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

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) 2018-2026, Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V. & The VVdeC 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

*/

#include "CommonDefX86.h"

#include "../SampleAdaptiveOffset.h"

#if defined(TARGET_SIMD_X86)  && ENABLE_SIMD_OPT_SAO

namespace vvdec

{

#  if USE_AVX2 && !defined( _mm256_set_m128i )

#    define VVCLIB_OWN_mm256_set_m128i

#    define _mm256_set_m128i( v0, v1 ) _mm256_inserti128_si256( _mm256_castsi128_si256( v1 ), ( v0 ), 1 )

#  endif

static bool isProcessDisabled( int xPos, int yPos, int numVerVirBndry, int numHorVirBndry, int verVirBndryPos[], int horVirBndryPos[] )

{

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

  {

    if( ( xPos == verVirBndryPos[i] ) || ( xPos == verVirBndryPos[i] - 1 ) )

    {

      return true;

    }

  }

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

  {

    if( ( yPos == horVirBndryPos[i] ) || ( yPos == horVirBndryPos[i] - 1 ) )

    {

      return true;

    }

  }

  return false;

}

Unexecuted instantiation: SampleAdaptiveOffset_sse41.cpp:vvdec::isProcessDisabled(int, int, int, int, int*, int*)

Unexecuted instantiation: SampleAdaptiveOffset_avx2.cpp:vvdec::isProcessDisabled(int, int, int, int, int*, int*)

static bool isHorProcessDisabled( int yPos, int numHorVirBndry, int horVirBndryPos[] )

{

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

  {

    if( ( yPos == horVirBndryPos[i] ) || ( yPos == horVirBndryPos[i] - 1 ) )

    {

      return true;

    }

  }

  return false;

}

Unexecuted instantiation: SampleAdaptiveOffset_sse41.cpp:vvdec::isHorProcessDisabled(int, int, int*)

Unexecuted instantiation: SampleAdaptiveOffset_avx2.cpp:vvdec::isHorProcessDisabled(int, int, int*)

static bool isVerProcessDisabled( int xPos, int numVerVirBndry, int verVirBndryPos[] )

{

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

  {

    if( ( xPos == verVirBndryPos[i] ) || ( xPos == verVirBndryPos[i] - 1 ) )

    {

      return true;

    }

  }

  return false;

}

Unexecuted instantiation: SampleAdaptiveOffset_sse41.cpp:vvdec::isVerProcessDisabled(int, int, int*)

Unexecuted instantiation: SampleAdaptiveOffset_avx2.cpp:vvdec::isVerProcessDisabled(int, int, int*)

template<X86_VEXT vext>

static void offsetBlock_SIMD_SAO_TYPE_BO( const int  channelBitDepth,

                                          int*       offset,

                                          int        startIdx,

                                          const Pel* srcBlk,

                                          Pel*       resBlk,

                                          ptrdiff_t  srcStride,

                                          ptrdiff_t  resStride,

                                          int        width,

                                          int        height )

{

  const Pel* srcLine = srcBlk;

  Pel*       resLine = resBlk;

  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];

  }

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

  {

    for( int x = 0; x < width; )

    {

#  ifdef USE_AVX2

      // AVX2

      if( width - x >= 16 && vext >= AVX2 )

      {

        __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 ) );

        __m256i 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 );

        x += 16;

      }

      else

#  endif

      {

        __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 );

        __m128i 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 );

        x += 8;

      }

    }

    srcLine += srcStride;

    resLine += resStride;

  }

}

Unexecuted instantiation: SampleAdaptiveOffset_sse41.cpp:void vvdec::offsetBlock_SIMD_SAO_TYPE_BO<(vvdec::x86_simd::X86_VEXT)1>(int, int*, int, short const*, short*, long, long, int, int)

SampleAdaptiveOffset_avx2.cpp:void vvdec::offsetBlock_SIMD_SAO_TYPE_BO<(vvdec::x86_simd::X86_VEXT)4>(int, int*, int, short const*, short*, long, long, int, int)

Line

Count

Source

112

5

{

113

5

  const Pel* srcLine = srcBlk;

114

5

  Pel*       resLine = resBlk;

115

116

5

  const int shiftBits        = channelBitDepth - NUM_SAO_BO_CLASSES_LOG2;

117

5

  int8_t    p_eo_offsets[16] = { 0 };

118

25

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

119

20

  {

120

20

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

121

20

  }

122

580

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

123

575

  {

124

4.54k

    for( int x = 0; x < width; )

125

3.96k

    {

126

3.96k

#  ifdef USE_AVX2

127

      // AVX2

128

3.96k

      if( width - x >= 16 && vext >= AVX2 )

129

3.96k

      {

130

3.96k

        __m256i vbaseoffset = _mm256_set1_epi16( startIdx - MAX_NUM_SAO_CLASSES );

131

3.96k

        __m256i vminus      = _mm256_set1_epi8( -1 );

132

3.96k

        __m256i vzero       = _mm256_set1_epi8( 0 );

133

134

3.96k

        __m256i vfour      = _mm256_set1_epi16( 4 );

135

3.96k

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

136

3.96k

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

137

138

3.96k

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

139

3.96k

        __m256i bands = _mm256_srai_epi16( vsrc, shiftBits );

140

3.96k

        bands         = _mm256_sub_epi16 ( bands, vbaseoffset );

141

3.96k

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

142

3.96k

        __m256i mask1 = _mm256_cmpgt_epi16( bands, vminus );

143

3.96k

        __m256i mask2 = _mm256_cmpgt_epi16( vfour, bands );

144

145

3.96k

        __m256i veoffsets = _mm256_shuffle_epi8( voffsettbl, bands );

146

3.96k

        veoffsets         = _mm256_slli_epi16( veoffsets, 8 );

147

3.96k

        veoffsets         = _mm256_srai_epi16( veoffsets, 8 );

148

149

3.96k

        veoffsets = _mm256_and_si256( veoffsets, mask1 );

150

3.96k

        veoffsets = _mm256_and_si256( veoffsets, mask2 );

151

152

3.96k

        vsrc = _mm256_add_epi16( vsrc, veoffsets );

153

3.96k

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

154

3.96k

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

155

156

3.96k

        x += 16;

157

3.96k

      }

158

4

      else

159

4

#  endif

160

4

      {

161

4

        __m128i vbaseoffset = _mm_set1_epi16( startIdx - MAX_NUM_SAO_CLASSES );

162

4

        __m128i vminus      = _mm_set1_epi8( -1 );

163

4

        __m128i vzero       = _mm_set1_epi8( 0 );

164

165

4

        __m128i vfour      = _mm_set1_epi16( 4 );

166

4

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

167

4

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

168

169

4

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

170

4

        __m128i bands = _mm_srai_epi16( vsrc, shiftBits );

171

4

        bands         = _mm_sub_epi16 ( bands, vbaseoffset );

172

4

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

173

4

        __m128i mask1 = _mm_cmpgt_epi16( bands, vminus );

174

4

        __m128i mask2 = _mm_cmplt_epi16( bands, vfour );

175

176

4

        __m128i veoffsets = _mm_shuffle_epi8( voffsettbl, bands );

177

4

        veoffsets         = _mm_slli_epi16( veoffsets, 8 );

178

4

        veoffsets         = _mm_srai_epi16( veoffsets, 8 );

179

180

4

        veoffsets = _mm_and_si128( veoffsets, mask1 );

181

4

        veoffsets = _mm_and_si128( veoffsets, mask2 );

182

183

4

        vsrc = _mm_add_epi16( vsrc, veoffsets );

184

4

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

185

4

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

186

187

4

        x += 8;

188

4

      }

189

3.96k

    }

190

575

    srcLine += srcStride;

191

575

    resLine += resStride;

192

575

  }

193

5

}

template<X86_VEXT vext>

static void offsetBlock_SIMD_SAO_TYPE_EO_0( const int            channelBitDepth,

                                            const ClpRng&        clpRng,

                                            int*                 offset,

                                            const Pel*           srcBlk,

                                            Pel*                 resBlk,

                                            ptrdiff_t            srcStride,

                                            ptrdiff_t            resStride,

                                            int                  width,

                                            int                  height,

                                            bool                 isLeftAvail,

                                            bool                 isRightAvail,

                                            bool                 isAboveAvail,

                                            bool                 isBelowAvail,

                                            bool                 isAboveLeftAvail,

                                            bool                 isAboveRightAvail,

                                            bool                 isBelowLeftAvail,

                                            bool                 isBelowRightAvail,

                                            std::vector<int8_t>* m_signLineBuf1,

                                            std::vector<int8_t>* m_signLineBuf2,

                                            bool                 isCtuCrossedByVirtualBoundaries,

                                            int                  horVirBndryPos[],

                                            int                  verVirBndryPos[],

                                            int                  numHorVirBndry,

                                            int                  numVerVirBndry,

                                            uint16_t             bndmask[MAX_CU_SIZE] )

{

  const Pel* srcLine = srcBlk;

  Pel*       resLine = resBlk;

  int    x, y, startX, endX, edgeType;

  int8_t signLeft, signRight;

  if( isLeftAvail && isRightAvail )

  {

    int8_t p_eo_offsets[16] = { 0 };

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

    {

      p_eo_offsets[i] = offset[i];

    }

#if defined( USE_AVX2 )

    // AVX2

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

    {

      __m256i vsrca, vsrcal, vsrcar, virBmask;

      __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 ) );

      if( isCtuCrossedByVirtualBoundaries )

      {

        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] );

            virBmask = _mm256_loadu_si256( (__m256i*)&bndmask[x] );

            vsrcal            = _mm256_subs_epi16( vsrca, vsrcal );

            vsrcar            = _mm256_subs_epi16( vsrca, vsrcar );

            __m256i vsignl    = _mm256_sign_epi16( vplusone, vsrcal );

            __m256i vsignr    = _mm256_sign_epi16( vplusone, vsrcar );

            __m256i vsign     = _mm256_adds_epi16( _mm256_adds_epi16( vsignl, vsignr ), vbaseoffset );

            __m256i veoffsets = _mm256_shuffle_epi8( voffsettbl, vsign );

            veoffsets         = _mm256_slli_epi16( veoffsets, 8 );

            veoffsets         = _mm256_srai_epi16( veoffsets, 8 );

            vsrcal = _mm256_add_epi16( vsrca, veoffsets );

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

            vsrcar = _mm256_blendv_epi8( vsrcal, vsrca, virBmask );

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

          }

          srcLine += srcStride;

          resLine += resStride;

        }

      }

      else

      {

        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_subs_epi16( vsrca, vsrcal );

            vsrcar            = _mm256_subs_epi16( vsrca, vsrcar );

            __m256i vsignl    = _mm256_sign_epi16( vplusone, vsrcal );

            __m256i vsignr    = _mm256_sign_epi16( vplusone, vsrcar );

            __m256i vsign     = _mm256_adds_epi16( _mm256_adds_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, virBmask;

      __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 );

      if( isCtuCrossedByVirtualBoundaries )

      {

        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] );

            virBmask          = _mm_loadu_si128( (__m128i*)&bndmask[x] );

            vsrcal            = _mm_subs_epi16( vsrca, vsrcal );

            vsrcar            = _mm_subs_epi16( vsrca, vsrcar );

            __m128i vsignl    = _mm_sign_epi16( vplusone, vsrcal );

            __m128i vsignr    = _mm_sign_epi16( vplusone, vsrcar );

            __m128i vsign     = _mm_adds_epi16( _mm_adds_epi16( vsignl, vsignr ), vbaseoffset );

            __m128i veoffsets = _mm_shuffle_epi8( voffsettbl, vsign );

            veoffsets         = _mm_slli_epi16( veoffsets, 8 );

            veoffsets         = _mm_srai_epi16( veoffsets, 8 );

            vsrcal = _mm_add_epi16( vsrca, veoffsets );

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

            vsrcar = _mm_blendv_epi8( vsrcal, vsrca, virBmask );

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

          }

          srcLine += srcStride;

          resLine += resStride;

        }

      }

      else

      {

        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_subs_epi16( vsrca, vsrcal );

            vsrcar            = _mm_subs_epi16( vsrca, vsrcar );

            __m128i vsignl    = _mm_sign_epi16( vplusone, vsrcal );

            __m128i vsignr    = _mm_sign_epi16( vplusone, vsrcar );

            __m128i vsign     = _mm_adds_epi16( _mm_adds_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 = isLeftAvail ? 0 : 1;

    endX   = isRightAvail ? 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] );

        if( isCtuCrossedByVirtualBoundaries && isVerProcessDisabled( x, numVerVirBndry, verVirBndryPos ) )

        {

          signLeft = -signRight;

          continue;

        }

        edgeType = signRight + signLeft;

        signLeft = -signRight;

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

      }

      srcLine += srcStride;

      resLine += resStride;

    }

  }

}

Unexecuted instantiation: SampleAdaptiveOffset_sse41.cpp:void vvdec::offsetBlock_SIMD_SAO_TYPE_EO_0<(vvdec::x86_simd::X86_VEXT)1>(int, vvdec::ClpRngTemplate<short> const&, int*, short const*, short*, long, long, int, int, bool, bool, bool, bool, bool, bool, bool, bool, std::__1::vector<signed char, std::__1::allocator<signed char> >*, std::__1::vector<signed char, std::__1::allocator<signed char> >*, bool, int*, int*, int, int, unsigned short*)

SampleAdaptiveOffset_avx2.cpp:void vvdec::offsetBlock_SIMD_SAO_TYPE_EO_0<(vvdec::x86_simd::X86_VEXT)4>(int, vvdec::ClpRngTemplate<short> const&, int*, short const*, short*, long, long, int, int, bool, bool, bool, bool, bool, bool, bool, bool, std::__1::vector<signed char, std::__1::allocator<signed char> >*, std::__1::vector<signed char, std::__1::allocator<signed char> >*, bool, int*, int*, int, int, unsigned short*)

Line

Count

Source

221

7

{

222

7

  const Pel* srcLine = srcBlk;

223

7

  Pel*       resLine = resBlk;

224

225

7

  int    x, y, startX, endX, edgeType;

226

7

  int8_t signLeft, signRight;

227

228

7

  if( isLeftAvail && isRightAvail )

229

4

  {

230

4

    int8_t p_eo_offsets[16] = { 0 };

231

24

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

232

20

    {

233

20

      p_eo_offsets[i] = offset[i];

234

20

    }

235

4

#if defined( USE_AVX2 )

236

    // AVX2

237

4

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

238

4

    {

239

240

4

      __m256i vsrca, vsrcal, vsrcar, virBmask;

241

242

4

      __m256i vbaseoffset = _mm256_set1_epi16( 2 );

243

4

      __m256i vplusone    = _mm256_set1_epi16( 1 );

244

4

      __m256i vzero       = _mm256_set1_epi8( 0 );

245

4

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

246

4

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

247

248

4

      if( isCtuCrossedByVirtualBoundaries )

249

0

      {

250

0

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

251

0

        {

252

0

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

253

0

          {

254

0

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

255

0

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

256

0

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

257

0

            virBmask = _mm256_loadu_si256( (__m256i*)&bndmask[x] );

258

259

0

            vsrcal            = _mm256_subs_epi16( vsrca, vsrcal );

260

0

            vsrcar            = _mm256_subs_epi16( vsrca, vsrcar );

261

0

            __m256i vsignl    = _mm256_sign_epi16( vplusone, vsrcal );

262

0

            __m256i vsignr    = _mm256_sign_epi16( vplusone, vsrcar );

263

0

            __m256i vsign     = _mm256_adds_epi16( _mm256_adds_epi16( vsignl, vsignr ), vbaseoffset );

264

0

            __m256i veoffsets = _mm256_shuffle_epi8( voffsettbl, vsign );

265

0

            veoffsets         = _mm256_slli_epi16( veoffsets, 8 );

266

0

            veoffsets         = _mm256_srai_epi16( veoffsets, 8 );

267

268

0

            vsrcal = _mm256_add_epi16( vsrca, veoffsets );

269

0

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

270

271

0

            vsrcar = _mm256_blendv_epi8( vsrcal, vsrca, virBmask );

272

273

0

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

274

0

          }

275

0

          srcLine += srcStride;

276

0

          resLine += resStride;

277

0

        }

278

0

      }

279

4

      else

280

4

      {

281

481

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

282

477

        {

283

4.27k

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

284

3.79k

          {

285

3.79k

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

286

3.79k

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

287

3.79k

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

288

3.79k

            vsrcal            = _mm256_subs_epi16( vsrca, vsrcal );

289

3.79k

            vsrcar            = _mm256_subs_epi16( vsrca, vsrcar );

290

3.79k

            __m256i vsignl    = _mm256_sign_epi16( vplusone, vsrcal );

291

3.79k

            __m256i vsignr    = _mm256_sign_epi16( vplusone, vsrcar );

292

3.79k

            __m256i vsign     = _mm256_adds_epi16( _mm256_adds_epi16( vsignl, vsignr ), vbaseoffset );

293

3.79k

            __m256i veoffsets = _mm256_shuffle_epi8( voffsettbl, vsign );

294

3.79k

            veoffsets         = _mm256_slli_epi16( veoffsets, 8 );

295

3.79k

            veoffsets         = _mm256_srai_epi16( veoffsets, 8 );

296

297

3.79k

            vsrca = _mm256_add_epi16( vsrca, veoffsets );

298

3.79k

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

299

300

3.79k

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

301

3.79k

          }

302

477

          srcLine += srcStride;

303

477

          resLine += resStride;

304

477

        }

305

4

      }

306

4

    }

307

0

    else

308

0

#  endif

309

0

    {

310

0

      __m128i vsrca, vsrcal, vsrcar, virBmask;

311

0

      __m128i vbaseoffset = _mm_set1_epi16( 2 );

312

0

      __m128i vplusone    = _mm_set1_epi16( 1 );

313

0

      __m128i vzero       = _mm_set1_epi8( 0 );

314

0

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

315

0

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

316

0

      if( isCtuCrossedByVirtualBoundaries )

317

0

      {

318

0

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

319

0

        {

320

0

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

321

0

          {

322

0

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

323

0

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

324

0

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

325

0

            virBmask          = _mm_loadu_si128( (__m128i*)&bndmask[x] );

326

0

            vsrcal            = _mm_subs_epi16( vsrca, vsrcal );

327

0

            vsrcar            = _mm_subs_epi16( vsrca, vsrcar );

328

0

            __m128i vsignl    = _mm_sign_epi16( vplusone, vsrcal );

329

0

            __m128i vsignr    = _mm_sign_epi16( vplusone, vsrcar );

330

0

            __m128i vsign     = _mm_adds_epi16( _mm_adds_epi16( vsignl, vsignr ), vbaseoffset );

331

0

            __m128i veoffsets = _mm_shuffle_epi8( voffsettbl, vsign );

332

0

            veoffsets         = _mm_slli_epi16( veoffsets, 8 );

333

0

            veoffsets         = _mm_srai_epi16( veoffsets, 8 );

334

335

0

            vsrcal = _mm_add_epi16( vsrca, veoffsets );

336

0

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

337

338

0

            vsrcar = _mm_blendv_epi8( vsrcal, vsrca, virBmask );

339

340

0

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

341

0

          }

342

0

          srcLine += srcStride;

343

0

          resLine += resStride;

344

0

        }

345

0

      }

346

0

      else

347

0

      {

348

0

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

349

0

        {

350

0

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

351

0

          {

352

0

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

353

0

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

354

0

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

355

0

            vsrcal            = _mm_subs_epi16( vsrca, vsrcal );

356

0

            vsrcar            = _mm_subs_epi16( vsrca, vsrcar );

357

0

            __m128i vsignl    = _mm_sign_epi16( vplusone, vsrcal );

358

0

            __m128i vsignr    = _mm_sign_epi16( vplusone, vsrcar );

359

0

            __m128i vsign     = _mm_adds_epi16( _mm_adds_epi16( vsignl, vsignr ), vbaseoffset );

360

0

            __m128i veoffsets = _mm_shuffle_epi8( voffsettbl, vsign );

361

0

            veoffsets         = _mm_slli_epi16( veoffsets, 8 );

362

0

            veoffsets         = _mm_srai_epi16( veoffsets, 8 );

363

364

0

            vsrca = _mm_add_epi16( vsrca, veoffsets );

365

0

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

366

0

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

367

0

          }

368

0

          srcLine += srcStride;

369

0

          resLine += resStride;

370

0

        }

371

0

      }

372

0

    }

373

4

  }

374

3

  else

375

3

  {

376

3

    offset += 2;

377

3

    startX = isLeftAvail ? 0 : 1;

378

3

    endX   = isRightAvail ? width : ( width - 1 );

379

387

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

380

384

    {

381

384

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

382

45.0k

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

383

44.6k

      {

384

44.6k

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

385

44.6k

        if( isCtuCrossedByVirtualBoundaries && isVerProcessDisabled( x, numVerVirBndry, verVirBndryPos ) )

386

0

        {

387

0

          signLeft = -signRight;

388

0

          continue;

389

0

        }

390

44.6k

        edgeType = signRight + signLeft;

391

44.6k

        signLeft = -signRight;

392

393

44.6k

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

394

44.6k

      }

395

396

384

      srcLine += srcStride;

397

384

      resLine += resStride;

398

384

    }

399

3

  }

400

7

}

template<X86_VEXT vext>

static void offsetBlock_SIMD_SAO_TYPE_EO_90( const int            channelBitDepth,

                                             const ClpRng&        clpRng,

                                             int*                 offset,

                                             const Pel*           srcBlk,

                                             Pel*                 resBlk,

                                             ptrdiff_t            srcStride,

                                             ptrdiff_t            resStride,

                                             int                  width,

                                             int                  height,

                                             bool                 isLeftAvail,

                                             bool                 isRightAvail,

                                             bool                 isAboveAvail,

                                             bool                 isBelowAvail,

                                             bool                 isAboveLeftAvail,

                                             bool                 isAboveRightAvail,

                                             bool                 isBelowLeftAvail,

                                             bool                 isBelowRightAvail,

                                             std::vector<int8_t>* m_signLineBuf1,

                                             std::vector<int8_t>* m_signLineBuf2,

                                             bool                 isCtuCrossedByVirtualBoundaries,

                                             int                  horVirBndryPos[],

                                             int                  verVirBndryPos[],

                                             int                  numHorVirBndry,

                                             int                  numVerVirBndry,

                                             uint16_t             bndmask[MAX_CU_SIZE] )

{

  const Pel* srcLine = srcBlk;

  Pel*       resLine = resBlk;

  int x, y, startY, endY;

  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( !isAboveAvail )

  {

    startY       = 1;

    srcLineAbove = srcLine;

    srcLine += srcStride;

    resLine += resStride;

    srcLineBelow = srcLine + srcStride;

  }

  endY = height;

  if( !isBelowAvail )

  {

    endY = height - 1;

  }

#  if defined( 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;

    if( isCtuCrossedByVirtualBoundaries )

    {

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

      {

        if( !isHorProcessDisabled( y, numHorVirBndry, horVirBndryPos ) )

        {

          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_subs_epi16( vsrca, vsrcat );

            vsrcab            = _mm256_subs_epi16( vsrca, vsrcab );

            __m256i vsignt    = _mm256_sign_epi16( vplusone, vsrcat );

            __m256i vsignb    = _mm256_sign_epi16( vplusone, vsrcab );

            __m256i vsign     = _mm256_adds_epi16( _mm256_adds_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

    {

      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_subs_epi16( vsrca, vsrcat );

          vsrcab            = _mm256_subs_epi16( vsrca, vsrcab );

          __m256i vsignt    = _mm256_sign_epi16( vplusone, vsrcat );

          __m256i vsignb    = _mm256_sign_epi16( vplusone, vsrcab );

          __m256i vsign     = _mm256_adds_epi16( _mm256_adds_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 );

    if( isCtuCrossedByVirtualBoundaries )

    {

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

      {

        if( !isHorProcessDisabled( y, numHorVirBndry, horVirBndryPos ) )

        {

          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_subs_epi16( vsrca, vsrcat );

            vsrcab            = _mm_subs_epi16( vsrca, vsrcab );

            __m128i vsignt    = _mm_sign_epi16( vplusone, vsrcat );

            __m128i vsignb    = _mm_sign_epi16( vplusone, vsrcab );

            __m128i vsign     = _mm_adds_epi16( _mm_adds_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

    {

      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_subs_epi16( vsrca, vsrcat );

          vsrcab            = _mm_subs_epi16( vsrca, vsrcab );

          __m128i vsignt    = _mm_sign_epi16( vplusone, vsrcat );

          __m128i vsignb    = _mm_sign_epi16( vplusone, vsrcab );

          __m128i vsign     = _mm_adds_epi16( _mm_adds_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;

      }

    }

  }

}

Unexecuted instantiation: SampleAdaptiveOffset_sse41.cpp:void vvdec::offsetBlock_SIMD_SAO_TYPE_EO_90<(vvdec::x86_simd::X86_VEXT)1>(int, vvdec::ClpRngTemplate<short> const&, int*, short const*, short*, long, long, int, int, bool, bool, bool, bool, bool, bool, bool, bool, std::__1::vector<signed char, std::__1::allocator<signed char> >*, std::__1::vector<signed char, std::__1::allocator<signed char> >*, bool, int*, int*, int, int, unsigned short*)

SampleAdaptiveOffset_avx2.cpp:void vvdec::offsetBlock_SIMD_SAO_TYPE_EO_90<(vvdec::x86_simd::X86_VEXT)4>(int, vvdec::ClpRngTemplate<short> const&, int*, short const*, short*, long, long, int, int, bool, bool, bool, bool, bool, bool, bool, bool, std::__1::vector<signed char, std::__1::allocator<signed char> >*, std::__1::vector<signed char, std::__1::allocator<signed char> >*, bool, int*, int*, int, int, unsigned short*)

Line

Count

Source

428

4

{

429

4

  const Pel* srcLine = srcBlk;

430

4

  Pel*       resLine = resBlk;

431

432

4

  int x, y, startY, endY;

433

434

4

  int8_t p_eo_offsets[16] = { 0 };

435

24

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

436

20

  {

437

20

    p_eo_offsets[i] = offset[i];

438

20

  }

439

4

  const Pel* srcLineAbove = srcLine - srcStride;

440

4

  const Pel* srcLineBelow = srcLine + srcStride;

441

4

  startY                  = 0;

442

4

  if( !isAboveAvail )

443

0

  {

444

0

    startY       = 1;

445

0

    srcLineAbove = srcLine;

446

0

    srcLine += srcStride;

447

0

    resLine += resStride;

448

0

    srcLineBelow = srcLine + srcStride;

449

0

  }

450

4

  endY = height;

451

4

  if( !isBelowAvail )

452

0

  {

453

0

    endY = height - 1;

454

0

  }

455

4

#  if defined( USE_AVX2 )

456

  // AVX2

457

4

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

458

4

  {

459

4

    __m256i vsrca, vsrcat, vsrcab;

460

461

4

    __m256i    vbaseoffset  = _mm256_set1_epi16( 2 );

462

4

    __m256i    vplusone     = _mm256_set1_epi16( 1 );

463

4

    __m256i    vzero        = _mm256_set1_epi8( 0 );

464

4

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

465

4

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

466

4

    const Pel* srcLineBelow = srcLine + srcStride;

467

468

4

    if( isCtuCrossedByVirtualBoundaries )

469

0

    {

470

0

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

471

0

      {

472

0

        if( !isHorProcessDisabled( y, numHorVirBndry, horVirBndryPos ) )

473

0

        {

474

0

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

475

0

          {

476

0

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

477

0

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

478

0

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

479

0

            vsrcat            = _mm256_subs_epi16( vsrca, vsrcat );

480

0

            vsrcab            = _mm256_subs_epi16( vsrca, vsrcab );

481

0

            __m256i vsignt    = _mm256_sign_epi16( vplusone, vsrcat );

482

0

            __m256i vsignb    = _mm256_sign_epi16( vplusone, vsrcab );

483

0

            __m256i vsign     = _mm256_adds_epi16( _mm256_adds_epi16( vsignt, vsignb ), vbaseoffset );

484

0

            __m256i veoffsets = _mm256_shuffle_epi8( voffsettbl, vsign );

485

0

            veoffsets         = _mm256_slli_epi16( veoffsets, 8 );

486

0

            veoffsets         = _mm256_srai_epi16( veoffsets, 8 );

487

488

0

            vsrca = _mm256_add_epi16( vsrca, veoffsets );

489

0

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

490

491

0

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

492

0

          }

493

0

        }

494

0

        srcLine += srcStride;

495

0

        srcLineBelow += srcStride;

496

0

        srcLineAbove += srcStride;

497

0

        resLine += resStride;

498

0

      }

499

0

    }

500

4

    else

501

4

    {

502

403

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

503

399

      {

504

3.08k

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

505

2.68k

        {

506

2.68k

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

507

2.68k

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

508

2.68k

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

509

2.68k

          vsrcat            = _mm256_subs_epi16( vsrca, vsrcat );

510

2.68k

          vsrcab            = _mm256_subs_epi16( vsrca, vsrcab );

511

2.68k

          __m256i vsignt    = _mm256_sign_epi16( vplusone, vsrcat );

512

2.68k

          __m256i vsignb    = _mm256_sign_epi16( vplusone, vsrcab );

513

2.68k

          __m256i vsign     = _mm256_adds_epi16( _mm256_adds_epi16( vsignt, vsignb ), vbaseoffset );

514

2.68k

          __m256i veoffsets = _mm256_shuffle_epi8( voffsettbl, vsign );

515

2.68k

          veoffsets         = _mm256_slli_epi16( veoffsets, 8 );

516

2.68k

          veoffsets         = _mm256_srai_epi16( veoffsets, 8 );

517