Each line is padded by the format, and errors are now reported and
stop the decoding. Around 5% speedup.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Around 3x speedup with 4 threads. Maybe more mb lines should be
batched per thread, but that's good enough for a first try.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Indicates a YCbCr->RGB transform at the block level. Although nothing
explicitly states it, this would assume the actual content is planar
RGB.
Currently unsupported, but the one sequence I found using it flagged
every mb that way, actually meaning the content was YCbCr, and thus
best left to the output format to decide what to do of it.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
It currently only applies to CID 1260, but this flag is dependent on
a higher-level flag located in the header.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
This bit is 1 in some samples, and seems to coincide with interlaced
mbs and CID1260. 2008 specs do not know about it, and maintain qscale
is 11 bits. This looks oversized, but may help larger bitdepths.
Currently, it leads to an obviously incorrect qscale value, meaning
its syntax is shifted by 1. However, reading 11 bits also leads to
obviously incorrect decoding: qscale seems to be 10 bits.
However, as most profiles still have 11bits qscale, the feature is
restricted to the CID1260 profile.
The encoder writes 12 bits of syntax, last and first bits always 0,
which is now somewhat inconsistent with the decoder, but ends up with
the same effect (progressive + reserved bit).
Partially fixes ticket #4876.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The sample position is made weird and non-nominal to force catching
such issues as default values or specialized operations hiding
issues in corner cases.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
For yuv420p, the chroma position is unilaterally overriden, even
if ffmpeg's command-line explicitly set it. To fix this, override
only if the value is the default one.
Regression since 1515bfb3.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
This is the remaining error, the output on the SPX samples,
respectively csi_miami_stereo_128_spx.eac3 and
csi_miami_5.1_256_spx.eac3, goes from:
stddev: 8.71 PSNR: 77.52 MAXDIFF: 235
stddev:24270.51 PSNR: 22.17 MAXDIFF:47166
to:
stddev: 0.12 PSNR:114.12 MAXDIFF: 1
stddev: 0.12 PSNR:114.73 MAXDIFF: 1
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
When the decoder is deactivated, the x86-optimized versions are
not compiled, resulting in a link error.
The C version is unaffected, as it is part of the idctdsp
subsystem.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The main difference consists in renaming properly labels, and
letting yasm select the gprs for skipping 1D transforms.
Previous-version-reviewed-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
When calling ff_alloc_entries, a number of entries are created.
They are never freed, as running fate with slice threading and
several frames on e.g. fate-hevc-conformance-ENTP_A_Qualcomm_1
would show.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Currently, a part of the loop has to be handled in plain C, because of DSP
function requirements of having batches of 16 elements to process.
Instead, pad the buffer with 0 where the DSP functions are overreading.
Signed-off-by: James Almer <jamrial@gmail.com>
The macro hell sometimes make it difficult to trace the source of
an error, so it is easier to analyze the preprocessed output.
This patch makes this automatical by specifying DBG=1 on the
command line: a file ffmpeg/dir/file.asm gets preprocessed to
builddir/dir/file.dbg.asm, which is then compiled.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
SSE2 instructions that are XMM-implementations of pre-existing MMX/MMX2
instructions did not issue warnings when used in SSE functions. Handle
it by also checking the register type when such instructions are used.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The 3*stride value stored in r3src can be loaded much later,
so use r3src instead of a dedicated gpr when possible.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The second stride is always the internal buffer one, MAX_PB_SIZE (times 2 to
get the value in bytes).
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The width parameter is now completely at the back, and actually
never used. This helps understanding the actual parameter list.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Their intent was to make the DSP work with wmalossless pro.
The later was fixed to work with the DSP.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The epel_hv functions were still relying on only epel_hv 8-wide
being the maximum width instanciated.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The buffer pointers would be otherwise overwritten, causing a
leak on e.g. PERSIST_RPARAM_A_RExt_Sony_1.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Use edge emu buffers
And enable the code unconditionally
Speed difference without USE_SAO_SMALL_BUFFER and with the new code:
Decicycles: 26772->26220 (BO32), 83803->80942 (BO64)
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This mimicks what is done for the other instruction sets.
Tested-by: James Almer <jamrial@gmail.com>
Tested-by: Mickaël Raulet <mraulet@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
For band filter, source and destination are aligned (except for 16x16 ctbs),
and otherwise, they are most often aligned. Overall, the total width is also
too small for amortizing memcpy.
Timings (using an intrinsic version of edge filters):
B/32 B/64 E/32 E/64
Before: 32045 93952 38925 126896
After: 26772 83803 33942 117182
Since the decoder has been fixed to output something similar to
other implementations, the encoder weight table needed this fix.
Reference PSNR values on a progressive sequence (from tiny_psnr)
from a chained encoding and decoding:
Full progressive: stddev: 0.74 PSNR: 50.69 MAXDIFF: 19
Before: stddev: 1.43 PSNR: 44.97 MAXDIFF: 17
This patch: stddev: 0.76 PSNR: 50.44 MAXDIFF: 14
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Before
2843 decicycles in ff_sbr_autocorrelate_sse3, 262086 runs, 58 skips
After
2693 decicycles in ff_sbr_autocorrelate_sse3, 262117 runs, 27 skips
Signed-off-by: James Almer <jamrial@gmail.com>
While the shortened format specifiers are valid C99 ones, they are
not supported by e.g. some Visual Studio runtimes (and thus mingw).
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The "faulty" samples actually sound fine when ignoring this issue.
For ticket #3886, more samples are decoded.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
For test images manually generated to contain only up prediction,
timing results:
8380x3032 255x185
before: 138635 1992
after: 139232 1996
Actually jumping to the proper version depending on the alignment:
8380x3032: 138767
A 0.5% speed improvement for gigantic images is not worth the code
duplication.
Fixes ticket #4148
Signed-off-by: Christophe Gisquet <christophe.gisquet@gmail.com>
Tested-by: Benoit Fouet <benoit.fouet@free.fr>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The format is now:
-bsf:X filter1[=opt1=str1/opt2=str2],filter2
ie the parameters are appended after the filter name using '='. As ','
has been reserved already for the list of filters, '/' is just an
example of token separation for now, but that could become part of the
API to avoid each bsf using its own tokenization.
The proper solution would be using AVOption, but this is overkill for now.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The encoder has its own tables and does not access the idct_factor
member of the DVVideoContext structure.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The coefficients must be in the appropriate zigzag scan order.
Also fix their values at the same time, as they were pretty wrong.
Fixes ticket #2970.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
It is derived from the actual equations of the specs. In
particular, it is closer to the inverse of what the encoder uses.
fate tests accordingly updated.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The bytes per row is a better indication of it.
Helps resolving ticket #3874 by fixing ffmpeg's encoder and transforming
the issue in a issue with non-compliant decoders. ffmpeg's one is ok,
but unfortunately, many others aren't handling correctly unusual chroma
samplings.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The extra bits have already been masked, so this was not doing anything.
Noticed-by: Justin Ruggles
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The user may not know how to select the profile, nor what he needs, in
particular to encode alpha.
Therefore, use an automatic selection as default, and warn when the
manually selected profile may cause issues.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
In some cases, 2 or 3 calls are performed to functions for unusual
widths. Instead, perform 2 calls for different widths to split the
workload.
The 8+16 and 4+8 widths for respectively 8 and more than 8 bits can't
be processed that way without modifications: some calls use unaligned
buffers, and having branches to handle this was resulting in no
micro-benchmark benefit.
For block_w == 12 (around 1% of the pixels of the sequence):
Before:
12758 decicycles in epel_uni, 4093 runs, 3 skips
19389 decicycles in qpel_uni, 8187 runs, 5 skips
22699 decicycles in epel_bi, 32743 runs, 25 skips
34736 decicycles in qpel_bi, 32733 runs, 35 skips
After:
11929 decicycles in epel_uni, 4096 runs, 0 skips
18131 decicycles in qpel_uni, 8184 runs, 8 skips
20065 decicycles in epel_bi, 32750 runs, 18 skips
31458 decicycles in qpel_bi, 32753 runs, 15 skips
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Not having allocated it is not a good reason to leave the object
in an undetermined state. Though a particular setting like the
AV_EF_* flags could be useful to control that behaviour.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
In some cases, in particular if several blocks are needed because of
the channel layout (e.g. 2.1), the information used to write the
trailing bits terminating the sample data was not reset.
This would cause potential desync on the decoder, although decoded
samples were actually mostly fine.
Fixes ticket #3879.
It was only validating that normal data wasn't filling the buffer.
However, extra data may be written afterwards.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Currently, the encoder will try to reduce it down to 150000, but the
decoder will complain starting at 131072 (WV_MAX_SAMPLES). Therefore,
change the loop limit.
Fixes ticket #3881.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Some streams were found to have what appeared to be truncated SPS.
Their syntax seem to be valid at least until the end of the VUI, so
try that syntax if the parsing would overflow the SPS in the
conforming syntax.
Fixes ticket #3872.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
bytestream2_* will not cause buffer overflow, but in that case, this means
the allocation would be incorrect and the encoded result invalid. Therefore,
assert no overflow occurred.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
With huge sampling rates, the table derivation method does not converge fast
enough. While fixing it using e.g. Newton-Rhapson-like methods (the curve is
nicely convex) is possible, it is much simpler to reject these cases.
The value of 96000 was arbitrarily chosen as a realistic value, though
1000000 would still work and converge.
Fixes ticket #3868.
Suggested-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
