With there being two hwaccels that use the CUDA pix_fmt now, just
relying on the pix_fmt to identify the selected hwaccel is not enough
anymore.

So this checks if the user explicitly selected a hwaccel, and only
accepts that one.

Some parts of the code are based on a patch by
Timo Rothenpieler <timo@rothenpieler.org>

Merges Libav commit b9129ec4.

Due to the name clash with our cuvid decoder, rename it to nvdec.

This commit also changes the Libav code to dynamic loading of the
cuda/cuvid libraries.
Signed-off-by: Timo Rothenpieler <timo@rothenpieler.org>

It allows us to specify what kind of audio parameter changes are allowed.

Should fix ticket #6721.
Signed-off-by: Marton Balint <cus@passwd.hu>

Signed-off-by: Peter Große <pegro@friiks.de>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Fixes looping files without audio or when using stream_copy, where
ist->nb_samples is not set since no decoding is done.

This fixes ticket #5719 and also fixes an endless loop with the sample
in ticket #6139.
Signed-off-by: Peter Große <pegro@friiks.de>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Typically only a small subset of the SDL texture formats are supported directly
by the SDL renderer drivers, the rest is software emulated. It's better if
libswscale does the format conversion to a hardware-accelerated texture format
instead of SDL.

This should fix video render slowdowns with some texture formats after
3bd2228d.
Signed-off-by: Marton Balint <cus@passwd.hu>

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

Since a7da1347, flush packets are passed
to process_input_packet() during stream copy.  This modifies the input
timestamp handling to ignore them - since they contain no data, timestamps
should not be affected.

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

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

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

It has no effect whatsoever since the major bump.
Replace the flag's documentation to reflect this as well.
Signed-off-by: James Almer <jamrial@gmail.com>

Deprecated since October 2015.

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

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

Deprecated (aka removed) in OSX 10.11, and we have a replacement for it
(VideoToolbox).

The private codec option will be used instead when available.

Otherwise the frame size of the codec is not set in the buffersink.

Fixes ticket #6603 and the following simpler case:

ffmpeg -c aac -filter_complex "sine=d=0.1,asetnsamples=1025" out.aac
Signed-off-by: Marton Balint <cus@passwd.hu>

This has been unused for a long time, and the original purpose has been
replaced by the per-stream hwaccel_flags.
Signed-off-by: Jun Zhao <jun.zhao@intel.com>
Signed-off-by: Mark Thompson <sw@jkqxz.net>

This is required for FLV files, for which duration_pts comes out to be zero.
Signed-off-by: Sasi Inguva <isasi@google.com>
Reviewed-by: Thomas Mundt <tmundt75@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

Since af1761f7 ffmpeg waits for a frame in each
stream before writing the output header. If we are using threaded decoding for
attached pictures, we have to read till EOF to be able to finally flush the
decoder and output the decoded frame. This essentially makes ffmpeg buffer all
non-attached picture packets, which will cause a "Too many packets buffered for
output stream" eventually.

By forcing single threaded decoding, we get a frame from a single packet as
well and we can avoid the error.

Fixes part of ticket #6375:
ffmpeg -i 46564100.mp3 -acodec libmp3lame -ab 128k -ac 2 out.mp3
Reviewed-by: Hendrik Leppkes <h.leppkes@gmail.com>
Signed-off-by: Marton Balint <cus@passwd.hu>

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