Old one was written with the assumption only even inputs would be given.
This very messy replacement supports even and odd inputs, and supports
AVX2 for extra speed. The buffers given are usually quite big (4k samples),
so the speedup is worth it.
The new SSE version is still faster than the old inline asm version by 33%.
Also checkasm is provided to make sure this monstrosity works.
This fixes some FATE tests.
Clang's static analyzer complains that leaving the variable
uninitialized could lead to a code path where the uninitialized value is
written to at the end of this function.
This patch simply zero-initializes that variable to avoid that.
Signed-off-by: Will Cassella <cassew@google.com>
Signed-off-by: James Almer <jamrial@gmail.com>
The aim of this test is to show the interleavement
of the file generated in the first pass; so make the
interleavement queue in the framecrc muxer in the second
pass as small as possible so that the framecrc muxer does not
fix wrong interleavement of the input file behind our backs.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
enc_dec is designed for raw input and output and computes
the PSNR between these two. The input of the shortest-sub
test is the idx file of a vobsub sub+idx combination
and the output is the output of framecrc of said vobsub
subtitle muxed into Matroska together with a synthesized
video. Calculating the PSNR between these two files makes
no sense, therefore switch to a transcode test, where
the ref file file contains the output of framecrc directly,
making the interleavement better visible in the ref file
at the cost of a larger ref file (>400 lines).
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
The MXF demuxer does not currently set AVStream::avg_frame_rate and ::r_frame_rate
when J2K essence is wrapped according to SMPTE ST 422.
Reviewed-by: Tomas Härdin <tjoppen@acc.umu.se>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Also covers writing mastering display metadata.
Reviewed-by: Tomas Härdin <tjoppen@acc.umu.se>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Basically reverts af15c17daa.
Flipping a picture by modifying the pointers is so common
that even users of direct rendering should take it into account.
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Up until now, libswscale/output.c used a macro to write
an output pixel which involved a call to av_pix_fmt_desc_get()
to find out whether the input pixel format is BE or LE
despite this being known at compile-time (there are templates
per pixfmt). Even worse, these calls are made in a loop,
so that e.g. there are eight calls to av_pix_fmt_desc_get()
for every pixel processed in yuv2rgba64_X_c_template()
for 64bit RGB formats.
This commit modifies these macros to ensure that isBE()
is evaluated at compile-time. This saved 41184B of .text
for me (GCC 11.2, -O3). Of course, it also improved performance.
E.g. ffmpeg_g -f lavfi -i testsrc2,format=yuva420p -pix_fmt rgba64le \
-threads 1 -t 1:00 -f null - (which uses yuv2rgba64le_X_c,
which is an invocation of yuv2rgba64_X_c_template() mentioned above),
performance improved from 95589 to 41387 decicycles for one call
to yuv2packedX; for the be variant the numbers went down from
76087 to 43024 decicycles.
Reviewed-by: Anton Khirnov <anton@khirnov.net>
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Up until now, libswscale/input.c used a macro to read
an input pixel which involved a call to av_pix_fmt_desc_get()
to find out whether the input pixel format is BE or LE
despite this being known at compile-time (there are templates
per pixfmt). Even worse, these calls are made in a loop,
so that e.g. there are six calls to av_pix_fmt_desc_get()
for every pair of UV pixel processed in
rgb64ToUV_half_c_template().
This commit modifies these macros to ensure that isBE()
is evaluated at compile-time. This saved 9743B of .text
for me (GCC 11.2, -O3). For a simple RGB64LE->YUV420P
transformation like
ffmpeg -f lavfi -i haldclutsrc,format=rgba64le -pix_fmt yuv420p \
-threads 1 -t 1:00 -f null -
the amount of decicycles spent in rgb64LEToUV_half_c
(which is created via the template mentioned above)
decreases from 19751 to 5341; for RGBA64BE the number
went down from 11945 to 5393. For shared builds (where
the call to av_pix_fmt_desc_get() is indirect) the old numbers
are 15230 for RGBA64BE and 27502 for RGBA64LE, whereas
the numbers with this patch are indistinguishable from
the numbers from a static build.
Also make the macros that are touched conform to the
usual convention of using uppercase names while just at it.
Reviewed-by: Anton Khirnov <anton@khirnov.net>
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Up until now, using NULL as key in av_dict_get() on a non-empty
AVDictionary would crash; using NULL as key in av_dict_set()
would also crash for a non-empty AVDictionary unless AV_DICT_MULTIKEY
was set; in case the dictionary was initially empty or AV_DICT_MULTIKEY
was set, it was even possible for av_dict_set() to succeed when
adding a NULL key, namely when one uses a value != NULL and
the AV_DICT_DONT_STRDUP_VAL flag. Using av_dict_get() on such
an AVDictionary will usually lead to crashes, though.
Fix this by actually checking for key in both functions; error out
if they are NULL.
While just at it, also stop relying on av_strdup(NULL) to return NULL
in av_dict_set().
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
The compiler cannot infer that the two float vectors do not alias,
causing unnecessary extra loads and serialisation. This patch caches
the two input values in local variables so that compiler can optimise
individual loop iterations.
While this probably never overflows, we are better safe than sorry.
The callback prototype should probably also use ptrdiff_t or size_t,
but I diggress (this would affect the DSP callback prototype).
Do this by setting AVCodecInternal.pad_samples.
This prevents reading into the frame's padding and writing
into the packet's padding.
This actually happened in our FATE tests (where the number of samples
is 2 mod 4), which therefore needed to be updated.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Some audio codecs work with atomic units that decode to a fixed
number of audio samples with this number being so small that it is
common to put multiple of these atoms into one packet. In these
cases it makes no sense to pad the last frame to the big frame_size,
so allow encoders to set the number of samples that they want
the last frame to be padded to instead.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
In particular, check that there is only one small last frame
in case the encoder has the AV_CODEC_CAP_SMALL_LAST_FRAME set.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
The APTX (HD) decoder decodes blocks of four (six) bytes to four
output samples. It makes no sense to handle incomplete blocks:
They would just lead to synchronization errors, in which case
the complete frame is discarded. So only handle complete blocks.
This also avoids reading from the packet's padding and writing
into the frame's padding.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Just because we try to put multiple units of block_align bytes
(the atomic units for APTX and APTX HD) into one packet
does not mean that packets with fewer units than the
one we wanted are corrupt; only those packets that are not
a multiple of block_align are.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This field was misunderstood: It gives the number of samples
in a packet, not the number of bytes. Its usage was wrong for APTX HD.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Currently the APTX (HD) codecs set frame_size if unset and check
whether it is divisible by block_size (corresponding to block_align
as used by other codecs). But this is based upon a misunderstanding
of the API: frame_size is not in bytes, but in samples.
Said value is also not intended to be set by the user at all,
but set by encoders and (possibly) decoders if the number of channels
in a frame is constant. The latter condition is not fulfilled here,
so only set it for encoders. Given that the encoder can handle any
number of samples as long as it is divisible by four and given that
it worked to set a custom frame size before, the encoders accept
any multiple of four; otherwise the value is set to the value
that it already had for APTX: 1024 samples (per channel).
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
APTX decodes four bytes of input to four stereo samples; APTX HD
does the same with six bytes of input. So it can be easily supported
in av_get_audio_frame_duration().
This fixes invalid durations and (derived) timestamps of demuxed
APTX HD packets and therefore fixed the timestamp in the aptx-hd
FATE test.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
We have de- and encoders for APTX and APTX HD, yet not FATE tests.
This commit therefore adds a transcoding test to utilize them.
Furthermore, during creating these tests it turned out that
the duration is set incorrectly for APTX HD. This will be fixed
in a future commit.
(Thanks to Andriy Gelman for finding an issue in an earlier version
that used a 192kHz input sample which does not work reliably accross
platforms.)
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This block was scheduled for removal, which means that no validation would have
taken place after the old API was removed.
It was algo going to mistakenly remove an unrelated bits_per_coded_sample
check.
Signed-off-by: James Almer <jamrial@gmail.com>
The opaque parameter for the callback is set in videotoolbox_start(),
called when the hwaccel is initialized. When frame threading is used,
avctx will be the context corresponding to the frame thread currently
doing the decoding. Using this same codec context in all subsequent
invocations of the decoder callback (even those triggered by a different
frame thread) is unsafe, and broken after
cc867f2c09, since each frame thread now
cleans up its hwaccel state after decoding each frame.
Fix this by passing hwaccel_priv_data as the opaque parameter, which
exists in a single instance forwarded between all frame threads.
The only other use of AVCodecContext in the decoder output callback is
as a logging context. For this purpose, store a logging context in
hwaccel_priv_data.
This is no longer used since 4608996772.
It also has no implementations other than the plain C one.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Since introducing the various packed formats used by VAAPI (and p012),
we've noticed that there's actually a gap in how
av_find_best_pix_fmt_of_2 works. It doesn't actually assign any value
to having the same bit depth as the source format, when comparing
against formats with a higher bit depth. This usually doesn't matter,
because av_get_padded_bits_per_pixel() will account for it.
However, as many of these formats use padding internally, we find that
av_get_padded_bits_per_pixel() actually returns the same value for the
10 bit, 12 bit, 16 bit flavours, etc. In these tied situations, we end
up just picking the first of the two provided formats, even if the
second one should be preferred because it matches the actual bit depth.
This bug already existed if you tried to compare yuv420p10 against p016
and p010, for example, but it simply hadn't come up before so we never
noticed.
But now, we actually got a situation in the VAAPI VP9 decoder where it
offers both p010 and p012 because Profile 3 could be either depth and
ends up picking p012 for 10 bit content due to the ordering of the
testing.
In addition, in the process of testing the fix, I realised we have the
same gap when it comes to chroma subsampling - we do not favour a
format that has exactly the same subsampling vs one with less
subsampling when all else is equal.
To fix this, I'm introducing a small score penalty if the bit depth or
subsampling doesn't exactly match the source format. This will break
the tie in favour of the format with the exact match, but not offset
any of the other scoring penalties we already have.
I have added a set of tests around these formats which will fail
without this fix.
Lynne
Will Cassella
Andreas Rheinhardt
Pierre-Anthony Lemieux
James Almer
Rémi Denis-Courmont
Anton Khirnov
Paul B Mahol
Philip Langdale