--- trunk/xvidcore/src/motion/motion_est.c	2003/02/15 15:22:19	851
+++ branches/dev-api-4/xvidcore/src/motion/motion_est.c	2003/03/29 01:29:53	958
@@ -46,6 +46,26 @@
 #include "../utils/emms.h"
 #include "../dct/fdct.h"
 
+/*****************************************************************************
+ * Modified rounding tables -- declared in motion.h
+ * Original tables see ISO spec tables 7-6 -> 7-9
+ ****************************************************************************/
+
+const uint32_t roundtab[16] =
+{0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2 };
+
+/* K = 4 */
+const uint32_t roundtab_76[16] =
+{ 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1 };
+
+/* K = 2 */
+const uint32_t roundtab_78[8] =
+{ 0, 0, 1, 1, 0, 0, 0, 1  };
+
+/* K = 1 */
+const uint32_t roundtab_79[4] =
+{ 0, 1, 0, 0 };
+
 #define INITIAL_SKIP_THRESH	(10)
 #define FINAL_SKIP_THRESH	(50)
 #define MAX_SAD00_FOR_SKIP	(20)
@@ -54,6 +74,10 @@
 #define CHECK_CANDIDATE(X,Y,D) { \
 CheckCandidate((X),(Y), (D), &iDirection, data ); }
 
+/*****************************************************************************
+ * Code
+ ****************************************************************************/
+
 static __inline uint32_t
 d_mv_bits(int x, int y, const VECTOR pred, const uint32_t iFcode, const int qpel, const int rrv)
 {
@@ -331,7 +355,7 @@
 	data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10;
 
 	if (data->chroma) sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3],
-													(yc >> 1) + roundtab_79[yc & 0x3], data);
+									   (yc >> 1) + roundtab_79[yc & 0x3], data);
 
 	if (sad < data->iMinSAD[0]) {
 		data->iMinSAD[0] = sad;
@@ -482,8 +506,9 @@
 	const uint8_t *ReferenceF, *ReferenceB;
 	VECTOR *current;
 
-	if ( (xf > data->max_dx) | (xf < data->min_dx)
-		| (yf > data->max_dy) | (yf < data->min_dy) ) return;
+	if ((xf > data->max_dx) || (xf < data->min_dx) ||
+		(yf > data->max_dy) || (yf < data->min_dy))
+		return;
 
 	if (!data->qpel_precision) {
 		ReferenceF = GetReference(xf, yf, data);
@@ -528,7 +553,7 @@
 	const uint8_t *ReferenceB;
 	VECTOR mvs, b_mvs;
 
-	if (( x > 31) | ( x < -32) | ( y > 31) | (y < -32)) return;
+	if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return;
 
 	for (k = 0; k < 4; k++) {
 		mvs.x = data->directmvF[k].x + x;
@@ -541,10 +566,11 @@
 			data->directmvB[k].y
 			: mvs.y - data->referencemv[k].y);
 
-		if ( (mvs.x > data->max_dx) | (mvs.x < data->min_dx)
-			| (mvs.y > data->max_dy) | (mvs.y < data->min_dy)
-			| (b_mvs.x > data->max_dx) | (b_mvs.x < data->min_dx)
-			| (b_mvs.y > data->max_dy) | (b_mvs.y < data->min_dy) ) return;
+		if ((mvs.x > data->max_dx)   || (mvs.x < data->min_dx)   ||
+			(mvs.y > data->max_dy)   || (mvs.y < data->min_dy)   ||
+			(b_mvs.x > data->max_dx) || (b_mvs.x < data->min_dx) ||
+			(b_mvs.y > data->max_dy) || (b_mvs.y < data->min_dy) )
+			return;
 
 		if (data->qpel) {
 			xcf += mvs.x/2; ycf += mvs.y/2;
@@ -697,7 +723,7 @@
 		}
 	}
 
-	bits += cbpy_tab[15-(cbp>>2)].len;
+	bits += xvid_cbpy_tab[15-(cbp>>2)].len;
 	bits += mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len;
 
 	if (bits < data->iMinSAD[0]) {
@@ -971,11 +997,15 @@
 	uint32_t mb_width = pParam->mb_width;
 	uint32_t mb_height = pParam->mb_height;
 	const uint32_t iEdgedWidth = pParam->edged_width;
-	const uint32_t MotionFlags = MakeGoodMotionFlags(current->motion_flags, current->global_flags);
+	const uint32_t MotionFlags = MakeGoodMotionFlags(current->motion_flags, current->vop_flags, current->vol_flags);
 
 	uint32_t x, y;
 	uint32_t iIntra = 0;
 	int32_t quant = current->quant, sad00;
+	int skip_thresh = \
+		INITIAL_SKIP_THRESH * \
+		(current->vop_flags & XVID_VOP_REDUCED ? 4:1) * \
+		(current->vop_flags & XVID_VOP_MODEDECISION_BITS ? 2:1);
 
 	// some pre-initialized thingies for SearchP
 	int32_t temp[8];
@@ -991,11 +1021,11 @@
 	Data.temp = temp;
 	Data.iFcode = current->fcode;
 	Data.rounding = pParam->m_rounding_type;
-	Data.qpel = pParam->m_quarterpel;
-	Data.chroma = MotionFlags & PMV_CHROMA16;
-	Data.rrv = current->global_flags & XVID_REDUCED;
+	Data.qpel = current->vol_flags & XVID_VOL_QUARTERPEL;
+	Data.chroma = MotionFlags & XVID_ME_CHROMA16;
+	Data.rrv = current->vop_flags & XVID_VOP_REDUCED;
 
-	if ((current->global_flags & XVID_REDUCED)) {
+	if ((current->vop_flags & XVID_VOP_REDUCED)) {
 		mb_width = (pParam->width + 31) / 32;
 		mb_height = (pParam->height + 31) / 32;
 		Data.qpel = 0;
@@ -1028,21 +1058,18 @@
 
 			sad00 = pMB->sad16;
 
-			if (!(current->global_flags & XVID_LUMIMASKING)) {
-				pMB->dquant = NO_CHANGE;
-			} else {
-				if (pMB->dquant != NO_CHANGE) {
-					quant += DQtab[pMB->dquant];
-					if (quant > 31) quant = 31;
-					else if (quant < 1) quant = 1;
-				}
+			if (pMB->dquant != 0) {
+				quant += DQtab[pMB->dquant];
+				if (quant > 31) quant = 31;
+				else if (quant < 1) quant = 1;
 			}
+			
 			pMB->quant = current->quant;
 
 //initial skip decision
 /* no early skip for GMC (global vector = skip vector is unknown!)  */
-			if (!(current->global_flags & XVID_GMC))	{ /* no fast SKIP for S(GMC)-VOPs */
-				if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH * (Data.rrv ? 4:1) )
+			if (!(current->vol_flags & XVID_VOL_GMC))	{ /* no fast SKIP for S(GMC)-VOPs */
+				if (pMB->dquant == 0 && sad00 < pMB->quant * skip_thresh)
 					if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) {
 						SkipMacroblockP(pMB, sad00);
 						continue;
@@ -1050,22 +1077,16 @@
 			}
 
 			SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
-						y, MotionFlags, current->global_flags, pMB->quant,
+						y, MotionFlags, current->vol_flags, pMB->quant,
 						&Data, pParam, pMBs, reference->mbs,
-						current->global_flags & XVID_INTER4V, pMB);
+						current->vop_flags & XVID_VOP_INTER4V, pMB);
 
 /* final skip decision, a.k.a. "the vector you found, really that good?" */
-			if (!(current->global_flags & XVID_GMC))	{
-				if ( pMB->dquant == NO_CHANGE && sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) {
-					if (!(current->global_flags & XVID_MODEDECISION_BITS)) {
-						if ( (100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH * (Data.rrv ? 4:1) )
-							if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv))
-								SkipMacroblockP(pMB, sad00);
-					} else { // BITS mode decision
-						if (pMB->sad16 > 10)
-							SkipMacroblockP(pMB, sad00);  // more than 10 bits would be used for this MB - skip
-
-					}
+			if (!(current->vol_flags & XVID_VOL_GMC || current->vop_flags & XVID_VOP_MODEDECISION_BITS)) {
+				if ( pMB->dquant == 0 && sad00 < pMB->quant * MAX_SAD00_FOR_SKIP) {
+					if ( (100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH * (Data.rrv ? 4:1) )
+						if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv))
+							SkipMacroblockP(pMB, sad00);
 				}
 			}
 			if (pMB->mode == MODE_INTRA)
@@ -1073,7 +1094,7 @@
 		}
 	}
 
-	if (current->global_flags & XVID_GMC )	/* GMC only for S(GMC)-VOPs */
+	if (current->vol_flags & XVID_VOL_GMC )	/* GMC only for S(GMC)-VOPs */
 	{
 		current->warp = GlobalMotionEst( pMBs, pParam, current, reference, pRefH, pRefV, pRefHV);
 	}
@@ -1148,13 +1169,13 @@
 		const int x, const int y,
 		const MBParam * const pParam,
 		const uint32_t MotionFlags,
-		const uint32_t GlobalFlags)
+		const uint32_t VopFlags)
 {
 
 	int mode = MODE_INTER;
 
-	if (!(GlobalFlags & XVID_MODEDECISION_BITS)) { //normal, fast, SAD-based mode decision
-		int intra = 0;
+	if (!(VopFlags & XVID_VOP_MODEDECISION_BITS)) { //normal, fast, SAD-based mode decision
+//		int intra = 0;
 		int sad;
 		int InterBias = MV16_INTER_BIAS;
 		if (inter4v == 0 || Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] +
@@ -1196,7 +1217,7 @@
 		int bits, intra, i;
 		VECTOR backup[5], *v;
 		Data->lambda16 = iQuant;
-		Data->lambda8 = pParam->m_quant_type;
+        Data->lambda8 = (pParam->vol_flags & XVID_VOL_MPEGQUANT)?1:0;
 
 		v = Data->qpel ? Data->currentQMV : Data->currentMV;
 		for (i = 0; i < 5; i++) {
@@ -1230,7 +1251,7 @@
 		const int x,
 		const int y,
 		const uint32_t MotionFlags,
-		const uint32_t GlobalFlags,
+		const uint32_t VopFlags,
 		const uint32_t iQuant,
 		SearchData * const Data,
 		const MBParam * const pParam,
@@ -1265,7 +1286,7 @@
 	Data->lambda8 = lambda_vec8[iQuant];
 	Data->qpel_precision = 0;
 
-	if (pMB->dquant != NO_CHANGE) inter4v = 0;
+	if (pMB->dquant != 0) inter4v = 0;
 
 	for(i = 0; i < 5; i++)
 		Data->currentMV[i].x = Data->currentMV[i].y = 0;
@@ -1280,7 +1301,7 @@
 	Data->iMinSAD[3] = pMB->sad8[2];
 	Data->iMinSAD[4] = pMB->sad8[3];
 
-	if ((!(GlobalFlags & XVID_MODEDECISION_BITS)) || (x | y)) {
+	if ((!(VopFlags & XVID_VOP_MODEDECISION_BITS)) || (x | y)) {
 		threshA = Data->temp[0]; // that's where we keep this SAD atm
 		if (threshA < 512) threshA = 512;
 		else if (threshA > 1024) threshA = 1024;
@@ -1306,12 +1327,12 @@
 	if ((Data->iMinSAD[0] <= threshA) ||
 			(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) &&
 			(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16))) {
-		if (!(GlobalFlags & XVID_MODEDECISION_BITS)) inter4v = 0;	}
+		if (!(VopFlags & XVID_VOP_MODEDECISION_BITS)) inter4v = 0;	}
 	else {
 
 		MainSearchFunc * MainSearchPtr;
-		if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
-		else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
+		if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = SquareSearch;
+		else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
 			else MainSearchPtr = DiamondSearch;
 
 		MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
@@ -1320,7 +1341,7 @@
 	note that this search is/might be done in halfpel positions,
 	which makes it more different than the diamond above */
 
-		if (MotionFlags & PMV_EXTSEARCH16) {
+		if (MotionFlags & XVID_ME_EXTSEARCH16) {
 			int32_t bSAD;
 			VECTOR startMV = Data->predMV, backupMV = Data->currentMV[0];
 			if (Data->rrv) {
@@ -1351,8 +1372,8 @@
 		}
 	}
 
-	if (MotionFlags & PMV_HALFPELREFINE16)
-		if ((!(MotionFlags & HALFPELREFINE16_BITS)) || Data->iMinSAD[0] < 200*(int)iQuant)
+	if (MotionFlags & XVID_ME_HALFPELREFINE16)
+		if ((!(MotionFlags & XVID_ME_HALFPELREFINE16_BITS)) || Data->iMinSAD[0] < 200*(int)iQuant)
 			SubpelRefine(Data);
 
 	for(i = 0; i < 5; i++) {
@@ -1360,20 +1381,22 @@
 		Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
 	}
 
-	if (MotionFlags & PMV_QUARTERPELREFINE16)
-		if ((!(MotionFlags & QUARTERPELREFINE16_BITS)) || (Data->iMinSAD[0] < 200*(int)iQuant)) {
-			Data->qpel_precision = 1;
-			get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
-					pParam->width, pParam->height, Data->iFcode, 1, 0);
+	if (MotionFlags & XVID_ME_QUARTERPELREFINE16) {
+		
+		get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
+				pParam->width, pParam->height, Data->iFcode, 1, 0);
 
+		if ((!(MotionFlags & XVID_ME_QUARTERPELREFINE16_BITS)) || (Data->iMinSAD[0] < 200*(int)iQuant)) {
+			Data->qpel_precision = 1;
 			SubpelRefine(Data);
 		}
+	}
 
-	if ((!(GlobalFlags & XVID_MODEDECISION_BITS)) && (Data->iMinSAD[0] < (int32_t)iQuant * 30)) inter4v = 0;
+	if ((!(VopFlags & XVID_VOP_MODEDECISION_BITS)) && (Data->iMinSAD[0] < (int32_t)iQuant * 30)) inter4v = 0;
 
-	if (inter4v && (!(GlobalFlags & XVID_MODEDECISION_BITS) ||
-			(!(MotionFlags & QUARTERPELREFINE8_BITS)) || (!(MotionFlags & HALFPELREFINE8_BITS)) ||
-			((!(MotionFlags & EXTSEARCH_BITS)) && (!(MotionFlags&PMV_EXTSEARCH8)) ))) {
+	if (inter4v && (!(VopFlags & XVID_VOP_MODEDECISION_BITS) ||
+			(!(MotionFlags & XVID_ME_QUARTERPELREFINE8_BITS)) || (!(MotionFlags & XVID_ME_HALFPELREFINE8_BITS)) ||
+			((!(MotionFlags & XVID_ME_EXTSEARCH_BITS)) && (!(MotionFlags&XVID_ME_EXTSEARCH8)) ))) {
 		// if decision is BITS-based and all refinement steps will be done in BITS domain, there is no reason to call this loop
 
 		SearchData Data8;
@@ -1384,7 +1407,7 @@
 		Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
 		Search8(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
 
-		if ((Data->chroma) && (!(GlobalFlags & XVID_MODEDECISION_BITS))) {
+		if ((Data->chroma) && (!(VopFlags & XVID_VOP_MODEDECISION_BITS))) {
 			// chroma is only used for comparsion to INTER. if the comparsion will be done in BITS domain, there is no reason to compute it
 			int sumx = 0, sumy = 0;
 			const int div = 1 + Data->qpel;
@@ -1400,7 +1423,7 @@
 		}
 	}
 
-	inter4v = ModeDecision(iQuant, Data, inter4v, pMB, pMBs, x, y, pParam, MotionFlags, GlobalFlags);
+	inter4v = ModeDecision(iQuant, Data, inter4v, pMB, pMBs, x, y, pParam, MotionFlags, VopFlags);
 
 	if (Data->rrv) {
 			Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x);
@@ -1459,7 +1482,7 @@
 
 	*(Data->iMinSAD) += (Data->lambda8 * i * (*Data->iMinSAD + NEIGH_8X8_BIAS))>>10;
 
-	if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8|PMV_QUARTERPELREFINE8)) {
+	if (MotionFlags & (XVID_ME_EXTSEARCH8|XVID_ME_HALFPELREFINE8|XVID_ME_QUARTERPELREFINE8)) {
 		if (Data->rrv) i = 2; else i = 1;
 
 		Data->Ref = OldData->Ref + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
@@ -1476,12 +1499,12 @@
 		if (!Data->rrv) CheckCandidate = CheckCandidate8;
 		else CheckCandidate = CheckCandidate16no4v;
 
-		if (MotionFlags & PMV_EXTSEARCH8 && (!(MotionFlags & EXTSEARCH_BITS))) {
+		if (MotionFlags & XVID_ME_EXTSEARCH8 && (!(MotionFlags & XVID_ME_EXTSEARCH_BITS))) {
 			int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
 
 			MainSearchFunc *MainSearchPtr;
-			if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
-				else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
+			if (MotionFlags & XVID_ME_USESQUARES8) MainSearchPtr = SquareSearch;
+				else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
 					else MainSearchPtr = DiamondSearch;
 
 			MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, 255);
@@ -1492,7 +1515,7 @@
 			}
 		}
 
-		if (MotionFlags & PMV_HALFPELREFINE8) {
+		if (MotionFlags & XVID_ME_HALFPELREFINE8) {
 			int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
 
 			SubpelRefine(Data); // perform halfpel refine of current best vector
@@ -1503,7 +1526,7 @@
 			}
 		}
 
-		if (Data->qpel && MotionFlags & PMV_QUARTERPELREFINE8) {
+		if (Data->qpel && MotionFlags & XVID_ME_QUARTERPELREFINE8) {
 				Data->qpel_precision = 1;
 				get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
 					pParam->width, pParam->height, Data->iFcode, 1, 0);
@@ -1627,8 +1650,8 @@
 		CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
 	}
 
-	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
-	else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
+	if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = SquareSearch;
+	else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
 		else MainSearchPtr = DiamondSearch;
 
 	MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
@@ -1799,13 +1822,14 @@
 		}
 	}
 
+	*Data->iMinSAD += Data->lambda16;
 	skip_sad = *Data->iMinSAD;
 
 //	DIRECT MODE DELTA VECTOR SEARCH.
 //	This has to be made more effective, but at the moment I'm happy it's running at all
 
-	if (MotionFlags & PMV_USESQUARES16) MainSearchPtr = SquareSearch;
-		else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
+	if (MotionFlags & XVID_ME_USESQUARES16) MainSearchPtr = SquareSearch;
+		else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
 			else MainSearchPtr = DiamondSearch;
 
 	MainSearchPtr(0, 0, Data, 255);
@@ -2014,9 +2038,9 @@
 	Data.currentMV = currentMV; Data.currentQMV = currentQMV;
 	Data.iMinSAD = &iMinSAD;
 	Data.lambda16 = lambda_vec16[frame->quant];
-	Data.qpel = pParam->m_quarterpel;
+	Data.qpel = pParam->vol_flags & XVID_VOL_QUARTERPEL;
 	Data.rounding = 0;
-	Data.chroma = frame->motion_flags & PMV_CHROMA8;
+	Data.chroma = frame->motion_flags & XVID_ME_CHROMA8;
 	Data.temp = temp;
 
 	Data.RefQ = f_refV->u; // a good place, also used in MC (for similar purpose)
@@ -2124,8 +2148,9 @@
 {
 
 	int i, mask;
+	int quarterpel = (pParam->vol_flags & XVID_VOL_QUARTERPEL)? 1: 0;
 	VECTOR pmv[3];
-	MACROBLOCK * pMB = &pMBs[x + y * pParam->mb_width];
+	MACROBLOCK * const pMB = &pMBs[x + y * pParam->mb_width];
 
 	for (i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR;
 
@@ -2139,7 +2164,7 @@
 			else Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0); //else median
 
 	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
-				pParam->width, pParam->height, Data->iFcode - pParam->m_quarterpel, 0, Data->rrv);
+        pParam->width, pParam->height, Data->iFcode - quarterpel, 0, 0);
 
 	Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
 	Data->Ref = pRef + (x + y * pParam->edged_width) * 16;
@@ -2151,43 +2176,48 @@
 	pmv[0].x = pmv[0].y = 0;
 
 	CheckCandidate32I(0, 0, 255, &i, Data);
+	Data->iMinSAD[1] -= 50;
+	Data->iMinSAD[2] -= 50;
+	Data->iMinSAD[3] -= 50;
+	Data->iMinSAD[4] -= 50;
 
-	if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP * 4) {
+	if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) {
 
 		if (!(mask = make_mask(pmv, 1)))
 			CheckCandidate32I(pmv[1].x, pmv[1].y, mask, &i, Data);
 		if (!(mask = make_mask(pmv, 2)))
 			CheckCandidate32I(pmv[2].x, pmv[2].y, mask, &i, Data);
 
-		if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP * 4) // diamond only if needed
+		if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) // diamond only if needed
 			DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i);
+	}
 
-		for (i = 0; i < 4; i++) {
-			MACROBLOCK * MB = &pMBs[x + (i&1) + (y+(i>>1)) * pParam->mb_width];
-			MB->mvs[0] = MB->mvs[1] = MB->mvs[2] = MB->mvs[3] = Data->currentMV[i];
-			MB->mode = MODE_INTER;
-			MB->sad16 = Data->iMinSAD[i+1];
-		}
+	for (i = 0; i < 4; i++) {
+		MACROBLOCK * MB = &pMBs[x + (i&1) + (y+(i>>1)) * pParam->mb_width];
+		MB->mvs[0] = MB->mvs[1] = MB->mvs[2] = MB->mvs[3] = Data->currentMV[i];
+		MB->mode = MODE_INTER;
+		MB->sad16 = Data->iMinSAD[i+1];
 	}
 }
 
-#define INTRA_BIAS		2500
-#define INTRA_THRESH	1500
-#define INTER_THRESH	1400
+#define INTRA_THRESH	2400
+#define INTER_THRESH	1100
 
 int
 MEanalysis(	const IMAGE * const pRef,
-			FRAMEINFO * const Current,
-			MBParam * const pParam,
-			int maxIntra, //maximum number if non-I frames
-			int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame
-			int bCount) // number of B frames in a row
+			const FRAMEINFO * const Current,
+			const MBParam * const pParam,
+			const int maxIntra, //maximum number if non-I frames
+			const int intraCount, //number of non-I frames after last I frame; 0 if we force P/B frame
+			const int bCount,  // number of B frames in a row
+			const int b_thresh)
 {
 	uint32_t x, y, intra = 0;
 	int sSAD = 0;
 	MACROBLOCK * const pMBs = Current->mbs;
 	const IMAGE * const pCurrent = &Current->image;
-	int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH;
+	int IntraThresh = INTRA_THRESH, InterThresh = INTER_THRESH + 10*b_thresh;
+	int s = 0, blocks = 0;
 
 	int32_t iMinSAD[5], temp[5];
 	VECTOR currentMV[5];
@@ -2196,26 +2226,32 @@
 	Data.currentMV = currentMV;
 	Data.iMinSAD = iMinSAD;
 	Data.iFcode = Current->fcode;
-	Data.rrv = Current->global_flags & XVID_REDUCED;
 	Data.temp = temp;
 	CheckCandidate = CheckCandidate32I;
 
 	if (intraCount != 0 && intraCount < 10) // we're right after an I frame
-		IntraThresh += 4 * (intraCount - 10) * (intraCount - 10);
+		IntraThresh += 8 * (intraCount - 10) * (intraCount - 10);
 	else
 		if ( 5*(maxIntra - intraCount) < maxIntra) // we're close to maximum. 2 sec when max is 10 sec
 			IntraThresh -= (IntraThresh * (maxIntra - 5*(maxIntra - intraCount)))/maxIntra;
 
-	InterThresh += 400 * (1 - bCount);
-	if (InterThresh < 300) InterThresh = 300;
+	InterThresh -= (350 - 8*b_thresh) * bCount;
+	if (InterThresh < 300 + 5*b_thresh) InterThresh = 300 + 5*b_thresh;
 
 	if (sadInit) (*sadInit) ();
 
 	for (y = 1; y < pParam->mb_height-1; y += 2) {
 		for (x = 1; x < pParam->mb_width-1; x += 2) {
 			int i;
+			blocks += 4;
 
 			if (bCount == 0) pMBs[x + y * pParam->mb_width].mvs[0] = zeroMV;
+			else { //extrapolation of the vector found for last frame
+				pMBs[x + y * pParam->mb_width].mvs[0].x = 
+					(pMBs[x + y * pParam->mb_width].mvs[0].x * (bCount+1) ) / bCount;
+				pMBs[x + y * pParam->mb_width].mvs[0].y = 
+					(pMBs[x + y * pParam->mb_width].mvs[0].y * (bCount+1) ) / bCount;
+			}
 
 			MEanalyzeMB(pRef->y, pCurrent->y, x, y, pParam, pMBs, &Data);
 
@@ -2227,19 +2263,24 @@
 									pParam->edged_width);
 					if (dev + IntraThresh < pMB->sad16) {
 						pMB->mode = MODE_INTRA;
-						if (++intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return I_VOP;
+						if (++intra > ((pParam->mb_height-2)*(pParam->mb_width-2))/2) return I_VOP;
 					}
 				}
+				if (pMB->mvs[0].x == 0 && pMB->mvs[0].y == 0) s++;
+
 				sSAD += pMB->sad16;
 			}
 		}
 	}
-	sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2);
-//	if (sSAD > IntraThresh + INTRA_BIAS) return I_VOP;
+
+	sSAD /= blocks;
+	s = (10*s) / blocks;
+
+	if (s > 5) sSAD += (s - 4) * (180 - 2*b_thresh); //static block - looks bad when in bframe...
+
 	if (sSAD > InterThresh ) return P_VOP;
 	emms();
 	return B_VOP;
-
 }
 
 
@@ -2271,12 +2312,16 @@
 	double meanx,meany;
 	int num,oldnum;
 
-	if (!MBmask) { fprintf(stderr,"Mem error\n"); return gmc;}
+	if (!MBmask) { fprintf(stderr,"Mem error\n"); 
+			       gmc.duv[0].x= gmc.duv[0].y = 
+				   		gmc.duv[1].x= gmc.duv[1].y = 			
+						gmc.duv[2].x= gmc.duv[2].y = 0;
+					return gmc; }
 
 // filter mask of all blocks
 
-	for (my = 1; my < MBh-1; my++)
-	for (mx = 1; mx < MBw-1; mx++)
+	for (my = 1; my < (uint32_t)MBh-1; my++)
+	for (mx = 1; mx < (uint32_t)MBw-1; mx++)
 	{
 		const int mbnum = mx + my * MBw;
 		const MACROBLOCK *pMB = &pMBs[mbnum];
@@ -2292,8 +2337,8 @@
 			MBmask[mbnum]=1;
 	}
 
-	for (my = 1; my < MBh-1; my++)
-	for (mx = 1; mx < MBw-1; mx++)
+	for (my = 1; my < (uint32_t)MBh-1; my++)
+	for (mx = 1; mx < (uint32_t)MBw-1; mx++)
 	{
 		const uint8_t *const pCur = current->image.y + 16*my*pParam->edged_width + 16*mx;
 
@@ -2301,9 +2346,9 @@
 		if (!MBmask[mbnum])
 			continue;
 
-		if (sad16 ( pCur, pCur+1 , pParam->edged_width, 65536) <= grad )
+		if (sad16 ( pCur, pCur+1 , pParam->edged_width, 65536) <= (uint32_t)grad )
 			MBmask[mbnum] = 0;
-		if (sad16 ( pCur, pCur+pParam->edged_width, pParam->edged_width, 65536) <= grad )
+		if (sad16 ( pCur, pCur+pParam->edged_width, pParam->edged_width, 65536) <= (uint32_t)grad )
 			MBmask[mbnum] = 0;
 
 	}
@@ -2314,8 +2359,8 @@
 
 	a = b = c = n = 0;
 	DtimesF[0] = DtimesF[1] = DtimesF[2] = DtimesF[3] = 0.;
-	for (my = 0; my < MBh; my++)
-		for (mx = 0; mx < MBw; mx++)
+	for (my = 0; my < (uint32_t)MBh; my++)
+		for (mx = 0; mx < (uint32_t)MBw; mx++)
 		{
 			const int mbnum = mx + my * MBw;
 			const MACROBLOCK *pMB = &pMBs[mbnum];
@@ -2352,8 +2397,8 @@
 
 	meanx = meany = 0.;
 	oldnum = 0;
-	for (my = 0; my < MBh; my++)
-		for (mx = 0; mx < MBw; mx++)
+	for (my = 0; my < (uint32_t)MBh; my++)
+		for (mx = 0; mx < (uint32_t)MBw; mx++)
 		{
 			const int mbnum = mx + my * MBw;
 			const MACROBLOCK *pMB = &pMBs[mbnum];
@@ -2381,8 +2426,8 @@
 	fprintf(stderr,"meanx = %8.5f  meany = %8.5f   %d\n",meanx,meany, oldnum);
 */
 	num = 0;
-	for (my = 0; my < MBh; my++)
-		for (mx = 0; mx < MBw; mx++)
+	for (my = 0; my < (uint32_t)MBh; my++)
+		for (mx = 0; mx < (uint32_t)MBw; mx++)
 		{
 			const int mbnum = mx + my * MBw;
 			const MACROBLOCK *pMB = &pMBs[mbnum];
@@ -2446,7 +2491,7 @@
 		if (Data->temp[0] == 0 && Data->temp[1] == 0 && Data->temp[2] == 0 && Data->temp[3] == 0)
 			return 0; //quick stop
 
-		if (MotionFlags & (HALFPELREFINE16_BITS | EXTSEARCH_BITS)) { //we have to prepare for halfpixel-precision search
+		if (MotionFlags & (XVID_ME_HALFPELREFINE16_BITS | XVID_ME_EXTSEARCH_BITS)) { //we have to prepare for halfpixel-precision search
 			for(i = 0; i < 5; i++) bsad[i] = Data->iMinSAD[i];
 			get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
 						pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv);
@@ -2464,12 +2509,12 @@
 		}
 	}
 
-	if (MotionFlags&EXTSEARCH_BITS) SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
+	if (MotionFlags&XVID_ME_EXTSEARCH_BITS) SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
 
-	if (MotionFlags&HALFPELREFINE16_BITS) SubpelRefine(Data);
+	if (MotionFlags&XVID_ME_HALFPELREFINE16_BITS) SubpelRefine(Data);
 
 	if (Data->qpel) {
-		if (MotionFlags&(EXTSEARCH_BITS | HALFPELREFINE16_BITS)) { // there was halfpel-precision search
+		if (MotionFlags&(XVID_ME_EXTSEARCH_BITS | XVID_ME_HALFPELREFINE16_BITS)) { // there was halfpel-precision search
 			for(i = 0; i < 5; i++) if (bsad[i] > Data->iMinSAD[i]) {
 				Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // we have found a better match
 				Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
@@ -2480,10 +2525,10 @@
 			get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
 					pParam->width, pParam->height, Data->iFcode, 1, 0);
 		}
-		if (MotionFlags&QUARTERPELREFINE16_BITS) SubpelRefine(Data);
+		if (MotionFlags&XVID_ME_QUARTERPELREFINE16_BITS) SubpelRefine(Data);
 	}
 
-	if (MotionFlags&CHECKPREDICTION_BITS) { //let's check vector equal to prediction
+	if (MotionFlags&XVID_ME_CHECKPREDICTION_BITS) { //let's check vector equal to prediction
 		VECTOR * v = Data->qpel ? Data->currentQMV : Data->currentMV;
 		if (!(Data->predMV.x == v->x && Data->predMV.y == v->y))
 			CheckCandidateBits16(Data->predMV.x, Data->predMV.y, 255, &iDirection, Data);
@@ -2544,7 +2589,7 @@
 		}
 
 		if (Data8->qpel) {
-			if (MotionFlags&HALFPELREFINE8_BITS || (MotionFlags&PMV_EXTSEARCH8 && MotionFlags&EXTSEARCH_BITS)) { // halfpixel motion search follows
+			if (MotionFlags&XVID_ME_HALFPELREFINE8_BITS || (MotionFlags&XVID_ME_EXTSEARCH8 && MotionFlags&XVID_ME_EXTSEARCH_BITS)) { // halfpixel motion search follows
 				int32_t s = *Data8->iMinSAD;
 				Data8->currentMV->x = Data8->currentQMV->x/2;
 				Data8->currentMV->y = Data8->currentQMV->y/2;
@@ -2555,10 +2600,10 @@
 				if (Data8->currentQMV->x & 1 || Data8->currentQMV->y & 1)
 					CheckCandidateBits8(Data8->currentMV->x, Data8->currentMV->y, 255, &iDirection, Data8);
 
-				if (MotionFlags & PMV_EXTSEARCH8 && MotionFlags & EXTSEARCH_BITS)
+				if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_BITS)
 					SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255);
 
-				if (MotionFlags & HALFPELREFINE8_BITS) SubpelRefine(Data8);
+				if (MotionFlags & XVID_ME_HALFPELREFINE8_BITS) SubpelRefine(Data8);
 
 				if(s > *Data8->iMinSAD) { //we have found a better match
 					Data8->currentQMV->x = 2*Data8->currentMV->x;
@@ -2570,13 +2615,13 @@
 							pParam->width, pParam->height, Data8->iFcode, 1, 0);
 
 			}
-			if (MotionFlags & QUARTERPELREFINE8_BITS) SubpelRefine(Data8);
+			if (MotionFlags & XVID_ME_QUARTERPELREFINE8_BITS) SubpelRefine(Data8);
 
 		} else // not qpel
-			if (MotionFlags & HALFPELREFINE8_BITS) SubpelRefine(Data8); //halfpel mode, halfpel refinement
+			if (MotionFlags & XVID_ME_HALFPELREFINE8_BITS) SubpelRefine(Data8); //halfpel mode, halfpel refinement
 
 		//checking vector equal to predicion
-		if (i != 0 && MotionFlags & CHECKPREDICTION_BITS) {
+		if (i != 0 && MotionFlags & XVID_ME_CHECKPREDICTION_BITS) {
 			const VECTOR * v = Data->qpel ? Data8->currentQMV : Data8->currentMV;
 			if (!(Data8->predMV.x == v->x && Data8->predMV.y == v->y))
 				CheckCandidateBits8(Data8->predMV.x, Data8->predMV.y, 255, &iDirection, Data8);
@@ -2630,7 +2675,7 @@
 				bits += CodeCoeffInter_CalcBits(coeff, scan_tables[0]);
 				cbp |= 1 << (5 - 5);
 			}
-			bits += cbpy_tab[15-(cbp>>2)].len;
+			bits += xvid_cbpy_tab[15-(cbp>>2)].len;
 			bits += mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len;
 		}
 	}
@@ -2680,9 +2725,9 @@
 
 		bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcc_tab[coeff[0] + 255].len;
 		if (t != 0) cbp |= 1 << (5 - 4);
-		Data->temp[4] = t;
 
 		if (bits < Data->iMinSAD[0]) {
+			iDcScaler = get_dc_scaler(iQuant, 1);
 			//chroma V
 			transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2);
 			fdct(in);
@@ -2693,16 +2738,9 @@
 			bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcc_tab[coeff[0] + 255].len;
 			if (t != 0) cbp |= 1 << (5 - 5);
 
-			Data->temp[5] = t;
-
-			bits += t = cbpy_tab[cbp>>2].len;
-			Data->temp[6] = t;
-
-			bits += t = mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp & 3) << 3)].len;
-			Data->temp[7] = t;
-
+			bits += xvid_cbpy_tab[cbp>>2].len;
+			bits += mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp & 3) << 3)].len;
 		}
 	}
-
 	return bits;
 }