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 7.1.1 has been released! | Please read ffmpeg.org/developer.html#Code-of-conduct
kimapr has quit [Remote host closed the connection]
<nevcairiel>
kasper93: seems like the job crashed with some weird memory error, which always causes the lock file to be left behind, probably unrelated VM nonsense
<BtbN>
Which job?
<nevcairiel>
One of my fate runners
dionisis has quit [Ping timeout: 252 seconds]
___nick___ has joined #ffmpeg-devel
<Xe>
Fwiw i know about the anubis issues, hacker news has been very insistently rude about it. I think I have a fix in the main branch.
<BtbN>
Which issue? Thatr Huawei got their bots through it?
<Xe>
no, the one where sometimes mobile firefox hits a race condition with cookies, opengraph passthrough, and other related torment
<Xe>
for huawei, ASN block them at the firewall level
<BtbN>
The Huawei thing is ridiculous anyway
<BtbN>
"We want to check if that code infringes our copyright!" *proceeds to craw Codeberg with tens of thousands of bots, instead of using git clone*
<Xe>
oh dude it's so sad
dionisis has joined #ffmpeg-devel
<BtbN>
Xe: Something else I ran into while setting up persistence: I only wanted to configure persistence, and keep using defaults for everything else. But the moment I have my own config file, I need to configure everything myself, I can't just add to the built in default.
<indecisiveturtle>
Lynne: Did some more testing and I think it was a bandwidth limitation or problem akin to that. Moving slice output buffer to device local memory and porting your transfer queue code to DMA it to the host made using 3 contexts consistently faster than 1, at least on NVIDIA
<indecisiveturtle>
At least on my system the vulkan prores encoder is around 3x to 4x faster than cpu one (encoding time of 41 seconds on cpu and 12 seconds with vulkan)
Warcop has quit [Remote host closed the connection]
<michaelni>
spend a few hours analyzing a use of uninitialized memory between libopus and ffmpeg. took a long time to just manage to replicate. the result, building with FORTIFY_SOURCE (which libopus does by default) is not compatible with memory sanitizer of course nothing warns you of that
<michaelni>
unless you read the asm code and find calls to __memcpy_chk in place of __msan_memcpy@PLT