This is now libvlc_track_description_list_release, as agreed at the VDD
technical discussion

This custom setting makes sure the codec actually fills the
OMX buffers with data, instead of opaque pointers to buffers
residing elsewhere.

This is needed in order to get sensible data output on
Nexus S, and most probably also on Galaxy S.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

OMX_SetConfig still lacks a good way to figure out the size of
the passed struct/variable, for now hardcoded to the size of the
only use case.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

The Samsung OMX components on e.g. Nexus S require major == 1
and minor == 0, otherwise they simply return errors on all
SetParameter/GetParameter calls.

Stagefright's OMX client always sets version 1.0.0, which thus
probably should be a safe choice for all android cases.

An alternative would be to store the version to set as a
variable in decoder_sys_t, and set that to 1.0.0 conditionally
only for Samsung OMX components, but that would make
OMX_INIT_COMMON more complex.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Also rename confusing vlc_free()

We're in 2011 guys !

This adds a fake OMX core implementation, relying the calls
via IOMX to the media server, which contains the actual
OMX core.

Building with IOMX requires private Android headers from the
Android source tree, namely the frameworks/base and
system/core repositories. (Either froyo or gingerbread
should work for building.) This API is not public, has no
ABI guarantees and isn't supported.

Linking also requires libraries extracted from a
froyo/gingerbread device or emulator.

Since there are no ABI guarantees, linking to this API might
make the .so fail to load on some devices, so for proper use
it should be in a dynamically loaded module, separate from the
rest of the VLC core and modules.

Since this can lead to crashes on unsupported devices, it should
only be used in production on whitelisted device/firmware
combinations that are known to work.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

It is useful in the IOMX wrapper.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Removal of tabs and trailing spaces
Fixing configuration with autoconf 2.67

API.

Also, distinguish between SSE and SSE2 CPUs.

This reverts commit d0dc1be9.

Must not block if we have no internet access

I'm not sure if these officially are NV12 or NV21 - on Nexus One
they correspond to NV21 at least.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

By pushing a sentinel/dummy buffer on the input buffer fifo
when the output needs reconfiguring, we avoid deadlocks on
HTC Sensation, where the input buffers aren't returned to
the caller until the decoder has been reconfigured.

Since all input buffers can be used up before the reconfigure event
is received, this earlier led to a deadlock where the decoder
waited for the caller to reconfigure it (as a response to the
reconfigure event), while the caller waited for an input buffer
to be freed to put the next input packet into.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

The nSliceHeight parameter in OMX apparently can indicate plane
height. This shows up on Nexus One with recent CyanogenMod
with the OpenMAX driver built from source (as opposed to the
older ones with a binary-only OpenMAX driver).
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

On OpenMAX devices supported earlier, the number of buffers has
stayed the same after reconfiguring, but on the Nexus One, the
number changes.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Earlier, an uninitialized string would be printed if
OMX_ComponentRoleEnum failed (or if it was unimplemented, see
omxil_utils.h), potentially leading to out of bounds reads.
Signed-off-by: Jean-Baptiste Kempf <jb@videolan.org>

Yet again...