Since we're doing two rotations, we can invert the sign on both.
Also adding a few comments for optimizing the FFT.

Also simplifying the code

Undoes commits f7547a4e and 72513f3c

This also happens to increase the accuracy since it appears that the new
ordering is optimal (at least for 20 ms frames), whereas the previous ordering
was pessimal.

IMDCT now uses the forward FFT.

This makes us lose a bit of precision in the first steps, but our gain is more
precise because it's only rounded once. Overall, SNR is slightly improved.

Peter Korsgaard authored 11 years ago and Jean-Marc Valin committed 11 years ago

Breaks configure when /bin/sh isn't bash with:

configure: Trying to force-enable ARMv6 media instructions...
checking if assembler supports ARMv6 media instructions on ARM... yes
configure: Trying to force-enable NEON instructions...
checking if assembler supports NEON instructions on ARM... yes
./configure.lineno: 12799: Bad substitution

Fix it by using the %% expansion to remove everything from the first
space instead.

Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
Signed-off-by: Jean-Marc Valin <jmvalin@jmvalin.ca>

Also moves the VSHR32() condition outside the loop just in case some compilers
don't optimize that properly.

This reduces the peak decoding error by removing small (inaudible) spikes in
the error at the frame boundaries. These were due to the frequency-domain bias
ending up as a small pulse in the middle of the IMDCT overlap. None of this
was ever audible, but fixing it is still cleaner.

Thanks to Kevin O'Connor for reporting this.

We released 1.1 with 1.1-beta as the hard-coded version.
Hopefully this only affects people doing their own windows builds
 without the git tools available.
Thanks to Ron Lee for the report.

Should make it easier when writing assembly.

Pedro Becerra authored 11 years ago and Jean-Marc Valin committed 11 years ago

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

Also adds an SSE-optimized celt_inner_prod().

This is a bit faster at -O2 because memcpy()/memmove()/memset() are
vectorized. The code is also cleaner.

gcc is better at optimizing it than the ?: version

NaN detection should now be able to catch values that would create NaNs
further down.