This reverts commit 96640312acefdfdd70ded1773353e34e85a3b637.

On AMD (ATI) graphics cards only one thread can use XOpenDisplay() and XCloseDisplay().

Use define VOUT_MAX_PICTURES from vout_internal.h instead of fixed number.

- subpicture_cache_destroy() set cache pointer to NULL
- CopyPictureToVAImage() invalidate image members

It is possible for the cache_counter to reach its maximum value. In that case abort().

Turns out vout_Close() does the same except for the vout_control_WaitEmpty().
The vout_control_WaitEmpty() is not needed according to Laurent Ailmar.

Wait till the stop command has been processed by the thread before continuing.

If the decoder fifo is full (growing to large), then instead of flushing and
producing a big artefact in the stream, just drop one buffer. This still limits
the size of the fifo, but does not produce a big artefact as the original code.

Also make the size configurable with a commandline option.

Make sure that when the VAAPI-XCB video output has been started no other picture chroma,
will be accepted. The punishment for not doing this is random crashes, since surfaces are
shared between display pictures and the decoding thread. The picture is the owner of the
surface. If not honoring this, then the decoder might write into surfaces being displayed.
This will result in random crashes since surface memory will be randomly overwritten.

When the vout is not reused the display thread was not properly stopped.

On AMD/ATI the X11 connection is shared between the decoding and displaying thread. Therefor close it only once. This is a workaround and might break NVIDIA GPU's.

Sync surfaces before they are destroyed. The surfaces can be used from more then
one thread context. Therefor make sure that all operations on the surface have
been performed. Then destroy the surface.

Fix version defines for avcodec.

The VA API output on AMD does not like to be recycled. Therefor add
an option to evade the vout recycle path.

The function vlc_va_get_conn() is only used inside va.c.

Conflicts:
	modules/codec/avcodec/video.c
	modules/stream_filter/httplive.c

The function is a bit ugly but it's difficult to do this using
pure arithmetic.

Fixes #6378
(cherry picked from commit 1e9dc296)
Signed-off-by: Pierre Ynard <linkfanel@yahoo.fr>

This will only be used if requested explicitly,
that is to say '--v4l2-chroma=h264'.

(cherry picked from commit 28db7725)

DBCS countries assign their currency symbol to '\', ASCII 92. This causes
an unexpected behavior when converting a directory separator on OS/2.
In case of GNU libiconv, it does not treat '\', ASCII 92, as a currency
symbol at all. So let's mimic GNU libiconv.
Signed-off-by: Rafaël Carré <funman@videolan.org>
(cherry picked from commit 8acdc1f5)
Signed-off-by: KO Myung-Hun <komh@chollian.net>
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

OS/2 supports UCS-2 only instead of UTF-16.
Signed-off-by: Rafaël Carré <funman@videolan.org>
(cherry picked from commit 2cc70e68)
Signed-off-by: KO Myung-Hun <komh@chollian.net>
Signed-off-by: Rémi Denis-Courmont <remi@remlab.net>

(cherry picked from commit 5785e63a)

Close #6403

This patch adds isScrollable() to a control, so that we can distinguish
controls that require mouse wheel (playtree, slider) from those that don't.
In the latter case, we can then forward events to vlc core.

this fixes #6457.
(cherry picked from commit f3afffbc)
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Without this flag, PulseAudio will tweak the buffer parameters, but
will not account for hardware latencies. As a consequence, the live
caching parameter is not really followed - the latency is longer.

Longer latency becomes a big problem when trying to synch with another
live source (such as a V4L2 video capture device). Then audio is too
late and gets discarded or messed up at output.
(cherry picked from commit d8a96604)

based on patch by Richard Grenville <pyxlcy@gmail.com>
(cherry picked from commit 66da497b)
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

(cherry picked from commit cd470819)
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Close #6451
(cherry picked from commit ef790e37)
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

According POSIX the va_list is defined in stdarg.h
http://pubs.opengroup.org/onlinepubs/009695399/basedefs/stdarg.h.html

On some systems (like uClibc) the detection of vasprintf will fail
unless stdarg.h is included and build will later fail due to a vasprintf
redeclaration.
Signed-off-by: Natanael Copa <ncopa@alpinelinux.org>
Signed-off-by: Rafaël Carré <funman@videolan.org>
(cherry picked from commit 2859c5db)
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Fixes undocumented/poorly documented Windows behavior. RemoveFilter function does a silent Release().
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>
(cherry picked from commit 3f6ccc0d)
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

(cherry picked from commit ea068351)

The slider is not displayed, anyway. But sometimes it flickers through
for a short moment which does not look good.
(cherry picked from commit 2fc5dc5f)
Signed-off-by: David Fuhrmann <david.fuhrmann@googlemail.com>

(cherry picked from commit d612827c)
Signed-off-by: David Fuhrmann <david.fuhrmann@googlemail.com>

Note that the origin values can also be negative when we have our window
on a secondary screen.
(cherry picked from commit 1cee2e3b)
Signed-off-by: David Fuhrmann <david.fuhrmann@googlemail.com>

(cherry picked from commit e6820b61)
Signed-off-by: David Fuhrmann <david.fuhrmann@googlemail.com>

(cherry picked from commit 3646cf43)
Signed-off-by: David Fuhrmann <david.fuhrmann@googlemail.com>

(cherry picked from commit bf916c35)
Signed-off-by: David Fuhrmann <david.fuhrmann@googlemail.com>