FFmpeg: libavcodec/ivi_dsp.c Source File

00001 /*
00002  * DSP functions for Indeo Video Interactive codecs (Indeo4 and Indeo5)
00003  *
00004  * Copyright (c) 2009-2011 Maxim Poliakovski
00005  *
00006  * This file is part of FFmpeg.
00007  *
00008  * FFmpeg is free software; you can redistribute it and/or
00009  * modify it under the terms of the GNU Lesser General Public
00010  * License as published by the Free Software Foundation; either
00011  * version 2.1 of the License, or (at your option) any later version.
00012  *
00013  * FFmpeg is distributed in the hope that it will be useful,
00014  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00015  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00016  * Lesser General Public License for more details.
00017  *
00018  * You should have received a copy of the GNU Lesser General Public
00019  * License along with FFmpeg; if not, write to the Free Software
00020  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00021  */
00022 
00029 #include "avcodec.h"
00030 #include "dsputil.h"
00031 #include "dwt.h"
00032 #include "ivi_common.h"
00033 #include "ivi_dsp.h"
00034 
00035 void ff_ivi_recompose53(const IVIPlaneDesc *plane, uint8_t *dst,
00036                         const int dst_pitch, const int num_bands)
00037 {
00038     int             x, y, indx;
00039     int32_t         p0, p1, p2, p3, tmp0, tmp1, tmp2;
00040     int32_t         b0_1, b0_2, b1_1, b1_2, b1_3, b2_1, b2_2, b2_3, b2_4, b2_5, b2_6;
00041     int32_t         b3_1, b3_2, b3_3, b3_4, b3_5, b3_6, b3_7, b3_8, b3_9;
00042     int32_t         pitch, back_pitch;
00043     const IDWTELEM *b0_ptr, *b1_ptr, *b2_ptr, *b3_ptr;
00044 
00045     /* all bands should have the same pitch */
00046     pitch = plane->bands[0].pitch;
00047 
00048     /* pixels at the position "y-1" will be set to pixels at the "y" for the 1st iteration */
00049     back_pitch = 0;
00050 
00051     /* get pointers to the wavelet bands */
00052     b0_ptr = plane->bands[0].buf;
00053     b1_ptr = plane->bands[1].buf;
00054     b2_ptr = plane->bands[2].buf;
00055     b3_ptr = plane->bands[3].buf;
00056 
00057     for (y = 0; y < plane->height; y += 2) {
00058 
00059         if (y+2 >= plane->height)
00060             pitch= 0;
00061         /* load storage variables with values */
00062         if (num_bands > 0) {
00063             b0_1 = b0_ptr[0];
00064             b0_2 = b0_ptr[pitch];
00065         }
00066 
00067         if (num_bands > 1) {
00068             b1_1 = b1_ptr[back_pitch];
00069             b1_2 = b1_ptr[0];
00070             b1_3 = b1_1 - b1_2*6 + b1_ptr[pitch];
00071         }
00072 
00073         if (num_bands > 2) {
00074             b2_2 = b2_ptr[0];     // b2[x,  y  ]
00075             b2_3 = b2_2;          // b2[x+1,y  ] = b2[x,y]
00076             b2_5 = b2_ptr[pitch]; // b2[x  ,y+1]
00077             b2_6 = b2_5;          // b2[x+1,y+1] = b2[x,y+1]
00078         }
00079 
00080         if (num_bands > 3) {
00081             b3_2 = b3_ptr[back_pitch]; // b3[x  ,y-1]
00082             b3_3 = b3_2;               // b3[x+1,y-1] = b3[x  ,y-1]
00083             b3_5 = b3_ptr[0];          // b3[x  ,y  ]
00084             b3_6 = b3_5;               // b3[x+1,y  ] = b3[x  ,y  ]
00085             b3_8 = b3_2 - b3_5*6 + b3_ptr[pitch];
00086             b3_9 = b3_8;
00087         }
00088 
00089         for (x = 0, indx = 0; x < plane->width; x+=2, indx++) {
00090             if (x+2 >= plane->width) {
00091                 b0_ptr --;
00092                 b1_ptr --;
00093                 b2_ptr --;
00094                 b3_ptr --;
00095             }
00096 
00097             /* some values calculated in the previous iterations can */
00098             /* be reused in the next ones, so do appropriate copying */
00099             b2_1 = b2_2; // b2[x-1,y  ] = b2[x,  y  ]
00100             b2_2 = b2_3; // b2[x  ,y  ] = b2[x+1,y  ]
00101             b2_4 = b2_5; // b2[x-1,y+1] = b2[x  ,y+1]
00102             b2_5 = b2_6; // b2[x  ,y+1] = b2[x+1,y+1]
00103             b3_1 = b3_2; // b3[x-1,y-1] = b3[x  ,y-1]
00104             b3_2 = b3_3; // b3[x  ,y-1] = b3[x+1,y-1]
00105             b3_4 = b3_5; // b3[x-1,y  ] = b3[x  ,y  ]
00106             b3_5 = b3_6; // b3[x  ,y  ] = b3[x+1,y  ]
00107             b3_7 = b3_8; // vert_HPF(x-1)
00108             b3_8 = b3_9; // vert_HPF(x  )
00109 
00110             p0 = p1 = p2 = p3 = 0;
00111 
00112             /* process the LL-band by applying LPF both vertically and horizontally */
00113             if (num_bands > 0) {
00114                 tmp0 = b0_1;
00115                 tmp2 = b0_2;
00116                 b0_1 = b0_ptr[indx+1];
00117                 b0_2 = b0_ptr[pitch+indx+1];
00118                 tmp1 = tmp0 + b0_1;
00119 
00120                 p0 =  tmp0 << 4;
00121                 p1 =  tmp1 << 3;
00122                 p2 = (tmp0 + tmp2) << 3;
00123                 p3 = (tmp1 + tmp2 + b0_2) << 2;
00124             }
00125 
00126             /* process the HL-band by applying HPF vertically and LPF horizontally */
00127             if (num_bands > 1) {
00128                 tmp0 = b1_2;
00129                 tmp1 = b1_1;
00130                 b1_2 = b1_ptr[indx+1];
00131                 b1_1 = b1_ptr[back_pitch+indx+1];
00132 
00133                 tmp2 = tmp1 - tmp0*6 + b1_3;
00134                 b1_3 = b1_1 - b1_2*6 + b1_ptr[pitch+indx+1];
00135 
00136                 p0 += (tmp0 + tmp1) << 3;
00137                 p1 += (tmp0 + tmp1 + b1_1 + b1_2) << 2;
00138                 p2 +=  tmp2 << 2;
00139                 p3 += (tmp2 + b1_3) << 1;
00140             }
00141 
00142             /* process the LH-band by applying LPF vertically and HPF horizontally */
00143             if (num_bands > 2) {
00144                 b2_3 = b2_ptr[indx+1];
00145                 b2_6 = b2_ptr[pitch+indx+1];
00146 
00147                 tmp0 = b2_1 + b2_2;
00148                 tmp1 = b2_1 - b2_2*6 + b2_3;
00149 
00150                 p0 += tmp0 << 3;
00151                 p1 += tmp1 << 2;
00152                 p2 += (tmp0 + b2_4 + b2_5) << 2;
00153                 p3 += (tmp1 + b2_4 - b2_5*6 + b2_6) << 1;
00154             }
00155 
00156             /* process the HH-band by applying HPF both vertically and horizontally */
00157             if (num_bands > 3) {
00158                 b3_6 = b3_ptr[indx+1];            // b3[x+1,y  ]
00159                 b3_3 = b3_ptr[back_pitch+indx+1]; // b3[x+1,y-1]
00160 
00161                 tmp0 = b3_1 + b3_4;
00162                 tmp1 = b3_2 + b3_5;
00163                 tmp2 = b3_3 + b3_6;
00164 
00165                 b3_9 = b3_3 - b3_6*6 + b3_ptr[pitch+indx+1];
00166 
00167                 p0 += (tmp0 + tmp1) << 2;
00168                 p1 += (tmp0 - tmp1*6 + tmp2) << 1;
00169                 p2 += (b3_7 + b3_8) << 1;
00170                 p3 +=  b3_7 - b3_8*6 + b3_9;
00171             }
00172 
00173             /* output four pixels */
00174             dst[x]             = av_clip_uint8((p0 >> 6) + 128);
00175             dst[x+1]           = av_clip_uint8((p1 >> 6) + 128);
00176             dst[dst_pitch+x]   = av_clip_uint8((p2 >> 6) + 128);
00177             dst[dst_pitch+x+1] = av_clip_uint8((p3 >> 6) + 128);
00178         }// for x
00179 
00180         dst += dst_pitch << 1;
00181 
00182         back_pitch = -pitch;
00183 
00184         b0_ptr += pitch + 1;
00185         b1_ptr += pitch + 1;
00186         b2_ptr += pitch + 1;
00187         b3_ptr += pitch + 1;
00188     }
00189 }
00190 
00191 void ff_ivi_recompose_haar(const IVIPlaneDesc *plane, uint8_t *dst,
00192                            const int dst_pitch, const int num_bands)
00193 {
00194     int             x, y, indx, b0, b1, b2, b3, p0, p1, p2, p3;
00195     const IDWTELEM *b0_ptr, *b1_ptr, *b2_ptr, *b3_ptr;
00196     int32_t         pitch;
00197 
00198     /* all bands should have the same pitch */
00199     pitch = plane->bands[0].pitch;
00200 
00201     /* get pointers to the wavelet bands */
00202     b0_ptr = plane->bands[0].buf;
00203     b1_ptr = plane->bands[1].buf;
00204     b2_ptr = plane->bands[2].buf;
00205     b3_ptr = plane->bands[3].buf;
00206 
00207     for (y = 0; y < plane->height; y += 2) {
00208         for (x = 0, indx = 0; x < plane->width; x += 2, indx++) {
00209             /* load coefficients */
00210             b0 = b0_ptr[indx]; //should be: b0 = (num_bands > 0) ? b0_ptr[indx] : 0;
00211             b1 = b1_ptr[indx]; //should be: b1 = (num_bands > 1) ? b1_ptr[indx] : 0;
00212             b2 = b2_ptr[indx]; //should be: b2 = (num_bands > 2) ? b2_ptr[indx] : 0;
00213             b3 = b3_ptr[indx]; //should be: b3 = (num_bands > 3) ? b3_ptr[indx] : 0;
00214 
00215             /* haar wavelet recomposition */
00216             p0 = (b0 + b1 + b2 + b3 + 2) >> 2;
00217             p1 = (b0 + b1 - b2 - b3 + 2) >> 2;
00218             p2 = (b0 - b1 + b2 - b3 + 2) >> 2;
00219             p3 = (b0 - b1 - b2 + b3 + 2) >> 2;
00220 
00221             /* bias, convert and output four pixels */
00222             dst[x]                 = av_clip_uint8(p0 + 128);
00223             dst[x + 1]             = av_clip_uint8(p1 + 128);
00224             dst[dst_pitch + x]     = av_clip_uint8(p2 + 128);
00225             dst[dst_pitch + x + 1] = av_clip_uint8(p3 + 128);
00226         }// for x
00227 
00228         dst += dst_pitch << 1;
00229 
00230         b0_ptr += pitch;
00231         b1_ptr += pitch;
00232         b2_ptr += pitch;
00233         b3_ptr += pitch;
00234     }// for y
00235 }
00236 
00238 #define IVI_HAAR_BFLY(s1, s2, o1, o2, t) \
00239     t  = (s1 - s2) >> 1;\
00240     o1 = (s1 + s2) >> 1;\
00241     o2 = t;\
00242 
00243 
00244 #define INV_HAAR8(s1, s5, s3, s7, s2, s4, s6, s8,\
00245                   d1, d2, d3, d4, d5, d6, d7, d8,\
00246                   t0, t1, t2, t3, t4, t5, t6, t7, t8) {\
00247     t1 = s1 << 1; t5 = s5 << 1;\
00248     IVI_HAAR_BFLY(t1, t5, t1, t5, t0); IVI_HAAR_BFLY(t1, s3, t1, t3, t0);\
00249     IVI_HAAR_BFLY(t5, s7, t5, t7, t0); IVI_HAAR_BFLY(t1, s2, t1, t2, t0);\
00250     IVI_HAAR_BFLY(t3, s4, t3, t4, t0); IVI_HAAR_BFLY(t5, s6, t5, t6, t0);\
00251     IVI_HAAR_BFLY(t7, s8, t7, t8, t0);\
00252     d1 = COMPENSATE(t1);\
00253     d2 = COMPENSATE(t2);\
00254     d3 = COMPENSATE(t3);\
00255     d4 = COMPENSATE(t4);\
00256     d5 = COMPENSATE(t5);\
00257     d6 = COMPENSATE(t6);\
00258     d7 = COMPENSATE(t7);\
00259     d8 = COMPENSATE(t8); }
00260 
00262 #define INV_HAAR4(s1, s3, s5, s7) {\
00263     HAAR_BFLY(s1, s5);  HAAR_BFLY(s1, s3);  HAAR_BFLY(s5, s7);\
00264     s1 = COMPENSATE(s1);\
00265     s3 = COMPENSATE(s3);\
00266     s5 = COMPENSATE(s5);\
00267     s7 = COMPENSATE(s7); }
00268 
00269 void ff_ivi_inverse_haar_8x8(const int32_t *in, int16_t *out, uint32_t pitch,
00270                              const uint8_t *flags)
00271 {
00272     int     i, shift, sp1, sp2, sp3, sp4;
00273     const int32_t *src;
00274     int32_t *dst;
00275     int     tmp[64];
00276     int     t0, t1, t2, t3, t4, t5, t6, t7, t8;
00277 
00278     /* apply the InvHaar8 to all columns */
00279 #define COMPENSATE(x) (x)
00280     src = in;
00281     dst = tmp;
00282     for (i = 0; i < 8; i++) {
00283         if (flags[i]) {
00284             /* pre-scaling */
00285             shift = !(i & 4);
00286             sp1 = src[ 0] << shift;
00287             sp2 = src[ 8] << shift;
00288             sp3 = src[16] << shift;
00289             sp4 = src[24] << shift;
00290             INV_HAAR8(    sp1,     sp2,     sp3,     sp4,
00291                       src[32], src[40], src[48], src[56],
00292                       dst[ 0], dst[ 8], dst[16], dst[24],
00293                       dst[32], dst[40], dst[48], dst[56],
00294                       t0, t1, t2, t3, t4, t5, t6, t7, t8);
00295         } else
00296             dst[ 0] = dst[ 8] = dst[16] = dst[24] =
00297             dst[32] = dst[40] = dst[48] = dst[56] = 0;
00298 
00299         src++;
00300         dst++;
00301     }
00302 #undef  COMPENSATE
00303 
00304     /* apply the InvHaar8 to all rows */
00305 #define COMPENSATE(x) (x)
00306     src = tmp;
00307     for (i = 0; i < 8; i++) {
00308         if (   !src[0] && !src[1] && !src[2] && !src[3]
00309             && !src[4] && !src[5] && !src[6] && !src[7]) {
00310             memset(out, 0, 8 * sizeof(out[0]));
00311         } else {
00312             INV_HAAR8(src[0], src[1], src[2], src[3],
00313                       src[4], src[5], src[6], src[7],
00314                       out[0], out[1], out[2], out[3],
00315                       out[4], out[5], out[6], out[7],
00316                       t0, t1, t2, t3, t4, t5, t6, t7, t8);
00317         }
00318         src += 8;
00319         out += pitch;
00320     }
00321 #undef  COMPENSATE
00322 }
00323 
00324 void ff_ivi_dc_haar_2d(const int32_t *in, int16_t *out, uint32_t pitch,
00325                        int blk_size)
00326 {
00327     int     x, y;
00328     int16_t dc_coeff;
00329 
00330     dc_coeff = (*in + 0) >> 3;
00331 
00332     for (y = 0; y < blk_size; out += pitch, y++) {
00333         for (x = 0; x < blk_size; x++)
00334             out[x] = dc_coeff;
00335     }
00336 }
00337 
00339 #define IVI_SLANT_BFLY(s1, s2, o1, o2, t) \
00340     t  = s1 - s2;\
00341     o1 = s1 + s2;\
00342     o2 = t;\
00343 
00344 
00345 #define IVI_IREFLECT(s1, s2, o1, o2, t) \
00346     t  = ((s1 + s2*2 + 2) >> 2) + s1;\
00347     o2 = ((s1*2 - s2 + 2) >> 2) - s2;\
00348     o1 = t;\
00349 
00350 
00351 #define IVI_SLANT_PART4(s1, s2, o1, o2, t) \
00352     t  = s2 + ((s1*4  - s2 + 4) >> 3);\
00353     o2 = s1 + ((-s1 - s2*4 + 4) >> 3);\
00354     o1 = t;\
00355 
00356 
00357 #define IVI_INV_SLANT8(s1, s4, s8, s5, s2, s6, s3, s7,\
00358                        d1, d2, d3, d4, d5, d6, d7, d8,\
00359                        t0, t1, t2, t3, t4, t5, t6, t7, t8) {\
00360     IVI_SLANT_PART4(s4, s5, t4, t5, t0);\
00361 \
00362     IVI_SLANT_BFLY(s1, t5, t1, t5, t0); IVI_SLANT_BFLY(s2, s6, t2, t6, t0);\
00363     IVI_SLANT_BFLY(s7, s3, t7, t3, t0); IVI_SLANT_BFLY(t4, s8, t4, t8, t0);\
00364 \
00365     IVI_SLANT_BFLY(t1, t2, t1, t2, t0); IVI_IREFLECT  (t4, t3, t4, t3, t0);\
00366     IVI_SLANT_BFLY(t5, t6, t5, t6, t0); IVI_IREFLECT  (t8, t7, t8, t7, t0);\
00367     IVI_SLANT_BFLY(t1, t4, t1, t4, t0); IVI_SLANT_BFLY(t2, t3, t2, t3, t0);\
00368     IVI_SLANT_BFLY(t5, t8, t5, t8, t0); IVI_SLANT_BFLY(t6, t7, t6, t7, t0);\
00369     d1 = COMPENSATE(t1);\
00370     d2 = COMPENSATE(t2);\
00371     d3 = COMPENSATE(t3);\
00372     d4 = COMPENSATE(t4);\
00373     d5 = COMPENSATE(t5);\
00374     d6 = COMPENSATE(t6);\
00375     d7 = COMPENSATE(t7);\
00376     d8 = COMPENSATE(t8);}
00377 
00379 #define IVI_INV_SLANT4(s1, s4, s2, s3, d1, d2, d3, d4, t0, t1, t2, t3, t4) {\
00380     IVI_SLANT_BFLY(s1, s2, t1, t2, t0); IVI_IREFLECT  (s4, s3, t4, t3, t0);\
00381 \
00382     IVI_SLANT_BFLY(t1, t4, t1, t4, t0); IVI_SLANT_BFLY(t2, t3, t2, t3, t0);\
00383     d1 = COMPENSATE(t1);\
00384     d2 = COMPENSATE(t2);\
00385     d3 = COMPENSATE(t3);\
00386     d4 = COMPENSATE(t4);}
00387 
00388 void ff_ivi_inverse_slant_8x8(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
00389 {
00390     int     i;
00391     const int32_t *src;
00392     int32_t *dst;
00393     int     tmp[64];
00394     int     t0, t1, t2, t3, t4, t5, t6, t7, t8;
00395 
00396 #define COMPENSATE(x) (x)
00397     src = in;
00398     dst = tmp;
00399     for (i = 0; i < 8; i++) {
00400         if (flags[i]) {
00401             IVI_INV_SLANT8(src[0], src[8], src[16], src[24], src[32], src[40], src[48], src[56],
00402                            dst[0], dst[8], dst[16], dst[24], dst[32], dst[40], dst[48], dst[56],
00403                            t0, t1, t2, t3, t4, t5, t6, t7, t8);
00404         } else
00405             dst[0] = dst[8] = dst[16] = dst[24] = dst[32] = dst[40] = dst[48] = dst[56] = 0;
00406 
00407             src++;
00408             dst++;
00409     }
00410 #undef COMPENSATE
00411 
00412 #define COMPENSATE(x) ((x + 1)>>1)
00413     src = tmp;
00414     for (i = 0; i < 8; i++) {
00415         if (!src[0] && !src[1] && !src[2] && !src[3] && !src[4] && !src[5] && !src[6] && !src[7]) {
00416             memset(out, 0, 8*sizeof(out[0]));
00417         } else {
00418             IVI_INV_SLANT8(src[0], src[1], src[2], src[3], src[4], src[5], src[6], src[7],
00419                            out[0], out[1], out[2], out[3], out[4], out[5], out[6], out[7],
00420                            t0, t1, t2, t3, t4, t5, t6, t7, t8);
00421         }
00422         src += 8;
00423         out += pitch;
00424     }
00425 #undef COMPENSATE
00426 }
00427 
00428 void ff_ivi_inverse_slant_4x4(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
00429 {
00430     int     i;
00431     const int32_t *src;
00432     int32_t *dst;
00433     int     tmp[16];
00434     int     t0, t1, t2, t3, t4;
00435 
00436 #define COMPENSATE(x) (x)
00437     src = in;
00438     dst = tmp;
00439     for (i = 0; i < 4; i++) {
00440         if (flags[i]) {
00441             IVI_INV_SLANT4(src[0], src[4], src[8], src[12],
00442                            dst[0], dst[4], dst[8], dst[12],
00443                            t0, t1, t2, t3, t4);
00444         } else
00445             dst[0] = dst[4] = dst[8] = dst[12] = 0;
00446 
00447             src++;
00448             dst++;
00449     }
00450 #undef COMPENSATE
00451 
00452 #define COMPENSATE(x) ((x + 1)>>1)
00453     src = tmp;
00454     for (i = 0; i < 4; i++) {
00455         if (!src[0] && !src[1] && !src[2] && !src[3]) {
00456             out[0] = out[1] = out[2] = out[3] = 0;
00457         } else {
00458             IVI_INV_SLANT4(src[0], src[1], src[2], src[3],
00459                            out[0], out[1], out[2], out[3],
00460                            t0, t1, t2, t3, t4);
00461         }
00462         src += 4;
00463         out += pitch;
00464     }
00465 #undef COMPENSATE
00466 }
00467 
00468 void ff_ivi_dc_slant_2d(const int32_t *in, int16_t *out, uint32_t pitch, int blk_size)
00469 {
00470     int     x, y;
00471     int16_t dc_coeff;
00472 
00473     dc_coeff = (*in + 1) >> 1;
00474 
00475     for (y = 0; y < blk_size; out += pitch, y++) {
00476         for (x = 0; x < blk_size; x++)
00477             out[x] = dc_coeff;
00478     }
00479 }
00480 
00481 void ff_ivi_row_slant8(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
00482 {
00483     int     i;
00484     int     t0, t1, t2, t3, t4, t5, t6, t7, t8;
00485 
00486 #define COMPENSATE(x) ((x + 1)>>1)
00487     for (i = 0; i < 8; i++) {
00488         if (!in[0] && !in[1] && !in[2] && !in[3] && !in[4] && !in[5] && !in[6] && !in[7]) {
00489             memset(out, 0, 8*sizeof(out[0]));
00490         } else {
00491             IVI_INV_SLANT8( in[0],  in[1],  in[2],  in[3],  in[4],  in[5],  in[6],  in[7],
00492                            out[0], out[1], out[2], out[3], out[4], out[5], out[6], out[7],
00493                            t0, t1, t2, t3, t4, t5, t6, t7, t8);
00494         }
00495         in += 8;
00496         out += pitch;
00497     }
00498 #undef COMPENSATE
00499 }
00500 
00501 void ff_ivi_dc_row_slant(const int32_t *in, int16_t *out, uint32_t pitch, int blk_size)
00502 {
00503     int     x, y;
00504     int16_t dc_coeff;
00505 
00506     dc_coeff = (*in + 1) >> 1;
00507 
00508     for (x = 0; x < blk_size; x++)
00509         out[x] = dc_coeff;
00510 
00511     out += pitch;
00512 
00513     for (y = 1; y < blk_size; out += pitch, y++) {
00514         for (x = 0; x < blk_size; x++)
00515             out[x] = 0;
00516     }
00517 }
00518 
00519 void ff_ivi_col_slant8(const int32_t *in, int16_t *out, uint32_t pitch, const uint8_t *flags)
00520 {
00521     int     i, row2, row4, row8;
00522     int     t0, t1, t2, t3, t4, t5, t6, t7, t8;
00523 
00524     row2 = pitch << 1;
00525     row4 = pitch << 2;
00526     row8 = pitch << 3;
00527 
00528 #define COMPENSATE(x) ((x + 1)>>1)
00529     for (i = 0; i < 8; i++) {
00530         if (flags[i]) {
00531             IVI_INV_SLANT8(in[0], in[8], in[16], in[24], in[32], in[40], in[48], in[56],
00532                            out[0], out[pitch], out[row2], out[row2 + pitch], out[row4],
00533                            out[row4 + pitch],  out[row4 + row2], out[row8 - pitch],
00534                            t0, t1, t2, t3, t4, t5, t6, t7, t8);
00535         } else {
00536             out[0] = out[pitch] = out[row2] = out[row2 + pitch] = out[row4] =
00537             out[row4 + pitch] =  out[row4 + row2] = out[row8 - pitch] = 0;
00538         }
00539 
00540         in++;
00541         out++;
00542     }
00543 #undef COMPENSATE
00544 }
00545 
00546 void ff_ivi_dc_col_slant(const int32_t *in, int16_t *out, uint32_t pitch, int blk_size)
00547 {
00548     int     x, y;
00549     int16_t dc_coeff;
00550 
00551     dc_coeff = (*in + 1) >> 1;
00552 
00553     for (y = 0; y < blk_size; out += pitch, y++) {
00554         out[0] = dc_coeff;
00555         for (x = 1; x < blk_size; x++)
00556             out[x] = 0;
00557     }
00558 }
00559 
00560 void ff_ivi_put_pixels_8x8(const int32_t *in, int16_t *out, uint32_t pitch,
00561                            const uint8_t *flags)
00562 {
00563     int     x, y;
00564 
00565     for (y = 0; y < 8; out += pitch, in += 8, y++)
00566         for (x = 0; x < 8; x++)
00567             out[x] = in[x];
00568 }
00569 
00570 void ff_ivi_put_dc_pixel_8x8(const int32_t *in, int16_t *out, uint32_t pitch,
00571                              int blk_size)
00572 {
00573     int     y;
00574 
00575     out[0] = in[0];
00576     memset(out + 1, 0, 7*sizeof(out[0]));
00577     out += pitch;
00578 
00579     for (y = 1; y < 8; out += pitch, y++)
00580         memset(out, 0, 8*sizeof(out[0]));
00581 }
00582 
00583 #define IVI_MC_TEMPLATE(size, suffix, OP) \
00584 void ff_ivi_mc_ ## size ##x## size ## suffix (int16_t *buf, const int16_t *ref_buf, \
00585                                               uint32_t pitch, int mc_type) \
00586 { \
00587     int     i, j; \
00588     const int16_t *wptr; \
00589 \
00590     switch (mc_type) { \
00591     case 0: /* fullpel (no interpolation) */ \
00592         for (i = 0; i < size; i++, buf += pitch, ref_buf += pitch) { \
00593             for (j = 0; j < size; j++) {\
00594                 OP(buf[j], ref_buf[j]); \
00595             } \
00596         } \
00597         break; \
00598     case 1: /* horizontal halfpel interpolation */ \
00599         for (i = 0; i < size; i++, buf += pitch, ref_buf += pitch) \
00600             for (j = 0; j < size; j++) \
00601                 OP(buf[j], (ref_buf[j] + ref_buf[j+1]) >> 1); \
00602         break; \
00603     case 2: /* vertical halfpel interpolation */ \
00604         wptr = ref_buf + pitch; \
00605         for (i = 0; i < size; i++, buf += pitch, wptr += pitch, ref_buf += pitch) \
00606             for (j = 0; j < size; j++) \
00607                 OP(buf[j], (ref_buf[j] + wptr[j]) >> 1); \
00608         break; \
00609     case 3: /* vertical and horizontal halfpel interpolation */ \
00610         wptr = ref_buf + pitch; \
00611         for (i = 0; i < size; i++, buf += pitch, wptr += pitch, ref_buf += pitch) \
00612             for (j = 0; j < size; j++) \
00613                 OP(buf[j], (ref_buf[j] + ref_buf[j+1] + wptr[j] + wptr[j+1]) >> 2); \
00614         break; \
00615     } \
00616 } \
00617 
00618 #define OP_PUT(a, b)  (a) = (b)
00619 #define OP_ADD(a, b)  (a) += (b)
00620 
00621 IVI_MC_TEMPLATE(8, _no_delta, OP_PUT)
00622 IVI_MC_TEMPLATE(8, _delta,    OP_ADD)
00623 IVI_MC_TEMPLATE(4, _no_delta, OP_PUT)
00624 IVI_MC_TEMPLATE(4, _delta,    OP_ADD)