01:56 alyssa: gfxstrand: AGX supports write masks for global stores but I don't really care
06:52 airlied: phew though I toasted my a380 and would have to leave the a770 in with it's oscillating fan speeds
11:03 mareko: alyssa: yes, shader_has_one_variant is suboptimal for GLES, or some of the clamp and lower flags shouldn't be set for GLES
11:18 mupuf: DavidHeidelberg[m], sergi: heya, do you guys have any plan to update vckts?
11:46 mareko: please somebody let alyssa know that I sent her the message above
11:46 mareko: when she's back
11:51 DavidHeidelberg[m]: mupuf: Hi, not yet
11:55 mupuf: DavidHeidelberg[m]: ok, and the process for it would be that we need to stress all drivers before merging?
12:03 DavidHeidelberg[m]: Yeah, and backport ~2 patches
12:58 mupuf: DavidHeidelberg[m]: ok, thanks for confirming
12:58 mupuf: how many runs do we need to do though? And when?
13:27 DavidHeidelberg[m]: mupuf: imho min. 3-5; optimally 20-30; 50+ is maybe overkill :D I usually check queue by https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests?scope=all&state=opened&assignee_username=marge-bot if marge.is_empty and ~ EU.night && US.night; :D
13:28 mupuf:can't do that, unfortunately
13:28 mupuf: DavidHeidelberg[m]: weekend is the best I can do
13:28 DavidHeidelberg[m]: mupuf: then weekend is usually also good
13:29 mupuf: I think we need to make the stress test not queue a new run as long as an MR is assigned to marge
13:29 mupuf: it will slow the stress test down, but also not DoS Mesa CI
13:29 mupuf: at least until we finally get time to work on fixing gitlab runners
13:30 DavidHeidelberg[m]: I have miniscript for that
13:31 mupuf: mind trying to upstream that then?
13:31 DavidHeidelberg[m]: ...somewhere
13:33 DavidHeidelberg[m]: mupuf: best I can now is: https://gitlab.freedesktop.org/okias/mesa/-/commit/6d70d30c49b7bf5ffce3824cf38e1019d351198f
13:34 mupuf: DavidHeidelberg[m]: thanks
13:36 DavidHeidelberg[m]: I want to upstream it, but there has been some ideas for improvements, which I should get to do
17:17 anholt: gallo: what are you up to on !21002?
17:20 gallo[m]: anholt: the performance tracking needs the performance jobs created to have data. The last time we had updates was on Jan 25. Sorry for the last marge assignment. I meant to assign it before, but I got confused with the UI
17:21 anholt: you just unassigned it from marge
17:23 gallo[m]: oh, sorry. I had the MR tab opened with the sidebar active, and then I clicked in the main div, and it triggered something. I did not realize it. My bad
17:38 alyssa: and that's when it hit me
17:38 alyssa: i have no idea how to install mesa
17:44 MrCooper: it's called blissful ignorance
17:48 danvet: MrCooper, I was about to say this is what distros are for :-P
17:49 alyssa: I used to just scp libgallium_dri.so around and yolo
17:56 Ristovski: alyssa: I assume you have a bouncer and read the message from marek? (he wanted to make sure you got it)
17:58 alyssa: Ristovski: :+1:
17:59 HdkR: alyssa: Just `sudo ninja install` right? :)
18:00 HdkR: Just like installing a new glibc </s>
18:06 alyssa: mareko: sounds like a good first task for someone on-boarding, opened up https://gitlab.freedesktop.org/mesa/mesa/-/issues/8209 and hopefully a student/newcomer bites :)
18:09 ccr: "it's a trap!"
18:13 alyssa: lol
19:03 karolherbst: looks like a 2 hour project, turns out to be a 4 months one
19:10 mattst88: definitely been there before
19:13 alyssa: mood
19:14 Ristovski: And just wait until you hit something like `=` instead of `==`! /s
19:38 zmike: I thought zink would take me an afternoon at most
19:39 alyssa: zmike: i mean kusma already did most of the work in 2019, what's left for you to finish, just a few gl exts? :p
19:39 zmike: yeah thankfully
19:42 ccr: ye olde 90%/10% rule
19:43 kusma: That's exactly what happened, yeah
19:47 jenatali: Ugh I broke the vectorizor for push consts just by adding alignment?
19:47 zf: does MESA_GLSL still work?
19:48 zf: I'm trying to set MESA_GLSL=dump and seeing nothing
19:54 zf: also, is there a way to get llvmpipe to print shader bytecode?
20:07 airlied: zf: make sure you have debug builds
20:07 zf: ah, I forget that some libraries need that, thanks
20:08 airlied: for llvmpipe GALLIVM_DEBUG=ir
20:08 zf: (coming from Wine, where nobody turns off debugging when distributing...)
20:08 zf: does that dump final CPU bytecode, or nir?
20:10 airlied: zf: dumps llvm ir
20:11 airlied: by cpu bytecode do you mean assembly?
20:11 airlied: then add asm to the debug flags
20:11 airlied: though ive hit some crashes in the past in the disasm
20:11 zf: assembly, yeah
20:12 zf: I mean, I can disassemble manually, bytecode would be fine :D
20:12 zf: obviously I'm not sure where the bug actually is yet
20:21 zf: thanks!
20:25 airlied: bbrezillon: do you have to use io pgtable btw? i think you will have some problems with fences vs mem allocs
20:53 hakzsam: could someone with gitlab admin act on https://gitlab.freedesktop.org/mesa/mesa/-/issues/8190 please?
21:37 zmike: DavidHeidelberg[m]: I'm seeing virgl-iris-traces job running from changes to a branch that only changes lavapipe files
21:37 zmike: do you know if this is intended
21:38 anholt: zmike: it's a test-manual-mr, so it should appear but be manual
21:39 zmike: ah ok
21:39 anholt: hopefully the manual gets removed quickly, otherwise I need to make a manual test-source-dep.yml list for it
21:39 anholt: (probably me, because test-source-dep is impenetrable, as I have recently demonstrated)
21:39 zmike: amen
21:39 gfxstrand: hakzsam: Done
21:40 hakzsam: thanks!
21:51 jenatali: 10 minutes for the sanity job is a bit of a pain...
21:53 gfxstrand: :-/
21:54 daniels: jenatali: queued?
21:54 jenatali: Yep
21:54 jenatali: Looks like there's no runner available
22:00 jenatali: gfxstrand: Would love an ack/review from you on the first 2 patches in !21029. Then I'd probably split them out and land them first
22:01 jenatali: I'd also love an ack/review on !20945, at least the first patch - the second (useful) patch I can get someone on my team to look
22:06 daniels: jenatali: looks like just NetworkManager doing really long-running tests again
22:06 daniels: one way to ameliorate that would be by getting them to be able to better parallelise their tests
22:06 gfxstrand: jenatali: Is adding align really all that's needed for the vectorizer to run?
22:07 jenatali: It causes progress to return true, since align gets updated by the vectorizer
22:15 gfxstrand: ah
22:31 zf: hmm, somehow the broken shader is also triggering "Illegal set constant without bind flag"
22:31 zf: I'll keep debugging, but I don't suppose that's interesting?
22:32 airlied: nope not interesting usually
22:32 zf: it seems somehow there's no fragment shader loop at all
22:33 airlied: all the triangles get culled?
22:34 zf: I don't know why they would; whatever's going wrong is specific to the fragment shader body
23:11 computermouth: Howdy folks, where's a good channel for asking insane mesa questions?
23:12 computermouth: for example: can I compile lavapipe for wasm, and have a portable swrast?
23:17 airlied: computermouth: this is a good place to ask crazy questions, but in terms of can you, the answer is can you?
23:20 computermouth: word, I'd been running into some problems with meson build flags always trying to build for x11 and requiring dri3. Figured I might ask the pro's to see if anyone had already done it
23:34 Ristovski: computermouth: Are you setting -Dplatforms when running meson?
23:36 computermouth: Ristovski: I hadn't, no.
23:36 Ristovski: computermouth: https://gitlab.freedesktop.org/mesa/mesa/-/blob/main/meson_options.txt#L22 by default its set to "auto" which will attempt to build for all platforms
23:37 Ristovski: (based on installed deps etc)
23:37 computermouth: I see, so at the least it sounds like I'd have to make a wasm platform, which maybe only builds static libs of the swrast targets
23:38 Ristovski: If you want to be explicit, you could do -Dauto_features=disabled, which would turn all of those to "disabled" and then explicitly enable anything/everything you need
23:38 computermouth: ah! that could be a good place to start, thanks!
23:39 Ristovski: do note swrast is no longer supported as of mesa 21.0 (superseded by llvmpipe/softpipe). It is still present in the source tree
23:41 computermouth: Oh interesting. I had been using the term synonymously. I was looking to build lavapipe specifically, and had found -Dvulkan-drivers=swrast, I figured that must be it
23:43 Ristovski: confusing nomenclature :P "swrast" is also the name of the old "classic" (non-gallium) software renderer
23:43 airlied: yeah we liked the swrast name :P
23:43 computermouth: The last time I hacked on mesa was pre-gallium, so I recognized it immediately haha
23:47 JoshuaAshton: ~~classic~~ vintage drivers
23:47 JoshuaAshton: FineWine(tm)
23:48 Ristovski: high-ranking drivers
23:48 Ristovski: hmm, doesn't vulkan-swrast depend on llvm though? I feel like that could be a major blocker for wasm stuff lol
23:50 JoshuaAshton: Why would llvm be a blocker? I've seen plenty of clang on the web things...
23:51 computermouth: I also thought it would be fine, but I'm not really an expert in wasm, vulkan, or llvm
23:51 computermouth: so feel free to tell me I'm a goofball
23:53 kisak: vulkan-swrast (arch package) provides lavapipe, lavapipe does use llvmpipe
23:55 Ristovski: Well I be damned https://soedirgo.github.io/llvm-wasm/ (fetches ~77MB of compressed wasm binaries, 199MB when decompressed)
23:55 Ristovski: oh oof, that demo isn't that good, it re-fetches the blobs on every recompile