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_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
 177 #define MC_CHROMA(cpu)\
 178 void x264_mc_chroma_##cpu( pixel *dstu, pixel *dstv, intptr_t i_dst, pixel *src, intptr_t i_src,\
 179                            int dx, int dy, int i_width, int i_height );
 180 MC_CHROMA(mmx2)
 181 MC_CHROMA(sse2)
 182 MC_CHROMA(ssse3)
 183 MC_CHROMA(ssse3_cache64)
 184 MC_CHROMA(avx)
 185 MC_CHROMA(avx2)
 186
 187 #define LOWRES(cpu)\
 188 void x264_frame_init_lowres_core_##cpu( pixel *src0, pixel *dst0, pixel *dsth, pixel *dstv, pixel *dstc,\
 189                                         intptr_t src_stride, intptr_t dst_stride, int width, int height );
 190 LOWRES(mmx2)
 191 LOWRES(cache32_mmx2)
 192 LOWRES(sse2)
 193 LOWRES(ssse3)
 194 LOWRES(avx)
 195 LOWRES(xop)
 196 LOWRES(avx2)
 197
 198 #define PIXEL_AVG_W(width,cpu)\
 199 void x264_pixel_avg2_w##width##_##cpu( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t );
 200 /* This declares some functions that don't exist, but that isn't a problem. */
 201 #define PIXEL_AVG_WALL(cpu)\
 202 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);
 203
 204 PIXEL_AVG_WALL(mmx2)
 205 PIXEL_AVG_WALL(cache32_mmx2)
 206 PIXEL_AVG_WALL(cache64_mmx2)
 207 PIXEL_AVG_WALL(cache64_sse2)
 208 PIXEL_AVG_WALL(sse2)
 209 PIXEL_AVG_WALL(cache64_ssse3)
 210 PIXEL_AVG_WALL(avx2)
 211
 212 #define PIXEL_AVG_WTAB(instr, name1, name2, name3, name4, name5)\
 213 static void (* const x264_pixel_avg_wtab_##instr[6])( pixel *, intptr_t, pixel *, intptr_t, pixel *, intptr_t ) =\
 214 {\
 215     NULL,\
 216     x264_pixel_avg2_w4_##name1,\
 217     x264_pixel_avg2_w8_##name2,\
 218     x264_pixel_avg2_w12_##name3,\
 219     x264_pixel_avg2_w16_##name4,\
 220     x264_pixel_avg2_w20_##name5,\
 221 };
 222
 223 #if HIGH_BIT_DEPTH
 224 /* we can replace w12/w20 with w10/w18 as only 9/17 pixels in fact are important */
 225 #define x264_pixel_avg2_w12_mmx2       x264_pixel_avg2_w10_mmx2
 226 #define x264_pixel_avg2_w20_mmx2       x264_pixel_avg2_w18_mmx2
 227 #define x264_pixel_avg2_w12_sse2         x264_pixel_avg2_w10_sse2
 228 #define x264_pixel_avg2_w20_sse2         x264_pixel_avg2_w18_sse2
 229 #define x264_pixel_avg2_w12_avx2         x264_pixel_avg2_w16_avx2
 230 #define x264_pixel_avg2_w20_avx2         x264_pixel_avg2_w18_avx2
 231 #else
 232 /* w16 sse2 is faster than w12 mmx as long as the cacheline issue is resolved */
 233 #define x264_pixel_avg2_w12_cache64_ssse3 x264_pixel_avg2_w16_cache64_ssse3
 234 #define x264_pixel_avg2_w12_cache64_sse2 x264_pixel_avg2_w16_cache64_sse2
 235 #define x264_pixel_avg2_w12_sse3         x264_pixel_avg2_w16_sse3
 236 #define x264_pixel_avg2_w12_sse2         x264_pixel_avg2_w16_sse2
 237 #endif // HIGH_BIT_DEPTH
 238
 239 PIXEL_AVG_WTAB(mmx2, mmx2, mmx2, mmx2, mmx2, mmx2)
 240 #if HIGH_BIT_DEPTH
 241 PIXEL_AVG_WTAB(sse2, mmx2, sse2, sse2, sse2, sse2)
 242 PIXEL_AVG_WTAB(avx2, mmx2, sse2, avx2, avx2, avx2)
 243 #else // !HIGH_BIT_DEPTH
 244 #if ARCH_X86
 245 PIXEL_AVG_WTAB(cache32_mmx2, mmx2, cache32_mmx2, cache32_mmx2, cache32_mmx2, cache32_mmx2)
 246 PIXEL_AVG_WTAB(cache64_mmx2, mmx2, cache64_mmx2, cache64_mmx2, cache64_mmx2, cache64_mmx2)
 247 #endif
 248 PIXEL_AVG_WTAB(sse2, mmx2, mmx2, sse2, sse2, sse2)
 249 PIXEL_AVG_WTAB(cache64_sse2, mmx2, cache64_mmx2, cache64_sse2, cache64_sse2, cache64_sse2)
 250 PIXEL_AVG_WTAB(cache64_ssse3, mmx2, cache64_mmx2, cache64_ssse3, cache64_ssse3, cache64_sse2)
 251 PIXEL_AVG_WTAB(cache64_ssse3_atom, mmx2, mmx2, cache64_ssse3, cache64_ssse3, sse2)
 252 PIXEL_AVG_WTAB(avx2, mmx2, mmx2, sse2, sse2, avx2)
 253 #endif // HIGH_BIT_DEPTH
 254
 255 #define MC_COPY_WTAB(instr, name1, name2, name3)\
 256 static void (* const x264_mc_copy_wtab_##instr[5])( pixel *, intptr_t, pixel *, intptr_t, int ) =\
 257 {\
 258     NULL,\
 259     x264_mc_copy_w4_##name1,\
 260     x264_mc_copy_w8_##name2,\
 261     NULL,\
 262     x264_mc_copy_w16_##name3,\
 263 };
 264
 265 MC_COPY_WTAB(mmx,mmx,mmx,mmx)
 266 #if HIGH_BIT_DEPTH
 267 MC_COPY_WTAB(sse,mmx,sse,sse)
 268 MC_COPY_WTAB(avx,mmx,sse,avx)
 269 #else
 270 MC_COPY_WTAB(sse,mmx,mmx,sse)
 271 #endif
 272
 273 #define MC_WEIGHT_WTAB(function, instr, name1, name2, w12version)\
 274     static void (* x264_mc_##function##_wtab_##instr[6])( pixel *, intptr_t, pixel *, intptr_t, const x264_weight_t *, int ) =\
 275 {\
 276     x264_mc_##function##_w4_##name1,\
 277     x264_mc_##function##_w4_##name1,\
 278     x264_mc_##function##_w8_##name2,\
 279     x264_mc_##function##_w##w12version##_##instr,\
 280     x264_mc_##function##_w16_##instr,\
 281     x264_mc_##function##_w20_##instr,\
 282 };
 283
 284 #if HIGH_BIT_DEPTH
 285 MC_WEIGHT_WTAB(weight,mmx2,mmx2,mmx2,12)
 286 MC_WEIGHT_WTAB(offsetadd,mmx2,mmx2,mmx2,12)
 287 MC_WEIGHT_WTAB(offsetsub,mmx2,mmx2,mmx2,12)
 288 MC_WEIGHT_WTAB(weight,sse2,mmx2,sse2,12)
 289 MC_WEIGHT_WTAB(offsetadd,sse2,mmx2,sse2,16)
 290 MC_WEIGHT_WTAB(offsetsub,sse2,mmx2,sse2,16)
 291
 292 static void x264_weight_cache_mmx2( x264_t *h, x264_weight_t *w )
 293 {
 294     if( w->i_scale == 1<<w->i_denom )
 295     {
 296         if( w->i_offset < 0 )
 297             w->weightfn = h->mc.offsetsub;
 298         else
 299             w->weightfn = h->mc.offsetadd;
 300         for( int i = 0; i < 8; i++ )
 301             w->cachea[i] = abs(w->i_offset<<(BIT_DEPTH-8));
 302         return;
 303     }
 304     w->weightfn = h->mc.weight;
 305     int den1 = 1<<w->i_denom;
 306     int den2 = w->i_scale<<1;
 307     int den3 = 1+(w->i_offset<<(BIT_DEPTH-8+1));
 308     for( int i = 0; i < 8; i++ )
 309     {
 310         w->cachea[i] = den1;
 311         w->cacheb[i] = i&1 ? den3 : den2;
 312     }
 313 }
 314 #else
 315 MC_WEIGHT_WTAB(weight,mmx2,mmx2,mmx2,12)
 316 MC_WEIGHT_WTAB(offsetadd,mmx2,mmx2,mmx2,12)
 317 MC_WEIGHT_WTAB(offsetsub,mmx2,mmx2,mmx2,12)
 318 MC_WEIGHT_WTAB(weight,sse2,mmx2,sse2,16)
 319 MC_WEIGHT_WTAB(offsetadd,sse2,mmx2,mmx2,16)
 320 MC_WEIGHT_WTAB(offsetsub,sse2,mmx2,mmx2,16)
 321 MC_WEIGHT_WTAB(weight,ssse3,ssse3,ssse3,16)
 322 MC_WEIGHT_WTAB(weight,avx2,ssse3,avx2,16)
 323
 324 static void x264_weight_cache_mmx2( x264_t *h, x264_weight_t *w )
 325 {
 326     int i;
 327     int16_t den1;
 328
 329     if( w->i_scale == 1<<w->i_denom )
 330     {
 331         if( w->i_offset < 0 )
 332             w->weightfn = h->mc.offsetsub;
 333         else
 334             w->weightfn = h->mc.offsetadd;
 335         memset( w->cachea, abs(w->i_offset), sizeof(w->cachea) );
 336         return;
 337     }
 338     w->weightfn = h->mc.weight;
 339     den1 = 1 << (w->i_denom - 1) | w->i_offset << w->i_denom;
 340     for( i = 0; i < 8; i++ )
 341     {
 342         w->cachea[i] = w->i_scale;
 343         w->cacheb[i] = den1;
 344     }
 345 }
 346
 347 static void x264_weight_cache_ssse3( x264_t *h, x264_weight_t *w )
 348 {
 349     int i, den1;
 350     if( w->i_scale == 1<<w->i_denom )
 351     {
 352         if( w->i_offset < 0 )
 353             w->weightfn = h->mc.offsetsub;
 354         else
 355             w->weightfn = h->mc.offsetadd;
 356
 357         memset( w->cachea, abs( w->i_offset ), sizeof(w->cachea) );
 358         return;
 359     }
 360     w->weightfn = h->mc.weight;
 361     den1 = w->i_scale << (8 - w->i_denom);
 362     for( i = 0; i < 8; i++ )
 363     {
 364         w->cachea[i] = den1;
 365         w->cacheb[i] = w->i_offset;
 366     }
 367 }
 368 #endif // !HIGH_BIT_DEPTH
 369
 370 #define MC_LUMA(name,instr1,instr2)\
 371 static void mc_luma_##name( pixel *dst,    intptr_t i_dst_stride,\
 372                             pixel *src[4], intptr_t i_src_stride,\
 373                             int mvx, int mvy,\
 374                             int i_width, int i_height, const x264_weight_t *weight )\
 375 {\
 376     int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
 377     int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
 378     pixel *src1 = src[x264_hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
 379     if( qpel_idx & 5 ) /* qpel interpolation needed */\
 380     {\
 381         pixel *src2 = src[x264_hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
 382         x264_pixel_avg_wtab_##instr1[i_width>>2](\
 383                 dst, i_dst_stride, src1, i_src_stride,\
 384                 src2, i_height );\
 385         if( weight->weightfn )\
 386             weight->weightfn[i_width>>2]( dst, i_dst_stride, dst, i_dst_stride, weight, i_height );\
 387     }\
 388     else if( weight->weightfn )\
 389         weight->weightfn[i_width>>2]( dst, i_dst_stride, src1, i_src_stride, weight, i_height );\
 390     else\
 391         x264_mc_copy_wtab_##instr2[i_width>>2](dst, i_dst_stride, src1, i_src_stride, i_height );\
 392 }
 393
 394 MC_LUMA(mmx2,mmx2,mmx)
 395 MC_LUMA(sse2,sse2,sse)
 396 #if HIGH_BIT_DEPTH
 397 MC_LUMA(avx2,avx2,avx)
 398 #else
 399 #if ARCH_X86
 400 MC_LUMA(cache32_mmx2,cache32_mmx2,mmx)
 401 MC_LUMA(cache64_mmx2,cache64_mmx2,mmx)
 402 #endif
 403 MC_LUMA(cache64_sse2,cache64_sse2,sse)
 404 MC_LUMA(cache64_ssse3,cache64_ssse3,sse)
 405 MC_LUMA(cache64_ssse3_atom,cache64_ssse3_atom,sse)
 406 #endif // !HIGH_BIT_DEPTH
 407
 408 #define GET_REF(name)\
 409 static pixel *get_ref_##name( pixel *dst,   intptr_t *i_dst_stride,\
 410                               pixel *src[4], intptr_t i_src_stride,\
 411                               int mvx, int mvy,\
 412                               int i_width, int i_height, const x264_weight_t *weight )\
 413 {\
 414     int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
 415     int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
 416     pixel *src1 = src[x264_hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
 417     if( qpel_idx & 5 ) /* qpel interpolation needed */\
 418     {\
 419         pixel *src2 = src[x264_hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
 420         x264_pixel_avg_wtab_##name[i_width>>2](\
 421                 dst, *i_dst_stride, src1, i_src_stride,\
 422                 src2, i_height );\
 423         if( weight->weightfn )\
 424             weight->weightfn[i_width>>2]( dst, *i_dst_stride, dst, *i_dst_stride, weight, i_height );\
 425         return dst;\
 426     }\
 427     else if( weight->weightfn )\
 428     {\
 429         weight->weightfn[i_width>>2]( dst, *i_dst_stride, src1, i_src_stride, weight, i_height );\
 430         return dst;\
 431     }\
 432     else\
 433     {\
 434         *i_dst_stride = i_src_stride;\
 435         return src1;\
 436     }\
 437 }
 438
 439 GET_REF(mmx2)
 440 GET_REF(sse2)
 441 GET_REF(avx2)
 442 #if !HIGH_BIT_DEPTH
 443 #if ARCH_X86
 444 GET_REF(cache32_mmx2)
 445 GET_REF(cache64_mmx2)
 446 #endif
 447 GET_REF(cache64_sse2)
 448 GET_REF(cache64_ssse3)
 449 GET_REF(cache64_ssse3_atom)
 450 #endif // !HIGH_BIT_DEPTH
 451
 452 #define HPEL(align, cpu, cpuv, cpuc, cpuh)\
 453 void x264_hpel_filter_v_##cpuv( pixel *dst, pixel *src, int16_t *buf, intptr_t stride, intptr_t width);\
 454 void x264_hpel_filter_c_##cpuc( pixel *dst, int16_t *buf, intptr_t width );\
 455 void x264_hpel_filter_h_##cpuh( pixel *dst, pixel *src, intptr_t width );\
 456 static void x264_hpel_filter_##cpu( pixel *dsth, pixel *dstv, pixel *dstc, pixel *src,\
 457                                     intptr_t stride, int width, int height, int16_t *buf )\
 458 {\
 459     intptr_t realign = (intptr_t)src & (align-1);\
 460     src -= realign;\
 461     dstv -= realign;\
 462     dstc -= realign;\
 463     dsth -= realign;\
 464     width += realign;\
 465     while( height-- )\
 466     {\
 467         x264_hpel_filter_v_##cpuv( dstv, src, buf+16, stride, width );\
 468         x264_hpel_filter_c_##cpuc( dstc, buf+16, width );\
 469         x264_hpel_filter_h_##cpuh( dsth, src, width );\
 470         dsth += stride;\
 471         dstv += stride;\
 472         dstc += stride;\
 473         src  += stride;\
 474     }\
 475     x264_sfence();\
 476 }
 477
 478 HPEL(8, mmx2, mmx2, mmx2, mmx2)
 479 #if HIGH_BIT_DEPTH
 480 HPEL(16, sse2, sse2, sse2, sse2)
 481 #else // !HIGH_BIT_DEPTH
 482 HPEL(16, sse2_amd, mmx2, mmx2, sse2)
 483 #if ARCH_X86_64
 484 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 );
 485 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 );
 486 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 );
 487 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 );
 488 #else
 489 HPEL(16, sse2, sse2, sse2, sse2)
 490 HPEL(16, ssse3, ssse3, ssse3, ssse3)
 491 HPEL(16, avx, avx, avx, avx)
 492 HPEL(32, avx2, avx2, avx2, avx2)
 493 #endif
 494 #endif // HIGH_BIT_DEPTH
 495
 496 #define PLANE_COPY(align, cpu)\
 497 static void x264_plane_copy_##cpu( pixel *dst, intptr_t i_dst, pixel *src, intptr_t i_src, int w, int h )\
 498 {\
 499     int c_w = (align) / sizeof(pixel) - 1;\
 500     if( w < 256 ) /* tiny resolutions don't want non-temporal hints. dunno the exact threshold. */\
 501         x264_plane_copy_c( dst, i_dst, src, i_src, w, h );\
 502     else if( !(w&c_w) )\
 503         x264_plane_copy_core_##cpu( dst, i_dst, src, i_src, w, h );\
 504     else\
 505     {\
 506         if( --h > 0 )\
 507         {\
 508             if( i_src > 0 )\
 509             {\
 510                 x264_plane_copy_core_##cpu( dst, i_dst, src, i_src, (w+c_w)&~c_w, h );\
 511                 dst += i_dst * h;\
 512                 src += i_src * h;\
 513             }\
 514             else\
 515                 x264_plane_copy_core_##cpu( dst+i_dst, i_dst, src+i_src, i_src, (w+c_w)&~c_w, h );\
 516         }\
 517         /* use plain memcpy on the last line (in memory order) to avoid overreading src. */\
 518         memcpy( dst, src, w*sizeof(pixel) );\
 519     }\
 520 }
 521
 522 PLANE_COPY(16, sse)
 523 PLANE_COPY(32, avx)
 524
 525 #define PLANE_COPY_SWAP(align, cpu)\
 526 static void x264_plane_copy_swap_##cpu( pixel *dst, intptr_t i_dst, pixel *src, intptr_t i_src, int w, int h )\
 527 {\
 528     int c_w = (align>>1) / sizeof(pixel) - 1;\
 529     if( !(w&c_w) )\
 530         x264_plane_copy_swap_core_##cpu( dst, i_dst, src, i_src, w, h );\
 531     else if( w > c_w )\
 532     {\
 533         if( --h > 0 )\
 534         {\
 535             if( i_src > 0 )\
 536             {\
 537                 x264_plane_copy_swap_core_##cpu( dst, i_dst, src, i_src, (w+c_w)&~c_w, h );\
 538                 dst += i_dst * h;\
 539                 src += i_src * h;\
 540             }\
 541             else\
 542                 x264_plane_copy_swap_core_##cpu( dst+i_dst, i_dst, src+i_src, i_src, (w+c_w)&~c_w, h );\
 543         }\
 544         x264_plane_copy_swap_core_##cpu( dst, 0, src, 0, w&~c_w, 1 );\
 545         for( int x = 2*(w&~c_w); x < 2*w; x += 2 )\
 546         {\
 547             dst[x]   = src[x+1];\
 548             dst[x+1] = src[x];\
 549         }\
 550     }\
 551     else\
 552         x264_plane_copy_swap_c( dst, i_dst, src, i_src, w, h );\
 553 }
 554
 555 PLANE_COPY_SWAP(16, ssse3)
 556 PLANE_COPY_SWAP(32, avx2)
 557
 558 #define PLANE_INTERLEAVE(cpu) \
 559 static void x264_plane_copy_interleave_##cpu( pixel *dst,  intptr_t i_dst,\
 560                                               pixel *srcu, intptr_t i_srcu,\
 561                                               pixel *srcv, intptr_t i_srcv, int w, int h )\
 562 {\
 563     int c_w = 16 / sizeof(pixel) - 1;\
 564     if( !(w&c_w) )\
 565         x264_plane_copy_interleave_core_##cpu( dst, i_dst, srcu, i_srcu, srcv, i_srcv, w, h );\
 566     else if( w > c_w && (i_srcu ^ i_srcv) >= 0 ) /* only works correctly for strides with identical signs */\
 567     {\
 568         if( --h > 0 )\
 569         {\
 570             if( i_srcu > 0 )\
 571             {\
 572                 x264_plane_copy_interleave_core_##cpu( dst, i_dst, srcu, i_srcu, srcv, i_srcv, (w+c_w)&~c_w, h );\
 573                 dst  += i_dst  * h;\
 574                 srcu += i_srcu * h;\
 575                 srcv += i_srcv * h;\
 576             }\
 577             else\
 578                 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 );\
 579         }\
 580         x264_plane_copy_interleave_c( dst, 0, srcu, 0, srcv, 0, w, 1 );\
 581     }\
 582     else\
 583         x264_plane_copy_interleave_c( dst, i_dst, srcu, i_srcu, srcv, i_srcv, w, h );\
 584 }
 585
 586 PLANE_INTERLEAVE(mmx2)
 587 PLANE_INTERLEAVE(sse2)
 588 #if HIGH_BIT_DEPTH
 589 PLANE_INTERLEAVE(avx)
 590 #endif
 591
 592 #if HAVE_X86_INLINE_ASM
 593 #define CLIP_ADD(s,x)\
 594 do\
 595 {\
 596     int temp;\
 597     asm("movd       %0, %%xmm0     \n"\
 598         "movd       %2, %%xmm1     \n"\
 599         "paddsw %%xmm1, %%xmm0     \n"\
 600         "movd   %%xmm0, %1         \n"\
 601         :"+m"(s), "=&r"(temp)\
 602         :"m"(x)\
 603     );\
 604     s = temp;\
 605 } while(0)
 606
 607 #define CLIP_ADD2(s,x)\
 608 do\
 609 {\
 610     asm("movd       %0, %%xmm0     \n"\
 611         "movd       %1, %%xmm1     \n"\
 612         "paddsw %%xmm1, %%xmm0     \n"\
 613         "movd   %%xmm0, %0         \n"\
 614         :"+m"(M32(s))\
 615         :"m"(M32(x))\
 616     );\
 617 } while(0)
 618 #else
 619 #define CLIP_ADD(s,x) (s) = X264_MIN((s)+(x),(1<<15)-1)
 620 #define CLIP_ADD2(s,x)\
 621 do\
 622 {\
 623     CLIP_ADD((s)[0], (x)[0]);\
 624     CLIP_ADD((s)[1], (x)[1]);\
 625 } while(0)
 626 #endif
 627
 628 #define PROPAGATE_LIST(cpu)\
 629 void x264_mbtree_propagate_list_internal_##cpu( int16_t (*mvs)[2], int16_t *propagate_amount,\
 630                                                 uint16_t *lowres_costs, int16_t *output,\
 631                                                 int bipred_weight, int mb_y, int len );\
 632 \
 633 static void x264_mbtree_propagate_list_##cpu( x264_t *h, uint16_t *ref_costs, int16_t (*mvs)[2],\
 634                                               int16_t *propagate_amount, uint16_t *lowres_costs,\
 635                                               int bipred_weight, int mb_y, int len, int list )\
 636 {\
 637     int16_t *current = h->scratch_buffer2;\
 638 \
 639     x264_mbtree_propagate_list_internal_##cpu( mvs, propagate_amount, lowres_costs,\
 640                                                current, bipred_weight, mb_y, len );\
 641 \
 642     unsigned stride = h->mb.i_mb_stride;\
 643     unsigned width = h->mb.i_mb_width;\
 644     unsigned height = h->mb.i_mb_height;\
 645 \
 646     for( unsigned i = 0; i < len; current += 32 )\
 647     {\
 648         int end = X264_MIN( i+8, len );\
 649         for( ; i < end; i++, current += 2 )\
 650         {\
 651             if( !(lowres_costs[i] & (1 << (list+LOWRES_COST_SHIFT))) )\
 652                 continue;\
 653 \
 654             unsigned mbx = current[0];\
 655             unsigned mby = current[1];\
 656             unsigned idx0 = mbx + mby * stride;\
 657             unsigned idx2 = idx0 + stride;\
 658 \
 659             /* Shortcut for the simple/common case of zero MV */\
 660             if( !M32( mvs[i] ) )\
 661             {\
 662                 CLIP_ADD( ref_costs[idx0], current[16] );\
 663                 continue;\
 664             }\
 665 \
 666             if( mbx < width-1 && mby < height-1 )\
 667             {\
 668                 CLIP_ADD2( ref_costs+idx0, current+16 );\
 669                 CLIP_ADD2( ref_costs+idx2, current+32 );\
 670             }\
 671             else\
 672             {\
 673                 /* Note: this takes advantage of unsigned representation to\
 674                  * catch negative mbx/mby. */\
 675                 if( mby < height )\
 676                 {\
 677                     if( mbx < width )\
 678                         CLIP_ADD( ref_costs[idx0+0], current[16] );\
 679                     if( mbx+1 < width )\
 680                         CLIP_ADD( ref_costs[idx0+1], current[17] );\
 681                 }\
 682                 if( mby+1 < height )\
 683                 {\
 684                     if( mbx < width )\
 685                         CLIP_ADD( ref_costs[idx2+0], current[32] );\
 686                     if( mbx+1 < width )\
 687                         CLIP_ADD( ref_costs[idx2+1], current[33] );\
 688                 }\
 689             }\
 690         }\
 691     }\
 692 }
 693
 694 PROPAGATE_LIST(ssse3)
 695 PROPAGATE_LIST(avx)
 696 #undef CLIP_ADD
 697 #undef CLIP_ADD2
 698
 699 void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
 700 {
 701     if( !(cpu&X264_CPU_MMX) )
 702         return;
 703
 704     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_mmx;
 705     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_mmx;
 706
 707     pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_mmx;
 708
 709     pf->copy_16x16_unaligned = x264_mc_copy_w16_mmx;
 710     pf->copy[PIXEL_16x16] = x264_mc_copy_w16_mmx;
 711     pf->copy[PIXEL_8x8]   = x264_mc_copy_w8_mmx;
 712     pf->copy[PIXEL_4x4]   = x264_mc_copy_w4_mmx;
 713     pf->memcpy_aligned  = x264_memcpy_aligned_mmx;
 714     pf->memzero_aligned = x264_memzero_aligned_mmx;
 715     pf->integral_init4v = x264_integral_init4v_mmx;
 716     pf->integral_init8v = x264_integral_init8v_mmx;
 717
 718     if( !(cpu&X264_CPU_MMX2) )
 719         return;
 720
 721     pf->prefetch_fenc_420 = x264_prefetch_fenc_420_mmx2;
 722     pf->prefetch_fenc_422 = x264_prefetch_fenc_422_mmx2;
 723     pf->prefetch_ref  = x264_prefetch_ref_mmx2;
 724
 725     pf->plane_copy_interleave = x264_plane_copy_interleave_mmx2;
 726     pf->store_interleave_chroma = x264_store_interleave_chroma_mmx2;
 727
 728     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_mmx2;
 729     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_mmx2;
 730     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_mmx2;
 731     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_mmx2;
 732     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_mmx2;
 733     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_mmx2;
 734     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_mmx2;
 735     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_mmx2;
 736     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_mmx2;
 737
 738     pf->mc_luma = mc_luma_mmx2;
 739     pf->get_ref = get_ref_mmx2;
 740     pf->mc_chroma = x264_mc_chroma_mmx2;
 741     pf->hpel_filter = x264_hpel_filter_mmx2;
 742     pf->weight = x264_mc_weight_wtab_mmx2;
 743     pf->weight_cache = x264_weight_cache_mmx2;
 744     pf->offsetadd = x264_mc_offsetadd_wtab_mmx2;
 745     pf->offsetsub = x264_mc_offsetsub_wtab_mmx2;
 746
 747     pf->frame_init_lowres_core = x264_frame_init_lowres_core_mmx2;
 748
 749     if( cpu&X264_CPU_SSE )
 750     {
 751         pf->memcpy_aligned  = x264_memcpy_aligned_sse;
 752         pf->memzero_aligned = x264_memzero_aligned_sse;
 753         pf->plane_copy = x264_plane_copy_sse;
 754     }
 755
 756 #if HIGH_BIT_DEPTH
 757 #if ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
 758     if( cpu&(X264_CPU_CACHELINE_32|X264_CPU_CACHELINE_64) )
 759         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
 760 #endif
 761
 762     if( !(cpu&X264_CPU_SSE2) )
 763         return;
 764
 765     pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
 766
 767     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_sse2;
 768     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_sse2;
 769
 770     pf->plane_copy_interleave   = x264_plane_copy_interleave_sse2;
 771     pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
 772
 773     if( cpu&X264_CPU_SSE2_IS_FAST )
 774     {
 775         pf->get_ref = get_ref_sse2;
 776         pf->mc_luma = mc_luma_sse2;
 777         pf->hpel_filter = x264_hpel_filter_sse2;
 778     }
 779
 780     pf->integral_init4v = x264_integral_init4v_sse2;
 781     pf->integral_init8v = x264_integral_init8v_sse2;
 782     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
 783     pf->store_interleave_chroma = x264_store_interleave_chroma_sse2;
 784     pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
 785     pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
 786
 787     if( cpu&X264_CPU_SSE2_IS_SLOW )
 788         return;
 789
 790     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
 791     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_sse2;
 792     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_sse2;
 793     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_sse2;
 794     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_sse2;
 795     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_sse2;
 796     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_sse2;
 797     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_sse2;
 798     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_sse2;
 799
 800     pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse;
 801     pf->weight = x264_mc_weight_wtab_sse2;
 802
 803     if( !(cpu&X264_CPU_STACK_MOD4) )
 804         pf->mc_chroma = x264_mc_chroma_sse2;
 805
 806     if( !(cpu&X264_CPU_SSSE3) )
 807         return;
 808
 809     pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
 810     pf->plane_copy_swap = x264_plane_copy_swap_ssse3;
 811     pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_ssse3;
 812     pf->mbtree_propagate_list = x264_mbtree_propagate_list_ssse3;
 813
 814     if( !(cpu&(X264_CPU_SLOW_SHUFFLE|X264_CPU_SLOW_ATOM|X264_CPU_SLOW_PALIGNR)) )
 815         pf->integral_init4v = x264_integral_init4v_ssse3;
 816
 817     if( !(cpu&X264_CPU_AVX) )
 818         return;
 819
 820     pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx;
 821     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_avx;
 822     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_avx;
 823     pf->plane_copy_interleave        = x264_plane_copy_interleave_avx;
 824     pf->plane_copy_deinterleave      = x264_plane_copy_deinterleave_avx;
 825     pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_avx;
 826     pf->store_interleave_chroma      = x264_store_interleave_chroma_avx;
 827     pf->copy[PIXEL_16x16]            = x264_mc_copy_w16_aligned_avx;
 828
 829     if( !(cpu&X264_CPU_STACK_MOD4) )
 830         pf->mc_chroma = x264_mc_chroma_avx;
 831
 832     if( cpu&X264_CPU_XOP )
 833         pf->frame_init_lowres_core = x264_frame_init_lowres_core_xop;
 834
 835     if( cpu&X264_CPU_AVX2 )
 836     {
 837         pf->mc_luma = mc_luma_avx2;
 838         pf->plane_copy_deinterleave_v210 = x264_plane_copy_deinterleave_v210_avx2;
 839     }
 840 #else // !HIGH_BIT_DEPTH
 841
 842 #if ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
 843     if( cpu&X264_CPU_CACHELINE_32 )
 844     {
 845         pf->mc_luma = mc_luma_cache32_mmx2;
 846         pf->get_ref = get_ref_cache32_mmx2;
 847         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
 848     }
 849     else if( cpu&X264_CPU_CACHELINE_64 )
 850     {
 851         pf->mc_luma = mc_luma_cache64_mmx2;
 852         pf->get_ref = get_ref_cache64_mmx2;
 853         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
 854     }
 855 #endif
 856
 857     if( !(cpu&X264_CPU_SSE2) )
 858         return;
 859
 860     pf->integral_init4v = x264_integral_init4v_sse2;
 861     pf->integral_init8v = x264_integral_init8v_sse2;
 862     pf->hpel_filter = x264_hpel_filter_sse2_amd;
 863     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
 864     pf->plane_copy_deinterleave_rgb = x264_plane_copy_deinterleave_rgb_sse2;
 865
 866     if( !(cpu&X264_CPU_SSE2_IS_SLOW) )
 867     {
 868         pf->weight = x264_mc_weight_wtab_sse2;
 869         if( !(cpu&X264_CPU_SLOW_ATOM) )
 870         {
 871             pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
 872             pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
 873         }
 874
 875         pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse;
 876         pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
 877         pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_sse2;
 878         pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_sse2;
 879         pf->avg[PIXEL_8x8]  = x264_pixel_avg_8x8_sse2;
 880         pf->avg[PIXEL_8x4]  = x264_pixel_avg_8x4_sse2;
 881         pf->hpel_filter = x264_hpel_filter_sse2;
 882         pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
 883         if( !(cpu&X264_CPU_STACK_MOD4) )
 884             pf->mc_chroma = x264_mc_chroma_sse2;
 885
 886         if( cpu&X264_CPU_SSE2_IS_FAST )
 887         {
 888             pf->store_interleave_chroma = x264_store_interleave_chroma_sse2; // FIXME sse2fast? sse2medium?
 889             pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_sse2;
 890             pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_sse2;
 891             pf->plane_copy_interleave   = x264_plane_copy_interleave_sse2;
 892             pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
 893             pf->mc_luma = mc_luma_sse2;
 894             pf->get_ref = get_ref_sse2;
 895             if( cpu&X264_CPU_CACHELINE_64 )
 896             {
 897                 pf->mc_luma = mc_luma_cache64_sse2;
 898                 pf->get_ref = get_ref_cache64_sse2;
 899             }
 900         }
 901     }
 902
 903     if( !(cpu&X264_CPU_SSSE3) )
 904         return;
 905
 906     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_ssse3;
 907     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_ssse3;
 908     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_ssse3;
 909     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_ssse3;
 910     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_ssse3;
 911     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_ssse3;
 912     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_ssse3;
 913     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_ssse3;
 914     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_ssse3;
 915     pf->plane_copy_swap = x264_plane_copy_swap_ssse3;
 916     pf->plane_copy_deinterleave_rgb = x264_plane_copy_deinterleave_rgb_ssse3;
 917     pf->mbtree_propagate_list = x264_mbtree_propagate_list_ssse3;
 918
 919     if( !(cpu&X264_CPU_SLOW_PSHUFB) )
 920     {
 921         pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_ssse3;
 922         pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_ssse3;
 923         pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_ssse3;
 924     }
 925
 926     if( !(cpu&X264_CPU_SLOW_PALIGNR) )
 927     {
 928 #if ARCH_X86_64
 929         if( !(cpu&X264_CPU_SLOW_ATOM) ) /* The 64-bit version is slower, but the 32-bit version is faster? */
 930 #endif
 931             pf->hpel_filter = x264_hpel_filter_ssse3;
 932         pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
 933     }
 934     if( !(cpu&X264_CPU_STACK_MOD4) )
 935         pf->mc_chroma = x264_mc_chroma_ssse3;
 936
 937     if( cpu&X264_CPU_CACHELINE_64 )
 938     {
 939         if( !(cpu&X264_CPU_STACK_MOD4) )
 940             pf->mc_chroma = x264_mc_chroma_ssse3_cache64;
 941         pf->mc_luma = mc_luma_cache64_ssse3;
 942         pf->get_ref = get_ref_cache64_ssse3;
 943         if( cpu&X264_CPU_SLOW_ATOM )
 944         {
 945             pf->mc_luma = mc_luma_cache64_ssse3_atom;
 946             pf->get_ref = get_ref_cache64_ssse3_atom;
 947         }
 948     }
 949
 950     pf->weight_cache = x264_weight_cache_ssse3;
 951     pf->weight = x264_mc_weight_wtab_ssse3;
 952
 953     if( !(cpu&(X264_CPU_SLOW_SHUFFLE|X264_CPU_SLOW_ATOM|X264_CPU_SLOW_PALIGNR)) )
 954         pf->integral_init4v = x264_integral_init4v_ssse3;
 955
 956     if( !(cpu&X264_CPU_SSE4) )
 957         return;
 958
 959     pf->integral_init4h = x264_integral_init4h_sse4;
 960     pf->integral_init8h = x264_integral_init8h_sse4;
 961
 962     if( !(cpu&X264_CPU_AVX) )
 963         return;
 964
 965     pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx;
 966     pf->integral_init8h = x264_integral_init8h_avx;
 967     pf->hpel_filter = x264_hpel_filter_avx;
 968
 969     if( !(cpu&X264_CPU_STACK_MOD4) )
 970         pf->mc_chroma = x264_mc_chroma_avx;
 971
 972     if( cpu&X264_CPU_XOP )
 973         pf->frame_init_lowres_core = x264_frame_init_lowres_core_xop;
 974
 975     if( cpu&X264_CPU_AVX2 )
 976     {
 977         pf->hpel_filter = x264_hpel_filter_avx2;
 978         pf->mc_chroma = x264_mc_chroma_avx2;
 979         pf->weight = x264_mc_weight_wtab_avx2;
 980         pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_avx2;
 981         pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_avx2;
 982         pf->integral_init8v = x264_integral_init8v_avx2;
 983         pf->integral_init4v = x264_integral_init4v_avx2;
 984         pf->integral_init8h = x264_integral_init8h_avx2;
 985         pf->integral_init4h = x264_integral_init4h_avx2;
 986         pf->frame_init_lowres_core = x264_frame_init_lowres_core_avx2;
 987     }
 988 #endif // HIGH_BIT_DEPTH
 989
 990     if( !(cpu&X264_CPU_AVX) )
 991         return;
 992     pf->memzero_aligned = x264_memzero_aligned_avx;
 993     pf->plane_copy = x264_plane_copy_avx;
 994     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_avx;
 995     pf->mbtree_propagate_list = x264_mbtree_propagate_list_avx;
 996
 997     if( cpu&X264_CPU_FMA4 )
 998         pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_fma4;
 999
1000     if( !(cpu&X264_CPU_AVX2) )
1001         return;
1002     pf->plane_copy_swap = x264_plane_copy_swap_avx2;
1003     pf->get_ref = get_ref_avx2;
1004     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_avx2;
1005 }