02:21 pmoreau: imirkin: Updating my userspace graphic stack to Git version. I'll retest Portal, see if I can reproduce the misrenderer portals on my G96 and send you screenshot + apitrace. :-)
04:05 pmoreau: imirkin: http://imgur.com/dG5lBBt
04:23 pmoreau: How does one make a useful apitrace? I made one which consists of one frame + its context but it ends up being ~485k calls...
07:18 hakzsam: mwk, --^
07:18 pmoreau: hakzsam: Cool!
07:18 hakzsam: yep
07:19 hakzsam: I started this tool two years ago, it's time to see it in envytools :)
07:23 pmoreau: :)
08:00 pmoreau: imirkin: Here is the apitrace: https://phabricator.pmoreau.org/F5697 Tell me if I need to trim it further or need more info
08:16 joi: hakzsam: parsing demmt output from C is... sad
08:19 hakzsam: joi, sure :) but if you tell me how to improve that part using demmt functions, I'll do
08:19 hakzsam: it was much simpler for me to do it that way
08:22 joi: I'd vote for merging it and fixing it later
08:23 hakzsam: sounds good
08:24 hakzsam: so, demmt exports some functions to parse a file directly? I didn't take a look at this
08:24 hakzsam: but It would be good to get rid of popen() ;)
08:24 joi: yeah, parsing core is shared between demmt and mmt_bin2dedma
08:25 hakzsam: ok good
08:26 hakzsam: once the pull request is merged, I'll fix it
08:34 joi: hakzsam: license/copyright headers are missing
08:34 joi: and cupti_trace.c
08:34 joi: and cupti_trace.c header is worrying...
08:37 hakzsam: why do you think it's worrying? this tool is based on cupti which is provided by NVIDIA..
08:41 joi: it's worrying because it does not tell where this code came from
08:42 joi: did original file have a license header?
08:45 hakzsam: I started this tool from scratch but some parts are inspired from cupti_query which has a one-line license header
08:45 hakzsam: I don't know if it needs to be added or not, I'm not a lawyer :)
08:47 joi: can I see the original file?
08:47 hakzsam: sure
08:48 hakzsam: http://paste.awesom.eu/rljG
08:49 joi: heh, it doesn't even have a license...
08:49 hakzsam: but what it's that copyright stuff then?
08:50 joi: it just states who is the copyright owner
08:51 hakzsam: so, we have the right to redistribute the code?
08:51 joi: in theory, if there's no other file which tells how it's licensed, you are not allowed to use it
08:53 hakzsam: in practice?
08:56 hakzsam: with the header I added to cupti_trace.c, everything is fine, right?
09:00 joi: no, every code file should have a copyright and license
09:00 hakzsam: joi, ?
09:00 hakzsam: cuda_sample.cu too?
09:00 joi: yes
09:02 hakzsam: joi, like this for cupti_trace.c http://paste.awesom.eu/Husf ?
09:04 joi: without license it's useless - nobody can use it...
09:04 pmoreau: Isn't without license == no license == do whatever you want with it?
09:05 mjg59: No
09:05 mjg59: Without license == no license == you can only do what local copyright law says you can do with it
09:05 pmoreau: Hum
09:06 hakzsam: joi, okay, so that tool has to be private ? it cannot be merged to envytools?
09:06 mjg59: The license grants you permissions that you wouldn't otherwise have
09:07 joi: hakzsam: you could contact nvidia and ask what is the license for this file
09:07 pmoreau: Ok
09:08 hakzsam: mmh okay
09:09 hakzsam: and do I need to remove the pull request since it's public now?
09:11 joi: I think closing it should be enough
09:18 hakzsam: but well, if I don't have the right to write source code which uses the cupti API, that's makes no sense
09:20 hakzsam: anyways, I could remove the query stuff part of cupti_trace
09:52 imirkin: hakzsam: where is the original from? there is probably a LICENSE file around it
09:53 hakzsam: /opt/cuda/extras/CUPTI/sample
09:53 hakzsam: and there is no file around for CUPTI samples
09:53 hakzsam: there is one for CUDA
09:56 imirkin: what package did that CUPTI dir come from?
09:56 hakzsam: cuda
09:57 imirkin: hrmph
09:58 hakzsam: don't worry, I'm removing the parts of that tool which uses code from NVIDIA ;)
10:00 imirkin: cool
10:00 imirkin: licensing stuff sucks :(
10:00 hakzsam: yes
10:01 imirkin: pmoreau: please make that trace available somewhere
10:01 imirkin: pmoreau: please also try older versions
10:01 imirkin: (of mesa)
10:01 pmoreau: Isn't the link working?
10:01 imirkin: oh. heh. it is. my brain isn't :)
10:02 pmoreau: :D
10:02 pmoreau: How old should I try?
10:02 imirkin: but do try older versions of mesa to see if this is something i broke recently
10:02 imirkin: oh, andd.... try the *very* latest git version
10:02 imirkin: i.e. after that query fix
10:03 imirkin: i dunno, mesa 10.1 should be good to try out
10:03 pmoreau: Ok
10:03 imirkin: if that's broken, try 9.2
10:03 pmoreau: How recent was the query fix? I pulled about 8-9h ago
10:03 pmoreau: Ok
10:03 imirkin: oh, it was older than that
10:06 imirkin: [and i have portal now, so i can test it myself too]
10:07 pmoreau: [But you would have to plug-in your card :p ]
10:07 imirkin: yes
10:07 imirkin: i should have done it when i rebooted the other day
10:08 imirkin: but... i have these hopes that i'll get back to tess at some point, and maybe images/counters, which require the nvc0
10:08 pmoreau: Well, having tess would be great too! Even if I won't be able to use that on my cards.
10:09 imirkin: i have a branch for it... waiting on the core bits to land
10:10 imirkin: glennk: any idea how portal works wrt showing the things inside its portals? is it just an extra render from the viewpoint of the other portal?
10:27 glennk: imirkin, try stepping through an apitrace frame
10:28 imirkin: glennk: i was hoping you'd know off-hand. that's what i'll do...
11:00 RSpliet: ugh... my motherboard hates NV50
11:01 RSpliet: unless it has more hidden power plugs I missed
11:01 imirkin: did you plug in enough PCIe power connectors?
11:01 imirkin: it needs like 37 of them
11:01 imirkin: just hook it up to a nuclear power plant, that should be enough
11:03 imirkin: The 8800 GTX was also very power hungry for its time, demanding up to 185 watts of power and requiring two 6-pin PCI-E power connectors to operate.
11:03 imirkin: heh
11:03 RSpliet: yeah, this GTS one only has one plug
11:04 RSpliet: (the NVA0 has 2... so in that perspective yes, 2 * NV50 = NVA0)
11:04 imirkin: ;)
11:04 RSpliet: there's a red light glowing on the GPU
11:05 imirkin: is it fire?
11:05 RSpliet: well, I hope it's just the warning light that it's running low on petrol
11:05 imirkin: pour some more in maybe? :)
11:06 RSpliet: wait... is there seriously a "low on power" light on there
11:07 pmoreau: There are one led on each of the 980's power plug: red if it's not connected, white otherwise
11:07 pmoreau: s/are/is
11:07 RSpliet: heh, well, this led is on the other side of the card, and... I'm fairly sure it's really really really plugged in
11:08 pmoreau: Maybe you have something like that?
11:08 pmoreau: :)
11:08 tobijk: if it boots you should be allright :D
11:08 RSpliet: it doesn't
11:09 imirkin: moar power!
11:09 tobijk: heh
11:10 RSpliet: it's already getting max power
11:10 RSpliet: meh
11:11 RSpliet: so far for my evening
11:11 imirkin: max power doesn't abbreviate!
11:11 imirkin: each letter is as important as the previous one! no, more! no... as.
11:12 RSpliet: *does not
11:25 hakzsam: imirkin, joi, I'm going to ask github staff to remove the previous pull request, the new one doesn't contain any source code provided by NVIDIA
11:25 hakzsam: https://github.com/hakzsam/envytools/commits/cupti_trace
11:25 hakzsam: but I still have to write a script for tracing all events
11:27 hakzsam: because I can't list cupti events anymore :)
11:31 hakzsam: request sent
11:31 hakzsam: bbl
11:38 hakzsam: oh, github staff is really fast :)
15:08 imirkin: pmoreau: hm, well that trace replays fine for me on nvc0
15:08 imirkin: pmoreau: i guess that's expected
15:08 pmoreau: It replays fine on my MCP79 as well; just the G96 doesn't want to display it correctly
15:09 imirkin: wtf
15:09 imirkin: that sucks.
15:09 pmoreau: yup
15:10 pmoreau: I'll try tomorrow some old Mesa version