| - erlehmann (QUIT: Quit: Just say no, then the virus can not enter your body without your consent.) (~erlehmann@46.114.39.56) | 03:48 | |
| ndufresne | mntmn: apparently VAAPI support have landed in Firefox (targetting 78) | 03:49 |
|---|---|---|
| ndufresne | That's interesting, we are pretty far from a just work scenario, but it means that ffmpeg drmhw context support have landed (requires Wayland and a compositor that supports DMABuf) | 03:51 |
| ndufresne | This is the same hwcontext used by kwiboo | 03:51 |
| ndufresne | https://mastransky.wordpress.com/2020/06/03/firefox-on-fedora-finally-gets-va-api-on-wayland/amp/?__twitter_impression=true | 03:51 |
| + futarisIRCcloud (uid222239@gateway/web/irccloud.com/x-ocloqoiaqzsvsolg) | 05:12 | |
| + B[] (~Thunderbi@219-89-237-57-adsl.sparkbb.co.nz) | 05:52 | |
| - futarisIRCcloud (QUIT: Quit: Connection closed for inactivity) (uid222239@gateway/web/irccloud.com/x-ocloqoiaqzsvsolg) | 07:22 | |
| - dckc (QUIT: Ping timeout: 260 seconds) (~DanC@user2.kumc.edu) | 09:39 | |
| + dckc (~DanC@136.33.249.0) | 09:43 | |
| + erlehmann (~erlehmann@46.114.39.91) | 10:39 | |
| + X704 (~tiwake@75-111-184-123.lbbkcmtk01.res.dyn.suddenlink.net) | 14:34 | |
| + nn_ (~rvense@93-163-17-86-static.dk.customer.tdc.net) | 14:36 | |
| - rvense (QUIT: Ping timeout: 246 seconds) (~rvense@93-163-17-86-static.dk.customer.tdc.net) | 14:36 | |
| - banna (QUIT: Ping timeout: 264 seconds) (~banna@banna.tech) | 14:36 | |
| - tiwake (QUIT: Ping timeout: 264 seconds) (~tiwake@75-111-184-123.lbbkcmtk01.res.dyn.suddenlink.net) | 14:36 | |
| - specing (QUIT: Ping timeout: 264 seconds) (~specing@unaffiliated/specing) | 14:36 | |
| * X704 -> tiwake | 14:36 | |
| + banna (~banna@banna.tech) | 14:37 | |
| + specing (~specing@unaffiliated/specing) | 14:38 | |
| * nn_ -> rvense | 15:01 | |
| mntmn | ndufresne: huh, what is the relationship of vaapi and v4l? | 15:16 |
| - dckc (QUIT: Ping timeout: 265 seconds) (~DanC@136.33.249.0) | 15:29 | |
| + dckc (~DanC@2605:a601:ac60:1000:c5ce:f004:94b0:8230) | 15:30 | |
| - dckc (QUIT: Client Quit) (~DanC@2605:a601:ac60:1000:c5ce:f004:94b0:8230) | 15:30 | |
| + dckc (~DanC@2605:a601:ac60:1000:5088:7f97:3044:800b) | 15:31 | |
| - dckc (QUIT: Ping timeout: 260 seconds) (~DanC@2605:a601:ac60:1000:5088:7f97:3044:800b) | 15:41 | |
| + dckc (~DanC@user2.kumc.edu) | 15:53 | |
| ndufresne | mntmn: the ffmpeg DRMHWContext | 15:58 |
| ndufresne | and because there exist a VA driver for the reform codecs driver | 15:59 |
| mntmn | ndufresne: how is that structured, does libva know how to use the v4l stuff in the kernel? | 16:06 |
| mntmn | i'm using a reform with kernel 5.7.0 right now, should i be able to see if libva picks something up? | 16:07 |
| mntmn | hmm libva info: VA-API version 1.7.0 | 16:09 |
| mntmn | libva error: vaGetDriverNameByIndex() failed with unknown libva error, driver_name = (null) | 16:09 |
| mntmn | related? https://github.com/intel/libva/pull/340/files | 16:10 |
| mntmn | hmm https://github.com/intel/libva/pull/332 | 16:11 |
| mntmn | ndufresne: is libva support needed for hantro to work in firefox? | 16:11 |
| + futarisIRCcloud (uid222239@gateway/web/irccloud.com/x-piumdvmqliwwsoht) | 16:13 | |
| ndufresne | mntmn: libva is an abstraction layer for strateless decoder/encoder, you need a VA driver for you HW | 16:29 |
| ndufresne | cphealy is currently using such a driver | 16:30 |
| ndufresne | the driver is called "VA V4L2 Request" | 16:30 |
| ndufresne | but I don't think it's really worth investing time in that one | 16:31 |
| mntmn | ndufresne: is the VA driver a userspace thing? | 16:31 |
| ndufresne | mntmn: software whise, the hard part is to add support the FFMPEG DRM HW Context | 16:31 |
| ndufresne | yes, VA drivers are userspace | 16:31 |
| mntmn | ndufresne: so they talk to v4l infrastructure in the kernel, right? | 16:31 |
| ndufresne | in the case of Intel i965, this code manages bunch of complex shaders, which implement decoder/encoder | 16:32 |
| mntmn | ndufresne: ah crazy, they use GPGPU for video decode? | 16:32 |
| ndufresne | starting from iHD, they got some discrete accelerators they driver too | 16:32 |
| ndufresne | because it's more power efficient | 16:32 |
| mntmn | ndufresne: are VA drivers accessed through libva or directly by something like firefox? | 16:33 |
| ndufresne | on AMD and NVidia, they have accelerators that they have placed inside the GPU, and they implement extra command stream to drive this | 16:33 |
| mntmn | ah. | 16:33 |
| ndufresne | libva is a really thin wrapper that implement the dlopen() bit, as drivers are in fact plugins | 16:33 |
| mntmn | ok, but does libva also need work to support the hantro VA driver or will it just work? | 16:34 |
| ndufresne | now, as I said, Firefox whise, that hard bit was to get the support for FFMPEG HW Context, which is a special mechanism to zero copy buffers (dmabuf) from the decoder to the application | 16:35 |
| mntmn | ok | 16:35 |
| ndufresne | as we speak, the ffmpeg v4l2 stateless backend is based on that, and the other v4l2 stateful backend is being ported | 16:35 |
| ndufresne | so the gap to support ffmpeg native implementation (which I think it more future proof) is getting really small | 16:35 |
| mntmn | what needs to be done? | 16:36 |
| mntmn | and who's gonna do it? :D | 16:36 |
| ndufresne | someone needs to check the code, but it would be nice to engage with the Rehad folks doing the VAAPI work (likely done for Lenevo Linux laptop) and see what's possible to collaborate | 16:36 |
| ndufresne | you could maybe make a GSoC project for that, e.g. you give away a reform to a GSoC student to get that done (along with many other possible project) that would imho fit well | 16:37 |
| mntmn | ah, i was just about to ask, what can we do | 16:37 |
| ndufresne | think of it, learning to program HW acceleration + Firefox, such a huge experience for students to start their career, it's a winwin | 16:38 |
| ndufresne | and because it's not a from scratch task, its actually doable, and I'm sure the folks at redhat can help that student | 16:39 |
| mntmn | looks like it's too late for us to host anything, so we would have to work with collabora or another company? https://summerofcode.withgoogle.com/how-it-works | 16:41 |
| mntmn | not really sure how gsoc works | 16:42 |
| ndufresne | hmm, yeah, I wonder when was the deadline | 16:43 |
| ndufresne | gsoc works by Open Source project offering project, which candiates can apply to, each candidate gets a mentor that will follow the work, and also update google making sure the worked did happen | 16:44 |
| ndufresne | you have to be accepted as a project of course | 16:45 |
| mntmn | yeah. so organization deadline is over. | 16:45 |
| ndufresne | maybe ping the Gnome folks, they are always there | 16:45 |
| ndufresne | maybe it's not too late to add a project within that scope ? or maybe Mozilla got in too ? | 16:46 |
| mntmn | yeah good ideas, will check | 16:47 |
| mntmn | i mean this is not depending on reform, anyone using hantro will benefit? also rk3399? | 16:47 |
| mntmn | so at least purism and pine, too? | 16:47 |
| + oomono (uid328183@gateway/web/irccloud.com/x-ldihwmtmkomhsjsp) | 16:49 | |
| + salotz (~user@c-24-127-178-151.hsd1.mi.comcast.net) | 16:50 | |
| blast007 | For GSoC, students and the projects they are working on have already been decided. A similar program is Outreachy, but I'm not familar with that one myself. | 17:32 |
| mntmn | blast007: ah, thanks for clarifying! | 17:37 |
| + reformer (~reformer@softboy.mntmn.com) | 18:42 | |
| jackhill | Outreachy has also already started for this round, but they also have a round during the other half of the year. | 19:18 |
| jackhill | Outreachy is great because it helps get people involved who otherwise not be involved with free software. Increasing the diversity of people in this movement is critical work, especially now. | 19:20 |
| jackhill | but their pockets aren't as deep as Google's, so some fundraising to cover the stipend might be needed. | 19:20 |
| - oomono (QUIT: Quit: Connection closed for inactivity) (uid328183@gateway/web/irccloud.com/x-ldihwmtmkomhsjsp) | 21:11 | |
| + oomono (uid328183@gateway/web/irccloud.com/x-vdrpxivmdgvvmwke) | 21:27 | |
| + csprng (~sprang@h72-0-167-252.bendor.broadband.dynamic.tds.net) | 21:55 | |
| - csprng (QUIT: Ping timeout: 265 seconds) (~sprang@h72-0-167-252.bendor.broadband.dynamic.tds.net) | 23:59 | |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!