[Ffmpeg-devel-irc] ffmpeg-devel.log.20150823

burek burek021 at gmail.com
Mon Aug 24 02:05:03 CEST 2015


[00:51:07 CEST] <cone-529> ffmpeg 03Rostislav Pehlivanov 07master:f2ba60d4e9ca: aacenc_tns: temporarily disable
[02:05:03 CEST] <cone-529> ffmpeg 03Rostislav Pehlivanov 07master:4f396a6f3024: fate: temporarily disable AAC TNS test
[02:59:19 CEST] <cone-529> ffmpeg 03Ganesh Ajjanagadde 07master:6638e4a950a0: avcodec/x86/mpegaudiodsp: correct asm guards
[03:20:29 CEST] <atomnuker> swb_convert from AV_SAMPLE_FMT_FLTP to whatever expects a normalized float input, right?
[03:24:23 CEST] <atomnuker> I need to convert spectral coefficients to int32_t because our LPC system only works with that format
[03:24:36 CEST] <atomnuker> and the spectral coefficients are not normalized
[03:28:43 CEST] <atomnuker> yeah, nevermind that, it'll be better if our LPC system can work with both floats and int32_t's
[03:30:49 CEST] <atomnuker> internally lpc.c uses doubles so it'll be easy
[04:41:27 CEST] <cone-529> ffmpeg 03Michael Niedermayer 07master:b4d68e7cdb3b: avformat: Remove use of AVFrac and AVStream->pts
[05:04:56 CEST] <jamrial> looks like 095347ffe broke doc/print_options in msvc
[05:15:40 CEST] <rcombs> AVFrak
[08:21:46 CEST] <rcombs> https://gist.github.com/d1d4a6080ef641764824 <-- lavf doesn't like seeking in this file much
[08:23:10 CEST] <rcombs> (that's the entirety of the Cues)
[08:40:07 CEST] <nevcairiel> jamrial: i wonder why host compiles dont use the compat symbols... maybe because it was never needed before? But that seems like a limited point of view =p
[08:46:20 CEST] <nevcairiel> hm yeah apparently configure  just isnt setup to track host compat objects
[08:52:39 CEST] <nevcairiel> could add some evil defines in print_options.c to work-around the problem, but that isn't very pretty either
[08:53:49 CEST] <rcombs> https://gist.github.com/1be9f707d6e57a2d5197 this look sane?
[08:57:36 CEST] <nevcairiel> that sounds look like a broken index though
[08:57:40 CEST] <nevcairiel> just a very sparse one
[08:57:53 CEST] <nevcairiel> :D
[08:58:54 CEST] <jamrial> nevcairiel: print_options.c already has an evil define, so adding another shouldn't be a problem :p
[08:59:36 CEST] <nevcairiel> well its more like a bunch, excluding avutil/internal.h and defining null deprecation macros
[08:59:40 CEST] <nevcairiel> so..3
[08:59:42 CEST] <jamrial> defining AVUTIL_INTERNAL_H in print_options.c so that header isn't included should fix it
[08:59:47 CEST] <jamrial> i think
[08:59:55 CEST] <nevcairiel> wouldnt it w hine about missing defines
[08:59:58 CEST] <jamrial> ah right
[13:07:35 CEST] <cone-952> ffmpeg 03Paul B Mahol 07master:6be5b05fb134: avfilter/vf_histogram: levels: support more input pixel formats
[17:10:29 CEST] <lglinskih> kierank: hi! now my test supports a lot of tests for aac, one test for mp3float, imc, amrnb, flv, h264, flac (last 3 are old). Choice was pretty random, just for testing of a program.
[17:10:56 CEST] <lglinskih> here is makefile: https://github.com/lglinskih/FFmpeg/blob/63bbb4994d03909c127b6748d45820f18dbe0dc6/tests/fate/api.mak
[17:11:28 CEST] <lglinskih> and the program: https://github.com/lglinskih/FFmpeg/blob/63bbb4994d03909c127b6748d45820f18dbe0dc6/tests/api/api-decode-test.c
[19:25:14 CEST] <BtbN> yay, vaapi hevc doesn't fail or assert anymore
[19:25:21 CEST] <BtbN> now it just spits out green or white frames
[19:27:06 CEST] <wm4> progress
[19:27:19 CEST] <JEEB> 20:25 <@BtbN> yay, vaapi hevc doesn't fail or assert anymore
[19:27:23 CEST] <JEEB> ugh
[19:27:33 CEST] <JEEB> sorry
[19:28:24 CEST] <BtbN> Well, i'm now down to filling in the remaining parameters i don't yet translate from ffmpeg hevc to vaapi
[19:28:34 CEST] <BtbN> And hope that will make it decode correctly
[19:31:24 CEST] <philipl> BtbN: and I thought parameter mapping for vdpau was a pain.
[19:31:55 CEST] <BtbN> https://github.com/BtbN/FFmpeg/tree/vaapi_hevc any help is apreciated ;)
[19:32:08 CEST] <BtbN> there are some parameters i have no idea where to get them from ffmpeg yet
[19:32:37 CEST] <philipl> The vdpau header explains where in the spec each field comes from. Doesn't look like vaapi has that. :-/
[19:32:48 CEST] <philipl> Is the gstreamer equivalent no use? Can you even verify it works?
[19:33:50 CEST] <BtbN> gstreamer names most flags in its hevc structs the exact same way libva does
[19:34:14 CEST] <BtbN> so they just do slice_hdr->collocated_from_l0_flag = slice_param->collocated_from_l0_flag;
[19:34:25 CEST] <philipl> heh. But you can trace back to how the values are obtained - though that's still frightfully convoluted, I'm sure.
[19:34:26 CEST] <BtbN> ffmpeg just doesn't have any flag like that.
[19:34:49 CEST] <BtbN> I'm also still not sure how to correctly calculate the slice header offset it wants
[19:35:02 CEST] <philipl> hrm
[19:35:12 CEST] <BtbN> gst does some calculation with emulation prevention bytes
[19:35:19 CEST] <BtbN> can't find anything like that in ffmpeg
[19:36:06 CEST] <BtbN> I suspect this calculation to be the reason it decodes only white/green
[19:36:27 CEST] <philipl> When in doubt read the spec, I fear.
[19:36:36 CEST] <philipl> colocated_from_l0 is in there, for example
[19:36:49 CEST] <BtbN> yes, the libva parameters are very close to the spec actualy
[19:37:01 CEST] <BtbN> the problem is finding where to get those flags in the ffmpeg hevc structures
[19:37:20 CEST] <philipl> Likely you need to find the code that's walking through the bitstream and save more bits.
[19:39:35 CEST] <BtbN> there is no notion of emulation prevention in the entire ffmpeg hevc code
[19:39:48 CEST] <BtbN> or i don't know what to search for
[19:40:02 CEST] <philipl> hevc.c:644
[19:40:16 CEST] <philipl> sh->collocated_list is collocated_from_l0_flag
[19:40:19 CEST] <philipl> for example
[19:41:16 CEST] <philipl> Find the flag you need in the pseudo-code, then find a flag you can find in our code and find the code that implements the pseudo-code, then look for the part that does the mysterious flag
[19:42:12 CEST] <philipl> So, I think: collocated_from_l0_flag = sh->collocated_list == L0;
[19:42:37 CEST] <philipl> No, it's just the value.
[19:42:42 CEST] <philipl> of sh->collocated_list
[19:42:43 CEST] <BtbN> It's just 1 bit anyway
[19:43:35 CEST] <BtbN> I still have to collect which picture flags i'm missing
[19:43:43 CEST] <BtbN> Only have a few ones handled yet
[19:44:00 CEST] <BtbN> slice flags seem almost complete
[19:46:23 CEST] <philipl> just have to grind through it.
[19:48:05 CEST] <BtbN> the problem when it finaly "does something" is that you can't propperly debug it decoding white/green frames.
[19:48:17 CEST] <BtbN> tracing an assert is easy
[20:00:56 CEST] <philipl> yeah. just wait until it works for some frames and not others. :-)
[20:08:03 CEST] <cone-775> ffmpeg 03Ganesh Ajjanagadde 07master:9aaac0410767: avdevice/lavfi: fix self assignment warning
[20:12:49 CEST] <jamrial> BBB, nevcairiel: so libav's hevc mc is apparently not optimal. many functions seem to be loading eight bytes at a time, even some that process 64 pixels 10bit
[20:13:15 CEST] <BBB> ours isnt optimal either
[20:13:23 CEST] <BBB> which is faster?
[20:13:23 CEST] <BBB> :)
[20:14:11 CEST] <jamrial> can't really test :p
[20:14:19 CEST] <jamrial> on my k10 theirs is, because they have sse2 mc :p
[20:16:00 CEST] <jamrial> some functions are also extremely bottlenecked by using only two xmm registers, so almost every instruction needs to wait for the previous to finish
[20:16:10 CEST] <nevcairiel> i think our mc is the only one that hasnt been thoroughly reviewed and/or rewritten, so it c ould go either way
[20:16:37 CEST] <nevcairiel> "the only" as in hevc simd
[20:42:48 CEST] <Daemon404> ... wut
[20:49:11 CEST] <nevcairiel> so no traction on the ML, lets increase the pressure and create an issue? =P
[20:53:29 CEST] <jamrial> maybe he's not subscribed to the ml and didn't see the replies
[23:12:54 CEST] <jamrial> pity the print_options.c issue. the change to options_table.h reduced the amount of warnings from ~140 to ~80
[23:15:30 CEST] <nevcairiel> are those at least any the bump would get rid of
[23:15:38 CEST] <kierank> lglinskih: ok
[23:15:41 CEST] <kierank> please add more formats :)
[23:17:14 CEST] <jamrial> probably
[23:19:43 CEST] <lglinskih> kierank: ok)
[23:21:29 CEST] <cone-861> ffmpeg 03Ganesh Ajjanagadde 07master:24e6729a049e: swresample/dither: use integer arithmetic
[23:39:38 CEST] <cone-861> ffmpeg 03Andreas Cadhalpun 07master:03e71cccba28: avcodec: remove FF_API_LOWRES around max_lowres
[23:39:39 CEST] <cone-861> ffmpeg 03Andreas Cadhalpun 07master:01594ebb077c: options_table: fix compatibility with MSVC
[23:39:40 CEST] <cone-861> ffmpeg 03Andreas Cadhalpun 07master:1bf76cd2db0a: rtpenc_chain: also copy AVFMT_FLAG_BITEXACT to new AVFormatContext
[00:00:00 CEST] --- Mon Aug 24 2015


More information about the Ffmpeg-devel-irc mailing list