This prevents a call to bytestream_get_be16() using a movzwl both before
and after the ror instruction, which is obviously inefficient. Arm uses
the same trick also.
Sintel decoding goes from (avg+SD) 9.856 +/- 0.003 to 9.797 +/- 0.003 sec.
Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
This should hopefully fix roundup issue 2586.
This commit only implements it in the demuxer, not in the
protocol handler. If desired, some of the code could be
refactored to be shared by both implementations.
Signed-off-by: Martin Storsjö <martin@martin.st>
This can later be extended to support other AES bit sizes,
encryption, other crypto algorithms, reading the key from a URL, etc.
In order to use it, the key and initialization vector has to be
passed via AVOptions. Since such options can't be passed to
protocols from the command line, the protocol is currently
only for libavformat internal use.
Signed-off-by: Martin Storsjö <martin@martin.st>
Setting this should fix the FATE build failure on DragonFly BSD,
since inet_aton only is visible if __BSD_VISIBLE is set.
Alternatively, a line defining __BSD_VISIBLE=1 could be
added at the top of os_support.c. For FreeBSD, similar lines
are required in libavdevice/bktr.c and libavdevice/oss_audio.c, too.
Signed-off-by: Martin Storsjö <martin@martin.st>
The current value is masking the POSIX error code EPIPE, which has a
different semantics.
This breaks API.
Signed-off-by: Stefano Sabatini <stefano.sabatini-lala@poste.it>
Signed-off-by: Anton Khirnov <anton@khirnov.net>
AVERROR_NUMEXPECTED is used only in the image muxer and demuxer, and
has a too much specific meaning, which is better explained through a
log message. Thus it can be replaced by AVERROR(EINVAL).
This breaks API.
Signed-off-by: Stefano Sabatini <stefano.sabatini-lala@poste.it>
Signed-off-by: Anton Khirnov <anton@khirnov.net>
The new error code is better than AVERROR(ENOENT), which has a
completely different semantics ("No such file or directory").
Signed-off-by: Stefano Sabatini <stefano.sabatini-lala@poste.it>
Signed-off-by: Anton Khirnov <anton@khirnov.net>
This makes seek_test to be rebuilt when its dependencies
has changed. The changes to the dependencies didn't usually matter
in practice, but the introduction of side data in AVPacket required
a recompilation.
Signed-off-by: Martin Storsjö <martin@martin.st>
After switching this from a statically allocated array to a
dynamically allocated one in the major bump, this needs explicit
freeing.
Signed-off-by: Martin Storsjö <martin@martin.st>
instead of unconditionally at the start of the function.
This fixes a bug where a successful call to ff_thread_init() masks errors that
occur after that point in the function. It also makes future bugs like this
less likely since the error code is now set near to the point in the code
where the error is found.
The problem with url_exist() is that it tries to open a resource in
RDONLY mode. If the file is a FIFO and there is already a reading
client, the open() call will hang.
By using avio_check() with access mode of 0, the second reading
process will check if the file exists without attempting to open it,
thus avoiding the lock.
Fix issue #1663.
Signed-off-by: Anton Khirnov <anton@khirnov.net>
Make AVIO_FLAG_ access constants work as flags, and in particular fix
the behavior of functions (such as avio_check()) which expect them to
be flags rather than modes.
This breaks API.
