* scalar optimization, vectorization (including new file for SSE2 code
and path selection mechanism).
* 0.5% AEC overall speedup for the straight C path.
* 3.0% AEC overall speedup for the SSE2 path.
Review URL: http://webrtc-codereview.appspot.com/46005
git-svn-id: http://webrtc.googlecode.com/svn/trunk@134 4adac7df-926f-26a2-2b94-8c16560cd09d
Lots of AEC CPU usage is coming from calls to 'rdft'. To optimize this,
deep changes (modification of memory layout, ...) have to be done and it
is not practical to do them in an utility library. Most of these changes
will occur in subsequent CLs.
The new file 'aec_core_rdft.c' is a copy of 'modules/audio_processing/
utility/fft4g.c' whose size has been significantly reduced by removing
all code non-necessary to compute rdft. The main entry point and utility
functions have also been modified to take into account the fact that all
'rdft' calls performed by AEC have a length of 128. This yields:
* 1.8% AEC overall speedup for the straight C path.
* 2.3% AEC overall speedup for the SSE2 path.
Review URL: http://webrtc-codereview.appspot.com/44008
git-svn-id: http://webrtc.googlecode.com/svn/trunk@126 4adac7df-926f-26a2-2b94-8c16560cd09d
* new file for SSE2 code, code selection through function pointers.
* structure change for array of complex numbers.
* 3.8% AEC overall speedup for straight C path.
* 8.8% AEC overall speedup for SSE2 path.
Review URL: http://webrtc-codereview.appspot.com/34002
git-svn-id: http://webrtc.googlecode.com/svn/trunk@36 4adac7df-926f-26a2-2b94-8c16560cd09d
* new file for SSE2 code, code selection through function pointers.
* structure change for array of complex numbers.
* 3.8% AEC overall speedup for straight C path.
* 8.8% AEC overall speedup for SSE2 path.
Review URL: http://webrtc-codereview.appspot.com/33003
git-svn-id: http://webrtc.googlecode.com/svn/trunk@26 4adac7df-926f-26a2-2b94-8c16560cd09d
