This is mathematically equivalent to what we were doing before, but
gives subtly different results due to rounding (rows first vs columns
first). Doing it this way makes our film grain database generation match
reference implementation and now produces bit-exact outputs in my
testing.
Rename the transposed variables to be a bit less confusing.
When linking the main tools, the object files to link are set up
via the variable OBJS-<name>, but for the tools, we've only
used the target's list of dependencies.
In most cases, this has been fine, but it has caused specifying
the libraries to link in a duplicate fashion; the linking command
has looked like this:
$CC -Llibavutil ... tools/tool.o libavutil/libavutil.a -lavutil
Normally, the libraries to link are handled with "-Llibavutil -lavutil";
when linking the main fftools, this is how they are linked.
In the case of the binaries under the "tools" directory (within the
make variable TOOLS), we've passed the full set of dependencies
to the linker, via $^, which does contain the names of the
dependency libraries as well.
When libraries are built as regular static libraries, or shared
unix libraries, this has all worked fine. When libraries are
built as DLLs for Windows, though, the norm is not to pass the
actual DLL to the linker, but an import library.
Mingw tools generally can handle linking directly against a DLL
as well, but MSVC tools don't support that, and error out with
a very cryptic error message:
libavdevice\avdevice.dll : fatal error LNK1107: invalid or corrupt file: cannot read at 0x2D8
By omitting these parts of the dependencies, linking of these tool
executables succeed in MSVC builds with shared libraries enabled.
Signed-off-by: Martin Storsjö <martin@martin.st>
This is slower than the Zbb version on real hardware due to register
strides. Proper support for vector byte-swap requires the Zvbb
extension, but it's much too early for me to worry about it.
Added in 80e9e63c94 for reasons
unknown to me.
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Put it into an encoder-specific context with a SnowContext
at its front. This also avoids having to include mpegvideo.h
in snow.c and snowdec.c.
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
We have to write an explicit BlockDuration element (and use
a BlockGroup instead of a SimpleBlock) in case the Track
has a DefaultDuration that is inconsistent with the duration
of the packet.
The matroska-h264-remux test uses a file with coded fields
where the duration of a Block is the duration of a field,
not of a frame, therefore this patch writes said BlockDuration
elements.
(When using a BlockGroup, one has to add ReferenceBlock elements
to distinguish keyframes from non-keyframes. Unfortunately,
the AV1 codec mapping [1] requires us to reference all references
and to really use the real references, which requires a lot of
effort for basically no gain. When BlockGroups are used with AV1,
the created files are most likely invalid, both before and after
this patch, but this patch makes this more likely to happen.)
[1]: https://github.com/ietf-wg-cellar/matroska-specification/blob/master/codec/av1.md
Reviewed-by: James Almer <jamrial@gmail.com>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
They were replaced by TX from libavutil; the tremendous work
to get to this point (both creating TX as well as porting
the users of the components removed in this commit) was
completely performed by Lynne alone.
Removing the subsystems from configure may break some command lines,
because the --disable-fft etc. options are no longer recognized.
Co-authored-by: Lynne <dev@lynne.ee>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
It more directly shows that ff_flac_decode_frame_header() does not
modify the AVCodecContext given to it at all; and it would not be
allowed to do so, given that it is used by the parser when it is
still unknown whether said frame header is even valid.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
In particular the encoder used only a small part of the context:
The new encoder context is only 128B here. It used to be 32992.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
When using frame threading, the FFV1 decoder's update_thread_context()
function copies the whole context and afterwards restores some allocated
fields with backups made earlier. Among these fields are the
ThreadFrames and the source context's ThreadFrames can change
concurrently without any synchronization, leading to data races which
are undefined behaviour even if they don't lead to problems in
practice (as the destination's own ThreadFrames are restored directly
thereafter).
Fix this by only copying the actually needed fields.
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
After the AVFrame has been wrapped into a buffer,
it is owned by the buffer and must not be freed manually
any more. Yet this happens on subsequent errors.
This bug was introduced in 6ca43a9675.
Reviewed-by: Timo Rothenpieler <timo@rothenpieler.org>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Suggested-by: Tomas Härdin <git@haerdin.se>
Fixes: 51896/clusterfuzz-testcase-minimized-ffmpeg_dem_MXF_fuzzer-5130394286817280
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The AVDCT API used by this filter does in no way depend
upon the FFT subsystem.
Reviewed-by: Lynne <dev@lynne.ee>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
