There is no way for vmem to know the required padding otherwise.
This partly reverts commit 260a3c10.

This seems to work just fine on Samsung Galaxy S3 and Sony Xperia Z1.
(This didn't use to work in earlier Galaxy S3 firmwares even though
it was exposed in MediaCodecList, but currently it seems to work
just fine.)
Signed-off-by: Martin Storsjö <martin@martin.st>

This is pretty much normal NV12, but with stride and plane
heights aligned to 32 (according to a commit on code aurora
forum). The actual values are signalled in the normal stride
and slice height fields, so therefore it can be treated as
normal NV12 as far as our code is concerned.

This is returned on the Sony Xperia Z1 (Snapdragon 800) -
mediacodec works (with a quirk mentioned in the previous commit),
while our omxil/iomx code fails to allocate buffers.
Signed-off-by: Martin Storsjö <martin@martin.st>

The decoder on Sony Xperia Z1 seems to return one frame before
the output format is set (and calling getOutputFormat() at this
point throws an IllegalStateException). Therefore just drop any
frames that are returned before the output format has been set.
Signed-off-by: Martin Storsjö <martin@martin.st>

Also do not define unused PROJECTM2_(CFLAGS|LIBS) substitutions.

Also do not hard-code -lm for the fingerprinter.

Input and output channel count should be anyway the same and it's
hopefully littlebit clearer than fmt_in and fmt_out mixups.

Ref #9200

This way we get the features of both worlds: from 0% to 100% you
get quick volume change with no saturation, and you can still get
amplification over 100%.

Fixes #9371

Fix build of files including <list> on case-insensitive filesystems

Bug appeared after c1452fef silently changed module selection rules

Broken in 79ced312

142e288b set LDFLAGS conditionally
Be sure that it is always set and concatenate the DARWIN flags

The modules concerned all had non unique names (e.g. flac for both
the decoder and the demuxer).

MODULE_NAME macro uses Makefile target to retrieve a unique name
This was broken in 20cdb051, as
explained by automake documentation:

 When using a per-target compilation flag, Automake will choose a
 different name for the intermediate object files.

This reverts commit 892d21f1.
Thsi makes no sense.

Signed-off-by: Rafaël Carré <funman@videolan.org>