2026-02-22
00:18pb17 has quit [Ping timeout: 480 seconds]
00:23pb17 has joined #asahi-gpu
00:44pac85I posted the findings of some re that includes shader dumps on #asahi-re but realizing but have been the wrong channel since most GPU stuff seems to be discussed here (and there seems to be different people). Would it be better to repost here including the dumps (or I can leave those out, they are small internal shaders which include new instructions)
01:21ACG_ has joined #asahi-gpu
01:37druze has joined #asahi-gpu
02:43nightstrike has quit [Quit: Connection closed for inactivity]
02:54pb17 has quit [Ping timeout: 480 seconds]
03:02skipwich has quit [Remote host closed the connection]
03:02skipwich has joined #asahi-gpu
03:12pb17 has joined #asahi-gpu
03:45[KongrooParadox] has joined #asahi-gpu
03:48KongrooParadox has quit [Ping timeout: 480 seconds]
03:55dougall_ has joined #asahi-gpu
03:57dougall has quit [Ping timeout: 480 seconds]
06:01hamza has quit [Remote host closed the connection]
06:02hamza has joined #asahi-gpu
06:34ddxtanx has quit [Ping timeout: 480 seconds]
06:41ddxtanx has joined #asahi-gpu
07:36billak has joined #asahi-gpu
07:36billak_ has joined #asahi-gpu
07:44billak has quit [Ping timeout: 480 seconds]
08:26billak_ has quit [Quit: Konversation terminated!]
08:53_master_sword_[m] has joined #asahi-gpu
09:29xXxMeowUwUKittenxXx[m] has joined #asahi-gpu
09:49HerbY_NL has joined #asahi-gpu
10:52HerbY_NL has quit [Quit: Leaving]
11:26pb17 has quit [Ping timeout: 480 seconds]
11:27pb17 has joined #asahi-gpu
12:50linksmith has joined #asahi-gpu
13:11linksmith has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
13:14pb17 has quit [Ping timeout: 480 seconds]
13:26pb17 has joined #asahi-gpu
13:41linksmith has joined #asahi-gpu
14:07oliverI looked into the compression of 16x16 tiles a little. I am confident that the actual compression of the data is exactly the same between gpu tiled and interchange. The only difference is in the order of 8x4px subtiles. with interchange they are ordered left to right, top to bottom. With gpu tiled the subtiles are in morton order.
14:10oliverAlso, unrelated, the value in the metadata/accel buf is the length in bytes of the corresponding 8x4px subtile minus one.
14:35pb17 has quit [Ping timeout: 480 seconds]
14:40linksmith has quit [Quit: My Mac has gone to sleep. ZZZzzz…]
14:44pb17 has joined #asahi-gpu
14:51alyssauniform registers, texture state registers, sampler state registers, coefficient registers
14:51alyssaThe driver puts the right control words and then the program has the environment accessible easily.
14:52alyssauniform registers are the spiciest - Likely they still exist in the form of something like AMD's SGPRs but you can't use them the traditional m1 way because that part of the control stream is gone
14:53alyssaBut that presents a chicken & egg problem - how do you get kernel arguments and things like that into the shader if you don't have the sideband?
14:54alyssaBased on the snippet you linked, the solution is pretty simple: dynamically generate a prolog that uses shader instructions to setup the state that would previously be done with USC control words.
14:55alyssaThe first two moves are hardbaking the address of the data to plumb in, which those last 4 moves (from the loaded data into the uniform registers) are dealing with
14:56alyssaAlso this looks like a preamble shader, which on M1 are independent binaries from the main shader, not sure if they merged that in M3
14:57alyssaLikely the unknown instructions relate to dynamic caching / setting up other bits of state that traditionally USC control words managed but is now shader managed
15:00alyssaBut if I did have to make a guess, one of those instructions does the equivalent of the M1 "USC Shared" struct.
15:00alyssaIf you wanted to confirm that hypothesis, compare against shaders that use threadgroup memory (and different amounts of it)
15:01alyssaPotentially one of the instructions adapts "USC Registers" - in this case, dynamically allocating GPRs from the cache for the rest of the program to use
15:02alyssaThat you can confirm by varying the number of registers used, which is easy with an unrollable loop
15:02alyssaint temp0, ...., tempN; for(..) { temp0 = temp1; temp1 = temp2; ... tempN-1 = tempN; tempN = temp0; } foo(tempN)
15:10dougall_ has quit [Ping timeout: 480 seconds]
15:13dougall has joined #asahi-gpu
15:31robertwendt has quit [Ping timeout: 480 seconds]
15:40robertwendt has joined #asahi-gpu
15:54kaazoo has joined #asahi-gpu
16:17pb17 has quit [Ping timeout: 480 seconds]
16:22pb17 has joined #asahi-gpu
16:34lederstrumpf has quit [Read error: No route to host]
16:35lederstrumpf has joined #asahi-gpu
17:22hamza has quit [Remote host closed the connection]
17:22hamza has joined #asahi-gpu
18:01marcin_ has joined #asahi-gpu
18:20marcin_ has quit [Remote host closed the connection]
18:25michaelros4 has quit [Remote host closed the connection]
18:25gi0 has joined #asahi-gpu
18:26gi0 has quit []
18:27gi0 has joined #asahi-gpu
18:43michaelros4 has joined #asahi-gpu
18:45michaelros4 has quit [Read error: Connection reset by peer]
18:45michaelros4 has joined #asahi-gpu
18:48emptynine has quit [Remote host closed the connection]
18:49emptynine has joined #asahi-gpu
18:50ciara has quit [Remote host closed the connection]
18:50ciara has joined #asahi-gpu
19:33pb17 has quit [Ping timeout: 480 seconds]
19:40pb17 has joined #asahi-gpu
19:58kaazoo has quit [Quit: Leaving.]
20:41hamr8 has quit [Ping timeout: 480 seconds]
21:00KongrooParadox has joined #asahi-gpu
21:03[KongrooParadox] has quit [Ping timeout: 480 seconds]
21:25pb17 has quit [Ping timeout: 480 seconds]
21:42pb17 has joined #asahi-gpu
22:06awb_dev has joined #asahi-gpu
22:08awb_dev has quit [Remote host closed the connection]
22:09awb_dev has joined #asahi-gpu
22:34awb_dev has quit [Remote host closed the connection]
22:35awb_dev has joined #asahi-gpu
22:35awb_dev has quit []
22:42WindowPain5 has joined #asahi-gpu
22:48WindowPain has quit [Ping timeout: 480 seconds]
22:48WindowPain5 is now known as WindowPain