avformat/matroskaenc: fix targets for attachment tags

Attachment tags were being written targeting non-existent streams in the
output file.
Also filter filename and mimetype entries, as they are standard elements
in the Attachment master.

Signed-off-by: James Almer <jamrial@gmail.com>

lavfi/pan: renormalize negative gain coefficients properly

The parser for the outdef will accept a negative value for the first
named channel's gain. As negative values effectively only invert the
phase of the signal, and not negate the level, the gains' absolute
values must be used to correctly accumulate the levels.

Signed-off-by: Moritz Barsnick <barsnick@gmx.net>
Reviewed-by: Nicolas George <george@nsup.org>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

configure: loongson disable mipsfpu and mipsdsp optimizations

The optimizations of mipsdsp are not supported by all loongson cpu.
The optimizations of mipsfpu and mipsdspr2 maybe supported by 3A2000/3A3000/3A4000 but not tested yet.
Loongson only support mmi (loongSIMD) optimizations now.

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/d3d11va: use the proper slice index

The slice index expected by D3D11VA is the one from the texture not from the
array or texture/slices.

In VLC the slices we provide the decoder don't start from 0 and thus pictures
appear in bogus order. With possible crashes and corruptions when using an
invalid index.

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avutil/hwcontext_vdpau: Fix warning: "GET_CALLBACK" redefined

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avformat/movenc: Check frame rate in mov_write_uuidprof_tag()

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avformat/movenc: Switch mov_write_uuidprof_tag() to avg_frame_rate

Using the stream timebase simply overflows
Fix integer overflow in psp framerate computation

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avformat/utils: Update codec_id before using it in the parser init

Fixes assertion failure

Fixes: input.avi
Found-by: 连一汉 <lianyihan@360.cn>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avfilter/firequalizer: fix blackman window calculation

Signed-off-by: Muhammad Faiz <mfcc64@gmail.com>

cmdutils: fix typos

Signed-off-by: Moritz Barsnick <barsnick@gmx.net>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

lavf: fix typos

Signed-off-by: Moritz Barsnick <barsnick@gmx.net>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

lavfi: fix typos

Signed-off-by: Moritz Barsnick <barsnick@gmx.net>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

lavc: fix typos

Signed-off-by: Moritz Barsnick <barsnick@gmx.net>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

tools: fix grammar error

Signed-off-by: Moritz Barsnick <barsnick@gmx.net>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

doc/examples/http_multiclient: Fix "doc/examples/http_multiclient.c:43:9: warning: passing argument 4 of av_opt_get from incompatible pointer type"

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Merge commit '398f015f077c6a2406deffd9e37ff34b9c7bb3bc'

* commit '398f015f077c6a2406deffd9e37ff34b9c7bb3bc':
  avconv: buffer the packets written while the muxer is not initialized

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>

avformat/matroskaenc: fix Tags master on seekable output if there are tags after the last stream duration

The dynamic AVIOContext would get closed pointing to the wrong position
in the buffer.
This is a regression since 650e17d88b63b5aca6e0a43483e89e64b0f7d2dd.

Reviewed-by: Dave Rice <dave@dericed.com>
Signed-off-by: James Almer <jamrial@gmail.com>

ffmpeg: Fix bitstream typo

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

ffmpeg: explicitly write headers for files with no streams

Recent changes to ffmpeg.c tied output file init to stream init, which broke
stream-less files, specifically ffmetadata output.

avformat/matroskaenc: Fix () error

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Merge commit '1c169782cae6c5c430ff62e7d7272dc9d0e8d527'

* commit '1c169782cae6c5c430ff62e7d7272dc9d0e8d527':
  avconv: explicitly postpone writing the header until all streams are initialized

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>

Merge commit '5b63b15663d31f50ce45d980b904a68795ad3f7a'

* commit '5b63b15663d31f50ce45d980b904a68795ad3f7a':
  lavfi: set the link hwframes context before configuring the dst input

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>

Merge commit '0c9c4004ed57de210b4d83c7b39bbfb00b86b9af'

* commit '0c9c4004ed57de210b4d83c7b39bbfb00b86b9af':
  omx: Don't return > 0 from omx_encode_frame

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>

Merge commit '31aa5335c390c83a6c3ea955b155067c36c4a2c4'

* commit '31aa5335c390c83a6c3ea955b155067c36c4a2c4':
  libopenh264enc: Fix inconsistent whitespace

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>

Merge commit '4f98bb7b6d0323d9ecc3bebd6e24d46a3a374bad'

* commit '4f98bb7b6d0323d9ecc3bebd6e24d46a3a374bad':
  msmpeg4: Remove commented-out debug logging code

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>

Merge commit 'eedbeb4c2737f28844157fae4bd87ed42a61bb1d'

* commit 'eedbeb4c2737f28844157fae4bd87ed42a61bb1d':
  msmpeg4: Remove some broken, commented-out cruft

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>

Merge commit '3c84eaae9da0dc450ae99c65bb6b9865e3ba7fad'

* commit '3c84eaae9da0dc450ae99c65bb6b9865e3ba7fad':
  h264: Eliminate unused but set variable

Noop, the variable doesn't exist in our code.

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>

Merge commit '4dbfcd07570a9e45e9597561023adb6da26f27f6'

* commit '4dbfcd07570a9e45e9597561023adb6da26f27f6':
  librtmp: Avoid an infiniloop setting connection arguments

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>

Merge commit 'fe498ef5144d3712b887f44a0c5e654add99ead7'

* commit 'fe498ef5144d3712b887f44a0c5e654add99ead7':
  hwcontext_vaapi: Return all formats for constraints without config

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>

Merge commit '11b8030309ee93d79b3a6cd4b83bf00757db1598'

* commit '11b8030309ee93d79b3a6cd4b83bf00757db1598':
  vaapi_encode: Fix fallback when input does not match any format

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>

Merge commit '8c3c7b8920033d61c7aa15a4465b759c84e5958f'

* commit '8c3c7b8920033d61c7aa15a4465b759c84e5958f':
  dxva2_h264: Remove an unused variable

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>

Merge commit 'dc7501e524dc3270335749302c7aa449973625f3'

* commit 'dc7501e524dc3270335749302c7aa449973625f3':
  checkasm: Issue emms after benchmarking functions

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>

Merge commit 'beb62dac629603eb074a44c44389c230b5caac7c'

* commit 'beb62dac629603eb074a44c44389c230b5caac7c':
  Use AVFrame.pts instead of deprecated pkt_pts.

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>

Merge commit '32c8359093d1ff4f45ed19518b449b3ac3769d27'

* commit '32c8359093d1ff4f45ed19518b449b3ac3769d27':
  lavc: export the timestamps when decoding in AVFrame.pts

Merged-by: Hendrik Leppkes <h.leppkes@gmail.com>

ffmpeg: remove unused and errorneous AVFrame timestamp check

Decoders have previously not used AVFrame.pts, and with the upcoming
deprecation of pkt_pts (in favor of pts), this would lead to an errorneous
interpration of timestamps.

tests: add -nostdin flag when calling ffmpeg

This fixes a long-standing issue where running FATE in parallel could result
in the terminal being left misconfigured, particularly if a test failed or
was canceled wtih ^C.

ffmpeg: don't reconfigure terminal if we're not taking input from stdin

avformat/matroskaenc: add an option to disable writting CRC32 elements

Also add missing Changelog entry.

Signed-off-by: James Almer <jamrial@gmail.com>

avformat/matroskaenc: write a CRC32 element on Info

Finishes implementing ticket #4347

Tested-by: Dave Rice <dave@dericed.com>
Tested-by: Jerome Martinez <jerome@mediaarea.net>
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: James Almer <jamrial@gmail.com>

avformat/matroskaenc: write a CRC32 element on Tags

Implements part of ticket #4347

Tested-by: Dave Rice <dave@dericed.com>
Tested-by: Jerome Martinez <jerome@mediaarea.net>
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: James Almer <jamrial@gmail.com>

avformat/matroskaenc: write a CRC32 element on Attachments

Implements part of ticket #4347

Tested-by: Dave Rice <dave@dericed.com>
Tested-by: Jerome Martinez <jerome@mediaarea.net>
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: James Almer <jamrial@gmail.com>

avformat/matroskaenc: write a CRC32 element on Chapters

Implements part of ticket #4347

Tested-by: Dave Rice <dave@dericed.com>
Tested-by: Jerome Martinez <jerome@mediaarea.net>
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: James Almer <jamrial@gmail.com>

avformat/matroskaenc: write a CRC32 element on Tracks

Implements part of ticket #4347

Tested-by: Dave Rice <dave@dericed.com>
Tested-by: Jerome Martinez <jerome@mediaarea.net>
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: James Almer <jamrial@gmail.com>

avformat/matroskaenc: write a CRC32 element on Cues

Implements part of ticket #4347

Tested-by: Dave Rice <dave@dericed.com>
Tested-by: Jerome Martinez <jerome@mediaarea.net>
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: James Almer <jamrial@gmail.com>

avformat/matroskaenc: write a CRC32 element on SeekHead

Implements part of ticket #4347

Tested-by: Dave Rice <dave@dericed.com>
Tested-by: Jerome Martinez <jerome@mediaarea.net>
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: James Almer <jamrial@gmail.com>

avformat/matroskaenc: write a CRC32 element on each Cluster

Implements part of ticket #4347

Tested-by: Dave Rice <dave@dericed.com>
Tested-by: Jerome Martinez <jerome@mediaarea.net>
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: James Almer <jamrial@gmail.com>

avformat/matroskaenc: always use a dynamic buffer when writting clusters

Tested-by: Dave Rice <dave@dericed.com>
Tested-by: Jerome Martinez <jerome@mediaarea.net>
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: James Almer <jamrial@gmail.com>

avformat/matroskaenc: print debug message with cluster offsets only if the output is seekable

Printing the dynamic buffer offset is useless.

Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: James Almer <jamrial@gmail.com>

avformat/matroskadec: check for more reserved values on some Colour elements

Signed-off-by: James Almer <jamrial@gmail.com>

avformat/matroskaenc: don't write an empty Colour master element

Signed-off-by: James Almer <jamrial@gmail.com>

configure: Also try -mstack-alignment for clang,

Fixes stack alignment on x86_32 FreeBSD.
Fixes ticket #5231.

lavc/mediacodecdec: fix size variable shadowing in ff_mediacodec_dec_decode

Fixes incompatible pointer type warning on 64-bit systems.

avcodec/qsvenc_h264: fix segfault when a53 SEI is not available

Signed-off-by: Nablet Developer <sdk@nablet.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

lavf/utils: avoid using programs for duration when there's only one

This allows us to be more selective about the streams we derive durations from
(specifically, ignoring text streams with outlier end times) in the common case

lavf/utils: ignore outlier subtitle and data stream end times as well

pixfmt: Use enum assignment for aliases

This way, Doxygen is happier as aliases are now grouped together, and
it never handled #define's in an enum well in the first place.

Y400A already exists as an enum assignment.

pixdesc: Order function prototypes semantically

avutil/mips/generic_macros_msa: rename macro variable which causes segfault for mips r6

Signed-off-by: Shivraj Patil <shivraj.patil@imgtec.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avformat/riffenc: added possibility to set first to ninth audio language for RIFF taged files (e.g. avi files)

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

af_hdcd: disable auto-convert by default

As all known valid HDCD sample formats and sample rates are now handled
by the filter, remove the scan that "invades the privacy" of the filter graph
and turn off autoconvert by default as requested by Nicolas George.

http://ffmpeg.org/pipermail/ffmpeg-devel/2016-August/197571.html

Signed-off-by: Burt P <pburt0@gmail.com>

af_hdcd: add experimental 20 and 24-bit decoding support

I don't have any legitimate 20 or 24-bit HDCD to test. It is known
that the PM Model Two would insert packets into 20 and 24-bit output,
but I have no idea what differences in behavior existed when decoding
20 or 24-bit. For now, as with 16-bit, PE (if enabled) will expand the
top 3dB into 9dB and LLE (gain adjust) will be applied if signaled.

Signed-off-by: Burt P <pburt0@gmail.com>

af_hdcd: hdcd_scan() and hdcd_integrate() handle stereo and single channel

New versions of hdcd_scan() and hdcd_integrate() that also do the
work of hdcd_scan_stereo() and hdcd_integrate_stereo().

Some code split into previously separate functions to remove
duplication is now merged back into each function in the single
place where it is used.

Signed-off-by: Burt P <pburt0@gmail.com>

af_hdcd: support s16p (WavPack) directly

The buffer is already being copied anyway, so interlace the planar
format during the copy and remove one use of auto-convert.

Signed-off-by: Burt P <pburt0@gmail.com>

af_hdcd: allow all HDCD sample rates

The PM Model Two could output HDCD-encoded audio in CD and all
DVD-Audio sample rates. (44100, 48000, 88200, 96000, 176400, and
192000 Hz)

Signed-off-by: Burt P <pburt0@gmail.com>

af_hdcd: add mono as a supported channel layout

Signed-off-by: Burt P <pburt0@gmail.com>

ChangeLog: Add latest CUVID changes

lavc/tiff: Print compression debug information.

lavc: set best effort timestamp if unset when using new decode API

Some API users (in particular ffmpeg.c) check the best effort timestamp
only.

doc/developer: remove duplicate policies and fix error

Fixes regression as of ee72b6d1

Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Josh de Kock <josh@itanimul.li>
Signed-off-by: James Almer <jamrial@gmail.com>

avformat/matroskaenc: don't reserve space for stream duration tags if the output is not seekable

The durations are never written in that situation.

Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: James Almer <jamrial@gmail.com>

avformat/matroska: write FlagInterlaced element in WebM

It's listed as supported in both https://www.webmproject.org/docs/container/
and https://matroska.org/technical/specs/index.html

Reviewed-by: Dave Rice <dave@dericed.com>
Signed-off-by: James Almer <jamrial@gmail.com>

doc/developer: add sections for policies

And sort policies into these sections.

Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Josh de Kock <josh@itanimul.li>

doc/developer: reword some of the policies

Explicitly state that FATE should pass, and code should work
for all reviewers who tested.

Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Josh de Kock <josh@itanimul.li>

lavc/ivi_dsp.c: fix warnings due to indentation

Signed-off-by: Adriano Pallavicino <adriano.pallavicino@gmail.com>
Signed-off-by: Josh de Kock <josh@itanimul.li>

avcodec/cuvid: don't align frame size

hwcontext_cuda was changed to take care of proper alignment internally

lavf/mpegtsenc: fix autobsf when the first NAL is 0x1<XX> bytes

configure: add linker export script support on Darwin

This isn't a "version script" in the usual sense, since it doesn't set symbol
versions directly. Instead, the version for the whole .dylib is set in the
linker flags, and we generate a list of symbol patterns to export. This allows
us to keep our local symbols (e.g. ff_*) local on the platform.

The Darwin linker's exported_symbols_list format is a bit different than the
one used by the GNU linker. It doesn't handle local symbols at all, since when
a list is provided, all unlisted symbols are local by default; thus, we remove
local sections. It doesn't handle per-version sections, so we remove the
headers and brackets. It expects symbols to be prefixed with an underscore.
It errors if a listed symbol with no wildcards is not present in the output,
so we append an asterisk to any symbol that doesn't already end in one.

avformat/hlsenc: support multi level path in m3u8 with filename

before patch:

localhost:osx liuqi$ tree 20160926/
20160926/
└── file-20160909
    └── 26
        └── 16
            └── 15
                ├── 46
                │   ├── 13.ts
                │   ├── 25.ts
                │   ├── 36.ts
                │   ├── 45.ts
                │   └── 54.ts
                └── 47
                    ├── 04.ts
                    ├── 14.ts
                    ├── 24.ts
                    └── 35.ts

6 directories, 9 files
localhost:osx liuqi$ cat out.m3u8
 #EXTM3U
 #EXT-X-VERSION:3
 #EXT-X-TARGETDURATION:12
 #EXT-X-MEDIA-SEQUENCE:0
 #EXTINF:11.360000,
 13.ts
 #EXTINF:10.800000,
 25.ts
 #EXTINF:9.120000,
 36.ts
 #EXTINF:8.760000,
 45.ts
 #EXTINF:10.200000,
 54.ts
 #EXTINF:10.720000,
 04.ts
 #EXTINF:9.600000,
 14.ts
 #EXTINF:10.600000,
 24.ts
 #EXTINF:8.760000,
 35.ts
 #EXT-X-ENDLIST

after patch:

localhost:osx liuqi$ tree 20160926/
20160926/
└── file-20160909
    └── 26
        └── 16
            └── 15
                ├── 46
                │   ├── 13.ts
                │   ├── 25.ts
                │   ├── 36.ts
                │   ├── 45.ts
                │   └── 54.ts
                └── 47
                    ├── 04.ts
                    ├── 14.ts
                    ├── 24.ts
                    └── 35.ts

6 directories, 9 files
localhost:osx liuqi$ cat out.m3u8
 #EXTM3U
 #EXT-X-VERSION:3
 #EXT-X-TARGETDURATION:12
 #EXT-X-MEDIA-SEQUENCE:0
 #EXTINF:11.360000,
 20160926/file-20160909/26/16/15/46/13.ts
 #EXTINF:10.800000,
 20160926/file-20160909/26/16/15/46/25.ts
 #EXTINF:9.120000,
 20160926/file-20160909/26/16/15/46/36.ts
 #EXTINF:8.760000,
 20160926/file-20160909/26/16/15/46/45.ts
 #EXTINF:10.200000,
 20160926/file-20160909/26/16/15/46/54.ts
 #EXTINF:10.720000,
 20160926/file-20160909/26/16/15/47/04.ts
 #EXTINF:9.600000,
 20160926/file-20160909/26/16/15/47/14.ts
 #EXTINF:10.600000,
 20160926/file-20160909/26/16/15/47/24.ts
 #EXTINF:8.760000,
 20160926/file-20160909/26/16/15/47/35.ts
 #EXT-X-ENDLIST

Signed-off-by: Steven Liu <lingjiujianke@gmail.com>

lavf/aviobuf.c: Adapt avio_accept and avio_handshake to new AVIOContext API

Signed-off-by: Stephan Holljes <klaxa1337@googlemail.com>

lavd/openal: don't return zero sized packet if no samples are available

Signed-off-by: Marton Balint <cus@passwd.hu>

lavc/utils: disallow zero sized packets with data set in avcodec_send_packet

Reviewed-by: wm4 <nfxjfg@googlemail.com>
Reviewed-by: Hendrik Leppkes <h.leppkes@gmail.com>
Signed-off-by: Marton Balint <cus@passwd.hu>

lavf/bink.c: fix warning due to misleading indentation

Signed-off-by: Adriano Pallavicino <adriano.pallavicino@gmail.com>
Signed-off-by: Josh de Kock <josh@itanimul.li>

MAINTAINERS: add myself for hwcontext_cuda

avutil/hwcontext_cuda: align allocated frames

lavc/mediacodecdec_h2645: fix nalu data_size type

fate: add bitexact decode flag to fate-svq3-watermark

Fixes failure with mmxext or 3dnow cpuflags enabled but ssse3 disabled

Signed-off-by: James Almer <jamrial@gmail.com>

fate: Add PSP copy test

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

doc/t2h: use container

Full width text is really difficult to read, this makes it more
more legible on larger (widescreen) screens. It also means we aren't
inventing our own container instead of using the bootstrap one.

Signed-off-by: Josh de Kock <josh@itanimul.li>

configure: define posix source on cygwin

Some function definitions are missing without it,
for example sigaction.

Fixes ticket #5868

ffplay: remove unused viddec_width/viddec_height

Reviewed-by: Josh de Kock <josh@itanimul.li>
Signed-off-by: Marton Balint <cus@passwd.hu>

ffplay: use decoder avctx for decoded subtitle width/height

Fixes ticket #5873.

Reviewed-by: Josh de Kock <josh@itanimul.li>
Signed-off-by: Marton Balint <cus@passwd.hu>

avcodec: fix arguments on xmm/neon clobber test wrappers

Signed-off-by: James Almer <jamrial@gmail.com>

Changelog: Compress slightly to improve readability.

lavc: remove libfaac wrapper

There is really no need for two aac wrappers, we already have
libfdk-aac which is better. Not to mention that faac doesn't
even support HEv1, or HEv2. It's also under a license which is
unusable for distribution, so it would only be useful to people
who will compile their own ffmpeg, only use it themselves (which
at that point should just use fdk-aac).

Signed-off-by: Josh de Kock <josh@itanimul.li>

avcodec: add missing xmm/neon clobber test wrappers for the new encode API

Reviewed-by: Ronald S. Bultje <rsbultje@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>

lavfi/metadata: allow deleting all metadata

Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Marton Balint <cus@passwd.hu>

lavfi/metadata: fix setting metadata values

Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Marton Balint <cus@passwd.hu>

ffmpeg: use new encode API

ffmpeg: use new decode API

This is a bit messy, mainly due to timestamp handling.

decode_video() relied on the fact that it could set dts on a flush/drain
packet. This is not possible with the old API, and won't be. (I think
doing this was very questionable with the old API. Flush packets should
not contain any information; they just cause a FIFO to be emptied.) This
is replaced with checking the best_effort_timestamp for AV_NOPTS_VALUE,
and using the suggested DTS in the drain case.

The modified tests (fate-cavs and others) still fails due to dropping
the last frame. This happens because the timestamp of the last frame
goes backwards (ffprobe -show_frames shows the same thing). I suspect
that this "worked" due to the best effort timestamp logic picking the
DTS over the decreasing PTS. Since this logic is in libavcodec (where
it probably shouldn't be), this can't be easily fixed. The timestamps
of the cavs samples are weird anyway, so I chose not to fix it.

Another strange thing is the timestamp handling in the video path of
process_input_packet (after the decode_video() call). It looks like
the code to increase next_dts and next_pts should be run every time
a frame is decoded - but it's needed even if output is skipped.

ffmpeg: move subframe warning to libavcodec

With the new decode API, doing this in ffmpeg.c is impractical. There
was resistance against removing the warning, so put it into libavcodec.

Not bothering with reducing the warning to verbose log level for
subsequent wanrings. The warning should be rare, and only happen when
developing new codecs for the old API.

Includes a change suggested by Michael Niedermayer.