--- mbcoding.c 2003/05/03 19:11:58 1.44.2.7 +++ mbcoding.c 2003/06/09 13:52:12 1.44.2.12 @@ -1,53 +1,27 @@ - /****************************************************************************** - * * - * This file is part of XviD, a free MPEG-4 video encoder/decoder * - * * - * XviD is an implementation of a part of one or more MPEG-4 Video tools * - * as specified in ISO/IEC 14496-2 standard. Those intending to use this * - * software module in hardware or software products are advised that its * - * use may infringe existing patents or copyrights, and any such use * - * would be at such party's own risk. The original developer of this * - * software module and his/her company, and subsequent editors and their * - * companies, will have no liability for use of this software or * - * modifications or derivatives thereof. * - * * - * XviD is free software; you can redistribute it and/or modify it * - * under the terms of the GNU General Public License as published by * - * the Free Software Foundation; either version 2 of the License, or * - * (at your option) any later version. * - * * - * XviD is distributed in the hope that it will be useful, but * - * WITHOUT ANY WARRANTY; without even the implied warranty of * - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * - * GNU General Public License for more details. * - * * - * You should have received a copy of the GNU General Public License * - * along with this program; if not, write to the Free Software * - * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA * - * * - ******************************************************************************/ - - /****************************************************************************** - * * - * mbcoding.c * - * * - * Copyright (C) 2002 - Michael Militzer * - * * - * For more information visit the XviD homepage: http://www.xvid.org * - * * - ******************************************************************************/ - - /****************************************************************************** - * * - * Revision history: * - * * - * 28.10.2002 GMC support - gruel * - * 28.06.2002 added check_resync_marker() * - * 14.04.2002 bframe encoding * - * 08.03.2002 initial version; isibaar * - * * - ******************************************************************************/ - +/***************************************************************************** + * + * XVID MPEG-4 VIDEO CODEC + * - MB coding - + * + * Copyright (C) 2002 Michael Militzer + * + * This program is free software ; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation ; either version 2 of the License, or + * (at your option) any later version. + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY ; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program ; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + * + * $Id: mbcoding.c,v 1.44.2.12 2003/06/09 13:52:12 edgomez Exp $ + * + ****************************************************************************/ #include #include @@ -88,8 +62,10 @@ const int code2 = sprite_trajectory_len[len].code; const int len2 = sprite_trajectory_len[len].len; -// printf("GMC=%d Code/Len = %d / %d ",val, code,len); -// printf("Code2 / Len2 = %d / %d \n",code2,len2); +#if 0 + printf("GMC=%d Code/Len = %d / %d ",val, code,len); + printf("Code2 / Len2 = %d / %d \n",code2,len2); +#endif BitstreamPutBits(bs, code2, len2); if (len) BitstreamPutBits(bs, code, len); @@ -648,7 +624,7 @@ cbpy = pMB->cbp >> 2; - // write mcbpc + /* write mcbpc */ if (frame->coding_type == I_VOP) { mcbpc = ((pMB->mode >> 1) & 3) | ((pMB->cbp & 3) << 2); BitstreamPutBits(bs, mcbpc_intra_tab[mcbpc].code, @@ -659,24 +635,24 @@ mcbpc_inter_tab[mcbpc].len); } - // ac prediction flag + /* ac prediction flag */ if (pMB->acpred_directions[0]) BitstreamPutBits(bs, 1, 1); else BitstreamPutBits(bs, 0, 1); - // write cbpy + /* write cbpy */ BitstreamPutBits(bs, xvid_cbpy_tab[cbpy].code, xvid_cbpy_tab[cbpy].len); - // write dquant + /* write dquant */ if (pMB->mode == MODE_INTRA_Q) BitstreamPutBits(bs, DQ_VALUE2INDEX(pMB->dquant), 2); - // write interlacing + /* write interlacing */ if (frame->vol_flags & XVID_VOL_INTERLACING) { BitstreamPutBit(bs, pMB->field_dct); } - // code block coeffs + /* code block coeffs */ for (i = 0; i < 6; i++) { if (i < 4) BitstreamPutBits(bs, dcy_tab[qcoeff[i * 64 + 0] + 255].code, @@ -720,40 +696,40 @@ mcbpc = (pMB->mode & 7) | ((pMB->cbp & 3) << 3); cbpy = 15 - (pMB->cbp >> 2); - // write mcbpc + /* write mcbpc */ BitstreamPutBits(bs, mcbpc_inter_tab[mcbpc].code, mcbpc_inter_tab[mcbpc].len); if ( (frame->coding_type == S_VOP) && (pMB->mode == MODE_INTER || pMB->mode == MODE_INTER_Q) ) - BitstreamPutBit(bs, pMB->mcsel); // mcsel: '0'=local motion, '1'=GMC + BitstreamPutBit(bs, pMB->mcsel); /* mcsel: '0'=local motion, '1'=GMC */ - // write cbpy + /* write cbpy */ BitstreamPutBits(bs, xvid_cbpy_tab[cbpy].code, xvid_cbpy_tab[cbpy].len); - // write dquant + /* write dquant */ if (pMB->mode == MODE_INTER_Q) BitstreamPutBits(bs, DQ_VALUE2INDEX(pMB->dquant), 2); - // interlacing + /* interlacing */ if (frame->vol_flags & XVID_VOL_INTERLACING) { if (pMB->cbp) { BitstreamPutBit(bs, pMB->field_dct); - DPRINTF(DPRINTF_MB,"codep: field_dct: %i", pMB->field_dct); + DPRINTF(XVID_DEBUG_MB,"codep: field_dct: %i\n", pMB->field_dct); } - // if inter block, write field ME flag + /* if inter block, write field ME flag */ if (pMB->mode == MODE_INTER || pMB->mode == MODE_INTER_Q) { BitstreamPutBit(bs, pMB->field_pred); - DPRINTF(DPRINTF_MB,"codep: field_pred: %i", pMB->field_pred); + DPRINTF(XVID_DEBUG_MB,"codep: field_pred: %i\n", pMB->field_pred); - // write field prediction references + /* write field prediction references */ if (pMB->field_pred) { BitstreamPutBit(bs, pMB->field_for_top); BitstreamPutBit(bs, pMB->field_for_bot); } } } - // code motion vector(s) if motion is local + /* code motion vector(s) if motion is local */ if (!pMB->mcsel) for (i = 0; i < (pMB->mode == MODE_INTER4V ? 4 : 1); i++) { CodeVector(bs, pMB->pmvs[i].x, frame->fcode, pStat); @@ -762,7 +738,7 @@ bits = BitstreamPos(bs); - // code block coeffs + /* code block coeffs */ for (i = 0; i < 6; i++) if (pMB->cbp & (1 << (5 - i))) { @@ -790,7 +766,7 @@ Statistics * pStat) { if (frame->coding_type != I_VOP) - BitstreamPutBit(bs, 0); // not_coded + BitstreamPutBit(bs, 0); /* not_coded */ if (pMB->mode == MODE_INTRA || pMB->mode == MODE_INTRA_Q) CodeBlockIntra(frame, pMB, qcoeff, bs, pStat); @@ -799,14 +775,15 @@ } -/* -// moved to mbcoding.h so that in can be 'static __inline' + +/* moved to mbcoding.h so that in can be 'static __inline' */ +#if 0 void MBSkip(Bitstream * bs) { - BitstreamPutBit(bs, 1); // not coded + BitstreamPutBit(bs, 1); /* not coded */ } -*/ +#endif /*************************************************************** * bframe encoding start @@ -864,7 +841,7 @@ BitstreamPutBit(bs, 1); return; - default:; // invalid + default:; /* invalid */ } } @@ -888,16 +865,16 @@ ------------------------------------------------------------------ */ if (mb->mode == MODE_DIRECT_NONE_MV) { - BitstreamPutBit(bs, 1); // skipped + BitstreamPutBit(bs, 1); /* skipped */ return; } - BitstreamPutBit(bs, 0); // not skipped + BitstreamPutBit(bs, 0); /* not skipped */ if (mb->cbp == 0) { - BitstreamPutBit(bs, 1); // cbp == 0 + BitstreamPutBit(bs, 1); /* cbp == 0 */ } else { - BitstreamPutBit(bs, 0); // cbp == xxx + BitstreamPutBit(bs, 0); /* cbp == xxx */ } put_bvop_mbtype(bs, mb->mode); @@ -907,12 +884,12 @@ } if (mb->mode != MODE_DIRECT && mb->cbp != 0) { - put_bvop_dbquant(bs, 0); // todo: mb->dquant = 0 + put_bvop_dbquant(bs, 0); /* todo: mb->dquant = 0 */ } switch (mb->mode) { case MODE_INTERPOLATE: - CodeVector(bs, mb->pmvs[1].x, vcode, pStat); //forward vector of interpolate mode + CodeVector(bs, mb->pmvs[1].x, vcode, pStat); /* forward vector of interpolate mode */ CodeVector(bs, mb->pmvs[1].y, vcode, pStat); case MODE_BACKWARD: vcode = bcode; @@ -921,8 +898,8 @@ CodeVector(bs, mb->pmvs[0].y, vcode, pStat); break; case MODE_DIRECT: - CodeVector(bs, mb->pmvs[3].x, 1, pStat); // fcode is always 1 for delta vector - CodeVector(bs, mb->pmvs[3].y, 1, pStat); // prediction is always (0,0) + CodeVector(bs, mb->pmvs[3].x, 1, pStat); /* fcode is always 1 for delta vector */ + CodeVector(bs, mb->pmvs[3].y, 1, pStat); /* prediction is always (0,0) */ default: break; } @@ -944,10 +921,12 @@ ***************************************************************/ -// for IVOP addbits == 0 -// for PVOP addbits == fcode - 1 -// for BVOP addbits == max(fcode,bcode) - 1 -// returns true or false +/* + * for IVOP addbits == 0 + * for PVOP addbits == fcode - 1 + * for BVOP addbits == max(fcode,bcode) - 1 + * returns true or false + */ int check_resync_marker(Bitstream * bs, int addbits) { @@ -1162,7 +1141,7 @@ level = BitstreamGetBits(bs, 8); if (level == 0 || level == 128) - DPRINTF(DPRINTF_ERROR, "Illegal LEVEL for ESCAPE mode 4: %d", level); + DPRINTF(XVID_DEBUG_ERROR, "Illegal LEVEL for ESCAPE mode 4: %d\n", level); return (level << 24) >> 24; } @@ -1218,17 +1197,19 @@ do { level = get_coeff(bs, &run, &last, 1, 0); if (run == -1) { - DPRINTF(DPRINTF_ERROR,"fatal: invalid run"); + DPRINTF(XVID_DEBUG_ERROR,"fatal: invalid run"); break; } coeff += run; block[scan[coeff]] = level; - DPRINTF(DPRINTF_COEFF,"block[%i] %i", scan[coeff], level); - //DPRINTF(DPRINTF_COEFF,"block[%i] %i %08x", scan[coeff], level, BitstreamShowBits(bs, 32)); + DPRINTF(XVID_DEBUG_COEFF,"block[%i] %i\n", scan[coeff], level); +#if 0 + DPRINTF(XVID_DEBUG_COEFF,"block[%i] %i %08x\n", scan[coeff], level, BitstreamShowBits(bs, 32)); +#endif if (level < -2047 || level > 2047) { - DPRINTF(DPRINTF_ERROR,"warning: intra_overflow %i", level); + DPRINTF(XVID_DEBUG_ERROR,"warning: intra_overflow %i\n", level); } coeff++; } while (!last); @@ -1251,18 +1232,18 @@ do { level = get_coeff(bs, &run, &last, 0, 0); if (run == -1) { - DPRINTF(DPRINTF_ERROR,"fatal: invalid run"); + DPRINTF(XVID_DEBUG_ERROR,"fatal: invalid run"); break; } p += run; block[scan[p]] = level; - DPRINTF(DPRINTF_COEFF,"block[%i] %i", scan[p], level); - // DPRINTF(DPRINTF_COEFF,"block[%i] %i %08x", scan[p], level, BitstreamShowBits(bs, 32)); + DPRINTF(XVID_DEBUG_COEFF,"block[%i] %i\n", scan[p], level); + /* DPRINTF(XVID_DEBUG_COEFF,"block[%i] %i %08x\n", scan[p], level, BitstreamShowBits(bs, 32)); */ if (level < -2047 || level > 2047) { - DPRINTF(DPRINTF_ERROR,"warning: inter overflow %i", level); + DPRINTF(XVID_DEBUG_ERROR,"warning: inter overflow %i\n", level); } p++; } while (!last); @@ -1270,287 +1251,6 @@ } - - - - - -/************************************************************************ - * Trellis based R-D optimal quantization * - * not really "bitstream" or "mbcoding" related, but needs VLC tables * - * * - ************************************************************************/ - - -int __inline -RunLevel_CalcBits_inter(const int16_t run, int16_t level) -{ - const int esc_length = 30; - - if (!((level+32) & -64)) - return coeff_VLC[0][0][level+32][run].len; - else - return esc_length; -} - -int __inline -RunLevelLast_CalcBits_inter(const int16_t run, const int16_t level) -{ - const int esc_length = 30; - - if (!((level+32) & -64)) - return coeff_VLC[0][1][level+32][run].len; - else - return esc_length; -} - - -int __inline -RunLevel_CalcBits_intra(const int16_t run, int16_t level) -{ - const int esc_length = 30; - int bits; - - level = abs(level); - if (!(level & -64)) { - bits = coeff_VLC[1][0][level][run].len; - if (bits!=128) - return bits; - } - return esc_length; -} - -int __inline -RunLevelLast_CalcBits_intra(const int16_t run, int16_t level) -{ - const int esc_length = 30; - int bits; - - level = abs(level); - if (!(level & -64)) { - bits = coeff_VLC[1][1][level][run].len; - if (bits!=128) - return bits; - } - return esc_length; -} - -/* based on ffmpeg's trellis quant, thanks! */ -/* (C) 2003 Michael Niedermayer */ - -int -dct_quantize_trellis_inter_h263_c (int16_t *qcoeff, const int16_t *data, int quant) -{ - -/* input: original quantized DCT coefficients (to calc distorion)*/ -/* already quantized DCT coefficients */ -/* quantizer */ -/* output: modified table of quantized DCT coefficients */ - -/* maybe combining quantize&Trellis would be faster (even that it disables MMX quant) */ - - int run_tab[65]; - int level_tab[65]; - int score_tab[65]; - int last_run = 0; - int last_level = 0; - int last_score = 0; - int last_i = 0; - int coeff[64]; - int coeff_count[64]; /* is a table useful for this 0-1 (or 1-2) table? */ - int last_non_zero, i; - - const uint16_t *const zigzag = &scan_tables[0][0]; - /* ordinary zigzag order, so it's not INTERLACE compatible, yet */ - - const int qmul = 2*quant; - const int qadd = ((quant-1)|1); - -/* quant is not needed anymore after this */ - - int score_limit = 0; - int left_limit = 0; - - const int lambda = (quant * quant * 123 + 64) >> 7; // default lagrangian - -/* control lambda through a ENVIRONMENT variable (for automatic optmization) */ - -/* - const int lfact=123; // better control of the lagrangian lambda - int lambda = (quant * quant * 123 + 64) >> 7; // default lagrangian - - const char * const trellis_lambda = getenv("TRELLIS_LAMBDA"); - if(trellis_lambda) - lfact = atoi(trellis_lambda); - if (lfact < 1) - lfact = 123; // why this value? Who knows? But 123 seems better than 109 = 0.85<<7 - - lambda = (quant * quant * lfact + 64) >> 7; // lagrangian -*/ - - last_non_zero = -1; - for (i = 0; i < 64; i++) - { - const int level = qcoeff[zigzag[i]]; - - if (level) { - last_non_zero = i; - - if (level>0) { - if (level==1) { - coeff[i] = 1; - coeff_count[i] = 0; - } else { - coeff[i] = level; - coeff_count[i] = 1; - } - } else { - if (level==-1) { - coeff[i] = -1; - coeff_count[i] = 0; - } else { - coeff[i] = level+1; // because we check coeff[i] and coeff[i]-1 - coeff_count[i] = 1; - } - } - } else { - coeff[i] = ((data[zigzag[i]]>>31)|1); /* +- 1 because of gap */ - coeff_count[i] = 0; - } - } - - if (last_non_zero < 0) - return last_non_zero; - - score_tab[0] = 0; - - for (i = 0; i <= last_non_zero; i++) { - int level, run, j; - const int dct_coeff = data[zigzag[i]]; - const int zero_distortion = dct_coeff * dct_coeff; - int best_score = 256 * 256 * 256 * 120; - - int distortion; - int dequant_err; - - last_score += zero_distortion; - - -/****************** level loop unrolled: first check coeff[i] *********/ - level = coeff[i]; - - if (level > 0) // coeff[i]==0 is not possible here - dequant_err = level * qmul + qadd - dct_coeff; - else - dequant_err = level * qmul - qadd - dct_coeff; - - distortion = dequant_err*dequant_err; - - for (run = 0; run <= i - left_limit; run++) { - - int score = distortion + lambda*RunLevel_CalcBits_inter(run, level) + score_tab[i - run]; - - if (score < best_score) - { - best_score = score_tab[i + 1] = score; - run_tab[i + 1] = run; - level_tab[i + 1] = level; - } - } - - for (run = 0; run <= i - left_limit; run++) { - int score = distortion + lambda*RunLevelLast_CalcBits_inter(run, level) + score_tab[i - run]; - - if (score < last_score) - { - last_score = score; - last_run = run; - last_level = level; - last_i = i + 1; - } - } - -/****************** level loop unrolled: if possible, check coeff[i]-1 *********/ - - if (coeff_count[i]) { - - level--; - dequant_err -= qmul; - distortion = dequant_err*dequant_err; - - for (run = 0; run <= i - left_limit; run++) { - int score = distortion + lambda*RunLevel_CalcBits_inter(run, level) + score_tab[i-run]; - - if (score < best_score) - { - best_score = score_tab[i + 1] = score; - run_tab[i + 1] = run; - level_tab[i + 1] = level; - } - } - - for (run = 0; run <= i - left_limit; run++) { - int score = distortion + lambda*RunLevelLast_CalcBits_inter(run, level) + score_tab[i-run]; - - if (score < last_score) - { - last_score = score; - last_run = run; - last_level = level; - last_i = i + 1; - } - - } - } // of check coeff[i]-1 - - -/****************** checking coeff[i]-2 doesn't isn't supported *********/ - -/****************** add distorsion for higher RUN (-> coeff[i]==0) *******/ - for (j = left_limit; j <= i; j++) - score_tab[j] += zero_distortion; - - score_limit += zero_distortion; - - if (score_tab[i + 1] < score_limit) - score_limit = score_tab[i + 1]; - - // there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level - // so we finalize only if we have no chance of getting lower than score_limit + 1*lambda anymore - - while (score_tab[left_limit] > score_limit + lambda) - left_limit++; - - - } // end of (i=0;i<=last_non_zero;i++) - - last_non_zero = last_i - 1; - if (last_non_zero < 0) - return last_non_zero; - - i = last_i; - - memset(qcoeff,0x00,64*sizeof(int16_t)); - - qcoeff[zigzag[last_non_zero]] = last_level; - i -= last_run + 1; - - for (; i > 0; i -= run_tab[i] + 1) - { - qcoeff[zigzag[i-1]] = level_tab[i]; - } - - return last_non_zero; -} - -int -dct_quantize_trellis_inter_mpeg_c (int16_t *qcoeff, const int16_t *data, int quant) -{ return 64; } - - - - - /***************************************************************************** * VLC tables and other constant arrays ****************************************************************************/