#
…

2023-04-26

00:16alyssa has joined #asahi-gpu
00:16alyssaapple's compiler implements * 9 with iadd lsl
00:17alyssa* 17 too
00:17alyssaimad for * 33
00:17alyssaimad for * 7
00:17alyssaimad for * -7
00:18alyssaoh, this is interesting, it uses iadd+lsl instead of bfi for small shifts
00:19alyssait does use isub lsl
00:20alyssabut not in lieu of imad
00:20alyssa(and will optimize isub lsl to imad, weirdly)
00:21alyssaup to lsl 4 for a single iadd
00:21alyssathen, weirdly, <<5 is implemented by chaining iadd(lsl 1) with iadd(lsl 4)
00:21alyssai fail to see why that would be a win
00:23alyssaalthough, if there are a lot more SCIB than IC units than it makes sense
00:23alyssa(those are functional units, in the perf counters)
00:23alyssa(different ALUs)
00:26alyssaI haven't done any benchmarking myself, but comparing Philip Turner's notes with dougall's, I guess there are 4 SCIB units per 1 IC unit on Apple M1
00:27alyssawith int32 addition and basic bitwise ops on the SCIB unit, but the variable barrel shifters on the IC unit
00:27alyssameaning from a raw throughput instruction, using 3 adds (with lsl) to save a real shift would still be a win
00:28alyssathat suggests x << 12 would be implemented with chained iadds, but x << 13 is a toss up, but x << 17 would be real bitwise
00:29alyssaIn reality, Apple's compiler chooses to use bfi for 9 and above, but x << 8 is chained iadds
00:29alyssaThat is, it will use 2 iadds to save a bfi, but will not use 3 iadds
00:29alyssaPossibly there are battling latency and i-cache concerns here, this might be a heuristic it has
00:30abd has joined #asahi-gpu
00:30alyssaPerformance of shifts by 10 is not of any importance whatsoever to me. But it's a nice test case for reasoning about the uarch.
00:35alyssaI would like to have cycle count estimates in my shader-db stats, but meh, I don't think we know enough about the uarch to do that accurately yet.
00:37alyssaWith no render targets, but depth/stencil
00:37alyssamsaa 4x: sample count 1, sample stride 1, bytes per 4096, layout 32x16
00:37nsklaus has quit [Ping timeout: 480 seconds]
00:38alyssamsaa 2x: sample count 1, sample stride 1, bytes per 8192
00:38alyssaditto if no msaa
00:45nsklaus has joined #asahi-gpu
00:53nsklaus has quit [Ping timeout: 480 seconds]
01:02nsklaus has joined #asahi-gpu
01:04bluetail has quit [Quit: The Lounge - https://thelounge.chat]
01:10nsklaus has quit [Ping timeout: 480 seconds]
01:18i509vcb has joined #asahi-gpu
01:19nsklaus has joined #asahi-gpu
01:27nsklaus has quit [Ping timeout: 480 seconds]
01:36nsklaus has joined #asahi-gpu
01:44nsklaus has quit [Ping timeout: 480 seconds]
01:51nsklaus has joined #asahi-gpu
02:01nsklaus has quit [Ping timeout: 480 seconds]
02:10nsklaus has joined #asahi-gpu
02:13chadmed_ has joined #asahi-gpu
02:18nsklaus has quit [Ping timeout: 480 seconds]
02:26nsklaus has joined #asahi-gpu
02:34nsklaus has quit [Ping timeout: 480 seconds]
02:42nsklaus has joined #asahi-gpu
02:50nsklaus has quit [Ping timeout: 480 seconds]
02:58nsklaus has joined #asahi-gpu
03:08nsklaus has quit [Ping timeout: 480 seconds]
03:12nsklaus has joined #asahi-gpu
03:21nsklaus has quit [Ping timeout: 480 seconds]
03:26PyroPeter has joined #asahi-gpu
03:28pyropeter3 has quit [Ping timeout: 480 seconds]
03:30nsklaus has joined #asahi-gpu
03:38nsklaus has quit [Ping timeout: 480 seconds]
03:40bluetail has joined #asahi-gpu
03:46nsklaus has joined #asahi-gpu
03:54nsklaus has quit [Ping timeout: 480 seconds]
04:01possiblemeatball has quit [Quit: Quit]
04:02nsklaus has joined #asahi-gpu
04:10nsklaus has quit [Ping timeout: 480 seconds]
04:19nsklaus has joined #asahi-gpu
04:27nsklaus has quit [Ping timeout: 480 seconds]
04:36nsklaus has joined #asahi-gpu
04:44nsklaus has quit [Ping timeout: 480 seconds]
04:52nsklaus has joined #asahi-gpu
04:59abd has quit [Remote host closed the connection]
05:01nsklaus has quit [Ping timeout: 480 seconds]
05:08nsklaus has joined #asahi-gpu
05:17nsklaus has quit [Ping timeout: 480 seconds]
05:25nsklaus has joined #asahi-gpu
05:34nsklaus has quit [Ping timeout: 480 seconds]
05:42nsklaus has joined #asahi-gpu
05:50nsklaus has quit [Ping timeout: 480 seconds]
05:59nsklaus has joined #asahi-gpu
06:07nsklaus has quit [Ping timeout: 480 seconds]
06:16pthariensflame has joined #asahi-gpu
06:16pthariensflame has quit []
06:16nsklaus has joined #asahi-gpu
06:25nsklaus has quit [Ping timeout: 480 seconds]
06:32chadmed_ has quit [Quit: Page closed]
06:36nsklaus has joined #asahi-gpu
06:36MajorBiscuit has quit [Ping timeout: 480 seconds]
06:44nsklaus has quit [Ping timeout: 480 seconds]
06:52nsklaus has joined #asahi-gpu
07:01nsklaus has quit [Ping timeout: 480 seconds]
07:02i509vcb has quit [Quit: Connection closed for inactivity]
07:09nsklaus has joined #asahi-gpu
07:17nsklaus has quit [Ping timeout: 480 seconds]
07:25nsklaus has joined #asahi-gpu
07:33nsklaus has quit [Ping timeout: 480 seconds]
07:44nsklaus has joined #asahi-gpu
07:52nsklaus has quit [Ping timeout: 480 seconds]
08:00nsklaus has joined #asahi-gpu
08:08nsklaus has quit [Ping timeout: 480 seconds]
08:08MajorBiscuit has joined #asahi-gpu
08:17nsklaus has joined #asahi-gpu
08:25nsklaus has quit [Ping timeout: 480 seconds]
08:33nsklaus has joined #asahi-gpu
08:38MajorBiscuit has quit [Ping timeout: 480 seconds]
08:40MajorBiscuit has joined #asahi-gpu
08:40MajorBiscuit has quit []
08:42nsklaus has quit [Ping timeout: 480 seconds]
08:52nsklaus has joined #asahi-gpu
09:01nsklaus has quit [Ping timeout: 480 seconds]
09:01nyilas has joined #asahi-gpu
09:10nsklaus has joined #asahi-gpu
09:18nsklaus has quit [Ping timeout: 480 seconds]
09:26nsklaus has joined #asahi-gpu
09:32Jamie has joined #asahi-gpu
09:32rhysmdnz has joined #asahi-gpu
09:33Jamie is now known as Guest12222
09:34nsklaus has quit [Ping timeout: 480 seconds]
09:37Guest12142 has quit [Ping timeout: 480 seconds]
09:37rhysmdnz1 has quit [Ping timeout: 480 seconds]
09:41nsklaus has joined #asahi-gpu
11:59cylm has joined #asahi-gpu
12:19mkurz has quit [Ping timeout: 480 seconds]
12:24possiblemeatball has joined #asahi-gpu
13:24mkurz has joined #asahi-gpu
13:26alyssalina: dEQP-GLES31.functional.texture.multisample.samples_1.use_texture_depth_2d_array
13:29alyssae4cdbeab814bdc2b468ec2375a74fc961a423213
13:37nyilas has quit [Remote host closed the connection]
13:39nyilas has joined #asahi-gpu
13:40nyilas has quit [Remote host closed the connection]
13:44nyilas has joined #asahi-gpu
13:44nyilas has quit [Remote host closed the connection]
14:02hightower2 has joined #asahi-gpu
14:12mkurz has quit [Ping timeout: 480 seconds]
14:44LinuxM1 has joined #asahi-gpu
14:50cylm has quit [Ping timeout: 480 seconds]
14:56mkurz has joined #asahi-gpu
15:05alyssalina: If you revert bf3027c3916 ("mesa/st: Normalize wrap modes for seamless cubes")
15:06alyssathen a bunch of dEQP-GLES3 cases will fail (listed in the commit message)
15:06alyssaif you can find a control register bit that you can toggle to make them pass it would make me very interesting
15:06alyssainterested
15:06alyssawould not change the level of interesting i am
15:09LinuxM1 has quit [Quit: Leaving]
15:16alyssa(which is to say, about 37% on a good day)
15:48c10l has quit [Quit: Bye o/]
15:56c10l has joined #asahi-gpu
16:06stickytoffee has quit [Read error: Connection reset by peer]
16:08stickytoffee has joined #asahi-gpu
16:49Guest12222 has quit [Ping timeout: 480 seconds]
16:49rhysmdnz has quit [Ping timeout: 480 seconds]
17:38linaalyssa: I don't know if this is a good thing or a bad thing, but almost all the unknowns I exposed seem to do nothing to dEQP2/3... ^^;;
17:38alyssaSounds like a neutral thing ^^
17:39alyssaAlso I already had lunch, get some sleep %_%
17:39linaThe only interesting things I found are that if large_tib and the TVB isn't large enough I get a message and the job hangs (means I need to update the min_tvb_size for that case) and one of the registers has two bits which need to be set or a ton of things fault. I have a feeling it might be related to cache control/flushing?
17:39linaI just ended the stream! I'll have a bit of dinner and sleep ^^
17:39linaI opened a PR with the UAPI changes anyway ^^
17:39alyssa1. I don't know exactly how eMRT interacts with partial renders. I'm trying not to think about eMRT until I've finished up everything else in gles3.1
17:40alyssa2. Shrug
17:40linaSpeaking of, I thought we were passing all of GLES3 but there's still that one test that needs eMRT?
17:40lina(the one with a too large TIB stride)
18:03alyssaeh, yeah, right
18:03alyssathat's in GLES3.0 in your CTS, it's in GLES3.1 in mine because I haven't updated in years :p
18:37possiblemeatball has quit [Quit: Quit]
19:02hightower2 has quit [Ping timeout: 480 seconds]
19:25hightower2 has joined #asahi-gpu
19:55Jamie has joined #asahi-gpu
19:56rhysmdnz has joined #asahi-gpu
19:56Jamie is now known as Guest12262
20:05maximbaz has quit [Quit: bye]
20:07maximbaz has joined #asahi-gpu
20:55A_L_I_C_E has joined #asahi-gpu
21:03A_L_I_C_E has quit [Quit: Quit]
21:16maximbaz has quit [Quit: bye]
21:18maximbaz has joined #asahi-gpu
21:20zocker has quit [Ping timeout: 480 seconds]
21:38zocker has joined #asahi-gpu
23:20cylm has joined #asahi-gpu
23:33cylm_ has joined #asahi-gpu
23:35cylm has quit [Ping timeout: 480 seconds]
…