modified time benchmark to repeat same buffer over and over to avoid IO bottlenecks and get more consistent numbers.

kiss_fft.c

@@ -206,6 +206,43 @@ void bfly3(
      }while(--m);
 }
 
+void bfly5(
+        kiss_fft_cpx * Fout,
+        int fstride,
+        const kiss_fft_state * st,
+        int m
+        )
+{
+    int u,k,q1,q;
+    const int p=5;
+    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;
+        for ( q1=0 ; q1<p ; ++q1 ) {
+            scratch[q1] = Fout[ k  ];
+            C_FIXDIV(scratch[q1],p);
+            k += m;
+        }
+
+        k=u;
+        for ( q1=0 ; q1<p ; ++q1 ) {
+            int twidx=0;
+            Fout[ k ] = scratch[0];
+            for (q=1;q<p;++q ) {
+                twidx += fstride * k;
+                if (twidx>=Norig) twidx-=Norig;
+                C_MUL(t,scratch[q] , twiddles[twidx] );
+                C_ADDTO( Fout[ k ] ,t);
+            }
+            k += m;
+        }
+    }
+}
+
 /* perform the butterfly for one stage of a mixed radix FFT */
 void bfly_generic(
         kiss_fft_cpx * Fout,
@@ -268,6 +305,7 @@ void fft_work(
         case 2: bfly2(Fout,fstride,st,m); break;
         case 3: bfly3(Fout,fstride,st,m); break;
         case 4: bfly4(Fout,fstride,st,m); break;
+        case 5: bfly5(Fout,fstride,st,m); break;
         default: bfly_generic(Fout,fstride,st,m,p); break;
     }
 }
@@ -304,7 +342,6 @@ void * kiss_fft_alloc(int nfft,int inverse_fft)
     st->scratch = (kiss_fft_cpx*)(st->tmpbuf + nfft);
     st->factors = (int*)(st->scratch + nfft); /* just after tmpbuf*/
 
-
     for (i=0;i<nfft;++i) {
         const double pi=3.14159265358979323846264338327;
         double phase = ( 2*pi /nfft ) * i;

test/Makefile

@@ -28,25 +28,26 @@ endif
 $(UTIL): $(UTILSRC)
 	gcc -o $@ $(CFLAGS) $(TYPEFLAGS) $(UTILSRC)
 
-RANDDAT=rand_$(DATATYPE)_$(NUMFFTS)_$(NFFT).dat
+RANDDAT=rand_$(DATATYPE)_$(NFFT).dat
 $(RANDDAT):
-	./rand_fft_data.py -n $(NUMFFTS) -N $(NFFT) -t $(DATATYPE) > $(RANDDAT)
+	./rand_fft_data.py -n 1 -N $(NFFT) -t $(DATATYPE) > $(RANDDAT)
 
 time: all $(RANDDAT)
 	@echo
 	@echo -n "#### timing $(NUMFFTS) x $(NFFT) point FFTs.  "; factor $(NFFT)
 	@if [ -x ~/fftw/st ] && [ $(DATATYPE) == "double" ]; then \
 		echo "#### FFTW FFT $(DATATYPE)"; \
-		time ~/fftw/st -n $(NFFT) < $(RANDDAT) > /dev/null;\
+		time ~/fftw/st -x $(NUMFFTS) -n $(NFFT) < $(RANDDAT) > /dev/null;\
 	fi
 	@echo "#### KISS FFT $(DATATYPE)"
-	@time ./$(UTIL) -n $(NFFT) < $(RANDDAT) > /dev/null
+	@time ./$(UTIL) -x $(NUMFFTS) -n $(NFFT) < $(RANDDAT) > /dev/null
 	@rm $(RANDDAT)
 
 POW2=256 512 1024 2048
 POW3=243 729 2187
+POW5=25 125 625 
 mtime: all
-	@for n in $(POW2) ;do \
+	@for n in $(POW5) ;do \
 		export NFFT=$$n;make time; \
 		done
 

test/speedtest.c

@@ -0,0 +1,56 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <fftw3.h>
+#include <getopt.h>
+
+
+int main(int argc,char ** argv)
+{
+    int nfft=1024;
+    int isinverse=0;
+    FILE *fin=stdin;
+    FILE *fout=stdout;
+    int times=1,i;
+
+    fftw_complex * in=NULL;
+    fftw_complex * out=NULL;
+    fftw_plan p;
+
+    while (1) {
+      int c = getopt (argc, argv, "n:ix:");
+      if (c == -1)
+        break;
+      switch (c) {
+      case 'n':
+        nfft = atoi (optarg);
+        break;
+      case 'x':
+        times = atoi (optarg);
+        break;
+      case 'i':
+        isinverse = 1;
+        break;
+      }
+    }
+    //fprintf(stderr,"sizeof(fftw_complex) = %d \n" , sizeof(fftw_complex) );
+    fprintf(stderr,"sizeof(fftw_complex[0]) = %d \n" , sizeof((*in)[0]) );
+
+    in=fftw_malloc(sizeof(fftw_complex) * nfft);
+    out=fftw_malloc(sizeof(fftw_complex) * nfft);
+    if ( isinverse )
+        p = fftw_plan_dft_1d(nfft, in, out, FFTW_BACKWARD, FFTW_ESTIMATE);
+    else    
+        p = fftw_plan_dft_1d(nfft, in, out, FFTW_FORWARD, FFTW_ESTIMATE);
+
+    while ( fread( in , sizeof(fftw_complex) , nfft , fin ) > 0 ) {
+        for (i=0;i<times;++i)
+            fftw_execute(p);
+        fwrite( out , sizeof(fftw_complex) , nfft , fout );
+    }
+
+    fftw_destroy_plan(p);
+    fftw_free(in); fftw_free(out);
+
+    return 0;
+}
+

tools/Makefile

@@ -28,25 +28,26 @@ endif
 $(UTIL): $(UTILSRC)
 	gcc -o $@ $(CFLAGS) $(TYPEFLAGS) $(UTILSRC)
 
-RANDDAT=rand_$(DATATYPE)_$(NUMFFTS)_$(NFFT).dat
+RANDDAT=rand_$(DATATYPE)_$(NFFT).dat
 $(RANDDAT):
-	./rand_fft_data.py -n $(NUMFFTS) -N $(NFFT) -t $(DATATYPE) > $(RANDDAT)
+	./rand_fft_data.py -n 1 -N $(NFFT) -t $(DATATYPE) > $(RANDDAT)
 
 time: all $(RANDDAT)
 	@echo
 	@echo -n "#### timing $(NUMFFTS) x $(NFFT) point FFTs.  "; factor $(NFFT)
 	@if [ -x ~/fftw/st ] && [ $(DATATYPE) == "double" ]; then \
 		echo "#### FFTW FFT $(DATATYPE)"; \
-		time ~/fftw/st -n $(NFFT) < $(RANDDAT) > /dev/null;\
+		time ~/fftw/st -x $(NUMFFTS) -n $(NFFT) < $(RANDDAT) > /dev/null;\
 	fi
 	@echo "#### KISS FFT $(DATATYPE)"
-	@time ./$(UTIL) -n $(NFFT) < $(RANDDAT) > /dev/null
+	@time ./$(UTIL) -x $(NUMFFTS) -n $(NFFT) < $(RANDDAT) > /dev/null
 	@rm $(RANDDAT)
 
 POW2=256 512 1024 2048
 POW3=243 729 2187
+POW5=25 125 625 
 mtime: all
-	@for n in $(POW2) ;do \
+	@for n in $(POW5) ;do \
 		export NFFT=$$n;make time; \
 		done
 

tools/fftutil.c

@@ -26,16 +26,20 @@ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
 
 
 
-void fft_file(FILE * fin,FILE * fout,int nfft,int isinverse,int useascii)
+void fft_file(FILE * fin,FILE * fout,int nfft,int isinverse,int useascii,int times)
 {
+    int i;
     void *st;
     kiss_fft_cpx * buf;
+    kiss_fft_cpx * bufout;
 
     buf = (kiss_fft_cpx*)malloc(sizeof(kiss_fft_cpx) * nfft );
+    bufout = (kiss_fft_cpx*)malloc(sizeof(kiss_fft_cpx) * nfft );
     st = kiss_fft_alloc( nfft ,isinverse );
 
     while ( fread( buf , sizeof(kiss_fft_cpx) * nfft ,1, fin ) > 0 ) {
-        kiss_fft( st , buf );
+        for (i=0;i<times;++i)
+            kiss_fft_io( st , buf ,bufout);
         if (useascii) {
             int i;
             for (i=0;i<nfft;++i) 
@@ -46,6 +50,7 @@ void fft_file(FILE * fin,FILE * fout,int nfft,int isinverse,int useascii)
     }
     free(st);
     free(buf);
+    free(bufout);
 }
 
 int main(int argc,char ** argv)
@@ -55,14 +60,16 @@ int main(int argc,char ** argv)
     FILE *fin=stdin;
     FILE *fout=stdout;
     int useascii=0;
-        
+    int times=1;
+
     while (1) {
-        int c=getopt(argc,argv,"n:ia");
+        int c=getopt(argc,argv,"n:iax:");
         if (c==-1) break;
         switch (c) {
             case 'a':useascii=1;break;
             case 'n':nfft = atoi(optarg);break;
             case 'i':isinverse=1;break;
+            case 'x':times=atoi(optarg);break;
         }
     }
 
@@ -78,7 +85,7 @@ int main(int argc,char ** argv)
         ++optind;
     }
 
-    fft_file(fin,fout,nfft,isinverse,useascii);
+    fft_file(fin,fout,nfft,isinverse,useascii,times);
 
     if (fout!=stdout) fclose(fout);
     if (fin!=stdin) fclose(fin);