libavcodec/h264pred_template.c

   1 /*
   2  * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
   3  * Copyright (c) 2003-2011 Michael Niedermayer <michaelni@gmx.at>
   4  *
   5  * This file is part of Libav.
   6  *
   7  * Libav is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * Libav is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with Libav; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 /**
  23  * @file
  24  * H.264 / AVC / MPEG-4 part10 prediction functions.
  25  * @author Michael Niedermayer <michaelni@gmx.at>
  26  */
  27
  28 #include "libavutil/intreadwrite.h"
  29
  30 #include "mathops.h"
  31
  32 #include "bit_depth_template.c"
  33
  34 static void FUNCC(pred4x4_vertical)(uint8_t *_src, const uint8_t *topright,
  35                                     ptrdiff_t _stride)
  36 {
  37     pixel *src = (pixel*)_src;
  38     int stride = _stride/sizeof(pixel);
  39     const pixel4 a= AV_RN4PA(src-stride);
  40
  41     AV_WN4PA(src+0*stride, a);
  42     AV_WN4PA(src+1*stride, a);
  43     AV_WN4PA(src+2*stride, a);
  44     AV_WN4PA(src+3*stride, a);
  45 }
  46
  47 static void FUNCC(pred4x4_horizontal)(uint8_t *_src, const uint8_t *topright,
  48                                       ptrdiff_t _stride)
  49 {
  50     pixel *src = (pixel*)_src;
  51     int stride = _stride/sizeof(pixel);
  52     AV_WN4PA(src+0*stride, PIXEL_SPLAT_X4(src[-1+0*stride]));
  53     AV_WN4PA(src+1*stride, PIXEL_SPLAT_X4(src[-1+1*stride]));
  54     AV_WN4PA(src+2*stride, PIXEL_SPLAT_X4(src[-1+2*stride]));
  55     AV_WN4PA(src+3*stride, PIXEL_SPLAT_X4(src[-1+3*stride]));
  56 }
  57
  58 static void FUNCC(pred4x4_dc)(uint8_t *_src, const uint8_t *topright,
  59                               ptrdiff_t _stride)
  60 {
  61     pixel *src = (pixel*)_src;
  62     int stride = _stride/sizeof(pixel);
  63     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
  64                    + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
  65     const pixel4 a = PIXEL_SPLAT_X4(dc);
  66
  67     AV_WN4PA(src+0*stride, a);
  68     AV_WN4PA(src+1*stride, a);
  69     AV_WN4PA(src+2*stride, a);
  70     AV_WN4PA(src+3*stride, a);
  71 }
  72
  73 static void FUNCC(pred4x4_left_dc)(uint8_t *_src, const uint8_t *topright,
  74                                    ptrdiff_t _stride)
  75 {
  76     pixel *src = (pixel*)_src;
  77     int stride = _stride/sizeof(pixel);
  78     const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
  79     const pixel4 a = PIXEL_SPLAT_X4(dc);
  80
  81     AV_WN4PA(src+0*stride, a);
  82     AV_WN4PA(src+1*stride, a);
  83     AV_WN4PA(src+2*stride, a);
  84     AV_WN4PA(src+3*stride, a);
  85 }
  86
  87 static void FUNCC(pred4x4_top_dc)(uint8_t *_src, const uint8_t *topright,
  88                                   ptrdiff_t _stride)
  89 {
  90     pixel *src = (pixel*)_src;
  91     int stride = _stride/sizeof(pixel);
  92     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
  93     const pixel4 a = PIXEL_SPLAT_X4(dc);
  94
  95     AV_WN4PA(src+0*stride, a);
  96     AV_WN4PA(src+1*stride, a);
  97     AV_WN4PA(src+2*stride, a);
  98     AV_WN4PA(src+3*stride, a);
  99 }
 100
 101 static void FUNCC(pred4x4_128_dc)(uint8_t *_src, const uint8_t *topright,
 102                                   ptrdiff_t _stride)
 103 {
 104     pixel *src = (pixel*)_src;
 105     int stride = _stride/sizeof(pixel);
 106     const pixel4 a = PIXEL_SPLAT_X4(1<<(BIT_DEPTH-1));
 107
 108     AV_WN4PA(src+0*stride, a);
 109     AV_WN4PA(src+1*stride, a);
 110     AV_WN4PA(src+2*stride, a);
 111     AV_WN4PA(src+3*stride, a);
 112 }
 113
 114 static void FUNCC(pred4x4_127_dc)(uint8_t *_src, const uint8_t *topright,
 115                                   ptrdiff_t _stride)
 116 {
 117     pixel *src = (pixel*)_src;
 118     int stride = _stride/sizeof(pixel);
 119     const pixel4 a = PIXEL_SPLAT_X4((1<<(BIT_DEPTH-1))-1);
 120
 121     AV_WN4PA(src+0*stride, a);
 122     AV_WN4PA(src+1*stride, a);
 123     AV_WN4PA(src+2*stride, a);
 124     AV_WN4PA(src+3*stride, a);
 125 }
 126
 127 static void FUNCC(pred4x4_129_dc)(uint8_t *_src, const uint8_t *topright,
 128                                   ptrdiff_t _stride)
 129 {
 130     pixel *src = (pixel*)_src;
 131     int stride = _stride/sizeof(pixel);
 132     const pixel4 a = PIXEL_SPLAT_X4((1<<(BIT_DEPTH-1))+1);
 133
 134     AV_WN4PA(src+0*stride, a);
 135     AV_WN4PA(src+1*stride, a);
 136     AV_WN4PA(src+2*stride, a);
 137     AV_WN4PA(src+3*stride, a);
 138 }
 139
 140
 141 #define LOAD_TOP_RIGHT_EDGE\
 142     const unsigned av_unused t4 = topright[0];\
 143     const unsigned av_unused t5 = topright[1];\
 144     const unsigned av_unused t6 = topright[2];\
 145     const unsigned av_unused t7 = topright[3];\
 146
 147 #define LOAD_DOWN_LEFT_EDGE\
 148     const unsigned av_unused l4 = src[-1+4*stride];\
 149     const unsigned av_unused l5 = src[-1+5*stride];\
 150     const unsigned av_unused l6 = src[-1+6*stride];\
 151     const unsigned av_unused l7 = src[-1+7*stride];\
 152
 153 #define LOAD_LEFT_EDGE\
 154     const unsigned av_unused l0 = src[-1+0*stride];\
 155     const unsigned av_unused l1 = src[-1+1*stride];\
 156     const unsigned av_unused l2 = src[-1+2*stride];\
 157     const unsigned av_unused l3 = src[-1+3*stride];\
 158
 159 #define LOAD_TOP_EDGE\
 160     const unsigned av_unused t0 = src[ 0-1*stride];\
 161     const unsigned av_unused t1 = src[ 1-1*stride];\
 162     const unsigned av_unused t2 = src[ 2-1*stride];\
 163     const unsigned av_unused t3 = src[ 3-1*stride];\
 164
 165 static void FUNCC(pred4x4_down_right)(uint8_t *_src, const uint8_t *topright,
 166                                       ptrdiff_t _stride)
 167 {
 168     pixel *src = (pixel*)_src;
 169     int stride = _stride/sizeof(pixel);
 170     const int lt= src[-1-1*stride];
 171     LOAD_TOP_EDGE
 172     LOAD_LEFT_EDGE
 173
 174     src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
 175     src[0+2*stride]=
 176     src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
 177     src[0+1*stride]=
 178     src[1+2*stride]=
 179     src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
 180     src[0+0*stride]=
 181     src[1+1*stride]=
 182     src[2+2*stride]=
 183     src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
 184     src[1+0*stride]=
 185     src[2+1*stride]=
 186     src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
 187     src[2+0*stride]=
 188     src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
 189     src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
 190 }
 191
 192 static void FUNCC(pred4x4_down_left)(uint8_t *_src, const uint8_t *_topright,
 193                                      ptrdiff_t _stride)
 194 {
 195     pixel *src = (pixel*)_src;
 196     const pixel *topright = (const pixel*)_topright;
 197     int stride = _stride/sizeof(pixel);
 198     LOAD_TOP_EDGE
 199     LOAD_TOP_RIGHT_EDGE
 200 //    LOAD_LEFT_EDGE
 201
 202     src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
 203     src[1+0*stride]=
 204     src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
 205     src[2+0*stride]=
 206     src[1+1*stride]=
 207     src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
 208     src[3+0*stride]=
 209     src[2+1*stride]=
 210     src[1+2*stride]=
 211     src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
 212     src[3+1*stride]=
 213     src[2+2*stride]=
 214     src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
 215     src[3+2*stride]=
 216     src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
 217     src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
 218 }
 219
 220 static void FUNCC(pred4x4_vertical_right)(uint8_t *_src,
 221                                           const uint8_t *topright,
 222                                           ptrdiff_t _stride)
 223 {
 224     pixel *src = (pixel*)_src;
 225     int stride = _stride/sizeof(pixel);
 226     const int lt= src[-1-1*stride];
 227     LOAD_TOP_EDGE
 228     LOAD_LEFT_EDGE
 229
 230     src[0+0*stride]=
 231     src[1+2*stride]=(lt + t0 + 1)>>1;
 232     src[1+0*stride]=
 233     src[2+2*stride]=(t0 + t1 + 1)>>1;
 234     src[2+0*stride]=
 235     src[3+2*stride]=(t1 + t2 + 1)>>1;
 236     src[3+0*stride]=(t2 + t3 + 1)>>1;
 237     src[0+1*stride]=
 238     src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
 239     src[1+1*stride]=
 240     src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
 241     src[2+1*stride]=
 242     src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
 243     src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
 244     src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
 245     src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
 246 }
 247
 248 static void FUNCC(pred4x4_vertical_left)(uint8_t *_src,
 249                                          const uint8_t *_topright,
 250                                          ptrdiff_t _stride)
 251 {
 252     pixel *src = (pixel*)_src;
 253     const pixel *topright = (const pixel*)_topright;
 254     int stride = _stride/sizeof(pixel);
 255     LOAD_TOP_EDGE
 256     LOAD_TOP_RIGHT_EDGE
 257
 258     src[0+0*stride]=(t0 + t1 + 1)>>1;
 259     src[1+0*stride]=
 260     src[0+2*stride]=(t1 + t2 + 1)>>1;
 261     src[2+0*stride]=
 262     src[1+2*stride]=(t2 + t3 + 1)>>1;
 263     src[3+0*stride]=
 264     src[2+2*stride]=(t3 + t4+ 1)>>1;
 265     src[3+2*stride]=(t4 + t5+ 1)>>1;
 266     src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
 267     src[1+1*stride]=
 268     src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
 269     src[2+1*stride]=
 270     src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
 271     src[3+1*stride]=
 272     src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
 273     src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
 274 }
 275
 276 static void FUNCC(pred4x4_horizontal_up)(uint8_t *_src, const uint8_t *topright,
 277                                          ptrdiff_t _stride)
 278 {
 279     pixel *src = (pixel*)_src;
 280     int stride = _stride/sizeof(pixel);
 281     LOAD_LEFT_EDGE
 282
 283     src[0+0*stride]=(l0 + l1 + 1)>>1;
 284     src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
 285     src[2+0*stride]=
 286     src[0+1*stride]=(l1 + l2 + 1)>>1;
 287     src[3+0*stride]=
 288     src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
 289     src[2+1*stride]=
 290     src[0+2*stride]=(l2 + l3 + 1)>>1;
 291     src[3+1*stride]=
 292     src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
 293     src[3+2*stride]=
 294     src[1+3*stride]=
 295     src[0+3*stride]=
 296     src[2+2*stride]=
 297     src[2+3*stride]=
 298     src[3+3*stride]=l3;
 299 }
 300
 301 static void FUNCC(pred4x4_horizontal_down)(uint8_t *_src,
 302                                            const uint8_t *topright,
 303                                            ptrdiff_t _stride)
 304 {
 305     pixel *src = (pixel*)_src;
 306     int stride = _stride/sizeof(pixel);
 307     const int lt= src[-1-1*stride];
 308     LOAD_TOP_EDGE
 309     LOAD_LEFT_EDGE
 310
 311     src[0+0*stride]=
 312     src[2+1*stride]=(lt + l0 + 1)>>1;
 313     src[1+0*stride]=
 314     src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
 315     src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
 316     src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
 317     src[0+1*stride]=
 318     src[2+2*stride]=(l0 + l1 + 1)>>1;
 319     src[1+1*stride]=
 320     src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
 321     src[0+2*stride]=
 322     src[2+3*stride]=(l1 + l2+ 1)>>1;
 323     src[1+2*stride]=
 324     src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
 325     src[0+3*stride]=(l2 + l3 + 1)>>1;
 326     src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
 327 }
 328
 329 static void FUNCC(pred16x16_vertical)(uint8_t *_src, ptrdiff_t _stride)
 330 {
 331     int i;
 332     pixel *src = (pixel*)_src;
 333     int stride = _stride/sizeof(pixel);
 334     const pixel4 a = AV_RN4PA(((pixel4*)(src-stride))+0);
 335     const pixel4 b = AV_RN4PA(((pixel4*)(src-stride))+1);
 336     const pixel4 c = AV_RN4PA(((pixel4*)(src-stride))+2);
 337     const pixel4 d = AV_RN4PA(((pixel4*)(src-stride))+3);
 338
 339     for(i=0; i<16; i++){
 340         AV_WN4PA(((pixel4*)(src+i*stride))+0, a);
 341         AV_WN4PA(((pixel4*)(src+i*stride))+1, b);
 342         AV_WN4PA(((pixel4*)(src+i*stride))+2, c);
 343         AV_WN4PA(((pixel4*)(src+i*stride))+3, d);
 344     }
 345 }
 346
 347 static void FUNCC(pred16x16_horizontal)(uint8_t *_src, ptrdiff_t stride)
 348 {
 349     int i;
 350     pixel *src = (pixel*)_src;
 351     stride /= sizeof(pixel);
 352
 353     for(i=0; i<16; i++){
 354         const pixel4 a = PIXEL_SPLAT_X4(src[-1+i*stride]);
 355
 356         AV_WN4PA(((pixel4*)(src+i*stride))+0, a);
 357         AV_WN4PA(((pixel4*)(src+i*stride))+1, a);
 358         AV_WN4PA(((pixel4*)(src+i*stride))+2, a);
 359         AV_WN4PA(((pixel4*)(src+i*stride))+3, a);
 360     }
 361 }
 362
 363 #define PREDICT_16x16_DC(v)\
 364     for(i=0; i<16; i++){\
 365         AV_WN4PA(src+ 0, v);\
 366         AV_WN4PA(src+ 4, v);\
 367         AV_WN4PA(src+ 8, v);\
 368         AV_WN4PA(src+12, v);\
 369         src += stride;\
 370     }
 371
 372 static void FUNCC(pred16x16_dc)(uint8_t *_src, ptrdiff_t stride)
 373 {
 374     int i, dc=0;
 375     pixel *src = (pixel*)_src;
 376     pixel4 dcsplat;
 377     stride /= sizeof(pixel);
 378
 379     for(i=0;i<16; i++){
 380         dc+= src[-1+i*stride];
 381     }
 382
 383     for(i=0;i<16; i++){
 384         dc+= src[i-stride];
 385     }
 386
 387     dcsplat = PIXEL_SPLAT_X4((dc+16)>>5);
 388     PREDICT_16x16_DC(dcsplat);
 389 }
 390
 391 static void FUNCC(pred16x16_left_dc)(uint8_t *_src, ptrdiff_t stride)
 392 {
 393     int i, dc=0;
 394     pixel *src = (pixel*)_src;
 395     pixel4 dcsplat;
 396     stride /= sizeof(pixel);
 397
 398     for(i=0;i<16; i++){
 399         dc+= src[-1+i*stride];
 400     }
 401
 402     dcsplat = PIXEL_SPLAT_X4((dc+8)>>4);
 403     PREDICT_16x16_DC(dcsplat);
 404 }
 405
 406 static void FUNCC(pred16x16_top_dc)(uint8_t *_src, ptrdiff_t stride)
 407 {
 408     int i, dc=0;
 409     pixel *src = (pixel*)_src;
 410     pixel4 dcsplat;
 411     stride /= sizeof(pixel);
 412
 413     for(i=0;i<16; i++){
 414         dc+= src[i-stride];
 415     }
 416
 417     dcsplat = PIXEL_SPLAT_X4((dc+8)>>4);
 418     PREDICT_16x16_DC(dcsplat);
 419 }
 420
 421 #define PRED16x16_X(n, v) \
 422 static void FUNCC(pred16x16_##n##_dc)(uint8_t *_src, ptrdiff_t stride)\
 423 {\
 424     int i;\
 425     pixel *src = (pixel*)_src;\
 426     stride /= sizeof(pixel);\
 427     PREDICT_16x16_DC(PIXEL_SPLAT_X4(v));\
 428 }
 429
 430 PRED16x16_X(127, (1<<(BIT_DEPTH-1))-1)
 431 PRED16x16_X(128, (1<<(BIT_DEPTH-1))+0)
 432 PRED16x16_X(129, (1<<(BIT_DEPTH-1))+1)
 433
 434 static inline void FUNCC(pred16x16_plane_compat)(uint8_t *_src,
 435                                                  ptrdiff_t _stride,
 436                                                  const int svq3,
 437                                                  const int rv40)
 438 {
 439   int i, j, k;
 440   int a;
 441   INIT_CLIP
 442   pixel *src = (pixel*)_src;
 443   int stride = _stride/sizeof(pixel);
 444   const pixel * const src0 = src +7-stride;
 445   const pixel *       src1 = src +8*stride-1;
 446   const pixel *       src2 = src1-2*stride;    // == src+6*stride-1;
 447   int H = src0[1] - src0[-1];
 448   int V = src1[0] - src2[ 0];
 449   for(k=2; k<=8; ++k) {
 450     src1 += stride; src2 -= stride;
 451     H += k*(src0[k] - src0[-k]);
 452     V += k*(src1[0] - src2[ 0]);
 453   }
 454   if(svq3){
 455     H = ( 5*(H/4) ) / 16;
 456     V = ( 5*(V/4) ) / 16;
 457
 458     /* required for 100% accuracy */
 459     i = H; H = V; V = i;
 460   }else if(rv40){
 461     H = ( H + (H>>2) ) >> 4;
 462     V = ( V + (V>>2) ) >> 4;
 463   }else{
 464     H = ( 5*H+32 ) >> 6;
 465     V = ( 5*V+32 ) >> 6;
 466   }
 467
 468   a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
 469   for(j=16; j>0; --j) {
 470     int b = a;
 471     a += V;
 472     for(i=-16; i<0; i+=4) {
 473       src[16+i] = CLIP((b    ) >> 5);
 474       src[17+i] = CLIP((b+  H) >> 5);
 475       src[18+i] = CLIP((b+2*H) >> 5);
 476       src[19+i] = CLIP((b+3*H) >> 5);
 477       b += 4*H;
 478     }
 479     src += stride;
 480   }
 481 }
 482
 483 static void FUNCC(pred16x16_plane)(uint8_t *src, ptrdiff_t stride)
 484 {
 485     FUNCC(pred16x16_plane_compat)(src, stride, 0, 0);
 486 }
 487
 488 static void FUNCC(pred8x8_vertical)(uint8_t *_src, ptrdiff_t _stride)
 489 {
 490     int i;
 491     pixel *src = (pixel*)_src;
 492     int stride = _stride/sizeof(pixel);
 493     const pixel4 a= AV_RN4PA(((pixel4*)(src-stride))+0);
 494     const pixel4 b= AV_RN4PA(((pixel4*)(src-stride))+1);
 495
 496     for(i=0; i<8; i++){
 497         AV_WN4PA(((pixel4*)(src+i*stride))+0, a);
 498         AV_WN4PA(((pixel4*)(src+i*stride))+1, b);
 499     }
 500 }
 501
 502 static void FUNCC(pred8x16_vertical)(uint8_t *_src, ptrdiff_t _stride)
 503 {
 504     int i;
 505     pixel *src = (pixel*)_src;
 506     int stride = _stride>>(sizeof(pixel)-1);
 507     const pixel4 a= AV_RN4PA(((pixel4*)(src-stride))+0);
 508     const pixel4 b= AV_RN4PA(((pixel4*)(src-stride))+1);
 509
 510     for(i=0; i<16; i++){
 511         AV_WN4PA(((pixel4*)(src+i*stride))+0, a);
 512         AV_WN4PA(((pixel4*)(src+i*stride))+1, b);
 513     }
 514 }
 515
 516 static void FUNCC(pred8x8_horizontal)(uint8_t *_src, ptrdiff_t stride)
 517 {
 518     int i;
 519     pixel *src = (pixel*)_src;
 520     stride /= sizeof(pixel);
 521
 522     for(i=0; i<8; i++){
 523         const pixel4 a = PIXEL_SPLAT_X4(src[-1+i*stride]);
 524         AV_WN4PA(((pixel4*)(src+i*stride))+0, a);
 525         AV_WN4PA(((pixel4*)(src+i*stride))+1, a);
 526     }
 527 }
 528
 529 static void FUNCC(pred8x16_horizontal)(uint8_t *_src, ptrdiff_t stride)
 530 {
 531     int i;
 532     pixel *src = (pixel*)_src;
 533     stride >>= sizeof(pixel)-1;
 534     for(i=0; i<16; i++){
 535         const pixel4 a = PIXEL_SPLAT_X4(src[-1+i*stride]);
 536         AV_WN4PA(((pixel4*)(src+i*stride))+0, a);
 537         AV_WN4PA(((pixel4*)(src+i*stride))+1, a);
 538     }
 539 }
 540
 541 #define PRED8x8_X(n, v)\
 542 static void FUNCC(pred8x8_##n##_dc)(uint8_t *_src, ptrdiff_t stride)\
 543 {\
 544     int i;\
 545     const pixel4 a = PIXEL_SPLAT_X4(v);\
 546     pixel *src = (pixel*)_src;\
 547     stride /= sizeof(pixel);\
 548     for(i=0; i<8; i++){\
 549         AV_WN4PA(((pixel4*)(src+i*stride))+0, a);\
 550         AV_WN4PA(((pixel4*)(src+i*stride))+1, a);\
 551     }\
 552 }
 553
 554 PRED8x8_X(127, (1<<(BIT_DEPTH-1))-1)
 555 PRED8x8_X(128, (1<<(BIT_DEPTH-1))+0)
 556 PRED8x8_X(129, (1<<(BIT_DEPTH-1))+1)
 557
 558 static void FUNCC(pred8x16_128_dc)(uint8_t *_src, ptrdiff_t stride)
 559 {
 560     FUNCC(pred8x8_128_dc)(_src, stride);
 561     FUNCC(pred8x8_128_dc)(_src+8*stride, stride);
 562 }
 563
 564 static void FUNCC(pred8x8_left_dc)(uint8_t *_src, ptrdiff_t stride)
 565 {
 566     int i;
 567     int dc0, dc2;
 568     pixel4 dc0splat, dc2splat;
 569     pixel *src = (pixel*)_src;
 570     stride /= sizeof(pixel);
 571
 572     dc0=dc2=0;
 573     for(i=0;i<4; i++){
 574         dc0+= src[-1+i*stride];
 575         dc2+= src[-1+(i+4)*stride];
 576     }
 577     dc0splat = PIXEL_SPLAT_X4((dc0 + 2)>>2);
 578     dc2splat = PIXEL_SPLAT_X4((dc2 + 2)>>2);
 579
 580     for(i=0; i<4; i++){
 581         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc0splat);
 582         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc0splat);
 583     }
 584     for(i=4; i<8; i++){
 585         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc2splat);
 586         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc2splat);
 587     }
 588 }
 589
 590 static void FUNCC(pred8x16_left_dc)(uint8_t *_src, ptrdiff_t stride)
 591 {
 592     FUNCC(pred8x8_left_dc)(_src, stride);
 593     FUNCC(pred8x8_left_dc)(_src+8*stride, stride);
 594 }
 595
 596 static void FUNCC(pred8x8_top_dc)(uint8_t *_src, ptrdiff_t stride)
 597 {
 598     int i;
 599     int dc0, dc1;
 600     pixel4 dc0splat, dc1splat;
 601     pixel *src = (pixel*)_src;
 602     stride /= sizeof(pixel);
 603
 604     dc0=dc1=0;
 605     for(i=0;i<4; i++){
 606         dc0+= src[i-stride];
 607         dc1+= src[4+i-stride];
 608     }
 609     dc0splat = PIXEL_SPLAT_X4((dc0 + 2)>>2);
 610     dc1splat = PIXEL_SPLAT_X4((dc1 + 2)>>2);
 611
 612     for(i=0; i<4; i++){
 613         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc0splat);
 614         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc1splat);
 615     }
 616     for(i=4; i<8; i++){
 617         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc0splat);
 618         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc1splat);
 619     }
 620 }
 621
 622 static void FUNCC(pred8x16_top_dc)(uint8_t *_src, ptrdiff_t stride)
 623 {
 624     int i;
 625     int dc0, dc1;
 626     pixel4 dc0splat, dc1splat;
 627     pixel *src = (pixel*)_src;
 628     stride >>= sizeof(pixel)-1;
 629
 630     dc0=dc1=0;
 631     for(i=0;i<4; i++){
 632         dc0+= src[i-stride];
 633         dc1+= src[4+i-stride];
 634     }
 635     dc0splat = PIXEL_SPLAT_X4((dc0 + 2)>>2);
 636     dc1splat = PIXEL_SPLAT_X4((dc1 + 2)>>2);
 637
 638     for(i=0; i<16; i++){
 639         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc0splat);
 640         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc1splat);
 641     }
 642 }
 643
 644 static void FUNCC(pred8x8_dc)(uint8_t *_src, ptrdiff_t stride)
 645 {
 646     int i;
 647     int dc0, dc1, dc2;
 648     pixel4 dc0splat, dc1splat, dc2splat, dc3splat;
 649     pixel *src = (pixel*)_src;
 650     stride /= sizeof(pixel);
 651
 652     dc0=dc1=dc2=0;
 653     for(i=0;i<4; i++){
 654         dc0+= src[-1+i*stride] + src[i-stride];
 655         dc1+= src[4+i-stride];
 656         dc2+= src[-1+(i+4)*stride];
 657     }
 658     dc0splat = PIXEL_SPLAT_X4((dc0 + 4)>>3);
 659     dc1splat = PIXEL_SPLAT_X4((dc1 + 2)>>2);
 660     dc2splat = PIXEL_SPLAT_X4((dc2 + 2)>>2);
 661     dc3splat = PIXEL_SPLAT_X4((dc1 + dc2 + 4)>>3);
 662
 663     for(i=0; i<4; i++){
 664         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc0splat);
 665         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc1splat);
 666     }
 667     for(i=4; i<8; i++){
 668         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc2splat);
 669         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc3splat);
 670     }
 671 }
 672
 673 static void FUNCC(pred8x16_dc)(uint8_t *_src, ptrdiff_t stride)
 674 {
 675     int i;
 676     int dc0, dc1, dc2, dc3, dc4;
 677     pixel4 dc0splat, dc1splat, dc2splat, dc3splat, dc4splat, dc5splat, dc6splat, dc7splat;
 678     pixel *src = (pixel*)_src;
 679     stride >>= sizeof(pixel)-1;
 680
 681     dc0=dc1=dc2=dc3=dc4=0;
 682     for(i=0;i<4; i++){
 683         dc0+= src[-1+i*stride] + src[i-stride];
 684         dc1+= src[4+i-stride];
 685         dc2+= src[-1+(i+4)*stride];
 686         dc3+= src[-1+(i+8)*stride];
 687         dc4+= src[-1+(i+12)*stride];
 688     }
 689     dc0splat = PIXEL_SPLAT_X4((dc0 + 4)>>3);
 690     dc1splat = PIXEL_SPLAT_X4((dc1 + 2)>>2);
 691     dc2splat = PIXEL_SPLAT_X4((dc2 + 2)>>2);
 692     dc3splat = PIXEL_SPLAT_X4((dc1 + dc2 + 4)>>3);
 693     dc4splat = PIXEL_SPLAT_X4((dc3 + 2)>>2);
 694     dc5splat = PIXEL_SPLAT_X4((dc1 + dc3 + 4)>>3);
 695     dc6splat = PIXEL_SPLAT_X4((dc4 + 2)>>2);
 696     dc7splat = PIXEL_SPLAT_X4((dc1 + dc4 + 4)>>3);
 697
 698     for(i=0; i<4; i++){
 699         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc0splat);
 700         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc1splat);
 701     }
 702     for(i=4; i<8; i++){
 703         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc2splat);
 704         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc3splat);
 705     }
 706     for(i=8; i<12; i++){
 707         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc4splat);
 708         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc5splat);
 709     }
 710     for(i=12; i<16; i++){
 711         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc6splat);
 712         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc7splat);
 713     }
 714 }
 715
 716 static void FUNC(pred8x8_mad_cow_dc_l0t)(uint8_t *src, ptrdiff_t stride)
 717 {
 718     FUNCC(pred8x8_top_dc)(src, stride);
 719     FUNCC(pred4x4_dc)(src, NULL, stride);
 720 }
 721
 722 static void FUNC(pred8x16_mad_cow_dc_l0t)(uint8_t *src, ptrdiff_t stride)
 723 {
 724     FUNCC(pred8x16_top_dc)(src, stride);
 725     FUNCC(pred4x4_dc)(src, NULL, stride);
 726 }
 727
 728 static void FUNC(pred8x8_mad_cow_dc_0lt)(uint8_t *src, ptrdiff_t stride)
 729 {
 730     FUNCC(pred8x8_dc)(src, stride);
 731     FUNCC(pred4x4_top_dc)(src, NULL, stride);
 732 }
 733
 734 static void FUNC(pred8x16_mad_cow_dc_0lt)(uint8_t *src, ptrdiff_t stride)
 735 {
 736     FUNCC(pred8x16_dc)(src, stride);
 737     FUNCC(pred4x4_top_dc)(src, NULL, stride);
 738 }
 739
 740 static void FUNC(pred8x8_mad_cow_dc_l00)(uint8_t *src, ptrdiff_t stride)
 741 {
 742     FUNCC(pred8x8_left_dc)(src, stride);
 743     FUNCC(pred4x4_128_dc)(src + 4*stride                  , NULL, stride);
 744     FUNCC(pred4x4_128_dc)(src + 4*stride + 4*sizeof(pixel), NULL, stride);
 745 }
 746
 747 static void FUNC(pred8x16_mad_cow_dc_l00)(uint8_t *src, ptrdiff_t stride)
 748 {
 749     FUNCC(pred8x16_left_dc)(src, stride);
 750     FUNCC(pred4x4_128_dc)(src + 4*stride                  , NULL, stride);
 751     FUNCC(pred4x4_128_dc)(src + 4*stride + 4*sizeof(pixel), NULL, stride);
 752 }
 753
 754 static void FUNC(pred8x8_mad_cow_dc_0l0)(uint8_t *src, ptrdiff_t stride)
 755 {
 756     FUNCC(pred8x8_left_dc)(src, stride);
 757     FUNCC(pred4x4_128_dc)(src                  , NULL, stride);
 758     FUNCC(pred4x4_128_dc)(src + 4*sizeof(pixel), NULL, stride);
 759 }
 760
 761 static void FUNC(pred8x16_mad_cow_dc_0l0)(uint8_t *src, ptrdiff_t stride)
 762 {
 763     FUNCC(pred8x16_left_dc)(src, stride);
 764     FUNCC(pred4x4_128_dc)(src                  , NULL, stride);
 765     FUNCC(pred4x4_128_dc)(src + 4*sizeof(pixel), NULL, stride);
 766 }
 767
 768 static void FUNCC(pred8x8_plane)(uint8_t *_src, ptrdiff_t _stride)
 769 {
 770   int j, k;
 771   int a;
 772   INIT_CLIP
 773   pixel *src = (pixel*)_src;
 774   int stride = _stride/sizeof(pixel);
 775   const pixel * const src0 = src +3-stride;
 776   const pixel *       src1 = src +4*stride-1;
 777   const pixel *       src2 = src1-2*stride;    // == src+2*stride-1;
 778   int H = src0[1] - src0[-1];
 779   int V = src1[0] - src2[ 0];
 780   for(k=2; k<=4; ++k) {
 781     src1 += stride; src2 -= stride;
 782     H += k*(src0[k] - src0[-k]);
 783     V += k*(src1[0] - src2[ 0]);
 784   }
 785   H = ( 17*H+16 ) >> 5;
 786   V = ( 17*V+16 ) >> 5;
 787
 788   a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
 789   for(j=8; j>0; --j) {
 790     int b = a;
 791     a += V;
 792     src[0] = CLIP((b    ) >> 5);
 793     src[1] = CLIP((b+  H) >> 5);
 794     src[2] = CLIP((b+2*H) >> 5);
 795     src[3] = CLIP((b+3*H) >> 5);
 796     src[4] = CLIP((b+4*H) >> 5);
 797     src[5] = CLIP((b+5*H) >> 5);
 798     src[6] = CLIP((b+6*H) >> 5);
 799     src[7] = CLIP((b+7*H) >> 5);
 800     src += stride;
 801   }
 802 }
 803
 804 static void FUNCC(pred8x16_plane)(uint8_t *_src, ptrdiff_t _stride)
 805 {
 806   int j, k;
 807   int a;
 808   INIT_CLIP
 809   pixel *src = (pixel*)_src;
 810   int stride = _stride>>(sizeof(pixel)-1);
 811   const pixel * const src0 = src +3-stride;
 812   const pixel *       src1 = src +8*stride-1;
 813   const pixel *       src2 = src1-2*stride;    // == src+6*stride-1;
 814   int H = src0[1] - src0[-1];
 815   int V = src1[0] - src2[ 0];
 816
 817   for (k = 2; k <= 4; ++k) {
 818       src1 += stride; src2 -= stride;
 819       H += k*(src0[k] - src0[-k]);
 820       V += k*(src1[0] - src2[ 0]);
 821   }
 822   for (; k <= 8; ++k) {
 823       src1 += stride; src2 -= stride;
 824       V += k*(src1[0] - src2[0]);
 825   }
 826
 827   H = (17*H+16) >> 5;
 828   V = (5*V+32) >> 6;
 829
 830   a = 16*(src1[0] + src2[8] + 1) - 7*V - 3*H;
 831   for(j=16; j>0; --j) {
 832     int b = a;
 833     a += V;
 834     src[0] = CLIP((b    ) >> 5);
 835     src[1] = CLIP((b+  H) >> 5);
 836     src[2] = CLIP((b+2*H) >> 5);
 837     src[3] = CLIP((b+3*H) >> 5);
 838     src[4] = CLIP((b+4*H) >> 5);
 839     src[5] = CLIP((b+5*H) >> 5);
 840     src[6] = CLIP((b+6*H) >> 5);
 841     src[7] = CLIP((b+7*H) >> 5);
 842     src += stride;
 843   }
 844 }
 845
 846 #define SRC(x,y) src[(x)+(y)*stride]
 847 #define PL(y) \
 848     const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
 849 #define PREDICT_8x8_LOAD_LEFT \
 850     const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
 851                      + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
 852     PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
 853     const int l7 av_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
 854
 855 #define PT(x) \
 856     const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
 857 #define PREDICT_8x8_LOAD_TOP \
 858     const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
 859                      + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
 860     PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
 861     const int t7 av_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
 862                      + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
 863
 864 #define PTR(x) \
 865     t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
 866 #define PREDICT_8x8_LOAD_TOPRIGHT \
 867     int t8, t9, t10, t11, t12, t13, t14, t15; \
 868     if(has_topright) { \
 869         PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
 870         t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
 871     } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
 872
 873 #define PREDICT_8x8_LOAD_TOPLEFT \
 874     const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
 875
 876 #define PREDICT_8x8_DC(v) \
 877     int y; \
 878     for( y = 0; y < 8; y++ ) { \
 879         AV_WN4PA(((pixel4*)src)+0, v); \
 880         AV_WN4PA(((pixel4*)src)+1, v); \
 881         src += stride; \
 882     }
 883
 884 static void FUNCC(pred8x8l_128_dc)(uint8_t *_src, int has_topleft,
 885                                    int has_topright, ptrdiff_t _stride)
 886 {
 887     pixel *src = (pixel*)_src;
 888     int stride = _stride/sizeof(pixel);
 889
 890     PREDICT_8x8_DC(PIXEL_SPLAT_X4(1<<(BIT_DEPTH-1)));
 891 }
 892 static void FUNCC(pred8x8l_left_dc)(uint8_t *_src, int has_topleft,
 893                                     int has_topright, ptrdiff_t _stride)
 894 {
 895     pixel *src = (pixel*)_src;
 896     int stride = _stride/sizeof(pixel);
 897
 898     PREDICT_8x8_LOAD_LEFT;
 899     const pixel4 dc = PIXEL_SPLAT_X4((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3);
 900     PREDICT_8x8_DC(dc);
 901 }
 902 static void FUNCC(pred8x8l_top_dc)(uint8_t *_src, int has_topleft,
 903                                    int has_topright, ptrdiff_t _stride)
 904 {
 905     pixel *src = (pixel*)_src;
 906     int stride = _stride/sizeof(pixel);
 907
 908     PREDICT_8x8_LOAD_TOP;
 909     const pixel4 dc = PIXEL_SPLAT_X4((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3);
 910     PREDICT_8x8_DC(dc);
 911 }
 912 static void FUNCC(pred8x8l_dc)(uint8_t *_src, int has_topleft,
 913                                int has_topright, ptrdiff_t _stride)
 914 {
 915     pixel *src = (pixel*)_src;
 916     int stride = _stride/sizeof(pixel);
 917
 918     PREDICT_8x8_LOAD_LEFT;
 919     PREDICT_8x8_LOAD_TOP;
 920     const pixel4 dc = PIXEL_SPLAT_X4((l0+l1+l2+l3+l4+l5+l6+l7
 921                                      +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4);
 922     PREDICT_8x8_DC(dc);
 923 }
 924 static void FUNCC(pred8x8l_horizontal)(uint8_t *_src, int has_topleft,
 925                                        int has_topright, ptrdiff_t _stride)
 926 {
 927     pixel *src = (pixel*)_src;
 928     int stride = _stride/sizeof(pixel);
 929     pixel4 a;
 930
 931     PREDICT_8x8_LOAD_LEFT;
 932 #define ROW(y) a = PIXEL_SPLAT_X4(l##y); \
 933                AV_WN4PA(src+y*stride, a); \
 934                AV_WN4PA(src+y*stride+4, a);
 935     ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
 936 #undef ROW
 937 }
 938 static void FUNCC(pred8x8l_vertical)(uint8_t *_src, int has_topleft,
 939                                      int has_topright, ptrdiff_t _stride)
 940 {
 941     int y;
 942     pixel *src = (pixel*)_src;
 943     int stride = _stride/sizeof(pixel);
 944     pixel4 a, b;
 945
 946     PREDICT_8x8_LOAD_TOP;
 947     src[0] = t0;
 948     src[1] = t1;
 949     src[2] = t2;
 950     src[3] = t3;
 951     src[4] = t4;
 952     src[5] = t5;
 953     src[6] = t6;
 954     src[7] = t7;
 955     a = AV_RN4PA(((pixel4*)src)+0);
 956     b = AV_RN4PA(((pixel4*)src)+1);
 957     for( y = 1; y < 8; y++ ) {
 958         AV_WN4PA(((pixel4*)(src+y*stride))+0, a);
 959         AV_WN4PA(((pixel4*)(src+y*stride))+1, b);
 960     }
 961 }
 962 static void FUNCC(pred8x8l_down_left)(uint8_t *_src, int has_topleft,
 963                                       int has_topright, ptrdiff_t _stride)
 964 {
 965     pixel *src = (pixel*)_src;
 966     int stride = _stride/sizeof(pixel);
 967     PREDICT_8x8_LOAD_TOP;
 968     PREDICT_8x8_LOAD_TOPRIGHT;
 969     SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
 970     SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
 971     SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
 972     SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
 973     SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
 974     SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
 975     SRC(0,6)=SRC(1,5)=SRC(2,4)=SRC(3,3)=SRC(4,2)=SRC(5,1)=SRC(6,0)= (t6 + 2*t7 + t8 + 2) >> 2;
 976     SRC(0,7)=SRC(1,6)=SRC(2,5)=SRC(3,4)=SRC(4,3)=SRC(5,2)=SRC(6,1)=SRC(7,0)= (t7 + 2*t8 + t9 + 2) >> 2;
 977     SRC(1,7)=SRC(2,6)=SRC(3,5)=SRC(4,4)=SRC(5,3)=SRC(6,2)=SRC(7,1)= (t8 + 2*t9 + t10 + 2) >> 2;
 978     SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
 979     SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
 980     SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
 981     SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
 982     SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
 983     SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
 984 }
 985 static void FUNCC(pred8x8l_down_right)(uint8_t *_src, int has_topleft,
 986                                        int has_topright, ptrdiff_t _stride)
 987 {
 988     pixel *src = (pixel*)_src;
 989     int stride = _stride/sizeof(pixel);
 990     PREDICT_8x8_LOAD_TOP;
 991     PREDICT_8x8_LOAD_LEFT;
 992     PREDICT_8x8_LOAD_TOPLEFT;
 993     SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
 994     SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
 995     SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
 996     SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
 997     SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
 998     SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
 999     SRC(0,1)=SRC(1,2)=SRC(2,3)=SRC(3,4)=SRC(4,5)=SRC(5,6)=SRC(6,7)= (l1 + 2*l0 + lt + 2) >> 2;
1000     SRC(0,0)=SRC(1,1)=SRC(2,2)=SRC(3,3)=SRC(4,4)=SRC(5,5)=SRC(6,6)=SRC(7,7)= (l0 + 2*lt + t0 + 2) >> 2;
1001     SRC(1,0)=SRC(2,1)=SRC(3,2)=SRC(4,3)=SRC(5,4)=SRC(6,5)=SRC(7,6)= (lt + 2*t0 + t1 + 2) >> 2;
1002     SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
1003     SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
1004     SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
1005     SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
1006     SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
1007     SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
1008 }
1009 static void FUNCC(pred8x8l_vertical_right)(uint8_t *_src, int has_topleft,
1010                                            int has_topright, ptrdiff_t _stride)
1011 {
1012     pixel *src = (pixel*)_src;
1013     int stride = _stride/sizeof(pixel);
1014     PREDICT_8x8_LOAD_TOP;
1015     PREDICT_8x8_LOAD_LEFT;
1016     PREDICT_8x8_LOAD_TOPLEFT;
1017     SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
1018     SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
1019     SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
1020     SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
1021     SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
1022     SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
1023     SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
1024     SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
1025     SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
1026     SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
1027     SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
1028     SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
1029     SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
1030     SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
1031     SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
1032     SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
1033     SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
1034     SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
1035     SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
1036     SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
1037     SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
1038     SRC(7,0)= (t6 + t7 + 1) >> 1;
1039 }
1040 static void FUNCC(pred8x8l_horizontal_down)(uint8_t *_src, int has_topleft,
1041                                             int has_topright, ptrdiff_t _stride)
1042 {
1043     pixel *src = (pixel*)_src;
1044     int stride = _stride/sizeof(pixel);
1045     PREDICT_8x8_LOAD_TOP;
1046     PREDICT_8x8_LOAD_LEFT;
1047     PREDICT_8x8_LOAD_TOPLEFT;
1048     SRC(0,7)= (l6 + l7 + 1) >> 1;
1049     SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
1050     SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
1051     SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
1052     SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
1053     SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
1054     SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
1055     SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
1056     SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
1057     SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
1058     SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
1059     SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
1060     SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
1061     SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
1062     SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
1063     SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
1064     SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
1065     SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
1066     SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
1067     SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
1068     SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
1069     SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
1070 }
1071 static void FUNCC(pred8x8l_vertical_left)(uint8_t *_src, int has_topleft,
1072                                           int has_topright, ptrdiff_t _stride)
1073 {
1074     pixel *src = (pixel*)_src;
1075     int stride = _stride/sizeof(pixel);
1076     PREDICT_8x8_LOAD_TOP;
1077     PREDICT_8x8_LOAD_TOPRIGHT;
1078     SRC(0,0)= (t0 + t1 + 1) >> 1;
1079     SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
1080     SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
1081     SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
1082     SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
1083     SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
1084     SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
1085     SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
1086     SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
1087     SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
1088     SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
1089     SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
1090     SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
1091     SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
1092     SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
1093     SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
1094     SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
1095     SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
1096     SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
1097     SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
1098     SRC(7,6)= (t10 + t11 + 1) >> 1;
1099     SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
1100 }
1101 static void FUNCC(pred8x8l_horizontal_up)(uint8_t *_src, int has_topleft,
1102                                           int has_topright, ptrdiff_t _stride)
1103 {
1104     pixel *src = (pixel*)_src;
1105     int stride = _stride/sizeof(pixel);
1106     PREDICT_8x8_LOAD_LEFT;
1107     SRC(0,0)= (l0 + l1 + 1) >> 1;
1108     SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
1109     SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
1110     SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
1111     SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
1112     SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
1113     SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
1114     SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
1115     SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
1116     SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
1117     SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
1118     SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
1119     SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
1120     SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
1121     SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
1122     SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
1123     SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
1124     SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
1125 }
1126
1127 static void FUNCC(pred8x8l_vertical_filter_add)(uint8_t *_src, int16_t *_block, int has_topleft,
1128                                                 int has_topright, ptrdiff_t _stride)
1129 {
1130     int i;
1131     pixel *src = (pixel*)_src;
1132     const dctcoef *block = (const dctcoef*)_block;
1133     pixel pix[8];
1134     int stride = _stride/sizeof(pixel);
1135     PREDICT_8x8_LOAD_TOP;
1136
1137     pix[0] = t0;
1138     pix[1] = t1;
1139     pix[2] = t2;
1140     pix[3] = t3;
1141     pix[4] = t4;
1142     pix[5] = t5;
1143     pix[6] = t6;
1144     pix[7] = t7;
1145
1146     for (i = 0; i < 8; i++) {
1147         pixel v = pix[i];
1148         src[0 * stride] = v += block[0];
1149         src[1 * stride] = v += block[8];
1150         src[2 * stride] = v += block[16];
1151         src[3 * stride] = v += block[24];
1152         src[4 * stride] = v += block[32];
1153         src[5 * stride] = v += block[40];
1154         src[6 * stride] = v += block[48];
1155         src[7 * stride] = v +  block[56];
1156         src++;
1157         block++;
1158     }
1159
1160     memset(_block, 0, sizeof(dctcoef) * 64);
1161 }
1162
1163 static void FUNCC(pred8x8l_horizontal_filter_add)(uint8_t *_src, int16_t *_block, int has_topleft,
1164                                                   int has_topright, ptrdiff_t _stride)
1165 {
1166     int i;
1167     pixel *src = (pixel*)_src;
1168     const dctcoef *block = (const dctcoef*)_block;
1169     pixel pix[8];
1170     int stride = _stride/sizeof(pixel);
1171     PREDICT_8x8_LOAD_LEFT;
1172
1173     pix[0] = l0;
1174     pix[1] = l1;
1175     pix[2] = l2;
1176     pix[3] = l3;
1177     pix[4] = l4;
1178     pix[5] = l5;
1179     pix[6] = l6;
1180     pix[7] = l7;
1181
1182     for (i = 0; i < 8; i++) {
1183         pixel v = pix[i];
1184         src[0] = v += block[0];
1185         src[1] = v += block[1];
1186         src[2] = v += block[2];
1187         src[3] = v += block[3];
1188         src[4] = v += block[4];
1189         src[5] = v += block[5];
1190         src[6] = v += block[6];
1191         src[7] = v +  block[7];
1192         src   += stride;
1193         block += 8;
1194     }
1195
1196     memset(_block, 0, sizeof(dctcoef) * 64);
1197 }
1198
1199 #undef PREDICT_8x8_LOAD_LEFT
1200 #undef PREDICT_8x8_LOAD_TOP
1201 #undef PREDICT_8x8_LOAD_TOPLEFT
1202 #undef PREDICT_8x8_LOAD_TOPRIGHT
1203 #undef PREDICT_8x8_DC
1204 #undef PTR
1205 #undef PT
1206 #undef PL
1207 #undef SRC
1208
1209 static void FUNCC(pred4x4_vertical_add)(uint8_t *_pix, int16_t *_block,
1210                                         ptrdiff_t stride)
1211 {
1212     int i;
1213     pixel *pix = (pixel*)_pix;
1214     const dctcoef *block = (const dctcoef*)_block;
1215     stride /= sizeof(pixel);
1216     pix -= stride;
1217     for(i=0; i<4; i++){
1218         pixel v = pix[0];
1219         pix[1*stride]= v += block[0];
1220         pix[2*stride]= v += block[4];
1221         pix[3*stride]= v += block[8];
1222         pix[4*stride]= v +  block[12];
1223         pix++;
1224         block++;
1225     }
1226
1227     memset(_block, 0, sizeof(dctcoef) * 16);
1228 }
1229
1230 static void FUNCC(pred4x4_horizontal_add)(uint8_t *_pix, int16_t *_block,
1231                                           ptrdiff_t stride)
1232 {
1233     int i;
1234     pixel *pix = (pixel*)_pix;
1235     const dctcoef *block = (const dctcoef*)_block;
1236     stride /= sizeof(pixel);
1237     for(i=0; i<4; i++){
1238         pixel v = pix[-1];
1239         pix[0]= v += block[0];
1240         pix[1]= v += block[1];
1241         pix[2]= v += block[2];
1242         pix[3]= v +  block[3];
1243         pix+= stride;
1244         block+= 4;
1245     }
1246
1247     memset(_block, 0, sizeof(dctcoef) * 16);
1248 }
1249
1250 static void FUNCC(pred8x8l_vertical_add)(uint8_t *_pix, int16_t *_block,
1251                                          ptrdiff_t stride)
1252 {
1253     int i;
1254     pixel *pix = (pixel*)_pix;
1255     const dctcoef *block = (const dctcoef*)_block;
1256     stride /= sizeof(pixel);
1257     pix -= stride;
1258     for(i=0; i<8; i++){
1259         pixel v = pix[0];
1260         pix[1*stride]= v += block[0];
1261         pix[2*stride]= v += block[8];
1262         pix[3*stride]= v += block[16];
1263         pix[4*stride]= v += block[24];
1264         pix[5*stride]= v += block[32];
1265         pix[6*stride]= v += block[40];
1266         pix[7*stride]= v += block[48];
1267         pix[8*stride]= v +  block[56];
1268         pix++;
1269         block++;
1270     }
1271
1272     memset(_block, 0, sizeof(dctcoef) * 64);
1273 }
1274
1275 static void FUNCC(pred8x8l_horizontal_add)(uint8_t *_pix, int16_t *_block,
1276                                            ptrdiff_t stride)
1277 {
1278     int i;
1279     pixel *pix = (pixel*)_pix;
1280     const dctcoef *block = (const dctcoef*)_block;
1281     stride /= sizeof(pixel);
1282     for(i=0; i<8; i++){
1283         pixel v = pix[-1];
1284         pix[0]= v += block[0];
1285         pix[1]= v += block[1];
1286         pix[2]= v += block[2];
1287         pix[3]= v += block[3];
1288         pix[4]= v += block[4];
1289         pix[5]= v += block[5];
1290         pix[6]= v += block[6];
1291         pix[7]= v +  block[7];
1292         pix+= stride;
1293         block+= 8;
1294     }
1295
1296     memset(_block, 0, sizeof(dctcoef) * 64);
1297 }
1298
1299 static void FUNCC(pred16x16_vertical_add)(uint8_t *pix, const int *block_offset,
1300                                           int16_t *block,
1301                                           ptrdiff_t stride)
1302 {
1303     int i;
1304     for(i=0; i<16; i++)
1305         FUNCC(pred4x4_vertical_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
1306 }
1307
1308 static void FUNCC(pred16x16_horizontal_add)(uint8_t *pix,
1309                                             const int *block_offset,
1310                                             int16_t *block,
1311                                             ptrdiff_t stride)
1312 {
1313     int i;
1314     for(i=0; i<16; i++)
1315         FUNCC(pred4x4_horizontal_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
1316 }
1317
1318 static void FUNCC(pred8x8_vertical_add)(uint8_t *pix, const int *block_offset,
1319                                         int16_t *block, ptrdiff_t stride)
1320 {
1321     int i;
1322     for(i=0; i<4; i++)
1323         FUNCC(pred4x4_vertical_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
1324 }
1325
1326 static void FUNCC(pred8x16_vertical_add)(uint8_t *pix, const int *block_offset,
1327                                          int16_t *block, ptrdiff_t stride)
1328 {
1329     int i;
1330     for(i=0; i<4; i++)
1331         FUNCC(pred4x4_vertical_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
1332     for(i=4; i<8; i++)
1333         FUNCC(pred4x4_vertical_add)(pix + block_offset[i+4], block + i*16*sizeof(pixel), stride);
1334 }
1335
1336 static void FUNCC(pred8x8_horizontal_add)(uint8_t *pix, const int *block_offset,
1337                                           int16_t *block,
1338                                           ptrdiff_t stride)
1339 {
1340     int i;
1341     for(i=0; i<4; i++)
1342         FUNCC(pred4x4_horizontal_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
1343 }
1344
1345 static void FUNCC(pred8x16_horizontal_add)(uint8_t *pix,
1346                                            const int *block_offset,
1347                                            int16_t *block, ptrdiff_t stride)
1348 {
1349     int i;
1350     for(i=0; i<4; i++)
1351         FUNCC(pred4x4_horizontal_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
1352     for(i=4; i<8; i++)
1353         FUNCC(pred4x4_horizontal_add)(pix + block_offset[i+4], block + i*16*sizeof(pixel), stride);
1354 }