[Ffmpeg-devel-irc] ffmpeg-devel.log.20180628

burek burek021 at gmail.com
Fri Jun 29 03:05:03 EEST 2018

[11:49:08 CEST] <gagandeep> kierank: i now need to start multi threading and have been reading about slice and frame threading
[12:10:09 CEST] <kierank> gagandeep: are all samples bitexact in single thread mode
[12:16:55 CEST] <gagandeep> other than the mountain sample all samples are bitexact
[12:17:17 CEST] <gagandeep> the mountain sample file is not getting decoded in the vlc part
[12:17:44 CEST] <gagandeep> i tried to see how to fix that but to not much result
[12:17:57 CEST] <gagandeep> though was able to get correct decompanding for codebook 0
[12:18:35 CEST] <gagandeep> the p frame sample i got david is bit exact
[12:19:10 CEST] <gagandeep> also mountain sample uses codebook 0 and david told me that they used to encode with it only when they didn't have codebook 1
[12:19:46 CEST] <gagandeep> so i feel right now i should do my other stuff and will try fixing error with codebook 0 alongside
[12:41:36 CEST] <kierank> gagandeep: is it a big difference
[12:41:50 CEST] <kierank> And did you figure out 3d transformer?
[12:41:59 CEST] <kierank> Transform with coefficients from last frame
[12:50:47 CEST] <gagandeep> kierank: the quant is like 64 and the vlc code i don't know how reads wrong value to be dequantized to the values a missed by a few integers
[12:51:18 CEST] <gagandeep> and also i don't know why vlc part is missing a value between zeros
[12:51:38 CEST] <gagandeep> let me send you a screenshot of raw data
[12:53:45 CEST] <kierank> You could ask michaelni
[12:54:39 CEST] <gagandeep> https://imgur.com/a/BTVrnEj
[16:28:41 CEST] <BBB> hey C experts, lets say I have a struct x { int a1, a2, a3, a4, a5, a6; int b1, b2, b3, b4, b5, b6; }; and I want to copy only the a entries in the struct (in reality, they are all different sizes/types/names etc.). I can obviously memcpy(dst, src, offsetof(struct x, b1)), but that relies on b1 not moving within the struct. does C support some kind of labeling so I can add a label halfway (like struct x { int a1, a2, a3, a4, a5, a6; half: int b1,
[16:28:41 CEST] <BBB> b3, b4, b5, b6; };), so I can memcpy using offsetof(struct x, half) as size?
[16:31:06 CEST] <jdarnley> BBB can you put all the As into one substruct?
[16:31:31 CEST] <BBB> everything is possible :-p
[16:33:10 CEST] <jdarnley> You might consider "struct x { struct a {}; struct b{} }" then you can just do y.a = x.a
[16:33:38 CEST] <ubitux> int field_separator;
[16:34:01 CEST] <ubitux> (separator_field if you speak a better english i guess)
[16:34:40 CEST] <BBB> right I could obviously just add a separator; does C support zero-sized separators?
[16:34:51 CEST] <BBB> I see a unique use case for void
[16:35:33 CEST] <jdarnley> Unlikely.  I recall something about a zero-length array is possible at the end but really am no expert.
[16:39:47 CEST] <BBB> oooo I remember that
[16:39:48 CEST] <BBB> brilliant
[16:40:09 CEST] <BBB> so evil, I LOVE IT <3
[16:40:32 CEST] <jdarnley> Are you in a C obfuscation contest?
[17:00:20 CEST] <atomnuker> BBB: cmon dude I'm itching to work on ffav1
[17:00:27 CEST] <BBB> me too
[17:00:36 CEST] <BBB> you know that the holdup is not on my side
[17:00:55 CEST] <atomnuker> erm actually no, I thought you started and have been working on it for a month now
[17:00:59 CEST] <BBB> I have
[17:01:03 CEST] <BBB> its essentially finished
[17:01:26 CEST] <atomnuker> like, completely?
[17:01:43 CEST] <BBB> a few features are missing, but most things are there and it works correctly
[17:01:57 CEST] <atomnuker> with p-frames and loop restoration and global motion
[17:02:10 CEST] <BBB> no loop restoration yet
[17:02:17 CEST] <BBB> cdef/lpf works
[17:02:20 CEST] <BBB> rst not yet
[17:02:26 CEST] <atomnuker> rst?
[17:03:04 CEST] <BBB> my abbreviation for restoration
[17:03:19 CEST] <atomnuker> oh ok, is there simd yet?
[17:03:33 CEST] <BBB> some
[17:03:57 CEST] <atomnuker> k, I think I know where the holdup is, I'll try to get things moving
[17:06:22 CEST] <BBB> all I know, its way above my pay grade
[17:08:42 CEST] <atomnuker> BBB: how fast is it compared to aomdec for 1/many threads?
[17:08:49 CEST] <BBB> havent tested yet
[17:08:59 CEST] <BBB> but I dont think thats a useful question until they are both fully simd'ed
[17:09:16 CEST] <BBB> oh look, Im twice as fast <write some simd> oh now the other one is twice as fast
[17:09:27 CEST] <BBB> its about design for expected end speed
[17:09:34 CEST] <cone-441> ffmpeg 03Gyan Doshi 07master:07987404a8b7: ffmpeg: make loglevel verbose for frame duration warning
[17:10:43 CEST] <atomnuker> well, aomdec can barely do 1080p50 realtime, and would be nice for firefox to have 1080p60 playback on reasonably modern machines
[17:11:19 CEST] <atomnuker> rumor has it google are planning to roll out av1 on youtube in september
[17:12:19 CEST] <BBB> all that is way above my pay grade
[17:12:27 CEST] <gnafu> atomnuker: Is that when the first video finishes encoding?
[17:12:33 CEST] <BBB> :D
[17:12:42 CEST] <BBB> one star for gnafu
[17:12:49 CEST] <gnafu> ^^
[17:13:59 CEST] <atomnuker> BBB: anyway, would still be really interesting to know how they compare currently, could you do that if possible?
[17:14:02 CEST] <jdarnley> hahaa
[17:14:55 CEST] <BBB> not right away, but Ill try in the near future, sure
[17:35:40 CEST] <cone-441> ffmpeg 03Mark Thompson 07master:46fb150a81ab: lavfi/colorspace: Add namespace prefix to global functions
[17:44:01 CEST] <Gramner> BBB: memcpy(dst, src, offsetof(struct x, a6) + sizeof(dst->a6)) ?
[18:12:25 CEST] <jdarnley> Is anyone here familiar with the draw horizontal band feature?  Can you answer a few questions of mine?
[18:13:31 CEST] <jdarnley> You only call the function for new image data, right?
[18:14:19 CEST] <jdarnley> I should specify that I am asking in the context of a decoder, not a library user.
[18:15:14 CEST] <jdarnley> You don't call it for lines 0-15 then lines 0-31 but rather 0-15 and 16-31.
[18:15:18 CEST] <jdarnley> right?
[18:19:11 CEST] <durandal_1707> that misfeature should be removed
[18:21:24 CEST] <atomnuker> its alright, I don't mind the API
[18:21:44 CEST] <jdarnley> Maybe.  I might prefer returning chunks of frames as they're completed but the feature already exists.
[18:22:37 CEST] <atomnuker> eh, it makes more sense and is faster to work with rows rather than blocks
[18:23:21 CEST] <jdarnley> I do mean whole lines.
[18:23:45 CEST] <jdarnley> I am trying to stop using "slice" when I don't mean specific codec features.
[18:35:57 CEST] <jdarnley> Another question.  If a decoder is using draw_horiz_band() what is it supposed to do at the end of a frame?  Can it just return it as normal and the user is supposed to handle it all?
[18:54:19 CEST] <jdarnley> Why would attach sample files for common formats for which we already have decoders?
[21:37:50 CEST] <cone-914> ffmpeg 03Michael Niedermayer 07master:75a2db552423: avcodec/dpx: Check elements in 12bps planar path
[21:37:50 CEST] <cone-914> ffmpeg 03Michael Niedermayer 07master:ed22dc22216f: avformat/movenc: Check that frame_types other than EAC3_FRAME_TYPE_INDEPENDENT have a supported substream id
[21:37:50 CEST] <cone-914> ffmpeg 03Michael Niedermayer 07master:00e8181bd97c: avcodec/ac3_parser: Check init_get_bits8() for failure
[21:37:50 CEST] <cone-914> ffmpeg 03Michael Niedermayer 07master:95556e27e2c1: avformat/movenc: Do not pass AVCodecParameters in avpriv_request_sample
[21:37:50 CEST] <cone-914> ffmpeg 03Michael Niedermayer 07master:2aa9047486db: avcodec/mpeg4videodec: Check read profile before setting it
[21:37:51 CEST] <cone-914> ffmpeg 03Michael Niedermayer 07master:310a49f71df7: tools/target_dec_fuzzer: Also optionally fuzz with a parser
[21:37:51 CEST] <cone-914> ffmpeg 03Michael Niedermayer 07master:872ea3dfe565: avformat/movenc: Use mov->fc consistently for av_log()
[21:37:52 CEST] <cone-914> ffmpeg 03John Stebbins 07master:08b1d1d81225: h264_slice: Fix return of incomplete frames from decoder
[23:28:16 CEST] <kierank> jdarnley: iiuc there is a test for draw_horiz_band
[23:37:24 CEST] <cone-914> ffmpeg 03Jun Zhao 07master:2730a0fdf47c: configure: fix check for opencl_vaapi_intel_media.
[23:37:25 CEST] <cone-914> ffmpeg 03Jun Zhao 07master:e6ac328b79f4: hwcontext_opencl: Remove the opencl_device_init in opencl_device_derive
[23:37:26 CEST] <cone-914> ffmpeg 03Jun Zhao 07master:6b453a81bcb5: hwcontext_opencl: remove an unused variable
[00:00:00 CEST] --- Fri Jun 29 2018

More information about the Ffmpeg-devel-irc mailing list