libavcodec/ivi_dsp.c

   1 /*
   2  * DSP functions for Indeo Video Interactive codecs (Indeo4 and Indeo5)
   3  *
   4  * Copyright (c) 2009-2011 Maxim Poliakovski
   5  *
   6  * This file is part of Libav.
   7  *
   8  * Libav is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2.1 of the License, or (at your option) any later version.
  12  *
  13  * Libav is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with Libav; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22
  23 /**
  24  * @file
  25  * DSP functions (inverse transforms, motion compensation, wavelet recompostions)
  26  * for Indeo Video Interactive codecs.
  27  */
  28
  29 #include "avcodec.h"
  30 #include "ivi_common.h"
  31 #include "ivi_dsp.h"
  32
  33 void ff_ivi_recompose53(const IVIPlaneDesc *plane, uint8_t *dst,
  34                         const int dst_pitch)
  35 {
  36     int             x, y, indx;
  37     int32_t         p0, p1, p2, p3, tmp0, tmp1, tmp2;
  38     int32_t         b0_1, b0_2, b1_1, b1_2, b1_3, b2_1, b2_2, b2_3, b2_4, b2_5, b2_6;
  39     int32_t         b3_1, b3_2, b3_3, b3_4, b3_5, b3_6, b3_7, b3_8, b3_9;
  40     int32_t         pitch, back_pitch;
  41     const short    *b0_ptr, *b1_ptr, *b2_ptr, *b3_ptr;
  42     const int       num_bands = 4;
  43
  44     /* all bands should have the same pitch */
  45     pitch = plane->bands[0].pitch;
  46
  47     /* pixels at the position "y-1" will be set to pixels at the "y" for the 1st iteration */
  48     back_pitch = 0;
  49
  50     /* get pointers to the wavelet bands */
  51     b0_ptr = plane->bands[0].buf;
  52     b1_ptr = plane->bands[1].buf;
  53     b2_ptr = plane->bands[2].buf;
  54     b3_ptr = plane->bands[3].buf;
  55
  56     for (y = 0; y < plane->height; y += 2) {
  57         /* load storage variables with values */
  58         if (num_bands > 0) {
  59             b0_1 = b0_ptr[0];
  60             b0_2 = b0_ptr[pitch];
  61         }
  62
  63         if (num_bands > 1) {
  64             b1_1 = b1_ptr[back_pitch];
  65             b1_2 = b1_ptr[0];
  66             b1_3 = b1_1 - b1_2*6 + b1_ptr[pitch];
  67         }
  68
  69         if (num_bands > 2) {
  70             b2_2 = b2_ptr[0];     // b2[x,  y  ]
  71             b2_3 = b2_2;          // b2[x+1,y  ] = b2[x,y]
  72             b2_5 = b2_ptr[pitch]; // b2[x  ,y+1]
  73             b2_6 = b2_5;          // b2[x+1,y+1] = b2[x,y+1]
  74         }
  75
  76         if (num_bands > 3) {
  77             b3_2 = b3_ptr[back_pitch]; // b3[x  ,y-1]
  78             b3_3 = b3_2;               // b3[x+1,y-1] = b3[x  ,y-1]
  79             b3_5 = b3_ptr[0];          // b3[x  ,y  ]
  80             b3_6 = b3_5;               // b3[x+1,y  ] = b3[x  ,y  ]
  81             b3_8 = b3_2 - b3_5*6 + b3_ptr[pitch];
  82             b3_9 = b3_8;
  83         }
  84
  85         for (x = 0, indx = 0; x < plane->width; x+=2, indx++) {
  86             /* some values calculated in the previous iterations can */
  87             /* be reused in the next ones, so do appropriate copying */
  88             b2_1 = b2_2; // b2[x-1,y  ] = b2[x,  y  ]
  89             b2_2 = b2_3; // b2[x  ,y  ] = b2[x+1,y  ]
  90             b2_4 = b2_5; // b2[x-1,y+1] = b2[x  ,y+1]
  91             b2_5 = b2_6; // b2[x  ,y+1] = b2[x+1,y+1]
  92             b3_1 = b3_2; // b3[x-1,y-1] = b3[x  ,y-1]
  93             b3_2 = b3_3; // b3[x  ,y-1] = b3[x+1,y-1]
  94             b3_4 = b3_5; // b3[x-1,y  ] = b3[x  ,y  ]
  95             b3_5 = b3_6; // b3[x  ,y  ] = b3[x+1,y  ]
  96             b3_7 = b3_8; // vert_HPF(x-1)
  97             b3_8 = b3_9; // vert_HPF(x  )
  98
  99             p0 = p1 = p2 = p3 = 0;
 100
 101             /* process the LL-band by applying LPF both vertically and horizontally */
 102             if (num_bands > 0) {
 103                 tmp0 = b0_1;
 104                 tmp2 = b0_2;
 105                 b0_1 = b0_ptr[indx+1];
 106                 b0_2 = b0_ptr[pitch+indx+1];
 107                 tmp1 = tmp0 + b0_1;
 108
 109                 p0 =  tmp0 << 4;
 110                 p1 =  tmp1 << 3;
 111                 p2 = (tmp0 + tmp2) << 3;
 112                 p3 = (tmp1 + tmp2 + b0_2) << 2;
 113             }
 114
 115             /* process the HL-band by applying HPF vertically and LPF horizontally */
 116             if (num_bands > 1) {
 117                 tmp0 = b1_2;
 118                 tmp1 = b1_1;
 119                 b1_2 = b1_ptr[indx+1];
 120                 b1_1 = b1_ptr[back_pitch+indx+1];
 121
 122                 tmp2 = tmp1 - tmp0*6 + b1_3;
 123                 b1_3 = b1_1 - b1_2*6 + b1_ptr[pitch+indx+1];
 124
 125                 p0 += (tmp0 + tmp1) << 3;
 126                 p1 += (tmp0 + tmp1 + b1_1 + b1_2) << 2;
 127                 p2 +=  tmp2 << 2;
 128                 p3 += (tmp2 + b1_3) << 1;
 129             }
 130
 131             /* process the LH-band by applying LPF vertically and HPF horizontally */
 132             if (num_bands > 2) {
 133                 b2_3 = b2_ptr[indx+1];
 134                 b2_6 = b2_ptr[pitch+indx+1];
 135
 136                 tmp0 = b2_1 + b2_2;
 137                 tmp1 = b2_1 - b2_2*6 + b2_3;
 138
 139                 p0 += tmp0 << 3;
 140                 p1 += tmp1 << 2;
 141                 p2 += (tmp0 + b2_4 + b2_5) << 2;
 142                 p3 += (tmp1 + b2_4 - b2_5*6 + b2_6) << 1;
 143             }
 144
 145             /* process the HH-band by applying HPF both vertically and horizontally */
 146             if (num_bands > 3) {
 147                 b3_6 = b3_ptr[indx+1];            // b3[x+1,y  ]
 148                 b3_3 = b3_ptr[back_pitch+indx+1]; // b3[x+1,y-1]
 149
 150                 tmp0 = b3_1 + b3_4;
 151                 tmp1 = b3_2 + b3_5;
 152                 tmp2 = b3_3 + b3_6;
 153
 154                 b3_9 = b3_3 - b3_6*6 + b3_ptr[pitch+indx+1];
 155
 156                 p0 += (tmp0 + tmp1) << 2;
 157                 p1 += (tmp0 - tmp1*6 + tmp2) << 1;
 158                 p2 += (b3_7 + b3_8) << 1;
 159                 p3 +=  b3_7 - b3_8*6 + b3_9;
 160             }
 161
 162             /* output four pixels */
 163             dst[x]             = av_clip_uint8((p0 >> 6) + 128);
 164             dst[x+1]           = av_clip_uint8((p1 >> 6) + 128);
 165             dst[dst_pitch+x]   = av_clip_uint8((p2 >> 6) + 128);
 166             dst[dst_pitch+x+1] = av_clip_uint8((p3 >> 6) + 128);
 167         }// for x
 168
 169         dst += dst_pitch << 1;
 170
 171         back_pitch = -pitch;
 172
 173         b0_ptr += pitch;
 174         b1_ptr += pitch;
 175         b2_ptr += pitch;
 176         b3_ptr += pitch;
 177     }
 178 }
 179
 180 void ff_ivi_recompose_haar(const IVIPlaneDesc *plane, uint8_t *dst,
 181                            const int dst_pitch)
 182 {
 183     int             x, y, indx, b0, b1, b2, b3, p0, p1, p2, p3;
 184     const short    *b0_ptr, *b1_ptr, *b2_ptr, *b3_ptr;
 185     int32_t         pitch;
 186
 187     /* all bands should have the same pitch */
 188     pitch = plane->bands[0].pitch;
 189
 190     /* get pointers to the wavelet bands */
 191     b0_ptr = plane->bands[0].buf;
 192     b1_ptr = plane->bands[1].buf;
 193     b2_ptr = plane->bands[2].buf;
 194     b3_ptr = plane->bands[3].buf;
 195
 196     for (y = 0; y < plane->height; y += 2) {
 197         for (x = 0, indx = 0; x < plane->width; x += 2, indx++) {
 198             /* load coefficients */
 199             b0 = b0_ptr[indx]; //should be: b0 = (num_bands > 0) ? b0_ptr[indx] : 0;
 200             b1 = b1_ptr[indx]; //should be: b1 = (num_bands > 1) ? b1_ptr[indx] : 0;
 201             b2 = b2_ptr[indx]; //should be: b2 = (num_bands > 2) ? b2_ptr[indx] : 0;
 202             b3 = b3_ptr[indx]; //should be: b3 = (num_bands > 3) ? b3_ptr[indx] : 0;
 203
 204             /* haar wavelet recomposition */
 205             p0 = (b0 + b1 + b2 + b3 + 2) >> 2;
 206             p1 = (b0 + b1 - b2 - b3 + 2) >> 2;
 207             p2 = (b0 - b1 + b2 - b3 + 2) >> 2;
 208             p3 = (b0 - b1 - b2 + b3 + 2) >> 2;
 209
 210             /* bias, convert and output four pixels */
 211             dst[x]                 = av_clip_uint8(p0 + 128);
 212             dst[x + 1]             = av_clip_uint8(p1 + 128);
 213             dst[dst_pitch + x]     = av_clip_uint8(p2 + 128);
 214             dst[dst_pitch + x + 1] = av_clip_uint8(p3 + 128);
 215         }// for x
 216
 217         dst += dst_pitch << 1;
 218
 219         b0_ptr += pitch;
 220         b1_ptr += pitch;
 221         b2_ptr += pitch;
 222         b3_ptr += pitch;
 223     }// for y
 224 }
 225
 226 /** butterfly operation for the inverse Haar transform */
 227 #define IVI_HAAR_BFLY(s1, s2, o1, o2, t) \
 228     t  = (s1 - s2) >> 1;\
 229     o1 = (s1 + s2) >> 1;\
 230     o2 = t;\
 231
 232 /** inverse 8-point Haar transform */
 233 #define INV_HAAR8(s1, s5, s3, s7, s2, s4, s6, s8,\
 234                   d1, d2, d3, d4, d5, d6, d7, d8,\
 235                   t0, t1, t2, t3, t4, t5, t6, t7, t8) {\
 236     t1 = s1 << 1; t5 = s5 << 1;\
 237     IVI_HAAR_BFLY(t1, t5, t1, t5, t0); IVI_HAAR_BFLY(t1, s3, t1, t3, t0);\
 238     IVI_HAAR_BFLY(t5, s7, t5, t7, t0); IVI_HAAR_BFLY(t1, s2, t1, t2, t0);\
 239     IVI_HAAR_BFLY(t3, s4, t3, t4, t0); IVI_HAAR_BFLY(t5, s6, t5, t6, t0);\
 240     IVI_HAAR_BFLY(t7, s8, t7, t8, t0);\
 241     d1 = COMPENSATE(t1);\
 242     d2 = COMPENSATE(t2);\
 243     d3 = COMPENSATE(t3);\
 244     d4 = COMPENSATE(t4);\
 245     d5 = COMPENSATE(t5);\
 246     d6 = COMPENSATE(t6);\
 247     d7 = COMPENSATE(t7);\
 248     d8 = COMPENSATE(t8); }
 249
 250 /** inverse 4-point Haar transform */
 251 #define INV_HAAR4(s1, s3, s5, s7) {\
 252     HAAR_BFLY(s1, s5);  HAAR_BFLY(s1, s3);  HAAR_BFLY(s5, s7);\
 253     s1 = COMPENSATE(s1);\
 254     s3 = COMPENSATE(s3);\
 255     s5 = COMPENSATE(s5);\
 256     s7 = COMPENSATE(s7); }
 257
 258 void ff_ivi_inverse_haar_8x8(const int32_t *in, int16_t *out, uint32_t pitch,
 259                              const uint8_t *flags)
 260 {
 261     int     i, shift, sp1, sp2, sp3, sp4;
 262     const int32_t *src;
 263     int32_t *dst;
 264     int     tmp[64];
 265     int     t0, t1, t2, t3, t4, t5, t6, t7, t8;
 266
 267     /* apply the InvHaar8 to all columns */
 268 #define COMPENSATE(x) (x)
 269     src = in;
 270     dst = tmp;
 271     for (i = 0; i < 8; i++) {
 272         if (flags[i]) {
 273             /* pre-scaling */
 274             shift = !(i & 4);
 275             sp1 = src[ 0] << shift;
 276             sp2 = src[ 8] << shift;
 277             sp3 = src[16] << shift;
 278             sp4 = src[24] << shift;
 279             INV_HAAR8(    sp1,     sp2,     sp3,     sp4,
 280                       src[32], src[40], src[48], src[56],
 281                       dst[ 0], dst[ 8], dst[16], dst[24],
 282                       dst[32], dst[40], dst[48], dst[56],
 283                       t0, t1, t2, t3, t4, t5, t6, t7, t8);
 284         } else
 285             dst[ 0] = dst[ 8] = dst[16] = dst[24] =
 286             dst[32] = dst[40] = dst[48] = dst[56] = 0;
 287
 288         src++;
 289         dst++;
 290     }
 291 #undef  COMPENSATE
 292
 293     /* apply the InvHaar8 to all rows */
 294 #define COMPENSATE(x) (x)
 295     src = tmp;
 296     for (i = 0; i < 8; i++) {
 297         if (   !src[0] && !src[1] && !src[2] && !src[3]
 298             && !src[4] && !src[5] && !src[6] && !src[7]) {
 299             memset(out, 0, 8 * sizeof(out[0]));
 300         } else {
 301             INV_HAAR8(src[0], src[1], src[2], src[3],
 302                       src[4], src[5], src[6], src[7],
 303                       out[0], out[1], out[2], out[3],
 304                       out[4], out[5], out[6], out[7],
 305                       t0, t1, t2, t3, t4, t5, t6, t7, t8);
 306         }
 307         src += 8;
 308         out += pitch;
 309     }
 310 #undef  COMPENSATE
 311 }
 312
 313 void ff_ivi_dc_haar_2d(const int32_t *in, int16_t *out, uint32_t pitch,
 314                        int blk_size)
 315 {
 316     int     x, y;
 317     int16_t dc_coeff;
 318
 319     dc_coeff = (*in + 0) >> 3;
 320
 321     for (y = 0; y < blk_size; out += pitch, y++) {
 322         for (x = 0; x < blk_size; x++)
 323             out[x] = dc_coeff;
 324     }
 325 }
 326
 327 /** butterfly operation for the inverse slant transform */
 328 #define IVI_SLANT_BFLY(s1, s2, o1, o2, t) \
 329     t  = s1 - s2;\
 330     o1 = s1 + s2;\
 331     o2 = t;\
 332
 333 /** This is a reflection a,b = 1/2, 5/4 for the inverse slant transform */
 334 #define IVI_IREFLECT(s1, s2, o1, o2, t) \
 335     t  = ((s1 + s2*2 + 2) >> 2) + s1;\
 336     o2 = ((s1*2 - s2 + 2) >> 2) - s2;\
 337     o1 = t;\
 338
 339 /** This is a reflection a,b = 1/2, 7/8 for the inverse slant transform */
 340 #define IVI_SLANT_PART4(s1, s2, o1, o2, t) \
 341     t  = s2 + ((s1*4  - s2 + 4) >> 3);\
 342     o2 = s1 + ((-s1 - s2*4 + 4) >> 3);\
 343     o1 = t;\
 344
 345 /** inverse slant8 transform */
 346 #define IVI_INV_SLANT8(s1, s4, s8, s5, s2, s6, s3, s7,\
 347                        d1, d2, d3, d4, d5, d6, d7, d8,\
 348                        t0, t1, t2, t3, t4, t5, t6, t7, t8) {\
 349     IVI_SLANT_PART4(s4, s5, t4, t5, t0);\
 350 \
 351     IVI_SLANT_BFLY(s1, t5, t1, t5, t0); IVI_SLANT_BFLY(s2, s6, t2, t6, t0);\
 352     IVI_SLANT_BFLY(s7, s3, t7, t3, t0); IVI_SLANT_BFLY(t4, s8, t4, t8, t0);\
 353 \
 354     IVI_SLANT_BFLY(t1, t2, t1, t2, t0); IVI_IREFLECT  (t4, t3, t4, t3, t0);\
 355     IVI_SLANT_BFLY(t5, t6, t5, t6, t0); IVI_IREFLECT  (t8, t7, t8, t7, t0);\
 356     IVI_SLANT_BFLY(t1, t4, t1, t4, t0); IVI_SLANT_BFLY(t2, t3, t2, t3, t0);\
 357     IVI_SLANT_BFLY(t5, t8, t5, t8, t0); IVI_SLANT_BFLY(t6, t7, t6, t7, t0);\
 358     d1 = COMPENSATE(t1);\
 359     d2 = COMPENSATE(t2);\
 360     d3 = COMPENSATE(t3);\
 361     d4 = COMPENSATE(t4);\
 362     d5 = COMPENSATE(t5);\
 363     d6 = COMPENSATE(t6);\
 364     d7 = COMPENSATE(t7);\
 365     d8 = COMPENSATE(t8);}
 366
 367 /** inverse slant4 transform */
 368 #define IVI_INV_SLANT4(s1, s4, s2, s3, d1, d2, d3, d4, t0, t1, t2, t3, t4) {\
 369     IVI_SLANT_BFLY(s1, s2, t1, t2, t0); IVI_IREFLECT  (s4, s3, t4, t3, t0);\
 370 \
 371     IVI_SLANT_BFLY(t1, t4, t1, t4, t0); IVI_SLANT_BFLY(t2, t3, t2, t3, t0);\
 372     d1 = COMPENSATE(t1);\
 373     d2 = COMPENSATE(t2);\
 374     d3 = COMPENSATE(t3);\
 375     d4 = COMPENSATE(t4);}
 376
 377 void ff_ivi_inverse_slant_8x8(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
 378 {
 379     int     i;
 380     const int32_t *src;
 381     int32_t *dst;
 382     int     tmp[64];
 383     int     t0, t1, t2, t3, t4, t5, t6, t7, t8;
 384
 385 #define COMPENSATE(x) (x)
 386     src = in;
 387     dst = tmp;
 388     for (i = 0; i < 8; i++) {
 389         if (flags[i]) {
 390             IVI_INV_SLANT8(src[0], src[8], src[16], src[24], src[32], src[40], src[48], src[56],
 391                            dst[0], dst[8], dst[16], dst[24], dst[32], dst[40], dst[48], dst[56],
 392                            t0, t1, t2, t3, t4, t5, t6, t7, t8);
 393         } else
 394             dst[0] = dst[8] = dst[16] = dst[24] = dst[32] = dst[40] = dst[48] = dst[56] = 0;
 395
 396             src++;
 397             dst++;
 398     }
 399 #undef COMPENSATE
 400
 401 #define COMPENSATE(x) ((x + 1)>>1)
 402     src = tmp;
 403     for (i = 0; i < 8; i++) {
 404         if (!src[0] && !src[1] && !src[2] && !src[3] && !src[4] && !src[5] && !src[6] && !src[7]) {
 405             memset(out, 0, 8*sizeof(out[0]));
 406         } else {
 407             IVI_INV_SLANT8(src[0], src[1], src[2], src[3], src[4], src[5], src[6], src[7],
 408                            out[0], out[1], out[2], out[3], out[4], out[5], out[6], out[7],
 409                            t0, t1, t2, t3, t4, t5, t6, t7, t8);
 410         }
 411         src += 8;
 412         out += pitch;
 413     }
 414 #undef COMPENSATE
 415 }
 416
 417 void ff_ivi_inverse_slant_4x4(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
 418 {
 419     int     i;
 420     const int32_t *src;
 421     int32_t *dst;
 422     int     tmp[16];
 423     int     t0, t1, t2, t3, t4;
 424
 425 #define COMPENSATE(x) (x)
 426     src = in;
 427     dst = tmp;
 428     for (i = 0; i < 4; i++) {
 429         if (flags[i]) {
 430             IVI_INV_SLANT4(src[0], src[4], src[8], src[12],
 431                            dst[0], dst[4], dst[8], dst[12],
 432                            t0, t1, t2, t3, t4);
 433         } else
 434             dst[0] = dst[4] = dst[8] = dst[12] = 0;
 435
 436             src++;
 437             dst++;
 438     }
 439 #undef COMPENSATE
 440
 441 #define COMPENSATE(x) ((x + 1)>>1)
 442     src = tmp;
 443     for (i = 0; i < 4; i++) {
 444         if (!src[0] && !src[1] && !src[2] && !src[3]) {
 445             out[0] = out[1] = out[2] = out[3] = 0;
 446         } else {
 447             IVI_INV_SLANT4(src[0], src[1], src[2], src[3],
 448                            out[0], out[1], out[2], out[3],
 449                            t0, t1, t2, t3, t4);
 450         }
 451         src += 4;
 452         out += pitch;
 453     }
 454 #undef COMPENSATE
 455 }
 456
 457 void ff_ivi_dc_slant_2d(const int32_t *in, int16_t *out, uint32_t pitch, int blk_size)
 458 {
 459     int     x, y;
 460     int16_t dc_coeff;
 461
 462     dc_coeff = (*in + 1) >> 1;
 463
 464     for (y = 0; y < blk_size; out += pitch, y++) {
 465         for (x = 0; x < blk_size; x++)
 466             out[x] = dc_coeff;
 467     }
 468 }
 469
 470 void ff_ivi_row_slant8(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
 471 {
 472     int     i;
 473     int     t0, t1, t2, t3, t4, t5, t6, t7, t8;
 474
 475 #define COMPENSATE(x) ((x + 1)>>1)
 476     for (i = 0; i < 8; i++) {
 477         if (!in[0] && !in[1] && !in[2] && !in[3] && !in[4] && !in[5] && !in[6] && !in[7]) {
 478             memset(out, 0, 8*sizeof(out[0]));
 479         } else {
 480             IVI_INV_SLANT8( in[0],  in[1],  in[2],  in[3],  in[4],  in[5],  in[6],  in[7],
 481                            out[0], out[1], out[2], out[3], out[4], out[5], out[6], out[7],
 482                            t0, t1, t2, t3, t4, t5, t6, t7, t8);
 483         }
 484         in += 8;
 485         out += pitch;
 486     }
 487 #undef COMPENSATE
 488 }
 489
 490 void ff_ivi_dc_row_slant(const int32_t *in, int16_t *out, uint32_t pitch, int blk_size)
 491 {
 492     int     x, y;
 493     int16_t dc_coeff;
 494
 495     dc_coeff = (*in + 1) >> 1;
 496
 497     for (x = 0; x < blk_size; x++)
 498         out[x] = dc_coeff;
 499
 500     out += pitch;
 501
 502     for (y = 1; y < blk_size; out += pitch, y++) {
 503         for (x = 0; x < blk_size; x++)
 504             out[x] = 0;
 505     }
 506 }
 507
 508 void ff_ivi_col_slant8(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
 509 {
 510     int     i, row2, row4, row8;
 511     int     t0, t1, t2, t3, t4, t5, t6, t7, t8;
 512
 513     row2 = pitch << 1;
 514     row4 = pitch << 2;
 515     row8 = pitch << 3;
 516
 517 #define COMPENSATE(x) ((x + 1)>>1)
 518     for (i = 0; i < 8; i++) {
 519         if (flags[i]) {
 520             IVI_INV_SLANT8(in[0], in[8], in[16], in[24], in[32], in[40], in[48], in[56],
 521                            out[0], out[pitch], out[row2], out[row2 + pitch], out[row4],
 522                            out[row4 + pitch],  out[row4 + row2], out[row8 - pitch],
 523                            t0, t1, t2, t3, t4, t5, t6, t7, t8);
 524         } else {
 525             out[0] = out[pitch] = out[row2] = out[row2 + pitch] = out[row4] =
 526             out[row4 + pitch] =  out[row4 + row2] = out[row8 - pitch] = 0;
 527         }
 528
 529         in++;
 530         out++;
 531     }
 532 #undef COMPENSATE
 533 }
 534
 535 void ff_ivi_dc_col_slant(const int32_t *in, int16_t *out, uint32_t pitch, int blk_size)
 536 {
 537     int     x, y;
 538     int16_t dc_coeff;
 539
 540     dc_coeff = (*in + 1) >> 1;
 541
 542     for (y = 0; y < blk_size; out += pitch, y++) {
 543         out[0] = dc_coeff;
 544         for (x = 1; x < blk_size; x++)
 545             out[x] = 0;
 546     }
 547 }
 548
 549 void ff_ivi_put_pixels_8x8(const int32_t *in, int16_t *out, uint32_t pitch,
 550                            const uint8_t *flags)
 551 {
 552     int     x, y;
 553
 554     for (y = 0; y < 8; out += pitch, in += 8, y++)
 555         for (x = 0; x < 8; x++)
 556             out[x] = in[x];
 557 }
 558
 559 void ff_ivi_put_dc_pixel_8x8(const int32_t *in, int16_t *out, uint32_t pitch,
 560                              int blk_size)
 561 {
 562     int     y;
 563
 564     out[0] = in[0];
 565     memset(out + 1, 0, 7*sizeof(out[0]));
 566     out += pitch;
 567
 568     for (y = 1; y < 8; out += pitch, y++)
 569         memset(out, 0, 8*sizeof(out[0]));
 570 }
 571
 572 #define IVI_MC_TEMPLATE(size, suffix, OP) \
 573 void ff_ivi_mc_ ## size ##x## size ## suffix (int16_t *buf, const int16_t *ref_buf, \
 574                                               uint32_t pitch, int mc_type) \
 575 { \
 576     int     i, j; \
 577     const int16_t *wptr; \
 578 \
 579     switch (mc_type) { \
 580     case 0: /* fullpel (no interpolation) */ \
 581         for (i = 0; i < size; i++, buf += pitch, ref_buf += pitch) { \
 582             for (j = 0; j < size; j++) {\
 583                 OP(buf[j], ref_buf[j]); \
 584             } \
 585         } \
 586         break; \
 587     case 1: /* horizontal halfpel interpolation */ \
 588         for (i = 0; i < size; i++, buf += pitch, ref_buf += pitch) \
 589             for (j = 0; j < size; j++) \
 590                 OP(buf[j], (ref_buf[j] + ref_buf[j+1]) >> 1); \
 591         break; \
 592     case 2: /* vertical halfpel interpolation */ \
 593         wptr = ref_buf + pitch; \
 594         for (i = 0; i < size; i++, buf += pitch, wptr += pitch, ref_buf += pitch) \
 595             for (j = 0; j < size; j++) \
 596                 OP(buf[j], (ref_buf[j] + wptr[j]) >> 1); \
 597         break; \
 598     case 3: /* vertical and horizontal halfpel interpolation */ \
 599         wptr = ref_buf + pitch; \
 600         for (i = 0; i < size; i++, buf += pitch, wptr += pitch, ref_buf += pitch) \
 601             for (j = 0; j < size; j++) \
 602                 OP(buf[j], (ref_buf[j] + ref_buf[j+1] + wptr[j] + wptr[j+1]) >> 2); \
 603         break; \
 604     } \
 605 } \
 606
 607 #define OP_PUT(a, b)  (a) = (b)
 608 #define OP_ADD(a, b)  (a) += (b)
 609
 610 IVI_MC_TEMPLATE(8, _no_delta, OP_PUT)
 611 IVI_MC_TEMPLATE(8, _delta,    OP_ADD)
 612 IVI_MC_TEMPLATE(4, _no_delta, OP_PUT)
 613 IVI_MC_TEMPLATE(4, _delta,    OP_ADD)