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

   1 /*****************************************************************************
   2  * predict-c.c: intra prediction
   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 "common/common.h"
  29 #include "predict.h"
  30 #include "pixel.h"
  31
  32  void x264_predict_16x16_v_mmx( pixel *src );
  33  void x264_predict_16x16_v_sse2( pixel *src );
  34  void x264_predict_16x16_h_mmxext( pixel *src );
  35  void x264_predict_16x16_h_sse2( uint16_t *src );
  36  void x264_predict_16x16_h_ssse3( uint8_t *src );
  37  void x264_predict_16x16_dc_core_mmxext( pixel *src, int i_dc_left );
  38  void x264_predict_16x16_dc_core_sse2( pixel *src, int i_dc_left );
  39  void x264_predict_16x16_dc_left_core_mmxext( pixel *src, int i_dc_left );
  40  void x264_predict_16x16_dc_left_core_sse2( pixel *src, int i_dc_left );
  41  void x264_predict_16x16_dc_top_mmxext( pixel *src );
  42  void x264_predict_16x16_dc_top_sse2( pixel *src );
  43  void x264_predict_16x16_dc_top_ssse3( uint16_t *src );
  44  void x264_predict_16x16_p_core_mmxext( uint8_t *src, int i00, int b, int c );
  45  void x264_predict_16x16_p_core_sse2( pixel *src, int i00, int b, int c );
  46  void x264_predict_8x8c_p_core_mmxext( uint8_t *src, int i00, int b, int c );
  47  void x264_predict_8x8c_p_core_sse2( pixel *src, int i00, int b, int c );
  48  void x264_predict_8x8c_dc_mmxext( pixel *src );
  49  void x264_predict_8x8c_dc_sse2( uint16_t *src );
  50  void x264_predict_8x8c_dc_top_mmxext( uint8_t *src );
  51  void x264_predict_8x8c_dc_top_sse2( uint16_t *src );
  52  void x264_predict_8x8c_v_mmx( pixel *src );
  53  void x264_predict_8x8c_v_sse2( uint16_t *src );
  54  void x264_predict_8x8c_h_mmxext( uint8_t *src );
  55  void x264_predict_8x8c_h_sse2( pixel *src );
  56  void x264_predict_8x8c_h_ssse3( uint8_t *src );
  57  void x264_predict_8x8_v_mmxext( uint8_t *src, uint8_t edge[33] );
  58  void x264_predict_8x8_v_sse2( uint16_t *src, uint16_t edge[33] );
  59  void x264_predict_8x8_h_mmxext( uint8_t *src, uint8_t edge[33] );
  60  void x264_predict_8x8_h_sse2( uint16_t *src, uint16_t edge[33] );
  61  void x264_predict_8x8_hd_mmxext( uint8_t *src, uint8_t edge[33] );
  62  void x264_predict_8x8_hu_mmxext( uint8_t *src, uint8_t edge[33] );
  63  void x264_predict_8x8_dc_mmxext( uint8_t *src, uint8_t edge[33] );
  64  void x264_predict_8x8_dc_sse2( uint16_t *src, uint16_t edge[33] );
  65  void x264_predict_8x8_dc_top_mmxext( uint8_t *src, uint8_t edge[33] );
  66  void x264_predict_8x8_dc_top_sse2( uint16_t *src, uint16_t edge[33] );
  67  void x264_predict_8x8_dc_left_mmxext( uint8_t *src, uint8_t edge[33] );
  68  void x264_predict_8x8_dc_left_sse2( uint16_t *src, uint16_t edge[33] );
  69  void x264_predict_8x8_ddl_mmxext( uint8_t *src, uint8_t edge[33] );
  70  void x264_predict_8x8_ddr_mmxext( uint8_t *src, uint8_t edge[33] );
  71  void x264_predict_8x8_ddl_sse2( pixel *src, pixel edge[33] );
  72  void x264_predict_8x8_ddr_sse2( pixel *src, pixel edge[33] );
  73  void x264_predict_8x8_vl_sse2( uint8_t *src, uint8_t edge[33] );
  74  void x264_predict_8x8_vr_mmxext( uint8_t *src, uint8_t edge[33] );
  75  void x264_predict_8x8_vr_sse2( pixel *src, pixel edge[33] );
  76  void x264_predict_8x8_vr_ssse3( uint16_t *src, uint16_t edge[33] );
  77  void x264_predict_8x8_hu_sse2( pixel *src, pixel edge[33] );
  78  void x264_predict_8x8_hu_ssse3( pixel *src, pixel edge[33] );
  79  void x264_predict_8x8_hd_sse2( pixel *src, pixel edge[33] );
  80  void x264_predict_8x8_hd_ssse3( pixel *src, pixel edge[33] );
  81  void x264_predict_8x8_filter_mmxext( uint8_t *src, uint8_t edge[33], int i_neighbor, int i_filters );
  82  void x264_predict_8x8_filter_sse2( uint16_t *src, uint16_t edge[33], int i_neighbor, int i_filters );
  83  void x264_predict_8x8_filter_ssse3( pixel *src, pixel edge[33], int i_neighbor, int i_filters );
  84  void x264_predict_4x4_ddl_mmxext( pixel *src );
  85  void x264_predict_4x4_ddl_sse2( uint16_t *src );
  86  void x264_predict_4x4_ddr_mmxext( pixel *src );
  87  void x264_predict_4x4_vl_mmxext( pixel *src );
  88  void x264_predict_4x4_vl_sse2( uint16_t *src );
  89  void x264_predict_4x4_vr_mmxext( uint8_t *src );
  90  void x264_predict_4x4_vr_sse2( uint16_t *src );
  91  void x264_predict_4x4_vr_ssse3( pixel *src );
  92  void x264_predict_4x4_hd_mmxext( pixel *src );
  93  void x264_predict_4x4_hd_sse2( uint16_t *src );
  94  void x264_predict_4x4_hd_ssse3( pixel *src );
  95  void x264_predict_4x4_dc_mmxext( pixel *src );
  96  void x264_predict_4x4_ddr_sse2( uint16_t *src );
  97  void x264_predict_4x4_ddr_ssse3( pixel *src );
  98  void x264_predict_4x4_hu_mmxext( pixel *src );
  99
 100 #define PREDICT_16x16_DC(name)\
 101 static void x264_predict_16x16_dc_##name( pixel *src )\
 102 {\
 103     uint32_t dc = 16;\
 104     for( int i = 0; i < 16; i += 2 )\
 105     {\
 106         dc += src[-1 + i * FDEC_STRIDE];\
 107         dc += src[-1 + (i+1) * FDEC_STRIDE];\
 108     }\
 109     x264_predict_16x16_dc_core_##name( src, dc );\
 110 }
 111
 112 PREDICT_16x16_DC( mmxext )
 113 PREDICT_16x16_DC( sse2 )
 114
 115 #define PREDICT_16x16_DC_LEFT(name)\
 116 static void x264_predict_16x16_dc_left_##name( pixel *src )\
 117 {\
 118     uint32_t dc = 8;\
 119     for( int i = 0; i < 16; i += 2 )\
 120     {\
 121         dc += src[-1 + i * FDEC_STRIDE];\
 122         dc += src[-1 + (i+1) * FDEC_STRIDE];\
 123     }\
 124     x264_predict_16x16_dc_left_core_##name( src, dc>>4 );\
 125 }
 126
 127 PREDICT_16x16_DC_LEFT( mmxext )
 128 PREDICT_16x16_DC_LEFT( sse2 )
 129
 130 #define PREDICT_P_SUM(j,i)\
 131     H += i * ( src[j+i - FDEC_STRIDE ]  - src[j-i - FDEC_STRIDE ] );\
 132     V += i * ( src[(j+i)*FDEC_STRIDE -1] - src[(j-i)*FDEC_STRIDE -1] );\
 133
 134 ALIGNED_16( static const int16_t pw_12345678[8] ) = {1,2,3,4,5,6,7,8};
 135 ALIGNED_16( static const int16_t pw_m87654321[8] ) = {-8,-7,-6,-5,-4,-3,-2,-1};
 136 ALIGNED_16( static const int16_t pw_m32101234[8] ) = {-3,-2,-1,0,1,2,3,4};
 137 ALIGNED_8( static const int8_t pb_12345678[8] ) = {1,2,3,4,5,6,7,8};
 138 ALIGNED_8( static const int8_t pb_m87654321[8] ) = {-8,-7,-6,-5,-4,-3,-2,-1};
 139 ALIGNED_8( static const int8_t pb_m32101234[8] ) = {-3,-2,-1,0,1,2,3,4};
 140
 141 #if !HIGH_BIT_DEPTH
 142 #define PREDICT_16x16_P(name)\
 143 static void x264_predict_16x16_p_##name( pixel *src )\
 144 {\
 145     int a, b, c;\
 146     int H = 0;\
 147     int V = 0;\
 148     int i00;\
 149     PREDICT_P_SUM(7,1) \
 150     PREDICT_P_SUM(7,2) \
 151     PREDICT_P_SUM(7,3) \
 152     PREDICT_P_SUM(7,4) \
 153     PREDICT_P_SUM(7,5) \
 154     PREDICT_P_SUM(7,6) \
 155     PREDICT_P_SUM(7,7) \
 156     PREDICT_P_SUM(7,8) \
 157     a = 16 * ( src[15*FDEC_STRIDE -1] + src[15 - FDEC_STRIDE] );\
 158     b = ( 5 * H + 32 ) >> 6;\
 159     c = ( 5 * V + 32 ) >> 6;\
 160     i00 = a - b * 7 - c * 7 + 16;\
 161     x264_predict_16x16_p_core_##name( src, i00, b, c );\
 162 }
 163 #ifndef ARCH_X86_64
 164 PREDICT_16x16_P( mmxext )
 165 #endif
 166 PREDICT_16x16_P( sse2   )
 167 #endif //!HIGH_BIT_DEPTH
 168
 169 #ifdef __GNUC__
 170 #if HIGH_BIT_DEPTH
 171 static void x264_predict_16x16_p_sse2( uint16_t *src )
 172 #else
 173 static void x264_predict_16x16_p_ssse3( uint8_t *src )
 174 #endif
 175 {
 176     int a, b, c, i00;
 177     int H, V;
 178 #if HIGH_BIT_DEPTH
 179     asm (
 180         "movdqu        -2+%1, %%xmm1 \n"
 181         "movdqa        16+%1, %%xmm0 \n"
 182         "pmaddwd          %2, %%xmm0 \n"
 183         "pmaddwd          %3, %%xmm1 \n"
 184         "paddd        %%xmm1, %%xmm0 \n"
 185         "movhlps      %%xmm0, %%xmm1 \n"
 186         "paddd        %%xmm1, %%xmm0 \n"
 187         "pshuflw $14, %%xmm0, %%xmm1 \n"
 188         "paddd        %%xmm1, %%xmm0 \n"
 189         "movd         %%xmm0, %0     \n"
 190         :"=r"(H)
 191         :"m"(src[-FDEC_STRIDE]), "m"(*pw_12345678), "m"(*pw_m87654321)
 192     );
 193 #else
 194     asm (
 195         "movq           %1, %%mm1 \n"
 196         "movq         8+%1, %%mm0 \n"
 197         "palignr $7, -8+%1, %%mm1 \n"
 198         "pmaddubsw      %2, %%mm0 \n"
 199         "pmaddubsw      %3, %%mm1 \n"
 200         "paddw       %%mm1, %%mm0 \n"
 201         "pshufw $14, %%mm0, %%mm1 \n"
 202         "paddw       %%mm1, %%mm0 \n"
 203         "pshufw  $1, %%mm0, %%mm1 \n"
 204         "paddw       %%mm1, %%mm0 \n"
 205         "movd        %%mm0, %0    \n"
 206         "movsx         %w0, %0    \n"
 207         :"=r"(H)
 208         :"m"(src[-FDEC_STRIDE]), "m"(*pb_12345678), "m"(*pb_m87654321)
 209     );
 210 #endif
 211     V = 8 * ( src[15*FDEC_STRIDE-1] - src[-1*FDEC_STRIDE-1] )
 212       + 7 * ( src[14*FDEC_STRIDE-1] - src[ 0*FDEC_STRIDE-1] )
 213       + 6 * ( src[13*FDEC_STRIDE-1] - src[ 1*FDEC_STRIDE-1] )
 214       + 5 * ( src[12*FDEC_STRIDE-1] - src[ 2*FDEC_STRIDE-1] )
 215       + 4 * ( src[11*FDEC_STRIDE-1] - src[ 3*FDEC_STRIDE-1] )
 216       + 3 * ( src[10*FDEC_STRIDE-1] - src[ 4*FDEC_STRIDE-1] )
 217       + 2 * ( src[ 9*FDEC_STRIDE-1] - src[ 5*FDEC_STRIDE-1] )
 218       + 1 * ( src[ 8*FDEC_STRIDE-1] - src[ 6*FDEC_STRIDE-1] );
 219     a = 16 * ( src[15*FDEC_STRIDE -1] + src[15 - FDEC_STRIDE] );
 220     b = ( 5 * H + 32 ) >> 6;
 221     c = ( 5 * V + 32 ) >> 6;
 222     i00 = a - b * 7 - c * 7 + 16;
 223     /* b*15 + c*15 can overflow: it's easier to just branch away in this rare case
 224      * than to try to consider it in the asm. */
 225     if( BIT_DEPTH > 8 && (i00 > 0x7fff || abs(b) > 1092 || abs(c) > 1092) )
 226         x264_predict_16x16_p_c( src );
 227     else
 228         x264_predict_16x16_p_core_sse2( src, i00, b, c );
 229 }
 230 #endif
 231
 232 #if !HIGH_BIT_DEPTH
 233
 234 #define PREDICT_8x8_P(name)\
 235 static void x264_predict_8x8c_p_##name( uint8_t *src )\
 236 {\
 237     int a, b, c;\
 238     int H = 0;\
 239     int V = 0;\
 240     int i00;\
 241     PREDICT_P_SUM(3,1)\
 242     PREDICT_P_SUM(3,2)\
 243     PREDICT_P_SUM(3,3)\
 244     PREDICT_P_SUM(3,4)\
 245     a = 16 * ( src[7*FDEC_STRIDE -1] + src[7 - FDEC_STRIDE] );\
 246     b = ( 17 * H + 16 ) >> 5;\
 247     c = ( 17 * V + 16 ) >> 5;\
 248     i00 = a -3*b -3*c + 16;\
 249     x264_predict_8x8c_p_core_##name( src, i00, b, c );\
 250 }
 251 #ifndef ARCH_X86_64
 252 PREDICT_8x8_P( mmxext )
 253 #endif
 254 PREDICT_8x8_P( sse2   )
 255
 256 #endif //!HIGH_BIT_DEPTH
 257
 258 #ifdef __GNUC__
 259 #if HIGH_BIT_DEPTH
 260 static void x264_predict_8x8c_p_sse2( uint16_t *src )
 261 #else
 262 static void x264_predict_8x8c_p_ssse3( uint8_t *src )
 263 #endif
 264 {
 265     int a, b, c, i00;
 266     int H, V;
 267 #if HIGH_BIT_DEPTH
 268     asm (
 269         "movdqa           %1, %%xmm0 \n"
 270         "pmaddwd          %2, %%xmm0 \n"
 271         "movhlps      %%xmm0, %%xmm1 \n"
 272         "paddd        %%xmm1, %%xmm0 \n"
 273         "pshuflw $14, %%xmm0, %%xmm1 \n"
 274         "paddd        %%xmm1, %%xmm0 \n"
 275         "movd         %%xmm0, %0     \n"
 276         :"=r"(H)
 277         :"m"(src[-FDEC_STRIDE]), "m"(*pw_m32101234)
 278     );
 279 #else
 280     asm (
 281         "movq           %1, %%mm0 \n"
 282         "pmaddubsw      %2, %%mm0 \n"
 283         "pshufw $14, %%mm0, %%mm1 \n"
 284         "paddw       %%mm1, %%mm0 \n"
 285         "pshufw  $1, %%mm0, %%mm1 \n"
 286         "paddw       %%mm1, %%mm0 \n"
 287         "movd        %%mm0, %0    \n"
 288         "movsx         %w0, %0    \n"
 289         :"=r"(H)
 290         :"m"(src[-FDEC_STRIDE]), "m"(*pb_m32101234)
 291     );
 292 #endif
 293     V = 1 * ( src[4*FDEC_STRIDE -1] - src[ 2*FDEC_STRIDE -1] )
 294       + 2 * ( src[5*FDEC_STRIDE -1] - src[ 1*FDEC_STRIDE -1] )
 295       + 3 * ( src[6*FDEC_STRIDE -1] - src[ 0*FDEC_STRIDE -1] )
 296       + 4 * ( src[7*FDEC_STRIDE -1] - src[-1*FDEC_STRIDE -1] );
 297     H += -4 * src[-1*FDEC_STRIDE -1];
 298     a = 16 * ( src[7*FDEC_STRIDE -1] + src[7 - FDEC_STRIDE] );
 299     b = ( 17 * H + 16 ) >> 5;
 300     c = ( 17 * V + 16 ) >> 5;
 301     i00 = a -3*b -3*c + 16;
 302     /* b*7 + c*7 can overflow: it's easier to just branch away in this rare case
 303      * than to try to consider it in the asm. */
 304     if( BIT_DEPTH > 8 && (i00 > 0x7fff || abs(b) > 2340 || abs(c) > 2340) )
 305         x264_predict_8x8c_p_c( src );
 306     else
 307         x264_predict_8x8c_p_core_sse2( src, i00, b, c );
 308 }
 309 #endif
 310 #if !HIGH_BIT_DEPTH
 311 #if ARCH_X86_64
 312 static void x264_predict_8x8c_dc_left( uint8_t *src )
 313 {
 314     int y;
 315     uint32_t s0 = 0, s1 = 0;
 316     uint64_t dc0, dc1;
 317
 318     for( y = 0; y < 4; y++ )
 319     {
 320         s0 += src[y * FDEC_STRIDE     - 1];
 321         s1 += src[(y+4) * FDEC_STRIDE - 1];
 322     }
 323     dc0 = (( s0 + 2 ) >> 2) * 0x0101010101010101ULL;
 324     dc1 = (( s1 + 2 ) >> 2) * 0x0101010101010101ULL;
 325
 326     for( y = 0; y < 4; y++ )
 327     {
 328         M64( src ) = dc0;
 329         src += FDEC_STRIDE;
 330     }
 331     for( y = 0; y < 4; y++ )
 332     {
 333         M64( src ) = dc1;
 334         src += FDEC_STRIDE;
 335     }
 336
 337 }
 338 #endif
 339
 340 #define PL(y) \
 341     UNUSED int l##y = edge[14-y];
 342 #define PT(x) \
 343     UNUSED int t##x = edge[16+x];
 344 #define PREDICT_8x8_LOAD_LEFT \
 345     PL(0) PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) PL(7)
 346 #define PREDICT_8x8_LOAD_TOP \
 347     PT(0) PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) PT(7)
 348
 349 #define SUMSUB(a,b,c,d,e,f,g,h)\
 350     t=a; a+=b; b-=t;\
 351     t=c; c+=d; d-=t;\
 352     t=e; e+=f; f-=t;\
 353     t=g; g+=h; h-=t;
 354
 355 #define INTRA_SA8D_X3(cpu)\
 356 void x264_intra_sa8d_x3_8x8_##cpu( uint8_t *fenc, uint8_t edge[33], int res[3] )\
 357 {\
 358     PREDICT_8x8_LOAD_TOP\
 359     PREDICT_8x8_LOAD_LEFT\
 360     int t;\
 361     ALIGNED_16( int16_t sa8d_1d[2][8] );\
 362     SUMSUB(l0,l4,l1,l5,l2,l6,l3,l7);\
 363     SUMSUB(l0,l2,l1,l3,l4,l6,l5,l7);\
 364     SUMSUB(l0,l1,l2,l3,l4,l5,l6,l7);\
 365     sa8d_1d[0][0] = l0;\
 366     sa8d_1d[0][1] = l1;\
 367     sa8d_1d[0][2] = l2;\
 368     sa8d_1d[0][3] = l3;\
 369     sa8d_1d[0][4] = l4;\
 370     sa8d_1d[0][5] = l5;\
 371     sa8d_1d[0][6] = l6;\
 372     sa8d_1d[0][7] = l7;\
 373     SUMSUB(t0,t4,t1,t5,t2,t6,t3,t7);\
 374     SUMSUB(t0,t2,t1,t3,t4,t6,t5,t7);\
 375     SUMSUB(t0,t1,t2,t3,t4,t5,t6,t7);\
 376     sa8d_1d[1][0] = t0;\
 377     sa8d_1d[1][1] = t1;\
 378     sa8d_1d[1][2] = t2;\
 379     sa8d_1d[1][3] = t3;\
 380     sa8d_1d[1][4] = t4;\
 381     sa8d_1d[1][5] = t5;\
 382     sa8d_1d[1][6] = t6;\
 383     sa8d_1d[1][7] = t7;\
 384     x264_intra_sa8d_x3_8x8_core_##cpu( fenc, sa8d_1d, res );\
 385 }
 386
 387 #if ARCH_X86_64
 388 INTRA_SA8D_X3(sse2)
 389 INTRA_SA8D_X3(ssse3)
 390 #else
 391 INTRA_SA8D_X3(mmxext)
 392 #endif
 393 #endif // !HIGH_BIT_DEPTH
 394
 395 /****************************************************************************
 396  * Exported functions:
 397  ****************************************************************************/
 398 void x264_predict_16x16_init_mmx( int cpu, x264_predict_t pf[7] )
 399 {
 400     if( !(cpu&X264_CPU_MMX) )
 401         return;
 402     pf[I_PRED_16x16_V]       = x264_predict_16x16_v_mmx;
 403     if( cpu&X264_CPU_MMXEXT )
 404     {
 405         pf[I_PRED_16x16_DC]      = x264_predict_16x16_dc_mmxext;
 406         pf[I_PRED_16x16_DC_TOP]  = x264_predict_16x16_dc_top_mmxext;
 407         pf[I_PRED_16x16_DC_LEFT] = x264_predict_16x16_dc_left_mmxext;
 408         pf[I_PRED_16x16_H]       = x264_predict_16x16_h_mmxext;
 409     }
 410 #if HIGH_BIT_DEPTH
 411     if( !(cpu&X264_CPU_SSE2) )
 412         return;
 413     pf[I_PRED_16x16_DC]      = x264_predict_16x16_dc_sse2;
 414     pf[I_PRED_16x16_DC_TOP]  = x264_predict_16x16_dc_top_sse2;
 415     pf[I_PRED_16x16_DC_LEFT] = x264_predict_16x16_dc_left_sse2;
 416     pf[I_PRED_16x16_V]       = x264_predict_16x16_v_sse2;
 417     pf[I_PRED_16x16_H]       = x264_predict_16x16_h_sse2;
 418     pf[I_PRED_16x16_P]       = x264_predict_16x16_p_sse2;
 419 #else
 420 #if !ARCH_X86_64
 421     pf[I_PRED_16x16_P]       = x264_predict_16x16_p_mmxext;
 422 #endif
 423     if( !(cpu&X264_CPU_SSE2) )
 424         return;
 425     pf[I_PRED_16x16_DC]      = x264_predict_16x16_dc_sse2;
 426     pf[I_PRED_16x16_V]       = x264_predict_16x16_v_sse2;
 427     if( cpu&X264_CPU_SSE2_IS_SLOW )
 428         return;
 429     pf[I_PRED_16x16_DC_TOP]  = x264_predict_16x16_dc_top_sse2;
 430     pf[I_PRED_16x16_DC_LEFT] = x264_predict_16x16_dc_left_sse2;
 431     pf[I_PRED_16x16_P]       = x264_predict_16x16_p_sse2;
 432     if( !(cpu&X264_CPU_SSSE3) )
 433         return;
 434     pf[I_PRED_16x16_H]       = x264_predict_16x16_h_ssse3;
 435 #ifdef __GNUC__
 436     pf[I_PRED_16x16_P]       = x264_predict_16x16_p_ssse3;
 437 #endif
 438 #endif // HIGH_BIT_DEPTH
 439 }
 440
 441 void x264_predict_8x8c_init_mmx( int cpu, x264_predict_t pf[7] )
 442 {
 443     if( !(cpu&X264_CPU_MMX) )
 444         return;
 445 #if HIGH_BIT_DEPTH
 446     pf[I_PRED_CHROMA_V]       = x264_predict_8x8c_v_mmx;
 447     if( !(cpu&X264_CPU_MMXEXT) )
 448         return;
 449     pf[I_PRED_CHROMA_DC]      = x264_predict_8x8c_dc_mmxext;
 450     if( !(cpu&X264_CPU_SSE2) )
 451         return;
 452     pf[I_PRED_CHROMA_V]       = x264_predict_8x8c_v_sse2;
 453     pf[I_PRED_CHROMA_DC]      = x264_predict_8x8c_dc_sse2;
 454     pf[I_PRED_CHROMA_DC_TOP]  = x264_predict_8x8c_dc_top_sse2;
 455     pf[I_PRED_CHROMA_H]       = x264_predict_8x8c_h_sse2;
 456     pf[I_PRED_CHROMA_P]       = x264_predict_8x8c_p_sse2;
 457 #else
 458 #if ARCH_X86_64
 459     pf[I_PRED_CHROMA_DC_LEFT] = x264_predict_8x8c_dc_left;
 460 #endif
 461     pf[I_PRED_CHROMA_V]       = x264_predict_8x8c_v_mmx;
 462     if( !(cpu&X264_CPU_MMXEXT) )
 463         return;
 464     pf[I_PRED_CHROMA_DC_TOP]  = x264_predict_8x8c_dc_top_mmxext;
 465     pf[I_PRED_CHROMA_H]       = x264_predict_8x8c_h_mmxext;
 466 #if !ARCH_X86_64
 467     pf[I_PRED_CHROMA_P]       = x264_predict_8x8c_p_mmxext;
 468 #endif
 469     pf[I_PRED_CHROMA_DC]      = x264_predict_8x8c_dc_mmxext;
 470     if( !(cpu&X264_CPU_SSE2) )
 471         return;
 472     pf[I_PRED_CHROMA_P]       = x264_predict_8x8c_p_sse2;
 473     if( !(cpu&X264_CPU_SSSE3) )
 474         return;
 475     pf[I_PRED_CHROMA_H]       = x264_predict_8x8c_h_ssse3;
 476 #ifdef __GNUC__
 477     pf[I_PRED_CHROMA_P]       = x264_predict_8x8c_p_ssse3;
 478 #endif
 479 #endif // HIGH_BIT_DEPTH
 480 }
 481
 482 void x264_predict_8x8_init_mmx( int cpu, x264_predict8x8_t pf[12], x264_predict_8x8_filter_t *predict_8x8_filter )
 483 {
 484     if( !(cpu&X264_CPU_MMXEXT) )
 485         return;
 486 #if HIGH_BIT_DEPTH
 487     if( !(cpu&X264_CPU_SSE2) )
 488         return;
 489     pf[I_PRED_8x8_V]      = x264_predict_8x8_v_sse2;
 490     pf[I_PRED_8x8_H]      = x264_predict_8x8_h_sse2;
 491     pf[I_PRED_8x8_DC]     = x264_predict_8x8_dc_sse2;
 492     pf[I_PRED_8x8_DC_TOP] = x264_predict_8x8_dc_top_sse2;
 493     pf[I_PRED_8x8_DC_LEFT]= x264_predict_8x8_dc_left_sse2;
 494     pf[I_PRED_8x8_DDL]    = x264_predict_8x8_ddl_sse2;
 495     pf[I_PRED_8x8_DDR]    = x264_predict_8x8_ddr_sse2;
 496     pf[I_PRED_8x8_VR]     = x264_predict_8x8_vr_sse2;
 497     pf[I_PRED_8x8_HD]     = x264_predict_8x8_hd_sse2;
 498     pf[I_PRED_8x8_HU]     = x264_predict_8x8_hu_sse2;
 499     *predict_8x8_filter   = x264_predict_8x8_filter_sse2;
 500     if( !(cpu&X264_CPU_SSSE3) )
 501         return;
 502     pf[I_PRED_8x8_HD]     = x264_predict_8x8_hd_ssse3;
 503     pf[I_PRED_8x8_HU]     = x264_predict_8x8_hu_ssse3;
 504     pf[I_PRED_8x8_VR]     = x264_predict_8x8_vr_ssse3;
 505     *predict_8x8_filter   = x264_predict_8x8_filter_ssse3;
 506 #else
 507     pf[I_PRED_8x8_V]      = x264_predict_8x8_v_mmxext;
 508     pf[I_PRED_8x8_H]      = x264_predict_8x8_h_mmxext;
 509     pf[I_PRED_8x8_DC]     = x264_predict_8x8_dc_mmxext;
 510     pf[I_PRED_8x8_DC_TOP] = x264_predict_8x8_dc_top_mmxext;
 511     pf[I_PRED_8x8_DC_LEFT]= x264_predict_8x8_dc_left_mmxext;
 512     pf[I_PRED_8x8_HD]     = x264_predict_8x8_hd_mmxext;
 513     *predict_8x8_filter   = x264_predict_8x8_filter_mmxext;
 514 #if ARCH_X86
 515     pf[I_PRED_8x8_DDL]  = x264_predict_8x8_ddl_mmxext;
 516     pf[I_PRED_8x8_DDR]  = x264_predict_8x8_ddr_mmxext;
 517     pf[I_PRED_8x8_VR]   = x264_predict_8x8_vr_mmxext;
 518     pf[I_PRED_8x8_HU]   = x264_predict_8x8_hu_mmxext;
 519 #endif
 520     if( !(cpu&X264_CPU_SSE2) )
 521         return;
 522     pf[I_PRED_8x8_DDL]  = x264_predict_8x8_ddl_sse2;
 523     pf[I_PRED_8x8_VL]   = x264_predict_8x8_vl_sse2;
 524     pf[I_PRED_8x8_VR]   = x264_predict_8x8_vr_sse2;
 525     pf[I_PRED_8x8_DDR]  = x264_predict_8x8_ddr_sse2;
 526     pf[I_PRED_8x8_HD]   = x264_predict_8x8_hd_sse2;
 527     pf[I_PRED_8x8_HU]   = x264_predict_8x8_hu_sse2;
 528     if( !(cpu&X264_CPU_SSSE3) )
 529         return;
 530     pf[I_PRED_8x8_HD]   = x264_predict_8x8_hd_ssse3;
 531     pf[I_PRED_8x8_HU]   = x264_predict_8x8_hu_ssse3;
 532     *predict_8x8_filter = x264_predict_8x8_filter_ssse3;
 533 #endif // HIGH_BIT_DEPTH
 534 }
 535
 536 void x264_predict_4x4_init_mmx( int cpu, x264_predict_t pf[12] )
 537 {
 538     if( !(cpu&X264_CPU_MMXEXT) )
 539         return;
 540     pf[I_PRED_4x4_DC]  = x264_predict_4x4_dc_mmxext;
 541     pf[I_PRED_4x4_DDL] = x264_predict_4x4_ddl_mmxext;
 542     pf[I_PRED_4x4_DDR] = x264_predict_4x4_ddr_mmxext;
 543     pf[I_PRED_4x4_VL]  = x264_predict_4x4_vl_mmxext;
 544     pf[I_PRED_4x4_HD]  = x264_predict_4x4_hd_mmxext;
 545     pf[I_PRED_4x4_HU]  = x264_predict_4x4_hu_mmxext;
 546 #if HIGH_BIT_DEPTH
 547     if( !(cpu&X264_CPU_SSE2) )
 548         return;
 549     pf[I_PRED_4x4_DDL] = x264_predict_4x4_ddl_sse2;
 550     pf[I_PRED_4x4_DDR] = x264_predict_4x4_ddr_sse2;
 551     pf[I_PRED_4x4_HD]  = x264_predict_4x4_hd_sse2;
 552     pf[I_PRED_4x4_VL]  = x264_predict_4x4_vl_sse2;
 553     pf[I_PRED_4x4_VR]  = x264_predict_4x4_vr_sse2;
 554 #else
 555     pf[I_PRED_4x4_VR]  = x264_predict_4x4_vr_mmxext;
 556 #endif // HIGH_BIT_DEPTH
 557     if( !(cpu&X264_CPU_SSSE3) )
 558         return;
 559     pf[I_PRED_4x4_DDR] = x264_predict_4x4_ddr_ssse3;
 560     pf[I_PRED_4x4_VR]  = x264_predict_4x4_vr_ssse3;
 561     pf[I_PRED_4x4_HD]  = x264_predict_4x4_hd_ssse3;
 562 }