Adding SSE version of fft filters. Complex (ccc) version working.

9 files changed, 53 insertions, 42 deletions

diff --git a/gnuradio-core/src/lib/filter/Makefile.am b/gnuradio-core/src/lib/filter/Makefile.am
index 2d7fb0db8..7ea8c582e 100644
--- a/gnuradio-core/src/lib/filter/Makefile.am
+++ b/gnuradio-core/src/lib/filter/Makefile.am
@@ -184,8 +184,10 @@ libfilter_la_common_SOURCES = 		\
 	$(GENERATED_CC)			\
 	gr_adaptive_fir_ccf.cc		\
 	gr_cma_equalizer_cc.cc		\
-	gri_fft_filter_fff.cc		\
-	gri_fft_filter_ccc.cc		\
+	gri_fft_filter_fff_generic.cc	\
+	gri_fft_filter_ccc_generic.cc	\
+	gri_fft_filter_fff_sse.cc	\
+	gri_fft_filter_ccc_sse.cc	\
 	gr_fft_filter_ccc.cc		\
 	gr_fft_filter_fff.cc		\
 	gr_goertzel_fc.cc		\
@@ -261,8 +263,10 @@ grinclude_HEADERS = 			\
 	gr_altivec.h			\
 	gr_cma_equalizer_cc.h		\
 	gr_cpu.h			\
-	gri_fft_filter_fff.h		\
-	gri_fft_filter_ccc.h		\
+	gri_fft_filter_fff_generic.h	\
+	gri_fft_filter_fff_sse.h	\
+	gri_fft_filter_ccc_generic.h	\
+	gri_fft_filter_ccc_sse.h	\
 	gr_fft_filter_ccc.h		\
 	gr_fft_filter_fff.h		\
 	gr_filter_delay_fc.h		\
diff --git a/gnuradio-core/src/lib/filter/gr_fft_filter_ccc.cc b/gnuradio-core/src/lib/filter/gr_fft_filter_ccc.cc
index b1f04fd0f..ad514ee37 100644
--- a/gnuradio-core/src/lib/filter/gr_fft_filter_ccc.cc
+++ b/gnuradio-core/src/lib/filter/gr_fft_filter_ccc.cc
@@ -30,7 +30,7 @@
 #endif
 
 #include <gr_fft_filter_ccc.h>
-#include <gri_fft_filter_ccc.h>
+#include <gri_fft_filter_ccc_sse.h>
 #include <gr_io_signature.h>
 #include <gri_fft.h>
 #include <math.h>
@@ -56,7 +56,7 @@ gr_fft_filter_ccc::gr_fft_filter_ccc (int decimation, const std::vector<gr_compl
     d_updated(false)
 {
   set_history(1);
-  d_filter = new gri_fft_filter_ccc(decimation, taps);
+  d_filter = new gri_fft_filter_ccc_sse(decimation, taps);
   d_nsamples = d_filter->set_taps(taps);
   set_output_multiple(d_nsamples);
 }
diff --git a/gnuradio-core/src/lib/filter/gr_fft_filter_ccc.h b/gnuradio-core/src/lib/filter/gr_fft_filter_ccc.h
index cfb9ff35d..18c10dfd3 100644
--- a/gnuradio-core/src/lib/filter/gr_fft_filter_ccc.h
+++ b/gnuradio-core/src/lib/filter/gr_fft_filter_ccc.h
@@ -28,7 +28,7 @@ class gr_fft_filter_ccc;
 typedef boost::shared_ptr<gr_fft_filter_ccc> gr_fft_filter_ccc_sptr;
 gr_fft_filter_ccc_sptr gr_make_fft_filter_ccc (int decimation, const std::vector<gr_complex> &taps);
 
-class gri_fft_filter_ccc;
+class gri_fft_filter_ccc_sse;
 
 /*!
  * \brief Fast FFT filter with gr_complex input, gr_complex output and gr_complex taps
@@ -41,7 +41,7 @@ class gr_fft_filter_ccc : public gr_sync_decimator
 
   int			   d_nsamples;
   bool			   d_updated;
-  gri_fft_filter_ccc	  *d_filter;  
+  gri_fft_filter_ccc_sse  *d_filter;  
   std::vector<gr_complex>  d_new_taps;
 
   /*!
diff --git a/gnuradio-core/src/lib/filter/gr_fft_filter_fff.cc b/gnuradio-core/src/lib/filter/gr_fft_filter_fff.cc
index 08c2c022f..eda3d2147 100644
--- a/gnuradio-core/src/lib/filter/gr_fft_filter_fff.cc
+++ b/gnuradio-core/src/lib/filter/gr_fft_filter_fff.cc
@@ -25,7 +25,7 @@
 #endif
 
 #include <gr_fft_filter_fff.h>
-#include <gri_fft_filter_fff.h>
+#include <gri_fft_filter_fff_generic.h>
 #include <gr_io_signature.h>
 #include <assert.h>
 #include <stdexcept>
@@ -49,7 +49,7 @@ gr_fft_filter_fff::gr_fft_filter_fff (int decimation, const std::vector<float> &
 {
   set_history(1);
 
-  d_filter = new gri_fft_filter_fff(decimation, taps);
+  d_filter = new gri_fft_filter_fff_generic(decimation, taps);
   d_nsamples = d_filter->set_taps(taps);
   set_output_multiple(d_nsamples);
 }
diff --git a/gnuradio-core/src/lib/filter/gr_fft_filter_fff.h b/gnuradio-core/src/lib/filter/gr_fft_filter_fff.h
index 04cb3c8f0..6c9fcc04b 100644
--- a/gnuradio-core/src/lib/filter/gr_fft_filter_fff.h
+++ b/gnuradio-core/src/lib/filter/gr_fft_filter_fff.h
@@ -28,7 +28,7 @@ class gr_fft_filter_fff;
 typedef boost::shared_ptr<gr_fft_filter_fff> gr_fft_filter_fff_sptr;
 gr_fft_filter_fff_sptr gr_make_fft_filter_fff (int decimation, const std::vector<float> &taps);
 
-class gri_fft_filter_fff;
+class gri_fft_filter_fff_generic;
 
 /*!
  * \brief Fast FFT filter with float input, float output and float taps
@@ -41,7 +41,7 @@ class gr_fft_filter_fff : public gr_sync_decimator
 
   int			   d_nsamples;
   bool			   d_updated;
-  gri_fft_filter_fff	  *d_filter;
+  gri_fft_filter_fff_generic  *d_filter;
   std::vector<float>	   d_new_taps;
 
   /*!
diff --git a/gnuradio-core/src/lib/filter/gri_fft_filter_ccc.cc b/gnuradio-core/src/lib/filter/gri_fft_filter_ccc_generic.cc
index 4c596630c..91049a92a 100644
--- a/gnuradio-core/src/lib/filter/gri_fft_filter_ccc.cc
+++ b/gnuradio-core/src/lib/filter/gri_fft_filter_ccc_generic.cc
@@ -24,19 +24,22 @@
 #include "config.h"
 #endif
 
-#include <gri_fft_filter_ccc.h>
+#include <gri_fft_filter_ccc_generic.h>
 #include <gri_fft.h>
 #include <assert.h>
 #include <stdexcept>
 #include <cstdio>
+#include <xmmintrin.h>
+#include <fftw3.h>
 
-gri_fft_filter_ccc::gri_fft_filter_ccc (int decimation, const std::vector<gr_complex> &taps)
+gri_fft_filter_ccc_generic::gri_fft_filter_ccc_generic (int decimation, 
+							const std::vector<gr_complex> &taps)
   : d_fftsize(-1), d_decimation(decimation), d_fwdfft(0), d_invfft(0)
 {
   set_taps(taps);
 }
 
-gri_fft_filter_ccc::~gri_fft_filter_ccc ()
+gri_fft_filter_ccc_generic::~gri_fft_filter_ccc_generic ()
 {
   delete d_fwdfft;
   delete d_invfft;
@@ -58,7 +61,7 @@ print_vector_complex(const std::string label, const std::vector<gr_complex> &x)
  * determines d_ntaps, d_nsamples, d_fftsize, d_xformed_taps
  */
 int
-gri_fft_filter_ccc::set_taps (const std::vector<gr_complex> &taps)
+gri_fft_filter_ccc_generic::set_taps (const std::vector<gr_complex> &taps)
 {
   int i = 0;
   compute_sizes(taps.size());
@@ -92,7 +95,7 @@ gri_fft_filter_ccc::set_taps (const std::vector<gr_complex> &taps)
 // determine and set d_ntaps, d_nsamples, d_fftsize
 
 void
-gri_fft_filter_ccc::compute_sizes(int ntaps)
+gri_fft_filter_ccc_generic::compute_sizes(int ntaps)
 {
   int old_fftsize = d_fftsize;
   d_ntaps = ntaps;
@@ -100,7 +103,7 @@ gri_fft_filter_ccc::compute_sizes(int ntaps)
   d_nsamples = d_fftsize - d_ntaps + 1;
 
   if (0)
-    fprintf(stderr, "gri_fft_filter_ccc: ntaps = %d, fftsize = %d, nsamples = %d\n",
+    fprintf(stderr, "gri_fft_filter_ccc_generic: ntaps = %d, fftsize = %d, nsamples = %d\n",
 	    d_ntaps, d_fftsize, d_nsamples);
 
   assert(d_fftsize == d_ntaps + d_nsamples -1 );
@@ -115,7 +118,7 @@ gri_fft_filter_ccc::compute_sizes(int ntaps)
 }
 
 int
-gri_fft_filter_ccc::filter (int nitems, const gr_complex *input, gr_complex *output)
+gri_fft_filter_ccc_generic::filter (int nitems, const gr_complex *input, gr_complex *output)
 {
   int dec_ctr = 0;
   int j = 0;
@@ -129,13 +132,14 @@ gri_fft_filter_ccc::filter (int nitems, const gr_complex *input, gr_complex *out
       d_fwdfft->get_inbuf()[j] = 0;
 
     d_fwdfft->execute();	// compute fwd xform
-
+    
     gr_complex *a = d_fwdfft->get_outbuf();
     gr_complex *b = &d_xformed_taps[0];
     gr_complex *c = d_invfft->get_inbuf();
 
-    for (j = 0; j < d_fftsize; j++)	// filter in the freq domain
+    for (j = 0; j < d_fftsize; j+=1) {	// filter in the freq domain
       c[j] = a[j] * b[j];
+    } 
     
     d_invfft->execute();	// compute inv xform
 
diff --git a/gnuradio-core/src/lib/filter/gri_fft_filter_ccc.h b/gnuradio-core/src/lib/filter/gri_fft_filter_ccc_generic.h
index a857e5ab1..3cd9105c7 100644
--- a/gnuradio-core/src/lib/filter/gri_fft_filter_ccc.h
+++ b/gnuradio-core/src/lib/filter/gri_fft_filter_ccc_generic.h
@@ -20,8 +20,8 @@
  * Boston, MA 02110-1301, USA.
  */
 
-#ifndef INCLUDED_GRI_FFT_FILTER_CCC_H
-#define INCLUDED_GRI_FFT_FILTER_CCC_H
+#ifndef INCLUDED_GRI_FFT_FILTER_CCC_GENERIC_H
+#define INCLUDED_GRI_FFT_FILTER_CCC_GENERIC_H
 
 #include <gr_complex.h>
 #include <vector>
@@ -32,7 +32,7 @@ class gri_fft_complex;
  * \brief Fast FFT filter with gr_complex input, gr_complex output and gr_complex taps
  * \ingroup filter_blk
  */
-class gri_fft_filter_ccc
+class gri_fft_filter_ccc_generic
 {
  private:
   int			   d_ntaps;
@@ -57,8 +57,8 @@ class gri_fft_filter_ccc
    * \param decimation The decimation rate of the filter (int)
    * \param taps       The filter taps (complex)
    */
-  gri_fft_filter_ccc (int decimation, const std::vector<gr_complex> &taps);
-  ~gri_fft_filter_ccc ();
+  gri_fft_filter_ccc_generic (int decimation, const std::vector<gr_complex> &taps);
+  ~gri_fft_filter_ccc_generic ();
 
   /*!
    * \brief Set new taps for the filter.
@@ -79,4 +79,4 @@ class gri_fft_filter_ccc
 
 };
 
-#endif /* INCLUDED_GRI_FFT_FILTER_CCC_H */
+#endif /* INCLUDED_GRI_FFT_FILTER_CCC_GENERIC_H */
diff --git a/gnuradio-core/src/lib/filter/gri_fft_filter_fff.cc b/gnuradio-core/src/lib/filter/gri_fft_filter_fff_generic.cc
index 246d876b2..5a0916669 100644
--- a/gnuradio-core/src/lib/filter/gri_fft_filter_fff.cc
+++ b/gnuradio-core/src/lib/filter/gri_fft_filter_fff_generic.cc
@@ -24,19 +24,20 @@
 #include "config.h"
 #endif
 
-#include <gri_fft_filter_fff.h>
+#include <gri_fft_filter_fff_generic.h>
 #include <gri_fft.h>
 #include <assert.h>
 #include <stdexcept>
 #include <cstdio>
 
-gri_fft_filter_fff::gri_fft_filter_fff (int decimation, const std::vector<float> &taps)
+gri_fft_filter_fff_generic::gri_fft_filter_fff_generic (int decimation, 
+							const std::vector<float> &taps)
   : d_fftsize(-1), d_decimation(decimation), d_fwdfft(0), d_invfft(0)
 {
   set_taps(taps);
 }
 
-gri_fft_filter_fff::~gri_fft_filter_fff ()
+gri_fft_filter_fff_generic::~gri_fft_filter_fff_generic ()
 {
   delete d_fwdfft;
   delete d_invfft;
@@ -46,7 +47,7 @@ gri_fft_filter_fff::~gri_fft_filter_fff ()
  * determines d_ntaps, d_nsamples, d_fftsize, d_xformed_taps
  */
 int
-gri_fft_filter_fff::set_taps (const std::vector<float> &taps)
+gri_fft_filter_fff_generic::set_taps (const std::vector<float> &taps)
 {
   int i = 0;
   compute_sizes(taps.size());
@@ -80,7 +81,7 @@ gri_fft_filter_fff::set_taps (const std::vector<float> &taps)
 // determine and set d_ntaps, d_nsamples, d_fftsize
 
 void
-gri_fft_filter_fff::compute_sizes(int ntaps)
+gri_fft_filter_fff_generic::compute_sizes(int ntaps)
 {
   int old_fftsize = d_fftsize;
   d_ntaps = ntaps;
@@ -88,7 +89,7 @@ gri_fft_filter_fff::compute_sizes(int ntaps)
   d_nsamples = d_fftsize - d_ntaps + 1;
 
   if (0)
-    fprintf(stderr, "gri_fft_filter_fff: ntaps = %d, fftsize = %d, nsamples = %d\n",
+    fprintf(stderr, "gri_fft_filter_fff_generic: ntaps = %d, fftsize = %d, nsamples = %d\n",
 	    d_ntaps, d_fftsize, d_nsamples);
 
   assert(d_fftsize == d_ntaps + d_nsamples -1 );
@@ -103,7 +104,7 @@ gri_fft_filter_fff::compute_sizes(int ntaps)
 }
 
 int
-gri_fft_filter_fff::filter (int nitems, const float *input, float *output)
+gri_fft_filter_fff_generic::filter (int nitems, const float *input, float *output)
 {
   int dec_ctr = 0;
   int j = 0;
@@ -122,9 +123,10 @@ gri_fft_filter_fff::filter (int nitems, const float *input, float *output)
     gr_complex *b = &d_xformed_taps[0];
     gr_complex *c = d_invfft->get_inbuf();
 
-    for (j = 0; j < d_fftsize/2+1; j++)	// filter in the freq domain
+    for (j = 0; j < d_fftsize/2+1; j++) {	// filter in the freq domain
       c[j] = a[j] * b[j];
-    
+    }      
+   
     d_invfft->execute();	// compute inv xform
 
     // add in the overlapping tail
diff --git a/gnuradio-core/src/lib/filter/gri_fft_filter_fff.h b/gnuradio-core/src/lib/filter/gri_fft_filter_fff_generic.h
index 2f6b7e45a..6c31632d5 100644
--- a/gnuradio-core/src/lib/filter/gri_fft_filter_fff.h
+++ b/gnuradio-core/src/lib/filter/gri_fft_filter_fff_generic.h
@@ -20,8 +20,8 @@
  * Boston, MA 02110-1301, USA.
  */
 
-#ifndef INCLUDED_GRI_FFT_FILTER_FFF_H
-#define INCLUDED_GRI_FFT_FILTER_FFF_H
+#ifndef INCLUDED_GRI_FFT_FILTER_FFF_GENERIC_H
+#define INCLUDED_GRI_FFT_FILTER_FFF_GENERIC_H
 
 #include <gr_complex.h>
 #include <vector>
@@ -29,7 +29,7 @@
 class gri_fft_real_fwd;
 class gri_fft_real_rev;
 
-class gri_fft_filter_fff
+class gri_fft_filter_fff_generic
 {
  private:
   int			   d_ntaps;
@@ -42,6 +42,7 @@ class gri_fft_filter_fff
   std::vector<gr_complex>  d_xformed_taps;	// Fourier xformed taps
   std::vector<float>	   d_new_taps;
 
+
   void compute_sizes(int ntaps);
   int tailsize() const { return d_ntaps - 1; }
   
@@ -54,8 +55,8 @@ class gri_fft_filter_fff
    * \param decimation The decimation rate of the filter (int)
    * \param taps       The filter taps (float)
    */
-  gri_fft_filter_fff (int decimation, const std::vector<float> &taps);
-  ~gri_fft_filter_fff ();
+  gri_fft_filter_fff_generic (int decimation, const std::vector<float> &taps);
+  ~gri_fft_filter_fff_generic ();
 
   /*!
    * \brief Set new taps for the filter.
@@ -76,4 +77,4 @@ class gri_fft_filter_fff
 
 };
 
-#endif /* INCLUDED_GRI_FFT_FILTER_FFF_H */
+#endif /* INCLUDED_GRI_FFT_FILTER_FFF_GENERIC_H */