| Age | Commit message (Collapse) | Author |
|---|
|
The introduced function give "frontend like" plugins a chance to
lock and unlock port rewiring. This protects such threads (when
combined with holding the frontend lock) from beeing blocked when
calling functions like xine_get_current_frame().
|
|
|
|
The idea is to allow only a "single" frontend to rewire ports at
a certain point in time. Regarding a stream, frontend_lock is used
for example to allow only a single frontend to change the speed.
Unfortunately, frontend_lock cannot be used as the rewire functions
are not stream related.
Therefore a new port_rewiring_lock was introduced and used at
appropriate locations. When an arbitrary thread now holds the
frontend_lock and the port_rewiring_lock, it is safe that acquiring
a port ticket in functions like xine_get_current_frame() will
never block the thread.
|
|
One can assume that a still frame is to show when there are no more
frames left to display.
The changed code uses _x_query_buffer_usage() to retrieve the number
of frames waiting to be displayed to integrate this information into
the decision.
|
|
|
|
There are some situations where bob deinterlacing doesn't make much
sense, as the effect doesn't work for example for slow motion. And
in fast motion mode, the sleep between displaying the two fields
lowers the reachable fast motion frame rate. Another annoying effect
is the reduction in vertical resolution for still images.
The changed code tries to detect such issues and disables bob
deinterlacing for such frames. The detection of still images is
still to come.
|
|
Bob deinterlacing is implemented as showing the top field, sleeping
for half the frame duration and showing the bottom field. Most
drivers tend to synchronize displaying a field on the VBI and thus
displaying a field may take up to half the frame duration in certain
cases.
According to the original code, the sleep took always half the
frame duration and therefore the second field could get displayed
too late. As a result, the driver was syncing to VBI most often,
so that things got even worse.
The changed code now calculates the sleep time in a way that the
second field gets displayed half the frame duration after the
first field. Moreover, it monitors how much time was spent to
display the first field and when this time exceeds 75 % of the
field time (= half the frame time), it skips displaying the second
field, as usually this is an indicator that the driver has no
more frame buffers left. So displaying the second field would just
make things go worse.
|
|
The current implementation keeps references to VO_NUM_RECENT_FRAMES
frames (for deinterlacing), but doesn't make any use of them.
As many XXMC capable devices only supply 8 frames at all, keeping
fewer frames referenced makes more available for decoding and thus
avoids frame drops by keeping the number of frames which are ready
for display more often above frame_drop_limit.
|
|
The current code misses the ability to switch back to an
unaccelerated context, e. g. when previously MPEG2 material
was displayed which is then followed by H.264 material. As
the latter is not handled as XINE_IMGFMT_XXMC there was no
way to leave the accelerated context and therefore the images
did not appear on screen.
|
|
This function shall be used to poll the number of remaining frames
from a certain point in time on until the reported numbers are all
0. At that point in time, the content on screen is identical to a
certain state of the stream, at which for example, a hardcopy may
be taken.
|
|
The current code has a race condition which can block arbitrary
threads that call for example xine_get_current_frame() until the
stream gets unpaused again. This can happen when the internal
ticket acquiration collides with a ticket revokation for example
when another thread is going to pause the stream.
There are a few situations where a port ticket needs to be
acquired for calling a port function but where it is absolutely
undesireable to get blocked for an undetermined period of time.
Therefore the ticket system should be extended by nonblocking
functions which allow ticket acquiration even when a ticket
revokation is in progress. And in the case where blocking is
not avoidable, it should simply be indicated that no ticket was
acquired. The caller can then choose to repeat the call at a
later point in time.
|
|
The _x_config_init() and _x_config_change_opt() functions should stay internal
and not be exported by the library, so remove XINE_PROTECTED to hide them.
|
|
The current code implements hardware (a shift register) in software
just to find the byte pattern 00 00 01 xx, which causes remarkable
CPU load on less powerful machines.
The new approach uses memchr() to find the 01 in the buffer, which
most often hits a start code. memchr() seems to be even faster then
implementing a real pattern search (i. e. by just looking at every
third byte to find 01). The new implementation causes significantly
fewer CPU load on less powerful machines.
|
|
The current code emits a frame when a non slice start code is seen.
For still frames, this is typically a sequence end code. But the
current code doesn't call parse_chunk() immediately because it waits
for a further start code to determine the chunk of data to pass to
parse_chunk(). But there isn't such a further start code for still
frames after the sequence end code and thus, the still frame will
not be emitted.
As sequence end code is the only start code which has no data
according to the MPEG specification, let's use this information
to call parse_chunk() immediately.
|
|
The current code cannot detect the absence of AFD once it has been
seen in the stream. As AFD can appear in user data after sequence,
group or picture start codes, the idea is to reset the stored AFD
value when processing the sequence start code. In the case where
AFD is seen in user data, it is stored internally, to have it ready
when the first slice is processed. At least at that time, AFD data
has been seen and can be analyzed for changes. At any change, the
AFD value will then be stored into a stream property. Doing this
only for changes avoids locks while writing the same value over and
over to the stream's property.
|
|
determination.
The current code detects MPEG2 when parsing the sequence extention
which appears only in MPEG2 streams. But this is to late for correct
aspect ratio determination which happens earlier and therefore
assumes MPEG1. The result is a totally wrong aspect ratio.
To fix this issue, the next start code (which is already available
at that time) is passed to parse_chunk() too, which can then be
used to detect a MPEG2 extension start code early enough before
the aspect ratio is determined.
|
|
|
|
|
|
|
|
xine_musepack_decoder.c in libxineadec.
--HG--
rename : src/libmusepack/xine_musepack_decoder.c => src/libxineadec/xine_musepack_decoder.c
|
|
the patching needed.
Contestually, move libmpcdec code in contrib/libmpcdec, so that it's not mixed together with xine code.
The use of MUSEPACK_LIBS/MUSEPACK_CFLAGS intermediate is to make simpler move to pkg-config checks if libmpcdec developers decide to install one.
|
|
|
|
|
|
extend the documentation available before.
|
|
|
|
|
|
as needed for some files where gcc runs out of registers otherwise.
|
|
This fixes linkage problems for people who want this but don't want X.
|
|
|
|
First of all, as we check the presence of the attribute in configure.ac and
we generate the XINE_PACKED definition there, there's no need to do the work
again on attributes.h.
Also remove #pragma pack support for now, as the trick is just not neat enough,
and as it is it's just noise.
Make Wine, libdvdnav and nosefart use XINE_PACKED rather than doing their own
checks, as they are less reliable.
|
|
32-bit value.
|
|
bswap.h already.
|
|
|
|
|
|
|
|
enabled in configuration.
As external FFmpeg does not install its own ffmpeg_config.h, we have
to assume that it supports every buffer type; this should probably
be checked for errors when opening the decode, but at least it should
work to the same extent it worked on 1.1.
|
|
The dvdata.h is never installed by FFmpeg, so when using the external
copy you have to still use the internal dvdata.h file at the moment.
|
|
This way there's no need to fake it when using the external copy.
|
|
|
|
|
|
Altivec code.
|
|
Two of the modified files are headers, but each contains definitions as well as
declarations and is only ever used once.
|
|
|
|
from the FFmpeg's build directory (as it's not easy to take it with relative path).
|
|
With the new way to build FFmpeg, there's no need for a special function
to use when building with internal FFmpeg, so just use the one already
provided by libavcodec.
Codecs that shouldn't be registered should just be disabled in the build.
|
|
it, we should put it in contrib/ instead.
|
|
|
|
|
|
to be video_overlay.
|
|
|
