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 / MPEG4 part10 prediction functions.
  25  * @author Michael Niedermayer <michaelni@gmx.at>
  26  */
  27
  28 #include "mathops.h"
  29
  30 #include "bit_depth_template.c"
  31
  32 static void FUNCC(pred4x4_vertical)(uint8_t *_src, const uint8_t *topright,
  33                                     ptrdiff_t _stride)
  34 {
  35     pixel *src = (pixel*)_src;
  36     int stride = _stride/sizeof(pixel);
  37     const pixel4 a= AV_RN4PA(src-stride);
  38
  39     AV_WN4PA(src+0*stride, a);
  40     AV_WN4PA(src+1*stride, a);
  41     AV_WN4PA(src+2*stride, a);
  42     AV_WN4PA(src+3*stride, a);
  43 }
  44
  45 static void FUNCC(pred4x4_horizontal)(uint8_t *_src, const uint8_t *topright,
  46                                       ptrdiff_t _stride)
  47 {
  48     pixel *src = (pixel*)_src;
  49     int stride = _stride/sizeof(pixel);
  50     AV_WN4PA(src+0*stride, PIXEL_SPLAT_X4(src[-1+0*stride]));
  51     AV_WN4PA(src+1*stride, PIXEL_SPLAT_X4(src[-1+1*stride]));
  52     AV_WN4PA(src+2*stride, PIXEL_SPLAT_X4(src[-1+2*stride]));
  53     AV_WN4PA(src+3*stride, PIXEL_SPLAT_X4(src[-1+3*stride]));
  54 }
  55
  56 static void FUNCC(pred4x4_dc)(uint8_t *_src, const uint8_t *topright,
  57                               ptrdiff_t _stride)
  58 {
  59     pixel *src = (pixel*)_src;
  60     int stride = _stride/sizeof(pixel);
  61     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
  62                    + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
  63     const pixel4 a = PIXEL_SPLAT_X4(dc);
  64
  65     AV_WN4PA(src+0*stride, a);
  66     AV_WN4PA(src+1*stride, a);
  67     AV_WN4PA(src+2*stride, a);
  68     AV_WN4PA(src+3*stride, a);
  69 }
  70
  71 static void FUNCC(pred4x4_left_dc)(uint8_t *_src, const uint8_t *topright,
  72                                    ptrdiff_t _stride)
  73 {
  74     pixel *src = (pixel*)_src;
  75     int stride = _stride/sizeof(pixel);
  76     const int dc= (  src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
  77     const pixel4 a = PIXEL_SPLAT_X4(dc);
  78
  79     AV_WN4PA(src+0*stride, a);
  80     AV_WN4PA(src+1*stride, a);
  81     AV_WN4PA(src+2*stride, a);
  82     AV_WN4PA(src+3*stride, a);
  83 }
  84
  85 static void FUNCC(pred4x4_top_dc)(uint8_t *_src, const uint8_t *topright,
  86                                   ptrdiff_t _stride)
  87 {
  88     pixel *src = (pixel*)_src;
  89     int stride = _stride/sizeof(pixel);
  90     const int dc= (  src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
  91     const pixel4 a = PIXEL_SPLAT_X4(dc);
  92
  93     AV_WN4PA(src+0*stride, a);
  94     AV_WN4PA(src+1*stride, a);
  95     AV_WN4PA(src+2*stride, a);
  96     AV_WN4PA(src+3*stride, a);
  97 }
  98
  99 static void FUNCC(pred4x4_128_dc)(uint8_t *_src, const uint8_t *topright,
 100                                   ptrdiff_t _stride)
 101 {
 102     pixel *src = (pixel*)_src;
 103     int stride = _stride/sizeof(pixel);
 104     const pixel4 a = PIXEL_SPLAT_X4(1<<(BIT_DEPTH-1));
 105
 106     AV_WN4PA(src+0*stride, a);
 107     AV_WN4PA(src+1*stride, a);
 108     AV_WN4PA(src+2*stride, a);
 109     AV_WN4PA(src+3*stride, a);
 110 }
 111
 112 static void FUNCC(pred4x4_127_dc)(uint8_t *_src, const uint8_t *topright,
 113                                   ptrdiff_t _stride)
 114 {
 115     pixel *src = (pixel*)_src;
 116     int stride = _stride/sizeof(pixel);
 117     const pixel4 a = PIXEL_SPLAT_X4((1<<(BIT_DEPTH-1))-1);
 118
 119     AV_WN4PA(src+0*stride, a);
 120     AV_WN4PA(src+1*stride, a);
 121     AV_WN4PA(src+2*stride, a);
 122     AV_WN4PA(src+3*stride, a);
 123 }
 124
 125 static void FUNCC(pred4x4_129_dc)(uint8_t *_src, const uint8_t *topright,
 126                                   ptrdiff_t _stride)
 127 {
 128     pixel *src = (pixel*)_src;
 129     int stride = _stride/sizeof(pixel);
 130     const pixel4 a = PIXEL_SPLAT_X4((1<<(BIT_DEPTH-1))+1);
 131
 132     AV_WN4PA(src+0*stride, a);
 133     AV_WN4PA(src+1*stride, a);
 134     AV_WN4PA(src+2*stride, a);
 135     AV_WN4PA(src+3*stride, a);
 136 }
 137
 138
 139 #define LOAD_TOP_RIGHT_EDGE\
 140     const unsigned av_unused t4 = topright[0];\
 141     const unsigned av_unused t5 = topright[1];\
 142     const unsigned av_unused t6 = topright[2];\
 143     const unsigned av_unused t7 = topright[3];\
 144
 145 #define LOAD_DOWN_LEFT_EDGE\
 146     const unsigned av_unused l4 = src[-1+4*stride];\
 147     const unsigned av_unused l5 = src[-1+5*stride];\
 148     const unsigned av_unused l6 = src[-1+6*stride];\
 149     const unsigned av_unused l7 = src[-1+7*stride];\
 150
 151 #define LOAD_LEFT_EDGE\
 152     const unsigned av_unused l0 = src[-1+0*stride];\
 153     const unsigned av_unused l1 = src[-1+1*stride];\
 154     const unsigned av_unused l2 = src[-1+2*stride];\
 155     const unsigned av_unused l3 = src[-1+3*stride];\
 156
 157 #define LOAD_TOP_EDGE\
 158     const unsigned av_unused t0 = src[ 0-1*stride];\
 159     const unsigned av_unused t1 = src[ 1-1*stride];\
 160     const unsigned av_unused t2 = src[ 2-1*stride];\
 161     const unsigned av_unused t3 = src[ 3-1*stride];\
 162
 163 static void FUNCC(pred4x4_down_right)(uint8_t *_src, const uint8_t *topright,
 164                                       ptrdiff_t _stride)
 165 {
 166     pixel *src = (pixel*)_src;
 167     int stride = _stride/sizeof(pixel);
 168     const int lt= src[-1-1*stride];
 169     LOAD_TOP_EDGE
 170     LOAD_LEFT_EDGE
 171
 172     src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
 173     src[0+2*stride]=
 174     src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
 175     src[0+1*stride]=
 176     src[1+2*stride]=
 177     src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
 178     src[0+0*stride]=
 179     src[1+1*stride]=
 180     src[2+2*stride]=
 181     src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
 182     src[1+0*stride]=
 183     src[2+1*stride]=
 184     src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
 185     src[2+0*stride]=
 186     src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
 187     src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
 188 }
 189
 190 static void FUNCC(pred4x4_down_left)(uint8_t *_src, const uint8_t *_topright,
 191                                      ptrdiff_t _stride)
 192 {
 193     pixel *src = (pixel*)_src;
 194     const pixel *topright = (const pixel*)_topright;
 195     int stride = _stride/sizeof(pixel);
 196     LOAD_TOP_EDGE
 197     LOAD_TOP_RIGHT_EDGE
 198 //    LOAD_LEFT_EDGE
 199
 200     src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
 201     src[1+0*stride]=
 202     src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
 203     src[2+0*stride]=
 204     src[1+1*stride]=
 205     src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
 206     src[3+0*stride]=
 207     src[2+1*stride]=
 208     src[1+2*stride]=
 209     src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
 210     src[3+1*stride]=
 211     src[2+2*stride]=
 212     src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
 213     src[3+2*stride]=
 214     src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
 215     src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
 216 }
 217
 218 static void FUNCC(pred4x4_vertical_right)(uint8_t *_src,
 219                                           const uint8_t *topright,
 220                                           ptrdiff_t _stride)
 221 {
 222     pixel *src = (pixel*)_src;
 223     int stride = _stride/sizeof(pixel);
 224     const int lt= src[-1-1*stride];
 225     LOAD_TOP_EDGE
 226     LOAD_LEFT_EDGE
 227
 228     src[0+0*stride]=
 229     src[1+2*stride]=(lt + t0 + 1)>>1;
 230     src[1+0*stride]=
 231     src[2+2*stride]=(t0 + t1 + 1)>>1;
 232     src[2+0*stride]=
 233     src[3+2*stride]=(t1 + t2 + 1)>>1;
 234     src[3+0*stride]=(t2 + t3 + 1)>>1;
 235     src[0+1*stride]=
 236     src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
 237     src[1+1*stride]=
 238     src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
 239     src[2+1*stride]=
 240     src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
 241     src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
 242     src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
 243     src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
 244 }
 245
 246 static void FUNCC(pred4x4_vertical_left)(uint8_t *_src,
 247                                          const uint8_t *_topright,
 248                                          ptrdiff_t _stride)
 249 {
 250     pixel *src = (pixel*)_src;
 251     const pixel *topright = (const pixel*)_topright;
 252     int stride = _stride/sizeof(pixel);
 253     LOAD_TOP_EDGE
 254     LOAD_TOP_RIGHT_EDGE
 255
 256     src[0+0*stride]=(t0 + t1 + 1)>>1;
 257     src[1+0*stride]=
 258     src[0+2*stride]=(t1 + t2 + 1)>>1;
 259     src[2+0*stride]=
 260     src[1+2*stride]=(t2 + t3 + 1)>>1;
 261     src[3+0*stride]=
 262     src[2+2*stride]=(t3 + t4+ 1)>>1;
 263     src[3+2*stride]=(t4 + t5+ 1)>>1;
 264     src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
 265     src[1+1*stride]=
 266     src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
 267     src[2+1*stride]=
 268     src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
 269     src[3+1*stride]=
 270     src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
 271     src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
 272 }
 273
 274 static void FUNCC(pred4x4_horizontal_up)(uint8_t *_src, const uint8_t *topright,
 275                                          ptrdiff_t _stride)
 276 {
 277     pixel *src = (pixel*)_src;
 278     int stride = _stride/sizeof(pixel);
 279     LOAD_LEFT_EDGE
 280
 281     src[0+0*stride]=(l0 + l1 + 1)>>1;
 282     src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
 283     src[2+0*stride]=
 284     src[0+1*stride]=(l1 + l2 + 1)>>1;
 285     src[3+0*stride]=
 286     src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
 287     src[2+1*stride]=
 288     src[0+2*stride]=(l2 + l3 + 1)>>1;
 289     src[3+1*stride]=
 290     src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
 291     src[3+2*stride]=
 292     src[1+3*stride]=
 293     src[0+3*stride]=
 294     src[2+2*stride]=
 295     src[2+3*stride]=
 296     src[3+3*stride]=l3;
 297 }
 298
 299 static void FUNCC(pred4x4_horizontal_down)(uint8_t *_src,
 300                                            const uint8_t *topright,
 301                                            ptrdiff_t _stride)
 302 {
 303     pixel *src = (pixel*)_src;
 304     int stride = _stride/sizeof(pixel);
 305     const int lt= src[-1-1*stride];
 306     LOAD_TOP_EDGE
 307     LOAD_LEFT_EDGE
 308
 309     src[0+0*stride]=
 310     src[2+1*stride]=(lt + l0 + 1)>>1;
 311     src[1+0*stride]=
 312     src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
 313     src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
 314     src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
 315     src[0+1*stride]=
 316     src[2+2*stride]=(l0 + l1 + 1)>>1;
 317     src[1+1*stride]=
 318     src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
 319     src[0+2*stride]=
 320     src[2+3*stride]=(l1 + l2+ 1)>>1;
 321     src[1+2*stride]=
 322     src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
 323     src[0+3*stride]=(l2 + l3 + 1)>>1;
 324     src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
 325 }
 326
 327 static void FUNCC(pred16x16_vertical)(uint8_t *_src, ptrdiff_t _stride)
 328 {
 329     int i;
 330     pixel *src = (pixel*)_src;
 331     int stride = _stride/sizeof(pixel);
 332     const pixel4 a = AV_RN4PA(((pixel4*)(src-stride))+0);
 333     const pixel4 b = AV_RN4PA(((pixel4*)(src-stride))+1);
 334     const pixel4 c = AV_RN4PA(((pixel4*)(src-stride))+2);
 335     const pixel4 d = AV_RN4PA(((pixel4*)(src-stride))+3);
 336
 337     for(i=0; i<16; i++){
 338         AV_WN4PA(((pixel4*)(src+i*stride))+0, a);
 339         AV_WN4PA(((pixel4*)(src+i*stride))+1, b);
 340         AV_WN4PA(((pixel4*)(src+i*stride))+2, c);
 341         AV_WN4PA(((pixel4*)(src+i*stride))+3, d);
 342     }
 343 }
 344
 345 static void FUNCC(pred16x16_horizontal)(uint8_t *_src, ptrdiff_t stride)
 346 {
 347     int i;
 348     pixel *src = (pixel*)_src;
 349     stride /= sizeof(pixel);
 350
 351     for(i=0; i<16; i++){
 352         const pixel4 a = PIXEL_SPLAT_X4(src[-1+i*stride]);
 353
 354         AV_WN4PA(((pixel4*)(src+i*stride))+0, a);
 355         AV_WN4PA(((pixel4*)(src+i*stride))+1, a);
 356         AV_WN4PA(((pixel4*)(src+i*stride))+2, a);
 357         AV_WN4PA(((pixel4*)(src+i*stride))+3, a);
 358     }
 359 }
 360
 361 #define PREDICT_16x16_DC(v)\
 362     for(i=0; i<16; i++){\
 363         AV_WN4PA(src+ 0, v);\
 364         AV_WN4PA(src+ 4, v);\
 365         AV_WN4PA(src+ 8, v);\
 366         AV_WN4PA(src+12, v);\
 367         src += stride;\
 368     }
 369
 370 static void FUNCC(pred16x16_dc)(uint8_t *_src, ptrdiff_t stride)
 371 {
 372     int i, dc=0;
 373     pixel *src = (pixel*)_src;
 374     pixel4 dcsplat;
 375     stride /= sizeof(pixel);
 376
 377     for(i=0;i<16; i++){
 378         dc+= src[-1+i*stride];
 379     }
 380
 381     for(i=0;i<16; i++){
 382         dc+= src[i-stride];
 383     }
 384
 385     dcsplat = PIXEL_SPLAT_X4((dc+16)>>5);
 386     PREDICT_16x16_DC(dcsplat);
 387 }
 388
 389 static void FUNCC(pred16x16_left_dc)(uint8_t *_src, ptrdiff_t stride)
 390 {
 391     int i, dc=0;
 392     pixel *src = (pixel*)_src;
 393     pixel4 dcsplat;
 394     stride /= sizeof(pixel);
 395
 396     for(i=0;i<16; i++){
 397         dc+= src[-1+i*stride];
 398     }
 399
 400     dcsplat = PIXEL_SPLAT_X4((dc+8)>>4);
 401     PREDICT_16x16_DC(dcsplat);
 402 }
 403
 404 static void FUNCC(pred16x16_top_dc)(uint8_t *_src, ptrdiff_t stride)
 405 {
 406     int i, dc=0;
 407     pixel *src = (pixel*)_src;
 408     pixel4 dcsplat;
 409     stride /= sizeof(pixel);
 410
 411     for(i=0;i<16; i++){
 412         dc+= src[i-stride];
 413     }
 414
 415     dcsplat = PIXEL_SPLAT_X4((dc+8)>>4);
 416     PREDICT_16x16_DC(dcsplat);
 417 }
 418
 419 #define PRED16x16_X(n, v) \
 420 static void FUNCC(pred16x16_##n##_dc)(uint8_t *_src, ptrdiff_t stride)\
 421 {\
 422     int i;\
 423     pixel *src = (pixel*)_src;\
 424     stride /= sizeof(pixel);\
 425     PREDICT_16x16_DC(PIXEL_SPLAT_X4(v));\
 426 }
 427
 428 PRED16x16_X(127, (1<<(BIT_DEPTH-1))-1)
 429 PRED16x16_X(128, (1<<(BIT_DEPTH-1))+0)
 430 PRED16x16_X(129, (1<<(BIT_DEPTH-1))+1)
 431
 432 static inline void FUNCC(pred16x16_plane_compat)(uint8_t *_src,
 433                                                  ptrdiff_t _stride,
 434                                                  const int svq3,
 435                                                  const int rv40)
 436 {
 437   int i, j, k;
 438   int a;
 439   INIT_CLIP
 440   pixel *src = (pixel*)_src;
 441   int stride = _stride/sizeof(pixel);
 442   const pixel * const src0 = src +7-stride;
 443   const pixel *       src1 = src +8*stride-1;
 444   const pixel *       src2 = src1-2*stride;    // == src+6*stride-1;
 445   int H = src0[1] - src0[-1];
 446   int V = src1[0] - src2[ 0];
 447   for(k=2; k<=8; ++k) {
 448     src1 += stride; src2 -= stride;
 449     H += k*(src0[k] - src0[-k]);
 450     V += k*(src1[0] - src2[ 0]);
 451   }
 452   if(svq3){
 453     H = ( 5*(H/4) ) / 16;
 454     V = ( 5*(V/4) ) / 16;
 455
 456     /* required for 100% accuracy */
 457     i = H; H = V; V = i;
 458   }else if(rv40){
 459     H = ( H + (H>>2) ) >> 4;
 460     V = ( V + (V>>2) ) >> 4;
 461   }else{
 462     H = ( 5*H+32 ) >> 6;
 463     V = ( 5*V+32 ) >> 6;
 464   }
 465
 466   a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
 467   for(j=16; j>0; --j) {
 468     int b = a;
 469     a += V;
 470     for(i=-16; i<0; i+=4) {
 471       src[16+i] = CLIP((b    ) >> 5);
 472       src[17+i] = CLIP((b+  H) >> 5);
 473       src[18+i] = CLIP((b+2*H) >> 5);
 474       src[19+i] = CLIP((b+3*H) >> 5);
 475       b += 4*H;
 476     }
 477     src += stride;
 478   }
 479 }
 480
 481 static void FUNCC(pred16x16_plane)(uint8_t *src, ptrdiff_t stride)
 482 {
 483     FUNCC(pred16x16_plane_compat)(src, stride, 0, 0);
 484 }
 485
 486 static void FUNCC(pred8x8_vertical)(uint8_t *_src, ptrdiff_t _stride)
 487 {
 488     int i;
 489     pixel *src = (pixel*)_src;
 490     int stride = _stride/sizeof(pixel);
 491     const pixel4 a= AV_RN4PA(((pixel4*)(src-stride))+0);
 492     const pixel4 b= AV_RN4PA(((pixel4*)(src-stride))+1);
 493
 494     for(i=0; i<8; i++){
 495         AV_WN4PA(((pixel4*)(src+i*stride))+0, a);
 496         AV_WN4PA(((pixel4*)(src+i*stride))+1, b);
 497     }
 498 }
 499
 500 static void FUNCC(pred8x16_vertical)(uint8_t *_src, ptrdiff_t _stride)
 501 {
 502     int i;
 503     pixel *src = (pixel*)_src;
 504     int stride = _stride>>(sizeof(pixel)-1);
 505     const pixel4 a= AV_RN4PA(((pixel4*)(src-stride))+0);
 506     const pixel4 b= AV_RN4PA(((pixel4*)(src-stride))+1);
 507
 508     for(i=0; i<16; i++){
 509         AV_WN4PA(((pixel4*)(src+i*stride))+0, a);
 510         AV_WN4PA(((pixel4*)(src+i*stride))+1, b);
 511     }
 512 }
 513
 514 static void FUNCC(pred8x8_horizontal)(uint8_t *_src, ptrdiff_t stride)
 515 {
 516     int i;
 517     pixel *src = (pixel*)_src;
 518     stride /= sizeof(pixel);
 519
 520     for(i=0; i<8; i++){
 521         const pixel4 a = PIXEL_SPLAT_X4(src[-1+i*stride]);
 522         AV_WN4PA(((pixel4*)(src+i*stride))+0, a);
 523         AV_WN4PA(((pixel4*)(src+i*stride))+1, a);
 524     }
 525 }
 526
 527 static void FUNCC(pred8x16_horizontal)(uint8_t *_src, ptrdiff_t stride)
 528 {
 529     int i;
 530     pixel *src = (pixel*)_src;
 531     stride >>= sizeof(pixel)-1;
 532     for(i=0; i<16; i++){
 533         const pixel4 a = PIXEL_SPLAT_X4(src[-1+i*stride]);
 534         AV_WN4PA(((pixel4*)(src+i*stride))+0, a);
 535         AV_WN4PA(((pixel4*)(src+i*stride))+1, a);
 536     }
 537 }
 538
 539 #define PRED8x8_X(n, v)\
 540 static void FUNCC(pred8x8_##n##_dc)(uint8_t *_src, ptrdiff_t stride)\
 541 {\
 542     int i;\
 543     const pixel4 a = PIXEL_SPLAT_X4(v);\
 544     pixel *src = (pixel*)_src;\
 545     stride /= sizeof(pixel);\
 546     for(i=0; i<8; i++){\
 547         AV_WN4PA(((pixel4*)(src+i*stride))+0, a);\
 548         AV_WN4PA(((pixel4*)(src+i*stride))+1, a);\
 549     }\
 550 }
 551
 552 PRED8x8_X(127, (1<<(BIT_DEPTH-1))-1)
 553 PRED8x8_X(128, (1<<(BIT_DEPTH-1))+0)
 554 PRED8x8_X(129, (1<<(BIT_DEPTH-1))+1)
 555
 556 static void FUNCC(pred8x16_128_dc)(uint8_t *_src, ptrdiff_t stride)
 557 {
 558     FUNCC(pred8x8_128_dc)(_src, stride);
 559     FUNCC(pred8x8_128_dc)(_src+8*stride, stride);
 560 }
 561
 562 static void FUNCC(pred8x8_left_dc)(uint8_t *_src, ptrdiff_t stride)
 563 {
 564     int i;
 565     int dc0, dc2;
 566     pixel4 dc0splat, dc2splat;
 567     pixel *src = (pixel*)_src;
 568     stride /= sizeof(pixel);
 569
 570     dc0=dc2=0;
 571     for(i=0;i<4; i++){
 572         dc0+= src[-1+i*stride];
 573         dc2+= src[-1+(i+4)*stride];
 574     }
 575     dc0splat = PIXEL_SPLAT_X4((dc0 + 2)>>2);
 576     dc2splat = PIXEL_SPLAT_X4((dc2 + 2)>>2);
 577
 578     for(i=0; i<4; i++){
 579         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc0splat);
 580         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc0splat);
 581     }
 582     for(i=4; i<8; i++){
 583         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc2splat);
 584         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc2splat);
 585     }
 586 }
 587
 588 static void FUNCC(pred8x16_left_dc)(uint8_t *_src, ptrdiff_t stride)
 589 {
 590     FUNCC(pred8x8_left_dc)(_src, stride);
 591     FUNCC(pred8x8_left_dc)(_src+8*stride, stride);
 592 }
 593
 594 static void FUNCC(pred8x8_top_dc)(uint8_t *_src, ptrdiff_t stride)
 595 {
 596     int i;
 597     int dc0, dc1;
 598     pixel4 dc0splat, dc1splat;
 599     pixel *src = (pixel*)_src;
 600     stride /= sizeof(pixel);
 601
 602     dc0=dc1=0;
 603     for(i=0;i<4; i++){
 604         dc0+= src[i-stride];
 605         dc1+= src[4+i-stride];
 606     }
 607     dc0splat = PIXEL_SPLAT_X4((dc0 + 2)>>2);
 608     dc1splat = PIXEL_SPLAT_X4((dc1 + 2)>>2);
 609
 610     for(i=0; i<4; i++){
 611         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc0splat);
 612         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc1splat);
 613     }
 614     for(i=4; i<8; i++){
 615         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc0splat);
 616         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc1splat);
 617     }
 618 }
 619
 620 static void FUNCC(pred8x16_top_dc)(uint8_t *_src, ptrdiff_t stride)
 621 {
 622     int i;
 623     int dc0, dc1;
 624     pixel4 dc0splat, dc1splat;
 625     pixel *src = (pixel*)_src;
 626     stride >>= sizeof(pixel)-1;
 627
 628     dc0=dc1=0;
 629     for(i=0;i<4; i++){
 630         dc0+= src[i-stride];
 631         dc1+= src[4+i-stride];
 632     }
 633     dc0splat = PIXEL_SPLAT_X4((dc0 + 2)>>2);
 634     dc1splat = PIXEL_SPLAT_X4((dc1 + 2)>>2);
 635
 636     for(i=0; i<16; i++){
 637         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc0splat);
 638         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc1splat);
 639     }
 640 }
 641
 642 static void FUNCC(pred8x8_dc)(uint8_t *_src, ptrdiff_t stride)
 643 {
 644     int i;
 645     int dc0, dc1, dc2;
 646     pixel4 dc0splat, dc1splat, dc2splat, dc3splat;
 647     pixel *src = (pixel*)_src;
 648     stride /= sizeof(pixel);
 649
 650     dc0=dc1=dc2=0;
 651     for(i=0;i<4; i++){
 652         dc0+= src[-1+i*stride] + src[i-stride];
 653         dc1+= src[4+i-stride];
 654         dc2+= src[-1+(i+4)*stride];
 655     }
 656     dc0splat = PIXEL_SPLAT_X4((dc0 + 4)>>3);
 657     dc1splat = PIXEL_SPLAT_X4((dc1 + 2)>>2);
 658     dc2splat = PIXEL_SPLAT_X4((dc2 + 2)>>2);
 659     dc3splat = PIXEL_SPLAT_X4((dc1 + dc2 + 4)>>3);
 660
 661     for(i=0; i<4; i++){
 662         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc0splat);
 663         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc1splat);
 664     }
 665     for(i=4; i<8; i++){
 666         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc2splat);
 667         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc3splat);
 668     }
 669 }
 670
 671 static void FUNCC(pred8x16_dc)(uint8_t *_src, ptrdiff_t stride)
 672 {
 673     int i;
 674     int dc0, dc1, dc2, dc3, dc4;
 675     pixel4 dc0splat, dc1splat, dc2splat, dc3splat, dc4splat, dc5splat, dc6splat, dc7splat;
 676     pixel *src = (pixel*)_src;
 677     stride >>= sizeof(pixel)-1;
 678
 679     dc0=dc1=dc2=dc3=dc4=0;
 680     for(i=0;i<4; i++){
 681         dc0+= src[-1+i*stride] + src[i-stride];
 682         dc1+= src[4+i-stride];
 683         dc2+= src[-1+(i+4)*stride];
 684         dc3+= src[-1+(i+8)*stride];
 685         dc4+= src[-1+(i+12)*stride];
 686     }
 687     dc0splat = PIXEL_SPLAT_X4((dc0 + 4)>>3);
 688     dc1splat = PIXEL_SPLAT_X4((dc1 + 2)>>2);
 689     dc2splat = PIXEL_SPLAT_X4((dc2 + 2)>>2);
 690     dc3splat = PIXEL_SPLAT_X4((dc1 + dc2 + 4)>>3);
 691     dc4splat = PIXEL_SPLAT_X4((dc3 + 2)>>2);
 692     dc5splat = PIXEL_SPLAT_X4((dc1 + dc3 + 4)>>3);
 693     dc6splat = PIXEL_SPLAT_X4((dc4 + 2)>>2);
 694     dc7splat = PIXEL_SPLAT_X4((dc1 + dc4 + 4)>>3);
 695
 696     for(i=0; i<4; i++){
 697         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc0splat);
 698         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc1splat);
 699     }
 700     for(i=4; i<8; i++){
 701         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc2splat);
 702         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc3splat);
 703     }
 704     for(i=8; i<12; i++){
 705         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc4splat);
 706         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc5splat);
 707     }
 708     for(i=12; i<16; i++){
 709         AV_WN4PA(((pixel4*)(src+i*stride))+0, dc6splat);
 710         AV_WN4PA(((pixel4*)(src+i*stride))+1, dc7splat);
 711     }
 712 }
 713
 714 static void FUNC(pred8x8_mad_cow_dc_l0t)(uint8_t *src, ptrdiff_t stride)
 715 {
 716     FUNCC(pred8x8_top_dc)(src, stride);
 717     FUNCC(pred4x4_dc)(src, NULL, stride);
 718 }
 719
 720 static void FUNC(pred8x16_mad_cow_dc_l0t)(uint8_t *src, ptrdiff_t stride)
 721 {
 722     FUNCC(pred8x16_top_dc)(src, stride);
 723     FUNCC(pred4x4_dc)(src, NULL, stride);
 724 }
 725
 726 static void FUNC(pred8x8_mad_cow_dc_0lt)(uint8_t *src, ptrdiff_t stride)
 727 {
 728     FUNCC(pred8x8_dc)(src, stride);
 729     FUNCC(pred4x4_top_dc)(src, NULL, stride);
 730 }
 731
 732 static void FUNC(pred8x16_mad_cow_dc_0lt)(uint8_t *src, ptrdiff_t stride)
 733 {
 734     FUNCC(pred8x16_dc)(src, stride);
 735     FUNCC(pred4x4_top_dc)(src, NULL, stride);
 736 }
 737
 738 static void FUNC(pred8x8_mad_cow_dc_l00)(uint8_t *src, ptrdiff_t stride)
 739 {
 740     FUNCC(pred8x8_left_dc)(src, stride);
 741     FUNCC(pred4x4_128_dc)(src + 4*stride                  , NULL, stride);
 742     FUNCC(pred4x4_128_dc)(src + 4*stride + 4*sizeof(pixel), NULL, stride);
 743 }
 744
 745 static void FUNC(pred8x16_mad_cow_dc_l00)(uint8_t *src, ptrdiff_t stride)
 746 {
 747     FUNCC(pred8x16_left_dc)(src, stride);
 748     FUNCC(pred4x4_128_dc)(src + 4*stride                  , NULL, stride);
 749     FUNCC(pred4x4_128_dc)(src + 4*stride + 4*sizeof(pixel), NULL, stride);
 750 }
 751
 752 static void FUNC(pred8x8_mad_cow_dc_0l0)(uint8_t *src, ptrdiff_t stride)
 753 {
 754     FUNCC(pred8x8_left_dc)(src, stride);
 755     FUNCC(pred4x4_128_dc)(src                  , NULL, stride);
 756     FUNCC(pred4x4_128_dc)(src + 4*sizeof(pixel), NULL, stride);
 757 }
 758
 759 static void FUNC(pred8x16_mad_cow_dc_0l0)(uint8_t *src, ptrdiff_t stride)
 760 {
 761     FUNCC(pred8x16_left_dc)(src, stride);
 762     FUNCC(pred4x4_128_dc)(src                  , NULL, stride);
 763     FUNCC(pred4x4_128_dc)(src + 4*sizeof(pixel), NULL, stride);
 764 }
 765
 766 static void FUNCC(pred8x8_plane)(uint8_t *_src, ptrdiff_t _stride)
 767 {
 768   int j, k;
 769   int a;
 770   INIT_CLIP
 771   pixel *src = (pixel*)_src;
 772   int stride = _stride/sizeof(pixel);
 773   const pixel * const src0 = src +3-stride;
 774   const pixel *       src1 = src +4*stride-1;
 775   const pixel *       src2 = src1-2*stride;    // == src+2*stride-1;
 776   int H = src0[1] - src0[-1];
 777   int V = src1[0] - src2[ 0];
 778   for(k=2; k<=4; ++k) {
 779     src1 += stride; src2 -= stride;
 780     H += k*(src0[k] - src0[-k]);
 781     V += k*(src1[0] - src2[ 0]);
 782   }
 783   H = ( 17*H+16 ) >> 5;
 784   V = ( 17*V+16 ) >> 5;
 785
 786   a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
 787   for(j=8; j>0; --j) {
 788     int b = a;
 789     a += V;
 790     src[0] = CLIP((b    ) >> 5);
 791     src[1] = CLIP((b+  H) >> 5);
 792     src[2] = CLIP((b+2*H) >> 5);
 793     src[3] = CLIP((b+3*H) >> 5);
 794     src[4] = CLIP((b+4*H) >> 5);
 795     src[5] = CLIP((b+5*H) >> 5);
 796     src[6] = CLIP((b+6*H) >> 5);
 797     src[7] = CLIP((b+7*H) >> 5);
 798     src += stride;
 799   }
 800 }
 801
 802 static void FUNCC(pred8x16_plane)(uint8_t *_src, ptrdiff_t _stride)
 803 {
 804   int j, k;
 805   int a;
 806   INIT_CLIP
 807   pixel *src = (pixel*)_src;
 808   int stride = _stride>>(sizeof(pixel)-1);
 809   const pixel * const src0 = src +3-stride;
 810   const pixel *       src1 = src +8*stride-1;
 811   const pixel *       src2 = src1-2*stride;    // == src+6*stride-1;
 812   int H = src0[1] - src0[-1];
 813   int V = src1[0] - src2[ 0];
 814
 815   for (k = 2; k <= 4; ++k) {
 816       src1 += stride; src2 -= stride;
 817       H += k*(src0[k] - src0[-k]);
 818       V += k*(src1[0] - src2[ 0]);
 819   }
 820   for (; k <= 8; ++k) {
 821       src1 += stride; src2 -= stride;
 822       V += k*(src1[0] - src2[0]);
 823   }
 824
 825   H = (17*H+16) >> 5;
 826   V = (5*V+32) >> 6;
 827
 828   a = 16*(src1[0] + src2[8] + 1) - 7*V - 3*H;
 829   for(j=16; j>0; --j) {
 830     int b = a;
 831     a += V;
 832     src[0] = CLIP((b    ) >> 5);
 833     src[1] = CLIP((b+  H) >> 5);
 834     src[2] = CLIP((b+2*H) >> 5);
 835     src[3] = CLIP((b+3*H) >> 5);
 836     src[4] = CLIP((b+4*H) >> 5);
 837     src[5] = CLIP((b+5*H) >> 5);
 838     src[6] = CLIP((b+6*H) >> 5);
 839     src[7] = CLIP((b+7*H) >> 5);
 840     src += stride;
 841   }
 842 }
 843
 844 #define SRC(x,y) src[(x)+(y)*stride]
 845 #define PL(y) \
 846     const int l##y = (SRC(-1,y-1) + 2*SRC(-1,y) + SRC(-1,y+1) + 2) >> 2;
 847 #define PREDICT_8x8_LOAD_LEFT \
 848     const int l0 = ((has_topleft ? SRC(-1,-1) : SRC(-1,0)) \
 849                      + 2*SRC(-1,0) + SRC(-1,1) + 2) >> 2; \
 850     PL(1) PL(2) PL(3) PL(4) PL(5) PL(6) \
 851     const int l7 av_unused = (SRC(-1,6) + 3*SRC(-1,7) + 2) >> 2
 852
 853 #define PT(x) \
 854     const int t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
 855 #define PREDICT_8x8_LOAD_TOP \
 856     const int t0 = ((has_topleft ? SRC(-1,-1) : SRC(0,-1)) \
 857                      + 2*SRC(0,-1) + SRC(1,-1) + 2) >> 2; \
 858     PT(1) PT(2) PT(3) PT(4) PT(5) PT(6) \
 859     const int t7 av_unused = ((has_topright ? SRC(8,-1) : SRC(7,-1)) \
 860                      + 2*SRC(7,-1) + SRC(6,-1) + 2) >> 2
 861
 862 #define PTR(x) \
 863     t##x = (SRC(x-1,-1) + 2*SRC(x,-1) + SRC(x+1,-1) + 2) >> 2;
 864 #define PREDICT_8x8_LOAD_TOPRIGHT \
 865     int t8, t9, t10, t11, t12, t13, t14, t15; \
 866     if(has_topright) { \
 867         PTR(8) PTR(9) PTR(10) PTR(11) PTR(12) PTR(13) PTR(14) \
 868         t15 = (SRC(14,-1) + 3*SRC(15,-1) + 2) >> 2; \
 869     } else t8=t9=t10=t11=t12=t13=t14=t15= SRC(7,-1);
 870
 871 #define PREDICT_8x8_LOAD_TOPLEFT \
 872     const int lt = (SRC(-1,0) + 2*SRC(-1,-1) + SRC(0,-1) + 2) >> 2
 873
 874 #define PREDICT_8x8_DC(v) \
 875     int y; \
 876     for( y = 0; y < 8; y++ ) { \
 877         AV_WN4PA(((pixel4*)src)+0, v); \
 878         AV_WN4PA(((pixel4*)src)+1, v); \
 879         src += stride; \
 880     }
 881
 882 static void FUNCC(pred8x8l_128_dc)(uint8_t *_src, int has_topleft,
 883                                    int has_topright, ptrdiff_t _stride)
 884 {
 885     pixel *src = (pixel*)_src;
 886     int stride = _stride/sizeof(pixel);
 887
 888     PREDICT_8x8_DC(PIXEL_SPLAT_X4(1<<(BIT_DEPTH-1)));
 889 }
 890 static void FUNCC(pred8x8l_left_dc)(uint8_t *_src, int has_topleft,
 891                                     int has_topright, ptrdiff_t _stride)
 892 {
 893     pixel *src = (pixel*)_src;
 894     int stride = _stride/sizeof(pixel);
 895
 896     PREDICT_8x8_LOAD_LEFT;
 897     const pixel4 dc = PIXEL_SPLAT_X4((l0+l1+l2+l3+l4+l5+l6+l7+4) >> 3);
 898     PREDICT_8x8_DC(dc);
 899 }
 900 static void FUNCC(pred8x8l_top_dc)(uint8_t *_src, int has_topleft,
 901                                    int has_topright, ptrdiff_t _stride)
 902 {
 903     pixel *src = (pixel*)_src;
 904     int stride = _stride/sizeof(pixel);
 905
 906     PREDICT_8x8_LOAD_TOP;
 907     const pixel4 dc = PIXEL_SPLAT_X4((t0+t1+t2+t3+t4+t5+t6+t7+4) >> 3);
 908     PREDICT_8x8_DC(dc);
 909 }
 910 static void FUNCC(pred8x8l_dc)(uint8_t *_src, int has_topleft,
 911                                int has_topright, ptrdiff_t _stride)
 912 {
 913     pixel *src = (pixel*)_src;
 914     int stride = _stride/sizeof(pixel);
 915
 916     PREDICT_8x8_LOAD_LEFT;
 917     PREDICT_8x8_LOAD_TOP;
 918     const pixel4 dc = PIXEL_SPLAT_X4((l0+l1+l2+l3+l4+l5+l6+l7
 919                                      +t0+t1+t2+t3+t4+t5+t6+t7+8) >> 4);
 920     PREDICT_8x8_DC(dc);
 921 }
 922 static void FUNCC(pred8x8l_horizontal)(uint8_t *_src, int has_topleft,
 923                                        int has_topright, ptrdiff_t _stride)
 924 {
 925     pixel *src = (pixel*)_src;
 926     int stride = _stride/sizeof(pixel);
 927     pixel4 a;
 928
 929     PREDICT_8x8_LOAD_LEFT;
 930 #define ROW(y) a = PIXEL_SPLAT_X4(l##y); \
 931                AV_WN4PA(src+y*stride, a); \
 932                AV_WN4PA(src+y*stride+4, a);
 933     ROW(0); ROW(1); ROW(2); ROW(3); ROW(4); ROW(5); ROW(6); ROW(7);
 934 #undef ROW
 935 }
 936 static void FUNCC(pred8x8l_vertical)(uint8_t *_src, int has_topleft,
 937                                      int has_topright, ptrdiff_t _stride)
 938 {
 939     int y;
 940     pixel *src = (pixel*)_src;
 941     int stride = _stride/sizeof(pixel);
 942     pixel4 a, b;
 943
 944     PREDICT_8x8_LOAD_TOP;
 945     src[0] = t0;
 946     src[1] = t1;
 947     src[2] = t2;
 948     src[3] = t3;
 949     src[4] = t4;
 950     src[5] = t5;
 951     src[6] = t6;
 952     src[7] = t7;
 953     a = AV_RN4PA(((pixel4*)src)+0);
 954     b = AV_RN4PA(((pixel4*)src)+1);
 955     for( y = 1; y < 8; y++ ) {
 956         AV_WN4PA(((pixel4*)(src+y*stride))+0, a);
 957         AV_WN4PA(((pixel4*)(src+y*stride))+1, b);
 958     }
 959 }
 960 static void FUNCC(pred8x8l_down_left)(uint8_t *_src, int has_topleft,
 961                                       int has_topright, ptrdiff_t _stride)
 962 {
 963     pixel *src = (pixel*)_src;
 964     int stride = _stride/sizeof(pixel);
 965     PREDICT_8x8_LOAD_TOP;
 966     PREDICT_8x8_LOAD_TOPRIGHT;
 967     SRC(0,0)= (t0 + 2*t1 + t2 + 2) >> 2;
 968     SRC(0,1)=SRC(1,0)= (t1 + 2*t2 + t3 + 2) >> 2;
 969     SRC(0,2)=SRC(1,1)=SRC(2,0)= (t2 + 2*t3 + t4 + 2) >> 2;
 970     SRC(0,3)=SRC(1,2)=SRC(2,1)=SRC(3,0)= (t3 + 2*t4 + t5 + 2) >> 2;
 971     SRC(0,4)=SRC(1,3)=SRC(2,2)=SRC(3,1)=SRC(4,0)= (t4 + 2*t5 + t6 + 2) >> 2;
 972     SRC(0,5)=SRC(1,4)=SRC(2,3)=SRC(3,2)=SRC(4,1)=SRC(5,0)= (t5 + 2*t6 + t7 + 2) >> 2;
 973     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;
 974     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;
 975     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;
 976     SRC(2,7)=SRC(3,6)=SRC(4,5)=SRC(5,4)=SRC(6,3)=SRC(7,2)= (t9 + 2*t10 + t11 + 2) >> 2;
 977     SRC(3,7)=SRC(4,6)=SRC(5,5)=SRC(6,4)=SRC(7,3)= (t10 + 2*t11 + t12 + 2) >> 2;
 978     SRC(4,7)=SRC(5,6)=SRC(6,5)=SRC(7,4)= (t11 + 2*t12 + t13 + 2) >> 2;
 979     SRC(5,7)=SRC(6,6)=SRC(7,5)= (t12 + 2*t13 + t14 + 2) >> 2;
 980     SRC(6,7)=SRC(7,6)= (t13 + 2*t14 + t15 + 2) >> 2;
 981     SRC(7,7)= (t14 + 3*t15 + 2) >> 2;
 982 }
 983 static void FUNCC(pred8x8l_down_right)(uint8_t *_src, int has_topleft,
 984                                        int has_topright, ptrdiff_t _stride)
 985 {
 986     pixel *src = (pixel*)_src;
 987     int stride = _stride/sizeof(pixel);
 988     PREDICT_8x8_LOAD_TOP;
 989     PREDICT_8x8_LOAD_LEFT;
 990     PREDICT_8x8_LOAD_TOPLEFT;
 991     SRC(0,7)= (l7 + 2*l6 + l5 + 2) >> 2;
 992     SRC(0,6)=SRC(1,7)= (l6 + 2*l5 + l4 + 2) >> 2;
 993     SRC(0,5)=SRC(1,6)=SRC(2,7)= (l5 + 2*l4 + l3 + 2) >> 2;
 994     SRC(0,4)=SRC(1,5)=SRC(2,6)=SRC(3,7)= (l4 + 2*l3 + l2 + 2) >> 2;
 995     SRC(0,3)=SRC(1,4)=SRC(2,5)=SRC(3,6)=SRC(4,7)= (l3 + 2*l2 + l1 + 2) >> 2;
 996     SRC(0,2)=SRC(1,3)=SRC(2,4)=SRC(3,5)=SRC(4,6)=SRC(5,7)= (l2 + 2*l1 + l0 + 2) >> 2;
 997     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;
 998     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;
 999     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;
1000     SRC(2,0)=SRC(3,1)=SRC(4,2)=SRC(5,3)=SRC(6,4)=SRC(7,5)= (t0 + 2*t1 + t2 + 2) >> 2;
1001     SRC(3,0)=SRC(4,1)=SRC(5,2)=SRC(6,3)=SRC(7,4)= (t1 + 2*t2 + t3 + 2) >> 2;
1002     SRC(4,0)=SRC(5,1)=SRC(6,2)=SRC(7,3)= (t2 + 2*t3 + t4 + 2) >> 2;
1003     SRC(5,0)=SRC(6,1)=SRC(7,2)= (t3 + 2*t4 + t5 + 2) >> 2;
1004     SRC(6,0)=SRC(7,1)= (t4 + 2*t5 + t6 + 2) >> 2;
1005     SRC(7,0)= (t5 + 2*t6 + t7 + 2) >> 2;
1006 }
1007 static void FUNCC(pred8x8l_vertical_right)(uint8_t *_src, int has_topleft,
1008                                            int has_topright, ptrdiff_t _stride)
1009 {
1010     pixel *src = (pixel*)_src;
1011     int stride = _stride/sizeof(pixel);
1012     PREDICT_8x8_LOAD_TOP;
1013     PREDICT_8x8_LOAD_LEFT;
1014     PREDICT_8x8_LOAD_TOPLEFT;
1015     SRC(0,6)= (l5 + 2*l4 + l3 + 2) >> 2;
1016     SRC(0,7)= (l6 + 2*l5 + l4 + 2) >> 2;
1017     SRC(0,4)=SRC(1,6)= (l3 + 2*l2 + l1 + 2) >> 2;
1018     SRC(0,5)=SRC(1,7)= (l4 + 2*l3 + l2 + 2) >> 2;
1019     SRC(0,2)=SRC(1,4)=SRC(2,6)= (l1 + 2*l0 + lt + 2) >> 2;
1020     SRC(0,3)=SRC(1,5)=SRC(2,7)= (l2 + 2*l1 + l0 + 2) >> 2;
1021     SRC(0,1)=SRC(1,3)=SRC(2,5)=SRC(3,7)= (l0 + 2*lt + t0 + 2) >> 2;
1022     SRC(0,0)=SRC(1,2)=SRC(2,4)=SRC(3,6)= (lt + t0 + 1) >> 1;
1023     SRC(1,1)=SRC(2,3)=SRC(3,5)=SRC(4,7)= (lt + 2*t0 + t1 + 2) >> 2;
1024     SRC(1,0)=SRC(2,2)=SRC(3,4)=SRC(4,6)= (t0 + t1 + 1) >> 1;
1025     SRC(2,1)=SRC(3,3)=SRC(4,5)=SRC(5,7)= (t0 + 2*t1 + t2 + 2) >> 2;
1026     SRC(2,0)=SRC(3,2)=SRC(4,4)=SRC(5,6)= (t1 + t2 + 1) >> 1;
1027     SRC(3,1)=SRC(4,3)=SRC(5,5)=SRC(6,7)= (t1 + 2*t2 + t3 + 2) >> 2;
1028     SRC(3,0)=SRC(4,2)=SRC(5,4)=SRC(6,6)= (t2 + t3 + 1) >> 1;
1029     SRC(4,1)=SRC(5,3)=SRC(6,5)=SRC(7,7)= (t2 + 2*t3 + t4 + 2) >> 2;
1030     SRC(4,0)=SRC(5,2)=SRC(6,4)=SRC(7,6)= (t3 + t4 + 1) >> 1;
1031     SRC(5,1)=SRC(6,3)=SRC(7,5)= (t3 + 2*t4 + t5 + 2) >> 2;
1032     SRC(5,0)=SRC(6,2)=SRC(7,4)= (t4 + t5 + 1) >> 1;
1033     SRC(6,1)=SRC(7,3)= (t4 + 2*t5 + t6 + 2) >> 2;
1034     SRC(6,0)=SRC(7,2)= (t5 + t6 + 1) >> 1;
1035     SRC(7,1)= (t5 + 2*t6 + t7 + 2) >> 2;
1036     SRC(7,0)= (t6 + t7 + 1) >> 1;
1037 }
1038 static void FUNCC(pred8x8l_horizontal_down)(uint8_t *_src, int has_topleft,
1039                                             int has_topright, ptrdiff_t _stride)
1040 {
1041     pixel *src = (pixel*)_src;
1042     int stride = _stride/sizeof(pixel);
1043     PREDICT_8x8_LOAD_TOP;
1044     PREDICT_8x8_LOAD_LEFT;
1045     PREDICT_8x8_LOAD_TOPLEFT;
1046     SRC(0,7)= (l6 + l7 + 1) >> 1;
1047     SRC(1,7)= (l5 + 2*l6 + l7 + 2) >> 2;
1048     SRC(0,6)=SRC(2,7)= (l5 + l6 + 1) >> 1;
1049     SRC(1,6)=SRC(3,7)= (l4 + 2*l5 + l6 + 2) >> 2;
1050     SRC(0,5)=SRC(2,6)=SRC(4,7)= (l4 + l5 + 1) >> 1;
1051     SRC(1,5)=SRC(3,6)=SRC(5,7)= (l3 + 2*l4 + l5 + 2) >> 2;
1052     SRC(0,4)=SRC(2,5)=SRC(4,6)=SRC(6,7)= (l3 + l4 + 1) >> 1;
1053     SRC(1,4)=SRC(3,5)=SRC(5,6)=SRC(7,7)= (l2 + 2*l3 + l4 + 2) >> 2;
1054     SRC(0,3)=SRC(2,4)=SRC(4,5)=SRC(6,6)= (l2 + l3 + 1) >> 1;
1055     SRC(1,3)=SRC(3,4)=SRC(5,5)=SRC(7,6)= (l1 + 2*l2 + l3 + 2) >> 2;
1056     SRC(0,2)=SRC(2,3)=SRC(4,4)=SRC(6,5)= (l1 + l2 + 1) >> 1;
1057     SRC(1,2)=SRC(3,3)=SRC(5,4)=SRC(7,5)= (l0 + 2*l1 + l2 + 2) >> 2;
1058     SRC(0,1)=SRC(2,2)=SRC(4,3)=SRC(6,4)= (l0 + l1 + 1) >> 1;
1059     SRC(1,1)=SRC(3,2)=SRC(5,3)=SRC(7,4)= (lt + 2*l0 + l1 + 2) >> 2;
1060     SRC(0,0)=SRC(2,1)=SRC(4,2)=SRC(6,3)= (lt + l0 + 1) >> 1;
1061     SRC(1,0)=SRC(3,1)=SRC(5,2)=SRC(7,3)= (l0 + 2*lt + t0 + 2) >> 2;
1062     SRC(2,0)=SRC(4,1)=SRC(6,2)= (t1 + 2*t0 + lt + 2) >> 2;
1063     SRC(3,0)=SRC(5,1)=SRC(7,2)= (t2 + 2*t1 + t0 + 2) >> 2;
1064     SRC(4,0)=SRC(6,1)= (t3 + 2*t2 + t1 + 2) >> 2;
1065     SRC(5,0)=SRC(7,1)= (t4 + 2*t3 + t2 + 2) >> 2;
1066     SRC(6,0)= (t5 + 2*t4 + t3 + 2) >> 2;
1067     SRC(7,0)= (t6 + 2*t5 + t4 + 2) >> 2;
1068 }
1069 static void FUNCC(pred8x8l_vertical_left)(uint8_t *_src, int has_topleft,
1070                                           int has_topright, ptrdiff_t _stride)
1071 {
1072     pixel *src = (pixel*)_src;
1073     int stride = _stride/sizeof(pixel);
1074     PREDICT_8x8_LOAD_TOP;
1075     PREDICT_8x8_LOAD_TOPRIGHT;
1076     SRC(0,0)= (t0 + t1 + 1) >> 1;
1077     SRC(0,1)= (t0 + 2*t1 + t2 + 2) >> 2;
1078     SRC(0,2)=SRC(1,0)= (t1 + t2 + 1) >> 1;
1079     SRC(0,3)=SRC(1,1)= (t1 + 2*t2 + t3 + 2) >> 2;
1080     SRC(0,4)=SRC(1,2)=SRC(2,0)= (t2 + t3 + 1) >> 1;
1081     SRC(0,5)=SRC(1,3)=SRC(2,1)= (t2 + 2*t3 + t4 + 2) >> 2;
1082     SRC(0,6)=SRC(1,4)=SRC(2,2)=SRC(3,0)= (t3 + t4 + 1) >> 1;
1083     SRC(0,7)=SRC(1,5)=SRC(2,3)=SRC(3,1)= (t3 + 2*t4 + t5 + 2) >> 2;
1084     SRC(1,6)=SRC(2,4)=SRC(3,2)=SRC(4,0)= (t4 + t5 + 1) >> 1;
1085     SRC(1,7)=SRC(2,5)=SRC(3,3)=SRC(4,1)= (t4 + 2*t5 + t6 + 2) >> 2;
1086     SRC(2,6)=SRC(3,4)=SRC(4,2)=SRC(5,0)= (t5 + t6 + 1) >> 1;
1087     SRC(2,7)=SRC(3,5)=SRC(4,3)=SRC(5,1)= (t5 + 2*t6 + t7 + 2) >> 2;
1088     SRC(3,6)=SRC(4,4)=SRC(5,2)=SRC(6,0)= (t6 + t7 + 1) >> 1;
1089     SRC(3,7)=SRC(4,5)=SRC(5,3)=SRC(6,1)= (t6 + 2*t7 + t8 + 2) >> 2;
1090     SRC(4,6)=SRC(5,4)=SRC(6,2)=SRC(7,0)= (t7 + t8 + 1) >> 1;
1091     SRC(4,7)=SRC(5,5)=SRC(6,3)=SRC(7,1)= (t7 + 2*t8 + t9 + 2) >> 2;
1092     SRC(5,6)=SRC(6,4)=SRC(7,2)= (t8 + t9 + 1) >> 1;
1093     SRC(5,7)=SRC(6,5)=SRC(7,3)= (t8 + 2*t9 + t10 + 2) >> 2;
1094     SRC(6,6)=SRC(7,4)= (t9 + t10 + 1) >> 1;
1095     SRC(6,7)=SRC(7,5)= (t9 + 2*t10 + t11 + 2) >> 2;
1096     SRC(7,6)= (t10 + t11 + 1) >> 1;
1097     SRC(7,7)= (t10 + 2*t11 + t12 + 2) >> 2;
1098 }
1099 static void FUNCC(pred8x8l_horizontal_up)(uint8_t *_src, int has_topleft,
1100                                           int has_topright, ptrdiff_t _stride)
1101 {
1102     pixel *src = (pixel*)_src;
1103     int stride = _stride/sizeof(pixel);
1104     PREDICT_8x8_LOAD_LEFT;
1105     SRC(0,0)= (l0 + l1 + 1) >> 1;
1106     SRC(1,0)= (l0 + 2*l1 + l2 + 2) >> 2;
1107     SRC(0,1)=SRC(2,0)= (l1 + l2 + 1) >> 1;
1108     SRC(1,1)=SRC(3,0)= (l1 + 2*l2 + l3 + 2) >> 2;
1109     SRC(0,2)=SRC(2,1)=SRC(4,0)= (l2 + l3 + 1) >> 1;
1110     SRC(1,2)=SRC(3,1)=SRC(5,0)= (l2 + 2*l3 + l4 + 2) >> 2;
1111     SRC(0,3)=SRC(2,2)=SRC(4,1)=SRC(6,0)= (l3 + l4 + 1) >> 1;
1112     SRC(1,3)=SRC(3,2)=SRC(5,1)=SRC(7,0)= (l3 + 2*l4 + l5 + 2) >> 2;
1113     SRC(0,4)=SRC(2,3)=SRC(4,2)=SRC(6,1)= (l4 + l5 + 1) >> 1;
1114     SRC(1,4)=SRC(3,3)=SRC(5,2)=SRC(7,1)= (l4 + 2*l5 + l6 + 2) >> 2;
1115     SRC(0,5)=SRC(2,4)=SRC(4,3)=SRC(6,2)= (l5 + l6 + 1) >> 1;
1116     SRC(1,5)=SRC(3,4)=SRC(5,3)=SRC(7,2)= (l5 + 2*l6 + l7 + 2) >> 2;
1117     SRC(0,6)=SRC(2,5)=SRC(4,4)=SRC(6,3)= (l6 + l7 + 1) >> 1;
1118     SRC(1,6)=SRC(3,5)=SRC(5,4)=SRC(7,3)= (l6 + 3*l7 + 2) >> 2;
1119     SRC(0,7)=SRC(1,7)=SRC(2,6)=SRC(2,7)=SRC(3,6)=
1120     SRC(3,7)=SRC(4,5)=SRC(4,6)=SRC(4,7)=SRC(5,5)=
1121     SRC(5,6)=SRC(5,7)=SRC(6,4)=SRC(6,5)=SRC(6,6)=
1122     SRC(6,7)=SRC(7,4)=SRC(7,5)=SRC(7,6)=SRC(7,7)= l7;
1123 }
1124 #undef PREDICT_8x8_LOAD_LEFT
1125 #undef PREDICT_8x8_LOAD_TOP
1126 #undef PREDICT_8x8_LOAD_TOPLEFT
1127 #undef PREDICT_8x8_LOAD_TOPRIGHT
1128 #undef PREDICT_8x8_DC
1129 #undef PTR
1130 #undef PT
1131 #undef PL
1132 #undef SRC
1133
1134 static void FUNCC(pred4x4_vertical_add)(uint8_t *_pix, const int16_t *_block,
1135                                         ptrdiff_t stride)
1136 {
1137     int i;
1138     pixel *pix = (pixel*)_pix;
1139     const dctcoef *block = (const dctcoef*)_block;
1140     stride /= sizeof(pixel);
1141     pix -= stride;
1142     for(i=0; i<4; i++){
1143         pixel v = pix[0];
1144         pix[1*stride]= v += block[0];
1145         pix[2*stride]= v += block[4];
1146         pix[3*stride]= v += block[8];
1147         pix[4*stride]= v +  block[12];
1148         pix++;
1149         block++;
1150     }
1151 }
1152
1153 static void FUNCC(pred4x4_horizontal_add)(uint8_t *_pix, const int16_t *_block,
1154                                           ptrdiff_t stride)
1155 {
1156     int i;
1157     pixel *pix = (pixel*)_pix;
1158     const dctcoef *block = (const dctcoef*)_block;
1159     stride /= sizeof(pixel);
1160     for(i=0; i<4; i++){
1161         pixel v = pix[-1];
1162         pix[0]= v += block[0];
1163         pix[1]= v += block[1];
1164         pix[2]= v += block[2];
1165         pix[3]= v +  block[3];
1166         pix+= stride;
1167         block+= 4;
1168     }
1169 }
1170
1171 static void FUNCC(pred8x8l_vertical_add)(uint8_t *_pix, const int16_t *_block,
1172                                          ptrdiff_t stride)
1173 {
1174     int i;
1175     pixel *pix = (pixel*)_pix;
1176     const dctcoef *block = (const dctcoef*)_block;
1177     stride /= sizeof(pixel);
1178     pix -= stride;
1179     for(i=0; i<8; i++){
1180         pixel v = pix[0];
1181         pix[1*stride]= v += block[0];
1182         pix[2*stride]= v += block[8];
1183         pix[3*stride]= v += block[16];
1184         pix[4*stride]= v += block[24];
1185         pix[5*stride]= v += block[32];
1186         pix[6*stride]= v += block[40];
1187         pix[7*stride]= v += block[48];
1188         pix[8*stride]= v +  block[56];
1189         pix++;
1190         block++;
1191     }
1192 }
1193
1194 static void FUNCC(pred8x8l_horizontal_add)(uint8_t *_pix, const int16_t *_block,
1195                                            ptrdiff_t stride)
1196 {
1197     int i;
1198     pixel *pix = (pixel*)_pix;
1199     const dctcoef *block = (const dctcoef*)_block;
1200     stride /= sizeof(pixel);
1201     for(i=0; i<8; i++){
1202         pixel v = pix[-1];
1203         pix[0]= v += block[0];
1204         pix[1]= v += block[1];
1205         pix[2]= v += block[2];
1206         pix[3]= v += block[3];
1207         pix[4]= v += block[4];
1208         pix[5]= v += block[5];
1209         pix[6]= v += block[6];
1210         pix[7]= v +  block[7];
1211         pix+= stride;
1212         block+= 8;
1213     }
1214 }
1215
1216 static void FUNCC(pred16x16_vertical_add)(uint8_t *pix, const int *block_offset,
1217                                           const int16_t *block,
1218                                           ptrdiff_t stride)
1219 {
1220     int i;
1221     for(i=0; i<16; i++)
1222         FUNCC(pred4x4_vertical_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
1223 }
1224
1225 static void FUNCC(pred16x16_horizontal_add)(uint8_t *pix,
1226                                             const int *block_offset,
1227                                             const int16_t *block,
1228                                             ptrdiff_t stride)
1229 {
1230     int i;
1231     for(i=0; i<16; i++)
1232         FUNCC(pred4x4_horizontal_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
1233 }
1234
1235 static void FUNCC(pred8x8_vertical_add)(uint8_t *pix, const int *block_offset,
1236                                         const int16_t *block, ptrdiff_t stride)
1237 {
1238     int i;
1239     for(i=0; i<4; i++)
1240         FUNCC(pred4x4_vertical_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
1241 }
1242
1243 static void FUNCC(pred8x16_vertical_add)(uint8_t *pix, const int *block_offset,
1244                                          const int16_t *block, ptrdiff_t stride)
1245 {
1246     int i;
1247     for(i=0; i<4; i++)
1248         FUNCC(pred4x4_vertical_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
1249     for(i=4; i<8; i++)
1250         FUNCC(pred4x4_vertical_add)(pix + block_offset[i+4], block + i*16*sizeof(pixel), stride);
1251 }
1252
1253 static void FUNCC(pred8x8_horizontal_add)(uint8_t *pix, const int *block_offset,
1254                                           const int16_t *block,
1255                                           ptrdiff_t stride)
1256 {
1257     int i;
1258     for(i=0; i<4; i++)
1259         FUNCC(pred4x4_horizontal_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
1260 }
1261
1262 static void FUNCC(pred8x16_horizontal_add)(uint8_t *pix,
1263                                            const int *block_offset,
1264                                            const int16_t *block, ptrdiff_t stride)
1265 {
1266     int i;
1267     for(i=0; i<4; i++)
1268         FUNCC(pred4x4_horizontal_add)(pix + block_offset[i], block + i*16*sizeof(pixel), stride);
1269     for(i=4; i<8; i++)
1270         FUNCC(pred4x4_horizontal_add)(pix + block_offset[i+4], block + i*16*sizeof(pixel), stride);
1271 }