Uninitialized coefficients were being used to generate exponents, some
of which actually ended up in the final stream. Even though, they were
just extra exponents that are not used by any decoder, it is still
better to have consistent output for testing. This also fixes valgrind
errors.
Channel coupling is an optional AC-3 feature that increases quality by
combining high frequency information from multiple channels into a
single channel. The per-channel high frequency information is sent with
less accuracy in both the frequency and time domains. This allows more
bits to be used for lower frequencies while preserving enough
information to reconstruct the high frequencies.
This makes the AC3 encoder use the shared fixed-point MDCT rather
than its own implementation. The checksum changes are due to
different rounding in the MDCT.
Signed-off-by: Mans Rullgard <mans@mansr.com>
These expressions are equivalent since levels is always odd, and
overflow is impossible due to the constraints set by the assert().
Signed-off-by: Mans Rullgard <mans@mansr.com>
These fields are only used in quantize_mantissas() and reset
on each call, no need to store them in the main context.
Signed-off-by: Mans Rullgard <mans@mansr.com>
This increases the accuracy of coefficients, leading to improved quality.
Rescaling of the coefficients to full 25-bit accuracy is done rather than
offsetting the exponent values. This requires coefficient scaling to be done
before determining the rematrixing strategy. Also, the rematrixing strategy
calculation must use 64-bit math to prevent overflow due to the higher
precision coefficients.
Currently it is always 4, but this change will allow it to be adjusted when
bandwidth-related features are added such as channel coupling, enhanced
channel coupling, and spectral extension.
The rematrixing strategy reuse flags are not reset between frames, so they
need to be initialized for all blocks, not just block 0.
Signed-off-by: Mans Rullgard <mans@mansr.com>
This is to match the value in every (E-)AC-3 file from commercial sources.
It has a negligible effect on audio quality.
Signed-off-by: Mans Rullgard <mans@mansr.com>
It is pretty hopeless that other considerable projects will adopt
libavutil alone in other projects. Projects that need small footprint
are better off with more specialized libraries such as gnulib or rather
just copy the necessary parts that they need. With this in mind, nobody
is helped by having libavutil and libavcore split. In order to ease
maintenance inside and around FFmpeg and to reduce confusion where to
put common code, avcore's functionality is merged (back) to avutil.
Signed-off-by: Reinhard Tartler <siretart@tauware.de>
This patch changes the exponent difference threshold in the exponent
strategy decision function of the AC-3 encoder. I tested lowering in
increments of 100. From 1000 down to 500 generally increased in quality
with each step, but 400 was generally much worse.
Signed-off-by: Mans Rullgard <mans@mansr.com>
exponent strategies for a single channel to compute_exp_strategy_ch().
This allows for removal of the temporary pointer arrays.
Originally committed as revision 26356 to svn://svn.ffmpeg.org/ffmpeg/trunk
This improves the audio quality significantly for stereo source with both the
fixed-point and floating-point AC-3 encoders.
Update acodec-ac3_fixed and seek-ac3_rm test references.
Originally committed as revision 26271 to svn://svn.ffmpeg.org/ffmpeg/trunk
instead of doing it separately in 2 different functions.
This makes float AC-3 encoding approx. 3-7% faster overall.
Also, the coefficient conversion can now be easily SIMD-optimized.
Originally committed as revision 26232 to svn://svn.ffmpeg.org/ffmpeg/trunk
accessing of structs and arrays inside the loop.
Approx. 30% faster in function extract_exponents().
Originally committed as revision 26226 to svn://svn.ffmpeg.org/ffmpeg/trunk
Fixed-point AC-3 encoder renamed to ac3_fixed.
Regression test acodec-ac3 renamed to acodec-ac3_fixed.
Regression test lavf-rm changed to use ac3_fixed encoder.
Originally committed as revision 26209 to svn://svn.ffmpeg.org/ffmpeg/trunk
maximum value of 1023.
This speeds up overall encoding depending on the content and bitrate.
The most improvement is with high bitrates and/or low complexity content.
Originally committed as revision 26181 to svn://svn.ffmpeg.org/ffmpeg/trunk
instead of 64. This will change output in some cases, but it happens to not
affect the AC-3 regression tests.
Originally committed as revision 26180 to svn://svn.ffmpeg.org/ffmpeg/trunk
This gives slightly better quality in PEAQ tests.
Code 3 gives a dBpb value of 2816 = -132dB (128 psd units = -6dB), which
corresponds to 22 bits. Since the exponents have an offset applied, the
16-bit source looks like 24-bit source to the bit allocation routine.
So using dBpb code=3 is a closer match to the exponent range.
Regression test refs updated for acodec-ac3, lavf-rm, and seek-ac3_rm.
Originally committed as revision 26144 to svn://svn.ffmpeg.org/ffmpeg/trunk
This is optional for encoders, but it's a good idea and has minimal impact
on performance.
This will change the output for some files, but it happens not to affect the
regression tests.
Originally committed as revision 26083 to svn://svn.ffmpeg.org/ffmpeg/trunk
This avoids a 16-bit overflow in mdct512() due to a -32768 value in costab.
References updated for acodec-ac3, lavf-rm, and seek-ac3_rm tests.
Thanks to Måns Rullgård for finding the bug.
Originally committed as revision 26071 to svn://svn.ffmpeg.org/ffmpeg/trunk
Fixes AC-3 encoding on OpenBSD 4.8 x86_32 and hopefully other similar
configurations.
Originally committed as revision 26070 to svn://svn.ffmpeg.org/ffmpeg/trunk
This allows encoding with lower bitrates by decreasing exponent bits first,
then decreasing bandwidth if the user did not specify a specific cutoff
frequency.
Originally committed as revision 26050 to svn://svn.ffmpeg.org/ffmpeg/trunk
We can do this because exponents are the only bit allocation parameters which
change from block-to-block currently.
Approx. 57% faster in function bit_alloc().
Approx. 25% faster overall encoding.
Originally committed as revision 26040 to svn://svn.ffmpeg.org/ffmpeg/trunk
allocation for each block.
24% faster in function bit_alloc(). Approx. 10% faster overall encoding.
Originally committed as revision 26039 to svn://svn.ffmpeg.org/ffmpeg/trunk
in encode_exponents_blk_ch() by removing the inner loops. This is about 30-40%
faster for the modified sections.
Originally committed as revision 26036 to svn://svn.ffmpeg.org/ffmpeg/trunk
longer required. This gets rid of the temp buffer as well as encoded_exp in
AC3EncodeContext. It also allows for skipping the exponent grouping for
EXP_D15. 56% faster in encode_exponents_blk_ch().
Originally committed as revision 26034 to svn://svn.ffmpeg.org/ffmpeg/trunk
This reduces the memory footprint when using less than 6 channels.
Modify bit allocation to swap the 2 buffers instead of using memcpy() and use
per-block pointers for bap. This is slightly faster (0.3%) in function
cbr_bit_allocation().
Originally committed as revision 26023 to svn://svn.ffmpeg.org/ffmpeg/trunk
Avoids memcpy that was used to store last samples for next frame.
Approx. 3% faster in function deinterleave_input_samples() and reduces memory
usage by 3kB.
Originally committed as revision 26021 to svn://svn.ffmpeg.org/ffmpeg/trunk
