Change the size specifiers to match the actual element sizes
of the data. This makes no practical difference with strict
alignment checking disabled (the default) other than somewhat
documenting the code. With strict alignment checking on, it
avoids trapping the unaligned loads.
Signed-off-by: Mans Rullgard <mans@mansr.com>
Converting the double to float for lrintf() loses precision when
the value is not exactly representable as a single-precision float.
Apart from being inaccurate, this causes discrepancies in some
configurations due to differences in rounding.
Note that the changed timestamp in the vc1-ism test is a bogus,
made-up value.
Signed-off-by: Mans Rullgard <mans@mansr.com>
Code mostly inspired by vp8's MC, however:
- its MMX2 horizontal filter is worse because it can't take advantage of
the coefficient redundancy
- that same coefficient redundancy allows better code for non-SSSE3 versions
Benchmark (rounded to tens of unit):
V8x8 H8x8 2D8x8 V16x16 H16x16 2D16x16
C 445 358 985 1785 1559 3280
MMX* 219 271 478 714 929 1443
SSE2 131 158 294 425 515 892
SSSE3 120 122 248 387 390 763
End result is overall around a 15% speedup for SSSE3 version (on 6 sequences);
all loop filter functions now take around 55% of decoding time, while luma MC
dsp functions are around 6%, chroma ones are 1.3% and biweight around 2.3%.
Signed-off-by: Diego Biurrun <diego@biurrun.de>
The vertically interpolating variants of these functions read
ahead one line to optimise the loop. On the last line processed,
this might be outside the buffer. Fix these invalid reads by
processing the last line outside the loop.
Signed-off-by: Mans Rullgard <mans@mansr.com>
According to the behaviour of librtmp, it is recommended to send this
message to the server after receiving the 'onBWDone' callback in order
to do bandwidth checking and improve compatibility with some servers.
Those functions are only useful inside filters. It is better to not
support user filters until the API is more stable.
This breaks audio filtering API and ABI in theory, but since it's
unusable right now this shouldn't be a problem.
There's no reason for it to be explicitly 32 bits. It's declared as a
plain int in all other places in Libav.
This breaks audio filtering API and ABI in theory, but since it's
unusable right now this shouldn't be a problem.
The additional parameters are just complicating the function interface.
Assume that a requested samples buffer will *always* have the format
specified in the requested link.
This breaks audio filtering API and ABI in theory, but since it's
unusable right now this shouldn't be a problem.
Signed-off-by: Anton Khirnov <anton@khirnov.net>
Remove AVFilterBufferRefAudioProps.size, and use nb_samples in its place
everywhere.
This is required as the size in the audio buffer may be aligned, so it
may not contain a well defined number of samples.
Also remove the useless planar parameter, which can be deduced from the
sample format.
This is technically an API and ABI break, but since the audio part of
lavfi is not usable now, this should not be a problem in practice.
Signed-off-by: Anton Khirnov <anton@khirnov.net>
This makes only tests actually using avconv depend on it.
The remaining tests already depend on what they need.
Signed-off-by: Mans Rullgard <mans@mansr.com>
Just like gcc 4.6 and later on ARM, gcc 4.8 on MIPS generates
inefficient code when a known-unaligned location is used as a
memory input operand. This applies the same fix as has been
previously done to the ARM version of the code.
Signed-off-by: Mans Rullgard <mans@mansr.com>
Michael Kostylev
Diego Biurrun
Mans Rullgard
Christophe Gisquet
Samuel Pitoiset
Sean McGovern
Justin Ruggles
Anton Khirnov
Stefano Sabatini