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

   1 /*****************************************************************************
   2  * mc-c.c: x86 motion compensation
   3  *****************************************************************************
   4  * Copyright (C) 2003-2016 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_swap_core_ssse3( pixel *, intptr_t, pixel *, intptr_t, int w, int h );
  97 void x264_plane_copy_swap_core_avx2 ( pixel *, intptr_t, pixel *, intptr_t, int w, int h );
  98 void x264_plane_copy_swap_c( pixel *, intptr_t, pixel *, intptr_t, int w, int h );
  99 void x264_plane_copy_interleave_core_mmx2( 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_sse2( 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_core_avx( 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_interleave_c( pixel *dst,  intptr_t i_dst,
 109                                    pixel *srcu, intptr_t i_srcu,
 110                                    pixel *srcv, intptr_t i_srcv, int w, int h );
 111 void x264_plane_copy_deinterleave_mmx( 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_sse2( pixel *dstu, intptr_t i_dstu,
 115                                         pixel *dstv, intptr_t i_dstv,
 116                                         pixel *src,  intptr_t i_src, int w, int h );
 117 void x264_plane_copy_deinterleave_ssse3( uint8_t *dstu, intptr_t i_dstu,
 118                                          uint8_t *dstv, intptr_t i_dstv,
 119                                          uint8_t *src,  intptr_t i_src, int w, int h );
 120 void x264_plane_copy_deinterleave_avx( uint16_t *dstu, intptr_t i_dstu,
 121                                        uint16_t *dstv, intptr_t i_dstv,
 122                                        uint16_t *src,  intptr_t i_src, int w, int h );
 123 void x264_plane_copy_deinterleave_rgb_sse2 ( pixel *dsta, intptr_t i_dsta,
 124                                              pixel *dstb, intptr_t i_dstb,
 125                                              pixel *dstc, intptr_t i_dstc,
 126                                              pixel *src,  intptr_t i_src, int pw, int w, int h );
 127 void x264_plane_copy_deinterleave_rgb_ssse3( pixel *dsta, intptr_t i_dsta,
 128                                              pixel *dstb, intptr_t i_dstb,
 129                                              pixel *dstc, intptr_t i_dstc,
 130                                              pixel *src,  intptr_t i_src, int pw, int w, int h );
 131 void x264_plane_copy_deinterleave_v210_ssse3( 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_avx  ( 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_plane_copy_deinterleave_v210_avx2 ( uint16_t *dstu, intptr_t i_dstu,
 138                                               uint16_t *dstv, intptr_t i_dstv,
 139                                               uint32_t *src,  intptr_t i_src, int w, int h );
 140 void x264_store_interleave_chroma_mmx2( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
 141 void x264_store_interleave_chroma_sse2( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
 142 void x264_store_interleave_chroma_avx ( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
 143 void x264_load_deinterleave_chroma_fenc_mmx ( pixel *dst, pixel *src, intptr_t i_src, int height );
 144 void x264_load_deinterleave_chroma_fenc_sse2( pixel *dst, pixel *src, intptr_t i_src, int height );
 145 void x264_load_deinterleave_chroma_fenc_ssse3( uint8_t *dst, uint8_t *src, intptr_t i_src, int height );
 146 void x264_load_deinterleave_chroma_fenc_avx( uint16_t *dst, uint16_t *src, intptr_t i_src, int height );
 147 void x264_load_deinterleave_chroma_fdec_mmx ( pixel *dst, pixel *src, intptr_t i_src, int height );
 148 void x264_load_deinterleave_chroma_fdec_sse2( pixel *dst, pixel *src, intptr_t i_src, int height );
 149 void x264_load_deinterleave_chroma_fdec_ssse3( uint8_t *dst, uint8_t *src, intptr_t i_src, int height );
 150 void x264_load_deinterleave_chroma_fdec_avx( uint16_t *dst, uint16_t *src, intptr_t i_src, int height );
 151 void *x264_memcpy_aligned_mmx( void *dst, const void *src, size_t n );
 152 void *x264_memcpy_aligned_sse( void *dst, const void *src, size_t n );
 153 void x264_memzero_aligned_mmx( void *dst, size_t n );
 154 void x264_memzero_aligned_sse( void *dst, size_t n );
 155 void x264_memzero_aligned_avx( void *dst, size_t n );
 156 void x264_integral_init4h_sse4( uint16_t *sum, uint8_t *pix, intptr_t stride );
 157 void x264_integral_init4h_avx2( uint16_t *sum, uint8_t *pix, intptr_t stride );
 158 void x264_integral_init8h_sse4( uint16_t *sum, uint8_t *pix, intptr_t stride );
 159 void x264_integral_init8h_avx ( uint16_t *sum, uint8_t *pix, intptr_t stride );
 160 void x264_integral_init8h_avx2( uint16_t *sum, uint8_t *pix, intptr_t stride );
 161 void x264_integral_init4v_mmx  ( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
 162 void x264_integral_init4v_sse2 ( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
 163 void x264_integral_init4v_ssse3( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
 164 void x264_integral_init4v_avx2( uint16_t *sum8, uint16_t *sum4, intptr_t stride );
 165 void x264_integral_init8v_mmx ( uint16_t *sum8, intptr_t stride );
 166 void x264_integral_init8v_sse2( uint16_t *sum8, intptr_t stride );
 167 void x264_integral_init8v_avx2( uint16_t *sum8, intptr_t stride );
 168 void x264_mbtree_propagate_cost_sse2( int16_t *dst, uint16_t *propagate_in, uint16_t *intra_costs,
 169                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
 170 void x264_mbtree_propagate_cost_avx ( int16_t *dst, uint16_t *propagate_in, uint16_t *intra_costs,
 171                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
 172 void x264_mbtree_propagate_cost_fma4( int16_t *dst, uint16_t *propagate_in, uint16_t *intra_costs,
 173                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
 174 void x264_mbtree_propagate_cost_avx2( int16_t *dst, uint16_t *propagate_in, uint16_t *intra_costs,
 175                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
 176 void x264_mbtree_fix8_pack_ssse3( uint16_t *dst, float *src, int count );
 177 void x264_mbtree_fix8_pack_avx2 ( uint16_t *dst, float *src, int count );
 178 void x264_mbtree_fix8_unpack_ssse3( float *dst, uint16_t *src, int count );
 179 void x264_mbtree_fix8_unpack_avx2 ( float *dst, uint16_t *src, int count );
 180
 181 #define MC_CHROMA(cpu)\
 182 void x264_mc_chroma_##cpu( pixel *dstu, pixel *dstv, intptr_t i_dst, pixel *src, intptr_t i_src,\
 183                            int dx, int dy, int i_width, int i_height );
 184 MC_CHROMA(mmx2)
 185 MC_CHROMA(sse2)
 186 MC_CHROMA(ssse3)
 187 MC_CHROMA(ssse3_cache64)
 188 MC_CHROMA(avx)
 189 MC_CHROMA(avx2)
 190
 191 #define LOWRES(cpu)\
 192 void x264_frame_init_lowres_core_##cpu( pixel *src0, pixel *dst0, pixel *dsth, pixel *dstv, pixel *dstc,\
 193                                         intptr_t src_stride, intptr_t dst_stride, int width, int height );
 194 LOWRES(mmx2)
 195 LOWRES(cache32_mmx2)
 196 LOWRES(sse2)
 197 LOWRES(ssse3)
 198 LOWRES(avx)
 199 LOWRES(xop)
 200 LOWRES(avx2)
 201
 202 #define PIXEL_AVG_W(width,cpu)\
 203 void x264_pixel_avg2_w##width##_##cpu( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t );
 204 /* This declares some functions that don't exist, but that isn't a problem. */
 205 #define PIXEL_AVG_WALL(cpu)\
 206 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);
 207
 208 PIXEL_AVG_WALL(mmx2)
 209 PIXEL_AVG_WALL(cache32_mmx2)
 210 PIXEL_AVG_WALL(cache64_mmx2)
 211 PIXEL_AVG_WALL(cache64_sse2)
 212 PIXEL_AVG_WALL(sse2)
 213 PIXEL_AVG_WALL(cache64_ssse3)
 214 PIXEL_AVG_WALL(avx2)
 215
 216 #define PIXEL_AVG_WTAB(instr, name1, name2, name3, name4, name5)\
 217 static void (* const x264_pixel_avg_wtab_##instr[6])( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t ) =\
 218 {\
 219     NULL,\
 220     x264_pixel_avg2_w4_##name1,\
 221     x264_pixel_avg2_w8_##name2,\
 222     x264_pixel_avg2_w12_##name3,\
 223     x264_pixel_avg2_w16_##name4,\
 224     x264_pixel_avg2_w20_##name5,\
 225 };
 226
 227 #if HIGH_BIT_DEPTH
 228 /* we can replace w12/w20 with w10/w18 as only 9/17 pixels in fact are important */
 229 #define x264_pixel_avg2_w12_mmx2       x264_pixel_avg2_w10_mmx2
 230 #define x264_pixel_avg2_w20_mmx2       x264_pixel_avg2_w18_mmx2
 231 #define x264_pixel_avg2_w12_sse2         x264_pixel_avg2_w10_sse2
 232 #define x264_pixel_avg2_w20_sse2         x264_pixel_avg2_w18_sse2
 233 #define x264_pixel_avg2_w12_avx2         x264_pixel_avg2_w16_avx2
 234 #define x264_pixel_avg2_w20_avx2         x264_pixel_avg2_w18_avx2
 235 #else
 236 /* w16 sse2 is faster than w12 mmx as long as the cacheline issue is resolved */
 237 #define x264_pixel_avg2_w12_cache64_ssse3 x264_pixel_avg2_w16_cache64_ssse3
 238 #define x264_pixel_avg2_w12_cache64_sse2 x264_pixel_avg2_w16_cache64_sse2
 239 #define x264_pixel_avg2_w12_sse3         x264_pixel_avg2_w16_sse3
 240 #define x264_pixel_avg2_w12_sse2         x264_pixel_avg2_w16_sse2
 241 #endif // HIGH_BIT_DEPTH
 242
 243 PIXEL_AVG_WTAB(mmx2, mmx2, mmx2, mmx2, mmx2, mmx2)
 244 #if HIGH_BIT_DEPTH
 245 PIXEL_AVG_WTAB(sse2, mmx2, sse2, sse2, sse2, sse2)
 246 PIXEL_AVG_WTAB(avx2, mmx2, sse2, avx2, avx2, avx2)
 247 #else // !HIGH_BIT_DEPTH
 248 #if ARCH_X86
 249 PIXEL_AVG_WTAB(cache32_mmx2, mmx2, cache32_mmx2, cache32_mmx2, cache32_mmx2, cache32_mmx2)
 250 PIXEL_AVG_WTAB(cache64_mmx2, mmx2, cache64_mmx2, cache64_mmx2, cache64_mmx2, cache64_mmx2)
 251 #endif
 252 PIXEL_AVG_WTAB(sse2, mmx2, mmx2, sse2, sse2, sse2)
 253 PIXEL_AVG_WTAB(cache64_sse2, mmx2, cache64_mmx2, cache64_sse2, cache64_sse2, cache64_sse2)
 254 PIXEL_AVG_WTAB(cache64_ssse3, mmx2, cache64_mmx2, cache64_ssse3, cache64_ssse3, cache64_sse2)
 255 PIXEL_AVG_WTAB(cache64_ssse3_atom, mmx2, mmx2, cache64_ssse3, cache64_ssse3, sse2)
 256 PIXEL_AVG_WTAB(avx2, mmx2, mmx2, sse2, sse2, avx2)
 257 #endif // HIGH_BIT_DEPTH
 258
 259 #define MC_COPY_WTAB(instr, name1, name2, name3)\
 260 static void (* const x264_mc_copy_wtab_##instr[5])( pixel *, intptr_t, pixel *, intptr_t, int ) =\
 261 {\
 262     NULL,\
 263     x264_mc_copy_w4_##name1,\
 264     x264_mc_copy_w8_##name2,\
 265     NULL,\
 266     x264_mc_copy_w16_##name3,\
 267 };
 268
 269 MC_COPY_WTAB(mmx,mmx,mmx,mmx)
 270 #if HIGH_BIT_DEPTH
 271 MC_COPY_WTAB(sse,mmx,sse,sse)
 272 MC_COPY_WTAB(avx,mmx,sse,avx)
 273 #else
 274 MC_COPY_WTAB(sse,mmx,mmx,sse)
 275 #endif
 276
 277 #define MC_WEIGHT_WTAB(function, instr, name1, name2, w12version)\
 278     static void (* x264_mc_##function##_wtab_##instr[6])( pixel *, intptr_t, pixel *, intptr_t, const x264_weight_t *, int ) =\
 279 {\
 280     x264_mc_##function##_w4_##name1,\
 281     x264_mc_##function##_w4_##name1,\
 282     x264_mc_##function##_w8_##name2,\
 283     x264_mc_##function##_w##w12version##_##instr,\
 284     x264_mc_##function##_w16_##instr,\
 285     x264_mc_##function##_w20_##instr,\
 286 };
 287
 288 #if HIGH_BIT_DEPTH
 289 MC_WEIGHT_WTAB(weight,mmx2,mmx2,mmx2,12)
 290 MC_WEIGHT_WTAB(offsetadd,mmx2,mmx2,mmx2,12)
 291 MC_WEIGHT_WTAB(offsetsub,mmx2,mmx2,mmx2,12)
 292 MC_WEIGHT_WTAB(weight,sse2,mmx2,sse2,12)
 293 MC_WEIGHT_WTAB(offsetadd,sse2,mmx2,sse2,16)
 294 MC_WEIGHT_WTAB(offsetsub,sse2,mmx2,sse2,16)
 295
 296 static void x264_weight_cache_mmx2( x264_t *h, x264_weight_t *w )
 297 {
 298     if( w->i_scale == 1<<w->i_denom )
 299     {
 300         if( w->i_offset < 0 )
 301             w->weightfn = h->mc.offsetsub;
 302         else
 303             w->weightfn = h->mc.offsetadd;
 304         for( int i = 0; i < 8; i++ )
 305             w->cachea[i] = abs(w->i_offset<<(BIT_DEPTH-8));
 306         return;
 307     }
 308     w->weightfn = h->mc.weight;
 309     int den1 = 1<<w->i_denom;
 310     int den2 = w->i_scale<<1;
 311     int den3 = 1+(w->i_offset<<(BIT_DEPTH-8+1));
 312     for( int i = 0; i < 8; i++ )
 313     {
 314         w->cachea[i] = den1;
 315         w->cacheb[i] = i&1 ? den3 : den2;
 316     }
 317 }
 318 #else
 319 MC_WEIGHT_WTAB(weight,mmx2,mmx2,mmx2,12)
 320 MC_WEIGHT_WTAB(offsetadd,mmx2,mmx2,mmx2,12)
 321 MC_WEIGHT_WTAB(offsetsub,mmx2,mmx2,mmx2,12)
 322 MC_WEIGHT_WTAB(weight,sse2,mmx2,sse2,16)
 323 MC_WEIGHT_WTAB(offsetadd,sse2,mmx2,mmx2,16)
 324 MC_WEIGHT_WTAB(offsetsub,sse2,mmx2,mmx2,16)
 325 MC_WEIGHT_WTAB(weight,ssse3,ssse3,ssse3,16)
 326 MC_WEIGHT_WTAB(weight,avx2,ssse3,avx2,16)
 327
 328 static void x264_weight_cache_mmx2( x264_t *h, x264_weight_t *w )
 329 {
 330     int i;
 331     int16_t den1;
 332
 333     if( w->i_scale == 1<<w->i_denom )
 334     {
 335         if( w->i_offset < 0 )
 336             w->weightfn = h->mc.offsetsub;
 337         else
 338             w->weightfn = h->mc.offsetadd;
 339         memset( w->cachea, abs(w->i_offset), sizeof(w->cachea) );
 340         return;
 341     }
 342     w->weightfn = h->mc.weight;
 343     den1 = 1 << (w->i_denom - 1) | w->i_offset << w->i_denom;
 344     for( i = 0; i < 8; i++ )
 345     {
 346         w->cachea[i] = w->i_scale;
 347         w->cacheb[i] = den1;
 348     }
 349 }
 350
 351 static void x264_weight_cache_ssse3( x264_t *h, x264_weight_t *w )
 352 {
 353     int i, den1;
 354     if( w->i_scale == 1<<w->i_denom )
 355     {
 356         if( w->i_offset < 0 )
 357             w->weightfn = h->mc.offsetsub;
 358         else
 359             w->weightfn = h->mc.offsetadd;
 360
 361         memset( w->cachea, abs( w->i_offset ), sizeof(w->cachea) );
 362         return;
 363     }
 364     w->weightfn = h->mc.weight;
 365     den1 = w->i_scale << (8 - w->i_denom);
 366     for( i = 0; i < 8; i++ )
 367     {
 368         w->cachea[i] = den1;
 369         w->cacheb[i] = w->i_offset;
 370     }
 371 }
 372 #endif // !HIGH_BIT_DEPTH
 373
 374 #define MC_LUMA(name,instr1,instr2)\
 375 static void mc_luma_##name( pixel *dst,    intptr_t i_dst_stride,\
 376                             pixel *src[4], intptr_t i_src_stride,\
 377                             int mvx, int mvy,\
 378                             int i_width, int i_height, const x264_weight_t *weight )\
 379 {\
 380     int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
 381     int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
 382     pixel *src1 = src[x264_hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
 383     if( qpel_idx & 5 ) /* qpel interpolation needed */\
 384     {\
 385         pixel *src2 = src[x264_hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
 386         x264_pixel_avg_wtab_##instr1[i_width>>2](\
 387                 dst, i_dst_stride, src1, i_src_stride,\
 388                 src2, i_height );\
 389         if( weight->weightfn )\
 390             weight->weightfn[i_width>>2]( dst, i_dst_stride, dst, i_dst_stride, weight, i_height );\
 391     }\
 392     else if( weight->weightfn )\
 393         weight->weightfn[i_width>>2]( dst, i_dst_stride, src1, i_src_stride, weight, i_height );\
 394     else\
 395         x264_mc_copy_wtab_##instr2[i_width>>2](dst, i_dst_stride, src1, i_src_stride, i_height );\
 396 }
 397
 398 MC_LUMA(mmx2,mmx2,mmx)
 399 MC_LUMA(sse2,sse2,sse)
 400 #if HIGH_BIT_DEPTH
 401 MC_LUMA(avx2,avx2,avx)
 402 #else
 403 #if ARCH_X86
 404 MC_LUMA(cache32_mmx2,cache32_mmx2,mmx)
 405 MC_LUMA(cache64_mmx2,cache64_mmx2,mmx)
 406 #endif
 407 MC_LUMA(cache64_sse2,cache64_sse2,sse)
 408 MC_LUMA(cache64_ssse3,cache64_ssse3,sse)
 409 MC_LUMA(cache64_ssse3_atom,cache64_ssse3_atom,sse)
 410 #endif // !HIGH_BIT_DEPTH
 411
 412 #define GET_REF(name)\
 413 static pixel *get_ref_##name( pixel *dst,   intptr_t *i_dst_stride,\
 414                               pixel *src[4], intptr_t i_src_stride,\
 415                               int mvx, int mvy,\
 416                               int i_width, int i_height, const x264_weight_t *weight )\
 417 {\
 418     int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
 419     int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
 420     pixel *src1 = src[x264_hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
 421     if( qpel_idx & 5 ) /* qpel interpolation needed */\
 422     {\
 423         pixel *src2 = src[x264_hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
 424         x264_pixel_avg_wtab_##name[i_width>>2](\
 425                 dst, *i_dst_stride, src1, i_src_stride,\
 426                 src2, i_height );\
 427         if( weight->weightfn )\
 428             weight->weightfn[i_width>>2]( dst, *i_dst_stride, dst, *i_dst_stride, weight, i_height );\
 429         return dst;\
 430     }\
 431     else if( weight->weightfn )\
 432     {\
 433         weight->weightfn[i_width>>2]( dst, *i_dst_stride, src1, i_src_stride, weight, i_height );\
 434         return dst;\
 435     }\
 436     else\
 437     {\
 438         *i_dst_stride = i_src_stride;\
 439         return src1;\
 440     }\
 441 }
 442
 443 GET_REF(mmx2)
 444 GET_REF(sse2)
 445 GET_REF(avx2)
 446 #if !HIGH_BIT_DEPTH
 447 #if ARCH_X86
 448 GET_REF(cache32_mmx2)
 449 GET_REF(cache64_mmx2)
 450 #endif
 451 GET_REF(cache64_sse2)
 452 GET_REF(cache64_ssse3)
 453 GET_REF(cache64_ssse3_atom)
 454 #endif // !HIGH_BIT_DEPTH
 455
 456 #define HPEL(align, cpu, cpuv, cpuc, cpuh)\
 457 void x264_hpel_filter_v_##cpuv( pixel *dst, pixel *src, int16_t *buf, intptr_t stride, intptr_t width);\
 458 void x264_hpel_filter_c_##cpuc( pixel *dst, int16_t *buf, intptr_t width );\
 459 void x264_hpel_filter_h_##cpuh( pixel *dst, pixel *src, intptr_t width );\
 460 static void x264_hpel_filter_##cpu( pixel *dsth, pixel *dstv, pixel *dstc, pixel *src,\
 461                                     intptr_t stride, int width, int height, int16_t *buf )\
 462 {\
 463     intptr_t realign = (intptr_t)src & (align-1);\
 464     src -= realign;\
 465     dstv -= realign;\
 466     dstc -= realign;\
 467     dsth -= realign;\
 468     width += realign;\
 469     while( height-- )\
 470     {\
 471         x264_hpel_filter_v_##cpuv( dstv, src, buf+16, stride, width );\
 472         x264_hpel_filter_c_##cpuc( dstc, buf+16, width );\
 473         x264_hpel_filter_h_##cpuh( dsth, src, width );\
 474         dsth += stride;\
 475         dstv += stride;\
 476         dstc += stride;\
 477         src  += stride;\
 478     }\
 479     x264_sfence();\
 480 }
 481
 482 HPEL(8, mmx2, mmx2, mmx2, mmx2)
 483 #if HIGH_BIT_DEPTH
 484 HPEL(16, sse2, sse2, sse2, sse2)
 485 #else // !HIGH_BIT_DEPTH
 486 HPEL(16, sse2_amd, mmx2, mmx2, sse2)
 487 #if ARCH_X86_64
 488 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 );
 489 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 );
 490 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 );
 491 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 );
 492 #else
 493 HPEL(16, sse2, sse2, sse2, sse2)
 494 HPEL(16, ssse3, ssse3, ssse3, ssse3)
 495 HPEL(16, avx, avx, avx, avx)
 496 HPEL(32, avx2, avx2, avx2, avx2)
 497 #endif
 498 #endif // HIGH_BIT_DEPTH
 499
 500 #define PLANE_COPY(align, cpu)\
 501 static void x264_plane_copy_##cpu( pixel *dst, intptr_t i_dst, pixel *src, intptr_t i_src, int w, int h )\
 502 {\
 503     int c_w = (align) / sizeof(pixel) - 1;\
 504     if( w < 256 ) /* tiny resolutions don't want non-temporal hints. dunno the exact threshold. */\
 505         x264_plane_copy_c( dst, i_dst, src, i_src, w, h );\
 506     else if( !(w&c_w) )\
 507         x264_plane_copy_core_##cpu( dst, i_dst, src, i_src, w, h );\
 508     else\
 509     {\
 510         if( --h > 0 )\
 511         {\
 512             if( i_src > 0 )\
 513             {\
 514                 x264_plane_copy_core_##cpu( dst, i_dst, src, i_src, (w+c_w)&~c_w, h );\
 515                 dst += i_dst * h;\
 516                 src += i_src * h;\
 517             }\
 518             else\
 519                 x264_plane_copy_core_##cpu( dst+i_dst, i_dst, src+i_src, i_src, (w+c_w)&~c_w, h );\
 520         }\
 521         /* use plain memcpy on the last line (in memory order) to avoid overreading src. */\
 522         memcpy( dst, src, w*sizeof(pixel) );\
 523     }\
 524 }
 525
 526 PLANE_COPY(16, sse)
 527 PLANE_COPY(32, avx)
 528
 529 #define PLANE_COPY_SWAP(align, cpu)\
 530 static void x264_plane_copy_swap_##cpu( pixel *dst, intptr_t i_dst, pixel *src, intptr_t i_src, int w, int h )\
 531 {\
 532     int c_w = (align>>1) / sizeof(pixel) - 1;\
 533     if( !(w&c_w) )\
 534         x264_plane_copy_swap_core_##cpu( dst, i_dst, src, i_src, w, h );\
 535     else if( w > c_w )\
 536     {\
 537         if( --h > 0 )\
 538         {\
 539             if( i_src > 0 )\
 540             {\
 541                 x264_plane_copy_swap_core_##cpu( dst, i_dst, src, i_src, (w+c_w)&~c_w, h );\
 542                 dst += i_dst * h;\
 543                 src += i_src * h;\
 544             }\
 545             else\
 546                 x264_plane_copy_swap_core_##cpu( dst+i_dst, i_dst, src+i_src, i_src, (w+c_w)&~c_w, h );\
 547         }\
 548         x264_plane_copy_swap_core_##cpu( dst, 0, src, 0, w&~c_w, 1 );\
 549         for( int x = 2*(w&~c_w); x < 2*w; x += 2 )\
 550         {\
 551             dst[x]   = src[x+1];\
 552             dst[x+1] = src[x];\
 553         }\
 554     }\
 555     else\
 556         x264_plane_copy_swap_c( dst, i_dst, src, i_src, w, h );\
 557 }
 558
 559 PLANE_COPY_SWAP(16, ssse3)
 560 PLANE_COPY_SWAP(32, avx2)
 561
 562 #define PLANE_INTERLEAVE(cpu) \
 563 static void x264_plane_copy_interleave_##cpu( pixel *dst,  intptr_t i_dst,\
 564                                               pixel *srcu, intptr_t i_srcu,\
 565                                               pixel *srcv, intptr_t i_srcv, int w, int h )\
 566 {\
 567     int c_w = 16 / sizeof(pixel) - 1;\
 568     if( !(w&c_w) )\
 569         x264_plane_copy_interleave_core_##cpu( dst, i_dst, srcu, i_srcu, srcv, i_srcv, w, h );\
 570     else if( w > c_w && (i_srcu ^ i_srcv) >= 0 ) /* only works correctly for strides with identical signs */\
 571     {\
 572         if( --h > 0 )\
 573         {\
 574             if( i_srcu > 0 )\
 575             {\
 576                 x264_plane_copy_interleave_core_##cpu( dst, i_dst, srcu, i_srcu, srcv, i_srcv, (w+c_w)&~c_w, h );\
 577                 dst  += i_dst  * h;\
 578                 srcu += i_srcu * h;\
 579                 srcv += i_srcv * h;\
 580             }\
 581             else\
 582                 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 );\
 583         }\
 584         x264_plane_copy_interleave_c( dst, 0, srcu, 0, srcv, 0, w, 1 );\
 585     }\
 586     else\
 587         x264_plane_copy_interleave_c( dst, i_dst, srcu, i_srcu, srcv, i_srcv, w, h );\
 588 }
 589
 590 PLANE_INTERLEAVE(mmx2)
 591 PLANE_INTERLEAVE(sse2)
 592 #if HIGH_BIT_DEPTH
 593 PLANE_INTERLEAVE(avx)
 594 #endif
 595
 596 #if HAVE_X86_INLINE_ASM
 597 #undef MC_CLIP_ADD
 598 #define MC_CLIP_ADD(s,x)\
 599 do\
 600 {\
 601     int temp;\
 602     asm("movd       %0, %%xmm0     \n"\
 603         "movd       %2, %%xmm1     \n"\
 604         "paddsw %%xmm1, %%xmm0     \n"\
 605         "movd   %%xmm0, %1         \n"\
 606         :"+m"(s), "=&r"(temp)\
 607         :"m"(x)\
 608     );\
 609     s = temp;\
 610 } while(0)
 611
 612 #undef MC_CLIP_ADD2
 613 #define MC_CLIP_ADD2(s,x)\
 614 do\
 615 {\
 616     asm("movd       %0, %%xmm0     \n"\
 617         "movd       %1, %%xmm1     \n"\
 618         "paddsw %%xmm1, %%xmm0     \n"\
 619         "movd   %%xmm0, %0         \n"\
 620         :"+m"(M32(s))\
 621         :"m"(M32(x))\
 622     );\
 623 } while(0)
 624 #endif
 625
 626 PROPAGATE_LIST(ssse3)
 627 PROPAGATE_LIST(avx)
 628
 629 void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
 630 {
 631     if( !(cpu&X264_CPU_MMX) )
 632         return;
 633
 634     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_mmx;
 635     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_mmx;
 636
 637     pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_mmx;
 638
 639     pf->copy_16x16_unaligned = x264_mc_copy_w16_mmx;
 640     pf->copy[PIXEL_16x16] = x264_mc_copy_w16_mmx;
 641     pf->copy[PIXEL_8x8]   = x264_mc_copy_w8_mmx;
 642     pf->copy[PIXEL_4x4]   = x264_mc_copy_w4_mmx;
 643     pf->memcpy_aligned  = x264_memcpy_aligned_mmx;
 644     pf->memzero_aligned = x264_memzero_aligned_mmx;
 645     pf->integral_init4v = x264_integral_init4v_mmx;
 646     pf->integral_init8v = x264_integral_init8v_mmx;
 647
 648     if( !(cpu&X264_CPU_MMX2) )
 649         return;
 650
 651     pf->prefetch_fenc_420 = x264_prefetch_fenc_420_mmx2;
 652     pf->prefetch_fenc_422 = x264_prefetch_fenc_422_mmx2;
 653     pf->prefetch_ref  = x264_prefetch_ref_mmx2;
 654
 655     pf->plane_copy_interleave = x264_plane_copy_interleave_mmx2;
 656     pf->store_interleave_chroma = x264_store_interleave_chroma_mmx2;
 657
 658     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_mmx2;
 659     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_mmx2;
 660     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_mmx2;
 661     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_mmx2;
 662     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_mmx2;
 663     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_mmx2;
 664     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_mmx2;
 665     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_mmx2;
 666     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_mmx2;
 667
 668     pf->mc_luma = mc_luma_mmx2;
 669     pf->get_ref = get_ref_mmx2;
 670     pf->mc_chroma = x264_mc_chroma_mmx2;
 671     pf->hpel_filter = x264_hpel_filter_mmx2;
 672     pf->weight = x264_mc_weight_wtab_mmx2;
 673     pf->weight_cache = x264_weight_cache_mmx2;
 674     pf->offsetadd = x264_mc_offsetadd_wtab_mmx2;
 675     pf->offsetsub = x264_mc_offsetsub_wtab_mmx2;
 676
 677     pf->frame_init_lowres_core = x264_frame_init_lowres_core_mmx2;
 678
 679     if( cpu&X264_CPU_SSE )
 680     {
 681         pf->memcpy_aligned  = x264_memcpy_aligned_sse;
 682         pf->memzero_aligned = x264_memzero_aligned_sse;
 683         pf->plane_copy = x264_plane_copy_sse;
 684     }
 685
 686 #if HIGH_BIT_DEPTH
 687 #if ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
 688     if( cpu&(X264_CPU_CACHELINE_32|X264_CPU_CACHELINE_64) )
 689         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
 690 #endif
 691
 692     if( !(cpu&X264_CPU_SSE2) )
 693         return;
 694
 695     pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
 696
 697     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_sse2;
 698     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_sse2;
 699
 700     pf->plane_copy_interleave   = x264_plane_copy_interleave_sse2;
 701     pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
 702
 703     if( cpu&X264_CPU_SSE2_IS_FAST )
 704     {
 705         pf->get_ref = get_ref_sse2;
 706         pf->mc_luma = mc_luma_sse2;
 707         pf->hpel_filter = x264_hpel_filter_sse2;
 708     }
 709
 710     pf->integral_init4v = x264_integral_init4v_sse2;
 711     pf->integral_init8v = x264_integral_init8v_sse2;
 712     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
 713     pf->store_interleave_chroma = x264_store_interleave_chroma_sse2;
 714     pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
 715     pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
 716
 717     if( cpu&X264_CPU_SSE2_IS_SLOW )
 718         return;
 719
 720     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
 721     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_sse2;
 722     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_sse2;
 723     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_sse2;
 724     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_sse2;
 725     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_sse2;
 726     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_sse2;
 727     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_sse2;
 728     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_sse2;
 729
 730     pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse;
 731     pf->weight = x264_mc_weight_wtab_sse2;
 732
 733     if( !(cpu&X264_CPU_STACK_MOD4) )
 734         pf->mc_chroma = x264_mc_chroma_sse2;
 735
 736     if( !(cpu&X264_CPU_SSSE3) )
 737         return;
 738
 739     pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
 740     pf->plane_copy_swap = x264_plane_copy_swap_ssse3;
 741     pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_ssse3;
 742     pf->mbtree_propagate_list = x264_mbtree_propagate_list_ssse3;
 743     pf->mbtree_fix8_pack      = x264_mbtree_fix8_pack_ssse3;
 744     pf->mbtree_fix8_unpack    = x264_mbtree_fix8_unpack_ssse3;
 745
 746     if( !(cpu&(X264_CPU_SLOW_SHUFFLE|X264_CPU_SLOW_ATOM|X264_CPU_SLOW_PALIGNR)) )
 747         pf->integral_init4v = x264_integral_init4v_ssse3;
 748
 749     if( !(cpu&X264_CPU_AVX) )
 750         return;
 751
 752     pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx;
 753     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_avx;
 754     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_avx;
 755     pf->plane_copy_interleave        = x264_plane_copy_interleave_avx;
 756     pf->plane_copy_deinterleave      = x264_plane_copy_deinterleave_avx;
 757     pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_avx;
 758     pf->store_interleave_chroma      = x264_store_interleave_chroma_avx;
 759     pf->copy[PIXEL_16x16]            = x264_mc_copy_w16_aligned_avx;
 760
 761     if( !(cpu&X264_CPU_STACK_MOD4) )
 762         pf->mc_chroma = x264_mc_chroma_avx;
 763
 764     if( cpu&X264_CPU_XOP )
 765         pf->frame_init_lowres_core = x264_frame_init_lowres_core_xop;
 766
 767     if( cpu&X264_CPU_AVX2 )
 768     {
 769         pf->mc_luma = mc_luma_avx2;
 770         pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_avx2;
 771     }
 772 #else // !HIGH_BIT_DEPTH
 773
 774 #if ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
 775     if( cpu&X264_CPU_CACHELINE_32 )
 776     {
 777         pf->mc_luma = mc_luma_cache32_mmx2;
 778         pf->get_ref = get_ref_cache32_mmx2;
 779         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
 780     }
 781     else if( cpu&X264_CPU_CACHELINE_64 )
 782     {
 783         pf->mc_luma = mc_luma_cache64_mmx2;
 784         pf->get_ref = get_ref_cache64_mmx2;
 785         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
 786     }
 787 #endif
 788
 789     if( !(cpu&X264_CPU_SSE2) )
 790         return;
 791
 792     pf->integral_init4v = x264_integral_init4v_sse2;
 793     pf->integral_init8v = x264_integral_init8v_sse2;
 794     pf->hpel_filter = x264_hpel_filter_sse2_amd;
 795     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
 796     pf->plane_copy_deinterleave_rgb = x264_plane_copy_deinterleave_rgb_sse2;
 797
 798     if( !(cpu&X264_CPU_SSE2_IS_SLOW) )
 799     {
 800         pf->weight = x264_mc_weight_wtab_sse2;
 801         if( !(cpu&X264_CPU_SLOW_ATOM) )
 802         {
 803             pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
 804             pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
 805         }
 806
 807         pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse;
 808         pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
 809         pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_sse2;
 810         pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_sse2;
 811         pf->avg[PIXEL_8x8]  = x264_pixel_avg_8x8_sse2;
 812         pf->avg[PIXEL_8x4]  = x264_pixel_avg_8x4_sse2;
 813         pf->hpel_filter = x264_hpel_filter_sse2;
 814         pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
 815         if( !(cpu&X264_CPU_STACK_MOD4) )
 816             pf->mc_chroma = x264_mc_chroma_sse2;
 817
 818         if( cpu&X264_CPU_SSE2_IS_FAST )
 819         {
 820             pf->store_interleave_chroma = x264_store_interleave_chroma_sse2; // FIXME sse2fast? sse2medium?
 821             pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_sse2;
 822             pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_sse2;
 823             pf->plane_copy_interleave   = x264_plane_copy_interleave_sse2;
 824             pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
 825             pf->mc_luma = mc_luma_sse2;
 826             pf->get_ref = get_ref_sse2;
 827             if( cpu&X264_CPU_CACHELINE_64 )
 828             {
 829                 pf->mc_luma = mc_luma_cache64_sse2;
 830                 pf->get_ref = get_ref_cache64_sse2;
 831             }
 832         }
 833     }
 834
 835     if( !(cpu&X264_CPU_SSSE3) )
 836         return;
 837
 838     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_ssse3;
 839     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_ssse3;
 840     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_ssse3;
 841     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_ssse3;
 842     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_ssse3;
 843     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_ssse3;
 844     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_ssse3;
 845     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_ssse3;
 846     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_ssse3;
 847     pf->plane_copy_swap = x264_plane_copy_swap_ssse3;
 848     pf->plane_copy_deinterleave_rgb = x264_plane_copy_deinterleave_rgb_ssse3;
 849     pf->mbtree_propagate_list = x264_mbtree_propagate_list_ssse3;
 850     pf->mbtree_fix8_pack      = x264_mbtree_fix8_pack_ssse3;
 851     pf->mbtree_fix8_unpack    = x264_mbtree_fix8_unpack_ssse3;
 852
 853     if( !(cpu&X264_CPU_SLOW_PSHUFB) )
 854     {
 855         pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_ssse3;
 856         pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_ssse3;
 857         pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_ssse3;
 858     }
 859
 860     if( !(cpu&X264_CPU_SLOW_PALIGNR) )
 861     {
 862 #if ARCH_X86_64
 863         if( !(cpu&X264_CPU_SLOW_ATOM) ) /* The 64-bit version is slower, but the 32-bit version is faster? */
 864 #endif
 865             pf->hpel_filter = x264_hpel_filter_ssse3;
 866         pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
 867     }
 868     if( !(cpu&X264_CPU_STACK_MOD4) )
 869         pf->mc_chroma = x264_mc_chroma_ssse3;
 870
 871     if( cpu&X264_CPU_CACHELINE_64 )
 872     {
 873         if( !(cpu&X264_CPU_STACK_MOD4) )
 874             pf->mc_chroma = x264_mc_chroma_ssse3_cache64;
 875         pf->mc_luma = mc_luma_cache64_ssse3;
 876         pf->get_ref = get_ref_cache64_ssse3;
 877         if( cpu&X264_CPU_SLOW_ATOM )
 878         {
 879             pf->mc_luma = mc_luma_cache64_ssse3_atom;
 880             pf->get_ref = get_ref_cache64_ssse3_atom;
 881         }
 882     }
 883
 884     pf->weight_cache = x264_weight_cache_ssse3;
 885     pf->weight = x264_mc_weight_wtab_ssse3;
 886
 887     if( !(cpu&(X264_CPU_SLOW_SHUFFLE|X264_CPU_SLOW_ATOM|X264_CPU_SLOW_PALIGNR)) )
 888         pf->integral_init4v = x264_integral_init4v_ssse3;
 889
 890     if( !(cpu&X264_CPU_SSE4) )
 891         return;
 892
 893     pf->integral_init4h = x264_integral_init4h_sse4;
 894     pf->integral_init8h = x264_integral_init8h_sse4;
 895
 896     if( !(cpu&X264_CPU_AVX) )
 897         return;
 898
 899     pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx;
 900     pf->integral_init8h = x264_integral_init8h_avx;
 901     pf->hpel_filter = x264_hpel_filter_avx;
 902
 903     if( !(cpu&X264_CPU_STACK_MOD4) )
 904         pf->mc_chroma = x264_mc_chroma_avx;
 905
 906     if( cpu&X264_CPU_XOP )
 907         pf->frame_init_lowres_core = x264_frame_init_lowres_core_xop;
 908
 909     if( cpu&X264_CPU_AVX2 )
 910     {
 911         pf->hpel_filter = x264_hpel_filter_avx2;
 912         pf->mc_chroma = x264_mc_chroma_avx2;
 913         pf->weight = x264_mc_weight_wtab_avx2;
 914         pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_avx2;
 915         pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_avx2;
 916         pf->integral_init8v = x264_integral_init8v_avx2;
 917         pf->integral_init4v = x264_integral_init4v_avx2;
 918         pf->integral_init8h = x264_integral_init8h_avx2;
 919         pf->integral_init4h = x264_integral_init4h_avx2;
 920         pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx2;
 921     }
 922 #endif // HIGH_BIT_DEPTH
 923
 924     if( !(cpu&X264_CPU_AVX) )
 925         return;
 926     pf->memzero_aligned = x264_memzero_aligned_avx;
 927     pf->plane_copy = x264_plane_copy_avx;
 928     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_avx;
 929     pf->mbtree_propagate_list = x264_mbtree_propagate_list_avx;
 930
 931     if( cpu&X264_CPU_FMA4 )
 932         pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_fma4;
 933
 934     if( !(cpu&X264_CPU_AVX2) )
 935         return;
 936     pf->plane_copy_swap = x264_plane_copy_swap_avx2;
 937     pf->get_ref = get_ref_avx2;
 938     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_avx2;
 939     pf->mbtree_fix8_pack      = x264_mbtree_fix8_pack_avx2;
 940     pf->mbtree_fix8_unpack    = x264_mbtree_fix8_unpack_avx2;
 941 }