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

[duby229]

Now even really short posts are being moderated. I give up -- this just isn't worth the aggravation.

Nothing I can do about the forum. Although I am right, I apologize for losing my temper. It's just one misstep after the next with AMD. and this one hits close to home because I bought a GCN card and recommended multiple because I knew it would get launch day support. Or at least I thought I knew.

Sorry. I will be more polite from now on. I hope you can at least try to understand. The drivers great, but OpenGL performance isn't. I have put a lot of hope into Vulkan. That feeling probably rubs you the wrong way. Again, sorry.

[bridgman]

Your "statements to the contrary" is exactly what I'm telling you is the problem. Just wait and see the public outlash when Vulkan gets released with no GCN 1.0 support and "unsupported" GCN 1.1 support. It is going to backfire on you.

EDIT: I didn't guess anything at all. I made the suggestion that you guys need to get all GCN generations supported in amdgpu. That is not a guess. An out of tree solution is not acceptable. You need to work with the kernel team and figure out a solution. If it's too late for Vulkan's launch day, then that is squarely AMD's fault. You must have known for a long time now.

EDIT: Out of tree branches are fine for development and testing, but not for supported hardware. Get it upstream.

Huh ? The Catalyst replacement (hybrid driver) was always going to be out of tree. We have said that multiple times.

Is there any chance you are confusing "upstream" with "enabled by default" ? We're pushing upstream very aggressively, but won't be able to enable by default until appropriate userspace has made its way into distro releases and onto user systems.

[bridgman]

Now even really short posts are being moderated. I give up -- this just isn't worth the aggravation.

EDIT - think I see a pattern... it's posts that cross a page break which get moderated. Every time I see a yellow pane with my post in it there's also a "this post is on the next page click here" message above it.

[bridgman]

What the heck do you mean by "unsupported" GCN 1.1 support ? We're talking about the Catalyst replacement, out-of-tree drivers with closed source userspace. THE UPSTREAM DEFAULTS HAVE ABSOLUTELY NOTHING TO DO WITH WHAT WE DEVELOP AND RELEASE OUT OF TREE !!!

[duby229]

If you are claiming I said these things, please provide links. I have asked multiple times now.



If you don't know, and you're not psychic, why do you post statements claiming to know our plans and criticizing us for them ?

If you're just saying that in the absence of public statements people are going to make guesses that's fair, but (a) none of the vendors are saying much until the NDA lifts other than discussions with developers also under NDA, and (b) it doesn't explain why you keep repeating those guesses even after we have made public statements to the contrary.

Your "statements to the contrary" is exactly what I'm telling you is the problem. Just wait and see the public outlash when Vulkan gets released with no GCN 1.0 support and "unsupported" GCN 1.1 support. It is going to backfire on you.

EDIT: I didn't guess anything at all. I made the suggestion that you guys need to get all GCN generations supported in amdgpu. That is not a guess. An out of tree solution is not acceptable. You need to work with the kernel team and figure out a solution. If it's too late for Vulkan's launch day, then that is squarely AMD's fault. You must have known for a long time now.

EDIT: Out of tree branches are fine for development and testing, but not for supported hardware. Get it upstream.

[bridgman]

What did I make up now? Apparently everything you say is something I made up. Pretty neat trick.

If you are claiming I said these things, please provide links. I have asked multiple times now.

Since it's plainly obvious that you have hostility towards phoronix, why doesn't AMD release press news at it's own site? But instead you choose to wait for an obviously hostile website to release obviously hostile news. You know it's going to happen. And yet AMD still maintains a policy of keeping it's mouth shut.

When was the planning started, who's part of it, what is the plan, etc, etc.... THESE ARE THINGS NOBODY KNOWS YET, EXCEPT YOU. We can't all be psychic.

If you don't know, and you're not psychic, why do you post statements claiming to know our plans and criticizing us for them ?

If you're just saying that in the absence of public statements people are going to make guesses that's fair, but (a) none of the vendors are saying much until the NDA lifts other than discussions with developers also under NDA, and (b) it doesn't explain why you keep repeating those guesses even after we have made public statements to the contrary.

[artivision]

Moderated again. The only part I didn't cover in previous post is :



This would be another case of you making stuff up then getting mad at us because you don't like what you made up. The planning was *started* long ago, it just didn't happen to be *finished* (for SI) by the time Chris & Graham talke to Michael.

Mr. Bridgman, tell as first when we can have Vulkan even for one GPU, and then you can tell as which one.

[duby229]

Moderated again. The only part I didn't cover in previous post is :



This would be another case of you making stuff up then getting mad at us because you don't like what you made up. The planning was *started* long ago, it just didn't happen to be *finished* (for SI) by the time Chris & Graham talke to Michael.

What did I make up now? Apparently everything you say is something I made up. Pretty neat trick.

Since it's plainly obvious that you have hostility towards phoronix, why doesn't AMD release press news at it's own site? But instead you choose to wait for an obviously hostile website to release obviously hostile news. You know it's going to happen. And yet AMD still maintains a policy of keeping it's mouth shut.

When was the planning started, who's part of it, what is the plan, etc, etc.... THESE ARE THINGS NOBODY KNOWS YET, EXCEPT YOU. We can't all be psychic.

It's damn surprising that I'm having to tell you this.

[bridgman]

Moderated again. The only part I didn't cover in previous post is :

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.

This would be another case of you making stuff up then getting mad at us because you don't like what you made up. The planning was *started* long ago, it just didn't happen to be *finished* (for SI) by the time Chris & Graham talke to Michael.

[bridgman]

You asked exactly what I thought you made up, and I answered (#1, #4 and #6). What answer were you expecting ?

A lot of your beliefs seem to be based on a statement you think twriter made, can you please provide a link ?

The rest of your post isn't making much sense. There is no "policy of one driver in kernel per hardware device", however there is (a) a policy of only one *default* driver per hardware device (otherwise system behaviour would depend on quirks of the init subsystem) and (b) a policy that new kernels need to be able to replace old kernels without breaking userspace.

One consequence of those policies is that you can't quickly change default drivers unless they have identical user-kernel interfaces, and since one of the reasons for starting a new driver was to change the user-kernel interface that does explain the current state of upstream defaults. I don't know how you twist that into "hostility" though -- I have never disagreed with the policies and have explained a few times why they are important.

That said, the policies do not affect what we can do with out of tree drivers (and we have already said multiple times that initial Vulkan support will be via out of tree drivers) and do not affect what we can do upstream in the future although they do make changing driver defaults more complicated and more time consuming.

The plan is what we have said multiple times. VI and CI on amdgpu, initially via out-of-tree driver builds, SI TBD but using libdrm-amdgpu interfaces.