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