common/predict.c

   1 /*****************************************************************************
   2  * predict.c: h264 encoder
   3  *****************************************************************************
   4  * Copyright (C) 2003-2008 x264 project
   5  *
   6  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
   7  *          Loren Merritt <lorenm@u.washington.edu>
   8  *          Jason Garrett-Glaser <darkshikari@gmail.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
  25 /* predict4x4 are inspired from ffmpeg h264 decoder */
  26
  27
  28 #include "common.h"
  29
  30 #ifdef HAVE_MMX
  31 #   include "x86/predict.h"
  32 #endif
  33 #ifdef ARCH_PPC
  34 #   include "ppc/predict.h"
  35 #endif
  36
  37 /****************************************************************************
  38  * 16x16 prediction for intra luma block
  39  ****************************************************************************/
  40
  41 #define PREDICT_16x16_DC(v) \
  42     for( i = 0; i < 16; i++ )\
  43     {\
  44         uint32_t *p = (uint32_t*)src;\
  45         *p++ = v;\
  46         *p++ = v;\
  47         *p++ = v;\
  48         *p++ = v;\
  49         src += FDEC_STRIDE;\
  50     }
  51
  52 static void predict_16x16_dc( uint8_t *src )
  53 {
  54     uint32_t dc = 0;
  55     int i;
  56
  57     for( i = 0; i < 16; i++ )
  58     {
  59         dc += src[-1 + i * FDEC_STRIDE];
  60         dc += src[i - FDEC_STRIDE];
  61     }
  62     dc = (( dc + 16 ) >> 5) * 0x01010101;
  63
  64     PREDICT_16x16_DC(dc);
  65 }
  66 static void predict_16x16_dc_left( uint8_t *src )
  67 {
  68     uint32_t dc = 0;
  69     int i;
  70
  71     for( i = 0; i < 16; i++ )
  72     {
  73         dc += src[-1 + i * FDEC_STRIDE];
  74     }
  75     dc = (( dc + 8 ) >> 4) * 0x01010101;
  76
  77     PREDICT_16x16_DC(dc);
  78 }
  79 static void predict_16x16_dc_top( uint8_t *src )
  80 {
  81     uint32_t dc = 0;
  82     int i;
  83
  84     for( i = 0; i < 16; i++ )
  85     {
  86         dc += src[i - FDEC_STRIDE];
  87     }
  88     dc = (( dc + 8 ) >> 4) * 0x01010101;
  89
  90     PREDICT_16x16_DC(dc);
  91 }
  92 static void predict_16x16_dc_128( uint8_t *src )
  93 {
  94     int i;
  95     PREDICT_16x16_DC(0x80808080);
  96 }
  97 static void predict_16x16_h( uint8_t *src )
  98 {
  99     int i;
 100
 101     for( i = 0; i < 16; i++ )
 102     {
 103         const uint32_t v = 0x01010101 * src[-1];
 104         uint32_t *p = (uint32_t*)src;
 105
 106         *p++ = v;
 107         *p++ = v;
 108         *p++ = v;
 109         *p++ = v;
 110
 111         src += FDEC_STRIDE;
 112
 113     }
 114 }
 115 static void predict_16x16_v( uint8_t *src )
 116 {
 117     uint32_t v0 = *(uint32_t*)&src[ 0-FDEC_STRIDE];
 118     uint32_t v1 = *(uint32_t*)&src[ 4-FDEC_STRIDE];
 119     uint32_t v2 = *(uint32_t*)&src[ 8-FDEC_STRIDE];
 120     uint32_t v3 = *(uint32_t*)&src[12-FDEC_STRIDE];
 121     int i;
 122
 123     for( i = 0; i < 16; i++ )
 124     {
 125         uint32_t *p = (uint32_t*)src;
 126         *p++ = v0;
 127         *p++ = v1;
 128         *p++ = v2;
 129         *p++ = v3;
 130         src += FDEC_STRIDE;
 131     }
 132 }
 133 static void predict_16x16_p( uint8_t *src )
 134 {
 135     int x, y, i;
 136     int a, b, c;
 137     int H = 0;
 138     int V = 0;
 139     int i00;
 140
 141     /* calculate H and V */
 142     for( i = 0; i <= 7; i++ )
 143     {
 144         H += ( i + 1 ) * ( src[ 8 + i - FDEC_STRIDE ] - src[6 -i -FDEC_STRIDE] );
 145         V += ( i + 1 ) * ( src[-1 + (8+i)*FDEC_STRIDE] - src[-1 + (6-i)*FDEC_STRIDE] );
 146     }
 147
 148     a = 16 * ( src[-1 + 15*FDEC_STRIDE] + src[15 - FDEC_STRIDE] );
 149     b = ( 5 * H + 32 ) >> 6;
 150     c = ( 5 * V + 32 ) >> 6;
 151
 152     i00 = a - b * 7 - c * 7 + 16;
 153
 154     for( y = 0; y < 16; y++ )
 155     {
 156         int pix = i00;
 157         for( x = 0; x < 16; x++ )
 158         {
 159             src[x] = x264_clip_uint8( pix>>5 );
 160             pix += b;
 161         }
 162         src += FDEC_STRIDE;
 163         i00 += c;
 164     }
 165 }
 166
 167
 168 /****************************************************************************
 169  * 8x8 prediction for intra chroma block
 170  ****************************************************************************/
 171
 172 static void predict_8x8c_dc_128( uint8_t *src )
 173 {
 174     int y;
 175
 176     for( y = 0; y < 8; y++ )
 177     {
 178         uint32_t *p = (uint32_t*)src;
 179         *p++ = 0x80808080;
 180         *p++ = 0x80808080;
 181         src += FDEC_STRIDE;
 182     }
 183 }
 184 static void predict_8x8c_dc_left( uint8_t *src )
 185 {
 186     int y;
 187     uint32_t dc0 = 0, dc1 = 0;
 188
 189     for( y = 0; y < 4; y++ )
 190     {
 191         dc0 += src[y * FDEC_STRIDE     - 1];
 192         dc1 += src[(y+4) * FDEC_STRIDE - 1];
 193     }
 194     dc0 = (( dc0 + 2 ) >> 2)*0x01010101;
 195     dc1 = (( dc1 + 2 ) >> 2)*0x01010101;
 196
 197     for( y = 0; y < 4; y++ )
 198     {
 199         uint32_t *p = (uint32_t*)src;
 200         *p++ = dc0;
 201         *p++ = dc0;
 202         src += FDEC_STRIDE;
 203     }
 204     for( y = 0; y < 4; y++ )
 205     {
 206         uint32_t *p = (uint32_t*)src;
 207         *p++ = dc1;
 208         *p++ = dc1;
 209         src += FDEC_STRIDE;
 210     }
 211
 212 }
 213 static void predict_8x8c_dc_top( uint8_t *src )
 214 {
 215     int y, x;
 216     uint32_t dc0 = 0, dc1 = 0;
 217
 218     for( x = 0; x < 4; x++ )
 219     {
 220         dc0 += src[x     - FDEC_STRIDE];
 221         dc1 += src[x + 4 - FDEC_STRIDE];
 222     }
 223     dc0 = (( dc0 + 2 ) >> 2)*0x01010101;
 224     dc1 = (( dc1 + 2 ) >> 2)*0x01010101;
 225
 226     for( y = 0; y < 8; y++ )
 227     {
 228         uint32_t *p = (uint32_t*)src;
 229         *p++ = dc0;
 230         *p++ = dc1;
 231         src += FDEC_STRIDE;
 232     }
 233 }
 234 static void predict_8x8c_dc( uint8_t *src )
 235 {
 236     int y;
 237     int s0 = 0, s1 = 0, s2 = 0, s3 = 0;
 238     uint32_t dc0, dc1, dc2, dc3;
 239     int i;
 240
 241     /*
 242           s0 s1
 243        s2
 244        s3
 245     */
 246     for( i = 0; i < 4; i++ )
 247     {
 248         s0 += src[i - FDEC_STRIDE];
 249         s1 += src[i + 4 - FDEC_STRIDE];
 250         s2 += src[-1 + i * FDEC_STRIDE];
 251         s3 += src[-1 + (i+4)*FDEC_STRIDE];
 252     }
 253     /*
 254        dc0 dc1
 255        dc2 dc3
 256      */
 257     dc0 = (( s0 + s2 + 4 ) >> 3)*0x01010101;
 258     dc1 = (( s1 + 2 ) >> 2)*0x01010101;
 259     dc2 = (( s3 + 2 ) >> 2)*0x01010101;
 260     dc3 = (( s1 + s3 + 4 ) >> 3)*0x01010101;
 261
 262     for( y = 0; y < 4; y++ )
 263     {
 264         uint32_t *p = (uint32_t*)src;
 265         *p++ = dc0;
 266         *p++ = dc1;
 267         src += FDEC_STRIDE;
 268     }
 269
 270     for( y = 0; y < 4; y++ )
 271     {
 272         uint32_t *p = (uint32_t*)src;
 273         *p++ = dc2;
 274         *p++ = dc3;
 275         src += FDEC_STRIDE;
 276     }
 277 }
 278 static void predict_8x8c_h( uint8_t *src )
 279 {
 280     int i;
 281
 282     for( i = 0; i < 8; i++ )
 283     {
 284         uint32_t v = 0x01010101 * src[-1];
 285         uint32_t *p = (uint32_t*)src;
 286         *p++ = v;
 287         *p++ = v;
 288         src += FDEC_STRIDE;
 289     }
 290 }
 291 static void predict_8x8c_v( uint8_t *src )
 292 {
 293     uint32_t v0 = *(uint32_t*)&src[0-FDEC_STRIDE];
 294     uint32_t v1 = *(uint32_t*)&src[4-FDEC_STRIDE];
 295     int i;
 296
 297     for( i = 0; i < 8; i++ )
 298     {
 299         uint32_t *p = (uint32_t*)src;
 300         *p++ = v0;
 301         *p++ = v1;
 302         src += FDEC_STRIDE;
 303     }
 304 }
 305 static void predict_8x8c_p( uint8_t *src )
 306 {
 307     int i;
 308     int x,y;
 309     int a, b, c;
 310     int H = 0;
 311     int V = 0;
 312     int i00;
 313
 314     for( i = 0; i < 4; i++ )
 315     {
 316         H += ( i + 1 ) * ( src[4+i - FDEC_STRIDE] - src[2 - i -FDEC_STRIDE] );
 317         V += ( i + 1 ) * ( src[-1 +(i+4)*FDEC_STRIDE] - src[-1+(2-i)*FDEC_STRIDE] );
 318     }
 319
 320     a = 16 * ( src[-1+7*FDEC_STRIDE] + src[7 - FDEC_STRIDE] );
 321     b = ( 17 * H + 16 ) >> 5;
 322     c = ( 17 * V + 16 ) >> 5;
 323     i00 = a -3*b -3*c + 16;
 324
 325     for( y = 0; y < 8; y++ )
 326     {
 327         int pix = i00;
 328         for( x = 0; x < 8; x++ )
 329         {
 330             src[x] = x264_clip_uint8( pix>>5 );
 331             pix += b;
 332         }
 333         src += FDEC_STRIDE;
 334         i00 += c;
 335     }
 336 }
 337
 338 /****************************************************************************
 339  * 4x4 prediction for intra luma block
 340  ****************************************************************************/
 341
 342 #define SRC(x,y) src[(x)+(y)*FDEC_STRIDE]
 343 #define SRC32(x,y) *(uint32_t*)&SRC(x,y)
 344
 345 #define PREDICT_4x4_DC(v)\
 346     SRC32(0,0) = SRC32(0,1) = SRC32(0,2) = SRC32(0,3) = v;
 347
 348 static void predict_4x4_dc_128( uint8_t *src )
 349 {
 350     PREDICT_4x4_DC(0x80808080);
 351 }
 352 static void predict_4x4_dc_left( uint8_t *src )
 353 {
 354     uint32_t dc = ((SRC(-1,0) + SRC(-1,1) + SRC(-1,2) + SRC(-1,3) + 2) >> 2) * 0x01010101;
 355     PREDICT_4x4_DC(dc);
 356 }
 357 static void predict_4x4_dc_top( uint8_t *src )
 358 {
 359     uint32_t dc = ((SRC(0,-1) + SRC(1,-1) + SRC(2,-1) + SRC(3,-1) + 2) >> 2) * 0x01010101;
 360     PREDICT_4x4_DC(dc);
 361 }
 362 static void predict_4x4_dc( uint8_t *src )
 363 {
 364     uint32_t dc = ((SRC(-1,0) + SRC(-1,1) + SRC(-1,2) + SRC(-1,3) +
 365                     SRC(0,-1) + SRC(1,-1) + SRC(2,-1) + SRC(3,-1) + 4) >> 3) * 0x01010101;
 366     PREDICT_4x4_DC(dc);
 367 }
 368 static void predict_4x4_h( uint8_t *src )
 369 {
 370     SRC32(0,0) = SRC(-1,0) * 0x01010101;
 371     SRC32(0,1) = SRC(-1,1) * 0x01010101;
 372     SRC32(0,2) = SRC(-1,2) * 0x01010101;
 373     SRC32(0,3) = SRC(-1,3) * 0x01010101;
 374 }
 375 static void predict_4x4_v( uint8_t *src )
 376 {
 377     PREDICT_4x4_DC(SRC32(0,-1));
 378 }
 379
 380 #define PREDICT_4x4_LOAD_LEFT\
 381     const int l0 = SRC(-1,0);\
 382     const int l1 = SRC(-1,1);\
 383     const int l2 = SRC(-1,2);\
 384     UNUSED const int l3 = SRC(-1,3);
 385
 386 #define PREDICT_4x4_LOAD_TOP\
 387     const int t0 = SRC(0,-1);\
 388     const int t1 = SRC(1,-1);\
 389     const int t2 = SRC(2,-1);\
 390     UNUSED const int t3 = SRC(3,-1);
 391
 392 #define PREDICT_4x4_LOAD_TOP_RIGHT\
 393     const int t4 = SRC(4,-1);\
 394     const int t5 = SRC(5,-1);\
 395     const int t6 = SRC(6,-1);\
 396     UNUSED const int t7 = SRC(7,-1);
 397
 398 #define F1(a,b)   (((a)+(b)+1)>>1)
 399 #define F2(a,b,c) (((a)+2*(b)+(c)+2)>>2)
 400
 401 static void predict_4x4_ddl( uint8_t *src )
 402 {
 403     PREDICT_4x4_LOAD_TOP
 404     PREDICT_4x4_LOAD_TOP_RIGHT
 405     SRC(0,0)= F2(t0,t1,t2);
 406     SRC(1,0)=SRC(0,1)= F2(t1,t2,t3);
 407     SRC(2,0)=SRC(1,1)=SRC(0,2)= F2(t2,t3,t4);
 408     SRC(3,0)=SRC(2,1)=SRC(1,2)=SRC(0,3)= F2(t3,t4,t5);
 409     SRC(3,1)=SRC(2,2)=SRC(1,3)= F2(t4,t5,t6);
 410     SRC(3,2)=SRC(2,3)= F2(t5,t6,t7);
 411     SRC(3,3)= F2(t6,t7,t7);
 412 }
 413 static void predict_4x4_ddr( uint8_t *src )
 414 {
 415     const int lt = SRC(-1,-1);
 416     PREDICT_4x4_LOAD_LEFT
 417     PREDICT_4x4_LOAD_TOP
 418     SRC(3,0)= F2(t3,t2,t1);
 419     SRC(2,0)=SRC(3,1)= F2(t2,t1,t0);
 420     SRC(1,0)=SRC(2,1)=SRC(3,2)= F2(t1,t0,lt);
 421     SRC(0,0)=SRC(1,1)=SRC(2,2)=SRC(3,3)= F2(t0,lt,l0);
 422     SRC(0,1)=SRC(1,2)=SRC(2,3)= F2(lt,l0,l1);
 423     SRC(0,2)=SRC(1,3)= F2(l0,l1,l2);
 424     SRC(0,3)= F2(l1,l2,l3);
 425 }
 426
 427 static void predict_4x4_vr( uint8_t *src )
 428 {
 429     const int lt = SRC(-1,-1);
 430     PREDICT_4x4_LOAD_LEFT
 431     PREDICT_4x4_LOAD_TOP
 432     SRC(0,3)= F2(l2,l1,l0);
 433     SRC(0,2)= F2(l1,l0,lt);
 434     SRC(0,1)=SRC(1,3)= F2(l0,lt,t0);
 435     SRC(0,0)=SRC(1,2)= F1(lt,t0);
 436     SRC(1,1)=SRC(2,3)= F2(lt,t0,t1);
 437     SRC(1,0)=SRC(2,2)= F1(t0,t1);
 438     SRC(2,1)=SRC(3,3)= F2(t0,t1,t2);
 439     SRC(2,0)=SRC(3,2)= F1(t1,t2);
 440     SRC(3,1)= F2(t1,t2,t3);
 441     SRC(3,0)= F1(t2,t3);
 442 }
 443
 444 static void predict_4x4_hd( uint8_t *src )
 445 {
 446     const int lt= SRC(-1,-1);
 447     PREDICT_4x4_LOAD_LEFT
 448     PREDICT_4x4_LOAD_TOP
 449     SRC(0,3)= F1(l2,l3);
 450     SRC(1,3)= F2(l1,l2,l3);
 451     SRC(0,2)=SRC(2,3)= F1(l1,l2);
 452     SRC(1,2)=SRC(3,3)= F2(l0,l1,l2);
 453     SRC(0,1)=SRC(2,2)= F1(l0,l1);
 454     SRC(1,1)=SRC(3,2)= F2(lt,l0,l1);
 455     SRC(0,0)=SRC(2,1)= F1(lt,l0);
 456     SRC(1,0)=SRC(3,1)= F2(t0,lt,l0);
 457     SRC(2,0)= F2(t1,t0,lt);
 458     SRC(3,0)= F2(t2,t1,t0);
 459 }
 460
 461 static void predict_4x4_vl( uint8_t *src )
 462 {
 463     PREDICT_4x4_LOAD_TOP
 464     PREDICT_4x4_LOAD_TOP_RIGHT
 465     SRC(0,0)= F1(t0,t1);
 466     SRC(0,1)= F2(t0,t1,t2);
 467     SRC(1,0)=SRC(0,2)= F1(t1,t2);
 468     SRC(1,1)=SRC(0,3)= F2(t1,t2,t3);
 469     SRC(2,0)=SRC(1,2)= F1(t2,t3);
 470     SRC(2,1)=SRC(1,3)= F2(t2,t3,t4);
 471     SRC(3,0)=SRC(2,2)= F1(t3,t4);
 472     SRC(3,1)=SRC(2,3)= F2(t3,t4,t5);
 473     SRC(3,2)= F1(t4,t5);
 474     SRC(3,3)= F2(t4,t5,t6);
 475 }
 476
 477 static void predict_4x4_hu( uint8_t *src )
 478 {
 479     PREDICT_4x4_LOAD_LEFT
 480     SRC(0,0)= F1(l0,l1);
 481     SRC(1,0)= F2(l0,l1,l2);
 482     SRC(2,0)=SRC(0,1)= F1(l1,l2);
 483     SRC(3,0)=SRC(1,1)= F2(l1,l2,l3);
 484     SRC(2,1)=SRC(0,2)= F1(l2,l3);
 485     SRC(3,1)=SRC(1,2)= F2(l2,l3,l3);
 486     SRC(3,2)=SRC(1,3)=SRC(0,3)=
 487     SRC(2,2)=SRC(2,3)=SRC(3,3)= l3;
 488 }
 489
 490 /****************************************************************************
 491  * 8x8 prediction for intra luma block
 492  ****************************************************************************/
 493
 494 #define PL(y) \
 495     edge[14-y] = F2(SRC(-1,y-1), SRC(-1,y), SRC(-1,y+1));
 496 #define PT(x) \
 497     edge[16+x] = F2(SRC(x-1,-1), SRC(x,-1), SRC(x+1,-1));
 498
 499 void x264_predict_8x8_filter( uint8_t *src, uint8_t edge[33], int i_neighbor, int i_filters )
 500 {
 501     /* edge[7..14] = l7..l0
 502      * edge[15] = lt
 503      * edge[16..31] = t0 .. t15
 504      * edge[32] = t15 */
 505
 506     int have_lt = i_neighbor & MB_TOPLEFT;
 507     if( i_filters & MB_LEFT )
 508     {
 509         edge[15] = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2;
 510         edge[14] = ((have_lt ? SRC(-1,-1) : SRC(-1,0))
 511                     + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2;
 512         PL(1) PL(2) PL(3) PL(4) PL(5) PL(6)
 513         edge[7] = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2;
 514     }
 515
 516     if( i_filters & MB_TOP )
 517     {
 518         int have_tr = i_neighbor & MB_TOPRIGHT;
 519         edge[16] = ((have_lt ? SRC(-1,-1) : SRC(0,-1))
 520                     + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2;
 521         PT(1) PT(2) PT(3) PT(4) PT(5) PT(6)
 522         edge[23] = ((have_tr ? SRC(8,-1) : SRC(7,-1))
 523                     + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2;
 524
 525         if( i_filters & MB_TOPRIGHT )
 526         {
 527             if( have_tr )
 528             {
 529                 PT(8) PT(9) PT(10) PT(11) PT(12) PT(13) PT(14)
 530                 edge[31] =
 531                 edge[32] = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2;
 532             }
 533             else
 534             {
 535                 *(uint64_t*)(edge+24) = SRC(7,-1) * 0x0101010101010101ULL;
 536                 edge[32] = SRC(7,-1);
 537             }
 538         }
 539     }
 540 }
 541
 542 #undef PL
 543 #undef PT
 544
 545 #define PL(y) \
 546     UNUSED const int l##y = edge[14-y];
 547 #define PT(x) \
 548     UNUSED const int t##x = edge[16+x];
 549 #define PREDICT_8x8_LOAD_TOPLEFT \
 550     const int lt = edge[15];
 551 #define PREDICT_8x8_LOAD_LEFT \
 552     PL(0) PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) PL(7)
 553 #define PREDICT_8x8_LOAD_TOP \
 554     PT(0) PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) PT(7)
 555 #define PREDICT_8x8_LOAD_TOPRIGHT \
 556     PT(8) PT(9) PT(10) PT(11) PT(12) PT(13) PT(14) PT(15)
 557
 558 #define PREDICT_8x8_DC(v) \
 559     int y; \
 560     for( y = 0; y < 8; y++ ) { \
 561         ((uint32_t*)src)[0] = \
 562         ((uint32_t*)src)[1] = v; \
 563         src += FDEC_STRIDE; \
 564     }
 565
 566 /* SIMD is much faster than C for all of these except HU and HD. */
 567 static void predict_8x8_dc_128( uint8_t *src, uint8_t edge[33] )
 568 {
 569     PREDICT_8x8_DC(0x80808080);
 570 }
 571 static void predict_8x8_dc_left( uint8_t *src, uint8_t edge[33] )
 572 {
 573     PREDICT_8x8_LOAD_LEFT
 574     const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3) * 0x01010101;
 575     PREDICT_8x8_DC(dc);
 576 }
 577 static void predict_8x8_dc_top( uint8_t *src, uint8_t edge[33] )
 578 {
 579     PREDICT_8x8_LOAD_TOP
 580     const uint32_t dc = ((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3) * 0x01010101;
 581     PREDICT_8x8_DC(dc);
 582 }
 583 static void predict_8x8_dc( uint8_t *src, uint8_t edge[33] )
 584 {
 585     PREDICT_8x8_LOAD_LEFT
 586     PREDICT_8x8_LOAD_TOP
 587     const uint32_t dc = ((l0+l1+l2+l3+l4+l5+l6+l7
 588                          +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4) * 0x01010101;
 589     PREDICT_8x8_DC(dc);
 590 }
 591 static void predict_8x8_h( uint8_t *src, uint8_t edge[33] )
 592 {
 593     PREDICT_8x8_LOAD_LEFT
 594 #define ROW(y) ((uint32_t*)(src+y*FDEC_STRIDE))[0] =\
 595                ((uint32_t*)(src+y*FDEC_STRIDE))[1] = 0x01010101U * l##y
 596     ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
 597 #undef ROW
 598 }
 599 static void predict_8x8_v( uint8_t *src, uint8_t edge[33] )
 600 {
 601     const uint64_t top = *(uint64_t*)(edge+16);
 602     int y;
 603     for( y = 0; y < 8; y++ )
 604         *(uint64_t*)(src+y*FDEC_STRIDE) = top;
 605 }
 606 static void predict_8x8_ddl( uint8_t *src, uint8_t edge[33] )
 607 {
 608     PREDICT_8x8_LOAD_TOP
 609     PREDICT_8x8_LOAD_TOPRIGHT
 610     SRC(0,0)= F2(t0,t1,t2);
 611     SRC(0,1)=SRC(1,0)= F2(t1,t2,t3);
 612     SRC(0,2)=SRC(1,1)=SRC(2,0)= F2(t2,t3,t4);
 613     SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= F2(t3,t4,t5);
 614     SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= F2(t4,t5,t6);
 615     SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= F2(t5,t6,t7);
 616     SRC(0,6)=SRC(1,5)=SRC(2,4)=SRC(3,3)=SRC(4,2)=SRC(5,1)=SRC(6,0)= F2(t6,t7,t8);
 617     SRC(0,7)=SRC(1,6)=SRC(2,5)=SRC(3,4)=SRC(4,3)=SRC(5,2)=SRC(6,1)=SRC(7,0)= F2(t7,t8,t9);
 618     SRC(1,7)=SRC(2,6)=SRC(3,5)=SRC(4,4)=SRC(5,3)=SRC(6,2)=SRC(7,1)= F2(t8,t9,t10);
 619     SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= F2(t9,t10,t11);
 620     SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= F2(t10,t11,t12);
 621     SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= F2(t11,t12,t13);
 622     SRC(5,7)=SRC(6,6)=SRC(7,5)= F2(t12,t13,t14);
 623     SRC(6,7)=SRC(7,6)= F2(t13,t14,t15);
 624     SRC(7,7)= F2(t14,t15,t15);
 625 }
 626 static void predict_8x8_ddr( uint8_t *src, uint8_t edge[33] )
 627 {
 628     PREDICT_8x8_LOAD_TOP
 629     PREDICT_8x8_LOAD_LEFT
 630     PREDICT_8x8_LOAD_TOPLEFT
 631     SRC(0,7)= F2(l7,l6,l5);
 632     SRC(0,6)=SRC(1,7)= F2(l6,l5,l4);
 633     SRC(0,5)=SRC(1,6)=SRC(2,7)= F2(l5,l4,l3);
 634     SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= F2(l4,l3,l2);
 635     SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= F2(l3,l2,l1);
 636     SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= F2(l2,l1,l0);
 637     SRC(0,1)=SRC(1,2)=SRC(2,3)=SRC(3,4)=SRC(4,5)=SRC(5,6)=SRC(6,7)= F2(l1,l0,lt);
 638     SRC(0,0)=SRC(1,1)=SRC(2,2)=SRC(3,3)=SRC(4,4)=SRC(5,5)=SRC(6,6)=SRC(7,7)= F2(l0,lt,t0);
 639     SRC(1,0)=SRC(2,1)=SRC(3,2)=SRC(4,3)=SRC(5,4)=SRC(6,5)=SRC(7,6)= F2(lt,t0,t1);
 640     SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= F2(t0,t1,t2);
 641     SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= F2(t1,t2,t3);
 642     SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= F2(t2,t3,t4);
 643     SRC(5,0)=SRC(6,1)=SRC(7,2)= F2(t3,t4,t5);
 644     SRC(6,0)=SRC(7,1)= F2(t4,t5,t6);
 645     SRC(7,0)= F2(t5,t6,t7);
 646
 647 }
 648 static void predict_8x8_vr( uint8_t *src, uint8_t edge[33] )
 649 {
 650     PREDICT_8x8_LOAD_TOP
 651     PREDICT_8x8_LOAD_LEFT
 652     PREDICT_8x8_LOAD_TOPLEFT
 653     SRC(0,6)= F2(l5,l4,l3);
 654     SRC(0,7)= F2(l6,l5,l4);
 655     SRC(0,4)=SRC(1,6)= F2(l3,l2,l1);
 656     SRC(0,5)=SRC(1,7)= F2(l4,l3,l2);
 657     SRC(0,2)=SRC(1,4)=SRC(2,6)= F2(l1,l0,lt);
 658     SRC(0,3)=SRC(1,5)=SRC(2,7)= F2(l2,l1,l0);
 659     SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= F2(l0,lt,t0);
 660     SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= F1(lt,t0);
 661     SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= F2(lt,t0,t1);
 662     SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= F1(t0,t1);
 663     SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= F2(t0,t1,t2);
 664     SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= F1(t1,t2);
 665     SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= F2(t1,t2,t3);
 666     SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= F1(t2,t3);
 667     SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= F2(t2,t3,t4);
 668     SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= F1(t3,t4);
 669     SRC(5,1)=SRC(6,3)=SRC(7,5)= F2(t3,t4,t5);
 670     SRC(5,0)=SRC(6,2)=SRC(7,4)= F1(t4,t5);
 671     SRC(6,1)=SRC(7,3)= F2(t4,t5,t6);
 672     SRC(6,0)=SRC(7,2)= F1(t5,t6);
 673     SRC(7,1)= F2(t5,t6,t7);
 674     SRC(7,0)= F1(t6,t7);
 675 }
 676 static void predict_8x8_hd( uint8_t *src, uint8_t edge[33] )
 677 {
 678     PREDICT_8x8_LOAD_TOP
 679     PREDICT_8x8_LOAD_LEFT
 680     PREDICT_8x8_LOAD_TOPLEFT
 681     int p1 = pack8to16(F1(l6,l7), F2(l5,l6,l7));
 682     int p2 = pack8to16(F1(l5,l6), F2(l4,l5,l6));
 683     int p3 = pack8to16(F1(l4,l5), F2(l3,l4,l5));
 684     int p4 = pack8to16(F1(l3,l4), F2(l2,l3,l4));
 685     int p5 = pack8to16(F1(l2,l3), F2(l1,l2,l3));
 686     int p6 = pack8to16(F1(l1,l2), F2(l0,l1,l2));
 687     int p7 = pack8to16(F1(l0,l1), F2(lt,l0,l1));
 688     int p8 = pack8to16(F1(lt,l0), F2(l0,lt,t0));
 689     int p9 = pack8to16(F2(t1,t0,lt), F2(t2,t1,t0));
 690     int p10 = pack8to16(F2(t3,t2,t1), F2(t4,t3,t2));
 691     int p11 = pack8to16(F2(t5,t4,t3), F2(t6,t5,t4));
 692     SRC32(0,7)= pack16to32(p1,p2);
 693     SRC32(0,6)= pack16to32(p2,p3);
 694     SRC32(4,7)=SRC32(0,5)= pack16to32(p3,p4);
 695     SRC32(4,6)=SRC32(0,4)= pack16to32(p4,p5);
 696     SRC32(4,5)=SRC32(0,3)= pack16to32(p5,p6);
 697     SRC32(4,4)=SRC32(0,2)= pack16to32(p6,p7);
 698     SRC32(4,3)=SRC32(0,1)= pack16to32(p7,p8);
 699     SRC32(4,2)=SRC32(0,0)= pack16to32(p8,p9);
 700     SRC32(4,1)= pack16to32(p9,p10);
 701     SRC32(4,0)= pack16to32(p10,p11);
 702 }
 703 static void predict_8x8_vl( uint8_t *src, uint8_t edge[33] )
 704 {
 705     PREDICT_8x8_LOAD_TOP
 706     PREDICT_8x8_LOAD_TOPRIGHT
 707     SRC(0,0)= F1(t0,t1);
 708     SRC(0,1)= F2(t0,t1,t2);
 709     SRC(0,2)=SRC(1,0)= F1(t1,t2);
 710     SRC(0,3)=SRC(1,1)= F2(t1,t2,t3);
 711     SRC(0,4)=SRC(1,2)=SRC(2,0)= F1(t2,t3);
 712     SRC(0,5)=SRC(1,3)=SRC(2,1)= F2(t2,t3,t4);
 713     SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= F1(t3,t4);
 714     SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= F2(t3,t4,t5);
 715     SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= F1(t4,t5);
 716     SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= F2(t4,t5,t6);
 717     SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= F1(t5,t6);
 718     SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= F2(t5,t6,t7);
 719     SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= F1(t6,t7);
 720     SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= F2(t6,t7,t8);
 721     SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= F1(t7,t8);
 722     SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= F2(t7,t8,t9);
 723     SRC(5,6)=SRC(6,4)=SRC(7,2)= F1(t8,t9);
 724     SRC(5,7)=SRC(6,5)=SRC(7,3)= F2(t8,t9,t10);
 725     SRC(6,6)=SRC(7,4)= F1(t9,t10);
 726     SRC(6,7)=SRC(7,5)= F2(t9,t10,t11);
 727     SRC(7,6)= F1(t10,t11);
 728     SRC(7,7)= F2(t10,t11,t12);
 729 }
 730 static void predict_8x8_hu( uint8_t *src, uint8_t edge[33] )
 731 {
 732     PREDICT_8x8_LOAD_LEFT
 733     int p1 = pack8to16(F1(l0,l1), F2(l0,l1,l2));
 734     int p2 = pack8to16(F1(l1,l2), F2(l1,l2,l3));
 735     int p3 = pack8to16(F1(l2,l3), F2(l2,l3,l4));
 736     int p4 = pack8to16(F1(l3,l4), F2(l3,l4,l5));
 737     int p5 = pack8to16(F1(l4,l5), F2(l4,l5,l6));
 738     int p6 = pack8to16(F1(l5,l6), F2(l5,l6,l7));
 739     int p7 = pack8to16(F1(l6,l7), F2(l6,l7,l7));
 740     int p8 = pack8to16(l7,l7);
 741     SRC32(0,0)= pack16to32(p1,p2);
 742     SRC32(0,1)= pack16to32(p2,p3);
 743     SRC32(4,0)=SRC32(0,2)= pack16to32(p3,p4);
 744     SRC32(4,1)=SRC32(0,3)= pack16to32(p4,p5);
 745     SRC32(4,2)=SRC32(0,4)= pack16to32(p5,p6);
 746     SRC32(4,3)=SRC32(0,5)= pack16to32(p6,p7);
 747     SRC32(4,4)=SRC32(0,6)= pack16to32(p7,p8);
 748     SRC32(4,5)=SRC32(4,6)= SRC32(0,7) = SRC32(4,7) = pack16to32(p8,p8);
 749 }
 750
 751 /****************************************************************************
 752  * Exported functions:
 753  ****************************************************************************/
 754 void x264_predict_16x16_init( int cpu, x264_predict_t pf[7] )
 755 {
 756     pf[I_PRED_16x16_V ]     = predict_16x16_v;
 757     pf[I_PRED_16x16_H ]     = predict_16x16_h;
 758     pf[I_PRED_16x16_DC]     = predict_16x16_dc;
 759     pf[I_PRED_16x16_P ]     = predict_16x16_p;
 760     pf[I_PRED_16x16_DC_LEFT]= predict_16x16_dc_left;
 761     pf[I_PRED_16x16_DC_TOP ]= predict_16x16_dc_top;
 762     pf[I_PRED_16x16_DC_128 ]= predict_16x16_dc_128;
 763
 764 #ifdef HAVE_MMX
 765     x264_predict_16x16_init_mmx( cpu, pf );
 766 #endif
 767
 768 #ifdef ARCH_PPC
 769     if( cpu&X264_CPU_ALTIVEC )
 770     {
 771         x264_predict_16x16_init_altivec( pf );
 772     }
 773 #endif
 774 }
 775
 776 void x264_predict_8x8c_init( int cpu, x264_predict_t pf[7] )
 777 {
 778     pf[I_PRED_CHROMA_V ]     = predict_8x8c_v;
 779     pf[I_PRED_CHROMA_H ]     = predict_8x8c_h;
 780     pf[I_PRED_CHROMA_DC]     = predict_8x8c_dc;
 781     pf[I_PRED_CHROMA_P ]     = predict_8x8c_p;
 782     pf[I_PRED_CHROMA_DC_LEFT]= predict_8x8c_dc_left;
 783     pf[I_PRED_CHROMA_DC_TOP ]= predict_8x8c_dc_top;
 784     pf[I_PRED_CHROMA_DC_128 ]= predict_8x8c_dc_128;
 785
 786 #ifdef HAVE_MMX
 787     x264_predict_8x8c_init_mmx( cpu, pf );
 788 #endif
 789 }
 790
 791 void x264_predict_8x8_init( int cpu, x264_predict8x8_t pf[12] )
 792 {
 793     pf[I_PRED_8x8_V]      = predict_8x8_v;
 794     pf[I_PRED_8x8_H]      = predict_8x8_h;
 795     pf[I_PRED_8x8_DC]     = predict_8x8_dc;
 796     pf[I_PRED_8x8_DDL]    = predict_8x8_ddl;
 797     pf[I_PRED_8x8_DDR]    = predict_8x8_ddr;
 798     pf[I_PRED_8x8_VR]     = predict_8x8_vr;
 799     pf[I_PRED_8x8_HD]     = predict_8x8_hd;
 800     pf[I_PRED_8x8_VL]     = predict_8x8_vl;
 801     pf[I_PRED_8x8_HU]     = predict_8x8_hu;
 802     pf[I_PRED_8x8_DC_LEFT]= predict_8x8_dc_left;
 803     pf[I_PRED_8x8_DC_TOP] = predict_8x8_dc_top;
 804     pf[I_PRED_8x8_DC_128] = predict_8x8_dc_128;
 805
 806 #ifdef HAVE_MMX
 807     x264_predict_8x8_init_mmx( cpu, pf );
 808 #endif
 809 }
 810
 811 void x264_predict_4x4_init( int cpu, x264_predict_t pf[12] )
 812 {
 813     pf[I_PRED_4x4_V]      = predict_4x4_v;
 814     pf[I_PRED_4x4_H]      = predict_4x4_h;
 815     pf[I_PRED_4x4_DC]     = predict_4x4_dc;
 816     pf[I_PRED_4x4_DDL]    = predict_4x4_ddl;
 817     pf[I_PRED_4x4_DDR]    = predict_4x4_ddr;
 818     pf[I_PRED_4x4_VR]     = predict_4x4_vr;
 819     pf[I_PRED_4x4_HD]     = predict_4x4_hd;
 820     pf[I_PRED_4x4_VL]     = predict_4x4_vl;
 821     pf[I_PRED_4x4_HU]     = predict_4x4_hu;
 822     pf[I_PRED_4x4_DC_LEFT]= predict_4x4_dc_left;
 823     pf[I_PRED_4x4_DC_TOP] = predict_4x4_dc_top;
 824     pf[I_PRED_4x4_DC_128] = predict_4x4_dc_128;
 825
 826 #ifdef HAVE_MMX
 827     x264_predict_4x4_init_mmx( cpu, pf );
 828 #endif
 829 }
 830