SSETypes.h

Go to the documentation of this file.

/********************************************************************
**                Image Component Library (ICL)                    **
**                                                                 **
** Copyright (C) 2006-2013 CITEC, University of Bielefeld          **
**                         Neuroinformatics Group                  **
** Website: www.iclcv.org and                                      **
**          http://opensource.cit-ec.de/projects/icl               **
**                                                                 **
** File   : ICLUtils/src/ICLUtils/SSETypes.h                       **
** Module : ICLUtils                                               **
** Authors: Sergius Gaulik                                         **
**                                                                 **
**                                                                 **
** GNU LESSER GENERAL PUBLIC LICENSE                               **
** This file may be used under the terms of the GNU Lesser General **
** Public License version 3.0 as published by the                  **
**                                                                 **
** Free Software Foundation and appearing in the file LICENSE.LGPL **
** included in the packaging of this file.  Please review the      **
** following information to ensure the license requirements will   **
** be met: http://www.gnu.org/licenses/lgpl-3.0.txt                **
**                                                                 **
** The development of this software was supported by the           **
** Excellence Cluster EXC 277 Cognitive Interaction Technology.    **
** The Excellence Cluster EXC 277 is a grant of the Deutsche       **
** Forschungsgemeinschaft (DFG) in the context of the German       **
** Excellence Initiative.                                          **
**                                                                 **
********************************************************************/

#pragma once

#if defined ICL_USE_SSE2 && (__SSE2__ || defined _M_X64  || (defined _M_IX86_FP && _M_IX86_FP >= 2))
        #include "emmintrin.h"
        #define ICL_HAVE_SSE2
        #if defined ICL_USE_SSE3 && (__SSE3__ || (defined _MSC_VER && _MSC_VER >= 1500))
                #include "pmmintrin.h"
                #define ICL_HAVE_SSE3
                #if defined ICL_USE_SSSE3 && (__SSSE3__ || (defined _MSC_VER && _MSC_VER >= 1500))
                        #include "tmmintrin.h"
                        #define ICL_HAVE_SSSE3
                #endif
        #endif
#endif

#include <ICLUtils/CompatMacros.h>
#include <ICLUtils/BasicTypes.h>

/* This header wraps the 128 bit SSE types and defines some basic
   functions for them. The idea is to create an easier and more intuitive
   way to work with SSE types.
   The type names are combinations of the basic type name and the number
   of these basic types in the SSE type. For example: icl8ux16 is a type
   with 16 icl8u values.
   The following examples will show some diffrent ways to use the SSE types:

   // ++++++++++++ black and white example ++++++++++++ //
    #include <ICLUtils/SSETypes.h>
    #include <ICLUtils/Time.h>

    using namespace std;
    using namespace icl;
    using namespace icl::utils;

    #define T_SIZE 111111111

    // use a threshold to set values of an array to 0 or 255
    void createBinaryValues(icl8u *d, icl8u *dEnd, const icl8u threshold) {
      for (; d < dEnd; ++d) *d = (*d < threshold) ? 0 : 255;
    }

    // use a threshold to set values of an array to 0 or 255
    void createBinaryValuesSSE(icl8u *d, icl8u *dEnd, const icl8u threshold) {
      // if we end up with less than 16 values at the end
      // we have to convert them value by value to prevent memory access violation
      icl8u *dSSEEnd = dEnd - 15;
      // some constants
      const icl8ux16 c128 = icl8ux16((icl8s)128);
      const icl8ux16 cT = icl8ux16((icl8s)threshold - 129);

      // convert 16 values at the same time
      for (; d < dSSEEnd; d += 16) {
        // load the first 16 values
        icl8ux16 v = icl8ux16(d);
        // subtract 128 from every value in v
        v -= c128;
        // if a function for the SSE wrapper types does not exist
        // we can mix the wrapper types with the actual SSE functions
        v = _mm_cmpgt_epi8(v, cT);
        // storeu stores the values from v in d
        // (store works only with 16 aligned memory,
        // but storeu does not have this restriction)
        v.storeu(d);
      }

      // convert 1 value at a time
      for (; d < dEnd; ++d) {
        *d = (*d < threshold) ? 0 : 255;
      }
    }

    int main(int n, char **ppc){
      icl::utils::Time t;
      icl8u *c = new icl8u[T_SIZE];

      for (unsigned int i = 0; i < T_SIZE; ++i)
        c[i] = rand() % 256;

      t = icl::utils::Time::now();
      createBinaryValues(c, c + T_SIZE, 123);
      t.showAge("without SSE");

      for (unsigned int i = 0; i < T_SIZE; ++i)
        c[i] = rand() % 256;

      t = icl::utils::Time::now();
      createBinaryValuesSSE(c, c + T_SIZE, 123);
      t.showAge("with SSE");

      delete c;

      return 0;
    }
   // ------------ black and white example ------------ //

   // ++++++++++++ rgb to gray ++++++++++++ //
   #include <ICLUtils/SSETypes.h>
   #include <ICLUtils/ClippedCast.h>
   #include <ICLUtils/Time.h>

   using namespace std;
   using namespace icl;
   using namespace icl::utils;

   #define T_SIZE 111111111

   void RGBtoGray(const icl8u *r, const icl8u *g, const icl8u *b, icl8u *gr, icl8u *grEnd) {
   for (; gr != grEnd; ++gr, ++r, ++g, ++b) *gr = clipped_cast<icl32f, icl8u>((*r + *g + *b) / 3.0f + 0.5f);
   }
   void SSERGBtoGray(const icl8u *r, const icl8u *g, const icl8u *b, icl8u *gr, icl8u *grEnd) {
   icl8u *grSSEEnd = grEnd - 15;

   for (; gr < grSSEEnd; gr += 16, r += 16, g += 16, b += 16) {
   // convert to icl16s for number higher than 255
   icl16sx16 vR(r);
   icl16sx16 vG(g);
   icl16sx16 vB(b);

   vR += vB;
   vR += vG;

   // convert to icl32s and then to icl32f type for floating point operations
   icl32fx16 vRes = icl32sx16(vR);

   vRes *= icl32fx16(1.0f / 3.0f);
   vRes.storeu(gr);
   }

   for (; gr != grEnd; ++gr, ++r, ++g, ++b) *gr = clipped_cast<icl8u, icl8u>((*r + *g + *b) / 3.0f + 0.5f);
   }

   int main(int n, char **ppc){
     icl::utils::Time t;
     icl8u *r  = new icl8u[T_SIZE];
     icl8u *g  = new icl8u[T_SIZE];
     icl8u *b  = new icl8u[T_SIZE];
     icl8u *gr = new icl8u[T_SIZE];

     for (unsigned int i = 0; i < T_SIZE; ++i) {
     r[i] = rand() % 256;
     g[i] = rand() % 256;
     b[i] = rand() % 256;
     }

     t = icl::utils::Time::now();
     RGBtoGray(r, g, b, gr, gr + T_SIZE);
     t.showAge("without SSE");

     t = icl::utils::Time::now();
     SSERGBtoGray(r, g, b, gr, gr + T_SIZE);
     t.showAge("with SSE");

     delete r;
     delete g;
     delete b;
     delete gr;

     return 0;
   }
   // ------------ rgb to gray ------------ //

*/

namespace icl{
  namespace utils{

    #ifdef ICL_HAVE_SSE2

      // ++ basic SSE types ++ //

      struct Icl128 {
        __m128 v0;
      };

      struct Icl128i {
        __m128i v0;

        inline Icl128i() {
        }

        inline Icl128i(const Icl128i &v) {
          v0 = v.v0;
        }

        inline Icl128i(const __m128i &v) {
          v0 = v;
        }

        inline Icl128i(const __m128i *v) {
          v0 = _mm_loadu_si128(v);
        }

        inline Icl128i& operator=(const Icl128i &v) {
          v0 = v.v0;
          return *this;
        }

        inline operator __m128i () const {
          return v0;
        }

        inline Icl128i& operator&=(const Icl128i &v) {
          v0 = _mm_and_si128(v0, v.v0);
          return *this;
        }

        inline Icl128i& operator|=(const Icl128i &v) {
          v0 = _mm_or_si128(v0, v.v0);
          return *this;
        }

        inline Icl128i& operator^=(const Icl128i &v) {
          v0 = _mm_xor_si128(v0, v.v0);
          return *this;
        }

        inline Icl128i& andnot(const Icl128i &v) {
          v0 = _mm_andnot_si128(v.v0, v0);
          return *this;
        }

        inline void store(__m128i *v) const {
          _mm_store_si128(v, v0);
        }

        inline void storeu(__m128i *v) const {
          _mm_storeu_si128(v, v0);
        }
      };

      struct Icl128d {
        __m128d v0;
      };

      struct Icl256 {
        __m128 v0; __m128 v1;
      };

      struct Icl256i {
        __m128i v0; __m128i v1;

        inline Icl256i() {
        }

        inline Icl256i(const Icl256i &v) {
          v0 = v.v0;
          v1 = v.v1;
        }

        inline Icl256i(const __m128i &vl, const __m128i &vh) {
          v0 = vl;
          v1 = vh;
        }

        inline Icl256i(const __m128i *v) {
          v0 = *v;
          v1 = *(v + 1);
        }

        inline Icl256i& operator=(const Icl256i &v) {
          v0 = v.v0;
          v1 = v.v1;
          return *this;
        }

        inline Icl256i& operator&=(const Icl256i &v) {
          v0 = _mm_and_si128(v0, v.v0);
          v1 = _mm_and_si128(v1, v.v1);
          return *this;
        }

        inline Icl256i& operator|=(const Icl256i &v) {
          v0 = _mm_or_si128(v0, v.v0);
          v1 = _mm_or_si128(v1, v.v1);
          return *this;
        }

        inline Icl256i& operator^=(const Icl256i &v) {
          v0 = _mm_xor_si128(v0, v.v0);
          v1 = _mm_xor_si128(v1, v.v1);
          return *this;
        }

        inline Icl256i& andnot(const Icl256i &v) {
          v0 = _mm_andnot_si128(v.v0, v0);
          v1 = _mm_andnot_si128(v.v1, v1);
          return *this;
        }

        inline void store(__m128i *v) const {
          _mm_store_si128(v, v0);
          _mm_store_si128(v + 1, v1);
        }

        inline void storeu(__m128i *v) const {
          _mm_storeu_si128(v, v0);
          _mm_storeu_si128(v + 1, v1);
        }
      };

      struct Icl256d {
        __m128d v0; __m128d v1;
      };

      struct Icl512 {
        __m128 v0; __m128 v1; __m128 v2; __m128 v3;
      };

      struct Icl512i {
        __m128i v0; __m128i v1; __m128i v2; __m128i v3;

        inline Icl512i() {
        }

        inline Icl512i(const Icl512i &v) {
          v0 = v.v0;
          v1 = v.v1;
          v2 = v.v2;
          v3 = v.v3;
        }

        inline Icl512i(const __m128i &vll, const __m128i &vlh,
          const __m128i &vhl, const __m128i &vhh) {
          v0 = vll;
          v1 = vlh;
          v2 = vhl;
          v3 = vhh;
        }

        inline Icl512i(const __m128i *v) {
          v0 = *v;
          v1 = *(v + 1);
          v0 = *(v + 2);
          v1 = *(v + 3);
        }

        inline Icl512i& operator=(const Icl512i &v) {
          v0 = v.v0;
          v1 = v.v1;
          v2 = v.v2;
          v3 = v.v3;
          return *this;
        }

        inline Icl512i& operator&=(const Icl512i &v) {
          v0 = _mm_and_si128(v0, v.v0);
          v1 = _mm_and_si128(v1, v.v1);
          v2 = _mm_and_si128(v2, v.v2);
          v3 = _mm_and_si128(v3, v.v3);
          return *this;
        }

        inline Icl512i& operator|=(const Icl512i &v) {
          v0 = _mm_or_si128(v0, v.v0);
          v1 = _mm_or_si128(v1, v.v1);
          v2 = _mm_or_si128(v2, v.v2);
          v3 = _mm_or_si128(v3, v.v3);
          return *this;
        }

        inline Icl512i& operator^=(const Icl512i &v) {
          v0 = _mm_xor_si128(v0, v.v0);
          v1 = _mm_xor_si128(v1, v.v1);
          v2 = _mm_xor_si128(v2, v.v2);
          v3 = _mm_xor_si128(v3, v.v3);
          return *this;
        }

        inline Icl512i& andnot(const Icl512i &v) {
          v0 = _mm_andnot_si128(v.v0, v0);
          v1 = _mm_andnot_si128(v.v1, v1);
          v2 = _mm_andnot_si128(v.v2, v2);
          v3 = _mm_andnot_si128(v.v3, v3);
          return *this;
        }

        inline void store(__m128i *v) const {
          _mm_store_si128(v, v0);
          _mm_store_si128(v + 1, v1);
          _mm_store_si128(v + 2, v2);
          _mm_store_si128(v + 3, v3);
        }

        inline void storeu(__m128i *v) const {
          _mm_storeu_si128(v, v0);
          _mm_storeu_si128(v + 1, v1);
          _mm_storeu_si128(v + 2, v2);
          _mm_storeu_si128(v + 3, v3);
        }
      };

      struct Icl512d {
        __m128d v0; __m128d v1; __m128d v2; __m128d v3;
      };

      struct Icl1024d {
        __m128d v0; __m128d v1; __m128d v2; __m128d v3;
        __m128d v4; __m128d v5; __m128d v6; __m128d v7;
      };

      // -- basic SSE types -- //


      // ++ advanced SSE types ++ //

      struct icl128 : Icl128 {
        inline icl128() {
        }

        inline icl128(const icl128 &v) {
          v0 = v.v0;
        }

        inline icl128(const __m128 &v) {
          v0 = v;
        }

        inline icl128(const icl32f *v) {
          v0 = _mm_loadu_ps(v);
        }

        inline icl128(const icl32f v) {
          v0 = _mm_set1_ps(v);
        }

        inline icl128(const Icl128 &v) {
          v0 = v.v0;
        }

        inline icl128(const Icl128i &v) {
          v0 = _mm_cvtepi32_ps(v.v0);
        }
/*
        inline icl128& operator=(const __m128 &v) {
          v0 = v;
          return *this;
        }

        inline icl128& operator=(const icl32f *v) {
          v0 = _mm_loadu_ps(v);
          return *this;
        }

        inline icl128& operator=(const icl32f v) {
          v0 = _mm_set1_ps(v);
          return *this;
        }
*/
        inline icl128& operator=(const icl128 &v) {
          v0 = v.v0;
          return *this;
        }

        inline icl128& operator=(const Icl128 &v) {
          v0 = v.v0;
          return *this;
        }
/*
        inline icl128& operator=(const Icl128i &v) {
          v0 = _mm_cvtepi32_ps(v.v0);
          return *this;
        }
*/
        inline operator __m128 () const {
          return v0;
        }

        inline icl128& operator+=(const Icl128 &v) {
          v0 = _mm_add_ps(v0, v.v0);
          return *this;
        }

        inline icl128& operator-=(const Icl128 &v) {
          v0 = _mm_sub_ps(v0, v.v0);
          return *this;
        }

        inline icl128& operator*=(const Icl128 &v) {
          v0 = _mm_mul_ps(v0, v.v0);
          return *this;
        }

        inline icl128& operator/=(const Icl128 &v) {
          v0 = _mm_div_ps(v0, v.v0);
          return *this;
        }

        inline icl128& operator&=(const Icl128 &v) {
          v0 = _mm_and_ps(v0, v.v0);
          return *this;
        }

        inline icl128& operator|=(const Icl128 &v) {
          v0 = _mm_or_ps(v0, v.v0);
          return *this;
        }

        inline icl128& operator^=(const Icl128 &v) {
          v0 = _mm_xor_ps(v0, v.v0);
          return *this;
        }

        inline icl128& andnot(const Icl128 &v) {
          v0 = _mm_andnot_ps(v.v0, v0);
          return *this;
        }

        inline icl128& rcp() {
                      v0 = _mm_rcp_ps(v0);
          return *this;
        }

        inline void store(icl32f *v) const {
                      _mm_store_ps(v, v0);
        }

        inline void storeu(icl32f *v) const {
                      _mm_storeu_ps(v, v0);
        }
      };

      struct icl256 : Icl256 {
        inline icl256() {
        }

        inline icl256(const icl256 &v) {
          v0 = v.v0;
          v1 = v.v1;
        }

        inline icl256(const __m128 &vl, const __m128 &vh) {
          v0 = vl;
          v1 = vh;
        }

        inline icl256(const __m128 *v) {
          v0 = *v;
          v1 = *(v + 1);
        }

        inline icl256(const icl32f v) {
          v0 = _mm_set1_ps(v);
          v1 = _mm_set1_ps(v);
        }

        inline icl256(const Icl256 &v) {
          v0 = v.v0;
          v1 = v.v1;
        }

        inline icl256(const Icl256i &v) {
          v0 = _mm_cvtepi32_ps(v.v0);
          v1 = _mm_cvtepi32_ps(v.v1);
        }
/*
        inline icl256& operator=(const __m128 *v) {
          v0 = *v;
          v1 = *(v + 1);
          return *this;
        }

        inline icl256& operator=(const icl32f v) {
          v0 = _mm_set1_ps(v);
          v1 = _mm_set1_ps(v);
          return *this;
        }
*/
        inline icl256& operator=(const icl256 &v) {
          v0 = v.v0;
          v1 = v.v1;
          return *this;
        }

        inline icl256& operator=(const Icl256 &v) {
          v0 = v.v0;
          v1 = v.v1;
          return *this;
        }
/*
        inline icl256& operator=(const Icl256i &v) {
          v0 = _mm_cvtepi32_ps(v.v0);
          v1 = _mm_cvtepi32_ps(v.v1);
          return *this;
        }
*/
        inline icl256& operator+=(const Icl256 &v) {
          v0 = _mm_add_ps(v0, v.v0);
          v1 = _mm_add_ps(v1, v.v1);
          return *this;
        }

        inline icl256& operator-=(const Icl256 &v) {
          v0 = _mm_sub_ps(v0, v.v0);
          v1 = _mm_sub_ps(v1, v.v1);
          return *this;
        }

        inline icl256& operator*=(const Icl256 &v) {
          v0 = _mm_mul_ps(v0, v.v0);
          v1 = _mm_mul_ps(v1, v.v1);
          return *this;
        }

        inline icl256& operator/=(const Icl256 &v) {
          v0 = _mm_div_ps(v0, v.v0);
          v1 = _mm_div_ps(v1, v.v1);
          return *this;
        }

        inline icl256& operator&=(const Icl256 &v) {
          v0 = _mm_and_ps(v0, v.v0);
          v1 = _mm_and_ps(v1, v.v1);
          return *this;
        }

        inline icl256& operator|=(const Icl256 &v) {
          v0 = _mm_or_ps(v0, v.v0);
          v1 = _mm_or_ps(v1, v.v1);
          return *this;
        }

        inline icl256& operator^=(const Icl256 &v) {
          v0 = _mm_xor_ps(v0, v.v0);
          v1 = _mm_xor_ps(v1, v.v1);
          return *this;
        }

        inline icl256& andnot(const Icl256 &v) {
          v0 = _mm_andnot_ps(v.v0, v0);
          v1 = _mm_andnot_ps(v.v1, v1);
          return *this;
        }

        inline icl256& rcp() {
          v0 = _mm_rcp_ps(v0);
          v1 = _mm_rcp_ps(v1);
          return *this;
        }

        inline void store(icl32f *v) const {
          _mm_store_ps(v, v0);
          _mm_store_ps(v + 4, v1);
        }

        inline void storeu(icl32f *v) const {
          _mm_storeu_ps(v, v0);
          _mm_storeu_ps(v + 4, v1);
        }
      };

      struct icl512 : Icl512 {
        inline icl512() {
        }

        inline icl512(const icl512 &v) {
          v0 = v.v0;
          v1 = v.v1;
          v2 = v.v2;
          v3 = v.v3;
        }

        inline icl512(const __m128 &vll, const __m128 &vlh,
          const __m128 &vhl, const __m128 &vhh) {
          v0 = vll;
          v1 = vlh;
          v2 = vhl;
          v3 = vhh;
        }

        inline icl512(const __m128 *v) {
          v0 = *v;
          v1 = *(v + 1);
          v2 = *(v + 2);
          v3 = *(v + 3);
        }

        inline icl512(const icl8u *v) {
          const __m128i vk0 = _mm_setzero_si128();
          __m128i vt0, vt1, vt2, vt3;

          vt3 = _mm_loadu_si128((__m128i*)v);

          vt1 = _mm_unpacklo_epi8(vt3, vk0);
          vt3 = _mm_unpackhi_epi8(vt3, vk0);

          vt0 = _mm_unpacklo_epi16(vt1, vk0);
          vt1 = _mm_unpackhi_epi16(vt1, vk0);
          vt2 = _mm_unpacklo_epi16(vt3, vk0);
          vt3 = _mm_unpackhi_epi16(vt3, vk0);

          v0 = _mm_cvtepi32_ps(vt0);
          v1 = _mm_cvtepi32_ps(vt1);
          v2 = _mm_cvtepi32_ps(vt2);
          v3 = _mm_cvtepi32_ps(vt3);
        }

        inline icl512(const icl32f *v) {
          v0 = _mm_loadu_ps(v);
          v1 = _mm_loadu_ps(v + 4);
          v2 = _mm_loadu_ps(v + 8);
          v3 = _mm_loadu_ps(v + 12);
        }

        inline icl512(const Icl512 &v) {
          v0 = v.v0;
          v1 = v.v1;
          v2 = v.v2;
          v3 = v.v3;
        }

        inline icl512(const Icl512i &v) {
          v0 = _mm_cvtepi32_ps(v.v0);
          v1 = _mm_cvtepi32_ps(v.v1);
          v2 = _mm_cvtepi32_ps(v.v2);
          v3 = _mm_cvtepi32_ps(v.v3);
        }

        inline icl512(const icl32f v) {
          v0 = _mm_set1_ps(v);
          v1 = _mm_set1_ps(v);
          v2 = _mm_set1_ps(v);
          v3 = _mm_set1_ps(v);
        }
/*
        inline icl512& operator=(const __m128 *v) {
          v0 = *v;
          v1 = *(v + 1);
          v2 = *(v + 2);
          v3 = *(v + 3);
          return *this;
        }

        inline icl512& operator=(const icl32f *v) {
          v0 = _mm_loadu_ps(v);
          v1 = _mm_loadu_ps(v + 4);
          v2 = _mm_loadu_ps(v + 8);
          v3 = _mm_loadu_ps(v + 12);
          return *this;
        }
*/
        inline icl512& operator=(const icl512 &v) {
          v0 = v.v0;
          v1 = v.v1;
          v2 = v.v2;
          v3 = v.v3;
          return *this;
        }

        inline icl512& operator=(const Icl512 &v) {
          v0 = v.v0;
          v1 = v.v1;
          v2 = v.v2;
          v3 = v.v3;
          return *this;
        }
/*
        inline icl512& operator=(const Icl512i &v) {
          v0 = _mm_cvtepi32_ps(v.v0);
          v1 = _mm_cvtepi32_ps(v.v1);
          v2 = _mm_cvtepi32_ps(v.v2);
          v3 = _mm_cvtepi32_ps(v.v3);
          return *this;
        }

        inline icl512& operator=(const icl32f v) {
          v0 = _mm_set1_ps(v);
          v1 = _mm_set1_ps(v);
          v2 = _mm_set1_ps(v);
          v3 = _mm_set1_ps(v);
          return *this;
        }
*/
        inline icl512& operator+=(const Icl512 &v) {
          v0 = _mm_add_ps(v0, v.v0);
          v1 = _mm_add_ps(v1, v.v1);
          v2 = _mm_add_ps(v2, v.v2);
          v3 = _mm_add_ps(v3, v.v3);
          return *this;
        }

        inline icl512& operator-=(const Icl512 &v) {
          v0 = _mm_sub_ps(v0, v.v0);
          v1 = _mm_sub_ps(v1, v.v1);
          v2 = _mm_sub_ps(v2, v.v2);
          v3 = _mm_sub_ps(v3, v.v3);
          return *this;
        }

        inline icl512& operator*=(const Icl512 &v) {
          v0 = _mm_mul_ps(v0, v.v0);
          v1 = _mm_mul_ps(v1, v.v1);
          v2 = _mm_mul_ps(v2, v.v2);
          v3 = _mm_mul_ps(v3, v.v3);
          return *this;
        }

        inline icl512& operator/=(const Icl512 &v) {
          v0 = _mm_div_ps(v0, v.v0);
          v1 = _mm_div_ps(v1, v.v1);
          v2 = _mm_div_ps(v2, v.v2);
          v3 = _mm_div_ps(v3, v.v3);
          return *this;
        }

        inline icl512& operator&=(const Icl512 &v) {
          v0 = _mm_and_ps(v0, v.v0);
          v1 = _mm_and_ps(v1, v.v1);
          v2 = _mm_and_ps(v2, v.v2);
          v3 = _mm_and_ps(v3, v.v3);
          return *this;
        }

        inline icl512& operator|=(const Icl512 &v) {
          v0 = _mm_or_ps(v0, v.v0);
          v1 = _mm_or_ps(v1, v.v1);
          v2 = _mm_or_ps(v2, v.v2);
          v3 = _mm_or_ps(v3, v.v3);
          return *this;
        }

        inline icl512& operator^=(const Icl512 &v) {
          v0 = _mm_xor_ps(v0, v.v0);
          v1 = _mm_xor_ps(v1, v.v1);
          v2 = _mm_xor_ps(v2, v.v2);
          v3 = _mm_xor_ps(v3, v.v3);
          return *this;
        }

        inline icl512& andnot(const Icl512 &v) {
          v0 = _mm_andnot_ps(v.v0, v0);
          v1 = _mm_andnot_ps(v.v1, v1);
          v2 = _mm_andnot_ps(v.v2, v2);
          v3 = _mm_andnot_ps(v.v3, v3);
          return *this;
        }

        inline icl512& rcp() {
          v0 = _mm_rcp_ps(v0);
          v1 = _mm_rcp_ps(v1);
          v2 = _mm_rcp_ps(v2);
          v3 = _mm_rcp_ps(v3);
          return *this;
        }

        inline void store(icl8u *v) const {
          //__m128 vMin = _mm_set1_ps(-2147483520.f);
          //__m128 vMax = _mm_set1_ps(2147483520.f);
          //v0 = _mm_cvtps_epi32(_mm_min_ps(_mm_max_ps(v.v0, vMin), vMax));
          //v1 = _mm_cvtps_epi32(_mm_min_ps(_mm_max_ps(v.v1, vMin), vMax));
          //v2 = _mm_cvtps_epi32(_mm_min_ps(_mm_max_ps(v.v2, vMin), vMax));
          //v3 = _mm_cvtps_epi32(_mm_min_ps(_mm_max_ps(v.v3, vMin), vMax));
          __m128i vt0 = _mm_cvtps_epi32(v0);
          __m128i vt1 = _mm_cvtps_epi32(v1);
          __m128i vt2 = _mm_cvtps_epi32(v2);
          __m128i vt3 = _mm_cvtps_epi32(v3);

          vt0 = _mm_packus_epi16(_mm_packs_epi32(vt0, vt1), _mm_packs_epi32(vt2, vt3));
          _mm_store_si128((__m128i*)v, vt0);
        }

        inline void storeu(icl8u *v) const {
          //__m128 vMin = _mm_set1_ps(-2147483520.f);
          //__m128 vMax = _mm_set1_ps(2147483520.f);
          //v0 = _mm_cvtps_epi32(_mm_min_ps(_mm_max_ps(v.v0, vMin), vMax));
          //v1 = _mm_cvtps_epi32(_mm_min_ps(_mm_max_ps(v.v1, vMin), vMax));
          //v2 = _mm_cvtps_epi32(_mm_min_ps(_mm_max_ps(v.v2, vMin), vMax));
          //v3 = _mm_cvtps_epi32(_mm_min_ps(_mm_max_ps(v.v3, vMin), vMax));
          __m128i vt0 = _mm_cvtps_epi32(v0);
          __m128i vt1 = _mm_cvtps_epi32(v1);
          __m128i vt2 = _mm_cvtps_epi32(v2);
          __m128i vt3 = _mm_cvtps_epi32(v3);

          vt0 = _mm_packus_epi16(_mm_packs_epi32(vt0, vt1), _mm_packs_epi32(vt2, vt3));
          _mm_storeu_si128((__m128i*)v, vt0);
        }

        inline void store(icl32f *v) const {
          _mm_store_ps(v, v0);
          _mm_store_ps(v + 4, v1);
          _mm_store_ps(v + 8, v2);
          _mm_store_ps(v + 12, v3);
        }

        inline void storeu(icl32f *v) const {
          _mm_storeu_ps(v, v0);
          _mm_storeu_ps(v + 4, v1);
          _mm_storeu_ps(v + 8, v2);
          _mm_storeu_ps(v + 12, v3);
        }
      };

      struct icl128i8u : Icl128i {
        inline icl128i8u() {
        }

        inline icl128i8u(const Icl128i &v) {
          v0 = v.v0;
        }

        inline icl128i8u(const icl128i8u &v) {
          v0 = v.v0;
        }

        inline icl128i8u(const __m128i &v) {
          v0 = v;
        }

        inline icl128i8u(const __m128i *v) {
          v0 = _mm_loadu_si128(v);
        }

        inline icl128i8u(const icl8s *v) {
          v0 = _mm_loadu_si128((__m128i*)v);
        }

        inline icl128i8u(const icl8u *v) {
          v0 = _mm_loadu_si128((__m128i*)v);
        }

        inline icl128i8u(const icl8s v) {
          v0 = _mm_set1_epi8(v);
        }

        inline icl128i8u(const Icl256i &v) {
          //v0 = _mm_packs_epi16(v.v0, v.v1); // for icl8s
          v0 = _mm_packus_epi16(v.v0, v.v1);
        }

        inline icl128i8u(const Icl512i &v) {
          //v0 = _mm_packs_epi16(_mm_packs_epi32(v.v0, v.v1), _mm_packs_epi32(v.v2, v.v3)); // for icl8s
          v0 = _mm_packus_epi16(_mm_packs_epi32(v.v0, v.v1), _mm_packs_epi32(v.v2, v.v3));
        }

        inline operator Icl128i () const {
          return *this;
        }

        inline icl128i8u& operator=(const icl128i8u &v) {
          v0 = v.v0;
          return *this;
        }

        inline icl128i8u& operator=(const Icl128i &v) {
          v0 = v.v0;
          return *this;
        }

        inline icl128i8u& operator+=(const icl128i8u &v) {
          v0 = _mm_add_epi8(v0, v.v0);
          return *this;
        }

        inline icl128i8u& operator-=(const icl128i8u &v) {
          v0 = _mm_sub_epi8(v0, v.v0);
          return *this;
        }

        inline void store(__m128i *v) const {
          _mm_store_si128(v, v0);
        }

        inline void storeu(__m128i *v) const {
          _mm_storeu_si128(v, v0);
        }

        inline void store(icl8s *v) const {
          _mm_store_si128((__m128i*)v, v0);
        }

        inline void storeu(icl8s *v) const {
          _mm_storeu_si128((__m128i*)v, v0);
        }

        inline void store(icl8u *v) const {
          _mm_store_si128((__m128i*)v, v0);
        }

        inline void storeu(icl8u *v) const {
          _mm_storeu_si128((__m128i*)v, v0);
        }
      };

      struct icl128i16s : Icl128i {
        inline icl128i16s() {
        }

        inline icl128i16s(const icl128i16s &v) {
          v0 = v.v0;
        }

        inline icl128i16s(const Icl128i &v) {
          v0 = v.v0;
        }

        inline icl128i16s(const __m128i &v) {
          v0 = v;
        }

        inline icl128i16s(const __m128i *v) {
          v0 = _mm_loadu_si128(v);
        }

        inline icl128i16s(const icl16s *v) {
          v0 = _mm_loadu_si128((__m128i*)v);
        }

        inline icl128i16s(const icl16u *v) {
          v0 = _mm_loadu_si128((__m128i*)v);
        }

        inline icl128i16s(const icl16s v) {
          v0 = _mm_set1_epi16(v);
        }

        inline icl128i16s(const Icl256i &v) {
          v0 = _mm_packs_epi32(v.v0, v.v1);
        }

        inline operator Icl128i () const {
          return *this;
        }

        inline icl128i16s& operator=(const icl128i16s &v) {
          v0 = v.v0;
          return *this;
        }

        inline icl128i16s& operator=(const Icl128i &v) {
          v0 = v.v0;
          return *this;
        }

        inline icl128i16s& operator+=(const icl128i16s &v) {
          v0 = _mm_add_epi16(v0, v.v0);
          return *this;
        }

        inline icl128i16s& operator-=(const icl128i16s &v) {
          v0 = _mm_sub_epi16(v0, v.v0);
          return *this;
        }

        inline void store(__m128i *v) const {
          _mm_store_si128(v, v0);
        }

        inline void storeu(__m128i *v) const {
          _mm_storeu_si128(v, v0);
        }

        inline void store(icl16s *v) const {
          _mm_store_si128((__m128i*)v, v0);
        }

        inline void storeu(icl16s *v) const {
          _mm_storeu_si128((__m128i*)v, v0);
        }

        inline void store(icl16u *v) const {
          _mm_store_si128((__m128i*)v, v0);
        }

        inline void storeu(icl16u *v) const {
          _mm_storeu_si128((__m128i*)v, v0);
        }
      };

      struct icl128i32s : Icl128i {
        inline icl128i32s() {
        }

        inline icl128i32s(const icl128i32s &v) {
          v0 = v.v0;
        }

        inline icl128i32s(const Icl128i &v) {
          v0 = v.v0;
        }

        inline icl128i32s(const __m128i &v) {
          v0 = v;
        }

        inline icl128i32s(const __m128i *v) {
          v0 = _mm_loadu_si128(v);
        }

        inline icl128i32s(const icl32s *v) {
          v0 = _mm_loadu_si128((__m128i*)v);
        }

        inline icl128i32s(const icl32u *v) {
          v0 = _mm_loadu_si128((__m128i*)v);
        }

        inline icl128i32s(const icl32s i0, const icl32s i1, const icl32s i2, const icl32s i3) {
          v0 = _mm_set_epi32(i3, i2, i1, i0);
        }

        inline icl128i32s(const icl32s v) {
          v0 = _mm_set1_epi32(v);
        }

        inline icl128i32s(const Icl128 &v) {
          //__m128 vMin = _mm_set1_ps(-2147483520.f);
          //__m128 vMax = _mm_set1_ps(2147483520.f);
          //v0 = _mm_cvtps_epi32(_mm_min_ps(_mm_max_ps(v.v0, vMin), vMax));
          v0 = _mm_cvtps_epi32(v.v0);
        }

        inline operator Icl128i () const {
          return *this;
        }

#ifndef __clang__
        inline icl128i32s& operator=(const icl128i32s &v) {
          v0 = v.v0;
          return *this;
        }

        inline icl128i32s& operator=(const Icl128i &v) {
          v0 = v.v0;
          return *this;
        }
#endif
        inline icl128i32s& operator+=(const icl128i32s &v) {
          v0 = _mm_add_epi32(v0, v.v0);
          return *this;
        }

        inline icl128i32s& operator-=(const icl128i32s &v) {
          v0 = _mm_sub_epi32(v0, v.v0);
          return *this;
        }

        inline void store(__m128i *v) const {
          _mm_store_si128(v, v0);
        }

        inline void storeu(__m128i *v) const {
          _mm_storeu_si128(v, v0);
        }

        inline void store(icl32s *v) const {
          _mm_store_si128((__m128i*)v, v0);
        }

        inline void storeu(icl32s *v) const {
          _mm_storeu_si128((__m128i*)v, v0);
        }

        inline void store(icl32u *v) const {
          _mm_store_si128((__m128i*)v, v0);
        }

        inline void storeu(icl32u *v) const {
          _mm_storeu_si128((__m128i*)v, v0);
        }
      };

      struct icl256i16s : Icl256i {
        inline icl256i16s() {
        }

        inline icl256i16s(const icl256i16s &v) {
          v0 = v.v0;
          v1 = v.v0;
        }

        inline icl256i16s(const Icl256i &v) {
          v0 = v.v0;
          v1 = v.v0;
        }

        inline icl256i16s(const __m128i &vl, const __m128i &vh) {
          v0 = vl;
          v1 = vh;
        }

        inline icl256i16s(const __m128i *v) {
          v0 = *v;
          v1 = *(v + 1);
        }

        inline icl256i16s(const icl16s *v) {
          v0 = _mm_loadu_si128((__m128i*)v);
          v1 = _mm_loadu_si128((__m128i*)(v + 8));
        }

        inline icl256i16s(const icl16s v) {
          v0 = _mm_set1_epi16(v);
          v1 = _mm_set1_epi16(v);
        }

        inline icl256i16s(const icl128i8u &v) {
          const __m128i vk0 = _mm_setzero_si128();
          v0 = _mm_unpacklo_epi8(v.v0, vk0);
          v1 = _mm_unpackhi_epi8(v.v0, vk0);
        }

        inline icl256i16s(const Icl512i &v) {
          v0 = _mm_packs_epi32(v.v0, v.v1);
          v1 = _mm_packs_epi32(v.v2, v.v3);
        }

        inline operator Icl256i () const {
          return *this;
        }

        inline icl256i16s& operator=(const icl256i16s &v) {
          v0 = v.v0;
          v1 = v.v1;
          return *this;
        }

        inline icl256i16s& operator=(const Icl256i &v) {
          v0 = v.v0;
          v1 = v.v1;
          return *this;
        }

        inline icl256i16s& operator+=(const icl256i16s &v) {
          v0 = _mm_add_epi16(v0, v.v0);
          v1 = _mm_add_epi16(v1, v.v1);
          return *this;
        }

        inline icl256i16s& operator-=(const icl256i16s &v) {
          v0 = _mm_sub_epi16(v0, v.v0);
          v1 = _mm_sub_epi16(v1, v.v1);
          return *this;
        }

        inline void store(__m128i *v) const {
          _mm_store_si128(v, v0);
          _mm_store_si128(v + 1, v1);
        }

        inline void storeu(__m128i *v) const {
          _mm_storeu_si128(v, v0);
          _mm_storeu_si128(v + 1, v1);
        }

        inline void store(icl16s *v) const {
          _mm_store_si128((__m128i*)v, v0);
          _mm_store_si128((__m128i*)(v + 8), v1);
        }

        inline void storeu(icl16s *v) const {
          _mm_storeu_si128((__m128i*)v, v0);
          _mm_storeu_si128((__m128i*)(v + 8), v1);
        }

        inline void store(icl16u *v) const {
          _mm_store_si128((__m128i*)v, v0);
          _mm_store_si128((__m128i*)(v + 8), v1);
        }

        inline void storeu(icl16u *v) const {
          _mm_storeu_si128((__m128i*)v, v0);
          _mm_storeu_si128((__m128i*)(v + 8), v1);
        }
      };

      struct icl256i32s : Icl256i {

        inline icl256i32s() {
        }

        inline icl256i32s(const icl256i32s &v) {
          v0 = v.v0;
          v1 = v.v1;
        }

        inline icl256i32s(const Icl256i &v) {
          v0 = v.v0;
          v1 = v.v1;
        }

        inline icl256i32s(const __m128i &vl, const __m128i &vh) {
          v0 = vl;
          v1 = vh;
        }

        inline icl256i32s(const __m128i *v) {
          v0 = *v;
          v1 = *(v + 1);
        }

        inline icl256i32s(const icl32s *v) {
          v0 = _mm_loadu_si128((__m128i*)v);
          v1 = _mm_loadu_si128((__m128i*)(v + 4));
        }

        inline icl256i32s(const icl32s v) {
          v0 = _mm_set1_epi32(v);
          v1 = _mm_set1_epi32(v);
        }

        inline icl256i32s(const Icl256 &v) {
          //__m128 vMin = _mm_set1_ps(-2147483520.f);
          //__m128 vMax = _mm_set1_ps(2147483520.f);
          //v0 = _mm_cvtps_epi32(_mm_min_ps(_mm_max_ps(v.v0, vMin), vMax));
          //v1 = _mm_cvtps_epi32(_mm_min_ps(_mm_max_ps(v.v1, vMin), vMax));
          v0 = _mm_cvtps_epi32(v.v0);
          v1 = _mm_cvtps_epi32(v.v1);
        }

        inline icl256i32s& operator=(const icl256i32s &v) {
          v0 = v.v0;
          v1 = v.v1;
          return *this;
        }

        inline icl256i32s& operator=(const Icl256i &v) {
          v0 = v.v0;
          v1 = v.v1;
          return *this;
        }

        inline icl256i32s& operator+=(const icl256i32s &v) {
          v0 = _mm_add_epi16(v0, v.v0);
          v1 = _mm_add_epi16(v1, v.v1);
          return *this;
        }

        inline icl256i32s& operator-=(const icl256i32s &v) {
          v0 = _mm_sub_epi16(v0, v.v0);
          v1 = _mm_sub_epi16(v1, v.v1);
          return *this;
        }

        inline void store(__m128i *v) const {
          _mm_store_si128(v, v0);
          _mm_store_si128(v + 1, v1);
        }

        inline void storeu(__m128i *v) const {
          _mm_storeu_si128(v, v0);
          _mm_storeu_si128(v + 1, v1);
        }

        inline void store(icl32s *v) const {
          _mm_store_si128((__m128i*)v, v0);
          _mm_store_si128((__m128i*)(v + 4), v1);
        }

        inline void storeu(icl32s *v) const {
          _mm_storeu_si128((__m128i*)v, v0);
          _mm_storeu_si128((__m128i*)(v + 4), v1);
        }

        inline void store(icl32u *v) const {
          _mm_store_si128((__m128i*)v, v0);
          _mm_store_si128((__m128i*)(v + 4), v1);
        }

        inline void storeu(icl32u *v) const {
          _mm_storeu_si128((__m128i*)v, v0);
          _mm_storeu_si128((__m128i*)(v + 4), v1);
        }
      };

      struct icl512i32s : Icl512i {
        inline icl512i32s() {
        }

        inline icl512i32s(const icl512i32s &v) {
          v0 = v.v0;
          v1 = v.v1;
          v2 = v.v2;
          v3 = v.v3;
        }

        inline icl512i32s(const Icl512i &v) {
          v0 = v.v0;
          v1 = v.v1;
          v2 = v.v2;
          v3 = v.v3;
        }

        inline icl512i32s(const __m128i &vll, const __m128i &vlh,
          const __m128i &vhl, const __m128i &vhh) {
          v0 = vll;
          v1 = vlh;
          v2 = vhl;
          v3 = vhh;
        }

        inline icl512i32s(const icl32s *v) {
          v0 = _mm_loadu_si128((__m128i*)v);
          v1 = _mm_loadu_si128((__m128i*)(v + 4));
          v2 = _mm_loadu_si128((__m128i*)(v + 8));
          v3 = _mm_loadu_si128((__m128i*)(v + 12));
        }

        inline icl512i32s(const Icl256i &v) {
          const __m128i vk0 = _mm_setzero_si128();
          v0 = _mm_unpacklo_epi16(v.v0, vk0);
          v1 = _mm_unpackhi_epi16(v.v0, vk0);
          v2 = _mm_unpacklo_epi16(v.v1, vk0);
          v3 = _mm_unpackhi_epi16(v.v1, vk0);
        }

        inline icl512i32s(const icl32s v) {
          v0 = _mm_set1_epi32(v);
          v1 = _mm_set1_epi32(v);
          v2 = _mm_set1_epi32(v);
          v3 = _mm_set1_epi32(v);
        }

        inline icl512i32s(const Icl512 &v) {
          //__m128 vMin = _mm_set1_ps(-2147483520.f);
          //__m128 vMax = _mm_set1_ps(2147483520.f);
          //v0 = _mm_cvtps_epi32(_mm_min_ps(_mm_max_ps(v.v0, vMin), vMax));
          //v1 = _mm_cvtps_epi32(_mm_min_ps(_mm_max_ps(v.v1, vMin), vMax));
          //v2 = _mm_cvtps_epi32(_mm_min_ps(_mm_max_ps(v.v2, vMin), vMax));
          //v3 = _mm_cvtps_epi32(_mm_min_ps(_mm_max_ps(v.v3, vMin), vMax));
          v0 = _mm_cvtps_epi32(v.v0);
          v1 = _mm_cvtps_epi32(v.v1);
          v2 = _mm_cvtps_epi32(v.v2);
          v3 = _mm_cvtps_epi32(v.v3);
        }

        inline icl512i32s& operator=(const icl512i32s &v) {
          v0 = v.v0;
          v1 = v.v1;
          v2 = v.v2;
          v3 = v.v3;
          return *this;
        }

        inline icl512i32s& operator=(const Icl512i &v) {
          v0 = v.v0;
          v1 = v.v1;
          v2 = v.v2;
          v3 = v.v3;
          return *this;
        }

        inline icl512i32s& operator+=(const icl512i32s &v) {
          v0 = _mm_add_epi32(v0, v.v0);
          v1 = _mm_add_epi32(v1, v.v1);
          v2 = _mm_add_epi32(v2, v.v2);
          v3 = _mm_add_epi32(v3, v.v3);
          return *this;
        }

        inline icl512i32s& operator-=(const icl512i32s &v) {
          v0 = _mm_sub_epi32(v0, v.v0);
          v1 = _mm_sub_epi32(v1, v.v1);
          v2 = _mm_sub_epi32(v2, v.v2);
          v3 = _mm_sub_epi32(v3, v.v3);
          return *this;
        }

        inline void store(icl32s *v) const {
          _mm_store_si128((__m128i*)v, v0);
          _mm_store_si128((__m128i*)(v + 4), v1);
          _mm_store_si128((__m128i*)(v + 8), v1);
          _mm_store_si128((__m128i*)(v + 12), v1);
        }

        inline void storeu(icl32s *v) const {
          _mm_storeu_si128((__m128i*)v, v0);
          _mm_storeu_si128((__m128i*)(v + 4), v1);
          _mm_storeu_si128((__m128i*)(v + 8), v1);
          _mm_storeu_si128((__m128i*)(v + 12), v1);
        }

        inline void store(icl32u *v) const {
          _mm_store_si128((__m128i*)v, v0);
          _mm_store_si128((__m128i*)(v + 4), v1);
          _mm_store_si128((__m128i*)(v + 8), v1);
          _mm_store_si128((__m128i*)(v + 12), v1);
        }

        inline void storeu(icl32u *v) const {
          _mm_storeu_si128((__m128i*)v, v0);
          _mm_storeu_si128((__m128i*)(v + 4), v1);
          _mm_storeu_si128((__m128i*)(v + 8), v1);
          _mm_storeu_si128((__m128i*)(v + 12), v1);
        }
      };

      struct icl128d : Icl128d {
        inline icl128d() {
        }

        inline icl128d(const __m128d &v) {
          v0 = v;
        }

        inline icl128d(const icl64f *v) {
          v0 = _mm_loadu_pd(v);
        }

        inline icl128d(const icl64f v) {
          v0 = _mm_set1_pd(v);
        }

        inline icl128d(const icl128d &v) {
          v0 = v.v0;
        }

        inline icl128d& operator=(const __m128d &v) {
          v0 = v;
          return *this;
        }

        inline icl128d& operator=(const icl64f *v) {
          v0 = _mm_loadu_pd(v);
          return *this;
        }

        inline icl128d& operator=(const icl128d &v) {
          v0 = v.v0;
          return *this;
        }

        inline operator __m128d () const {
          return v0;
        }

        inline icl128d& operator+=(const Icl128d &v) {
          v0 = _mm_add_pd(v0, v.v0);
          return *this;
        }

        inline icl128d& operator-=(const Icl128d &v) {
          v0 = _mm_sub_pd(v0, v.v0);
          return *this;
        }

        inline icl128d& operator*=(const Icl128d &v) {
          v0 = _mm_mul_pd(v0, v.v0);
          return *this;
        }

        inline icl128d& operator/=(const Icl128d &v) {
          v0 = _mm_div_pd(v0, v.v0);
          return *this;
        }

        inline icl128d& operator&=(const Icl128d &v) {
          v0 = _mm_and_pd(v0, v.v0);
          return *this;
        }

        inline icl128d& operator|=(const Icl128d &v) {
          v0 = _mm_or_pd(v0, v.v0);
          return *this;
        }

        inline icl128d& operator^=(const Icl128d &v) {
          v0 = _mm_xor_pd(v0, v.v0);
          return *this;
        }

        inline icl128d& andnot(const Icl128d &v) {
          v0 = _mm_andnot_pd(v.v0, v0);
          return *this;
        }

        inline void store(icl64f *v) const {
                      _mm_store_pd(v, v0);
        }

        inline void storeu(icl64f *v) const {
                      _mm_storeu_pd(v, v0);
        }
      };

      struct icl256d : Icl512d {
        // TODO
      };

      // type for 8 icl64f values
      struct icl512d : Icl512d {
        // TODO
      };

      struct icl1024d : Icl1024d {
        // TODO
      };

      // -- advanced SSE types -- //


      // ++ operations on SSE types ++ //

      // ++ arithmetic operations ++ //

      inline icl128 operator+(const icl128 &lv, const icl128 &rv) {
        icl128 ret = lv;
        return ret += rv;
      }

      inline icl128 operator-(const icl128 &lv, const icl128 &rv) {
        icl128 ret = lv;
        return ret -= rv;
      }

      inline icl128 operator*(const icl128 &lv, const icl128 &rv) {
        icl128 ret = lv;
        return ret *= rv;
      }

#ifndef __clang__
      inline icl128 operator/(const icl128 &lv, const icl128 &rv) {
        icl128 ret = lv;
        return ret /= rv;
      }
#endif

      inline icl256 operator+(const icl256 &lv, const icl256 &rv) {
        icl256 ret = lv;
        return ret += rv;
      }

      inline icl256 operator-(const icl256 &lv, const icl256 &rv) {
        icl256 ret = lv;
        return ret -= rv;
      }

      inline icl256 operator*(const icl256 &lv, const icl256 &rv) {
        icl256 ret = lv;
        return ret *= rv;
      }

      inline icl256 operator/(const icl256 &lv, const icl256 &rv) {
        icl256 ret = lv;
        return ret /= rv;
      }

      inline icl512 operator+(const icl512 &lv, const icl512 &rv) {
        icl512 ret = lv;
        return ret += rv;
      }

      inline icl512 operator-(const icl512 &lv, const icl512 &rv) {
        icl512 ret = lv;
        return ret -= rv;
      }

      inline icl512 operator*(const icl512 &lv, const icl512 &rv) {
        icl512 ret = lv;
        return ret *= rv;
      }

      inline icl512 operator/(const icl512 &lv, const icl512 &rv) {
        icl512 ret = lv;
        return ret /= rv;
      }

      // -- arithmetic operations -- //

      // ++ comparison operations ++ //

      inline icl128 operator==(const icl128 &lv, const icl128 &rv) {
        icl128 ret;
        ret.v0 = _mm_cmpeq_ps(lv.v0, rv.v0);
        return ret;
      }

      inline icl128 operator!=(const icl128 &lv, const icl128 &rv) {
        icl128 ret;
        ret.v0 = _mm_cmpneq_ps(lv.v0, rv.v0);
        return ret;
      }

      inline icl128 operator<(const icl128 &lv, const icl128 &rv) {
        icl128 ret;
        ret.v0 = _mm_cmplt_ps(lv.v0, rv.v0);
        return ret;
      }

      inline icl128 operator>(const icl128 &lv, const icl128 &rv) {
        icl128 ret;
        ret.v0 = _mm_cmpgt_ps(lv.v0, rv.v0);
        return ret;
      }

      inline icl128 operator<=(const icl128 &lv, const icl128 &rv) {
        icl128 ret;
        ret.v0 = _mm_cmple_ps(lv.v0, rv.v0);
        return ret;
      }

      inline icl128 operator>=(const icl128 &lv, const icl128 &rv) {
        icl128 ret;
        ret.v0 = _mm_cmpge_ps(lv.v0, rv.v0);
        return ret;
      }

      inline icl256 operator==(const icl256 &lv, const icl256 &rv) {
        icl256 ret;
        ret.v0 = _mm_cmpeq_ps(lv.v0, rv.v0);
        ret.v1 = _mm_cmpeq_ps(lv.v1, rv.v1);
        return ret;
      }

      inline icl256 operator!=(const icl256 &lv, const icl256 &rv) {
        icl256 ret;
        ret.v0 = _mm_cmpneq_ps(lv.v0, rv.v0);
        ret.v1 = _mm_cmpneq_ps(lv.v1, rv.v1);
        return ret;
      }

      inline icl256 operator<(const icl256 &lv, const icl256 &rv) {
        icl256 ret;
        ret.v0 = _mm_cmplt_ps(lv.v0, rv.v0);
        ret.v1 = _mm_cmplt_ps(lv.v1, rv.v1);
        return ret;
      }

      inline icl256 operator>(const icl256 &lv, const icl256 &rv) {
        icl256 ret;
        ret.v0 = _mm_cmpgt_ps(lv.v0, rv.v0);
        ret.v1 = _mm_cmpgt_ps(lv.v1, rv.v1);
        return ret;
      }

      inline icl256 operator<=(const icl256 &lv, const icl256 &rv) {
        icl256 ret;
        ret.v0 = _mm_cmple_ps(lv.v0, rv.v0);
        ret.v1 = _mm_cmple_ps(lv.v1, rv.v1);
        return ret;
      }

      inline icl256 operator>=(const icl256 &lv, const icl256 &rv) {
        icl256 ret;
        ret.v0 = _mm_cmpge_ps(lv.v0, rv.v0);
        ret.v1 = _mm_cmpge_ps(lv.v1, rv.v1);
        return ret;
      }

      inline icl512 operator==(const icl512 &lv, const icl512 &rv) {
        icl512 ret;
        ret.v0 = _mm_cmpeq_ps(lv.v0, rv.v0);
        ret.v1 = _mm_cmpeq_ps(lv.v1, rv.v1);
        ret.v2 = _mm_cmpeq_ps(lv.v2, rv.v2);
        ret.v3 = _mm_cmpeq_ps(lv.v3, rv.v3);
        return ret;
      }

      inline icl512 operator!=(const icl512 &lv, const icl512 &rv) {
        icl512 ret;
        ret.v0 = _mm_cmpneq_ps(lv.v0, rv.v0);
        ret.v1 = _mm_cmpneq_ps(lv.v1, rv.v1);
        ret.v2 = _mm_cmpneq_ps(lv.v2, rv.v2);
        ret.v3 = _mm_cmpneq_ps(lv.v3, rv.v3);
        return ret;
      }

      inline icl512 operator<(const icl512 &lv, const icl512 &rv) {
        icl512 ret;
        ret.v0 = _mm_cmplt_ps(lv.v0, rv.v0);
        ret.v1 = _mm_cmplt_ps(lv.v1, rv.v1);
        ret.v2 = _mm_cmplt_ps(lv.v2, rv.v2);
        ret.v3 = _mm_cmplt_ps(lv.v3, rv.v3);
        return ret;
      }

      inline icl512 operator>(const icl512 &lv, const icl512 &rv) {
        icl512 ret;
        ret.v0 = _mm_cmpgt_ps(lv.v0, rv.v0);
        ret.v1 = _mm_cmpgt_ps(lv.v1, rv.v1);
        ret.v2 = _mm_cmpgt_ps(lv.v2, rv.v2);
        ret.v3 = _mm_cmpgt_ps(lv.v3, rv.v3);
        return ret;
      }

      inline icl512 operator<=(const icl512 &lv, const icl512 &rv) {
        icl512 ret;
        ret.v0 = _mm_cmple_ps(lv.v0, rv.v0);
        ret.v1 = _mm_cmple_ps(lv.v1, rv.v1);
        ret.v2 = _mm_cmple_ps(lv.v2, rv.v2);
        ret.v3 = _mm_cmple_ps(lv.v3, rv.v3);
        return ret;
      }

      inline icl512 operator>=(const icl512 &lv, const icl512 &rv) {
        icl512 ret;
        ret.v0 = _mm_cmpge_ps(lv.v0, rv.v0);
        ret.v1 = _mm_cmpge_ps(lv.v1, rv.v1);
        ret.v2 = _mm_cmpge_ps(lv.v2, rv.v2);
        ret.v3 = _mm_cmpge_ps(lv.v3, rv.v3);
        return ret;
      }

      // -- comparison operations -- //

      // ++ logical operations ++ //

      inline icl128 operator&(const icl128 &lv, const icl128 &rv) {
        icl128 ret;
        ret.v0 = _mm_and_ps(lv.v0, rv.v0);
        return ret;
      }

      inline icl128 operator|(const icl128 &lv, const icl128 &rv) {
        icl128 ret;
        ret.v0 = _mm_or_ps(lv.v0, rv.v0);
        return ret;
      }

      inline icl128 operator^(const icl128 &lv, const icl128 &rv) {
        icl128 ret;
        ret.v0 = _mm_xor_ps(lv.v0, rv.v0);
        return ret;
      }

      inline icl128 andnot(const icl128 &lv, const icl128 &rv) {
        icl128 ret;
        ret.v0 = _mm_andnot_ps(rv.v0, lv.v0);
        return ret;
      }

      inline Icl128i operator&(const Icl128i &lv, const Icl128i &rv) {
        Icl128i ret;
        ret.v0 = _mm_and_si128(lv.v0, rv.v0);
        return ret;
      }

      inline Icl128i operator|(const Icl128i &lv, const Icl128i &rv) {
        Icl128i ret;
        ret.v0 = _mm_or_si128(lv.v0, rv.v0);
        return ret;
      }

      inline Icl128i operator^(const Icl128i &lv, const Icl128i &rv) {
        Icl128i ret;
        ret.v0 = _mm_xor_si128(lv.v0, rv.v0);
        return ret;
      }

      inline Icl128i andnot(const Icl128i &lv, const Icl128i &rv) {
        Icl128i ret;
        ret.v0 = _mm_andnot_si128(rv.v0, lv.v0);
        return ret;
      }

      inline icl256 operator&(const icl256 &lv, const icl256 &rv) {
        icl256 ret;
        ret.v0 = _mm_and_ps(lv.v0, rv.v0);
        ret.v1 = _mm_and_ps(lv.v1, rv.v1);
        return ret;
      }

      inline icl256 operator|(const icl256 &lv, const icl256 &rv) {
        icl256 ret;
        ret.v0 = _mm_or_ps(lv.v0, rv.v0);
        ret.v1 = _mm_or_ps(lv.v1, rv.v1);
        return ret;
      }

      inline icl256 operator^(const icl256 &lv, const icl256 &rv) {
        icl256 ret;
        ret.v0 = _mm_xor_ps(lv.v0, rv.v0);
        ret.v1 = _mm_xor_ps(lv.v1, rv.v1);
        return ret;
      }

      inline icl256 andnot(const icl256 &lv, const icl256 &rv) {
        icl256 ret;
        ret.v0 = _mm_andnot_ps(rv.v0, lv.v0);
        ret.v1 = _mm_andnot_ps(rv.v1, lv.v1);
        return ret;
      }

      inline Icl256i operator&(const Icl256i &lv, const Icl256i &rv) {
        Icl256i ret;
        ret.v0 = _mm_and_si128(lv.v0, rv.v0);
        ret.v1 = _mm_and_si128(lv.v1, rv.v1);
        return ret;
      }

      inline Icl256i operator|(const Icl256i &lv, const Icl256i &rv) {
        Icl256i ret;
        ret.v0 = _mm_or_si128(lv.v0, rv.v0);
        ret.v1 = _mm_or_si128(lv.v1, rv.v1);
        return ret;
      }

      inline Icl256i operator^(const Icl256i &lv, const Icl256i &rv) {
        Icl256i ret;
        ret.v0 = _mm_xor_si128(lv.v0, rv.v0);
        ret.v1 = _mm_xor_si128(lv.v1, rv.v1);
        return ret;
      }

      inline Icl256i andnot(const Icl256i &lv, const Icl256i &rv) {
        Icl256i ret;
        ret.v0 = _mm_andnot_si128(rv.v0, lv.v0);
        ret.v1 = _mm_andnot_si128(rv.v1, lv.v1);
        return ret;
      }

      inline icl512 operator&(const icl512 &lv, const icl512 &rv) {
        icl512 ret;
        ret.v0 = _mm_and_ps(lv.v0, rv.v0);
        ret.v1 = _mm_and_ps(lv.v1, rv.v1);
        ret.v2 = _mm_and_ps(lv.v2, rv.v2);
        ret.v3 = _mm_and_ps(lv.v3, rv.v3);
        return ret;
      }

      inline icl512 operator|(const icl512 &lv, const icl512 &rv) {
        icl512 ret;
        ret.v0 = _mm_or_ps(lv.v0, rv.v0);
        ret.v1 = _mm_or_ps(lv.v1, rv.v1);
        ret.v2 = _mm_or_ps(lv.v2, rv.v2);
        ret.v3 = _mm_or_ps(lv.v3, rv.v3);
        return ret;
      }

      inline icl512 operator^(const icl512 &lv, const icl512 &rv) {
        icl512 ret;
        ret.v0 = _mm_xor_ps(lv.v0, rv.v0);
        ret.v1 = _mm_xor_ps(lv.v1, rv.v1);
        ret.v2 = _mm_xor_ps(lv.v2, rv.v2);
        ret.v3 = _mm_xor_ps(lv.v3, rv.v3);
        return ret;
      }

      inline icl512 andnot(const icl512 &lv, const icl512 &rv) {
        icl512 ret;
        ret.v0 = _mm_andnot_ps(rv.v0, lv.v0);
        ret.v1 = _mm_andnot_ps(rv.v1, lv.v1);
        ret.v2 = _mm_andnot_ps(rv.v2, lv.v2);
        ret.v3 = _mm_andnot_ps(rv.v3, lv.v3);
        return ret;
      }

      inline Icl512i operator&(const Icl512i &lv, const Icl512i &rv) {
        Icl512i ret;
        ret.v0 = _mm_and_si128(lv.v0, rv.v0);
        ret.v1 = _mm_and_si128(lv.v1, rv.v1);
        ret.v2 = _mm_and_si128(lv.v2, rv.v2);
        ret.v3 = _mm_and_si128(lv.v3, rv.v3);
        return ret;
      }

      inline Icl512i operator|(const Icl512i &lv, const Icl512i &rv) {
        Icl512i ret;
        ret.v0 = _mm_or_si128(lv.v0, rv.v0);
        ret.v1 = _mm_or_si128(lv.v1, rv.v1);
        ret.v2 = _mm_or_si128(lv.v2, rv.v2);
        ret.v3 = _mm_or_si128(lv.v3, rv.v3);
        return ret;
      }

      inline Icl512i operator^(const Icl512i &lv, const Icl512i &rv) {
        Icl512i ret;
        ret.v0 = _mm_xor_si128(lv.v0, rv.v0);
        ret.v1 = _mm_xor_si128(lv.v1, rv.v1);
        ret.v2 = _mm_xor_si128(lv.v2, rv.v2);
        ret.v3 = _mm_xor_si128(lv.v3, rv.v3);
        return ret;
      }

      inline Icl512i andnot(const Icl512i &lv, const Icl512i &rv) {
        Icl512i ret;
        ret.v0 = _mm_andnot_si128(rv.v0, lv.v0);
        ret.v1 = _mm_andnot_si128(rv.v1, lv.v1);
        ret.v2 = _mm_andnot_si128(rv.v2, lv.v2);
        ret.v3 = _mm_andnot_si128(rv.v3, lv.v3);
        return ret;
      }

      // -- logical operations -- //

      // ++ shift operetions ++ //

      inline Icl128i& operator<<(Icl128i &v, const int i) {
        v.v0 = _mm_slli_epi32(v.v0, i);
        return v;
      }

      inline Icl128i& operator>>(Icl128i &v, const int i) {
        v.v0 = _mm_srai_epi32(v.v0, i);
        return v;
      }

      inline Icl256i& operator<<(Icl256i &v, const int i) {
        v.v0 = _mm_slli_epi32(v.v0, i);
        v.v1 = _mm_slli_epi32(v.v1, i);
        return v;
      }

      inline Icl256i& operator>>(Icl256i &v, const int i) {
        v.v0 = _mm_srai_epi32(v.v0, i);
        v.v1 = _mm_srai_epi32(v.v1, i);
        return v;
      }

      inline Icl512i& operator<<(Icl512i &v, const int i) {
        v.v0 = _mm_slli_epi32(v.v0, i);
        v.v1 = _mm_slli_epi32(v.v1, i);
        v.v2 = _mm_slli_epi32(v.v2, i);
        v.v3 = _mm_slli_epi32(v.v3, i);
        return v;
      }

      inline Icl512i& operator>>(Icl512i &v, const int i) {
        v.v0 = _mm_srai_epi32(v.v0, i);
        v.v1 = _mm_srai_epi32(v.v1, i);
        v.v2 = _mm_srai_epi32(v.v2, i);
        v.v3 = _mm_srai_epi32(v.v3, i);
        return v;
      }

      // -- shift operations -- //

      // ++ min-max operations ++ //

      inline icl128i8u min(const icl128i8u &lv, const icl128i8u &rv) {
        icl128i8u ret;
        ret.v0 = _mm_min_epu8(lv.v0, rv.v0);
        return ret;
      }

      inline icl128i8u max(const icl128i8u &lv, const icl128i8u &rv) {
        icl128i8u ret;
        ret.v0 = _mm_max_epu8(lv.v0, rv.v0);
        return ret;
      }

      inline icl128i16s min(const icl128i16s &lv, const icl128i16s &rv) {
        icl128i16s ret;
        ret.v0 = _mm_min_epi16(lv.v0, rv.v0);
        return ret;
      }

      inline icl128i16s max(const icl128i16s &lv, const icl128i16s &rv) {
        icl128i16s ret;
        ret.v0 = _mm_max_epi16(lv.v0, rv.v0);
        return ret;
      }

      inline icl256i16s min(const icl256i16s &lv, const icl256i16s &rv) {
        icl256i16s ret;
        ret.v0 = _mm_min_epi16(lv.v0, rv.v0);
        ret.v1 = _mm_min_epi16(lv.v1, rv.v1);
        return ret;
      }

      inline icl256i16s max(const icl256i16s &lv, const icl256i16s &rv) {
        icl256i16s ret;
        ret.v0 = _mm_max_epi16(lv.v0, rv.v0);
        ret.v1 = _mm_max_epi16(lv.v1, rv.v1);
        return ret;
      }

      inline icl128 min(const icl128 &lv, const icl128 &rv) {
        icl128 ret;
        ret.v0 = _mm_min_ps(lv.v0, rv.v0);
        return ret;
      }

      inline icl128 max(const icl128 &lv, const icl128 &rv) {
        icl128 ret;
        ret.v0 = _mm_max_ps(lv.v0, rv.v0);
        return ret;
      }

      inline icl256 min(const icl256 &lv, const icl256 &rv) {
        icl256 ret;
        ret.v0 = _mm_min_ps(lv.v0, rv.v0);
        ret.v1 = _mm_min_ps(lv.v1, rv.v1);
        return ret;
      }

      inline icl256 max(const icl256 &lv, const icl256 &rv) {
        icl256 ret;
        ret.v0 = _mm_max_ps(lv.v0, rv.v0);
        ret.v1 = _mm_max_ps(lv.v1, rv.v1);
        return ret;
      }

      inline icl512 min(const icl512 &lv, const icl512 &rv) {
        icl512 ret;
        ret.v0 = _mm_min_ps(lv.v0, rv.v0);
        ret.v1 = _mm_min_ps(lv.v1, rv.v1);
        ret.v2 = _mm_min_ps(lv.v2, rv.v2);
        ret.v3 = _mm_min_ps(lv.v3, rv.v3);
        return ret;
      }

      inline icl512 max(const icl512 &lv, const icl512 &rv) {
        icl512 ret;
        ret.v0 = _mm_max_ps(lv.v0, rv.v0);
        ret.v1 = _mm_max_ps(lv.v1, rv.v1);
        ret.v2 = _mm_max_ps(lv.v2, rv.v2);
        ret.v3 = _mm_max_ps(lv.v3, rv.v3);
        return ret;
      }

      // -- min-max operations -- //


      // ++ absosulte values ++ //

    #ifdef ICL_HAVE_SSE3
      inline icl128i8u abs(const icl128i8u &v) {
        icl128i8u ret;
        ret.v0 = _mm_abs_epi8(v.v0);
        return ret;
      }

      inline icl128i16s abs(const icl128i16s &v) {
        icl128i16s ret;
        ret.v0 = _mm_abs_epi16(v.v0);
        return ret;
      }

      inline icl128i32s abs(const icl128i32s &v) {
        icl128i32s ret;
        ret.v0 = _mm_abs_epi32(v.v0);
        return ret;
      }

      inline icl256i16s abs(const icl256i16s &v) {
        icl256i16s ret;
        ret.v0 = _mm_abs_epi16(v.v0);
        ret.v1 = _mm_abs_epi16(v.v1);
        return ret;
      }

      inline icl256i32s abs(const icl256i32s &v) {
        icl256i32s ret;
        ret.v0 = _mm_abs_epi32(v.v0);
        ret.v1 = _mm_abs_epi32(v.v1);
        return ret;
      }

      inline icl512i32s abs(const icl512i32s &v) {
        icl512i32s ret;
        ret.v0 = _mm_abs_epi32(v.v0);
        ret.v1 = _mm_abs_epi32(v.v1);
        ret.v2 = _mm_abs_epi32(v.v2);
        ret.v3 = _mm_abs_epi32(v.v3);
        return ret;
      }
    #else
      // TODO: without SSE3
    #endif

      inline icl128 abs(const icl128 &v) {
        icl128 ret;
        ret.v0 = _mm_andnot_ps(icl128(-0.0f), v.v0);
        return ret;
      }

      inline icl256 abs(const icl256 &v) {
        icl128 tmp(-0.0f);
        icl256 ret;
        ret.v0 = _mm_andnot_ps(tmp.v0, v.v0);
        ret.v1 = _mm_andnot_ps(tmp.v0, v.v1);
        return ret;
      }

      inline icl512 abs(const icl512 &v) {
        icl128 tmp(-0.0f);
        icl512 ret;
        ret.v0 = _mm_andnot_ps(tmp.v0, v.v0);
        ret.v1 = _mm_andnot_ps(tmp.v0, v.v1);
        ret.v2 = _mm_andnot_ps(tmp.v0, v.v2);
        ret.v3 = _mm_andnot_ps(tmp.v0, v.v3);
        return ret;
      }

      // -- absosulte values -- //


      // ++ squared root ++ //

      inline icl128 sqrt(const icl128 &v) {
        icl128 r;
        r.v0 = _mm_sqrt_ps(v.v0);
        return r;
      }

      inline icl256 sqrt(const icl256 &v) {
        icl256 r;
        r.v0 = _mm_sqrt_ps(v.v0);
        r.v1 = _mm_sqrt_ps(v.v1);
        return r;
      }

      inline icl512 sqrt(const icl512 &v) {
        icl512 r;
        r.v0 = _mm_sqrt_ps(v.v0);
        r.v1 = _mm_sqrt_ps(v.v1);
        r.v2 = _mm_sqrt_ps(v.v2);
        r.v3 = _mm_sqrt_ps(v.v3);
        return r;
      }

      inline icl128d sqrt(const icl128d &v) {
        icl128d r;
        r.v0 = _mm_sqrt_pd(v.v0);
        return r;
      }

      inline icl256d sqrt(const icl256d &v) {
        icl256d r;
        r.v0 = _mm_sqrt_pd(v.v0);
        r.v1 = _mm_sqrt_pd(v.v1);
        return r;
      }

      inline icl512d sqrt(const icl512d &v) {
        icl512d r;
        r.v0 = _mm_sqrt_pd(v.v0);
        r.v1 = _mm_sqrt_pd(v.v1);
        r.v2 = _mm_sqrt_pd(v.v2);
        r.v3 = _mm_sqrt_pd(v.v3);
        return r;
      }

      inline icl1024d sqrt(const icl1024d &v) {
        icl1024d r;
        r.v0 = _mm_sqrt_pd(v.v0);
        r.v1 = _mm_sqrt_pd(v.v1);
        r.v2 = _mm_sqrt_pd(v.v2);
        r.v3 = _mm_sqrt_pd(v.v3);
        r.v4 = _mm_sqrt_pd(v.v4);
        r.v5 = _mm_sqrt_pd(v.v5);
        r.v6 = _mm_sqrt_pd(v.v6);
        r.v7 = _mm_sqrt_pd(v.v7);
        return r;
      }

      // -- squared root -- //


      // ++ cube root ++ //

      inline icl128 cbrt(const icl128 &v) {
        icl128i32s tmp = icl128i32s(_mm_castps_si128(v));
        tmp = tmp / icl128i32s(3) + icl128i32s(709921077);
        icl128 a = icl128(_mm_castsi128_ps(tmp));
        icl128 a3 = a * a * a;
        return a * (a3 + v + v) * (a3 + a3 + v).rcp();
      }

      inline icl256 cbrt(const icl256 &v) {
        __m128i t0 = _mm_castps_si128(v.v0);
        __m128i t1 = _mm_castps_si128(v.v1);
        icl256i32s tmp = icl256i32s(t0, t1);
        tmp = tmp / icl256i32s(3) + icl256i32s(709921077);
        icl256 a = icl256(_mm_castsi128_ps(tmp.v0),
                          _mm_castsi128_ps(tmp.v1));
        icl256 a3 = a * a * a;
        return a * (a3 + v + v) * (a3 + a3 + v).rcp();
      }

      inline icl512 cbrt(const icl512 &v) {
        __m128i t0 = _mm_castps_si128(v.v0);
        __m128i t1 = _mm_castps_si128(v.v1);
        __m128i t2 = _mm_castps_si128(v.v2);
        __m128i t3 = _mm_castps_si128(v.v3);
        icl512i32s tmp = icl512i32s(t0, t1, t2, t3);
        tmp = tmp / icl512i32s(3) + icl512i32s(709921077);
        icl512 a = icl512(_mm_castsi128_ps(tmp.v0),
                          _mm_castsi128_ps(tmp.v1),
                          _mm_castsi128_ps(tmp.v2),
                          _mm_castsi128_ps(tmp.v3));
        icl512 a3 = a * a * a;
        return a * (a3 + v + v) * (a3 + a3 + v).rcp();
      }

      // -- cube root -- //

      // -- operations on SSE types -- //

      typedef icl128 icl32fx4;
      typedef icl256 icl32fx8;
      typedef icl512 icl32fx16;
      typedef icl128i8u icl8ux16;
      typedef icl128i16s icl16sx8;
      typedef icl128i32s icl32sx4;
      typedef icl256i16s icl16sx16;
      typedef icl256i32s icl32sx8;
      typedef icl512i32s icl32sx16;
      typedef icl128d icl64fx2;
      typedef icl256d icl64fx4;
      typedef icl512d icl64fx8;
      typedef icl1024d icl64fx16;

    #endif

  } // namespace utils
}