--- trunk/xvidcore/src/motion/motion_est.c	2003/02/19 21:59:30	876
+++ trunk/xvidcore/src/motion/motion_est.c	2003/04/05 15:54:59	973
@@ -81,30 +81,28 @@
 static __inline uint32_t
 d_mv_bits(int x, int y, const VECTOR pred, const uint32_t iFcode, const int qpel, const int rrv)
 {
-	int xb, yb;
-	x = qpel ? x<<1 : x;
-	y = qpel ? y<<1 : y;
+	int bits;
+	const int q = (1 << (iFcode - 1)) - 1;
+
+	x <<= qpel;
+	y <<= qpel;
 	if (rrv) { x = RRV_MV_SCALEDOWN(x); y = RRV_MV_SCALEDOWN(y); }
 
 	x -= pred.x;
+	bits = (x != 0 ? iFcode:0);
+	x = abs(x);
+	x += q;
+	x >>= (iFcode - 1);
+	bits += mvtab[x];
+
 	y -= pred.y;
+	bits += (y != 0 ? iFcode:0);
+	y = abs(y);
+	y += q;
+	y >>= (iFcode - 1);
+	bits += mvtab[y];
 
-	if (x) {
-		x = ABS(x);
-		x += (1 << (iFcode - 1)) - 1;
-		x >>= (iFcode - 1);
-		if (x > 32) x = 32;
-		xb = mvtab[x] + iFcode;
-	} else xb = 1;
-
-	if (y) {
-		y = ABS(y);
-		y += (1 << (iFcode - 1)) - 1;
-		y >>= (iFcode - 1);
-		if (y > 32) y = 32;
-		yb = mvtab[y] + iFcode;
-	} else yb = 1;
-	return xb + yb;
+	return bits;
 }
 
 static int32_t ChromaSAD2(int fx, int fy, int bx, int by, const SearchData * const data)
@@ -249,8 +247,15 @@
 	ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data);
 	ref1 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
 	switch( ((x&1)<<1) + (y&1) ) {
-	case 0: // pure halfpel position
-		return (uint8_t *) ref1;
+	case 3: // x and y in qpel resolution - the "corners" (top left/right and
+			// bottom left/right) during qpel refinement
+		ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
+		ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
+		ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data);
+		ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
+		ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
+		ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
+		interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
 		break;
 
 	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
@@ -265,16 +270,9 @@
 		interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding, 8);
 		break;
 
-	default: // x and y in qpel resolution - the "corners" (top left/right and
-			 // bottom left/right) during qpel refinement
-		ref2 = GetReferenceB(halfpel_x, y - halfpel_y, dir, data);
-		ref3 = GetReferenceB(x - halfpel_x, halfpel_y, dir, data);
-		ref4 = GetReferenceB(x - halfpel_x, y - halfpel_y, dir, data);
-		ref2 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
-		ref3 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
-		ref4 += 8 * (block&1) + 8 * (block>>1) * iEdgedWidth;
-		interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
-		break;
+	default: // pure halfpel position
+		return (uint8_t *) ref1;
+	
 	}
 	return Reference;
 }
@@ -319,7 +317,7 @@
 		interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
 		break;
 
-	case 0: // pure halfpel position
+	default: // pure halfpel position
 		return (uint8_t *) ref1;
 	}
 	return Reference;
@@ -355,7 +353,7 @@
 	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;
@@ -379,12 +377,18 @@
 {
 	int32_t sad; uint32_t t;
 	const uint8_t * Reference;
+	VECTOR * current;
 
 	if ( (x > data->max_dx) || (x < data->min_dx)
 		|| (y > data->max_dy) || (y < data->min_dy) ) return;
 
-	if (!data->qpel_precision) Reference = GetReference(x, y, data);
-	else Reference = Interpolate8x8qpel(x, y, 0, 0, data);
+	if (!data->qpel_precision) {
+		Reference = GetReference(x, y, data);
+		current = data->currentMV;
+	} else { // x and y are in 1/4 precision
+		Reference = Interpolate8x8qpel(x, y, 0, 0, data);
+		current = data->currentQMV;
+	}
 
 	sad = sad8(data->Cur, Reference, data->iEdgedWidth);
 	t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
@@ -393,7 +397,7 @@
 
 	if (sad < *(data->iMinSAD)) {
 		*(data->iMinSAD) = sad;
-		data->currentMV->x = x; data->currentMV->y = y;
+		current->x = x; current->y = y;
 		*dir = Direction;
 	}
 }
@@ -405,7 +409,7 @@
 	uint32_t t;
 	const uint8_t * Reference;
 
-	if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) || //non-zero integer value
+	if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) || //non-zero even value
 		(x > data->max_dx) || (x < data->min_dx)
 		|| (y > data->max_dy) || (y < data->min_dy) ) return;
 
@@ -440,8 +444,8 @@
 	uint32_t t;
 	VECTOR * current;
 
-	if ( (x > data->max_dx) | ( x < data->min_dx)
-		| (y > data->max_dy) | (y < data->min_dy) ) return;
+	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
 
@@ -506,8 +510,9 @@
 	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);
@@ -552,7 +557,7 @@
 	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;
@@ -565,10 +570,11 @@
 			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;
@@ -610,7 +616,7 @@
 	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;
 
 	mvs.x = data->directmvF[0].x + x;
 	b_mvs.x = ((x == 0) ?
@@ -622,10 +628,10 @@
 		data->directmvB[0].y
 		: mvs.y - data->referencemv[0].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 = 4*(mvs.x/2); ycf = 4*(mvs.y/2);
@@ -659,7 +665,7 @@
 CheckCandidateBits16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
 {
 
-	static int16_t in[64], coeff[64];
+	int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64;
 	int32_t bits = 0, sum;
 	VECTOR * current;
 	const uint8_t * ptr;
@@ -721,7 +727,7 @@
 		}
 	}
 
-	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]) {
@@ -744,7 +750,7 @@
 CheckCandidateBits8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
 {
 
-	static int16_t in[64], coeff[64];
+	int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64;
 	int32_t sum, bits;
 	VECTOR * current;
 	const uint8_t * ptr;
@@ -950,21 +956,22 @@
 							const uint32_t stride, const uint32_t iQuant, int rrv)
 
 {
+	int offset = (x + y*stride)*8;
 	if(!rrv) {
-		uint32_t sadC = sad8(current->u + x*8 + y*stride*8,
-						reference->u + x*8 + y*stride*8, stride);
+		uint32_t sadC = sad8(current->u + offset,
+						reference->u + offset, stride);
 		if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
-		sadC += sad8(current->v + (x + y*stride)*8,
-						reference->v + (x + y*stride)*8, stride);
+		sadC += sad8(current->v + offset,
+						reference->v + offset, stride);
 		if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
 		return 1;
 
 	} else {
-		uint32_t sadC = sad16(current->u + x*16 + y*stride*16,
-						reference->u + x*16 + y*stride*16, stride, 256*4096);
+		uint32_t sadC = sad16(current->u + 2*offset,
+						reference->u + 2*offset, stride, 256*4096);
 		if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
-		sadC += sad16(current->v + (x + y*stride)*16,
-						reference->v + (x + y*stride)*16, stride, 256*4096);
+		sadC += sad16(current->v + 2*offset,
+						reference->v + 2*offset, stride, 256*4096);
 		if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP*4) return 0;
 		return 1;
 	}
@@ -1000,12 +1007,16 @@
 	uint32_t x, y;
 	uint32_t iIntra = 0;
 	int32_t quant = current->quant, sad00;
+	int skip_thresh = INITIAL_SKIP_THRESH * 
+		(current->global_flags & XVID_REDUCED ? 4:1) * 
+		(current->global_flags & XVID_MODEDECISION_BITS ? 2:1);
 
 	// some pre-initialized thingies for SearchP
 	int32_t temp[8];
 	VECTOR currentMV[5];
 	VECTOR currentQMV[5];
 	int32_t iMinSAD[5];
+	DECLARE_ALIGNED_MATRIX(dct_space, 2, 64, int16_t, CACHE_LINE);
 	SearchData Data;
 	memset(&Data, 0, sizeof(SearchData));
 	Data.iEdgedWidth = iEdgedWidth;
@@ -1018,6 +1029,7 @@
 	Data.qpel = pParam->m_quarterpel;
 	Data.chroma = MotionFlags & PMV_CHROMA16;
 	Data.rrv = current->global_flags & XVID_REDUCED;
+	Data.dctSpace = dct_space;
 
 	if ((current->global_flags & XVID_REDUCED)) {
 		mb_width = (pParam->width + 31) / 32;
@@ -1066,7 +1078,7 @@
 //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 (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * skip_thresh)
 					if (Data.chroma || SkipDecisionP(pCurrent, pRef, x, y, iEdgedWidth/2, pMB->quant, Data.rrv)) {
 						SkipMacroblockP(pMB, sad00);
 						continue;
@@ -1079,17 +1091,11 @@
 						current->global_flags & XVID_INTER4V, pMB);
 
 /* final skip decision, a.k.a. "the vector you found, really that good?" */
-			if (!(current->global_flags & XVID_GMC))	{
+			if (!(current->global_flags & XVID_GMC || current->global_flags & XVID_MODEDECISION_BITS)) {
 				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 ( (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)
@@ -1178,13 +1184,12 @@
 	int mode = MODE_INTER;
 
 	if (!(GlobalFlags & XVID_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] +
 			Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant) {
-				mode = 0; //inter
-				sad = Data->iMinSAD[0];
+			mode = MODE_INTER;
+			sad = Data->iMinSAD[0];
 		} else {
 			mode = MODE_INTER4V;
 			sad = Data->iMinSAD[1] + Data->iMinSAD[2] +
@@ -1211,7 +1216,7 @@
 				dev16(Data->Cur + 8*Data->iEdgedWidth, Data->iEdgedWidth) +
 				dev16(Data->Cur+8+8*Data->iEdgedWidth, Data->iEdgedWidth);
 
-			if (deviation < (sad - InterBias))  return MODE_INTRA;// intra
+			if (deviation < (sad - InterBias)) return MODE_INTRA;
 		}
 		return mode;
 
@@ -1232,8 +1237,8 @@
 		if (bits == 0) return MODE_INTER; // quick stop
 
 		if (inter4v) {
-			int inter4v = CountMBBitsInter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup);
-			if (inter4v < bits) { Data->iMinSAD[0] = bits = inter4v; mode = MODE_INTER4V; }
+			int bits_inter4v = CountMBBitsInter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup);
+			if (bits_inter4v < bits) { Data->iMinSAD[0] = bits = bits_inter4v; mode = MODE_INTER4V; }
 		}
 
 
@@ -1291,8 +1296,7 @@
 
 	if (pMB->dquant != NO_CHANGE) inter4v = 0;
 
-	for(i = 0; i < 5; i++)
-		Data->currentMV[i].x = Data->currentMV[i].y = 0;
+	memset(Data->currentMV, 0, 5*sizeof(VECTOR));
 
 	if (Data->qpel) Data->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
 	else Data->predMV = pmv[0];
@@ -1384,14 +1388,16 @@
 		Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
 	}
 
-	if (MotionFlags & PMV_QUARTERPELREFINE16)
+	if (MotionFlags & PMV_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 & 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);
-
 			SubpelRefine(Data);
 		}
+	}
 
 	if ((!(GlobalFlags & XVID_MODEDECISION_BITS)) && (Data->iMinSAD[0] < (int32_t)iQuant * 30)) inter4v = 0;
 
@@ -1712,9 +1718,9 @@
 
 	for (k = 0; k < 4; k++) {
 		dy += Data->directmvF[k].y / div;
-		dx += Data->directmvF[0].x / div;
-		b_dy += Data->directmvB[0].y / div;
-		b_dx += Data->directmvB[0].x / div;
+		dx += Data->directmvF[k].x / div;
+		b_dy += Data->directmvB[k].y / div;
+		b_dx += Data->directmvB[k].x / div;
 	}
 
 	dy = (dy >> 3) + roundtab_76[dy & 0xf];
@@ -1734,7 +1740,13 @@
 					b_Ref->v + (y*8 + b_dy/2) * stride + x*8 + b_dx/2,
 					stride);
 
-	if (sum < 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) pMB->mode = MODE_DIRECT_NONE_MV; //skipped
+	if (sum < 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) {
+		pMB->mode = MODE_DIRECT_NONE_MV; //skipped
+		for (k = 0; k < 4; k++) { 
+			pMB->qmvs[k] = pMB->mvs[k];
+			pMB->b_qmvs[k] = pMB->b_mvs[k];
+		}
+	}
 }
 
 static __inline uint32_t
@@ -1823,6 +1835,7 @@
 		}
 	}
 
+	*Data->iMinSAD += Data->lambda16;
 	skip_sad = *Data->iMinSAD;
 
 //	DIRECT MODE DELTA VECTOR SEARCH.
@@ -2149,7 +2162,7 @@
 
 	int i, mask;
 	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;
 
@@ -2163,7 +2176,7 @@
 			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 - pParam->m_quarterpel, 0, 0);
 
 	Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
 	Data->Ref = pRef + (x + y * pParam->edged_width) * 16;
@@ -2175,43 +2188,48 @@
 	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];
@@ -2220,26 +2238,32 @@
 	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);
 
@@ -2251,19 +2275,24 @@
 									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;
-
 }
 
 
@@ -2531,7 +2560,7 @@
 	int cbp = 0, bits = 0, t = 0, i, iDirection;
 	SearchData Data2, *Data8 = &Data2;
 	int sumx = 0, sumy = 0;
-	int16_t in[64], coeff[64];
+	int16_t *in = Data->dctSpace, *coeff = Data->dctSpace + 64;
 
 	memcpy(Data8, Data, sizeof(SearchData));
 	CheckCandidate = CheckCandidateBits8;
@@ -2658,7 +2687,7 @@
 				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;
 		}
 	}
@@ -2673,7 +2702,7 @@
 	int bits = 1; //this one is ac/dc prediction flag. always 1.
 	int cbp = 0, i, t, dc = 0, b_dc = 1024;
 	const uint32_t iQuant = Data->lambda16;
-	int16_t in[64], coeff[64];
+	int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64;
 
 	for(i = 0; i < 4; i++) {
 		uint32_t iDcScaler = get_dc_scaler(iQuant, 1);
@@ -2708,9 +2737,9 @@
 
 		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);
@@ -2721,16 +2750,9 @@
 			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;
 }