00:00 karolherbst: imirkin: ha, no, parsing json in php is a big pain
00:00 karolherbst: maybe there are good libraries though
00:00 karolherbst: but you always use some crazy functions
00:00 karolherbst: ohh wait
00:00 karolherbst: json_decode and then I should have a object with all that crap
00:06 imirkin: right.
00:40 karolherbst: nearly done :D
00:40 karolherbst: well php seems to be good enough for it
00:40 karolherbst: and it can also be executed locally without having a webserver, so this might be good
00:49 karolherbst: imirkin: https://github.com/karolherbst/SteamDevGameOverviewer :)
00:51 imirkin: is it up anywhere?
00:52 karolherbst: no
00:52 karolherbst: but wait a second
00:54 karolherbst: https://htmlpreview.github.io/?https://gist.githubusercontent.com/karolherbst/b297487421b213ac15d95181a09157e9/raw/052d4eb4959f4b084aa5e6c9f2a618106da360bb/test.html
00:54 karolherbst: ohh somewhere below it messed up
00:55 karolherbst: sadly the steam api doesn't give me the name of the game directly :O
00:56 karolherbst: ahh kate was stupid
00:57 imirkin: yes... it's the tool's fault...
00:59 karolherbst: https://htmlpreview.github.io/?https://gist.githubusercontent.com/karolherbst/b297487421b213ac15d95181a09157e9/raw/051262686a957fc90ce3e0ee9f8a806ac4c4de12/test.html
00:59 karolherbst: well we just need an awesome CSS style, some search boxes and filters :)
01:06 karolherbst: ohh &include_appinfo=1
01:06 karolherbst: hehe
01:22 Celti: karolherbst: Yes, I used bumblebee on here for a while, but it had a weird issue where it would stop being able to bring up the discrete card after suspend.
01:25 martm: when we landed to a program, that made me to read again, and i understand that world has pretty many programmers to be honest, and most of the 50tools i inspected will show right away where the problem is in the library without getting into deadlock
01:26 martm: if that is a racy condition somewhere, most tools will exactly show where it is
01:28 martm: pretty much shocked in a good way, how much information is about this on the net, and how many solutions
01:33 karolherbst: argh, is this web development community really so aweful or why do I find just stupid stuff when searching for "filter dataset with CSS and HTML" ...
02:54 karolherbst: imirkin: hah! done
08:44 martm: http://www.cse.psu.edu/~gxt29/paper/mcfi.pdf
10:17 martm: i won't have time today, as it took me alot to filter those compiler based automatic solutions to thread-safety, i think there should not be many worries
10:18 martm: if i get some time i will try to fix bugs mupuf: i was not counting that nouveau's cpu code is occationally thread unsafe, that is not a whole lot of big problem in the list, dx part is large project too
10:38 RSpliet: karolherbst: you had the INA3221 running right?
10:39 RSpliet: I've just booted the GTX780, but I'm a bit scared of making it do work with my 500W PSU :-P
11:22 bozhan: kraolherbst
11:41 karolherbst: who has by the way a Steam library with some games? I would like to have the Steamid for those. I wrote a tool which checks which game everybody has and lists it, so that we can easily know who owns what
11:43 Celti: I'm Celti on Steam. ^^
11:45 karolherbst: I would prefer if nouveau devs would tell me, but that might work too
11:49 martm: karolherbst: i delt with thread-safety issues, but i have bookmarks in the laptop that my mom took with her, i think i am capable of fixing it + doing the scheduler
12:10 pmoreau: karolherbst: Oh, the number are game ids! I thought you were trying to establish a game score ranking among devs! :-D
12:12 karolherbst: :D
12:12 karolherbst: noooo
12:12 pmoreau: karolherbst: I'm kurtelborg on Steam.
12:12 pmoreau: But my profile seems to be private…
12:13 karolherbst: doesn't matter
12:13 karolherbst: ohh wait
12:13 karolherbst: I don't get your games :D
12:14 pmoreau: Right, I'll try to make my profile public
12:15 RSpliet: karolherbst: got the GTX780 running at "full speed" ;-)
12:15 RSpliet: core clock reads 908MHz, which seems to correspond with the clock in the marketing materials
12:15 karolherbst: RSpliet: awesome :)
12:15 RSpliet: (not the boost speed... but meh)
12:15 karolherbst: RSpliet: but you use my branch right?
12:16 RSpliet: a week-and-a-half old check-out of it
12:16 karolherbst: there should be a boost file, and the 0 and 1 value should correspond with the marketing stuff
12:16 RSpliet: yeah, my branch might be older, but I did boot with config=NvBoost=2
12:16 karolherbst: isn't there a boost file in debugfs?
12:16 karolherbst: I don't think this interface is that old :/
12:17 karolherbst: meh... php complaints in foreach if the array is empty? the heck
12:18 RSpliet: oh week-and-a-half old was an arsepull, didn't verify how old my check-out was
12:18 karolherbst: ohh no, it is fine
12:18 RSpliet: March 15th, so more like 4 weeks
12:18 karolherbst: :D
12:19 karolherbst: pmoreau: ok, yeah so I don't get your games :D
12:19 RSpliet: time flies like an arrow...
12:21 pmoreau: karolherbst: Try again ;-)
12:21 karolherbst: pmoreau: you changed it while I was writing my fallback code for privates :D
12:21 karolherbst: but yeah, it works now
12:21 pmoreau: :-D
12:22 karolherbst: wanna make it private again for a moment?
12:22 pmoreau: Sure
12:22 pmoreau: Done
12:23 karolherbst: mhh, I still get your games
12:23 pmoreau: (Though it wasn't in "private" mode, it was in "friends only" mode :-))
12:23 karolherbst: maybe it takes some time
12:23 pmoreau: Try again?
12:24 karolherbst: k, worked
12:24 karolherbst: I just have a list with all private profiles at the begining
12:28 pmoreau: karolherbst: Ping me when I should revert to public.
12:29 karolherbst: well you can revert it again if you want :) I am done with the private stuff
12:29 karolherbst: currently I improve the search field
12:29 karolherbst: because now you can only search by game name, but I want to allow multiple data fields
12:31 pmoreau: Done
12:32 pmoreau: Do you have a link to the webpage?
12:32 karolherbst: not yet
12:32 karolherbst: I don't even develop on a web server
12:32 karolherbst: just php cli call
12:32 pmoreau: Ok
12:39 karolherbst: pmoreau: but the stuff is here, you would just need an API key and the account ids :D https://github.com/karolherbst/SteamDevGameOverviewer
12:41 karolherbst: I can give you my config.php file if you want then
12:42 karolherbst: RSpliet: well my current branch just improves stuff generally, so there is no point in testing, except for finding regressions
12:43 RSpliet: karolherbst: yeah, and I currently don't have a lot of time for regression tests :-P
12:43 RSpliet: but I'm very glad to see stuff works so well in general
12:43 karolherbst: RSpliet: but I don't have the vbios of your nvf0 :D
12:43 RSpliet: karolherbst: I'll send it to you tomorrow
12:43 karolherbst: k
12:43 karolherbst: thanks
12:43 RSpliet: currently running a little experiment
12:43 RSpliet: it has an INA3221
12:43 karolherbst: right
12:43 karolherbst: it should have
12:43 karolherbst: :D
12:43 RSpliet: header-generated GPIO voltages
12:43 RSpliet: memory at 6008MHz
12:44 karolherbst: I was more interested in the CSTEP and base clocks
12:44 RSpliet: oh yes
12:44 RSpliet: I don't recall those by heart, give me until tomorrow :-P
13:23 RSpliet: hmm, very strange.. the per/GPC context size appears to be {42KiB, 36KiB, 53KiB, 53KiB, 42KiB}
13:24 RSpliet: I could look into that, as it hints at the mapping being {2, 1, 3, 3, 2} SMs, totalling to 11 instead of 12 SMXs
13:25 karolherbst: on the 780?
13:25 RSpliet: yes
13:26 RSpliet: it's only a hint, as there might just be a weird TPC mapping instead
13:26 karolherbst: ohh wait
13:26 karolherbst: mhh no, doesn't make sense
13:31 karolherbst: RSpliet: ahh right, you should know that. Where do I get the total GPC count?
13:31 karolherbst: nvatiming reads nva_rd32(cnum, 0x121c78), but this only returns garbage
13:31 karolherbst: and nva_rd32(cnum, 0x121c74) for the parts
13:31 RSpliet: 0x22430
13:32 karolherbst: ahh good
13:32 karolherbst: and what is this "PART" thing?
13:33 RSpliet: idk
13:33 RSpliet: https://github.com/RSpliet/envytools/blob/2b148c61189185bd74c46b569601f1e33acf8c83/nva/nvactxctl.c
13:34 karolherbst: 22438 :)
13:34 karolherbst: "GPC[0] set 0: 50631776 Hz" :)
13:36 mr: Hi.... Iam testing the latest 4.6rc4 kernel and the nouveau kernel module. I installed the latest /lib/firmwares from git. After rebooting, sceen goes black, monitor goes to sleep.
13:36 mr: KernelMessages:
13:36 mr: [ 2.587816] nouveau 0000:01:00.0: fifo: read fault at 00ffba0000 engine 1f [] client 12 [PMU] reason 0d [REGION_VIOLATION] on channel -1 [0000000000 unknown]
13:36 mr: [ 2.687724] nouveau 0000:01:00.0: timeout at drivers/gpu/drm/nouveau/nvkm/subdev/secboot/base.c:145/nvkm_secboot_falcon_run()!
13:36 mr: [ 2.687736] nouveau 0000:01:00.0: fifo: write fault at 000030d000 engine 05 [BAR3] client 08 [HOST_CPU_NB] reason 0d [REGION_VIOLATION] on channel -1 [00ffbf5000 unknown]
13:36 mr: [ 2.687752] nouveau 0000:01:00.0: fifo: write fault at 0000011000 engine 05 [BAR3] client 08 [HOST_CPU_NB] reason 0d [REGION_VIOLATION] on channel -1 [00ffbf5000 unknown]
13:37 mr: Any solution ?
13:38 imirkin: mr: that's a GM20x right?
13:38 mr: yes
13:38 imirkin: that'll probably be gnurou's domain... dunno if he's still around
13:39 imirkin: seems like the secure boot stuff doesn't come up properly =/
13:39 RSpliet: nope, the TPCs are mapped 2,2,3,3,2 over the GPCs
13:39 karolherbst: RSpliet: wanna see if this nvatiming commits makes it possible to read out the GPC clocks on your gpu too?
13:40 RSpliet: karolherbst: maybe later,thanks
13:40 mr: Here is the complete dmesg: http://pastebin.com/pWD1wuTL
13:41 pmoreau: mr: Is it a laptop?
13:41 mr: no, asus z97 mainboard
13:42 imirkin: mr: file a bug on bugs.freedesktop.org xorg -> Driver/nouveau, and hopefully gnurou will be able to have a look
13:43 RSpliet: mwk: how should I read MP_IN_GPC_NUMBER_MAPn ?
13:43 pmoreau: You can probably link it to https://bugs.freedesktop.org/show_bug.cgi?id=94725
13:44 pmoreau: (Still opening a new bug report, but add a link to the other bug)
13:45 imirkin: please don't
13:45 imirkin: that will just confuse matters
13:48 karolherbst: mhhh, this explains why nvatiming can't check the pcopy speed for me: 00104000: deadbeef
13:48 karolherbst: ...
13:50 pmoreau: mhh, the other one is an optimus setup, right
13:52 pmoreau: Oh! I read resuming instead of rebooting… --"
13:52 mwk: RSpliet: I don't recall that register at all, I'm afraid
13:52 mr: Bug reportet : https://bugs.freedesktop.org/show_bug.cgi?id=94990
13:52 mr: Hopefully someone can help... :-)
13:53 RSpliet: mwk: that's a shame. do you happen to know how to find the SMX -> GPC mapping?
13:54 mwk: no
13:54 karolherbst: nobody really touched nvatiming in a long time as it seems
13:54 karolherbst: any thing PDISPLAY related changed since nvd9?
14:03 karolherbst: mhhh
14:03 karolherbst: on nouveau the video engines are a bit slower
14:04 karolherbst: but I guess they are scaled down through cstates too on nouveau
14:04 karolherbst: maybe not
14:04 karolherbst: ahh yes, they are
14:04 karolherbst: but they shouldn't
14:14 gnurou: mr: thanks - I will definitely want to look at this one
14:14 mr: gnurou, thanx :-)
14:21 RSpliet: gnurou: do you happen to have any idea where to find the smx -> gpc mapping?
14:22 RSpliet: Tegra K doesn't have more than one SMX iirc, so I can't blame you if you don't :-P
14:22 gnurou: RSpliet: mmm please don't blame me >_<
14:22 RSpliet: hahaha I'm not :-D
14:22 gnurou: sadly I do not have this information at hand
14:23 RSpliet: darn! I'll search on :-)
14:24 karolherbst: gnurou: could you try to ping somebody about this? https://github.com/Gnurou/nouveau/issues/13
14:24 karolherbst: gnurou: even a that's 99.5% close is good enough :D
14:24 gnurou: karolherbst: ok - I think I know who I need to ping, which help here
14:24 karolherbst: :) awesome thanks
14:25 karolherbst: the last comment contains the current coefficients I use
14:25 karolherbst: and if he doesn't know what min/max is, that they are the first two 32bit values in the table entries
14:25 karolherbst: we just call it min/max, no idea how you call it
14:29 imirkin: karolherbst: on the off chance you're not aware of this
14:29 imirkin: 2^24 = 16777216
14:29 karolherbst: yeah I know
14:30 karolherbst: hex(1677721) = 0x199999 ..
14:30 imirkin: well the main point is that you have stuff like 16.7772163 in your equation
14:30 karolherbst: this value was kind of leaked by a stupid manufacturer using the mode1 table for static voltages....
14:30 karolherbst: so I am pretty sure it is kind of right
14:31 karolherbst: imirkin: yeah well, 16777216 is too low though
14:31 imirkin: ok.
14:31 karolherbst: ohh or maybe not?
14:31 karolherbst: ohh wait
14:31 karolherbst: you mean something slightly different
14:32 imirkin: it could be e.g. c0 * 1000000 / (1 << 24) or something
14:32 karolherbst: yeah, let me try that
14:33 imirkin: (i might have missed the number of 0's)
14:33 imirkin: anyways, that seems way more likely of a coefficient than some random floating-point value
14:39 karolherbst: right
14:54 RSpliet: and then there's probably some Hz/KHz/MHz rounding magic?
14:55 andi-: the nouveau git repo linked on freedesktop.org seems to be slightly out of date (the latest stuff I can find is about 5 months old). Did it move?
14:55 pmoreau: Oh, there is a SPIR-V 1.1 in the works, with OpenCL-C++ support apparently… :-)
14:55 andi-: (kernel driver that is)
14:55 karolherbst: andi-: yeah, githubg
14:55 pmoreau: https://github.com/skeggsb/nouveau
14:55 andi-: karolherbst: thanks
14:57 karolherbst: imirkin: yeah, this seems to work much better :)
14:58 karolherbst: (c0 * 1000000) / (2 << 23)
14:58 imirkin: karolherbst: aka 1 << 24
14:58 imirkin: karolherbst: aka >> 24
14:58 pmoreau: Wut, and OpenCL 2.2 provisional spec released?! O_O
14:58 karolherbst: ohh right
14:58 imirkin: aka no division
14:58 RSpliet: pmoreau: cool, what's the novelties?
14:58 imirkin: (just make sure to use 64-bit things)
14:59 imirkin: dump SPIR, take in SPIR-V, let's hope
15:00 pmoreau: RSpliet: Absolutely no idea! I just saw someone pushed a spirv-1.1 branch to the KhronosGroup/SPIR repo (https://github.com/KhronosGroup/SPIR/commit/dbc9604552d8e37eb9e88341572a3eded4662740), and that same person pushed the first version of libclcxx
15:00 pmoreau: RSpliet: Which prompted me to have a look at khronos.org to see news about SPIR-V 1.1, and then I saw OpenCL 2.2 :-)
15:01 RSpliet: their announcement is mostly C++ and pipe storage types
15:01 karolherbst: imirkin: how no division?
15:02 RSpliet: constructors/destructors... cool if I want to port that malloc lib to CL 2.2 :-P
15:02 pmoreau: karolherbst: If you have (c0 * 1000000) >> 24
15:03 karolherbst: ahh right
15:07 RSpliet: karolherbst: wait, is that one million or ten million?
15:08 karolherbst: 10^6
15:08 RSpliet: ah, they have a GDC of 64
15:08 imirkin: 16
15:09 imirkin: errr
15:09 imirkin: dunno what GDC is :)
15:09 karolherbst: https://github.com/Gnurou/nouveau/issues/13#issuecomment-211079164 :)
15:09 RSpliet: greatest common divisor
15:09 imirkin: that's more frequently spelled GCD
15:09 karolherbst: RSpliet: :D
15:10 karolherbst: k, then (c0 * 15625) >> 18
15:10 RSpliet: exactly
15:11 karolherbst: uhh
15:11 karolherbst: that is something 5^
15:11 karolherbst: ahh close
15:11 karolherbst: no
15:11 karolherbst: 5^6
15:11 imirkin: 625 = 5^4
15:11 imirkin: nice
15:11 karolherbst: yeah
15:11 karolherbst: well
15:12 karolherbst: it makes sense
15:12 karolherbst: because we just pulled out 2^6
15:12 karolherbst: :D
15:12 RSpliet: anyway, if c0 doesn't exceed ~274000, you can get by with 32-bit arithmetic now :-P
15:14 karolherbst: it does
15:14 karolherbst: and that by a far amount
15:14 karolherbst: 0x119999a
15:14 karolherbst: for example
15:14 RSpliet: heh, yes
15:15 karolherbst: yeah
15:15 imirkin: 64-bit arith is cheap on 64-bit cpu's
15:15 karolherbst: yeah well, on 32bit you don't have a kepler usually anyway
15:15 imirkin: exactly.
15:15 RSpliet: except for the Tegra K1, but that thing is special
15:15 karolherbst: right
15:15 karolherbst: it uses it's own code
15:15 imirkin: and even there, shifting bits, not such a big deal
15:15 imirkin: or 32*32 -> 64
15:16 RSpliet: and not critical path and everything
15:16 karolherbst: ((s64)info.arg[3] * volt->speedo * temp * 15625) >> 18;
15:16 karolherbst: :)
15:16 karolherbst: speedo is in a range of 0x600-0x800
15:16 RSpliet: hmm... did they use any other "prime ^ 6" numbers? :-P
15:17 karolherbst: well there are some numbers left
15:17 karolherbst: 15.472735253
15:17 karolherbst: 68738.15231
15:17 karolherbst: 47015102.518084794
15:19 karolherbst: but with the new formular: https://gist.github.com/karolherbst/d18f3c511e66d8cffd6f09cbcc62edc6
15:20 karolherbst: all mode 1 entries with min==max
15:20 karolherbst: the volt= thing is calculated according to (c0 * 15625) >> 18
15:20 karolherbst: seems to be right somewhat
15:20 imirkin: probably not a big change from what you had before
15:20 karolherbst: no
15:20 karolherbst: something around 1%
15:20 karolherbst: ohh wait
15:21 karolherbst: yeah before I had only some values -1
15:21 karolherbst: but before I found this vbios, I had 1% lower values
15:22 imirkin: 1 << 36 == 68719476736
15:23 imirkin: which is sorta close to your 68738.15231 number
15:25 imirkin: karolherbst: could your temperature factor be wrong? did you ever consider it might be (T + x) * factor?
15:25 karolherbst: it could be
15:28 karolherbst: imirkin: but then "c3 * S * T * 5^6 >> 18" wouldn't work anymore
15:28 karolherbst: and I REed that with T=1
15:28 karolherbst: so any value added to T should have a huge impact
15:29 karolherbst: but it behaved like the current formular anyway
15:32 karolherbst: RSpliet: what about 1000000 and 1<<36 ?
15:32 karolherbst: ohh 2...
15:33 karolherbst: and more 2s..
15:34 karolherbst: 16125 >> 30
15:34 karolherbst: ohh wait, I did something wrong
15:34 karolherbst: and tehre we are again :D
15:35 karolherbst: 15625 >> 30
15:35 karolherbst: (c4 * S^2 * 5^6) >> 30
15:36 pmoreau: Eh eh
15:36 karolherbst: but this value could get really huge
15:36 imirkin: that seems much nicer.
15:37 karolherbst: 29514712500000
15:37 imirkin: 64-bit numbers can take it.
15:37 karolherbst: yeah
15:37 karolherbst: but that was with c4 being 546
15:37 imirkin: 0..18446744073709551615
15:37 karolherbst: but it should be fine
15:39 imirkin: anyways, nice round numbers like >> 30 make me much happier than "multiply by random constant"
15:39 imirkin: (or worse, divide)
15:40 karolherbst: maybe the other thing is 5^6 >> 10
15:40 karolherbst: (543 * 5 * 15.472735253) : 42008
15:41 karolherbst: (543 * 5 * 5**6) >> 10 : 41427
15:42 imirkin: maybe.
15:42 imirkin: why 543?
15:42 karolherbst: random value for c2
15:42 imirkin: oh ok
15:42 karolherbst: and 5 is random value for T
15:42 imirkin: anyways, i recommend playing around with it
15:42 karolherbst: now the last value
15:43 imirkin: that's a 1% diff btw
15:43 karolherbst: yeah
15:43 karolherbst: this is the usual error range I am in currently
15:44 karolherbst: mhh no idea for the last one
15:45 karolherbst: maybe the voltage calculation behaves better now
15:45 karolherbst: the last part isn't used on my gpu anyway
15:48 karolherbst: yeah
15:48 karolherbst: that looks much better
15:48 karolherbst: really good
15:50 karolherbst: https://gist.github.com/karolherbst/3cd4fc1301945787f1c0c4e1b6466da4
15:50 karolherbst: now I remove my +0.1% voltage hack
15:51 karolherbst: \o/
15:52 karolherbst: mupuf: without the hack, nouveau calculates a higher voltage everytime nvidia downvolts ms after
15:52 karolherbst: like here: https://gist.github.com/karolherbst/3cd4fc1301945787f1c0c4e1b6466da4#file-gistfile1-txt-L26-L27
15:53 karolherbst: imirkin, RSpliet: thanks a lot, I think we are much closer now :)
15:53 karolherbst: and this really looks right
15:53 karolherbst: ohh it volts up in this case, because I droped the tempreature...
15:54 karolherbst: yeah, besdies c6, I think we've got it now
15:54 imirkin: cool
15:55 karolherbst: gnurou: I've updated the last comment, now everything should be like on your side :D except the c5 thing
15:57 imirkin: karolherbst: wtf is 47015102.518084794 ? need to convert that too...
15:57 karolherbst: yeah
15:57 karolherbst: I know
15:57 karolherbst: that's why I said except c5
15:58 imirkin: could that one be off by a lot?
15:58 karolherbst: but this isn't used on my GPU and I didn't find anything good yet
15:58 karolherbst: imirkin: could be
15:58 karolherbst: I only know that the T^2 is right
15:58 karolherbst: maybe it is S^3
15:58 karolherbst: or S
15:59 imirkin: could it be 1<<42? 4398046511104
15:59 karolherbst: mhh
15:59 karolherbst: so without the 5^6
15:59 imirkin: not randomly picked... note that 30 - 18 = 12.
15:59 karolherbst: yeah coult be
15:59 imirkin: oh right... hm.
15:59 karolherbst: I always thought it might be a mix out of the both other ones
15:59 imirkin: there's an extra 6 in there
15:59 imirkin: 48 281474976710656
15:59 imirkin: not quite =/
16:01 imirkin: hm. so S = 24, S^2 = 36
16:01 imirkin: T = 16, S * T = 24 as well =/
16:02 imirkin: so i don't think there's a hard pattern...
16:02 karolherbst: yeahm that doesn't help much
16:02 karolherbst: look ak c0, c1 and c3
16:02 karolherbst: ohh
16:02 karolherbst: c4
16:03 imirkin: imho 1<<42 is the most likely candidate.
16:03 imirkin: dunno tho
16:04 karolherbst: yeah, no clue either
16:35 karolherbst: imirkin: any idea about 28085.485375199998 ?
16:36 imirkin: 48 281474976710656
16:36 imirkin: 38 274877906944
16:36 imirkin: karolherbst: in a python shell: for i in range(64): print i, 1<<i
16:37 imirkin: 58 288230376151711744
16:37 karolherbst: mhh
16:38 imirkin: and if you're WAY off, 28 268435456
16:39 karolherbst: I like 38 most, let's check
16:40 karolherbst: 78125 mhh
16:40 karolherbst: >> 31
16:40 karolherbst: 5^7 >> 31
16:40 karolherbst: not _that_ bad
16:42 karolherbst: 48 looks bad
16:42 karolherbst: or maybe not
16:42 karolherbst: 5^10 >> 38
16:43 karolherbst: maybe 28 is right
16:44 karolherbst: 5^4 >> 24
16:45 karolherbst: maybe it is (c5 * S * T^2 *5^6) >> 29 in the end
16:45 karolherbst: but I would need a gpu to test this on
16:46 karolherbst: hakzsam: which gpus are in reator currently?
16:46 hakzsam: gk208 and gfXXX
16:47 hakzsam: you are lucky because I need to RE some bits of MADSP before testing images on gk208, so have fun :)
16:47 hakzsam: will probably need reator later today
16:48 karolherbst: mhhh
16:48 karolherbst: no, I would have to fake the vbios on the gk208 too
16:48 karolherbst: then I can also do it on my gpu
16:49 hakzsam: okay
16:50 mupuf: that looks indeed much nicer now
16:52 karolherbst: yeah
16:52 karolherbst: with the old values the error wasn't constant
16:52 karolherbst: but no I have no error at all :D
16:52 karolherbst: *now
16:52 karolherbst: and my pwm voltage step is 6.25mV
16:53 karolherbst: and I have never a error bigger than this
16:53 karolherbst: *an
16:53 karolherbst: only the last one is bothering me now.. c5 * S^2 * T^2 / 47015102.518084794
16:53 karolherbst: but I guess I spend some time on this today then
16:54 karolherbst: maybe this isn't speedo affected at all
16:54 imirkin: or maybe one of those values isn't ^2?
16:54 karolherbst: mhh wait, it should be
16:54 karolherbst: yeah
16:54 karolherbst: imirkin: 28085.485375199998
16:54 karolherbst: that would be this value then ,)
16:54 imirkin: try sticking values of diff magnitudes and see what happens
16:55 karolherbst: ohhhh
16:55 karolherbst: :O
16:55 karolherbst: (c5 * T^2 * 5^6) >> 18
16:55 karolherbst: :O
16:55 imirkin: that seems easier.
16:55 karolherbst: https://github.com/Gnurou/nouveau/issues/13#issuecomment-211079164
16:56 karolherbst: looks really nice now
16:56 karolherbst: and makes kind of sense
16:56 imirkin: that way you have a full binomial expansion of (S + T)^2
16:56 karolherbst: yeah
16:56 karolherbst: and the offset makes sense
16:56 karolherbst: kind of
16:56 imirkin: 2-variable polynomial
16:56 imirkin: wtvr
16:56 imirkin: 2-variable 2nd order poly
16:56 karolherbst: S: +12
16:57 karolherbst: mhh
16:57 karolherbst: no
16:57 karolherbst: doesn't amtter
16:57 imirkin: anyways, i def like these equations a LOT better :)
16:57 karolherbst: yeah
16:57 karolherbst: me too
16:59 karolherbst: I push that into envytools then too
17:01 karolherbst: and even if that's not quite right, we get close enough
17:04 karolherbst: gnurou: okay, so I need a "this is right" or "this isn't right" now :) I hope it is the former
17:05 karolherbst: imirkin: ohh maybe it will make more sense if I pull the 5^6 out :) let's see
17:07 karolherbst: huh
17:07 karolherbst: ohh yeah, no, it is fine
17:08 karolherbst: well no clue about why it is 10,18 and 30 and no idea why 5^6, but maybe that's just the way it is
17:08 imirkin: well, micro is a convenient unit
17:09 imirkin: hence 10^6
17:09 imirkin: and you just broke it up as 5^6 * 2^6
17:09 imirkin: and removed 6 from the divisor's power
17:14 karolherbst: ahh right
17:16 karolherbst: but that still doesn't tell us why it is 16, 24, 36 then...
17:27 andi-: I've a new notebook (t460p) with a GM108M, i've seen that it is WIP, is there something I (as a non GPU aware software delveoper) can do to help ?
17:31 karolherbst: I think mmiotraces are good
17:33 imirkin: we already have traces
17:33 imirkin: someone just needs to sit down and go over the golden values we use in nouveau and compare them to what's in the trace
17:33 imirkin: unfortunately this is labor-intensive and not straight-forward
17:33 imirkin: since the nouveau tables are splattered over 75 differnet files in 75 different little mini-tables
17:33 andi-: :/
17:33 imirkin: in a imho misguided attempt to "share" the init across chips
17:34 imirkin: andi-: however in the meanwhile, most people have had no issues using the GM107 golden register values, so it's an easy patch to make the GM108 appear as a GM107
17:35 imirkin: andi-: that said if what you're looking for is performance, you're better off sticking to the blob driver for now, as reclocking is not enabled for the GM10x's
17:36 andi-: Not looking for performance just trying to get the setup "completed" :-)
17:36 andi-: Currenlty i'm just using the intel driver which proibably is enough for most of what I do day2day
17:37 imirkin: andi-: https://bugs.freedesktop.org/show_bug.cgi?id=89558
17:37 imirkin: has instructions for how to get it going
17:38 imirkin: nouveau should be able to suspend your gpu and save battery life, which is probably the primary benefit here.
17:38 andi-: ahh yeah that would have been my next question :D
17:38 andi-: even if I can't use it i'd like to suspend it. I'll work through the bugreport before questioning further
17:38 andi-: thank you imirkin :-)
17:57 martm: it's arguable that if i had correct go during those years for studying linux operating system, but today is my 15nth year with running linux
17:58 martm: most of it were the times when my life was fucked up, i was bit scattered by problems, today it starts to shine in world, like a use of os that is really strong or can be made easily to be the consumers choise of a system a primary one
17:59 martm: fucking hell what do i write again, shattered by..correction
18:00 martm: no scatter gather DMA there i.e sgdma
18:02 martm: so there are guys like me here others, that use desktop 15years in a row with open systems under linux, and feel like this is exactly for my taste
18:06 martm: knowing about brittish culture even too, and have learned and delt with all major nations, i am not sure if you are bullying with me or not, but...
18:08 martm: i have been connecting with french and germans and such brittish very extensively, also chinease and russians, we know about how calim also got his computer education, i know austrias too, ill recap what is that strong university that teached or taught so big programmer
18:40 karolherbst: imirkin: mhh seems like the formular is close enough :/ but there might be something odd though
18:41 karolherbst: https://gist.github.com/karolherbst/98708b57d0194aaf51f6a21b3124204c
18:41 karolherbst: but it still looks as expected...
18:41 karolherbst: just the inaccuracy is killing us here
18:41 karolherbst: at high temperature this c5 part behaves really brutal :D
18:42 karolherbst: ohhh
18:42 karolherbst: the downvolt after "943750, 937222, -6528, 99.308291, 15, 24, 75" is interessting
18:43 karolherbst: step size is 6250
18:43 karolherbst: mhh something even closer: 6320
18:44 karolherbst: yeah, I think this is right
18:49 karolherbst: gnurou: I am like 98% sure that the coefficients are right now :)
18:52 karolherbst: yeah, an no downvolt at -6232..
18:52 karolherbst: this has to be right
18:59 martm: and folks today is my 33rd birthday, so to tie two things together it's almmost worth couple of zips of vodka...
19:00 imirkin: hehe
19:00 imirkin: *has to*
19:04 karolherbst: imirkin: yeah, but it just looks like it
19:05 karolherbst: imirkin: with the earlier stuff I always saw some kind of error and some oddness, but this is completly gone now
19:06 imirkin: =]
19:06 imirkin: hopefully that improves stability for some people
19:08 karolherbst: mhh
19:08 karolherbst: doubtfull
19:08 karolherbst: well compared to v3 that is
19:08 karolherbst: or, wait
19:08 karolherbst: no, in v3 I didn't had the speedo yet
19:09 karolherbst: well the old stuff was pretty good to already, and so I don't think it actually increases stability by a noticable amount at all
19:09 karolherbst: there are more severe issues, like on some nve7 users needs a 200mv higher voltage (than nvidia) on nouveau to be able to reclock
19:09 karolherbst: ...
19:15 karolherbst: and I think I will now leave the branches for each send patch series without any changes and increment the branch v number (I hope)
21:01 martm: calim: i know inria and such, bordeux ..and also heidelberg and several others in europe, we do not talk about brittish, cause those are pure kings
21:01 martm: but... this austrias university is strong too
21:02 martm: in europes ground nobel prize has been won the most time over a hundred by brittish
21:02 martm: united kingdom, great britain
21:03 martm: but in worlds sense, the answer is easy, united states has won it more then 300times
21:05 martm: the part what you do not understand prolly , is that british and united states will never give a fuck about estonia, neither they ever gave
21:06 martm: there is the source of minor conflict what we face today, our stars like me, they can not rely on brittish nor yanks
21:09 martm: but at the same time i do not deny that saint bastions hostpital from brittish and all their medicine isn't the worlds best almost
21:12 martm: we all know that states , united states surgeons have their own aces, and those ones are as big then brittish and french
21:13 martm: lets not forget about shapiro md.
21:13 martm: who kinda escaped lance armostrong from certain death and several others on the web who got second oprtuinity to live
21:14 martm: it was because yanks jewish pure ace resected the brain tumour
21:15 martm: and do not fuck with me, i know estonian who as done it as successfully as shapiro
21:16 martm: so we have our own stars in the shadow of brits
21:16 martm: and if it's going in good leadership almost as i am in life, we have chanches in every area
21:20 martm: you do not give me shit about portuguese here, before portuguese ronaldo, we knew brazilian ronalda, and i know the french surgeon who saved his career
21:39 martm: apart from portuguese and their science, it is something to mention but not too much, i'll go to my first rant of the evening
21:40 martm: i think linus torvalds that one of the biggest estonians stars is a linux junky
21:41 martm: linus is very intelligent in a sense that he speaks two languages pure, swedish and finnish that i have confirmed
21:41 martm: finnish was beatiful
21:41 martm: and he says that swedish is stronger
21:41 martm: and english is almost good
21:42 martm: but now about music, i love shura from united kingdom recently
21:42 martm: and kygo from norway
21:44 karolherbst: https://htmlpreview.github.io/?https://gist.githubusercontent.com/karolherbst/b297487421b213ac15d95181a09157e9/raw/ade57a6efc671f5096b520409cd8428e5143d6d4/test.html :)
21:45 karolherbst: now also shows Linux support :)
21:59 mooch3: hey guys, i'm here from an nv4e right now
22:00 mooch3: i want to know if much has changed in pfifo between nv4 and nv4e. i need to know if testing on nv4e can enlighten what goes on on nv4
22:00 karolherbst: with nv4 you mean nv04?
22:01 mooch3: yes
22:01 karolherbst: and what do you mean by nv4e?
22:01 mooch3: i'm currently attempting to get nt4 to boot while using nv4 drivers, but it's currently hanging on pfifo-related things
22:01 mooch3: i mean c51, or geforce go 6100
22:02 karolherbst: ohh MCP51
22:02 mooch3: yeah, this laptop is old, i know
22:03 karolherbst: mh I have no idea, I was just surprised by the naming
22:03 mooch3: i prefer to use the nvxx naming scheme. :)
22:04 karolherbst: ahh now I see there the 4e comes from
22:04 karolherbst: right
22:04 karolherbst: well
22:04 karolherbst: I would say, that a lot has changed
22:04 mooch3: i do know that nv20 is fairly similar, so i've been using xqemu so far, but really, xqemu doesn't implement lots of stuff
22:04 karolherbst: because there are like 3 generations between those two
22:47 martm: airlied: is in both of the unions, great britain and united kingdom..or what the fuck
22:47 martm: this is a cocky bastard, but he is quite lovely
22:47 martm: some irish soul
22:57 hakzsam: imirkin, well, images on gk110 are almost done, I still have a few number of things to fix up in both envydis and codegen but they will wait for tomorrow :)
23:00 imirkin: nice
23:01 karolherbst: awesome :)
23:26 karolherbst: imirkin: okay, I looked a bit into the BB issue, and the last instructions are removed inside FlatteningPass
23:26 karolherbst: and before that there are some joins left
23:28 karolherbst: like this: https://gist.github.com/karolherbst/731f56fbda8619a986e963527736f30e
23:29 imirkin: yeah
23:29 karolherbst: the thing is
23:29 karolherbst: they branches are always empty
23:30 karolherbst: this is before any optimization is done: https://gist.github.com/karolherbst/731f56fbda8619a986e963527736f30e#file-gistfile2-txt
23:30 karolherbst: ohhh
23:30 karolherbst: it is even in the TGSI
23:30 karolherbst: the heck
23:30 karolherbst: https://gist.github.com/karolherbst/731f56fbda8619a986e963527736f30e#file-tgsi
23:31 karolherbst: no idea what ELSE :0 means
23:31 karolherbst: but this looks pretty useless to have
23:33 karolherbst: imirkin: shouldn't that be optimized away in the TGSI already?
23:42 vita_cell2: karolherbst