Previously, if the absolute delay dropped by a longer interval than the
packetization time, we could end up repeatedly waiting into the future,
and never dequeue anything.

Signed-off-by: Felix Paul Kühne <fkuehne@videolan.org>

use the compiler to check -iquote support
faad use inline asm => no thumb support

This reverts commit 2d235298.

When reloading, correctly update the view.

Maybe playlist_ServicesDiscoveryAdd could return the playlist_item_t* ?
This is ugly, and should be fixed.

And correct decoration.

The idea behind this is to simplify the playlist_model and have a better SD control.

This conversion is always needed between mixer and output.

 * always use NEON VCVT
 * avoid pointless interlace/deinterlace in load/store

This patch adds the following keywords for skins developpers :
      - vlc.canRecord      (boolean / refers to can-record)
      - vlc.isRecording    (boolean / refers to record)
      - vlc.toggleRecord() (function / toggle recording)
      - vlc.nextFrame()    (function / next frame)

Sending one frame per block had way to high CPU overhead.

In some circumstances QTreeView does not behave properly if the column index 0 can hide/reappear.

Precisely, when column 0 is hidden and an item get's subitems, those subitems can not be D&D-ed.
If column 0 is then unhidden, trying to D&D the subitems will fail or result in unexpected behaviour.

So that tree branches appear on it.

Currently, detection is done at build time. Feel free to add run-time
detect if you know how.

While VFP/NEON do support floating point instructions, they appear to
be noticeably slower than fixed point.

You don't want it to be selected automatically.