RADV Sees Experimental Fragment Shader Interlock - Important For Emulators, D3D12

[user1]

The thing is, that outcome wasn't going to be known, and you can't look at things in hindsight to say "I told you so" because I'm sure even you didn't know with certainty that things would pan out this way.

Yes, you're right, I myself also had no clue what would come out of it. But I'd say the same logic could be applied in 1991 when Linus started developing the Linux kernel. By the same logic one could say that this is a wasted effort (even if Linus himself didn't expect it to become something serious) and that Linus should've waited for or contributed to the GNU Hurd kernel instead. As you can see, when something new looks pointless from first glance, it may drastically change in the future.

[schmidtbag]

Yes, you're right, I myself also had no clue what would come out of it. But I'd say the same logic could be applied in 1991 when Linus started developing the Linux kernel. By the same logic one could say that this is a wasted effort (even if Linus himself didn't expect it to become something serious) and that Linus should've waited for or contributed to the GNU Hurd kernel instead. As you can see, when something new looks pointless from first glance, it may drastically change in the future.

I get where you're coming from I don't find the two to be similar enough to be comparable. For one thing, it seems Hurd didn't exist at the time, or at least wasn't made public soon enough for Linus to know, let alone be able to contribute. So, then-obscure developer Linus made Freax as an unprofessional hobby, and his product fit a niche that didn't exist at the time. I don't entirely understand why Linux got so popular compared to something like Hurd; perhaps because it was less political and therefore something people were more willing to work with.

Now, compare to Dave Arlie, who in 2016 was already a renowned professional developer, and (as far as I'm aware) didn't create RADV as a hobby. I would be shocked if he didn't know amdvlk was being developed at the time. Considering his history and affiliation, I'd be a little surprised if AMD were to turn down contributions from him. So, it never really became understood to me why he made it.

Where RADV and Linux are comparable is the fact that RADV ended up being more popular for what I assume to be the same reason: it was politically easier to work with. RADV was much more open to 3rd party contributions, and it has served them well. Perhaps the licensing was why he created RADV.

Anyway, of course anything seemingly pointless can drastically change the future, but think of how much more different the future would have been if a different GPU got its first Vulkan driver instead. If amdvlk got patches from Valve, Red Hat, and people like Triang3l, it might have come out better than RADV is today, since it'd have been a collaborative effort rather than a competitive one, all while another GPU would have Vulkan support, further enhancing the Linux experience across a wider range of devices. I do acknowledge that is a very big "if", though.
So here's the thing (and I'm not trying to be antagonistic here): had amdvlk turned out to be better than RADV, what would you say about all this? That outcome was well within the realm of possibility (no offense to Arlie).

[user1]

Anyway, of course anything seemingly pointless can drastically change the future, but think of how much more different the future would have been if a different GPU got its first Vulkan driver instead. If amdvlk got patches from Valve, Red Hat, and people like Triang3l, it might have come out better than RADV is today, since it'd have been a collaborative effort rather than a competitive one, all while another GPU would have Vulkan support, further enhancing the Linux experience across a wider range of devices. I do acknowledge that is a very big "if", though.
So here's the thing (and I'm not trying to be antagonistic here): had amdvlk turned out to be better than RADV, what would you say about all this? That outcome was well within the realm of possibility (no offense to Arlie).

Thing is, even if in alternative reality radv didn't exist, I think AMDVLK would've definitely received more care for Linux gamer's needs than now (I think even Bridgman mentioned this at some point). But I still highly doubt it would've been as good as radv.
First, there would've been no ACO compiler and AMD is known to be highly committed to LLVM which especially on AMDVLK is painfully slow to compile shaders. (this is a fundamental issue with LLVM and one of the reasons for creating ACO).
Second, let's not forget that AMDVLK was originally proprietary software only, which means open sourcing it has some caveats. And the main caveat is that it's still mostly developed behind closed doors (probably because AMD still releases both the proprietary variant and open source variant of AMDVLK and the proprietary variant is much more widely used on WIndows) hence the reason it's much less open to 3rd party contributions (you said it yourself in your previous paragraph). If licensing issues don't stand in the way, the only one who can theoretically change AMDVLK's development model to more resemble Mesa's is only AMD itself (this is also highly questionable, again, because of the proprietary variant that shares the same codebase). Not Valve, not Red Hat and not Triang3l are able to change anything about AMDVLK's development model, hence the much more limited 3rd party contribution opportunities. David Airlie himself mentioned that the big difference between AMDVLK and radv is that the former is "open - released" (open source code snapshots are thrown over the fence as seen in AMDVLK github page), while the latter is "open - developed".

[RejectModernity]

More likely it could happen the other way - Russia will block access to something like Github or Gitlab because they aren't under their censorship (shooting themselves in the foot with that is expected, considering everything else Putin is doing). Many developers are leaving Russia while they can.

Nah, Russia bans only Soros-funded brainwashing media.

[shmerl]

Nah, Russia bans only Soros-funded brainwashing media.

They ban anything that's not Putler approved. Not yet on China level, but getting there, including asking China to help with censorship. Putin likes stone age, that's what he'll get technology wise.

[Quackdoc]

That's sadly my understanding, too.
Greetings,

Dieter

it's sad, but my next gpu certainly won't be AMD. when I look at the features that even skylake igpus have that AMD doesn't... I don't know what features will be missing that Nvidia will support. I will probably be going nvidia or intel, Intel has actually been a pretty nice experience so far on the A380, albiet ofc, bugs. very promising cards though

[RejectModernity]

They ban anything that's not Putler approved. Not yet on China level, but getting there, including asking China to help with censorship. Putin likes stone age, that's what he'll get technology wise.

Stone age doesn't produce supersonic rockets. You're delusional and should join Reddit battalion and help Ukraine to fight evil Putler.

[shmerl]

Stone age doesn't produce supersonic rockets.

It doesn't indeed, that's why his imaginary wunderwaffe mostly works in propaganda cartoons and he has to turn to his friends in other fascist states for help. What else do you expect if all normal scientists don't have any future there. Brain drain and negative selection are very real. And those who do remain are more interested in filling their pockets like all common thieves, not in any kind of progress.

[agd5f]

it's sad, but my next gpu certainly won't be AMD. when I look at the features that even skylake igpus have that AMD doesn't... I don't know what features will be missing that Nvidia will support.

Do you have any specific examples? Driver extension support is comparable, arguably AMD even supports more extensions: https://mesamatrix.net/

[Quackdoc]

Do you have any specific examples? Driver extension support is comparable, arguably AMD even supports more extensions: https://mesamatrix.net/

mind you with context, this is specifically talking about polaris. off the top of my head VK_EXT_image_drm_format_modifier. this is needed for wlroots vulkan renderers as well as MPV's libplacebo zerocopy rendering. I realize that newer cards have much better current vulkan support. however this makes me particularly worried about longevity of the cards.

https://gitlab.freedesktop.org/mesa/mesa/-/issues/5882

https://github.com/GPUOpen-Drivers/AMDVLK/issues/317

other notable GPUs that support this feature are nvidia's GTX 750ti, intel skylake's igpu etc