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
minimal has quit [Quit: Leaving]
acryo has quit [Remote host closed the connection]
<thardin>
ePirat: I hurt my back yesterday, so it's looking unlikely that I will be able to travel
<thardin>
but it's still up in the air so maybe, if I recover faster
<JEEB>
ouch, I hope the recovery will be straightforward
j45_ has joined #ffmpeg-devel
j45 has quit [Ping timeout: 248 seconds]
j45_ is now known as j45
j45 has quit [Changing host]
j45 has joined #ffmpeg-devel
acryo has quit [Remote host closed the connection]
acryo has joined #ffmpeg-devel
acryo has joined #ffmpeg-devel
<thardin>
it's already better than yesterday, so there is some hope
<kasper93>
wbs: your wine32 completally borked o_O
<wbs>
kasper93: interesting; will tend to it
<wbs>
kasper93: fascinating. it looks like a regression in clang, affecting the clang-cl build. but I also build the latest nightly wine with the latest clang, and use that for the msvc test, and that also seems to have broken. will investigate and fix in one way or another
<JEEB>
\o/
<TheVibeCoder>
its 2025.
<TheVibeCoder>
and ffmpeg still on ml
<kierank>
TheVibeCoder: glorious forjego is coming
jamrial has joined #ffmpeg-devel
<kierank>
until it forks for the third time
Everything has joined #ffmpeg-devel
<jamrial>
kierank: people are giving a lot of arguments in favor of it and against gitlab's policies, so it's not going to happen on a whim...
<jamrial>
you could argue against it in the thread too
<kasper93>
ah, clang too if you change argument types to u8
Everything has quit [Quit: leaving]
Flat has quit [Quit: Rip internet]
<haasn>
Down to 55x after making the C ver better, oh well
<JEEB>
still quite nice numbers
Flat has joined #ffmpeg-devel
<haasn>
Lynne: can you teach me how to use avx513 masks to eliminate the need for scalar code to handle the tail?
<haasn>
Like say you want to write add_u8(uint8_t *a, b, size_t bytes)
<haasn>
Where bytes is not a multiple of the mmsize
<haasn>
How would you idiomatically write this as avx512?
<haasn>
Or xor_u8 for single
<haasn>
Simplicity*
<another|>
haasn: How many -O's did you use?
<haasn>
another|: defaults, which is O3 iirc
<another|>
ay
<haasn>
I tried also with march=znver5 -ftree-vectorize but it was barely faster
<haasn>
Anyway, the new functions are stupidly fast, like memcpy fast - I hit 70k fps at 4K
<haasn>
Which is even faster than the GPU / Vulkan version the client was using before :)
sr55 is now known as s55
s55 has quit [Changing host]
s55 has joined #ffmpeg-devel
<Lynne>
haasn: you should be able to just use kmovq with a static table and an offset, I think
<Lynne>
not sure what the most idiomatic way to handle tails is, maybe Gramner does
<haasn>
ChatGPT suggests pretty much the same
<haasn>
I was hoping there would be something neat like the RVV setvl helpers
<Lynne>
avx10?
Everything has joined #ffmpeg-devel
cone-178 has joined #ffmpeg-devel
<cone-178>
ffmpeg Timo Rothenpieler master:87b09f3931fe: avformat/tls_openssl: set dtls remote addr in listen mode
<cone-178>
ffmpeg Timo Rothenpieler master:db0adbbd3fcf: avformat/tls_openssl: don't abort if dtls has no key/cert set
<cone-178>
ffmpeg Timo Rothenpieler master:b6e808d28b78: avformat/tls_openssl: initialize DTLS context with correct method
<cone-178>
ffmpeg Timo Rothenpieler master:951013e6030c: avformat/tls_openssl: force dtls handshake to be blocking
<cone-178>
ffmpeg Timo Rothenpieler master:576f44016af1: avformat/tls_openssl: set default MTU if none is set
<cone-178>
ffmpeg Timo Rothenpieler master:f3355a1fff6f: avformat/tls_openssl: properly limit written size to data mtu
<cone-178>
ffmpeg Timo Rothenpieler master:95fd0840feb2: avformat/tls_openssl: don't hardcode ciphers and curves for dtls
<cone-178>
ffmpeg Timo Rothenpieler master:5edbfc4bae46: avformat/tls_openssl: clean up peer verify logic in dtls mode
<cone-178>
ffmpeg Timo Rothenpieler master:31abcc1505a8: avformar/tls_openssl: use correct info callback in DTLS mode
<cone-178>
ffmpeg Timo Rothenpieler master:49badc8689e3: avformat/tls_openssl: don't enable read_ahead in dtls mode
<cone-178>
ffmpeg Timo Rothenpieler master:5339db2cf4fc: avformat/tls_openssl: properly free generated/read keys and certificates
<cone-178>
ffmpeg Timo Rothenpieler master:40ce64eae7ee: avformat/tls_openssl: don't expose deprecated EC_KEY outside of its function
<cone-178>
ffmpeg Timo Rothenpieler master:454f161b4bd3: avformat/tls_openssl: make generating fingerprints optional
<cone-178>
ffmpeg Timo Rothenpieler master:483e5091694b: avformat/tls_openssl: automatically generate self-signed certificate when none is provided in listen mode
<cone-178>
ffmpeg Timo Rothenpieler master:9015d595a115: avformat/flvdec: don't skip backwards or over EOF
<BtbN>
michaelni: nothing to backport for that patch, the offending code does not exist anywhere but on master