00:23fdobridge: <gfxstrand> Does GL work?
00:30fdobridge: <azkali> Yup
00:48fdobridge: <gfxstrand> Okay. IDK why NVK isn't working then.
00:48fdobridge: <gfxstrand> I could probably figure it out if I got myself some Linux on my switch.
00:49fdobridge: <Conan Kudo (ニール・ゴンパ)> has the GSP licensing stuff been worked out yet?
00:49fdobridge: <gfxstrand> I believe GSP is redistributable but I'm not sure on the details.
00:50fdobridge: <Conan Kudo (ニール・ゴンパ)> this is the latest I'm aware of: https://github.com/NVIDIA/open-gpu-kernel-modules/issues/62#issuecomment-1338149520
00:50fdobridge: <gfxstrand> If it isn't yet, that sounds like something Red Hat can sort out.
00:52fdobridge: <gfxstrand> Yeah, that sounds like something for Red Hat to push on.
00:54fdobridge: <Conan Kudo (ニール・ゴンパ)> I guess I'll ask @Mr Fall🐧 once they're online then
00:55fdobridge: <karolherbst🐧🦀> ohh I think I might have forgotten to ask 🙃
00:55fdobridge: <Conan Kudo (ニール・ゴンパ)> uhh how are you still online?
00:55fdobridge: <Conan Kudo (ニール・ゴンパ)> isn't it like brain-dead o-clock in the morning there?
01:03fdobridge: <Mohamexiety> flexible sleep schedule 😆
02:29fdobridge: <Conan Kudo (ニール・ゴンパ)> can you fire off a request asap-ish then, please? 🙂
02:30fdobridge: <karolherbst🐧🦀> yeah, will do
02:36fdobridge: <gfxstrand> @Mr Fall🐧 doesn't sleep. Or if he does, he sleeps like a computer with WoL attached to Discord.
02:36fdobridge: <karolherbst🐧🦀> wait, I do sleep
02:36fdobridge: <karolherbst🐧🦀> I'm a normal human being
02:37fdobridge: <Conan Kudo (ニール・ゴンパ)> people say the same thing about me 🤣
02:40fdobridge: <karolherbst🐧🦀> anyway, i'ts _only_ 4:40 am here 😄
02:51fdobridge: <Conan Kudo (ニール・ゴンパ)> people freak out when I say stuff and they know it's like 2~4 am my time
02:52fdobridge: <Conan Kudo (ニール・ゴンパ)> but unfortunately, it's the hazard of being an American OSS developer in community open source projects
02:53fdobridge: <Conan Kudo (ニール・ゴンパ)> there's increasingly a dearth of folks in my timezones in community open source projects... most are in Europe, with a sprinkle of them in Japan and Australia
03:15fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> How can I attach WoL to Discord? 😅
10:27fdobridge: <🌺 ¿butterflies? 🌸> imagine a world where nvgpu has fully stable abis
11:54RSpliet: nvuser: those are rather bland control registers from the looks of it. the coefficients are more interesting. Are those in 137004?
11:55RSpliet: can't remember. Anyway, peeking in that region doesn't harm. Just don't poke :-P
16:25fdobridge: <DadSchoorse> will nak fuse ffma in the backend or will it just use the late algebraic option?
17:02fdobridge: <gfxstrand> Probably in NIR somehow. TBD exactly what that looks like, though. The late algebraic option is kinda meh. There's an intel-specific pass that's better. Might steal that.
17:42nvuser: RSpliet: aha, for sure I won't try to write anything there.. well I just did ./nvapeek on each PLL listed in gf100 xml document, I'm not sure if those are where I should have looked
17:43nvuser: Changing the nvidia control panel perf values changed something with those values, so I guess that means they do something for kepler
17:44nvuser: the ones I posted are with max perf (1006 MHz core clock)
17:45fdobridge: <phomes> I reworked the conditional rendering MR again. It precalculates if condtional rendering is active in a separate mme. This makes it only use a single temporary register in draw, clear, and dispatch calls. Maybe it is good enough for pre-turing?
17:50nvuser: RSpliet: 137004 was not listed there that I could tell, where I pasted the 0x137<etc> values are the ones I found, the bottom values are the output for each respectively
17:55nvuser: oh hm, the xml does seem to suggest 'coef' offsets, maybe I should append those and get that output as well
18:00nvuser: okay I'll give it another go heh
18:01fdobridge: <ByLaws> @phomes any reason for not using the actual conditional rendering regs?
18:06fdobridge: <ByLaws> I guess you need a temp buffer for inverted
18:06fdobridge: <ByLaws> I guess youd need a temp buffer for inverted with them (edited)
18:14fdobridge: <phomes> @ByLaws no reason other than me not knowing about them 🙂 I am still learning new things every time I decide spend a little time on mesa
18:14fdobridge: <ByLaws> Ah
18:15fdobridge: <ByLaws> SET_RENDER_ENABLE_B_OFFSET
18:15fdobridge: <ByLaws> With SET_RENDER_ENABLE_C_MODE_CONDITIONAL, if data at that address is zero or won't draw
18:15fdobridge: <ByLaws> Or clear
18:16fdobridge: <ByLaws> I.e. draw methods just won't do anything
18:16fdobridge: <ByLaws> Even if they are called
18:16fdobridge: <ByLaws> Idk if you'd wanna look into since your current approach makes inverted easier
18:17nvuser: okay, I'll put that down to pebkac.. I think I have the right values now, will just double check before posting
18:18fdobridge: <ByLaws> Otherwise you'd have to use I2m with a fifo ref to upload the data + a zero constant
18:18fdobridge: <ByLaws> Then use SET_RENDER_ENABLE_C_MODE_RENDER_IF_EQUAL
18:19fdobridge: <ByLaws> Idk if that helps at all
18:19fdobridge: <phomes> @ByLaws thanks 🙂 I will see if I can get that to work. Maybe they can fix the remaining fails. My method would only work for things that are called in mme's
18:19fdobridge: <ByLaws> With SET_RENDER_ENABLE_C_MODE_CONDITIONAL, if data at that address is zero it won't draw (edited)
18:20fdobridge: <ByLaws> Also wanna ensure NVB197_SET_RENDER_ENABLE_OVERRIDE is 0
18:20fdobridge: <ByLaws> And SET_RENDER_ENABLE_CONTROL_CONDITIONAL_LOAD_CONSTANT_BUFFER_FALSE is set I imagine
18:24nvuser: RSpliet: I guess you mean the coef offsets are appended to the PLL source, so doing that for each clock this is what I have: https://dpaste.com/CR2CJRWBP
18:25nvuser: replacing last hex digit with 4 is how I got coef for each
18:27nvuser: Is CLK0 the core clock aka 0f?
18:54fdobridge: <gfxstrand> PSA for anyone playing with NAK: You'll need a magic meson branch: https://github.com/dcbaker/meson/tree/submit/rust-cargo-subproject
18:56fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I was able to compile it with regular arch-meson
18:56fdobridge: <![NVK Whacker] Echo (she) 🇱🇹>
18:56fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> But vkcube doesn't work yet so I think NAK is not worth messing around right now
18:56fdobridge: <gfxstrand> Really? Interesting...
18:58fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Yes
20:00nvuser: let me know if that's everything needed
20:00nvuser: cheers
20:47fdobridge: <gfxstrand> @ASDQueerFromEU Did you get any steam games running stably with GSP?
20:48fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Even something as simple as SuperTuxKart (with Vulkan renderer) causes the `fail ttm_validate` error 🐸
20:53fdobridge: <gfxstrand> kk
20:53fdobridge: <gfxstrand> @Mr Fall🐧 power management should be good on a 980M, no?
21:00fdobridge: <gfxstrand> @ASDQueerFromEU Was there a game or two that you thought ran pretty well? Looking for something that would make a good demo.
21:08fdobridge: <Mohamexiety> airlied and Esdras usually used Talos iirc
21:09fdobridge: <Esdras Tarsis> I tested Talos without gsp
21:10fdobridge: <Esdras Tarsis> Does it need to be a 3D game?
21:11fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I don't have Talos
21:13fdobridge: <karolherbst🐧🦀> uhm... well... kinda
21:13fdobridge: <karolherbst🐧🦀> if you patch the kernel that is
21:14fdobridge: <karolherbst🐧🦀> could just enable it, but we don't because there is no reliable way to identify if the GPU has a proper fan or we mess up
21:14fdobridge: <karolherbst🐧🦀> so it's currently not enabled
21:18fdobridge: <gfxstrand> oh
21:18fdobridge: <gfxstrand> Let's assume this is a $400 laptop that I'm not going to cry about if it burns. 🙂
21:19fdobridge: <gfxstrand> Also, we're looking at Alienware machines. They should have good fans.
21:20fdobridge: <Esdras Tarsis> Epic gave it away for free
21:20fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Let's double-check if I have it
21:24fdobridge: <karolherbst🐧🦀> yeah, it's not about the fans really, it's just that power management in nouveau is quite yolo
21:24fdobridge: <karolherbst🐧🦀> we don't even have protection against overheating in palce
21:24fdobridge: <karolherbst🐧🦀> but I do have some patches
21:24fdobridge: <karolherbst🐧🦀> let me dig them up tomorrow
21:30fdobridge: <gfxstrand> ok
21:33fdobridge: <gfxstrand> I'm fine with a bit of YOLO. Like, if I just have to find a frequency that works and lock it there, that's fine.
21:50fdobridge: <Conan Kudo (ニール・ゴンパ)> did you do this? 🙂
21:51fdobridge: <karolherbst🐧🦀> we'll talk about it tomorrow with Nvidia
21:54fdobridge: <Conan Kudo (ニール・ゴンパ)> Sweet!