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

   1 /*****************************************************************************
   2  * predict-c.c: intra prediction
   3  *****************************************************************************
   4  * Copyright (C) 2003-2016 x264 project
   5  *
   6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
   7  *          Loren Merritt <lorenm@u.washington.edu>
   8  *          Fiona Glaser <fiona@x264.com>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License as published by
  12  * the Free Software Foundation; either version 2 of the License, or
  13  * (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  * GNU General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with this program; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
  23  *
  24  * This program is also available under a commercial proprietary license.
  25  * For more information, contact us at licensing@x264.com.
  26  *****************************************************************************/
  27
  28 #include "common/common.h"
  29 #include "predict.h"
  30 #include "pixel.h"
  31
  32 #define PREDICT_16x16_DC(name)\
  33 void x264_predict_16x16_dc_##name( pixel *src )\
  34 {\
  35     uint32_t dc = 16;\
  36     for( int i = 0; i < 16; i += 2 )\
  37     {\
  38         dc += src[-1 + i * FDEC_STRIDE];\
  39         dc += src[-1 + (i+1) * FDEC_STRIDE];\
  40     }\
  41     x264_predict_16x16_dc_core_##name( src, dc );\
  42 }
  43
  44 PREDICT_16x16_DC( mmx2 )
  45 PREDICT_16x16_DC( sse2 )
  46 PREDICT_16x16_DC( avx2 )
  47
  48 #define PREDICT_16x16_DC_LEFT(name)\
  49 static void x264_predict_16x16_dc_left_##name( pixel *src )\
  50 {\
  51     uint32_t dc = 8;\
  52     for( int i = 0; i < 16; i += 2 )\
  53     {\
  54         dc += src[-1 + i * FDEC_STRIDE];\
  55         dc += src[-1 + (i+1) * FDEC_STRIDE];\
  56     }\
  57     x264_predict_16x16_dc_left_core_##name( src, dc>>4 );\
  58 }
  59
  60 PREDICT_16x16_DC_LEFT( mmx2 )
  61 PREDICT_16x16_DC_LEFT( sse2 )
  62 PREDICT_16x16_DC_LEFT( avx2 )
  63
  64 #define PREDICT_P_SUM(j,i)\
  65     H += i * ( src[j+i - FDEC_STRIDE ]  - src[j-i - FDEC_STRIDE ] );\
  66     V += i * ( src[(j+i)*FDEC_STRIDE -1] - src[(j-i)*FDEC_STRIDE -1] );
  67
  68 #if HAVE_X86_INLINE_ASM
  69 #if HIGH_BIT_DEPTH
  70 ALIGNED_16( static const int16_t pw_12345678[8] ) = {1,2,3,4,5,6,7,8};
  71 ALIGNED_16( static const int16_t pw_m87654321[8] ) = {-8,-7,-6,-5,-4,-3,-2,-1};
  72 ALIGNED_16( static const int16_t pw_m32101234[8] ) = {-3,-2,-1,0,1,2,3,4};
  73 #else // !HIGH_BIT_DEPTH
  74 ALIGNED_8( static const int8_t pb_12345678[8] ) = {1,2,3,4,5,6,7,8};
  75 ALIGNED_8( static const int8_t pb_m87654321[8] ) = {-8,-7,-6,-5,-4,-3,-2,-1};
  76 ALIGNED_8( static const int8_t pb_m32101234[8] ) = {-3,-2,-1,0,1,2,3,4};
  77 #endif // HIGH_BIT_DEPTH
  78 #endif // HAVE_X86_INLINE_ASM
  79
  80 #define PREDICT_16x16_P_CORE\
  81     int H = 0;\
  82     int V = 0;\
  83     PREDICT_P_SUM(7,1)\
  84     PREDICT_P_SUM(7,2)\
  85     PREDICT_P_SUM(7,3)\
  86     PREDICT_P_SUM(7,4)\
  87     PREDICT_P_SUM(7,5)\
  88     PREDICT_P_SUM(7,6)\
  89     PREDICT_P_SUM(7,7)\
  90     PREDICT_P_SUM(7,8)
  91
  92 #define PREDICT_16x16_P_END(name)\
  93     int a = 16 * ( src[15*FDEC_STRIDE -1] + src[15 - FDEC_STRIDE] );\
  94     int b = ( 5 * H + 32 ) >> 6;\
  95     int c = ( 5 * V + 32 ) >> 6;\
  96     int i00 = a - b * 7 - c * 7 + 16;\
  97     /* b*15 + c*15 can overflow: it's easier to just branch away in this rare case
  98      * than to try to consider it in the asm. */\
  99     if( BIT_DEPTH > 8 && (i00 > 0x7fff || abs(b) > 1092 || abs(c) > 1092) )\
 100         x264_predict_16x16_p_c( src );\
 101     else\
 102         x264_predict_16x16_p_core_##name( src, i00, b, c );
 103
 104 #define PREDICT_16x16_P(name, name2)\
 105 static void x264_predict_16x16_p_##name( pixel *src )\
 106 {\
 107     PREDICT_16x16_P_CORE\
 108     PREDICT_16x16_P_END(name2)\
 109 }
 110
 111 #if HAVE_X86_INLINE_ASM
 112 #if HIGH_BIT_DEPTH
 113 #define PREDICT_16x16_P_ASM\
 114     asm (\
 115         "movdqu           %1, %%xmm1 \n"\
 116         "movdqa           %2, %%xmm0 \n"\
 117         "pmaddwd          %3, %%xmm0 \n"\
 118         "pmaddwd          %4, %%xmm1 \n"\
 119         "paddd        %%xmm1, %%xmm0 \n"\
 120         "movhlps      %%xmm0, %%xmm1 \n"\
 121         "paddd        %%xmm1, %%xmm0 \n"\
 122         "pshuflw $14, %%xmm0, %%xmm1 \n"\
 123         "paddd        %%xmm1, %%xmm0 \n"\
 124         "movd         %%xmm0, %0     \n"\
 125         :"=r"(H)\
 126         :"m"(src[-FDEC_STRIDE-1]), "m"(src[-FDEC_STRIDE+8]),\
 127          "m"(*pw_12345678), "m"(*pw_m87654321)\
 128     );
 129 #else // !HIGH_BIT_DEPTH
 130 #define PREDICT_16x16_P_ASM\
 131     asm (\
 132         "movq           %1, %%mm1 \n"\
 133         "movq           %2, %%mm0 \n"\
 134         "palignr $7,    %3, %%mm1 \n"\
 135         "pmaddubsw      %4, %%mm0 \n"\
 136         "pmaddubsw      %5, %%mm1 \n"\
 137         "paddw       %%mm1, %%mm0 \n"\
 138         "pshufw $14, %%mm0, %%mm1 \n"\
 139         "paddw       %%mm1, %%mm0 \n"\
 140         "pshufw  $1, %%mm0, %%mm1 \n"\
 141         "paddw       %%mm1, %%mm0 \n"\
 142         "movd        %%mm0, %0    \n"\
 143         "movswl        %w0, %0    \n"\
 144         :"=r"(H)\
 145         :"m"(src[-FDEC_STRIDE]), "m"(src[-FDEC_STRIDE+8]),\
 146          "m"(src[-FDEC_STRIDE-8]), "m"(*pb_12345678), "m"(*pb_m87654321)\
 147     );
 148 #endif // HIGH_BIT_DEPTH
 149
 150 #define PREDICT_16x16_P_CORE_INLINE\
 151     int H, V;\
 152     PREDICT_16x16_P_ASM\
 153     V = 8 * ( src[15*FDEC_STRIDE-1] - src[-1*FDEC_STRIDE-1] )\
 154       + 7 * ( src[14*FDEC_STRIDE-1] - src[ 0*FDEC_STRIDE-1] )\
 155       + 6 * ( src[13*FDEC_STRIDE-1] - src[ 1*FDEC_STRIDE-1] )\
 156       + 5 * ( src[12*FDEC_STRIDE-1] - src[ 2*FDEC_STRIDE-1] )\
 157       + 4 * ( src[11*FDEC_STRIDE-1] - src[ 3*FDEC_STRIDE-1] )\
 158       + 3 * ( src[10*FDEC_STRIDE-1] - src[ 4*FDEC_STRIDE-1] )\
 159       + 2 * ( src[ 9*FDEC_STRIDE-1] - src[ 5*FDEC_STRIDE-1] )\
 160       + 1 * ( src[ 8*FDEC_STRIDE-1] - src[ 6*FDEC_STRIDE-1] );
 161
 162 #define PREDICT_16x16_P_INLINE(name, name2)\
 163 static void x264_predict_16x16_p_##name( pixel *src )\
 164 {\
 165     PREDICT_16x16_P_CORE_INLINE\
 166     PREDICT_16x16_P_END(name2)\
 167 }
 168 #else // !HAVE_X86_INLINE_ASM
 169 #define PREDICT_16x16_P_INLINE(name, name2) PREDICT_16x16_P(name, name2)
 170 #endif // HAVE_X86_INLINE_ASM
 171
 172 #if HIGH_BIT_DEPTH
 173 PREDICT_16x16_P_INLINE( sse2, sse2 )
 174 #else // !HIGH_BIT_DEPTH
 175 #if !ARCH_X86_64
 176 PREDICT_16x16_P( mmx2, mmx2 )
 177 #endif // !ARCH_X86_64
 178 PREDICT_16x16_P( sse2, sse2 )
 179 #if HAVE_X86_INLINE_ASM
 180 PREDICT_16x16_P_INLINE( ssse3, sse2 )
 181 #endif // HAVE_X86_INLINE_ASM
 182 PREDICT_16x16_P_INLINE( avx, avx )
 183 #endif // HIGH_BIT_DEPTH
 184 PREDICT_16x16_P_INLINE( avx2, avx2 )
 185
 186 #define PREDICT_8x16C_P_CORE\
 187     int H = 0, V = 0;\
 188     for( int i = 0; i < 4; i++ )\
 189         H += ( i + 1 ) * ( src[4 + i - FDEC_STRIDE] - src[2 - i - FDEC_STRIDE] );\
 190     for( int i = 0; i < 8; i++ )\
 191         V += ( i + 1 ) * ( src[-1 + (i+8)*FDEC_STRIDE] - src[-1 + (6-i)*FDEC_STRIDE] );
 192
 193 #if HIGH_BIT_DEPTH
 194 #define PREDICT_8x16C_P_END(name)\
 195     int a = 16 * ( src[-1 + 15*FDEC_STRIDE] + src[7 - FDEC_STRIDE] );\
 196     int b = ( 17 * H + 16 ) >> 5;\
 197     int c = ( 5 * V + 32 ) >> 6;\
 198     x264_predict_8x16c_p_core_##name( src, a, b, c );
 199 #else // !HIGH_BIT_DEPTH
 200 #define PREDICT_8x16C_P_END(name)\
 201     int a = 16 * ( src[-1 + 15*FDEC_STRIDE] + src[7 - FDEC_STRIDE] );\
 202     int b = ( 17 * H + 16 ) >> 5;\
 203     int c = ( 5 * V + 32 ) >> 6;\
 204     int i00 = a -3*b -7*c + 16;\
 205     x264_predict_8x16c_p_core_##name( src, i00, b, c );
 206 #endif // HIGH_BIT_DEPTH
 207
 208 #define PREDICT_8x16C_P(name)\
 209 static void x264_predict_8x16c_p_##name( pixel *src )\
 210 {\
 211     PREDICT_8x16C_P_CORE\
 212     PREDICT_8x16C_P_END(name)\
 213 }
 214
 215 #if !ARCH_X86_64 && !HIGH_BIT_DEPTH
 216 PREDICT_8x16C_P( mmx2 )
 217 #endif // !ARCH_X86_64 && !HIGH_BIT_DEPTH
 218 PREDICT_8x16C_P( sse2 )
 219 PREDICT_8x16C_P( avx )
 220 PREDICT_8x16C_P( avx2 )
 221
 222 #define PREDICT_8x8C_P_CORE\
 223     int H = 0;\
 224     int V = 0;\
 225     PREDICT_P_SUM(3,1)\
 226     PREDICT_P_SUM(3,2)\
 227     PREDICT_P_SUM(3,3)\
 228     PREDICT_P_SUM(3,4)
 229
 230 #if HIGH_BIT_DEPTH
 231 #define PREDICT_8x8C_P_END(name)\
 232     int a = 16 * ( src[7*FDEC_STRIDE -1] + src[7 - FDEC_STRIDE] );\
 233     int b = ( 17 * H + 16 ) >> 5;\
 234     int c = ( 17 * V + 16 ) >> 5;\
 235     x264_predict_8x8c_p_core_##name( src, a, b, c );
 236 #else // !HIGH_BIT_DEPTH
 237 #define PREDICT_8x8C_P_END(name)\
 238     int a = 16 * ( src[7*FDEC_STRIDE -1] + src[7 - FDEC_STRIDE] );\
 239     int b = ( 17 * H + 16 ) >> 5;\
 240     int c = ( 17 * V + 16 ) >> 5;\
 241     int i00 = a -3*b -3*c + 16;\
 242     x264_predict_8x8c_p_core_##name( src, i00, b, c );
 243 #endif // HIGH_BIT_DEPTH
 244
 245 #define PREDICT_8x8C_P(name, name2)\
 246 static void x264_predict_8x8c_p_##name( pixel *src )\
 247 {\
 248     PREDICT_8x8C_P_CORE\
 249     PREDICT_8x8C_P_END(name2)\
 250 }
 251
 252 #if HAVE_X86_INLINE_ASM
 253 #if HIGH_BIT_DEPTH
 254 #define PREDICT_8x8C_P_ASM\
 255     asm (\
 256         "movdqa           %1, %%xmm0 \n"\
 257         "pmaddwd          %2, %%xmm0 \n"\
 258         "movhlps      %%xmm0, %%xmm1 \n"\
 259         "paddd        %%xmm1, %%xmm0 \n"\
 260         "pshuflw $14, %%xmm0, %%xmm1 \n"\
 261         "paddd        %%xmm1, %%xmm0 \n"\
 262         "movd         %%xmm0, %0     \n"\
 263         :"=r"(H)\
 264         :"m"(src[-FDEC_STRIDE]), "m"(*pw_m32101234)\
 265     );
 266 #else // !HIGH_BIT_DEPTH
 267 #define PREDICT_8x8C_P_ASM\
 268     asm (\
 269         "movq           %1, %%mm0 \n"\
 270         "pmaddubsw      %2, %%mm0 \n"\
 271         "pshufw $14, %%mm0, %%mm1 \n"\
 272         "paddw       %%mm1, %%mm0 \n"\
 273         "pshufw  $1, %%mm0, %%mm1 \n"\
 274         "paddw       %%mm1, %%mm0 \n"\
 275         "movd        %%mm0, %0    \n"\
 276         "movswl        %w0, %0    \n"\
 277         :"=r"(H)\
 278         :"m"(src[-FDEC_STRIDE]), "m"(*pb_m32101234)\
 279     );
 280 #endif // HIGH_BIT_DEPTH
 281
 282 #define PREDICT_8x8C_P_CORE_INLINE\
 283     int H, V;\
 284     PREDICT_8x8C_P_ASM\
 285     V = 1 * ( src[4*FDEC_STRIDE -1] - src[ 2*FDEC_STRIDE -1] )\
 286       + 2 * ( src[5*FDEC_STRIDE -1] - src[ 1*FDEC_STRIDE -1] )\
 287       + 3 * ( src[6*FDEC_STRIDE -1] - src[ 0*FDEC_STRIDE -1] )\
 288       + 4 * ( src[7*FDEC_STRIDE -1] - src[-1*FDEC_STRIDE -1] );\
 289     H += -4 * src[-1*FDEC_STRIDE -1];
 290
 291 #define PREDICT_8x8C_P_INLINE(name, name2)\
 292 static void x264_predict_8x8c_p_##name( pixel *src )\
 293 {\
 294     PREDICT_8x8C_P_CORE_INLINE\
 295     PREDICT_8x8C_P_END(name2)\
 296 }
 297 #else // !HAVE_X86_INLINE_ASM
 298 #define PREDICT_8x8C_P_INLINE(name, name2) PREDICT_8x8C_P(name, name2)
 299 #endif // HAVE_X86_INLINE_ASM
 300
 301 #if HIGH_BIT_DEPTH
 302 PREDICT_8x8C_P_INLINE( sse2, sse2 )
 303 #else  //!HIGH_BIT_DEPTH
 304 #if !ARCH_X86_64
 305 PREDICT_8x8C_P( mmx2, mmx2 )
 306 #endif // !ARCH_X86_64
 307 PREDICT_8x8C_P( sse2, sse2 )
 308 #if HAVE_X86_INLINE_ASM
 309 PREDICT_8x8C_P_INLINE( ssse3, sse2 )
 310 #endif // HAVE_X86_INLINE_ASM
 311 #endif // HIGH_BIT_DEPTH
 312 PREDICT_8x8C_P_INLINE( avx, avx )
 313 PREDICT_8x8C_P_INLINE( avx2, avx2 )
 314
 315 #if ARCH_X86_64 && !HIGH_BIT_DEPTH
 316 static void x264_predict_8x8c_dc_left( uint8_t *src )
 317 {
 318     int y;
 319     uint32_t s0 = 0, s1 = 0;
 320     uint64_t dc0, dc1;
 321
 322     for( y = 0; y < 4; y++ )
 323     {
 324         s0 += src[y * FDEC_STRIDE     - 1];
 325         s1 += src[(y+4) * FDEC_STRIDE - 1];
 326     }
 327     dc0 = (( s0 + 2 ) >> 2) * 0x0101010101010101ULL;
 328     dc1 = (( s1 + 2 ) >> 2) * 0x0101010101010101ULL;
 329
 330     for( y = 0; y < 4; y++ )
 331     {
 332         M64( src ) = dc0;
 333         src += FDEC_STRIDE;
 334     }
 335     for( y = 0; y < 4; y++ )
 336     {
 337         M64( src ) = dc1;
 338         src += FDEC_STRIDE;
 339     }
 340 }
 341 #endif // ARCH_X86_64 && !HIGH_BIT_DEPTH
 342
 343 /****************************************************************************
 344  * Exported functions:
 345  ****************************************************************************/
 346 void x264_predict_16x16_init_mmx( int cpu, x264_predict_t pf[7] )
 347 {
 348     if( !(cpu&X264_CPU_MMX2) )
 349         return;
 350     pf[I_PRED_16x16_DC]      = x264_predict_16x16_dc_mmx2;
 351     pf[I_PRED_16x16_DC_TOP]  = x264_predict_16x16_dc_top_mmx2;
 352     pf[I_PRED_16x16_DC_LEFT] = x264_predict_16x16_dc_left_mmx2;
 353     pf[I_PRED_16x16_V]       = x264_predict_16x16_v_mmx2;
 354     pf[I_PRED_16x16_H]       = x264_predict_16x16_h_mmx2;
 355 #if HIGH_BIT_DEPTH
 356     if( !(cpu&X264_CPU_SSE) )
 357         return;
 358     pf[I_PRED_16x16_V]       = x264_predict_16x16_v_sse;
 359     if( !(cpu&X264_CPU_SSE2) )
 360         return;
 361     pf[I_PRED_16x16_DC]      = x264_predict_16x16_dc_sse2;
 362     pf[I_PRED_16x16_DC_TOP]  = x264_predict_16x16_dc_top_sse2;
 363     pf[I_PRED_16x16_DC_LEFT] = x264_predict_16x16_dc_left_sse2;
 364     pf[I_PRED_16x16_H]       = x264_predict_16x16_h_sse2;
 365     pf[I_PRED_16x16_P]       = x264_predict_16x16_p_sse2;
 366     if( !(cpu&X264_CPU_AVX) )
 367         return;
 368     pf[I_PRED_16x16_V]       = x264_predict_16x16_v_avx;
 369     if( !(cpu&X264_CPU_AVX2) )
 370         return;
 371     pf[I_PRED_16x16_H]       = x264_predict_16x16_h_avx2;
 372 #else
 373 #if !ARCH_X86_64
 374     pf[I_PRED_16x16_P]       = x264_predict_16x16_p_mmx2;
 375 #endif
 376     if( !(cpu&X264_CPU_SSE) )
 377         return;
 378     pf[I_PRED_16x16_V]       = x264_predict_16x16_v_sse;
 379     if( !(cpu&X264_CPU_SSE2) )
 380         return;
 381     pf[I_PRED_16x16_DC]      = x264_predict_16x16_dc_sse2;
 382     if( cpu&X264_CPU_SSE2_IS_SLOW )
 383         return;
 384     pf[I_PRED_16x16_DC_TOP]  = x264_predict_16x16_dc_top_sse2;
 385     pf[I_PRED_16x16_DC_LEFT] = x264_predict_16x16_dc_left_sse2;
 386     pf[I_PRED_16x16_P]       = x264_predict_16x16_p_sse2;
 387     if( !(cpu&X264_CPU_SSSE3) )
 388         return;
 389     if( !(cpu&X264_CPU_SLOW_PSHUFB) )
 390         pf[I_PRED_16x16_H]       = x264_predict_16x16_h_ssse3;
 391 #if HAVE_X86_INLINE_ASM
 392     pf[I_PRED_16x16_P]       = x264_predict_16x16_p_ssse3;
 393 #endif
 394     if( !(cpu&X264_CPU_AVX) )
 395         return;
 396     pf[I_PRED_16x16_P]       = x264_predict_16x16_p_avx;
 397 #endif // HIGH_BIT_DEPTH
 398
 399     if( cpu&X264_CPU_AVX2 )
 400     {
 401         pf[I_PRED_16x16_P]       = x264_predict_16x16_p_avx2;
 402         pf[I_PRED_16x16_DC]      = x264_predict_16x16_dc_avx2;
 403         pf[I_PRED_16x16_DC_TOP]  = x264_predict_16x16_dc_top_avx2;
 404         pf[I_PRED_16x16_DC_LEFT] = x264_predict_16x16_dc_left_avx2;
 405     }
 406 }
 407
 408 void x264_predict_8x8c_init_mmx( int cpu, x264_predict_t pf[7] )
 409 {
 410     if( !(cpu&X264_CPU_MMX) )
 411         return;
 412 #if HIGH_BIT_DEPTH
 413     pf[I_PRED_CHROMA_V]       = x264_predict_8x8c_v_mmx;
 414     if( !(cpu&X264_CPU_MMX2) )
 415         return;
 416     pf[I_PRED_CHROMA_DC]      = x264_predict_8x8c_dc_mmx2;
 417     pf[I_PRED_CHROMA_H]       = x264_predict_8x8c_h_mmx2;
 418     if( !(cpu&X264_CPU_SSE) )
 419         return;
 420     pf[I_PRED_CHROMA_V]       = x264_predict_8x8c_v_sse;
 421     if( !(cpu&X264_CPU_SSE2) )
 422         return;
 423     pf[I_PRED_CHROMA_DC]      = x264_predict_8x8c_dc_sse2;
 424     pf[I_PRED_CHROMA_DC_TOP]  = x264_predict_8x8c_dc_top_sse2;
 425     pf[I_PRED_CHROMA_H]       = x264_predict_8x8c_h_sse2;
 426     pf[I_PRED_CHROMA_P]       = x264_predict_8x8c_p_sse2;
 427     if( !(cpu&X264_CPU_AVX) )
 428         return;
 429     pf[I_PRED_CHROMA_P]       = x264_predict_8x8c_p_avx;
 430     if( !(cpu&X264_CPU_AVX2) )
 431         return;
 432     pf[I_PRED_CHROMA_H]   = x264_predict_8x8c_h_avx2;
 433 #else
 434 #if ARCH_X86_64
 435     pf[I_PRED_CHROMA_DC_LEFT] = x264_predict_8x8c_dc_left;
 436 #endif
 437     pf[I_PRED_CHROMA_V]       = x264_predict_8x8c_v_mmx;
 438     if( !(cpu&X264_CPU_MMX2) )
 439         return;
 440     pf[I_PRED_CHROMA_DC_TOP]  = x264_predict_8x8c_dc_top_mmx2;
 441     pf[I_PRED_CHROMA_H]       = x264_predict_8x8c_h_mmx2;
 442 #if !ARCH_X86_64
 443     pf[I_PRED_CHROMA_P]       = x264_predict_8x8c_p_mmx2;
 444 #endif
 445     pf[I_PRED_CHROMA_DC]      = x264_predict_8x8c_dc_mmx2;
 446     if( !(cpu&X264_CPU_SSE2) )
 447         return;
 448     pf[I_PRED_CHROMA_P]       = x264_predict_8x8c_p_sse2;
 449     if( !(cpu&X264_CPU_SSSE3) )
 450         return;
 451     pf[I_PRED_CHROMA_H]       = x264_predict_8x8c_h_ssse3;
 452 #if HAVE_X86_INLINE_ASM
 453     pf[I_PRED_CHROMA_P]       = x264_predict_8x8c_p_ssse3;
 454 #endif
 455     if( !(cpu&X264_CPU_AVX) )
 456         return;
 457     pf[I_PRED_CHROMA_P]       = x264_predict_8x8c_p_avx;
 458 #endif // HIGH_BIT_DEPTH
 459
 460     if( cpu&X264_CPU_AVX2 )
 461     {
 462         pf[I_PRED_CHROMA_P]   = x264_predict_8x8c_p_avx2;
 463     }
 464 }
 465
 466 void x264_predict_8x16c_init_mmx( int cpu, x264_predict_t pf[7] )
 467 {
 468     if( !(cpu&X264_CPU_MMX) )
 469         return;
 470 #if HIGH_BIT_DEPTH
 471     if( !(cpu&X264_CPU_MMX2) )
 472         return;
 473     pf[I_PRED_CHROMA_DC]      = x264_predict_8x16c_dc_mmx2;
 474     pf[I_PRED_CHROMA_H]       = x264_predict_8x16c_h_mmx2;
 475     if( !(cpu&X264_CPU_SSE) )
 476         return;
 477     pf[I_PRED_CHROMA_V]       = x264_predict_8x16c_v_sse;
 478     if( !(cpu&X264_CPU_SSE2) )
 479         return;
 480     pf[I_PRED_CHROMA_DC_TOP]  = x264_predict_8x16c_dc_top_sse2;
 481     pf[I_PRED_CHROMA_DC]      = x264_predict_8x16c_dc_sse2;
 482     pf[I_PRED_CHROMA_H]       = x264_predict_8x16c_h_sse2;
 483     pf[I_PRED_CHROMA_P]       = x264_predict_8x16c_p_sse2;
 484     if( !(cpu&X264_CPU_AVX) )
 485         return;
 486     pf[I_PRED_CHROMA_P]       = x264_predict_8x16c_p_avx;
 487     if( !(cpu&X264_CPU_AVX2) )
 488         return;
 489     pf[I_PRED_CHROMA_H]   = x264_predict_8x16c_h_avx2;
 490 #else
 491     pf[I_PRED_CHROMA_V]       = x264_predict_8x16c_v_mmx;
 492     if( !(cpu&X264_CPU_MMX2) )
 493         return;
 494     pf[I_PRED_CHROMA_DC_TOP]  = x264_predict_8x16c_dc_top_mmx2;
 495     pf[I_PRED_CHROMA_DC]      = x264_predict_8x16c_dc_mmx2;
 496     pf[I_PRED_CHROMA_H]       = x264_predict_8x16c_h_mmx2;
 497 #if !ARCH_X86_64
 498     pf[I_PRED_CHROMA_P]       = x264_predict_8x16c_p_mmx2;
 499 #endif
 500     if( !(cpu&X264_CPU_SSE2) )
 501         return;
 502     pf[I_PRED_CHROMA_P]       = x264_predict_8x16c_p_sse2;
 503     if( !(cpu&X264_CPU_SSSE3) )
 504         return;
 505     pf[I_PRED_CHROMA_H]       = x264_predict_8x16c_h_ssse3;
 506     if( !(cpu&X264_CPU_AVX) )
 507         return;
 508     pf[I_PRED_CHROMA_P]       = x264_predict_8x16c_p_avx;
 509 #endif // HIGH_BIT_DEPTH
 510
 511     if( cpu&X264_CPU_AVX2 )
 512     {
 513         pf[I_PRED_CHROMA_P]   = x264_predict_8x16c_p_avx2;
 514     }
 515 }
 516
 517 void x264_predict_8x8_init_mmx( int cpu, x264_predict8x8_t pf[12], x264_predict_8x8_filter_t *predict_8x8_filter )
 518 {
 519     if( !(cpu&X264_CPU_MMX2) )
 520         return;
 521 #if HIGH_BIT_DEPTH
 522     if( !(cpu&X264_CPU_SSE) )
 523         return;
 524     pf[I_PRED_8x8_V]      = x264_predict_8x8_v_sse;
 525     if( !(cpu&X264_CPU_SSE2) )
 526         return;
 527     pf[I_PRED_8x8_H]      = x264_predict_8x8_h_sse2;
 528     pf[I_PRED_8x8_DC]     = x264_predict_8x8_dc_sse2;
 529     pf[I_PRED_8x8_DC_TOP] = x264_predict_8x8_dc_top_sse2;
 530     pf[I_PRED_8x8_DC_LEFT]= x264_predict_8x8_dc_left_sse2;
 531     pf[I_PRED_8x8_DDL]    = x264_predict_8x8_ddl_sse2;
 532     pf[I_PRED_8x8_DDR]    = x264_predict_8x8_ddr_sse2;
 533     pf[I_PRED_8x8_VL]     = x264_predict_8x8_vl_sse2;
 534     pf[I_PRED_8x8_VR]     = x264_predict_8x8_vr_sse2;
 535     pf[I_PRED_8x8_HD]     = x264_predict_8x8_hd_sse2;
 536     pf[I_PRED_8x8_HU]     = x264_predict_8x8_hu_sse2;
 537     *predict_8x8_filter   = x264_predict_8x8_filter_sse2;
 538     if( !(cpu&X264_CPU_SSSE3) )
 539         return;
 540     pf[I_PRED_8x8_DDL]    = x264_predict_8x8_ddl_ssse3;
 541     pf[I_PRED_8x8_DDR]    = x264_predict_8x8_ddr_ssse3;
 542     pf[I_PRED_8x8_HD]     = x264_predict_8x8_hd_ssse3;
 543     pf[I_PRED_8x8_HU]     = x264_predict_8x8_hu_ssse3;
 544     pf[I_PRED_8x8_VL]     = x264_predict_8x8_vl_ssse3;
 545     pf[I_PRED_8x8_VR]     = x264_predict_8x8_vr_ssse3;
 546     *predict_8x8_filter   = x264_predict_8x8_filter_ssse3;
 547     if( cpu&X264_CPU_CACHELINE_64 )
 548     {
 549         pf[I_PRED_8x8_DDL]= x264_predict_8x8_ddl_ssse3_cache64;
 550         pf[I_PRED_8x8_DDR]= x264_predict_8x8_ddr_ssse3_cache64;
 551     }
 552     if( !(cpu&X264_CPU_AVX) )
 553         return;
 554     pf[I_PRED_8x8_HD]     = x264_predict_8x8_hd_avx;
 555     pf[I_PRED_8x8_HU]     = x264_predict_8x8_hu_avx;
 556     pf[I_PRED_8x8_VL]     = x264_predict_8x8_vl_avx;
 557     pf[I_PRED_8x8_VR]     = x264_predict_8x8_vr_avx;
 558     *predict_8x8_filter   = x264_predict_8x8_filter_avx;
 559 #else
 560     pf[I_PRED_8x8_V]      = x264_predict_8x8_v_mmx2;
 561     pf[I_PRED_8x8_H]      = x264_predict_8x8_h_mmx2;
 562     pf[I_PRED_8x8_DC]     = x264_predict_8x8_dc_mmx2;
 563     pf[I_PRED_8x8_DC_TOP] = x264_predict_8x8_dc_top_mmx2;
 564     pf[I_PRED_8x8_DC_LEFT]= x264_predict_8x8_dc_left_mmx2;
 565     pf[I_PRED_8x8_HD]     = x264_predict_8x8_hd_mmx2;
 566     pf[I_PRED_8x8_VL]     = x264_predict_8x8_vl_mmx2;
 567     *predict_8x8_filter   = x264_predict_8x8_filter_mmx2;
 568 #if ARCH_X86
 569     pf[I_PRED_8x8_DDL]  = x264_predict_8x8_ddl_mmx2;
 570     pf[I_PRED_8x8_DDR]  = x264_predict_8x8_ddr_mmx2;
 571     pf[I_PRED_8x8_VR]   = x264_predict_8x8_vr_mmx2;
 572     pf[I_PRED_8x8_HU]   = x264_predict_8x8_hu_mmx2;
 573 #endif
 574     if( !(cpu&X264_CPU_SSE2) )
 575         return;
 576     pf[I_PRED_8x8_DDL]  = x264_predict_8x8_ddl_sse2;
 577     pf[I_PRED_8x8_VL]   = x264_predict_8x8_vl_sse2;
 578     pf[I_PRED_8x8_VR]   = x264_predict_8x8_vr_sse2;
 579     pf[I_PRED_8x8_DDR]  = x264_predict_8x8_ddr_sse2;
 580     pf[I_PRED_8x8_HD]   = x264_predict_8x8_hd_sse2;
 581     pf[I_PRED_8x8_HU]   = x264_predict_8x8_hu_sse2;
 582     if( !(cpu&X264_CPU_SSSE3) )
 583         return;
 584     if( !(cpu&X264_CPU_SLOW_PALIGNR) )
 585     {
 586         pf[I_PRED_8x8_DDL]  = x264_predict_8x8_ddl_ssse3;
 587         pf[I_PRED_8x8_VR]   = x264_predict_8x8_vr_ssse3;
 588     }
 589     pf[I_PRED_8x8_HU]   = x264_predict_8x8_hu_ssse3;
 590     *predict_8x8_filter = x264_predict_8x8_filter_ssse3;
 591     if( !(cpu&X264_CPU_AVX) )
 592         return;
 593     pf[I_PRED_8x8_DDL]  = x264_predict_8x8_ddl_avx;
 594     pf[I_PRED_8x8_DDR]  = x264_predict_8x8_ddr_avx;
 595     pf[I_PRED_8x8_VL]   = x264_predict_8x8_vl_avx;
 596     pf[I_PRED_8x8_VR]   = x264_predict_8x8_vr_avx;
 597     pf[I_PRED_8x8_HD]   = x264_predict_8x8_hd_avx;
 598 #endif // HIGH_BIT_DEPTH
 599 }
 600
 601 void x264_predict_4x4_init_mmx( int cpu, x264_predict_t pf[12] )
 602 {
 603     if( !(cpu&X264_CPU_MMX2) )
 604         return;
 605     pf[I_PRED_4x4_DC]  = x264_predict_4x4_dc_mmx2;
 606     pf[I_PRED_4x4_DDL] = x264_predict_4x4_ddl_mmx2;
 607     pf[I_PRED_4x4_DDR] = x264_predict_4x4_ddr_mmx2;
 608     pf[I_PRED_4x4_VL]  = x264_predict_4x4_vl_mmx2;
 609     pf[I_PRED_4x4_HD]  = x264_predict_4x4_hd_mmx2;
 610     pf[I_PRED_4x4_HU]  = x264_predict_4x4_hu_mmx2;
 611 #if HIGH_BIT_DEPTH
 612     if( !(cpu&X264_CPU_SSE2) )
 613         return;
 614     pf[I_PRED_4x4_DDL] = x264_predict_4x4_ddl_sse2;
 615     pf[I_PRED_4x4_DDR] = x264_predict_4x4_ddr_sse2;
 616     pf[I_PRED_4x4_HD]  = x264_predict_4x4_hd_sse2;
 617     pf[I_PRED_4x4_VL]  = x264_predict_4x4_vl_sse2;
 618     pf[I_PRED_4x4_VR]  = x264_predict_4x4_vr_sse2;
 619     if( !(cpu&X264_CPU_SSSE3) )
 620         return;
 621     pf[I_PRED_4x4_DDR] = x264_predict_4x4_ddr_ssse3;
 622     pf[I_PRED_4x4_VR]  = x264_predict_4x4_vr_ssse3;
 623     pf[I_PRED_4x4_HD]  = x264_predict_4x4_hd_ssse3;
 624     if( !(cpu&X264_CPU_AVX) )
 625         return;
 626     pf[I_PRED_4x4_DDL] = x264_predict_4x4_ddl_avx;
 627     pf[I_PRED_4x4_DDR] = x264_predict_4x4_ddr_avx;
 628     pf[I_PRED_4x4_HD]  = x264_predict_4x4_hd_avx;
 629     pf[I_PRED_4x4_VL]  = x264_predict_4x4_vl_avx;
 630     pf[I_PRED_4x4_VR]  = x264_predict_4x4_vr_avx;
 631     if( !(cpu&X264_CPU_AVX2) )
 632         return;
 633     pf[I_PRED_4x4_H]  = x264_predict_4x4_h_avx2;
 634 #else
 635     pf[I_PRED_4x4_VR]  = x264_predict_4x4_vr_mmx2;
 636     if( !(cpu&X264_CPU_SSSE3) )
 637         return;
 638     pf[I_PRED_4x4_DDR] = x264_predict_4x4_ddr_ssse3;
 639     pf[I_PRED_4x4_VR]  = x264_predict_4x4_vr_ssse3;
 640     pf[I_PRED_4x4_HD]  = x264_predict_4x4_hd_ssse3;
 641     if( cpu&X264_CPU_CACHELINE_64 )
 642         pf[I_PRED_4x4_VR] = x264_predict_4x4_vr_ssse3_cache64;
 643 #endif // HIGH_BIT_DEPTH
 644 }