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

   1 /*****************************************************************************
   2  * mc-c.c: x86 motion compensation
   3  *****************************************************************************
   4  * Copyright (C) 2003-2015 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_sse( pixel *, intptr_t, pixel *, intptr_t, int w, int h );
  94 void x264_plane_copy_core_avx( pixel *, intptr_t, pixel *, intptr_t, int w, int h );
  95 void x264_plane_copy_c( pixel *, intptr_t, pixel *, intptr_t, int w, int h );
  96 void x264_plane_copy_interleave_core_mmx2( pixel *dst,  intptr_t i_dst,
  97                                            pixel *srcu, intptr_t i_srcu,
  98                                            pixel *srcv, intptr_t i_srcv, int w, int h );
  99 void x264_plane_copy_interleave_core_sse2( pixel *dst,  intptr_t i_dst,
 100                                            pixel *srcu, intptr_t i_srcu,
 101                                            pixel *srcv, intptr_t i_srcv, int w, int h );
 102 void x264_plane_copy_interleave_core_avx( pixel *dst,  intptr_t i_dst,
 103                                           pixel *srcu, intptr_t i_srcu,
 104                                           pixel *srcv, intptr_t i_srcv, int w, int h );
 105 void x264_plane_copy_interleave_c( pixel *dst,  intptr_t i_dst,
 106                                    pixel *srcu, intptr_t i_srcu,
 107                                    pixel *srcv, intptr_t i_srcv, int w, int h );
 108 void x264_plane_copy_deinterleave_mmx( pixel *dstu, intptr_t i_dstu,
 109                                        pixel *dstv, intptr_t i_dstv,
 110                                        pixel *src,  intptr_t i_src, int w, int h );
 111 void x264_plane_copy_deinterleave_sse2( pixel *dstu, intptr_t i_dstu,
 112                                         pixel *dstv, intptr_t i_dstv,
 113                                         pixel *src,  intptr_t i_src, int w, int h );
 114 void x264_plane_copy_deinterleave_ssse3( uint8_t *dstu, intptr_t i_dstu,
 115                                          uint8_t *dstv, intptr_t i_dstv,
 116                                          uint8_t *src,  intptr_t i_src, int w, int h );
 117 void x264_plane_copy_deinterleave_avx( uint16_t *dstu, intptr_t i_dstu,
 118                                        uint16_t *dstv, intptr_t i_dstv,
 119                                        uint16_t *src,  intptr_t i_src, int w, int h );
 120 void x264_plane_copy_deinterleave_rgb_sse2 ( pixel *dsta, intptr_t i_dsta,
 121                                              pixel *dstb, intptr_t i_dstb,
 122                                              pixel *dstc, intptr_t i_dstc,
 123                                              pixel *src,  intptr_t i_src, int pw, int w, int h );
 124 void x264_plane_copy_deinterleave_rgb_ssse3( pixel *dsta, intptr_t i_dsta,
 125                                              pixel *dstb, intptr_t i_dstb,
 126                                              pixel *dstc, intptr_t i_dstc,
 127                                              pixel *src,  intptr_t i_src, int pw, int w, int h );
 128 void x264_plane_copy_deinterleave_v210_ssse3( uint16_t *dstu, intptr_t i_dstu,
 129                                               uint16_t *dstv, intptr_t i_dstv,
 130                                               uint32_t *src,  intptr_t i_src, int w, int h );
 131 void x264_plane_copy_deinterleave_v210_avx  ( uint16_t *dstu, intptr_t i_dstu,
 132                                               uint16_t *dstv, intptr_t i_dstv,
 133                                               uint32_t *src,  intptr_t i_src, int w, int h );
 134 void x264_plane_copy_deinterleave_v210_avx2 ( uint16_t *dstu, intptr_t i_dstu,
 135                                               uint16_t *dstv, intptr_t i_dstv,
 136                                               uint32_t *src,  intptr_t i_src, int w, int h );
 137 void x264_store_interleave_chroma_mmx2( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
 138 void x264_store_interleave_chroma_sse2( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
 139 void x264_store_interleave_chroma_avx ( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
 140 void x264_load_deinterleave_chroma_fenc_mmx ( pixel *dst, pixel *src, intptr_t i_src, int height );
 141 void x264_load_deinterleave_chroma_fenc_sse2( pixel *dst, pixel *src, intptr_t i_src, int height );
 142 void x264_load_deinterleave_chroma_fenc_ssse3( uint8_t *dst, uint8_t *src, intptr_t i_src, int height );
 143 void x264_load_deinterleave_chroma_fenc_avx( uint16_t *dst, uint16_t *src, intptr_t i_src, int height );
 144 void x264_load_deinterleave_chroma_fdec_mmx ( pixel *dst, pixel *src, intptr_t i_src, int height );
 145 void x264_load_deinterleave_chroma_fdec_sse2( pixel *dst, pixel *src, intptr_t i_src, int height );
 146 void x264_load_deinterleave_chroma_fdec_ssse3( uint8_t *dst, uint8_t *src, intptr_t i_src, int height );
 147 void x264_load_deinterleave_chroma_fdec_avx( uint16_t *dst, uint16_t *src, intptr_t i_src, int height );
 148 void *x264_memcpy_aligned_mmx( void *dst, const void *src, size_t n );
 149 void *x264_memcpy_aligned_sse( void *dst, const void *src, size_t n );
 150 void x264_memzero_aligned_mmx( void *dst, size_t n );
 151 void x264_memzero_aligned_sse( void *dst, size_t n );
 152 void x264_memzero_aligned_avx( void *dst, size_t n );
 153 void x264_integral_init4h_sse4( uint16_t *sum, uint8_t *pix, intptr_t stride );
 154 void x264_integral_init4h_avx2( uint16_t *sum, uint8_t *pix, intptr_t stride );
 155 void x264_integral_init8h_sse4( uint16_t *sum, uint8_t *pix, intptr_t stride );
 156 void x264_integral_init8h_avx ( uint16_t *sum, uint8_t *pix, intptr_t stride );
 157 void x264_integral_init8h_avx2( uint16_t *sum, uint8_t *pix, intptr_t stride );
 158 void x264_integral_init4v_mmx  ( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
 159 void x264_integral_init4v_sse2 ( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
 160 void x264_integral_init4v_ssse3( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
 161 void x264_integral_init4v_avx2( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
 162 void x264_integral_init8v_mmx ( uint16_t *sum8, intptr_t stride );
 163 void x264_integral_init8v_sse2( uint16_t *sum8, intptr_t stride );
 164 void x264_integral_init8v_avx2( uint16_t *sum8, intptr_t stride );
 165 void x264_mbtree_propagate_cost_sse2( int16_t *dst, uint16_t *propagate_in, uint16_t *intra_costs,
 166                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
 167 void x264_mbtree_propagate_cost_avx ( int16_t *dst, uint16_t *propagate_in, uint16_t *intra_costs,
 168                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
 169 void x264_mbtree_propagate_cost_fma4( int16_t *dst, uint16_t *propagate_in, uint16_t *intra_costs,
 170                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
 171 void x264_mbtree_propagate_cost_avx2( int16_t *dst, uint16_t *propagate_in, uint16_t *intra_costs,
 172                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
 173
 174 #define MC_CHROMA(cpu)\
 175 void x264_mc_chroma_##cpu( pixel *dstu, pixel *dstv, intptr_t i_dst, pixel *src, intptr_t i_src,\
 176                            int dx, int dy, int i_width, int i_height );
 177 MC_CHROMA(mmx2)
 178 MC_CHROMA(sse2)
 179 MC_CHROMA(ssse3)
 180 MC_CHROMA(ssse3_cache64)
 181 MC_CHROMA(avx)
 182 MC_CHROMA(avx2)
 183
 184 #define LOWRES(cpu)\
 185 void x264_frame_init_lowres_core_##cpu( pixel *src0, pixel *dst0, pixel *dsth, pixel *dstv, pixel *dstc,\
 186                                         intptr_t src_stride, intptr_t dst_stride, int width, int height );
 187 LOWRES(mmx2)
 188 LOWRES(cache32_mmx2)
 189 LOWRES(sse2)
 190 LOWRES(ssse3)
 191 LOWRES(avx)
 192 LOWRES(xop)
 193 LOWRES(avx2)
 194
 195 #define PIXEL_AVG_W(width,cpu)\
 196 void x264_pixel_avg2_w##width##_##cpu( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t );
 197 /* This declares some functions that don't exist, but that isn't a problem. */
 198 #define PIXEL_AVG_WALL(cpu)\
 199 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);
 200
 201 PIXEL_AVG_WALL(mmx2)
 202 PIXEL_AVG_WALL(cache32_mmx2)
 203 PIXEL_AVG_WALL(cache64_mmx2)
 204 PIXEL_AVG_WALL(cache64_sse2)
 205 PIXEL_AVG_WALL(sse2)
 206 PIXEL_AVG_WALL(cache64_ssse3)
 207 PIXEL_AVG_WALL(avx2)
 208
 209 #define PIXEL_AVG_WTAB(instr, name1, name2, name3, name4, name5)\
 210 static void (* const x264_pixel_avg_wtab_##instr[6])( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t ) =\
 211 {\
 212     NULL,\
 213     x264_pixel_avg2_w4_##name1,\
 214     x264_pixel_avg2_w8_##name2,\
 215     x264_pixel_avg2_w12_##name3,\
 216     x264_pixel_avg2_w16_##name4,\
 217     x264_pixel_avg2_w20_##name5,\
 218 };
 219
 220 #if HIGH_BIT_DEPTH
 221 /* we can replace w12/w20 with w10/w18 as only 9/17 pixels in fact are important */
 222 #define x264_pixel_avg2_w12_mmx2       x264_pixel_avg2_w10_mmx2
 223 #define x264_pixel_avg2_w20_mmx2       x264_pixel_avg2_w18_mmx2
 224 #define x264_pixel_avg2_w12_sse2         x264_pixel_avg2_w10_sse2
 225 #define x264_pixel_avg2_w20_sse2         x264_pixel_avg2_w18_sse2
 226 #define x264_pixel_avg2_w12_avx2         x264_pixel_avg2_w16_avx2
 227 #define x264_pixel_avg2_w20_avx2         x264_pixel_avg2_w18_avx2
 228 #else
 229 /* w16 sse2 is faster than w12 mmx as long as the cacheline issue is resolved */
 230 #define x264_pixel_avg2_w12_cache64_ssse3 x264_pixel_avg2_w16_cache64_ssse3
 231 #define x264_pixel_avg2_w12_cache64_sse2 x264_pixel_avg2_w16_cache64_sse2
 232 #define x264_pixel_avg2_w12_sse3         x264_pixel_avg2_w16_sse3
 233 #define x264_pixel_avg2_w12_sse2         x264_pixel_avg2_w16_sse2
 234 #endif // HIGH_BIT_DEPTH
 235
 236 PIXEL_AVG_WTAB(mmx2, mmx2, mmx2, mmx2, mmx2, mmx2)
 237 #if HIGH_BIT_DEPTH
 238 PIXEL_AVG_WTAB(sse2, mmx2, sse2, sse2, sse2, sse2)
 239 PIXEL_AVG_WTAB(avx2, mmx2, sse2, avx2, avx2, avx2)
 240 #else // !HIGH_BIT_DEPTH
 241 #if ARCH_X86
 242 PIXEL_AVG_WTAB(cache32_mmx2, mmx2, cache32_mmx2, cache32_mmx2, cache32_mmx2, cache32_mmx2)
 243 PIXEL_AVG_WTAB(cache64_mmx2, mmx2, cache64_mmx2, cache64_mmx2, cache64_mmx2, cache64_mmx2)
 244 #endif
 245 PIXEL_AVG_WTAB(sse2, mmx2, mmx2, sse2, sse2, sse2)
 246 PIXEL_AVG_WTAB(cache64_sse2, mmx2, cache64_mmx2, cache64_sse2, cache64_sse2, cache64_sse2)
 247 PIXEL_AVG_WTAB(cache64_ssse3, mmx2, cache64_mmx2, cache64_ssse3, cache64_ssse3, cache64_sse2)
 248 PIXEL_AVG_WTAB(cache64_ssse3_atom, mmx2, mmx2, cache64_ssse3, cache64_ssse3, sse2)
 249 PIXEL_AVG_WTAB(avx2, mmx2, mmx2, sse2, sse2, avx2)
 250 #endif // HIGH_BIT_DEPTH
 251
 252 #define MC_COPY_WTAB(instr, name1, name2, name3)\
 253 static void (* const x264_mc_copy_wtab_##instr[5])( pixel *, intptr_t, pixel *, intptr_t, int ) =\
 254 {\
 255     NULL,\
 256     x264_mc_copy_w4_##name1,\
 257     x264_mc_copy_w8_##name2,\
 258     NULL,\
 259     x264_mc_copy_w16_##name3,\
 260 };
 261
 262 MC_COPY_WTAB(mmx,mmx,mmx,mmx)
 263 #if HIGH_BIT_DEPTH
 264 MC_COPY_WTAB(sse,mmx,sse,sse)
 265 MC_COPY_WTAB(avx,mmx,sse,avx)
 266 #else
 267 MC_COPY_WTAB(sse,mmx,mmx,sse)
 268 #endif
 269
 270 #define MC_WEIGHT_WTAB(function, instr, name1, name2, w12version)\
 271     static void (* x264_mc_##function##_wtab_##instr[6])( pixel *, intptr_t, pixel *, intptr_t, const x264_weight_t *, int ) =\
 272 {\
 273     x264_mc_##function##_w4_##name1,\
 274     x264_mc_##function##_w4_##name1,\
 275     x264_mc_##function##_w8_##name2,\
 276     x264_mc_##function##_w##w12version##_##instr,\
 277     x264_mc_##function##_w16_##instr,\
 278     x264_mc_##function##_w20_##instr,\
 279 };
 280
 281 #if HIGH_BIT_DEPTH
 282 MC_WEIGHT_WTAB(weight,mmx2,mmx2,mmx2,12)
 283 MC_WEIGHT_WTAB(offsetadd,mmx2,mmx2,mmx2,12)
 284 MC_WEIGHT_WTAB(offsetsub,mmx2,mmx2,mmx2,12)
 285 MC_WEIGHT_WTAB(weight,sse2,mmx2,sse2,12)
 286 MC_WEIGHT_WTAB(offsetadd,sse2,mmx2,sse2,16)
 287 MC_WEIGHT_WTAB(offsetsub,sse2,mmx2,sse2,16)
 288
 289 static void x264_weight_cache_mmx2( x264_t *h, x264_weight_t *w )
 290 {
 291     if( w->i_scale == 1<<w->i_denom )
 292     {
 293         if( w->i_offset < 0 )
 294             w->weightfn = h->mc.offsetsub;
 295         else
 296             w->weightfn = h->mc.offsetadd;
 297         for( int i = 0; i < 8; i++ )
 298             w->cachea[i] = abs(w->i_offset<<(BIT_DEPTH-8));
 299         return;
 300     }
 301     w->weightfn = h->mc.weight;
 302     int den1 = 1<<w->i_denom;
 303     int den2 = w->i_scale<<1;
 304     int den3 = 1+(w->i_offset<<(BIT_DEPTH-8+1));
 305     for( int i = 0; i < 8; i++ )
 306     {
 307         w->cachea[i] = den1;
 308         w->cacheb[i] = i&1 ? den3 : den2;
 309     }
 310 }
 311 #else
 312 MC_WEIGHT_WTAB(weight,mmx2,mmx2,mmx2,12)
 313 MC_WEIGHT_WTAB(offsetadd,mmx2,mmx2,mmx2,12)
 314 MC_WEIGHT_WTAB(offsetsub,mmx2,mmx2,mmx2,12)
 315 MC_WEIGHT_WTAB(weight,sse2,mmx2,sse2,16)
 316 MC_WEIGHT_WTAB(offsetadd,sse2,mmx2,mmx2,16)
 317 MC_WEIGHT_WTAB(offsetsub,sse2,mmx2,mmx2,16)
 318 MC_WEIGHT_WTAB(weight,ssse3,ssse3,ssse3,16)
 319 MC_WEIGHT_WTAB(weight,avx2,ssse3,avx2,16)
 320
 321 static void x264_weight_cache_mmx2( x264_t *h, x264_weight_t *w )
 322 {
 323     int i;
 324     int16_t den1;
 325
 326     if( w->i_scale == 1<<w->i_denom )
 327     {
 328         if( w->i_offset < 0 )
 329             w->weightfn = h->mc.offsetsub;
 330         else
 331             w->weightfn = h->mc.offsetadd;
 332         memset( w->cachea, abs(w->i_offset), sizeof(w->cachea) );
 333         return;
 334     }
 335     w->weightfn = h->mc.weight;
 336     den1 = 1 << (w->i_denom - 1) | w->i_offset << w->i_denom;
 337     for( i = 0; i < 8; i++ )
 338     {
 339         w->cachea[i] = w->i_scale;
 340         w->cacheb[i] = den1;
 341     }
 342 }
 343
 344 static void x264_weight_cache_ssse3( x264_t *h, x264_weight_t *w )
 345 {
 346     int i, den1;
 347     if( w->i_scale == 1<<w->i_denom )
 348     {
 349         if( w->i_offset < 0 )
 350             w->weightfn = h->mc.offsetsub;
 351         else
 352             w->weightfn = h->mc.offsetadd;
 353
 354         memset( w->cachea, abs( w->i_offset ), sizeof(w->cachea) );
 355         return;
 356     }
 357     w->weightfn = h->mc.weight;
 358     den1 = w->i_scale << (8 - w->i_denom);
 359     for( i = 0; i < 8; i++ )
 360     {
 361         w->cachea[i] = den1;
 362         w->cacheb[i] = w->i_offset;
 363     }
 364 }
 365 #endif // !HIGH_BIT_DEPTH
 366
 367 #define MC_LUMA(name,instr1,instr2)\
 368 static void mc_luma_##name( pixel *dst,    intptr_t i_dst_stride,\
 369                             pixel *src[4], intptr_t i_src_stride,\
 370                             int mvx, int mvy,\
 371                             int i_width, int i_height, const x264_weight_t *weight )\
 372 {\
 373     int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
 374     int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
 375     pixel *src1 = src[x264_hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
 376     if( qpel_idx & 5 ) /* qpel interpolation needed */\
 377     {\
 378         pixel *src2 = src[x264_hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
 379         x264_pixel_avg_wtab_##instr1[i_width>>2](\
 380                 dst, i_dst_stride, src1, i_src_stride,\
 381                 src2, i_height );\
 382         if( weight->weightfn )\
 383             weight->weightfn[i_width>>2]( dst, i_dst_stride, dst, i_dst_stride, weight, i_height );\
 384     }\
 385     else if( weight->weightfn )\
 386         weight->weightfn[i_width>>2]( dst, i_dst_stride, src1, i_src_stride, weight, i_height );\
 387     else\
 388         x264_mc_copy_wtab_##instr2[i_width>>2](dst, i_dst_stride, src1, i_src_stride, i_height );\
 389 }
 390
 391 MC_LUMA(mmx2,mmx2,mmx)
 392 MC_LUMA(sse2,sse2,sse)
 393 #if HIGH_BIT_DEPTH
 394 MC_LUMA(avx2,avx2,avx)
 395 #else
 396 #if ARCH_X86
 397 MC_LUMA(cache32_mmx2,cache32_mmx2,mmx)
 398 MC_LUMA(cache64_mmx2,cache64_mmx2,mmx)
 399 #endif
 400 MC_LUMA(cache64_sse2,cache64_sse2,sse)
 401 MC_LUMA(cache64_ssse3,cache64_ssse3,sse)
 402 MC_LUMA(cache64_ssse3_atom,cache64_ssse3_atom,sse)
 403 #endif // !HIGH_BIT_DEPTH
 404
 405 #define GET_REF(name)\
 406 static pixel *get_ref_##name( pixel *dst,   intptr_t *i_dst_stride,\
 407                               pixel *src[4], intptr_t i_src_stride,\
 408                               int mvx, int mvy,\
 409                               int i_width, int i_height, const x264_weight_t *weight )\
 410 {\
 411     int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
 412     int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
 413     pixel *src1 = src[x264_hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
 414     if( qpel_idx & 5 ) /* qpel interpolation needed */\
 415     {\
 416         pixel *src2 = src[x264_hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
 417         x264_pixel_avg_wtab_##name[i_width>>2](\
 418                 dst, *i_dst_stride, src1, i_src_stride,\
 419                 src2, i_height );\
 420         if( weight->weightfn )\
 421             weight->weightfn[i_width>>2]( dst, *i_dst_stride, dst, *i_dst_stride, weight, i_height );\
 422         return dst;\
 423     }\
 424     else if( weight->weightfn )\
 425     {\
 426         weight->weightfn[i_width>>2]( dst, *i_dst_stride, src1, i_src_stride, weight, i_height );\
 427         return dst;\
 428     }\
 429     else\
 430     {\
 431         *i_dst_stride = i_src_stride;\
 432         return src1;\
 433     }\
 434 }
 435
 436 GET_REF(mmx2)
 437 GET_REF(sse2)
 438 GET_REF(avx2)
 439 #if !HIGH_BIT_DEPTH
 440 #if ARCH_X86
 441 GET_REF(cache32_mmx2)
 442 GET_REF(cache64_mmx2)
 443 #endif
 444 GET_REF(cache64_sse2)
 445 GET_REF(cache64_ssse3)
 446 GET_REF(cache64_ssse3_atom)
 447 #endif // !HIGH_BIT_DEPTH
 448
 449 #define HPEL(align, cpu, cpuv, cpuc, cpuh)\
 450 void x264_hpel_filter_v_##cpuv( pixel *dst, pixel *src, int16_t *buf, intptr_t stride, intptr_t width);\
 451 void x264_hpel_filter_c_##cpuc( pixel *dst, int16_t *buf, intptr_t width );\
 452 void x264_hpel_filter_h_##cpuh( pixel *dst, pixel *src, intptr_t width );\
 453 static void x264_hpel_filter_##cpu( pixel *dsth, pixel *dstv, pixel *dstc, pixel *src,\
 454                                     intptr_t stride, int width, int height, int16_t *buf )\
 455 {\
 456     intptr_t realign = (intptr_t)src & (align-1);\
 457     src -= realign;\
 458     dstv -= realign;\
 459     dstc -= realign;\
 460     dsth -= realign;\
 461     width += realign;\
 462     while( height-- )\
 463     {\
 464         x264_hpel_filter_v_##cpuv( dstv, src, buf+16, stride, width );\
 465         x264_hpel_filter_c_##cpuc( dstc, buf+16, width );\
 466         x264_hpel_filter_h_##cpuh( dsth, src, width );\
 467         dsth += stride;\
 468         dstv += stride;\
 469         dstc += stride;\
 470         src  += stride;\
 471     }\
 472     x264_sfence();\
 473 }
 474
 475 HPEL(8, mmx2, mmx2, mmx2, mmx2)
 476 #if HIGH_BIT_DEPTH
 477 HPEL(16, sse2, sse2, sse2, sse2)
 478 #else // !HIGH_BIT_DEPTH
 479 HPEL(16, sse2_amd, mmx2, mmx2, sse2)
 480 #if ARCH_X86_64
 481 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 );
 482 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 );
 483 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 );
 484 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 );
 485 #else
 486 HPEL(16, sse2, sse2, sse2, sse2)
 487 HPEL(16, ssse3, ssse3, ssse3, ssse3)
 488 HPEL(16, avx, avx, avx, avx)
 489 HPEL(32, avx2, avx2, avx2, avx2)
 490 #endif
 491 #endif // HIGH_BIT_DEPTH
 492
 493 #define PLANE_COPY(align, cpu)\
 494 static void x264_plane_copy_##cpu( pixel *dst, intptr_t i_dst, pixel *src, intptr_t i_src, int w, int h )\
 495 {\
 496     int c_w = (align) / sizeof(pixel) - 1;\
 497     if( w < 256 ) /* tiny resolutions don't want non-temporal hints. dunno the exact threshold. */\
 498         x264_plane_copy_c( dst, i_dst, src, i_src, w, h );\
 499     else if( !(w&c_w) )\
 500         x264_plane_copy_core_##cpu( dst, i_dst, src, i_src, w, h );\
 501     else\
 502     {\
 503         if( --h > 0 )\
 504         {\
 505             if( i_src > 0 )\
 506             {\
 507                 x264_plane_copy_core_##cpu( dst, i_dst, src, i_src, (w+c_w)&~c_w, h );\
 508                 dst += i_dst * h;\
 509                 src += i_src * h;\
 510             }\
 511             else\
 512                 x264_plane_copy_core_##cpu( dst+i_dst, i_dst, src+i_src, i_src, (w+c_w)&~c_w, h );\
 513         }\
 514         /* use plain memcpy on the last line (in memory order) to avoid overreading src. */\
 515         memcpy( dst, src, w*sizeof(pixel) );\
 516     }\
 517 }
 518
 519 PLANE_COPY(16, sse)
 520 PLANE_COPY(32, avx)
 521
 522 #define PLANE_INTERLEAVE(cpu) \
 523 static void x264_plane_copy_interleave_##cpu( pixel *dst,  intptr_t i_dst,\
 524                                               pixel *srcu, intptr_t i_srcu,\
 525                                               pixel *srcv, intptr_t i_srcv, int w, int h )\
 526 {\
 527     int c_w = 16 / sizeof(pixel) - 1;\
 528     if( !(w&c_w) )\
 529         x264_plane_copy_interleave_core_##cpu( dst, i_dst, srcu, i_srcu, srcv, i_srcv, w, h );\
 530     else if( w > c_w && (i_srcu ^ i_srcv) >= 0 ) /* only works correctly for strides with identical signs */\
 531     {\
 532         if( --h > 0 )\
 533         {\
 534             if( i_srcu > 0 )\
 535             {\
 536                 x264_plane_copy_interleave_core_##cpu( dst, i_dst, srcu, i_srcu, srcv, i_srcv, (w+c_w)&~c_w, h );\
 537                 dst  += i_dst  * h;\
 538                 srcu += i_srcu * h;\
 539                 srcv += i_srcv * h;\
 540             }\
 541             else\
 542                 x264_plane_copy_interleave_core_##cpu( dst+i_dst, i_dst, srcu+i_srcu, i_srcu, srcv+i_srcv, i_srcv, (w+c_w)&~c_w, h );\
 543         }\
 544         x264_plane_copy_interleave_c( dst, 0, srcu, 0, srcv, 0, w, 1 );\
 545     }\
 546     else\
 547         x264_plane_copy_interleave_c( dst, i_dst, srcu, i_srcu, srcv, i_srcv, w, h );\
 548 }
 549
 550 PLANE_INTERLEAVE(mmx2)
 551 PLANE_INTERLEAVE(sse2)
 552 #if HIGH_BIT_DEPTH
 553 PLANE_INTERLEAVE(avx)
 554 #endif
 555
 556 #if HAVE_X86_INLINE_ASM
 557 #define CLIP_ADD(s,x)\
 558 do\
 559 {\
 560     int temp;\
 561     asm("movd       %0, %%xmm0     \n"\
 562         "movd       %2, %%xmm1     \n"\
 563         "paddsw %%xmm1, %%xmm0     \n"\
 564         "movd   %%xmm0, %1         \n"\
 565         :"+m"(s), "=&r"(temp)\
 566         :"m"(x)\
 567     );\
 568     s = temp;\
 569 } while(0)
 570
 571 #define CLIP_ADD2(s,x)\
 572 do\
 573 {\
 574     asm("movd       %0, %%xmm0     \n"\
 575         "movd       %1, %%xmm1     \n"\
 576         "paddsw %%xmm1, %%xmm0     \n"\
 577         "movd   %%xmm0, %0         \n"\
 578         :"+m"(M32(s))\
 579         :"m"(M32(x))\
 580     );\
 581 } while(0)
 582 #else
 583 #define CLIP_ADD(s,x) (s) = X264_MIN((s)+(x),(1<<15)-1)
 584 #define CLIP_ADD2(s,x)\
 585 do\
 586 {\
 587     CLIP_ADD((s)[0], (x)[0]);\
 588     CLIP_ADD((s)[1], (x)[1]);\
 589 } while(0)
 590 #endif
 591
 592 #define PROPAGATE_LIST(cpu)\
 593 void x264_mbtree_propagate_list_internal_##cpu( int16_t (*mvs)[2], int16_t *propagate_amount,\
 594                                                 uint16_t *lowres_costs, int16_t *output,\
 595                                                 int bipred_weight, int mb_y, int len );\
 596 \
 597 static void x264_mbtree_propagate_list_##cpu( x264_t *h, uint16_t *ref_costs, int16_t (*mvs)[2],\
 598                                               int16_t *propagate_amount, uint16_t *lowres_costs,\
 599                                               int bipred_weight, int mb_y, int len, int list )\
 600 {\
 601     int16_t *current = h->scratch_buffer2;\
 602 \
 603     x264_mbtree_propagate_list_internal_##cpu( mvs, propagate_amount, lowres_costs,\
 604                                                current, bipred_weight, mb_y, len );\
 605 \
 606     unsigned stride = h->mb.i_mb_stride;\
 607     unsigned width = h->mb.i_mb_width;\
 608     unsigned height = h->mb.i_mb_height;\
 609 \
 610     for( unsigned i = 0; i < len; current += 32 )\
 611     {\
 612         int end = X264_MIN( i+8, len );\
 613         for( ; i < end; i++, current += 2 )\
 614         {\
 615             if( !(lowres_costs[i] & (1 << (list+LOWRES_COST_SHIFT))) )\
 616                 continue;\
 617 \
 618             unsigned mbx = current[0];\
 619             unsigned mby = current[1];\
 620             unsigned idx0 = mbx + mby * stride;\
 621             unsigned idx2 = idx0 + stride;\
 622 \
 623             /* Shortcut for the simple/common case of zero MV */\
 624             if( !M32( mvs[i] ) )\
 625             {\
 626                 CLIP_ADD( ref_costs[idx0], current[16] );\
 627                 continue;\
 628             }\
 629 \
 630             if( mbx < width-1 && mby < height-1 )\
 631             {\
 632                 CLIP_ADD2( ref_costs+idx0, current+16 );\
 633                 CLIP_ADD2( ref_costs+idx2, current+32 );\
 634             }\
 635             else\
 636             {\
 637                 /* Note: this takes advantage of unsigned representation to\
 638                  * catch negative mbx/mby. */\
 639                 if( mby < height )\
 640                 {\
 641                     if( mbx < width )\
 642                         CLIP_ADD( ref_costs[idx0+0], current[16] );\
 643                     if( mbx+1 < width )\
 644                         CLIP_ADD( ref_costs[idx0+1], current[17] );\
 645                 }\
 646                 if( mby+1 < height )\
 647                 {\
 648                     if( mbx < width )\
 649                         CLIP_ADD( ref_costs[idx2+0], current[32] );\
 650                     if( mbx+1 < width )\
 651                         CLIP_ADD( ref_costs[idx2+1], current[33] );\
 652                 }\
 653             }\
 654         }\
 655     }\
 656 }
 657
 658 PROPAGATE_LIST(ssse3)
 659 PROPAGATE_LIST(avx)
 660 #undef CLIP_ADD
 661 #undef CLIP_ADD2
 662
 663 void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
 664 {
 665     if( !(cpu&X264_CPU_MMX) )
 666         return;
 667
 668     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_mmx;
 669     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_mmx;
 670
 671     pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_mmx;
 672
 673     pf->copy_16x16_unaligned = x264_mc_copy_w16_mmx;
 674     pf->copy[PIXEL_16x16] = x264_mc_copy_w16_mmx;
 675     pf->copy[PIXEL_8x8]   = x264_mc_copy_w8_mmx;
 676     pf->copy[PIXEL_4x4]   = x264_mc_copy_w4_mmx;
 677     pf->memcpy_aligned  = x264_memcpy_aligned_mmx;
 678     pf->memzero_aligned = x264_memzero_aligned_mmx;
 679     pf->integral_init4v = x264_integral_init4v_mmx;
 680     pf->integral_init8v = x264_integral_init8v_mmx;
 681
 682     if( !(cpu&X264_CPU_MMX2) )
 683         return;
 684
 685     pf->prefetch_fenc_420 = x264_prefetch_fenc_420_mmx2;
 686     pf->prefetch_fenc_422 = x264_prefetch_fenc_422_mmx2;
 687     pf->prefetch_ref  = x264_prefetch_ref_mmx2;
 688
 689     pf->plane_copy_interleave = x264_plane_copy_interleave_mmx2;
 690     pf->store_interleave_chroma = x264_store_interleave_chroma_mmx2;
 691
 692     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_mmx2;
 693     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_mmx2;
 694     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_mmx2;
 695     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_mmx2;
 696     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_mmx2;
 697     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_mmx2;
 698     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_mmx2;
 699     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_mmx2;
 700     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_mmx2;
 701
 702     pf->mc_luma = mc_luma_mmx2;
 703     pf->get_ref = get_ref_mmx2;
 704     pf->mc_chroma = x264_mc_chroma_mmx2;
 705     pf->hpel_filter = x264_hpel_filter_mmx2;
 706     pf->weight = x264_mc_weight_wtab_mmx2;
 707     pf->weight_cache = x264_weight_cache_mmx2;
 708     pf->offsetadd = x264_mc_offsetadd_wtab_mmx2;
 709     pf->offsetsub = x264_mc_offsetsub_wtab_mmx2;
 710
 711     pf->frame_init_lowres_core = x264_frame_init_lowres_core_mmx2;
 712
 713     if( cpu&X264_CPU_SSE )
 714     {
 715         pf->memcpy_aligned  = x264_memcpy_aligned_sse;
 716         pf->memzero_aligned = x264_memzero_aligned_sse;
 717         pf->plane_copy = x264_plane_copy_sse;
 718     }
 719
 720 #if HIGH_BIT_DEPTH
 721 #if ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
 722     if( cpu&(X264_CPU_CACHELINE_32|X264_CPU_CACHELINE_64) )
 723         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
 724 #endif
 725
 726     if( !(cpu&X264_CPU_SSE2) )
 727         return;
 728
 729     pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
 730
 731     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_sse2;
 732     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_sse2;
 733
 734     pf->plane_copy_interleave   = x264_plane_copy_interleave_sse2;
 735     pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
 736
 737     if( cpu&X264_CPU_SSE2_IS_FAST )
 738     {
 739         pf->get_ref = get_ref_sse2;
 740         pf->mc_luma = mc_luma_sse2;
 741         pf->hpel_filter = x264_hpel_filter_sse2;
 742     }
 743
 744     pf->integral_init4v = x264_integral_init4v_sse2;
 745     pf->integral_init8v = x264_integral_init8v_sse2;
 746     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
 747     pf->store_interleave_chroma = x264_store_interleave_chroma_sse2;
 748     pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
 749     pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
 750
 751     if( cpu&X264_CPU_SSE2_IS_SLOW )
 752         return;
 753
 754     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
 755     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_sse2;
 756     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_sse2;
 757     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_sse2;
 758     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_sse2;
 759     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_sse2;
 760     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_sse2;
 761     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_sse2;
 762     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_sse2;
 763
 764     pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse;
 765     pf->weight = x264_mc_weight_wtab_sse2;
 766
 767     if( !(cpu&X264_CPU_STACK_MOD4) )
 768         pf->mc_chroma = x264_mc_chroma_sse2;
 769
 770     if( !(cpu&X264_CPU_SSSE3) )
 771         return;
 772
 773     pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
 774     pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_ssse3;
 775     pf->mbtree_propagate_list = x264_mbtree_propagate_list_ssse3;
 776
 777     if( !(cpu&(X264_CPU_SLOW_SHUFFLE|X264_CPU_SLOW_ATOM|X264_CPU_SLOW_PALIGNR)) )
 778         pf->integral_init4v = x264_integral_init4v_ssse3;
 779
 780     if( !(cpu&X264_CPU_AVX) )
 781         return;
 782
 783     pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx;
 784     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_avx;
 785     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_avx;
 786     pf->plane_copy_interleave        = x264_plane_copy_interleave_avx;
 787     pf->plane_copy_deinterleave      = x264_plane_copy_deinterleave_avx;
 788     pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_avx;
 789     pf->store_interleave_chroma      = x264_store_interleave_chroma_avx;
 790     pf->copy[PIXEL_16x16]            = x264_mc_copy_w16_aligned_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->mc_luma = mc_luma_avx2;
 801         pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_avx2;
 802     }
 803 #else // !HIGH_BIT_DEPTH
 804
 805 #if ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
 806     if( cpu&X264_CPU_CACHELINE_32 )
 807     {
 808         pf->mc_luma = mc_luma_cache32_mmx2;
 809         pf->get_ref = get_ref_cache32_mmx2;
 810         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
 811     }
 812     else if( cpu&X264_CPU_CACHELINE_64 )
 813     {
 814         pf->mc_luma = mc_luma_cache64_mmx2;
 815         pf->get_ref = get_ref_cache64_mmx2;
 816         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
 817     }
 818 #endif
 819
 820     if( !(cpu&X264_CPU_SSE2) )
 821         return;
 822
 823     pf->integral_init4v = x264_integral_init4v_sse2;
 824     pf->integral_init8v = x264_integral_init8v_sse2;
 825     pf->hpel_filter = x264_hpel_filter_sse2_amd;
 826     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
 827     pf->plane_copy_deinterleave_rgb = x264_plane_copy_deinterleave_rgb_sse2;
 828
 829     if( !(cpu&X264_CPU_SSE2_IS_SLOW) )
 830     {
 831         pf->weight = x264_mc_weight_wtab_sse2;
 832         if( !(cpu&X264_CPU_SLOW_ATOM) )
 833         {
 834             pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
 835             pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
 836         }
 837
 838         pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse;
 839         pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
 840         pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_sse2;
 841         pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_sse2;
 842         pf->avg[PIXEL_8x8]  = x264_pixel_avg_8x8_sse2;
 843         pf->avg[PIXEL_8x4]  = x264_pixel_avg_8x4_sse2;
 844         pf->hpel_filter = x264_hpel_filter_sse2;
 845         pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
 846         if( !(cpu&X264_CPU_STACK_MOD4) )
 847             pf->mc_chroma = x264_mc_chroma_sse2;
 848
 849         if( cpu&X264_CPU_SSE2_IS_FAST )
 850         {
 851             pf->store_interleave_chroma = x264_store_interleave_chroma_sse2; // FIXME sse2fast? sse2medium?
 852             pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_sse2;
 853             pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_sse2;
 854             pf->plane_copy_interleave   = x264_plane_copy_interleave_sse2;
 855             pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
 856             pf->mc_luma = mc_luma_sse2;
 857             pf->get_ref = get_ref_sse2;
 858             if( cpu&X264_CPU_CACHELINE_64 )
 859             {
 860                 pf->mc_luma = mc_luma_cache64_sse2;
 861                 pf->get_ref = get_ref_cache64_sse2;
 862             }
 863         }
 864     }
 865
 866     if( !(cpu&X264_CPU_SSSE3) )
 867         return;
 868
 869     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_ssse3;
 870     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_ssse3;
 871     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_ssse3;
 872     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_ssse3;
 873     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_ssse3;
 874     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_ssse3;
 875     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_ssse3;
 876     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_ssse3;
 877     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_ssse3;
 878     pf->plane_copy_deinterleave_rgb = x264_plane_copy_deinterleave_rgb_ssse3;
 879     pf->mbtree_propagate_list = x264_mbtree_propagate_list_ssse3;
 880
 881     if( !(cpu&X264_CPU_SLOW_PSHUFB) )
 882     {
 883         pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_ssse3;
 884         pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_ssse3;
 885         pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_ssse3;
 886     }
 887
 888     if( !(cpu&X264_CPU_SLOW_PALIGNR) )
 889     {
 890 #if ARCH_X86_64
 891         if( !(cpu&X264_CPU_SLOW_ATOM) ) /* The 64-bit version is slower, but the 32-bit version is faster? */
 892 #endif
 893             pf->hpel_filter = x264_hpel_filter_ssse3;
 894         pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
 895     }
 896     if( !(cpu&X264_CPU_STACK_MOD4) )
 897         pf->mc_chroma = x264_mc_chroma_ssse3;
 898
 899     if( cpu&X264_CPU_CACHELINE_64 )
 900     {
 901         if( !(cpu&X264_CPU_STACK_MOD4) )
 902             pf->mc_chroma = x264_mc_chroma_ssse3_cache64;
 903         pf->mc_luma = mc_luma_cache64_ssse3;
 904         pf->get_ref = get_ref_cache64_ssse3;
 905         if( cpu&X264_CPU_SLOW_ATOM )
 906         {
 907             pf->mc_luma = mc_luma_cache64_ssse3_atom;
 908             pf->get_ref = get_ref_cache64_ssse3_atom;
 909         }
 910     }
 911
 912     pf->weight_cache = x264_weight_cache_ssse3;
 913     pf->weight = x264_mc_weight_wtab_ssse3;
 914
 915     if( !(cpu&(X264_CPU_SLOW_SHUFFLE|X264_CPU_SLOW_ATOM|X264_CPU_SLOW_PALIGNR)) )
 916         pf->integral_init4v = x264_integral_init4v_ssse3;
 917
 918     if( !(cpu&X264_CPU_SSE4) )
 919         return;
 920
 921     pf->integral_init4h = x264_integral_init4h_sse4;
 922     pf->integral_init8h = x264_integral_init8h_sse4;
 923
 924     if( !(cpu&X264_CPU_AVX) )
 925         return;
 926
 927     pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx;
 928     pf->integral_init8h = x264_integral_init8h_avx;
 929     pf->hpel_filter = x264_hpel_filter_avx;
 930
 931     if( !(cpu&X264_CPU_STACK_MOD4) )
 932         pf->mc_chroma = x264_mc_chroma_avx;
 933
 934     if( cpu&X264_CPU_XOP )
 935         pf->frame_init_lowres_core = x264_frame_init_lowres_core_xop;
 936
 937     if( cpu&X264_CPU_AVX2 )
 938     {
 939         pf->hpel_filter = x264_hpel_filter_avx2;
 940         pf->mc_chroma = x264_mc_chroma_avx2;
 941         pf->weight = x264_mc_weight_wtab_avx2;
 942         pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_avx2;
 943         pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_avx2;
 944         pf->integral_init8v = x264_integral_init8v_avx2;
 945         pf->integral_init4v = x264_integral_init4v_avx2;
 946         pf->integral_init8h = x264_integral_init8h_avx2;
 947         pf->integral_init4h = x264_integral_init4h_avx2;
 948         pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx2;
 949     }
 950 #endif // HIGH_BIT_DEPTH
 951
 952     if( !(cpu&X264_CPU_AVX) )
 953         return;
 954     pf->memzero_aligned = x264_memzero_aligned_avx;
 955     pf->plane_copy = x264_plane_copy_avx;
 956     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_avx;
 957     pf->mbtree_propagate_list = x264_mbtree_propagate_list_avx;
 958
 959     if( cpu&X264_CPU_FMA4 )
 960         pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_fma4;
 961
 962     if( !(cpu&X264_CPU_AVX2) )
 963         return;
 964     pf->get_ref = get_ref_avx2;
 965     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_avx2;
 966 }