Something appeards to be very wrong with codec registration.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@39885 a95241bf-73f2-0310-859d-f6bbb57e9c96
* replaced .. with $(DOTDOT) in Jamfiles
* added __STDC_CONSTANT_MACROS define for C++ files (required for UINT64_C)
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@39537 a95241bf-73f2-0310-859d-f6bbb57e9c96
Initialize the stream's IO buffer, otherwise it could crash in ~StreamBase().
It's only initialized when the media format was actually successfully probed.
Fix#6595. Thanks.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@39374 a95241bf-73f2-0310-859d-f6bbb57e9c96
few warnings. Thanks! I did not apply the hunks about moving
a logging function in the common accelerant code to be static.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@39320 a95241bf-73f2-0310-859d-f6bbb57e9c96
pointer in the media_format user data section.
* In the AVCodecEncoder, optionally use the AVCodecContext pointer
from the AVFormatWriter instead of its own instance. The problems
I am investigating are not improved by this, but it may be needed
anyway.
* Map the bitrate for audio to a fixed table. Certain encoders will
refuse to use a non-standard bitrate, like the currently enabled
AC-3 encoder.
* Fixed tracing output in _EncodeAudio().
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@39039 a95241bf-73f2-0310-859d-f6bbb57e9c96
find a way to do this properly, though. The pixel format
selected by AVCodecEncoder should match here.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@39037 a95241bf-73f2-0310-859d-f6bbb57e9c96
that they can modify the media_format passed in. For example they
can store information in the user_data section. I don't actually
use this anymore, but it may come in handy again.
AVFormatWriter:
* Adjust the AVCodecContext flags not only for video, but also
for audio streams (as the API example does). This mechanism
may not yet work, since the AVCodecEncoder actually uses a
different AVCodecContext instance.
* Use the encodeInfo->flags and specify the key frame flag
for the AVPacket. This finally makes videos encoded on Haiku
seekable.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@39035 a95241bf-73f2-0310-859d-f6bbb57e9c96
hard to see the big picture and tell what actually works
in the higher levels of the code.
* Coding style cleanup.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@39033 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Don't hardcode PIX_FMT_YUV420P, but use the color space
from the list of codec supported ones which has the best
quality.
* Maintain media_encode_info->flags by specifying whether
a chunk contains a keyframe. According to the libavformat
API example, all audio chunks are keyframes. For video
chunks, use the fContext->coded_frame->key_frame flag.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@39032 a95241bf-73f2-0310-859d-f6bbb57e9c96
certain video sizes (720x576 for example) and with 48 kHz raw
audio. Clockwerk actually uses a mechanism which worked on BeOS
to check if an encoder would accept a certain media_format, but
this does not yet work on Haiku and thus the format is available
even when trying to render later will fail.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38978 a95241bf-73f2-0310-859d-f6bbb57e9c96
be enabled in config.h. That's the case now, but I did this change
later and didn't realize it would fix MPEG4 encoding.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38976 a95241bf-73f2-0310-859d-f6bbb57e9c96
and quality in audio and video AVCodecContext setup code
paths.
* Disabled some debug output
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38975 a95241bf-73f2-0310-859d-f6bbb57e9c96
to the beginning of the stream by bytes, if all else
fails.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38853 a95241bf-73f2-0310-859d-f6bbb57e9c96
ByteIOContext. libavformat may reallocate it
on demand, we need to use the matching allocation
methods.
* Init the ByteIOContext with the proper "write flag".
This solves a busy loop when writing the trailer of
MKV files, since the first buffer was initially skipped
and the MKV muxer can not seek back in the stream
where it wants.
* Get rid of the fCalculatePTS member, and calculate
PTS of audio packets as well. I don't remember why
I prevented that, however VLC complains about audio
packets having wrong PTS (with or without this change)
Our own MediaPlayer plays videos generated by (a modified)
Clockwerk at least once, but seeking subsequently fails.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38851 a95241bf-73f2-0310-859d-f6bbb57e9c96
and there is also ituh263enc.c, of which I am unsure how it
fits into the picture. However the mpeg4 encoder is using a "RL"
table of the h263 encoder, and it appears it was forgotten to
make sure it's initialized when splitting these files. Should
be upstreamed. Fixes a crash when trying to use the MPEG4 encoder.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38847 a95241bf-73f2-0310-859d-f6bbb57e9c96
Some test clips with sub-title tracks would hang MediaPlayer
without this fix here.
* Optimize FindKeyFrame() and Seek(), check the range between
last requested/reported frame and bail out early with the
same result. Seems to fix MediaPlayer starting to drop frames
when it got caught up in a keyframe finding party...
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38808 a95241bf-73f2-0310-859d-f6bbb57e9c96
in many situations, FindKeyframe() was unable to reliably
predict what frame Seek() would be able to seek to.
* Refactored a new base class StreamBase from the old
StreamCookie, renamed StreamCookie to just Stream.
* In FindKeyframe(), Stream will create a "ghost" StreamBase
instance. That one will be used to actually seek in the
stream without modifying the AVFormatContext of the real
Stream. From that we can tell what position we can /really/
seek to. For AVIs mostly, it is important to still use
av_index_search_timestamp(), since for many AVIs I tested,
reading the next packet after seeking did not produce a
timestamp, however the index entry contained just the correct
one. If the next packet does contain a PTS, it will still
override the index timestamp, though.
* Contrary to my previous belief, there was still a locking
problem with how MediaPlayer used the BMediaTracks. The video
decoding thread and the playback manager both used
FindKeyframe() without holding the same lock. We support this
now by using one BLocker per Stream. (The source BDataIO is
still protected by another single lock.) With the new ghost
stream stuff, the locking problem became much more of a problem,
previously the FindKeyframe() had a much rarer race condition
which would only trip when the decoding thread would cause new
index entries to be inserted into the index.
* Use the same ByteIOContext buffer size that avformat would be
using if it initialized the ByteIOContext through other API.
* Don't leak the probe buffer in case of error.
* Don't leak the ByteIOContext buffer in the end.
* Do not discard other stream packets anymore, this makes the
ASF demuxer happy and ASF files can now be seeked as well as
with ffplay itself.
With these changes, all my MPEG test streams work. Some could be seeked
before, but would show bad artifacts. Some streams would completely loose
video after seeking once. My MPEG2 test stream works much better now,
although audio is slightly out of sync, unfortunately. All my test AVIs
work as good as before, MP4 and MKV still work perfectly. The single
test ASF I got is now perfectly seekable.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38807 a95241bf-73f2-0310-859d-f6bbb57e9c96
the new video decoding function. This just avoids a warning
generated from the libavcodec sources. The function used before
did the exact same thing...
* Maintain fStartTime correctly in _DecodeVideo(). Don't overwrite
it with a calculated starttime in Decode(). This will allow drift
to bubble up to the higher layers.
* Do not use the previously required hack to close and reopen the
AVCodec after seeking. avcodec_flush_buffers() seems to work
fine now, and for certain stream types (MPEG1, MPEG2 video for
example) the keyframe is correctly used after seeking.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38806 a95241bf-73f2-0310-859d-f6bbb57e9c96
FFmpeg. It's a bit sad, but this obsoletes pretty much all other decoder
and reader plugins. Some of them were built on external libraries as well
(AC3 (not part of default image anyway, since it's GPL), APE, MusePack),
so it's not really a big difference to using FFmpeg as external library.
The format matching is greatly simplified by using B_MISC_FORMAT_FAMILY
for everything but raw audio, and the actual FFmpeg CodecID as codec tag.
The downside of this is that the AVFormatReader can no longer be used with
other decoder plugins, but it would be easy to add special cases for native
decoders we wish to support. Obviously the out of the box support for file
formats and decoders has greatly increased with this change, so there has
to be a pretty good reason now for writing a "native" decoder or reader.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38786 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Enabled SWF demuxer in FFmpeg (though the one test file
I have is compressed, which is not supported by the FFmpeg
SWF demuxer...).
* Removed testing TODOs for demuxers that I tested.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38761 a95241bf-73f2-0310-859d-f6bbb57e9c96
itself generates.
* For GCC2, FFmpeg uses -fPIC instead of -DPIC. Also disable SSE for
GCC2, since that fixes a crash in the SSE version of clear_block().
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38747 a95241bf-73f2-0310-859d-f6bbb57e9c96
speed up VLC table (whatever it is) initialization. Somehow the code
does not work on GCC2. Use the old code we had previously in SVN
minus a flag (INIT_VLC_USE_STATIC) which is no longer used in FFmpeg.
This fixes the GCC2 build of FFmpeg and closes Ã#6643, #6582 and #656.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38746 a95241bf-73f2-0310-859d-f6bbb57e9c96
different config.h, of which I am adapting the differences in this
commit. Does not change a thing for the broken GCC2 build of FFmpeg.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38745 a95241bf-73f2-0310-859d-f6bbb57e9c96
mpgs, the video does not recover, though.
* Remember the last reported keyframe information, so we
avoid rounding artifacts. Not as effective, since we
cannot use stream time-base for seeking, but have to use
it for finding the keyframes.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38706 a95241bf-73f2-0310-859d-f6bbb57e9c96
SeekedTo(), since it's only informative to decoders. They
can't modify the seeked frame/time. This also mirrors what
all existing decoders were doing in Seek(). BMediaTrack
is simplified accordingly (resolved two TODOs).
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38705 a95241bf-73f2-0310-859d-f6bbb57e9c96
if the input format requires it. Practically,
it does not work.
* Implement the new meta-data API.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38686 a95241bf-73f2-0310-859d-f6bbb57e9c96
the stream is building the index on the fly once.
This allows to seek back to earlier positions, since
then the index will contain entries for later in the
stream and the logic to detect auto-generated indices
was broken.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38678 a95241bf-73f2-0310-859d-f6bbb57e9c96
muxer by the "default" stream. When I previously tried this,
I mistakenly remembered AV_TIME_BASE to be 1000, but it's
1000000, the same as the native bigtime_t time representation.
Luckily, we can still set all other streams (including the
"default" stream) to be discarded when obtaining chunks
packets.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38677 a95241bf-73f2-0310-859d-f6bbb57e9c96
it can be any of these values, hope the priority is right.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38659 a95241bf-73f2-0310-859d-f6bbb57e9c96
frame rate (perhaps it changed in 0.6?). This fixes
playback of several MP4 clips I have for testing.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38658 a95241bf-73f2-0310-859d-f6bbb57e9c96
the counters, so it's not the average for the entire
decoding time, but for the last five frames. This gives
a more accurate picture of what's going on.
* Added NOTE about possibly removing the SWS version of the
colorspace conversion code unless it's used for otherwise
unsupported conversions. David's code is about 40% faster
in my tests (nice job!).
* Free the sws context in NegotiateVideoFormat, if necessary.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38651 a95241bf-73f2-0310-859d-f6bbb57e9c96
of support for these codecs and demuxer in the FFmpeg plugin. The
ogg test streams I downloaded play fine now. For example, Big Buck Bunny
would play without video before and ogg files natively encoded with the
FFmpeg plugin wouldn't play at all. Since the removed plugins were not
maintained and were based on external libs themselves, I didn't see the
point in keeping them.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38646 a95241bf-73f2-0310-859d-f6bbb57e9c96
* Enabled ogg muxer in MuxerTable.
ogg/vorbis creation successfully tested with MediaConverter.
ogg/theora needs more testing, it seems to work, but I need
to switch from the other vorbis/theora/ogg plugins to the
FFmpeg built-in support, otherwise the current theora stream
is not supported by the old plugin.
git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@38641 a95241bf-73f2-0310-859d-f6bbb57e9c96
