Move a number of files and functions which are only used by fixed point builds from silk/ to silk/fixed/.

I attempted to not break the msvc projects, but I can't test them.

Koen Vos authored 13 years ago and Jean-Marc Valin committed 13 years ago

- compile warning in opus_decoder.c
- decoder state reduced by ~3 kB (by merging buffers, as Tim suggested)
- some minor decoder optimizations (only the PLC is non-bit exact, so should be ok)

We check the DC gain before doing the full RC calculations

Koen Vos authored 13 years ago and Jean-Marc Valin committed 13 years ago

- increases the max pitch lag by 1 (the thing Tim pointed out).  this brings the decoder in sync with the old one
- avoids that the first stereo frame is collapsed to mono

SILK now reports an "effective width" that takes into account side prediction.

We now make sure that side is given >0 bits when it isn't disabled.

1) averaging the output of the left and right resampling states when
   switching to mono
2) averaging the the delay buffers from left and right when switching

Adds SILK delay compensation that depends on encode and decode sampling
rate, as well as SILK internal coding rate. This ensures that the SILK
part of Opus is always in sync with the CELT part no matter what the
sampling rates are. It also increases the resampling delay to 1.15 ms
(was previously 0.48 ms).

Koen Vos authored 13 years ago and Jean-Marc Valin committed 13 years ago

Simplifies mono/stereo switching in SILK
Fixes a quantization mismatch between encoder and decoder
Constrains the pitch lags in the same way in the encoder and decoder

Koen Vos authored 13 years ago and Jean-Marc Valin committed 13 years ago

Fixes a mono-stereo switching glitch and fixes a side prediction
bug where coding side would result in worse panned-mono behaviour

Gregory Maxwell authored 13 years ago and Jean-Marc Valin committed 13 years ago

There were assert failures for test cases with mode switching and LBRR enabled.
(some similar problems remain)

Tim identified that the assert was falsely firing while encoding LBRR frames but
the flag doesn't matter while encoding the bitstream it only actually matters in
silk/process_NLSFs.c.

More assert updates, make the initialization in silk_warped_autocorrelation_FLP more explicit to make tools happy, and allow resetting to auto with OPUS_SET_FORCE_CHANNELS.

Gregory Maxwell authored 13 years ago and Jean-Marc Valin committed 13 years ago

Add the noreturn attribute on the assert functions to aid static analysis, improve test_repacketizer error handling, and silence 19 clang static analysis errors with additional assertions.

Gets rid of some more glitches caused by the decoder resampler

Delaying stereo->mono switching decisions so that SILK can do a smooth
downmix. Also, wrote proper float/fixed code for the hybrid variable
stereo collapse, including a smooth downmix for stereo<->mono switching

Jean-Marc Valin authored 13 years ago and Jean-Marc Valin committed 13 years ago

LPC_inverse_pred_gain_QA() was failing to detect unstable filters
because of insufficient numerical precision. Precision was increased
to Q24 for LPC coefs and the reciprocal now uses all the precision
available (variable shift right from the start).

(Fs_kHz is always 8/12/16 in Opus and also enforced by the assert on line 112)

There's still a range coder mismatch on the first frame when using FEC.

Timothy B. Terriberry authored 13 years ago and Jean-Marc Valin committed 13 years ago

The SILK bitstream allowed coding 0 pulses in a shell block, but a
 non-zero number of LSb's, meaning some excitation coefficients
 could be non-zero, but would not have a corresponding sign.
To fix this without breaking already-encoded bitstreams, this patch
 adds a set of sign PDFs for the 0 pulses case.

This is occasionally more efficient than the normal encoding if
 there are a large number of coefficients with positive signs,
 since these cost more than 1 bit when using the > 0 pulse PDFs.
It only saves 0.33 bits per second (on average: it does better at
 high rates), but that's probably enough to justify the two
 redundant ways of coding things (and it's too late now to remove
 the second one entirely, anyway).
This patch does not include the encoder modifications required to
 check if this coding method is more efficient and switch to it.

We no longer reset the decoder predictors when switching only the
frame size and not the sampling rate. The encoder already doesn't reset.

For these transitions, we now start the left and right resamplers
from the same state.

Result should be bit-identical on most machines/compilers, minus the undefined
behaviour

Koen Vos authored 13 years ago and Jean-Marc Valin committed 13 years ago

Also some MSVC build fixes