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

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

{

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

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

#  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

115

5

{

116

5

  const Pel* srcLine = srcBlk;

117

5

  Pel*       resLine = resBlk;

118

119

5

  const int shiftBits        = channelBitDepth - NUM_SAO_BO_CLASSES_LOG2;

120

5

  int8_t    p_eo_offsets[16] = { 0 };

121

25

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

122

20

  {

123

20

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

124

20

  }

125

593

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

126

588

  {

127

4.76k

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

128

4.17k

    {

129

4.17k

#  ifdef USE_AVX2

130

      // AVX2

131

4.25k

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

132

4.27k

      {

133

4.27k

        __m256i vbaseoffset = _mm256_set1_epi16( startIdx - MAX_NUM_SAO_CLASSES );

134

4.27k

        __m256i vminus      = _mm256_set1_epi8( -1 );

135

4.27k

        __m256i vzero       = _mm256_set1_epi8( 0 );

136

137

4.27k

        __m256i vfour      = _mm256_set1_epi16( 4 );

138

4.27k

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

139

4.27k

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

140

141

4.27k

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

142

4.27k

        __m256i bands = _mm256_srai_epi16( vsrc, shiftBits );

143

4.27k

        bands         = _mm256_sub_epi16 ( bands, vbaseoffset );

144

4.27k

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

145

4.27k

        __m256i mask1 = _mm256_cmpgt_epi16( bands, vminus );

146

4.27k

        __m256i mask2 = _mm256_cmpgt_epi16( vfour, bands );

147

148

4.27k

        __m256i veoffsets = _mm256_shuffle_epi8( voffsettbl, bands );

149

4.27k

        veoffsets         = _mm256_slli_epi16( veoffsets, 8 );

150

4.27k

        veoffsets         = _mm256_srai_epi16( veoffsets, 8 );

151

152

4.27k

        veoffsets = _mm256_and_si256( veoffsets, mask1 );

153

4.27k

        veoffsets = _mm256_and_si256( veoffsets, mask2 );

154

155

4.27k

        vsrc = _mm256_add_epi16( vsrc, veoffsets );

156

4.27k

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

157

4.27k

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

158

159

4.27k

        x += 16;

160

4.27k

      }

161

18.4E

      else

162

18.4E

#  endif

163

18.4E

      {

164

18.4E

        __m128i vbaseoffset = _mm_set1_epi16( startIdx - MAX_NUM_SAO_CLASSES );

165

18.4E

        __m128i vminus      = _mm_set1_epi8( -1 );

166

18.4E

        __m128i vzero       = _mm_set1_epi8( 0 );

167

168

18.4E

        __m128i vfour      = _mm_set1_epi16( 4 );

169

18.4E

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

170

18.4E

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

171

172

18.4E

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

173

18.4E

        __m128i bands = _mm_srai_epi16( vsrc, shiftBits );

174

18.4E

        bands         = _mm_sub_epi16 ( bands, vbaseoffset );

175

18.4E

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

176

18.4E

        __m128i mask1 = _mm_cmpgt_epi16( bands, vminus );

177

18.4E

        __m128i mask2 = _mm_cmplt_epi16( bands, vfour );

178

179

18.4E

        __m128i veoffsets = _mm_shuffle_epi8( voffsettbl, bands );

180

18.4E

        veoffsets         = _mm_slli_epi16( veoffsets, 8 );

181

18.4E

        veoffsets         = _mm_srai_epi16( veoffsets, 8 );

182

183

18.4E

        veoffsets = _mm_and_si128( veoffsets, mask1 );

184

18.4E

        veoffsets = _mm_and_si128( veoffsets, mask2 );

185

186

18.4E

        vsrc = _mm_add_epi16( vsrc, veoffsets );

187

18.4E

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

188

18.4E

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

189

190

18.4E

        x += 8;

191

18.4E

      }

192

4.17k

    }

193

588

    srcLine += srcStride;

194

588

    resLine += resStride;

195

588

  }

196

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

224

7

{

225

7

  const Pel* srcLine = srcBlk;

226

7

  Pel*       resLine = resBlk;

227

228

7

  int    x, y, startX, endX, edgeType;

229

7

  int8_t signLeft, signRight;

230

231

7

  if( isLeftAvail && isRightAvail )

232

4

  {

233

4

    int8_t p_eo_offsets[16] = { 0 };

234

24

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

235

20

    {

236

20

      p_eo_offsets[i] = offset[i];

237

20

    }

238

4

#if defined( USE_AVX2 )

239

    // AVX2

240

4

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

241

4

    {

242

243

4

      __m256i vsrca, vsrcal, vsrcar, virBmask;

244

245

4

      __m256i vbaseoffset = _mm256_set1_epi16( 2 );

246

4

      __m256i vplusone    = _mm256_set1_epi16( 1 );

247

4

      __m256i vzero       = _mm256_set1_epi8( 0 );

248

4

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

249

4

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

250

251

4

      if( isCtuCrossedByVirtualBoundaries )

252

0

      {

253

0

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

254

0

        {

255

0

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

256

0

          {

257

0

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

258

0

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

259

0

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

260

0

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

261

262

0

            vsrcal            = _mm256_subs_epi16( vsrca, vsrcal );

263

0

            vsrcar            = _mm256_subs_epi16( vsrca, vsrcar );

264

0

            __m256i vsignl    = _mm256_sign_epi16( vplusone, vsrcal );

265

0

            __m256i vsignr    = _mm256_sign_epi16( vplusone, vsrcar );

266

0

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

267

0

            __m256i veoffsets = _mm256_shuffle_epi8( voffsettbl, vsign );

268

0

            veoffsets         = _mm256_slli_epi16( veoffsets, 8 );

269

0

            veoffsets         = _mm256_srai_epi16( veoffsets, 8 );

270

271

0

            vsrcal = _mm256_add_epi16( vsrca, veoffsets );

272

0

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

273

274

0

            vsrcar = _mm256_blendv_epi8( vsrcal, vsrca, virBmask );

275

276

0

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

277

0

          }

278

0

          srcLine += srcStride;

279

0

          resLine += resStride;

280

0

        }

281

0

      }

282

4

      else

283

4

      {

284

418

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

285

414

        {

286

3.48k

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

287

3.06k

          {

288

3.06k

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

289

3.06k

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

290

3.06k

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

291

3.06k

            vsrcal            = _mm256_subs_epi16( vsrca, vsrcal );

292

3.06k

            vsrcar            = _mm256_subs_epi16( vsrca, vsrcar );

293

3.06k

            __m256i vsignl    = _mm256_sign_epi16( vplusone, vsrcal );

294

3.06k

            __m256i vsignr    = _mm256_sign_epi16( vplusone, vsrcar );

295

3.06k

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

296

3.06k

            __m256i veoffsets = _mm256_shuffle_epi8( voffsettbl, vsign );

297

3.06k

            veoffsets         = _mm256_slli_epi16( veoffsets, 8 );

298

3.06k

            veoffsets         = _mm256_srai_epi16( veoffsets, 8 );

299

300

3.06k

            vsrca = _mm256_add_epi16( vsrca, veoffsets );

301

3.06k

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

302

303

3.06k

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

304

3.06k

          }

305

414

          srcLine += srcStride;

306

414

          resLine += resStride;

307

414

        }

308

4

      }

309

4

    }

310

0

    else

311

0

#  endif

312

0

    {

313

0

      __m128i vsrca, vsrcal, vsrcar, virBmask;

314

0

      __m128i vbaseoffset = _mm_set1_epi16( 2 );

315

0

      __m128i vplusone    = _mm_set1_epi16( 1 );

316

0

      __m128i vzero       = _mm_set1_epi8( 0 );

317

0

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

318

0

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

319

0

      if( isCtuCrossedByVirtualBoundaries )

320

0

      {

321

0

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

322

0

        {

323

0

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

324

0

          {

325

0

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

326

0

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

327

0

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

328

0

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

329

0

            vsrcal            = _mm_subs_epi16( vsrca, vsrcal );

330

0

            vsrcar            = _mm_subs_epi16( vsrca, vsrcar );

331

0

            __m128i vsignl    = _mm_sign_epi16( vplusone, vsrcal );

332

0

            __m128i vsignr    = _mm_sign_epi16( vplusone, vsrcar );

333

0

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

334

0

            __m128i veoffsets = _mm_shuffle_epi8( voffsettbl, vsign );

335

0

            veoffsets         = _mm_slli_epi16( veoffsets, 8 );

336

0

            veoffsets         = _mm_srai_epi16( veoffsets, 8 );

337

338

0

            vsrcal = _mm_add_epi16( vsrca, veoffsets );

339

0

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

340

341

0

            vsrcar = _mm_blendv_epi8( vsrcal, vsrca, virBmask );

342

343

0

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

344

0

          }

345

0

          srcLine += srcStride;

346

0

          resLine += resStride;

347

0

        }

348

0

      }

349

0

      else

350

0

      {

351

0

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

352

0

        {

353

0

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

354

0

          {

355

0

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

356

0

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

357

0

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

358

0

            vsrcal            = _mm_subs_epi16( vsrca, vsrcal );

359

0

            vsrcar            = _mm_subs_epi16( vsrca, vsrcar );

360

0

            __m128i vsignl    = _mm_sign_epi16( vplusone, vsrcal );

361

0

            __m128i vsignr    = _mm_sign_epi16( vplusone, vsrcar );

362

0

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

363

0

            __m128i veoffsets = _mm_shuffle_epi8( voffsettbl, vsign );

364

0

            veoffsets         = _mm_slli_epi16( veoffsets, 8 );

365

0

            veoffsets         = _mm_srai_epi16( veoffsets, 8 );

366

367

0

            vsrca = _mm_add_epi16( vsrca, veoffsets );

368

0

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

369

0

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

370

0

          }

371

0

          srcLine += srcStride;

372

0

          resLine += resStride;

373

0

        }

374

0

      }

375

0

    }

376

4

  }

377

3

  else

378

3

  {

379

3

    offset += 2;

380

3

    startX = isLeftAvail ? 0 : 1;

381

3

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

382

387

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

383

384

    {

384

384

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

385

44.7k

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

386

44.3k

      {

387

44.3k

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

388

44.3k

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

389

0

        {

390

0

          signLeft = -signRight;

391

0

          continue;

392

0

        }

393

44.3k

        edgeType = signRight + signLeft;

394

44.3k

        signLeft = -signRight;

395

396

44.3k

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

397

44.3k

      }

398

399

384

      srcLine += srcStride;

400

384

      resLine += resStride;

401

384

    }

402

3

  }

403

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

431

4

{

432

4

  const Pel* srcLine = srcBlk;

433

4

  Pel*       resLine = resBlk;

434

435

4

  int x, y, startY, endY;

436

437

4

  int8_t p_eo_offsets[16] = { 0 };

438

24

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

439

20

  {

440

20

    p_eo_offsets[i] = offset[i];

441

20

  }

442

4

  const Pel* srcLineAbove = srcLine - srcStride;

443

4

  const Pel* srcLineBelow = srcLine + srcStride;

444

4

  startY                  = 0;

445

4

  if( !isAboveAvail )

446

0

  {

447

0

    startY       = 1;

448

0

    srcLineAbove = srcLine;

449

0

    srcLine += srcStride;

450

0

    resLine += resStride;

451

0

    srcLineBelow = srcLine + srcStride;

452

0

  }

453

4

  endY = height;

454

4

  if( !isBelowAvail )

455

0

  {

456

0

    endY = height - 1;

457

0

  }

458

4

#  if defined( USE_AVX2 )

459

  // AVX2

460

4

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

461

4

  {

462

4

    __m256i vsrca, vsrcat, vsrcab;

463

464

4

    __m256i    vbaseoffset  = _mm256_set1_epi16( 2 );

465

4

    __m256i    vplusone     = _mm256_set1_epi16( 1 );

466

4

    __m256i    vzero        = _mm256_set1_epi8( 0 );

467

4

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

468

4

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

469

4

    const Pel* srcLineBelow = srcLine + srcStride;

470

471

4

    if( isCtuCrossedByVirtualBoundaries )

472

0

    {

473

0

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

474

0

      {

475

0

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

476

0

        {

477

0

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

478

0

          {

479

0

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

480

0

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

481

0

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

482

0

            vsrcat            = _mm256_subs_epi16( vsrca, vsrcat );

483

0

            vsrcab            = _mm256_subs_epi16( vsrca, vsrcab );

484

0

            __m256i vsignt    = _mm256_sign_epi16( vplusone, vsrcat );

485

0

            __m256i vsignb    = _mm256_sign_epi16( vplusone, vsrcab );

486

0

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

487

0

            __m256i veoffsets = _mm256_shuffle_epi8( voffsettbl, vsign );

488

0

            veoffsets         = _mm256_slli_epi16( veoffsets, 8 );

489

0

            veoffsets         = _mm256_srai_epi16( veoffsets, 8 );

490

491

0

            vsrca = _mm256_add_epi16( vsrca, veoffsets );

492

0

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

493

494

0

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

495

0

          }

496

0

        }

497

0

        srcLine += srcStride;

498

0

        srcLineBelow += srcStride;

499

0

        srcLineAbove += srcStride;

500

0

        resLine += resStride;

501

0

      }

502

0

    }

503

4

    else

504

4

    {

505

438

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

506

434

      {

507

3.67k

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

508

3.23k

        {

509

3.23k

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

510

3.23k

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

511

3.23k

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

512

3.23k

          vsrcat            = _mm256_subs_epi16( vsrca, vsrcat );

513

3.23k

          vsrcab            = _mm256_subs_epi16( vsrca, vsrcab );

514

3.23k

          __m256i vsignt    = _mm256_sign_epi16( vplusone, vsrcat );

515

3.23k

          __m256i vsignb    = _mm256_sign_epi16( vplusone, vsrcab );

516

3.23k

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

517

3.23k

          __m256i veoffsets = _mm256_shuffle_epi8( voffsettbl, vsign );