#
…

2021-04-10

00:02odmir has joined #asahi-gpu
00:05odmir has quit [Remote host closed the connection]
00:05odmir has joined #asahi-gpu
00:30odmir has quit [Remote host closed the connection]
00:31odmir has joined #asahi-gpu
00:47odmir has quit [Ping timeout: 268 seconds]
01:03odmir has joined #asahi-gpu
01:23Emantor has joined #asahi-gpu
01:36odmir has quit [Ping timeout: 240 seconds]
01:51solarkraft has joined #asahi-gpu
02:20odmir has joined #asahi-gpu
02:25odmir has quit [Ping timeout: 240 seconds]
02:51odmir has joined #asahi-gpu
02:57odmir has quit [Remote host closed the connection]
02:58odmir has joined #asahi-gpu
02:59phiologe has joined #asahi-gpu
03:20Necrosporus has quit [Read error: Connection reset by peer]
03:21Necrosporus has joined #asahi-gpu
03:22odmir has quit [Remote host closed the connection]
04:17phiologe has quit [*.net *.split]
04:17Emantor has quit [*.net *.split]
04:17morelightning[m] has quit [*.net *.split]
04:17hypergenesis[m] has quit [*.net *.split]
04:17Bastian[m] has quit [*.net *.split]
04:17mrkajetanp has quit [*.net *.split]
04:17Augur[m] has quit [*.net *.split]
04:17wicast has quit [*.net *.split]
04:17the-mentor3 has quit [*.net *.split]
04:17Necrosporus has quit [*.net *.split]
04:17josiahmendes[m] has quit [*.net *.split]
04:17rockinrobstar[m] has quit [*.net *.split]
04:17bylaws has quit [*.net *.split]
04:17emily has quit [*.net *.split]
04:17Jasper[m] has quit [*.net *.split]
04:18phiologe has joined #asahi-gpu
04:18Emantor has joined #asahi-gpu
04:18Necrosporus has joined #asahi-gpu
04:18the-mentor3 has joined #asahi-gpu
04:18wicast has joined #asahi-gpu
04:18Augur[m] has joined #asahi-gpu
04:18morelightning[m] has joined #asahi-gpu
04:18rockinrobstar[m] has joined #asahi-gpu
04:18hypergenesis[m] has joined #asahi-gpu
04:18josiahmendes[m] has joined #asahi-gpu
04:18bylaws has joined #asahi-gpu
04:18Bastian[m] has joined #asahi-gpu
04:18emily has joined #asahi-gpu
04:18Jasper[m] has joined #asahi-gpu
04:18mrkajetanp has joined #asahi-gpu
04:18stemnic has joined #asahi-gpu
04:18kit_ty_kate has joined #asahi-gpu
04:18tbodt has joined #asahi-gpu
04:18Hakon has joined #asahi-gpu
04:18comex has joined #asahi-gpu
04:18prusnak has joined #asahi-gpu
04:18daniels has joined #asahi-gpu
04:18NekomimiScience has joined #asahi-gpu
04:18GaveUp has joined #asahi-gpu
04:18Lightsword has joined #asahi-gpu
04:18XeR has joined #asahi-gpu
04:18radex1 has joined #asahi-gpu
04:18gruetzkopf has joined #asahi-gpu
04:18ar has joined #asahi-gpu
04:18JTL has joined #asahi-gpu
04:18cyrozap has joined #asahi-gpu
04:18DarkShadow44 has joined #asahi-gpu
04:18larbob has joined #asahi-gpu
04:18dottedmag has joined #asahi-gpu
04:18DrWhax has joined #asahi-gpu
04:18krzywix- has joined #asahi-gpu
04:18tomtastic has joined #asahi-gpu
04:18marcan has joined #asahi-gpu
04:18zkrx has joined #asahi-gpu
04:18bpye has joined #asahi-gpu
04:18mxw39 has joined #asahi-gpu
04:18Z750 has joined #asahi-gpu
04:18tardyp has joined #asahi-gpu
04:18HotSwap has joined #asahi-gpu
04:18Yuzu has joined #asahi-gpu
04:18x56_ has joined #asahi-gpu
04:18DragoonAethis has joined #asahi-gpu
04:18Stary has joined #asahi-gpu
04:18tarzeau_ has joined #asahi-gpu
04:18Simonx22 has joined #asahi-gpu
04:18riatre has joined #asahi-gpu
04:18narmstrong has joined #asahi-gpu
04:18jn__ has joined #asahi-gpu
04:18puhitaku has joined #asahi-gpu
04:18nkaretnikov has joined #asahi-gpu
04:18sbingner has joined #asahi-gpu
04:18linkmauve has joined #asahi-gpu
04:18jkkm has joined #asahi-gpu
04:18Baughn has joined #asahi-gpu
04:18robinp has joined #asahi-gpu
04:18varad has joined #asahi-gpu
04:18zarvox has joined #asahi-gpu
04:18commandoline has joined #asahi-gpu
04:18robher has joined #asahi-gpu
04:18mrkajetanp has quit [Max SendQ exceeded]
04:19winocm has quit [Ping timeout: 245 seconds]
04:19Bastian[m] has quit [Ping timeout: 258 seconds]
04:19josiahmendes[m] has quit [Ping timeout: 244 seconds]
04:19rockinrobstar[m] has quit [Ping timeout: 244 seconds]
04:20emily has quit [Ping timeout: 244 seconds]
04:20davidrysk[m] has quit [Ping timeout: 252 seconds]
04:20meiji163[m] has quit [Ping timeout: 260 seconds]
04:20cepheus has quit [Ping timeout: 245 seconds]
04:20mofux[m] has quit [Ping timeout: 245 seconds]
04:20blazra has quit [Ping timeout: 245 seconds]
04:20uartman[m] has quit [Ping timeout: 245 seconds]
04:20zzoon has quit [Ping timeout: 252 seconds]
04:20Lockna[m] has quit [Ping timeout: 252 seconds]
04:20user1tt[m] has quit [Ping timeout: 252 seconds]
04:20bylaws has quit [Ping timeout: 245 seconds]
04:21brentr123[m] has quit [Ping timeout: 248 seconds]
04:21HeN has quit [Ping timeout: 260 seconds]
04:21konradybcio has quit [Ping timeout: 245 seconds]
04:21Jasper[m] has quit [Ping timeout: 245 seconds]
04:21izzyisles[m] has quit [Ping timeout: 246 seconds]
04:21Augur[m] has quit [Ping timeout: 258 seconds]
04:21simjnd[m] has quit [Ping timeout: 276 seconds]
04:21svenpeter has quit [Ping timeout: 276 seconds]
04:21JJJollyjim has quit [Ping timeout: 276 seconds]
04:21morelightning[m] has quit [Ping timeout: 258 seconds]
04:21hypergenesis[m] has quit [Ping timeout: 258 seconds]
04:21rkjnsn[m] has quit [Ping timeout: 246 seconds]
04:22HeN has joined #asahi-gpu
04:23_alice has quit [Ping timeout: 252 seconds]
04:56svenpeter has joined #asahi-gpu
05:25mrkajetanp has joined #asahi-gpu
05:25meiji163[m] has joined #asahi-gpu
05:25cepheus has joined #asahi-gpu
05:25emily has joined #asahi-gpu
05:25mofux[m] has joined #asahi-gpu
05:25user1tt[m] has joined #asahi-gpu
05:25uartman[m] has joined #asahi-gpu
05:25Bastian[m] has joined #asahi-gpu
05:25rkjnsn[m] has joined #asahi-gpu
05:25Lockna[m] has joined #asahi-gpu
05:25Augur[m] has joined #asahi-gpu
05:25konradybcio has joined #asahi-gpu
05:25izzyisles[m] has joined #asahi-gpu
05:25winocm has joined #asahi-gpu
05:25rockinrobstar[m] has joined #asahi-gpu
05:25Jasper[m] has joined #asahi-gpu
05:25_alice has joined #asahi-gpu
05:25davidrysk[m] has joined #asahi-gpu
05:25bylaws has joined #asahi-gpu
05:25blazra has joined #asahi-gpu
05:25simjnd[m] has joined #asahi-gpu
05:25brentr123[m] has joined #asahi-gpu
05:25JJJollyjim has joined #asahi-gpu
05:25josiahmendes[m] has joined #asahi-gpu
05:25hypergenesis[m] has joined #asahi-gpu
05:25zzoon has joined #asahi-gpu
05:25morelightning[m] has joined #asahi-gpu
05:57vlixa has quit [Remote host closed the connection]
09:00Glanzmann has joined #asahi-gpu
10:31vlixa has joined #asahi-gpu
10:59Glanzmann has quit [Quit: leaving]
11:03glibc has joined #asahi-gpu
11:10glibcDo you have suggestions of "relatively" easy tasks to help the project?
11:58robinpglibc: your probably better off asking in the main channel #asahi. I think at the moment they are just getting through the low level initial support - I'm hoping that there will be more JJ once they get that sorted
12:06jn__glibc: it somewhat depends on your field of interest, too — what kind of work are you particularly interested in?
12:07glibceither GPU reverse engineering, or driver work
12:07jn__i'll wait for bloom or someone else to come around with good ideas :)
13:41vlixa has quit [Remote host closed the connection]
13:58bloomdougall: I have to hand it to you, no clue how you figured out the control flow stuff <3
14:11bloomeven have spilling disassembled correctly, nice!
14:36_whitelogger has joined #asahi-gpu
14:39vlixa has joined #asahi-gpu
15:24odmir has joined #asahi-gpu
15:29odmir has quit [Ping timeout: 260 seconds]
15:50DarkShadow44bloom: Would you mind sharing a shader that uses that? Would be helpful to understand that.
15:55bloomCan you find the pattern in this?
15:56bloomSpecifically, can you write a function that takes in the left number (340) and outputs the right one (6), fitting so well that it can predict the results for future data?
16:00simjnd[m] has quit [Quit: Idle for 30+ days]
16:05glibc@bloom: thank you for your suggestion. I will have a look.
16:13jn__bloom: how much data do you have for the validation of functions?
16:23jn__i have a fitted function and it seems reasonably compact, but it could be overfitted still
16:26DarkShadow44jn__: mind showing?
16:30DarkShadow44hm interesting
16:32bloomjn__: what's f2gen?
16:33jn__it generates the boundaries (0x20, 0x40, etc.) along with the output numbers (2,3,4,..)
16:33bloomit's certainly a possibility :)
16:33bloomIt's the division-by-3 that's baffling me
16:33bloomclearly you are right, that's in the data. but why?
16:35jn__my approximation is more or less a floating point with a mantissa of between one and two bits
16:35jn__or, wait
16:35jn__actually two bits of mantissa?
16:36jn__no, less
16:36jn__else 0x30 would be in it
16:37bloomfwiw, the equivalent field on Mali is https://cgit.freedesktop.org/mesa/mesa/tree/src/panfrost/lib/pan_scratch.c#n70 explanation in the comments
16:37bloomuh. apparently that comment is wrnog, should be ceil(log2(s / 16)) but I digress
16:39bloomceil(log2(ceil(s / 16))) I guess it was
16:39glibcah, the input number is TLS byte size in shader?
16:40bloomOn Mali, the idea is that the hardware allocates a power-of-two number of 16-byte slots
16:40bloomso we need to round up the number of 16-byte slots, and then round up the power-of-two
16:40bloomtable.py shows a clear logarithmic pattern (as jn__ also identified) so this doesn't seem far off from AGX
16:40bloombut the coefficients get all weird for small n
16:41bloom (36, 2),
16:41bloom (20, 1),
16:42bloom^ This should be bugging you
16:42bloomIf it's log2, then why is the there are threshold betweeen 20 and 36?
16:42glibcactually what bothers me is the 372/388 boundary :) 20/36 -> computation could simply be in unit of 32 bytes
16:42bloom(372, 6),
16:42bloom(388, 7)
16:42bloomglibc: yeah
16:43bloomglibc: The "explanation" there is if you divide by 3, that becomes a 124/129 boundary -- fine, 128
16:43bloomBut again... why 3?!
16:43* jn__finds this table much easier to look at in hex
16:44bloom----Oh shit
16:44bloomjn__: ty
16:45bloomit's not about division by 3 at all, gah
16:45jn__bloom: the pattern i fitted is like 1 - 2 3 4 - 8 12 16 -- one big step (double the number), two small steps (increment the two most significant bits), repeat
16:45bloomjn__: Yeah, I think you got it spot on.
16:45bloomsomething like upper bits are powers of 4, lower bits are 2/3
16:59DarkShadow44bloom: Out of curiousity, how does one get those test numbers? I'd be interested in the result of (for example) 800
17:00bloomDarkShadow44: modifying a shader to vary its register pressure, disassembling the shader to get the amount spilled (X), and parsing the command stream to get the amount reported (Y)
17:01glibchow about having a function-local array, dynamically indexed by a uniform/attribute?
17:02bloomworth checking as well, and would probably be easier :)
17:02glibcalright :)
17:06DarkShadow44bloom: Using pandecode_cmdstream after a computer shader?
17:06bloomvert in this case
17:06DarkShadow44doesn't work for computer shader yet?
17:51bloomAnd for my next trick, I shall pull a rabbit out of a hat!
17:54bloom("Go back to sleep, Bloom.")
17:55vijfhoek has quit [*.net *.split]
17:55chrisf has quit [*.net *.split]
17:55Shiz has quit [*.net *.split]
17:55ch3sh1r3c4t has quit [*.net *.split]
17:55anuejn has quit [*.net *.split]
17:55Ziemas has quit [*.net *.split]
17:55rbenua has quit [*.net *.split]
17:55bloom has quit [*.net *.split]
17:56ch3sh1r3c4t has joined #asahi-gpu
17:56Shiz has joined #asahi-gpu
17:56chrisf has joined #asahi-gpu
17:56bloom has joined #asahi-gpu
17:56anuejn has joined #asahi-gpu
17:56Ziemas has joined #asahi-gpu
17:56vijfhoek has joined #asahi-gpu
17:56rbenua has joined #asahi-gpu
18:11odmir has joined #asahi-gpu
18:26DarkShadow44dougall: I found some issue with your documentation
18:26DarkShadow44Would you mind taking a look at this: https://pastebin.com/0f6ey9vu
19:30odmir has quit [Remote host closed the connection]
19:31odmir has joined #asahi-gpu
19:35odmir has quit [Ping timeout: 240 seconds]
20:01odmir has joined #asahi-gpu
20:36odmir has quit [Ping timeout: 265 seconds]
20:51odmir has joined #asahi-gpu
21:23odmir has quit [Ping timeout: 240 seconds]
21:33aratuk has joined #asahi-gpu
21:37aratuk has quit [Client Quit]
22:03mxw39 has quit [Quit: Konversation terminated!]
22:04mxw39 has joined #asahi-gpu
22:15odmir has joined #asahi-gpu
22:40dougallDarkShadow44: good catch, thanks - i'll try to look later today, but i'm guessing 0b00 is the correct encoding there https://github.com/dougallj/applegpu/blob/main/applegpu.py#L859-L860
22:45odmir has quit [Remote host closed the connection]
22:46odmir has joined #asahi-gpu
22:51odmir has quit [Ping timeout: 268 seconds]
23:05odmir has joined #asahi-gpu
23:11* bloomhas a WIP XMLification of applegpu's disassembler
23:13bloomThough tbh I am thinking we'd be happier just pulling applegpu.py into mesa
…