--- branches/dev-api-3/xvidcore/src/motion/motion_est.c	2002/10/05 21:39:39	580
+++ branches/dev-api-3/xvidcore/src/motion/motion_est.c	2002/10/17 13:27:22	600
@@ -53,7 +53,7 @@
 (*CheckCandidate)((const int)(X),(const int)(Y), (D), &iDirection, data ); }
 
 #define GET_REFERENCE(X, Y, REF) { \
-	switch ( ((X&1)<<1) + (Y&1) ) \
+	switch ( (((X)&1)<<1) + ((Y)&1) ) \
 	{ \
 		case 0 : REF = data->Ref + (X)/2 + ((Y)/2)*(data->iEdgedWidth); break; \
 		case 1 : REF = data->RefV + (X)/2 + (((Y)-1)/2)*(data->iEdgedWidth); break; \
@@ -95,7 +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;
 
@@ -109,9 +108,13 @@
 		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);
+
+	if(data->quarterpel)
+		t = d_mv_bits(2*x - data->predQMV.x, 2*y - data->predQMV.y, data->iFcode);
+	else
+		t = d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
 
-	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;
 
@@ -148,8 +151,13 @@
 		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);
+	if(data->quarterpel)
+		sad = lambda_vec16[data->iQuant] *
+		      d_mv_bits(2*x - data->predQMV.x, 2*y - data->predQMV.y, data->iFcode);
+	else
+		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, MV_MAX_ERROR);
 
 	if (sad < *(data->iMinSAD)) {
@@ -165,7 +173,6 @@
 // Important: This is no general usable routine! x and y must be +/-1 (qpel resolution!)
 // around currentMV!
 {
-	int32_t * const sad = data->temp;
 	int t;
 	uint8_t * Reference = (uint8_t *) data->RefQ;
 	const uint8_t *ref1, *ref2, *ref3, *ref4;
@@ -217,7 +224,7 @@
 		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->predQMV.x, y - data->predQMV.y, data->iFcode);
 	data->temp[0] += lambda_vec16[data->iQuant] * t;
@@ -226,7 +233,7 @@
 	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; }
+	/*	*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; }
@@ -303,7 +310,7 @@
 	if (sad < *(data->iMinSAD)) {
 		*(data->iMinSAD) = sad;
 		data->currentQMV[0].x = x; data->currentQMV[0].y = y;
-		*dir = Direction;
+//		*dir = Direction;
 	}
 }
 
@@ -488,7 +495,11 @@
 	}
 
 	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);
+	
+	if(data->quarterpel)
+		sad += lambda_vec8[data->iQuant] * d_mv_bits(2*x - data->predQMV.x, 2*y - data->predQMV.y, data->iFcode);
+	else
+		sad += lambda_vec8[data->iQuant] * d_mv_bits(x - data->predMV.x, y - data->predMV.y, data->iFcode);
 
 	if (sad < *(data->iMinSAD)) {
 		*(data->iMinSAD) = sad;
@@ -810,9 +821,10 @@
 	Data.temp = temp;
 	Data.iFcode = current->fcode;
 	Data.rounding = pParam->m_rounding_type;
+	Data.quarterpel = pParam->m_quarterpel;
 
 	if((qimage = (uint8_t *) malloc(32 * pParam->edged_width)) == NULL)
-		return 0; // allocate some mem for qpel interpolated blocks
+		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!
 
@@ -996,7 +1008,11 @@
 	for(i = 0;  i < 5; i++)
 		Data->currentMV[i].x = Data->currentMV[i].y = 0;
 
-	i = d_mv_bits(Data->predMV.x, Data->predMV.y, Data->iFcode);
+	if(Data->quarterpel)
+		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);
+	
 	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];
@@ -1084,7 +1100,10 @@
 			CheckCandidate = CheckCandidate16_qpel;
 		else
 			CheckCandidate = CheckCandidate16no4v_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); // get real range
+
 		QuarterpelRefine(Data);
 	}
 
@@ -1144,22 +1163,22 @@
 	Data->iMinSAD = OldData->iMinSAD + 1 + block;
 	Data->currentMV = OldData->currentMV + 1 + block;
 	Data->currentQMV = OldData->currentQMV + 1 + block;
+	Data->quarterpel = OldData->quarterpel;
 
-	if (block != 0) {
-		if(pParam->m_quarterpel) {
-			*(Data->iMinSAD) += lambda_vec8[Data->iQuant] *
-									d_mv_bits(	Data->currentQMV->x - Data->predQMV.x, 
-												Data->currentQMV->y - Data->predQMV.y,
-												Data->iFcode);
-		}
-		else {
+	if(Data->quarterpel) // add d_mv_bits[qpel] everywhere but not in 0 (it's already there)
+	{
+		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);
-		}
-	}
+								d_mv_bits(Data->currentQMV->x - Data->predQMV.x, 
+								 		  Data->currentQMV->y - Data->predQMV.y,
+										  Data->iFcode);
 
+	} else // add d_mv_bits[hpel] everywhere but not in 0 (it's already there)
+		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 (MotionFlags & (PMV_EXTSEARCH8|PMV_HALFPELREFINE8)) {
 
@@ -1186,7 +1205,7 @@
 
 			(*MainSearchPtr)(Data->currentMV->x, Data->currentMV->y, Data, 255);
 
-			if(*(Data->iMinSAD) < temp_sad) {
+			if(*(Data->iMinSAD) < temp_sad) { //found a better match?
 					Data->currentQMV->x = 2 * Data->currentMV->x; // update our qpel vector
 					Data->currentQMV->y = 2 * Data->currentMV->y;
 			}
@@ -1203,13 +1222,15 @@
 			}
 		}
 
-		if(pParam->m_quarterpel) {
-			if((!(Data->currentQMV->x & 1)) && (!(Data->currentQMV->y & 1)) &&
-				(MotionFlags & PMV_QUARTERPELREFINE8)) {
-				
-				CheckCandidate = CheckCandidate8_qpel;
-				QuarterpelRefine(Data);
-			}
+		if((Data->quarterpel) && (!(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, 0); // get real range
+
+			QuarterpelRefine(Data);
 		}
 	}
 
@@ -1225,7 +1246,7 @@
 	pMB->mvs[block] = *(Data->currentMV);
 	pMB->qmvs[block] = *(Data->currentQMV);
 
-	pMB->sad8[block] =  4 * (*Data->iMinSAD);
+	pMB->sad8[block] =  4 * (*Data->iMinSAD); // Isibaar: why?
 }
 
 /* B-frames code starts here */
@@ -1426,7 +1447,7 @@
 
 // skip decision
 	if (*Data->iMinSAD - 2 * lambda_vec16[Data->iQuant] < (int32_t)Data->iQuant * SKIP_THRESH_B) {
-		//checking chroma. everything copied from MC
+		//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;
 
@@ -1736,59 +1757,101 @@
 
 /* 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->predMV = get_pmv2(pMBs, pParam->mb_width, 0, x/2 , y/2, block);
+	Data->predQMV = get_qpmv2(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;
+	Data->quarterpel = OldData->quarterpel;
 
-	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 (block != 0) {
+		if(pParam->m_quarterpel) {
+			*(Data->iMinSAD) += lambda_vec8[Data->iQuant] *
+									d_mv_bits(	Data->currentQMV->x - Data->predQMV.x, 
+												Data->currentQMV->y - Data->predQMV.y,
+												Data->iFcode);
+		}
+		else {
+			*(Data->iMinSAD) += lambda_vec8[Data->iQuant] *
+									d_mv_bits(	Data->currentMV->x - Data->predMV.x, 
+												Data->currentMV->y - Data->predMV.y,
+												Data->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));
+	Data->RefQ = OldData->RefQ;
 
-	get_range(&Data.min_dx, &Data.max_dx, &Data.min_dy, &Data.max_dy, x, y, 8,
+	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);
 
-	if (pMB->mode == MODE_INTER4V) {
-		int dummy;
-		CheckCandidate8(pMB->mvs[block].x, pMB->mvs[block].y, 0, &dummy, &Data); }
+	CheckCandidate = CheckCandidate8;
 
+	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(*(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) {
+		temp_sad = *(Data->iMinSAD); // store current MinSAD
 
-	if (MotionFlags & PMV_HALFPELREFINE8) HalfpelRefine(&Data);
+		HalfpelRefine(Data); // perform halfpel refine of current best vector
 
-	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(*(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((Data->quarterpel) && (!(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, 0); // get real range
+
+		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->mvs[block] = *(Data->currentMV);
+	pMB->qmvs[block] = *(Data->currentQMV);
+
+	pMB->sad8[block] =  4 * (*Data->iMinSAD);
 }
 
 
@@ -1797,6 +1860,7 @@
 				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,
@@ -1814,6 +1878,7 @@
 	int i, t;
 	MainSearchFunc * MainSearchPtr;
 
+	Data->predQMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, 0);
 	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);
@@ -1823,6 +1888,8 @@
 	Data->RefH = pRefH + (x + iEdgedWidth*y) * 16;
 	Data->RefV = pRefV + (x + iEdgedWidth*y) * 16;
 	Data->RefHV = pRefHV + (x + iEdgedWidth*y) * 16;
+	Data->RefQ = pRefQ;
+
 	Data->iQuant = iQuant;
 
 	if (!(MotionFlags & PMV_HALFPEL16)) {
@@ -1868,9 +1935,33 @@
 
 	if (MotionFlags & PMV_HALFPELREFINE16) HalfpelRefine(Data);
 
-	if (inter4v)
-		for(i = 0; i < 4; i++)
-			Search8hinted(Data, 2*x+(i&1), 2*y+(i>>1), MotionFlags, pParam, pMB, pMBs, i);
+	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)) {
+		if(inter4v)
+			CheckCandidate = CheckCandidate16_qpel;
+		else
+			CheckCandidate = CheckCandidate16no4v_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); // get real range
+
+		QuarterpelRefine(Data);
+	}
+
+	if (inter4v) {
+		SearchData Data8;
+		Data8.iFcode = Data->iFcode;
+		Data8.iQuant = Data->iQuant;
+		Data8.iEdgedWidth = Data->iEdgedWidth;
+		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) ||
 		(Data->iMinSAD[0] < Data->iMinSAD[1] + Data->iMinSAD[2] + Data->iMinSAD[3] + 
@@ -1908,15 +1999,24 @@
 	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];
+	VECTOR 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!
 
 	if (sadInit) (*sadInit) ();
 
@@ -1940,13 +2040,14 @@
 					pMB->quant = quant;
 				}
 
-			SearchPhinted(pRef->y, pRefH->y, pRefV->y, pRefHV->y, pCurrent, x,
+			SearchPhinted(pRef->y, pRefH->y, pRefV->y, pRefHV->y, qimage, pCurrent, x,
 							y, current->motion_flags, pMB->quant,
 							pParam, pMBs, current->global_flags & XVID_INTER4V, pMB,
 							&Data);
 
 		}
 	}
+	free(qimage);
 }
 
 static __inline int