#
…

2023-06-07

00:03brolin has quit [Ping timeout: 480 seconds]
00:14* alyssais talking to a maintainer of another (non-Mesa/Linux project)
00:14alyssawiki style has come up, lol
00:14alyssa"I don't need review, you do" is unbalanced
00:15alyssa"I need review just like you" is one way to balance but
00:15alyssa"nobody needs to pre-merge review, look post-merge" is another (~:
00:22brolin has joined #asahi-gpu
00:42flibitijibibo has quit [Quit: Leaving]
00:46brolin has quit [Ping timeout: 480 seconds]
01:23brolin has joined #asahi-gpu
01:28hightower3 has joined #asahi-gpu
01:35hightower2 has quit [Ping timeout: 480 seconds]
01:45Method has quit [Ping timeout: 480 seconds]
02:04brolin has quit [Ping timeout: 480 seconds]
02:35stsmwg has joined #asahi-gpu
02:36cylm has joined #asahi-gpu
02:40r0ni has quit [Quit: Textual IRC Client: www.textualapp.com]
03:32pyropeter1 has joined #asahi-gpu
03:33eidial has joined #asahi-gpu
03:34PyroPeter_ has quit [Ping timeout: 480 seconds]
03:39eidial has quit [Remote host closed the connection]
05:25drubrkletern has joined #asahi-gpu
05:40cylm has quit [Ping timeout: 480 seconds]
05:44Method has joined #asahi-gpu
05:44cr1901 has quit [Remote host closed the connection]
05:44cr1901 has joined #asahi-gpu
05:51drubrkletern has quit [Remote host closed the connection]
05:55cr1901 has quit [Remote host closed the connection]
05:55cr1901 has joined #asahi-gpu
06:26chadmedi feel dirty for not reviewing jannau's ebuilds before slapping that rebase and merge button
06:28chadmedone day our luck is going to run out an 300 issues are going to be posted about some silly lil ebuild nuking peoples root fs
06:32jannauchadmed: -EWRONGCHAN, but it's mostly just renaming ebuilds for version bumps
06:32chadmedah i was just adding $0.02 to alyssa's musings on review :P
07:17marcannobody reviews my kernel code and apparently I broke suspend in the latest release, sooooo :P
07:20jannauI look at them post merge but I wouldn't have spend enough time on brcmfmac to spot that the change breaks suspend
07:34hightower3 has quit [Ping timeout: 480 seconds]
07:35chadmed has quit [Remote host closed the connection]
07:39chadmed has joined #asahi-gpu
07:52nimprod3l has joined #asahi-gpu
08:07svenI try to not look at Broadcom stuff if i can avoid it for my sanity ;)
08:46austriancoderalyssa: I am always open to learn something
08:50hightower2 has joined #asahi-gpu
08:57nimprod3l has quit [Quit: Leaving]
09:50nsklaus has joined #asahi-gpu
11:04hays has quit [Remote host closed the connection]
11:05cylm has joined #asahi-gpu
11:30hays has joined #asahi-gpu
11:30Brandon has joined #asahi-gpu
11:33Brandon is now known as heliumlake
11:56c10l48 has quit []
12:03c10l48 has joined #asahi-gpu
12:06ColnMustard has joined #asahi-gpu
12:08heliumlake has quit [Remote host closed the connection]
12:10ColnMustard has left #asahi-gpu [Leaving]
12:18brandon has joined #asahi-gpu
12:18brandon is now known as heliumlake
12:20heliumlake has quit []
12:33i509vcb has quit [Quit: Connection closed for inactivity]
12:52mkurz has quit [Ping timeout: 480 seconds]
13:29mkurz has joined #asahi-gpu
13:31r0ni has joined #asahi-gpu
13:41mkurz has quit [Remote host closed the connection]
13:49SinSinati5 has joined #asahi-gpu
13:53stipa has joined #asahi-gpu
14:04SinSinati5 has quit [Remote host closed the connection]
15:05i509vcb has joined #asahi-gpu
15:13c10l48 has quit [Ping timeout: 480 seconds]
15:25nimprod3l has joined #asahi-gpu
15:42alyssaearly frag tests + side fx: first instruction is a `sample_mask ~0, 1`, which I guess triggers tests to happen
15:43alyssaafter that normal
15:43alyssa(before that is wait pix 512, 3 and fencing)
15:43alyssa 3c: f5aa memory_barrier 2, 2, 10
15:43alyssa 3e: f5ae memory_barrier 3, 2, 10
15:43alyssa 40: f5a9 memory_barrier 2, 1, 10
15:43alyssa 42: f5ad memory_barrier 3, 1, 10
15:43alyssafencing needed for image_write -> texture_load
15:43alyssainterestingly no wait is required
15:44alyssamaybe there's a builtin wait
15:45nimprod3l has quit [Quit: Leaving]
16:02cylm has quit [Ping timeout: 480 seconds]
16:03alyssaGuest2261: indeed
16:12alyssaa control barrier with a texture fence:
16:12alyssa 4e: f596 memory_barrier 1, 2, 9
16:12alyssa 50: f5aa memory_barrier 2, 2, 10
16:12alyssa 52: f5ae memory_barrier 3, 2, 10
16:12alyssa 54: 6800 threadgroup_barrier
16:12alyssa 56: f5a9 memory_barrier 2, 1, 10
16:12alyssa 58: f5ad memory_barrier 3, 1, 10
16:12alyssathe same basic ops there
16:16possiblemeatball has joined #asahi-gpu
16:22c10l48 has joined #asahi-gpu
16:23zzywysmFYI Apple dropped a huge patch on top of wine that adds DirectX12 support: https://github.com/apple/homebrew-apple/blob/main/Formula/game-porting-toolkit.rb
16:24alyssaI'm aware.
16:25alyssalife goes on for us *shrug*
16:26alyssaultimately FOSS will win out technically, usually does
16:28alyssaupstream community-backed FOSS, even
16:28alyssathere are good reasons gamers prefer radv to amdvlk, despite the latter even being open source (!)
16:28TellowKrinkleAFAIK the open source part of the patch is mostly proton stuff. The DX12 support is not included, it's a closed source DX-Metal translator from Apple
16:28zzywysmmerely thought it would make lina's livestreamed games testing more interesting
16:28alyssathat was my read as well
16:29alyssaTellowKrinkle's
16:29alyssaat any rate, it's not super on topic here
16:29TellowKrinkleHow? Lina isn't going to be running their DX-Metal translator in Linux...
16:31TellowKrinkle(Also vkd3d is a thing, it works pretty well with Vulkan drivers that aren't MoltenVK)
16:31alyssavkd3d + agxv + fex + linux is the dream
16:31alyssahow will vkd3d + agxv + fex + linux stack up against d3dmetal + apple's metal driver + rosetta + macOS?
16:31alyssawell, we'll know in a few years I suppsoe.
16:35* eric_engestromlikes this future, much better than the "have to keep a windows partition around just for gaming" from a couple of decades ago
16:41alyssaI like the former better than the latter >:
17:36c10l48 has quit []
17:44c10l48 has joined #asahi-gpu
17:45LinuxM2 has joined #asahi-gpu
17:47mkurz has joined #asahi-gpu
17:59stsmwg has quit [Quit: Lost terminal]
18:02Armlin has joined #asahi-gpu
18:03yuka has quit [Remote host closed the connection]
18:04yuka has joined #asahi-gpu
18:05c10l48 has quit [Ping timeout: 480 seconds]
18:06c10l48 has joined #asahi-gpu
18:17Armlin has quit []
18:31c10l48 has quit []
18:39c10l48 has joined #asahi-gpu
18:42possiblemeatball has quit [Quit: Quit]
18:46c10l48 has quit [Read error: Connection reset by peer]
18:46c10l48 has joined #asahi-gpu
18:56hightower2 has quit [Ping timeout: 480 seconds]
18:58brolin has joined #asahi-gpu
18:59c10l48 has quit [Ping timeout: 480 seconds]
19:09c10l48 has joined #asahi-gpu
19:25c10l48 has quit [Ping timeout: 480 seconds]
19:26hightower2 has joined #asahi-gpu
19:30c10l48 has joined #asahi-gpu
19:37c10l484 has joined #asahi-gpu
19:37c10l48 has quit [Read error: Connection reset by peer]
19:40ChaosPrincessalyssa: so, the main feature of geometry shaders is the ability to write out multiple vertices. But their max count is known and capped ahead of time. So on hw level, what is the difference between a multi-output vertex shader and a geometry one?
19:50alyssaChaosPrincess: By multi-output I assume you mean Metal's "amplification"?
19:50alyssaAs far as I know, that's limited to outputting 2 vertices (instead of 1) from the VS
19:50alyssawhich makes it ~useless for our pruposes
19:50alyssa(Geometry shaders need like 256 min-max vertices)
19:53hightower3 has joined #asahi-gpu
19:54hightower3 has quit []
20:00LinuxM2 has quit [Quit: Leaving]
20:04ChaosPrincessOkay, not quite multi output with vertex but more like compute ones writing to an array
20:05alyssaRight, geometry shaders on AGX are conceptually implemented as compute shaders writing to an array and then fed into the rasterizer with a passthrough vertex shader
20:05alyssaSimilar for mesh shaders and tessellation shaders
20:05ChaosPrincessSeems too "simple"
20:06alyssaFrom what i can tell, D3DMetal is doing geometry-on-mesh-on-vertex which sounds like it has "interesting" performance characteristics
20:06ChaosPrincessthere is bound to be some weird corner case that makes it ass
20:06alyssaChaosPrincess: Well, yeah. Devil's in the details.
20:06alyssaPiles and piles of em
20:06alyssabut the basic idea isn't too bad
20:06brolin has quit [Ping timeout: 480 seconds]
20:06alyssaYou get lots of weirdness in all the non-corner cases, actually
20:06alyssalike... geometry shaders output triangle *strips*, not triangles
20:06alyssaand they can end the strip and start a new strip whenever they want
20:07ChaosPrincessCan agx take triangle strips as vertex shader inputs?
20:07alyssayes, that's fine
20:07alyssabut either you need to do extra copying (synthesizing vertices that don't actually exist to feed in triangles), or you need to generate an index buffer with primitive restart so you can feed in strips
20:08alyssaboth are decidedly more complicated than "just write to an array"
20:08ChaosPrincessSo i take it you also need to write indices from "geometry"
20:08alyssasimilar issues on the input side
20:08alyssaif you feed a geometry shader with inputs with an index buffer, well if you're doing compute, you get to do primitive assembly yourself
20:08alyssahave fun reading the index buffer yourself in compute
20:08alyssanone of these problems are impossible to solve
20:08alyssabut there are lots and lots of them
20:09alyssaand then.. transform feedback
20:09alyssabut i don't want to talk about that right now o_o
20:09ChaosPrincessIs primitive assembly "after vertex" or "before raster"?
20:10ChaosPrincessGuess im assuming its a fixed function block - c/d?
20:26c10l484 has quit [Read error: Connection reset by peer]
20:27c10l484 has joined #asahi-gpu
20:30aafeke_ has joined #asahi-gpu
20:34alyssaforce-early-z has depth unchanged, pass type "Translucent punch through", tri merging enabled
20:34alyssalate-z has depth any, pass type Punch through, tri merging disabled
20:34alyssawith fragcoord z input
20:34alyssazs_emit at the end of the shader
20:35alyssawith late-z and no output, still zs_emit at very end
20:36alyssaunconditional discard 1/2 set
20:36alyssaunknown 1:0 cleared
20:36alyssa(from 1)
20:36alyssaunknown 3:0 next to cf bindings cleared
20:36alyssaand unknown 7 cleared to 0
20:36alyssain occlusion query 2
20:41TellowKrinkleIs tri merging mixing pixels from different triangles in one quad? Do they have to disable that for anything that needs derivatives?
20:42alyssaTellowKrinkle: It's not 100% clear how tri merging works in the hw, but yes, disabled for anything needing derivatives
20:44alyssafor force early with no rt, just sample mask at the start
20:44alyssapass type trans punch through
20:44alyssatri merging enabled
20:45alyssaunconditional discard 1/2 still set
20:50alyssawell all those bits are set so, um,
20:51alyssaoh i can't
21:25aafeke_ has quit [Quit: aafeke_]
21:26aafeke_ has joined #asahi-gpu
21:43aafeke_ has quit [Quit: aafeke_]
22:08alyssaDisturbingly, it seems txf does not ignore the sampler (-:
22:08alyssain particular, the hw applies the wrap mode (-:
22:49compassion18 has joined #asahi-gpu
22:51alyssaand also disturbingly, I still can't find where bindless samplers are in memory
22:51alyssathey just.. don't exist?
22:51alyssahow is this memory possible mapped?
22:51alyssalina: I think this might be one for you. wrap.dylib is giving me spooky action
22:52compassion1 has quit [Ping timeout: 480 seconds]
22:52compassion18 is now known as compassion1
22:52alyssaIn Metal, if you have a sampler in an argument buffer and use it
22:52alyssawhere does that sampler live in GPU memory?!
22:52alyssaThere's some global heap of sampler descriptors *somewhere*
22:52alyssabut where
22:53alyssai'm dumping all known memory and not seeing anything happen
23:01cylm has joined #asahi-gpu
23:30brolin has joined #asahi-gpu
23:37alyssa2023-06-07 19:37:07.887488-0400 app[4245:43641] Execution of the command buffer was aborted due to an error during execution. Invalid Resource (00000009:kIOGPUCommandBufferCallbackErrorInvalidResource)
23:37alyssathat looks, possibly relevant
23:38alyssaIDK. -ELATE
23:38alyssalina: Definitely one for you I think
23:38brolin has quit [Ping timeout: 480 seconds]
23:38alyssawe're specifically interested in samplers encoded to argument buffers, where does the actual sampler descriptor end up in memory?
23:39alyssaThis probably will correspond to UAPI stuff
23:42mkurz has quit [Ping timeout: 480 seconds]
23:54Z750 has quit [Quit: bye]
23:55Z750 has joined #asahi-gpu
23:55Z750 has quit []
23:55Z750 has joined #asahi-gpu
…