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

   1 /*****************************************************************************
   2  * mc-c.c: x86 motion compensation
   3  *****************************************************************************
   4  * Copyright (C) 2003-2011 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
  40 DECL_SUF( x264_pixel_avg_16x16, ( pixel *, int, pixel *, int, pixel *, int, int ))
  41 DECL_SUF( x264_pixel_avg_16x8,  ( pixel *, int, pixel *, int, pixel *, int, int ))
  42 DECL_SUF( x264_pixel_avg_8x16,  ( pixel *, int, pixel *, int, pixel *, int, int ))
  43 DECL_SUF( x264_pixel_avg_8x8,   ( pixel *, int, pixel *, int, pixel *, int, int ))
  44 DECL_SUF( x264_pixel_avg_8x4,   ( pixel *, int, pixel *, int, pixel *, int, int ))
  45 DECL_SUF( x264_pixel_avg_4x16,  ( pixel *, int, pixel *, int, pixel *, int, int ))
  46 DECL_SUF( x264_pixel_avg_4x8,   ( pixel *, int, pixel *, int, pixel *, int, int ))
  47 DECL_SUF( x264_pixel_avg_4x4,   ( pixel *, int, pixel *, int, pixel *, int, int ))
  48 DECL_SUF( x264_pixel_avg_4x2,   ( pixel *, int, pixel *, int, pixel *, int, int ))
  49
  50 #define MC_WEIGHT(w,type) \
  51     void x264_mc_weight_w##w##_##type( pixel *,int, pixel *,int, const x264_weight_t *,int );
  52
  53 #define MC_WEIGHT_OFFSET(w,type) \
  54     void x264_mc_offsetadd_w##w##_##type( pixel *,int, pixel *,int, const x264_weight_t *,int ); \
  55     void x264_mc_offsetsub_w##w##_##type( pixel *,int, pixel *,int, const x264_weight_t *,int ); \
  56     MC_WEIGHT(w,type)
  57
  58 MC_WEIGHT_OFFSET( 4, mmx2 )
  59 MC_WEIGHT_OFFSET( 8, mmx2 )
  60 MC_WEIGHT_OFFSET( 12, mmx2 )
  61 MC_WEIGHT_OFFSET( 16, mmx2 )
  62 MC_WEIGHT_OFFSET( 20, mmx2 )
  63 MC_WEIGHT_OFFSET( 12, sse2 )
  64 MC_WEIGHT_OFFSET( 16, sse2 )
  65 MC_WEIGHT_OFFSET( 20, sse2 )
  66 #if HIGH_BIT_DEPTH
  67 MC_WEIGHT_OFFSET( 8, sse2 )
  68 #endif
  69 MC_WEIGHT( 8, sse2  )
  70 MC_WEIGHT( 4, ssse3 )
  71 MC_WEIGHT( 8, ssse3 )
  72 MC_WEIGHT( 12, ssse3 )
  73 MC_WEIGHT( 16, ssse3 )
  74 MC_WEIGHT( 20, ssse3 )
  75 MC_WEIGHT( 4, avx )
  76 MC_WEIGHT( 8, avx )
  77 MC_WEIGHT( 12, avx )
  78 MC_WEIGHT( 16, avx )
  79 MC_WEIGHT( 20, avx )
  80 #undef MC_OFFSET
  81 #undef MC_WEIGHT
  82
  83 void x264_mc_copy_w4_mmx( pixel *, int, pixel *, int, int );
  84 void x264_mc_copy_w8_mmx( pixel *, int, pixel *, int, int );
  85 void x264_mc_copy_w8_sse2( pixel *, int, pixel *, int, int );
  86 void x264_mc_copy_w16_mmx( pixel *, int, pixel *, int, int );
  87 void x264_mc_copy_w16_sse2( pixel *, int, pixel *, int, int );
  88 void x264_mc_copy_w16_aligned_sse2( pixel *, int, pixel *, int, int );
  89 void x264_prefetch_fenc_420_mmx2( pixel *, int, pixel *, int, int );
  90 void x264_prefetch_fenc_422_mmx2( pixel *, int, pixel *, int, int );
  91 void x264_prefetch_ref_mmx2( pixel *, int, int );
  92 void x264_plane_copy_core_mmx2( pixel *, int, pixel *, int, int w, int h);
  93 void x264_plane_copy_c( pixel *, int, pixel *, int, int w, int h );
  94 void x264_plane_copy_interleave_core_mmx2( pixel *dst, int i_dst,
  95                                              pixel *srcu, int i_srcu,
  96                                              pixel *srcv, int i_srcv, int w, int h );
  97 void x264_plane_copy_interleave_core_sse2( pixel *dst, int i_dst,
  98                                            pixel *srcu, int i_srcu,
  99                                            pixel *srcv, int i_srcv, int w, int h );
 100 void x264_plane_copy_interleave_core_avx( pixel *dst, int i_dst,
 101                                            pixel *srcu, int i_srcu,
 102                                            pixel *srcv, int i_srcv, int w, int h );
 103 void x264_plane_copy_interleave_c( pixel *dst, int i_dst,
 104                                    pixel *srcu, int i_srcu,
 105                                    pixel *srcv, int i_srcv, int w, int h );
 106 void x264_plane_copy_deinterleave_mmx( pixel *dstu, int i_dstu,
 107                                        pixel *dstv, int i_dstv,
 108                                        pixel *src, int i_src, int w, int h );
 109 void x264_plane_copy_deinterleave_sse2( pixel *dstu, int i_dstu,
 110                                         pixel *dstv, int i_dstv,
 111                                         pixel *src, int i_src, int w, int h );
 112 void x264_plane_copy_deinterleave_ssse3( uint8_t *dstu, int i_dstu,
 113                                          uint8_t *dstv, int i_dstv,
 114                                          uint8_t *src, int i_src, int w, int h );
 115 void x264_plane_copy_deinterleave_avx( uint16_t *dstu, int i_dstu,
 116                                          uint16_t *dstv, int i_dstv,
 117                                          uint16_t *src, int i_src, int w, int h );
 118 void x264_store_interleave_chroma_mmx2( pixel *dst, int i_dst, pixel *srcu, pixel *srcv, int height );
 119 void x264_store_interleave_chroma_sse2( pixel *dst, int i_dst, pixel *srcu, pixel *srcv, int height );
 120 void x264_store_interleave_chroma_avx( pixel *dst, int i_dst, pixel *srcu, pixel *srcv, int height );
 121 void x264_load_deinterleave_chroma_fenc_mmx( pixel *dst, pixel *src, int i_src, int height );
 122 void x264_load_deinterleave_chroma_fenc_sse2( pixel *dst, pixel *src, int i_src, int height );
 123 void x264_load_deinterleave_chroma_fenc_ssse3( uint8_t *dst, uint8_t *src, int i_src, int height );
 124 void x264_load_deinterleave_chroma_fenc_avx( uint16_t *dst, uint16_t *src, int i_src, int height );
 125 void x264_load_deinterleave_chroma_fdec_mmx( pixel *dst, pixel *src, int i_src, int height );
 126 void x264_load_deinterleave_chroma_fdec_sse2( pixel *dst, pixel *src, int i_src, int height );
 127 void x264_load_deinterleave_chroma_fdec_ssse3( uint8_t *dst, uint8_t *src, int i_src, int height );
 128 void x264_load_deinterleave_chroma_fdec_avx( uint16_t *dst, uint16_t *src, int i_src, int height );
 129 void *x264_memcpy_aligned_mmx( void * dst, const void * src, size_t n );
 130 void *x264_memcpy_aligned_sse2( void * dst, const void * src, size_t n );
 131 void x264_memzero_aligned_mmx( void * dst, int n );
 132 void x264_memzero_aligned_sse2( void * dst, int n );
 133 void x264_integral_init4h_sse4( uint16_t *sum, uint8_t *pix, int stride );
 134 void x264_integral_init8h_sse4( uint16_t *sum, uint8_t *pix, int stride );
 135 void x264_integral_init8h_avx ( uint16_t *sum, uint8_t *pix, int stride );
 136 void x264_integral_init4v_mmx( uint16_t *sum8, uint16_t *sum4, int stride );
 137 void x264_integral_init4v_sse2( uint16_t *sum8, uint16_t *sum4, int stride );
 138 void x264_integral_init8v_mmx( uint16_t *sum8, int stride );
 139 void x264_integral_init8v_sse2( uint16_t *sum8, int stride );
 140 void x264_integral_init4v_ssse3( uint16_t *sum8, uint16_t *sum4, int stride );
 141 void x264_mbtree_propagate_cost_sse2( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
 142                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
 143 void x264_mbtree_propagate_cost_avx( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
 144                                      uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
 145 void x264_mbtree_propagate_cost_fma4( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
 146                                       uint16_t *inter_costs, uint16_t *inv_qscales, float *fps_factor, int len );
 147
 148 #define MC_CHROMA(cpu)\
 149 void x264_mc_chroma_##cpu( pixel *dstu, pixel *dstv, int i_dst,\
 150                            pixel *src, int i_src,\
 151                            int dx, int dy, int i_width, int i_height );
 152 MC_CHROMA(mmx2)
 153 MC_CHROMA(sse2)
 154 MC_CHROMA(sse2_misalign)
 155 MC_CHROMA(ssse3)
 156 MC_CHROMA(ssse3_cache64)
 157 MC_CHROMA(avx)
 158 MC_CHROMA(avx_cache64)
 159
 160 #define LOWRES(cpu)\
 161 void x264_frame_init_lowres_core_##cpu( pixel *src0, pixel *dst0, pixel *dsth, pixel *dstv, pixel *dstc,\
 162                                         int src_stride, int dst_stride, int width, int height );
 163 LOWRES(mmx2)
 164 LOWRES(cache32_mmx2)
 165 LOWRES(sse2)
 166 LOWRES(ssse3)
 167
 168 #define PIXEL_AVG_W(width,cpu)\
 169 void x264_pixel_avg2_w##width##_##cpu( pixel *, int, pixel *, int, pixel *, int );
 170 /* This declares some functions that don't exist, but that isn't a problem. */
 171 #define PIXEL_AVG_WALL(cpu)\
 172 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);
 173
 174 PIXEL_AVG_WALL(mmx2)
 175 PIXEL_AVG_WALL(cache32_mmx2)
 176 PIXEL_AVG_WALL(cache64_mmx2)
 177 PIXEL_AVG_WALL(cache64_sse2)
 178 PIXEL_AVG_WALL(sse2)
 179 PIXEL_AVG_WALL(sse2_misalign)
 180 PIXEL_AVG_WALL(cache64_ssse3)
 181
 182 #define PIXEL_AVG_WTAB(instr, name1, name2, name3, name4, name5)\
 183 static void (* const x264_pixel_avg_wtab_##instr[6])( pixel *, int, pixel *, int, pixel *, int ) =\
 184 {\
 185     NULL,\
 186     x264_pixel_avg2_w4_##name1,\
 187     x264_pixel_avg2_w8_##name2,\
 188     x264_pixel_avg2_w12_##name3,\
 189     x264_pixel_avg2_w16_##name4,\
 190     x264_pixel_avg2_w20_##name5,\
 191 };
 192
 193 #if HIGH_BIT_DEPTH
 194 /* we can replace w12/w20 with w10/w18 as only 9/17 pixels in fact are important */
 195 #define x264_pixel_avg2_w12_mmx2       x264_pixel_avg2_w10_mmx2
 196 #define x264_pixel_avg2_w20_mmx2       x264_pixel_avg2_w18_mmx2
 197 #define x264_pixel_avg2_w12_sse2         x264_pixel_avg2_w10_sse2
 198 #define x264_pixel_avg2_w20_sse2         x264_pixel_avg2_w18_sse2
 199 #else
 200 /* w16 sse2 is faster than w12 mmx as long as the cacheline issue is resolved */
 201 #define x264_pixel_avg2_w12_cache64_ssse3 x264_pixel_avg2_w16_cache64_ssse3
 202 #define x264_pixel_avg2_w12_cache64_sse2 x264_pixel_avg2_w16_cache64_sse2
 203 #define x264_pixel_avg2_w12_sse3         x264_pixel_avg2_w16_sse3
 204 #define x264_pixel_avg2_w12_sse2         x264_pixel_avg2_w16_sse2
 205 #endif // HIGH_BIT_DEPTH
 206
 207 PIXEL_AVG_WTAB(mmx2, mmx2, mmx2, mmx2, mmx2, mmx2)
 208 #if HIGH_BIT_DEPTH
 209 PIXEL_AVG_WTAB(sse2, mmx2, sse2, sse2, sse2, sse2)
 210 #else // !HIGH_BIT_DEPTH
 211 #if ARCH_X86
 212 PIXEL_AVG_WTAB(cache32_mmx2, mmx2, cache32_mmx2, cache32_mmx2, cache32_mmx2, cache32_mmx2)
 213 PIXEL_AVG_WTAB(cache64_mmx2, mmx2, cache64_mmx2, cache64_mmx2, cache64_mmx2, cache64_mmx2)
 214 #endif
 215 PIXEL_AVG_WTAB(sse2, mmx2, mmx2, sse2, sse2, sse2)
 216 PIXEL_AVG_WTAB(sse2_misalign, mmx2, mmx2, sse2, sse2, sse2_misalign)
 217 PIXEL_AVG_WTAB(cache64_sse2, mmx2, cache64_mmx2, cache64_sse2, cache64_sse2, cache64_sse2)
 218 PIXEL_AVG_WTAB(cache64_ssse3, mmx2, cache64_mmx2, cache64_ssse3, cache64_ssse3, cache64_sse2)
 219 #endif // HIGH_BIT_DEPTH
 220
 221 #define MC_COPY_WTAB(instr, name1, name2, name3)\
 222 static void (* const x264_mc_copy_wtab_##instr[5])( pixel *, int, pixel *, int, int ) =\
 223 {\
 224     NULL,\
 225     x264_mc_copy_w4_##name1,\
 226     x264_mc_copy_w8_##name2,\
 227     NULL,\
 228     x264_mc_copy_w16_##name3,\
 229 };
 230
 231 MC_COPY_WTAB(mmx,mmx,mmx,mmx)
 232 #if HIGH_BIT_DEPTH
 233 MC_COPY_WTAB(sse2,mmx,sse2,sse2)
 234 #else
 235 MC_COPY_WTAB(sse2,mmx,mmx,sse2)
 236 #endif
 237
 238 #define MC_WEIGHT_WTAB(function, instr, name1, name2, w12version)\
 239     static void (* x264_mc_##function##_wtab_##instr[6])( pixel *, int, pixel *, int, const x264_weight_t *, int ) =\
 240 {\
 241     x264_mc_##function##_w4_##name1,\
 242     x264_mc_##function##_w4_##name1,\
 243     x264_mc_##function##_w8_##name2,\
 244     x264_mc_##function##_w##w12version##_##instr,\
 245     x264_mc_##function##_w16_##instr,\
 246     x264_mc_##function##_w20_##instr,\
 247 };
 248
 249 #if HIGH_BIT_DEPTH
 250 MC_WEIGHT_WTAB(weight,mmx2,mmx2,mmx2,12)
 251 MC_WEIGHT_WTAB(offsetadd,mmx2,mmx2,mmx2,12)
 252 MC_WEIGHT_WTAB(offsetsub,mmx2,mmx2,mmx2,12)
 253 MC_WEIGHT_WTAB(weight,sse2,mmx2,sse2,12)
 254 MC_WEIGHT_WTAB(offsetadd,sse2,mmx2,sse2,16)
 255 MC_WEIGHT_WTAB(offsetsub,sse2,mmx2,sse2,16)
 256
 257 static void x264_weight_cache_mmx2( x264_t *h, x264_weight_t *w )
 258 {
 259     if( w->i_scale == 1<<w->i_denom )
 260     {
 261         if( w->i_offset < 0 )
 262             w->weightfn = h->mc.offsetsub;
 263         else
 264             w->weightfn = h->mc.offsetadd;
 265         for( int i = 0; i < 8; i++ )
 266             w->cachea[i] = abs(w->i_offset<<(BIT_DEPTH-8));
 267         return;
 268     }
 269     w->weightfn = h->mc.weight;
 270     int den1 = 1<<w->i_denom;
 271     int den2 = w->i_scale<<1;
 272     int den3 = 1+(w->i_offset<<(BIT_DEPTH-8+1));
 273     for( int i = 0; i < 8; i++ )
 274     {
 275         w->cachea[i] = den1;
 276         w->cacheb[i] = i&1 ? den3 : den2;
 277     }
 278 }
 279 #else
 280 MC_WEIGHT_WTAB(weight,mmx2,mmx2,mmx2,12)
 281 MC_WEIGHT_WTAB(offsetadd,mmx2,mmx2,mmx2,12)
 282 MC_WEIGHT_WTAB(offsetsub,mmx2,mmx2,mmx2,12)
 283 MC_WEIGHT_WTAB(weight,sse2,mmx2,sse2,16)
 284 MC_WEIGHT_WTAB(offsetadd,sse2,mmx2,mmx2,16)
 285 MC_WEIGHT_WTAB(offsetsub,sse2,mmx2,mmx2,16)
 286 MC_WEIGHT_WTAB(weight,ssse3,ssse3,ssse3,16)
 287
 288 static void x264_weight_cache_mmx2( x264_t *h, x264_weight_t *w )
 289 {
 290     int i;
 291     int16_t den1;
 292
 293     if( w->i_scale == 1<<w->i_denom )
 294     {
 295         if( w->i_offset < 0 )
 296             w->weightfn = h->mc.offsetsub;
 297         else
 298             w->weightfn = h->mc.offsetadd;
 299         memset( w->cachea, abs(w->i_offset), sizeof(w->cachea) );
 300         return;
 301     }
 302     w->weightfn = h->mc.weight;
 303     den1 = 1 << (w->i_denom - 1) | w->i_offset << w->i_denom;
 304     for( i = 0; i < 8; i++ )
 305     {
 306         w->cachea[i] = w->i_scale;
 307         w->cacheb[i] = den1;
 308     }
 309 }
 310
 311 static void x264_weight_cache_ssse3( x264_t *h, x264_weight_t *w )
 312 {
 313     int i, den1;
 314     if( w->i_scale == 1<<w->i_denom )
 315     {
 316         if( w->i_offset < 0 )
 317             w->weightfn = h->mc.offsetsub;
 318         else
 319             w->weightfn = h->mc.offsetadd;
 320
 321         memset( w->cachea, abs( w->i_offset ), sizeof(w->cachea) );
 322         return;
 323     }
 324     w->weightfn = h->mc.weight;
 325     den1 = w->i_scale << (8 - w->i_denom);
 326     for( i = 0; i < 8; i++ )
 327     {
 328         w->cachea[i] = den1;
 329         w->cacheb[i] = w->i_offset;
 330     }
 331 }
 332 #endif // !HIGH_BIT_DEPTH
 333
 334 static const uint8_t hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
 335 static const uint8_t hpel_ref1[16] = {0,0,0,0,2,2,3,2,2,2,3,2,2,2,3,2};
 336
 337 #define MC_LUMA(name,instr1,instr2)\
 338 static void mc_luma_##name( pixel *dst,    int i_dst_stride,\
 339                   pixel *src[4], int i_src_stride,\
 340                   int mvx, int mvy,\
 341                   int i_width, int i_height, const x264_weight_t *weight )\
 342 {\
 343     int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
 344     int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
 345     pixel *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
 346     if( qpel_idx & 5 ) /* qpel interpolation needed */\
 347     {\
 348         pixel *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
 349         x264_pixel_avg_wtab_##instr1[i_width>>2](\
 350                 dst, i_dst_stride, src1, i_src_stride,\
 351                 src2, i_height );\
 352         if( weight->weightfn )\
 353             weight->weightfn[i_width>>2]( dst, i_dst_stride, dst, i_dst_stride, weight, i_height );\
 354     }\
 355     else if( weight->weightfn )\
 356         weight->weightfn[i_width>>2]( dst, i_dst_stride, src1, i_src_stride, weight, i_height );\
 357     else\
 358         x264_mc_copy_wtab_##instr2[i_width>>2](dst, i_dst_stride, src1, i_src_stride, i_height );\
 359 }
 360
 361 MC_LUMA(mmx2,mmx2,mmx)
 362 MC_LUMA(sse2,sse2,sse2)
 363 #if !HIGH_BIT_DEPTH
 364 #if ARCH_X86
 365 MC_LUMA(cache32_mmx2,cache32_mmx2,mmx)
 366 MC_LUMA(cache64_mmx2,cache64_mmx2,mmx)
 367 #endif
 368 MC_LUMA(cache64_sse2,cache64_sse2,sse2)
 369 MC_LUMA(cache64_ssse3,cache64_ssse3,sse2)
 370 #endif // !HIGH_BIT_DEPTH
 371
 372 #define GET_REF(name)\
 373 static pixel *get_ref_##name( pixel *dst,   int *i_dst_stride,\
 374                          pixel *src[4], int i_src_stride,\
 375                          int mvx, int mvy,\
 376                          int i_width, int i_height, const x264_weight_t *weight )\
 377 {\
 378     int qpel_idx = ((mvy&3)<<2) + (mvx&3);\
 379     int offset = (mvy>>2)*i_src_stride + (mvx>>2);\
 380     pixel *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;\
 381     if( qpel_idx & 5 ) /* qpel interpolation needed */\
 382     {\
 383         pixel *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);\
 384         x264_pixel_avg_wtab_##name[i_width>>2](\
 385                 dst, *i_dst_stride, src1, i_src_stride,\
 386                 src2, i_height );\
 387         if( weight->weightfn )\
 388             weight->weightfn[i_width>>2]( dst, *i_dst_stride, dst, *i_dst_stride, weight, i_height );\
 389         return dst;\
 390     }\
 391     else if( weight->weightfn )\
 392     {\
 393         weight->weightfn[i_width>>2]( dst, *i_dst_stride, src1, i_src_stride, weight, i_height );\
 394         return dst;\
 395     }\
 396     else\
 397     {\
 398         *i_dst_stride = i_src_stride;\
 399         return src1;\
 400     }\
 401 }
 402
 403 GET_REF(mmx2)
 404 GET_REF(sse2)
 405 #if !HIGH_BIT_DEPTH
 406 #if ARCH_X86
 407 GET_REF(cache32_mmx2)
 408 GET_REF(cache64_mmx2)
 409 #endif
 410 GET_REF(sse2_misalign)
 411 GET_REF(cache64_sse2)
 412 GET_REF(cache64_ssse3)
 413 #endif // !HIGH_BIT_DEPTH
 414
 415 #define HPEL(align, cpu, cpuv, cpuc, cpuh)\
 416 void x264_hpel_filter_v_##cpuv( pixel *dst, pixel *src, int16_t *buf, int stride, int width);\
 417 void x264_hpel_filter_c_##cpuc( pixel *dst, int16_t *buf, int width );\
 418 void x264_hpel_filter_h_##cpuh( pixel *dst, pixel *src, int width );\
 419 static void x264_hpel_filter_##cpu( pixel *dsth, pixel *dstv, pixel *dstc, pixel *src,\
 420                              int stride, int width, int height, int16_t *buf )\
 421 {\
 422     int realign = (intptr_t)src & (align-1);\
 423     src -= realign;\
 424     dstv -= realign;\
 425     dstc -= realign;\
 426     dsth -= realign;\
 427     width += realign;\
 428     while( height-- )\
 429     {\
 430         x264_hpel_filter_v_##cpuv( dstv, src, buf+8, stride, width );\
 431         x264_hpel_filter_c_##cpuc( dstc, buf+8, width );\
 432         x264_hpel_filter_h_##cpuh( dsth, src, width );\
 433         dsth += stride;\
 434         dstv += stride;\
 435         dstc += stride;\
 436         src  += stride;\
 437     }\
 438     x264_sfence();\
 439 }
 440
 441 HPEL(8, mmx2, mmx2, mmx2, mmx2)
 442 #if HIGH_BIT_DEPTH
 443 HPEL(16, sse2, sse2, sse2, sse2)
 444 #else // !HIGH_BIT_DEPTH
 445 HPEL(16, sse2_amd, mmx2, mmx2, sse2)
 446 #if ARCH_X86_64
 447 void x264_hpel_filter_sse2( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, int stride, int width, int height, int16_t *buf );
 448 void x264_hpel_filter_ssse3( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, int stride, int width, int height, int16_t *buf );
 449 void x264_hpel_filter_avx( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src, int stride, int width, int height, int16_t *buf );
 450 #else
 451 HPEL(16, sse2, sse2, sse2, sse2)
 452 HPEL(16, ssse3, ssse3, ssse3, ssse3)
 453 HPEL(16, avx, avx, avx, ssse3)
 454 #endif
 455 HPEL(16, sse2_misalign, sse2, sse2_misalign, sse2)
 456 #endif // HIGH_BIT_DEPTH
 457
 458 static void x264_plane_copy_mmx2( pixel *dst, int i_dst, pixel *src, int i_src, int w, int h )
 459 {
 460     int c_w = 16/sizeof(pixel) - 1;
 461     if( w < 256 ) { // tiny resolutions don't want non-temporal hints. dunno the exact threshold.
 462         x264_plane_copy_c( dst, i_dst, src, i_src, w, h );
 463     } else if( !(w&c_w) ) {
 464         x264_plane_copy_core_mmx2( dst, i_dst, src, i_src, w, h );
 465     } else if( i_src > 0 ) {
 466         // have to use plain memcpy on the last line (in memory order) to avoid overreading src
 467         x264_plane_copy_core_mmx2( dst, i_dst, src, i_src, (w+c_w)&~c_w, h-1 );
 468         memcpy( dst+i_dst*(h-1), src+i_src*(h-1), w*sizeof(pixel) );
 469     } else {
 470         memcpy( dst, src, w*sizeof(pixel) );
 471         x264_plane_copy_core_mmx2( dst+i_dst, i_dst, src+i_src, i_src, (w+c_w)&~c_w, h-1 );
 472     }
 473 }
 474
 475 #define PLANE_INTERLEAVE(cpu) \
 476 static void x264_plane_copy_interleave_##cpu( pixel *dst, int i_dst,\
 477                                               pixel *srcu, int i_srcu,\
 478                                               pixel *srcv, int i_srcv, int w, int h )\
 479 {\
 480     if( !(w&15) ) {\
 481         x264_plane_copy_interleave_core_##cpu( dst, i_dst, srcu, i_srcu, srcv, i_srcv, w, h );\
 482     } else if( w < 16 || (i_srcu ^ i_srcv) ) {\
 483         x264_plane_copy_interleave_c( dst, i_dst, srcu, i_srcu, srcv, i_srcv, w, h );\
 484     } else if( i_srcu > 0 ) {\
 485         x264_plane_copy_interleave_core_##cpu( dst, i_dst, srcu, i_srcu, srcv, i_srcv, (w+15)&~15, h-1 );\
 486         x264_plane_copy_interleave_c( dst+i_dst*(h-1), 0, srcu+i_srcu*(h-1), 0, srcv+i_srcv*(h-1), 0, w, 1 );\
 487     } else {\
 488         x264_plane_copy_interleave_c( dst, 0, srcu, 0, srcv, 0, w, 1 );\
 489         x264_plane_copy_interleave_core_##cpu( dst+i_dst, i_dst, srcu+i_srcu, i_srcu, srcv+i_srcv, i_srcv, (w+15)&~15, h-1 );\
 490     }\
 491 }
 492
 493 PLANE_INTERLEAVE(mmx2)
 494 PLANE_INTERLEAVE(sse2)
 495 #if HIGH_BIT_DEPTH
 496 PLANE_INTERLEAVE(avx)
 497 #endif
 498
 499 void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
 500 {
 501     if( !(cpu&X264_CPU_MMX) )
 502         return;
 503
 504     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_mmx;
 505     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_mmx;
 506
 507     pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_mmx;
 508
 509     pf->copy_16x16_unaligned = x264_mc_copy_w16_mmx;
 510     pf->copy[PIXEL_16x16] = x264_mc_copy_w16_mmx;
 511     pf->copy[PIXEL_8x8]   = x264_mc_copy_w8_mmx;
 512     pf->copy[PIXEL_4x4]   = x264_mc_copy_w4_mmx;
 513     pf->memcpy_aligned  = x264_memcpy_aligned_mmx;
 514     pf->memzero_aligned = x264_memzero_aligned_mmx;
 515     pf->integral_init4v = x264_integral_init4v_mmx;
 516     pf->integral_init8v = x264_integral_init8v_mmx;
 517
 518     if( !(cpu&X264_CPU_MMX2) )
 519         return;
 520
 521     pf->prefetch_fenc_420 = x264_prefetch_fenc_420_mmx2;
 522     pf->prefetch_fenc_422 = x264_prefetch_fenc_422_mmx2;
 523     pf->prefetch_ref  = x264_prefetch_ref_mmx2;
 524
 525     pf->plane_copy = x264_plane_copy_mmx2;
 526     pf->plane_copy_interleave = x264_plane_copy_interleave_mmx2;
 527     pf->store_interleave_chroma = x264_store_interleave_chroma_mmx2;
 528
 529     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_mmx2;
 530     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_mmx2;
 531     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_mmx2;
 532     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_mmx2;
 533     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_mmx2;
 534     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_mmx2;
 535     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_mmx2;
 536     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_mmx2;
 537     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_mmx2;
 538
 539     pf->mc_luma = mc_luma_mmx2;
 540     pf->get_ref = get_ref_mmx2;
 541     pf->mc_chroma = x264_mc_chroma_mmx2;
 542     pf->hpel_filter = x264_hpel_filter_mmx2;
 543     pf->weight = x264_mc_weight_wtab_mmx2;
 544     pf->weight_cache = x264_weight_cache_mmx2;
 545     pf->offsetadd = x264_mc_offsetadd_wtab_mmx2;
 546     pf->offsetsub = x264_mc_offsetsub_wtab_mmx2;
 547
 548     pf->frame_init_lowres_core = x264_frame_init_lowres_core_mmx2;
 549
 550 #if HIGH_BIT_DEPTH
 551 #if ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
 552     if( cpu&(X264_CPU_CACHELINE_32|X264_CPU_CACHELINE_64) )
 553         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
 554 #endif
 555
 556     if( !(cpu&X264_CPU_SSE2) )
 557         return;
 558
 559     pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
 560
 561     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_sse2;
 562     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_sse2;
 563
 564     pf->plane_copy_interleave   = x264_plane_copy_interleave_sse2;
 565     pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
 566
 567     if( cpu&X264_CPU_SSE2_IS_FAST )
 568     {
 569         pf->get_ref = get_ref_sse2;
 570         pf->mc_luma = mc_luma_sse2;
 571         pf->hpel_filter = x264_hpel_filter_sse2;
 572     }
 573
 574     pf->memcpy_aligned  = x264_memcpy_aligned_sse2;
 575     pf->memzero_aligned = x264_memzero_aligned_sse2;
 576     pf->integral_init4v = x264_integral_init4v_sse2;
 577     pf->integral_init8v = x264_integral_init8v_sse2;
 578     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
 579     pf->store_interleave_chroma = x264_store_interleave_chroma_sse2;
 580     pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
 581     pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
 582
 583     if( cpu&X264_CPU_SSE2_IS_SLOW )
 584         return;
 585
 586     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
 587     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_sse2;
 588     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_sse2;
 589     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_sse2;
 590     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_sse2;
 591     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_sse2;
 592     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_sse2;
 593     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_sse2;
 594     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_sse2;
 595
 596     pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse2;
 597     pf->weight = x264_mc_weight_wtab_sse2;
 598
 599     if( !(cpu&X264_CPU_STACK_MOD4) )
 600         pf->mc_chroma = x264_mc_chroma_sse2;
 601
 602     if( !(cpu&X264_CPU_SSSE3) )
 603         return;
 604
 605     pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
 606
 607     if( (cpu&X264_CPU_SHUFFLE_IS_FAST) && !(cpu&X264_CPU_SLOW_ATOM) )
 608         pf->integral_init4v = x264_integral_init4v_ssse3;
 609
 610     if( !(cpu&X264_CPU_AVX) )
 611         return;
 612
 613     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_avx;
 614     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_avx;
 615     pf->plane_copy_interleave        = x264_plane_copy_interleave_avx;
 616     pf->plane_copy_deinterleave      = x264_plane_copy_deinterleave_avx;
 617     pf->store_interleave_chroma      = x264_store_interleave_chroma_avx;
 618
 619     if( !(cpu&X264_CPU_STACK_MOD4) )
 620         pf->mc_chroma = x264_mc_chroma_avx;
 621 #else // !HIGH_BIT_DEPTH
 622
 623 #if ARCH_X86 // all x86_64 cpus with cacheline split issues use sse2 instead
 624     if( cpu&X264_CPU_CACHELINE_32 )
 625     {
 626         pf->mc_luma = mc_luma_cache32_mmx2;
 627         pf->get_ref = get_ref_cache32_mmx2;
 628         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
 629     }
 630     else if( cpu&X264_CPU_CACHELINE_64 )
 631     {
 632         pf->mc_luma = mc_luma_cache64_mmx2;
 633         pf->get_ref = get_ref_cache64_mmx2;
 634         pf->frame_init_lowres_core = x264_frame_init_lowres_core_cache32_mmx2;
 635     }
 636 #endif
 637
 638     if( !(cpu&X264_CPU_SSE2) )
 639         return;
 640
 641     pf->memcpy_aligned = x264_memcpy_aligned_sse2;
 642     pf->memzero_aligned = x264_memzero_aligned_sse2;
 643     pf->integral_init4v = x264_integral_init4v_sse2;
 644     pf->integral_init8v = x264_integral_init8v_sse2;
 645     pf->hpel_filter = x264_hpel_filter_sse2_amd;
 646     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_sse2;
 647
 648     if( cpu&X264_CPU_SSE2_IS_SLOW )
 649         return;
 650
 651     pf->weight = x264_mc_weight_wtab_sse2;
 652     if( !(cpu&X264_CPU_SLOW_ATOM) )
 653     {
 654         pf->offsetadd = x264_mc_offsetadd_wtab_sse2;
 655         pf->offsetsub = x264_mc_offsetsub_wtab_sse2;
 656     }
 657
 658     pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_sse2;
 659     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_sse2;
 660     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_sse2;
 661     pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_sse2;
 662     pf->avg[PIXEL_8x8]  = x264_pixel_avg_8x8_sse2;
 663     pf->avg[PIXEL_8x4]  = x264_pixel_avg_8x4_sse2;
 664     pf->hpel_filter = x264_hpel_filter_sse2;
 665     if( cpu&X264_CPU_SSE_MISALIGN )
 666         pf->hpel_filter = x264_hpel_filter_sse2_misalign;
 667     pf->frame_init_lowres_core = x264_frame_init_lowres_core_sse2;
 668     if( !(cpu&X264_CPU_STACK_MOD4) )
 669         pf->mc_chroma = x264_mc_chroma_sse2;
 670
 671     if( cpu&X264_CPU_SSE2_IS_FAST )
 672     {
 673         pf->store_interleave_chroma = x264_store_interleave_chroma_sse2; // FIXME sse2fast? sse2medium?
 674         pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_sse2;
 675         pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_sse2;
 676         pf->plane_copy_interleave   = x264_plane_copy_interleave_sse2;
 677         pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_sse2;
 678         pf->mc_luma = mc_luma_sse2;
 679         pf->get_ref = get_ref_sse2;
 680         if( cpu&X264_CPU_CACHELINE_64 )
 681         {
 682             pf->mc_luma = mc_luma_cache64_sse2;
 683             pf->get_ref = get_ref_cache64_sse2;
 684         }
 685         if( cpu&X264_CPU_SSE_MISALIGN )
 686         {
 687             pf->get_ref = get_ref_sse2_misalign;
 688             if( !(cpu&X264_CPU_STACK_MOD4) )
 689                 pf->mc_chroma = x264_mc_chroma_sse2_misalign;
 690         }
 691     }
 692
 693     if( !(cpu&X264_CPU_SSSE3) )
 694         return;
 695
 696     pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_ssse3;
 697     pf->avg[PIXEL_16x8]  = x264_pixel_avg_16x8_ssse3;
 698     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_ssse3;
 699     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_ssse3;
 700     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_ssse3;
 701     pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_ssse3;
 702     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_ssse3;
 703     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_ssse3;
 704     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_ssse3;
 705
 706     pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_ssse3;
 707     pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_ssse3;
 708     pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_ssse3;
 709
 710     pf->hpel_filter = x264_hpel_filter_ssse3;
 711     pf->frame_init_lowres_core = x264_frame_init_lowres_core_ssse3;
 712     if( !(cpu&X264_CPU_STACK_MOD4) )
 713         pf->mc_chroma = x264_mc_chroma_ssse3;
 714
 715     if( cpu&X264_CPU_CACHELINE_64 )
 716     {
 717         if( !(cpu&X264_CPU_STACK_MOD4) )
 718             pf->mc_chroma = x264_mc_chroma_ssse3_cache64;
 719         pf->mc_luma = mc_luma_cache64_ssse3;
 720         pf->get_ref = get_ref_cache64_ssse3;
 721
 722         /* ssse3 weight is slower on Nehalem, so only assign here. */
 723         pf->weight_cache = x264_weight_cache_ssse3;
 724         pf->weight = x264_mc_weight_wtab_ssse3;
 725     }
 726
 727     if( (cpu&X264_CPU_SHUFFLE_IS_FAST) && !(cpu&X264_CPU_SLOW_ATOM) )
 728         pf->integral_init4v = x264_integral_init4v_ssse3;
 729
 730     if( !(cpu&X264_CPU_SSE4) )
 731         return;
 732
 733     pf->integral_init4h = x264_integral_init4h_sse4;
 734     pf->integral_init8h = x264_integral_init8h_sse4;
 735
 736     if( !(cpu&X264_CPU_AVX) )
 737         return;
 738
 739     pf->integral_init8h = x264_integral_init8h_avx;
 740     pf->hpel_filter = x264_hpel_filter_avx;
 741     if( !(cpu&X264_CPU_STACK_MOD4) )
 742         pf->mc_chroma = x264_mc_chroma_avx;
 743 #endif // HIGH_BIT_DEPTH
 744
 745     if( !(cpu&X264_CPU_AVX) )
 746         return;
 747     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_avx;
 748
 749     if( !(cpu&X264_CPU_FMA4) )
 750         return;
 751     pf->mbtree_propagate_cost = x264_mbtree_propagate_cost_fma4;
 752 }