#
…

2022-10-26

00:01chengsun has quit [Quit: Quit]
00:01chengsun has joined #asahi-gpu
00:01Etrien___ has quit [Ping timeout: 480 seconds]
00:09ella-0 has quit [Read error: Connection reset by peer]
00:59sadams0978 has joined #asahi-gpu
01:00sadams0978 has quit [Quit: Konversation terminated!]
01:45Etrien has quit [Read error: Connection reset by peer]
01:46Etrien has joined #asahi-gpu
01:57pthariensflame has joined #asahi-gpu
02:06pthariensflame has quit [Quit: Textual IRC Client: www.textualapp.com]
02:15alyssaIt's really interesting comparing statistics for a given compiler shader *across* instruction sets
02:15alyssahere's a chunky shader from glmark2 -bterrain
02:15alyssashaders/glmark/1-22.shader_test - MESA_SHADER_FRAGMENT shader: 1294 inst, 8600 bytes, 50 halfregs, 1 threads, 0 loops, 0:0 spills:fills
02:16alyssashaders/glmark/1-22.shader_test - MESA_SHADER_FRAGMENT shader: 1249 inst, 16.015625 cycles, 16.015625 fma, 3.453125 cvt, 0.000000 sfu, 0.125000 v, 0.000000 t, 0.000000 ls, 632 quadwords, 1 threads, 0 loops, 0:0 spills:fills
02:16alyssatop is Apple M1, bottom is Mali-G57
02:16alyssaoff the bat you'll notice that the statistics for Mali are a lot more advanced, because we actually understand the uarch there
02:16alyssa(AGX will get similar stuff in time~)
02:16alyssabut here's an obvious one:
02:17alyssaAGX is 8600 bytes
02:17alyssaMali is ~10,000 bytes
02:17alyssa(quadwords = 16 bytes)
02:17alyssaThis is interesting, a Mali-G57 instruction can actually do *more* than an AGX instruction
02:18alyssathere are fewer of them (1249 vs 1294), but they're bigger
02:18alyssaWhy? mali-g57 uses fixed-length instructions, all are 8 bytes, and many are sparse
02:18alyssaAGX uses variable length instructions, in this shader instructions are an average of 6.65 bytes
02:20alyssaSimilarly, register pressure
02:20alyssait's not explicitly printed for Mali-G57, but "1 thread" means that it's using at least 33 dword registers
02:20alyssawhereas AGX is using 50/2 = 25 dword registers
02:20alyssawhy would the same program use so many fewer registers on AGX?
02:21alyssaone answer is "I did a better job at register allocation for AGX than Mali"
02:32chengsun_ has joined #asahi-gpu
02:35chengsun has quit [Ping timeout: 480 seconds]
02:37chengsun_ has quit [Read error: Connection reset by peer]
02:37chengsun has joined #asahi-gpu
02:50chengsun_ has joined #asahi-gpu
02:52chengsun has quit [Ping timeout: 480 seconds]
02:55chengsun_ has quit [Quit: Quit]
02:57chengsun has joined #asahi-gpu
03:13chengsun_ has joined #asahi-gpu
03:16chengsun has quit [Ping timeout: 480 seconds]
03:22chengsun_ has quit [Ping timeout: 480 seconds]
03:58ella-0 has joined #asahi-gpu
04:00Graypup_ has quit [Quit: meow]
04:01Graypup_ has joined #asahi-gpu
04:31bluetail has joined #asahi-gpu
04:55Etrien has quit [Read error: Connection reset by peer]
04:55Etrien has joined #asahi-gpu
05:37SSJ_GZ has joined #asahi-gpu
05:44Etrien has quit [Read error: Connection reset by peer]
05:45Etrien has joined #asahi-gpu
05:55linaNice!! <3
06:12Etrien__ has joined #asahi-gpu
06:17Etrien has quit [Ping timeout: 480 seconds]
06:26Etrien__ has quit [Read error: Connection reset by peer]
06:26Etrien has joined #asahi-gpu
08:05sharonmary6[m] has quit []
08:05MatrixTravelerbot[m]1 has quit []
08:05psydroid[m] has quit []
08:05arisu has quit []
08:05Ella[m] has quit []
08:05Lucy[m] has quit [Quit: Bridge terminating on SIGTERM]
08:05Soroush has quit []
08:06Dcow has joined #asahi-gpu
08:11capta1nt0ad has joined #asahi-gpu
08:44subatomic has joined #asahi-gpu
08:46Dcow_ has joined #asahi-gpu
08:53Dcow has quit [Ping timeout: 480 seconds]
08:57subatomic has quit [Quit: Textual IRC Client: www.textualapp.com]
09:00capta1nt0ad has quit [Quit: Konversation terminated!]
09:01chengsun has joined #asahi-gpu
09:03geochip has joined #asahi-gpu
09:48geochip has quit [Quit: leaving]
09:59chadmed has joined #asahi-gpu
10:31chadmed has quit [Quit: Konversation terminated!]
10:32chadmed has joined #asahi-gpu
10:34chadmed has quit []
10:40kov has quit [Quit: Coyote finally caught me]
10:40kov has joined #asahi-gpu
11:00chadmed has joined #asahi-gpu
11:48chadmed has quit [Read error: No route to host]
12:07chadmed has joined #asahi-gpu
13:23Gaspare has joined #asahi-gpu
13:37chadmed has quit [Read error: No route to host]
13:40n1c has quit [Quit: ZNC 1.8.2+deb1+focal2 - https://znc.in]
13:41n1c has joined #asahi-gpu
13:46Gaspare has quit [Read error: Connection reset by peer]
15:05linaSo the M1 Ultra just needed a bunch of new buffers and some larger bigger ones, initdata changes, and a couple constants changed, and then it worked.
15:05linaThere's a Z acceleration/hierarchical Z thing that showed up, and then 5 new buffers adjacent to the tiler?
15:06linaI get the feeling they're actually trying to balance work within single jobs between dies, which would explain why they need some new buffers to transfer things around.
15:07linaParallelizing fragment processing is trivial, but fragment is not, since it interacts with the tiler/sorting stuff
15:09alyssalina: as I texted you, the "Z acceleration" buffer is probably just from depth compression, which Apple's driver is aggressive about enabling and took me a lot of time to figure out how to disable
15:09alyssait does visually look like hier-z but I'm not convinced it actually is
15:14linaalyssa: I've never seen it enabled on the M1 Mini, ever. I have zero hits for those pointers in all my historical hypervisor logs. But on this one, it showed up, and mesa was faulting without it... so I'm not sure you disabled it ^^
15:20alyssaVery curious
15:20alyssaI'd love to see the Mesa patch if you've pushed
15:20alyssa(the diff from t8103 mesa I mean)
15:21linaLet me do that!
15:22linaThe size is just random though, haven't worked out how to calculate any of it yet.
15:29linaThe stencil one is just a guess though, haven't actually seen it yet
15:48linaLooks like it's 1/32 compression and it uses POT addressing, so align zbuffer size to POT and divide by 32 for the accel buffer size
15:48linaIt seems every 8x4 block of Z pixels maps to one accel buffer byte, 0x03 means clear.
15:49linaStill need to look at the deflake buffer sizes... 2 of them are obvious due to adjacency, but do you remember how you figured out the third bound?
16:45alyssalina: guess
16:45alyssaor maybe not
16:45alyssano, adjacency as well
16:45alyssafinding the next allocation in the same BO and using that as an upper bound
17:24Gaspare has joined #asahi-gpu
18:01Gaspare has quit [Ping timeout: 480 seconds]
18:31Dcow_ has quit [Remote host closed the connection]
18:31Dcow has joined #asahi-gpu
18:40Dcow has quit [Ping timeout: 480 seconds]
19:11Dcow has joined #asahi-gpu
19:55yuyichao_ has quit [Remote host closed the connection]
19:55yuyichao_ has joined #asahi-gpu
20:27rwhitby has joined #asahi-gpu
20:41Dcow has quit [Remote host closed the connection]
20:42Dcow has joined #asahi-gpu
20:43Dcow has quit [Remote host closed the connection]
20:43Dcow has joined #asahi-gpu
21:05rwhitby has quit [Quit: rwhitby]
21:15Etrien__ has joined #asahi-gpu
21:18phireI suspect the accleration buffer stuff is just because whatever alyssa did to force a linear buffer, has regressed for some reason
21:20Etrien has quit [Ping timeout: 480 seconds]
21:30SSJ_GZ has quit [Read error: No route to host]
21:31Etrien__ has quit [Read error: Connection reset by peer]
21:31Etrien has joined #asahi-gpu
21:57alyssas/linear/uncompressed/, IIRC it's stil twiddled
21:57alyssaMetal really doesn't like rendering to linear
21:59alyssaIt's probably not a *big* deal to support properly in mesa but that patch is not the way to do it
22:06phireso it's just compressed vs uncompressed?
22:16alyssaYes, I think so
23:30jannauhah, dcp supports XRGB afterall, just not as separate pixelformat but via a flag in dcp_surface (either unk1 or unk2)
23:32jannaubut it took me far too long to realize that fbcon displayed just a transparent terminal
23:47phireclassic bug
23:56alyssajannau: Woof
23:56alyssaHow'd you discover the flag?
…