--- motion_est.c	2002/09/25 21:27:55	1.44.2.2
+++ motion_est.c	2002/10/21 13:07:33	1.44.2.12
@@ -37,6 +37,7 @@
 #include "../prediction/mbprediction.h"
 #include "../global.h"
 #include "../utils/timer.h"
+#include "../image/interpolate8x8.h"
 #include "motion_est.h"
 #include "motion.h"
 #include "sad.h"
@@ -51,6 +52,16 @@
 #define CHECK_CANDIDATE(X,Y,D) { \
 (*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); }
 
+#define GET_REFERENCE(X, Y, REF) { \
+	switch ( (((X)&1)<<1) + ((Y)&1) ) \
+	{ \
+		case 0 : REF = (uint8_t *)data->Ref + (X)/2 + ((Y)/2)*(data->iEdgedWidth); break; \
+		case 1 : REF = (uint8_t *)data->RefV + (X)/2 + (((Y)-1)/2)*(data->iEdgedWidth); break; \
+		case 2 : REF = (uint8_t *)data->RefH + ((X)-1)/2 + ((Y)/2)*(data->iEdgedWidth); break; \
+		default : REF = (uint8_t *)data->RefHV + ((X)-1)/2 + (((Y)-1)/2)*(data->iEdgedWidth); break; \
+	} \
+}
+
 #define iDiamondSize 2
 
 static __inline int
@@ -84,7 +95,6 @@
 static void 
 CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
 {
-	int32_t * const sad = data->temp;
 	int t;
 	const uint8_t * Reference;
 
@@ -98,11 +108,50 @@
 		default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
 	}
 	
-	data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, sad+1);
+	data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
 
 	t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
-	data->temp[0] += lambda_vec16[data->iQuant] * t;
-	data->temp[1] += lambda_vec8[data->iQuant] * t;
+	data->temp[0] += data->lambda16 * t;
+	data->temp[1] += data->lambda8 * t;
+
+	if (data->temp[0] < data->iMinSAD[0]) {
+		data->iMinSAD[0] = data->temp[0];
+		data->currentMV[0].x = x; data->currentMV[0].y = y;
+		*dir = Direction; }
+
+	if (data->temp[1] < data->iMinSAD[1]) {
+		data->iMinSAD[1] = data->temp[1]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
+	if (data->temp[2] < data->iMinSAD[2]) {
+		data->iMinSAD[2] = data->temp[2]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
+	if (data->temp[3] < data->iMinSAD[3]) {
+		data->iMinSAD[3] = data->temp[3]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
+	if (data->temp[4] < data->iMinSAD[4]) {
+		data->iMinSAD[4] = data->temp[4]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
+
+}
+
+static void 
+CheckCandidate16Q(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
+{
+	//this function expects X and Y in halfpel positions, but assumes that qpel positions are our goal
+	int t;
+	const uint8_t * Reference;
+
+	if (( x > data->max_dx) || ( x < data->min_dx)
+		|| ( y > data->max_dy) || (y < data->min_dy)) return;
+
+	switch ( ((x&1)<<1) + (y&1) ) {
+		case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
+		case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
+		case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
+		default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
+	}
+	
+	data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
+
+	t = d_mv_bits(2 * x - data->predQMV.x, 2 * y - data->predQMV.y, data->iFcode);
+	data->temp[0] += data->lambda16 * t;
+	data->temp[1] += data->lambda8 * t;
 
 	if (data->temp[0] < data->iMinSAD[0]) {
 		data->iMinSAD[0] = data->temp[0];
@@ -137,9 +186,8 @@
 		default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
 	}
 
-	sad = lambda_vec16[data->iQuant] *
-			d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
-	sad += sad16(data->Cur, Reference, data->iEdgedWidth, 256*4096);
+	sad = data->lambda16 * d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
+	sad += sad16(data->Cur, Reference, data->iEdgedWidth, MV_MAX_ERROR);
 
 	if (sad < *(data->iMinSAD)) {
 		*(data->iMinSAD) = sad;
@@ -148,6 +196,81 @@
 }
 
 static void 
+CheckCandidate16_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
+
+// CheckCandidate16 variant which expects x and y in quarter pixel resolution
+// Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!)
+// around currentMV!
+{
+	int t;
+	uint8_t * Reference = (uint8_t *)data->RefQ;
+	const uint8_t *ref1, *ref2, *ref3, *ref4;
+	VECTOR halfpelMV = *(data->currentMV);
+
+	int32_t iEdgedWidth = data->iEdgedWidth;
+	uint32_t rounding = data->rounding;
+
+	if (( x > data->max_dx) || ( x < data->min_dx)
+		|| ( y > data->max_dy) || (y < data->min_dy)) return;
+
+	GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1); // this refenrence is used in all cases
+	switch( ((x&1)<<1) + (y&1) )
+	{
+	case 0: // pure halfpel position - shouldn't happen during a refinement step
+		GET_REFERENCE(halfpelMV.x, halfpelMV.y, Reference); 
+		break;
+
+	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
+		GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
+		interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
+		interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding);
+		interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding);
+		interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding);
+		break;
+
+	case 2: // x qpel, y halfpel - left or right during qpel refinement
+		GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2);
+		interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
+		interpolate8x8_avg2(Reference+8, ref1+8, ref2+8, iEdgedWidth, rounding);
+		interpolate8x8_avg2(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, iEdgedWidth, rounding);
+		interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding);
+		break;
+
+	default: // x and y in qpel resolution - the "corners" (top left/right and
+			 // bottom left/right) during qpel refinement
+		GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
+		GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3);
+		GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4);
+
+		interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
+		interpolate8x8_avg4(Reference+8, ref1+8, ref2+8, ref3+8, ref4+8, iEdgedWidth, rounding);
+		interpolate8x8_avg4(Reference+8*iEdgedWidth, ref1+8*iEdgedWidth, ref2+8*iEdgedWidth, ref3+8*iEdgedWidth, ref4+8*iEdgedWidth, iEdgedWidth, rounding);
+		interpolate8x8_avg4(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, ref3+8*iEdgedWidth+8, ref4+8*iEdgedWidth+8, iEdgedWidth, rounding);
+		break;
+	}
+	
+	data->temp[0] = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp+1);
+
+	t = d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode);
+	data->temp[0] += data->lambda16 * t;
+	data->temp[1] += data->lambda8 * t;
+
+	if (data->temp[0] < data->iMinSAD[0]) {
+		data->iMinSAD[0] = data->temp[0];
+		data->currentQMV[0].x = x; data->currentQMV[0].y = y;
+	/*	*dir = Direction;*/ }
+
+	if (data->temp[1] < data->iMinSAD[1]) {
+		data->iMinSAD[1] = data->temp[1]; data->currentQMV[1].x = x; data->currentQMV[1].y = y; }
+	if (data->temp[2] < data->iMinSAD[2]) {
+		data->iMinSAD[2] = data->temp[2]; data->currentQMV[2].x = x; data->currentQMV[2].y = y; }
+	if (data->temp[3] < data->iMinSAD[3]) {
+		data->iMinSAD[3] = data->temp[3]; data->currentQMV[3].x = x; data->currentQMV[3].y = y; }
+	if (data->temp[4] < data->iMinSAD[4]) {
+		data->iMinSAD[4] = data->temp[4]; data->currentQMV[4].x = x; data->currentQMV[4].y = y; }
+}
+
+static void 
 CheckCandidate16no4vI(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
 {
 	int32_t sad;
@@ -155,10 +278,7 @@
 	if (( x > data->max_dx) || ( x < data->min_dx)
 		|| ( y > data->max_dy) || (y < data->min_dy)) return;
 
-	sad = lambda_vec16[data->iQuant] *
-			d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
-
-	sad += sad16(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth),
+	sad = sad16(data->Cur, data->Ref + x/2 + (y/2)*(data->iEdgedWidth),
 					data->iEdgedWidth, 256*4096);
 
 	if (sad < *(data->iMinSAD)) {
@@ -193,7 +313,7 @@
 		default : ReferenceB = data->bRefHV + (xb-1)/2 + ((yb-1)/2)*(data->iEdgedWidth); break;
 	}
 
-	sad = lambda_vec16[data->iQuant] *
+	sad = data->lambda16 *
 			( d_mv_bits(xf - data->predMV.x, yf - data->predMV.y, data->iFcode) + 
 			  d_mv_bits(xb - data->bpredMV.x, yb - data->bpredMV.y, data->iFcode) );
 
@@ -216,7 +336,7 @@
 
 	if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return;
 
-	sad = lambda_vec16[data->iQuant] * d_mv_bits(x, y, 1);
+	sad = data->lambda16 * d_mv_bits(x, y, 1);
 
 	for (k = 0; k < 4; k++) {
 		mvs.x = data->directmvF[k].x + x;
@@ -269,9 +389,9 @@
 	const uint8_t *ReferenceB;
 	VECTOR mvs, b_mvs;
 
-	if (( x > 31) || ( x < -31) || ( y > 31) || (y < -31)) return;
+	if (( x > 31) || ( x < -32) || ( y > 31) || (y < -32)) return;
 	
-		sad = lambda_vec16[data->iQuant] * d_mv_bits(x, y, 1);
+		sad = data->lambda16 * d_mv_bits(x, y, 1);
 
 	mvs.x = data->directmvF[0].x + x;
 	b_mvs.x = ((x == 0) ?
@@ -328,7 +448,33 @@
 	}
 
 	sad = sad8(data->Cur, Reference, data->iEdgedWidth);
-	sad += lambda_vec8[data->iQuant] * d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
+	sad += data->lambda8 * d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
+
+	if (sad < *(data->iMinSAD)) {
+		*(data->iMinSAD) = sad;
+		data->currentMV->x = x; data->currentMV->y = y;
+		*dir = Direction; }
+}
+
+static void
+CheckCandidate8Q(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
+{
+	int32_t sad;
+	const uint8_t * Reference;
+
+	if (( x > data->max_dx) || ( x < data->min_dx)
+		|| ( y > data->max_dy) || (y < data->min_dy)) return;
+
+	switch ( ((x&1)<<1) + (y&1) )
+	{
+		case 0 : Reference = data->Ref + x/2 + (y/2)*(data->iEdgedWidth); break;
+		case 1 : Reference = data->RefV + x/2 + ((y-1)/2)*(data->iEdgedWidth); break;
+		case 2 : Reference = data->RefH + (x-1)/2 + (y/2)*(data->iEdgedWidth); break;
+		default : Reference = data->RefHV + (x-1)/2 + ((y-1)/2)*(data->iEdgedWidth); break;
+	}
+
+	sad = sad8(data->Cur, Reference, data->iEdgedWidth);
+	sad += data->lambda8 * d_mv_bits(2 * x - data->predQMV.x, 2 * y - data->predQMV.y, data->iFcode);
 
 	if (sad < *(data->iMinSAD)) {
 		*(data->iMinSAD) = sad;
@@ -336,6 +482,62 @@
 		*dir = Direction; }
 }
 
+static void
+CheckCandidate8_qpel(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
+// CheckCandidate16no4v variant which expects x and y in quarter pixel resolution
+// Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!)
+// around currentMV!
+
+{
+	int32_t sad;
+	uint8_t *Reference = (uint8_t *) data->RefQ;
+	const uint8_t *ref1, *ref2, *ref3, *ref4;
+	VECTOR halfpelMV = *(data->currentMV);
+
+	int32_t iEdgedWidth = data->iEdgedWidth;
+	uint32_t rounding = data->rounding;
+
+	if (( x > data->max_dx) || ( x < data->min_dx)
+		|| ( y > data->max_dy) || (y < data->min_dy)) return;
+
+	GET_REFERENCE(halfpelMV.x, halfpelMV.y, ref1);
+	switch( ((x&1)<<1) + (y&1) )
+	{
+	case 0: // pure halfpel position - shouldn't happen during a refinement step
+		GET_REFERENCE(halfpelMV.x, halfpelMV.y, Reference); 
+		break;
+
+	case 1: // x halfpel, y qpel - top or bottom during qpel refinement
+		GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
+
+		interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
+		break;
+
+	case 2: // x qpel, y halfpel - left or right during qpel refinement
+		GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref2);
+
+		interpolate8x8_avg2(Reference, ref1, ref2, iEdgedWidth, rounding);
+		break;
+
+	default: // x and y in qpel resolution - the "corners" (top left/right and
+			 // bottom left/right) during qpel refinement
+		GET_REFERENCE(halfpelMV.x, y - halfpelMV.y, ref2);
+		GET_REFERENCE(x - halfpelMV.x, halfpelMV.y, ref3);
+		GET_REFERENCE(x - halfpelMV.x, y - halfpelMV.y, ref4);
+
+		interpolate8x8_avg4(Reference, ref1, ref2, ref3, ref4, iEdgedWidth, rounding);
+		break;
+	}
+
+	sad = sad8(data->Cur, Reference, data->iEdgedWidth);
+	sad += data->lambda8 * d_mv_bits(x - data->predQMV.x, y - data->predQMV.y, data->iFcode);
+
+	if (sad < *(data->iMinSAD)) {
+		*(data->iMinSAD) = sad;
+		data->currentQMV->x = x; data->currentQMV->y = y;
+		*dir = Direction; }
+}
+
 /* CHECK_CANDIATE FUNCTIONS END */
 
 /* MAINSEARCH FUNCTIONS START */
@@ -504,6 +706,28 @@
 	CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0);
 }
 
+
+static void
+QuarterpelRefine(const SearchData * const data)
+{
+/* Perform quarter pixel refinement*/
+
+	VECTOR backupMV = *(data->currentQMV);
+	int iDirection; //not needed
+
+	CHECK_CANDIDATE(backupMV.x - 1, backupMV.y - 1, 0);
+	CHECK_CANDIDATE(backupMV.x + 1, backupMV.y - 1, 0);
+	CHECK_CANDIDATE(backupMV.x - 1, backupMV.y + 1, 0);
+	CHECK_CANDIDATE(backupMV.x + 1, backupMV.y + 1, 0);
+
+	CHECK_CANDIDATE(backupMV.x - 1, backupMV.y, 0);
+	CHECK_CANDIDATE(backupMV.x + 1, backupMV.y, 0);
+
+	CHECK_CANDIDATE(backupMV.x, backupMV.y + 1, 0);
+	CHECK_CANDIDATE(backupMV.x, backupMV.y - 1, 0);
+
+}
+
 static __inline int
 SkipDecisionP(const IMAGE * current, const IMAGE * reference,
 							const int x, const int y,
@@ -517,8 +741,8 @@
 	uint32_t sadC = sad8(current->u + x*8 + y*(iEdgedWidth/2)*8,
 					reference->u + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2);
 	if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
-	sadC += sad8(current->v + x*8 + y*(iEdgedWidth/2)*8,
-					reference->v + x*8 + y*(iEdgedWidth/2)*8, iEdgedWidth/2);
+	sadC += sad8(current->v + (x + y*(iEdgedWidth/2))*8,
+					reference->v + (x + y*(iEdgedWidth/2))*8, iEdgedWidth/2);
 	if (sadC > iQuant * MAX_CHROMA_SAD_FOR_SKIP) return 0;
 
 	return 1;
@@ -530,6 +754,10 @@
 	pMB->mode = MODE_NOT_CODED;
 	pMB->mvs[0].x = pMB->mvs[1].x = pMB->mvs[2].x = pMB->mvs[3].x = 0;
 	pMB->mvs[0].y = pMB->mvs[1].y = pMB->mvs[2].y = pMB->mvs[3].y = 0;
+
+	pMB->qmvs[0].x = pMB->qmvs[1].x = pMB->qmvs[2].x = pMB->qmvs[3].x = 0;
+	pMB->qmvs[0].y = pMB->qmvs[1].y = pMB->qmvs[2].y = pMB->qmvs[3].y = 0;
+
 	pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
 }
 
@@ -550,18 +778,27 @@
 
 	uint32_t x, y;
 	uint32_t iIntra = 0;
-	int32_t InterBias;
+	int32_t InterBias, quant = current->quant;
+	uint8_t *qimage;
 
 	// some pre-initialized thingies for SearchP
 	int32_t temp[5];
 	VECTOR currentMV[5];
+	VECTOR currentQMV[5];
 	int32_t iMinSAD[5];
 	SearchData Data;
 	Data.iEdgedWidth = pParam->edged_width;
 	Data.currentMV = currentMV;
+	Data.currentQMV = currentQMV;
 	Data.iMinSAD = iMinSAD;
 	Data.temp = temp;
 	Data.iFcode = current->fcode;
+	Data.rounding = pParam->m_rounding_type;
+
+	if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL)
+		return 1; // allocate some mem for qpel interpolated blocks
+				  // somehow this is dirty since I think we shouldn't use malloc outside
+				  // encoder_create() - so please fix me!
 
 	if (sadInit) (*sadInit) ();
 
@@ -576,27 +813,36 @@
 
 			if (!(current->global_flags & XVID_LUMIMASKING)) {
 				pMB->dquant = NO_CHANGE;
-				pMB->quant = current->quant; }
+				pMB->quant = current->quant; 
+			} else {
+				if (pMB->dquant != NO_CHANGE) {
+					quant += DQtab[pMB->dquant];
+					if (quant > 31) quant = 31;
+					else if (quant < 1) quant = 1;
+				}
+				pMB->quant = quant;
+			}
 
 //initial skip decision
 
-			if ((pMB->dquant == NO_CHANGE) && (sad00 <= MAX_SAD00_FOR_SKIP * pMB->quant)
-				&& (SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) ) {
-				if (pMB->sad16 < pMB->quant * INITIAL_SKIP_THRESH) {
-						SkipMacroblockP(pMB, sad00);
-						continue;
+			if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * INITIAL_SKIP_THRESH)
+				if (SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) {
+					SkipMacroblockP(pMB, sad00);
+					continue;
 				}
-			} else sad00 = 256*4096; // skip not allowed - for final skip decision
 
-			SearchP(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
+			SearchP(pRef->y, pRefH->y, pRefV->y, pRefHV->y, qimage, pCurrent, x,
 						y, current->motion_flags, pMB->quant,
 						&Data, pParam, pMBs, reference->mbs,
 						current->global_flags & XVID_INTER4V, pMB);
 
 /* final skip decision, a.k.a. "the vector you found, really that good?" */
-			if (sad00 < pMB->quant * MAX_SAD00_FOR_SKIP)
-				if ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH)
-				{ SkipMacroblockP(pMB, sad00); continue; }
+			if (pMB->dquant == NO_CHANGE && sad00 < pMB->quant * MAX_SAD00_FOR_SKIP
+				&& ((100*pMB->sad16)/(sad00+1) > FINAL_SKIP_THRESH) )
+				if (SkipDecisionP(pCurrent, pRef, x, y, pParam->edged_width, pMB->quant)) {
+					SkipMacroblockP(pMB, sad00);
+					continue;
+				}
 	
 /* finally, intra decision */
 
@@ -613,16 +859,19 @@
 						  pParam->edged_width);
 
 				if (deviation < (pMB->sad16 - InterBias)) {
-					if (++iIntra >= iLimit) return 1;
+					if (++iIntra >= iLimit) { free(qimage); return 1; }
 					pMB->mode = MODE_INTRA;
 					pMB->mvs[0] = pMB->mvs[1] = pMB->mvs[2] =
 							pMB->mvs[3] = zeroMV;
+					pMB->qmvs[0] = pMB->qmvs[1] = pMB->qmvs[2] =
+							pMB->qmvs[3] = zeroMV;
 					pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] =
 						pMB->sad8[3] = 0;
 				}
 			}
 		}
 	}
+	free(qimage);
 	return 0;
 }
 
@@ -684,6 +933,7 @@
 		const uint8_t * const pRefH,
 		const uint8_t * const pRefV,
 		const uint8_t * const pRefHV,
+		const uint8_t * const pRefQ,
 		const IMAGE * const pCur,
 		const int x,
 		const int y,
@@ -702,55 +952,61 @@
 
 	get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp);  //has to be changed to get_pmv(2)()
 	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
-				pParam->width, pParam->height, Data->iFcode);
+				pParam->width, pParam->height, Data->iFcode, pParam->m_quarterpel);
 
 	Data->predMV = pmv[0];
+
 	Data->Cur = pCur->y + (x + y * Data->iEdgedWidth) * 16;
 	Data->Ref = pRef + (x + Data->iEdgedWidth*y)*16;
 	Data->RefH = pRefH + (x + Data->iEdgedWidth*y) * 16;
 	Data->RefV = pRefV + (x + Data->iEdgedWidth*y) * 16;
 	Data->RefHV = pRefHV + (x + Data->iEdgedWidth*y) * 16;
+	Data->RefQ = pRefQ;
 
-	Data->iQuant = iQuant;
+	Data->lambda16 = lambda_vec16[iQuant];
+	Data->lambda8 = lambda_vec8[iQuant];
 
 	if (!(MotionFlags & PMV_HALFPEL16)) {
 		Data->min_dx = EVEN(Data->min_dx);
 		Data->max_dx = EVEN(Data->max_dx);
 		Data->min_dy = EVEN(Data->min_dy);
 		Data->max_dy = EVEN(Data->max_dy); }
+	
+	if (pMB->dquant != NO_CHANGE) inter4v = 0;
 
-	for(i = 0;  i < 5; i++) Data->iMinSAD[i] = 256*4096;
-
-	if (inter4v) CheckCandidate = CheckCandidate16;
-	else CheckCandidate = CheckCandidate16no4v;
-
-	(*CheckCandidate)(Data->predMV.x, Data->predMV.y, 0, &iDirection, Data);
+	for(i = 0;  i < 5; i++)
+		Data->currentMV[i].x = Data->currentMV[i].y = 0;
 
-	for(i = 0;  i < 5; i++) Data->currentMV[i].x = Data->currentMV[i].y = 0;
+	if (pParam->m_quarterpel) {
+		Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
+		i = d_mv_bits(Data->predQMV.x, Data->predQMV.y, Data->iFcode);
+	} else i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode);
 
-	i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode);
 	Data->iMinSAD[0] = pMB->sad16 + lambda_vec16[iQuant] * i;
 	Data->iMinSAD[1] = pMB->sad8[0] + lambda_vec8[iQuant] * i;
 	Data->iMinSAD[2] = pMB->sad8[1];
 	Data->iMinSAD[3] = pMB->sad8[2];
 	Data->iMinSAD[4] = pMB->sad8[3];
 
-	if (pMB->dquant != NO_CHANGE) inter4v = 0;
-
 	if ((x == 0) && (y == 0)) threshA = 512;
 	else {
-		threshA = Data->temp[0] + 20;
+		threshA = Data->temp[0]; // that's when we keep this SAD atm
 		if (threshA < 512) threshA = 512;
 		if (threshA > 1024) threshA = 1024; }
 
 	PreparePredictionsP(pmv, x, y, pParam->mb_width, pParam->mb_height,
 					prevMBs + x + y * pParam->mb_width);
 
+	if (pParam->m_quarterpel) CheckCandidate = CheckCandidate16Q;
+	else 
+		if (inter4v) CheckCandidate = CheckCandidate16;
+		else CheckCandidate = CheckCandidate16no4v;
+
 /* main loop. checking all predictions */
 
 	for (i = 1; i < 7; i++) {
 		if (!(mask = make_mask(pmv, i)) ) continue;
-		CheckCandidate16(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
+		(*CheckCandidate)(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
 		if (Data->iMinSAD[0] <= threshA) break;
 	}
 
@@ -779,7 +1035,7 @@
 			if (!(MVequal(startMV, backupMV))) {
 				bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
 
-				CheckCandidate16(startMV.x, startMV.y, 255, &iDirection, Data);
+				(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
 				(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
 				if (bSAD < Data->iMinSAD[0]) {
 					Data->currentMV[0] = backupMV;
@@ -792,7 +1048,7 @@
 			if (!(MVequal(startMV, backupMV))) {
 				bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
 
-				CheckCandidate16(startMV.x, startMV.y, 255, &iDirection, Data);
+				(*CheckCandidate)(startMV.x, startMV.y, 255, &iDirection, Data);
 				(*MainSearchPtr)(startMV.x, startMV.y, Data, 255);
 				if (bSAD < Data->iMinSAD[0]) {
 					Data->currentMV[0] = backupMV;
@@ -803,11 +1059,26 @@
 
 	if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data);
 
+	for(i = 0; i < 5; i++) {
+		Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
+		Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
+	}
+
+	if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) {
+
+		CheckCandidate = CheckCandidate16_qpel;
+		get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
+				pParam->width, pParam->height, Data->iFcode, 0);
+
+		QuarterpelRefine(Data);
+	}
+
 	if (inter4v) {
 		SearchData Data8;
 		Data8.iFcode = Data->iFcode;
-		Data8.iQuant = Data->iQuant;
+		Data8.lambda8 = Data->lambda8;
 		Data8.iEdgedWidth = Data->iEdgedWidth;
+		Data8.RefQ = Data->RefQ;
 		Search8(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
 		Search8(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
 		Search8(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
@@ -822,18 +1093,26 @@
 		pMB->mvs[0] = pMB->mvs[1]
 			= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
 
+		pMB->qmvs[0] = pMB->qmvs[1]
+			= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
+
 		pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] =
 			pMB->sad8[2] = pMB->sad8[3] =  Data->iMinSAD[0];
 
-		pMB->pmvs[0].x = Data->currentMV[0].x - Data->predMV.x;
-		pMB->pmvs[0].y = Data->currentMV[0].y - Data->predMV.y;
+		if(pParam->m_quarterpel) {
+			pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predQMV.x;
+			pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predQMV.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 {
 // INTER4V MODE; all other things are already set in Search8 
 		pMB->mode = MODE_INTER4V;
 		pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + 
 			Data->iMinSAD[3] + Data->iMinSAD[4] + IMV16X16 * iQuant;
 	}
-
 }
 
 static void
@@ -846,15 +1125,25 @@
 		const int block,
 		SearchData * const Data)
 {
-	Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
 	Data->iMinSAD = OldData->iMinSAD + 1 + block;
 	Data->currentMV = OldData->currentMV + 1 + block;
+	Data->currentQMV = OldData->currentQMV + 1 + block;
 
-	if (block != 0)
-		*(Data->iMinSAD) += lambda_vec8[Data->iQuant] *
-								d_mv_bits(	Data->currentMV->x - Data->predMV.x, 
-											Data->currentMV->y - Data->predMV.y,
-											Data->iFcode);
+	if(pParam->m_quarterpel) {
+		Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
+		if (block != 0)	*(Data->iMinSAD) += Data->lambda8 *
+									d_mv_bits(	Data->currentQMV->x - Data->predQMV.x, 
+												Data->currentQMV->y - Data->predQMV.y,
+												Data->iFcode);
+		CheckCandidate = CheckCandidate8Q;
+	} else {
+		Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
+		if (block != 0)	*(Data->iMinSAD) += Data->lambda8 *
+									d_mv_bits(	Data->currentMV->x - Data->predMV.x, 
+												Data->currentMV->y - Data->predMV.y,
+												Data->iFcode);
+		CheckCandidate = CheckCandidate8;
+	}
 
 	if (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8)) {
 
@@ -866,27 +1155,59 @@
 		Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1));
 		
 		get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
-				pParam->width, pParam->height, OldData->iFcode);
-
-		CheckCandidate = CheckCandidate8;
+				pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel);
 
 		if (MotionFlags & PMV_EXTSEARCH8) {
-
+			int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
+			
 			MainSearchFunc *MainSearchPtr;
 			if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
 				else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
 					else MainSearchPtr = DiamondSearch;
 
-			(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);	}
+			(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
+
+			if(*(Data->iMinSAD) < temp_sad) {
+					Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
+					Data->currentQMV->y = 2 * Data->currentMV->y;
+			}
+		}
+
+		if (MotionFlags & PMV_HALFPELREFINE8) {
+			int32_t temp_sad = *(Data->iMinSAD); // store current MinSAD
 
-		if (MotionFlags & PMV_HALFPELREFINE8) HalfpelRefine(Data);
+			HalfpelRefine(Data); // perform halfpel refine of current best vector
+
+			if(*(Data->iMinSAD) < temp_sad) { // we have found a better match
+				Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
+				Data->currentQMV->y = 2 * Data->currentMV->y;
+			}
+		}
+
+		if(pParam->m_quarterpel) {
+			if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) &&
+				(MotionFlags & PMV_QUARTERPELREFINE8)) {
+			CheckCandidate = CheckCandidate8_qpel;
+			get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
+				pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel);
+			QuarterpelRefine(Data);
+			}
+		}
+	}
+
+	if(pParam->m_quarterpel) {
+		pMB->pmvs[block].x = Data->currentQMV->x - Data->predQMV.x;
+		pMB->pmvs[block].y = Data->currentQMV->y - Data->predQMV.y;
+	}
+	else {
+		pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
+		pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
 	}
 
-	pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
-	pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
 	pMB->mvs[block] = *(Data->currentMV);
-	pMB->sad8[block] =  4 * (*Data->iMinSAD);
+	pMB->qmvs[block] = *(Data->currentQMV);
 
+	pMB->sad8[block] =  4 * (*Data->iMinSAD);
 }
 
 /* B-frames code starts here */
@@ -947,59 +1268,44 @@
 			const IMAGE * const pCur,
 			const int x, const int y,
 			const uint32_t MotionFlags,
-			const uint32_t iQuant,
 			const uint32_t iFcode,
 			const MBParam * const pParam,
 			MACROBLOCK * const pMB,
 			const VECTOR * const predMV,
 			int32_t * const best_sad,
-			const int32_t mode_current)
+			const int32_t mode_current,
+			SearchData * const Data)
 {
 
 	const int32_t iEdgedWidth = pParam->edged_width;
 	
 	int i, iDirection, mask;
-	VECTOR currentMV, pmv[7];
+	VECTOR pmv[7];
 	MainSearchFunc *MainSearchPtr;
-	int32_t iMinSAD = MV_MAX_ERROR;
-	SearchData Data;
-
-	Data.iMinSAD = &iMinSAD;
-	Data.Cur = pCur->y + (x + y * iEdgedWidth) * 16;
-	Data.iEdgedWidth = iEdgedWidth;
-	Data.currentMV = &currentMV;
-	Data.iMinSAD = &iMinSAD;
-	Data.Ref = pRef + (x + y * iEdgedWidth) * 16;
-	Data.RefH = pRefH + (x + y * iEdgedWidth) * 16;
-	Data.RefV = pRefV + (x + y * iEdgedWidth) * 16;
-	Data.RefHV = pRefHV + (x + y * iEdgedWidth) * 16;
-
-	Data.iQuant = iQuant;
-	Data.iFcode = iFcode;
-	Data.predMV = *predMV;
+	*Data->iMinSAD = MV_MAX_ERROR;
+	Data->iFcode = iFcode;
 
-	get_range(&Data.min_dx, &Data.max_dx, &Data.min_dy, &Data.max_dy, x, y, 16,
-				pParam->width, pParam->height, iFcode);
+	Data->Ref = pRef + (x + y * iEdgedWidth) * 16;
+	Data->RefH = pRefH + (x + y * iEdgedWidth) * 16;
+	Data->RefV = pRefV + (x + y * iEdgedWidth) * 16;
+	Data->RefHV = pRefHV + (x + y * iEdgedWidth) * 16;
 
-	if (!(MotionFlags & PMV_HALFPEL16)) {
-		Data.min_dx = EVEN(Data.min_dx);
-		Data.max_dx = EVEN(Data.max_dx);
-		Data.min_dy = EVEN(Data.min_dy);
-		Data.max_dy = EVEN(Data.max_dy); } // no-halpel and b-frames. do we need it?
+	Data->predMV = *predMV;
 
+	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
+				pParam->width, pParam->height, iFcode, pParam->m_quarterpel);
 
-	pmv[0] = Data.predMV;
-	PreparePredictionsBF(pmv, x, y, pParam->mb_width,
-					pMB, mode_current);
+	pmv[0] = Data->predMV;
+	PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current);
 
-	currentMV.x = currentMV.y = 0;
+	Data->currentMV->x = Data->currentMV->y = 0;
 
 	CheckCandidate = CheckCandidate16no4v;
 
 // main loop. checking all predictions
 	for (i = 0; i < 8; i++) {
 		if (!(mask = make_mask(pmv, i)) ) continue;
-		CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, &Data);
+		CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
 	}
 
 	if (MotionFlags & PMV_USESQUARES16)
@@ -1008,23 +1314,22 @@
 		MainSearchPtr = AdvDiamondSearch;
 		else MainSearchPtr = DiamondSearch;
 
-	(*MainSearchPtr)(currentMV.x, currentMV.y, &Data, 255);
+	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
 
-	if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(&Data);
+	HalfpelRefine(Data);
 
 // three bits are needed to code backward mode. four for forward
 // we treat the bits just like they were vector's
-	if (mode_current == MODE_FORWARD) iMinSAD +=  4 * lambda_vec16[iQuant];
-	else iMinSAD +=  3 * lambda_vec16[iQuant];
-
+	if (mode_current == MODE_FORWARD) *Data->iMinSAD +=  4 * Data->lambda16 * 2;
+	else *Data->iMinSAD +=  3 * Data->lambda16 * 2;
 
-	if (iMinSAD < *best_sad) {
-		*best_sad = iMinSAD;
+	if (*Data->iMinSAD < *best_sad) {
+		*best_sad = *Data->iMinSAD;
 		pMB->mode = mode_current;
-		pMB->pmvs[0].x = currentMV.x - predMV->x;
-		pMB->pmvs[0].y = currentMV.y - predMV->y;
-		if (mode_current == MODE_FORWARD) pMB->mvs[0] = currentMV;
-		else pMB->b_mvs[0] = currentMV;
+		pMB->pmvs[0].x = Data->currentMV->x - predMV->x;
+		pMB->pmvs[0].y = Data->currentMV->y - predMV->y;
+		if (mode_current == MODE_FORWARD) pMB->mvs[0] = *(Data->currentMV+2) = *Data->currentMV;
+		else pMB->b_mvs[0] = *(Data->currentMV+1) = *Data->currentMV; //we store currmv for interpolate search
 	}
 	
 }
@@ -1041,56 +1346,44 @@
 				const IMAGE * const pCur,
 				const int x, const int y,
 				const uint32_t MotionFlags,
-				const uint32_t iQuant,
 				const int32_t TRB, const int32_t TRD,
 				const MBParam * const pParam,
 				MACROBLOCK * const pMB,
 				const MACROBLOCK * const b_mb,
-				int32_t * const best_sad)
+				int32_t * const best_sad,
+				SearchData * const Data)
 
 {
-	const uint32_t iEdgedWidth = pParam->edged_width;
-	int32_t iMinSAD = 266*4096, skip_sad;
+	int32_t skip_sad;
 	int k;
-	VECTOR currentMV;
+
 	MainSearchFunc *MainSearchPtr;
-	SearchData Data;
 
-	Data.iMinSAD = &iMinSAD;
-	Data.Cur = pCur->y + x * 16 + y * 16 * iEdgedWidth;
-	Data.iEdgedWidth = iEdgedWidth;
-	Data.currentMV = &currentMV;
-	Data.iQuant = iQuant;
-	Data.referencemv = b_mb->mvs;
+	*Data->iMinSAD = 256*4096;
+	Data->referencemv = b_mb->mvs;
 
-	Data.Ref= f_Ref->y + (x + iEdgedWidth*y) * 16;
-	Data.RefH = f_RefH + (x + iEdgedWidth*y) * 16;
-	Data.RefV = f_RefV + (x + iEdgedWidth*y) * 16;
-	Data.RefHV = f_RefHV + (x + iEdgedWidth*y) * 16;
-	Data.bRef = b_Ref->y + (x + iEdgedWidth*y) * 16;
-	Data.bRefH = b_RefH + (x + iEdgedWidth*y) * 16;
-	Data.bRefV = b_RefV + (x + iEdgedWidth*y) * 16;
-	Data.bRefHV = b_RefHV + (x + iEdgedWidth*y) * 16;
-/*
-//What we do here is a complicated version of CheckCandidateDirect(0,0);
-get_range(&Data.min_dx, &Data.max_dx, &Data.min_dy, &Data.max_dy, x, y, 16, pParam->width, pParam->height, 19);
-
-*/
-	Data.max_dx = 2 * pParam->width - 2 * (x) * 16;
-	Data.max_dy = 2 * pParam->height - 2 * (y) * 16;
-	Data.min_dx = -(2 * 16 + 2 * (x) * 16);
-	Data.min_dy = -(2 * 16 + 2 * (y) * 16);
+	Data->Ref = f_Ref->y + (x + Data->iEdgedWidth*y) * 16;
+	Data->RefH = f_RefH + (x + Data->iEdgedWidth*y) * 16;
+	Data->RefV = f_RefV + (x + Data->iEdgedWidth*y) * 16;
+	Data->RefHV = f_RefHV + (x + Data->iEdgedWidth*y) * 16;
+	Data->bRef = b_Ref->y + (x + Data->iEdgedWidth*y) * 16;
+	Data->bRefH = b_RefH + (x + Data->iEdgedWidth*y) * 16;
+	Data->bRefV = b_RefV + (x + Data->iEdgedWidth*y) * 16;
+	Data->bRefHV = b_RefHV + (x + Data->iEdgedWidth*y) * 16;
+
+	Data->max_dx = 2 * pParam->width - 2 * (x) * 16;
+	Data->max_dy = 2 * pParam->height - 2 * (y) * 16;
+	Data->min_dx = -(2 * 16 + 2 * (x) * 16);
+	Data->min_dy = -(2 * 16 + 2 * (y) * 16);
 
 	for (k = 0; k < 4; k++) {
-		pMB->mvs[k].x = Data.directmvF[k].x = ((TRB * Data.referencemv[k].x) / TRD);
-		pMB->b_mvs[k].x = Data.directmvB[k].x = ((TRB - TRD) * Data.referencemv[k].x) / TRD;
-		pMB->mvs[k].y = Data.directmvF[k].y = ((TRB * Data.referencemv[k].y) / TRD);
-		pMB->b_mvs[k].y = Data.directmvB[k].y = ((TRB - TRD) * Data.referencemv[k].y) / TRD;
-
-		if (( pMB->mvs[k].x > Data.max_dx ) || ( pMB->mvs[k].x < Data.min_dx )
-			|| ( pMB->mvs[k].y > Data.max_dy ) || ( pMB->mvs[k].y < Data.min_dy )
-			|| ( pMB->b_mvs[k].x > Data.max_dx ) || ( pMB->b_mvs[k].x < Data.min_dx )
-			|| ( pMB->b_mvs[k].y > Data.max_dy ) || ( pMB->b_mvs[k].y < Data.min_dy )) {
+		pMB->mvs[k].x = Data->directmvF[k].x = ((TRB * Data->referencemv[k].x) / TRD);
+		pMB->b_mvs[k].x = Data->directmvB[k].x = ((TRB - TRD) * Data->referencemv[k].x) / TRD;
+		pMB->mvs[k].y = Data->directmvF[k].y = ((TRB * Data->referencemv[k].y) / TRD);
+		pMB->b_mvs[k].y = Data->directmvB[k].y = ((TRB - TRD) * Data->referencemv[k].y) / TRD;
+
+		if ( ( pMB->b_mvs[k].x > Data->max_dx ) || ( pMB->b_mvs[k].x < Data->min_dx )
+			|| ( pMB->b_mvs[k].y > Data->max_dy ) || ( pMB->b_mvs[k].y < Data->min_dy )) {
 
 			*best_sad = 256*4096; // in that case, we won't use direct mode
 			pMB->mode = MODE_DIRECT; // just to make sure it doesn't say "MODE_DIRECT_NONE_MV"
@@ -1100,8 +1393,8 @@
 		if (b_mb->mode != MODE_INTER4V) {
 			pMB->mvs[1] = pMB->mvs[2] = pMB->mvs[3] = pMB->mvs[0];
 			pMB->b_mvs[1] = pMB->b_mvs[2] = pMB->b_mvs[3] = pMB->b_mvs[0];
-			Data.directmvF[1] = Data.directmvF[2] = Data.directmvF[3] = Data.directmvF[0];
-			Data.directmvB[1] = Data.directmvB[2] = Data.directmvB[3] = Data.directmvB[0];
+			Data->directmvF[1] = Data->directmvF[2] = Data->directmvF[3] = Data->directmvF[0];
+			Data->directmvB[1] = Data->directmvB[2] = Data->directmvB[3] = Data->directmvB[0];
 			break;
 		}
 	}
@@ -1110,11 +1403,11 @@
 		CheckCandidate = CheckCandidateDirect;
 	else CheckCandidate = CheckCandidateDirectno4v;
 
-	(*CheckCandidate)(0, 0, 255, &k, &Data);
+	(*CheckCandidate)(0, 0, 255, &k, Data);
 
 // skip decision
-	if (iMinSAD - 2 * lambda_vec16[iQuant] < (int32_t)iQuant * SKIP_THRESH_B) {
-		//checking chroma. everything copied from MC
+	if (*Data->iMinSAD - 2 * Data->lambda16 < (uint32_t)pMB->quant * SKIP_THRESH_B) {
+		//possible skip - checking chroma. everything copied from MC
 		//this is not full chroma compensation, only it's fullpel approximation. should work though
 		int sum, dx, dy, b_dx, b_dy;
 
@@ -1130,22 +1423,22 @@
 		sum = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y;
 		b_dy = (sum == 0 ? 0 : SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2));
 
-		sum = sad8bi(pCur->u + 8*x + 8*y*(iEdgedWidth/2),
-					f_Ref->u + (y*8 + dy/2) * (iEdgedWidth/2) + x*8 + dx/2,
-					b_Ref->u + (y*8 + b_dy/2) * (iEdgedWidth/2) + x*8 + b_dx/2,
-					iEdgedWidth/2);
-		sum += sad8bi(pCur->v + 8*x + 8*y*(iEdgedWidth/2),
-					f_Ref->v + (y*8 + dy/2) * (iEdgedWidth/2) + x*8 + dx/2,
-					b_Ref->v + (y*8 + b_dy/2) * (iEdgedWidth/2) + x*8 + b_dx/2,
-					iEdgedWidth/2);
+		sum = sad8bi(pCur->u + 8*x + 8*y*(Data->iEdgedWidth/2),
+					f_Ref->u + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2,
+					b_Ref->u + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2,
+					Data->iEdgedWidth/2);
+		sum += sad8bi(pCur->v + 8*x + 8*y*(Data->iEdgedWidth/2),
+					f_Ref->v + (y*8 + dy/2) * (Data->iEdgedWidth/2) + x*8 + dx/2,
+					b_Ref->v + (y*8 + b_dy/2) * (Data->iEdgedWidth/2) + x*8 + b_dx/2,
+					Data->iEdgedWidth/2);
 
-		if ((uint32_t) sum < MAX_CHROMA_SAD_FOR_SKIP * Data.iQuant) {
+		if (sum < MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) {
 			pMB->mode = MODE_DIRECT_NONE_MV;
-			return iMinSAD; 
+			return *Data->iMinSAD; 
 		}
 	}
 
-	skip_sad = iMinSAD;
+	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
@@ -1154,37 +1447,38 @@
 		else if (MotionFlags & PMV_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
 			else MainSearchPtr = DiamondSearch;
 
-	(*MainSearchPtr)(0, 0, &Data, 255);
+	(*MainSearchPtr)(0, 0, Data, 255);
 
-	HalfpelRefine(&Data);
+	HalfpelRefine(Data);
 
-	iMinSAD +=  1 * lambda_vec16[iQuant]; // one bit is needed to code direct mode. we treat this bit just like it was vector's
-	*best_sad = iMinSAD;
+	*Data->iMinSAD +=  1 * Data->lambda16 * 2; // one bit is needed to code direct mode
+	*best_sad = *Data->iMinSAD;
 
 	if (b_mb->mode == MODE_INTER4V) 
 		pMB->mode = MODE_DIRECT;
 	else pMB->mode = MODE_DIRECT_NO4V; //for faster compensation
 
-	pMB->pmvs[3] = currentMV;
+	pMB->pmvs[3] = *Data->currentMV;
 
 	for (k = 0; k < 4; k++) {
-		pMB->mvs[k].x = Data.directmvF[k].x + currentMV.x;
-		pMB->b_mvs[k].x = ((currentMV.x == 0)
-							? Data.directmvB[k].x
-							: pMB->mvs[k].x - Data.referencemv[k].x);
-		pMB->mvs[k].y = (Data.directmvF[k].y + currentMV.y);
-		pMB->b_mvs[k].y = ((currentMV.y == 0)
-							? Data.directmvB[k].y
-							: pMB->mvs[k].y - Data.referencemv[k].y);
+		pMB->mvs[k].x = Data->directmvF[k].x + Data->currentMV->x;
+		pMB->b_mvs[k].x = ((Data->currentMV->x == 0)
+							? Data->directmvB[k].x
+							: pMB->mvs[k].x - Data->referencemv[k].x);
+		pMB->mvs[k].y = (Data->directmvF[k].y + Data->currentMV->y);
+		pMB->b_mvs[k].y = ((Data->currentMV->y == 0)
+							? Data->directmvB[k].y
+							: pMB->mvs[k].y - Data->referencemv[k].y);
 		if (b_mb->mode != MODE_INTER4V) {
 			pMB->mvs[3] = pMB->mvs[2] = pMB->mvs[1] = pMB->mvs[0];
 			pMB->b_mvs[3] = pMB->b_mvs[2] = pMB->b_mvs[1] = pMB->b_mvs[0];
 			break;
 		}
 	}
-	return 0;//skip_sad;
+	return skip_sad;
 }
 
+
 static __inline void
 SearchInterpolate(const uint8_t * const f_Ref,
 				const uint8_t * const f_RefH,
@@ -1199,78 +1493,70 @@
 				const uint32_t fcode,
 				const uint32_t bcode,
 				const uint32_t MotionFlags,
-				const uint32_t iQuant,
 				const MBParam * const pParam,
 				const VECTOR * const f_predMV,
 				const VECTOR * const b_predMV,
 				MACROBLOCK * const pMB,
-				int32_t * const best_sad)
+				int32_t * const best_sad,
+				SearchData * const fData)
 
 {
-/* Interpolated MC motion vector search, this is tedious and more complicated because there are
-   two values for everything, always one for backward and one for forward ME. Still, we don't gain
-   much from this search, maybe it should simply be skipped and simply current i_sad16 value used
-   as "optimal". */
 
 	const int32_t iEdgedWidth = pParam->edged_width;
 
 	int iDirection, i, j;
-	int32_t iMinSAD = 256*4096;
-	VECTOR currentMV[3];
-	SearchData fData, bData;
+	SearchData bData;
 
-	fData.iMinSAD = bData.iMinSAD = &iMinSAD;
+	*(bData.iMinSAD = fData->iMinSAD) = 4096*256;
+	bData.Cur = fData->Cur;
+	fData->iEdgedWidth = bData.iEdgedWidth = iEdgedWidth;
+	bData.currentMV = fData->currentMV + 1;
+	bData.lambda16 = fData->lambda16;
+	fData->iFcode = bData.bFcode = fcode; fData->bFcode = bData.iFcode = bcode;
+
+	bData.bRef = fData->Ref = f_Ref + (x + y * iEdgedWidth) * 16;
+	bData.bRefH = fData->RefH = f_RefH + (x + y * iEdgedWidth) * 16;
+	bData.bRefV = fData->RefV = f_RefV + (x + y * iEdgedWidth) * 16;
+	bData.bRefHV = fData->RefHV = f_RefHV + (x + y * iEdgedWidth) * 16;
+	bData.Ref = fData->bRef = b_Ref + (x + y * iEdgedWidth) * 16;
+	bData.RefH = fData->bRefH = b_RefH + (x + y * iEdgedWidth) * 16;
+	bData.RefV = fData->bRefV = b_RefV + (x + y * iEdgedWidth) * 16;
+	bData.RefHV = fData->bRefHV = b_RefHV + (x + y * iEdgedWidth) * 16;
+
+	bData.bpredMV = fData->predMV = *f_predMV;
+	fData->bpredMV = bData.predMV = *b_predMV;
+
+	fData->currentMV[0] = fData->currentMV[3]; //forward search stored it's vector here. backward stored it in the place it's needed
+	get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, pParam->m_quarterpel);
+	get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, pParam->m_quarterpel);
+
+	if (fData->currentMV[0].x > fData->max_dx) fData->currentMV[0].x = fData->max_dx;
+	if (fData->currentMV[0].x < fData->min_dx) fData->currentMV[0].x = fData->min_dy;
+	if (fData->currentMV[0].y > fData->max_dy) fData->currentMV[0].y = fData->max_dx;
+	if (fData->currentMV[0].y > fData->min_dy) fData->currentMV[0].y = fData->min_dy;
+
+	if (fData->currentMV[1].x > bData.max_dx) fData->currentMV[1].x = bData.max_dx;
+	if (fData->currentMV[1].x < bData.min_dx) fData->currentMV[1].x = bData.min_dy;
+	if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dx;
+	if (fData->currentMV[1].y > bData.min_dy) fData->currentMV[1].y = bData.min_dy;
 
-	fData.Cur = bData.Cur = pCur->y + (x + y * iEdgedWidth) * 16;
-	fData.iEdgedWidth = bData.iEdgedWidth = iEdgedWidth;
-	fData.currentMV = currentMV; bData.currentMV = currentMV + 1;
-	fData.iQuant = bData.iQuant = iQuant;
-	fData.iFcode = bData.bFcode = fcode; fData.bFcode = bData.iFcode = bcode;
-
-	bData.bRef = fData.Ref = f_Ref + (x + y * iEdgedWidth) * 16;
-	bData.bRefH = fData.RefH = f_RefH + (x + y * iEdgedWidth) * 16;
-	bData.bRefV = fData.RefV = f_RefV + (x + y * iEdgedWidth) * 16;
-	bData.bRefHV = fData.RefHV = f_RefHV + (x + y * iEdgedWidth) * 16;
-	bData.Ref = fData.bRef = b_Ref + (x + y * iEdgedWidth) * 16;
-	bData.RefH = fData.bRefH = b_RefH + (x + y * iEdgedWidth) * 16;
-	bData.RefV = fData.bRefV = b_RefV + (x + y * iEdgedWidth) * 16;
-	bData.RefHV = fData.bRefHV = b_RefHV + (x + y * iEdgedWidth) * 16;
-
-	bData.bpredMV = fData.predMV = *f_predMV;
-	fData.bpredMV = bData.predMV = *b_predMV;
-
-	currentMV[0] = pMB->mvs[0];
-	currentMV[1] = pMB->b_mvs[0];
-	get_range(&fData.min_dx, &fData.max_dx, &fData.min_dy, &fData.max_dy, x, y, 16, pParam->width, pParam->height, fcode);
-	get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode);
-
-	if (currentMV[0].x > fData.max_dx) currentMV[0].x = fData.max_dx;
-	if (currentMV[0].x < fData.min_dx) currentMV[0].x = fData.min_dy;
-	if (currentMV[0].y > fData.max_dy) currentMV[0].y = fData.max_dx;
-	if (currentMV[0].y > fData.min_dy) currentMV[0].y = fData.min_dy;
-
-	if (currentMV[1].x > bData.max_dx) currentMV[1].x = bData.max_dx;
-	if (currentMV[1].x < bData.min_dx) currentMV[1].x = bData.min_dy;
-	if (currentMV[1].y > bData.max_dy) currentMV[1].y = bData.max_dx;
-	if (currentMV[1].y > bData.min_dy) currentMV[1].y = bData.min_dy;
-
-	CheckCandidateInt(currentMV[0].x, currentMV[0].y, 255, &iDirection, &fData);
+	CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData);
 
 //diamond. I wish we could use normal mainsearch functions (square, advdiamond)
 
 	do {
 		iDirection = 255;
 		// forward MV moves
-		i = currentMV[0].x; j = currentMV[0].y;
+		i = fData->currentMV[0].x; j = fData->currentMV[0].y;
 
-		CheckCandidateInt(i + 1, j, 0, &iDirection, &fData);
-		CheckCandidateInt(i, j + 1, 0, &iDirection, &fData);
-		CheckCandidateInt(i - 1, j, 0, &iDirection, &fData);
-		CheckCandidateInt(i, j - 1, 0, &iDirection, &fData);
+		CheckCandidateInt(i + 1, j, 0, &iDirection, fData);
+		CheckCandidateInt(i, j + 1, 0, &iDirection, fData);
+		CheckCandidateInt(i - 1, j, 0, &iDirection, fData);
+		CheckCandidateInt(i, j - 1, 0, &iDirection, fData);
 
 		// backward MV moves
-		i = currentMV[1].x; j = currentMV[1].y;
-		currentMV[2] = currentMV[0];
+		i = fData->currentMV[1].x; j = fData->currentMV[1].y;
+		fData->currentMV[2] = fData->currentMV[0];
 
 		CheckCandidateInt(i + 1, j, 0, &iDirection, &bData);
 		CheckCandidateInt(i, j + 1, 0, &iDirection, &bData);
@@ -1280,11 +1566,12 @@
 	} while (!(iDirection));
 
 // two bits are needed to code interpolate mode. we treat the bits just like they were vector's
-	iMinSAD +=  2 * lambda_vec16[iQuant];
-	if (iMinSAD < *best_sad) {
-		*best_sad = iMinSAD;
-		pMB->mvs[0] = currentMV[0];
-		pMB->b_mvs[0] = currentMV[1];
+	*fData->iMinSAD +=  2 * fData->lambda16 * 2; 
+
+	if (*fData->iMinSAD < *best_sad) {
+		*best_sad = *fData->iMinSAD;
+		pMB->mvs[0] = fData->currentMV[0];
+		pMB->b_mvs[0] = fData->currentMV[1];
 		pMB->mode = MODE_INTERPOLATE;
 
 		pMB->pmvs[1].x = pMB->mvs[0].x - f_predMV->x;
@@ -1294,6 +1581,7 @@
 	}
 }
 
+
 void
 MotionEstimationBVOP(MBParam * const pParam,
 					 FRAMEINFO * const frame,
@@ -1322,6 +1610,16 @@
 	const int32_t TRB = time_pp - time_bp;
 	const int32_t TRD = time_pp;
 
+// some pre-inintialized data for the rest of the search
+
+	SearchData Data;
+	int32_t iMinSAD;
+	VECTOR currentMV[3];
+	Data.iEdgedWidth = pParam->edged_width;
+	Data.currentMV = currentMV;
+	Data.iMinSAD = &iMinSAD;
+	Data.lambda16 = lambda_vec16[frame->quant];
+
 	// note: i==horizontal, j==vertical
 
 	for (j = 0; j < pParam->mb_height; j++) {
@@ -1338,6 +1636,8 @@
 				continue;
 			}
 
+			Data.Cur = frame->image.y + (j * Data.iEdgedWidth + i) * 16;
+			pMB->quant = frame->quant;
 /* direct search comes first, because it (1) checks for SKIP-mode
 	and (2) sets very good predictions for forward and backward search */
 
@@ -1346,11 +1646,11 @@
 									&frame->image,
 									i, j,
 									frame->motion_flags,
-									frame->quant,
 									TRB, TRD,
 									pParam,
 									pMB, b_mb,
-									&best_sad);
+									&best_sad,
+									&Data);
 
 			if (pMB->mode == MODE_DIRECT_NONE_MV) { n_count++; continue; }
 
@@ -1361,17 +1661,17 @@
 			SearchBF(f_ref->y, f_refH->y, f_refV->y, f_refHV->y,
 						&frame->image, i, j,
 						frame->motion_flags,
-						frame->quant, frame->fcode, pParam,
+						frame->fcode, pParam,
 						pMB, &f_predMV, &best_sad,
-						MODE_FORWARD);
+						MODE_FORWARD, &Data);
 
 			// backward search
 			SearchBF(b_ref->y, b_refH->y, b_refV->y, b_refHV->y,
 						&frame->image, i, j,
 						frame->motion_flags,
-						frame->quant, frame->bcode, pParam,
+						frame->bcode, pParam,
 						pMB, &b_predMV, &best_sad,
-						MODE_BACKWARD);
+						MODE_BACKWARD, &Data);
 
 			// interpolate search comes last, because it uses data from forward and backward as prediction
 
@@ -1381,9 +1681,10 @@
 						i, j,
 						frame->fcode, frame->bcode,
 						frame->motion_flags,
-						frame->quant, pParam,
+						pParam,
 						&f_predMV, &b_predMV,
-						pMB, &best_sad);
+						pMB, &best_sad,
+						&Data);
 
 			switch (pMB->mode) {
 				case MODE_FORWARD:
@@ -1416,61 +1717,92 @@
 
 /* Hinted ME starts here */
 
-static __inline void 
-Search8hinted(	const SearchData * const OldData,
-				const int x, const int y,
-				const uint32_t MotionFlags,
-				const MBParam * const pParam,
-				MACROBLOCK * const pMB,
-				const MACROBLOCK * const pMBs,
-				const int block)
+static void
+Search8hinted(const SearchData * const OldData,
+			const int x, const int y,
+			const uint32_t MotionFlags,
+			const MBParam * const pParam,
+			MACROBLOCK * const pMB,
+			const MACROBLOCK * const pMBs,
+			const int block,
+			SearchData * const Data)
 {
-	SearchData Data;
+	int32_t temp_sad;
 	MainSearchFunc *MainSearchPtr;
+	Data->iMinSAD = OldData->iMinSAD + 1 + block;
+	Data->currentMV = OldData->currentMV + 1 + block;
+	Data->currentQMV = OldData->currentQMV + 1 + block;
+	Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
 
-	Data.predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
-	Data.iMinSAD = OldData->iMinSAD + 1 + block;
-	Data.currentMV = OldData->currentMV+1+block;
-	Data.iFcode = OldData->iFcode;
-	Data.iQuant = OldData->iQuant;
-
-	Data.Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1));
-	Data.RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1));
-	Data.RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1));
-	Data.RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1));
-	Data.iEdgedWidth = pParam->edged_width;
-	Data.Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1));
-
-	CheckCandidate = CheckCandidate8;
-
-	if (block != 0)
-		*(Data.iMinSAD) += lambda_vec8[Data.iQuant] * 
-								d_mv_bits(	Data.currentMV->x - Data.predMV.x, 
-											Data.currentMV->y - Data.predMV.y, 
-											Data.iFcode);
-
+	if(pParam->m_quarterpel) {
+		Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
+		if (block != 0)	*(Data->iMinSAD) += Data->lambda8 *
+									d_mv_bits(	Data->currentQMV->x - Data->predQMV.x, 
+												Data->currentQMV->y - Data->predQMV.y,
+												Data->iFcode);
+		CheckCandidate = CheckCandidate8Q;
+	} else {
+		if (block != 0)	*(Data->iMinSAD) += Data->lambda8 *
+									d_mv_bits(	Data->currentMV->x - Data->predMV.x, 
+												Data->currentMV->y - Data->predMV.y,
+												Data->iFcode);
+		CheckCandidate = CheckCandidate8;
+	}
 
-	get_range(&Data.min_dx, &Data.max_dx, &Data.min_dy, &Data.max_dy, x, y, 8,
-				pParam->width, pParam->height, OldData->iFcode);
+	Data->Ref = OldData->Ref + 8 * ((block&1) + pParam->edged_width*(block>>1));
+	Data->RefH = OldData->RefH + 8 * ((block&1) + pParam->edged_width*(block>>1));
+	Data->RefV = OldData->RefV + 8 * ((block&1) + pParam->edged_width*(block>>1));
+	Data->RefHV = OldData->RefHV + 8 * ((block&1) + pParam->edged_width*(block>>1));
 
-	if (pMB->mode == MODE_INTER4V) {
-		int dummy;
-		CheckCandidate8(pMB->mvs[block].x, pMB->mvs[block].y, 0, &dummy, &Data); }
+	Data->Cur = OldData->Cur + 8 * ((block&1) + pParam->edged_width*(block>>1));
+		
+	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
+				pParam->width, pParam->height, OldData->iFcode, pParam->m_quarterpel);
 
+	temp_sad = *(Data->iMinSAD); // store current MinSAD		
+			
 	if (MotionFlags & PMV_USESQUARES8) MainSearchPtr = SquareSearch;
 		else if (MotionFlags & PMV_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
 			else MainSearchPtr = DiamondSearch;
 
-	(*MainSearchPtr)(Data.currentMV->x, Data.currentMV->y, &Data, 255);
+	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
 
-	if (MotionFlags & PMV_HALFPELREFINE8) HalfpelRefine(&Data);
+	if(*(Data->iMinSAD) < temp_sad) {
+		Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
+		Data->currentQMV->y = 2 * Data->currentMV->y;
+	}
 
-	pMB->pmvs[block].x = Data.currentMV->x - Data.predMV.x;
-	pMB->pmvs[block].y = Data.currentMV->y - Data.predMV.y;
-	pMB->mvs[block] = *(Data.currentMV);
-	pMB->sad8[block] =  4 * (*(Data.iMinSAD));
-}
+	if (MotionFlags & PMV_HALFPELREFINE8) {
+		temp_sad = *(Data->iMinSAD); // store current MinSAD
+
+		HalfpelRefine(Data); // perform halfpel refine of current best vector
 
+		if(*(Data->iMinSAD) < temp_sad) { // we have found a better match
+			Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
+			Data->currentQMV->y = 2 * Data->currentMV->y;
+		}
+	}
+
+	if(pParam->m_quarterpel) {
+		if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) &&
+			(MotionFlags & PMV_QUARTERPELREFINE8)) {
+			get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 8,
+				pParam->width, pParam->height, Data->iFcode, 0);
+			CheckCandidate = CheckCandidate8_qpel;
+			QuarterpelRefine(Data);
+		} 
+	pMB->pmvs[block].x = Data->currentQMV->x - Data->predQMV.x;
+	pMB->pmvs[block].y = Data->currentQMV->y - Data->predQMV.y;
+	} else { 
+		pMB->pmvs[block].x = Data->currentMV->x - Data->predMV.x;
+		pMB->pmvs[block].y = Data->currentMV->y - Data->predMV.y;
+	}
+
+	pMB->mvs[block] = *(Data->currentMV);
+	pMB->qmvs[block] = *(Data->currentQMV);
+
+	pMB->sad8[block] =  4 * (*Data->iMinSAD);
+}
 
 static void
 SearchPhinted (	const uint8_t * const pRef,
@@ -1482,69 +1814,61 @@
 				const int y,
 				const uint32_t MotionFlags,
 				const uint32_t iQuant,
-				const uint32_t iFcode,
 				const MBParam * const pParam,
 				const MACROBLOCK * const pMBs,
 				int inter4v,
-				MACROBLOCK * const pMB)
+				MACROBLOCK * const pMB,
+				SearchData * const Data)
 {
 
 	const int32_t iEdgedWidth = pParam->edged_width;
  
-	int i;
-	VECTOR currentMV[5];
-	int32_t iMinSAD[5];
-	int32_t temp[5];
+	int i, t;
 	MainSearchFunc * MainSearchPtr;
-	SearchData Data;
 
-	Data.predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0);
-	get_range(&Data.min_dx, &Data.max_dx, &Data.min_dy, &Data.max_dy, x, y, 16,
-				pParam->width, pParam->height, iFcode);
+	Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0);
+	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);
+
+	Data->Cur = pCur->y + (x + y * iEdgedWidth) * 16;
+	Data->Ref = pRef + (x + iEdgedWidth*y)*16;
+	Data->RefH = pRefH + (x + iEdgedWidth*y) * 16;
+	Data->RefV = pRefV + (x + iEdgedWidth*y) * 16;
+	Data->RefHV = pRefHV + (x + iEdgedWidth*y) * 16;
+	Data->lambda16 = lambda_vec16[iQuant];
+	Data->lambda8 = lambda_vec8[iQuant];
 
-	Data.Cur = pCur->y + (x + y * iEdgedWidth) * 16;
-	Data.iEdgedWidth = iEdgedWidth;
-	Data.currentMV = currentMV;
-	Data.iMinSAD = iMinSAD;
-	Data.Ref = pRef + (x + iEdgedWidth*y)*16;
-	Data.RefH = pRefH + (x + iEdgedWidth*y) * 16;
-	Data.RefV = pRefV + (x + iEdgedWidth*y) * 16;
-	Data.RefHV = pRefHV + (x + iEdgedWidth*y) * 16;
-	Data.temp = temp;
-	Data.iQuant = iQuant;
-	Data.iFcode = iFcode;
-	
 	if (!(MotionFlags & PMV_HALFPEL16)) {
-		Data.min_dx = EVEN(Data.min_dx);
-		Data.max_dx = EVEN(Data.max_dx);
-		Data.min_dy = EVEN(Data.min_dy);
-		Data.max_dy = EVEN(Data.max_dy); 
+		Data->min_dx = EVEN(Data->min_dx);
+		Data->max_dx = EVEN(Data->max_dx);
+		Data->min_dy = EVEN(Data->min_dy);
+		Data->max_dy = EVEN(Data->max_dy); 
 	}
 
-	for(i = 0; i < 5; i++) iMinSAD[i] = MV_MAX_ERROR;
+	for(i = 0; i < 5; i++) Data->iMinSAD[i] = MV_MAX_ERROR;
 
 	if (pMB->dquant != NO_CHANGE) inter4v = 0;
 
-	if (inter4v)
-		CheckCandidate = CheckCandidate16;
-	else CheckCandidate = CheckCandidate16no4v;
-
+	if(pParam->m_quarterpel) CheckCandidate = CheckCandidate16Q;
+	else
+		if (inter4v) CheckCandidate = CheckCandidate16;
+		else CheckCandidate = CheckCandidate16no4v;
 
 	pMB->mvs[0].x = EVEN(pMB->mvs[0].x);
 	pMB->mvs[0].y = EVEN(pMB->mvs[0].y);
-	if (pMB->mvs[0].x > Data.max_dx) pMB->mvs[0].x = Data.max_dx; // this is in case iFcode changed
-	if (pMB->mvs[0].x < Data.min_dx) pMB->mvs[0].x = Data.min_dx;
-	if (pMB->mvs[0].y > Data.max_dy) pMB->mvs[0].y = Data.max_dy;
-	if (pMB->mvs[0].y < Data.min_dy) pMB->mvs[0].y = Data.min_dy;
+	if (pMB->mvs[0].x > Data->max_dx) pMB->mvs[0].x = Data->max_dx; // this is in case iFcode changed
+	if (pMB->mvs[0].x < Data->min_dx) pMB->mvs[0].x = Data->min_dx;
+	if (pMB->mvs[0].y > Data->max_dy) pMB->mvs[0].y = Data->max_dy;
+	if (pMB->mvs[0].y < Data->min_dy) pMB->mvs[0].y = Data->min_dy;
 
-	CheckCandidate16(pMB->mvs[0].x, pMB->mvs[0].y, 0, &i, &Data);
+	(*CheckCandidate)(pMB->mvs[0].x, pMB->mvs[0].y, 0, &t, Data);
 
 	if (pMB->mode == MODE_INTER4V)
 		for (i = 1; i < 4; i++) { // all four vectors will be used as four predictions for 16x16 search
 			pMB->mvs[i].x = EVEN(pMB->mvs[i].x);
 			pMB->mvs[i].y = EVEN(pMB->mvs[i].y);
 			if (!(make_mask(pMB->mvs, i))) 
-				CheckCandidate16(pMB->mvs[i].x, pMB->mvs[i].y, 0, &i, &Data);
+				(*CheckCandidate)(pMB->mvs[i].x, pMB->mvs[i].y, 0, &t, Data);
 		}
 
 	if (MotionFlags & PMV_USESQUARES16)
@@ -1553,31 +1877,61 @@
 		MainSearchPtr = AdvDiamondSearch;
 		else MainSearchPtr = DiamondSearch;
 
-	(*MainSearchPtr)(currentMV->x, currentMV->y, &Data, 255);
+	(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
 
-	if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(&Data);
+	if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data);
+
+	for(i = 0; i < 5; i++) {
+		Data->currentQMV[i].x = 2 * Data->currentMV[i].x; // initialize qpel vectors
+		Data->currentQMV[i].y = 2 * Data->currentMV[i].y;
+	}
+
+	if((pParam->m_quarterpel) && (MotionFlags & PMV_QUARTERPELREFINE16)) {
+		get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 16,
+				pParam->width, pParam->height, Data->iFcode, 0);
+		CheckCandidate = CheckCandidate16_qpel;
+		QuarterpelRefine(Data);
+	}
 
-	if (inter4v)
-		for(i = 0; i < 4; i++)
-			Search8hinted(&Data, 2*x+(i&1), 2*y+(i>>1), MotionFlags, pParam, pMB, pMBs, i);
+	if (inter4v) {
+		SearchData Data8;
+		Data8.iFcode = Data->iFcode;
+		Data8.lambda8 = Data->lambda8;
+		Data8.iEdgedWidth = Data->iEdgedWidth;
+		Data8.RefQ = Data->RefQ;
+		Search8hinted(Data, 2*x, 2*y, MotionFlags, pParam, pMB, pMBs, 0, &Data8);
+		Search8hinted(Data, 2*x + 1, 2*y, MotionFlags, pParam, pMB, pMBs, 1, &Data8);
+		Search8hinted(Data, 2*x, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 2, &Data8);
+		Search8hinted(Data, 2*x + 1, 2*y + 1, MotionFlags, pParam, pMB, pMBs, 3, &Data8);
+	}
 
 	if (!(inter4v) ||
-		(iMinSAD[0] < iMinSAD[1] + iMinSAD[2] + iMinSAD[3] + iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
+		(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] + 
+							Data->iMinSAD[4] + IMV16X16 * (int32_t)iQuant )) {
 // INTER MODE
-
 		pMB->mode = MODE_INTER;
 		pMB->mvs[0] = pMB->mvs[1]
-			= pMB->mvs[2] = pMB->mvs[3] = currentMV[0];
+			= pMB->mvs[2] = pMB->mvs[3] = Data->currentMV[0];
+
+		pMB->qmvs[0] = pMB->qmvs[1]
+			= pMB->qmvs[2] = pMB->qmvs[3] = Data->currentQMV[0];
 
 		pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] =
-			pMB->sad8[2] = pMB->sad8[3] =  iMinSAD[0];
+			pMB->sad8[2] = pMB->sad8[3] =  Data->iMinSAD[0];
 
-		pMB->pmvs[0].x = currentMV[0].x - Data.predMV.x;
-		pMB->pmvs[0].y = currentMV[0].y - Data.predMV.y;
+		if(pParam->m_quarterpel) {
+			pMB->pmvs[0].x = Data->currentQMV[0].x - Data->predQMV.x;
+			pMB->pmvs[0].y = Data->currentQMV[0].y - Data->predQMV.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 {
 // INTER4V MODE; all other things are already set in Search8hinted
 		pMB->mode = MODE_INTER4V;
-		pMB->sad16 = iMinSAD[1] + iMinSAD[2] + iMinSAD[3] + iMinSAD[4] + IMV16X16 * iQuant;
+		pMB->sad16 = Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] 
+						+ Data->iMinSAD[4] + IMV16X16 * iQuant;
 	}
 
 }
@@ -1595,7 +1949,26 @@
 	const IMAGE *const pRef = &reference->image;
 
 	uint32_t x, y;
-	
+	uint8_t * qimage;
+	int32_t temp[5], quant = current->quant;
+	int32_t iMinSAD[5];
+	VECTOR currentMV[5], currentQMV[5];
+	SearchData Data;
+	Data.iEdgedWidth = pParam->edged_width;
+	Data.currentMV = currentMV;
+	Data.currentQMV = currentQMV;
+	Data.iMinSAD = iMinSAD;
+	Data.temp = temp;
+	Data.iFcode = current->fcode;
+	Data.rounding = pParam->m_rounding_type;
+
+	if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL)
+		return; // allocate some mem for qpel interpolated blocks
+				  // somehow this is dirty since I think we shouldn't use malloc outside
+				  // encoder_create() - so please fix me!
+
+	Data.RefQ = qimage;
+
 	if (sadInit) (*sadInit) ();
 
 	for (y = 0; y < pParam->mb_height; y++)	{
@@ -1606,17 +1979,27 @@
 //intra mode is copied from the first pass. At least for the time being
 			if  ((pMB->mode == MODE_INTRA) || (pMB->mode == MODE_NOT_CODED) ) continue;
 
+
 			if (!(current->global_flags & XVID_LUMIMASKING)) {
 				pMB->dquant = NO_CHANGE;
 				pMB->quant = current->quant; }
+			else {
+				if (pMB->dquant != NO_CHANGE) {
+					quant += DQtab[pMB->dquant];
+					if (quant > 31) quant = 31;
+					else if (quant < 1) quant = 1;
+				}
+				pMB->quant = quant;
+			}
 
 			SearchPhinted(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
 							y, current->motion_flags, pMB->quant,
-							current->fcode, pParam, pMBs,
-							current->global_flags & XVID_INTER4V, pMB);
+							pParam, pMBs, current->global_flags & XVID_INTER4V, pMB,
+							&Data);
 
 		}
 	}
+	free(qimage);
 }
 
 static __inline int
@@ -1624,50 +2007,42 @@
 				const uint8_t * const pCur,
 				const int x,
 				const int y,
-				const uint32_t iFcode,
 				const MBParam * const pParam,
 				const MACROBLOCK * const pMBs,
-				MACROBLOCK * const pMB)
+				MACROBLOCK * const pMB,
+				SearchData * const Data)
 {
 
-	const int32_t iEdgedWidth = pParam->edged_width;
-	int i, mask;
-	VECTOR currentMV, pmv[3];
-	int32_t iMinSAD = MV_MAX_ERROR;
-	SearchData Data;
+	int i = 255, mask;
+	VECTOR pmv[3];
 
-	Data.predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0);
-	get_range(&Data.min_dx, &Data.max_dx, &Data.min_dy, &Data.max_dy, x, y, 16,
-				pParam->width, pParam->height, iFcode);
+	*(Data->iMinSAD) = MV_MAX_ERROR;
+	Data->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x, y, 0);
+	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);
 
-	Data.Cur = pCur + (x + y * iEdgedWidth) * 16;
-	Data.iEdgedWidth = iEdgedWidth;
-	Data.currentMV = &currentMV;
-	Data.iMinSAD = &iMinSAD;
-	Data.Ref = pRef + (x + iEdgedWidth*y)*16;
-	Data.iQuant = 2;
-	Data.iFcode = iFcode;
+	Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
+	Data->Ref = pRef + (x + y * pParam->edged_width) * 16;
 	
 	CheckCandidate = CheckCandidate16no4vI;
 
 	pmv[1].x = EVEN(pMB->mvs[0].x);
 	pmv[1].y = EVEN(pMB->mvs[0].y);
-	pmv[0].x = EVEN(Data.predMV.x);
-	pmv[0].y = EVEN(Data.predMV.y);
+	pmv[0].x = EVEN(Data->predMV.x);
+	pmv[0].y = EVEN(Data->predMV.y);
 	pmv[2].x = pmv[2].y = 0;
 
-	CheckCandidate16no4vI(pmv[0].x, pmv[0].y, 255, &i, &Data);
+	CheckCandidate16no4vI(pmv[0].x, pmv[0].y, 255, &i, Data);
 	if (!(mask = make_mask(pmv, 1)))
-		CheckCandidate16no4vI(pmv[1].x, pmv[1].y, mask, &i, &Data);
+		CheckCandidate16no4vI(pmv[1].x, pmv[1].y, mask, &i, Data);
 	if (!(mask = make_mask(pmv, 2)))
-		CheckCandidate16no4vI(0, 0, mask, &i, &Data);
+		CheckCandidate16no4vI(0, 0, mask, &i, Data);
 
-	DiamondSearch(currentMV.x, currentMV.y, &Data, i);
+	DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i);
 
-	pMB->mvs[0] = pMB->mvs[1]
-			= pMB->mvs[2] = pMB->mvs[3] = currentMV; // all, for future get_pmv()
+	pMB->mvs[0] = *Data->currentMV;
 
-	return iMinSAD;
+	return *(Data->iMinSAD);
 }
 
 #define INTRA_THRESH	1350
@@ -1682,28 +2057,32 @@
 {
 	uint32_t x, y, intra = 0;
 	int sSAD = 0;
-	
+
+	VECTOR currentMV;
+	int32_t iMinSAD;
+	SearchData Data;
+	Data.iEdgedWidth = pParam->edged_width;
+	Data.currentMV = &currentMV;
+	Data.iMinSAD = &iMinSAD;
+	Data.iFcode = iFcode;
+
 	if (sadInit) (*sadInit) ();
 
-	for (y = 0; y < pParam->mb_height-1; y++) {
-		for (x = 0; x < pParam->mb_width; x++) {
+	for (y = 1; y < pParam->mb_height-1; y++) {
+		for (x = 1; x < pParam->mb_width-1; x++) {
 			int sad, dev;
-
 			MACROBLOCK *pMB = &pMBs[x + y * pParam->mb_width];
 
 			sad = MEanalyzeMB(pRef->y, pCurrent->y, x, y,
-								iFcode, pParam, pMBs, pMB);
+								pParam, pMBs, pMB, &Data);
 			
-			if ( x != 0 && y != 0 && x != pParam->mb_width-1 ) { //no edge macroblocks, they just don't work
-				if (sad > INTRA_THRESH) {
-					dev = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
-								  pParam->edged_width);
-					if (dev + INTRA_THRESH < sad) intra++; 
-					if (intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return 2;  // I frame
+			if (sad > INTRA_THRESH) {
+				dev = dev16(pCurrent->y + (x + y * pParam->edged_width) * 16,
+							  pParam->edged_width);
+				if (dev + INTRA_THRESH < sad) intra++; 
+				if (intra > (pParam->mb_height-2)*(pParam->mb_width-2)/2) return 2;  // I frame
 				}
-				sSAD += sad;
-			}
-
+			sSAD += sad;
 		}
 	}
 	sSAD /= (pParam->mb_height-2)*(pParam->mb_width-2);
@@ -1712,3 +2091,33 @@
 	return 0; // B frame
 
 }
+
+int
+FindFcode(	const MBParam * const pParam,
+			const FRAMEINFO * const current)
+{
+	uint32_t x, y;
+	int max = 0, min = 0, i;
+
+	for (y = 0; y < pParam->mb_height; y++) {
+		for (x = 0; x < pParam->mb_width; x++) {
+		
+			MACROBLOCK *pMB = &current->mbs[x + y * pParam->mb_width];
+			for(i = 0; i < (pMB->mode == MODE_INTER4V ? 4:1); i++) {
+				if (pMB->mvs[i].x > max) max = pMB->mvs[i].x;
+				if (pMB->mvs[i].y > max) max = pMB->mvs[i].y;
+
+				if (pMB->mvs[i].x < min) min = pMB->mvs[i].x;
+				if (pMB->mvs[i].y < min) min = pMB->mvs[i].y;
+			}
+		}
+	}
+
+	min = -min;
+	max += 1;
+	if (min > max) max = min;
+	if (pParam->m_quarterpel) max *= 2;
+
+	for (i = 1; (max > 32 << (i - 1)); i++);
+	return i;
+}