git.sesse.net Git - x264/blob - common/x86/mc-c.c

   1 /*****************************************************************************
   2  * mc-c.c: x86 motion compensation
   3  *****************************************************************************
   4  * Copyright (C) 2003-2014 x264 project
   5  *
   6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
   7  *          Loren Merritt <lorenm@u.washington.edu>
   8  *          Fiona Glaser <fiona@x264.com>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License as published by
  12  * the Free Software Foundation; either version 2 of the License, or
  13  * (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  * GNU General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with this program; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
  23  *
  24  * This program is also available under a commercial proprietary license.
  25  * For more information, contact us at licensing@x264.com.
  26  *****************************************************************************/
  27
  28 #include <stdlib.h>
  29 #include <stdio.h>
  30 #include <string.h>
  31
  32 #include "common/common.h"
  33 #include "mc.h"
  34
  35 #define DECL_SUF( func, args )\
  36     void func##_mmx2 args;\
  37     void func##_sse2 args;\
  38     void func##_ssse3 args;\
  39     void func##_avx2 args;
  40
  41 DECL_SUF( x264_pixel_avg_16x16, ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
  42 DECL_SUF( x264_pixel_avg_16x8,  ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
  43 DECL_SUF( x264_pixel_avg_8x16,  ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
  44 DECL_SUF( x264_pixel_avg_8x8,   ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
  45 DECL_SUF( x264_pixel_avg_8x4,   ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
  46 DECL_SUF( x264_pixel_avg_4x16,  ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
  47 DECL_SUF( x264_pixel_avg_4x8,   ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
  48 DECL_SUF( x264_pixel_avg_4x4,   ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
  49 DECL_SUF( x264_pixel_avg_4x2,   ( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t, int ))
  50
  51 #define MC_WEIGHT(w,type) \
  52     void x264_mc_weight_w##w##_##type( pixel *, intptr_t, pixel *, intptr_t, const x264_weight_t *, int );
  53
  54 #define MC_WEIGHT_OFFSET(w,type) \
  55     void x264_mc_offsetadd_w##w##_##type( pixel *, intptr_t, pixel *, intptr_t, const x264_weight_t *, int ); \
  56     void x264_mc_offsetsub_w##w##_##type( pixel *, intptr_t, pixel *, intptr_t, const x264_weight_t *, int ); \
  57     MC_WEIGHT(w,type)
  58
  59 MC_WEIGHT_OFFSET( 4, mmx2 )
  60 MC_WEIGHT_OFFSET( 8, mmx2 )
  61 MC_WEIGHT_OFFSET( 12, mmx2 )
  62 MC_WEIGHT_OFFSET( 16, mmx2 )
  63 MC_WEIGHT_OFFSET( 20, mmx2 )
  64 MC_WEIGHT_OFFSET( 12, sse2 )
  65 MC_WEIGHT_OFFSET( 16, sse2 )
  66 MC_WEIGHT_OFFSET( 20, sse2 )
  67 #if HIGH_BIT_DEPTH
  68 MC_WEIGHT_OFFSET( 8, sse2 )
  69 #endif
  70 MC_WEIGHT( 8, sse2  )
  71 MC_WEIGHT( 4, ssse3 )
  72 MC_WEIGHT( 8, ssse3 )
  73 MC_WEIGHT( 12, ssse3 )
  74 MC_WEIGHT( 16, ssse3 )
  75 MC_WEIGHT( 20, ssse3 )
  76 MC_WEIGHT( 8, avx2 )
  77 MC_WEIGHT( 16, avx2 )
  78 MC_WEIGHT( 20, avx2 )
  79 #undef MC_OFFSET
  80 #undef MC_WEIGHT
  81
  82 void x264_mc_copy_w4_mmx ( pixel *, intptr_t, pixel *, intptr_t, int );
  83 void x264_mc_copy_w8_mmx ( pixel *, intptr_t, pixel *, intptr_t, int );
  84 void x264_mc_copy_w8_sse ( pixel *, intptr_t, pixel *, intptr_t, int );
  85 void x264_mc_copy_w16_mmx( pixel *, intptr_t, pixel *, intptr_t, int );
  86 void x264_mc_copy_w16_sse( pixel *, intptr_t, pixel *, intptr_t, int );
  87 void x264_mc_copy_w16_aligned_sse( pixel *, intptr_t, pixel *, intptr_t, int );
  88 void x264_mc_copy_w16_avx( uint16_t *, intptr_t, uint16_t *, intptr_t, int );
  89 void x264_mc_copy_w16_aligned_avx( uint16_t *, intptr_t, uint16_t *, intptr_t, int );
  90 void x264_prefetch_fenc_420_mmx2( pixel *, intptr_t, pixel *, intptr_t, int );
  91 void x264_prefetch_fenc_422_mmx2( pixel *, intptr_t, pixel *, intptr_t, int );
  92 void x264_prefetch_ref_mmx2( pixel *, intptr_t, int );
  93 void x264_plane_copy_core_mmx2( pixel *, intptr_t, pixel *, intptr_t, int w, int h );
  94 void x264_plane_copy_c( pixel *, intptr_t, pixel *, intptr_t, int w, int h );
  95 void x264_plane_copy_interleave_core_mmx2( pixel *dst,  intptr_t i_dst,
  96                                            pixel *srcu, intptr_t i_srcu,
  97                                            pixel *srcv, intptr_t i_srcv, int w, int h );
  98 void x264_plane_copy_interleave_core_sse2( pixel *dst,  intptr_t i_dst,
  99                                            pixel *srcu, intptr_t i_srcu,
 100                                            pixel *srcv, intptr_t i_srcv, int w, int h );
 101 void x264_plane_copy_interleave_core_avx( pixel *dst,  intptr_t i_dst,
 102                                           pixel *srcu, intptr_t i_srcu,
 103                                           pixel *srcv, intptr_t i_srcv, int w, int h );
 104 void x264_plane_copy_interleave_c( pixel *dst,  intptr_t i_dst,
 105                                    pixel *srcu, intptr_t i_srcu,
 106                                    pixel *srcv, intptr_t i_srcv, int w, int h );
 107 void x264_plane_copy_deinterleave_mmx( pixel *dstu, intptr_t i_dstu,
 108                                        pixel *dstv, intptr_t i_dstv,
 109                                        pixel *src,  intptr_t i_src, int w, int h );
 110 void x264_plane_copy_deinterleave_sse2( pixel *dstu, intptr_t i_dstu,
 111                                         pixel *dstv, intptr_t i_dstv,
 112                                         pixel *src,  intptr_t i_src, int w, int h );
 113 void x264_plane_copy_deinterleave_ssse3( uint8_t *dstu, intptr_t i_dstu,
 114                                          uint8_t *dstv, intptr_t i_dstv,
 115                                          uint8_t *src,  intptr_t i_src, int w, int h );
 116 void x264_plane_copy_deinterleave_avx( uint16_t *dstu, intptr_t i_dstu,
 117                                        uint16_t *dstv, intptr_t i_dstv,
 118                                        uint16_t *src,  intptr_t i_src, int w, int h );
 119 void x264_plane_copy_deinterleave_v210_ssse3( uint16_t *dstu, intptr_t i_dstu,
 120                                               uint16_t *dstv, intptr_t i_dstv,
 121                                               uint32_t *src,  intptr_t i_src, int w, int h );
 122 void x264_plane_copy_deinterleave_v210_avx  ( uint16_t *dstu, intptr_t i_dstu,
 123                                               uint16_t *dstv, intptr_t i_dstv,
 124                                               uint32_t *src,  intptr_t i_src, int w, int h );
 125 void x264_plane_copy_deinterleave_v210_avx2 ( uint16_t *dstu, intptr_t i_dstu,
 126                                               uint16_t *dstv, intptr_t i_dstv,
 127                                               uint32_t *src,  intptr_t i_src, int w, int h );
 128 void x264_store_interleave_chroma_mmx2( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
 129 void x264_store_interleave_chroma_sse2( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
 130 void x264_store_interleave_chroma_avx ( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
 131 void x264_load_deinterleave_chroma_fenc_mmx ( pixel *dst, pixel *src, intptr_t i_src, int height );
 132 void x264_load_deinterleave_chroma_fenc_sse2( pixel *dst, pixel *src, intptr_t i_src, int height );
 133 void x264_load_deinterleave_chroma_fenc_ssse3( uint8_t *dst, uint8_t *src, intptr_t i_src, int height );
 134 void x264_load_deinterleave_chroma_fenc_avx( uint16_t *dst, uint16_t *src, intptr_t i_src, int height );
 135 void x264_load_deinterleave_chroma_fdec_mmx ( pixel *dst, pixel *src, intptr_t i_src, int height );
 136 void x264_load_deinterleave_chroma_fdec_sse2( pixel *dst, pixel *src, intptr_t i_src, int height );
 137 void x264_load_deinterleave_chroma_fdec_ssse3( uint8_t *dst, uint8_t *src, intptr_t i_src, int height );
 138 void x264_load_deinterleave_chroma_fdec_avx( uint16_t *dst, uint16_t *src, intptr_t i_src, int height );
 139 void *x264_memcpy_aligned_mmx( void *dst, const void *src, size_t n );
 140 void *x264_memcpy_aligned_sse( void *dst, const void *src, size_t n );
 141 void x264_memzero_aligned_mmx( void *dst, size_t n );
 142 void x264_memzero_aligned_sse( void *dst, size_t n );
 143 void x264_memzero_aligned_avx( void *dst, size_t n );
 144 void x264_integral_init4h_sse4( uint16_t *sum, uint8_t *pix, intptr_t stride );
 145 void x264_integral_init4h_avx2( uint16_t *sum, uint8_t *pix, intptr_t stride );
 146 void x264_integral_init8h_sse4( uint16_t *sum, uint8_t *pix, intptr_t stride );
 147 void x264_integral_init8h_avx ( uint16_t *sum, uint8_t *pix, intptr_t stride );
 148 void x264_integral_init8h_avx2( uint16_t *sum, uint8_t *pix, intptr_t stride );
 149 void x264_integral_init4v_mmx  ( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
 150 void x264_integral_init4v_sse2 ( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
 151 void x264_integral_init4v_ssse3( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
 152 void x264_integral_init4v_avx2( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
 153 void x264_integral_init8v_mmx ( uint16_t *sum8, intptr_t stride );
 154 void x264_integral_init8v_sse2( uint16_t *sum8, intptr_t stride );
 155 void x264_integral_init8v_avx2( uint16_t *sum8, intptr_t stride );
 156 void x264_mbtree_propagate_cost_sse2( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
 157                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
 158 void x264_mbtree_propagate_cost_avx ( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
 159                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
 160 void x264_mbtree_propagate_cost_fma4( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
 161                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
 162 void x264_mbtree_propagate_cost_avx2_fma3( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
 163                                            uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
 164
 165 #define MC_CHROMA(cpu)\
 166 void x264_mc_chroma_##cpu( pixel *dstu, pixel *dstv, intptr_t i_dst, pixel *src, intptr_t i_src,\
 167                            int dx, int dy, int i_width, int i_height );
 168 MC_CHROMA(mmx2)
 169 MC_CHROMA(sse2)
 170 MC_CHROMA(ssse3)
 171 MC_CHROMA(ssse3_cache64)
 172 MC_CHROMA(avx)
 173 MC_CHROMA(avx2)
 174
 175 #define LOWRES(cpu)\
 176 void x264_frame_init_lowres_core_##cpu( pixel *src0, pixel *dst0, pixel *dsth, pixel *dstv, pixel *dstc,\
 177                                         intptr_t src_stride, intptr_t dst_stride, int width, int height );
 178 LOWRES(mmx2)
 179 LOWRES(cache32_mmx2)
 180 LOWRES(sse2)
 181 LOWRES(ssse3)
 182 LOWRES(avx)
 183 LOWRES(xop)
 184 LOWRES(avx2)
 185
 186 #define PIXEL_AVG_W(width,cpu)\
 187 void x264_pixel_avg2_w##width##_##cpu( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t );
 188 /* This declares some functions that don't exist, but that isn't a problem. */
 189 #define PIXEL_AVG_WALL(cpu)\
 190 PIXEL_AVG_W(4,cpu); PIXEL_AVG_W(8,cpu); PIXEL_AVG_W(10,cpu); PIXEL_AVG_W(12,cpu); PIXEL_AVG_W(16,cpu); PIXEL_AVG_W(18,cpu); PIXEL_AVG_W(20,cpu);
 191
 192 PIXEL_AVG_WALL(mmx2)
 193 PIXEL_AVG_WALL(cache32_mmx2)
 194 PIXEL_AVG_WALL(cache64_mmx2)
 195 PIXEL_AVG_WALL(cache64_sse2)
 196 PIXEL_AVG_WALL(sse2)
 197 PIXEL_AVG_WALL(cache64_ssse3)
 198 PIXEL_AVG_WALL(avx2)
 199
 200 #define PIXEL_AVG_WTAB(instr, name1, name2, name3, name4, name5)\
 201 static void (* const x264_pixel_avg_wtab_##instr[6])( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t ) =\
 202 {\
 203     NULL,\
 204     x264_pixel_avg2_w4_##name1,\
 205     x264_pixel_avg2_w8_##name2,\
 206     x264_pixel_avg2_w12_##name3,\
 207     x264_pixel_avg2_w16_##name4,\
 208     x264_pixel_avg2_w20_##name5,\
 209 };
 210
 211 #if HIGH_BIT_DEPTH
 212 /* we can replace w12/w20 with w10/w18 as only 9/17 pixels in fact are important */
 213 #define x264_pixel_avg2_w12_mmx2       x264_pixel_avg2_w10_mmx2
 214 #define x264_pixel_avg2_w20_mmx2       x264_pixel_avg2_w18_mmx2
 215 #define x264_pixel_avg2_w12_sse2         x264_pixel_avg2_w10_sse2
 216 #define x264_pixel_avg2_w20_sse2         x264_pixel_avg2_w18_sse2
 217 #define x264_pixel_avg2_w12_avx2         x264_pixel_avg2_w16_avx2
 218 #define x264_pixel_avg2_w20_avx2         x264_pixel_avg2_w18_avx2
 219 #else
 220 /* w16 sse2 is faster than w12 mmx as long as the cacheline issue is resolved */
 221 #define x264_pixel_avg2_w12_cache64_ssse3 x264_pixel_avg2_w16_cache64_ssse3
 222 #define x264_pixel_avg2_w12_cache64_sse2 x264_pixel_avg2_w16_cache64_sse2
 223 #define x264_pixel_avg2_w12_sse3         x264_pixel_avg2_w16_sse3
 224 #define x264_pixel_avg2_w12_sse2         x264_pixel_avg2_w16_sse2
 225 #endif // HIGH_BIT_DEPTH
 226
 227 PIXEL_AVG_WTAB(mmx2, mmx2, mmx2, mmx2, mmx2, mmx2)
 228 #if HIGH_BIT_DEPTH
 229 PIXEL_AVG_WTAB(sse2, mmx2, sse2, sse2, sse2, sse2)
 230 PIXEL_AVG_WTAB(avx2, mmx2, sse2, avx2, avx2, avx2)
 231 #else // !HIGH_BIT_DEPTH
 232 #if ARCH_X86
 233 PIXEL_AVG_WTAB(cache32_mmx2, mmx2, cache32_mmx2, cache32_mmx2, cache32_mmx2, cache32_mmx2)
 234 PIXEL_AVG_WTAB(cache64_mmx2, mmx2, cache64_mmx2, cache64_mmx2, cache64_mmx2, cache64_mmx2)
 235 #endif
 236 PIXEL_AVG_WTAB(sse2, mmx2, mmx2, sse2, sse2, sse2)
 237 PIXEL_AVG_WTAB(cache64_sse2, mmx2, cache64_mmx2, cache64_sse2, cache64_sse2, cache64_sse2)
 238 PIXEL_AVG_WTAB(cache64_ssse3, mmx2, cache64_mmx2, cache64_ssse3, cache64_ssse3, cache64_sse2)
 239 PIXEL_AVG_WTAB(cache64_ssse3_atom, mmx2, mmx2, cache64_ssse3, cache64_ssse3, sse2)
 240 PIXEL_AVG_WTAB(avx2, mmx2, mmx2, sse2, sse2, avx2)
 241 #endif // HIGH_BIT_DEPTH
 242
 243 #define MC_COPY_WTAB(instr, name1, name2, name3)\
 244 static void (* const x264_mc_copy_wtab_##instr[5])( pixel *, intptr_t, pixel *, intptr_t, int ) =\
 245 {\
 246     NULL,\
 247     x264_mc_copy_w4_##name1,\
 248     x264_mc_copy_w8_##name2,\
 249     NULL,\
 250     x264_mc_copy_w16_##name3,\
 251 };
 252
 253 MC_COPY_WTAB(mmx,mmx,mmx,mmx)
 254 #if HIGH_BIT_DEPTH
 255 MC_COPY_WTAB(sse,mmx,sse,sse)
 256 MC_COPY_WTAB(avx,mmx,sse,avx)
 257 #else
 258 MC_COPY_WTAB(sse,mmx,mmx,sse)
 259 #endif
 260
 261 #define MC_WEIGHT_WTAB(function, instr, name1, name2, w12version)\
 262     static void (* x264_mc_##function##_wtab_##instr[6])( pixel *, intptr_t, pixel *, intptr_t, const x264_weight_t *, int ) =\
 263 {\
 264     x264_mc_##function##_w4_##name1,\
 265     x264_mc_##function##_w4_##name1,\
 266     x264_mc_##function##_w8_##name2,\
 267     x264_mc_##function##_w##w12version##_##instr,\
 268     x264_mc_##function##_w16_##instr,\
 269     x264_mc_##function##_w20_##instr,\
 270 };
 271
 272 #if HIGH_BIT_DEPTH
 273 MC_WEIGHT_WTAB(weight,mmx2,mmx2,mmx2,12)
 274 MC_WEIGHT_WTAB(offsetadd,mmx2,mmx2,mmx2,12)
 275 MC_WEIGHT_WTAB(offsetsub,mmx2,mmx2,mmx2,12)
 276 MC_WEIGHT_WTAB(weight,sse2,mmx2,sse2,12)
 277 MC_WEIGHT_WTAB(offsetadd,sse2,mmx2,sse2,16)
 278 MC_WEIGHT_WTAB(offsetsub,sse2,mmx2,sse2,16)
 279
 280 static void x264_weight_cache_mmx2( x264_t *h, x264_weight_t *w )
 281 {
 282     if( w->i_scale == 1<<w->i_denom )
 283     {
 284         if( w->i_offset < 0 )
 285             w->weightfn = h->mc.offsetsub;
 286         else
 287             w->weightfn = h->mc.offsetadd;
 288         for( int i = 0; i < 8; i++ )
 289             w->cachea[i] = abs(w->i_offset<<(BIT_DEPTH-8));
 290         return;
 291     }
 292     w->weightfn = h->mc.weight;
 293     int den1 = 1<<w->i_denom;
 294     int den2 = w->i_scale<<1;
 295     int den3 = 1+(w->i_offset<<(BIT_DEPTH-8+1));
 296     for( int i = 0; i < 8; i++ )
 297     {
 298         w->cachea[i] = den1;
 299         w->cacheb[i] = i&1 ? den3 : den2;
 300     }
 301 }
 302 #else
 303 MC_WEIGHT_WTAB(weight,mmx2,mmx2,mmx2,12)
 304 MC_WEIGHT_WTAB(offsetadd,mmx2,mmx2,mmx2,12)
 305 MC_WEIGHT_WTAB(offsetsub,mmx2,mmx2,mmx2,12)
 306 MC_WEIGHT_WTAB(weight,sse2,mmx2,sse2,16)
 307 MC_WEIGHT_WTAB(offsetadd,sse2,mmx2,mmx2,16)
 308 MC_WEIGHT_WTAB(offsetsub,sse2,mmx2,mmx2,16)
 309 MC_WEIGHT_WTAB(weight,ssse3,ssse3,ssse3,16)
 310 MC_WEIGHT_WTAB(weight,avx2,ssse3,avx2,16)
 311
 312 static void x264_weight_cache_mmx2( x264_t *h, x264_weight_t *w )
 313 {
 314     int i;
 315     int16_t den1;
 316
 317     if( w->i_scale == 1<<w->i_denom )
 318     {
 319         if( w->i_offset < 0 )
 320             w->weightfn = h->mc.offsetsub;
 321         else
 322             w->weightfn = h->mc.offsetadd;
 323         memset( w->cachea, abs(w->i_offset), sizeof(w->cachea) );
 324         return;
 325     }
 326     w->weightfn = h->mc.weight;
 327     den1 = 1 << (w->i_denom - 1) | w->i_offset << w->i_denom;
 328     for( i = 0; i < 8; i++ )
 329     {
 330         w->cachea[i] = w->i_scale;
 331         w->cacheb[i] = den1;
 332     }
 333 }
 334
 335 static void x264_weight_cache_ssse3( x264_t *h, x264_weight_t *w )
 336 {
 337     int i, den1;
 338     if( w->i_scale == 1<<w->i_denom )
 339     {
 340         if( w->i_offset < 0 )
 341             w->weightfn = h->mc.offsetsub;
 342         else
 343             w->weightfn = h->mc.offsetadd;
 344
 345         memset( w->cachea, abs( w->i_offset ), sizeof(w->cachea) );
 346         return;
 347     }
 348     w->weightfn = h->mc.weight;
 349     den1 = w->i_scale << (8 - w->i_denom);
 350     for( i = 0; i < 8; i++ )
 351     {
 352         w->cachea[i] = den1;
 353         w->cacheb[i] = w->i_offset;
 354     }
 355 }
 356 #endif // !HIGH_BIT_DEPTH
 357
 358 static const uint8_t hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
 359 static const uint8_t hpel_ref1[16] = {0,0,0,0,2,2,3,2,2,2,3,2,2,2,3,2};
 360
 361 #define MC_LUMA(name,instr1,instr2)\
 362 static void mc_luma_##name( pixel *dst,    intptr_t i_dst_stride,\
 363                             pixel *src[4], intptr_t i_src_stride,\
 364                             int mvx, int mvy,\
 365                             int i_width, int i_height, const x264_weight_t *weight )\
 366 {\
 367     int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
 368     int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
 369     pixel *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
 370     if( qpel_idx & 5 ) /* qpel interpolation needed */\
 371     {\
 372         pixel *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
 373         x264_pixel_avg_wtab_##instr1[i_width>>2](\
 374                 dst, i_dst_stride, src1, i_src_stride,\
 375                 src2, i_height );\
 376         if( weight->weightfn )\
 377             weight->weightfn[i_width>>2]( dst, i_dst_stride, dst, i_dst_stride, weight, i_height );\
 378     }\
 379     else if( weight->weightfn )\
 380         weight->weightfn[i_width>>2]( dst, i_dst_stride, src1, i_src_stride, weight, i_height );\
 381     else\
 382         x264_mc_copy_wtab_##instr2[i_width>>2](dst, i_dst_stride, src1, i_src_stride, i_height );\
 383 }
 384
 385 MC_LUMA(mmx2,mmx2,mmx)
 386 MC_LUMA(sse2,sse2,sse)
 387 #if HIGH_BIT_DEPTH
 388 MC_LUMA(avx2,avx2,avx)
 389 #else
 390 #if ARCH_X86
 391 MC_LUMA(cache32_mmx2,cache32_mmx2,mmx)
 392 MC_LUMA(cache64_mmx2,cache64_mmx2,mmx)
 393 #endif
 394 MC_LUMA(cache64_sse2,cache64_sse2,sse)
 395 MC_LUMA(cache64_ssse3,cache64_ssse3,sse)
 396 MC_LUMA(cache64_ssse3_atom,cache64_ssse3_atom,sse)
 397 #endif // !HIGH_BIT_DEPTH
 398
 399 #define GET_REF(name)\
 400 static pixel *get_ref_##name( pixel *dst,   intptr_t *i_dst_stride,\
 401                               pixel *src[4], intptr_t i_src_stride,\
 402                               int mvx, int mvy,\
 403                               int i_width, int i_height, const x264_weight_t *weight )\
 404 {\
 405     int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
 406     int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
 407     pixel *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
 408     if( qpel_idx & 5 ) /* qpel interpolation needed */\
 409     {\
 410         pixel *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
 411         x264_pixel_avg_wtab_##name[i_width>>2](\
 412                 dst, *i_dst_stride, src1, i_src_stride,\
 413                 src2, i_height );\
 414         if( weight->weightfn )\
 415             weight->weightfn[i_width>>2]( dst, *i_dst_stride, dst, *i_dst_stride, weight, i_height );\
 416         return dst;\
 417     }\
 418     else if( weight->weightfn )\
 419     {\
 420         weight->weightfn[i_width>>2]( dst, *i_dst_stride, src1, i_src_stride, weight, i_height );\
 421         return dst;\
 422     }\
 423     else\
 424     {\
 425         *i_dst_stride = i_src_stride;\
 426         return src1;\
 427     }\
 428 }
 429
 430 GET_REF(mmx2)
 431 GET_REF(sse2)
 432 GET_REF(avx2)
 433 #if !HIGH_BIT_DEPTH
 434 #if ARCH_X86
 435 GET_REF(cache32_mmx2)
 436 GET_REF(cache64_mmx2)
 437 #endif
 438 GET_REF(cache64_sse2)
 439 GET_REF(cache64_ssse3)
 440 GET_REF(cache64_ssse3_atom)
 441 #endif // !HIGH_BIT_DEPTH
 442
 443 #define HPEL(align, cpu, cpuv, cpuc, cpuh)\
 444 void x264_hpel_filter_v_##cpuv( pixel *dst, pixel *src, int16_t *buf, intptr_t stride, intptr_t width);\
 445 void x264_hpel_filter_c_##cpuc( pixel *dst, int16_t *buf, intptr_t width );\
 446 void x264_hpel_filter_h_##cpuh( pixel *dst, pixel *src, intptr_t width );\
 447 static void x264_hpel_filter_##cpu( pixel *dsth, pixel *dstv, pixel *dstc, pixel *src,\
 448                                     intptr_t stride, int width, int height, int16_t *buf )\
 449 {\
 450     intptr_t realign = (intptr_t)src & (align-1);\
 451     src -= realign;\
 452     dstv -= realign;\
 453     dstc -= realign;\
 454     dsth -= realign;\
 455     width += realign;\
 456     while( height-- )\
 457     {\
 458         x264_hpel_filter_v_##cpuv( dstv, src, buf+16, stride, width );\
 459         x264_hpel_filter_c_##cpuc( dstc, buf+16, width );\
 460         x264_hpel_filter_h_##cpuh( dsth, src, width );\
 461         dsth += stride;\
 462         dstv += stride;\
 463         dstc += stride;\
 464         src  += stride;\
 465     }\
 466     x264_sfence();\
 467 }
 468
 469 HPEL(8, mmx2, mmx2, mmx2, mmx2)
 470 #if HIGH_BIT_DEPTH
 471 HPEL(16, sse2, sse2, sse2, sse2)
 472 #else // !HIGH_BIT_DEPTH
 473 HPEL(16, sse2_amd, mmx2, mmx2, sse2)
 474 #if ARCH_X86_64
 475 void x264_hpel_filter_sse2 ( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, intptr_t stride, int width, int height, int16_t *buf );
 476 void x264_hpel_filter_ssse3( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, intptr_t stride, int width, int height, int16_t *buf );
 477 void x264_hpel_filter_avx  ( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, intptr_t stride, int width, int height, int16_t *buf );
 478 void x264_hpel_filter_avx2 ( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, intptr_t stride, int width, int height, int16_t *buf );
 479 #else
 480 HPEL(16, sse2, sse2, sse2, sse2)
 481 HPEL(16, ssse3, ssse3, ssse3, ssse3)
 482 HPEL(16, avx, avx, avx, avx)
 483 HPEL(32, avx2, avx2, avx2, avx2)
 484 #endif
 485 #endif // HIGH_BIT_DEPTH
 486
 487 static void x264_plane_copy_mmx2( pixel *dst, intptr_t i_dst, pixel *src, intptr_t i_src, int w, int h )
 488 {
 489     int c_w = 16/sizeof(pixel) - 1;
 490     if( w < 256 ) { // tiny resolutions don't want non-temporal hints. dunno the exact threshold.
 491         x264_plane_copy_c( dst, i_dst, src, i_src, w, h );
 492     } else if( !(w&c_w) ) {
 493         x264_plane_copy_core_mmx2( dst, i_dst, src, i_src, w, h );
 494     } else if( i_src > 0 ) {
 495         // have to use plain memcpy on the last line (in memory order) to avoid overreading src
 496         x264_plane_copy_core_mmx2( dst, i_dst, src, i_src, (w+c_w)&~c_w, h-1 );
 497         memcpy( dst+i_dst*(h-1), src+i_src*(h-1), w*sizeof(pixel) );
 498     } else {
 499         memcpy( dst, src, w*sizeof(pixel) );
 500         x264_plane_copy_core_mmx2( dst+i_dst, i_dst, src+i_src, i_src, (w+c_w)&~c_w, h-1 );
 501     }
 502 }
 503
 504 #define PLANE_INTERLEAVE(cpu) \
 505 static void x264_plane_copy_interleave_##cpu( pixel *dst,  intptr_t i_dst,\
 506                                               pixel *srcu, intptr_t i_srcu,\
 507                                               pixel *srcv, intptr_t i_srcv, int w, int h )\
 508 {\
 509     if( !(w&15) ) {\
 510         x264_plane_copy_interleave_core_##cpu( dst, i_dst, srcu, i_srcu, srcv, i_srcv, w, h );\
 511     } else if( w < 16 || (i_srcu ^ i_srcv) ) {\
 512         x264_plane_copy_interleave_c( dst, i_dst, srcu, i_srcu, srcv, i_srcv, w, h );\
 513     } else if( i_srcu > 0 ) {\
 514         x264_plane_copy_interleave_core_##cpu( dst, i_dst, srcu, i_srcu, srcv, i_srcv, (w+15)&~15, h-1 );\
 515         x264_plane_copy_interleave_c( dst+i_dst*(h-1), 0, srcu+i_srcu*(h-1), 0, srcv+i_srcv*(h-1), 0, w, 1 );\
 516     } else {\
 517         x264_plane_copy_interleave_c( dst, 0, srcu, 0, srcv, 0, w, 1 );\
 518         x264_plane_copy_interleave_core_##cpu( dst+i_dst, i_dst, srcu+i_srcu, i_srcu, srcv+i_srcv, i_srcv, (w+15)&~15, h-1 );\
 519     }\
 520 }
 521
 522 PLANE_INTERLEAVE(mmx2)
 523 PLANE_INTERLEAVE(sse2)
 524 #if HIGH_BIT_DEPTH
 525 PLANE_INTERLEAVE(avx)
 526 #endif
 527
 528 void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
 529 {
 530     if( !(cpu&X264_CPU_MMX) )
 531         return;
 532
 533     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_mmx;
 534     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_mmx;
 535
 536     pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_mmx;
 537
 538     pf->copy_16x16_unaligned = x264_mc_copy_w16_mmx;
 539     pf->copy[PIXEL_16x16] = x264_mc_copy_w16_mmx;
 540     pf->copy[PIXEL_8x8]   = x264_mc_copy_w8_mmx;
 541     pf->copy[PIXEL_4x4]   = x264_mc_copy_w4_mmx;
 542     pf->memcpy_aligned  = x264_memcpy_aligned_mmx;
 543     pf->memzero_aligned = x264_memzero_aligned_mmx;
 544     pf->integral_init4v = x264_integral_init4v_mmx;
 545     pf->integral_init8v = x264_integral_init8v_mmx;
 546
 547     if( !(cpu&X264_CPU_MMX2) )
 548         return;
 549
 550     pf->prefetch_fenc_420 = x264_prefetch_fenc_420_mmx2;
 551     pf->prefetch_fenc_422 = x264_prefetch_fenc_422_mmx2;
 552     pf->prefetch_ref  = x264_prefetch_ref_mmx2;
 553
 554     pf->plane_copy = x264_plane_copy_mmx2;
 555     pf->plane_copy_interleave = x264_plane_copy_interleave_mmx2;
 556     pf->store_interleave_chroma = x264_store_interleave_chroma_mmx2;
 557
 558     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_mmx2;
 559     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_mmx2;
 560     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_mmx2;
 561     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_mmx2;
 562     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_mmx2;
 563     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_mmx2;
 564     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_mmx2;
 565     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_mmx2;
 566     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_mmx2;
 567
 568     pf->mc_luma = mc_luma_mmx2;
 569     pf->get_ref = get_ref_mmx2;
 570     pf->mc_chroma = x264_mc_chroma_mmx2;
 571     pf->hpel_filter = x264_hpel_filter_mmx2;
 572     pf->weight = x264_mc_weight_wtab_mmx2;
 573     pf->weight_cache = x264_weight_cache_mmx2;
 574     pf->offsetadd = x264_mc_offsetadd_wtab_mmx2;
 575     pf->offsetsub = x264_mc_offsetsub_wtab_mmx2;
 576
 577     pf->frame_init_lowres_core = x264_frame_init_lowres_core_mmx2;
 578
 579     if( cpu&X264_CPU_SSE )
 580     {
 581         pf->memcpy_aligned  = x264_memcpy_aligned_sse;
 582         pf->memzero_aligned = x264_memzero_aligned_sse;
 583     }
 584
 585 #if HIGH_BIT_DEPTH
 586 #if ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
 587     if( cpu&(X264_CPU_CACHELINE_32|X264_CPU_CACHELINE_64) )
 588         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
 589 #endif
 590
 591     if( !(cpu&X264_CPU_SSE2) )
 592         return;
 593
 594     pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
 595
 596     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_sse2;
 597     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_sse2;
 598
 599     pf->plane_copy_interleave   = x264_plane_copy_interleave_sse2;
 600     pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
 601
 602     if( cpu&X264_CPU_SSE2_IS_FAST )
 603     {
 604         pf->get_ref = get_ref_sse2;
 605         pf->mc_luma = mc_luma_sse2;
 606         pf->hpel_filter = x264_hpel_filter_sse2;
 607     }
 608
 609     pf->integral_init4v = x264_integral_init4v_sse2;
 610     pf->integral_init8v = x264_integral_init8v_sse2;
 611     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
 612     pf->store_interleave_chroma = x264_store_interleave_chroma_sse2;
 613     pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
 614     pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
 615
 616     if( cpu&X264_CPU_SSE2_IS_SLOW )
 617         return;
 618
 619     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
 620     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_sse2;
 621     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_sse2;
 622     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_sse2;
 623     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_sse2;
 624     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_sse2;
 625     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_sse2;
 626     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_sse2;
 627     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_sse2;
 628
 629     pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse;
 630     pf->weight = x264_mc_weight_wtab_sse2;
 631
 632     if( !(cpu&X264_CPU_STACK_MOD4) )
 633         pf->mc_chroma = x264_mc_chroma_sse2;
 634
 635     if( !(cpu&X264_CPU_SSSE3) )
 636         return;
 637
 638     pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
 639     pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_ssse3;
 640
 641     if( !(cpu&(X264_CPU_SLOW_SHUFFLE|X264_CPU_SLOW_ATOM|X264_CPU_SLOW_PALIGNR)) )
 642         pf->integral_init4v = x264_integral_init4v_ssse3;
 643
 644     if( !(cpu&X264_CPU_AVX) )
 645         return;
 646
 647     pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx;
 648     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_avx;
 649     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_avx;
 650     pf->plane_copy_interleave        = x264_plane_copy_interleave_avx;
 651     pf->plane_copy_deinterleave      = x264_plane_copy_deinterleave_avx;
 652     pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_avx;
 653     pf->store_interleave_chroma      = x264_store_interleave_chroma_avx;
 654     pf->copy[PIXEL_16x16]            = x264_mc_copy_w16_aligned_avx;
 655
 656     if( !(cpu&X264_CPU_STACK_MOD4) )
 657         pf->mc_chroma = x264_mc_chroma_avx;
 658
 659     if( cpu&X264_CPU_XOP )
 660         pf->frame_init_lowres_core = x264_frame_init_lowres_core_xop;
 661
 662     if( cpu&X264_CPU_AVX2 )
 663     {
 664         pf->mc_luma = mc_luma_avx2;
 665         pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_avx2;
 666     }
 667 #else // !HIGH_BIT_DEPTH
 668
 669 #if ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
 670     if( cpu&X264_CPU_CACHELINE_32 )
 671     {
 672         pf->mc_luma = mc_luma_cache32_mmx2;
 673         pf->get_ref = get_ref_cache32_mmx2;
 674         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
 675     }
 676     else if( cpu&X264_CPU_CACHELINE_64 )
 677     {
 678         pf->mc_luma = mc_luma_cache64_mmx2;
 679         pf->get_ref = get_ref_cache64_mmx2;
 680         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
 681     }
 682 #endif
 683
 684     if( !(cpu&X264_CPU_SSE2) )
 685         return;
 686
 687     pf->integral_init4v = x264_integral_init4v_sse2;
 688     pf->integral_init8v = x264_integral_init8v_sse2;
 689     pf->hpel_filter = x264_hpel_filter_sse2_amd;
 690     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
 691
 692     if( !(cpu&X264_CPU_SSE2_IS_SLOW) )
 693     {
 694         pf->weight = x264_mc_weight_wtab_sse2;
 695         if( !(cpu&X264_CPU_SLOW_ATOM) )
 696         {
 697             pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
 698             pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
 699         }
 700
 701         pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse;
 702         pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
 703         pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_sse2;
 704         pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_sse2;
 705         pf->avg[PIXEL_8x8]  = x264_pixel_avg_8x8_sse2;
 706         pf->avg[PIXEL_8x4]  = x264_pixel_avg_8x4_sse2;
 707         pf->hpel_filter = x264_hpel_filter_sse2;
 708         pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
 709         if( !(cpu&X264_CPU_STACK_MOD4) )
 710             pf->mc_chroma = x264_mc_chroma_sse2;
 711
 712         if( cpu&X264_CPU_SSE2_IS_FAST )
 713         {
 714             pf->store_interleave_chroma = x264_store_interleave_chroma_sse2; // FIXME sse2fast? sse2medium?
 715             pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_sse2;
 716             pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_sse2;
 717             pf->plane_copy_interleave   = x264_plane_copy_interleave_sse2;
 718             pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
 719             pf->mc_luma = mc_luma_sse2;
 720             pf->get_ref = get_ref_sse2;
 721             if( cpu&X264_CPU_CACHELINE_64 )
 722             {
 723                 pf->mc_luma = mc_luma_cache64_sse2;
 724                 pf->get_ref = get_ref_cache64_sse2;
 725             }
 726         }
 727     }
 728
 729     if( !(cpu&X264_CPU_SSSE3) )
 730         return;
 731
 732     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_ssse3;
 733     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_ssse3;
 734     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_ssse3;
 735     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_ssse3;
 736     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_ssse3;
 737     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_ssse3;
 738     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_ssse3;
 739     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_ssse3;
 740     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_ssse3;
 741
 742     if( !(cpu&X264_CPU_SLOW_PSHUFB) )
 743     {
 744         pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_ssse3;
 745         pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_ssse3;
 746         pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_ssse3;
 747     }
 748
 749     if( !(cpu&X264_CPU_SLOW_PALIGNR) )
 750     {
 751 #if ARCH_X86_64
 752         if( !(cpu&X264_CPU_SLOW_ATOM) ) /* The 64-bit version is slower, but the 32-bit version is faster? */
 753 #endif
 754             pf->hpel_filter = x264_hpel_filter_ssse3;
 755         pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
 756     }
 757     if( !(cpu&X264_CPU_STACK_MOD4) )
 758         pf->mc_chroma = x264_mc_chroma_ssse3;
 759
 760     if( cpu&X264_CPU_CACHELINE_64 )
 761     {
 762         if( !(cpu&X264_CPU_STACK_MOD4) )
 763             pf->mc_chroma = x264_mc_chroma_ssse3_cache64;
 764         pf->mc_luma = mc_luma_cache64_ssse3;
 765         pf->get_ref = get_ref_cache64_ssse3;
 766         if( cpu&X264_CPU_SLOW_ATOM )
 767         {
 768             pf->mc_luma = mc_luma_cache64_ssse3_atom;
 769             pf->get_ref = get_ref_cache64_ssse3_atom;
 770         }
 771     }
 772
 773     pf->weight_cache = x264_weight_cache_ssse3;
 774     pf->weight = x264_mc_weight_wtab_ssse3;
 775
 776     if( !(cpu&(X264_CPU_SLOW_SHUFFLE|X264_CPU_SLOW_ATOM|X264_CPU_SLOW_PALIGNR)) )
 777         pf->integral_init4v = x264_integral_init4v_ssse3;
 778
 779     if( !(cpu&X264_CPU_SSE4) )
 780         return;
 781
 782     pf->integral_init4h = x264_integral_init4h_sse4;
 783     pf->integral_init8h = x264_integral_init8h_sse4;
 784
 785     if( !(cpu&X264_CPU_AVX) )
 786         return;
 787
 788     pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx;
 789     pf->integral_init8h = x264_integral_init8h_avx;
 790     pf->hpel_filter = x264_hpel_filter_avx;
 791
 792     if( !(cpu&X264_CPU_STACK_MOD4) )
 793         pf->mc_chroma = x264_mc_chroma_avx;
 794
 795     if( cpu&X264_CPU_XOP )
 796         pf->frame_init_lowres_core = x264_frame_init_lowres_core_xop;
 797
 798     if( cpu&X264_CPU_AVX2 )
 799     {
 800         pf->hpel_filter = x264_hpel_filter_avx2;
 801         pf->mc_chroma = x264_mc_chroma_avx2;
 802         pf->weight = x264_mc_weight_wtab_avx2;
 803         pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_avx2;
 804         pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_avx2;
 805         pf->integral_init8v = x264_integral_init8v_avx2;
 806         pf->integral_init4v = x264_integral_init4v_avx2;
 807         pf->integral_init8h = x264_integral_init8h_avx2;
 808         pf->integral_init4h = x264_integral_init4h_avx2;
 809         pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx2;
 810     }
 811 #endif // HIGH_BIT_DEPTH
 812
 813     if( !(cpu&X264_CPU_AVX) )
 814         return;
 815     pf->memzero_aligned = x264_memzero_aligned_avx;
 816     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_avx;
 817
 818     if( cpu&X264_CPU_FMA4 )
 819         pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_fma4;
 820
 821     if( !(cpu&X264_CPU_AVX2) )
 822         return;
 823     pf->get_ref = get_ref_avx2;
 824
 825     if( cpu&X264_CPU_FMA3 )
 826         pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_avx2_fma3;
 827 }