Close #6640

Close #6556

This should work-around distorsion due to excessively short latency
computation in PulseAudio, e.g. NVIDIA HDMI Audio.

In the longer term, VLC should probably set minreq to 40ms and tlength
to a larger value or even -1. But that would require more testing.

Close #7543

This does not actually seek to the right position yet
Ref #6895

It could still be better. If the disconnection is unrecoverable,
I suspect a flood of identical errors will be emitted continuously
as the core keeps pushing audio buffers.

Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

The MTA is fully taken care of by mmdevice.c.

vout_display_opengl_Delete will call OpenglLock, so it is a bad idea to previously release sys->glView

Close #7292

MSDN mostly fails to explain COM requirements and totally fails to
explain thread-safety requirements. This patch takes the most
pessimistic approach.

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

VLC_GENERIC was particularly bad for RISC processors.

Actual support for changing device requires (yet another) addition to
the audio_output_t callbacks set, I think.

Apparently, there is no need for COM to do that.

This reverts commit 7f807d9b.

It fixes skins2 crashing when an input is started.

Ref #7842

Since NO_NEON is not used anymore, leaving it undefined on ARMv6 will result in HAVE_NEON showing up, resulting in crashes.

ARMv7 Android builds are now built to tegra2 with a special dynamic NEON loader for libav.
Signed-off-by: Konstantin Pavlov <thresh@videolan.org>

Patch taken from Gentoo portage.
http://sources.gentoo.org/cgi-bin/viewvc.cgi/gentoo-x86/media-libs/libmad/files/libmad-0.15.1b-gcc44-mips-h-constraint-removal.patch?revision=1.1

Signed-off-by: Konstantin Pavlov <thresh@videolan.org>

Ref #7838

Remove libvlc hack, and fixes missing input state changes