00:26HdkR: Kind of looks like a tagged pointer. Ignoring the top bits the pointer could be valid :D
00:48fdobridge: <gfxstrand> I'm gonna run again and we'll see how long it takes this one to die.
00:49fdobridge: <gfxstrand> Bottom nibble is `0xb` which is a little odd for a pointer
01:13fdobridge: <karolherbst🐧🦀> could run with kasan or so and see if something triggers
06:59fdobridge: <marysaka> Do you want me to test that on my laptop or with my Turing card in my desktop?
07:01fdobridge: <marysaka> I have been meaning to put back a SSD in that one laptop so I could use that as an excuse to motivate myself to do so :nanikore:
08:33fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> 69 is ROBUST_CHANNEL_GR_CLASS_ERROR (according to OGK)
08:52fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> SuperTuxKart has this (but still loads into the game): `[47243.430946] __vm_enough_memory: pid: 19695, comm: supertuxkart, not enough memory for the allocation`
08:59fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I don't know if this was on RADV or NVK though
09:41fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Hopefully we can achieve a 27x performance improvement on NVK 🚀
09:51fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I mean 22x actuallu
09:51fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> *actually
10:50fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> And it only happens on DXVK v1.10.0+
12:03fdobridge: <karolherbst🐧🦀> @airlied is the `[PATCH] nouveau/gsp/r535: Fix a NULL vs error pointer bug` thing on the ML already taken care of?
12:13fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> The patch definitely does fix an incorrect if statement
12:14fdobridge: <karolherbst🐧🦀> sure, but I wonder if a fix like that already landed, because I think _something_ landed
12:14fdobridge: <karolherbst🐧🦀> not sure if that was the same
12:18fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I think it was this: https://github.com/torvalds/linux/commit/c5f9362307c685fe6a90d344bf81579578fd25d8
12:20fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Which is unrelated to the GSP thing
13:08fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Specifically 0709c5f5c7e4f5f25e92e7cef263bc2edf9128b4
14:58fdobridge: <gfxstrand> I've got a slightly less hellish Turing card in my other box. I can test with that.
14:58fdobridge: <gfxstrand> One of my runs yesterday did survive:
14:58fdobridge: <gfxstrand> `Pass: 411767, Fail: 13, Crash: 9, Warn: 4, Skip: 3577069, Flake: 6, Duration: 59:49`
15:32fdobridge: <theevilskeleton> https://cdn.discordapp.com/attachments/1034184951790305330/1171834698813677690/image.png?ex=655e1f1b&is=654baa1b&hm=45b664ae5b9b157700ae407b52ea2212c4b9ac523a1b8d4babea91b73d52b181&
15:32fdobridge: <theevilskeleton> https://cdn.discordapp.com/attachments/1034184951790305330/1171834704387911792/image.png?ex=655e1f1d&is=654baa1d&hm=db4aedfdb8ebb0f071e6447d75d037a6ffb9c8593b3f4ab14a08b2136b4cf6e7&
15:32fdobridge: <theevilskeleton> Currently rewriting the nouveau wiki's template and CSS
15:33fdobridge: <theevilskeleton> I'm just throwing ideas for myself, but the style is subject to change
15:34fdobridge: <theevilskeleton> I talked to Karol in DMs and he wanted a redesign + accessibility fixes for a long time
16:14fdobridge: <gfxstrand> @marysaka do you mind if we move your sm50 branch to nouveau/mesa?
16:27fdobridge: <marysaka> sure! that would be better
16:27fdobridge: <marysaka> will need a rebase
16:28fdobridge: <gfxstrand> Rebased and pushed as nak/sm50
16:54TimurTabi: airlied: I see that the copr build succeeded. What do I do with it? I'm not familiar with copr. I see that I can download the src rpm. Is there a binary?
16:57fdobridge: <gfxstrand> @marysaka what CTS tests were you using to test your SM50 stuff?
17:00fdobridge: <marysaka> only some from compute I think but I was mostly trying to get encoding somewhat okay with scheduling 😅
17:07TimurTabi: airlied: FYI the 0001 patchfile doesn't apply to the src tarball because there's a merge conflict in then WHENCE file, since mine is based off the linux-firmware tree.
17:07fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> https://github.com/doitsujin/dxvk/commit/0709c5f5c7e4f5f25e92e7cef263bc2edf9128b4
17:15fdobridge: <airlied> TimurTabi: the builds are fedora rpms inside each subdir, however I found a bug with them, but I don't think it's caused by your patch, but not sure, getting an extra gsp symlink to nowhere in some of the subdirs, will dig in a bit next week when my travel is settled down
17:15TimurTabi: airlied: I reported that to Mario.
17:16TimurTabi: His latest change to copy-firmware.sh adds that bogus symlink.
17:16fdobridge: <airlied> @karolherbst I've asked dakr to pick up the patches once drm-misc-fixes rolls forward to include rc1
17:16fdobridge: <airlied> ah cool
17:17fdobridge: <airlied> hi from Honolulu everyone 🙂 (I was not expected to travel via Hawaii, but such is life)
18:14fdobridge: <gfxstrand> @karolherbst Does Maxwell not support vector attribute I/O?
18:15fdobridge: <karolherbst🐧🦀> you mean loads/stores on it?
18:15fdobridge: <gfxstrand> Yeah, ALD/AST
18:17fdobridge: <gfxstrand> There's an instruction field for it in the codegen encoder but it doesn't seem to be working
18:17fdobridge: <gfxstrand> I'm not doing indirects or anything
18:18fdobridge: <marysaka> I'm pretty certain it has those and they should work :nya_confused:
18:19fdobridge: <gfxstrand> Oh, well this is fun... My test passes AND shader traps in dmesg...
18:21fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> The CBP welcomes you
18:27fdobridge: <karolherbst🐧🦀> what's the shader trap?
18:38fdobridge: <gfxstrand> ```
18:38fdobridge: <gfxstrand> [63714.325700] nouveau 0000:65:00.0: gr: TRAP ch 6 [007f777000 deqp-vk[91345]]
18:38fdobridge: <gfxstrand> [63714.325717] nouveau 0000:65:00.0: gr: GPC0/TPC0/MP trap: global 00000000 [] warp 3c000e [OOR_ADDR]
18:38fdobridge: <gfxstrand> ```
19:09benjaminl: hmm, I think I had ALD/AST working on Maxwell in an NAK branch
19:10benjaminl: can look for it later today
19:53fdobridge: <gfxstrand> Oh, I forgot to look at your branch
19:56fdobridge: <gfxstrand> I'm going to start pulling some stuff together
20:44fdobridge: <gfxstrand> They work fine. I just needed `NAK_DEBUG=serialize`. Story of my life... 🙄
20:50benjaminl: haha yeah
20:50benjaminl: ime sched is broken *most of the time* on sm50, so I ended up just hardcoding it to use serialize
20:51benjaminl: https://gitlab.freedesktop.org/benjaminl/mesa/-/commit/f6c3c8e918af2b55c72bbd1bc63aecc9f1b19e09
20:51benjaminl: the idea being we could undo this later after fixing all the sched problems
20:56fdobridge: <benjaminl> also, sounds like you got it working, but if it's still useful here are the encoding I was using earlier:
20:56fdobridge: <benjaminl> ```rust
20:56fdobridge: <benjaminl> fn encode_ald(&mut self, op: &OpALd) {
20:56fdobridge: <benjaminl> self.set_opcode(0xefd8);
20:56fdobridge: <benjaminl>
20:56fdobridge: <benjaminl> self.set_dst(op.dst);
20:56fdobridge: <benjaminl> self.set_reg_src(8..16, op.vtx);
20:56fdobridge: <benjaminl> self.set_field(20..30, op.access.addr);
20:56fdobridge: <benjaminl> self.set_reg_src(39..47, op.offset);
20:56fdobridge: <benjaminl> self.set_field(47..49, op.access.comps - 1);
20:56fdobridge: <benjaminl> self.set_bit(31, op.access.patch);
20:57fdobridge: <benjaminl>
20:57fdobridge: <benjaminl> assert_eq!(op.access.flags, 0x0);
20:57fdobridge: <benjaminl> }
20:57fdobridge: <benjaminl>
20:57fdobridge: <benjaminl> fn encode_ast(&mut self, op: &OpASt) {
20:57fdobridge: <benjaminl> self.set_opcode(0xeff0);
20:57fdobridge: <benjaminl>
20:57fdobridge: <benjaminl> self.set_reg_src(0..8, op.data);
20:57fdobridge: <benjaminl> self.set_reg_src(8..16, op.vtx);
20:57fdobridge: <benjaminl> self.set_field(20..30, op.access.addr);
20:57fdobridge: <benjaminl> self.set_reg_src(39..47, op.offset);
20:57fdobridge: <benjaminl> self.set_field(47..49, op.access.comps - 1);
20:57fdobridge: <benjaminl> self.set_bit(31, op.access.patch);
20:57fdobridge: <benjaminl>
20:57fdobridge: <benjaminl> assert_eq!(op.access.flags, 0x0);
20:57fdobridge: <benjaminl> }
20:57fdobridge: <benjaminl> ```
20:57fdobridge: <benjaminl> looks like they never made it onto the actual MR branch
21:38fdobridge: <gfxstrand> Yeah, I've got it now
21:57TimurTabi: airlied: GSP-RM firmware images have been merged into linux-firmware.git: https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git/log/
21:57TimurTabi: I really hope I got it right.
22:16fdobridge: <gfxstrand> ```
22:16fdobridge: <gfxstrand> Test run totals:
22:16fdobridge: <gfxstrand> Passed: 5962/12160 (49.0%)
22:16fdobridge: <gfxstrand> Failed: 7/12160 (0.1%)
22:16fdobridge: <gfxstrand> Not supported: 6191/12160 (50.9%)
22:16fdobridge: <gfxstrand> Warnings: 0/12160 (0.0%)
22:16fdobridge: <gfxstrand> Waived: 0/12160 (0.0%)
22:16fdobridge: <gfxstrand> ```
22:16fdobridge: <gfxstrand> That's `dEQP-VK.ssbo.*` on Maxwell with the current (just pushed) NAK branch. Now we have something we can regression test. 😁
22:16fdobridge: <gfxstrand> @marysaka @vdpafaor @dwlsalmeida ^^
22:17fdobridge: <marysaka> great 🎉
22:23fdobridge: <gfxstrand> IDK why 7 tests are failing. I should take a look at that
23:04fdobridge: <gfxstrand> @karolherbst What's this `flagsSrc/Def` thing?
23:06fdobridge: <karolherbst🐧🦀> ohh
23:06fdobridge: <karolherbst🐧🦀> it's where the carry is
23:06fdobridge: <karolherbst🐧🦀> I think
23:06fdobridge: <karolherbst🐧🦀> previous gens had a reg of flags
23:06fdobridge: <karolherbst🐧🦀> and the carry was one of those
23:22TimurTabi: airlied: dumb Red Hat question. Why can't I install rdfind in rawhide?
23:22TimurTabi: # yum install rdfind
23:22TimurTabi: Last metadata expiration check: 1:22:31 ago on Wed 08 Nov 2023 03:56:06 PM CST.
23:22TimurTabi: No match for argument: rdfind
23:22TimurTabi: Error: Unable to find a match: rdfind
23:50TimurTabi: I found an rpm on rpmfind.net and it works now. Just seems kinda weird that it's literally the only package I couldn't install with yum.