Nils Wallménius authored 12 years ago and Jean-Marc Valin committed 12 years ago

Gets rid of one stack buffer.

Signed-off-by: Jean-Marc Valin <jmvalin@jmvalin.ca>

Adds more sizes to test_unit_mdct.c

Nils Wallménius authored 12 years ago and Jean-Marc Valin committed 12 years ago

Place output of imdct post-rotate and deshuffle in out buffer to avoid copying and simplify the code.

Signed-off-by: Jean-Marc Valin <jmvalin@jmvalin.ca>

Nils Wallménius authored 12 years ago and Jean-Marc Valin committed 12 years ago

Signed-off-by: Jean-Marc Valin <jmvalin@jmvalin.ca>

Was causing problems with the PLC

Previous code was so broken that it's`not worth listing the changes.
Aside from quality, one obvious improvement is a 3.75k reduction in stack size

We always define OPUS_BUILD, so this code has no effect.
Likewise there is no reason for it to be conditional.
According to Jean-Marc it's left over from when we had
an alternate CELT_BUILD define.

This also adds some extra casts to shut up compiler warnings
 reported on MSVC 6 where there is implicit truncation for the
 arguments of bitexact_cos().
Lacking access to CLZ/BSR will make the code a fair bit slower but
 that is better than failing to compile.

Instead of using the average energy over the frame, we use a half-way
point between the average energy and the max energy.

Nils Wallménius authored 12 years ago and Jean-Marc Valin committed 12 years ago

Remove two now redundant zeroing loops

Signed-off-by: Jean-Marc Valin <jmvalin@jmvalin.ca>

shorter dependencies

Also, moves the dynalloc code to a separate function

Commits ae0e2ca8 and 6906210c split encode and decode
support into separate files to help reduce the footprint
of decode-only builds.

This commit adds the new source files to the MSVC
project files, fixing that build.

max_decay was being computed based on the small number of bits allocated
to CELT. It's now set to the max value (16) when in hybrid mode.

Also forces SPREAD_NORMAL in hybrid mode

This prevents issues on speech where tf would undo what transient_analysis()
found

No need to allocate for the low-band

Reduce size of "norm" because we never need it for the last band

Delays stack allocations as much as possible, and makes some of these
allocations conditional.

Doing the overlap coying in the same OPUS_MOVE() as the rest of the synthesis
memory. Also, there was a missing OPUS_MOVE() in the PLC for the hybrid/CNG
case.

Nils Wallménius authored 12 years ago and Jean-Marc Valin committed 12 years ago

Use the output buffer directly for imdct output, avoiding use of an
intermediary buffer and copying from it. This works because the output
buffer is offset to always be continous with the overlap buffer in the
decoder. Adjust encoder resynth code to have the same buffer layout.

Signed-off-by: Jean-Marc Valin <jmvalin@jmvalin.ca>

This one is explicitly based on a simple temporal masking model