michaelni changed the topic of #ffmpeg-devel to: Welcome to the FFmpeg development channel | Questions about using FFmpeg or developing with libav* libs should be asked in #ffmpeg | This channel is publicly logged | FFmpeg 6.1 has been released! | Please read ffmpeg.org/developer.html#Code-of-conduct
cone-137 has joined #ffmpeg-devel
<cone-137>
ffmpeg Jun Zhao master:ab3bd5ead047: avdevice/decklink_dec: add explicit specifier
<cone-137>
ffmpeg Jun Zhao master:c961ac4b0ca2: vulkan_decode: fix the print format of VkDeviceSize
TD--Linux is now known as TD-Linux
<compn>
wonder why that guy thought you were from usa
gust82 has quit [Remote host closed the connection]
gust82 has joined #ffmpeg-devel
MrZeus has quit [Ping timeout: 255 seconds]
MrZeus has joined #ffmpeg-devel
odrling has quit [Ping timeout: 245 seconds]
odrling has joined #ffmpeg-devel
gust82 has quit [Remote host closed the connection]
gust82 has joined #ffmpeg-devel
microchip_ has quit [Remote host closed the connection]
microchip_ has joined #ffmpeg-devel
wangbin has joined #ffmpeg-devel
lexano has quit [Ping timeout: 256 seconds]
thilo has quit [Ping timeout: 256 seconds]
thilo has joined #ffmpeg-devel
thilo has quit [Changing host]
thilo has joined #ffmpeg-devel
MrZeus has quit [Ping timeout: 276 seconds]
wangbin has quit [Quit: Connection closed]
wangbin has joined #ffmpeg-devel
jamrial has quit []
lemourin has quit [Read error: Connection reset by peer]
lemourin has joined #ffmpeg-devel
cone-137 has quit [Quit: transmission timeout]
mkver has quit [Ping timeout: 256 seconds]
derpydoo has quit [Ping timeout: 256 seconds]
durandal_1707 has quit [Ping timeout: 264 seconds]
durandal_1707 has joined #ffmpeg-devel
AbleBacon has quit [Read error: Connection reset by peer]
wangbin has quit [Quit: Connection closed]
kurosu has joined #ffmpeg-devel
gust82 has quit [Remote host closed the connection]
gust82 has joined #ffmpeg-devel
gust82 has quit [Remote host closed the connection]
gust82 has joined #ffmpeg-devel
gust82 has quit [Remote host closed the connection]
<jdek>
>Around 2023-06-15, a project known as FFmpeg decided that it would be a great idea to clone GMP in their CI scripts, meaning that every one of their commits requested a compressed clone from the GMP servers. But, by Github's design, hundreds of FFmpeg forks automatically followed suit, themselves cloning the GMP repo.
<jdek>
which CI scripts are they referring to even
<JEEB>
BtbN's auto-builder
<JEEB>
of which there are dozens of forks
<wbs>
but scheduled pipelines don't run on forks automatically, afaik?
<jdek>
I didn't think they did either
<BtbN>
Nope, you need to manually turn it on. Their description of it is greatly exagarated
<BtbN>
It was like 10~20 parallel requests tops, spread out over an hour
<BtbN>
The main problem apparently was that their hg host was misconfigured or something
<elenril>
I'm getting random resets, which I thought was memory
<Lynne>
if you want more unstable, do export RUSTICL_FEATURES=fp16,fp64
<JEEB>
so yea, just needed to look for a newer piece of hardware
<elenril>
but now I'm less sure
<Lynne>
if you want to stress it, just run nlmeans_vulkan on 1080p
\\Mr_C\\ has joined #ffmpeg-devel
<Lynne>
btw, fun fact, hip runs on an *alternative* kernel module, not the main amdgpu kernel module
<elenril>
why do they even have two
<Lynne>
so it's a completetly different codepath, that pretty much no one outside of amd touches, unlike the main kernel module, which everyone who works on mesa touches
<elenril>
do they want this stuff to be a mess?
<Lynne>
speed? dunno, I bet it's slower
<elenril>
is this like a weird stockholm syndrome thing where they cannot imagine having non-shit drivers
<Lynne>
at least it's abstracted away enough so that you can write non-shit drivers, they even give you the docs
<elenril>
but then why pay people to write a separate shit driver branch
<courmisch>
surely the great proprietary driver has some superior qualities that cannot be achieved in a opensoarish process
<courmisch>
such as using the company approved coding style
<Lynne>
oh they don't pay a single team, they pay three (amdvlk, amdgpu-pro, which may or may not be just amdgpu-pro with llvm ripped out and the proprietary compiler used, and hip)
\\Mr_C\\ has quit [Remote host closed the connection]
<elenril>
proprietary compiler?
<elenril>
even intel gave up on that
<Lynne>
it's the one they use in their windows drivers, may as well use it here too
<Lynne>
but the biggest issue with amd isn't the drivers, mesa is pretty much optimal for everything now
<Lynne>
it's the stability, they've spent so much time optimizing everything that they didn't actually think about how it may be developed on
<Lynne>
so if you overread, you crash the GPU, and with it, the entire session (contents of VRAM, which means your composer and all programs using the GPU directly or indirectly, are gone)
<Lynne>
it's like it's 1987 and it runs on dos running on x86 real mode
\\Mr_C\\ has joined #ffmpeg-devel
<elenril>
doesn't the kernel object to that?
<Lynne>
what happens is the kernel recreates the GPU session, and anything that uses the old session is just frozen
<elenril>
so is the bug in mesa?
<Lynne>
becuse it endlessly blocks during ioctls (IIRC)
<elenril>
or in the kernel?
Krowl has quit [Read error: Connection reset by peer]
<Lynne>
no, the GPU, it doesn't give you tools to do proper context isolation
<Lynne>
not even virtualization can fix this
<elenril>
as in the hardware itself?
<Lynne>
yup
<elenril>
sad
<elenril>
what do i need driver-wise for vulkan video decoding?
<Lynne>
RADV_PERFTEST=video_decode
<elenril>
>Device does not support the VK_KHR_video_decode_queue extension!
<elenril>
guess I should upgrade that to 23.3.0~rc4-1
<elenril>
neat, video decoding now segfaults
<elenril>
av:hevc:df0[15455]: segfault at 18 ip 00007f5e900ad5a9 sp 00007f5deffedfa0 error 4 in libavcodec.so.60.31.102[7f5e8fe84000+aad000] likely on CPU 9 (core 1, socket 0)
<elenril>
but nlmeans_vulkan seems to be actually doing something
<Lynne>
where does decoding segfault in lavc?
<Sebastinas>
jamrial: Thanks, will update the patch in the next upload.
<elenril>
Lynne: ff_hwaccel_frame_priv_alloc (avctx=<optimized out>, hwaccel_picture_private=hwaccel_picture_private@entry=0x555555a62228) at src/libavcodec/decode.c:1853
gust82 has quit [Remote host closed the connection]
<elenril>
(in 6.1 from debian)
gust82 has joined #ffmpeg-devel
<jamrial>
elenril: yeah, see my message to Sebastinas above
<jamrial>
Lynne: if that version of the fix is good, please push it
Krowl has joined #ffmpeg-devel
<Sebastinas>
The old patch at least fixed crashes with nvidia xD
<Lynne>
I haven't been able to replicate the crash though, so I don't know
<Lynne>
not on ffmpeg or mpv
<Lynne>
it seems to work fine, but someone with vlc should test it
<Sebastinas>
I can do that later tonight.
<Lynne>
elenril: can you test it? not sure how you're hitting that path
<elenril>
seems debian has a broken version of that patch applied
<JEEB>
Sebastinas: btw if I recall you were related to debian/ubuntu? do you happen to know when the next merge freezes are on either side?
<elenril>
that does not initialize frames_ctx
<Sebastinas>
JEEB: Debian only. Next freeze will start early 2025.
<JEEB>
alrighty
<JEEB>
thanks
Krowl has quit [Read error: Connection reset by peer]
<jdek>
how is cone-123 setup to log here for ffmpeg.git? should add it for ffmpeg-web.git as well
<BtbN>
I think it's coming from videolan git
<Lynne>
kierank: could you ping dave rice anyway? I'd just like to hear his thoughts on it
<jdek>
honestly I thought source.ffmpeg.org was videolan git, but they're separate repos? or is http://git.videolan.org 's gitweb just not configured to show ffmpeg-web
<durandal_1707>
Lynne: dont you have email address?
<Lynne>
durandal_1707: oh, yeah, I forgot I do
<kierank>
jdek: i was confused by this too
<jdek>
would really like to have an overview of our infra...