[Ffmpeg-devel-irc] ffmpeg-devel.log.20140506

burek burek021 at gmail.com
Wed May 7 02:05:02 CEST 2014


[01:46] <cone-764> ffmpeg.git 03Michael Niedermayer 07master:5fc4c0097239: avfilter/af_bs2b: Fix incompatible pointer type warnings
[01:46] <cone-764> ffmpeg.git 03Michael Niedermayer 07master:2baf1c8c6080: avcodec/snowenc: Fix "incompatible pointer type" warning
[05:44] <cone-542> ffmpeg.git 03Michael Niedermayer 07master:141a84688f58: avcodec/mpeg4videodec: fix mcbpc error messages
[05:44] <cone-542> ffmpeg.git 03Michael Niedermayer 07master:7db0f9bef16d: avcodec/mpeg4videoenc: fix alternative implementation of mpeg4_encode_dc()
[09:44] <ubitux> i didn't know ioq3 was using opus
[09:44] <ubitux> that's pretty fun
[10:47] <j-b> Compn: as far as I know, I received, from Dolby, DTS, VIA, Bluray, One-Red, One-Blue, Philips, MPEG-LA
[14:07] <plepere> BBB, do you have any feedback to give me about my patch ?
[14:07] <BBB> not yet... I looked at earlier one and that was pretty good, so I don't mind if we just commit it
[14:07] <BBB> let's look now
[14:10] <BBB> what is the mcbuffer argument to the dsp functions?
[14:10] <BBB> I guess that's gone right?
[14:10] <BBB> so that's not an issue
[14:11] <BBB> lol @ intptr_t vs ptrdiff_t (that was a big flamewar back then and you're doing exactly what I wanted to do - use intptr_t - and the decision was to use ptrdiff_t :-p)
[14:11] <BBB> don't fix/change it, it's fine
[14:11] <BBB> just funny :)
[14:11] <plepere> ok ok. haha
[14:11] <BBB> c patch is ok
[14:11] <plepere> :)
[14:12] <plepere> the ASM is my work, so it's the most interesting for me
[14:13] <plepere> also, I've got the AVX2 machine back so I'll try to get on that if the current ASM is validated
[14:13] <BBB> I think it looks very similar to what you send privately a while ago, I see most of the comments were addressed, e.g. the weird 12/24 px functions are gone, etc.
[14:13] <BBB> so this is basically ok
[14:14] <plepere> thanks. :)
[14:14] <BBB> can we get it 1-2 instructions faster? sure, most probably, but this is a massive improvement, and the foundation (api/structure) is good, so no comments from me
[14:14] <BBB> what's next? lf or idct?
[14:14] <plepere> the LF is already done, I'll have to see if the author wants it patched.
[14:15] <BBB> lf isn't in ffmpeg yet is it?
[14:15] <plepere> no
[14:15] <plepere> we have it in openHEVC
[14:15] <BBB> ok, look forward to seeing that move over
[14:16] <plepere> yes
[14:16] <BBB> nice work, I really like this code
[14:16] <plepere> but once we've got the base MC, the rest of the ASM will come smoothly
[14:16] <BBB> do you guys have speed comparisons of other hevc decoders vs. your code?
[14:16] <BBB> or planning to get that once all asm is done?
[14:16] <JEEB> HEVC asm, nice
[14:17] <BBB> seeing same-quality or same-size hevc vs vp9 or hevc vs x264 would be nice also
[14:17] <BBB> but that's stuff for later I guess
[14:17] <plepere> I don't have other decoder speeds. I heard that the competitors were surprised to see that we could do 2 4K at 60fps decoding and display on a single PC.
[14:17] Action: BBB run to work now
[14:17] <BBB> bye
[14:17] <BBB> well that's mt right?
[14:17] <BBB> most decoders aren't mt
[14:18] <BBB> (frame-mt, not wpp/tile/slice)
[14:18] <plepere> yes, but it's still quite a burden on PCs to decode so much data
[14:18] <BBB> right
[14:18] <BBB> that's a good number, nice - I'll compare a little in a few weeks
[14:18] <BBB> for now - work!
[14:18] <plepere> tell me wxhen the patch is validated, the rest will follow shortly.
[14:18] <plepere> thank you
[14:18] <plepere> good lucj
[14:19] <plepere> *luck
[14:37] <cone-318> ffmpeg.git 03Lukasz Marek 07master:38b722e7d71c: gitignore: add tools/uncoded_frame
[14:37] <cone-318> ffmpeg.git 03Matt Oliver 07master:ac9869ffb070: x86/mpegaudiodsp.c: msvc compilation error without sse/avx_external
[15:47] <cone-318> ffmpeg.git 03Marton Balint 07master:a82e87618fc0: mpegts: remove uneeded buf_size check
[15:47] <cone-318> ffmpeg.git 03Marton Balint 07master:9ba8debce994: mpegts: only emit new packets if data buffer exists
[15:47] <cone-318> ffmpeg.git 03Marton Balint 07master:e2752de34458: mpegts: fix indentation after last commit
[15:47] <cone-318> ffmpeg.git 03Marton Balint 07master:d7ca91498843: mpegts: factorize pes packet state reset function
[15:48] <cone-318> ffmpeg.git 03Marton Balint 07master:a78274329c0b: mpegts: unref buffer in reset_pes_packet_state
[15:48] <cone-318> ffmpeg.git 03Marton Balint 07master:6c5375100c4f: mpegts: always reset pes packet state on new packet
[15:48] <cone-318> ffmpeg.git 03Michael Niedermayer 07master:760a9ef45184: Merge remote-tracking branch 'cus/stable'
[18:26] <cone-318> ffmpeg.git 03Mickaël Raulet 07master:83976e40e896: hevc: C code update for new motion compensation
[18:26] <cone-318> ffmpeg.git 03plepere 07master:7a2491c436af: HEVC : added assembly MC functions
[18:26] <cone-318> ffmpeg.git 03Michael Niedermayer 07master:3b3db02f2e5c: avcodec/x86/hevcdsp_init: fix build on 32bit
[18:26] <cone-318> ffmpeg.git 03Michael Niedermayer 07master:7be230b5fa58: avcodec/x86/Makefile: remove duplicate line
[18:29] <jamrial> michaelni: you missed my email about hevc asm
[18:30] <JEEB> oh, someone wrote asm that works on both 32bit and 64bit?
[18:30] <JEEB> kind of surprised
[18:31] <nevcairiel> nah, its all 64-bit
[18:31] <nevcairiel> although a bunch of the functions look like they could just work on 32-bit
[18:32] <cone-318> ffmpeg.git 03Michael Niedermayer 07master:fc7d0d82017d: avcodec/x86/hevcdsp_init: fix SSE4 checks
[18:33] <jamrial> Thanks
[19:13] <j-b> move DSD-to-PCM conversion from libavcodec/dsddec into libswresample
[19:14] <j-b> what does this mean?
[19:20] <av500> j-b: its tech talk
[19:44] <j-b> av500: sure, but aren't all decoders a converter to PCM?
[19:48] <nevcairiel> one could argue that, but DSD is technically special, as its a true alternative to PCM and not just a way to store PCM compressed
[19:52] <j-b> nevcairiel: meaning that I need libswresample then to decode it
[20:01] <jamrial> seems that a big chunk of the hevc asm is either ssse3 or sse2 after all
[20:01] <jamrial> hevcdps_init.c macros will need quite a lot of changes to get that working, lol
[20:05] <jamrial> and like nevcairiel said, many can work just fine in x86
[20:07] <jamrial> just found one that's using registers xmm0 to xmm4, then xmm9, xmm14 and xmm15 for no reason
[20:08] <jamrial> this is terrible for win64
[20:08] <jamrial> i may be missing something, though
[20:08] <nevcairiel> a lot of the functions declare way high xmm usage counts, i've been wondering if they really all need that many
[20:11] <jamrial> for example http://pastebin.com/Ywt7iWhh
[20:14] <jamrial> it only needs seven xmm registers, but it's using x86_64 registers
[20:50] <wm4> shouldn't this DSD thing be a codec?
[20:52] <j-b> wm4: thanks :)
[20:55] <kierank> http://www.euclidiq.com/euclidiq-announces-the-results-of-recent-mpeg-2-compression-testing/
[20:55] <kierank> lol comparing to ffmpeg mpeg2
[20:55] <thardin> best encoder
[20:55] <iive> what happened with x262?
[20:55] <JEEB> lavc mpeg-2 might be good, but it sure means you'd need to know all of those non-documented options :D
[20:56] <JEEB> that said, I've never tested it and have no idea about how good it actually is
[20:56] <wm4> why can't it have sane presets?
[20:56] <JEEB> because no-one cares about it I guess
[20:56] <JEEB> just like the mpeg4 encoder
[20:58] <Skyler_> I think it had the problem of having so many options and being so modular and trying to do everything
[21:00] <kierank> iive: kind of works ok, some people use it
[21:00] <kierank> needs some quality improvement though
[21:22] <j-b> jb
[21:22] <j-b> oops, wrong window :)
[21:25] <kierank> Ed balls
[21:26] <iive> people, stop pasting your passwords in this publicly logged and indexed channel.
[21:26] <iive> ;)
[21:47] <nevcairiel> i totally expected msvc to break with the new asm, oh well
[21:47] <JEEB> sometimes surprises do happen
[21:47] <j-b> iive: nice pwd :)
[21:47] <nevcairiel> its not only msvc though, also mingw on windows
[21:48] <nevcairiel> and only 64-bit, so it must be causing issues with the x64 abi
[21:48] <nevcairiel> on the other hand, the asm isnt in 32-bit.... :P
[21:48] <JEEB> :D
[21:49] <nevcairiel> i would guess the number of regs is wrong on a bunch of functions, which only showed up on windows in previous cases
[21:57] <Daemon404> :D
[00:00] --- Wed May  7 2014


More information about the Ffmpeg-devel-irc mailing list