2025-08-22 10:25
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 8.0 has been released! | Please read ffmpeg.org/developer.html#Code-of-conduct
00:15
<
Traneptora >
BtbN: according to my waterfall diagram most of that 1200 ms is waiting
00:19
<
kierank >
mine was entirely on the page
00:19
<
kierank >
all other assets cached
00:21
<
Traneptora >
page itself waits for 411 ms here
00:22
<
Traneptora >
to load git master commit list is 2.7s
00:22
<
Traneptora >
it wasn't always this slow, so I'm guessing something changed
00:22
<
BtbN >
stuff that calls git commands in the background is slow, yeah
00:23
<
BtbN >
though the next F5
_should_ be fast, since it caches that stuff
00:23
<
Traneptora >
pushed refresh without cache purge, 1988 ms
00:23
<
BtbN >
it caches it server side
00:23
<
BtbN >
so even Ctrl+F5 should see a speedup on the second attempt
00:23
<
Traneptora >
yea, another f5, and it was 1979 ms
00:23
<
Traneptora >
which is still like 2s
00:23
<
Traneptora >
even with server-side cache
00:24
<
Traneptora >
1900 ms is pretty slow
00:24
<
Traneptora >
there may be an issue with the cache
00:24
<
BtbN >
given the cache is nice and full, according to htop, that seems unlikely
00:24
<
Traneptora >
well, the observed behavior is that 1900ms every time I load git master
00:25
<
Traneptora >
including via f5 with only the page gen (no assets need to be repulled)
00:25
<
BtbN >
it's not calling any git commands at least
00:25
<
BtbN >
every time I hit F5 there's a 100% CPU spike of mariadb though
00:25
<
BtbN >
so that seems to be the limiting factor
00:25
<
Traneptora >
perhaps that's the bottleneck yea
00:26
<
BtbN >
the commit list and PR list does pull info from
_a lot_ of different sources into one place, so that's probably why
00:26
<
BtbN >
though it should still be able to do that much faster with this small of a DB
00:26
<
BtbN >
The DB having to do heavy lifting for any read-only load like that usually hints at missing indices
00:47
<
kasper93 >
we have only one repo on this instance and dozen users, it should have no problems
00:47
<
kasper93 >
unless you run it on rpi :)
01:04
ubitux has quit [Ping timeout: 260 seconds]
01:04
ubitux has joined #ffmpeg-devel
01:10
minimal has quit [Quit: Leaving]
01:37
Teukka` has joined #ffmpeg-devel
01:38
Martchus_ has joined #ffmpeg-devel
01:38
ramiro_ has joined #ffmpeg-devel
01:39
ramiro has quit [Ping timeout: 244 seconds]
01:39
Martchus has quit [Ping timeout: 244 seconds]
01:39
michaelni has quit [Ping timeout: 244 seconds]
01:39
Teukka has quit [Ping timeout: 244 seconds]
01:39
michaelni has joined #ffmpeg-devel
02:14
Kimapr_ has quit [*.net *.split]
02:14
lemourin has quit [*.net *.split]
02:14
witchymary has quit [*.net *.split]
02:14
Flat has quit [*.net *.split]
02:14
zsoltiv__ has quit [*.net *.split]
02:14
odrling has quit [*.net *.split]
02:14
rossy has quit [*.net *.split]
02:14
jessidhia has quit [*.net *.split]
02:14
nevcairiel has quit [*.net *.split]
02:14
Thulinma has quit [*.net *.split]
02:14
funman has quit [*.net *.split]
02:20
Kimapr_ has joined #ffmpeg-devel
02:20
Flat has joined #ffmpeg-devel
02:20
lemourin has joined #ffmpeg-devel
02:20
witchymary has joined #ffmpeg-devel
02:20
zsoltiv__ has joined #ffmpeg-devel
02:20
odrling has joined #ffmpeg-devel
02:20
jessidhia has joined #ffmpeg-devel
02:20
nevcairiel has joined #ffmpeg-devel
02:20
Thulinma has joined #ffmpeg-devel
02:20
funman has joined #ffmpeg-devel
02:26
derpydoo has joined #ffmpeg-devel
02:26
derpydoo has quit [Client Quit]
02:29
derpydoo has joined #ffmpeg-devel
02:47
derpydoo has quit [Ping timeout: 244 seconds]
03:00
arbitercoin has joined #ffmpeg-devel
03:15
rossy has joined #ffmpeg-devel
03:29
<
mkver >
rcombs: What happened?
03:31
<
rcombs >
mkver: dug deeper and realized my testing had been incomplete and I'd missed a critical piece of the underlying code; I was under the impression that that field was passed through from input to output, but under some circumstances it actually gets set internally to the *encode duration*, which we don't want
03:32
<
rcombs >
so, reverted while looking into a correct solution
03:32
jamrial has quit []
03:37
<
rcombs >
probably just gonna want to set `p_app_private`
03:44
Martchus has joined #ffmpeg-devel
03:45
Martchus_ has quit [Ping timeout: 258 seconds]
03:54
<
rcombs >
hmmm, I'm not entirely clear on how lifetimes are supposed to work with… actually most codec APIs now that I look deeper
03:55
<
rcombs >
like, if you pass a frame to svt_av1_enc_send_picture, is it guaranteed that that frame is only read from during the call, and that you're allowed to free its memory immediately afterwards?
03:57
<
rcombs >
looking at the implementation, it appears that it always makes a copy of the frame, but like. why? that seems crazy-inefficient
04:02
<
rcombs >
the x264 wrapper handles the opaque-lifetime issue by querying the encoder for the maximum possible delay and allocating an array of that size, but SVT-AV1 doesn't appear to expose that information
04:03
<
rcombs >
so maybe the answer is… a linked list of opaque buffers to free when the encoder is closed?
04:12
Kimapr_ has quit [Remote host closed the connection]
04:12
Kimapr_ has joined #ffmpeg-devel
04:17
mkver has quit [Ping timeout: 260 seconds]
05:03
psykose has quit [Remote host closed the connection]
05:04
psykose has joined #ffmpeg-devel
05:34
System_Error has quit [Remote host closed the connection]
05:41
System_Error has joined #ffmpeg-devel
05:57
MisterMinister has joined #ffmpeg-devel
06:07
rvalue- has joined #ffmpeg-devel
06:08
rvalue has quit [Ping timeout: 244 seconds]
06:14
rvalue- is now known as rvalue
06:16
redzic4 has joined #ffmpeg-devel
06:16
fennewald has quit [Ping timeout: 256 seconds]
06:16
Fenrir has quit [Ping timeout: 256 seconds]
06:17
graphitemaster has quit [Ping timeout: 256 seconds]
06:17
abdo- has joined #ffmpeg-devel
06:18
abdo has quit [Ping timeout: 256 seconds]
06:18
tortoise has quit [Ping timeout: 256 seconds]
06:18
haxar has quit [Ping timeout: 256 seconds]
06:19
redzic has quit [Ping timeout: 256 seconds]
06:19
redzic4 is now known as redzic
06:20
tortoise has joined #ffmpeg-devel
06:23
Fenrir has joined #ffmpeg-devel
06:23
graphitemaster has joined #ffmpeg-devel
06:31
fennewald has joined #ffmpeg-devel
06:33
haxar has joined #ffmpeg-devel
06:52
System_Error has quit [Remote host closed the connection]
06:59
System_Error has joined #ffmpeg-devel
07:01
fjlogger has quit [Read error: Connection reset by peer]
07:01
System_Error has quit [Remote host closed the connection]
07:07
System_Error has joined #ffmpeg-devel
07:44
MisterMinister has quit [Ping timeout: 255 seconds]
07:45
av500 has quit [Remote host closed the connection]
07:46
av500 has joined #ffmpeg-devel
08:00
av500 has quit [Remote host closed the connection]
08:01
av500 has joined #ffmpeg-devel
09:01
mkver has joined #ffmpeg-devel
09:23
mkver has quit [Ping timeout: 260 seconds]
09:23
mkver has joined #ffmpeg-devel
09:29
fjlogger has joined #ffmpeg-devel
10:08
microlappy has joined #ffmpeg-devel
10:11
microlappy has quit [Remote host closed the connection]
10:12
microlappy has joined #ffmpeg-devel
10:12
microlappy has quit [Remote host closed the connection]
10:37
j45_ has joined #ffmpeg-devel
10:38
j45 has quit [Ping timeout: 260 seconds]
10:38
j45_ is now known as j45
10:38
j45 has joined #ffmpeg-devel
10:38
j45 has quit [Changing host]
11:28
jamrial has joined #ffmpeg-devel
11:58
<
Gramner >
mkver: from what I can understand after some googling is that the "VEX temporary storage exhausted" error valgrind gives is some internal bug/issue within valgrind itself and not a problem with the code in question
11:58
<
Gramner >
not sure what to do about that though
12:12
putacho has joined #ffmpeg-devel
12:15
microchip_ has quit [Ping timeout: 248 seconds]
12:15
putacho is now known as microchip_
12:41
<
mkver >
Gramner, ubitux: valgrind works here with --track-origins=yes --vex-guest-max-insns=32 (but no higher) or without track-origins and --vex-guest-max-insns=56 (but not higher).
13:01
_whitelogger has joined #ffmpeg-devel
13:41
<
ubitux >
mkver: do I have to upgrade? I haven't done the update in a while
13:42
<
mkver >
ubitux: You could add that workaround to your --target-exec for this box.
14:29
_whitelogger has joined #ffmpeg-devel
15:10
ramiro_ is now known as ramiro
15:19
secondcreek has joined #ffmpeg-devel
15:42
bsFFFFFF has joined #ffmpeg-devel
15:57
<
Lynne >
haasn: ping on exposing temperature in libplacebo
16:02
<
kasper93 >
isn't it already in options?
16:07
<
Lynne >
kinda, but instead of kelvin you give it a 0-1 factor, which starts at the quite high 3500K
16:07
<
haasn >
kasper93: the option we have is a bit of a hack
16:08
<
haasn >
it's more subjective and also not colorimetrically accurate when not using bt709 primaries iirc
16:15
arbitercoin has quit [Ping timeout: 250 seconds]
16:16
minimal has joined #ffmpeg-devel
16:50
<
haasn >
jamrial: should this be a helper function? I imagine this is not the only file using the same logic
16:50
<
haasn >
av_rescale_pts
17:04
<
jamrial >
haasn: i guess
17:04
<
jamrial >
a macro like TS2T maybe
17:14
<
haasn >
Inline functions > macros wherever possible (imho)
17:18
kasper93 has quit [Ping timeout: 265 seconds]
17:22
kasper93 has joined #ffmpeg-devel
17:44
leo60228 has joined #ffmpeg-devel
17:45
vriska has quit [Ping timeout: 256 seconds]
18:14
System_Error has quit [Ping timeout: 272 seconds]
18:23
System_Error has joined #ffmpeg-devel
19:46
jamrial has quit [Read error: Connection reset by peer]
20:09
mkver has quit [Ping timeout: 244 seconds]
20:09
j45_ has joined #ffmpeg-devel
20:10
j45 has quit [Ping timeout: 260 seconds]
20:10
j45_ is now known as j45
20:10
j45 has joined #ffmpeg-devel
20:10
j45 has quit [Changing host]
20:39
jamrial_ has joined #ffmpeg-devel
21:16
System_Error has quit [Ping timeout: 272 seconds]
21:20
bsFFFFFF has quit [Quit: bsFFFFFF]
21:24
mkver has joined #ffmpeg-devel
21:40
<
desmond-netint >
Why are there no hwaccel tests in FATE? Outside of some (near useless) api tests, nothing tests qsv,nvenc,vaapi,vulkan,etc.
22:05
Kimapr_ has quit [Remote host closed the connection]
22:05
Kimapr has joined #ffmpeg-devel
22:21
<
BtbN >
Cause nobody wrote any, and it's unclear how to do it properly
22:21
<
BtbN >
cause they can't just run whenever the feature is built in
22:27
<
kasper93 >
and historically hwdec apis were really simple, where there is not much to test.
22:27
<
kasper93 >
now with lower level apis it make more sense to test them possibly
22:51
<
Compn >
lack of people willing to run hardware fate boxes?
22:51
<
Compn >
lack of hardware?
22:52
<
Compn >
some devs still using the same laptops they used in uni /s
22:53
<
Compn >
wont someone donate some new puters to ff devs?
23:00
<
Compn >
and then of course power bill for 10+ hardware gpu computers
23:00
<
Compn >
well maybe not that many
23:03
Hassan016 has joined #ffmpeg-devel
23:05
_av500_ has joined #ffmpeg-devel
23:05
vriska has joined #ffmpeg-devel
23:06
leo60228 has quit [*.net *.split]
23:06
fjlogger has quit [*.net *.split]
23:06
av500 has quit [*.net *.split]
23:06
ramiro has quit [*.net *.split]
23:06
Hassan01 has quit [*.net *.split]
23:06
Kwiboo has quit [*.net *.split]
23:06
Hassan016 is now known as Hassan01
23:07
Kwiboo has joined #ffmpeg-devel
23:08
ramiro has joined #ffmpeg-devel
23:10
Chagalle has quit [Ping timeout: 260 seconds]
23:11
Chagall has joined #ffmpeg-devel
23:12
minimal has quit [Quit: Leaving]
23:12
<
desmond-netint >
qsv wouldn't be difficult hardware-wise. You just need the right Intel CPU. The others need a dGPU. An iGPU should work.
23:16
<
Compn >
i was thinking of the old days where each card had its own thing :D
23:17
RT|AO has quit [Ping timeout: 260 seconds]
23:17
RT|AO has joined #ffmpeg-devel
23:17
<
Traneptora >
mkver: I see what you meant about max packet size, pushed another change. I think existing behavior was incorrect wrt the ICC profile
23:20
fjlogger has joined #ffmpeg-devel
23:28
<
desmond-netint >
I'm studying how FATE works. The FATE tests are fairly basic. Is this by design?
23:57
<
BtbN >
basic in what sense? They're effectively all integration tests
23:57
<
BtbN >
FFmpeg has no unit tests