--- estimation_rd_based.c 2003/11/13 23:11:24 1.1.2.9 +++ estimation_rd_based.c 2004/07/08 07:12:54 1.6 @@ -20,7 +20,7 @@ * along with this program ; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA * - * $Id: estimation_rd_based.c,v 1.1.2.9 2003/11/13 23:11:24 edgomez Exp $ + * $Id: estimation_rd_based.c,v 1.6 2004/07/08 07:12:54 syskin Exp $ * ****************************************************************************/ @@ -55,7 +55,8 @@ const uint32_t quant, const int quant_type, uint32_t * cbp, const int block, - const uint16_t * scan_table) + const uint16_t * scan_table, + const uint16_t * mpeg_quant_matrices) { int sum; int bits; @@ -63,15 +64,15 @@ fdct(data); - if (quant_type) sum = quant_h263_inter(coeff, data, quant); - else sum = quant_mpeg_inter(coeff, data, quant); + if (quant_type) sum = quant_h263_inter(coeff, data, quant, mpeg_quant_matrices); + else sum = quant_mpeg_inter(coeff, data, quant, mpeg_quant_matrices); if (sum > 0) { *cbp |= 1 << (5 - block); bits = BITS_MULT * CodeCoeffInter_CalcBits(coeff, scan_table); - if (quant_type) dequant_h263_inter(dqcoeff, coeff, quant); - else dequant_mpeg_inter(dqcoeff, coeff, quant); + if (quant_type) dequant_h263_inter(dqcoeff, coeff, quant, mpeg_quant_matrices); + else dequant_mpeg_inter(dqcoeff, coeff, quant, mpeg_quant_matrices); distortion = sse8_16bit(data, dqcoeff, 8*sizeof(int16_t)); } else { @@ -107,7 +108,8 @@ const uint32_t quant, const int quant_type, unsigned int bits[2], - unsigned int cbp[2]) + unsigned int cbp[2], + const uint16_t * mpeg_quant_matrices) { int direction; int16_t *pCurrent; @@ -118,11 +120,11 @@ fdct(coeff); if (quant_type) { - quant_h263_intra(qcoeff, coeff, quant, iDcScaler); - dequant_h263_intra(dqcoeff, qcoeff, quant, iDcScaler); + quant_h263_intra(qcoeff, coeff, quant, iDcScaler, mpeg_quant_matrices); + dequant_h263_intra(dqcoeff, qcoeff, quant, iDcScaler, mpeg_quant_matrices); } else { - quant_mpeg_intra(qcoeff, coeff, quant, iDcScaler); - dequant_mpeg_intra(dqcoeff, qcoeff, quant, iDcScaler); + quant_mpeg_intra(qcoeff, coeff, quant, iDcScaler, mpeg_quant_matrices); + dequant_mpeg_intra(dqcoeff, qcoeff, quant, iDcScaler, mpeg_quant_matrices); } predict_acdc(pMB-(x+mb_width*y), x, y, mb_width, block, qcoeff, @@ -133,6 +135,7 @@ /* store current coeffs to pred_values[] for future prediction */ pCurrent[0] = qcoeff[0] * iDcScaler; + pCurrent[0] = CLIP(pCurrent[0], -2048, 2047); for (i = 1; i < 8; i++) { pCurrent[i] = qcoeff[i]; pCurrent[i + 7] = qcoeff[i * 8]; @@ -172,14 +175,15 @@ static void -CheckCandidateRD16(const int x, const int y, const SearchData * const data, const unsigned int Direction) +CheckCandidateRD16(const int x, const int y, SearchData * const data, const unsigned int Direction) { int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; int32_t rd = 0; VECTOR * current; const uint8_t * ptr; - int i, cbp = 0, t, xc, yc; + int i, t, xc, yc; + unsigned cbp = 0; if ( (x > data->max_dx) || (x < data->min_dx) || (y > data->max_dy) || (y < data->min_dy) ) return; @@ -197,7 +201,7 @@ for(i = 0; i < 4; i++) { int s = 8*((i&1) + (i>>1)*data->iEdgedWidth); transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth); - rd += data->temp[i] = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, i, data->scan_table); + rd += data->temp[i] = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, i, data->scan_table, data->mpeg_quant_matrices); } rd += t = BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); @@ -222,33 +226,33 @@ /* chroma U */ ptr = interpolate8x8_switch2(data->RefQ, data->RefP[4], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); transfer_8to16subro(in, data->CurU, ptr, data->iEdgedWidth/2); - rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 4, data->scan_table); + rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 4, data->scan_table, data->mpeg_quant_matrices); if (rd >= data->iMinSAD[0]) return; /* chroma V */ ptr = interpolate8x8_switch2(data->RefQ, data->RefP[5], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding); transfer_8to16subro(in, data->CurV, ptr, data->iEdgedWidth/2); - rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5, data->scan_table); + rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5, data->scan_table, data->mpeg_quant_matrices); rd += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len; if (rd < data->iMinSAD[0]) { data->iMinSAD[0] = rd; current[0].x = x; current[0].y = y; - *data->dir = Direction; + data->dir = Direction; *data->cbp = cbp; } } static void -CheckCandidateRD8(const int x, const int y, const SearchData * const data, const unsigned int Direction) +CheckCandidateRD8(const int x, const int y, SearchData * const data, const unsigned int Direction) { int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64; int32_t rd; VECTOR * current; const uint8_t * ptr; - int cbp = 0; + unsigned int cbp = 0; if ( (x > data->max_dx) || (x < data->min_dx) || (y > data->max_dy) || (y < data->min_dy) ) return; @@ -262,14 +266,14 @@ } transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth); - rd = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5, data->scan_table); + rd = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5, data->scan_table, data->mpeg_quant_matrices); rd += BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0); if (rd < data->iMinSAD[0]) { *data->cbp = cbp; data->iMinSAD[0] = rd; current[0].x = x; current[0].y = y; - *data->dir = Direction; + data->dir = Direction; } } @@ -309,7 +313,7 @@ xvid_me_SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, 255, CheckCandidateRD16); if (MotionFlags&XVID_ME_HALFPELREFINE16_RD) - xvid_me_SubpelRefine(Data, CheckCandidateRD16); + xvid_me_SubpelRefine(Data->currentMV[0], Data, CheckCandidateRD16, 0); if (Data->qpel) { if (MotionFlags&(XVID_ME_EXTSEARCH_RD | XVID_ME_HALFPELREFINE16_RD)) { /* there was halfpel-precision search */ @@ -323,8 +327,12 @@ get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, pParam->width, pParam->height, Data->iFcode, 2, 0); } - if (MotionFlags&XVID_ME_QUARTERPELREFINE16_RD) - xvid_me_SubpelRefine(Data, CheckCandidateRD16); + if (MotionFlags & XVID_ME_QUARTERPELREFINE16_RD) { + if (MotionFlags & XVID_ME_FASTREFINE16) + FullRefine_Fast(Data, CheckCandidateRD16, 0); + else + xvid_me_SubpelRefine(Data->currentQMV[0], Data, CheckCandidateRD16, 0); + } } if (MotionFlags&XVID_ME_CHECKPREDICTION_RD) { /* let's check vector equal to prediction */ @@ -336,14 +344,14 @@ } static int -findRD_inter4v(const SearchData * const Data, +findRD_inter4v(SearchData * const Data, MACROBLOCK * const pMB, const MACROBLOCK * const pMBs, const int x, const int y, const MBParam * const pParam, const uint32_t MotionFlags, const VECTOR * const backup) { - int cbp = 0, bits = 0, t = 0, i; + unsigned int cbp = 0, bits = 0, t = 0, i; SearchData Data2, *Data8 = &Data2; int sumx = 0, sumy = 0; int16_t *in = Data->dctSpace, *coeff = Data->dctSpace + 64; @@ -353,9 +361,9 @@ for (i = 0; i < 4; i++) { /* for all luma blocks */ - Data8->iMinSAD = Data->iMinSAD + i + 1; - Data8->currentMV = Data->currentMV + i + 1; - Data8->currentQMV = Data->currentQMV + i + 1; + *Data8->iMinSAD = *(Data->iMinSAD + i + 1); + *Data8->currentMV = *(Data->currentMV + i + 1); + *Data8->currentQMV = *(Data->currentQMV + i + 1); Data8->Cur = Data->Cur + 8*((i&1) + (i>>1)*Data->iEdgedWidth); Data8->RefP[0] = Data->RefP[0] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); Data8->RefP[2] = Data->RefP[2] + 8*((i&1) + (i>>1)*Data->iEdgedWidth); @@ -387,10 +395,14 @@ } if (Data8->qpel) { + int bsad = Data8->iMinSAD[0]; + int bx = Data8->currentQMV->x; + int by = Data8->currentQMV->y; + + Data8->currentMV->x = Data8->currentQMV->x/2; + Data8->currentMV->y = Data8->currentQMV->y/2; + if (MotionFlags&XVID_ME_HALFPELREFINE8_RD || (MotionFlags&XVID_ME_EXTSEARCH8 && MotionFlags&XVID_ME_EXTSEARCH_RD)) { /* halfpixel motion search follows */ - int32_t s = *Data8->iMinSAD; - Data8->currentMV->x = Data8->currentQMV->x/2; - Data8->currentMV->y = Data8->currentQMV->y/2; Data8->qpel_precision = 0; get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3, pParam->width, pParam->height, Data8->iFcode - 1, 1, 0); @@ -402,11 +414,12 @@ xvid_me_SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255, CheckCandidateRD8); if (MotionFlags & XVID_ME_HALFPELREFINE8_RD) - xvid_me_SubpelRefine(Data8, CheckCandidateRD8); + xvid_me_SubpelRefine(Data->currentMV[0], Data8, CheckCandidateRD8, 0); - if(s > *Data8->iMinSAD) { /* we have found a better match */ - Data8->currentQMV->x = 2*Data8->currentMV->x; - Data8->currentQMV->y = 2*Data8->currentMV->y; + if(bsad > *Data8->iMinSAD) { /* we have found a better match */ + bx = Data8->currentQMV->x = 2*Data8->currentMV->x; + by = Data8->currentQMV->y = 2*Data8->currentMV->y; + bsad = Data8->iMinSAD[0]; } Data8->qpel_precision = 1; @@ -414,8 +427,19 @@ pParam->width, pParam->height, Data8->iFcode, 2, 0); } - if (MotionFlags & XVID_ME_QUARTERPELREFINE8_RD) - xvid_me_SubpelRefine(Data8, CheckCandidateRD8); + + if (MotionFlags & XVID_ME_QUARTERPELREFINE8_RD) { + if (MotionFlags & XVID_ME_FASTREFINE8) + FullRefine_Fast(Data8, CheckCandidateRD8, 0); + else xvid_me_SubpelRefine(Data->currentQMV[0], Data8, CheckCandidateRD8, 0); + } + + if (bsad <= Data->iMinSAD[0]) { + /* we have not found a better match */ + Data8->iMinSAD[0] = bsad; + Data8->currentQMV->x = bx; + Data8->currentQMV->y = by; + } } else { /* not qpel */ @@ -423,7 +447,7 @@ xvid_me_SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255, CheckCandidateRD8); if (MotionFlags & XVID_ME_HALFPELREFINE8_RD) - xvid_me_SubpelRefine(Data8, CheckCandidateRD8); /* halfpel refinement */ + xvid_me_SubpelRefine(Data->currentMV[0], Data8, CheckCandidateRD8, 0); /* halfpel refinement */ } /* checking vector equal to predicion */ @@ -464,14 +488,14 @@ /* chroma U */ ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[4], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); transfer_8to16subro(in, Data->CurU, ptr, Data->iEdgedWidth/2); - bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4, Data->scan_table); + bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4, Data->scan_table, Data->mpeg_quant_matrices); if (bits >= *Data->iMinSAD) return bits; /* chroma V */ ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefP[5], 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding); transfer_8to16subro(in, Data->CurV, ptr, Data->iEdgedWidth/2); - bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5, Data->scan_table); + bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5, Data->scan_table, Data->mpeg_quant_matrices); bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len; @@ -480,12 +504,12 @@ } static int -findRD_intra(const SearchData * const Data, MACROBLOCK * pMB, +findRD_intra(SearchData * const Data, MACROBLOCK * pMB, const int x, const int y, const int mb_width) { - int cbp[2] = {0, 0}, bits[2], i; - int bits1 = BITS_MULT*1, bits2 = BITS_MULT*1; /* this one is ac/dc prediction flag bit */ - int distortion = 0; + unsigned int cbp[2] = {0, 0}, bits[2], i; + unsigned int bits1 = BITS_MULT*1, bits2 = BITS_MULT*1; /* this one is ac/dc prediction flag bit */ + unsigned int distortion = 0; int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64, * dqcoeff = Data->dctSpace + 128; const uint32_t iQuant = Data->iQuant; @@ -497,7 +521,7 @@ distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, i, in, coeff, dqcoeff, - predictors[i], iQuant, Data->quant_type, bits, cbp); + predictors[i], iQuant, Data->quant_type, bits, cbp, Data->mpeg_quant_matrices); bits1 += distortion + BITS_MULT * bits[0]; bits2 += distortion + BITS_MULT * bits[1]; @@ -511,7 +535,7 @@ /*chroma U */ transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2); distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, 4, in, coeff, dqcoeff, - predictors[4], iQuant, Data->quant_type, bits, cbp); + predictors[4], iQuant, Data->quant_type, bits, cbp, Data->mpeg_quant_matrices); bits1 += distortion + BITS_MULT * bits[0]; bits2 += distortion + BITS_MULT * bits[1]; @@ -521,7 +545,7 @@ /* chroma V */ transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2); distortion = Block_CalcBitsIntra(pMB, x, y, mb_width, 5, in, coeff, dqcoeff, - predictors[5], iQuant, Data->quant_type, bits, cbp); + predictors[5], iQuant, Data->quant_type, bits, cbp, Data->mpeg_quant_matrices); bits1 += distortion + BITS_MULT * bits[0]; bits2 += distortion + BITS_MULT * bits[1]; @@ -536,16 +560,16 @@ static int -findRD_gmc(const SearchData * const Data, const IMAGE * const vGMC, const int x, const int y) +findRD_gmc(SearchData * const Data, const IMAGE * const vGMC, const int x, const int y) { int bits = BITS_MULT*1; /* this one is mcsel */ - int cbp = 0, i; + unsigned int cbp = 0, i; int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64; for(i = 0; i < 4; i++) { int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth); transfer_8to16subro(in, Data->Cur + s, vGMC->y + s + 16*(x+y*Data->iEdgedWidth), Data->iEdgedWidth); - bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, i, Data->scan_table); + bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, i, Data->scan_table, Data->mpeg_quant_matrices); if (bits >= Data->iMinSAD[0]) return bits; } @@ -553,13 +577,13 @@ /*chroma U */ transfer_8to16subro(in, Data->CurU, vGMC->u + 8*(x+y*(Data->iEdgedWidth/2)), Data->iEdgedWidth/2); - bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4, Data->scan_table); + bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 4, Data->scan_table, Data->mpeg_quant_matrices); if (bits >= Data->iMinSAD[0]) return bits; /* chroma V */ transfer_8to16subro(in, Data->CurV , vGMC->v + 8*(x+y*(Data->iEdgedWidth/2)), Data->iEdgedWidth/2); - bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5, Data->scan_table); + bits += Block_CalcBits(coeff, in, Data->dctSpace + 128, Data->iQuant, Data->quant_type, &cbp, 5, Data->scan_table, Data->mpeg_quant_matrices); bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len; @@ -587,10 +611,9 @@ int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0); const uint32_t iQuant = pMB->quant; - int min_rd, intra_rd, i, cbp, c[2] = {0, 0}; + int min_rd, intra_rd, i, cbp; VECTOR backup[5], *v; Data->iQuant = iQuant; - Data->cbp = c; Data->scan_table = VopFlags & XVID_VOP_ALTERNATESCAN ? scan_tables[2] : scan_tables[0]; @@ -694,7 +717,7 @@ const uint32_t iQuant = pMB->quant; const int skip_possible = (coding_type == P_VOP) && (pMB->dquant == 0); int sad; - int min_rd = -1, intra_rd, i, cbp = 63, c[2] = {0, 0}; + int min_rd = -1, intra_rd, i, cbp = 63; VECTOR backup[5], *v; int sad_backup[5]; int InterBias = MV16_INTER_BIAS; @@ -748,7 +771,6 @@ } } else { /* Rate-Distortion INTER<->INTER4V */ Data->iQuant = iQuant; - Data->cbp = c; v = Data->qpel ? Data->currentQMV : Data->currentMV; /* final skip decision, a.k.a. "the vector you found, really that good?" */ @@ -849,7 +871,6 @@ } else { /* Rate-Distortion INTRA<->INTER */ if(min_rd < 0) { Data->iQuant = iQuant; - Data->cbp = c; v = Data->qpel ? Data->currentQMV : Data->currentMV; for (i = 0; i < 5; i++) {