The buffer is used for an additional pass of frame compression, so videos
can be coded without ever using it (and some are coded so indeed, e.g. in
Woodruff and the Schnibble of Azimuth game).

Indeo 4 has quantiser range 0-31 instead of 0-23 for Indeo 5, and clipping
quantiser leads to incorrect quantisation and DC prediction on low-quality
videos.

This fixes bug 259.

A few of the h264qpel neon functions are shared with other
hpeldsp functions in this file.

This fixes standalone compilation of the h264 decoder on arm.
Signed-off-by: Martin Storsjö <martin@martin.st>

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

This fixes build errors in cases where FFT is disabled.
Signed-off-by: Martin Storsjö <martin@martin.st>

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

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

Odd chunk size makes no sense for stereo and incomplete chunks are
not supported.

Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org

Simplify a little, assume empty frames are acceptable and
do not pointlessly reinit the bytestream2 contexts using
possibly wrong size values.

Reported-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org

CC: libav-stable@libav.org

While at it drop useless parentheses.

The decoder did not start predicted frames with missing references but
called the end of frame handling on them anyway. This caused an
assertion in the VA API HW accelorator since it tried to render a
picture with uninitialized buffers.

This way we don't require the caller to parse the block header. It also
allows the channel configuration to change mid-stream.

This way we don't require the caller to parse the block header. It also
allows the sample rate to change mid-stream.

Currently if no custom block VLC is specified for band, a default block VLC
will be used, while the global block VLC stored in the context should be used
instead.

This fixes decoding of one sample I have.

That involves fixing INV_HAAR4() macro and changing it to work with different
input and output like INV_HAAR8() instead of in-place transform.

Also add a note about the feature in the changelog.
Signed-off-by: Martin Storsjö <martin@martin.st>

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

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

Now it can use different references for those blocks and even use
averaging.

This fixes several chroma artifacts in several videos.
Signed-off-by: Martin Storsjö <martin@martin.st>

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

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

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

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

Use the intensity-compensated reference frame for subsequent
fields/B-frames.

Since we currently don't change the reference frame we have to
maintain lookup tables for intensity compensation in the following
dependent frames.
Signed-off-by: Martin Storsjö <martin@martin.st>

This simplifies the code since copying MVs to the reference is not needed
anymore (and maybe something about fixing artifacts).

Also remove the unused mv_f_last.

Fixes a small number of artifacts in black_screen_VC-1.mkv
and several more artifacts in other videos.
Signed-off-by: Martin Storsjö <martin@martin.st>

This allows us to get rid of demuxer-specific hacks in the decoder and
will allow streamcopy from matroska once we have a wavpack muxer.

Currently the demuxer shaves the blocks and exports only the
information that is useful to the decoder.

Exporting the blocks just as they are stored is simpler to understand
and will make remuxing wavpack easier.

We don't know whether it will be float until we look at the data.

The decoder always returns output if an error does not occur.

Such blocks are not valid.

Number of samples in the first block is checked to be strictly positive
earlier in wavpack_decode_frame() and number of samples in all the other
blocks is checked to be equal to the first one.