01:56airlied: Lyude: nice!
04:53fdobridge: <!DodoNVK (she) 🇱🇹> HLSL mentioned 🐸
06:21airlied: Lyude: just hit the alloc fail on a laptop here
09:53fdobridge: <redsheep> For those of you working on the game tracker thanks for getting so much in here! I am not sure who Danayer is, but if you see this it's okay to edit again.
09:55fdobridge: <redsheep> I just finished sorting out some issues I was starting to see. The individual reports really need statuses of their own, so I added that to the reports sheet. Given that sorting the reports breaks all of the links from the game sheet having two sheets just makes it too brittle, so I have removed it.
09:55fdobridge: <redsheep> The idea of having one overarching status per game is not really very maintainable
09:57fdobridge: <redsheep> In order to make it work I had to add statuses for games with multiple entries, if you see a status on one of your reports that you disagree with go ahead and change it.
11:50fdobridge: <Sid> @redsheep I discovered, it wasn't NVK causing loading slowdowns
11:50fdobridge: <Sid> it was proton experimental bleeding-edge-debug
11:51fdobridge: <Sid> even though I had logging disabled
11:52fdobridge: <Sid> because I saw the same slowdowns on nv prop when I tested that out just for the sake of it
11:54fdobridge: <redsheep> I haven't been using bleeding edge. Maybe we were seeing different issues, I am not even 100% certain what I saw was a genuine regression.
11:55fdobridge: <Sid> funky
13:02fdobridge: <gfxstrand> Thanks to everyone who's testing stuff out! I'm sure we'll eventually outgrow a spreadsheet but it's a good place to start.
13:03fdobridge: <gfxstrand> We may eventually want to move to protondb but I'll need to figure out how to effectively get data out of it first.
13:03fdobridge: <gfxstrand> For now, this gives us a spot for us and I think that's fine. We'll see how it goes.
13:07fdobridge: <Sid> we'll make our own protondb, with blackjack and hookers
13:07fdobridge: <Sid> :wolfFIRE:
13:07fdobridge: <Sid> (protondb isn't open source, there's not way to get data out of it other than scrape it, for now)
13:13fdobridge: <marysaka> I was updating my test bench to play some stuffs and do some testing here
13:13fdobridge: <ahuillet> so, what game am I testing the driver with this weekend
13:13fdobridge: <redsheep> Something better than the spreadsheet would be nice, but I think as the sheet has already shown pretty well (without me even adding most of my data yet) I don't think it will be long before we're no longer in a state where tracking outside of full issues makes sense
13:14fdobridge: <ahuillet> sometimes the simplest tool is the best...
13:14fdobridge: <Sid> yeah, I haven't added all my reports as well
13:14fdobridge: <Sid> just got the non-working ones in asap before I forgot about them
13:15fdobridge: <redsheep> Yeah same, I've tested like 50 games and most of them don't crash.
13:15fdobridge: <gfxstrand> It's hosted by Valve and I know people...
13:16fdobridge: <gfxstrand> IDK what query tools they have for it, though.
13:17fdobridge: <Sid> oh, Valve adopted it?
13:17fdobridge: <Sid> as far as I knew it was a passion project by a community member 😅
13:17fdobridge: <Sid> wait, no, it's still that
13:17fdobridge: <Sid> https://cdn.discordapp.com/attachments/1034184951790305330/1233406662031970374/image.png?ex=662cfaf7&is=662ba977&hm=a70dd3a24a738ea7dab2301a34ae1418b830f9d2fd5de3b9673730b5d5fa316f&
13:18fdobridge: <Sid> :o
13:18fdobridge: <Sid> https://github.com/bdefore/protondb-data
13:23fdobridge: <gfxstrand> Oh, okay, then.
13:23fdobridge: <gfxstrand> I guess I'm mistaken.
13:24fdobridge: <ahuillet> on the flip side, seems like the data is public, though I'm not sure what you had in mind of doing with it
13:24fdobridge: <Sid> yeah, data is public, snapshots released every month
13:24fdobridge: <gfxstrand> Mostly, I want to get a sense of where we're at right now.
13:25fdobridge: <gfxstrand> But I think it's better to do that with targeted testing by folks with decent sized steam libraries.
13:25fdobridge: <Sid> yeah, protondb isn't the most reliable imo, what with people putting in placebo launch options and env vars all over the place 😅
13:25fdobridge: <gfxstrand> Too many reports and things quickly become unactionable.
13:26fdobridge: <gfxstrand> We're already at the point where I don't really know what to do next. 😂
13:27fdobridge: <redsheep> Interlock?
13:27fdobridge: <gfxstrand> I mean I have a plan. I'm in vacation mode until late next week and then I'll be trying to land modifiers and start working on my perf checklist.
13:27fdobridge: <Sid> vkd3d-proton tracker issue would be a good place to start as well
13:27fdobridge: <Sid> :wolfFIRE:
13:27fdobridge: <gfxstrand> That's not too far down the list.
13:27fdobridge: <phomes_> gfxstrand: testing reminded me about this issue I hit in zero_vram(). There is an assert if the bo is above a certain size. I hit that on multiple games.
13:28fdobridge: <phomes_> Do you remember the reason for this limit? https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/src/nouveau/vulkan/nvk_device_memory.c#L58
13:28fdobridge: <redsheep> I always love to hear performance mentioned
13:28fdobridge: <phomes_> oh. Vacation mode. Never mind 🙂
13:28fdobridge: <redsheep> I do think performance is #1 in my mind at this point
13:28fdobridge: <Sid> also might be worth looking at how to change our thingy to have compute-only queues as well
13:28fdobridge: <gfxstrand> Just me being lazy.
13:29Lyude: airlied_: gotcha, I should be sending out the patch for it today
13:29fdobridge: <gfxstrand> Nah, you're fine. I'm sitting here chatting on my phone while waiting for a plane.
13:29fdobridge: <gfxstrand> Y'all don't need to worry about bothering me. I'm pretty good at ignoring people when I need a break.
13:30fdobridge: <Sid> hope you have a good vacation
13:30fdobridge: <Sid> :saigeheart:
13:30fdobridge: <gfxstrand> I will
13:30fdobridge: <gfxstrand> Yeah, next on the list is a cbuf0 rework. Then I'm going to take a crack at figuring out compression.
13:31fdobridge: <gfxstrand> I really hope the kernel doesn't need fixes for that... 😬
13:32fdobridge: <gfxstrand> But I also want to figure out modifiers ASAP so we can drop the bad Zink paths.
13:32fdobridge: <gfxstrand> That's the last step for making Zink competent for GL, I think.
13:33fdobridge: <redsheep> That and descriptor buffer like zmike said
13:33fdobridge: <gfxstrand> Yeah, maybe. 🤷🏻♀️
13:33fdobridge: <redsheep> But, that's already being worked iirc
13:33fdobridge: <gfxstrand> We need to do it for sure but it's not as fundamental, IMO.
13:33fdobridge: <esdrastarsis> descriptor buffer is pain
13:33fdobridge: <gfxstrand> That, too
13:34fdobridge: <redsheep> Is it valid to just... Make up our own thing that zink or even vkd3d can use instead of descriptor buffer?
13:39fdobridge: <Sid> that sounds like making things painful for everyone else
13:40fdobridge: <redsheep> Maybe but if that extension is a really poor match for the hardware it might be a better idea to try to avoid code hitting that path
13:50fdobridge: <gfxstrand> I think we'll figure it out
14:05fdobridge: <rinlovesyou> So nice to hear, really looking forward to ditching proprietary for good once things run faster :3
14:05fdobridge: <rinlovesyou>
14:05fdobridge: <rinlovesyou> Enjoy the rest of your vacation!
14:31fdobridge: <gfxstrand> https://tenor.com/view/vacation-vacation-time-cuba-drinks-pool-gif-15936207
14:31fdobridge: <zmike.> you only need descriptor buffers if you care about async separate shader compiles
14:32fdobridge: <zmike.> if you're fine with random stuttering in games that use separate shaders, keep on keeping on
14:34fdobridge: <gfxstrand> I'm not fine with that long term but IDK if descriptor buffer is the way to fix it in the short term. We'll get there.
14:34fdobridge: <gfxstrand> Not that I have any love of descriptor sets...
14:39fdobridge: <mohamexiety> enjoy your vacation and hope you have a good time! ❤️
14:44fdobridge: <dadschoorse> surely there will be yet another vk descriptor extension that will fix everything for all IHVs and ISVs, right?
14:44fdobridge: <zmike.> and then another one after that to Really fix it
14:49fdobridge: <mohamexiety> only for a new vendor to come up with an architecture that is so cursed it's incompatible with that one, too
15:44Lyude: airlied_: fixes posted