--- trunk/xvidcore/src/motion/estimation_bvop.c 2004/07/21 12:50:30 1515 +++ trunk/xvidcore/src/motion/estimation_bvop.c 2004/12/08 12:43:48 1567 @@ -21,7 +21,7 @@ * 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_bvop.c,v 1.11 2004-07-21 12:50:30 syskin Exp $ + * $Id: estimation_bvop.c,v 1.18 2004-12-08 12:43:48 syskin Exp $ * ****************************************************************************/ @@ -137,8 +137,8 @@ xcb = xb/2; ycb = yb/2; } - t = d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0) - + d_mv_bits(xb, yb, data->bpredMV, data->iFcode, data->qpel^data->qpel_precision, 0); + t = d_mv_bits(xf, yf, data->predMV, data->iFcode, data->qpel^data->qpel_precision) + + d_mv_bits(xb, yb, data->bpredMV, data->iFcode, data->qpel^data->qpel_precision); sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); sad += (data->lambda16 * t * sad)>>10; @@ -207,7 +207,7 @@ if (sad > *(data->iMinSAD)) return; } - sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10; + sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0) * sad)>>10; if (data->chroma && sad < *data->iMinSAD) sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], @@ -265,7 +265,7 @@ done: sad = sad16bi(data->Cur, ReferenceF, ReferenceB, data->iEdgedWidth); - sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0, 0) * sad)>>10; + sad += (data->lambda16 * d_mv_bits(x, y, zeroMV, 1, 0) * sad)>>10; if (data->chroma && sad < *data->iMinSAD) sad += ChromaSAD2((xcf >> 3) + roundtab_76[xcf & 0xf], @@ -291,8 +291,6 @@ if ( (x > data->max_dx) || ( x < data->min_dx) || (y > data->max_dy) || (y < data->min_dy) ) return; - if (data->rrv && (!(x&1) && x !=0) | (!(y&1) && y !=0) ) return; /* non-zero even value */ - if (data->qpel_precision) { /* x and y are in 1/4 precision */ Reference = xvid_me_interpolate16x16qpel(x, y, 0, data); current = data->currentQMV; @@ -303,7 +301,7 @@ xc = x; yc = y; } t = d_mv_bits(x, y, data->predMV, data->iFcode, - data->qpel^data->qpel_precision, data->rrv); + data->qpel^data->qpel_precision); sad = sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096); sad += (data->lambda16 * t * sad)>>10; @@ -445,7 +443,7 @@ Data->predMV = *predMV; get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4, - pParam->width, pParam->height, iFcode - Data->qpel, 1, 0); + pParam->width, pParam->height, iFcode - Data->qpel, 1); pmv[0] = Data->predMV; if (Data->qpel) { @@ -492,7 +490,7 @@ if(MotionFlags & XVID_ME_FASTREFINE16) { /* fast */ 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); + pParam->width, pParam->height, Data->iFcode, 2); FullRefine_Fast(Data, CheckCandidate16no4v, 0); } else { @@ -507,7 +505,7 @@ Data->currentQMV->y = 2*Data->currentMV->y; } 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); + pParam->width, pParam->height, Data->iFcode, 2); Data->qpel_precision = 1; xvid_me_SubpelRefine(Data->currentQMV[0], Data, CheckCandidate16no4v, 0); /* qpel part */ } @@ -522,8 +520,6 @@ { int k; - pMB->mode = MODE_DIRECT; /* just to initialize it */ - if (!Data->chroma) { int dx = 0, dy = 0, b_dx = 0, b_dy = 0; int32_t sum; @@ -543,15 +539,15 @@ b_dx = (b_dx >> 3) + roundtab_76[b_dx & 0xf]; sum = sad8bi(Data->CurU, - Data->RefP[4] + (dy/2) * stride + dx/2, - Data->b_RefP[4] + (b_dy/2) * stride + b_dx/2, + Data->RefP[4] + (dy/2) * (int)stride + dx/2, + Data->b_RefP[4] + (b_dy/2) * (int)stride + b_dx/2, stride); if (sum >= MAX_CHROMA_SAD_FOR_SKIP * (int)Data->iQuant) return; /* no skip */ sum += sad8bi(Data->CurV, - Data->RefP[5] + (dy/2) * stride + dx/2, - Data->b_RefP[5] + (b_dy/2) * stride + b_dx/2, + Data->RefP[5] + (dy/2) * (int)stride + dx/2, + Data->b_RefP[5] + (b_dy/2) * (int)stride + b_dx/2, stride); if (sum >= MAX_CHROMA_SAD_FOR_SKIP * (int)Data->iQuant) return; /* no skip */ @@ -566,6 +562,10 @@ for (k = 0; k < 4; k++) { pMB->qmvs[k] = pMB->mvs[k] = Data->directmvF[k]; pMB->b_qmvs[k] = pMB->b_mvs[k] = Data->directmvB[k]; + if (Data->qpel) { + pMB->mvs[k].x /= 2; pMB->mvs[k].y /= 2; /* it's a hint for future searches */ + pMB->b_mvs[k].x /= 2; pMB->b_mvs[k].y /= 2; + } } } @@ -625,8 +625,10 @@ return *Data->iMinSAD; /* skipped */ } - skip_sad = 4*MAX(MAX(Data->iMinSAD[1],Data->iMinSAD[2]), MAX(Data->iMinSAD[3],Data->iMinSAD[4])); - if (Data->chroma) skip_sad += Data->chromaSAD; + if (Data->chroma && Data->chromaSAD >= MAX_CHROMA_SAD_FOR_SKIP * (int)Data->iQuant) /* chroma doesn't allow skip */ + skip_sad = 256*4096; + else + skip_sad = 4*MAX(MAX(Data->iMinSAD[1],Data->iMinSAD[2]), MAX(Data->iMinSAD[3],Data->iMinSAD[4])); Data->currentMV[1].x = Data->directmvF[0].x + Data->currentMV->x; /* hints for forward and backward searches */ Data->currentMV[1].y = Data->directmvF[0].y + Data->currentMV->y; @@ -639,6 +641,8 @@ Data->directmvB[0].y : Data->currentMV[1].y - Data->referencemv[0].y); + *best_sad = Data->iMinSAD[0]; + return skip_sad; } @@ -707,8 +711,8 @@ Data->currentMV[0] = startF; Data->currentMV[1] = startB; - get_range(f_range, f_range+1, f_range+2, f_range+3, x, y, 4, pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, 0); - get_range(b_range, b_range+1, b_range+2, b_range+3, x, y, 4, pParam->width, pParam->height, Data->bFcode - Data->qpel, 1, 0); + get_range(f_range, f_range+1, f_range+2, f_range+3, x, y, 4, pParam->width, pParam->height, Data->iFcode - Data->qpel, 1); + get_range(b_range, b_range+1, b_range+2, b_range+3, x, y, 4, pParam->width, pParam->height, Data->bFcode - Data->qpel, 1); if (Data->currentMV[0].x > f_range[1]) Data->currentMV[0].x = f_range[1]; if (Data->currentMV[0].x < f_range[0]) Data->currentMV[0].x = f_range[0]; @@ -737,8 +741,8 @@ int i, j; int b_range[4], f_range[4]; - get_range(f_range, f_range+1, f_range+2, f_range+3, x, y, 4, pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, 0); - get_range(b_range, b_range+1, b_range+2, b_range+3, x, y, 4, pParam->width, pParam->height, Data->bFcode - Data->qpel, 1, 0); + get_range(f_range, f_range+1, f_range+2, f_range+3, x, y, 4, pParam->width, pParam->height, Data->iFcode - Data->qpel, 1); + get_range(b_range, b_range+1, b_range+2, b_range+3, x, y, 4, pParam->width, pParam->height, Data->bFcode - Data->qpel, 1); /* diamond */ do { @@ -768,7 +772,7 @@ if (Data->qpel) { Data->qpel_precision = 1; 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); + x, y, 4, pParam->width, pParam->height, Data->iFcode, 2); Data->currentQMV[0].x = 2 * Data->currentMV[0].x; Data->currentQMV[0].y = 2 * Data->currentMV[0].y; @@ -779,7 +783,7 @@ xvid_me_SubpelRefine(Data->currentQMV[0], Data, CheckCandidateInt, 1); get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, - x, y, 4, pParam->width, pParam->height, Data->bFcode, 2, 0); + x, y, 4, pParam->width, pParam->height, Data->bFcode, 2); xvid_me_SubpelRefine(Data->currentQMV[1], Data, CheckCandidateInt, 2); } @@ -914,6 +918,32 @@ } } +static __inline void +maxMotionBVOP(int * const MVmaxF, int * const MVmaxB, const MACROBLOCK * const pMB, const int qpel) +{ + if (pMB->mode == MODE_FORWARD || pMB->mode == MODE_INTERPOLATE) { + const VECTOR * const mv = qpel ? pMB->qmvs : pMB->mvs; + int max = *MVmaxF; + if (mv[0].x > max) max = mv[0].x; + else if (-mv[0].x - 1 > max) max = -mv[0].x - 1; + if (mv[0].y > max) max = mv[0].y; + else if (-mv[0].y - 1 > max) max = -mv[0].y - 1; + + *MVmaxF = max; + } + + if (pMB->mode == MODE_BACKWARD || pMB->mode == MODE_INTERPOLATE) { + const VECTOR * const mv = qpel ? pMB->b_qmvs : pMB->b_mvs; + int max = *MVmaxB; + if (mv[0].x > max) max = mv[0].x; + else if (-mv[0].x - 1 > max) max = -mv[0].x - 1; + if (mv[0].y > max) max = mv[0].y; + else if (-mv[0].y - 1 > max) max = -mv[0].y - 1; + *MVmaxB = max; + } +} + + void MotionEstimationBVOP(MBParam * const pParam, FRAMEINFO * const frame, @@ -933,13 +963,15 @@ const IMAGE * const b_refHV) { uint32_t i, j; - int32_t best_sad, sad2; + int32_t best_sad = 256*4096; + int32_t sad2; uint32_t skip_sad; const MACROBLOCK * const b_mbs = b_reference->mbs; VECTOR f_predMV, b_predMV; + int MVmaxF = 0, MVmaxB = 0; const int32_t TRB = time_pp - time_bp; const int32_t TRD = time_pp; DECLARE_ALIGNED_MATRIX(dct_space, 3, 64, int16_t, CACHE_LINE); @@ -963,9 +995,8 @@ memcpy(&Data_b, &Data_d, sizeof(SearchData)); memcpy(&Data_i, &Data_d, sizeof(SearchData)); - Data_f.iFcode = Data_i.iFcode = frame->fcode; - Data_b.iFcode = Data_i.bFcode = frame->bcode; - + Data_f.iFcode = Data_i.iFcode = frame->fcode = b_reference->fcode; + Data_b.iFcode = Data_i.bFcode = frame->bcode = b_reference->fcode; for (j = 0; j < pParam->mb_height; j++) { @@ -1025,14 +1056,16 @@ if ((Data_d.iMinSAD[0] <= 2*best_sad) && (!frame->motion_flags&XVID_ME_SKIP_DELTASEARCH)) SearchDirect_final(frame->motion_flags, b_mb, &best_sad, &Data_d); - /* final skip decision */ - if ( (skip_sad < Data_d.iQuant * MAX_SAD00_FOR_SKIP ) + if ( (skip_sad < 2 * Data_d.iQuant * MAX_SAD00_FOR_SKIP ) && ((100*best_sad)/(skip_sad+1) > FINAL_SKIP_THRESH) ) { + Data_d.chromaSAD = 0; /* green light for chroma check */ + SkipDecisionB(pMB, &Data_d); + if (pMB->mode == MODE_DIRECT_NONE_MV) { /* skipped? */ - pMB->sad16 = best_sad; + pMB->sad16 = skip_sad; continue; } } @@ -1043,6 +1076,11 @@ else ModeDecision_BVOP_SAD(&Data_d, &Data_b, &Data_f, &Data_i, pMB, b_mb, &f_predMV, &b_predMV); + maxMotionBVOP(&MVmaxF, &MVmaxB, pMB, Data_d.qpel); + } } + + frame->fcode = getMinFcode(MVmaxF); + frame->bcode = getMinFcode(MVmaxB); }