01:07pmoreau: RSpliet: A few questions about ram reclocking:
01:07pmoreau: 1) reclocking for nvaa/ac is supported, but I can't find all the training functions / related stuff in the ramnvaa file. Does it use the one from ramnva3?
01:08pmoreau: 2) is there some reason why ram reclocking would be different given the chipset? should it not be dependent only on the ram type?
01:55mlankhorst: pmoreau: it differs per ram type and can differ per chipset too :P
01:56mlankhorst: gf100 and other fermis all seem to have slight differences even for gddr5
02:03mupuf_: pmoreau: as for 1), why would there be memory training functions on vram-less chispets? :)
02:05mlankhorst: or that
02:38pmoreau: mupuf_: Right, I forgot about that... --'
02:39pmoreau: mlankhorst: I thought this would be the kind of things "defined" in the ram type spec. But my hardware knowledge is quite empty :/
02:42Leftmost: Clocking isn't really a property of the RAM but a property of the bus, which is up to the chipset.
02:45mupuf_: pmoreau: well, there is the RAM
02:46mupuf_: and all manufacturers are a bit different
02:46mupuf_: but then you have PFB ... which is driving the ram
02:46mupuf_: and pfb can be chipset-dependent
02:47pmoreau: I figured out the last part :D
02:52pmoreau: Is there some documentation available about reclocking theory? I had a (really) quick look to the radeon docs but couldn't find some.
03:07mlankhorst: pmoreau: hahaha good luck ;-)
03:10pmoreau: mlankhorst: Thanks! :)
03:11mlankhorst: pmoreau: although maybe in the tegra docs you could find some info
03:11mlankhorst: although that's more about general reclocking, memory reclocking won't be covered there
03:12pmoreau: I guess I'll start by playing with Ilia's patches for ramnv50 and see how it goes
03:12Eliasvan: Hi everyone, I think I succeeded in getting mmiodumps of the NVD7!
03:12Eliasvan: for the reclocking
03:13Eliasvan: I sent the results to mmio.dumps at gmail dot com
03:13pmoreau: mlankhorst: Which documentation are you talking about?
03:13pmoreau: Hi! :)
03:13Eliasvan: pmoreau: hey :)
03:18Eliasvan: imirkin: is it possible to take a quick look at my dumps, just to see whether the vbios modifications had any results on the traces? (I've no experience in analysing mmiodumps, especially regarding reclocking)
03:35pmoreau: Eliasvan: Do you know about the demmio tool from envytools? It parses the mmio dump with the lookup command, making the trace easier to analyse
03:38Eliasvan: pmoreau: yes, I heard about it; the problem is that I don't have time right now to analyse (however I would be glad to do it if I had enough time)
03:39Eliasvan: just a question: who receives the mails at mmio.dumps at gmail dot com?
03:40pmoreau: Every developer can access it through the webmail.
03:40pmoreau: I don't know if some added some redirection to their own mailbox
03:41pmoreau: s/through the webmail//
03:41Eliasvan: ah, ok, in that case enough people will receive the mail, and if they find time, the can notify me via mail whether the traces are useful
03:42Eliasvan: then I will log out now, thanks again for all help, and success with all work!
03:42pmoreau: See you!
05:11zeael: Hi, I'm using nouveau for the 9400M in my MacBook Pro 5,5 and there are a few issues, most notably flickering on my 23'' Dell monitor connected via DisplayPort. I suppose I should not file a bug report because this is a known issue, as stated in the troubleshootig page?
06:00pmoreau: zeael: Which kernel version are yo using?
06:01zeael: built it from the 3.19 branch
06:01zeael: because the latest patches finally enable me to use nouveau on my laptop :-)
06:02pmoreau: Ah, you already have it then
06:02pmoreau: Do you also have a 9600M GT in it or not?
06:02zeael: only the 9400m
06:03pmoreau: Do you have a dmesg output?
06:05pmoreau: I should be able to give it a try this evening. But I remember having this issue a few years ago, until I disabled Nouveau on it.
06:06zeael: It works with a DP -> VGA adapter though (at least on a different monitor)
06:07zeael: Give me a few minutes, I'll get to my monitor and make logs
06:07pmoreau: Ah, damn it, that's the only configuration I have at the moment... Left my DP -> HDMI at home :/
06:07zeael: Which debug options do I need?
06:07zeael: Ah okay..
06:08pmoreau: Add nouveau.debug=debug and drm=0xf, that should be enough material I hope
06:08pmoreau: (I'll find it back in two weeks, don't worry ;) )
06:09zeael: so i'll be right back
06:33zeael: Alright. First, on the monitor in question even the DP->VGA adapter doesn't work, I get the same flickering. So the problem should only apply to certain monitors? Maybe that's why I didn't find much on the internet...
06:34zeael: I have a dmesg for the DP->DP case
06:34pmoreau: I read about the troubleshooting case a few seconds after you left
06:35pmoreau: Might be worth to check if reclocking solves it, as reclocking is supported on NVAC cards! :)
06:35zeael: Okay, that'd be great!
06:35pmoreau: You need to boot with nouveau.pstate=1
06:36pmoreau: Then as root, you can cat /sys/class/drm/card0/device/pstate to get the available values. Simply echo one of the values to the same file to changed to that level
06:39zeael: Okay. So I try those values and one or some of them could make it work?
06:39pmoreau: The higher ones 07, 0f should make it work (hopefully)
06:39pmoreau: Otherwise I'll get back to that card
06:40zeael: Great, thanks. I'll try that and report
06:47zeael: no luck
06:47zeael: values are 03, 05, 0e, 0f and AC
06:48zeael: 03 and 05 make it worse, as I expected, 0e and 0f look the same to me and AC I can't set
06:48pmoreau: AC is the current one ;)
06:50pmoreau: Could you please link the dmesg you grabbed?
06:51zeael: Shall I include changing of pstates?
06:51pmoreau: Well, if you also have that one, it won't hurt.
06:51pmoreau: But do not restart to generate it
06:52zeael: I just give you the current one
06:52pmoreau: (or another website)
06:56pmoreau: zeael: If you have some time, what might be interesting is testing with the blob, see if it works, and for which clocks.
07:01zeael: here you are
07:01zeael: yeah of course
07:06pmoreau: Meh... I forgot the .debug part for drm: it should have been drm.debug=0xf, sorry
07:07zeael: No problem. You'll get a new one in a minute :-)
07:07pmoreau: Hopefully the DP port is linked to my NVAC, so I should be able to have the same environment as you
07:08pmoreau: Except that I have an NV96 along the NVAC
07:09RSpliet: zeael: is that one monitor with issues? the one with 1920x1080?
07:12RSpliet: pmoreau: do you have more info about this problem?
07:13pmoreau: RSpliet: I remember having this issue like 3 years ago, before I decided blacklisting Nouveau
07:13pmoreau: But haven't retested with the patches
07:13RSpliet: pmoreau: no I mean zeaels problems
07:13zeael: Wow...drm is flooding my dmesg
07:13pmoreau: That's what I meant
07:14zeael: By what?
07:14pmoreau: RSpliet: I used 3 years ago to connect an external screen to my laptop and had the issue. But since I had troubles with the NVAC and blacklisted Nouveau, I never tried again to connect an external display under Linux
07:15zeael: i see
07:15zeael: This is 1M of dmesg
07:15RSpliet: zeael: you're experiencing issues with the 1920x1080 monitor I presume... what happens if you disable the internal panel?
07:16zeael: At boot time?
07:16RSpliet: zeael: shouldn't matter much
07:16RSpliet: as long as it's really really disabled, and not just a disabled backlight :p...
07:16zeael: So I can try xrandr?
07:17zeael: Oh I just posted the link to the pastebin error page, sorry :D
07:18zeael: It is still flickering, albeit more statically
07:19RSpliet: hmm... odd; I'm sure I can drive a 1920x1080 panel from an NVAC :)
07:19pmoreau: But it's THE nvac ;)
07:19RSpliet: pmoreau: well...
07:19pmoreau: The one who never works
07:19RSpliet: one option is figuring out how the hell the linebuffer works
07:20RSpliet: that should be useful for increasing bandwidth and obviate bandwidth spikes
07:20pmoreau: But I'm grateful to have received this laptop: probably wouldn't have ended doing Nouveau stuff otherwise
07:21RSpliet: so much time
07:21RSpliet: so much to do :D
07:21pmoreau: Yeah... :(
07:22pmoreau: zeael: Could you please open a bug report on bugs.freedesktop.org?
07:22zeael: 5000 lines do the trick
07:22zeael: Okay I will.
07:51zeael: filed it
07:51pmoreau: I'll test that tonight
07:55zeael: Okay. I really appreciate the effort you put into this, because as I said, only the latest patches made nouveau work for me! Can I provide more information?
07:57pmoreau: You're welcome. :) It took me quite a year to fix the issue on the NVAC, as I was completely new to Nouveau.
08:00pmoreau: RSpliet: Does the linebuffer configuration appear in mmio traces?
08:11zeael: By the way, is it safe to use pstate=1 and can I benefit from it?
08:12pmoreau: You can benefit from it by going to the lowest clock to save some battery, or go higher ones for more performances
08:12pmoreau: I guess it is safe, but RSpliet knows better as he added support for it :)
08:13pmoreau: (I didn't experience any troubles by using it)
08:14zeael: But at the moment there are no tools to set states dynamically, I suppose?
08:14pmoreau: Not yet
08:15pmoreau: Once hakzsam finishes adding perf counters, we should be able to use them to determine whether we should change clocks or not.
08:16RSpliet: zeael? yes, changing the clocks is safe
08:16RSpliet: we have excellent thermal support thanks to mupuf :p
08:16pmoreau: Maybe there is already support for automatically downlocking if the GPU runs too hot
08:16RSpliet: pmoreau: there is some hw support for that as far as I know
08:16RSpliet: but NVAC's don't overheat :D
08:17pmoreau: Right :)
08:18pmoreau: (Was refering to this https://trello.com/c/WKKbOXeI I guess it is not implemented yet)
08:18zeael: Hm, could have saved the life of the GPU in my deceased Thinkpad?
08:18zeael: It had the Quadro NVS140m which were known to die sooner or later
08:19zeael: And I ran nouveau but didn't set clocks
08:45zeael: Okay I'll log out then. I'd be glad to help if more information or testing is needed. Wish you much success!
09:38imirkin: Leftmost: if you're really brave, you could try tackling ARB_clear_texture or ARB_copy_image support in gallium. those are more involved and require clear thinking around interface design. but they'll get you to explore lots of gallium details, and won't be really hardware-dependent.
10:59Utal: i need a mentor for EVoC .....can anyone help me on this....
11:00imirkin: first you need a project
11:03Utal: ya i have checked out the gsoc ideas page and the todo list .... i have figured out one project ....but it will be helpful if u give me suggession ....
11:04imirkin: this is the current list: https://trello.com/b/ZudRDiTL/nouveau
11:05imirkin: [for nouveau, obviously]
11:07buhman: ooh, can https://bugs.freedesktop.org/show_bug.cgi?id=70354 be on gsoc? :3
11:08imirkin: buhman: ;)
11:09imirkin: trying any way you can, eh
11:09imirkin: i think the main trick would be getting the relevant hw in e.g. skeggsb's hands
11:09imirkin: i bet it'd get fixed in an hour
11:13Utal: can u suggest me any documentation that will help me get started with nouveau...i am quite new to this...
11:13imirkin: Utal: do you have any nvidia hw?
11:14Utal: ya ...........i have worked c and opengl
11:14imirkin: which nvidia hw do you have?
11:20Utal: imirkin:GeForce GT 755M
11:20imirkin: what is that? GK208?
11:22Utal: imirkin: GK208
11:24imirkin: Utal: well, an EVoC is like 3 months, right? if you have (or are interested in) compiler experience, you could try to get compute (e.g. opencl) working
11:26buhman: Utal: ^ dewit.
11:26imirkin: but it'll be something probably useful for ARB_compute_shaders as well
11:27Utal: i am interetest so can u suggest me any documentation...
11:27imirkin: unfortunately not really
11:32Utal: imirkin: r u a mentor
11:32imirkin: like do i have a card from the Mentor's Union as an accredited mentor? no.
11:33imirkin: but for EVoC's, pretty much anyone can be a mentor
11:34Utal: so what do u suggest me to work on ...
11:35imirkin: i suggest you find something you're interested in
11:37Utal: i was willing to work on some mesa related project...do u have any suggession on that........
11:44imirkin: Utal: i (nor anyone else i'm pretty sure) will come up with a project for you. you have to look around and find something you're interested in.
11:45imirkin: erm, that ended up being phrased funny. i meant to say that i won't be doing it, nor will anyone else.
11:45imirkin: once you do find something you're interested in, finding a mentor is unlikely to be an issue.
11:47Leftmost: imirkin, could do, but I think it may make sense to poke a couple very small things first so that I have an idea of how things fit together.
11:48imirkin: Leftmost: well that's the thing. i think in the end those extensions will be very little code.
11:48imirkin: Leftmost: but they _will_ require a very clear understanding of what's going on.
11:50Leftmost: Right. It's the latter I'm not sure I have.
11:50imirkin: you definitely don't
11:51imirkin: but making that stuff work will cause you to gain it.
11:52Leftmost: Fair enough. I assume that a good place to start would be looking at the bits implemented in mesa, other gallium extensions, and the specs.
11:53imirkin: Leftmost: well, pick one, read the extension, and then come back here and we can talk about it
11:53Leftmost: Alright, thanks.
13:02pmoreau: Too bad, the external screen doesn't get any signal from my laptop, though xrandr can see it.
13:03imirkin: how are you hooking it up?
13:03pmoreau: Guess I'll have to wait and use the DP -> HDMI adpator instead
13:03pmoreau: DP -> VGA adaptor
13:17pmoreau: I works now but only for a few seconds
13:17imirkin: you're on latest nouveau i assume?
13:18pmoreau: On the blob right now
13:18pmoreau: But otherwise yes, quite recent version of Nouveau
13:18imirkin: does it work with blob?
13:21pmoreau: I can get to work for a few seconds, then it drops and I need to reset it to access my windows (which become invisibles)
13:23pmoreau: Seems there were some fighting between nvidia-settings and xrandr :D
13:23pmoreau: And xrandr was the winner!
13:38pmoreau: So the blob is jumping from perflvl 0 to 2 when adding the external screen, good to know.
13:38imirkin: with multi-screen blob usually (always?) goes to highest perf level
13:39imirkin: coz you can't switch between vblank periods since the 2 screens aren't sync'd up
13:39imirkin: [without glitches]
13:39pmoreau: It was flirting between the perflvl 2 and 3, but after setting the ext screen down and up again, it stabilised at perflvl 2
13:40RSpliet: imirkin: you can if you properly configure the linebuffer
13:40imirkin: RSpliet: oh. neat.
13:41RSpliet: or well
13:41RSpliet: you don't need to switch between vblank if that's configured
13:41imirkin: it's big enough to handle a reclocking cycle?
13:42RSpliet: I have no idea, haven't figured out how the thing works, but this is apparently how NVIDIA handles it
13:42pmoreau: Shouldn't it be only related to memory clock only? Why up the graphics and processor clocks as weel, especially as it report a use of the GPU of 0%
13:42RSpliet: pmoreau: with older GPUs, NVIDIA link together the clocks in a perflvl
13:42imirkin: nvac has no memory speed though ;)
13:43imirkin: you should know that better than anyone :p
13:43RSpliet: NVAC doesn't flicker on reclock
13:43pmoreau: That's the point, why would you need to ups the other clocks?
13:50RSpliet: pmoreau: not sure what clock the copy engine is attached to
13:51RSpliet: and it still need to render the image for a large screen surface
13:51RSpliet: although their reason might just as well be "habit"
13:52RSpliet: perhaps even a bit in the VBIOS
13:52RSpliet: there's plenty of bits we haven't re'd yet
14:08pmoreau: Back to Nouveau, and no signal
14:09imirkin: pre-nva3 i think DP was a little dodgy
14:09imirkin: a blob trace lighting it up wouldn't hurt
14:10pmoreau: I should be able to do that
14:10imirkin: i guess there was an external encoder? not sure. i think some nva0's had that...
14:11imirkin: dunno about the mac situation
15:05pmoreau: Finally managed to get a trace with the external screen working
15:06pmoreau: Should I look for something or send it to mmio.dumps (or maybe even both :D )?
15:22imirkin: pmoreau: both
15:23pmoreau: I hope xz -9 will be enough to compress it
15:23imirkin: that's what i always do...
15:23imirkin: make sure to _not_ demmio it first
15:23imirkin: the demmio'd stuff is much bigger
15:24pmoreau: I had to try so many times to get the external display up, the trace ended up being 517MB
15:24pmoreau: The demmio version is only 30MB bigger
15:26imirkin: yeah, but xz'ing isn't as effected
15:27tobijk: hey pmoreau, you already had time to test my patch on your hardware? (though it shouldn't really be hw dependent) but a second confirm that does not regress something would be nice :)
15:28pmoreau: tobijk: Not yet. I can launch one test now that I don't need the monitor anymore
15:29tobijk: its not time critical, test when you're in the mood
15:29pmoreau: I can just launch it and go to sleep if I'm not in the mood ;)
15:30pmoreau: imirkin: What should I be looking for?
15:30imirkin: pmoreau: the MARK statement you put in before plugging the connector it
15:31pmoreau: I should have done that? O:-)
15:32imirkin: seems like it would have made reading the trace a whole lot easier
15:32imirkin: you can look for SOR/PIOR setting too
15:32imirkin: and/or read the nouveau code to see what _it_ does, and look for the blob doing comparable things
15:33imirkin: i think LVDS is set up in a different way, so it shouldn't be too confusing wrt other stuff
15:33imirkin: although maybe it's also handled via TMDS... i forget
15:33imirkin: (and/or never knew)
17:18RSpliet: imirkin: sorry for the misunderstanding
17:18RSpliet: shortForm is a bitmask of which instruction can emit 4-byte instructions. getMinEncodingSize has a shortcut in the top that checks the bit and if unset always returns 8
17:19RSpliet: I should have added MAD to the comment line above
18:56imirkin: RSpliet: yeah, i looked it up afterwards