that the SI code might land a tad sooner than 88 days, but I wouldn't have thought 16. Seriously though John, fantastic job, and congratulations to your team. I'm sure you haven't whipped up 114 KSLOC in 16 days and that it's been pretty far along in-house this entire time, but you didn't want to make any public promises in case there were some unforeseen delays (software is usually/always late ). So it's understandable that you didn't come right out and tell me back in March "hey dude, don't worry, it's coming out in like 2 weeks now". Because the moment you say something like that, Alex D. would stroll into your office and ask for another 3 months' time before publishing the code.

Note that I tried very hard not to make my post offensive or derogatory towards AMD. How'd I do? I think I did a decent job. I was just laying out the facts as I could see them at the time, coupled with a healthy dose of doubt.

But let me say that this turnaround time has significantly exceeded my expectations. Even if the driver is not at all usable today, the initial push is basically where the flood gates open -- we both know this. This is when the testing starts happening on all different chipsets; this is where power users, enthusiasts and developers start fixing application-specific bugs; this is where we start to polish the code. It's downright fantastic that this work is going to happen in April, rather than June or July, because that essentially means that our chances are very good for getting this work into distros a whole 6 months earlier. And for many people, that's about 15 to 25% of the total time that they'll be using this graphics card. That's VERY significant.

Thanks. Hat's off to you and your team.

[bridgman]

You know what they say about pessimists, right? No matter what happens you'll be pleasantly surprised.

Works for me

Yeah, SI was the first GPU generation where the developers were able to start well before product launch (3-6 months depending on how you define "start"). Unfortunately it was also the most significant HW change since r600. Alex had simple kernel driver tests working in December, first triangle via radeondemo was either late Jan or early Feb, and Michel was drawing untextured triangles with the Gallium3D code in March. Tom was working on LLVM but alternating between SI graphics and VLIW/SI compute (clover).

IIRC a normal push is maybe 20k-40k LOC for mesa plus ddx, this was over 110k for mesa alone. Probably the biggest patch I've ever seen

For the next GPU generation the driver work is starting even earlier relative to HW launch. Starting earlier means even more work for the developers (you have to test on simulators rather than real hardware, for example) so development time will probably be even longer, but relatively more of it will happen before the hardware launches.

[Thatguy]

"life isn't fair whaaaaaaaaaa"

Everytime you post, a Dictator in africa kills a family of infants, who have malaria and are starving to death. See what you did now ?

STFU please

[Thatguy]

they only do have malaria because they are to stupid to heal malaria with natural herbs for example with Artemisinin in the Artemisia vulgaris what kills effective Malaria source: http://de.wikipedia.org/wiki/Einj%C3...%C3%9F#Nutzung they are also to stupid to just wipe out the mosquito the source of malaria. 27 country's on earth are free of malaria today only because of a final wipe out with DDT http://de.wikipedia.org/wiki/Dichlor...ltrichlorethan
and you don't need DDT for a final strike against the mosquito you also can use natural Pyrethrum flowers like "Tanacetum cinerariifolium"
you can also use the Bacillus thuringiensis for a final strike http://en.wikipedia.org/wiki/Bacillus_thuringiensis
i think the best for a final solution against the mosquito's are the fungus : Metarhizium anisopliae http://de.wikipedia.org/wiki/Metarhizium_anisopliae

And they only do have a Dictator because they are to stupid to start a civil war against the Dictator and kill him.

what i dit is fighting stupidity like your writing about "Malaria" and "Dictators"

The only thing your prove, is that someone should take the keyboard from you and beat your ass to death with it.