getting ready for next release

Makefile

@@ -2,9 +2,9 @@ message:
 	@echo "Nothing to make here.  Move on down to sample_code for ... you guessed it! Sample Code!"
 
 tarball: clean
-	tar --exclude CVS -czf kiss_fft.tar.gz .
+	tar --exclude CVS --exclude .cvsignore -cvzf kiss_fft.tar.gz .
 
 clean:
 	cd sample_code && make clean
-	rm -f kiss_fft.tar.gz *~
+	rm -f kiss_fft.tar.gz *~ *.pyc
 

README

@@ -1,4 +1,4 @@
-KISS FFT - A power-of-two Fast Fourier Transform based up on the principle, 
+KISS FFT - A mixed-radix Fast Fourier Transform based up on the principle, 
 "Keep It Simple, Stupid."
 
     There are many great fft libraries already around.  Kiss FFT is not trying
@@ -26,8 +26,8 @@ two functions you'll need to use. Code definitions are in kiss_fft.c, along
 with sample usage code.
 
 
-    The code can be easily recompiled to work with 16bit fixed point data,
+    The code can be compiled to use float, double or 16bit short samples.
-or various floating point types.  The default is float.
+The default is float.
 
 
 BACKGROUND:
@@ -35,7 +35,7 @@ BACKGROUND:
     I started coding this because I couldn't find a fixed point FFT that didn't 
 use assembly code.  I started with floating point numbers so I could get the 
 theory straight before working on fixed point issues.  In the end, I had a 
-little bit of code that could be recompiled easily to do ffts with short, float,
+little bit of code that could be recompiled easily to do ffts with short, float
 or double (other types should be easy too).  
 
     Once I got my FFT working, I wanted to get some performance numbers against 
@@ -46,7 +46,8 @@ During this process, I learned:
     1. FFT_BRANDX has 500 times as many lines of code as Kiss 
         (and that's just the C code).
     2. It took me an embarrassingly long time to get FFT_BRANDX working.
-    3. FFT_BRANDX is almost 3 times faster than Kiss
+    3. A simple program using FFT_BRANDX is 500K. A similar program using kiss_fft is 18k.
+    4. FFT_BRANDX is about 3-4 times faster than Kiss
 
     It is wonderful that free, highly optimized libraries like FFT_BRANDX exist.
 But such libraries carry a huge burden of complexity necessary to extract every 
@@ -62,13 +63,12 @@ PERFORMANCE:
 
 DO NOT:
     ... use Kiss if you need the absolute fastest fft in the world
-    ... use Kiss if you need mixed radix FFTs 
     ... ask me to add features that will bloat the code
 
 UNDER THE HOOD:
 
-    Kiss uses a complex-only, frequency decimation, radix 2, in-place FFT. Bit reversed 
+    Kiss uses a complex-only, time decimation, mixed-radix , out-of-place FFT. 
-addressing is corrected as the last step in the transform.  No scaling is done.
+No scaling is done.
 
 LICENSE:
     BSD, see COPYING for details. Basically, "free to use, give credit where due, no guarantees"
@@ -76,7 +76,6 @@ LICENSE:
 TODO:
     *) Add sample code for parallel ffts (stereo) packed into re,im components of time sequence.
     *) Add simple windowing function, e.g. Hamming : w(i)=.54-.46*cos(2pi*i/(n-1))
-    *) Could mixed-radix FFTs be made simple enough to stand by the KISS principle?
     *) Make the fixed point scaling and bit shifts more easily configurable.
     *) Document/revisit the input/output fft scaling
     *) See if the fixed point code can be optimized a little without adding complexity.

kiss_fft.c

@@ -23,13 +23,6 @@ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
    typedef struct { kiss_fft_scalar r; kiss_fft_scalar i; }kiss_fft_cpx; */
 #include "kiss_fft.h"
 
-/*  Try static inline. Who knows? You might get a speed improvement. I didn't. YMMV   */
-#if 0
-#  define FUNCDECL static inline
-#else
-#  define FUNCDECL
-#endif
-
 typedef struct {
     int nfft;
     int inverse;
@@ -47,25 +40,15 @@ typedef struct {
       }while(0)
 #   define C_FIXDIV(c,div) \
     do{ (c).r /= div; (c).i /=div; }while(0)
-#define C_MUL_SCALAR(m,s) \
-     do{ (m).r = ( (m).r * (s) + (1<<14) ) >> 15;\
-         (m).i = ( (m).i * (s) + (1<<14) ) >> 15;\
-     }while(0)
 
 #else  /* not FIXED_POINT*/
 
-#define C_MUL_SCALAR(m,s) \
-     do{ (m).r *= (s);\
-         (m).i *= (s);\
-     }while(0)
 #   define C_FIXDIV(c,div) /* NOOP */
 #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)
 #endif
 
-#define  C_ADD( res, a,b)\
-    do {    (res).r=(a).r+(b).r;  (res).i=(a).i+(b).i;  }while(0)
 #define  C_SUB( res, a,b)\
     do {    (res).r=(a).r-(b).r;  (res).i=(a).i-(b).i;  }while(0)
 #define C_ADDTO( res , a)\
@@ -93,7 +76,6 @@ kiss_fft_cpx cexp(double phase)
     return x;
 }
 
-FUNCDECL
 void bfly2(
         kiss_fft_cpx * Fout,
         int fstride,
@@ -116,52 +98,6 @@ void bfly2(
     }while (--m);
 }
 
-FUNCDECL
-void bfly3(
-        kiss_fft_cpx * Fout,
-        int fstride,
-        const kiss_fft_state * st,
-        int m
-        )
-{
-    kiss_fft_cpx *Fout0,*Fout1,*Fout2;
-    
-    int u;
-    kiss_fft_cpx * scratch = st->scratch;
-    kiss_fft_cpx *tw1,*tw2;
-    tw2 = tw1 = st->twiddles;
-
-    Fout0=Fout;
-    Fout1=Fout0+m;
-    Fout2=Fout0+2*m;
-
-    scratch[3] = st->twiddles[ fstride*m ];
-
-    for ( u=0; u<m; ++u ) {
-        C_FIXDIV(*Fout0,3);
-        C_FIXDIV(*Fout1,3);
-        C_FIXDIV(*Fout2,3);
-        scratch[0] = *Fout0;
-        C_MUL(scratch[1],*Fout1 , *tw1 ); 
-        tw1 += fstride;
-        C_MUL(scratch[2],*Fout2 , *tw2 );
-        tw2 += 2*fstride;
-        C_ADD(scratch[5],scratch[1],scratch[2]);
-        C_SUB(scratch[6],scratch[1],scratch[2]);
-        C_ADDTO(*Fout0,scratch[5]);
-        C_MUL_SCALAR(scratch[5], scratch[3].r );
-        C_MUL_SCALAR(scratch[6], scratch[3].i );
-        scratch[4].r = scratch[5].r - scratch[6].i;
-        scratch[4].i = scratch[5].i + scratch[6].r;
-        C_ADD( *Fout1, scratch[0] , scratch[4] );
-        *Fout2 = *Fout1;
-        Fout2->r += 2*scratch[6].i;
-        Fout2->i -= 2*scratch[6].r;
-        ++Fout0;++Fout1;++Fout2;
-    }
-}
-
-FUNCDECL
 void bfly4(
         kiss_fft_cpx * Fout,
         int fstride,
@@ -214,8 +150,7 @@ void bfly4(
     }while(--m);
 }
         
-FUNCDECL
+void bfly_generic(
-void bflyp(
         kiss_fft_cpx * Fout,
         int fstride,
         const kiss_fft_state * st,
@@ -227,6 +162,7 @@ void bflyp(
     kiss_fft_cpx * scratch = st->scratch;
     kiss_fft_cpx * twiddles = st->twiddles;
     kiss_fft_cpx t;
+    int Norig = st->nfft;
 
     for ( u=0; u<m; ++u ) {
         k=u;
@@ -241,7 +177,6 @@ void bflyp(
             int twidx=0;
             Fout[ k ] = scratch[0];
             for (q=1;q<p;++q ) {
-                int Norig = st->nfft;
                 twidx += fstride * k;
                 if (twidx>=Norig) twidx-=Norig;
                 C_MUL(t,scratch[q] , twiddles[twidx] );
@@ -252,7 +187,6 @@ void bflyp(
     }
 }
 
-FUNCDECL
 void fft_work(
         kiss_fft_cpx * Fout,
         const kiss_fft_cpx * f,
@@ -274,10 +208,9 @@ void fft_work(
     }
 
     switch (p) {
-        case 2: bfly2(Fout,fstride,st,m); break;
-        case 3: bfly3(Fout,fstride,st,m); break;
         case 4: bfly4(Fout,fstride,st,m); break;
-        default: bflyp(Fout,fstride,st,m,p); break;
+        case 2: bfly2(Fout,fstride,st,m); break;
+        default: bfly_generic(Fout,fstride,st,m,p); break;
     }
 }
 
@@ -324,7 +257,7 @@ void * kiss_fft_alloc(int nfft,int inverse_fft)
 
     while (nfft>1) {
         /* If you add a new radix, don't forget to put it here */
-        const int primes[] = {4,2,3,-1};
+        const int primes[] = {4,2,-1};
         int p=nfft;
         i=0;
         while ( primes[i] != -1 ) {
@@ -345,10 +278,13 @@ void * kiss_fft_alloc(int nfft,int inverse_fft)
 
 void kiss_fft(const void * cfg,kiss_fft_cpx *f)
 {
-    int n;
     const kiss_fft_state * st = cfg;
-
+    memcpy(st->tmpbuf,f,sizeof(kiss_fft_cpx)*st->nfft);
-    n = st->nfft;
-    memcpy(st->tmpbuf,f,sizeof(kiss_fft_cpx)*n);
     fft_work( f, st->tmpbuf, 1, st->factors,st );
 }
+
+void kiss_fft_io(const void * cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout)
+{
+    const kiss_fft_state * st = cfg;
+    fft_work( fout, fin, 1, st->factors,st );
+}

kiss_fft.h

@@ -2,10 +2,7 @@
 #define KISS_FFT_H
 
 #ifdef FIXED_POINT
-
 # define kiss_fft_scalar short
-# define FIXED_POINT_FRAC_BITS 15
-
 #else
 # ifndef kiss_fft_scalar
 #   define kiss_fft_scalar float
@@ -21,7 +18,7 @@ typedef struct {
 /* 
  *  fft_alloc
  *  
- *  Initialize a radix 2 FFT (or IFFT) algorithm.
+ *  Initialize a FFT (or IFFT) algorithm.
  *
  *  The return value from fft_alloc is a cfg buffer used internally
  *  by the fft routine.
@@ -42,6 +39,9 @@ void* kiss_fft_alloc(int nfft,int inverse_fft);
  * */
 void kiss_fft( const void* cfg_from_alloc , kiss_fft_cpx *f ); /* call for each buffer */
 
+/* two buffer version */
+void kiss_fft_io(const void * cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout);
+
 /* when done with the cfg for a given fft size and direction, simply free it*/
 #define kiss_fft_free free
 

test/Makefile

@@ -1,9 +1,17 @@
-NFFT=840
+
+ifeq "$(NFFT)" ""
+	NFFT=1024
+endif
+
 ALLUTILS=kfft kffts kfftd
+
+ifeq "$(NUMFFTS)" ""
 	NUMFFTS=10000
+endif
+
 UTILSRC=../kiss_fft.c fftutil.c
 
-CFLAGS=-Wall -O3 -lm -I.. -ansi -DFUNCDECL= -pedantic
+CFLAGS=-Wall -O3 -lm -I.. -ansi -pedantic
 
 all: $(ALLUTILS)
 
@@ -17,7 +25,7 @@ kfftd: $(UTILSRC)
 
 time: all
 	@echo 
-	@echo "#### timing $(NUMFFTS) x $(NFFT) point FFTs"
+	@echo -n "#### timing $(NUMFFTS) x $(NFFT) point FFTs.  "; factor $(NFFT)
 	@echo "#### DOUBLE"
 	@time -f 'Elapsed:%E user:%U sys:%S' \
 	 sh -c 'dd if=/dev/zero bs=$$((16*$(NFFT))) count=$(NUMFFTS) 2>/dev/null |./kfftd>/dev/null'
@@ -28,6 +36,14 @@ time: all
 	@time -f 'Elapsed:%E user:%U sys:%S' \
 	 sh -c 'dd if=/dev/zero bs=$$((4*$(NFFT))) count=$(NUMFFTS) 2>/dev/null |./kffts>/dev/null'
 
+
+POW2=256 512 1024 2048
+POW3=243 729 2187
+mtime: all
+	@for n in $(POW2) ;do \
+		export NFFT=$$n;make time; \
+		done
+
 snr: all
 	@which octave
 	@echo

test/freqpeak.c

@@ -1,87 +0,0 @@
-#include <stdio.h>
-#include <string.h>
-#include <memory.h>
-#include <malloc.h>
-#include <stdio.h>
-#include <math.h>
-#include "kiss_fft.h"
-
-#define NFFT 1024
-int main(int argc, char ** argv)
-{
-    int k;
-    void * st;
-    float fs=44100;
-
-    short sampsin[2*NFFT];
-    float lmag2[NFFT/2];
-    float rmag2[NFFT/2];
-    int peakr=0,peakl=0;
-    int removedc=1;
-
-    kiss_fft_cpx cbuf[NFFT];
-    int nbufs=0;
-
-    st = kiss_fft_alloc(NFFT,0);
-
-    memset( lmag2 , 0 , sizeof(lmag2) );
-    memset( rmag2 , 0 , sizeof(rmag2) );
-
-    while ( fread( sampsin , sizeof(short) * 2*NFFT, 1 , stdin ) == 1 ) {
-
-        //perform two ffts in parallel by packing the left&right channels into the real and imaginary
-        for (k=0;k<NFFT;++k) {
-            cbuf[k].r = sampsin[2*k];
-            cbuf[k].i = sampsin[2*k+1];
-        }
-
-        if (removedc) {
-            float dcr=0,dci=0;
-            for (k=0;k<NFFT;++k){
-                dcr += cbuf[k].r;
-                dci += cbuf[k].i;
-            }
-            dcr /= NFFT;
-            dci /= NFFT;
-
-            for (k=0;k<NFFT;++k) {
-                cbuf[k].r -= dcr;
-                cbuf[k].i -= dci;
-            }
-        }
-
-        // perform the fft on the L+R packed buffer
-        kiss_fft( st , cbuf );
-
-        // get the half-symmetric FFTs for the Left and Right Channels
-        for (k=0;k<NFFT/2;++k) {
-            int k2 = (NFFT-k)%NFFT;
-            kiss_fft_cpx r,l;
-
-            r.r = (cbuf[k].r + cbuf[k2].r) * 0.5;
-            r.i = (cbuf[k].i - cbuf[k2].i) * 0.5;
-
-            l.r = (cbuf[k].i + cbuf[k2].i) * 0.5;
-            l.i = (cbuf[k].r - cbuf[k2].r) * 0.5;
-
-            rmag2[k] += r.r * r.r + r.i * r.i;
-            lmag2[k] += l.r * l.r + l.i * l.i;
-        }
-        ++nbufs;
-    }
-    free(st);
-
-    for (k=0;k<NFFT/2;++k) {
-        if (rmag2[peakr] < rmag2[k])
-            peakr = k;
-        if (lmag2[peakl] < lmag2[k])
-            peakl = k;
-    }
-
-    printf("%d buffers\n",nbufs);
-
-    printf("peak frequency R:%.3fdB @ %.1f Hz\n",10*log10(rmag2[peakr]) , peakr*fs/NFFT);
-    printf("               L:%.3fdB @ %.1f Hz\n",10*log10(lmag2[peakl]) , peakl*fs/NFFT);
-
-    return 0;
-}

test/sigdiff.c

@@ -1,80 +0,0 @@
-/*
-Copyright (c) 2003, 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 <stdlib.h>
-#include <math.h>
-#include <stdio.h>
-#include <string.h>
-
-
-#ifndef BUFSIZE
-# define BUFSIZE 1024
-#endif
-
-int main(int argc, char ** argv)
-{
-    FILE * f1=stdin;
-    FILE * f2=stdin;
-
-    int i,n;
-    TYPE buf1[BUFSIZE];
-    TYPE buf2[BUFSIZE];
-
-    double sigpower=0;
-    double noisepower=0;
-    double snrdb=0;
-    double scale=0;
-    long ntotal=0;
-
-    if (argc>1 && strcmp( argv[1] ,"-") != 0 ) {
-        f1 = fopen(argv[1],"rb");
-    }
-    if (argc>2 && strcmp( argv[2] ,"-") != 0 ) {
-        f2 = fopen(argv[2],"rb");
-    }
-
-    // TODO LEFT OFF HERE
-    while ( ( n = fread( buf1 , sizeof(TYPE) , BUFSIZE , f1 ) ) > 0 ) {
-        if ( fread( buf2 , sizeof(TYPE) , BUFSIZE , f2 ) != n ) {
-            fprintf(stderr,"premature end of file 2\n");
-            exit(1);
-        }
-        for (i=0;i<n;++i) {
-            double s=buf1[i];
-            double n = s - buf2[i];
-            sigpower += s*s;
-            noisepower += n*n;
-            if (s!=0) {
-                ++ntotal;
-                scale += buf2[i] / s;
-            }
-        }
-    }
-    
-    if  ( fread( buf2 , sizeof(TYPE) , BUFSIZE , f2 ) > 0 ) {
-        fprintf(stderr,"premature end of file 1\n");
-        exit(1);
-    }
-    scale /= ntotal;
-    
-    if (noisepower>0)
-        snrdb = 10*log10( sigpower / noisepower );
-    else
-        snrdb = 200;
-
-    printf("snr = %.2f dB\n",snrdb);
-    printf("average output/input = %.5e\n",scale);
-
-    return 0;
-}

test/typeconvert.c

@@ -1,46 +0,0 @@
-/*
-Copyright (c) 2003, 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 <stdlib.h>
-#include <math.h>
-#include <stdio.h>
-#include <string.h>
-
-
-#define BUFSIZE 1024
-int main(int argc, char ** argv )
-{
-    int i,n;
-    TYPE1 buf1[BUFSIZE];
-    TYPE2 buf2[BUFSIZE];
-    double mult=1.0;
-
-    for (i=1;i<argc;++i){
-        if (strcmp("-m",argv[i])==0){
-            mult *= atof(argv[i+1] );
-            ++i;
-        }else if (strcmp("-d",argv[i])==0){
-            mult /= atof(argv[i+1] );
-            ++i;
-        }
-    }
-
-    while ( ( n = fread( buf1 , sizeof(TYPE1) , BUFSIZE , stdin ) ) > 0 ) {
-        for (i=0;i<n;++i)
-            buf2[i] = (TYPE1)(mult * buf1[i]);
-        fwrite( buf2 , sizeof(TYPE2) , n , stdout );
-    }
-
-    return 0;
-}

tools/Makefile

@@ -1,9 +1,17 @@
-NFFT=840
+
+ifeq "$(NFFT)" ""
+	NFFT=1024
+endif
+
 ALLUTILS=kfft kffts kfftd
+
+ifeq "$(NUMFFTS)" ""
 	NUMFFTS=10000
+endif
+
 UTILSRC=../kiss_fft.c fftutil.c
 
-CFLAGS=-Wall -O3 -lm -I.. -ansi -DFUNCDECL= -pedantic
+CFLAGS=-Wall -O3 -lm -I.. -ansi -pedantic
 
 all: $(ALLUTILS)
 
@@ -17,7 +25,7 @@ kfftd: $(UTILSRC)
 
 time: all
 	@echo 
-	@echo "#### timing $(NUMFFTS) x $(NFFT) point FFTs"
+	@echo -n "#### timing $(NUMFFTS) x $(NFFT) point FFTs.  "; factor $(NFFT)
 	@echo "#### DOUBLE"
 	@time -f 'Elapsed:%E user:%U sys:%S' \
 	 sh -c 'dd if=/dev/zero bs=$$((16*$(NFFT))) count=$(NUMFFTS) 2>/dev/null |./kfftd>/dev/null'
@@ -28,6 +36,14 @@ time: all
 	@time -f 'Elapsed:%E user:%U sys:%S' \
 	 sh -c 'dd if=/dev/zero bs=$$((4*$(NFFT))) count=$(NUMFFTS) 2>/dev/null |./kffts>/dev/null'
 
+
+POW2=256 512 1024 2048
+POW3=243 729 2187
+mtime: all
+	@for n in $(POW2) ;do \
+		export NFFT=$$n;make time; \
+		done
+
 snr: all
 	@which octave
 	@echo