04:39 OftenTimeConsuming: I have a 780 Ti with 0f powermode enabled and for most software it works great. I'm getting terrible performance (10fps) running certain not too graphically demanding games like ultimate stunts, xmoto and minetest? I was wondering if it was a CPU bound thing, but ustunts is a very old game that's designed to run on much slower CPUs and GPUs that what I have, plus I get 120 fps if I
04:39 OftenTimeConsuming: set the viewing distance ridiculously low. Why is this?
06:21 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> Is it possible to add userptr support in the new UAPI?
13:02 fdobridge: <g​fxstrand> It's a separate thing. Should be possible to add, though.
13:03 fdobridge: <g​fxstrand> But, really, I should just finish my damn extension. 🙄
13:04 fdobridge: <g​fxstrand> Userptr is a really bad way to accomplish what they're trying to accomplish.
13:33 fdobridge: <g​fxstrand> Well, the new UAPI is definitely slower. 66us/submit vs. 54us/submit
13:33 fdobridge: <g​fxstrand> https://gitlab.freedesktop.org/mesa/crucible/-/merge_requests/141
13:34 fdobridge: <g​fxstrand> For reference, ANV is around 30-40 us/submit
13:37 fdobridge: <g​fxstrand> @airlied ^^
13:38 fdobridge: <g​fxstrand> I haven't dug into why yet
13:56 fdobridge: <g​fxstrand> strace is being useless... 🙄
13:57 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> I know Wine has an implementation of that WIP extension so drivers just need to support it
14:08 fdobridge: <g​fxstrand> There's driver implementations already. It's waiting for someone to write CTS tests
14:08 fdobridge: <g​fxstrand> That someone might have to be me. 😭
14:09 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> I was only able to find an ANV one
14:09 fdobridge: <g​fxstrand> I thought someone wired it up for RADV, too.
14:09 fdobridge: <g​fxstrand> Should be easy enough to hook up for NVK.
14:09 fdobridge: <g​fxstrand> Again, the real blocker is tests.
14:10 fdobridge: <g​fxstrand> That's why the extension hasn't shipped. 😕
14:11 fdobridge: <g​fxstrand> Sometimes CTS tests magically appear if you ask nicely and wait a while.
14:11 fdobridge: <g​fxstrand> Sometimes they don't
14:11 fdobridge: <g​fxstrand> They haven't. 😭
14:29 fdobridge: <g​fxstrand> If I allocate 1000 memory objects, it jumps to 259us/submit. That's bad.
14:29 fdobridge: <g​fxstrand> @airlied ^^
14:30 fdobridge: <g​fxstrand> Good news is that the new UAPI scales way better. The old one jumps to 1134us/submit with 1k memory objects.
14:30 fdobridge: <g​fxstrand> But with VM_BIND, the number of memory objects literally shouldn't matter past the first submit.
14:39 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> NFS Underground on NVK + DXVK + WoW64 + winewayland 💪
14:40 fdobridge: <g​fxstrand> I sure hope all that BO walking we're clearly doing isn't taking locks. 😉
14:40 fdobridge: <g​fxstrand> I sure hope all that BO walking we're clearly doing isn't taking locks. 😉 (edited)
14:58 fdobridge: <e​sdrastarsis> @karolherbst I think your SLM MR broke nvk build, default isn't returning anything in the `max_warps_per_mp_for_sm` switch case
14:58 fdobridge: <e​sdrastarsis> I fixed returning 0
15:19 fdobridge: <k​arolherbst🐧🦀> make it return the biggest known number
15:19 fdobridge: <k​arolherbst🐧🦀> 0 will break every GPU, returning 64 will just waste VRAM, might break on some newer GPUs having more
15:20 fdobridge: <k​arolherbst🐧🦀> anyway.. rb
15:23 fdobridge: <k​arolherbst🐧🦀> or should I open the MR for that?
15:26 fdobridge: <e​sdrastarsis> Yes, I don't have a freedesktop account yet 🐸
15:55 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> `Image Type operand of OpTypeSampledImage should not have a Dim of Buffer.` 🤔
16:03 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1132342141302091787/Screenshot_20230722_190225.png
17:35 fdobridge: <k​arolherbst🐧🦀> @esdrastarsis https://gitlab.freedesktop.org/nouveau/mesa/-/merge_requests/237
17:50 fdobridge: <g​fxstrand> Merged
17:50 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> Why nvk_get_buffer_alignment() has UNUSED arguments?
17:52 fdobridge: <g​fxstrand> Because we'll need the physical device if we want to make it all correct. They're different on different generations. Right now it's making some pretty rubbish assumptions. @eanholt has an MR that maybe improves it some, maybe, but IDK that it's right.
17:53 fdobridge: <g​fxstrand> At least now we have the infrastructure required to make it right. We just need to work through the details and actually return the right values.
17:53 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> I doubt VkBufferCreateFlags is needed though
17:55 fdobridge: <g​fxstrand> It is. We need it for sparse. Take a look at the new UAPI MR.
18:02 fdobridge: <g​fxstrand> I have a plan! Really, I do! 😅
18:05 fdobridge: <g​fxstrand> I would protest by saying I'm not a crazy woman but you all know I am. 🙃
18:27 fdobridge: <e​sdrastarsis> Nice
18:31 fdobridge: <k​arolherbst🐧🦀> NSFU was so scuffed
18:31 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> Not Safe For UBOs?
18:32 fdobridge: <k​arolherbst🐧🦀> need fo speed underground
18:34 fdobridge: <p​ixelcluster> ~~need sor fpeed~~
18:35 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> Does NVK/nouveau support host-visible VRAM?
18:35 fdobridge: <k​arolherbst🐧🦀> no
18:36 fdobridge: <k​arolherbst🐧🦀> ohh.. will this be part of hte new uapi btw...
18:36 fdobridge: <k​arolherbst🐧🦀> though I guess adding it later probably won't require changing the uapi except maybe a version bump
18:36 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> That will be useful for a :plagfrog: thing
19:10 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> `[384506.375461] nouveau 0000:01:00.0: gamescope[3807935]: validate: -22`
19:48 fdobridge: <g​fxstrand> I thought we could on Turing? We don't but I thought we could.
20:19 fdobridge: <k​arolherbst🐧🦀> in theory we can on all gens, but the kernel never supported it.
20:47 fdobridge: <a​irlied> @gfxstrand so baseline is higher but scales better? I expect validating all the vm bos takes time
20:49 fdobridge: <g​fxstrand> Baseline scales worse, actually.
20:49 fdobridge: <g​fxstrand> But VM_BIND shouldn't have any scaling problems. We shouldn't be validating on every exec
20:49 fdobridge: <g​fxstrand> See email
21:04 fdobridge: <a​irlied> Pretty sure it is using a single dma resv, I think it's validating that has overheads, I think we discussed avoiding validate if the no bo in the vm has moved
21:06 fdobridge: <a​irlied> But i let Danilo dig in and let us know
21:07 airlied: dakr: ^
21:21 fdobridge: <g​fxstrand> How? Swapping out for an individual one on first export? IDK how that would even work.
21:30 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> Anyway here's what I have so far :triangle_nvk:
21:30 fdobridge: <!​[NVK Whacker] Echo (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1132424520020537394/nvk-gamescope.patch
22:02 fdobridge: <e​sdrastarsis> Were you able to get gsp to work?