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 7.0.1 has been released! | Please read ffmpeg.org/developer.html#Code-of-conduct
lexano has quit [Ping timeout: 256 seconds]
<Lynne>
jamrial: I don't think my lpc_autocorr *does anything* for mmsize == 32, I'm pretty sure I have an error
<Lynne>
getting rid of that shufpd on m0 should break it, but it doesn't, which means the upper part shouldn't be doing much
<jamrial>
Lynne: courmisch also mentioned the checkasm is not comparing the last value in the buffer
Krowl has quit [Read error: Connection reset by peer]
unlord has quit [Changing host]
unlord has joined #ffmpeg-devel
IndecisiveTurtle has joined #ffmpeg-devel
Krowl has joined #ffmpeg-devel
rvalue has joined #ffmpeg-devel
dionysis has joined #ffmpeg-devel
<Sean_McG>
hi
IndecisiveTurtle has quit [Remote host closed the connection]
<rajivharlalka>
hello
<rajivharlalka>
saw some old chats around migrating to something other git platform, couldn't find the final decision 👀
<rajivharlalka>
s/something/some
<another|>
there isn't one
cone-264 has quit [Quit: transmission timeout]
System_Error has quit [Remote host closed the connection]
IndecisiveTurtle has joined #ffmpeg-devel
System_Error has joined #ffmpeg-devel
Krowl has quit [Read error: Connection reset by peer]
<IndecisiveTurtle>
Gsoc coding period has begun, nice xD
<Lynne>
IndecisiveTurtle: welcome back
System_Error has quit [Remote host closed the connection]
<IndecisiveTurtle>
Thanks, I was checking the chats here almost every night, hope I didn't miss anything important :p
<IndecisiveTurtle>
I have setup ffmpeg dev environment and saw a bit how codecs are structured. I've also read about more than half of the VC2 spec so far
<Lynne>
great, could you start with exporting all header writing to a separate file, and making a boilerplate encoder, and generating headers?
System_Error has joined #ffmpeg-devel
<IndecisiveTurtle>
I took the structures and enums I saw and placed them in a vec2enc_common.h file as you had suggested
<IndecisiveTurtle>
Was VC2EncContext and it's dependencies and that VC2BaseVideoFormat table
<Lynne>
you need to do the same with the functions
ngaullier has quit [Quit: Leaving]
<IndecisiveTurtle>
So add declarations of all of the functions in the header as well, got it
<Lynne>
you can name it vc2_headers.c/h and add it to both encoders in libavcodec/Makefile
<IndecisiveTurtle>
And I assume the vulkan implementation will be a separate vc2enc_vulkan.c file?
<Lynne>
yes
<IndecisiveTurtle>
From what I saw, each encoder has 3 main functions in this case vc2_encode_frame, vc2_encode_init, vc2_encode_end that call the rest of them. So the plan is to define a new FFCodec for vulkan, add implementations while re-using common functionality in the headers?
<Lynne>
yup
Krowl has joined #ffmpeg-devel
natto has quit [Quit: a.]
natto has joined #ffmpeg-devel
Guest13 has joined #ffmpeg-devel
Guest13 has quit [Client Quit]
cone-584 has joined #ffmpeg-devel
<cone-584>
ffmpeg Andreas Rheinhardt master:41e13228451c: tests/fate/source-check: Relax BSD licence check
<cone-584>
ffmpeg Andreas Rheinhardt master:f1337e5dd9e1: doc/mips: Update list of files with MIPS copyright notice
<cone-584>
ffmpeg Diego Felix de Souza master:aead61451cc4: avcodec/nvenc_av1: Correct CQ range for AV1
HarshK23 has quit [Quit: Connection closed for inactivity]
Krowl has quit [Read error: Connection reset by peer]
<courmisch>
unlord: so, fun fact, the SiFive h.264 code is broken
<courmisch>
corrolary: the h264 checkasm is not good enough
rvalue has quit [Ping timeout: 260 seconds]
<courmisch>
other fun fact, as suspect, 4 GiB is not enough to run FATE with 8 jobs
<courmisch>
+ed
<wbs>
courmisch: do you use --disable-large-tests already?
<courmisch>
wbs: no but I think LD is the memory bottleneck
<wbs>
ah, I see
<wbs>
I mostly crosss compile, so only the fate tests are limiting. with that option, I can run tests with -j4 on a 1GB target device
<jamrial>
courmisch: if your problem is ld, then run make fate-api separate from make fate -j8
<jamrial>
afaik that target has a lot of tools that all link to every library
<Lynne>
courmisch: was it an incomplete checkasm check?
<Sean_McG>
I wonder if 'mold' would help, if it's the linker
<courmisch>
relax relocations are a big PITA when it comes to linker speed and memory use
<another|>
Sean_McG: probably not
mkver has quit [Ping timeout: 264 seconds]
ccawley2011_ has joined #ffmpeg-devel
ccawley2011 has quit [Ping timeout: 240 seconds]
<courmisch>
jamrial: not sure how that would work with fate.sh
<jamrial>
oh right, you meant not manually called
<courmisch>
I guess I'll just ignore the problem and let someone give me better hardware or set a better FATE runner up
<courmisch>
and until then no vector tests in FATE still
<wbs>
in my fate setups, I don't use fate.sh as is from the repo, but I have a copy of the file, which I invoke - that way, it's easy to tweak it to do different make invocations in different sequences
<courmisch>
this is conditional code so it might be that the C code gets a fake boost from branch prediction
<courmisch>
jamrial: but of course, you can have a competition with BBB - who resists the temptation to write an SSE2 version the longest
<BBB>
it's also that mmx is legit slow :)
<wbs>
but benchmarking loop filters is always tricky for many reasons. the C code can skip early per pixel, while SIMD can't - so the actual speedup depends on how large percentage of pixels are skipped or not
<wbs>
plus when benchmarking, running the same function over and over on the same data, it won't take the same codepaths each time as the data it operates on will change. but that should progress the same for C and SIMD at least