avutil/libm: use FFMIN instead of fmin()

MSVC apparently doesnt support fmin() either

Suggested/Found-by: ubitux, Daemon404, nevcairiel
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avformat/mov: fix "warning: variable altitude set but not used" warning

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avutil/libm: fix fminf typo

Found-by: ubitux
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

build: add fminf() emulation

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Merge remote-tracking branch 'cehoyos/master'

* cehoyos/master:
  Allow >2G for skip_initial_bytes.

Merged-by: Michael Niedermayer <michaelni@gmx.at>

wma lossless: reuse scalarproduct_and_madd_int16

This is done by padding the coefficient buffer with 0s, because the order
may be only a multiple of 4, and the DSP function requires batches of 8.

However, no sample with such a case was found, so request one if it uses
that kind of order.

Approximate relative speedup depending on instruction set:
plain C: -6%
mmxext:  51%
sse2:    54%

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '2f7065190ad48744014a02288799d03adfa613e0'

* commit '2f7065190ad48744014a02288799d03adfa613e0':
  libfdk-aac: Relicense the library wrappers to the ISC license

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'e7d6d0bf3c5cc1bc048b0ddbc169a91862568e0c'

* commit 'e7d6d0bf3c5cc1bc048b0ddbc169a91862568e0c':
  mov: Export geotag metadata fields

Conflicts:
libavformat/mov.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>

libfdk-aac: Relicense the library wrappers to the ISC license

This reduces the number of different licenses used within libav,
and is preferrable since it has less ambiguous wordings than
the BSD license with respect to the duties of the user of the code.

Fraunhofer have now indicated that they're allowed to contribute
code under this license as well.

Signed-off-by: Martin Storsjö <martin@martin.st>

mov: Export geotag metadata fields

The '?xyz' form is used by android devices (and according to apple
mailing list archives, also by older iOS devices). The 'loci' field
(defined in 3GPP 26.244) is used by recent iOS devices.

Even though the loci field can contain an altitude, it was plain
0 in my sample. Just export longitude and latitude, in a string
format matching the one used by the '?xyz' metadata field.

Signed-off-by: Martin Storsjö <martin@martin.st>

Allow >2G for skip_initial_bytes.

x86/dsputil: move some mmx init code inside dsputil_init_mmx()

This reduces differences with the fork

Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/mjpegdec: Improve intel jpeg flip heuristic

Fixes Ticket3698

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/aic: fix quantization table permutation

Fixes Ticket3700

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/mpegaudiodec: Reset dither and mdct state

This makes the mp3 decoder produce the same result when repeatly flushing and decoding

Suggested-by: Dale Curtis <dalecurtis@chromium.org>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

apedsp: move to llauddsp

APE is not the sole codec using scalarproduct_and_madd_int16.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'd816e125fe1fa5c909b706d47904a4d6c1799996'

* commit 'd816e125fe1fa5c909b706d47904a4d6c1799996':
  dict: const correctness for av_dict_get() and av_dict_copy()

Conflicts:
libavutil/dict.c
libavutil/dict.h

See: e12a73246d8ce7d0fc4036522688934e26de4bb1
Merged-by: Michael Niedermayer <michaelni@gmx.at>

ffmpeg: prevent pts < dts to be passed through to the muxer on stream copy

Fixes Ticket3658

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

dict: const correctness for av_dict_get() and av_dict_copy()

avfilter: new multimedia filter avf_showcqt.c

this filter is the same as showspectrum but with constant Q transform,
so frequency is spaced logarithmically

tests: add adpcm trellis tests

adpcm_ima_qt does not produce reproducible results, so it is temporarily
disabled (see #3701).

Signed-off-by: Timothy Gu <timothygu99@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

tests: Add ac3_fixed decoder test

Signed-off-by: Timothy Gu <timothygu99@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

tests/swr: move the results out of the functions

This prevents all results from being declared whenever the function is called.

Signed-off-by: Timothy Gu <timothygu99@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avutil/ppc/intreadwrite: try to fix intreadwrite for ppc64le

Untested due to lack of hardware

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

huffyuvdec: remove superfluous check in BGR code

Before: 154861 decicycles
After:  152912 decicycles

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

huffyuvdec: remove superfluous check

An invalid entry already has the property of having a negative number
of bits, so remove the check on the reserved value, and rearrange the
code as a consequence.

346800 decicycles in 422, 262079 runs, 65 skips
168197 decicycles in gray, 262077 runs, 67 skips
Overall time: 7.878s

319076 decicycles in 422, 262096 runs, 48 skips
159875 decicycles in gray, 262057 runs, 87 skips
Overall time: 7.394s

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

huffyuvdec: trick for plane decoding, 8bits

10496 -> 9105 cycles.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

huffyuvdec: clean up macro

It's no longer used inside another specific macro, so rename it.
Also remove duplicated definition and realign code.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/dcadec: Check dca_dmixtable index

Found-by: Niels Möller <nisse@lysator.liu.se>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/dcadec: remove fishy FFMAX()

These where intended to maintain the previous behavior before dca_dmix_code()
but it is unclear (to me) which way is correct and no sample seem to trigger
the case, also they are incomplete for the purprose of error checking

Found-by: Niels Möller <nisse@lysator.liu.se>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Merge remote-tracking branch 'cehoyos/master'

* cehoyos/master:
  Fix rc_max_rate documentation.
  Fix some fate filter tests with --target-path.

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Fix rc_max_rate documentation.

The mpegvideo decoder sets rc_max_rate since 25b7aa98.

Fix some fate filter tests with --target-path.

Merge commit '39ec5e1cf8444f827c42effb76e5694e091bbff3'

* commit '39ec5e1cf8444f827c42effb76e5694e091bbff3':
  avconv: Report the codec and the encoder separately

Merged-by: Michael Niedermayer <michaelni@gmx.at>

avfilter/bufferqueue: Increase buffer queue size

Avoid buffer overruns when processing some MOV files with the amerge
filter. Files produced by Adobe Premiere Pro CC have up to one second of audio
not interleaved. With common settings (<= 48kHz) that makes up to 47 frames so
a queue length of 64 makes sense.

Fixes ticket #3510.

Reviewed-by: Nicolas George <george@nsup.org>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/dvbsubdec: add some basic av_log debuging support

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

huffyuvdec: implement trick for BGR

Before:
179436 decicycles in rgb, 32735 runs, 33 skips
Stripped object size: 23188

After:
154861 decicycles in rgb, 32738 runs, 30 skips
Stripped object size: 22736

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

huffyuvdec: implement trick

When the joint table does not contain a valid entry, the decoding restarts
from scratch. By implementing the trick of jumping to the 2nd level of the
individual table (and inlining the whole), a speed improvement of 5-10%
is possible.

On a 1000-frames YUV4:2:0 video, before:
362851 decicycles in 422, 262094 runs, 50 skips
182488 decicycles in gray, 262087 runs, 57 skips
Object size: 23584
Overall time: 8.377

After:
346800 decicycles in 422, 262079 runs, 65 skips
168197 decicycles in gray, 262077 runs, 67 skips
Object size: 23188
Overall time: 7.878

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avconv: Report the codec and the encoder separately

Bug-Id: 694

Merge commit '95b7fa1729b93bbb3f4fb85a5c0cb53cf970c3c7'

* commit '95b7fa1729b93bbb3f4fb85a5c0cb53cf970c3c7':
  oggenc: Support flushing the muxer

Conflicts:
libavformat/version.h

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'd5a55981986ac5d1a31aef3a8d16eaff8534a412'

* commit 'd5a55981986ac5d1a31aef3a8d16eaff8534a412':
  build: check if AS supports the '.func' directive

Merged-by: Michael Niedermayer <michaelni@gmx.at>

oggenc: Support flushing the muxer

This allows the caller to write all buffered data to disk, allowing
the caller to know at what byte position in the file a certain
packet starts (any packet written after the flush will be located
after that byte position).

Signed-off-by: Martin Storsjö <martin@martin.st>

avcodec/x86/dsputilenc_mmx: fix build without yasm

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avformat/avidec: allow rounding errors between scale/rate and timebase

Fixes Ticket3670

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/h264: in the absence of recovery points, be more tolerant on accepting plain I frames

Fixes: Ticket3652
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/lagarithrac: lag_get_rac: use normal division

maybe 0.3% faster and simpler code

avcodec/lagarithrac: increase LUT from 256 to 1024 bytes

about 2 cycles (~1%) faster lag_get_rac()

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/lagarithrac.h/lag_get_rac: drop apparently unneeded operations

x86/motion_est: sad_{x, y}2_mmxext functions are bitexact

Only the xy2 functions aren't.

Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

motion-test: force C functions for the reference C context

It was instead using the highest available asm functions, which completely
kills the point of being a reference C context.

Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Merge remote-tracking branch 'cehoyos/master'

* cehoyos/master:
  Fix fate-aac-ln-encode with --target-path (second try).

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Fix fate-aac-ln-encode with --target-path (second try).

x86: dsputilenc: convert hf_noise*_mmx to yasm

Signed-off-by: Timothy Gu <timothygu99@gmail.com>
Several bugfixes by: Christophe Gisquet <christophe.gisquet@gmail.com>
See: [FFmpeg-devel] [WIP] [PATCH 4/4] x86: dsputilenc: convert hf_noise*_mmx to yasm
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/libopenjpegenc: Add frame-threading capability

This gives a >50% speed up when encoding with libopenjpeg.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

fix various typos

Signed-off-by: Lou Logan <lou@lrcd.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

ffmpeg: try to use the more fitting of the 2 encoder contexts

This fixes the interactively enabled encoder debug by pressing 'D'

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

doc/muxers: Document problems of omitting nut syncpoints as well as alternatives

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '896a5bff64264f4d01ed98eacc97a67260c1e17e'

* commit '896a5bff64264f4d01ed98eacc97a67260c1e17e':
  arm: check if AS supports .dn

Conflicts:
configure
libavcodec/arm/vc1dsp_init_neon.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '68a06b3a639ee21c78532ee4c50c3366bf890ff7'

* commit '68a06b3a639ee21c78532ee4c50c3366bf890ff7':
  aarch64: use '#' for whole line asm comments

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '6a0fa4d86f2b3e9060a1153b39fa3bfe36f1b617'

* commit '6a0fa4d86f2b3e9060a1153b39fa3bfe36f1b617':
  aarch64: remove optional :pg_hi21: for adrp instruction

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'fd2981ea92d9a776fcb1a13377dce1c8a7db7b5e'

* commit 'fd2981ea92d9a776fcb1a13377dce1c8a7db7b5e':
  aarch64: add darwin style PAGE/PAGEOFF relocations

Merged-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/vp3: free tables before allocating new ones

Fixes memleak on seeking

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avcodec/vp3: factorize out free_tables()

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

ffmpeg: use av_log_ask_for_sample() instead of duplicating avpriv_request_sample()

Loose mplayer in MPlayer/incoming/

Found-by: Clément Bœsch <u@pkh.me>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Add direct link to ML in missing_feature_sample()

Reviewed-by: Clément Bœsch <u@pkh.me>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

ffmpeg: request sample if the decoder & demuxer has_b_frames mismatch

We cannot use avpriv_request_sample() as this is private to the libs
or rather it would be a bad usage example

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

ffmpeg: remove "has_b_frames" backpassing

It seems working without this now for the files i tested it with, if this causes
a regression, dont hesitate to put the line back or open a ticket or fix (if possible)
the parser

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avformat/asfdec: Use a parser for mpeg4-asp to find has_b_frames

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

build: check if AS supports the '.func' directive

Not supported by Clang's integrated assembler. Since it just adds
debug information it can safely omitted.

arm: check if AS supports .dn

Move the GNU as check before the arch specific asm checks since the .dn
check requires gas compatible assembler.

Disable the VC-1 motion compensation NEON asm which is the only part
using that directive. The integrated assembler in the upcoming clang 3.5
does not support .dn/.qn without plans to change that. Too much effort
to implement it while it is rarely used.

http://llvm.org/bugs/show_bug.cgi?id=18199.

aarch64: use '#' for whole line asm comments

Both gnu as and clang treat lines starting with '#' as comments if they
aren't consumed by the C-style preprocessor.
Using '//' does not work with clang since comments are removed before
macro expansion.

aarch64: remove optional :pg_hi21: for adrp instruction

Clang's integrated assembler does not support it.

aarch64: add darwin style PAGE/PAGEOFF relocations

Fix fate-opus and fate-aac-ln-encode with --target-path.

avcodec/utils/ff_init_buffer_info: Favor color information from AVFrame if available

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Do not limit blocksize when reading PCM from aiff.

Fixes ticket #3695.

doc/bsf: document noise bitstream filter

Partially addresses ticket #3198.

Signed-off-by: Lou Logan <lou@lrcd.com>
Signed-off-by: Timothy Gu <timothygu99@gmail.com>

avformat/utils/avformat_find_stream_info: Move max_analyze_duration setup out of the packet loop

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

bfin: add README

Based on sh4/README
Suggested-by: jamrial
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '880e2aa23645ed9871c66ee1cbd00f93c72d2d73'

* commit '880e2aa23645ed9871c66ee1cbd00f93c72d2d73':
  Remove all Blackfin architecture optimizations

Conflicts:
libavcodec/bfin/dsputil.S
libavcodec/bfin/dsputil_init.c
libavcodec/bfin/fdct_bfin.S
libavcodec/bfin/hpel_pixels_no_rnd.S
libavcodec/bfin/hpeldsp_init.c
libavcodec/bfin/idct_bfin.S
libavcodec/bfin/mathops.h
libavcodec/bfin/pixels.S
libavcodec/bfin/pixels.h
libavcodec/bfin/vp3dsp.S
libavcodec/bfin/vp3dsp_init.c
libavutil/bfin/asm.h
libavutil/bfin/attributes.h
libswscale/bfin/internal_bfin.S
libswscale/bfin/swscale_bfin.c
libswscale/bfin/yuv2rgb_bfin.c
libswscale/swscale_internal.h
libswscale/version.h

If someone wants to maintain blackfin support in FFmpeg, please contact
ffmpeg-devel@ffmpeg.org

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit 'b88cc5cca111132b42c2ee99662bfefe7652e3da'

* commit 'b88cc5cca111132b42c2ee99662bfefe7652e3da':
  bink: Rename BinkDSPContext member so as not to clash with BlockDSPContext

Conflicts:
libavcodec/bink.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>

avutil/mathematics/av_add_stable: Avoid av_cmp_q() call

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

av_add_stable: Add fast special case where step can be represented exactly

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avutil/mathematics/av_add_stable: check for the common case of inc=1

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avutil/mathematics/av_add_stable: avoid unneeded variable

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Remove all Blackfin architecture optimizations

Blackfin is a painful platform to work with, no test machines are available
and the range of multimedia applications is dubious. Thus it only represents
a maintenance burden.

avformat/utils: Call ff_rfps_add_frame() only for video

This avoids some unneeded computations

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

swr: move compensation_distance handling to swri_resample caller.

I think there's an off-by-one in terms of the switchpoint where we
switch from dst_incr to ideal_dst_incr, I don't think that's a massive
issue, but just be aware of that. It's probably trivial to prevent but
I don't care.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
I could not reproduce any off by 1 error, results are bit exact (michael)

avformat/mux: Dont leave stale side data pointers in ff_interleave_add_packet()

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

avformat/mux: fix memleak of side data in flac fate tests

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

fate: fate-hevc-conformance-INITQP_B_Sony_1 belongs to the 10bit kind

Fixes the test on big endian

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

bink: Rename BinkDSPContext member so as not to clash with BlockDSPContext

Report "unknown codec" instead of "unspecifed size" if the codec is unknown.

tests: Replace fieldmatch test with pixfmt

Signed-off-by: Timothy Gu <timothygu99@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

fate: add simple test for async

This only checks that things havnt changed, the values provide little
help in determining if a change is good or bad.

Improvements welcome!

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

swr/resample_template: prevent end_index from overflowing and add check for delta_frac overflow

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Rewrite main resampling loop (common and linear).

This removes a branch at a performance-sensitive point (in the middle
of the loop). In fate-swr-resample-s32p-8000-2626, this makes the code
about 10% faster. It also simplifies the loops, allowing us to rewrite
it in yasm at some later point.

The compensation_distance != 0 code and index < 0 code are still kind
of hairy. For compensation_distance != 0, this should likely be handled
in the caller, so that it calls swri_resample twice (once until the
dst_incr switch-point, and once with the remainder of the samples). For
index < 0, the code should probably be rewritten to break out of the
loop once sample_index >= 0, and then resume (e.g. as a tail-call) to
the common or linear resampling loops.

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '59cb5747ec3c5cd842b94e574c37889521c97cc4'

* commit '59cb5747ec3c5cd842b94e574c37889521c97cc4':
  rtmpproto: read metadata to set correct FLV header

Merged-by: Michael Niedermayer <michaelni@gmx.at>

Merge commit '3b18857ab301d2a0b3e86e9d85eed76f0798a29c'

* commit '3b18857ab301d2a0b3e86e9d85eed76f0798a29c':
  rtmppkt: Add method to read an AMF string that is not prefixed by its type

Merged-by: Michael Niedermayer <michaelni@gmx.at>