vp9: always keep s->bytesperpixel and ctx->pix_fmt in sync.

Fixes mozilla bug 1229128.

git-howto: mention how to clone ffmpeg-web

Signed-off-by: Andreas Cadhalpun <Andreas.Cadhalpun@googlemail.com>

Makefile: add cleanup of pkg-config files created by configure to distclean

Signed-off-by: Andreas Cadhalpun <Andreas.Cadhalpun@googlemail.com>

lavf/rtpenc_jpeg: Less strict check for standard Huffman tables.

There can be one or more Huffman table segments DHT.

Reported-by: Andrey Utkin

avfilter/af_stereotools: remove floor usage

Signed-off-by: Paul B Mahol <onemda@gmail.com>

avfilter: add audio pulsator filter

Signed-off-by: Paul B Mahol <onemda@gmail.com>

AAC encoder: improve SF range utilization

This patch does 4 things, all of which interact and thus it
woudln't be possible to commit them separately without causing
either quality regressions or assertion failures.

Fate comparison targets don't all reflect improvements in
quality, yet listening tests show substantially improved quality
and stability.

1. Increase SF range utilization.

The spec requires SF delta values to be constrained within the
range -60..60. The previous code was applying that range to
the whole SF array and not only the deltas of consecutive values,
because doing so requires smarter code: zeroing or otherwise
skipping a band may invalidate lots of SF choices.

This patch implements that logic to allow the coders to utilize
the full dynamic range of scalefactors, increasing quality quite
considerably, and fixing delta-SF-related assertion failures,
since now the limitation is enforced rather than asserted.

2. PNS tweaks

The previous modification makes big improvements in twoloop's
efficiency, and every time that happens PNS logic needs to be
tweaked accordingly to avoid it from stepping all over twoloop's
decisions. This patch includes modifications of the sort.

3. Account for lowpass cutoff during PSY analysis

The closer PSY's allocation is to final allocation the better
the quality is, and given these modifications, twoloop is now
very efficient at avoiding holes. Thus, to compute accurate
thresholds, PSY needs to account for the lowpass applied
implicitly during twoloop (by zeroing high bands).

This patch makes twoloop set the cutoff in psymodel's context
the first time it runs, and makes PSY account for it during
threshold computation, making PE and threshold computations
closer to the final allocation and thus achieving better
subjective quality.

4. Tweaks to RC lambda tracking loop in relation to PNS

Without this tweak some corner cases cause quality regressions.
Basically, lambda needs to react faster to overall bitrate
efficiency changes since now PNS can be quite successful in
enforcing maximum bitrates, when PSY allocates too many bits
to the lower bands, suppressing the signals RC logic uses to
lower lambda in those cases and causing aggressive PNS.

This tweak makes PNS much less aggressive, though it can still
use some further tweaks.

Also update MIPS specializations and adjust fuzz

Also in lavc/mips/aacpsy_mips.h: remove trailing whitespace

avformat/matroskadec: Fix sample_aspect_ratio for stereo matroska content

matroskaenc.c applies divisors to the display width/height when generating
stereo content. This patch adds the corresponding multipliers to matroskadec.c
so that the original sample aspect ratio can be recovered.

Signed-off-by: wm4 <nfxjfg@googlemail.com>

avformat/mp3dec: prefer "fast_seek" to TOC seek for CBR files.

"Fast seek" uses linear interpolation to find the position of the
requested seek time. For CBR this is more direct than using the
mp3 TOC and bypassing the TOC avoids problems with TOC precision.
(see https://crbug.com/545914#c13)

For VBR, fast seek is not precise, so continue to prefer the TOC
when available (the lesser of two evils).

Also, some re-ordering of the logic in mp3_seek to simplify and
give usetoc=1 precedence over fastseek flag.

Signed-off-by: wm4 <nfxjfg@googlemail.com>

avcodec/mpegaudio_tablegen: Add (), fix fate

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

avfilter/af_volumedetect: use log10 instead of hardcoded constant

This is likely more precise and conveys the intent better.

Reviewed-by: Mark Harris <mark.hsj@gmail.com>
Signed-off-by: Ganesh Ajjanagadde <gajjanagadde@gmail.com>

doc/filters/drawtext: fix centering example

Signed-off-by: Andrey Utkin <andrey.od.utkin@gmail.com>
Signed-off-by: Lou Logan <lou@lrcd.com>

avcodec/cbrt_tablegen: speed up dynamic table creation

On systems having cbrt, there is no reason to use the slow pow function.

Sample benchmark (x86-64, Haswell, GNU/Linux):
new:
5124920 decicycles in cbrt_tableinit,       1 runs,      0 skips

old:
12321680 decicycles in cbrt_tableinit,       1 runs,      0 skips

Reviewed-by: Ronald S. Bultje <rsbultje@gmail.com>
Signed-off-by: Ganesh Ajjanagadde <gajjanagadde@gmail.com>

avcodec/mpegaudio_tablegen: more dynamic table creation speedups

This further speeds up runtime initialization, with identical generated tables.

Sample benchmark (x86-64, Haswell, GNU/Linux):

old:
34441423 decicycles in mpegaudio_tableinit,    8192 runs,      0 skips

new:
10776291 decicycles in mpegaudio_tableinit,    8192 runs,      0 skips

Most low hanging fruit is taken care of here. For some idea, note that
83,064 array elements totalling 233,722 bytes need to be initialized.
Thus, with this patch, we average ~ 12.9 cycles per element or ~ 4.6
cycles per byte.

Reviewed-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>
Signed-off-by: Ganesh Ajjanagadde <gajjanagadde@gmail.com>

avcodec/mpegaudio_tablegen: speed up dynamic table creation

This does some miscellaneous stuff mainly avoiding the usage of pow to
achieve significant speedups. This is not speed critical, but is
unnecessary latency and cycles wasted for a user.

All tables tested and are identical to the old ones
(bit-exact even in floating point case).

Sample benchmark (x86-64, Haswell, GNU/Linux):
old:
102329530 decicycles in mpegaudio_tableinit,       1 runs,      0 skips

new:
34111900 decicycles in mpegaudio_tableinit,       1 runs,      0 skips

Reviewed-by: Ronald S. Bultje <rsbultje@gmail.com>
Signed-off-by: Ganesh Ajjanagadde <gajjanagadde@gmail.com>

avutil/tablegen: add tablegen libm compatibility shims

This is useful for build-time table generation (--enable-hardcoded-tables),
by providing compat shims for hosts that have broken libms.

This file is deliberately kept minimal; functions can always be added on
an as-needed basis.

Reviewed-by: Clément Bœsch <u@pkh.me>
Reviewed-by: Ronald S. Bultje <rsbultje@gmail.com>
Signed-off-by: Ganesh Ajjanagadde <gajjanagadde@gmail.com>

avcodec/ppc/fdctdsp: use more accurate constants

Whoever wrote this stuff had a pretty bad libm - digits differ pretty
quickly.

Reviewed-by: Ronald S. Bultje <rsbultje@gmail.com>
Signed-off-by: Ganesh Ajjanagadde <gajjanagadde@gmail.com>

avfilter/vf_shuffleframes: Assert that the case of an uninitialized ret does not occur

Fixes CID1258479

Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avfilter/vf_histogram: remove deprecated stuff

Remove all modes except levels mode.
Users should already switch to other filters with
extended funcionality: vectorscope and waveform.

Signed-off-by: Paul B Mahol <onemda@gmail.com>

avutil/mathematics: Do not treat INT64_MIN as positive in av_rescale_rnd

The code expects actual positive numbers and gives completely wrong
results if INT64_MIN is treated as positive
Instead clip it into the valid range that is add 1 and treat it as
negative

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

avutil/integer: Fix av_mod_i() with negative dividend

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

avformat/dump: Fix integer overflow in av_dump_format()

Fixes part of mozilla bug 1229167

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

avcodec/ac3: always use hardcoded tables

The table in question is a 253 byte one. In fact, it turns out that
dynamic generation of the table results in an increased binary size.

Code compiled with GCC 5.2.0, x86-64 (size in bytes), before and after
patch:
old: 62321064 libavcodec/libavcodec.so.57
new: 62320536 libavcodec/libavcodec.so.57

Thus, it always make sense to statically allocate this.

Tested with FATE with/without --enable-hardcoded-tables.

Reviewed-by: wm4 <nfxjfg@googlemail.com>
Signed-off-by: Ganesh Ajjanagadde <gajjanagadde@gmail.com>

avfilter/af_compand: use hypot()

Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Ganesh Ajjanagadde <gajjanagadde@gmail.com>

qsvenc: write a53 caption data to SEI

Signed-off-by: Will Kelleher <wkelleher@gogoair.com>
Previous version reviewed-by: Ivan Uskov <ivan.uskov@nablet.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

ffserver: Clear avio context after closing it in rtp_new_av_stream()

Suggested-by: "Reynaldo H. Verdejo Pinochet" <reynaldo@osg.samsung.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

ffserver: Clear avio context after closing it

Fixes: ==13287== Invalid read of size 4
==13287==    at 0x45161A: flush_buffer (aviobuf.c:143)
==13287==    by 0x451971: avio_flush (aviobuf.c:200)
==13287==    by 0x512CCF: av_write_trailer (mux.c:1016)
==13287==    by 0x41A5E0: close_connection (ffserver.c:853)
==13287==    by 0x421EDC: rtsp_cmd_interrupt (ffserver.c:3245)
==13287==    by 0x420B9C: rtsp_parse_request (ffserver.c:2854)
==13287==    by 0x41A9C2: handle_connection (ffserver.c:930)
==13287==    by 0x41A04B: http_server (ffserver.c:700)
==13287==    by 0x423A60: main (ffserver.c:3897)
==13287==  Address 0xb6cd258 is 88 bytes inside a block of size 192 free'd
==13287==    at 0x4C2B5D9: free (vg_replace_malloc.c:446)
==13287==    by 0x1004DAC: av_free (mem.c:239)
==13287==    by 0x454835: avio_close_dyn_buf (aviobuf.c:1170)
==13287==    by 0x41F385: http_prepare_data (ffserver.c:2368)
==13287==    by 0x41F59B: http_send_data (ffserver.c:2416)
==13287==    by 0x41ABE2: handle_connection (ffserver.c:986)
==13287==    by 0x41A04B: http_server (ffserver.c:700)
==13287==    by 0x423A60: main (ffserver.c:3897)

Reviewed-by: "Reynaldo H. Verdejo Pinochet" <reynaldo@osg.samsung.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avfilter/af_sidechaincompress: do not require writable input frames for acompressor filter

Signed-off-by: Paul B Mahol <onemda@gmail.com>

avfilter/af_sidechaincompress: add level_in and level_sc options

Signed-off-by: Paul B Mahol <onemda@gmail.com>

avcodec/vp3: always set pix_fmt in theora_decode_header()

Fixes assertion failure
Fixes: d0bb0662da342ec65f8f2a081222e6b9/signal_sigabrt_7ffff6ae7cc9_5471_82964f0a9ac2f4d3d59390c15473f6f7.ogg

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/mpeg4videodec: Check available data before reading custom matrix

Fixes: out of array read
Fixes: 76c515fc3779d1b838667c61ea13ce92/asan_heap-oob_1fc0d07_8913_794a4629a264ebdb25b58d3a94ed1785.bit

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/dnxhddec: Disable 12bit by default

The DC VLC table used is too small, fixing this requires a sample,
thus request a sample.
Some samples are said to work even though the table has the wrong size, thus
this is left enabled if the user enables experimental features.

Fixes: 2abd25478c62a675f335fac00b467023/asan_static-oob_10aff98_1227_8811480c6ef1e970a7977ceb7e5a9958.mxf

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Approved-by: kurosu
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

lavf/mxfdec: Set width to actual coded_width for AVCI50.

Fixes ticket #5029.
Reported and analyzed by videolan trac user HenkDemper in vlc ticket #15762.

avfilter/af_sidechaincompress: fix output gain for rms(default) detection

Signed-off-by: Paul B Mahol <onemda@gmail.com>

AAC encoder: make pe.min a local minimum

As noted in a comment, pe.min in the reference encoder
is centered around current pe. The bit reservoir algo
needs pe.min to be a local minimum, because it can only
account for local PE variations. If it's set to a global
minimum as was being done, bit reservoir logic doesn't
work as efficiently.

This patch tries to forget old minimums and converge to
a local minimum without losing the stability of the
previous solution. Listening tests until now suggest this
solves numerous RC issues.

ffserver: add NULL context to ff_rtsp_parse_line().

Needed after f62fe53/2c17fb6.

avcodec/libdcadec: require first public release

Reviewed-by: Hendrik Leppkes <h.leppkes@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>

configure: Add user options to enable gcrypt/gmp for rtmp(t)e support.

Signed-off-by: Matt Oliver <protogonoi@gmail.com>

Merge commit '0e2395293bff089536b97131b32fea8b58bca0ba'

* commit '0e2395293bff089536b97131b32fea8b58bca0ba':
  nut: Mark non-fatal errors as warnings

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

Merge commit '62f72b40c0b0d2cd6a2b81977287fa01d9f4ca6d'

* commit '62f72b40c0b0d2cd6a2b81977287fa01d9f4ca6d':
  nut: Provide more information on failure

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

Merge commit '2c17fb61ced2059034856a6c6cd303014aed01fe'

* commit '2c17fb61ced2059034856a6c6cd303014aed01fe':
  rtsp: Log getaddrinfo failures

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

Merge commit '12b14382861fbf19378afcddaa19cd9a949a86a3'

* commit '12b14382861fbf19378afcddaa19cd9a949a86a3':
  udp: Provide additional information on getaddrinfo failure

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

Merge commit '34af7813f77e2b5b9dcb34f20bcf53bfcc0ba9c9'

* commit '34af7813f77e2b5b9dcb34f20bcf53bfcc0ba9c9':
  udp: Use the logging context

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

Merge commit '98063bcf15eb1e9bf9c8758c83c88d51cbb7ace7'

* commit '98063bcf15eb1e9bf9c8758c83c88d51cbb7ace7':
  rtsp: Do not assume getnameinfo cannot fail

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

Merge commit '5b70fb8fee4af3b13f29a2dc7222fd3c9782f79b'

* commit '5b70fb8fee4af3b13f29a2dc7222fd3c9782f79b':
  movenc-test: Fix integer overflows

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

Merge commit 'fdd5c48ebdec489ec9e84eee547fefa50c3ad53c'

* commit 'fdd5c48ebdec489ec9e84eee547fefa50c3ad53c':
  texturedsp: Explicitly cast RGBA parameters to unsigned

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

Merge commit 'eef38316cab68aff080cf2b83c2427a734194e8d'

* commit 'eef38316cab68aff080cf2b83c2427a734194e8d':
  texturedspenc: Avoid using separate variables

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

Merge commit '7831fb90503142e32cc3c9be43bc3f9d342ded6b'

* commit '7831fb90503142e32cc3c9be43bc3f9d342ded6b':
  textureencdsp: cosmetics: Use normal static const for tables

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

Merge commit '99cb833fc2d9874c62fffbcd3347fae660de0fe5'

* commit '99cb833fc2d9874c62fffbcd3347fae660de0fe5':
  sgi: Correctly propagate meaningful error values

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

Merge commit '823fa7004571cb8404ca5785f9fa6e85f0f9f3d3'

* commit '823fa7004571cb8404ca5785f9fa6e85f0f9f3d3':
  fate: Rework sgi tests into a suite and add the missing ones

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

developer.texi: Call out K&R coding style.

K&R coding style is implied but not listed in 'Coding Rules'.

Signed-off-by: Rick Kern <kernrj@gmail.com>

Merge commit '4a0918cae6394e503b17c71f8f171b4a795eb849'

* commit '4a0918cae6394e503b17c71f8f171b4a795eb849':
  sgienc: Support encoding high bit depth images with RLE

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

avfilter/avf_showvolume: Fix "warning: comparison of unsigned expression < 0 is always false"

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

Merge commit '59e3f4e598ae381600ef54e1f6e6a8a5270ac245'

* commit '59e3f4e598ae381600ef54e1f6e6a8a5270ac245':
  sgienc: Use a local RLE encoding function

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

Merge commit '62cc8f4d79dad119e8efeaae080a58a8dcb1e89d'

* commit '62cc8f4d79dad119e8efeaae080a58a8dcb1e89d':
  sgienc: Port to bytestream2

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

Merge commit '99f40fd02f1870c2dee70977b672da9cb198f0be'

* commit '99f40fd02f1870c2dee70977b672da9cb198f0be':
  sgienc: Do not end RLE lines with 0s

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

Merge commit '4d05e9392f84702e3c833efa86e84c7f1cf5f612'

* commit '4d05e9392f84702e3c833efa86e84c7f1cf5f612':
  libvpxenc: remove some unused ctrl id mappings

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

Merge commit '6b7df14251c49331ae7efd4811ec21f4415ccdb5'

* commit '6b7df14251c49331ae7efd4811ec21f4415ccdb5':
  doc: Document better how to use MSYS2

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

Merge commit '8375dc1dd101d51baa430f34c0bcadfa37873896'

* commit '8375dc1dd101d51baa430f34c0bcadfa37873896':
  asfdec: handle the case when the stream index has an invalid value better

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

Merge commit 'c12c085be7e86880924249e5cb3f898e45dee134'

* commit 'c12c085be7e86880924249e5cb3f898e45dee134':
  dcadec: Do not check for overreads in auxiliary data

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

Merge commit 'c8fa647811371885be421a84a2388529857fed23'

* commit 'c8fa647811371885be421a84a2388529857fed23':
  oggparsespeex: Fix unchecked malloc

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

Merge commit 'afdff8008149515afebf9992eae84be7d76e6b1e'

* commit 'afdff8008149515afebf9992eae84be7d76e6b1e':
  configure: Clearly state that MSYS native builds are discouraged

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

Merge commit '462a54e2291e1fa18e1f1254d09739dfbb795617'

* commit '462a54e2291e1fa18e1f1254d09739dfbb795617':
  lavc: Deprecate avctx.rtp_callback field

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

Merge commit '8a9de5c52401aa1e87bdb147524831c640d1ec53'

* commit '8a9de5c52401aa1e87bdb147524831c640d1ec53':
  doc/encoders: document the QSV option mappings

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

Merge commit 'fc4c27c4edfc6a5f9bc7c696e823652474a65ce8'

* commit 'fc4c27c4edfc6a5f9bc7c696e823652474a65ce8':
  qsvenc: expose additional encoding options

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

Merge commit 'e7d7cf86dcaba8eaaed62c80172ff0aff2588c2a'

* commit 'e7d7cf86dcaba8eaaed62c80172ff0aff2588c2a':
  qsvenc: support more RC methods

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

avfilter/avf_showvolume: fix several issues

Fix color fading: previously color could fade to red when
volume level for red color was actually never reached.
Display volume value on right side.
Use red color only if clipping is needed.

Signed-off-by: Paul B Mahol <onemda@gmail.com>

avformat/mlpdec: consider all valid mlp access units when probing

Fixes probing of truehd/mlp files with a lot of frames in between the
major sync frames. The spec allows a distance of up to 128 frames in
between major sync frames, which leads to the probing code not reaching
the desired score.

avcodec/h264_refs: Check that long references match before use

Fixes out of array read
Fixes: 59bb925e90201fa0f87f0a31945d43b5/asan_heap-oob_4a52e5_3388_66027f11e3d072f1e02401ecc6193361.jvt

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/utils: Clear dimensions in ff_get_buffer() on failure

Fixes out of array access
Fixes: 482d8f2fd17c9f532b586458a33f267c/asan_heap-oob_4a52b6_7417_1d08d477736d66cdadd833d146bb8bae.mov

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/utils: Use 64bit for aspect ratio calculation in avcodec_string()

Fixes integer overflow
Fixes: 3a45b2ae02f2cf12b7bd99543cdcdae5/asan_heap-oob_1dff502_8022_899f75e1e81046ebd7b6c2394a1419f4.mov

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

doc/filters: extend agate documentation

Signed-off-by: Paul B Mahol <onemda@gmail.com>

avfilter: add audio compressor filter

Signed-off-by: Paul B Mahol <onemda@gmail.com>

avfilter: add compensation delay line filter

Signed-off-by: Paul B Mahol <onemda@gmail.com>

avcodec/hevc: Check max ctb addresses for WPP

Fixes out of array read
Fixes: 2f95ddd996db8a6281d2e18c184595a7/asan_heap-oob_192fe91_3330_58e4441181e30a66c19f743dcb392347.bit

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/aac_tablegen: get rid of hardcoded tables entirely

Commit 96786a12f6df26990bbe7c0ca4592b3731724469 makes runtime
initialization cheap.

Tested with FATE, with/without --enable-hardcoded-tables.

Reviewed-by: Rostislav Pehlivanov <atomnuker@gmail.com>
Signed-off-by: Ganesh Ajjanagadde <gajjanagadde@gmail.com>

avcodec/vp3: Fix several memleaks

Fixes: 1536b9b096a8f95b742bae9d3d761cc6/signal_sigsegv_294aaed_4460_b209bd1e7cebe458b53072a44191316d.ogg

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/vp3: Clear context on reinitialization failure

Fixes null pointer dereference
Fixes: 1536b9b096a8f95b742bae9d3d761cc6/signal_sigsegv_294aaed_2039_8d1797aeb823ea43858d0fa45c9eb899.ogv

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/hevc: allocate entries unconditionally

Fixes out of array access
Fixes: 08664a2a7921ef48172f26495c7455be/asan_heap-oob_23036c6_3301_523388ef84285a0270caf67a43247b59.bit

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/pthread_slice: Allow calling ff_alloc_entries() multiple times to readjust the entry count

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

avcodec/hevc_cabac: Fix multiple integer overflows

Fixes: 04ec80eefa77aecd7a49a442cc02baea/asan_heap-oob_19544fa_3303_1905796cd9d8e15f86d664332caabc00.bit

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

aacenc: fix broken build with hardcoded tables

ff_aac_tableinit is a macro in the case of hardcoded tables, so wrap
that up in a function (similar to how the decoder template does it) and
use that as the argument for ff_thread_once().

Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>

avcodec/jpeg2000dwt: Check ndeclevels before calling dwt_encode*()

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

avcodec/jpeg2000dwt: Check ndeclevels before calling dwt_decode*()

Fixes out of array access
Fixes: 01859c9a9ac6cd60a008274123275574/asan_heap-oob_1dff571_8250_50d3d1611e294c3519fd1fa82198b69b.avi

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/aac_tablegen: make exp2_lut static

Suggested-by: Clément Bœsch <u@pkh.me>
Signed-off-by: Ganesh Ajjanagadde <gajjanagadde@gmail.com>

avcodec/hevc: Check entry_point_offsets

Fixes out of array read
Fixes: 007c4a36608ebdf27ee260ad60a81184/asan_heap-oob_32076b4_2243_116b1cb29d91cc4974d6680e3d10bd91.bit

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

aac: temporarily un-share aac_table_init AVOnce variable

AAC-Fixed decoder segfaulted. This commit makes the aac encoder
and decoder init the table twice in case of transcoding again.

Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>

aactab.h: fix comment

Previous commit broke it.

Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>

aactab.h: update and correct comment

Tables in that file have been used by both for a long time now.

Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>

aacenc: make threadsafe

Since the ff_aac_tableinit() can be called by both the encoder and
the decoder (in case of transcoding) this commit shares the AVOnce
variable to prevent this.

Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>

avcodec/cabac: Check initial cabac decoder state

Fixes integer overflows
Fixes: 1430e9c43fae47a24c179c7c54f94918/signal_sigsegv_421427_2340_591e9810c7b09efe501ad84638c9e9f8.264

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Found-by: xiedingbao (Ticket4727)
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/cabac_functions: Fix "left shift of negative value -31767"

Fixes: 1430e9c43fae47a24c179c7c54f94918/signal_sigsegv_421427_2340_591e9810c7b09efe501ad84638c9e9f8.264

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Found-by: xiedingbao (Ticket4727)
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/aac_tablegen: speed up table initialization

This speeds up aac_tablegen to a ludicruous degree (~97%), i.e to the point
where it can be argued that runtime initialization can always be done instead of
hard-coded tables. The only cost is essentially a trivial increase in
the stack size.

Even if one does not care about this, the patch also improves accuracy
as detailed below.

Performance:
Benchmark obtained by looping 10^4 times over ff_aac_tableinit.

Sample benchmark (x86-64, Haswell, GNU/Linux):
old:
1295292 decicycles in ff_aac_tableinit,     512 runs,      0 skips
1275981 decicycles in ff_aac_tableinit,    1024 runs,      0 skips
1272932 decicycles in ff_aac_tableinit,    2048 runs,      0 skips
1262164 decicycles in ff_aac_tableinit,    4096 runs,      0 skips
1256720 decicycles in ff_aac_tableinit,    8192 runs,      0 skips

new:
21112 decicycles in ff_aac_tableinit,     511 runs,      1 skips
21269 decicycles in ff_aac_tableinit,    1023 runs,      1 skips
21352 decicycles in ff_aac_tableinit,    2043 runs,      5 skips
21386 decicycles in ff_aac_tableinit,    4080 runs,     16 skips
21299 decicycles in ff_aac_tableinit,    8173 runs,     19 skips

Accuracy:
The previous code was resulting in needless loss of
accuracy due to the pow being called in succession. As an illustration
of this:
ff_aac_pow34sf_tab[3]
old : 0.000000000007598092294225
new : 0.000000000007598091426864
real: 0.000000000007598091778545

truncated to float
old : 0.000000000007598092294225
new : 0.000000000007598091426864
real: 0.000000000007598091426864

showing that the old value was not correctly rounded. This affects a
large number of elements of the array.

Patch tested with FATE.

Reviewed-by: Rostislav Pehlivanov <atomnuker@gmail.com>
Signed-off-by: Ganesh Ajjanagadde <gajjanagadde@gmail.com>

lavf/utils: avoid decoding a frame to get the codec parameters

Avoid decoding a frame to get the codec parameters while the codec
supports FF_CODEC_CAP_SKIP_FRAME_FILL_PARAM. This is particulary useful
to avoid decoding twice images (once in avformat_find_stream_info and
once when the actual decode is made).

aac_ltp: actually signal LTP as off during EIGHT_SHORT windows

This hugely reduces the echo which was introduced with the previous
commit (though likely because previously everything was broken).
Makes LTP actually worthwhile now.

Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>

aac_ltp: split, reorder and improve prediction algorithm

This commit attempts to mirror what the decoder does more closely
in addition to fixing some shortcomings.

avcodec/faandct: remove L suffixes for floating point literal

Should fix issues with ppc, tested by bug reporter.

Reported-by: John Warburton <john@johnwarburton.net>
Signed-off-by: Ganesh Ajjanagadde <gajjanagadde@gmail.com>

fate: add FF_CODEC_CAP_SKIP_FRAME_FILL_PARAM tests

avfilter/vsrc_mptestsrc: use hypot()

Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Ganesh Ajjanagadde <gajjanagadde@gmail.com>

avfilter/af_dynaudnorm: remove wasteful pow

This removes wasteful pow(x, 2.0) that although not terribly important
for speed, is still useless.

Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Ganesh Ajjanagadde <gajjanagadde@gmail.com>