changeset 64:6cb2b3cd5356

* Refactor FFT a little bit so as to separate construction and processing rather than have a single static method -- will make it easier to use a different implementation * pull in KissFFT implementation (not hooked up yet)
author cannam
date Wed, 13 May 2009 09:19:12 +0000
parents 0dcbce5d7dce
children d1d65fff5356
files dsp/chromagram/Chromagram.cpp dsp/chromagram/Chromagram.h dsp/chromagram/ConstantQ.cpp dsp/mfcc/MFCC.cpp dsp/mfcc/MFCC.h dsp/onsets/DetectionFunction.cpp dsp/phasevocoder/PhaseVocoder.cpp dsp/phasevocoder/PhaseVocoder.h dsp/segmentation/ClusterMeltSegmenter.cpp dsp/segmentation/ClusterMeltSegmenter.h dsp/tempotracking/DownBeat.cpp dsp/tempotracking/DownBeat.h dsp/transforms/FFT.cpp dsp/transforms/FFT.h dsp/transforms/kissfft/COPYING dsp/transforms/kissfft/_kiss_fft_guts.h dsp/transforms/kissfft/kiss_fft.c dsp/transforms/kissfft/kiss_fft.h dsp/transforms/kissfft/kiss_fftr.c dsp/transforms/kissfft/kiss_fftr.h
diffstat 20 files changed, 1026 insertions(+), 98 deletions(-) [+]
line wrap: on
line diff
--- a/dsp/chromagram/Chromagram.cpp	Tue May 12 21:04:25 2009 +0000
+++ b/dsp/chromagram/Chromagram.cpp	Wed May 13 09:19:12 2009 +0000
@@ -34,9 +34,6 @@
     // Create array for chroma result
     m_chromadata = new double[ m_BPO ];
 
-    // Initialise FFT object	
-    m_FFT = new FFT;
-
     // Create Config Structure for ConstantQ operator
     CQConfig ConstantQConfig;
 
@@ -55,6 +52,9 @@
     m_frameSize = m_ConstantQ->getfftlength();
     m_hopSize = m_ConstantQ->gethop();
 
+    // Initialise FFT object	
+    m_FFT = new FFTReal(m_frameSize);
+
     m_FFTRe = new double[ m_frameSize ];
     m_FFTIm = new double[ m_frameSize ];
     m_CQRe  = new double[ m_uK ];
@@ -135,7 +135,7 @@
     m_window->cut(m_windowbuf);
 
     // FFT of current frame
-    m_FFT->process(m_frameSize, 0, m_windowbuf, NULL, m_FFTRe, m_FFTIm);
+    m_FFT->process(0, m_windowbuf, m_FFTRe, m_FFTIm);
 
     return process(m_FFTRe, m_FFTIm);
 }
--- a/dsp/chromagram/Chromagram.h	Tue May 12 21:04:25 2009 +0000
+++ b/dsp/chromagram/Chromagram.h	Wed May 13 09:19:12 2009 +0000
@@ -61,7 +61,7 @@
     unsigned int m_frameSize;
     unsigned int m_hopSize;
 
-    FFT* m_FFT;
+    FFTReal* m_FFT;
     ConstantQ* m_ConstantQ;
 
     double* m_FFTRe;
--- a/dsp/chromagram/ConstantQ.cpp	Tue May 12 21:04:25 2009 +0000
+++ b/dsp/chromagram/ConstantQ.cpp	Wed May 13 09:19:12 2009 +0000
@@ -103,7 +103,7 @@
     //add it to the sparse kernels matrix
     double squareThreshold = m_CQThresh * m_CQThresh;
 
-    FFT m_FFT;
+    FFT m_FFT(m_FFTLength);
 	
     for (unsigned k = m_uK; k--; ) 
     {
@@ -138,7 +138,7 @@
         }
     
 	//do fft of hammingWindow
-	m_FFT.process( m_FFTLength, 0, hammingWindowRe, hammingWindowIm, transfHammingWindowRe, transfHammingWindowIm );
+	m_FFT.process( 0, hammingWindowRe, hammingWindowIm, transfHammingWindowRe, transfHammingWindowIm );
 
 		
 	for (unsigned j=0; j<( m_FFTLength ); j++) 
--- a/dsp/mfcc/MFCC.cpp	Tue May 12 21:04:25 2009 +0000
+++ b/dsp/mfcc/MFCC.cpp	Wed May 13 09:19:12 2009 +0000
@@ -31,6 +31,7 @@
   
     /* FFT and analysis window sizes */
     fftSize           = config.fftsize;
+    fft               = new FFTReal(fftSize);
 
     totalFilters      = linearFilters + logFilters;
     logPower          = config.logpower;
@@ -145,7 +146,6 @@
     window      = new Window<double>(config.window, fftSize);
 
     /* Allocate memory for the FFT */
-    imagIn      = (double*)calloc(fftSize, sizeof(double));
     realOut     = (double*)calloc(fftSize, sizeof(double));
     imagOut     = (double*)calloc(fftSize, sizeof(double));
 
@@ -185,9 +185,10 @@
     free(fftMag);
     
     /* Free the FFT */
-    free(imagIn);
     free(realOut);
     free(imagOut);
+
+    delete fft;
 }
 
 
@@ -204,7 +205,7 @@
     window->cut(inputData);
   
     /* Calculate the fft on the input frame */
-    FFT::process(fftSize, 0, inputData, imagIn, realOut, imagOut);
+    fft->process(0, inputData, realOut, imagOut);
 
     free(inputData);
 
--- a/dsp/mfcc/MFCC.h	Tue May 12 21:04:25 2009 +0000
+++ b/dsp/mfcc/MFCC.h	Wed May 13 09:19:12 2009 +0000
@@ -13,6 +13,8 @@
 
 #include "base/Window.h"
 
+class FFTReal;
+
 struct MFCCConfig {
     int FS;
     int fftsize;
@@ -76,11 +78,11 @@
     Window<double> *window;
     
     /* For the FFT */
-    double *imagIn;		// always zero
     double *realOut;
     double *imagOut;
     double *fftMag;
     double *earMag;
+    FFTReal *fft;
 
     /* Set if user want C0 */
     int WANT_C0;
--- a/dsp/onsets/DetectionFunction.cpp	Tue May 12 21:04:25 2009 +0000
+++ b/dsp/onsets/DetectionFunction.cpp	Wed May 13 09:19:12 2009 +0000
@@ -58,7 +58,9 @@
     m_magPeaks = new double[ m_halfLength ];
     memset(m_magPeaks,0, m_halfLength*sizeof(double));
 
-    m_phaseVoc = new PhaseVocoder;
+    // See note in process(const double *) below
+    int actualLength = MathUtilities::previousPowerOfTwo(m_dataLength);
+    m_phaseVoc = new PhaseVocoder(actualLength);
 
     m_DFWindowedFrame = new double[ m_dataLength ];
     m_magnitude = new double[ m_halfLength ];
@@ -104,7 +106,7 @@
         }
     }
 
-    m_phaseVoc->process(actualLength, m_DFWindowedFrame, m_magnitude, m_thetaAngle);
+    m_phaseVoc->process(m_DFWindowedFrame, m_magnitude, m_thetaAngle);
 
     if (m_whiten) whiten();
 
--- a/dsp/phasevocoder/PhaseVocoder.cpp	Tue May 12 21:04:25 2009 +0000
+++ b/dsp/phasevocoder/PhaseVocoder.cpp	Wed May 13 09:19:12 2009 +0000
@@ -16,14 +16,19 @@
 // Construction/Destruction
 //////////////////////////////////////////////////////////////////////
 
-PhaseVocoder::PhaseVocoder()
+PhaseVocoder::PhaseVocoder(unsigned int n) :
+    m_n(n)
 {
-
+    m_fft = new FFTReal(m_n);
+    m_realOut = new double[m_n];
+    m_imagOut = new double[m_n];
 }
 
 PhaseVocoder::~PhaseVocoder()
 {
-
+    delete [] m_realOut;
+    delete [] m_imagOut;
+    delete m_fft;
 }
 
 void PhaseVocoder::FFTShift(unsigned int size, double *src)
@@ -36,30 +41,14 @@
     }
 }
 
-void PhaseVocoder::process(unsigned int size, double *src, double *mag, double *theta)
+void PhaseVocoder::process(double *src, double *mag, double *theta)
 {
+    FFTShift( m_n, src);
+	
+    m_fft->process(0, src, m_realOut, m_imagOut);
 
-    // Primary Interface to Phase Vocoder
-    realOut = new double[ size ];
-    imagOut = new double[ size ];
-
-    FFTShift( size, src);
-	
-    coreFFT( size, src, 0, realOut, imagOut);
-
-    getMagnitude( size/2, mag, realOut, imagOut);
-    getPhase( size/2, theta, realOut, imagOut);
-
-    delete [] realOut;
-    delete [] imagOut;
-}
-
-
-void PhaseVocoder::coreFFT( unsigned int NumSamples, double *RealIn, double* ImagIn, double *RealOut, double *ImagOut)
-{
-    // This function is taken from a standard freeware implementation defined in FFT.h
-    // TODO: Use  FFTW
-    FFT::process( NumSamples,0, RealIn, ImagIn, RealOut, ImagOut );
+    getMagnitude( m_n/2, mag, m_realOut, m_imagOut);
+    getPhase( m_n/2, theta, m_realOut, m_imagOut);
 }
 
 void PhaseVocoder::getMagnitude(unsigned int size, double *mag, double *real, double *imag)
--- a/dsp/phasevocoder/PhaseVocoder.h	Tue May 12 21:04:25 2009 +0000
+++ b/dsp/phasevocoder/PhaseVocoder.h	Wed May 13 09:19:12 2009 +0000
@@ -11,23 +11,26 @@
 #ifndef PHASEVOCODER_H
 #define PHASEVOCODER_H
 
+class FFTReal;
 
 class PhaseVocoder  
 {
 public:
-    PhaseVocoder();
+    PhaseVocoder( unsigned int size );
     virtual ~PhaseVocoder();
 
-    void process( unsigned int size, double* src, double* mag, double* theta);
+    void process( double* src, double* mag, double* theta);
 
 protected:
     void getPhase(unsigned int size, double *theta, double *real, double *imag);
-    void coreFFT( unsigned int NumSamples, double *RealIn, double* ImagIn, double *RealOut, double *ImagOut);
+//    void coreFFT( unsigned int NumSamples, double *RealIn, double* ImagIn, double *RealOut, double *ImagOut);
     void getMagnitude( unsigned int size, double* mag, double* real, double* imag);
     void FFTShift( unsigned int size, double* src);
 
-    double* imagOut;
-    double* realOut;
+    unsigned int m_n;
+    FFTReal *m_fft;
+    double *m_imagOut;
+    double *m_realOut;
 
 };
 
--- a/dsp/segmentation/ClusterMeltSegmenter.cpp	Tue May 12 21:04:25 2009 +0000
+++ b/dsp/segmentation/ClusterMeltSegmenter.cpp	Wed May 13 09:19:12 2009 +0000
@@ -22,6 +22,7 @@
 
 ClusterMeltSegmenter::ClusterMeltSegmenter(ClusterMeltSegmenterParams params) :
     window(NULL),
+    fft(NULL),
     constq(NULL),
     mfcc(NULL),
     featureType(params.featureType),
@@ -73,6 +74,8 @@
         constq->sparsekernel();
         
         ncoeff = constq->getK();
+
+        fft = new FFTReal(constq->getfftlength());
         
     } else if (featureType == FEATURE_TYPE_MFCC) {
 
@@ -107,6 +110,7 @@
     delete window;
     delete constq;
     delete decimator;
+    delete fft;
 }
 
 int
@@ -200,7 +204,7 @@
 
         window->cut(frame);
         
-        FFT::process(fftsize, false, frame, 0, real, imag);
+        fft->process(false, frame, real, imag);
         
         constq->process(real, imag, cqre, cqim);
 	
--- a/dsp/segmentation/ClusterMeltSegmenter.h	Tue May 12 21:04:25 2009 +0000
+++ b/dsp/segmentation/ClusterMeltSegmenter.h	Wed May 13 09:19:12 2009 +0000
@@ -20,6 +20,7 @@
 class Decimator;
 class ConstantQ;
 class MFCC;
+class FFTReal;
 
 class ClusterMeltSegmenterParams
 // defaults are sensible for 11025Hz with 0.2 second hopsize
@@ -71,6 +72,7 @@
     void extractFeaturesMFCC(const double *, int);
 
     Window<double> *window;
+    FFTReal *fft;
     ConstantQ* constq; 
     MFCC* mfcc;
     model_t* model;				// the HMM
--- a/dsp/tempotracking/DownBeat.cpp	Tue May 12 21:04:25 2009 +0000
+++ b/dsp/tempotracking/DownBeat.cpp	Wed May 13 09:19:12 2009 +0000
@@ -43,6 +43,7 @@
     m_beatframe = new double[m_beatframesize];
     m_fftRealOut = new double[m_beatframesize];
     m_fftImagOut = new double[m_beatframesize];
+    m_fft = new FFTReal(m_beatframesize);
 }
 
 DownBeat::~DownBeat()
@@ -54,6 +55,7 @@
     delete[] m_beatframe;
     delete[] m_fftRealOut;
     delete[] m_fftImagOut;
+    delete m_fft;
 }
 
 void
@@ -186,8 +188,7 @@
 
         // Now FFT beat frame
         
-        FFT::process(m_beatframesize, false,
-                     m_beatframe, 0, m_fftRealOut, m_fftImagOut);
+        m_fft->process(false, m_beatframe, m_fftRealOut, m_fftImagOut);
         
         // Calculate magnitudes
 
--- a/dsp/tempotracking/DownBeat.h	Tue May 12 21:04:25 2009 +0000
+++ b/dsp/tempotracking/DownBeat.h	Wed May 13 09:19:12 2009 +0000
@@ -17,6 +17,8 @@
 
 using std::vector;
 
+class FFTReal;
+
 /**
  * This class takes an input audio signal and a sequence of beat
  * locations (calculated e.g. by TempoTrackV2) and estimates which of
@@ -119,6 +121,7 @@
     size_t m_buffill;
     size_t m_beatframesize;
     double *m_beatframe;
+    FFTReal *m_fft;
     double *m_fftRealOut;
     double *m_fftImagOut;
     d_vec_t m_beatsd;
--- a/dsp/transforms/FFT.cpp	Tue May 12 21:04:25 2009 +0000
+++ b/dsp/transforms/FFT.cpp	Wed May 13 09:19:12 2009 +0000
@@ -15,13 +15,21 @@
 
 #include <iostream>
 
-//////////////////////////////////////////////////////////////////////
-// Construction/Destruction
-//////////////////////////////////////////////////////////////////////
+#define USE_BUILTIN_FFT 1
 
-FFT::FFT()
+#ifdef USE_BUILTIN_FFT
+
+FFT::FFT(unsigned int n) :
+    m_n(n),
+    m_private(0)
 {
-
+    if( !MathUtilities::isPowerOfTwo(m_n) )
+    {
+        std::cerr << "ERROR: FFT: Non-power-of-two FFT size "
+                  << m_n << " not supported in this implementation"
+                  << std::endl;
+	return;
+    }
 }
 
 FFT::~FFT()
@@ -29,14 +37,61 @@
 
 }
 
-void FFT::process(unsigned int p_nSamples, bool p_bInverseTransform,
-                  const double *p_lpRealIn, const double *p_lpImagIn,
-                  double *p_lpRealOut, double *p_lpImagOut)
+FFTReal::FFTReal(unsigned int n) :
+    m_n(n),
+    m_private(0)
 {
+    m_private = new FFT(m_n);
+}
 
+FFTReal::~FFTReal()
+{
+    delete (FFT *)m_private;
+}
+
+void
+FFTReal::process(bool inverse,
+                 const double *realIn,
+                 double *realOut, double *imagOut)
+{
+    ((FFT *)m_private)->process(inverse, realIn, 0, realOut, imagOut);
+}
+
+static unsigned int numberOfBitsNeeded(unsigned int p_nSamples)
+{	
+    int i;
+
+    if( p_nSamples < 2 )
+    {
+	return 0;
+    }
+
+    for ( i=0; ; i++ )
+    {
+	if( p_nSamples & (1 << i) ) return i;
+    }
+}
+
+static unsigned int reverseBits(unsigned int p_nIndex, unsigned int p_nBits)
+{
+    unsigned int i, rev;
+
+    for(i=rev=0; i < p_nBits; i++)
+    {
+	rev = (rev << 1) | (p_nIndex & 1);
+	p_nIndex >>= 1;
+    }
+
+    return rev;
+}
+
+void
+FFT::process(bool p_bInverseTransform,
+             const double *p_lpRealIn, const double *p_lpImagIn,
+             double *p_lpRealOut, double *p_lpImagOut)
+{
     if(!p_lpRealIn || !p_lpRealOut || !p_lpImagOut) return;
 
-
     unsigned int NumBits;
     unsigned int i, j, k, n;
     unsigned int BlockSize, BlockEnd;
@@ -44,20 +99,20 @@
     double angle_numerator = 2.0 * M_PI;
     double tr, ti;
 
-    if( !MathUtilities::isPowerOfTwo(p_nSamples) )
+    if( !MathUtilities::isPowerOfTwo(m_n) )
     {
         std::cerr << "ERROR: FFT::process: Non-power-of-two FFT size "
-                  << p_nSamples << " not supported in this implementation"
+                  << m_n << " not supported in this implementation"
                   << std::endl;
 	return;
     }
 
     if( p_bInverseTransform ) angle_numerator = -angle_numerator;
 
-    NumBits = numberOfBitsNeeded ( p_nSamples );
+    NumBits = numberOfBitsNeeded ( m_n );
 
 
-    for( i=0; i < p_nSamples; i++ )
+    for( i=0; i < m_n; i++ )
     {
 	j = reverseBits ( i, NumBits );
 	p_lpRealOut[j] = p_lpRealIn[i];
@@ -66,7 +121,7 @@
 
 
     BlockEnd = 1;
-    for( BlockSize = 2; BlockSize <= p_nSamples; BlockSize <<= 1 )
+    for( BlockSize = 2; BlockSize <= m_n; BlockSize <<= 1 )
     {
 	double delta_angle = angle_numerator / (double)BlockSize;
 	double sm2 = -sin ( -2 * delta_angle );
@@ -76,7 +131,7 @@
 	double w = 2 * cm1;
 	double ar[3], ai[3];
 
-	for( i=0; i < p_nSamples; i += BlockSize )
+	for( i=0; i < m_n; i += BlockSize )
 	{
 
 	    ar[2] = cm2;
@@ -116,9 +171,9 @@
 
     if( p_bInverseTransform )
     {
-	double denom = (double)p_nSamples;
+	double denom = (double)m_n;
 
-	for ( i=0; i < p_nSamples; i++ )
+	for ( i=0; i < m_n; i++ )
 	{
 	    p_lpRealOut[i] /= denom;
 	    p_lpImagOut[i] /= denom;
@@ -126,30 +181,9 @@
     }
 }
 
-unsigned int FFT::numberOfBitsNeeded(unsigned int p_nSamples)
-{	
-    int i;
+#else
 
-    if( p_nSamples < 2 )
-    {
-	return 0;
-    }
+#include "kissfft/kiss_fft.h"
+#include "kissfft/kiss_fftr.h"
 
-    for ( i=0; ; i++ )
-    {
-	if( p_nSamples & (1 << i) ) return i;
-    }
-}
-
-unsigned int FFT::reverseBits(unsigned int p_nIndex, unsigned int p_nBits)
-{
-    unsigned int i, rev;
-
-    for(i=rev=0; i < p_nBits; i++)
-    {
-	rev = (rev << 1) | (p_nIndex & 1);
-	p_nIndex >>= 1;
-    }
-
-    return rev;
-}
+#endif
--- a/dsp/transforms/FFT.h	Tue May 12 21:04:25 2009 +0000
+++ b/dsp/transforms/FFT.h	Wed May 13 09:19:12 2009 +0000
@@ -4,7 +4,6 @@
     QM DSP Library
 
     Centre for Digital Music, Queen Mary, University of London.
-    This file is based on Don Cross's public domain FFT implementation.
 */
 
 #ifndef FFT_H
@@ -13,16 +12,31 @@
 class FFT  
 {
 public:
-    static void process(unsigned int nSamples, bool bInverseTransform,
-                        const double *lpRealIn, const double *lpImagIn,
-                        double *lpRealOut, double *lpImagOut);
-    FFT();
-    virtual ~FFT();
+    FFT(unsigned int nsamples);
+    ~FFT();
 
-protected:
-    static unsigned int reverseBits(unsigned int nIndex, unsigned int nBits);
-    static unsigned int numberOfBitsNeeded( unsigned int nSamples );
-    static bool isPowerOfTwo( unsigned int nX );
+    void process(bool inverse,
+                 const double *realIn, const double *imagIn,
+                 double *realOut, double *imagOut);
+    
+private:
+    unsigned int m_n;
+    void *m_private;
 };
 
+class FFTReal
+{
+public:
+    FFTReal(unsigned int nsamples);
+    ~FFTReal();
+
+    void process(bool inverse,
+                 const double *realIn,
+                 double *realOut, double *imagOut);
+
+private:
+    unsigned int m_n;
+    void *m_private;
+};    
+
 #endif
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/dsp/transforms/kissfft/COPYING	Wed May 13 09:19:12 2009 +0000
@@ -0,0 +1,11 @@
+Copyright (c) 2003-2004 Mark Borgerding
+
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
+    * Neither the author nor the names of any contributors may be used to endorse or promote products derived from this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/dsp/transforms/kissfft/_kiss_fft_guts.h	Wed May 13 09:19:12 2009 +0000
@@ -0,0 +1,150 @@
+/*
+Copyright (c) 2003-2004, Mark Borgerding
+
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
+    * Neither the author nor the names of any contributors may be used to endorse or promote products derived from this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+/* kiss_fft.h
+   defines kiss_fft_scalar as either short or a float type
+   and defines
+   typedef struct { kiss_fft_scalar r; kiss_fft_scalar i; }kiss_fft_cpx; */
+#include "kiss_fft.h"
+#include <limits.h>
+
+#define MAXFACTORS 32
+/* e.g. an fft of length 128 has 4 factors 
+ as far as kissfft is concerned
+ 4*4*4*2
+ */
+
+struct kiss_fft_state{
+    int nfft;
+    int inverse;
+    int factors[2*MAXFACTORS];
+    kiss_fft_cpx twiddles[1];
+};
+
+/*
+  Explanation of macros dealing with complex math:
+
+   C_MUL(m,a,b)         : m = a*b
+   C_FIXDIV( c , div )  : if a fixed point impl., c /= div. noop otherwise
+   C_SUB( res, a,b)     : res = a - b
+   C_SUBFROM( res , a)  : res -= a
+   C_ADDTO( res , a)    : res += a
+ * */
+#ifdef FIXED_POINT
+#if (FIXED_POINT==32)
+# define FRACBITS 31
+# define SAMPPROD int64_t
+#define SAMP_MAX 2147483647
+#else
+# define FRACBITS 15
+# define SAMPPROD int32_t 
+#define SAMP_MAX 32767
+#endif
+
+#define SAMP_MIN -SAMP_MAX
+
+#if defined(CHECK_OVERFLOW)
+#  define CHECK_OVERFLOW_OP(a,op,b)  \
+	if ( (SAMPPROD)(a) op (SAMPPROD)(b) > SAMP_MAX || (SAMPPROD)(a) op (SAMPPROD)(b) < SAMP_MIN ) { \
+		fprintf(stderr,"WARNING:overflow @ " __FILE__ "(%d): (%d " #op" %d) = %ld\n",__LINE__,(a),(b),(SAMPPROD)(a) op (SAMPPROD)(b) );  }
+#endif
+
+
+#   define smul(a,b) ( (SAMPPROD)(a)*(b) )
+#   define sround( x )  (kiss_fft_scalar)( ( (x) + (1<<(FRACBITS-1)) ) >> FRACBITS )
+
+#   define S_MUL(a,b) sround( smul(a,b) )
+
+#   define C_MUL(m,a,b) \
+      do{ (m).r = sround( smul((a).r,(b).r) - smul((a).i,(b).i) ); \
+          (m).i = sround( smul((a).r,(b).i) + smul((a).i,(b).r) ); }while(0)
+
+#   define DIVSCALAR(x,k) \
+	(x) = sround( smul(  x, SAMP_MAX/k ) )
+
+#   define C_FIXDIV(c,div) \
+	do {    DIVSCALAR( (c).r , div);  \
+		DIVSCALAR( (c).i  , div); }while (0)
+
+#   define C_MULBYSCALAR( c, s ) \
+    do{ (c).r =  sround( smul( (c).r , s ) ) ;\
+        (c).i =  sround( smul( (c).i , s ) ) ; }while(0)
+
+#else  /* not FIXED_POINT*/
+
+#   define S_MUL(a,b) ( (a)*(b) )
+#define C_MUL(m,a,b) \
+    do{ (m).r = (a).r*(b).r - (a).i*(b).i;\
+        (m).i = (a).r*(b).i + (a).i*(b).r; }while(0)
+#   define C_FIXDIV(c,div) /* NOOP */
+#   define C_MULBYSCALAR( c, s ) \
+    do{ (c).r *= (s);\
+        (c).i *= (s); }while(0)
+#endif
+
+#ifndef CHECK_OVERFLOW_OP
+#  define CHECK_OVERFLOW_OP(a,op,b) /* noop */
+#endif
+
+#define  C_ADD( res, a,b)\
+    do { \
+	    CHECK_OVERFLOW_OP((a).r,+,(b).r)\
+	    CHECK_OVERFLOW_OP((a).i,+,(b).i)\
+	    (res).r=(a).r+(b).r;  (res).i=(a).i+(b).i; \
+    }while(0)
+#define  C_SUB( res, a,b)\
+    do { \
+	    CHECK_OVERFLOW_OP((a).r,-,(b).r)\
+	    CHECK_OVERFLOW_OP((a).i,-,(b).i)\
+	    (res).r=(a).r-(b).r;  (res).i=(a).i-(b).i; \
+    }while(0)
+#define C_ADDTO( res , a)\
+    do { \
+	    CHECK_OVERFLOW_OP((res).r,+,(a).r)\
+	    CHECK_OVERFLOW_OP((res).i,+,(a).i)\
+	    (res).r += (a).r;  (res).i += (a).i;\
+    }while(0)
+
+#define C_SUBFROM( res , a)\
+    do {\
+	    CHECK_OVERFLOW_OP((res).r,-,(a).r)\
+	    CHECK_OVERFLOW_OP((res).i,-,(a).i)\
+	    (res).r -= (a).r;  (res).i -= (a).i; \
+    }while(0)
+
+
+#ifdef FIXED_POINT
+#  define KISS_FFT_COS(phase)  floor(.5+SAMP_MAX * cos (phase))
+#  define KISS_FFT_SIN(phase)  floor(.5+SAMP_MAX * sin (phase))
+#  define HALF_OF(x) ((x)>>1)
+#elif defined(USE_SIMD)
+#  define KISS_FFT_COS(phase) _mm_set1_ps( cos(phase) )
+#  define KISS_FFT_SIN(phase) _mm_set1_ps( sin(phase) )
+#  define HALF_OF(x) ((x)*_mm_set1_ps(.5))
+#else
+#  define KISS_FFT_COS(phase) (kiss_fft_scalar) cos(phase)
+#  define KISS_FFT_SIN(phase) (kiss_fft_scalar) sin(phase)
+#  define HALF_OF(x) ((x)*.5)
+#endif
+
+#define  kf_cexp(x,phase) \
+	do{ \
+		(x)->r = KISS_FFT_COS(phase);\
+		(x)->i = KISS_FFT_SIN(phase);\
+	}while(0)
+
+
+/* a debugging function */
+#define pcpx(c)\
+    fprintf(stderr,"%g + %gi\n",(double)((c)->r),(double)((c)->i) )
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/dsp/transforms/kissfft/kiss_fft.c	Wed May 13 09:19:12 2009 +0000
@@ -0,0 +1,399 @@
+/*
+Copyright (c) 2003-2004, Mark Borgerding
+
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
+    * Neither the author nor the names of any contributors may be used to endorse or promote products derived from this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+
+#include "_kiss_fft_guts.h"
+/* The guts header contains all the multiplication and addition macros that are defined for
+ fixed or floating point complex numbers.  It also delares the kf_ internal functions.
+ */
+
+static kiss_fft_cpx *scratchbuf=NULL;
+static size_t nscratchbuf=0;
+static kiss_fft_cpx *tmpbuf=NULL;
+static size_t ntmpbuf=0;
+
+#define CHECKBUF(buf,nbuf,n) \
+    do { \
+        if ( nbuf < (size_t)(n) ) {\
+            free(buf); \
+            buf = (kiss_fft_cpx*)KISS_FFT_MALLOC(sizeof(kiss_fft_cpx)*(n)); \
+            nbuf = (size_t)(n); \
+        } \
+   }while(0)
+
+
+static void kf_bfly2(
+        kiss_fft_cpx * Fout,
+        const size_t fstride,
+        const kiss_fft_cfg st,
+        int m
+        )
+{
+    kiss_fft_cpx * Fout2;
+    kiss_fft_cpx * tw1 = st->twiddles;
+    kiss_fft_cpx t;
+    Fout2 = Fout + m;
+    do{
+        C_FIXDIV(*Fout,2); C_FIXDIV(*Fout2,2);
+
+        C_MUL (t,  *Fout2 , *tw1);
+        tw1 += fstride;
+        C_SUB( *Fout2 ,  *Fout , t );
+        C_ADDTO( *Fout ,  t );
+        ++Fout2;
+        ++Fout;
+    }while (--m);
+}
+
+static void kf_bfly4(
+        kiss_fft_cpx * Fout,
+        const size_t fstride,
+        const kiss_fft_cfg st,
+        const size_t m
+        )
+{
+    kiss_fft_cpx *tw1,*tw2,*tw3;
+    kiss_fft_cpx scratch[6];
+    size_t k=m;
+    const size_t m2=2*m;
+    const size_t m3=3*m;
+
+    tw3 = tw2 = tw1 = st->twiddles;
+
+    do {
+        C_FIXDIV(*Fout,4); C_FIXDIV(Fout[m],4); C_FIXDIV(Fout[m2],4); C_FIXDIV(Fout[m3],4);
+
+        C_MUL(scratch[0],Fout[m] , *tw1 );
+        C_MUL(scratch[1],Fout[m2] , *tw2 );
+        C_MUL(scratch[2],Fout[m3] , *tw3 );
+
+        C_SUB( scratch[5] , *Fout, scratch[1] );
+        C_ADDTO(*Fout, scratch[1]);
+        C_ADD( scratch[3] , scratch[0] , scratch[2] );
+        C_SUB( scratch[4] , scratch[0] , scratch[2] );
+        C_SUB( Fout[m2], *Fout, scratch[3] );
+        tw1 += fstride;
+        tw2 += fstride*2;
+        tw3 += fstride*3;
+        C_ADDTO( *Fout , scratch[3] );
+
+        if(st->inverse) {
+            Fout[m].r = scratch[5].r - scratch[4].i;
+            Fout[m].i = scratch[5].i + scratch[4].r;
+            Fout[m3].r = scratch[5].r + scratch[4].i;
+            Fout[m3].i = scratch[5].i - scratch[4].r;
+        }else{
+            Fout[m].r = scratch[5].r + scratch[4].i;
+            Fout[m].i = scratch[5].i - scratch[4].r;
+            Fout[m3].r = scratch[5].r - scratch[4].i;
+            Fout[m3].i = scratch[5].i + scratch[4].r;
+        }
+        ++Fout;
+    }while(--k);
+}
+
+static void kf_bfly3(
+         kiss_fft_cpx * Fout,
+         const size_t fstride,
+         const kiss_fft_cfg st,
+         size_t m
+         )
+{
+     size_t k=m;
+     const size_t m2 = 2*m;
+     kiss_fft_cpx *tw1,*tw2;
+     kiss_fft_cpx scratch[5];
+     kiss_fft_cpx epi3;
+     epi3 = st->twiddles[fstride*m];
+
+     tw1=tw2=st->twiddles;
+
+     do{
+         C_FIXDIV(*Fout,3); C_FIXDIV(Fout[m],3); C_FIXDIV(Fout[m2],3);
+
+         C_MUL(scratch[1],Fout[m] , *tw1);
+         C_MUL(scratch[2],Fout[m2] , *tw2);
+
+         C_ADD(scratch[3],scratch[1],scratch[2]);
+         C_SUB(scratch[0],scratch[1],scratch[2]);
+         tw1 += fstride;
+         tw2 += fstride*2;
+
+         Fout[m].r = Fout->r - HALF_OF(scratch[3].r);
+         Fout[m].i = Fout->i - HALF_OF(scratch[3].i);
+
+         C_MULBYSCALAR( scratch[0] , epi3.i );
+
+         C_ADDTO(*Fout,scratch[3]);
+
+         Fout[m2].r = Fout[m].r + scratch[0].i;
+         Fout[m2].i = Fout[m].i - scratch[0].r;
+
+         Fout[m].r -= scratch[0].i;
+         Fout[m].i += scratch[0].r;
+
+         ++Fout;
+     }while(--k);
+}
+
+static void kf_bfly5(
+        kiss_fft_cpx * Fout,
+        const size_t fstride,
+        const kiss_fft_cfg st,
+        int m
+        )
+{
+    kiss_fft_cpx *Fout0,*Fout1,*Fout2,*Fout3,*Fout4;
+    int u;
+    kiss_fft_cpx scratch[13];
+    kiss_fft_cpx * twiddles = st->twiddles;
+    kiss_fft_cpx *tw;
+    kiss_fft_cpx ya,yb;
+    ya = twiddles[fstride*m];
+    yb = twiddles[fstride*2*m];
+
+    Fout0=Fout;
+    Fout1=Fout0+m;
+    Fout2=Fout0+2*m;
+    Fout3=Fout0+3*m;
+    Fout4=Fout0+4*m;
+
+    tw=st->twiddles;
+    for ( u=0; u<m; ++u ) {
+        C_FIXDIV( *Fout0,5); C_FIXDIV( *Fout1,5); C_FIXDIV( *Fout2,5); C_FIXDIV( *Fout3,5); C_FIXDIV( *Fout4,5);
+        scratch[0] = *Fout0;
+
+        C_MUL(scratch[1] ,*Fout1, tw[u*fstride]);
+        C_MUL(scratch[2] ,*Fout2, tw[2*u*fstride]);
+        C_MUL(scratch[3] ,*Fout3, tw[3*u*fstride]);
+        C_MUL(scratch[4] ,*Fout4, tw[4*u*fstride]);
+
+        C_ADD( scratch[7],scratch[1],scratch[4]);
+        C_SUB( scratch[10],scratch[1],scratch[4]);
+        C_ADD( scratch[8],scratch[2],scratch[3]);
+        C_SUB( scratch[9],scratch[2],scratch[3]);
+
+        Fout0->r += scratch[7].r + scratch[8].r;
+        Fout0->i += scratch[7].i + scratch[8].i;
+
+        scratch[5].r = scratch[0].r + S_MUL(scratch[7].r,ya.r) + S_MUL(scratch[8].r,yb.r);
+        scratch[5].i = scratch[0].i + S_MUL(scratch[7].i,ya.r) + S_MUL(scratch[8].i,yb.r);
+
+        scratch[6].r =  S_MUL(scratch[10].i,ya.i) + S_MUL(scratch[9].i,yb.i);
+        scratch[6].i = -S_MUL(scratch[10].r,ya.i) - S_MUL(scratch[9].r,yb.i);
+
+        C_SUB(*Fout1,scratch[5],scratch[6]);
+        C_ADD(*Fout4,scratch[5],scratch[6]);
+
+        scratch[11].r = scratch[0].r + S_MUL(scratch[7].r,yb.r) + S_MUL(scratch[8].r,ya.r);
+        scratch[11].i = scratch[0].i + S_MUL(scratch[7].i,yb.r) + S_MUL(scratch[8].i,ya.r);
+        scratch[12].r = - S_MUL(scratch[10].i,yb.i) + S_MUL(scratch[9].i,ya.i);
+        scratch[12].i = S_MUL(scratch[10].r,yb.i) - S_MUL(scratch[9].r,ya.i);
+
+        C_ADD(*Fout2,scratch[11],scratch[12]);
+        C_SUB(*Fout3,scratch[11],scratch[12]);
+
+        ++Fout0;++Fout1;++Fout2;++Fout3;++Fout4;
+    }
+}
+
+/* perform the butterfly for one stage of a mixed radix FFT */
+static void kf_bfly_generic(
+        kiss_fft_cpx * Fout,
+        const size_t fstride,
+        const kiss_fft_cfg st,
+        int m,
+        int p
+        )
+{
+    int u,k,q1,q;
+    kiss_fft_cpx * twiddles = st->twiddles;
+    kiss_fft_cpx t;
+    int Norig = st->nfft;
+
+    CHECKBUF(scratchbuf,nscratchbuf,p);
+
+    for ( u=0; u<m; ++u ) {
+        k=u;
+        for ( q1=0 ; q1<p ; ++q1 ) {
+            scratchbuf[q1] = Fout[ k  ];
+            C_FIXDIV(scratchbuf[q1],p);
+            k += m;
+        }
+
+        k=u;
+        for ( q1=0 ; q1<p ; ++q1 ) {
+            int twidx=0;
+            Fout[ k ] = scratchbuf[0];
+            for (q=1;q<p;++q ) {
+                twidx += fstride * k;
+                if (twidx>=Norig) twidx-=Norig;
+                C_MUL(t,scratchbuf[q] , twiddles[twidx] );
+                C_ADDTO( Fout[ k ] ,t);
+            }
+            k += m;
+        }
+    }
+}
+
+static
+void kf_work(
+        kiss_fft_cpx * Fout,
+        const kiss_fft_cpx * f,
+        const size_t fstride,
+        int in_stride,
+        int * factors,
+        const kiss_fft_cfg st
+        )
+{
+    kiss_fft_cpx * Fout_beg=Fout;
+    const int p=*factors++; /* the radix  */
+    const int m=*factors++; /* stage's fft length/p */
+    const kiss_fft_cpx * Fout_end = Fout + p*m;
+
+    if (m==1) {
+        do{
+            *Fout = *f;
+            f += fstride*in_stride;
+        }while(++Fout != Fout_end );
+    }else{
+        do{
+            kf_work( Fout , f, fstride*p, in_stride, factors,st);
+            f += fstride*in_stride;
+        }while( (Fout += m) != Fout_end );
+    }
+
+    Fout=Fout_beg;
+
+    switch (p) {
+        case 2: kf_bfly2(Fout,fstride,st,m); break;
+        case 3: kf_bfly3(Fout,fstride,st,m); break; 
+        case 4: kf_bfly4(Fout,fstride,st,m); break;
+        case 5: kf_bfly5(Fout,fstride,st,m); break; 
+        default: kf_bfly_generic(Fout,fstride,st,m,p); break;
+    }
+}
+
+/*  facbuf is populated by p1,m1,p2,m2, ...
+    where 
+    p[i] * m[i] = m[i-1]
+    m0 = n                  */
+static 
+void kf_factor(int n,int * facbuf)
+{
+    int p=4;
+    double floor_sqrt;
+    floor_sqrt = floor( sqrt((double)n) );
+
+    /*factor out powers of 4, powers of 2, then any remaining primes */
+    do {
+        while (n % p) {
+            switch (p) {
+                case 4: p = 2; break;
+                case 2: p = 3; break;
+                default: p += 2; break;
+            }
+            if (p > floor_sqrt)
+                p = n;          /* no more factors, skip to end */
+        }
+        n /= p;
+        *facbuf++ = p;
+        *facbuf++ = n;
+    } while (n > 1);
+}
+
+/*
+ *
+ * User-callable function to allocate all necessary storage space for the fft.
+ *
+ * The return value is a contiguous block of memory, allocated with malloc.  As such,
+ * It can be freed with free(), rather than a kiss_fft-specific function.
+ * */
+kiss_fft_cfg kiss_fft_alloc(int nfft,int inverse_fft,void * mem,size_t * lenmem )
+{
+    kiss_fft_cfg st=NULL;
+    size_t memneeded = sizeof(struct kiss_fft_state)
+        + sizeof(kiss_fft_cpx)*(nfft-1); /* twiddle factors*/
+
+    if ( lenmem==NULL ) {
+        st = ( kiss_fft_cfg)KISS_FFT_MALLOC( memneeded );
+    }else{
+        if (mem != NULL && *lenmem >= memneeded)
+            st = (kiss_fft_cfg)mem;
+        *lenmem = memneeded;
+    }
+    if (st) {
+        int i;
+        st->nfft=nfft;
+        st->inverse = inverse_fft;
+
+        for (i=0;i<nfft;++i) {
+            const double pi=3.141592653589793238462643383279502884197169399375105820974944;
+            double phase = -2*pi*i / nfft;
+            if (st->inverse)
+                phase *= -1;
+            kf_cexp(st->twiddles+i, phase );
+        }
+
+        kf_factor(nfft,st->factors);
+    }
+    return st;
+}
+
+
+
+    
+void kiss_fft_stride(kiss_fft_cfg st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout,int in_stride)
+{
+    if (fin == fout) {
+        CHECKBUF(tmpbuf,ntmpbuf,st->nfft);
+        kf_work(tmpbuf,fin,1,in_stride, st->factors,st);
+        memcpy(fout,tmpbuf,sizeof(kiss_fft_cpx)*st->nfft);
+    }else{
+        kf_work( fout, fin, 1,in_stride, st->factors,st );
+    }
+}
+
+void kiss_fft(kiss_fft_cfg cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout)
+{
+    kiss_fft_stride(cfg,fin,fout,1);
+}
+
+
+/* not really necessary to call, but if someone is doing in-place ffts, they may want to free the 
+   buffers from CHECKBUF
+ */ 
+void kiss_fft_cleanup(void)
+{
+    free(scratchbuf);
+    scratchbuf = NULL;
+    nscratchbuf=0;
+    free(tmpbuf);
+    tmpbuf=NULL;
+    ntmpbuf=0;
+}
+
+int kiss_fft_next_fast_size(int n)
+{
+    while(1) {
+        int m=n;
+        while ( (m%2) == 0 ) m/=2;
+        while ( (m%3) == 0 ) m/=3;
+        while ( (m%5) == 0 ) m/=5;
+        if (m<=1)
+            break; /* n is completely factorable by twos, threes, and fives */
+        n++;
+    }
+    return n;
+}
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/dsp/transforms/kissfft/kiss_fft.h	Wed May 13 09:19:12 2009 +0000
@@ -0,0 +1,108 @@
+#ifndef KISS_FFT_H
+#define KISS_FFT_H
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <math.h>
+#include <memory.h>
+#include <malloc.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifdef USE_SIMD
+# include <xmmintrin.h>
+# define kiss_fft_scalar __m128
+#define KISS_FFT_MALLOC(nbytes) memalign(16,nbytes)
+#else	
+#define KISS_FFT_MALLOC malloc
+#endif	
+
+
+#ifdef FIXED_POINT
+#include <sys/types.h>	
+# if (FIXED_POINT == 32)
+#  define kiss_fft_scalar int32_t
+# else	
+#  define kiss_fft_scalar int16_t
+# endif
+#else
+# ifndef kiss_fft_scalar
+/*  default is float */
+/* #   define kiss_fft_scalar float */
+/* ... but doubles for QM library ... */
+#define kiss_fft_scalar double
+# endif
+#endif
+
+typedef struct {
+    kiss_fft_scalar r;
+    kiss_fft_scalar i;
+}kiss_fft_cpx;
+
+typedef struct kiss_fft_state* kiss_fft_cfg;
+
+/* 
+ *  kiss_fft_alloc
+ *  
+ *  Initialize a FFT (or IFFT) algorithm's cfg/state buffer.
+ *
+ *  typical usage:      kiss_fft_cfg mycfg=kiss_fft_alloc(1024,0,NULL,NULL);
+ *
+ *  The return value from fft_alloc is a cfg buffer used internally
+ *  by the fft routine or NULL.
+ *
+ *  If lenmem is NULL, then kiss_fft_alloc will allocate a cfg buffer using malloc.
+ *  The returned value should be free()d when done to avoid memory leaks.
+ *  
+ *  The state can be placed in a user supplied buffer 'mem':
+ *  If lenmem is not NULL and mem is not NULL and *lenmem is large enough,
+ *      then the function places the cfg in mem and the size used in *lenmem
+ *      and returns mem.
+ *  
+ *  If lenmem is not NULL and ( mem is NULL or *lenmem is not large enough),
+ *      then the function returns NULL and places the minimum cfg 
+ *      buffer size in *lenmem.
+ * */
+
+kiss_fft_cfg kiss_fft_alloc(int nfft,int inverse_fft,void * mem,size_t * lenmem); 
+
+/*
+ * kiss_fft(cfg,in_out_buf)
+ *
+ * Perform an FFT on a complex input buffer.
+ * for a forward FFT,
+ * fin should be  f[0] , f[1] , ... ,f[nfft-1]
+ * fout will be   F[0] , F[1] , ... ,F[nfft-1]
+ * Note that each element is complex and can be accessed like
+    f[k].r and f[k].i
+ * */
+void kiss_fft(kiss_fft_cfg cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout);
+
+/*
+ A more generic version of the above function. It reads its input from every Nth sample.
+ * */
+void kiss_fft_stride(kiss_fft_cfg cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout,int fin_stride);
+
+/* If kiss_fft_alloc allocated a buffer, it is one contiguous 
+   buffer and can be simply free()d when no longer needed*/
+#define kiss_fft_free free
+
+/*
+ Cleans up some memory that gets managed internally. Not necessary to call, but it might clean up 
+ your compiler output to call this before you exit.
+*/
+void kiss_fft_cleanup(void);
+	
+
+/*
+ * Returns the smallest integer k, such that k>=n and k has only "fast" factors (2,3,5)
+ */
+int kiss_fft_next_fast_size(int n);
+
+#ifdef __cplusplus
+} 
+#endif
+
+#endif
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/dsp/transforms/kissfft/kiss_fftr.c	Wed May 13 09:19:12 2009 +0000
@@ -0,0 +1,159 @@
+/*
+Copyright (c) 2003-2004, Mark Borgerding
+
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
+
+    * Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
+    * Neither the author nor the names of any contributors may be used to endorse or promote products derived from this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#include "kiss_fftr.h"
+#include "_kiss_fft_guts.h"
+
+struct kiss_fftr_state{
+    kiss_fft_cfg substate;
+    kiss_fft_cpx * tmpbuf;
+    kiss_fft_cpx * super_twiddles;
+#ifdef USE_SIMD    
+    long pad;
+#endif    
+};
+
+kiss_fftr_cfg kiss_fftr_alloc(int nfft,int inverse_fft,void * mem,size_t * lenmem)
+{
+    int i;
+    kiss_fftr_cfg st = NULL;
+    size_t subsize, memneeded;
+
+    if (nfft & 1) {
+        fprintf(stderr,"Real FFT optimization must be even.\n");
+        return NULL;
+    }
+    nfft >>= 1;
+
+    kiss_fft_alloc (nfft, inverse_fft, NULL, &subsize);
+    memneeded = sizeof(struct kiss_fftr_state) + subsize + sizeof(kiss_fft_cpx) * ( nfft * 2);
+
+    if (lenmem == NULL) {
+        st = (kiss_fftr_cfg) KISS_FFT_MALLOC (memneeded);
+    } else {
+        if (*lenmem >= memneeded)
+            st = (kiss_fftr_cfg) mem;
+        *lenmem = memneeded;
+    }
+    if (!st)
+        return NULL;
+
+    st->substate = (kiss_fft_cfg) (st + 1); /*just beyond kiss_fftr_state struct */
+    st->tmpbuf = (kiss_fft_cpx *) (((char *) st->substate) + subsize);
+    st->super_twiddles = st->tmpbuf + nfft;
+    kiss_fft_alloc(nfft, inverse_fft, st->substate, &subsize);
+
+    for (i = 0; i < nfft; ++i) {
+        double phase =
+            -3.14159265358979323846264338327 * ((double) i / nfft + .5);
+        if (inverse_fft)
+            phase *= -1;
+        kf_cexp (st->super_twiddles+i,phase);
+    }
+    return st;
+}
+
+void kiss_fftr(kiss_fftr_cfg st,const kiss_fft_scalar *timedata,kiss_fft_cpx *freqdata)
+{
+    /* input buffer timedata is stored row-wise */
+    int k,ncfft;
+    kiss_fft_cpx fpnk,fpk,f1k,f2k,tw,tdc;
+
+    if ( st->substate->inverse) {
+        fprintf(stderr,"kiss fft usage error: improper alloc\n");
+        exit(1);
+    }
+
+    ncfft = st->substate->nfft;
+
+    /*perform the parallel fft of two real signals packed in real,imag*/
+    kiss_fft( st->substate , (const kiss_fft_cpx*)timedata, st->tmpbuf );
+    /* The real part of the DC element of the frequency spectrum in st->tmpbuf
+     * contains the sum of the even-numbered elements of the input time sequence
+     * The imag part is the sum of the odd-numbered elements
+     *
+     * The sum of tdc.r and tdc.i is the sum of the input time sequence. 
+     *      yielding DC of input time sequence
+     * The difference of tdc.r - tdc.i is the sum of the input (dot product) [1,-1,1,-1... 
+     *      yielding Nyquist bin of input time sequence
+     */
+ 
+    tdc.r = st->tmpbuf[0].r;
+    tdc.i = st->tmpbuf[0].i;
+    C_FIXDIV(tdc,2);
+    CHECK_OVERFLOW_OP(tdc.r ,+, tdc.i);
+    CHECK_OVERFLOW_OP(tdc.r ,-, tdc.i);
+    freqdata[0].r = tdc.r + tdc.i;
+    freqdata[ncfft].r = tdc.r - tdc.i;
+#ifdef USE_SIMD    
+    freqdata[ncfft].i = freqdata[0].i = _mm_set1_ps(0);
+#else
+    freqdata[ncfft].i = freqdata[0].i = 0;
+#endif
+
+    for ( k=1;k <= ncfft/2 ; ++k ) {
+        fpk    = st->tmpbuf[k]; 
+        fpnk.r =   st->tmpbuf[ncfft-k].r;
+        fpnk.i = - st->tmpbuf[ncfft-k].i;
+        C_FIXDIV(fpk,2);
+        C_FIXDIV(fpnk,2);
+
+        C_ADD( f1k, fpk , fpnk );
+        C_SUB( f2k, fpk , fpnk );
+        C_MUL( tw , f2k , st->super_twiddles[k]);
+
+        freqdata[k].r = HALF_OF(f1k.r + tw.r);
+        freqdata[k].i = HALF_OF(f1k.i + tw.i);
+        freqdata[ncfft-k].r = HALF_OF(f1k.r - tw.r);
+        freqdata[ncfft-k].i = HALF_OF(tw.i - f1k.i);
+    }
+}
+
+void kiss_fftri(kiss_fftr_cfg st,const kiss_fft_cpx *freqdata,kiss_fft_scalar *timedata)
+{
+    /* input buffer timedata is stored row-wise */
+    int k, ncfft;
+
+    if (st->substate->inverse == 0) {
+        fprintf (stderr, "kiss fft usage error: improper alloc\n");
+        exit (1);
+    }
+
+    ncfft = st->substate->nfft;
+
+    st->tmpbuf[0].r = freqdata[0].r + freqdata[ncfft].r;
+    st->tmpbuf[0].i = freqdata[0].r - freqdata[ncfft].r;
+    C_FIXDIV(st->tmpbuf[0],2);
+
+    for (k = 1; k <= ncfft / 2; ++k) {
+        kiss_fft_cpx fk, fnkc, fek, fok, tmp;
+        fk = freqdata[k];
+        fnkc.r = freqdata[ncfft - k].r;
+        fnkc.i = -freqdata[ncfft - k].i;
+        C_FIXDIV( fk , 2 );
+        C_FIXDIV( fnkc , 2 );
+
+        C_ADD (fek, fk, fnkc);
+        C_SUB (tmp, fk, fnkc);
+        C_MUL (fok, tmp, st->super_twiddles[k]);
+        C_ADD (st->tmpbuf[k],     fek, fok);
+        C_SUB (st->tmpbuf[ncfft - k], fek, fok);
+#ifdef USE_SIMD        
+        st->tmpbuf[ncfft - k].i *= _mm_set1_ps(-1.0);
+#else
+        st->tmpbuf[ncfft - k].i *= -1;
+#endif
+    }
+    kiss_fft (st->substate, st->tmpbuf, (kiss_fft_cpx *) timedata);
+}
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/dsp/transforms/kissfft/kiss_fftr.h	Wed May 13 09:19:12 2009 +0000
@@ -0,0 +1,46 @@
+#ifndef KISS_FTR_H
+#define KISS_FTR_H
+
+#include "kiss_fft.h"
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+    
+/* 
+ 
+ Real optimized version can save about 45% cpu time vs. complex fft of a real seq.
+
+ 
+ 
+ */
+
+typedef struct kiss_fftr_state *kiss_fftr_cfg;
+
+
+kiss_fftr_cfg kiss_fftr_alloc(int nfft,int inverse_fft,void * mem, size_t * lenmem);
+/*
+ nfft must be even
+
+ If you don't care to allocate space, use mem = lenmem = NULL 
+*/
+
+
+void kiss_fftr(kiss_fftr_cfg cfg,const kiss_fft_scalar *timedata,kiss_fft_cpx *freqdata);
+/*
+ input timedata has nfft scalar points
+ output freqdata has nfft/2+1 complex points
+*/
+
+void kiss_fftri(kiss_fftr_cfg cfg,const kiss_fft_cpx *freqdata,kiss_fft_scalar *timedata);
+/*
+ input freqdata has  nfft/2+1 complex points
+ output timedata has nfft scalar points
+*/
+
+#define kiss_fftr_free free
+
+#ifdef __cplusplus
+}
+#endif
+#endif