avcodec/videotoolboxenc: fix encoding frame crash on iOS 11

On iOS 11, encoding a frame may return error with log
"Error encoding frame 0", which means vtenc_output_callback
is called with status=0 and sample_buffer=NULL. Then the
encoding session will be crashed on next callback wether or not
closing the codec context.

Let us look through the link below introducing VTCompressionOutputCallback,

https://developer.apple.com/documentation/videotoolbox/vtcompressionoutputcallback?language=objc

"status=0" (noErr) means compression was successful.
"sampleBuffer=NULL" means the frame was dropped when compression
was successful (status=0) or compression was not successful (status!=0).

So we should not set AVERROR_EXTERNAL on "status=0" and "sample_buffer=NULL"
as it is not a error.

The fix is that we only set AVERROR_EXTERNAL with status value non zero.
When sample_buffer is NULL and status value is zero, we simply return
with no other operation.

This crash often occurs on iOS 11 for example encoding 720p@25fps.

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

avfilter/vf_stereo3d: assert that out variable is valid

avfilter/vf_stereo3d: merge same code in case branches

avfilter/vf_shuffleplanes: remove not needed line

avfilter/vf_ciescope: remove dead assignments

avfilter/window_func: clarify intention in dolph window calculation

avfilter/vf_v360: add missing av_assert0()

avfilter/vf_avgblur: remove dupe assignment

avfilter/af_headphone: return on error immediately

avfilter/x86/vf_360: add most of >8 depth asm

avformat/vividas: Remove align offset which is always masked off

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

avformat/vividas: remove dead assignment

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

avformat/cdxl: Fix integer overflow in intermediate

Fixes: signed integer overflow: 65535 * 65312 cannot be represented in type 'int'
Fixes: 16704/clusterfuzz-testcase-minimized-ffmpeg_DEMUXER_fuzzer-6294115603447808

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/cfhd: Check that cropped size is smaller than full

Fixes: signed integer overflow: 57342 * 120830 cannot be represented in type 'int'
Fixes: 16426/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_CFHD_fuzzer-5758744817827840

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg

tools/target_dec_fuzzer: increase snows threshold

snow allows creating large output from tiny input, like other
wavelet codecs

Fixes: Timeout (82sec -> 1.5sec)
Fixes: 9520/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_SNOW_fuzzer-6286612576075776

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/hevcdec: repeat character in skiped

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

repeat an even number of characters in occured

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

avformat/hcom: Tell the compiler about set but not read variables

This avoids 3 warnings

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

doc/examples/decode_audio: Fix "warning: ISO C90 forbids mixed declarations and code"

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

avcodec/gdv: Replace assert() checking bitstream by if()

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

libavcodec/utils: Free threads on init failure

Fixes: Multiple memleaks
Fixes: ffmpeg-memory-leak

Found-by: Francis Provencher <francis@protekresearchlab.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

tools/target_dec_fuzzer: Adjust motionpixels threshold

Fixes: Timeout (151sec -> 173ms)
Fixes: 16053/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_MOTIONPIXELS_fuzzer-5647069169057792
Fixes: 16053/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_MOTIONPIXELS_fuzzer-5108957126852608

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avutil/mips: refactor msa SLDI_Bn_0 and SLDI_Bn macros.

Changing details as following:
1. The previous order of parameters are irregular and difficult to
   understand. Adjust the order of the parameters according to the
   rule: (RTYPE, input registers, input mask/input index/..., output registers).
   Most of the existing msa macros follow the rule.
2. Remove the redundant macro SLDI_Bn_0 and use SLDI_Bn instead.

Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

lavc/videotoolboxdec: fix crop handling when multithreaded

This was partially fixed by 233cd89, but it made changes to AVFrame fields
from within end_frame, which doesn't work consistently when multithreading
is enabled. This is what the post_process function is for.

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2: fix compile with older videodev2.h

Some of these symbols are only defined in newer kernel releases.

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/libdav1d: use the library default for the filmgrain option

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

avformat/rmdec.c: fix brackets around arguments

Regression since 78f52b4fe3

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

avformat/rmdec.c: fix left shift of negative value in rm_sync()

Fixes ticket 8143.

Reviewed-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>

avfilter/vf_stereo3d: add tb (top-bottom) aliases

avfilter/vf_v360: refactor dimensions and offsets calculations

doc/filters: add another v360 example

avfilter/vf_v360: implement stereo 3D support

avfilter/vf_v360: reduce by one cosf call less

avfilter/vf_v360: define variables in for loops consistently

avfilter/vf_v360: call tanf() once

avfilter/vf_v360: implement diagonal field of view

avformat/hlsenc: make fix code style of hls_write_trailer

change vs->avf to oc

avformat/dashdec: reindent code at parse_manifest

lavf/hls: fix memory leak in error handling path for option tmp

fix memory leak in error handling path for option tmp.

Reviewed-by: Steven Liu <lq@onvideo.cn>
Signed-off-by: Jun Zhao <barryjzhao@tencent.com>

lavf/hls: remove redundancy seekable setting

ffio_init_context with NULL seek callback will setting seekable with 0, so
remove the redundancy seekable setting.

Reviewed-by: Steven Liu <lq@onvideo.cn>
Signed-off-by: Jun Zhao <barryjzhao@tencent.com>

lavf/hls: drop unnecessary check before ff_format_io_close

ff_format_io_close will check the AVIOContext pointer pb, so drop
the unnecessary check before ff_format_io_close.

Reviewed-by: Steven Liu <lq@onvideo.cn>
Signed-off-by: Jun Zhao <barryjzhao@tencent.com>

lavf/hls: refine the log message

refine the log message, it's will help the debugging

Reviewed-by: Steven Liu <lq@onvideo.cn>
Signed-off-by: Jun Zhao <barryjzhao@tencent.com>

avformat/hlsenc: Fix memleak when using single_file

This commit fixes a memleak in the hls muxer when one uses a single file
as output. It has been forgotten to free the temporary buffers used to write
the packets so that the size of the leaks basically amounts to the size
of the output file. This commit adds the necessary free.

Reviewed-by: Steven Liu <lq@onvideo.cn>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>

hlsenc: flush segments to guarantuee atomic single file hls

avcodec/mjpegdec: Restore non bayer checks in ljpeg_decode_rgb_scan()

Fixes: out of array write
Fixes: 17088/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_JPEGLS_fuzzer-5654877765632000

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/mjpegdec: Only allow 0x11110000 pix_fmt_id for bayer mode

Fixes: NULL pointer dereference
Fixes: assertion failure
Fixes: 17003/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_MJPEG_fuzzer-5696929253556224
Fixes: 17039/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_MJPEG_fuzzer-5651008134316032

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avformat/sccdec: remove not needed code

avfilter/vf_v360: use always float version instead of double floating-point function

avfilter/vf_v360: fix flat projection field of view calculation

cmdutils: promote report level if loglevel is higher

avfilter/vf_v360: add dfisheye output

avfilter/vf_v360: fix setting default height for stereographic

avfilter/vf_v360: fix errors is comments

avfilter/vf_v360: fix M_PI_4 usage consistency

avformat/dashdec: add startNumber parser for segmentlist

and get start_number for compute current segment number.

fix ticket: 7976

Signed-off-by: Steven Liu <lq@chinaffmpeg.org>

avformat/hlsenc: fix memleak at hls_write_trailer

Found-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: Steven Liu <lq@chinaffmpeg.org>

avformat/hlsenc: print warning at the end when upload the last segment failed

Signed-off-by: Steven Liu <lq@chinaffmpeg.org>

avformat/hlsenc: merge fmp4 and mpegts segment type m3u8 list AVIOConext

hlsenc has been merge fmp4 and mpegts workflow before
so it can merge m3u8 list AVIOContext now.

Signed-off-by: Steven Liu <lq@chinaffmpeg.org>

x85/opusdsp: enable the functions on all FMA3 CPUs

It's not using ymm registers, so limiting it to CPUs with fast AVX
is not necessary.

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

x86/opusdps: clear the high bits from some gprs

Fixes checkasm on systems like win64.

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

avcodec/htmlsubtitles: Avoid locale dependant isdigit()

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

avcodec/mpeg12enc: Add FF_CODEC_CAP_INIT_CLEANUP

Fixes: Multiple memleaks
Fixes: ffmpeg-memory-leak

Found-by: Francis Provencher <francis@protekresearchlab.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/alsdec: Check k from being outside what our implementation can handle

The specification does not seem to list what the maximum valid
value is

Fixes: shift exponent 32 is too large for 32-bit type 'unsigned int'
Fixes: 16268/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_ALS_fuzzer-5638164544225280

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/takdec: Fix integer overflow in decorrelate()

Fixes: signed integer overflow: -2424832 - 2145653689 cannot be represented in type 'int'
Fixes: 16138/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_TAK_fuzzer-5643451346976768

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/aacps: Fix integer overflows in hybrid_synthesis()

Fixes: signed integer overflow: -822667928 + -1399761199 cannot be represented in type 'int'
Fixes: 15756/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_AAC_FIXED_fuzzer-5645182051024896

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/mpeg4videodec: Fix integer overflow in mpeg4_decode_studio_block()

Fixes: signed integer overflow: 24023040 * 112 cannot be represented in type 'int'
Fixes: 16570/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_MPEG4_fuzzer-5173275211071488

Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Reviewed-by: Kieran Kunhya <kierank@obe.tv>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

avcodec/vp56rac: delay signaling an error on truncated input

A threshold of 1 is sufficient for simple_dump_cut.webm, 10 is used
just to be sure the next truncated file doesnt cause the same issue

Obvious alternative fixes are to simply accept that the file is broken or to
write some advanced error concealment or to
simply accept that the decoder wont stop at the end of input.

Fixes: Ticket 8069 (artifacts not the differing md5 which was there before 1afd246960202917e244c844c534e9c1e3c323f5)
Fixes: simple_dump_cut.webm
Fixes: regression of 1afd246960202917e244c844c534e9c1e3c323f5

fate-vp5 changes because the last frame is truncated and now handled
differently.

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

avcodec/v4l2_context: set frame SAR using VIDIOC_CROPCAP

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2_buffers: use correct timebase for encoder/decoder

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2_m2m_dec: fix compile

internal.h header required for FF_CODEC_CAP_SETS_PKT_DTS

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2_m2m_dec: fix dropped packets while decoding

* FFmpeg retrieves a packet from the bitstream
* It attempts to get an input buffer (from its own list or by dequeuing one from the driver)
* If no input buffer is found, the bitstream packet is dropped instead of scheduled for trying again later

It's an issue that showed especially at high speeds (like using `-f null -` as output parameters).

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2_context: expose timeout for dequeue_frame

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2_m2m_dec: set pkt_dts on decoded frames to NOPTS

Without this ffmpeg will attempt to copy the dts from the
most recently enqueued packet into the most recently dequeued
frame, which does not account for the buffering inside v4l2
and is not accurate.

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2_buffers: read height/width from the proper context

Frames are generally dequeued into capture buffers, so using
the output properties here was incorrect. It happened to work
fine for decoding, since the output/capture buffers have the same
dimensions.

For the v4l2 scaler, the dimensions can be different between output
and capture. Using the buffer's associated context makes this code
work correctly regardless of where the frame is coming from.

Signed-off-by: Aman Gupta <aman@tmm1.net>

avfilter/vf_v360: remove pointless special case when allocating stuff

avfilter/vf_v360: refactor FoV calculation for stereographic

Move in out of loop.

avcodec/h2645_parse: simplify memset call

Removed (new_size - pkt->nals_allocated) because this value is always 1
during the call.

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

avfilter/vf_v360: extend max limit to vertical fov

avfilter/vf_v360: really fix FoV selection

Big thanks to Michael Koch for providing actual formula.

lavfi/sinc: fix memory leak in error handling path

free work&pi_wraps in error handling path to avoid memory leak.

Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Jun Zhao <barryjzhao@tencent.com>

lavfi/anequalizer: fix memory leak in error handling path

free the pad.name in error handling path to avoid memory leak.

Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Jun Zhao <barryjzhao@tencent.com>

checkasm: add opusdsp tests

opusdsp: adjust and optimize C function to match assembly

The C and asm versions behaved differently _outside_ of the codec.

The C version returned pre-multiplied 'state' for the next execution
to use right away, while the assembly version outputted non-multiplied
'state' for the next execution to multiply to save instructions.
Since the initial state when initialized or seeking is always 0,
and since C and asm versions were never mixed, there was no issue.

However, comparing outputs directly in checkasm doesn't work without
dividing the initial state by CELT_EMPH_COEFF and multiplying the
returned state by CELT_EMPH_COEFF for the assembly function.

Since its actually faster to do this in C as well, copy the behavior the
asm versions use. As a reminder, the initial state 0 is divided by
CELT_EMPH_COEFF on seek and init (just in case in the future this is
changed, its technically more correct to init with CELT_EMPH_COEFF than 0,
however when seeking this will result in more audiable pops, unlike with 0
where the output gets in sync over a few samples).

avcodec/v4l2_buffers: split out v4l2_buf_increase_ref helper

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2_buffers: split out V4L2Buffer generation into helper method

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2_buffers: split out AVFrame generation into helper method

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2_buffers: teach ff_v4l2_buffer_avframe_to_buf about contiguous planar formats

This fixes h264_v4l2m2m encoding on the Raspberry Pi

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2_m2m_enc: fix indentation and add M2MENC_CLASS macro

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2_m2m_dec: fix indentation and add M2MDEC_CLASS macro

This just makes the M2MDEC_CLASS similar to how it is done in rkmpp. It looks
clean and has proper indentation

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2_m2m: decouple v4l2_m2m helpers from AVCodecContext

This will allow re-use of the m2m backend with AVFilterContext

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2_buffers: Add handling for NV21 and YUV420P

The single planar support was for NV12 only.
Add NV21 and YUV420P support.

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2_m2m: fix av_pix_fmt changing when multiple /dev/video* devices are probed

On the RPI, three different /dev/video devices exist (decoder, scaler, encoder).

When probing the devices in order, the originally requested pix fmt
would be mutated causing the wrong one to be chosen when a matching
device was finally found.

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2_m2m: disable info logging during device probe

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/v4l2_m2m_enc: add support for -force_key_frames

Signed-off-by: Aman Gupta <aman@tmm1.net>

avfilter/vf_v360: remove w/h changing for flat output

Fixes crashes with bigger h/v fov.

avfilter/vf_v360: increase h_fov max limit

avfilter/vf_v360: fix fov calculation for stereographic output

avcodec/mediacodec_surface: define and use FFANativeWindow to mimic NDK interface

This will make it easy to switch to ANativeWindow_fromSurface
and ANativeWindow_release in the future.

Signed-off-by: Aman Gupta <aman@tmm1.net>

avcodec/mediacodecdec: warn when input buffers are not configured with proper size

In rare circumstances, if the codec is not configured with the
proper parameters the input buffers can be allocated with a size
that's too small to hold an individual packet. Since MediaCodec
expects exactly one incoming buffer with a given PTS, it is not
valid to split data for a given PTS across two input buffers.

See https://developer.android.com/reference/android/media/MediaCodec#data-processing:

  > Do not submit multiple input buffers with the same timestamp

Signed-off-by: Aman Gupta <aman@tmm1.net>

doc/filters: extend flip options in v360 filter