/* -----------------------------------------------------------------------------

The copyright in this software is being made available under the Clear BSD

License, included below. No patent rights, trademark rights and/or 

other Intellectual Property Rights other than the copyrights concerning 

the Software are granted under this license.

The Clear BSD License

Copyright (c) 2019-2026, Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V. & The VVenC Authors.

All rights reserved.

Redistribution and use in source and binary forms, with or without modification,

are permitted (subject to the limitations in the disclaimer below) provided that

the following conditions are met:

     * Redistributions of source code must retain the above copyright notice,

     this list of conditions and the following disclaimer.

     * 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.

     * Neither the name of the copyright holder nor the names of its

     contributors may be used to endorse or promote products derived from this

     software without specific prior written permission.

NO EXPRESS OR IMPLIED LICENSES TO ANY PARTY'S PATENT RIGHTS ARE GRANTED BY

THIS LICENSE. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND

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 THE COPYRIGHT HOLDER OR

CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,

EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,

PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR

BUSINESS INTERRUPTION) 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.

------------------------------------------------------------------------------------------- */

/** \file     SampleAdaptiveOffsetX86.h

    \brief    SAO filter class

*/

#pragma once

#include "CommonDefX86.h"

#include "SampleAdaptiveOffset.h"

#if defined(TARGET_SIMD_X86)  && ENABLE_SIMD_OPT_SAO

//! \ingroup CommonLib

//! \{

namespace vvenc {

#define SAO_NUM_OFFSETS                     4 /* number of SAO offset values */

#define SAO_EO_NUM_CATEGORIES               (SAO_NUM_OFFSETS + 1) /* number of different eo categories */

template <X86_VEXT vext>

void offsetBlock_SIMD( const int     channelBitDepth,

                                        const ClpRng& clpRng,

                                        int           typeIdx,

                                        int*          offset,

                                        int           startIdx,

                                        const Pel*    srcBlk,

                                        Pel*          resBlk,

                                        ptrdiff_t     srcStride,

                                        ptrdiff_t     resStride,

                                        int           width,

                                        int           height,

                                        uint8_t       availMask,

//                                        bool          isLeftAvail,

//                                        bool          isRightAvail,

//                                        bool          isAboveAvail,

//                                        bool          isBelowAvail,

//                                        bool          isAboveLeftAvail,

//                                        bool          isAboveRightAvail,

//                                        bool          isBelowLeftAvail,

//                                        bool          isBelowRightAvail,

                                        std::vector<int8_t> &signLineBuf1,

                                        std::vector<int8_t> &signLineBuf2)

{

  int x,y, startX, startY, endX, endY, edgeType;

  int firstLineStartX, firstLineEndX, lastLineStartX, lastLineEndX;

  int8_t signLeft, signRight, signDown;

  const Pel* srcLine = srcBlk;

  Pel* resLine = resBlk;

  switch(typeIdx)

  {

  case SAO_TYPE_EO_0:

  {

    if (availMask&LeftAvail && availMask&RightAvail)

    {

      int8_t p_eo_offsets[16] = {0,};

      for (int i = 0; i < SAO_EO_NUM_CATEGORIES; i++)

      {

        p_eo_offsets[i] = offset[i];

      }

#ifdef USE_AVX2

      // AVX2

      if( ( width & 15 ) == 0 && vext >= AVX2 )

      {

        __m256i vsrca,vsrcal,vsrcar;

        __m256i vbaseoffset = _mm256_set1_epi16(2) ;

        __m256i vplusone = _mm256_set1_epi16(1);

        __m256i vzero = _mm256_set1_epi8(0);

        __m256i vibdimax = _mm256_set1_epi16((1<<channelBitDepth) -1 );

        __m256i voffsettbl =  _mm256_broadcastsi128_si256(_mm_loadu_si128((__m128i*)p_eo_offsets));

        for (y=0; y< height; y++)

        {

          for (x=0; x< width; x+=16)

          {

            vsrca = _mm256_loadu_si256((__m256i*)&srcLine[x]);

            vsrcal = _mm256_loadu_si256((__m256i*)&srcLine[x-1]);

            vsrcar = _mm256_loadu_si256((__m256i*)&srcLine[x+1]);

            vsrcal = _mm256_sub_epi16(vsrca, vsrcal);

            vsrcar = _mm256_sub_epi16(vsrca, vsrcar);

            __m256i vsignl = _mm256_sign_epi16(vplusone, vsrcal);

            __m256i vsignr = _mm256_sign_epi16(vplusone, vsrcar);

            __m256i vsign = _mm256_add_epi16(_mm256_add_epi16(vsignl, vsignr), vbaseoffset);

            __m256i veoffsets = _mm256_shuffle_epi8(voffsettbl, vsign);

            veoffsets = _mm256_slli_epi16 (veoffsets,8);

            veoffsets = _mm256_srai_epi16 (veoffsets,8);

            vsrca = _mm256_add_epi16(vsrca, veoffsets);

            vsrca    = _mm256_min_epi16(_mm256_max_epi16(vsrca, vzero), vibdimax);

            _mm256_storeu_si256((__m256i*)&resLine[x], vsrca);

          }

          srcLine  += srcStride;

          resLine += resStride;

        }

      }

      else

#endif

      {

        __m128i vsrca,vsrcal,vsrcar;

        __m128i vbaseoffset = _mm_set1_epi16(2) ;

        __m128i vplusone = _mm_set1_epi16(1);

        __m128i vzero = _mm_set1_epi8(0);

        __m128i vibdimax = _mm_set1_epi16((1<<channelBitDepth) -1 );

        __m128i voffsettbl = _mm_loadu_si128((__m128i*)p_eo_offsets);

        for (y=0; y< height; y++)

        {

          for (x=0; x< width; x+=8)

          {

            vsrca = _mm_loadu_si128((__m128i*)&srcLine[x]);

            vsrcal = _mm_loadu_si128((__m128i*)&srcLine[x-1]);

            vsrcar = _mm_loadu_si128((__m128i*)&srcLine[x+1]);

            vsrcal = _mm_sub_epi16(vsrca, vsrcal);

            vsrcar = _mm_sub_epi16(vsrca, vsrcar);

            __m128i vsignl = _mm_sign_epi16(vplusone, vsrcal);

            __m128i vsignr = _mm_sign_epi16(vplusone, vsrcar);

            __m128i vsign = _mm_add_epi16(_mm_add_epi16(vsignl, vsignr), vbaseoffset);

            __m128i veoffsets = _mm_shuffle_epi8(voffsettbl, vsign);

            veoffsets = _mm_slli_epi16 (veoffsets,8);

            veoffsets = _mm_srai_epi16 (veoffsets,8);

            vsrca = _mm_add_epi16(vsrca, veoffsets);

            vsrca    = _mm_min_epi16(_mm_max_epi16(vsrca, vzero), vibdimax);

            _mm_store_si128((__m128i*)&resLine[x], vsrca);

          }

          srcLine  += srcStride;

          resLine += resStride;

        }

      }

    }

    else

    {

      offset += 2;

      startX = availMask&LeftAvail ? 0 : 1;

      endX   = availMask&RightAvail ? width : (width -1);

      for (y=0; y< height; y++)

      {

        signLeft = (int8_t)sgn(srcLine[startX] - srcLine[startX-1]);

        for (x=startX; x< endX; x++)

        {

          signRight = (int8_t)sgn(srcLine[x] - srcLine[x+1]);

          edgeType =  signRight + signLeft;

          signLeft  = -signRight;

          resLine[x] = ClipPel<int>( srcLine[x] + offset[edgeType], clpRng);

        }

        srcLine  += srcStride;

        resLine += resStride;

      }

    }

  }

  break;

  case SAO_TYPE_EO_90:

  {

    int8_t p_eo_offsets[16] = {0,};

    for (int i = 0; i < SAO_EO_NUM_CATEGORIES; i++)

    {

      p_eo_offsets[i] = offset[i];

    }

    const Pel* srcLineAbove= srcLine- srcStride;

    const Pel* srcLineBelow= srcLine+ srcStride;

    startY=0;

    if (!(availMask&AboveAvail))

    {

      startY=1;

      srcLineAbove= srcLine;

      srcLine  += srcStride;

      resLine += resStride;

      srcLineBelow= srcLine+ srcStride;

    }

    endY=height;

    if (!(availMask&BelowAvail))

    {

      endY=height-1;

    }

#ifdef USE_AVX2

    // AVX2

    if( ( width & 15 ) == 0 && ( vext >= AVX2 ) )

    {

      __m256i vsrca,vsrcat,vsrcab;

      __m256i vbaseoffset = _mm256_set1_epi16(2) ;

      __m256i vplusone = _mm256_set1_epi16(1);

      __m256i vzero = _mm256_set1_epi8(0);

      __m256i vibdimax = _mm256_set1_epi16((1<<channelBitDepth) -1 );

      __m256i voffsettbl =  _mm256_broadcastsi128_si256(_mm_loadu_si128((__m128i*)p_eo_offsets));

      const Pel* srcLineBelow= srcLine+ srcStride;

      for (y=startY; y< endY; y++)

      {

        for (x=0; x< width; x+=16)

        {

          vsrca = _mm256_loadu_si256((__m256i*)&srcLine[x]);

          vsrcat = _mm256_loadu_si256((__m256i*)&srcLineAbove[x]);

          vsrcab = _mm256_loadu_si256((__m256i*)&srcLineBelow[x]);

          vsrcat = _mm256_sub_epi16(vsrca, vsrcat);

          vsrcab = _mm256_sub_epi16(vsrca, vsrcab);

          __m256i vsignt = _mm256_sign_epi16(vplusone, vsrcat);

          __m256i vsignb = _mm256_sign_epi16(vplusone, vsrcab);

          __m256i vsign = _mm256_add_epi16(_mm256_add_epi16(vsignt, vsignb), vbaseoffset);

          __m256i veoffsets = _mm256_shuffle_epi8(voffsettbl, vsign);

          veoffsets = _mm256_slli_epi16 (veoffsets,8);

          veoffsets = _mm256_srai_epi16 (veoffsets,8);

          vsrca = _mm256_add_epi16(vsrca, veoffsets);

          vsrca    = _mm256_min_epi16(_mm256_max_epi16(vsrca, vzero), vibdimax);

          _mm256_storeu_si256((__m256i*)&resLine[x], vsrca);

        }

        srcLine  += srcStride;

        srcLineBelow += srcStride;

        srcLineAbove += srcStride;

        resLine += resStride;

      }

    }

    else

#endif

    {

      __m128i vsrca,vsrcat,vsrcab;

      __m128i vbaseoffset = _mm_set1_epi16(2) ;

      __m128i vplusone = _mm_set1_epi16(1);

      __m128i vzero = _mm_set1_epi8(0);

      __m128i vibdimax = _mm_set1_epi16((1<<channelBitDepth) -1 );

      __m128i voffsettbl = _mm_loadu_si128((__m128i*)p_eo_offsets);

      for (y=startY; y< endY; y++)

      {

        for (x=0; x< width; x+=8)

        {

          vsrca = _mm_loadu_si128((__m128i*)&srcLine[x]);

          vsrcat = _mm_loadu_si128((__m128i*)&srcLineAbove[x]);

          vsrcab = _mm_loadu_si128((__m128i*)&srcLineBelow[x]);

          vsrcat = _mm_sub_epi16(vsrca, vsrcat);

          vsrcab = _mm_sub_epi16(vsrca, vsrcab);

          __m128i vsignt = _mm_sign_epi16(vplusone, vsrcat);

          __m128i vsignb = _mm_sign_epi16(vplusone, vsrcab);

          __m128i vsign = _mm_add_epi16(_mm_add_epi16(vsignt, vsignb), vbaseoffset);

          __m128i veoffsets = _mm_shuffle_epi8(voffsettbl, vsign);

          veoffsets = _mm_slli_epi16 (veoffsets,8);

          veoffsets = _mm_srai_epi16 (veoffsets,8);

          vsrca = _mm_add_epi16(vsrca, veoffsets);

          vsrca    = _mm_min_epi16(_mm_max_epi16(vsrca, vzero), vibdimax);

          _mm_store_si128((__m128i*)&resLine[x], vsrca);

        }

        srcLine  += srcStride;

        srcLineBelow += srcStride;

        srcLineAbove += srcStride;

        resLine += resStride;

      }

    }

  }

  break;

  case SAO_TYPE_EO_135:

  {

//    if (isLeftAvail && isRightAvail && isAboveLeftAvail && isBelowRightAvail )

    if((LeftAvail|RightAvail|AboveLeftAvail|BelowRightAvail) == (int)(availMask&(LeftAvail|RightAvail|AboveLeftAvail|BelowRightAvail)))

    {

      int8_t p_eo_offsets[16] = {0,};

      for (int i = 0; i < SAO_EO_NUM_CATEGORIES; i++)

      {

        p_eo_offsets[i] = offset[i];

      }

      const Pel* srcLineAbove= srcLine- srcStride;

      const Pel* srcLineBelow= srcLine+ srcStride;

      startY=0;

      if (!(availMask&AboveAvail))

      {

        startY=1;

        srcLineAbove= srcLine;

        srcLine  += srcStride;

        resLine += resStride;

        srcLineBelow= srcLine+ srcStride;

      }

      endY=height;

      if (!(availMask&BelowAvail))

      {

        endY=height-1;

      }

#ifdef USE_AVX2

      // AVX2

      if( ( width & 15 ) == 0 && vext >= AVX2 )

      {

        __m256i vsrca,vsrcat,vsrcab;

        __m256i vbaseoffset = _mm256_set1_epi16(2) ;

        __m256i vplusone = _mm256_set1_epi16(1);

        __m256i vzero = _mm256_set1_epi8(0);

        __m256i vibdimax = _mm256_set1_epi16((1<<channelBitDepth) -1 );

        __m256i voffsettbl =  _mm256_broadcastsi128_si256(_mm_loadu_si128((__m128i*)p_eo_offsets));

        const Pel* srcLineBelow= srcLine+ srcStride;

        for (y=startY; y< endY; y++)

        {

          for (x=0; x< width; x+=16)

          {

            vsrca = _mm256_loadu_si256((__m256i*)&srcLine[x]);

            vsrcat = _mm256_loadu_si256((__m256i*)&srcLineAbove[x-1]);

            vsrcab = _mm256_loadu_si256((__m256i*)&srcLineBelow[x+1]);

            vsrcat = _mm256_sub_epi16(vsrca, vsrcat);

            vsrcab = _mm256_sub_epi16(vsrca, vsrcab);

            __m256i vsignt = _mm256_sign_epi16(vplusone, vsrcat);

            __m256i vsignb = _mm256_sign_epi16(vplusone, vsrcab);

            __m256i vsign = _mm256_add_epi16(_mm256_add_epi16(vsignt, vsignb), vbaseoffset);

            __m256i veoffsets = _mm256_shuffle_epi8(voffsettbl, vsign);

            veoffsets = _mm256_slli_epi16 (veoffsets,8);

            veoffsets = _mm256_srai_epi16 (veoffsets,8);

            vsrca = _mm256_add_epi16(vsrca, veoffsets);

            vsrca    = _mm256_min_epi16(_mm256_max_epi16(vsrca, vzero), vibdimax);

            _mm256_storeu_si256((__m256i*)&resLine[x], vsrca);

          }

          srcLine  += srcStride;

          srcLineBelow += srcStride;

          srcLineAbove += srcStride;

          resLine += resStride;

        }

      }

      else

#endif

      {

        __m128i vsrca,vsrcat,vsrcab;

        __m128i vbaseoffset = _mm_set1_epi16(2) ;

        __m128i vplusone = _mm_set1_epi16(1);

        __m128i vzero = _mm_set1_epi8(0);

        __m128i vibdimax = _mm_set1_epi16((1<<channelBitDepth) -1 );

        __m128i voffsettbl = _mm_loadu_si128((__m128i*)p_eo_offsets);

        for (y=startY; y< endY; y++)

        {

          for (x=0; x< width; x+=8)

          {

            vsrca = _mm_loadu_si128((__m128i*)&srcLine[x]);

            vsrcat = _mm_loadu_si128((__m128i*)&srcLineAbove[x-1]);

            vsrcab = _mm_loadu_si128((__m128i*)&srcLineBelow[x+1]);

            vsrcat = _mm_sub_epi16(vsrca, vsrcat);

            vsrcab = _mm_sub_epi16(vsrca, vsrcab);

            __m128i vsignt = _mm_sign_epi16(vplusone, vsrcat);

            __m128i vsignb = _mm_sign_epi16(vplusone, vsrcab);

            __m128i vsign = _mm_add_epi16(_mm_add_epi16(vsignt, vsignb), vbaseoffset);

            __m128i veoffsets = _mm_shuffle_epi8(voffsettbl, vsign);

            veoffsets = _mm_slli_epi16 (veoffsets,8);

            veoffsets = _mm_srai_epi16 (veoffsets,8);

            vsrca = _mm_add_epi16(vsrca, veoffsets);

            vsrca    = _mm_min_epi16(_mm_max_epi16(vsrca, vzero), vibdimax);

            _mm_store_si128((__m128i*)&resLine[x], vsrca);

          }

          srcLine  += srcStride;

          srcLineBelow += srcStride;

          srcLineAbove += srcStride;

          resLine += resStride;

        }

      }

    }

    else

    {

      offset += 2;

      int8_t *signUpLine, *signDownLine, *signTmpLine;

      signUpLine  = &signLineBuf1[0];

      signDownLine= &signLineBuf2[0];

      startX = availMask&LeftAvail ? 0 : 1 ;

      endX   = availMask&RightAvail ? width : (width-1);

      //prepare 2nd line's upper sign

      const Pel* srcLineBelow= srcLine+ srcStride;

      for (x=startX; x< endX+1; x++)

      {

        signUpLine[x] = (int8_t)sgn(srcLineBelow[x] - srcLine[x- 1]);

      }

      //1st line

      const Pel* srcLineAbove= srcLine- srcStride;

      firstLineStartX = availMask&AboveLeftAvail ? 0 : 1;

      firstLineEndX   = availMask&AboveAvail? endX: 1;

      for(x= firstLineStartX; x< firstLineEndX; x++)

      {

        edgeType  =  sgn(srcLine[x] - srcLineAbove[x- 1]) - signUpLine[x+1];

        resLine[x] = ClipPel<int>( srcLine[x] + offset[edgeType], clpRng);

      }

      srcLine  += srcStride;

      resLine  += resStride;

      //middle lines

      for (y= 1; y< height-1; y++)

      {

        srcLineBelow= srcLine+ srcStride;

        for (x=startX; x<endX; x++)

        {

          signDown =  (int8_t)sgn(srcLine[x] - srcLineBelow[x+ 1]);

          edgeType =  signDown + signUpLine[x];

          resLine[x] = ClipPel<int>( srcLine[x] + offset[edgeType], clpRng);

          signDownLine[x+1] = -signDown;

        }

        signDownLine[startX] = (int8_t)sgn(srcLineBelow[startX] - srcLine[startX-1]);

        signTmpLine  = signUpLine;

        signUpLine   = signDownLine;

        signDownLine = signTmpLine;

        srcLine += srcStride;

        resLine += resStride;

      }

      //last line

      srcLineBelow= srcLine+ srcStride;

      lastLineStartX = availMask&BelowAvail ? startX : (width -1);

      lastLineEndX   = availMask&BelowRightAvail ? width : (width -1);

      for(x= lastLineStartX; x< lastLineEndX; x++)

      {

        edgeType =  sgn(srcLine[x] - srcLineBelow[x+ 1]) + signUpLine[x];

        resLine[x] = ClipPel<int>( srcLine[x] + offset[edgeType], clpRng);

      }

    }

  }

  break;

  case SAO_TYPE_EO_45:

  {

//    if (isLeftAvail && isRightAvail && isAboveLeftAvail && isBelowRightAvail )

    if((LeftAvail|RightAvail|AboveLeftAvail|BelowRightAvail) == ((int)availMask&(LeftAvail|RightAvail|AboveLeftAvail|BelowRightAvail)))

    {

      int8_t p_eo_offsets[16] = {0,};

      for (int i = 0; i < SAO_EO_NUM_CATEGORIES; i++)

      {

        p_eo_offsets[i] = offset[i];

      }

      const Pel* srcLineAbove= srcLine- srcStride;

      const Pel* srcLineBelow= srcLine+ srcStride;

      startY=0;

      if (!(availMask&AboveAvail))

      {

        startY=1;

        srcLineAbove= srcLine;

        srcLine  += srcStride;

        resLine += resStride;

        srcLineBelow= srcLine+ srcStride;

      }

      endY=height;

      if (!(availMask&BelowAvail))

      {

        endY=height-1;

      }

#ifdef USE_AVX2

      // AVX2

      if( ( width & 15 ) == 0 && vext >= AVX2 )

      {

        __m256i vsrca,vsrcat,vsrcab;

        __m256i vbaseoffset = _mm256_set1_epi16(2) ;

        __m256i vplusone = _mm256_set1_epi16(1);

        __m256i vzero = _mm256_set1_epi8(0);

        __m256i vibdimax = _mm256_set1_epi16((1<<channelBitDepth) -1 );

        __m256i voffsettbl =  _mm256_broadcastsi128_si256(_mm_loadu_si128((__m128i*)p_eo_offsets));

        const Pel* srcLineBelow= srcLine+ srcStride;

        for (y=startY; y< endY; y++)

        {

          for (x=0; x< width; x+=16)

          {

            vsrca = _mm256_loadu_si256((__m256i*)&srcLine[x]);

            vsrcat = _mm256_loadu_si256((__m256i*)&srcLineAbove[x+1]);

            vsrcab = _mm256_loadu_si256((__m256i*)&srcLineBelow[x-1]);

            vsrcat = _mm256_sub_epi16(vsrca, vsrcat);

            vsrcab = _mm256_sub_epi16(vsrca, vsrcab);

            __m256i vsignt = _mm256_sign_epi16(vplusone, vsrcat);

            __m256i vsignb = _mm256_sign_epi16(vplusone, vsrcab);

            __m256i vsign = _mm256_add_epi16(_mm256_add_epi16(vsignt, vsignb), vbaseoffset);

            __m256i veoffsets = _mm256_shuffle_epi8(voffsettbl, vsign);

            veoffsets = _mm256_slli_epi16 (veoffsets,8);

            veoffsets = _mm256_srai_epi16 (veoffsets,8);

            vsrca = _mm256_add_epi16(vsrca, veoffsets);

            vsrca    = _mm256_min_epi16(_mm256_max_epi16(vsrca, vzero), vibdimax);

            _mm256_storeu_si256((__m256i*)&resLine[x], vsrca);

          }

          srcLine  += srcStride;

          srcLineBelow += srcStride;

          srcLineAbove += srcStride;

          resLine += resStride;

        }

      }

      else

#endif

      {

        __m128i vsrca,vsrcat,vsrcab;

        __m128i vbaseoffset = _mm_set1_epi16(2) ;

        __m128i vplusone = _mm_set1_epi16(1);

        __m128i vzero = _mm_set1_epi8(0);

        __m128i vibdimax = _mm_set1_epi16((1<<channelBitDepth) -1 );

        __m128i voffsettbl = _mm_loadu_si128((__m128i*)p_eo_offsets);

        for (y=startY; y< endY; y++)

        {

          for (x=0; x< width; x+=8)

          {

            vsrca = _mm_loadu_si128((__m128i*)&srcLine[x]);

            vsrcat = _mm_loadu_si128((__m128i*)&srcLineAbove[x+1]);

            vsrcab = _mm_loadu_si128((__m128i*)&srcLineBelow[x-1]);

            vsrcat = _mm_sub_epi16(vsrca, vsrcat);

            vsrcab = _mm_sub_epi16(vsrca, vsrcab);

            __m128i vsignt = _mm_sign_epi16(vplusone, vsrcat);

            __m128i vsignb = _mm_sign_epi16(vplusone, vsrcab);

            __m128i vsign = _mm_add_epi16(_mm_add_epi16(vsignt, vsignb), vbaseoffset);

            __m128i veoffsets = _mm_shuffle_epi8(voffsettbl, vsign);

            veoffsets = _mm_slli_epi16 (veoffsets,8);

            veoffsets = _mm_srai_epi16 (veoffsets,8);

            vsrca = _mm_add_epi16(vsrca, veoffsets);

            vsrca    = _mm_min_epi16(_mm_max_epi16(vsrca, vzero), vibdimax);

            _mm_store_si128((__m128i*)&resLine[x], vsrca);

          }

          srcLine  += srcStride;

          srcLineBelow += srcStride;

          srcLineAbove += srcStride;

          resLine += resStride;

        }

      }

    }

    else

    {

      offset += 2;

      int8_t *signUpLine = &signLineBuf1[1];

      startX = availMask&LeftAvail ? 0 : 1;

      endX   = availMask&RightAvail ? width : (width -1);

      //prepare 2nd line upper sign

      const Pel* srcLineBelow= srcLine+ srcStride;

      for (x=startX-1; x< endX; x++)

      {

        signUpLine[x] = (int8_t)sgn(srcLineBelow[x] - srcLine[x+1]);

      }

      //first line

      const Pel* srcLineAbove= srcLine- srcStride;

      firstLineStartX = availMask&AboveAvail ? startX : (width -1 );

      firstLineEndX   = availMask&AboveRightAvail ? width : (width-1);

      for(x= firstLineStartX; x< firstLineEndX; x++)

      {

        edgeType = sgn(srcLine[x] - srcLineAbove[x+1]) -signUpLine[x-1];

        resLine[x] = ClipPel<int>(srcLine[x] + offset[edgeType], clpRng);

      }

      srcLine += srcStride;

      resLine += resStride;

      //middle lines

      for (y= 1; y< height-1; y++)

      {

        srcLineBelow= srcLine+ srcStride;

        for(x= startX; x< endX; x++)

        {

          signDown =  (int8_t)sgn(srcLine[x] - srcLineBelow[x-1]);

          edgeType =  signDown + signUpLine[x];

          resLine[x] = ClipPel<int>(srcLine[x] + offset[edgeType], clpRng);

          signUpLine[x-1] = -signDown;

        }

        signUpLine[endX-1] = (int8_t)sgn(srcLineBelow[endX-1] - srcLine[endX]);

        srcLine  += srcStride;

        resLine += resStride;

      }

      //last line

      srcLineBelow= srcLine+ srcStride;

      lastLineStartX = availMask&BelowLeftAvail ? 0 : 1;

      lastLineEndX   = availMask&BelowAvail ? endX : 1;

      for(x= lastLineStartX; x< lastLineEndX; x++)

      {

        edgeType = sgn(srcLine[x] - srcLineBelow[x-1]) + signUpLine[x];

        resLine[x] = ClipPel<int>(srcLine[x] + offset[edgeType], clpRng);

      }

    }

  }

  break;

  case SAO_TYPE_BO:

  {

    const int shiftBits = channelBitDepth - NUM_SAO_BO_CLASSES_LOG2;

    int8_t p_eo_offsets[16] = {0,};

    for (int i = 0; i < 4; i++)

    {

      p_eo_offsets[i] = offset[( startIdx + i ) % MAX_NUM_SAO_CLASSES];

    }

#ifdef USE_AVX2

    // AVX2

    if( ( width & 15 ) == 0 && vext >= AVX2 )

    {

      __m256i vsrc;

      __m256i vbaseoffset = _mm256_set1_epi16(startIdx - MAX_NUM_SAO_CLASSES) ;

      __m256i vminus = _mm256_set1_epi8(-1);

      __m256i vzero = _mm256_set1_epi8(0);

      __m256i vfour = _mm256_set1_epi16(4);

      __m256i vibdimax = _mm256_set1_epi16((1<<channelBitDepth) -1 );

      __m256i voffsettbl =  _mm256_broadcastsi128_si256(_mm_loadu_si128((__m128i*)p_eo_offsets));

      for (y=0; y< height; y++)

      {

        for (x=0; x< width; x+=16)

        {

          vsrc = _mm256_loadu_si256((__m256i*)&srcLine[x]);

          __m256i bands = _mm256_srai_epi16(vsrc, shiftBits);

          bands = _mm256_sub_epi16(bands, vbaseoffset);

          bands = _mm256_and_si256(bands, _mm256_set1_epi16( MAX_NUM_SAO_CLASSES - 1 )); // modulo 32 = modulo NUM_SAO_BO_CLASSES_LOG2

          __m256i mask1 = _mm256_cmpgt_epi16(bands,vminus);

          __m256i mask2 = _mm256_cmpgt_epi16(vfour,bands);

          __m256i veoffsets = _mm256_shuffle_epi8(voffsettbl, bands);

          veoffsets = _mm256_slli_epi16 (veoffsets,8);

          veoffsets = _mm256_srai_epi16 (veoffsets,8);

          veoffsets = _mm256_and_si256(veoffsets,mask1);

          veoffsets = _mm256_and_si256(veoffsets,mask2);

          vsrc = _mm256_add_epi16(vsrc, veoffsets);

          vsrc    = _mm256_min_epi16(_mm256_max_epi16(vsrc, vzero), vibdimax);

          _mm256_storeu_si256((__m256i*)&resLine[x], vsrc);

        }

        srcLine  += srcStride;

        resLine += resStride;

      }

    }

    else

#endif

    {

      __m128i vsrc;

      __m128i vbaseoffset = _mm_set1_epi16(startIdx - MAX_NUM_SAO_CLASSES) ;

      __m128i vminus = _mm_set1_epi8(-1);

      __m128i vzero = _mm_set1_epi8(0);

      __m128i vfour = _mm_set1_epi16(4);

      __m128i vibdimax = _mm_set1_epi16((1<<channelBitDepth) -1 );

      __m128i voffsettbl = _mm_loadu_si128((__m128i*)p_eo_offsets);

      for (y=0; y< height; y++)

      {

        for (x=0; x< width; x+=8)

        {

          vsrc = _mm_loadu_si128((__m128i*)&srcLine[x]);

          __m128i bands = _mm_srai_epi16(vsrc, shiftBits);

          bands = _mm_sub_epi16(bands, vbaseoffset);

          bands = _mm_and_si128(bands, _mm_set1_epi16( MAX_NUM_SAO_CLASSES - 1 )); // modulo 32 = modulo NUM_SAO_BO_CLASSES_LOG2

          __m128i mask1 = _mm_cmpgt_epi16(bands,vminus);

          __m128i mask2 = _mm_cmplt_epi16(bands,vfour);

          __m128i veoffsets = _mm_shuffle_epi8(voffsettbl, bands);

          veoffsets = _mm_slli_epi16 (veoffsets,8);

          veoffsets = _mm_srai_epi16 (veoffsets,8);

          veoffsets = _mm_and_si128(veoffsets,mask1);

          veoffsets = _mm_and_si128(veoffsets,mask2);

          vsrc = _mm_add_epi16(vsrc, veoffsets);

          vsrc    = _mm_min_epi16(_mm_max_epi16(vsrc, vzero), vibdimax);

          _mm_store_si128((__m128i*)&resLine[x], vsrc);

        }

        srcLine  += srcStride;

        resLine += resStride;

      }

    }

  }

  break;

  default:

  {

    THROW("Not a supported SAO types\n");

  }

  }

#if USE_AVX2

  _mm256_zeroupper();

#endif

}

Unexecuted instantiation: void vvenc::offsetBlock_SIMD<(vvenc::x86_simd::X86_VEXT)1>(int, vvenc::ClpRng const&, int, int*, int, short const*, short*, long, long, int, int, unsigned char, std::__1::vector<signed char, std::__1::allocator<signed char> >&, std::__1::vector<signed char, std::__1::allocator<signed char> >&)

Unexecuted instantiation: void vvenc::offsetBlock_SIMD<(vvenc::x86_simd::X86_VEXT)4>(int, vvenc::ClpRng const&, int, int*, int, short const*, short*, long, long, int, int, unsigned char, std::__1::vector<signed char, std::__1::allocator<signed char> >&, std::__1::vector<signed char, std::__1::allocator<signed char> >&)

template <X86_VEXT vext>

void calcSaoStatisticsBo_SIMD(int width,int endX,int endY,Pel*  srcLine,Pel*  orgLine,int srcStride,int orgStride,int channelBitDepth, int64_t *count,int64_t  *diff)

{

  if ( width % 16 == 0 )

  {

    int iNaRight=width-endX;

    int x;

    int i_bo_range_shift = channelBitDepth - NUM_SAO_BO_CLASSES_LOG2;

    __m128i vzero = _mm_setzero_si128();

    for (int y=0; y<endY; y++)

    {

      for (x=0; x<endX-16; x+=16)

      {

        __m128i vsrca, vsrcb;

        __m128i vdiffa,vdiffb;

        if (sizeof(Pel) == 1){

          __m128i vsrc = _mm_loadu_si128((__m128i*)&srcLine[x]);

          vsrca = _mm_unpacklo_epi8(vsrc, vzero);

          vsrcb = _mm_unpackhi_epi8(vsrc, vzero);

          __m128i vorg  = _mm_loadu_si128((__m128i*)&orgLine[x]);

          __m128i vorga = _mm_unpacklo_epi8(vorg, vzero);

          __m128i vorgb = _mm_unpackhi_epi8(vorg, vzero);

          vdiffa = _mm_sub_epi16(vorga, vsrca);

          vdiffb = _mm_sub_epi16(vorgb, vsrcb);

        }

        else

        {

          vsrca = _mm_loadu_si128((__m128i*)&srcLine[x]);

          vsrcb = _mm_loadu_si128((__m128i*)&srcLine[x+8]);

          __m128i vorga = _mm_loadu_si128((__m128i*)&orgLine[x]);

          __m128i vorgb = _mm_loadu_si128((__m128i*)&orgLine[x+8]);

          vdiffa = _mm_sub_epi16(vorga, vsrca);

          vdiffb = _mm_sub_epi16(vorgb, vsrcb);

        }

        __m128i vbanda = _mm_srai_epi16(vsrca, i_bo_range_shift);

        __m128i vbandb = _mm_srai_epi16(vsrcb, i_bo_range_shift);

        int iBand;

        // since gcc 4.6 synopsis of _mm_extract_epi16 has changed to (int)(unsigned short)_mm_extract_epi16()

        // therefore cast result to short to have signed values

        short iDiff;

        iBand = _mm_extract_epi16(vbanda, 0);

        iDiff = (short)_mm_extract_epi16(vdiffa, 0);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

        iBand = _mm_extract_epi16(vbanda, 1);

        iDiff = (short)_mm_extract_epi16(vdiffa, 1);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

        iBand = _mm_extract_epi16(vbanda, 2);

        iDiff = (short)_mm_extract_epi16(vdiffa, 2);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

        iBand = _mm_extract_epi16(vbanda, 3);

        iDiff = (short)_mm_extract_epi16(vdiffa, 3);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

        iBand = _mm_extract_epi16(vbanda, 4);

        iDiff = (short)_mm_extract_epi16(vdiffa, 4);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

        iBand = _mm_extract_epi16(vbanda, 5);

        iDiff = (short)_mm_extract_epi16(vdiffa, 5);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

        iBand = _mm_extract_epi16(vbanda, 6);

        iDiff = (short)_mm_extract_epi16(vdiffa, 6);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

        iBand = _mm_extract_epi16(vbanda, 7);

        iDiff = (short)_mm_extract_epi16(vdiffa, 7);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

        iBand = _mm_extract_epi16(vbandb, 0);

        iDiff = (short)_mm_extract_epi16(vdiffb, 0);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

        iBand = _mm_extract_epi16(vbandb, 1);

        iDiff = (short)_mm_extract_epi16(vdiffb, 1);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

        iBand = _mm_extract_epi16(vbandb, 2);

        iDiff = (short)_mm_extract_epi16(vdiffb, 2);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

        iBand = _mm_extract_epi16(vbandb, 3);

        iDiff = (short)_mm_extract_epi16(vdiffb, 3);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

        iBand = _mm_extract_epi16(vbandb, 4);

        iDiff = (short)_mm_extract_epi16(vdiffb, 4);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

        iBand = _mm_extract_epi16(vbandb, 5);

        iDiff = (short)_mm_extract_epi16(vdiffb, 5);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

        iBand = _mm_extract_epi16(vbandb, 6);

        iDiff = (short)_mm_extract_epi16(vdiffb, 6);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

        iBand = _mm_extract_epi16(vbandb, 7);

        iDiff = (short)_mm_extract_epi16(vdiffb, 7);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

      }

      //last colum

      {

        __m128i vsrca, vsrcb;

        __m128i vdiffa,vdiffb;

        if (sizeof(Pel) == 1){

          __m128i vsrc = _mm_loadu_si128((__m128i*)&srcLine[x]);

          vsrca = _mm_unpacklo_epi8(vsrc, vzero);

          vsrcb = _mm_unpackhi_epi8(vsrc, vzero);

          __m128i vorg  = _mm_loadu_si128((__m128i*)&orgLine[x]);

          __m128i vorga = _mm_unpacklo_epi8(vorg, vzero);

          __m128i vorgb = _mm_unpackhi_epi8(vorg, vzero);

          vdiffa = _mm_sub_epi16(vorga, vsrca);

          vdiffb = _mm_sub_epi16(vorgb, vsrcb);

        }

        else

        {

          vsrca = _mm_loadu_si128((__m128i*)&srcLine[x]);

          vsrcb = _mm_loadu_si128((__m128i*)&srcLine[x+8]);

          __m128i vorga = _mm_loadu_si128((__m128i*)&orgLine[x]);

          __m128i vorgb = _mm_loadu_si128((__m128i*)&orgLine[x+8]);

          vdiffa = _mm_sub_epi16(vorga, vsrca);

          vdiffb = _mm_sub_epi16(vorgb, vsrcb);

        }

        __m128i vbanda = _mm_srai_epi16(vsrca, i_bo_range_shift);

        __m128i vbandb = _mm_srai_epi16(vsrcb, i_bo_range_shift);

        int iBand;

        // since gcc 4.6 synopsis of _mm_extract_epi16 has changed to (int)(unsigned short)_mm_extract_epi16()

        // therefore cast result to short to have signed values

        short iDiff;

        iBand = _mm_extract_epi16(vbanda, 0);

        iDiff = (short)_mm_extract_epi16(vdiffa, 0);

        diff[iBand]  += iDiff;

        count[iBand] += 1;

        if (iNaRight<15)

        {

          iBand = _mm_extract_epi16(vbanda, 1);

          iDiff = (short)_mm_extract_epi16(vdiffa, 1);

          diff[iBand]  += iDiff;

          count[iBand] += 1;