#
…

2023-02-14

00:09mkurz has quit [Ping timeout: 480 seconds]
00:15mkurz has joined #asahi-gpu
01:01DarkShadow4444 has quit [Quit: ZNC - https://znc.in]
01:01DarkShadow44 has joined #asahi-gpu
02:03hightower2 has quit [Remote host closed the connection]
02:03hightower2 has joined #asahi-gpu
03:40alyssa 16: 9e138c0202840100 imadd r4_r5.cache, r6.cache, 32, u2
03:40alyssaI *think* the u2 should be u2_u3..
03:41alyssafrom context of the shader I'm staring at
03:41alyssa(Spoiler alert: bindless image writes are 100% a thing and it's a straightforward mechanism and I just need a bit more staring to get the details)
03:47alyssaor not, uh
03:47alyssaI see what's happening. cute.
03:48alyssathey implement (u32 * u32) + u64 in two instructions
03:48alyssalo_hi = mad(u32, u32, u64_lo)
03:48alyssahi = hi + u64_hi
03:49alyssai just mised the second add due to the scheduling
03:51alyssaok, so there's full bindless image_write, so what? well..
03:52alyssanext question is whether there's proper bindless image load i missed
04:08alyssaI still feel like I'm missing something
04:09alyssaare these addresses 32-bits or not?!
04:11alyssamaybe I just need to try to wire this stuff up and see what breaks.
04:31pthariensflame has joined #asahi-gpu
04:31pthariensflame has quit []
04:47TellowKrinkleI think the bindless register references are register pairs and we're not decompiling correctly
04:47TellowKrinkle It usually sets up a valid pair before each of those calls
04:48TellowKrinkleActually maybe not
05:00c10l9 has quit []
05:01c10l9 has joined #asahi-gpu
06:00skmp__ has quit []
06:02md_ has joined #asahi-gpu
06:10i509vcbhey ella-0, I recall hearing you have gotten vkcube to "work" on agxv. vkcube obviously requires WSI and agxv is probably some time away from working wsi so I imagine that showcase was probably from some fork of vkcube. I was wondering if you had the vkcube fork somewhere (if it even exists) to experiment with?
06:11i509vcbI am interested in trying to see if I *could* help with agxv, but I guess I am a bit unfamiliar with the hardware and I imagine some stuff is probably blocking larger open contribution on that
06:13md_ has quit [Quit: Konversation terminated!]
06:15md_ has joined #asahi-gpu
06:17md_ has quit []
06:44bisko has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
07:41bisko has joined #asahi-gpu
08:37MajorBiscuit has joined #asahi-gpu
08:44bisko has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
09:13stickytoffee has quit [Quit: brb]
09:33seeeath has joined #asahi-gpu
09:34bisko has joined #asahi-gpu
09:41seeeath has quit [Ping timeout: 480 seconds]
10:53stickytoffee has joined #asahi-gpu
12:49cylm has joined #asahi-gpu
13:41Cyrinux9 has quit []
13:44Cyrinux9 has joined #asahi-gpu
13:52jannau_ has quit []
13:52jannau has joined #asahi-gpu
13:54seeeath has joined #asahi-gpu
14:41seeeath has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
14:42seeeath has joined #asahi-gpu
16:07possiblemeatball has joined #asahi-gpu
16:08tertu has quit [Ping timeout: 480 seconds]
16:18tertu has joined #asahi-gpu
17:09stickytoffee has quit [Ping timeout: 480 seconds]
17:39stickytoffee has joined #asahi-gpu
17:39stickytoffee has quit []
17:45possiblemeatball has quit [Quit: Quit]
17:51MajorBiscuit has quit [Ping timeout: 480 seconds]
18:04seeeath has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
18:05seeeath has joined #asahi-gpu
18:45linuxgemini1 has quit []
19:38kesslerd has joined #asahi-gpu
19:58mkurz_ has joined #asahi-gpu
19:58kesslerd has quit [Ping timeout: 480 seconds]
20:01mkurz__ has joined #asahi-gpu
20:04mkurz has quit [Ping timeout: 480 seconds]
20:04alyssa has left #asahi-gpu [#asahi-gpu]
20:07mkurz_ has quit [Ping timeout: 480 seconds]
20:22mkurz__ has quit [Read error: No route to host]
20:22mkurz__ has joined #asahi-gpu
20:24mkurz__ has quit []
20:45zalyx has quit [Ping timeout: 480 seconds]
21:32alyssa has joined #asahi-gpu
21:32alyssaan unknown ALU instruction? on M1?
21:32alyssalikelier than you would think
21:35alyssaer no it's a funny encoding of convert
21:36mkurz has joined #asahi-gpu
21:38alyssathis doesn't make any sense, why are they converting a float from uint to float
21:38alyssawhat kind of witchcraft is this
21:38mkurz has quit []
21:40alyssabe91c8280000 is the spicy convert
21:40alyssafor anyone following along at home
21:41alyssathis is for alpha to coverage with msaa 4x
21:41alyssaFirst, they calculate the alpha and clamp it to [0.0, 1.0]
21:41alyssaThen, they multiply it with 255.0
21:41alyssaSo now they have a 32-bit float in the range of [0.0, 255.0]
21:41kesslerd has joined #asahi-gpu
21:42alyssaThen they... reinterpret the float as a 32-bit integer and convert it to a float?
21:44alyssaRange (0.0, 255.0] corresponds to an exponent of [-126, 7]
21:45alyssaencoded as [1, 134]
21:45alyssasign bit is clear
21:45alyssaand then adding back 0 we get an exponent range of [0, 134]
21:45seeeath has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
21:46alyssamantissa is free (save for denormals which have been flushed), so we end up encoding [0, 135 << 23)
21:46seeeath has joined #asahi-gpu
21:46zalyx has joined #asahi-gpu
21:47alyssaso end up with a float in the range [0, 1132462080.0) i guess
21:47alyssathis, this doesn't make sense to me why they would do that, which makes me think i'm missing something
21:56dottedmagAbility of Apple engineers to make bugs?
21:57alyssadottedmag: No, I don't have any reason to think it's wrong
21:57alyssaIt just doesn't make sense what's happening
21:59alyssasee https://developer.apple.com/documentation/metal/mtlrenderpipelinedescriptor?language=objc for the pseudo code, of course the actual expression is implementation-defined
22:23kesslerd has quit [Ping timeout: 480 seconds]
22:33seeeath has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
22:33seeeath has joined #asahi-gpu
22:42hightower3 has joined #asahi-gpu
22:44hightower4 has joined #asahi-gpu
22:49hightower2 has quit [Ping timeout: 480 seconds]
22:52hightower3 has quit [Ping timeout: 480 seconds]
23:26seeeath has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
23:27seeeath has joined #asahi-gpu
23:39seeeath has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
23:40seeeath has joined #asahi-gpu
…