| + nemunaire (~nemunaire@82-64-151-41.subs.proxad.net) | 00:39 | |
| chartreuse | Personally I would have wised for more ram, perhaps with a sodimm socket. But I understand that an ideal open hardware system module wasn't out there. | 00:40 |
|---|---|---|
| chartreuse | *wished | 00:40 |
| chartreuse | 4GB should be plenty enough for now, even if my current laptop runs 16. | 00:41 |
| mntmn | chartreuse: yeah, with what i'm doing with reform i haven't really had a problem with the 4 GB, but others might. i'm positive that future SoCs/SoMs can get us more memory. | 00:43 |
| chartreuse | Yeah, that's a really great feature with the reform, being able to swap out the system module. Maybe it needs a 6502 module instead :) | 00:44 |
| mntmn | :D | 00:45 |
| mntmn | i wanna make (or maybe someone else will be faster) an FPGA module, with that you could experiment with any available smaller cpu core | 00:46 |
| swivel | my x220 would OOM with some regularity on 4G, just firefox with too many tabs | 00:58 |
| - embden (QUIT: Ping timeout: 260 seconds) (~embden@mue-88-130-57-065.dsl.tropolys.de) | 02:02 | |
| Jookia1 | congrats on the funding | 02:19 |
| Jookia1 | isn't the whole point of virtual memory and swap that you don't need tons of RAM | 02:26 |
| swivel | swap sure, but I don't run any swap on my flash-based systems | 02:29 |
| swivel | even without that my ssd is already pretty aged according to the smart info | 02:29 |
| ex-parrot | I generally like to run a small swap and set vm.swappiness to low | 04:15 |
| ex-parrot | best of both worlds then | 04:15 |
| swivel | nowadays i at least run my web browser in an nspawn container which makes resource control a lot easier, i limit its memory to 2G and it'll thrash and OOM without affecting the rest of the system | 04:18 |
| swivel | but i also have 16G on the x220 now | 04:18 |
| ex-parrot | last year I got a nitrocaster kit for my X230 | 04:20 |
| ex-parrot | very good upgrade but fiddly | 04:20 |
| swivel | ive been meaning to swap my x220 keyboard onto the x230 i have just haven't gotten around to it, it'll be nice to have a vulkan-capable laptop | 04:20 |
| swivel | what resolution display did you get with the nitrocaster? | 04:21 |
| ex-parrot | sadly I actually like the X230 keyboard | 04:21 |
| ex-parrot | 1080p | 04:21 |
| swivel | nice, lots better than the original display | 04:21 |
| ex-parrot | yeah, so much better | 04:21 |
| swivel | my hands are pretty small so both the x220 and x230 keyboards suck for me, but i'll always prefer the classic thinkpad key style | 04:22 |
| swivel | the X40 was the best keyboard for me, hoping the reform will be more like that | 04:23 |
| ex-parrot | mmm the T/X 40 was (is?) pure bliss | 05:06 |
| swivel | i still have my X40 but haven't used it for anything in ages, if it had a SXGA+ LED-backlit display i'd prolly still use it for coding | 05:12 |
| ex-parrot | I swapped a 1050 panel in to mine | 05:16 |
| ex-parrot | wait no, I swapped the 1050 panel in to the T60 from a T40 | 05:16 |
| technomancy | such a good panel | 05:17 |
| technomancy | legendary | 05:17 |
| ex-parrot | for the time certainly | 05:17 |
| ex-parrot | mine has got fairly dim now | 05:17 |
| ex-parrot | I had to add an i2c PROM to it for the edid as well | 05:17 |
| swivel | i did the swap on my old X61S from the SXGA+ tablet, even replaced the CCFL with LED, but the results were mediocre and after a couple years it failed | 05:20 |
| ex-parrot | :( | 05:21 |
| swivel | it's a perfect resolution for that size though, i just wish they were available from factory w/LED | 05:22 |
| ex-parrot | agreed | 05:22 |
| technomancy | decent brightness, reasonable resolution, matte; pick any 2 =\ | 05:24 |
| ex-parrot | the panel I got in my X230 now is amazing | 05:28 |
| ex-parrot | and it just came from aliexpress | 05:28 |
| ex-parrot | the original novena panels were quite nice too | 05:28 |
| technomancy | not a big fan of the aspect ratio but other than that it's good | 05:49 |
| ex-parrot | mm the 1050 panels in the T series have the best aspect ratio :) | 06:10 |
| + jfred (~jfred@li1925-247.members.linode.com) | 06:19 | |
| + B[] (~Thunderbi@125-239-58-43-fibre.sparkbb.co.nz) | 06:21 | |
| * Jookia1 -> Jookia | 06:46 | |
| Jookia | swivel: you can use systemd-run to limit resources since it puts it in its own cgroup | 06:46 |
| swivel | Jookia: yeah, there's some feature overlap there w/systemd-nspawn, i like using nspawn containers for a number of reasons, and have kind of stopped installing much directly on the host | 06:48 |
| Jookia | ah ok | 06:48 |
| swivel | it's nice to run something internet-facing like firefox where it can't access my home dir/ssh keys etc ;) | 06:48 |
| - paulk-leonov (QUIT: Ping timeout: 246 seconds) (~paulk-leo@leonov.paulk.fr) | 06:50 | |
| + paulk-leonov (~paulk-leo@leonov.paulk.fr) | 06:51 | |
| Jookia | yeah i use multiple users for that atm | 06:58 |
| - Jookia (QUIT: Ping timeout: 240 seconds) (~Jookia@gateway/tor-sasl/jookia) | 09:11 | |
| + Jookia (~Jookia@gateway/tor-sasl/jookia) | 09:13 | |
| + embden (~embden@mue-88-130-57-191.dsl.tropolys.de) | 10:09 | |
| - B[] (QUIT: Ping timeout: 256 seconds) (~Thunderbi@125-239-58-43-fibre.sparkbb.co.nz) | 13:14 | |
| - erlehmann (QUIT: Ping timeout: 264 seconds) (~erlehmann@46.114.39.202) | 16:29 | |
| + erlehmann (~erlehmann@46.114.33.49) | 16:41 | |
| ndufresne | mntmn: just saw you report with the HW decoder, quite some progress, simple question, do you get these glitches with all videos ? what ezequiel is proposing to to record the decoding into a file, that is simply to rule out potential GL issues | 17:09 |
| ndufresne | mntmn: I know @cphealy carries some patches for implicit fences, so there is known glitches when doing v4l2/gpu there, but it didn't look like this | 17:10 |
| mntmn | ndufresne: oh i had similar glitches with kms-sink, so i’m confident these are decode errors. you can see smearing of motion vectors etc | 17:12 |
| mntmn | ndufresne: these videos come from ffmpeg’s standard settings when you encode to .mp4 with libx264 | 17:12 |
| ndufresne | if you have a link to a sample, I can git it a try here | 17:13 |
| ndufresne | would be nice is this issue is also visible through ffmpeg | 17:14 |
| ndufresne | this is of course all very early, I'm basically picking up broken files, one after another and fixing | 17:14 |
| mntmn | sure, let me see if i can make you a minimal example video | 17:18 |
| ndufresne | cool | 17:18 |
| ndufresne | you already got Khodi running on top of https://github.com/Kwiboo/FFmpeg ? | 17:19 |
| mntmn | ndufresne: when i use gst-build, that also seems to yield libavcodec and friends, right? how can i build the ffmpeg tools in that environment? | 17:21 |
| mntmn | (my system ffmpeg doesn't work when the gst-build devenv is active, crashes) | 17:22 |
| ndufresne | unless you pass --wrap-mode=nofallback to meson, or if you have ffmpeg-devel already, it will use some ffmpeg port, but it does not have anything v4l2 or HW accelerated in the build | 17:22 |
| ndufresne | we don't support ffmpeg HW accel in gst, as the buffer management was too weak until recently | 17:23 |
| ndufresne | we could possibly change, but usually we have more flexible implementation | 17:23 |
| ndufresne | "work when the gst-build devenv is active, crashes", would need to be looked at, works on Fedora AARCH64 here | 17:24 |
| mntmn | ndufresne: here is a video that i just generated purely with ffmpeg that is not decoded correctly http://dump.mntmn.com/testvid.mp4 | 17:25 |
| mntmn | regarding ffmpeg crash, probably debian bullseye's ffmpeg is a bit too old and i should build it myself | 17:26 |
| mntmn | ndufresne: the video was generated using the command ffmpeg -f lavfi -i testsrc=duration=10:size=1920x1080:rate=25 testvid.mp4 | 17:27 |
| ndufresne | mntmn: thanks, I'll give that a look, see if reproduce first | 17:45 |
| ndufresne | mntmn: would you mind doing an extra test ? gst-play-1.0 --videosink="capsfilter caps=video/x-raw,format=YUY2 ! glimagesink" testvid.mp4 ? | 18:00 |
| ndufresne | hmm, wait, might need playbin3 for that to work | 18:01 |
| ndufresne | mntmn: in gst build, when the build is gone, you do ./gst-uninstalled.py, that give you a shell to use what you just built, if ffmpeg was build as a subproject, then the related plugin will be available | 18:03 |
| ndufresne | just do gst-inspect-1.0 libav | 18:04 |
| ndufresne | you'll see all ffmpeg based features | 18:04 |
| mntmn | ndufresne: thanks! were you able to play back my test video? | 18:06 |
| ndufresne | not yet, working, will happen after lunch | 18:08 |
| ndufresne | it's nice an sunny, got a go out, spending too much time inside, you all know why | 18:08 |
| buckket | /scripts | 18:09 |
| + cphealy (~cphealy@2605:e000:160b:911f:4ffb:a92e:a44b:20d3) | 18:10 | |
| mntmn | ndufresne: haha yes indeed | 18:10 |
| mntmn | buckket: script scrapt | 18:11 |
| + B[] (~Thunderbi@125-239-58-43-fibre.sparkbb.co.nz) | 18:31 | |
| mntmn | ndufresne: the problem is related to h264 profiles. i just reencoded the video to baseline profile and it decodes fine then. | 18:43 |
| + cyrillu[m] (cyrillumat@gateway/shell/matrix.org/x-kgmrzzbqkanycmgz) | 19:12 | |
| ndufresne | mntmn: so it's clean here with ezequielg kernel | 19:33 |
| ndufresne | for the reference, gst-launch-1.0 filesrc location=videos/testvid.mp4 ! parsebin ! v4l2slh264dec ! videoconvert ! jpegenc ! filesink location=test.jpeg | 19:33 |
| ndufresne | We'll check that out, I suspect some patches didn't land yet | 19:33 |
| ndufresne | (getting patches into the kernel is a really slow process, each patches takes months) | 19:34 |
| mntmn | ndufresne: oh, my test video works on your end, yeah? | 19:37 |
| ndufresne | yes | 19:38 |
| mntmn | wow ok | 19:38 |
| ndufresne | mntmn: so you are on Linus tree ? | 19:38 |
| mntmn | ndufresne: yes | 19:41 |
| mntmn | from around ~ apr 29 | 19:42 |
| ndufresne | so I'm building 5.7 rc4 here to see | 19:44 |
| mntmn | cool | 19:45 |
| mntmn | ndufresne: this is what i get with your command http://dump.mntmn.com/test.jpg | 19:46 |
| mntmn | ndufresne: the buffer includes content from another video i played previously :D | 19:46 |
| ndufresne | cute | 19:46 |
| mntmn | indeed | 19:47 |
| ndufresne | hmm, hantro does not even probe on linus master here | 20:02 |
| mntmn | in my case that needed some non-obvious kernel config option | 20:03 |
| ndufresne | yes, the config issue is fixed in linuxtv/master now | 20:05 |
| ndufresne | but here the DTS changes for imx8mq evk board is not in yet | 20:06 |
| mntmn | ndufresne: ok, do you need them? | 20:06 |
| ndufresne | I got them in another branch, I'm just susprised these two things didn't make it in the same merge window | 20:07 |
| mntmn | ok :) | 20:07 |
| cphealy | You need staging drivers and then you should see: CONFIG_VIDEO_HANTRO_IMX8M | 20:09 |
| + ezequielg (sid363064@gateway/web/irccloud.com/x-epvwplxwfjmvfqqz) | 20:13 | |
| ezequielg | so we have hantro :) | 20:13 |
| ezequielg | \o/ | 20:14 |
| mntmn | :D | 20:14 |
| mntmn | yes | 20:14 |
| ndufresne | you had to pull some CAMERA config at some point, that was a bug though | 20:14 |
| ndufresne | the .jpeg is bigger, that's a bad sign ;-P | 20:15 |
| ezequielg | mntmn: what's your kernel tree? would be nice to check if we have any fixes or anything. | 20:15 |
| ndufresne | (currupted output don't compress well) | 20:15 |
| ndufresne | ezequielg: that's what I was looking | 20:16 |
| mntmn | ezequielg: linus master from 2 weeks ago with some patches on top, https://gist.github.com/mntmn/c1005bd9f4cbc53e71058f0a45084da3 | 20:16 |
| ndufresne | so I got linus master from today running, pulling the result now (my board is in the office) | 20:16 |
| ndufresne | mntmn: I confirm that decoding is broken on 5.7 rc5 | 20:17 |
| ndufresne | so now the question is which patches are needed | 20:18 |
| ezequielg | mntmn: define "broken" :) | 20:18 |
| ndufresne | ezequielg: unless you have a 5.7 rebase handy ? | 20:18 |
| ndufresne | ezequielg: very corrupted ;-P | 20:18 |
| ezequielg | on all profiles/samples ? | 20:18 |
| ezequielg | any other hint? | 20:18 |
| mntmn | ezequielg: Baseline seems to work | 20:19 |
| ndufresne | no idea, just tested one file | 20:19 |
| mntmn | (i mean for me) | 20:19 |
| ndufresne | I have a display less running remotely board for testing, so it takes a bit of time | 20:19 |
| mntmn | High 4:4:4 Predictive and High did not work for me | 20:19 |
| ndufresne | High 4:4:4 is not supported at all | 20:20 |
| ndufresne | 4:2:2 is not implemented, only 4:2:0 is currently implemented | 20:20 |
| ndufresne | and now I notice that the testvid.mp4 is high 4:4:4... how come it worked ... | 20:22 |
| ezequielg | mntmn: might be a good idea to push your sample. | 20:23 |
| ndufresne | he did | 20:23 |
| ndufresne | 11:25 <@mntmn> ndufresne: here is a video that i just generated purely with ffmpeg that is not decoded correctly http://dump.mntmn.com/testvid.mp4 | 20:24 |
| ezequielg | doh. | 20:24 |
| ezequielg | oky doky -- i have to finish a few things, but i can take a look shortly. saturday at most. | 20:24 |
| ndufresne | interesting, it's not the same file as I downloaded few minutes ago | 20:24 |
| mntmn | ok? :D | 20:25 |
| mntmn | the video with glitches/smears: High, yuvj420p(pc) | 20:26 |
| mntmn | the high 4:4:4, testvid.mp4, doesn't render at all, it just generates noise for me (as expected?) | 20:26 |
| ndufresne | yes, so high 4:4:4 is not support by the g1 at least | 20:32 |
| mntmn | ok | 20:32 |
| ndufresne | I have discussed this issue with ezequielg, we think we need a patch to make this hard failure, and also to expose the supported profiles | 20:32 |
| ndufresne | and I'll add the code in gst too, so it can fallback to software | 20:33 |
| mntmn | awesome | 20:33 |
| ndufresne | for 4:2:2, it's a limitation of the driver | 20:33 |
| mntmn | any idea why yuvj420p(pc) would kind of work but cause these smears and glitches? | 20:33 |
| mntmn | (if that's the cause) | 20:33 |
| ndufresne | is it as bad as what you show me, or just diagonal stripes, with sometimes black flashes ? | 20:34 |
| mntmn | ndufresne: http://dump.mntmn.com/IMG_8224.mp4 | 20:34 |
| mntmn | ndufresne: and this is what i got from the 4:4:4 (probably expected) http://dump.mntmn.com/IMG_8223.mp4 | 20:35 |
| mntmn | ndufresne: it seems like it ~mostly~ works | 20:36 |
| mntmn | but there's some corruption that is then smeared around, too | 20:36 |
| ndufresne | strange, normally you don't get anything for 4:4:4, unless I'm missing something | 20:36 |
| mntmn | ndufresne: the content you see in the 4:4:4 is not from the video itself, but old buffers from the previous video | 20:36 |
| mntmn | it keeps something around | 20:36 |
| mntmn | (not sure where exactly) | 20:37 |
| ndufresne | ok make sense, there is not enough bits in the register to tell the HW we are doing 4:4:4, right now it just get truncated, that really need fixing | 20:37 |
| mntmn | ok cool | 20:38 |
| ndufresne | as for 4:2:2: we pass a too small memory buffer, so of course it will go wrong | 20:40 |
| ndufresne | ok, to recap, High 4:4:4 need to be filtered out | 20:41 |
| ndufresne | 4:2:2 is missing implementation, so we need to temporily filter this out | 20:42 |
| mntmn | ok | 20:42 |
| ndufresne | and gst need to expose the profiles, so that software fallback can work in playbin/playbin3 | 20:42 |
| ndufresne | as for g2 (VP9 and HEVC) those will be 4:2:0 only, HW limitation, but they will have 10bit support | 20:43 |
| ndufresne | (no driver yet) | 20:43 |
| mntmn | got it | 20:43 |
| mntmn | is G2 very different from the G1 driver implementation work wise? | 20:44 |
| ndufresne | we'll try and improve this, thanks a lot for testing and reporting, not always simple to sort things out | 20:44 |
| ndufresne | mntmn: same workflow, different registers, and a lot of registers as this is complex codecs (HEVC and VP9) | 20:44 |
| ezequielg | ndufresne: note that the hantro driver is exposing some profiles through the V4L2_CID_MPEG_VIDEO_H264_PROFILE control. maybe we should double check if it's not exposing anything it doesn't support. | 20:45 |
| mntmn | cool, thank you for your hard work, i'm glad to test new things when you have them | 20:45 |
| ndufresne | ezequielg: ok, so it expose profiles but ignores the pps ? | 20:45 |
| ndufresne | sorry sps ? | 20:45 |
| ezequielg | most likely, yes. | 20:45 |
| ezequielg | that's definitely pending. | 20:45 |
| ndufresne | well, means it worth that I implement profile in gst already, as it's better then giving user corrupted frames | 20:46 |
| ndufresne | let me check the list | 20:46 |
| ndufresne | (trying to remember my v4l2-ctl foo) | 20:46 |
| ezequielg | well, it uses profile_idc, but it's not rejecting what can't support. | 20:46 |
| ndufresne | ok, so we expose baseline, constrained-baseline, main and high | 20:47 |
| ndufresne | in theory these should work, though there is one counter example | 20:47 |
| ndufresne | and baseline is a lie, the hardware supports it, but that would require a new kernel interface, and afaic, we'll never expose that | 20:48 |
| ndufresne | at the same time, most baseline files are secretly contrained | 20:48 |
| ndufresne | and there is no software fallback, since ffmpeg does not support the advance features in baseline | 20:49 |
| ndufresne | mntmn: when I thought it worked, I was basically lucky and the decoder picked up frames from the previous decode ;-P | 20:52 |
| ndufresne | so the "random content" was valid images | 20:52 |
| mntmn | ahhh | 20:54 |
| mntmn | :D | 20:54 |
| ndufresne | we will let you know when we have updates, I'll definatly add profiles filtering in gst, as this will help | 20:55 |
| ndufresne | you can launch epiphany (the gnome browser) to test what happens on the web, maybe youtube ? I never tested htat | 20:56 |
| mntmn | it uses gstreamer? | 20:56 |
| ndufresne | epiphany is using webkitgtk, which uses gstreamer yes | 20:56 |
| ndufresne | cervo, the Mozilla experimental browser (written in rust) is also GStreamer base | 20:57 |
| mntmn | oh | 20:57 |
| ndufresne | as for Firefox, it's ffmpeg based, and chromium has it's own native implementation for that codec, if you can figure-out the build knob | 20:57 |
| ndufresne | it's used on RK3288 and rK3399 chromebook | 20:58 |
| ndufresne | for Firefox, some work is needed, I'm not sure they have DMAbuf/EGLImage support yet, so pretty far | 20:58 |
| ndufresne | when webkitgtk goes gtk4, that will make a huge difference for embedded GPU, I'm really looking forward | 21:00 |
| ndufresne | mntmn: stupid question, but what's the window manage in your demos ? | 21:01 |
| mntmn | ndufresne: it's sway! | 21:01 |
| mntmn | https://swaywm.org/ | 21:01 |
| mntmn | (wayland) | 21:02 |
| ndufresne | ah, I know I had seen it before, they are the folks behind wlroots iirc | 21:02 |
| mntmn | yep | 21:02 |
| mntmn | so, epiphany is generally very sluggish here, and it does not pick up the hw decoding, just tried it | 21:02 |
| ndufresne | hmm, ok, well, most of web content is VP9, and that's on the g2 | 21:03 |
| ndufresne | might be why | 21:03 |
| ndufresne | we'll get there at some point for sure | 21:04 |
| ezequielg | sway <3 | 21:07 |
| ndufresne | hehe, then you might like to know that it's developers are looking into gstreamer waylandsink, as sending directly you decoded frames to the compositor have showed to yield better video streaming performance ;-P | 21:15 |
| ndufresne | of course, wlroot needs a minimum support, like scaling ;-P | 21:15 |
| mntmn | good good :) | 21:18 |
| - ehmry (QUIT: Quit: https://quassel-irc.org - Chat comfortably. Anywhere.) (~quassel@2a03:3b40:fe:ab::1) | 23:02 | |
| + ehmry (~quassel@2a03:3b40:fe:ab::1) | 23:10 | |
| ex-parrot | morning | 23:25 |
| mntmn | yo! | 23:33 |
| ex-parrot | how's it going mntmn? | 23:44 |
| mntmn | good good, getting back to KiCAD today to finish D-4 motherboard revision | 23:44 |
| ex-parrot | \o/ | 23:44 |
| ex-parrot | extremely misc question: on the "real" reform is the bottom panel frosted acrylic or clear or something else? | 23:45 |
| ex-parrot | also, I am super pleased that reform is done in KiCAD | 23:47 |
| ex-parrot | also mntmn, not sure if I ever mention, but I started following your work after Interim. that was very inspirational! thank you for all your good work so far in the community | 23:55 |
| ex-parrot | mentioned* | 23:55 |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!
