bschaeferhello, getting a fun failed to mmap a buffer, only when running it through SDL: http://paste.ubuntu.com/7317028/. demo multiwin works fine :(03:26
bschaeferthe mir surface is created, and is valid03:26
duflubschaefer: Weird buffer dimensions/attributes? Tested other software buffer demos like "progress" or "fingerpaint"?03:34
bschaeferduflu, i tested outt he pixel formate, which was fine03:37
bschaeferduflu, alright03:37
bschaeferduflu, hmm they appear to be working, i check what other info in the surface creation bit i could be messing up03:39
duflubschaefer: Either that or a buffer/structure overrun could be corrupting some critical field03:39
bschaeferduflu, good idea!03:40
duflubschaefer: Please also update this :)  https://bugs.launchpad.net/mir/+bug/124794303:41
ubot5Launchpad bug 1247943 in Mir "Mir server failing to start when compiled for 32bit" [High,New]03:41
duflubschaefer: Oh, which driver?03:41
bschaeferduflu, that was me a bit ago, on intel03:42
* bschaefer updates03:43
bschaeferduflu, hows all else going?03:45
duflubschaefer: Busy. And this week is book-ended by two public holidays in Oz03:50
duflubschaefer: How's your end?03:50
duflusunny Seattle :)03:51
bschaeferduflu, raining atm :)03:51
bschaeferduflu, Oz, sounds like fun!03:51
dufluFollow the yellow... lawn?03:52
bschaeferduflu, summer coming to a nice warm end?03:52
* bschaefer is just now starting to hit some summer days03:52
bschaeferduflu, valgrid, i've not dug through it yet.... lots of info.. http://paste.ubuntu.com/7319558/03:55
bschaeferdid mir move to using ShmMemory?03:56
* bschaefer doesn't remember that being around a while ago03:56
duflubschaefer: Just search for your own code to narrow it down. Then again, a simple overrun clobbering just Mir structures would not be detected as an overrun within the context of the client process :P03:57
duflubschaefer: Yeah Mir's software buffers implementation changed some months ago03:57
bschaeferduflu, ill check it out :)03:57
bschaeferit has to be my issue haha03:57
bschaeferduflu, thats for the help!03:57
duflubschaefer: No problem03:58
dufluWhee, new Ubuntu release, new flood of teething bug reports04:12
racarr__UTOPIC UNICORN04:46
RAOFuTopic unicorn :)04:52
dufluWhoa. Utopic Unicorn05:39
* duflu wonders what this means for Ubuntu and its relevance to reality05:40
=== chihchun_afk is now known as chihchun
anpokRAOF: ping07:57
RAOFanpok: Pong07:57
anpoktwo days ago it was unsure wether your 1hz branh would make it in so I too the timer parts07:58
anpokand proposed them as a separate branch07:58
anpoknow it seems like the 1hz branch would make it after 0.1.907:58
anpokor I am not sure..07:58
anpokI need the functionality for the input stuff07:59
anpokand did not want to duplicate anything07:59
RAOFI can rebase on your branch if you like.07:59
anpokok - since i would adress findings as far as possible that come up in https://code.launchpad.net/~andreas-pokorny/mir/add-timer-to-main-loop/+merge/21688108:00
anpokthat would be cool08:00
RAOFanpok: Yeah. Go for your life; address Alan's concerns, and I'll merge on top.08:00
* RAOF -> EOD08:03
dufluHmm, my monitor exposes 50Hz modes on DisplayPort but not DVI?08:14
dandraderanpok, got the qt compositor working with lp:~andreas-pokorny/mir/input-sender-split12:28
dandraderanpok, good work!12:28
anpokgreat news!12:42
kgunndandrader: anpok ....that is really nice work!12:52
kgunnanpok: think we could try to get that landed onto dev branch in the next few days ?....it'd be really great if we could bring that along with our 0.1.9 tag12:53
kgunnnote: i'm not going to tag until we address a couple of regressions12:53
anpokah ok - i thought we would wait for after 0.1.9. I will try to clean things up12:54
kgunnanpok: i'll take your lead, i would like it in 0.1.9 but if you think it injects too much risk that's worth considering...12:57
anpokthe way it is built right now, makes it work kind of in parallel to the existing dispatcher code. hence the risk is low..12:58
anpokof course at some point i will remove the old send path..12:58
kgunnanpok: thanks...that's good to hear, since we're at the beginning of cycle i'd like to get it into 0.1.9, risks are easier to swallow :)13:04
kgunnso its worth waiting on for a day or 2 for sure13:04
=== dandrader is now known as dandrader|bank
alf_kgunn: which regressions do we need fixes for before 0.1.9?14:13
kgunnalf_: we have this list...14:14
kgunnbut in reality there are 2 that i'd consider musts14:14
kgunn....and this one kinda troubles me 129451014:15
alf_kgunn: 1308843 is not as bad as it sounds, and doesn't seem to affect the phone... I would propose ignoring it for 0.1.9, because it is going to be fixed by the "1Hz" branch, but we can't land the 1Hz branch for 0.1.9 because we will need to become more confident in it.14:19
alf_kgunn: unless we say that we postpone 0.1.9 for some time until 1Hz has been tested properly14:23
kgunnalf_: hmmm, i understand now...it does stink b/c we definitely want to land this sooner than later14:26
kgunnalf_: how come it doesn't manifest on the phone ? nested ?14:26
alf_kgunn: I can't see it on any of my computers, so I don't have first hand experience, but my understanding is that the effect this has depends on the vsync timings of the outputs (in this case the real output and the fake 60Hz one)14:31
kgunnalf_: right, i was kind of assuming that the fake 60 hz will never be perfect for any display, timing will always be off somewhere14:32
kgunni guess, if the fake is a bit slower than the screen you're  ok?14:33
=== alan_g is now known as alan_g|tea
alf_kgunn: Right. So one thing we could potentially do for 0.1.9 and should be relatively safe is reduce the fake output rate to something sufficiently small e.g. 10Hz.14:45
racarr__Guten morgen14:46
alf_kgunn: oops, that won't work14:46
anpokracarr__: Guten Tag14:47
racarr__*races breakfast with standup*14:48
=== alan_g|tea is now known as alan_g
josharensonkgunn, current graphics dashboard is broken down by resolution and GPU manufacturer. Keep that format, or categorize by device?16:36
=== dandrader|bank is now known as dandrader
kgunnjosharenson: mmm, can we expand ?...to include formfactor {phone, tablet, desktop}17:00
kgunnwe should keep screen res17:00
kgunnand gpu17:00
kgunnfor sure17:00
kgunnor... if not formfactor, device name.... mako, manta, macbook, dell###17:01
josharensonkgunn, I can probably add that... the website looks like it was originally written to show desktop performance17:01
=== alan_g is now known as alan_g|EOD
josharensonits organized only by resolution and gpu, but can probably add anything17:02
josharensonkgunn, I'll work on resolution and gpu for now (cause thats what there) and add other criteria later. I'll let you know if there are issues with that...17:03
kdubalf_, once I fix https://code.launchpad.net/~kdub/mir/plumb-android-shader-creation/+merge/216779/comments/515830, would you be okay with a top-approve?17:07
racarr__A moment of silence for null cursor *bows head*17:24
josharensonrobotfuel, I'm working on extending the qa-dashboard/graphics page and was wondering what the significance of the "ps" prefix to the names of all gpus?17:25
robotfueljosharenson:  it's product-strategy the name of the group that owned the machines.17:30
kgunnalf_: hey, so looks like i can tag the tip of devel...17:31
kgunnanything else outstanding that needs to merge ?17:31
kgunnanyone ^17:31
josharensonrobotfuel, ah thanks... any reason to preserve that convention? even if other machines?17:31
kgunnotherwise i'll do it this afternoon....and apologize to duflu at the same time :)17:31
robotfueljosharenson: no it was the name of the machine, that's all. ps was absorbed in to ubuntu engineering.17:32
josharensonrobotfuel, ack17:33
kdubkgunn, I'm okay to merge17:37
racarr__for some reason that breaks my brain17:41
racarr__(random x server function lol)17:41
kdubthey just mis-spelled meek... see meekEnqueue17:41
kdublike a polite equeuement, obviously17:41
racarr__common locking pattern, reader, writer, and meek writer.17:45
AlbertAkgunn: https://code.launchpad.net/~vanvugt/mir/fix-1308941/+merge/21702117:46
AlbertAkgunn: that should probably land so 1308941 is taken care of17:46
kgunnAlbertA: thanks...i'll keep an eye on it17:52
racarr__yay shift key now outputs18:07
racarr__m6 hwich must mean18:07
racarr__xmir input is close to working18:07
racarr__as in now just mapping is off18:07
racarr__very off apparently18:07
racarr__but besides that getting close lol18:07
anpokracarr__: mapping as in mapping of different keyboard types?18:14
racarr__well just somewhere int he mapping of scan code to key code18:30
racarr__something has gone wrong18:30
racarr__as usual the answer was add or subtract 8 from scan code18:32
racarr__in this case its add 8 when going from mir scan code to x scan code18:32
AlbertAracarr__: can I add that comment on all bugs? "Did you try adding 8?" ;)19:09
AlbertAor 42 could work too...19:09
racarr__AlbertA: on all input related bugs I think its a fair comment lol20:33
racarr__probably top 10 lunches of the year so far.20:35
=== dandrader is now known as dandrader|afk
=== dandrader|afk is now known as dandrader
josharensonrobotfuel, just wanted your opinion... after looking at all the code, it seems creating a standalone 'performance test' category, rather than nesting under 'graphics' seems like a good idea... thoughts?21:16
racarr__hahaha watching the xmir cursor follow21:23
racarr__the mir cursor21:23
racarr__is my new hobby21:23
racarr__in other news increasming amounts of xmir input21:23
kgunnjosharenson: camako and i were talking about comparing mir & SF21:38
josharensonkgunn ok21:38
kgunnquestion is, do you know if Qt on the QPA for SF just renders as if the Qt engine were a full screen app on SF ?21:38
racarr__Everything (including the shell) is in fact a full screen app on SF right?21:39
racarr__no in process shell or any such..suchness21:39
josharensonkgunn, are you trying to determine something about bypass?21:40
josharensonkgunn, cause I'm not sure, but if you dump layer contents, it should be obvious21:40
kgunnjosharenson: not really about bypass...more like trying to satisfy the question of is mir on par with sf21:40
josharensonkgunn which I'm not sure how to do in ubuntu.... only android21:40
kgunnkdub: would know21:41
kgunnracarr__: true...no in process shell21:41
kgunnracarr__: also true...no nested21:41
kgunnmakes comparing more difficult me thinks21:41
kdubkgunn, I know the stuff we have to improve, yes21:41
robotfueljosharenson: there are already people working on performance tests, ping cgoldburg or nuclearbob21:42
kdubkgunn, or did you mean something else?21:42
kgunnkdub: something else...but i think racarr__ negated it21:42
kdubright, at the end of the day, its tough to get an apples-to-apples of the whole stack21:43
kgunnrobotfuel: do you know what kind of performance tests ? are they to compare sf & mir ?21:43
kduband some of the key differences are in driver details that are abstracted from us21:43
robotfuelkgunn: no all kinds of performance tests.21:43
kdubso its tricky to negate the goodness or badness of the particular driver21:43
kgunnkdub: camako josharenson ...i'm starting to wonder, would a simple native gl test written specifically native to sf and then native to mir be the only wya ?21:43
robotfuelkgunn: not sf vs mir21:44
camakokgunn: So we are measuring graphics perf?21:45
kdubkgunn, even at that, there's some serious headaches21:45
kdublike if you have "setprop debug.mdpcomp.maxlayer" set to two vs four on mako, you might get different results that take a while to figure out21:46
kdubor if you happen to run on tegra3, the hybris pthread issue might bite21:46
camakokdub: what is mdpcomp.maxlayer switch for?21:47
camakohw overlays?21:48
josharensonwhat exactly are you trying to measure? just general performance?21:48
kdubits just a mako switch for if the hwc happens to select the mdp core, to limit the number of layers it will accept :)21:48
kdubjust making the point that there are sorts of little gotchas there that a quagmires to sort out21:49
camakokgunn: any non-trivial graphics apps should produce the same result (fps?) on mature platforms21:50
camakoemphasis on "should" and "mature"21:50
kdubcamako, right, agree21:51
kdubbut also, the result is fps+power21:51
camakokdub: yes, cpu bw, mem bw, etc...21:51
camakoI guess without messing w any switches, comp at gl level would make sense to answer "is mir up to par with Android?"21:53
camakoAssuming we don't already know the answe :-)21:53
racarr__We dont know the answer wrt power.21:53
racarr__But of course both are capable of rendering rectangles at 60hz in isolated conditions21:54
camakoI see21:54
racarr__interesting sort of questions to me are like21:54
kdublike, and the other side is, are we capable of explaining any deltas?21:54
racarr__say same EGL app rendering at vsync on mir and SF and measure power21:55
racarr__hmm? You mean deltas over time or deltas21:55
racarr__v. surface flinger21:55
kdubvs surfaceflinger21:55
camakothen I guess gl perf would make sense21:56
camakogl perf comp*21:56
camakojosharenson: kgunn mentioned you were (or would) work on this, have you started this comp effort bw sf and mir?21:57
kdubbut like, do we have control over the gl drivers? we're just comparing (switching logic + ipc roundtrip) then21:58
josharensonI have a test that runs glmark2....21:58
josharensoncamako, but if we just need a simple gl app...21:58
camakokdub, correct... Do we care to have control over the gl drivers since what we develop is outside those22:00
racarr__ok if you ignore everything that isnt a 3button+2axis mouse or a us english keyboard22:01
racarr__and dont click on buttons at the same time22:01
racarr__xmir input is workin perfectly :D22:01
camakokdub, like a stock system-level comparison.. what an avg user would experience...22:01
racarr__as part of the overall process of ubuntu22:02
racarr__its good to have gl performance benchmarks22:02
kdublike, we can plumb mir/unity8 to get that information, and its good for analyzing, but to compare to surfaceflinger22:02
racarr__in terms of comparing mir to other compositors22:02
racarr__I dont think its useful22:02
kdubwe have to plumb surfaceflinger at the same points22:02
camakofps, cpu %, power consumption don't require mir/unity specific plumbing, do they22:05
kdubwell, not to measure those, but if we investigate, and figure out the ipc is slow or something, we do22:06
camakokdub, I agree that comp at that granularity is not useful.22:08
camakokdub: I don't know what info kgunn wanted to get out of it, but I think it'd be useful to get those readings for a simple app on mir vs SF for comps22:08
camakoOr may be not... :-)22:08
camakothose readings being fps, cpu%, power22:10
kdubsure, but we're already in the ballpark though22:10
camakokdub, someone measured already?22:11
kdubcamako, measured what? :)22:15
=== chihchun is now known as chihchun_afk

Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!