found out about little-documented FFTW_PRESERVE_INPUT which has effect on c2r transforms

3 files changed, 68 insertions, 54 deletions

diff --git a/unsupported/Eigen/FFT b/unsupported/Eigen/FFT
index 97b1b2cdf..454308fb7 100644
--- a/unsupported/Eigen/FFT
+++ b/unsupported/Eigen/FFT
@@ -152,18 +152,20 @@ class FFT
         m_impl.fwd(dst,src,nfft);
     }
 
+    /*
     inline 
-    void fwd2(Complex * dst, const Complex * src, int nrows,int ncols)
+    void fwd2(Complex * dst, const Complex * src, int n0,int n1)
     {
-      m_impl.fwd2(dst,src,nrows,ncols);
+      m_impl.fwd2(dst,src,n0,n1);
     }
+    */
 
     template <typename _Input>
     inline
     void fwd( std::vector<Complex> & dst, const std::vector<_Input> & src) 
     {
       if ( NumTraits<_Input>::IsComplex == 0 && HasFlag(HalfSpectrum) )
-        dst.resize( (src.size()>>1)+1);
+        dst.resize( (src.size()>>1)+1); // half the bins + Nyquist bin
       else
         dst.resize(src.size());
       fwd(&dst[0],&src[0],static_cast<int>(src.size()));
@@ -197,22 +199,22 @@ class FFT
     inline
     void inv( Complex * dst, const Complex * src, int nfft)
     {
-        m_impl.inv( dst,src,nfft );
-        if ( HasFlag( Unscaled ) == false)
-          scale(dst,1./nfft,nfft);
+      m_impl.inv( dst,src,nfft );
+      if ( HasFlag( Unscaled ) == false)
+        scale(dst,1./nfft,nfft); // scale the time series
     }
 
     inline
     void inv( Scalar * dst, const Complex * src, int nfft)
     {
-        m_impl.inv( dst,src,nfft );
-        if ( HasFlag( Unscaled ) == false)
-          scale(dst,1./nfft,nfft);
+      m_impl.inv( dst,src,nfft );
+      if ( HasFlag( Unscaled ) == false)
+        scale(dst,1./nfft,nfft); // scale the time series
     }
 
     template<typename OutputDerived, typename ComplexDerived>
     inline
-    void inv( MatrixBase<OutputDerived> & dst, const MatrixBase<ComplexDerived> & src)
+    void inv( MatrixBase<OutputDerived> & dst, const MatrixBase<ComplexDerived> & src, int nfft=-1)
     {
         EIGEN_STATIC_ASSERT_VECTOR_ONLY(OutputDerived)
         EIGEN_STATIC_ASSERT_VECTOR_ONLY(ComplexDerived)
@@ -222,10 +224,12 @@ class FFT
         EIGEN_STATIC_ASSERT(int(OutputDerived::Flags)&int(ComplexDerived::Flags)&DirectAccessBit,
                             THIS_METHOD_IS_ONLY_FOR_EXPRESSIONS_WITH_DIRECT_MEMORY_ACCESS_SUCH_AS_MAP_OR_PLAIN_MATRICES)
 
-        int nfft = src.size();
-        int nout = HasFlag(HalfSpectrum) ? ((nfft>>1)+1) : nfft;
-        dst.derived().resize( nout );
+        if (nfft<1) {
+          nfft = ( NumTraits<typename OutputDerived::Scalar>::IsComplex == 0 && HasFlag(HalfSpectrum) ) ? 2*(src.size()-1) : src.size();
+        }
+        dst.derived().resize( nfft );
         if (src.stride() != 1) {
+          // if the vector is strided, then we need to copy it to a packed temporary
           Matrix<typename ComplexDerived::Scalar,1,Dynamic> tmp = src;
           inv( &dst[0],&tmp[0], nfft);
         }else{
@@ -235,25 +239,25 @@ class FFT
 
     template <typename _Output>
     inline
-    void inv( std::vector<_Output> & dst, const std::vector<Complex> & src)
+    void inv( std::vector<_Output> & dst, const std::vector<Complex> & src,int nfft=-1)
     {
-      if ( NumTraits<_Output>::IsComplex == 0 && HasFlag(HalfSpectrum) )
-        dst.resize( 2*(src.size()-1) );
-      else
-        dst.resize( src.size() );
-      inv( &dst[0],&src[0],static_cast<int>(dst.size()) );
+      if (nfft<1)
+        nfft = ( NumTraits<_Output>::IsComplex == 0 && HasFlag(HalfSpectrum) ) ? 2*(src.size()-1) : src.size();
+      dst.resize( nfft );
+      inv( &dst[0],&src[0],nfft);
     }
 
 
+    /*
+    // TODO: multi-dimensional FFTs
     inline 
-    void inv2(Complex * dst, const Complex * src, int nrows,int ncols)
+    void inv2(Complex * dst, const Complex * src, int n0,int n1)
     {
-      m_impl.inv2(dst,src,nrows,ncols);
+      m_impl.inv2(dst,src,n0,n1);
       if ( HasFlag( Unscaled ) == false)
-          scale(dst,1./(nrows*ncols),nrows*ncols);
+          scale(dst,1./(n0*n1),n0*n1);
     }
-
-    // TODO: multi-dimensional FFTs
+  */
 
     inline
     impl_type & impl() {return m_impl;}
diff --git a/unsupported/Eigen/src/FFT/ei_fftw_impl.h b/unsupported/Eigen/src/FFT/ei_fftw_impl.h
index 411ff7425..c1f777e6d 100644
--- a/unsupported/Eigen/src/FFT/ei_fftw_impl.h
+++ b/unsupported/Eigen/src/FFT/ei_fftw_impl.h
@@ -71,34 +71,34 @@
 
       inline
       void fwd(complex_type * dst,complex_type * src,int nfft) {
-          if (m_plan==NULL) m_plan = fftwf_plan_dft_1d(nfft,src,dst, FFTW_FORWARD, FFTW_ESTIMATE);
+          if (m_plan==NULL) m_plan = fftwf_plan_dft_1d(nfft,src,dst, FFTW_FORWARD, FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftwf_execute_dft( m_plan, src,dst);
       }
       inline
       void inv(complex_type * dst,complex_type * src,int nfft) {
-          if (m_plan==NULL) m_plan = fftwf_plan_dft_1d(nfft,src,dst, FFTW_BACKWARD , FFTW_ESTIMATE);
+          if (m_plan==NULL) m_plan = fftwf_plan_dft_1d(nfft,src,dst, FFTW_BACKWARD , FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftwf_execute_dft( m_plan, src,dst);
       }
       inline
       void fwd(complex_type * dst,scalar_type * src,int nfft) {
-          if (m_plan==NULL) m_plan = fftwf_plan_dft_r2c_1d(nfft,src,dst,FFTW_ESTIMATE);
+          if (m_plan==NULL) m_plan = fftwf_plan_dft_r2c_1d(nfft,src,dst,FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftwf_execute_dft_r2c( m_plan,src,dst);
       }
       inline
       void inv(scalar_type * dst,complex_type * src,int nfft) {
           if (m_plan==NULL)
-              m_plan = fftwf_plan_dft_c2r_1d(nfft,src,dst,FFTW_ESTIMATE);
+              m_plan = fftwf_plan_dft_c2r_1d(nfft,src,dst,FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftwf_execute_dft_c2r( m_plan, src,dst);
       }
 
       inline 
-      void fwd2( complex_type * dst,complex_type * src,int nrows,int ncols) {
-          if (m_plan==NULL) m_plan = fftwf_plan_dft_2d(ncols,nrows,src,dst,FFTW_FORWARD,FFTW_ESTIMATE);
+      void fwd2( complex_type * dst,complex_type * src,int n0,int n1) {
+          if (m_plan==NULL) m_plan = fftwf_plan_dft_2d(n0,n1,src,dst,FFTW_FORWARD,FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftwf_execute_dft( m_plan, src,dst);
       }
       inline 
-      void inv2( complex_type * dst,complex_type * src,int nrows,int ncols) {
-          if (m_plan==NULL) m_plan = fftwf_plan_dft_2d(ncols,nrows,src,dst,FFTW_BACKWARD,FFTW_ESTIMATE);
+      void inv2( complex_type * dst,complex_type * src,int n0,int n1) {
+          if (m_plan==NULL) m_plan = fftwf_plan_dft_2d(n0,n1,src,dst,FFTW_BACKWARD,FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftwf_execute_dft( m_plan, src,dst);
       }
 
@@ -114,33 +114,33 @@
 
       inline
       void fwd(complex_type * dst,complex_type * src,int nfft) {
-          if (m_plan==NULL) m_plan = fftw_plan_dft_1d(nfft,src,dst, FFTW_FORWARD, FFTW_ESTIMATE);
+          if (m_plan==NULL) m_plan = fftw_plan_dft_1d(nfft,src,dst, FFTW_FORWARD, FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftw_execute_dft( m_plan, src,dst);
       }
       inline
       void inv(complex_type * dst,complex_type * src,int nfft) {
-          if (m_plan==NULL) m_plan = fftw_plan_dft_1d(nfft,src,dst, FFTW_BACKWARD , FFTW_ESTIMATE);
+          if (m_plan==NULL) m_plan = fftw_plan_dft_1d(nfft,src,dst, FFTW_BACKWARD , FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftw_execute_dft( m_plan, src,dst);
       }
       inline
       void fwd(complex_type * dst,scalar_type * src,int nfft) {
-          if (m_plan==NULL) m_plan = fftw_plan_dft_r2c_1d(nfft,src,dst,FFTW_ESTIMATE);
+          if (m_plan==NULL) m_plan = fftw_plan_dft_r2c_1d(nfft,src,dst,FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftw_execute_dft_r2c( m_plan,src,dst);
       }
       inline
       void inv(scalar_type * dst,complex_type * src,int nfft) {
           if (m_plan==NULL)
-              m_plan = fftw_plan_dft_c2r_1d(nfft,src,dst,FFTW_ESTIMATE);
+              m_plan = fftw_plan_dft_c2r_1d(nfft,src,dst,FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftw_execute_dft_c2r( m_plan, src,dst);
       }
       inline 
-      void fwd2( complex_type * dst,complex_type * src,int nrows,int ncols) {
-          if (m_plan==NULL) m_plan = fftw_plan_dft_2d(ncols,nrows,src,dst,FFTW_FORWARD,FFTW_ESTIMATE);
+      void fwd2( complex_type * dst,complex_type * src,int n0,int n1) {
+          if (m_plan==NULL) m_plan = fftw_plan_dft_2d(n0,n1,src,dst,FFTW_FORWARD,FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftw_execute_dft( m_plan, src,dst);
       }
       inline 
-      void inv2( complex_type * dst,complex_type * src,int nrows,int ncols) {
-          if (m_plan==NULL) m_plan = fftw_plan_dft_2d(ncols,nrows,src,dst,FFTW_BACKWARD,FFTW_ESTIMATE);
+      void inv2( complex_type * dst,complex_type * src,int n0,int n1) {
+          if (m_plan==NULL) m_plan = fftw_plan_dft_2d(n0,n1,src,dst,FFTW_BACKWARD,FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftw_execute_dft( m_plan, src,dst);
       }
   };
@@ -155,33 +155,33 @@
 
       inline
       void fwd(complex_type * dst,complex_type * src,int nfft) {
-          if (m_plan==NULL) m_plan = fftwl_plan_dft_1d(nfft,src,dst, FFTW_FORWARD, FFTW_ESTIMATE);
+          if (m_plan==NULL) m_plan = fftwl_plan_dft_1d(nfft,src,dst, FFTW_FORWARD, FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftwl_execute_dft( m_plan, src,dst);
       }
       inline
       void inv(complex_type * dst,complex_type * src,int nfft) {
-          if (m_plan==NULL) m_plan = fftwl_plan_dft_1d(nfft,src,dst, FFTW_BACKWARD , FFTW_ESTIMATE);
+          if (m_plan==NULL) m_plan = fftwl_plan_dft_1d(nfft,src,dst, FFTW_BACKWARD , FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftwl_execute_dft( m_plan, src,dst);
       }
       inline
       void fwd(complex_type * dst,scalar_type * src,int nfft) {
-          if (m_plan==NULL) m_plan = fftwl_plan_dft_r2c_1d(nfft,src,dst,FFTW_ESTIMATE);
+          if (m_plan==NULL) m_plan = fftwl_plan_dft_r2c_1d(nfft,src,dst,FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftwl_execute_dft_r2c( m_plan,src,dst);
       }
       inline
       void inv(scalar_type * dst,complex_type * src,int nfft) {
           if (m_plan==NULL)
-              m_plan = fftwl_plan_dft_c2r_1d(nfft,src,dst,FFTW_ESTIMATE);
+              m_plan = fftwl_plan_dft_c2r_1d(nfft,src,dst,FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftwl_execute_dft_c2r( m_plan, src,dst);
       }
       inline 
-      void fwd2( complex_type * dst,complex_type * src,int nrows,int ncols) {
-          if (m_plan==NULL) m_plan = fftwl_plan_dft_2d(ncols,nrows,src,dst,FFTW_FORWARD,FFTW_ESTIMATE);
+      void fwd2( complex_type * dst,complex_type * src,int n0,int n1) {
+          if (m_plan==NULL) m_plan = fftwl_plan_dft_2d(n0,n1,src,dst,FFTW_FORWARD,FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftwl_execute_dft( m_plan, src,dst);
       }
       inline 
-      void inv2( complex_type * dst,complex_type * src,int nrows,int ncols) {
-          if (m_plan==NULL) m_plan = fftwl_plan_dft_2d(ncols,nrows,src,dst,FFTW_BACKWARD,FFTW_ESTIMATE);
+      void inv2( complex_type * dst,complex_type * src,int n0,int n1) {
+          if (m_plan==NULL) m_plan = fftwl_plan_dft_2d(n0,n1,src,dst,FFTW_BACKWARD,FFTW_ESTIMATE|FFTW_PRESERVE_INPUT);
           fftwl_execute_dft( m_plan, src,dst);
       }
   };
@@ -214,9 +214,9 @@
 
       // 2-d complex-to-complex
       inline
-      void fwd2(Complex * dst, const Complex * src, int nrows,int ncols)
+      void fwd2(Complex * dst, const Complex * src, int n0,int n1)
       {
-          get_plan(nrows,ncols,false,dst,src).fwd2(ei_fftw_cast(dst), ei_fftw_cast(src) ,nrows,ncols);
+          get_plan(n0,n1,false,dst,src).fwd2(ei_fftw_cast(dst), ei_fftw_cast(src) ,n0,n1);
       }
 
       // inverse complex-to-complex
@@ -235,9 +235,9 @@
 
       // 2-d complex-to-complex
       inline
-      void inv2(Complex * dst, const Complex * src, int nrows,int ncols)
+      void inv2(Complex * dst, const Complex * src, int n0,int n1)
       {
-        get_plan(nrows,ncols,true,dst,src).inv2(ei_fftw_cast(dst), ei_fftw_cast(src) ,nrows,ncols);
+        get_plan(n0,n1,true,dst,src).inv2(ei_fftw_cast(dst), ei_fftw_cast(src) ,n0,n1);
       }
 
 
@@ -258,11 +258,11 @@
       }
 
       inline
-      PlanData & get_plan(int nrows,int ncols,bool inverse,void * dst,const void * src)
+      PlanData & get_plan(int n0,int n1,bool inverse,void * dst,const void * src)
       {
           bool inplace = (dst==src);
           bool aligned = ( (reinterpret_cast<size_t>(src)&15) | (reinterpret_cast<size_t>(dst)&15) ) == 0;
-          int64_t key = ( ( (((int64_t)ncols) << 30)|(nrows<<3 ) | (inverse<<2) | (inplace<<1) | aligned ) << 1 ) + 1;
+          int64_t key = ( ( (((int64_t)n0) << 30)|(n1<<3 ) | (inverse<<2) | (inplace<<1) | aligned ) << 1 ) + 1;
           return m_plans[key];
       }
   };
diff --git a/unsupported/Eigen/src/FFT/ei_kissfft_impl.h b/unsupported/Eigen/src/FFT/ei_kissfft_impl.h
index dbd92132e..5db1bf37d 100644
--- a/unsupported/Eigen/src/FFT/ei_kissfft_impl.h
+++ b/unsupported/Eigen/src/FFT/ei_kissfft_impl.h
@@ -291,6 +291,16 @@ struct ei_kissfft_impl
       get_plan(nfft,false).work(0, dst, src, 1,1);
     }
 
+  inline
+    void fwd2( Complex * dst,const Complex *src,int n0,int n1)
+    {
+    }
+
+  inline
+    void inv2( Complex * dst,const Complex *src,int n0,int n1)
+    {
+    }
+
   // real-to-complex forward FFT
   // perform two FFTs of src even and src odd
   // then twiddle to recombine them into the half-spectrum format