如上一篇Quantization所说，可以在编码端通过设置offset来调整量化后的值，从而趋向于期望的量化值，而且在逆量化公式可以看出offset值在逆量化的时候是不会用到的。
目前来说，确定offset的算法有三种：static offset、around offset、trellis offset。
Static Offset
H.264参考模型建议：当帧内预测时：$f = frac{1}{3}igtriangleup$;当帧间预测时$f = frac{1}{6}igtriangleup$。这种采用固定的比例作为量化的offset。
Around Offset
量化的时候加上Offset，目的是为了通过Offset的调整，使量化能趋向于得到最优结果。那么如何才是最优？当然是对量化后level进行反量化后，得到的数值与量化前的数值保持一致。当然这是不可能的，不过我们可以对第i次的量化结果，反馈到第i+1次量化计算中。通过这种自行反馈的方式，调整量化offset，使其趋向于最优的量化结果。
Around Offset会采用当前位置的上一次量化结果对这次的量化offset进行调整。
$ M_{i+1} = M_i + weight imes ((coeff_i – level_i << Qbits_i) >> (Qbits_i + 1))$
$ f_{i+1} = M_{i+1} << Qbits_{i+1}$
//Q_offsets.c //fi+1 = Mi+1 << Qbitsi+1 static inline void update_q_offset4x4(LevelQuantParams **q_params, short *offsetList, int q_bits) { int i, j; short *p_list = &offsetList[0]; for (j = 0; j < 4; j++) { for (i = 0; i < 4; i++) { q_params[j][i].OffsetComp = (int) *p_list++ << q_bits; } } } /*! ************************************************************************ * rief * Calculation of the quantization offset parameters at the frame level * * par Input: * none * * par Output: * none ************************************************************************ */ void CalculateOffset4x4Param (VideoParameters *p_Vid) { QuantParameters *p_Quant = p_Vid->p_Quant; int k; int qp_per, qp; int img_type = ((p_Vid->type == SI_SLICE) ? I_SLICE : (p_Vid->type == SP_SLICE ? P_SLICE : p_Vid->type)); int max_qp_scale = imax(p_Vid->bitdepth_luma_qp_scale, p_Vid->bitdepth_chroma_qp_scale); int max_qp = 51 + max_qp_scale; InputParameters *p_Inp = p_Vid->p_Inp; p_Vid->AdaptRndWeight = p_Inp->AdaptRndWFactor [p_Vid->nal_reference_idc != 0][img_type]; p_Vid->AdaptRndCrWeight = p_Inp->AdaptRndCrWFactor[p_Vid->nal_reference_idc != 0][img_type]; if (img_type == I_SLICE ) { for (qp = 0; qp < max_qp + 1; qp++) { k = p_Quant->qp_per_matrix [qp]; qp_per = Q_BITS + k - OffsetBits; k = p_Inp->AdaptRoundingFixed ? 0 : qp; // Intra4x4 luma update_q_offset4x4(p_Quant->q_params_4x4[0][1][qp], p_Quant->OffsetList4x4[k][ 0], qp_per); // Intra4x4 chroma u update_q_offset4x4(p_Quant->q_params_4x4[1][1][qp], p_Quant->OffsetList4x4[k][ 1], qp_per); // Intra4x4 chroma v update_q_offset4x4(p_Quant->q_params_4x4[2][1][qp], p_Quant->OffsetList4x4[k][ 2], qp_per); } } else if (img_type == B_SLICE) { for (qp = 0; qp < max_qp + 1; qp++) { k = p_Quant->qp_per_matrix [qp]; qp_per = Q_BITS + k - OffsetBits; k = p_Inp->AdaptRoundingFixed ? 0 : qp; // Inter4x4 luma update_q_offset4x4(p_Quant->q_params_4x4[0][0][qp], p_Quant->OffsetList4x4[k][12], qp_per); // Intra4x4 luma update_q_offset4x4(p_Quant->q_params_4x4[0][1][qp], p_Quant->OffsetList4x4[k][ 6], qp_per); // Inter4x4 chroma u update_q_offset4x4(p_Quant->q_params_4x4[1][0][qp], p_Quant->OffsetList4x4[k][13], qp_per); // Intra4x4 chroma u update_q_offset4x4(p_Quant->q_params_4x4[1][1][qp], p_Quant->OffsetList4x4[k][ 7], qp_per); // Inter4x4 chroma v update_q_offset4x4(p_Quant->q_params_4x4[2][0][qp], p_Quant->OffsetList4x4[k][14], qp_per); // Intra4x4 chroma v update_q_offset4x4(p_Quant->q_params_4x4[2][1][qp], p_Quant->OffsetList4x4[k][ 8], qp_per); } } else { for (qp = 0; qp < max_qp + 1; qp++) { k = p_Quant->qp_per_matrix [qp]; qp_per = Q_BITS + k - OffsetBits; k = p_Inp->AdaptRoundingFixed ? 0 : qp; // Inter4x4 luma update_q_offset4x4(p_Quant->q_params_4x4[0][0][qp], p_Quant->OffsetList4x4[k][ 9], qp_per); // Intra4x4 luma update_q_offset4x4(p_Quant->q_params_4x4[0][1][qp], p_Quant->OffsetList4x4[k][ 3], qp_per); // Inter4x4 chroma u update_q_offset4x4(p_Quant->q_params_4x4[1][0][qp], p_Quant->OffsetList4x4[k][10], qp_per); // Intra4x4 chroma u update_q_offset4x4(p_Quant->q_params_4x4[1][1][qp], p_Quant->OffsetList4x4[k][ 4], qp_per); // Inter4x4 chroma v update_q_offset4x4(p_Quant->q_params_4x4[2][0][qp], p_Quant->OffsetList4x4[k][11], qp_per); // Intra4x4 chroma v update_q_offset4x4(p_Quant->q_params_4x4[2][1][qp], p_Quant->OffsetList4x4[k][ 5], qp_per); } } } //Q_around.c //Mi+1 = Mi + k /*! ************************************************************************ * rief * update rounding offsets based on JVT-N011 ************************************************************************ */ void update_offset_params(Macroblock *currMB, int mode, byte luma_transform_size_8x8_flag) { VideoParameters *p_Vid = currMB->p_Vid; InputParameters *p_Inp = currMB->p_Inp; int is_inter = (mode != I4MB)&&(mode != I16MB) && (mode != I8MB); int luma_pos = AdaptRndPos[(is_inter<<1) + luma_transform_size_8x8_flag][p_Vid->type]; int i,j; int qp = currMB->qp + p_Vid->bitdepth_luma_qp_scale; int cur_qp = p_Inp->AdaptRoundingFixed ? 0 : qp; int temp = 0; QuantParameters *p_Quant = p_Vid->p_Quant; int offsetRange = 1 << (OffsetBits - 1); int blk_mask = 0x03 + (luma_transform_size_8x8_flag<<2); int blk_shift = 2 + luma_transform_size_8x8_flag; short **offsetList = luma_transform_size_8x8_flag ? p_Quant->OffsetList8x8[cur_qp] : p_Quant->OffsetList4x4[cur_qp]; short *cur_offset_list = offsetList[luma_pos]; int **fAdjust = luma_transform_size_8x8_flag ? p_Vid->ARCofAdj8x8[0][mode] : p_Vid->ARCofAdj4x4[0][mode]; if (mode == IPCM) return; if( (p_Vid->active_sps->chroma_format_idc == YUV444) && (p_Inp->separate_colour_plane_flag != 0) ) { if( luma_transform_size_8x8_flag ) // 8x8 luma_pos += 5 * p_Vid->colour_plane_id; else // 4x4 luma_pos += p_Vid->colour_plane_id; cur_offset_list = offsetList[luma_pos]; } for (j=0; j < MB_BLOCK_SIZE; j++) { int j_pos = ((j & blk_mask)<<blk_shift); for (i=0; i < MB_BLOCK_SIZE; i++) { temp = j_pos + (i & blk_mask); cur_offset_list[temp] = (short) iClip3(0, offsetRange, cur_offset_list[temp] + (short) fAdjust[j][i]); } } if(p_Vid->P444_joined) { int **fAdjustCbCr; int uv; for(uv = 0; uv < 2; uv++) { luma_pos = AdaptRndPos[(is_inter<<1) + luma_transform_size_8x8_flag][p_Vid->type]; fAdjustCbCr = luma_transform_size_8x8_flag ? p_Vid->ARCofAdj8x8[uv + 1][mode] : p_Vid->ARCofAdj4x4[uv + 1][mode]; if(luma_transform_size_8x8_flag ) // 8x8 luma_pos += 5 * (uv+1); else // 4x4 luma_pos += (uv+1); cur_offset_list = offsetList[luma_pos]; for (j=0; j < MB_BLOCK_SIZE; j++) { int j_pos = ((j & blk_mask)<<blk_shift); for (i=0; i < MB_BLOCK_SIZE; i++) { temp = j_pos + (i & blk_mask); cur_offset_list[temp] = (short) iClip3(0, offsetRange, cur_offset_list[temp] + (short) fAdjustCbCr[j][i]); } } } } if ((p_Inp->AdaptRndChroma) && (p_Vid->yuv_format == YUV420 || p_Vid->yuv_format == YUV422 )) { int u_pos = AdaptRndCrPos[is_inter][p_Vid->type]; int v_pos = u_pos + 1; int k, jpos, uv = 1; for (k = u_pos; k <= v_pos; k++) { int **fAdjustChroma = (luma_transform_size_8x8_flag && mode == P8x8 )? p_Vid->ARCofAdj4x4[uv][4] : p_Vid->ARCofAdj4x4[uv][mode]; uv++; cur_offset_list = p_Quant->OffsetList4x4[cur_qp][k]; for (j = 0; j < p_Vid->mb_cr_size_y; j++) { jpos = ((j & 0x03)<<2); for (i = 0; i < p_Vid->mb_cr_size_x; i++) { temp = jpos + (i & 0x03); cur_offset_list[temp] = (short) iClip3(0, offsetRange, cur_offset_list[temp] + (short) fAdjustChroma[j][i]); } } } } } //Quant4x4_around.c //k = weight * ((coeff - level<<Qbits) >> Qbits+1) /*! ************************************************************************ * rief * Quantization process for All coefficients for a 4x4 block * ************************************************************************ */ int quant_4x4_around(Macroblock *currMB, int **tblock, struct quant_methods *q_method) { VideoParameters *p_Vid = currMB->p_Vid; QuantParameters *p_Quant = p_Vid->p_Quant; Slice *currSlice = currMB->p_Slice; Boolean is_cavlc = (Boolean) (currSlice->symbol_mode == CAVLC); int AdaptRndWeight = p_Vid->AdaptRndWeight; int block_x = q_method->block_x; int qp = q_method->qp; int* ACL = &q_method->ACLevel[0]; int* ACR = &q_method->ACRun[0]; LevelQuantParams **q_params_4x4 = q_method->q_params; const byte (*pos_scan)[2] = q_method->pos_scan; const byte *c_cost = q_method->c_cost; int *coeff_cost = q_method->coeff_cost; LevelQuantParams *q_params = NULL; int **fadjust4x4 = q_method->fadjust; int i,j, coeff_ctr; int *m7; int scaled_coeff; int level, run = 0; int nonzero = FALSE; int qp_per = p_Quant->qp_per_matrix[qp]; int q_bits = Q_BITS + qp_per; const byte *p_scan = &pos_scan[0][0]; int* padjust4x4; // Quantization for (coeff_ctr = 0; coeff_ctr < 16; ++coeff_ctr) { i = *p_scan++; // horizontal position j = *p_scan++; // vertical position padjust4x4 = &fadjust4x4[j][block_x + i]; m7 = &tblock[j][block_x + i]; if (*m7 != 0) { q_params = &q_params_4x4[j][i]; scaled_coeff = iabs (*m7) * q_params->ScaleComp; level = (scaled_coeff + q_params->OffsetComp) >> q_bits; if (level != 0) { if (is_cavlc) level = imin(level, CAVLC_LEVEL_LIMIT); *padjust4x4 = rshift_rnd_sf((AdaptRndWeight * (scaled_coeff - (level << q_bits))), q_bits + 1); *coeff_cost += (level > 1) ? MAX_VALUE : c_cost[run]; level = isignab(level, *m7); *m7 = rshift_rnd_sf(((level * q_params->InvScaleComp) << qp_per), 4); // inverse scale can be alternative performed as follows to ensure 16bit // arithmetic is satisfied. // *m7 = (qp_per<4) ? rshift_rnd_sf((level*q_params->InvScaleComp),4-qp_per) : (level*q_params->InvScaleComp)<<(qp_per-4); *ACL++ = level; *ACR++ = run; // reset zero level counter run = 0; nonzero = TRUE; } else { *padjust4x4 = 0; *m7 = 0; ++run; } } else { *padjust4x4 = 0; ++run; } } *ACL = 0; return nonzero; }
trellis offset
trellis offset其实用trellis quantization来描述更为准确，因为这种量化方式不会用到offset。Trellis就是采用Rdoq来得到最佳量化值，即取0、level还是level+1会达到最优的量化结果。由于不会用到offset，因此得到的level统一都是取下整，这样的话需要进行Rdo的候选level有三个：0、level、level+1。三个候选值还是稍微多了，可以采用以下方式进行筛选。
Rdoq当中包含Rdo这三个字母，这意味它依赖编码后的码流长度以及残差来选择最优结果，不过由于Rdoq处于编码途中，因此无法得到确切的编码后码流长度以及残差，因此只能通过预测值来，即上述候选值来进行计算。计算码流长度涉及到熵编码，而熵编码是以8x8或4x4为单位进行的，但是由于当前像素进行预测时，其后面的像素还没有进行预测，所以进行Rdoq时，当前像素之前的像素点采用的是预测后的level，而当前像素点之后的像素点采用level[num-1]的像素点。
Rdoq实际上就是对于当前像素点所在的block进行Rdo：在该block上，当前像素采用的是0、level还是level+1才能得到最优的结果。
/*! ************************************************************************ * rief * Quantization process for All coefficients for a 4x4 block * ************************************************************************ */ int quant_4x4_trellis(Macroblock *currMB, int **tblock, struct quant_methods *q_method) { int block_x = q_method->block_x; int* ACL = &q_method->ACLevel[0]; int* ACR = &q_method->ACRun[0]; Slice *currSlice = currMB->p_Slice; QuantParameters *p_Quant = currMB->p_Vid->p_Quant; int qp = q_method->qp; LevelQuantParams **q_params_4x4 = q_method->q_params; const byte (*pos_scan)[2] = q_method->pos_scan; const byte *c_cost = q_method->c_cost; int *coeff_cost = q_method->coeff_cost; Boolean is_cavlc = (Boolean) (currSlice->symbol_mode == CAVLC); int i,j, coeff_ctr; int *m7; int level, run = 0; int nonzero = FALSE; int qp_per = p_Quant->qp_per_matrix[qp]; const byte *p_scan = &pos_scan[0][0]; int levelTrellis[16]; /* rdoq_4x4 * To decide witch level to use * 0:0 1:level 2:level+1 (lowerint == 0) */ currSlice->rdoq_4x4(currMB, tblock, q_method, levelTrellis); // Quantization for (coeff_ctr = 0; coeff_ctr < 16; ++coeff_ctr) { i = *p_scan++; // horizontal position j = *p_scan++; // vertical position m7 = &tblock[j][block_x + i]; if (*m7 != 0) { /* scaled_coeff = iabs (*m7) * q_params_4x4[j][i].ScaleComp; level = (scaled_coeff + q_params_4x4[j][i].OffsetComp) >> q_bits; */ level = levelTrellis[coeff_ctr]; if (level != 0) { if (is_cavlc) level = imin(level, CAVLC_LEVEL_LIMIT); *coeff_cost += (level > 1) ? MAX_VALUE : c_cost[run]; level = isignab(level, *m7); *m7 = rshift_rnd_sf(((level * q_params_4x4[j][i].InvScaleComp) << qp_per), 4); *ACL++ = level; *ACR++ = run; // reset zero level counter run = 0; nonzero = TRUE; } else { *m7 = 0; ++run; } } else { ++run; } } *ACL = 0; return nonzero; } /*! ************************************************************************ * rief * Rate distortion optimized Quantization process for * all coefficients in a 4x4 block (CAVLC) * ************************************************************************ */ void rdoq_4x4_CAVLC(Macroblock *currMB, int **tblock, struct quant_methods *q_method, int levelTrellis[]) { VideoParameters *p_Vid = currMB->p_Vid; int block_x = q_method->block_x; int block_y = q_method->block_y; LevelQuantParams **q_params_4x4 = q_method->q_params; const byte (*pos_scan)[2] = q_method->pos_scan; const byte *p_scan = &pos_scan[0][0]; int qp = q_method->qp; QuantParameters *p_Quant = currMB->p_Vid->p_Quant; int qp_per = p_Quant->qp_per_matrix[qp]; int qp_rem = p_Quant->qp_rem_matrix[qp]; levelDataStruct levelData[16]; double lambda_md = p_Vid->lambda_rdoq[p_Vid->type][p_Vid->masterQP]; int type = LUMA_4x4; int pos_x = block_x >> BLOCK_SHIFT; int pos_y = block_y >> BLOCK_SHIFT; int b8 = 2*(pos_y >> 1) + (pos_x >> 1); int b4 = 2*(pos_y & 0x01) + (pos_x & 0x01); init_trellis_data_4x4_CAVLC(currMB, tblock, block_x, qp_per, qp_rem, q_params_4x4, p_scan, &levelData[0], type); est_RunLevel_CAVLC(currMB, levelData, levelTrellis, LUMA, b8, b4, 16, lambda_md); } /*! **************************************************************************** * rief * Initialize levelData **************************************************************************** */ void init_trellis_data_4x4_CAVLC(Macroblock *currMB, int **tblock, int block_x, int qp_per, int qp_rem, LevelQuantParams **q_params, const byte *p_scan, levelDataStruct *dataLevel, int type) { Slice *currSlice = currMB->p_Slice; int i, j, coeff_ctr; int *m7; int end_coeff_ctr = ( ( type == LUMA_4x4 ) ? 16 : 15 ); int q_bits = Q_BITS + qp_per; int q_offset = ( 1 << (q_bits - 1) ); int scaled_coeff, level, lowerInt, k; double err, estErr; for (coeff_ctr = 0; coeff_ctr < end_coeff_ctr; coeff_ctr++) { i = *p_scan++; // horizontal position j = *p_scan++; // vertical position m7 = &tblock[j][block_x + i]; if (*m7 == 0) { dataLevel->levelDouble = 0; dataLevel->level[0] = 0; dataLevel->noLevels = 1; err = 0.0; dataLevel->errLevel[0] = 0.0; dataLevel->pre_level = 0; dataLevel->sign = 0; } else { estErr = ((double) estErr4x4[qp_rem][j][i]) / currSlice->norm_factor_4x4; scaled_coeff = iabs(*m7) * q_params[j][i].ScaleComp; dataLevel->levelDouble = scaled_coeff; level = (scaled_coeff >> q_bits); lowerInt = ((scaled_coeff - (level << q_bits)) < q_offset )? 1 : 0; dataLevel->level[0] = 0; if (level == 0 && lowerInt == 1) { dataLevel->noLevels = 1; } else if (level == 0 && lowerInt == 0) { dataLevel->level[1] = 1; dataLevel->noLevels = 2; } else if (level > 0 && lowerInt == 1) { dataLevel->level[1] = level; dataLevel->noLevels = 2; } else { dataLevel->level[1] = level; dataLevel->level[2] = level + 1; dataLevel->noLevels = 3; } for (k = 0; k < dataLevel->noLevels; k++) { err = (double)(dataLevel->level[k] << q_bits) - (double)scaled_coeff; dataLevel->errLevel[k] = (err * err * estErr); } if(dataLevel->noLevels == 1) dataLevel->pre_level = 0; else dataLevel->pre_level = (iabs (*m7) * q_params[j][i].ScaleComp + q_params[j][i].OffsetComp) >> q_bits; dataLevel->sign = isign(*m7); } dataLevel++; } } /*! **************************************************************************** * rief * estimate run and level for CAVLC **************************************************************************** */ void est_RunLevel_CAVLC(Macroblock *currMB, levelDataStruct *levelData, int *levelTrellis, int block_type, int b8, int b4, int coeff_num, double lambda) { int k, lastnonzero = -1, coeff_ctr; int level_to_enc[16] = {0}, sign_to_enc[16] = {0}; int cstat, bestcstat = 0; int nz_coeff=0; double lagr, lagrAcc = 0, minlagr = 0; VideoParameters *p_Vid = currMB->p_Vid; int subblock_x = ((b8 & 0x1) == 0) ? (((b4 & 0x1) == 0) ? 0 : 1) : (((b4 & 0x1) == 0) ? 2 : 3); // horiz. position for coeff_count context int subblock_y = (b8 < 2) ? ((b4 < 2) ? 0 : 1) :((b4 < 2) ? 2 : 3); // vert. position for coeff_count context int nnz; levelDataStruct *dataLevel = &levelData[0]; if (block_type != CHROMA_AC) nnz = predict_nnz(currMB, LUMA, subblock_x, subblock_y); else nnz = predict_nnz_chroma(currMB, currMB->subblock_x >> 2, (currMB->subblock_y >> 2) + 4); for (coeff_ctr=0;coeff_ctr < coeff_num;coeff_ctr++) { levelTrellis[coeff_ctr] = 0; for(k=0; k < dataLevel->noLevels; k++) { dataLevel->errLevel[k] /= 32768; } lagrAcc += dataLevel->errLevel[imax(0, dataLevel->noLevels - 1)]; level_to_enc[coeff_ctr] = dataLevel->pre_level; sign_to_enc[coeff_ctr] = dataLevel->sign; if(dataLevel->noLevels > 1) { dataLevel->coeff_ctr = coeff_ctr; lastnonzero = coeff_ctr; } else dataLevel->coeff_ctr = -1; dataLevel++; } if(lastnonzero != -1) { //sort the coefficients based on their absolute value qsort(levelData, lastnonzero + 1, sizeof(levelDataStruct), cmp); dataLevel = &levelData[lastnonzero]; for(coeff_ctr = lastnonzero; coeff_ctr >= 0; coeff_ctr--) // go over all coeff { if(dataLevel->noLevels == 1) { dataLevel--; continue; } lagrAcc -= dataLevel->errLevel[dataLevel->noLevels-1]; for(cstat=0; cstat<dataLevel->noLevels; cstat++) // go over all states of cur coeff k { level_to_enc[dataLevel->coeff_ctr] = dataLevel->level[cstat]; lagr = lagrAcc + dataLevel->errLevel[cstat]; lagr += lambda * est_CAVLC_bits( p_Vid, level_to_enc, sign_to_enc, nnz, block_type); if(cstat==0 || lagr<minlagr) { minlagr = lagr; bestcstat = cstat; } } lagrAcc += dataLevel->errLevel[bestcstat]; level_to_enc[dataLevel->coeff_ctr] = dataLevel->level[bestcstat]; dataLevel--; } for(coeff_ctr = 0; coeff_ctr <= lastnonzero; coeff_ctr++) { levelTrellis[coeff_ctr] = level_to_enc[coeff_ctr]; if (level_to_enc[coeff_ctr] != 0) nz_coeff++; } } p_Vid->nz_coeff [p_Vid->current_mb_nr ][subblock_x][subblock_y] = nz_coeff; } /*! ************************************************************************ * rief * estimate CAVLC bits ************************************************************************ */ int est_CAVLC_bits (VideoParameters *p_Vid, int level_to_enc[16], int sign_to_enc[16], int nnz, int block_type) { int no_bits = 0; SyntaxElement se; int coeff_ctr, scan_pos = 0; int k, level = 1, run = -1, vlcnum; int numcoeff = 0, lastcoeff = 0, numtrailingones = 0; int numones = 0, totzeros = 0, zerosleft, numcoef; int numcoeff_vlc; int level_two_or_higher; int max_coeff_num = 0, cdc = (block_type == CHROMA_DC ? 1 : 0); int yuv = p_Vid->yuv_format - 1; static const int incVlc[] = {0, 3, 6, 12, 24, 48, 32768}; // maximum vlc = 6 int pLevel[16] = {0}; int pRun[16] = {0}; static const int Token_lentab[3][4][17] = { { { 1, 6, 8, 9,10,11,13,13,13,14,14,15,15,16,16,16,16}, { 0, 2, 6, 8, 9,10,11,13,13,14,14,15,15,15,16,16,16}, { 0, 0, 3, 7, 8, 9,10,11,13,13,14,14,15,15,16,16,16}, { 0, 0, 0, 5, 6, 7, 8, 9,10,11,13,14,14,15,15,16,16} }, { { 2, 6, 6, 7, 8, 8, 9,11,11,12,12,12,13,13,13,14,14}, { 0, 2, 5, 6, 6, 7, 8, 9,11,11,12,12,13,13,14,14,14}, { 0, 0, 3, 6, 6, 7, 8, 9,11,11,12,12,13,13,13,14,14}, { 0, 0, 0, 4, 4, 5, 6, 6, 7, 9,11,11,12,13,13,13,14} }, { { 4, 6, 6, 6, 7, 7, 7, 7, 8, 8, 9, 9, 9,10,10,10,10}, { 0, 4, 5, 5, 5, 5, 6, 6, 7, 8, 8, 9, 9, 9,10,10,10}, { 0, 0, 4, 5, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9,10,10,10}, { 0, 0, 0, 4, 4, 4, 4, 4, 5, 6, 7, 8, 8, 9,10,10,10} } }; static const int Totalzeros_lentab[TOTRUN_NUM][16] = { { 1,3,3,4,4,5,5,6,6,7,7,8,8,9,9,9}, { 3,3,3,3,3,4,4,4,4,5,5,6,6,6,6}, { 4,3,3,3,4,4,3,3,4,5,5,6,5,6}, { 5,3,4,4,3,3,3,4,3,4,5,5,5}, { 4,4,4,3,3,3,3,3,4,5,4,5}, { 6,5,3,3,3,3,3,3,4,3,6}, { 6,5,3,3,3,2,3,4,3,6}, { 6,4,5,3,2,2,3,3,6}, { 6,6,4,2,2,3,2,5}, { 5,5,3,2,2,2,4}, { 4,4,3,3,1,3}, { 4,4,2,1,3}, { 3,3,1,2}, { 2,2,1}, { 1,1}, }; static const int Run_lentab[TOTRUN_NUM][16] = { {1,1}, {1,2,2}, {2,2,2,2}, {2,2,2,3,3}, {2,2,3,3,3,3}, {2,3,3,3,3,3,3}, {3,3,3,3,3,3,3,4,5,6,7,8,9,10,11}, }; static const int Token_lentab_cdc[3][4][17] = { //YUV420 {{ 2, 6, 6, 6, 6, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, { 0, 1, 6, 7, 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, { 0, 0, 3, 7, 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, { 0, 0, 0, 6, 7, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}}, //YUV422 {{ 1, 7, 7, 9, 9,10,11,12,13, 0, 0, 0, 0, 0, 0, 0, 0}, { 0, 2, 7, 7, 9,10,11,12,12, 0, 0, 0, 0, 0, 0, 0, 0}, { 0, 0, 3, 7, 7, 9,10,11,12, 0, 0, 0, 0, 0, 0, 0, 0}, { 0, 0, 0, 5, 6, 7, 7,10,11, 0, 0, 0, 0, 0, 0, 0, 0}}, //YUV444 {{ 1, 6, 8, 9,10,11,13,13,13,14,14,15,15,16,16,16,16}, { 0, 2, 6, 8, 9,10,11,13,13,14,14,15,15,15,16,16,16}, { 0, 0, 3, 7, 8, 9,10,11,13,13,14,14,15,15,16,16,16}, { 0, 0, 0, 5, 6, 7, 8, 9,10,11,13,14,14,15,15,16,16}} }; static const int Totalzeros_lentab_cdc[3][TOTRUN_NUM][16] = { //YUV420 {{ 1,2,3,3}, { 1,2,2}, { 1,1}}, //YUV422 {{ 1,3,3,4,4,4,5,5}, { 3,2,3,3,3,3,3}, { 3,3,2,2,3,3}, { 3,2,2,2,3}, { 2,2,2,2}, { 2,2,1}, { 1,1}}, //YUV444 {{ 1,3,3,4,4,5,5,6,6,7,7,8,8,9,9,9}, { 3,3,3,3,3,4,4,4,4,5,5,6,6,6,6}, { 4,3,3,3,4,4,3,3,4,5,5,6,5,6}, { 5,3,4,4,3,3,3,4,3,4,5,5,5}, { 4,4,4,3,3,3,3,3,4,5,4,5}, { 6,5,3,3,3,3,3,3,4,3,6}, { 6,5,3,3,3,2,3,4,3,6}, { 6,4,5,3,2,2,3,3,6}, { 6,6,4,2,2,3,2,5}, { 5,5,3,2,2,2,4}, { 4,4,3,3,1,3}, { 4,4,2,1,3}, { 3,3,1,2}, { 2,2,1}, { 1,1}} }; max_coeff_num = ( (block_type == CHROMA_DC) ? p_Vid->num_cdc_coeff : ( (block_type == LUMA_INTRA16x16AC || block_type == CB_INTRA16x16AC || block_type == CR_INTRA16x16AC || block_type == CHROMA_AC) ? 15 : 16) ); //convert zigzag scan to (run, level) pairs for (coeff_ctr = 0; coeff_ctr < max_coeff_num; coeff_ctr++) { run++; level = level_to_enc[coeff_ctr]; if (level != 0) { pLevel[scan_pos] = isignab(level, sign_to_enc[coeff_ctr]); pRun [scan_pos] = run; ++scan_pos; run = -1; // reset zero level counter } } level = 1; for(k = 0; (k < max_coeff_num) && level != 0; k++) { level = pLevel[k]; // level run = pRun[k]; // run if (level) { totzeros += run; // lets add run always (even if zero) to avoid conditional if (iabs(level) == 1) { numones ++; numtrailingones ++; numtrailingones = imin(numtrailingones, 3); // clip to 3 } else { numtrailingones = 0; } numcoeff ++; lastcoeff = k; } } if (!cdc) { numcoeff_vlc = (nnz < 2) ? 0 : ((nnz < 4) ? 1 : ((nnz < 8) ? 2 : 3)); } else { // chroma DC (has its own VLC) // numcoeff_vlc not relevant numcoeff_vlc = 0; } se.value1 = numcoeff; se.value2 = numtrailingones; se.len = numcoeff_vlc; /* use len to pass vlcnum */ if (!cdc) { if (se.len == 3) no_bits += 6; // 4 + 2 bit FLC else no_bits += Token_lentab[se.len][se.value2][se.value1]; } else no_bits += Token_lentab_cdc[yuv][se.value2][se.value1]; if (!numcoeff) return no_bits; else { if (numtrailingones) no_bits += numtrailingones; // encode levels level_two_or_higher = (numcoeff > 3 && numtrailingones == 3) ? 0 : 1; vlcnum = (numcoeff > 10 && numtrailingones < 3) ? 1 : 0; for (k = lastcoeff - numtrailingones; k >= 0; k--) { level = pLevel[k]; // level se.value1 = level; if (level_two_or_higher) { level_two_or_higher = 0; if (se.value1 > 0) se.value1 --; else se.value1 ++; } // encode level if (vlcnum == 0) estSyntaxElement_Level_VLC1(&se); else estSyntaxElement_Level_VLCN(&se, vlcnum); // update VLC table if (iabs(level) > incVlc[vlcnum]) vlcnum++; if ((k == lastcoeff - numtrailingones) && iabs(level) > 3) vlcnum = 2; no_bits += se.len; } // encode total zeroes if (numcoeff < max_coeff_num) { se.value1 = totzeros; vlcnum = numcoeff-1; se.len = vlcnum; if (!cdc) no_bits += Totalzeros_lentab[se.len][se.value1]; else no_bits += Totalzeros_lentab_cdc[yuv][se.len][se.value1]; } // encode run before each coefficient zerosleft = totzeros; numcoef = numcoeff; for (k = lastcoeff; k >= 0; k--) { run = pRun[k]; // run se.value1 = run; // for last coeff, run is remaining totzeros // when zerosleft is zero, remaining coeffs have 0 run if ((!zerosleft) || (numcoeff <= 1 )) break; if (numcoef > 1 && zerosleft) { vlcnum = imin(zerosleft - 1, RUNBEFORE_NUM_M1); se.len = vlcnum; no_bits += Run_lentab[se.len][se.value1]; zerosleft -= run; numcoef --; } } } return no_bits; }