00:37 imirkin: skeggsb_: ok, i'm going to try to finish it up then
00:37 imirkin: grabbing DOW3 now, and i already have hitman which iirc was the other game
00:48 koz_: imirkin: Does Hitman run OK under nouveau?
00:48 imirkin: koz_: no clue.
00:48 imirkin: also what's your definition of 'OK'?
00:49 koz_: imirkin: 'Can be played without hideous visual glitches?'.
00:49 imirkin: i'll try to see if i notice anything
00:49 imirkin: note that all my nvidia hw is bottom-of-the-line
00:49 koz_: s/glitches?'./glitches.'?
00:49 koz_: imirkin: It might fall below spec then.
00:49 imirkin: which is fine from a correctness standpoint, but definitely not usable for estimating perf
00:50 imirkin: currently i have a GT 730 plugged in, for example
00:50 imirkin: with DDR3 VRAM
00:50 koz_: imirkin: Yeah, I guess for correctness, it's fine.
00:50 koz_: I've had a great time with nouveau on a bunch of things.
00:51 koz_: Although none of them are super-duper demanding.
00:51 imirkin: i guess the FX 3700 i got is actually not bad, and the GT 240 i have is good too.
00:51 imirkin: but both are older
00:51 imirkin: and neither reclock in nouveau ATM
00:51 imirkin: (rspliet......)
00:52 koz_: imirkin: I have a GT680, and so far, I can confirm that a bunch of things work. Hand of Fate for example works really nicely.
00:52 koz_: I had brownout issues, but that was my PSU being too weak.
00:52 imirkin: cool
00:52 koz_: (it was OK before nouveau could hit boost frequencies)
00:52 imirkin: i don't really play games anymore, and spending hundreds on video cards seemed ... unnecessary.
00:52 imirkin: although of course by now i probably have spent a bunch on all the various boards i have
00:53 koz_: imirkin: I got mine second hand for just over a hundred.
00:56 imirkin: ah nice
00:57 imirkin: i think the most i've spent on a board is $30 or so
00:57 koz_: That's... extremely cheap. Old hardware?
00:57 imirkin: generally yea
00:58 imirkin: hakzsam: do you remember why we still do nve4_set_surface_info for GM107+?
00:58 imirkin: hakzsam: is there some data in there that we need and is not accessible via the TIC query methods?
01:58 mangix: any way to add extra CFLAGS to compilation of nouveau?
01:58 mangix: defining CFLAGS doesn't seem to work
02:00 imirkin: should work.
02:00 imirkin: (works when i do it...)
02:00 imirkin: perhaps we define it differently?
02:16 mangix: hrm lots of mismatching declarations in the code
02:16 Guest79244: i have a doubt, does the jetson tx1 SoC pushes negative 5.5 volts on boot at some part?
09:45 karolherbst: koz_: no it doesn't
09:45 karolherbst: random crashes which might be fixed by supporting bindless
09:46 koz_: karolherbst: Sorry, what was this in reference to?
09:46 karolherbst: hitman
09:46 koz_: Ah. Thanks for letting me know.
09:47 karolherbst: I looked into it and all the issues I've found aren't trivially fixed
09:47 koz_: karolherbst: I see. Well, I'm not in a hurry to play it, so that's OK.
10:43 vpelletier: imirkin_: hey. I just booted my g5 with the geforce in another slow, not expecting much. Turns out, the PCIe 16x is on one controler, and (at least) the 8x is on the other one, the one ethernet cards are internally wired to
10:43 vpelletier:'s mind: blown
10:43 vpelletier: sadly, I'm still getting the "switching to fbcon" line
10:44 vpelletier: still only one interrupt served for nvkm
10:45 vpelletier: but this time, geforce's MSI address is the same as ethernet cards: Address: 00000000fee00000 Data: 0008
10:45 vpelletier: (data is different, of course)
10:47 vpelletier: which I guess means there may be nothing wrong with the controler, but something wrong with how nouveau (?) tells the card to trigger MSI
11:29 dboyan_: imirkin: I admit that I've been a little stuck here this week. I was hacking another piece of hardware used in my lab. But I'll work on nouveau for the whole weekend.
11:30 dboyan_: Top-down scheduling doesn't seem to work very well, I'll try scheduling bottom-up
11:53 hakzsam: imirkin: I guess it's for imageSize() (OP_SUQ)?
11:53 hakzsam: we read the image sizes from the driver constant buffer
11:54 hakzsam: I think it's the only reason, but other stuff should be in the TIC header
12:08 karolherbst: dboyan_: will you concentrate on pre RA scheduling first for now or do you actually plan to write a post RA scheduler as well or maybe even a RA scheduler?
12:09 karolherbst: or make RA less dumb?
12:19 dboyan_: karolherbst: still pre-RA i guess, I'll probably also touch RA
12:37 manili_: Hey guys, how is it possible to join Wayland IRC (it says: you need to be identified with services) ?
12:39 pmoreau: manili_: This is not a support channel for IRC issues, but have a look at https://superuser.com/questions/106965/how-do-i-identify-with-services-on-irc
12:41 pmoreau: manili: And if it is on the freenode server as well, https://freenode.net/kb/answer/registration
12:42 manili: pmoreau: I'm so sorry. I thought it needs kind of invitation.
12:43 pmoreau: I don't think so... But I never tried joining
12:44 manili: OK. Thanks a lot for help.
12:44 pmoreau: manili: I can join, and I never was invited. So you most likely need to register a nickname.
13:50 imirkin_: hakzsam: ok. i think we can read stuff out of the TIC header directly with queries
13:51 imirkin_: hakzsam: so we should change that... i'll give it a shot before adding bindless on gm107
13:51 imirkin_: i got stuck on images a little, have to change a bunch of tgsi -> nvir
13:52 hakzsam: imirkin_: you can add nve4 support only for now
13:52 hakzsam: that would be enough for testing dow3
13:52 imirkin_: you mean texture only?
13:52 hakzsam: bindless only
13:52 imirkin_: [re-read]
13:52 hakzsam: because gm107 is quite different from nve4
13:52 imirkin_: not really.
13:52 hakzsam: images are
13:52 imirkin_: not really.
13:53 imirkin_: not different enough.
13:53 imirkin_: anyways, i was planning on handling the whole thing
13:53 hakzsam: because the code is well written :p
13:53 hakzsam: your call
13:53 imirkin_: i'm not complaining.
13:53 hakzsam: let me know when you have something which works
13:53 hakzsam: I could try dow3 on my gm107
13:54 imirkin_: will do
13:54 imirkin_: i have a gm107 too btw
13:54 imirkin_: so far i'm just hacking on the GK208
13:54 hakzsam: but you don't have dow3?
13:54 imirkin_: i do now
13:54 hakzsam: cool
13:54 imirkin_: it's on the feral plan
13:54 hakzsam: feral sent you a key?
13:59 karolherbst: uhh, then I can test dow3 as well, ncie
14:00 karolherbst: imirkin_: do you think we should just support bindless for fixing those hitman texture issues or should we try to get the normal texture support working as well? The issue was that we ran out of tics/tscs
14:01 imirkin_: skeggsb_: --^
14:01 karolherbst: or should those tics/tscs be dynamically allocated according to the needs?
14:01 imirkin_: karolherbst: should probably just bump up our table size
14:01 imirkin_: karolherbst: bindless won't solve the issue of having too many tic's/tsc's
14:01 hakzsam: and bindless hitman is somehow buggy
14:02 imirkin_: [in fact bindless will exacerbate that issue]
14:02 imirkin_: karolherbst: there should be no way to run out of table size there...
14:02 imirkin_: at most 32 * 6 slots should be locked without bindless
14:02 imirkin_: which is well below the 2048 limit
14:05 karolherbst: imirkin_: I quadroubled the table and still hit the issue
14:05 hakzsam: with only 2048 handles, you are not going to launch bindless hitman...
14:05 imirkin_: well, like i said, there's some bug
14:05 imirkin_: which is entirely unrelated to table size
14:05 karolherbst: most likely, yes
14:06 karolherbst: maybe I take another look and try to figure that out
14:06 imirkin_: from my understanding, at most 32*6 entries should be lockable at a time
14:07 karolherbst: so 192 in total?
14:08 imirkin_: that's if you have all 5 graphics stages and compute, and each one has 32 textures bound
14:08 karolherbst: mhhh
14:08 imirkin_: [*different* textures]
14:08 imirkin_: so ... slightly unlikely.
14:08 imirkin_: but it's the hard max.
14:08 karolherbst: our macros are set to 2048 entries, no clue if they map 1:1 here though
14:09 imirkin_: no
14:09 imirkin_: the table is for the TIC and TSC entries
14:09 imirkin_: and then those entries are referenced in a constbuf
14:09 imirkin_: when that reference is made, those entries are locked
14:09 imirkin_: but you can have valid textures uploaded into TIC/TSC entries that are not locked
14:09 imirkin_: however when trying to add another texture, those might get kicked out
14:10 imirkin_: whereas the locked ones can't
14:10 karolherbst: ohhhh, I see
14:13 karolherbst: imirkin_: anyway, i can get bigger than NVC0_TIC_MAX_ENTRIES here and I don't really know why: https://github.com/karolherbst/mesa/blob/master/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c#L1279
15:20 imirkin_: vpelletier: wait, is this a PCIe card or a PCI-X card?
15:25 imirkin_: karolherbst: what's screen->tic.next ?
15:25 imirkin_: karolherbst: if it were negative, then this logic might not work
15:25 vpelletier: imirkin_: pci-express. did I write pci-x somewhere by mistake ?
15:25 imirkin_: vpelletier: yeah, in your email to benh...
15:26 imirkin_: vpelletier: i assumed PCIe as well, but just double-checking :)
15:26 imirkin_: since PCI-X would bring all kinds of extra differences
15:26 karolherbst: imirkin_: will check after work
15:30 imirkin_: karolherbst: and it could become negative as a result of some rogue write... no way for that to happen "normally"
15:34 karolherbst: imirkin_: who knows, I just know that i wasn't _much_ above the size, as it grew normally, it just become slightly too big
15:36 imirkin_: oh
15:36 imirkin_: well... throw a lot of asserts in there
15:36 karolherbst: yeah
15:37 karolherbst: it became better after doubling the TIC/TSC size
15:37 karolherbst: uhm max_entries I meant
15:38 imirkin_: that ... has nothing to do with the problem
15:38 karolherbst: well kind of, because a rogue write should kind of have the same affect
15:38 imirkin_: well, perhaps you and i have a diff way of going about things
15:39 karolherbst: and with better I meant it became _a_lot_ better to a point the game is playable for a few minutes
15:39 imirkin_: but i spend time to understand the issue while changing as little as possible
15:39 imirkin_: once i do so, a fix becomes immediately apparent (or at least an understanding of the problem)
15:39 karolherbst: imirkin_: yeah, that's for sure
15:43 karolherbst: imirkin_: by the way, do you still plan to find a nice way to handle that pow thing?
15:43 imirkin_: probably not.....
15:44 karolherbst: understandable. It gave me quite a few headaches and the code I came up with wasn't working for all exponents to begin with
15:46 imirkin_: i gotta pick and choose what i work on
15:46 imirkin_: since my time to futz with nouveau has become MUCH more limited
15:46 imirkin_: seems like e.g. bindless is a better use of my time than pow lowering
15:46 karolherbst: makes sense
15:47 karolherbst: would be still good if you find some time to review things, otherwise we end up with a lot of devs, but nobody to review stuff (or push it in the end)
15:48 karolherbst: but maybe I should bother pmoreau or hakzsam with that now, since you are also occupied with gsoc stuff
15:49 hakzsam: karolherbst: about what?
15:50 pmoreau: karolherbst: I'll have a look at the pow series over the weekend. I was at a conference this week, and did not had too much time to do any reviewing or try stuff.
15:58 karolherbst: hakzsam: precise in TGSI and pow lowering
15:59 karolherbst: hakzsam: the former is needed for fixing rendering issues in TombRaider with Nouveau
15:59 hakzsam: okay
15:59 hakzsam: I will try to remember :)
15:59 karolherbst: will send out a v4 and put you in CC then :p
16:00 hakzsam: ok
16:03 pmoreau: imirkin_: Do you need me to send a v2 for the split patch, containing all the info? Or can I kindly ask you or hakzsam to push it, when you have time?
16:04 hakzsam: pmoreau: I can push this for you if you want
16:05 hakzsam: (I haven't looked at the patch)
16:06 imirkin_: pmoreau: either way is fine. i forgot you didn't have push rights.
16:06 pmoreau: hakzsam: If you feel like reviewing at the same time as well. :-) Then, don't forget to include Ilia's Rb + the CC and Fixes tags.
16:06 TiTan4T: hey, I installed nouveau with: https://wiki.archlinux.org/index.php/PRIME My first card is a Intel Graphics, which works well and my second card is a GTX 940MX, which don't work well. I installed glmark2 and compared. My intel card get 2080 Score and nvidia 266
16:06 pmoreau: imirkin_: One day, maybe one day I'll have them. ;-)
16:06 imirkin_: TiTan4T: sounds expected... your GTX 940MX is probably clocked to the lowest freq
16:07 imirkin_: TiTan4T: you can change clocks by echo'ing levels to /sys/kernel/debug/dri/1/pstate
16:07 Lyude: bleh, mupuf if you answered my question (if not that's cool) I didn't see the answer since my bouncer lost connection to freenode for a while
16:07 imirkin_: Lyude: there are logs
16:07 Lyude: oh right
16:07 Lyude: i forgot we have those
16:07 TiTan4T: 07: core 405 MHz memory 810 MHz 0a: core 270-1189 MHz memory 1600 MHz 0f: core 270-1189 MHz memory 5010 MHz AC: core 0 MHz memory 0 MHz
16:08 imirkin_: TiTan4T: ok, well the AC: is showing that the gpu is currently off
16:08 pmoreau: Is 940MX a GM108? We don't have that model in the code names table.
16:08 imirkin_: unfortunately we don't currently support setting the freq across gpu suspends, but if you run an application on the GPU and then echo 0f > the-file
16:08 TiTan4T: https://pastebin.com/CuHUCKET
16:08 imirkin_: then things should speed up
16:08 karolherbst: ohh right, skeggsb_ please look at my clocking fixes to fix clocking on prime systems as well? :/
16:10 imirkin_: pmoreau: 940MX tends to be GM107 or GM108 yeah
16:10 TiTan4T: imirkin_: What are correct values?
16:10 TiTan4T: I have no idea
16:10 imirkin_: TiTan4T: there are no *correct* values...
16:10 karolherbst: TiTan4T: start something on the GPU, then you see real values
16:10 imirkin_: all but the last line indicate a perf level
16:10 imirkin_: the last line indicates the current setting
16:10 TiTan4T: Wait I start the benchmark
16:11 karolherbst: TiTan4T: you are on a 4.10 kernel or newer, right?
16:11 TiTan4T: https://pastebin.com/TrpuFzJw
16:11 karolherbst: TiTan4T: now echo 0f into that
16:11 TiTan4T: [titan@Unibook2 ~]$ uname -r 4.11.6-1-ARCH
16:14 TiTan4T: one moment need reboot
16:15 pmoreau: TiTan4T: Could you please run `lspci -d 10de:` and tell me whether it says GM107 or GM108? I'll add that model to the table.
16:15 imirkin_: lspci doesn't know about it
16:16 TiTan4T_: the echo crasht my card
16:16 karolherbst: TiTan4T: you can only do it while something runs on the GPU
16:16 imirkin_: =/
16:16 TiTan4T_: 02:00.0 3D controller: NVIDIA Corporation GM108M [GeForce 940MX] (rev a2)
16:16 pmoreau: Thanks
16:16 imirkin_: ah, i guess lspci knows about this one :)
16:17 TiTan4T_: You say i should start benchmark and try to echo 0f in it?
16:17 imirkin_: yes.
16:17 TiTan4T_: like echo 0f | sudo tee /sys/kernel/debug/dri/1/pstate
16:17 imirkin_: i thought i said that.
16:17 pmoreau: imirkin_: It probably is found within the card name, or some other field that lspci reads.
16:17 karolherbst: TiTan4T: yeah, if nothing runs on the GPU, the echo call might hang stuff, because we have bugs in there while the GPU is off
16:17 imirkin_: pmoreau: lspci is just a database
16:18 imirkin_: pmoreau: in pci.ids. either it's there or it's not.
16:18 pmoreau: Ok
16:19 TiTan4T_: okay, now its run better. What is the max value, which I can pipe into that file? Is it 0f ?
16:19 karolherbst: TiTan4T_: exactly
16:19 karolherbst: and it resets everytime the GPU goes into suspend
16:19 karolherbst: but there are patches to fix all this
16:19 imirkin_: TiTan4T_: the list of available levels is listed when you cat the file
16:19 karolherbst: they just lack review
16:20 TiTan4T_: ah
16:20 TiTan4T_: of vs AC
16:20 TiTan4T_: 0f
16:20 imirkin_: AC = current state
16:20 TiTan4T_: 0f: core 270-1189 MHz memory 5010 MHz AC DC * AC: core 1063 MHz memory 5009 MHz
16:20 TiTan4T_: okay
16:21 imirkin_: the "*" is just there to fool you, ignore it :)
16:21 TiTan4T_: okay, so 0f is the highest performance
16:21 karolherbst: well, it's kind of correct though
16:21 TiTan4T_: What do you mean with there are patches? There will be patches or I can add fixes?
16:21 imirkin_: karolherbst: "kind of" being the operative comment
16:23 karolherbst: well, it's in 270-1189 MHz
16:24 karolherbst: TiTan4T_: I wrote some patches to fix all this, but they didn't got merged until now
16:24 TiTan4T_: k
16:24 TiTan4T_: But some fact. My intel HD still has more points in the benchmark
16:24 karolherbst: depends on the benchmark
16:24 karolherbst: glxinfo is none
16:24 karolherbst: :p
16:24 karolherbst: *glxgears
16:24 TiTan4T_: glmark2
16:25 pmoreau: mupuf: Do you have the powers to configure patchwork Nouveau to change the status to accepted if merged into Ben's tree? For example, Mikko's patches have been accepted but are still marked as new: https://patchwork.freedesktop.org/series/25558/
16:25 karolherbst: TiTan4T_: every benchmark with high FPS output can't be reliably used on prime setups
16:25 karolherbst: TiTan4T_: because the more FPS you have, the more you have to pump through PCIe
16:26 karolherbst: pmoreau: all my patches for mesa are still marked as new or in progress :p
16:27 TiTan4T_: When the card will deactivate and I have to reset the flag?
16:28 TiTan4T_: Actual I quit the benchmark, and started it with Intel GPU again, but the flag is still setted
16:28 karolherbst: it takes a few seconds
16:28 pmoreau: karolherbst: You have some accepted ones ;-) https://patchwork.freedesktop.org/project/mesa/patches/?submitter=15829&state=3&q=&archive=&delegate= https://patchwork.freedesktop.org/project/mesa/patches/?submitter=16071&state=3&q=&archive=&delegate=
16:28 karolherbst: fake news
16:28 karolherbst: odd
16:28 TiTan4T_: I restart and check one other bug
16:28 karolherbst: well wtvr
16:29 TiTan4T_: If I right click somewhere and hover over the menu i have there a sort of tearing
16:29 TiTan4T_: But somebody strange
16:31 karolherbst: pmoreau: my MAD patches aren't listed as accepted
16:31 karolherbst: ohh not true
16:31 karolherbst: there are just there several times
16:33 TiTan4T: hm its still a bug
16:33 pmoreau: It sometimes fails to figure out there are different versions of the same series.
16:34 imirkin_: TiTan4T: note that the intel driver has a sizeable team of full-time developers with full access to documentation, while nouveau is a rag-tag team of partial volunteers (and 1 full-timer trying to keep the ship afloat)
16:35 karolherbst: I like "rag-tag"
16:35 TiTan4T: If there are no performance issues in games it's all okay ;9
16:36 karolherbst: there are
16:36 karolherbst: saints row is the worst
16:36 TiTan4T: So I will have worse performance on my nvidia card, then on intel?
16:36 karolherbst: depends on the game
16:36 karolherbst: the 940M isn't an exactly fast GPU
16:36 imirkin_: at least it's got DDR5
16:36 karolherbst: true
16:37 imirkin_: that alone should be a win over intel in more complex scenarios
16:37 imirkin_: (e.g. try glxspheres)
16:37 karolherbst: glxspheres is also pretty much PCIe bound
16:37 imirkin_: really? hmm
16:38 karolherbst: unigine heaven
16:38 karolherbst: intel runs like shit there
16:38 imirkin_: i thought the framerate would be low enough
16:38 karolherbst: well
16:38 karolherbst: I get around 500 fps
16:38 imirkin_: ah yeah
16:38 imirkin_: your gpu is faster than mine ;)
16:38 karolherbst: yes
16:39 karolherbst: on all gputest benchmarks nouveau should be faster as well
16:39 karolherbst: especially pixmark_piano
16:40 TiTan4T: Where I get glxsphere?
16:41 imirkin_: grab unigine heaven :)
16:41 TiTan4T: ah
16:41 imirkin_: https://benchmark.unigine.com/heaven
16:41 TiTan4T: optirun
16:41 imirkin_: or one of the UE4 linux demos
16:41 TiTan4T: optirun glxspheres64 oder 32?
16:41 imirkin_: https://wiki.unrealengine.com/Linux_Demos
16:42 pmoreau: TiTan4T: There are AUR packages for unigine, and for glxspheres, it is the virtualgl package.
16:42 TiTan4T: jo
16:42 TiTan4T: I'm installing right now
16:44 TiTan4T: glxspheres64 has intel and nvidia 60 fps
16:47 pmoreau: `vblank_mode=0 glxspheres64`, to disable V-sync, IIRC
16:50 TiTan4T: both ca. 165 fps
16:51 imirkin_: that sounds very low
16:51 imirkin_: did you remember to bump clocks on nvidia?
17:00 TiTan4T: yes
17:00 TiTan4T: The star is still on 0f
17:02 orbea: TiTan4T: read the AC line at the bottom, that tells you the current state
17:04 TiTan4T: Nothing changed
17:04 TiTan4T: 0f is the current state
17:05 TiTan4T: but its look like, nvidia card doesn't perform better then intel
17:05 TiTan4T: in a few benchmarks
17:12 imirkin_: well, make sure you try a real benchmark before you decide that, e.g. unigine heaven
17:13 TiTan4T: I tryed unigine heaven ...
17:14 TiTan4T: tried
17:25 karolherbst: :3 http://beta.speedtest.net/result/6400930598.png
17:25 orbea:wants....
17:46 Lyude: Has anyone else been hitting this bug on drm-next? specifically with reloading nouveau and suspend/resume with nouveau https://paste.fedoraproject.org/paste/VDR5s69oLfsnXlTnNvBfwA
17:47 karolherbst: Lyude: https://gist.github.com/karolherbst/a11001df7f12907f074fbdfb12cbf05d
17:47 Lyude: karolherbst: hm?
17:48 karolherbst: cloning linux
17:48 karolherbst: we talked about it how long it takes
17:48 Lyude: ahh, yeah I never had any problems with kernel.org, it was github that was having issues
17:48 Lyude: but I left it sitting and it eventually managed to do it
17:48 karolherbst: ohh, I could speedtest githuib as well
17:49 karolherbst: mhh, it's equally fast
18:05 pmoreau: Lyude: I haven’t seen that one, but the NVIDIA card never goes back automatically to sleep after a cycle of suspend/resume.
18:06 Lyude: hm, i am going to debug this and figure out what's going on. it looks like a bad memory access somewhere but kasan didn't pick it up last time I tried, might have forgotten to enable some slub debugging option somewhere..
19:12 karolherbst: hakzsam: do you know how to keep hitman doing stuff if put out of focus?
19:13 karolherbst: seriously, HitmanPro has one of the most annoying engine features
19:27 karolherbst: imirkin: seems like that texture crash is gone for no apperant reason
19:47 karolherbst: why do you even make stuff so undebugable....
20:56 rhyskidd: So I've been able to verify a GP108 / nv138 vbios through envytools
20:56 rhyskidd: pushing the PR branch now so it runs through the CI and others can take a look
21:24 rhyskidd: dboyan_: Let me know if you want any debugging help with GP107/nv137 MMIO timeout issues https://bugs.freedesktop.org/show_bug.cgi?id=100228