AMD's Vulkan Driver Will Only Work With The AMDGPU Kernel Driver

[duby229]

Curses, moderated again. In the meantime:



We never talked about not supporting GCN 1.0 and/or 1.1, so #1 is either made up or based on misreading Michael's article.

#2 #3 and #5 are your opinions, I don't agree with them but I don't have a problem with you saying them. I have said multiple times that (c) is more likely but (b) might get us a broadly deployable solution 6 months faster, which is attractive.

#4 seems made up, both parts

#6 again seems made up, or at least based on misreading Michael's article.

And not a single answer.

Well, yeah, Of course I do expect AMD to support those cards in some way eventually, But I'm pretty sure the implication was launch day support. I may not have said launch day support, but it was obviously implied. You can mince words all day, but the value and meaning remains.

I guess in a way my interpretation of what you said is in fact my own opinion. However what you said is not my opinion, it is what you said. What you said leaves me no choice but to believe that you blame the Linux devs for their policy of one driver in kernel per hardware device. -AND- you have used that hostility as an excuse repeatedly in this thread.

So, what the the hell -IS- the plan then? Wait and see? How long? You must have known a long time ago, why didn't the plan get started when you realized it would be a problem? Why wait until a negative article gets released?

EDIT: I know, AMD always waits to fix problems until a negative article gets released.

[bridgman]

Curses, moderated again. In the meantime:

What exactly do you think I made up?

1. Do you refuse to recognize that lack of out of box support for GCN 1.0 and GCN 1.1 will in fact be a problem. And the lack of such -will- backfire on you?
2. That Option b: is so highly unlikely that it can be discounted immediately?
3. That the only option that makes sense for AMD to implement is option c:?
4. That AMD is actively hostile against Option c: and blames kernel policies for their hostilities?
5. It's a slippery slope that will leave you with no options. Right now at least you have c: and under the most unlikeliest terms, perhaps b:.

... and you left out:

6. Their doesn't seem to be any real plan to implement any solution. As of right now it's looking most likely that GCN owners will be sol.

We never talked about not supporting GCN 1.0 and/or 1.1, so #1 is either made up or based on misreading Michael's article.

#2 #3 and #5 are your opinions, I don't agree with them but I don't have a problem with you saying them. I have said multiple times that (c) is more likely but (b) might get us a broadly deployable solution 6 months faster, which is attractive.

#4 seems made up, both parts

#6 again seems made up, or at least based on misreading Michael's article.

Going off to shovel snow for a while before the temperature drops again and it all turns into ice.

[bridgman]

What exactly do you think I made up?

Do you refuse to recognize that lack of out of box support for GCN 1.0 and GCN 1.1 will in fact be a problem. And the lack of such -will- backfire on you?

Since we never said anything about not having out of box support for GCN 1.0 and 1.1 then either you made this up or misread Michael's article.

That Option b: is so highly unlikely that it can be discounted immediately?
That the only option that makes sense for AMD to implement is option c:?

These are just opinions. I don't agree with them but I'm OK with you making them up as long as you don't get upset with *us* as a result of *your* opinions.

That AMD is actively hostile against Option c: and blames kernel policies for their hostilities?

This one you are totally making up.

The only things we have said are (a) adding SI support to amdgpu will be somewhat messy because the SI HW handled by the kernel driver is more like NI than like CI, and (b) the CURRENT state of upstream defaults (which have ABSOLUTELY NOTHING to do with our future plans for supporting SI/CI in amdgpu) is a consequence of Linus's policy that new kernels need to be able to replace older kernels without breaking userspace dependencies.

You also forgot the most important one:

Their doesn't seem to be any real plan to implement any solution. As of right now it's looking most likely that GCN owners will be sol.

This seems to be completely made up as well (or, again, based on misreading Michael's article).

[duby229]

OK, I don't think I can help here. You are just making stuff up, posting it over and over again, and ignoring what we actually say.

What exactly do you think I made up?

Do you refuse to recognize that lack of out of box support for GCN 1.0 and GCN 1.1 will in fact be a problem. And the lack of such -will- backfire on you?
That Option b: is so highly unlikely that it can be discounted immediately?
That the only option that makes sense for AMD to implement is option c:?
That AMD is actively hostile against Option c: and blames kernel policies for their hostilities?

It's a slippery slope that will leave you with no options. Right now at least you have c: and under the most unlikeliest terms, perhaps b:.

EDIT: Also just the fact by itself that you made up option a:, which never was an option, and option b:, which is so unlikely that it almost certainly won't happen. And then you put the only real option at the back of the list and tried to minimize it. It must be sabotage.

[bridgman]

So where we are right now is, refusal to recognize that there is a problem, that the problem is people with GCN 1.0 and GCN 1.1 cards will -not- have out of box Vulkan support. Of the 2 realistic solutions being considered, one of them is highly unlikely, and the other (the one that actually makes sense) AMD is actively hostile against and puts the blame on the kernel developers for it. Their doesn't seem to be any real plan to implement any solution. As of right now it's looking most likely that GCN owners will be sol.

The more I look at responses in this thread, the more I'm convinced that it's a matter of sabotage. I just can't imagine someone dumb enough that they couldn't see right through it. It must be sabotage.

OK, I don't think I can help here. You are just making stuff up, posting it over and over again, and ignoring what we actually say.

[bridgman]

No, In fact I was replying to you. Do I really need to quote you? I can of course, but why?

Yes, please. A link to the post is probably more important than a quote.

As to "why", you are attributing something to me that I don't remember ever saying or thinking, so there's a good chance you are interpreting one of my comments in a way I did not intend.

[duby229]

I'm pretty sure that the only person talking about relying on community contributions is you.

No, In fact I was replying to you. Do I really need to quote you? I can of course, but why?

So where we are right now is, refusal to recognize that there is a problem, that the problem is people with GCN 1.0 and GCN 1.1 cards will -not- have out of box Vulkan support. Of the 2 realistic solutions being considered, one of them is highly unlikely, and the other (the one that actually makes sense) AMD is actively hostile against and puts the blame on the kernel developers for it. Their doesn't seem to be any real plan to implement any solution. As of right now it's looking most likely that GCN owners will be sol.

The more I look at responses in this thread, the more I'm convinced that it's a matter of sabotage. I just can't imagine someone dumb enough that they couldn't see right through it. It must be sabotage.

[bridgman]

Show me a history of AMD's drivers being developed successfully by anyone but AMD themselves? You remember before 2007 don't you?

There's actually been quite a lot of good work by non-AMD developers - you can't really trivialize the situation into "AMD doesn't do anything" or "AMD does everything". The truth is somewhere in between -- AMD is the largest contributor, but there is also a strong and active community.

That balance between vendor and community seems to be fairly unusual... perhaps that's why people have trouble recognizing it.

Do you really want to let you drivers fall into that position again? That is exactly what relying on community contributions will lead to.

I guess my advice would be to prove me wrong. I don't think the community outside of AMD devs will actually do it. But if you think it can be done and you can do it, then please, by all means go for it. At best it's wishful thinking and at worst it's sabotage. I don't know if someone is just dumb or evil.

With respect, I'm pretty sure that the only person talking about relying on community contributions is you.

[bridgman]

Auugh, now even my short posts are being moderated.

[bridgman]

Ok, fine. In that case, for an end user, b: is -exactly- the same as c:... Again, the bottom line is out of box Vulkan support.

Ding ding ding !!! Exactly.

I was reasonably certain that Mr, Writers posts at the minimum implied that AMD would not pursue option b: themselves, but would allow community contributions to be considered. Which is not at all what you just said.

The only post from Tim I could find was in response to a question about "whether AMD could implement pre-VI support in upstream amdgpu but have it disabled via Kconfig", where Tim said "that is exactly what we have today for CI". Nothing related to community contributions being required.

Any chance you could point me to the post ?