--- trunk/xvidcore/src/motion/motion_est.c	2003/02/15 15:22:19	851
+++ branches/dev-api-4/xvidcore/src/motion/motion_est.c	2003/04/10 13:05:54	982
@@ -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)
@@ -54,36 +74,39 @@
 #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)
 {
-	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;
@@ -91,50 +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;
 	}
 
@@ -144,38 +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;
 	}
@@ -187,28 +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 *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 *
@@ -225,8 +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
@@ -241,16 +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;
 }
@@ -269,7 +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);
@@ -295,7 +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;
@@ -331,7 +335,7 @@
 	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;
@@ -355,12 +359,18 @@
 {
 	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);
@@ -369,7 +379,7 @@
 
 	if (sad < *(data->iMinSAD)) {
 		*(data->iMinSAD) = sad;
-		data->currentMV->x = x; data->currentMV->y = y;
+		current->x = x; current->y = y;
 		*dir = Direction;
 	}
 }
@@ -381,7 +391,7 @@
 	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;
 
@@ -416,8 +426,8 @@
 	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
 
@@ -455,7 +465,7 @@
 	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),
+	sad = sad32v_c(data->Cur, data->RefP[0] + x/2 + (y/2)*(data->iEdgedWidth),
 							data->iEdgedWidth, data->temp+1);
 
 	if (sad < *(data->iMinSAD)) {
@@ -482,8 +492,9 @@
 	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 +539,7 @@
 	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 +552,11 @@
 			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;
@@ -586,7 +598,7 @@
 	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) ?
@@ -598,10 +610,10 @@
 		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);
@@ -635,7 +647,7 @@
 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;
@@ -673,7 +685,7 @@
 		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);
+		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);
 		fdct(in);
 		if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16);
@@ -685,7 +697,7 @@
 
 		if (bits < data->iMinSAD[0]) {
 			//chroma V
-			ptr = interpolate8x8_switch2(data->RefQ + 64, data->RefCV, 0, 0, xc, yc,  data->iEdgedWidth/2, data->rounding);
+			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);
 			fdct(in);
 			if (data->lambda8 == 0) sum = quant_inter(coeff, in, data->lambda16);
@@ -697,7 +709,7 @@
 		}
 	}
 
-	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]) {
@@ -720,7 +732,7 @@
 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;
@@ -926,21 +938,22 @@
 							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;
 	}
@@ -971,17 +984,22 @@
 	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];
 	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;
@@ -991,11 +1009,12 @@
 	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 ? 1:0);
+	Data.chroma = MotionFlags & XVID_ME_CHROMA16;
+	Data.rrv = (current->vop_flags & XVID_VOP_REDUCED ? 1:0);
+	Data.dctSpace = dct_space;
 
-	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 +1047,18 @@
 
 			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 +1066,16 @@
 			}
 
 			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)
@@ -1073,10 +1083,9 @@
 		}
 	}
 
-	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);
-	}
+
 	return 0;
 }
 
@@ -1148,19 +1157,18 @@
 		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 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] +
@@ -1187,7 +1195,7 @@
 				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;
 
@@ -1196,7 +1204,7 @@
 		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++) {
@@ -1208,8 +1216,8 @@
 		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; }
 		}
 
 
@@ -1230,7 +1238,7 @@
 		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,
@@ -1254,21 +1262,20 @@
 	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->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;
+	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];
@@ -1280,7 +1287,7 @@
 	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 +1313,12 @@
 	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 +1327,7 @@
 	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 +1358,8 @@
 		}
 	}
 
-	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 +1367,22 @@
 		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 +1393,7 @@
 		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 +1409,7 @@
 		}
 	}
 
-	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);
@@ -1459,15 +1468,16 @@
 
 	*(Data->iMinSAD) += (Data->lambda8 * i * (*Data->iMinSAD + NEIGH_8X8_BIAS))>>10;
 
-	if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8|PMV_QUARTERPELREFINE8)) {
-		if (Data->rrv) i = 2; else i = 1;
+	if (MotionFlags & (XVID_ME_EXTSEARCH8|XVID_ME_HALFPELREFINE8|XVID_ME_QUARTERPELREFINE8)) {
+
+		if (Data->rrv) i = 16; else i = 8;
 
-		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));
+		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,
@@ -1476,12 +1486,12 @@
 		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 +1502,7 @@
 			}
 		}
 
-		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 +1513,7 @@
 			}
 		}
 
-		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);
@@ -1601,12 +1611,12 @@
 	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;
 
@@ -1627,8 +1637,8 @@
 		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);
@@ -1688,9 +1698,9 @@
 
 	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];
@@ -1710,7 +1720,13 @@
 					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
@@ -1738,18 +1754,18 @@
 	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);
@@ -1799,13 +1815,14 @@
 		}
 	}
 
+	*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);
@@ -1879,19 +1896,19 @@
 	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;
@@ -2014,9 +2031,9 @@
 	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 +2141,9 @@
 {
 
 	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;
 
@@ -2139,10 +2157,10 @@
 			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;
+	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);
@@ -2152,42 +2170,43 @@
 
 	CheckCandidate32I(0, 0, 255, &i, Data);
 
-	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	1300
 
 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 +2215,32 @@
 	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 +2252,24 @@
 									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 > 4) sSAD += (s - 3) * (300 - 2*b_thresh); //static block - looks bad when in bframe...
+
 	if (sSAD > InterThresh ) return P_VOP;
 	emms();
 	return B_VOP;
-
 }
 
 
@@ -2271,12 +2301,16 @@
 	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];
@@ -2285,15 +2319,15 @@
 		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;
 	}
 
-	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 +2335,9 @@
 		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 +2348,8 @@
 
 	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];
@@ -2337,7 +2371,7 @@
 
 	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];
@@ -2352,8 +2386,8 @@
 
 	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];
@@ -2363,8 +2397,8 @@
 				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 */
@@ -2381,8 +2415,8 @@
 	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];
@@ -2391,8 +2425,8 @@
 			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++;
@@ -2446,7 +2480,7 @@
 		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 +2498,12 @@
 		}
 	}
 
-	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 +2514,10 @@
 			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);
@@ -2503,7 +2537,7 @@
 	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;
@@ -2513,10 +2547,10 @@
 		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);
@@ -2544,7 +2578,7 @@
 		}
 
 		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 +2589,10 @@
 				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 +2604,13 @@
 							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);
@@ -2609,7 +2643,7 @@
 		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);
+		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);
 		fdct(in);
 		if (Data->lambda8 == 0) i = quant_inter(coeff, in, Data->lambda16);
@@ -2621,7 +2655,7 @@
 
 		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);
+			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);
 			fdct(in);
 			if (Data->lambda8 == 0) i = quant_inter(coeff, in, Data->lambda16);
@@ -2630,7 +2664,7 @@
 				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;
 		}
 	}
@@ -2643,25 +2677,20 @@
 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;
+	int cbp = 0, i, t, dc = 1024, b_dc;
 	const uint32_t iQuant = Data->lambda16;
-	int16_t in[64], coeff[64];
+	int16_t *in = Data->dctSpace, * coeff = Data->dctSpace + 64;
+	uint32_t iDcScaler = get_dc_scaler(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->lambda8 == 0) quant_intra(coeff, in, iQuant, iDcScaler);
+		else quant4_intra(coeff, in, iQuant, iDcScaler);
 
 		bits += t = CodeCoeffIntra_CalcBits(coeff, scan_tables[0]) + dcy_tab[coeff[0] + 255].len;;
 		Data->temp[i] = t;
@@ -2670,7 +2699,7 @@
 	}
 
 	if (bits < Data->iMinSAD[0]) { // INTRA still looks good, let's add chroma
-		uint32_t iDcScaler = get_dc_scaler(iQuant, 0);
+		iDcScaler = get_dc_scaler(iQuant, 0);
 		//chroma U
 		transfer_8to16copy(in, Data->CurU, Data->iEdgedWidth/2);
 		fdct(in);
@@ -2680,7 +2709,6 @@
 
 		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]) {
 			//chroma V
@@ -2693,16 +2721,9 @@
 			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;
 }