FFTConvolution.h

Go to the documentation of this file.

//     ____   ______ __
//    / __ \ / ____// /
//   / /_/ // /    / /
//  / ____// /___ / /___   PixInsight Class Library
// /_/     \____//_____/   PCL 2.7.0
// ----------------------------------------------------------------------------
// pcl/FFTConvolution.h - Released 2024-06-18T15:48:54Z
// ----------------------------------------------------------------------------
// This file is part of the PixInsight Class Library (PCL).
// PCL is a multiplatform C++ framework for development of PixInsight modules.
//
// Copyright (c) 2003-2024 Pleiades Astrophoto S.L. All Rights Reserved.
//
// Redistribution and use in both source and binary forms, with or without
// modification, is permitted provided that the following conditions are met:
//
// 1. All redistributions of source code must retain the above copyright
//    notice, this list of conditions and the following disclaimer.
//
// 2. All 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.
//
// 3. Neither the names "PixInsight" and "Pleiades Astrophoto", nor the names
//    of their contributors, may be used to endorse or promote products derived
//    from this software without specific prior written permission. For written
//    permission, please contact info@pixinsight.com.
//
// 4. All products derived from this software, in any form whatsoever, must
//    reproduce the following acknowledgment in the end-user documentation
//    and/or other materials provided with the product:
//
//    "This product is based on software from the PixInsight project, developed
//    by Pleiades Astrophoto and its contributors (https://pixinsight.com/)."
//
//    Alternatively, if that is where third-party acknowledgments normally
//    appear, this acknowledgment must be reproduced in the product itself.
//
// THIS SOFTWARE IS PROVIDED BY PLEIADES ASTROPHOTO AND ITS 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 PLEIADES ASTROPHOTO OR ITS
// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
// EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, BUSINESS
// INTERRUPTION; PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; AND LOSS OF USE,
// DATA OR PROFITS) 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.
// ----------------------------------------------------------------------------
 
#ifndef __PCL_FFTConvolution_h
#define __PCL_FFTConvolution_h
 
 
#include <pcl/Defs.h>
#include <pcl/Diagnostics.h>
 
#include <pcl/AutoPointer.h>
#include <pcl/ImageTransformation.h>
#include <pcl/KernelFilter.h>
#include <pcl/ParallelProcess.h>
 
namespace pcl
{
 
// ----------------------------------------------------------------------------
 
class PCL_CLASS FFTConvolution : public ImageTransformation, public ParallelProcess
{
public:
 
   FFTConvolution() = default;
 
   FFTConvolution( const KernelFilter& filter )
      : m_filter( filter.Clone() )
   {
      PCL_CHECK( bool( m_filter ) )
   }
 
   FFTConvolution( const ImageVariant& f )
      : m_image( f )
   {
      PCL_CHECK( bool( m_image ) )
   }
 
   FFTConvolution( const FFTConvolution& x )
      : ImageTransformation( x )
      , ParallelProcess( x )
      , m_image( x.m_image )
      , m_outputRealCmp( x.m_outputRealCmp )
   {
      if ( x.m_filter )
         m_filter = x.m_filter->Clone();
   }
 
   FFTConvolution( FFTConvolution&& ) = default;
 
   ~FFTConvolution() override
   {
   }
 
   FFTConvolution& operator =( const FFTConvolution& x )
   {
      if ( &x != this )
      {
         (void)ImageTransformation::operator =( x );
         (void)ParallelProcess::operator =( x );
         DestroyFilter();
         if ( x.m_filter )
            m_filter = x.m_filter->Clone();
         else
            m_image = x.m_image;
         m_outputRealCmp = x.m_outputRealCmp;
      }
      return *this;
   }
 
   FFTConvolution& operator =( FFTConvolution&& ) = default;
 
   bool UsingFilter() const
   {
      return m_filter;
   }
 
   const KernelFilter& Filter() const
   {
      PCL_PRECONDITION( bool( m_filter ) )
      return *m_filter;
   }
 
   const ImageVariant& FilterImage() const
   {
      return m_image;
   }
 
   void SetFilter( const KernelFilter& filter )
   {
      DestroyFilter();
      m_filter = filter.Clone();
      PCL_CHECK( bool( m_filter ) )
   }
 
   void SetFilter( const ImageVariant& filter )
   {
      DestroyFilter();
      m_image = filter;
      PCL_CHECK( bool( m_image ) )
   }
 
   const ComplexImage* ResponseFunctionDFT() const
   {
      return m_h;
   }
 
   bool IsRealComponentOutputEnabled() const
   {
      return m_outputRealCmp;
   }
 
   bool IsMagnitudeOutputEnabled() const
   {
      return !m_outputRealCmp;
   }
 
   void EnableRealComponentOutput( bool enable = true )
   {
      m_outputRealCmp = enable;
   }
 
   void EnableMagnitudeOutput( bool enable = true )
   {
      m_outputRealCmp = !enable;
   }
 
   static constexpr int FasterThanNonseparableFilterSize( int numThreads )
   {
#ifdef __PCL_COMPATIBILITY
 
      // No vectorization
 
      if ( numThreads >= 32 )
         return 17;
      if ( numThreads >= 24 )
         return 15;
      if ( numThreads >= 12 )
         return 13;
      if ( numThreads >= 8 )
         return 11;
      return 9;
 
#else
 
      // Vectorization with SSE4.2 / AVX2 and FMA instructions
 
      if ( numThreads >= 32 )
         return 29;
      if ( numThreads >= 24 )
         return 27;
      if ( numThreads >= 16 )
         return 25;
      if ( numThreads >= 12 )
         return 21;
      if ( numThreads >= 8 )
         return 19;
      if ( numThreads >= 4 )
         return 17;
      if ( numThreads >= 2 )
         return 15;
      return 13;
 
#endif
   }
 
   static constexpr int FasterThanSeparableFilterSize( int numThreads )
   {
#ifdef __PCL_COMPATIBILITY
 
      // No vectorization
 
      if ( numThreads >= 24 )
         return 191;
      if ( numThreads >= 16 )
         return 141;
      if ( numThreads >= 12 )
         return 135;
      if ( numThreads >= 8 )
         return 95;
      if ( numThreads >= 4 )
         return 75;
      return 61;
 
#else
 
      // Vectorization with SSE4.2 / AVX2 and FMA instructions
 
      if ( numThreads >= 16 )
         return 901;
      if ( numThreads >= 12 )
         return 831;
      if ( numThreads >= 8 )
         return 741;
      return 395;
 
#endif
   }
 
protected:
 
   /*
    * The response function for convolution is defined as either a KernelFilter
    * instance or as an image. In the latter case, the image transported by the
    * ImageVariant must remain valid while this object is actively used.
    */
   AutoPointer<KernelFilter> m_filter;
   ImageVariant              m_image;
   bool                      m_outputRealCmp = false;
 
   /*
    * Internal DFT of the response function. Initially zero. This matrix is
    * generated/reused/regenerated as this object is applied to convolve
    * different target images. It is destroyed when a new filter is specified.
    */
   mutable AutoPointer<ComplexImage> m_h;
 
   /*
    * In-place Fourier-based 2-D convolution algorithm.
    */
   void Apply( pcl::Image& ) const override;
   void Apply( pcl::DImage& ) const override;
   void Apply( pcl::UInt8Image& ) const override;
   void Apply( pcl::UInt16Image& ) const override;
   void Apply( pcl::UInt32Image& ) const override;
   void Apply( pcl::ComplexImage& ) const override;
   void Apply( pcl::DComplexImage& ) const override;
 
   friend class PCL_FFTConvolutionEngine;
 
private:
 
   void DestroyFilter()
   {
      m_filter.Destroy();
      m_image.Free();
      m_h.Destroy();
   }
 
   void Validate() const;
   void ValidateFilter() const;
};
 
// ----------------------------------------------------------------------------
 
} // pcl
 
#endif   // __PCL_FFTConvolution_h
 
// ----------------------------------------------------------------------------
// EOF pcl/FFTConvolution.h - Released 2024-06-18T15:48:54Z