[Ffmpeg-devel-irc] ffmpeg-devel.log.20131117

burek burek021 at gmail.com
Mon Nov 18 02:05:02 CET 2013


[00:01] <cone-755> ffmpeg.git 03Anton Khirnov 07master:5b9c3b450520: Replace all instances of avcodec_alloc_frame() with av_frame_alloc().
[00:01] <cone-755> ffmpeg.git 03Michael Niedermayer 07master:0ee905e24307: Merge commit '5b9c3b4505206143d85398c1410949319fa1180f'
[00:10] <cone-755> ffmpeg.git 03Anton Khirnov 07master:b9fb59d2ab05: lavc: deprecate avcodec_alloc_frame().
[00:11] <cone-755> ffmpeg.git 03Michael Niedermayer 07master:29c83d23e8ed: Merge commit 'b9fb59d2ab05fdfe89d3fb0d7ecbbd91e560f57d'
[00:20] <cone-755> ffmpeg.git 03Anton Khirnov 07master:13e9cc9ce064: a64multienc: use the AVFrame API properly.
[00:20] <cone-755> ffmpeg.git 03Anton Khirnov 07master:6139f481ac9f: asvenc: use the AVFrame API properly.
[00:20] <cone-755> ffmpeg.git 03Michael Niedermayer 07master:cc4a6435638f: Merge commit '6139f481ac9feb1bee4e7d04789fb15d7f24ebbf'
[00:36] <cone-755> ffmpeg.git 03Anton Khirnov 07master:33329c6e872f: avs: use the AVFrame API properly.
[00:36] <cone-755> ffmpeg.git 03Anton Khirnov 07master:a83fd5010c4e: bethsoftvid: use the AVFrame API properly.
[00:36] <cone-755> ffmpeg.git 03Anton Khirnov 07master:670c74518f2d: c93: use the AVFrame API properly.
[00:36] <cone-755> ffmpeg.git 03Anton Khirnov 07master:cec5ce49229d: cdxl: remove an unused variable
[00:37] <cone-755> ffmpeg.git 03Michael Niedermayer 07master:21c41e76d7c1: Merge commit 'cec5ce49229d61e4eb1f331a6d0dff3aa24f6655'
[00:58] <cone-755> ffmpeg.git 03Anton Khirnov 07master:4a8a35bc1ffe: cinepak: use the AVFrame API properly.
[00:58] <cone-755> ffmpeg.git 03Anton Khirnov 07master:845020ed89ca: cljr: use the AVFrame API properly.
[00:58] <cone-755> ffmpeg.git 03Anton Khirnov 07master:224b6edd8bea: qtrle: use the AVFrame API properly.
[00:58] <cone-755> ffmpeg.git 03Anton Khirnov 07master:a639ea7f4bc4: escape124: use the AVFrame API properly.
[00:58] <cone-755> ffmpeg.git 03Michael Niedermayer 07master:bfb1f44d246f: Merge commit 'a639ea7f4bc44bf6bfa452675558a342924a66a9'
[01:14] <cone-755> ffmpeg.git 03Anton Khirnov 07master:ef2a99c7f4c4: msvideo1: use the AVFrame API properly.
[01:14] <cone-755> ffmpeg.git 03Anton Khirnov 07master:01de3c1dd57a: cin video: use the AVFrame API properly.
[01:14] <cone-755> ffmpeg.git 03Anton Khirnov 07master:55f954e7678f: qpeg: use the AVFrame API properly.
[01:14] <cone-755> ffmpeg.git 03Anton Khirnov 07master:2d2a92f72199: dxa: use the AVFrame API properly.
[01:14] <cone-755> ffmpeg.git 03Michael Niedermayer 07master:5219afc09d8e: Merge commit '2d2a92f72199823a92e4e226c32e42a27ec801c0'
[01:28] <cone-755> ffmpeg.git 03Anton Khirnov 07master:b7462a3904d7: jvdec: use the AVFrame API properly.
[01:28] <cone-755> ffmpeg.git 03Anton Khirnov 07master:952f943d40e9: FATE: add a jv test
[01:28] <cone-755> ffmpeg.git 03Michael Niedermayer 07master:ddfdcd2b5ecc: Merge commit 'b7462a3904d71ff799584faf5b875cad59ca2f31'
[01:28] <cone-755> ffmpeg.git 03Michael Niedermayer 07master:10ac6f3ce9ae: Merge commit '952f943d40e9dbcd68bc851460cc99268091a294'
[01:43] <cone-755> ffmpeg.git 03Anton Khirnov 07master:c5dfb9033f52: msrle: use the AVFrame API properly.
[01:43] <cone-755> ffmpeg.git 03Anton Khirnov 07master:4b8a1941465c: iff: use the AVFrame API properly.
[01:43] <cone-755> ffmpeg.git 03Anton Khirnov 07master:79d501a860bd: indeo2: use the AVFrame API properly.
[01:43] <cone-755> ffmpeg.git 03Anton Khirnov 07master:2e09096da912: kgv1: use the AVFrame API properly.
[01:43] <cone-755> ffmpeg.git 03Michael Niedermayer 07master:9f890a165666: Merge commit '2e09096da912f563c4dd889a8f25c314529bbaa6'
[01:59] <cone-755> ffmpeg.git 03Anton Khirnov 07master:5c96f02901d8: mmvideo: use the AVFrame API properly.
[01:59] <cone-755> ffmpeg.git 03Anton Khirnov 07master:84099f51f397: smacker: use the AVFrame API properly.
[01:59] <cone-755> ffmpeg.git 03Anton Khirnov 07master:d100f9e7cd6d: flicvideo: use the AVFrame API properly.
[01:59] <cone-755> ffmpeg.git 03Anton Khirnov 07master:a837c4f2df96: zmbvenc: use the AVFrame API properly.
[01:59] <cone-755> ffmpeg.git 03Michael Niedermayer 07master:fe3808eddee8: Merge commit 'a837c4f2df96a30bf9aa4115b426d608487c7101'
[02:18] <cone-755> ffmpeg.git 03Anton Khirnov 07master:6886a85f5ca9: zerocodec: use the AVFrame API properly.
[02:18] <cone-755> ffmpeg.git 03Anton Khirnov 07master:f3cd23fbc849: xxan: use the AVFrame API properly.
[02:18] <cone-755> ffmpeg.git 03Anton Khirnov 07master:04f30711d8d2: xan: use the AVFrame API properly.
[02:18] <cone-755> ffmpeg.git 03Anton Khirnov 07master:3c8ea9d4a74f: vmnc: use the AVFrame API properly.
[02:18] <cone-755> ffmpeg.git 03Michael Niedermayer 07master:8af7774c7aca: Merge commit '3c8ea9d4a74fd4d7493d40c818ca64ee492709f3'
[02:33] <cone-755> ffmpeg.git 03Anton Khirnov 07master:aca214783af9: vmdvideo: use the AVFrame API properly.
[02:33] <cone-755> ffmpeg.git 03Anton Khirnov 07master:e52f38bc2ea6: motionpixels: use the AVFrame API properly.
[02:33] <cone-755> ffmpeg.git 03Anton Khirnov 07master:e8ef8a328c81: rpza: use the AVFrame API properly.
[02:33] <cone-755> ffmpeg.git 03Anton Khirnov 07master:4a4841d4e0f0: fraps: use the AVFrame API properly.
[02:33] <cone-755> ffmpeg.git 03Michael Niedermayer 07master:4362f272c0ae: Merge commit '4a4841d4e0f0dc50998511bf6c48b518012024db'
[02:40] <cone-755> ffmpeg.git 03Anton Khirnov 07master:42d40fd28cea: truemotion1: use the AVFrame API properly.
[02:40] <cone-755> ffmpeg.git 03Anton Khirnov 07master:32f7cf9bc38a: truemotion2: use the AVFrame API properly.
[02:40] <cone-755> ffmpeg.git 03Anton Khirnov 07master:68af7f54a6cd: smc: use the AVFrame API properly.
[02:40] <cone-755> ffmpeg.git 03Anton Khirnov 07master:508b37557bf3: tiertexseqv: use the AVFrame API properly.
[02:40] <cone-755> ffmpeg.git 03Michael Niedermayer 07master:72df87088c8a: Merge commit '508b37557bf36eae83c18e64d42f27b44a321d81'
[02:52] <smarter> BBB: hevs use 7-taps and 8-taps for luma (which is qpel), and 4-taps for chroma (which is eighth-pel)
[02:52] <smarter> *hevc
[02:54] <smarter> pretty sure there's nothing to share in the loop filter
[02:55] <smarter> and I'm not really familiar with vp9 intra but I doubt it's similar enough
[02:55] <smarter> maybe with a lot of conditionals
[02:55] <smarter> like, doesn't vp9 intra ignore the top-right pixels for some reasons ?
[02:56] <smarter> the way hevc does the angular mode is pretty specific, I'd be worried about lawsuits if VP9 did the same :]
[02:56] <smarter> *modes
[02:56] <cone-755> ffmpeg.git 03Anton Khirnov 07master:acaffdca21f6: mss1: use the AVFrame API properly.
[02:56] <cone-755> ffmpeg.git 03Anton Khirnov 07master:e9198f61db90: mss2: use the AVFrame API properly.
[02:56] <cone-755> ffmpeg.git 03Anton Khirnov 07master:207909911d31: mss3: use the AVFrame API properly.
[02:56] <cone-755> ffmpeg.git 03Anton Khirnov 07master:730bac7bab3c: mss4: use the AVFrame API properly.
[02:56] <cone-755> ffmpeg.git 03Michael Niedermayer 07master:8d193a24f2da: Merge commit '730bac7bab3c7dcd9fcb7c70f154e5f4cfaef9a7'
[02:58] <smarter> [14:07:35] <BBB> I always tended to split dsp stuff in its own header since arch-specific code depends on that header
[02:58] <smarter> [14:07:41] <BBB> and you obviously don't want them to include all of vp9.h
[02:58] <smarter> why not?
[03:00] <smarter> BBB: if you care about code reuse, the {luma,chroma}_mc functions in hevc.c are very much inspired by vp8_mc_{luma,chroma}
[03:01] <smarter> and I expect vp9 to be similar
[03:02] <smarter> so it might make sense to share code
[03:06] <BBB> smarter: but they're 8tap, not 4/6tap, right?
[03:06] <BBB> oh, you mean decoder
[03:06] <BBB> those are identical in all decoders, but they're not simd'able, so not easily sharable
[03:07] <BBB> smarter: vp9.h includes a lot more than stuff that is relevant for dsp
[03:07] <BBB> smarter: dsp stuff is best left alone, even if just as an academic exercise
[03:07] <BBB> (alone as in "in it's own files"
[03:07] <BBB> )
[03:08] <BBB> smarter: and for intra pred, you've never seen h264pred.c, it's a ifdef jungle
[03:08] <BBB> smarter: but it works
[03:09] <smarter> looking at it now, I don't see too many ifdef
[03:10] <BBB> check the function assignments
[03:10] <BBB> if (codec == ...) func = this; else if (codec == ...) func = that; else ..
[03:10] <smarter> ah right :)
[03:10] <BBB> of course the simd functions themselves are not if'ed, but that'd be silly for performance critical infrastructure
[03:11] <BBB> so you could do that for vp9/hevc as well if they share stuff conceptually
[03:11] <BBB> and some implementation is shared
[03:11] <BBB> is hevc 8tap for luma and bilin for chroma?
[03:11] <BBB> or is it similar to the staged qpel in h264?
[03:12] <smarter> I wrote about that just before
[03:12] <smarter> [02:52:08] <smarter> BBB: hevs use 7-taps and 8-taps for luma (which is qpel), and 4-taps for chroma (which is eighth-pel)
[03:12] <smarter> there's no staged thing
[03:12] <BBB> ohright sorry
[03:12] <BBB> 4tap
[03:12] <BBB> that's weird
[03:12] <BBB> ohwell
[03:12] <smarter> reasoning is that chroma is less important
[03:12] <BBB> yes
[03:13] <BBB> but 4pel is weird
[03:13] <BBB> h264's bilin was sort of logical
[03:13] <BBB> 4pel is sort of halfway in between
[03:13] <BBB> ohwell I guess it makes sense
[03:13] <smarter> have you seen pred_angular in hevcpred_template.c ?
[03:14] <BBB> no, I heard there's 40 modes or so
[03:14] <smarter> yeah, but they're all in one function with tons of conditionals
[03:14] <BBB> ugh
[03:14] <BBB> dude
[03:14] <BBB> fix that :)
[03:15] <smarter> not much to fix in the C code
[03:15] <smarter> yasm macros should help
[03:16] <smarter> but I doubt that vp9 angular modes are similar enough to that
[03:18] <smarter> (there's also the incredible mess that is the intra_pred function in hevcpred_template.c to make the missing border samples available)
[03:22] <BBB> yeah that's a mess in libvpx also
[03:22] <BBB> ffvp9 did that quite nicely imo
[03:24] <cone-755> ffmpeg.git 03Anton Khirnov 07master:6a08d7fcc528: sgienc: use the AVFrame API properly.
[03:24] <cone-755> ffmpeg.git 03Anton Khirnov 07master:2a85a192c710: bmpenc: use the AVFrame API properly.
[03:24] <cone-755> ffmpeg.git 03Anton Khirnov 07master:4e7e8055a7b6: dpxenc: use the AVFrame API properly.
[03:24] <cone-755> ffmpeg.git 03Anton Khirnov 07master:e4155f15b35c: eamad: use the AVFrame API properly.
[03:24] <cone-755> ffmpeg.git 03Michael Niedermayer 07master:9ad477d9098b: Merge commit 'e4155f15b35c4272a235f5521d2dc6c2aabdd462'
[03:37] <cone-755> ffmpeg.git 03Anton Khirnov 07master:042aec41ae0b: flashsv: use the AVFrame API properly.
[03:37] <cone-755> ffmpeg.git 03Anton Khirnov 07master:6792559f8aa8: vc1: use the AVFrame API properly.
[03:37] <cone-755> ffmpeg.git 03Anton Khirnov 07master:c8a525197fb5: ulti: use the AVFrame API properly.
[03:37] <cone-755> ffmpeg.git 03Anton Khirnov 07master:d48c20630214: qtrleenc: use the AVFrame API properly.
[03:37] <cone-755> ffmpeg.git 03Michael Niedermayer 07master:92cbd7756872: Merge commit 'd48c20630214a4effcc920e93a5044bee4e2002e'
[09:10] <Zeranoe> Any ideas on this one: http://ffmpeg.zeranoe.com/forum/viewtopic.php?f=7&t=1468
[11:27] <JEEB> hmm, I think I'm relatively close to the demuxing and decoding example, but it seems like I'm getting some errors with some samples. Which either means that I'm missing extradata (although I'm using the lavc context provided by lavf, which I think should have that?), or that I'm set way too strict on the decoding side of things.
[11:28] <ubitux> can you reproduce the issue with the example?
[11:28] <JEEB> lemme build that one, too
[11:29] <JEEB> it's integrated into the makefile, right?
[11:29] <ubitux> see doc/examples/README
[11:29] <JEEB> k
[11:37] <JEEB> argh, should have just opted for in-tree compilation and not having a suffix
[11:39] <JEEB> and I didn't build swresample...
[11:39] <JEEB> at least I got the .o for the demuxing and decoding example
[11:50] <JEEB> great, finally linked it manually
[11:54] <cone-904> ffmpeg.git 03Anton Khirnov 07master:0ea430c75b8d: lclenc: use the AVFrame API properly.
[11:54] <cone-904> ffmpeg.git 03Michael Niedermayer 07master:85b7b0c519f8: Merge commit '0ea430c75b8d90449d2878ad84669a2da2ad3cbc'
[11:55] <JEEB> the example seems to work
[11:57] <JEEB> also it seems to output a ffmpeg-like input info screen, funky
[11:58] <ubitux> av_dump_ thing?
[11:59] <JEEB> yea
[11:59] <JEEB> the difference pretty much is that I grab the lavc context from av_find_best_stream, while the example uses open_codec_context
[12:01] <ubitux> open_codec_context does a av_find_best_stream
[12:01] <ubitux> and then find a decoder and open it
[12:01] <JEEB> lol
[12:01] <ubitux> (you noticed it's a local function right?)
[12:02] <JEEB> yes, I've just woken up
[12:02] <JEEB> I did read through it some times last night
[12:02] <JEEB> https://github.com/jeeb/matroska_thumbnails/blob/master/cli_test/cli_test.cpp#L61
[12:02] <JEEB> yeah, rather similar
[12:03] <JEEB> (that's basically a sample thing I've been doing to test basic one-frame decoding and swscale
[12:06] <ubitux> (you need to unref the frame if you use the ref counting system)
[12:06] <JEEB> yes, but I only decode one
[12:06] <JEEB> aka until the first can_has_picture
[12:07] <JEEB> http://up-cat.net/p/4b250f33
[12:07] <JEEB> I have a single H.264 sample that fails like this, and a H.263 sample
[12:07] <JEEB> a HEVC sample works
[12:07] <JEEB> (and a VOB with MPEG-2 works)
[12:10] <JEEB> with -v debug with that sample I get [h264 @ 04bf51e0] no picture [h264 @ 04c4bb00] no picture
[12:10] <JEEB> in ffmpeg
[12:10] <JEEB> which might be related?
[12:10] <JEEB> or I just fail at extradata somewhere (´
[12:10] <nevcairiel> that just happens on a few badly cut streams
[12:11] <JEEB> well, the question is -- how does the demuxing and decoding sample then brawl through this just fine?
[12:11] <JEEB> although I guess it might something as simple as the decoding loop being more OK with certain kind of failures?
[12:12] <JEEB> although I didn't notice anything special
[12:12] <cone-904> ffmpeg.git 03Anton Khirnov 07master:78c6c9d6c1ac: pngenc: use the AVFrame API properly.
[12:12] <JEEB> which thus again leads to it most probably being PEBKAC of my own
[12:12] <cone-904> ffmpeg.git 03Anton Khirnov 07master:f26f71139faf: tiffenc: use the AVFrame API properly.
[12:13] <cone-904> ffmpeg.git 03Anton Khirnov 07master:fb26d761676f: targaenc: use the AVFrame API properly.
[12:13] <cone-904> ffmpeg.git 03Anton Khirnov 07master:45bde93eefa7: sunrastenc: use the AVFrame API properly.
[12:13] <cone-904> ffmpeg.git 03Michael Niedermayer 07master:3ea168edeb7a: Merge commit '45bde93eefa78c1bdb0936109fbd2e2fb27fbfe7'
[12:14] <nevcairiel> the nal size messages basically mean that it was trying to read some mp4 nals with those sizes, and the sizes are just wrong
[12:15] <JEEB> yes
[12:16] <nevcairiel> so, is it actually a mp4/mkv?
[12:16] <JEEB> but since the demuxing_decoding example doesn't get that I'm going to assume that I'm doing something wrong here
[12:16] <JEEB> it is mkv
[12:16] <nevcairiel> your file doesnt happen to have audio, does it?
[12:16] <JEEB> it has audio, yes
[12:17] <nevcairiel> because you blindly decode any packet with the video decoder
[12:17] <nevcairiel> give it audio, it stops liking you
[12:17] Action: JEEB facepalms
[12:18] <JEEB> so I basically have to check what av_read_frame returns
[12:18] <nevcairiel> at least its an easy fix, compare packet.stream_index or what its called to avstream.index or something
[12:20] <JEEB> I somehow was just going through writing the code last night and didn't even think about what av_read_frame returns
[12:32] <cone-904> ffmpeg.git 03Anton Khirnov 07master:219b35f5d169: gif: use the AVFrame API properly.
[12:32] <cone-904> ffmpeg.git 03Anton Khirnov 07master:394ef4d18f37: svq1enc: use the AVFrame API properly.
[12:32] <cone-904> ffmpeg.git 03Anton Khirnov 07master:d2287740d932: libx264: use the AVFrame API properly.
[12:32] <cone-904> ffmpeg.git 03Anton Khirnov 07master:57e7b3a89f5a: dnxhdenc: use the AVFrame API properly.
[12:32] <cone-904> ffmpeg.git 03Michael Niedermayer 07master:5b0c70c2499e: Merge commit '57e7b3a89f5a0879ad039e8f04273b48649799a8'
[12:36] <JEEB> oh I bet I'm going to be having fun
[12:36] <JEEB> http://up-cat.net/p/76a4f768
[12:36] <JEEB> this change causes the output binary to crash at launch
[12:36] <JEEB> and I'm not even joking
[12:38] <JEEB> (ok, it seems to work in cmd.exe now while it crashed in msys's shell...)
[12:40] <JEEB> (also debugging the crashing executable would crash visual studio :D )
[12:41] <cone-904> ffmpeg.git 03Anton Khirnov 07master:d56a114adc54: flashsvenc: use the AVFrame API properly.
[12:41] <cone-904> ffmpeg.git 03Anton Khirnov 07master:5290e02a0a49: nuv: use the AVFrame API properly.
[12:41] <cone-904> ffmpeg.git 03Anton Khirnov 07master:b18c7c8d3ddf: libxavs: use the AVFrame API properly.
[12:41] <cone-904> ffmpeg.git 03Anton Khirnov 07master:97168b204a0b: eatgv: use the AVFrame API properly.
[12:41] <cone-904> ffmpeg.git 03Michael Niedermayer 07master:a0c0629dd963: Merge commit '97168b204a0b6b79bb6c5f0d40efdf7fc2262476'
[12:56] <{V}> JEEB, because it tried to av_free_packet an AVPacket that's on the stack?
[12:56] <JEEB> nope
[12:56] Action: {V} might be jumping to conclusions based on the word free
[12:57] <JEEB> the actual crashes were happening later
[12:57] <JEEB> it's just that the msys shell didn't flush
[12:57] <JEEB> + seemed to do something weird with MSVS debugging
[13:03] <cone-904> ffmpeg.git 03Anton Khirnov 07master:372262853b31: libschroedingerenc: use the AVFrame API properly.
[13:03] <cone-904> ffmpeg.git 03Anton Khirnov 07master:7ca97aa7ee46: roqvideo: remove unused variables
[13:03] <cone-904> ffmpeg.git 03Anton Khirnov 07master:863a670e483d: pcxenc: use the AVFrame API properly.
[13:03] <cone-904> ffmpeg.git 03Anton Khirnov 07master:ffe04c330335: libxvid: use the AVFrame API properly.
[13:03] <cone-904> ffmpeg.git 03Michael Niedermayer 07master:f4f7888bab70: Merge commit 'ffe04c330335add4c6d70ab0bb98e6b3f4f7abfa'
[13:58] <cone-904> ffmpeg.git 03Anton Khirnov 07master:14b35bf065a3: huffyuv: use the AVFrame API properly.
[13:58] <cone-904> ffmpeg.git 03Anton Khirnov 07master:706a92926cca: jpegls: use the AVFrame API properly.
[13:58] <cone-904> ffmpeg.git 03Anton Khirnov 07master:a6064b12b481: ffv1: use the AVFrame API properly.
[13:58] <cone-904> ffmpeg.git 03Anton Khirnov 07master:1c01b0253ebf: mpegvideo_enc: use the AVFrame API properly.
[13:58] <cone-904> ffmpeg.git 03Michael Niedermayer 07master:be1e6e7503b2: Merge remote-tracking branch 'qatar/master'
[14:26] <{V}> https://trac.ffmpeg.org/ticket/3143 looks like the pointers s->intra_ac_vlc_length and s->inter_ac_vlc_length don't get set for h261
[15:14] <cone-904> ffmpeg.git 03Michael Niedermayer 07master:331a90cec42f: avcodec/h261: trellis quantization support
[15:14] <cone-904> ffmpeg.git 03Michael Niedermayer 07master:459c7cb25645: fate: add h261 trellis quantization test
[16:09] <cone-904> ffmpeg.git 03Michael Niedermayer 07master:74a8fe9483ac: doc/ffmpeg: document qscale and video stream special case
[16:54] <cone-904> ffmpeg.git 03Michael Niedermayer 07master:bbe66ef91247: avutil: rename lls to lls2
[16:54] <cone-904> ffmpeg.git 03Michael Niedermayer 07master:c3814ab654a9: rename new lls code to lls2 to avoid conflict with the old which has a different ABI
[16:54] <cone-904> ffmpeg.git 03Michael Niedermayer 07master:b382d09d29be: avutil: reintroduce lls1 as the 52 ABI needs it
[17:41] <{V}> michaelni, noticed this while compiling, probably a result from the AVFrame related changes: libavcodec/qpeg.c:328:5: warning: passing argument 1 of 'avcodec_get_frame_defaults' from incompatible pointer type
[17:41] <{V}> libavcodec/avcodec.h:3379:6: note: expected 'struct AVFrame *' but argument is of type 'struct AVFrame **'
[17:43] <nevcairiel> luckily it shouldnt be calling that function at all anymore :)
[17:48] <{V}> ah good
[17:48] <nevcairiel> there is probably a bunch decoders left that need touching to be updated to the new api
[20:06] <cone-904> ffmpeg.git 03Anssi Hannula 07release/2.1:b56e9beeb838: lavf/spdifdec: fix demuxing of AAC in IEC 61937
[20:06] <cone-904> ffmpeg.git 03Michael Niedermayer 07release/2.1:5c1e9d3722db: avcodec/jpeglsdec: check err value for ls_get_code_runterm()
[20:06] <cone-904> ffmpeg.git 03Michael Niedermayer 07release/2.1:807d85400c73: avformat/thp: fix variable types to avoid overflows
[20:06] <cone-904> ffmpeg.git 03Michael Niedermayer 07release/2.1:51d1e79cc18f: avformat/thp: force moving forward
[20:06] <cone-904> ffmpeg.git 03Michael Niedermayer 07release/2.1:e40d01f45dfd: avcodec/cabac: support UNCHECKED_BITSTREAM_READER = 0
[20:07] <cone-904> ffmpeg.git 03Michael Niedermayer 07release/2.1:8c0064798259: avformat/utils: never decrease has_b_frames in compute_pkt_fields()
[20:07] <cone-904> ffmpeg.git 03Michael Niedermayer 07release/2.1:f16751175324: avcodec/ffv1enc: Check high bpp RGB against coder type too
[20:07] <cone-904> ffmpeg.git 03Michael Niedermayer 07release/2.1:1497633924c3: h264: Do not treat the initial frame special in handling of frame gaps
[20:07] <cone-904> ffmpeg.git 03Kostya Shishkov 07release/2.1:bc89c2902bdd: mpegvideo: Fix swapping of UV planes for VCR2
[20:07] <cone-904> ffmpeg.git 03Diego Biurrun 07release/2.1:425517eecbb0: mpeg12dec: Remove incomplete and wrong UV swapping code for VCR2
[20:07] <cone-904> ffmpeg.git 03Michael Niedermayer 07release/2.1:4b846f0ccf80: ffmpeg: Do not fill gap before the first decodable frame on single stream input files unless the user explicitly requests it.
[20:07] <cone-904> ffmpeg.git 03Michael Niedermayer 07release/2.1:d89e14bf54c8: avutil: rename lls to lls2
[20:07] <cone-904> ffmpeg.git 03Michael Niedermayer 07release/2.1:b276b913a175: rename new lls code to lls2 to avoid conflict with the old which has a different ABI
[20:07] <cone-904> ffmpeg.git 03Michael Niedermayer 07release/2.1:e5e048bbf7d3: avutil: reintroduce lls1 as the 52 ABI needs it
[20:26] <ubitux> BBB: do we really need to optimize the vert and horizontal intra pred?
[20:27] <ubitux> http://pastie.org/8487507  some of them don't look worth the optimization
[20:27] <ubitux> unless i'm missing something
[20:56] <ubitux> http://pastie.org/8487548 why doesn't gcc warn for this :(
[21:00] <ubitux> -Wsequence-point was supposed to warn on this
[21:22] <michaelni> {V}, thx, ill fix it in a moment
[21:33] <cone-904> ffmpeg.git 03Michael Niedermayer 07master:9f630e834670: avcodec/mpegaudioenc: prevent overflow in fixed point code
[21:33] <cone-904> ffmpeg.git 03Michael Niedermayer 07master:c017f15dde57: qpeg: remove faulty avcodec_get_frame_defaults()
[22:26] <BBB> ubitux: yes! see h264
[22:26] <BBB> ubitux: simd makes everything faster :)
[22:27] <BBB> if you really don't think it can be done faster then it's fine, but if it was done in h264, we can do it too
[22:27] <ubitux> well
[22:27] <ubitux> how do you make vert_4x4_c faster? oO
[22:28] <BBB> check h264pred
[22:28] <ubitux> 4x4 in h264 does a lot of other things
[22:29] <BBB> hm, h264 doesn't optimize vert_4x4, it seems
[22:29] <BBB> well maybe that can't be optimized
[22:29] <BBB> dunno
[22:29] <BBB> but there's 8x8, so check that
[22:44] <ubitux> BBB: http://pastie.org/8487752
[22:44] <ubitux> opinion on this?
[22:45] <BBB> I think that's fine... weird that we missed that
[22:45] <BBB> good catch
[22:45] <BBB> make sure to measure that it's faster ;) (I'd expect it to be, but always make sure)
[22:46] <ubitux> will do
[22:53] <iive> it might not be faster. the short version modifies mm0 and all instructions after it depend on that modification. so the instruction would be delayed in the pipeline until mm0 is finished.
[22:54] <iive> in the long version it could eventually execute 4 instructions at once.
[22:54] <iive> the short version could be faster in the psubw mm0,mm7 is moved up in the function body.
[22:57] <BBB> iive: that's why you measure it
[23:01] <ubitux> it's faster
[23:02] <ubitux> i'm moving it up to see if it's even faster
[23:04] <ubitux> currently:
[23:04] <ubitux> 411 decicycles in ff_pred4x4_tm_vp8_8_ssse3, 8388289 runs, 319 skips
[23:04] <ubitux> after my patch:
[23:04] <ubitux> 389 decicycles in ff_pred4x4_tm_vp8_8_ssse3, 8388308 runs, 300 skips
[23:05] <ubitux> and if moved upper (just before the lea):
[23:05] <ubitux> 399 decicycles in ff_pred4x4_tm_vp8_8_ssse3, 8388295 runs, 313 skips
[23:05] <ubitux> so i'll keep my version
[23:05] <iive> yep.
[23:12] <cone-904> ffmpeg.git 03Clément BSsch 07master:e0434cfcfcba: avcodec/x86: remove 3 sub in pred4x4_tm_vp8_8.
[00:00] --- Mon Nov 18 2013


More information about the Ffmpeg-devel-irc mailing list