--- branches/dev-api-4/xvidcore/src/motion/motion_est.c	2003/03/29 01:29:53	958
+++ branches/dev-api-4/xvidcore/src/motion/motion_est.c	2003/05/14 18:46:39	1022
@@ -81,33 +81,32 @@
 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)
+static int32_t ChromaSAD2(const int fx, const int fy, const int bx, const int by,
+							const SearchData * const data)
 {
 	int sad;
 	const uint32_t stride = data->iEdgedWidth/2;
@@ -115,50 +114,44 @@
 		* f_refv = data->RefQ + 8,
 		* b_refu = data->RefQ + 16,
 		* b_refv = data->RefQ + 24;
+	int offset = (fx>>1) + (fy>>1)*stride;
 
 	switch (((fx & 1) << 1) | (fy & 1))	{
 		case 0:
-			fx = fx / 2; fy = fy / 2;
-			f_refu = (uint8_t*)data->RefCU + fy * stride + fx, stride;
-			f_refv = (uint8_t*)data->RefCV + fy * stride + fx, stride;
+			f_refu = (uint8_t*)data->RefP[4] + offset;
+			f_refv = (uint8_t*)data->RefP[5] + offset;
 			break;
 		case 1:
-			fx = fx / 2; fy = (fy - 1) / 2;
-			interpolate8x8_halfpel_v(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding);
-			interpolate8x8_halfpel_v(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding);
+			interpolate8x8_halfpel_v(f_refu, data->RefP[4] + offset, stride, data->rounding);
+			interpolate8x8_halfpel_v(f_refv, data->RefP[5] + offset, stride, data->rounding);
 			break;
 		case 2:
-			fx = (fx - 1) / 2; fy = fy / 2;
-			interpolate8x8_halfpel_h(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding);
-			interpolate8x8_halfpel_h(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding);
+			interpolate8x8_halfpel_h(f_refu, data->RefP[4] + offset, stride, data->rounding);
+			interpolate8x8_halfpel_h(f_refv, data->RefP[5] + offset, stride, data->rounding);
 			break;
 		default:
-			fx = (fx - 1) / 2; fy = (fy - 1) / 2;
-			interpolate8x8_halfpel_hv(f_refu, data->RefCU + fy * stride + fx, stride, data->rounding);
-			interpolate8x8_halfpel_hv(f_refv, data->RefCV + fy * stride + fx, stride, data->rounding);
+			interpolate8x8_halfpel_hv(f_refu, data->RefP[4] + offset, stride, data->rounding);
+			interpolate8x8_halfpel_hv(f_refv, data->RefP[5] + offset, stride, data->rounding);
 			break;
 	}
 
+	offset = (bx>>1) + (by>>1)*stride;
 	switch (((bx & 1) << 1) | (by & 1))	{
 		case 0:
-			bx = bx / 2; by = by / 2;
-			b_refu = (uint8_t*)data->b_RefCU + by * stride + bx, stride;
-			b_refv = (uint8_t*)data->b_RefCV + by * stride + bx, stride;
+			b_refu = (uint8_t*)data->b_RefP[4] + offset;
+			b_refv = (uint8_t*)data->b_RefP[5] + offset;
 			break;
 		case 1:
-			bx = bx / 2; by = (by - 1) / 2;
-			interpolate8x8_halfpel_v(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding);
-			interpolate8x8_halfpel_v(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding);
+			interpolate8x8_halfpel_v(b_refu, data->b_RefP[4] + offset, stride, data->rounding);
+			interpolate8x8_halfpel_v(b_refv, data->b_RefP[5] + offset, stride, data->rounding);
 			break;
 		case 2:
-			bx = (bx - 1) / 2; by = by / 2;
-			interpolate8x8_halfpel_h(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding);
-			interpolate8x8_halfpel_h(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding);
+			interpolate8x8_halfpel_h(b_refu, data->b_RefP[4] + offset, stride, data->rounding);
+			interpolate8x8_halfpel_h(b_refv, data->b_RefP[5] + offset, stride, data->rounding);
 			break;
 		default:
-			bx = (bx - 1) / 2; by = (by - 1) / 2;
-			interpolate8x8_halfpel_hv(b_refu, data->b_RefCU + by * stride + bx, stride, data->rounding);
-			interpolate8x8_halfpel_hv(b_refv, data->b_RefCV + by * stride + bx, stride, data->rounding);
+			interpolate8x8_halfpel_hv(b_refu, data->b_RefP[4] + offset, stride, data->rounding);
+			interpolate8x8_halfpel_hv(b_refv, data->b_RefP[5] + offset, stride, data->rounding);
 			break;
 	}
 
@@ -168,38 +161,34 @@
 	return sad;
 }
 
-
 static int32_t
-ChromaSAD(int dx, int dy, const SearchData * const data)
+ChromaSAD(const int dx, const int dy, const SearchData * const data)
 {
 	int sad;
 	const uint32_t stride = data->iEdgedWidth/2;
+	int offset = (dx>>1) + (dy>>1)*stride;
 
 	if (dx == data->temp[5] && dy == data->temp[6]) return data->temp[7]; //it has been checked recently
 	data->temp[5] = dx; data->temp[6] = dy; // backup
 
 	switch (((dx & 1) << 1) | (dy & 1))	{
 		case 0:
-			dx = dx / 2; dy = dy / 2;
-			sad = sad8(data->CurU, data->RefCU + dy * stride + dx, stride);
-			sad += sad8(data->CurV, data->RefCV + dy * stride + dx, stride);
+			sad = sad8(data->CurU, data->RefP[4] + offset, stride);
+			sad += sad8(data->CurV, data->RefP[5] + offset, stride);
 			break;
 		case 1:
-			dx = dx / 2; dy = (dy - 1) / 2;
-			sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + (dy+1) * stride + dx, stride);
-			sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + (dy+1) * stride + dx, stride);
+			sad = sad8bi(data->CurU, data->RefP[4] + offset, data->RefP[4] + offset + stride, stride);
+			sad += sad8bi(data->CurV, data->RefP[5] + offset, data->RefP[5] + offset + stride, stride);
 			break;
 		case 2:
-			dx = (dx - 1) / 2; dy = dy / 2;
-			sad = sad8bi(data->CurU, data->RefCU + dy * stride + dx, data->RefCU + dy * stride + dx+1, stride);
-			sad += sad8bi(data->CurV, data->RefCV + dy * stride + dx, data->RefCV + dy * stride + dx+1, stride);
+			sad = sad8bi(data->CurU, data->RefP[4] + offset, data->RefP[4] + offset + 1, stride);
+			sad += sad8bi(data->CurV, data->RefP[5] + offset, data->RefP[5] + offset + 1, stride);
 			break;
 		default:
-			dx = (dx - 1) / 2; dy = (dy - 1) / 2;
-			interpolate8x8_halfpel_hv(data->RefQ, data->RefCU + dy * stride + dx, stride, data->rounding);
+			interpolate8x8_halfpel_hv(data->RefQ, data->RefP[4] + offset, stride, data->rounding);
 			sad = sad8(data->CurU, data->RefQ, stride);
 
-			interpolate8x8_halfpel_hv(data->RefQ, data->RefCV + dy * stride + dx, stride, data->rounding);
+			interpolate8x8_halfpel_hv(data->RefQ, data->RefP[5] + offset, stride, data->rounding);
 			sad += sad8(data->CurV, data->RefQ, stride);
 			break;
 	}
@@ -211,28 +200,19 @@
 GetReferenceB(const int x, const int y, const uint32_t dir, const SearchData * const data)
 {
 //	dir : 0 = forward, 1 = backward
-	switch ( (dir << 2) | ((x&1)<<1) | (y&1) ) {
-		case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth);
-		case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
-		case 2 : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
-		case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
-		case 4 : return data->bRef + x/2 + (y/2)*(data->iEdgedWidth);
-		case 5 : return data->bRefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
-		case 6 : return data->bRefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);
-		default : return data->bRefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
-	}
+	const uint8_t *const *const direction = ( dir == 0 ? data->RefP : data->b_RefP );
+	const int picture = ((x&1)<<1) | (y&1);
+	const int offset = (x>>1) + (y>>1)*data->iEdgedWidth;
+	return direction[picture] + offset;
 }
 
 // this is a simpler copy of GetReferenceB, but as it's __inline anyway, we can keep the two separate
 static __inline const uint8_t *
 GetReference(const int x, const int y, const SearchData * const data)
 {
-	switch ( ((x&1)<<1) | (y&1) ) {
-		case 0 : return data->Ref + x/2 + (y/2)*(data->iEdgedWidth);
-		case 3 : return data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth);
-		case 1 : return data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth);
-		default : return data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth);	//case 2
-	}
+	const int picture = ((x&1)<<1) | (y&1);
+	const int offset = (x>>1) + (y>>1)*data->iEdgedWidth;
+	return data->RefP[picture] + offset;
 }
 
 static uint8_t *
@@ -249,8 +229,15 @@
 	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 +252,9 @@
 		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;
 }
@@ -293,7 +273,7 @@
 	ref1 = GetReferenceB(halfpel_x, halfpel_y, dir, data);
 	switch( ((x&1)<<1) + (y&1) ) {
 	case 3: // x and y in qpel resolution - the "corners" (top left/right and
-			 // bottom left/right) during qpel refinement
+			// 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);
@@ -319,7 +299,7 @@
 		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;
@@ -371,7 +351,6 @@
 		data->iMinSAD[3] = data->temp[3]; current[3].x = x; current[3].y = y; }
 	if (data->temp[4] < data->iMinSAD[4]) {
 		data->iMinSAD[4] = data->temp[4]; current[4].x = x; current[4].y = y; }
-
 }
 
 static void
@@ -379,12 +358,18 @@
 {
 	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);
@@ -393,19 +378,18 @@
 
 	if (sad < *(data->iMinSAD)) {
 		*(data->iMinSAD) = sad;
-		data->currentMV->x = x; data->currentMV->y = y;
+		current->x = x; current->y = y;
 		*dir = Direction;
 	}
 }
 
-
 static void
 CheckCandidate32(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
 {
 	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 +424,8 @@
 	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
 
@@ -479,8 +463,8 @@
 	if ( (x > data->max_dx) || (x < data->min_dx)
 		|| (y > data->max_dy) || (y < data->min_dy) ) return;
 
-	sad = sad32v_c(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth),
-							data->iEdgedWidth, data->temp+1);
+	sad = sad32v_c(data->Cur, data->RefP[0] + (x>>1) + (y>>1)*(data->iEdgedWidth),
+					data->iEdgedWidth, data->temp+1);
 
 	if (sad < *(data->iMinSAD)) {
 		*(data->iMinSAD) = sad;
@@ -612,7 +596,7 @@
 	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) ?
@@ -624,10 +608,10 @@
 		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);
@@ -661,8 +645,8 @@
 CheckCandidateBits16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
 {
 
-	static int16_t in[64], coeff[64];
-	int32_t bits = 0, sum;
+	int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64;
+	int32_t bits = 0;
 	VECTOR * current;
 	const uint8_t * ptr;
 	int i, cbp = 0, t, xc, yc;
@@ -683,47 +667,30 @@
 	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);
-		fdct(in);
-		if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16);
-		else sum = quant4_inter(coeff, in, data->lambda16);
-		if (sum > 0) {
-			cbp |= 1 << (5 - i);
-			bits += data->temp[i] = CodeCoeffInter_CalcBits(coeff, scan_tables[0]);
-		} else data->temp[i] = 0;
+		bits += data->temp[i] = Block_CalcBits(coeff, in, data->iQuant, data->quant_type, &cbp, i, 0);
 	}
 
 	bits += t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
 
-	if (bits < data->iMinSAD[0]) { // there is still a chance, adding chroma
-		xc = (xc >> 1) + roundtab_79[xc & 0x3];
-		yc = (yc >> 1) + roundtab_79[yc & 0x3];
-
-		//chroma U
-		ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefCU, 0, 0, xc, yc,  data->iEdgedWidth/2, data->rounding);
-		transfer_8to16subro(in, ptr, data->CurU, data->iEdgedWidth/2);
-		fdct(in);
-		if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16);
-		else sum = quant4_inter(coeff, in, data->lambda16);
-		if (sum > 0) {
-			cbp |= 1 << (5 - 4);
-			bits += CodeCoeffInter_CalcBits(coeff, scan_tables[0]);
-		}
-
-		if (bits < data->iMinSAD[0]) {
-			//chroma V
-			ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefCV, 0, 0, xc, yc,  data->iEdgedWidth/2, data->rounding);
-			transfer_8to16subro(in, ptr, data->CurV, data->iEdgedWidth/2);
-			fdct(in);
-			if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16);
-			else sum = quant4_inter(coeff, in, data->lambda16);
-			if (sum > 0) {
-				cbp |= 1 << (5 - 5);
-				bits += CodeCoeffInter_CalcBits(coeff, scan_tables[0]);
-			}
-		}
-	}
-
 	bits += xvid_cbpy_tab[15-(cbp>>2)].len;
+
+	if (bits >= data->iMinSAD[0]) return;
+
+	//chroma
+	xc = (xc >> 1) + roundtab_79[xc & 0x3];
+	yc = (yc >> 1) + roundtab_79[yc & 0x3];
+
+	//chroma U
+	ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefP[4], 0, 0, xc, yc,  data->iEdgedWidth/2, data->rounding);
+	transfer_8to16subro(in, ptr, data->CurU, data->iEdgedWidth/2);
+	bits += Block_CalcBits(coeff, in, data->iQuant, data->quant_type, &cbp, 4, 0);
+	if (bits >= data->iMinSAD[0]) return;
+
+	//chroma V
+	ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefP[5], 0, 0, xc, yc,  data->iEdgedWidth/2, data->rounding);
+	transfer_8to16subro(in, ptr, data->CurV, data->iEdgedWidth/2);
+	bits += Block_CalcBits(coeff, in, data->iQuant, data->quant_type, &cbp, 5, 0);
+
 	bits += mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len;
 
 	if (bits < data->iMinSAD[0]) {
@@ -746,11 +713,11 @@
 CheckCandidateBits8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
 {
 
-	static int16_t in[64], coeff[64];
-	int32_t sum, bits;
+	int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64;
+	int32_t bits;
 	VECTOR * current;
 	const uint8_t * ptr;
-	int cbp;
+	int cbp = 0;
 
 	if ( (x > data->max_dx) || (x < data->min_dx)
 		|| (y > data->max_dy) || (y < data->min_dy) ) return;
@@ -764,15 +731,8 @@
 	}
 
 	transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth);
-	fdct(in);
-	if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16);
-	else sum = quant4_inter(coeff, in, data->lambda16);
-	if (sum > 0) {
-		bits = CodeCoeffInter_CalcBits(coeff, scan_tables[0]);
-		cbp = 1;
-	} else cbp = bits = 0;
-
-	bits += sum = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
+	bits = Block_CalcBits(coeff, in, data->iQuant, data->quant_type, &cbp, 5, 0);
+	bits += d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
 
 	if (bits < data->iMinSAD[0]) {
 		data->temp[0] = cbp;
@@ -952,21 +912,22 @@
 							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;
 	}
@@ -981,6 +942,122 @@
 	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
 }
 
+static __inline void
+ModeDecision(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 uint32_t VopFlags,
+			const uint32_t VolFlags,
+			const IMAGE * const pCurrent,
+			const IMAGE * const pRef)
+{
+	int mode = MODE_INTER;
+	int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0);
+	const uint32_t iQuant = pMB->quant;
+
+	const int skip_possible = (!(VolFlags & XVID_VOL_GMC)) && (pMB->dquant == 0);
+
+	if (!(VopFlags & XVID_VOP_MODEDECISION_BITS)) { //normal, fast, SAD-based mode decision
+		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 = MODE_INTER;
+			sad = Data->iMinSAD[0];
+		} else {
+			mode = MODE_INTER4V;
+			sad = Data->iMinSAD[1] + Data->iMinSAD[2] +
+						Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant;
+			Data->iMinSAD[0] = sad;
+		}
+
+		/* final skip decision, a.k.a. "the vector you found, really that good?" */
+		if (skip_possible && (pMB->sad16 < (int)iQuant * MAX_SAD00_FOR_SKIP))
+			if ( (100*sad)/(pMB->sad16+1) > FINAL_SKIP_THRESH)
+				if (Data->chroma || SkipDecisionP(pCurrent, pRef, x, y, Data->iEdgedWidth/2, iQuant, Data->rrv)) {
+					mode = MODE_NOT_CODED;
+					sad = 0;
+				}
+
+		/* intra decision */
+
+		if (iQuant > 8) InterBias += 100 * (iQuant - 8); // to make high quants work
+		if (y != 0)
+			if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80;
+		if (x != 0)
+			if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80;
+
+		if (Data->chroma) InterBias += 50; // dev8(chroma) ???
+		if (Data->rrv) InterBias *= 4;
+
+		if (InterBias < pMB->sad16) {
+			int32_t deviation;
+			if (!Data->rrv) deviation = dev16(Data->Cur, Data->iEdgedWidth);
+			else deviation = dev16(Data->Cur, Data->iEdgedWidth) +
+				dev16(Data->Cur+16, Data->iEdgedWidth) +
+				dev16(Data->Cur + 16*Data->iEdgedWidth, Data->iEdgedWidth) +
+				dev16(Data->Cur+16+16*Data->iEdgedWidth, Data->iEdgedWidth);
+
+			if (deviation < (sad - InterBias)) mode = MODE_INTRA;
+		}
+
+	} else { // BITS
+
+		int bits, intra, i;
+		VECTOR backup[5], *v;
+		Data->iQuant = iQuant;
+
+		v = Data->qpel ? Data->currentQMV : Data->currentMV;
+		for (i = 0; i < 5; i++) {
+			Data->iMinSAD[i] = 256*4096;
+			backup[i] = v[i];
+		}
+
+		bits = CountMBBitsInter(Data, pMBs, x, y, pParam, MotionFlags);
+		if (bits == 0)
+			mode = MODE_INTER; // quick stop
+		else {
+			if (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; }
+			}
+
+			intra = CountMBBitsIntra(Data);
+
+			if (intra < bits) { *Data->iMinSAD = bits = intra; mode = MODE_INTRA; }
+		}
+	}
+
+	if (Data->rrv) {
+			Data->currentMV[0].x = RRV_MV_SCALEDOWN(Data->currentMV[0].x);
+			Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y);
+	}
+
+	if (mode == MODE_INTER) {
+		pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
+		pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
+
+		if(Data->qpel) {
+			pMB->qmvs[0] = pMB->qmvs[1]
+				= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
+			pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x;
+			pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y;
+		} else {
+			pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
+			pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
+		}
+
+	} else if (mode == MODE_INTER4V)
+		pMB->sad16 = Data->iMinSAD[0];
+	else // INTRA, NOT_CODED
+		SkipMacroblockP(pMB, 0);
+
+	pMB->mode = mode;
+}
+
 bool
 MotionEstimation(MBParam * const pParam,
 				 FRAMEINFO * const current,
@@ -1012,6 +1089,7 @@
 	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;
@@ -1021,9 +1099,11 @@
 	Data.temp = temp;
 	Data.iFcode = current->fcode;
 	Data.rounding = pParam->m_rounding_type;
-	Data.qpel = current->vol_flags & XVID_VOL_QUARTERPEL;
+	Data.qpel = (current->vol_flags & XVID_VOL_QUARTERPEL ? 1:0);
 	Data.chroma = MotionFlags & XVID_ME_CHROMA16;
-	Data.rrv = current->vop_flags & XVID_VOP_REDUCED;
+	Data.rrv = (current->vop_flags & XVID_VOP_REDUCED ? 1:0);
+	Data.dctSpace = dct_space;
+	Data.quant_type = !(pParam->vol_flags & XVID_VOL_MPEGQUANT);
 
 	if ((current->vop_flags & XVID_VOP_REDUCED)) {
 		mb_width = (pParam->width + 31) / 32;
@@ -1063,8 +1143,7 @@
 				if (quant > 31) quant = 31;
 				else if (quant < 1) quant = 1;
 			}
-			
-			pMB->quant = current->quant;
+			pMB->quant = quant;
 
 //initial skip decision
 /* no early skip for GMC (global vector = skip vector is unknown!)  */
@@ -1077,18 +1156,13 @@
 			}
 
 			SearchP(pRef, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
-						y, MotionFlags, current->vol_flags, pMB->quant,
-						&Data, pParam, pMBs, reference->mbs,
-						current->vop_flags & XVID_VOP_INTER4V, pMB);
-
-/* final skip decision, a.k.a. "the vector you found, really that good?" */
-			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);
-				}
-			}
+					y, MotionFlags, current->vop_flags, current->vol_flags,
+					&Data, pParam, pMBs, reference->mbs, pMB);
+
+			ModeDecision(&Data, pMB, pMBs, x, y, pParam,
+						 MotionFlags, current->vop_flags, current->vol_flags,
+						 pCurrent, pRef);
+
 			if (pMB->mode == MODE_INTRA)
 				if (++iIntra > iLimit) return 1;
 		}
@@ -1161,87 +1235,6 @@
 	}
 }
 
-static int
-ModeDecision(const uint32_t iQuant, SearchData * const Data,
-		int inter4v,
-		MACROBLOCK * const pMB,
-		const MACROBLOCK * const pMBs,
-		const int x, const int y,
-		const MBParam * const pParam,
-		const uint32_t MotionFlags,
-		const uint32_t VopFlags)
-{
-
-	int mode = MODE_INTER;
-
-	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] +
-			Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant) {
-				mode = 0; //inter
-				sad = Data->iMinSAD[0];
-		} else {
-			mode = MODE_INTER4V;
-			sad = Data->iMinSAD[1] + Data->iMinSAD[2] +
-						Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant;
-			Data->iMinSAD[0] = sad;
-		}
-
-		/* intra decision */
-
-		if (iQuant > 8) InterBias += 100 * (iQuant - 8); // to make high quants work
-		if (y != 0)
-			if ((pMB - pParam->mb_width)->mode == MODE_INTRA ) InterBias -= 80;
-		if (x != 0)
-			if ((pMB - 1)->mode == MODE_INTRA ) InterBias -= 80;
-
-		if (Data->chroma) InterBias += 50; // to compensate bigger SAD
-		if (Data->rrv) InterBias *= 4;
-
-		if (InterBias < pMB->sad16) {
-			int32_t deviation;
-			if (!Data->rrv) deviation = dev16(Data->Cur, Data->iEdgedWidth);
-			else deviation = dev16(Data->Cur, Data->iEdgedWidth) +
-				dev16(Data->Cur+8, Data->iEdgedWidth) +
-				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
-		}
-		return mode;
-
-	} else {
-
-		int bits, intra, i;
-		VECTOR backup[5], *v;
-		Data->lambda16 = iQuant;
-        Data->lambda8 = (pParam->vol_flags & XVID_VOL_MPEGQUANT)?1:0;
-
-		v = Data->qpel ? Data->currentQMV : Data->currentMV;
-		for (i = 0; i < 5; i++) {
-			Data->iMinSAD[i] = 256*4096;
-			backup[i] = v[i];
-		}
-
-		bits = CountMBBitsInter(Data, pMBs, x, y, pParam, MotionFlags);
-		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; }
-		}
-
-
-		intra = CountMBBitsIntra(Data);
-
-		if (intra < bits) { *Data->iMinSAD = bits = intra; return MODE_INTRA; }
-
-		return mode;
-	}
-}
-
 static void
 SearchP(const IMAGE * const pRef,
 		const uint8_t * const pRefH,
@@ -1252,17 +1245,17 @@
 		const int y,
 		const uint32_t MotionFlags,
 		const uint32_t VopFlags,
-		const uint32_t iQuant,
+		const uint32_t VolFlags,
 		SearchData * const Data,
 		const MBParam * const pParam,
 		const MACROBLOCK * const pMBs,
 		const MACROBLOCK * const prevMBs,
-		int inter4v,
 		MACROBLOCK * const pMB)
 {
 
 	int i, iDirection = 255, mask, threshA;
 	VECTOR pmv[7];
+	int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0);
 
 	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);
@@ -1275,21 +1268,18 @@
 	Data->CurV = pCur->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
 	Data->CurU = pCur->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
 
-	Data->Ref = pRef->y + (x + Data->iEdgedWidth*y) * 16*i;
-	Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16*i;
-	Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16*i;
-	Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16*i;
-	Data->RefCV = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
-	Data->RefCU = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
+	Data->RefP[0] = pRef->y + (x + Data->iEdgedWidth*y) * 16*i;
+	Data->RefP[2] = pRefH + (x + Data->iEdgedWidth*y) * 16*i;
+	Data->RefP[1] = pRefV + (x + Data->iEdgedWidth*y) * 16*i;
+	Data->RefP[3] = pRefHV + (x + Data->iEdgedWidth*y) * 16*i;
+	Data->RefP[4] = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8*i;
+	Data->RefP[5] = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8*i;
 
-	Data->lambda16 = lambda_vec16[iQuant];
-	Data->lambda8 = lambda_vec8[iQuant];
+	Data->lambda16 = lambda_vec16[pMB->quant];
+	Data->lambda8 = lambda_vec8[pMB->quant];
 	Data->qpel_precision = 0;
 
-	if (pMB->dquant != 0) 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];
@@ -1326,8 +1316,8 @@
 
 	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 (!(VopFlags & XVID_VOP_MODEDECISION_BITS)) inter4v = 0;	}
+			(Data->iMinSAD[0] < (prevMBs+x+y*pParam->mb_width)->sad16)))
+		inter4v = 0;
 	else {
 
 		MainSearchFunc * MainSearchPtr;
@@ -1373,7 +1363,6 @@
 	}
 
 	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++) {
@@ -1385,20 +1374,14 @@
 		
 		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);
-		}
+		Data->qpel_precision = 1;
+		SubpelRefine(Data);
 	}
 
-	if ((!(VopFlags & XVID_VOP_MODEDECISION_BITS)) && (Data->iMinSAD[0] < (int32_t)iQuant * 30)) inter4v = 0;
-
-	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
+	if ((!(VopFlags & XVID_VOP_MODEDECISION_BITS)) && (Data->iMinSAD[0] < (int32_t)pMB->quant * 30))
+		inter4v = 0;
 
+	if (inter4v) {
 		SearchData Data8;
 		memcpy(&Data8, Data, sizeof(SearchData)); //quick copy of common data
 
@@ -1410,49 +1393,22 @@
 		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;
-			const VECTOR * const mv = Data->qpel ? pMB->qmvs : pMB->mvs;
 
-			for (i = 0; i < 4; i++) {
-				sumx += mv[i].x / div;
-				sumy += mv[i].y / div;
-			}
+			if (Data->qpel)
+				for (i = 1; i < 5; i++) {
+					sumx += Data->currentQMV[i].x/2;
+					sumy += Data->currentQMV[i].y/2;
+				}
+			else
+				for (i = 1; i < 5; i++) {
+					sumx += Data->currentMV[i].x;
+					sumy += Data->currentMV[i].y;
+				}
 
 			Data->iMinSAD[1] += ChromaSAD(	(sumx >> 3) + roundtab_76[sumx & 0xf],
 											(sumy >> 3) + roundtab_76[sumy & 0xf], Data);
 		}
-	}
-
-	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);
-			Data->currentMV[0].y = RRV_MV_SCALEDOWN(Data->currentMV[0].y);
-	}
-
-	if (inter4v == MODE_INTER) {
-		pMB->mode = MODE_INTER;
-		pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
-		pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = Data->iMinSAD[0];
-
-		if(Data->qpel) {
-			pMB->qmvs[0] = pMB->qmvs[1]
-				= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
-			pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predMV.x;
-			pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predMV.y;
-		} else {
-			pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
-			pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
-		}
-
-	} else if (inter4v == MODE_INTER4V) {
-		pMB->mode = MODE_INTER4V;
-		pMB->sad16 = Data->iMinSAD[0];
-	} else { // INTRA mode
-		SkipMacroblockP(pMB, 0); // not skip, but similar enough
-		pMB->mode = MODE_INTRA;
-	}
-
+	} else Data->iMinSAD[1] = 4096*256;
 }
 
 static void
@@ -1483,14 +1439,15 @@
 	*(Data->iMinSAD) += (Data->lambda8 * i * (*Data->iMinSAD + NEIGH_8X8_BIAS))>>10;
 
 	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));
-		Data->RefH = OldData->RefH + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
-		Data->RefV = OldData->RefV + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
-		Data->RefHV = OldData->RefHV + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
+		if (Data->rrv) i = 16; else i = 8;
+
+		Data->RefP[0] = OldData->RefP[0] + i * ((block&1) + Data->iEdgedWidth*(block>>1));
+		Data->RefP[1] = OldData->RefP[1] + i * ((block&1) + Data->iEdgedWidth*(block>>1));
+		Data->RefP[2] = OldData->RefP[2] + i * ((block&1) + Data->iEdgedWidth*(block>>1));
+		Data->RefP[3] = OldData->RefP[3] + i * ((block&1) + Data->iEdgedWidth*(block>>1));
 
-		Data->Cur = OldData->Cur + i * 8 * ((block&1) + Data->iEdgedWidth*(block>>1));
+		Data->Cur = OldData->Cur + i * ((block&1) + Data->iEdgedWidth*(block>>1));
 		Data->qpel_precision = 0;
 
 		get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
@@ -1624,12 +1581,12 @@
 	Data->qpel_precision = 0;
 	Data->temp[5] = Data->temp[6] = Data->temp[7] = 256*4096; // reset chroma-sad cache
 
-	Data->Ref = pRef->y + (x + y * Data->iEdgedWidth) * 16;
-	Data->RefH = pRefH + (x + y * Data->iEdgedWidth) * 16;
-	Data->RefV = pRefV + (x + y * Data->iEdgedWidth) * 16;
-	Data->RefHV = pRefHV + (x + y * Data->iEdgedWidth) * 16;
-	Data->RefCU = pRef->u + (x + y * Data->iEdgedWidth/2) * 8;
-	Data->RefCV = pRef->v + (x + y * Data->iEdgedWidth/2) * 8;
+	Data->RefP[0] = pRef->y + (x + Data->iEdgedWidth*y) * 16;
+	Data->RefP[2] = pRefH + (x + Data->iEdgedWidth*y) * 16;
+	Data->RefP[1] = pRefV + (x + Data->iEdgedWidth*y) * 16;
+	Data->RefP[3] = pRefHV + (x + Data->iEdgedWidth*y) * 16;
+	Data->RefP[4] = pRef->u + (x + y * (Data->iEdgedWidth/2)) * 8;
+	Data->RefP[5] = pRef->v + (x + y * (Data->iEdgedWidth/2)) * 8;
 
 	Data->predMV = *predMV;
 
@@ -1711,9 +1668,9 @@
 
 	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];
@@ -1733,7 +1690,13 @@
 					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
@@ -1761,18 +1724,18 @@
 	MainSearchFunc *MainSearchPtr;
 
 	*Data->iMinSAD = 256*4096;
-	Data->Ref = f_Ref->y + k;
-	Data->RefH = f_RefH + k;
-	Data->RefV = f_RefV + k;
-	Data->RefHV = f_RefHV + k;
-	Data->bRef = b_Ref->y + k;
-	Data->bRefH = b_RefH + k;
-	Data->bRefV = b_RefV + k;
-	Data->bRefHV = b_RefHV + k;
-	Data->RefCU = f_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8;
-	Data->RefCV = f_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8;
-	Data->b_RefCU = b_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8;
-	Data->b_RefCV = b_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8;
+	Data->RefP[0] = f_Ref->y + k;
+	Data->RefP[2] = f_RefH + k;
+	Data->RefP[1] = f_RefV + k;
+	Data->RefP[3] = f_RefHV + k;
+	Data->b_RefP[0] = b_Ref->y + k;
+	Data->b_RefP[2] = b_RefH + k;
+	Data->b_RefP[1] = b_RefV + k;
+	Data->b_RefP[3] = b_RefHV + k;
+	Data->RefP[4] = f_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8;
+	Data->RefP[5] = f_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8;
+	Data->b_RefP[4] = b_Ref->u + (x + (Data->iEdgedWidth/2) * y) * 8;
+	Data->b_RefP[5] = b_Ref->v + (x + (Data->iEdgedWidth/2) * y) * 8;
 
 	k = Data->qpel ? 4 : 2;
 	Data->max_dx = k * (pParam->width - x * 16);
@@ -1903,19 +1866,19 @@
 	fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode;
 
 	i = (x + y * fData->iEdgedWidth) * 16;
-	bData.bRef = fData->Ref = f_Ref->y + i;
-	bData.bRefH = fData->RefH = f_RefH + i;
-	bData.bRefV = fData->RefV = f_RefV + i;
-	bData.bRefHV = fData->RefHV = f_RefHV + i;
-	bData.Ref = fData->bRef = b_Ref->y + i;
-	bData.RefH = fData->bRefH = b_RefH + i;
-	bData.RefV = fData->bRefV = b_RefV + i;
-	bData.RefHV = fData->bRefHV = b_RefHV + i;
-	bData.b_RefCU = fData->RefCU = f_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8;
-	bData.b_RefCV = fData->RefCV = f_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8;
-	bData.RefCU = fData->b_RefCU = b_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8;
-	bData.RefCV = fData->b_RefCV = b_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8;
 
+	bData.b_RefP[0] = fData->RefP[0] = f_Ref->y + i;
+	bData.b_RefP[2] = fData->RefP[2] = f_RefH + i;
+	bData.b_RefP[1] = fData->RefP[1] = f_RefV + i;
+	bData.b_RefP[3] = fData->RefP[3] = f_RefHV + i;
+	bData.RefP[0] = fData->b_RefP[0] = b_Ref->y + i;
+	bData.RefP[2] = fData->b_RefP[2] = b_RefH + i;
+	bData.RefP[1] = fData->b_RefP[1] = b_RefV + i;
+	bData.RefP[3] = fData->b_RefP[3] = b_RefHV + i;
+	bData.b_RefP[4] = fData->RefP[4] = f_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8;
+	bData.b_RefP[5] = fData->RefP[5] = f_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8;
+	bData.RefP[4] = fData->b_RefP[4] = b_Ref->u + (x + (fData->iEdgedWidth/2) * y) * 8;
+	bData.RefP[5] = fData->b_RefP[5] = b_Ref->v + (x + (fData->iEdgedWidth/2) * y) * 8;
 
 	bData.bpredMV = fData->predMV = *f_predMV;
 	fData->bpredMV = bData.predMV = *b_predMV;
@@ -2167,7 +2130,7 @@
         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;
+	Data->RefP[0] = pRef + (x + y * pParam->edged_width) * 16;
 
 	pmv[1].x = EVEN(pMB->mvs[0].x);
 	pmv[1].y = EVEN(pMB->mvs[0].y);
@@ -2176,10 +2139,6 @@
 	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) {
 
@@ -2200,8 +2159,8 @@
 	}
 }
 
-#define INTRA_THRESH	2400
-#define INTER_THRESH	1100
+#define INTRA_THRESH	1800
+#define INTER_THRESH	1200
 
 int
 MEanalysis(	const IMAGE * const pRef,
@@ -2230,10 +2189,10 @@
 	CheckCandidate = CheckCandidate32I;
 
 	if (intraCount != 0 && intraCount < 10) // we're right after an I frame
-		IntraThresh += 8 * (intraCount - 10) * (intraCount - 10);
+		IntraThresh += 15 * (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;
+			IntraThresh -= (IntraThresh * (maxIntra - 8*(maxIntra - intraCount)))/maxIntra;
 
 	InterThresh -= (350 - 8*b_thresh) * bCount;
 	if (InterThresh < 300 + 5*b_thresh) InterThresh = 300 + 5*b_thresh;
@@ -2247,9 +2206,9 @@
 
 			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 =
 					(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 =
 					(pMBs[x + y * pParam->mb_width].mvs[0].y * (bCount+1) ) / bCount;
 			}
 
@@ -2276,7 +2235,7 @@
 	sSAD /= blocks;
 	s = (10*s) / blocks;
 
-	if (s > 5) sSAD += (s - 4) * (180 - 2*b_thresh); //static block - looks bad when in bframe...
+	if (s > 4) sSAD += (s - 2) * (60 - 2*b_thresh); //static block - looks bad when in bframe...
 
 	if (sSAD > InterThresh ) return P_VOP;
 	emms();
@@ -2312,9 +2271,9 @@
 	double meanx,meany;
 	int num,oldnum;
 
-	if (!MBmask) { fprintf(stderr,"Mem error\n"); 
-			       gmc.duv[0].x= gmc.duv[0].y = 
-				   		gmc.duv[1].x= gmc.duv[1].y = 			
+	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; }
 
@@ -2330,10 +2289,10 @@
 		if (pMB->mode == MODE_INTRA || pMB->mode == MODE_NOT_CODED)
 			continue;
 
-		if ( ( (ABS(mv.x -   (pMB-1)->mvs[0].x) < deltax) && (ABS(mv.y -   (pMB-1)->mvs[0].y) < deltay) )
-		&&   ( (ABS(mv.x -   (pMB+1)->mvs[0].x) < deltax) && (ABS(mv.y -   (pMB+1)->mvs[0].y) < deltay) )
-		&&   ( (ABS(mv.x - (pMB-MBw)->mvs[0].x) < deltax) && (ABS(mv.y - (pMB-MBw)->mvs[0].y) < deltay) )
-		&&   ( (ABS(mv.x - (pMB+MBw)->mvs[0].x) < deltax) && (ABS(mv.y - (pMB+MBw)->mvs[0].y) < deltay) ) )
+		if ( ( (abs(mv.x -   (pMB-1)->mvs[0].x) < deltax) && (abs(mv.y -   (pMB-1)->mvs[0].y) < deltay) )
+		&&   ( (abs(mv.x -   (pMB+1)->mvs[0].x) < deltax) && (abs(mv.y -   (pMB+1)->mvs[0].y) < deltay) )
+		&&   ( (abs(mv.x - (pMB-MBw)->mvs[0].x) < deltax) && (abs(mv.y - (pMB-MBw)->mvs[0].y) < deltay) )
+		&&   ( (abs(mv.x - (pMB+MBw)->mvs[0].x) < deltax) && (abs(mv.y - (pMB+MBw)->mvs[0].y) < deltay) ) )
 			MBmask[mbnum]=1;
 	}
 
@@ -2382,7 +2341,7 @@
 
 	denom = a*a+b*b-c*n;
 
-/* Solve the system:     sol = (D'*E*D)^{-1} D'*E*F   */
+/* Solve the system:	sol = (D'*E*D)^{-1} D'*E*F   */
 /* D'*E*F has been calculated in the same loop as matrix */
 
 	sol[0] = -c*DtimesF[0] + a*DtimesF[1] + b*DtimesF[2];
@@ -2408,8 +2367,8 @@
 				continue;
 
 			oldnum++;
-			meanx += ABS(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - mv.x );
-			meany += ABS(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - mv.y );
+			meanx += fabs(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - mv.x );
+			meany += fabs(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - mv.y );
 		}
 
 	if (4*meanx > oldnum)	/* better fit than 0.25 is useless */
@@ -2436,8 +2395,8 @@
 			if (!MBmask[mbnum])
 				continue;
 
-			if  ( ( ABS(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - mv.x ) > meanx )
-			   || ( ABS(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - mv.y ) > meany ) )
+			if  ( ( fabs(( sol[0] + (16*mx+8)*sol[1] + (16*my+8)*sol[2] ) - mv.x ) > meanx )
+				|| ( fabs(( sol[3] - (16*mx+8)*sol[2] + (16*my+8)*sol[1] ) - mv.y ) > meany ) )
 				MBmask[mbnum]=0;
 			else
 				num++;
@@ -2548,20 +2507,22 @@
 	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;
+	uint8_t * ptr;
 
 	memcpy(Data8, Data, sizeof(SearchData));
 	CheckCandidate = CheckCandidateBits8;
 
-	for (i = 0; i < 4; i++) {
+	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->Cur = Data->Cur + 8*((i&1) + (i>>1)*Data->iEdgedWidth);
-		Data8->Ref = Data->Ref + 8*((i&1) + (i>>1)*Data->iEdgedWidth);
-		Data8->RefH = Data->RefH + 8*((i&1) + (i>>1)*Data->iEdgedWidth);
-		Data8->RefV = Data->RefV + 8*((i&1) + (i>>1)*Data->iEdgedWidth);
-		Data8->RefHV = Data->RefHV + 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);
+		Data8->RefP[1] = Data->RefP[1] + 8*((i&1) + (i>>1)*Data->iEdgedWidth);
+		Data8->RefP[3] = Data->RefP[3] + 8*((i&1) + (i>>1)*Data->iEdgedWidth);
 
 		if(Data->qpel) {
 			Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i);
@@ -2580,11 +2541,9 @@
 
 		Data8->qpel_precision = Data8->qpel;
 		// checking the vector which has been found by SAD-based 8x8 search (if it's different than the one found so far)
-		if (Data8->qpel) {
-			if (!(Data8->currentQMV->x == backup[i+1].x && Data8->currentQMV->y == backup[i+1].y))
-				CheckCandidateBits8(backup[i+1].x, backup[i+1].y, 255, &iDirection, Data8);
-		} else {
-			if (!(Data8->currentMV->x == backup[i+1].x && Data8->currentMV->y == backup[i+1].y))
+		{
+			VECTOR *v = Data8->qpel ? Data8->currentQMV : Data8->currentMV;
+			if (!MVequal (*v, backup[i+1]) )
 				CheckCandidateBits8(backup[i+1].x, backup[i+1].y, 255, &iDirection, Data8);
 		}
 
@@ -2603,7 +2562,8 @@
 				if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_BITS)
 					SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255);
 
-				if (MotionFlags & XVID_ME_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;
@@ -2617,18 +2577,24 @@
 			}
 			if (MotionFlags & XVID_ME_QUARTERPELREFINE8_BITS) SubpelRefine(Data8);
 
-		} else // not qpel
-			if (MotionFlags & XVID_ME_HALFPELREFINE8_BITS) SubpelRefine(Data8); //halfpel mode, halfpel refinement
+		} else { // not qpel
+
+			if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_BITS) //extsearch
+				SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255);
+
+			if (MotionFlags & XVID_ME_HALFPELREFINE8_BITS)
+				SubpelRefine(Data8); //halfpel refinement
+		}
 
 		//checking vector equal to predicion
 		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))
+			if (!MVequal(*v, Data8->predMV))
 				CheckCandidateBits8(Data8->predMV.x, Data8->predMV.y, 255, &iDirection, Data8);
 		}
 
 		bits += *Data8->iMinSAD;
-		if (bits >= Data->iMinSAD[0]) break; // no chances for INTER4V
+		if (bits >= Data->iMinSAD[0]) return bits; // no chances for INTER4V
 
 		// MB structures for INTER4V mode; we have to set them here, we don't have predictor anywhere else
 		if(Data->qpel) {
@@ -2646,39 +2612,28 @@
 		pMB->mvs[i] = *Data8->currentMV;
 		pMB->sad8[i] = 4 * *Data8->iMinSAD;
 		if (Data8->temp[0]) cbp |= 1 << (5 - i);
-	}
 
-	if (bits < *Data->iMinSAD) { // there is still a chance for inter4v mode. let's check chroma
-		const uint8_t * ptr;
-		sumx = (sumx >> 3) + roundtab_76[sumx & 0xf];
-		sumy = (sumy >> 3) + roundtab_76[sumy & 0xf];
-
-		//chroma U
-		ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefCU, 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding);
-		transfer_8to16subro(in, Data->CurU, ptr, Data->iEdgedWidth/2);
-		fdct(in);
-		if (Data->lambda8 == 0) i = quant_inter(coeff, in, Data->lambda16);
-		else i = quant4_inter(coeff, in, Data->lambda16);
-		if (i > 0) {
-			bits += CodeCoeffInter_CalcBits(coeff, scan_tables[0]);
-			cbp |= 1 << (5 - 4);
-		}
-
-		if (bits < *Data->iMinSAD) { // still possible
-			//chroma V
-			ptr = interpolate8x8_switch2(Data->RefQ + 64, Data->RefCV, 0, 0, sumx, sumy, Data->iEdgedWidth/2, Data->rounding);
-			transfer_8to16subro(in, Data->CurV, ptr, Data->iEdgedWidth/2);
-			fdct(in);
-			if (Data->lambda8 == 0) i = quant_inter(coeff, in, Data->lambda16);
-			else i = quant4_inter(coeff, in, Data->lambda16);
-			if (i > 0) {
-				bits += CodeCoeffInter_CalcBits(coeff, scan_tables[0]);
-				cbp |= 1 << (5 - 5);
-			}
-			bits += xvid_cbpy_tab[15-(cbp>>2)].len;
-			bits += mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len;
-		}
-	}
+	} // /for all luma blocks
+
+	bits += xvid_cbpy_tab[15-(cbp>>2)].len;
+
+	// let's check chroma
+	sumx = (sumx >> 3) + roundtab_76[sumx & 0xf];
+	sumy = (sumy >> 3) + roundtab_76[sumy & 0xf];
+
+	//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->iQuant, Data->quant_type, &cbp, 4, 0);
+
+	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->iQuant, Data->quant_type, &cbp, 5, 0);
+
+	bits += mcbpc_inter_tab[(MODE_INTER4V & 7) | ((cbp & 3) << 3)].len;
 
 	return bits;
 }
@@ -2688,59 +2643,53 @@
 CountMBBitsIntra(const SearchData * const Data)
 {
 	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];
+	int cbp = 0, i, t, dc = 1024, b_dc;
+	int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64;
+	uint32_t iDcScaler = get_dc_scaler(Data->iQuant, 1);
 
 	for(i = 0; i < 4; i++) {
-		uint32_t iDcScaler = get_dc_scaler(iQuant, 1);
-
 		int s = 8*((i&1) + (i>>1)*Data->iEdgedWidth);
 		transfer_8to16copy(in, Data->Cur + s, Data->iEdgedWidth);
 		fdct(in);
-		b_dc = dc;
-		dc = in[0];
-		in[0] -= b_dc;
-		if (Data->lambda8 == 0) quant_intra_c(coeff, in, iQuant, iDcScaler);
-		else quant4_intra_c(coeff, in, iQuant, iDcScaler);
-
-		b_dc = dc;
-		dc = coeff[0];
-		if (i != 0) coeff[0] -= b_dc;
+		b_dc = in[0];
+		in[0] -= dc;
+		dc = b_dc;
+		if (Data->quant_type == 0) quant_intra(coeff, in, Data->iQuant, iDcScaler);
+		else quant4_intra(coeff, in, Data->iQuant, iDcScaler);
 
-		bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcy_tab[coeff[0] + 255].len;;
+		bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcy_tab[coeff[0] + 255].len;
 		Data->temp[i] = t;
 		if (t != 0)  cbp |= 1 << (5 - i);
-		if (bits >= Data->iMinSAD[0]) break;
+		if (bits >= Data->iMinSAD[0]) return bits;
 	}
 
-	if (bits < Data->iMinSAD[0]) { // INTRA still looks good, let's add chroma
-		uint32_t iDcScaler = get_dc_scaler(iQuant, 0);
-		//chroma U
-		transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2);
-		fdct(in);
-		in[0] -= 1024;
-		if (Data->lambda8 == 0) quant_intra(coeff, in, iQuant, iDcScaler);
-		else quant4_intra(coeff, in, iQuant, iDcScaler);
-
-		bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcc_tab[coeff[0] + 255].len;
-		if (t != 0) cbp |= 1 << (5 - 4);
-
-		if (bits < Data->iMinSAD[0]) {
-			iDcScaler = get_dc_scaler(iQuant, 1);
-			//chroma V
-			transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2);
-			fdct(in);
-			in[0] -= 1024;
-			if (Data->lambda8 == 0) quant_intra(coeff, in, iQuant, iDcScaler);
-			else quant4_intra(coeff, in, iQuant, iDcScaler);
+	bits += xvid_cbpy_tab[cbp>>2].len;
 
-			bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcc_tab[coeff[0] + 255].len;
-			if (t != 0) cbp |= 1 << (5 - 5);
+	iDcScaler = get_dc_scaler(Data->iQuant, 0);
+
+	//chroma U
+	transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2);
+	fdct(in);
+	in[0] -= 1024;
+	if (Data->quant_type == 0) quant_intra(coeff, in, Data->iQuant, iDcScaler);
+	else quant4_intra(coeff, in, Data->iQuant, iDcScaler);
+
+	bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcc_tab[coeff[0] + 255].len;
+	if (t != 0) cbp |= 1 << (5 - 4);
+
+	if (bits >= Data->iMinSAD[0]) return bits;
+
+	//chroma V
+	transfer_8to16copy(in, Data->CurV, Data->iEdgedWidth/2);
+	fdct(in);
+	in[0] -= 1024;
+	if (Data->quant_type == 0) quant_intra(coeff, in, Data->iQuant, iDcScaler);
+	else quant4_intra(coeff, in, Data->iQuant, iDcScaler);
+
+	bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcc_tab[coeff[0] + 255].len;
+	if (t != 0) cbp |= 1 << (5 - 5);
+
+	bits += mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp & 3) << 3)].len;
 
-			bits += xvid_cbpy_tab[cbp>>2].len;
-			bits += mcbpc_inter_tab[(MODE_INTRA & 7) | ((cbp & 3) << 3)].len;
-		}
-	}
 	return bits;
 }