--- branches/dev-api-4/xvidcore/src/motion/motion_est.c	2003/08/02 15:08:48	1107
+++ branches/dev-api-4/xvidcore/src/motion/motion_est.c	2003/08/28 11:14:04	1133
@@ -21,7 +21,7 @@
  *  along with this program ; if not, write to the Free Software
  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  *
- * $Id: motion_est.c,v 1.58.2.24 2003-08-02 15:08:39 edgomez Exp $
+ * $Id: motion_est.c,v 1.58.2.30 2003-08-28 11:14:04 syskin Exp $
  *
  ****************************************************************************/
 
@@ -70,7 +70,7 @@
 #define MAX_CHROMA_SAD_FOR_SKIP	(22)
 
 #define CHECK_CANDIDATE(X,Y,D) { \
-CheckCandidate((X),(Y), (D), &iDirection, data ); }
+CheckCandidate((X),(Y), data, (D) ); }
 
 
 /*****************************************************************************
@@ -109,49 +109,39 @@
 {
 	int sad;
 	const uint32_t stride = data->iEdgedWidth/2;
-	uint8_t * f_refu = data->RefQ,
-		* f_refv = data->RefQ + 8,
-		* b_refu = data->RefQ + 16,
-		* b_refv = data->RefQ + 24;
+	uint8_t *f_refu, *f_refv, *b_refu, *b_refv;
+
+	const INTERPOLATE8X8_PTR interpolate8x8_halfpel[] = {
+		NULL,
+		interpolate8x8_halfpel_v,
+		interpolate8x8_halfpel_h,
+		interpolate8x8_halfpel_hv
+	};
+
 	int offset = (fx>>1) + (fy>>1)*stride;
+	int filter = ((fx & 1) << 1) | (fy & 1);
 
-	switch (((fx & 1) << 1) | (fy & 1))	{
-		case 0:
-			f_refu = (uint8_t*)data->RefP[4] + offset;
-			f_refv = (uint8_t*)data->RefP[5] + offset;
-			break;
-		case 1:
-			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:
-			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:
-			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;
+	if (filter != 0) {
+		f_refu = data->RefQ;
+		f_refv = data->RefQ + 8;
+		interpolate8x8_halfpel[filter](f_refu, data->RefP[4] + offset, stride, data->rounding);
+		interpolate8x8_halfpel[filter](f_refv, data->RefP[5] + offset, stride, data->rounding);
+	} else {
+		f_refu = (uint8_t*)data->RefP[4] + offset;
+		f_refv = (uint8_t*)data->RefP[5] + offset;
 	}
 
 	offset = (bx>>1) + (by>>1)*stride;
-	switch (((bx & 1) << 1) | (by & 1))	{
-		case 0:
-			b_refu = (uint8_t*)data->b_RefP[4] + offset;
-			b_refv = (uint8_t*)data->b_RefP[5] + offset;
-			break;
-		case 1:
-			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:
-			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:
-			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;
+	filter = ((bx & 1) << 1) | (by & 1);
+
+	if (filter != 0) {
+		b_refu = data->RefQ + 16;
+		b_refv = data->RefQ + 24;
+		interpolate8x8_halfpel[filter](b_refu, data->b_RefP[4] + offset, stride, data->rounding);
+		interpolate8x8_halfpel[filter](b_refv, data->b_RefP[5] + offset, stride, data->rounding);
+	} else {
+		b_refu = (uint8_t*)data->b_RefP[4] + offset;
+		b_refv = (uint8_t*)data->b_RefP[5] + offset;
 	}
 
 	sad = sad8bi(data->CurU, b_refu, f_refu, stride);
@@ -166,6 +156,7 @@
 	int sad;
 	const uint32_t stride = data->iEdgedWidth/2;
 	int offset = (dx>>1) + (dy>>1)*stride;
+	int next = 1;
 
 	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 */
@@ -176,12 +167,10 @@
 			sad += sad8(data->CurV, data->RefP[5] + offset, stride);
 			break;
 		case 1:
-			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;
+			next = stride;
 		case 2:
-			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);
+			sad = sad8bi(data->CurU, data->RefP[4] + offset, data->RefP[4] + offset + next, stride);
+			sad += sad8bi(data->CurV, data->RefP[5] + offset, data->RefP[5] + offset + next, stride);
 			break;
 		default:
 			interpolate8x8_halfpel_hv(data->RefQ, data->RefP[4] + offset, stride, data->rounding);
@@ -301,6 +290,7 @@
 		interpolate8x8_avg2(Reference+8*iEdgedWidth+8, ref1+8*iEdgedWidth+8, ref2+8*iEdgedWidth+8, iEdgedWidth, rounding, 8);
 		break;
 
+
 	default: /* pure halfpel position */
 		return (uint8_t *) ref1;
 	}
@@ -310,7 +300,7 @@
 /* CHECK_CANDIATE FUNCTIONS START */
 
 static void
-CheckCandidate16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
+CheckCandidate16(const int x, const int y, const SearchData * const data, const int Direction)
 {
 	int xc, yc;
 	const uint8_t * Reference;
@@ -330,11 +320,11 @@
 		current = data->currentQMV;
 	}
 
-	sad = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
+	sad = sad16v(data->Cur, Reference, data->iEdgedWidth, data->temp);
 	t = d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
 
 	sad += (data->lambda16 * t * sad)>>10;
-	data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10;
+	data->temp[0] += (data->lambda8 * t * (data->temp[0] + NEIGH_8X8_BIAS))>>10;
 
 	if (data->chroma && sad < data->iMinSAD[0])
 		sad += ChromaSAD((xc >> 1) + roundtab_79[xc & 0x3],
@@ -343,21 +333,21 @@
 	if (sad < data->iMinSAD[0]) {
 		data->iMinSAD[0] = sad;
 		current[0].x = x; current[0].y = y;
-		*dir = Direction;
+		*data->dir = Direction;
 	}
 
-	if (data->temp[1] < data->iMinSAD[1]) {
-		data->iMinSAD[1] = data->temp[1]; current[1].x = x; current[1].y = y; }
-	if (data->temp[2] < data->iMinSAD[2]) {
-		data->iMinSAD[2] = data->temp[2]; current[2].x = x; current[2].y = y; }
-	if (data->temp[3] < data->iMinSAD[3]) {
-		data->iMinSAD[3] = data->temp[3]; current[3].x = x; current[3].y = y; }
-	if (data->temp[4] < data->iMinSAD[4]) {
-		data->iMinSAD[4] = data->temp[4]; current[4].x = x; current[4].y = y; }
+	if (data->temp[0] < data->iMinSAD[1]) {
+		data->iMinSAD[1] = data->temp[0]; current[1].x = x; current[1].y = y; }
+	if (data->temp[1] < data->iMinSAD[2]) {
+		data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; }
+	if (data->temp[2] < data->iMinSAD[3]) {
+		data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; }
+	if (data->temp[3] < data->iMinSAD[4]) {
+		data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; }
 }
 
 static void
-CheckCandidate8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
+CheckCandidate8(const int x, const int y, const SearchData * const data, const int Direction)
 {
 	int32_t sad; uint32_t t;
 	const uint8_t * Reference;
@@ -382,15 +372,16 @@
 	if (sad < *(data->iMinSAD)) {
 		*(data->iMinSAD) = sad;
 		current->x = x; current->y = y;
-		*dir = Direction;
+		*data->dir = Direction;
 	}
 }
 
 static void
-CheckCandidate32(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
+CheckCandidate32(const int x, const int y, const SearchData * const data, const int Direction)
 {
 	uint32_t t;
 	const uint8_t * Reference;
+	int sad;
 
 	if ( (!(x&1) && x !=0) || (!(y&1) && y !=0) || /* non-zero even value */
 		(x > data->max_dx) || (x < data->min_dx)
@@ -399,28 +390,29 @@
 	Reference = GetReference(x, y, data);
 	t = d_mv_bits(x, y, data->predMV, data->iFcode, 0, 1);
 
-	data->temp[0] = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp + 1);
+	sad = sad32v_c(data->Cur, Reference, data->iEdgedWidth, data->temp);
 
-	data->temp[0] += (data->lambda16 * t * data->temp[0]) >> 10;
-	data->temp[1] += (data->lambda8 * t * (data->temp[1] + NEIGH_8X8_BIAS))>>10;
+	sad += (data->lambda16 * t * sad) >> 10;
+	data->temp[0] += (data->lambda8 * t * (data->temp[0] + NEIGH_8X8_BIAS))>>10;
 
-	if (data->temp[0] < data->iMinSAD[0]) {
-		data->iMinSAD[0] = data->temp[0];
+	if (sad < data->iMinSAD[0]) {
+		data->iMinSAD[0] = sad;
 		data->currentMV[0].x = x; data->currentMV[0].y = y;
-		*dir = Direction; }
+		*data->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; }
+	if (data->temp[0] < data->iMinSAD[1]) {
+		data->iMinSAD[1] = data->temp[0]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
+	if (data->temp[1] < data->iMinSAD[2]) {
+		data->iMinSAD[2] = data->temp[1]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
+	if (data->temp[2] < data->iMinSAD[3]) {
+		data->iMinSAD[3] = data->temp[2]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
+	if (data->temp[3] < data->iMinSAD[4]) {
+		data->iMinSAD[4] = data->temp[3]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
 }
 
 static void
-CheckCandidate16no4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
+CheckCandidate16no4v(const int x, const int y, const SearchData * const data, const int Direction)
 {
 	int32_t sad, xc, yc;
 	const uint8_t * Reference;
@@ -454,12 +446,12 @@
 	if (sad < *(data->iMinSAD)) {
 		*(data->iMinSAD) = sad;
 		current->x = x; current->y = y;
-		*dir = Direction;
+		*data->dir = Direction;
 	}
 }
 
 static void
-CheckCandidate16I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
+CheckCandidate16I(const int x, const int y, const SearchData * const data, const int Direction)
 {
 	int sad;
 //	int xc, yc;
@@ -482,12 +474,12 @@
 	if (sad < data->iMinSAD[0]) {
 		data->iMinSAD[0] = sad;
 		data->currentMV[0].x = x; data->currentMV[0].y = y;
-		*dir = Direction;
+		*data->dir = Direction;
 	}
 }
 
 static void
-CheckCandidate32I(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
+CheckCandidate32I(const int x, const int y, const SearchData * const data, const int Direction)
 {
 	/* maximum speed - for P/B/I decision */
 	int32_t sad;
@@ -496,26 +488,26 @@
 		|| (y > data->max_dy) || (y < data->min_dy) ) return;
 
 	sad = sad32v_c(data->Cur, data->RefP[0] + (x>>1) + (y>>1)*((int)data->iEdgedWidth),
-					data->iEdgedWidth, data->temp+1);
+					data->iEdgedWidth, data->temp);
 
 	if (sad < *(data->iMinSAD)) {
 		*(data->iMinSAD) = sad;
 		data->currentMV[0].x = x; data->currentMV[0].y = y;
-		*dir = Direction;
+		*data->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; }
+	if (data->temp[0] < data->iMinSAD[1]) {
+		data->iMinSAD[1] = data->temp[0]; data->currentMV[1].x = x; data->currentMV[1].y = y; }
+	if (data->temp[1] < data->iMinSAD[2]) {
+		data->iMinSAD[2] = data->temp[1]; data->currentMV[2].x = x; data->currentMV[2].y = y; }
+	if (data->temp[2] < data->iMinSAD[3]) {
+		data->iMinSAD[3] = data->temp[2]; data->currentMV[3].x = x; data->currentMV[3].y = y; }
+	if (data->temp[3] < data->iMinSAD[4]) {
+		data->iMinSAD[4] = data->temp[3]; data->currentMV[4].x = x; data->currentMV[4].y = y; }
 
 }
 
 static void
-CheckCandidateInt(const int xf, const int yf, const int Direction, int * const dir, const SearchData * const data)
+CheckCandidateInt(const int xf, const int yf, const SearchData * const data, const int Direction)
 {
 	int32_t sad, xb, yb, xcf, ycf, xcb, ycb;
 	uint32_t t;
@@ -557,12 +549,12 @@
 	if (sad < *(data->iMinSAD)) {
 		*(data->iMinSAD) = sad;
 		current->x = xf; current->y = yf;
-		*dir = Direction;
+		*data->dir = Direction;
 	}
 }
 
 static void
-CheckCandidateDirect(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
+CheckCandidateDirect(const int x, const int y, const SearchData * const data, const int Direction)
 {
 	int32_t sad = 0, xcf = 0, ycf = 0, xcb = 0, ycb = 0;
 	uint32_t k;
@@ -618,12 +610,12 @@
 	if (sad < *(data->iMinSAD)) {
 		*(data->iMinSAD) = sad;
 		data->currentMV->x = x; data->currentMV->y = y;
-		*dir = Direction;
+		*data->dir = Direction;
 	}
 }
 
 static void
-CheckCandidateDirectno4v(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
+CheckCandidateDirectno4v(const int x, const int y, const SearchData * const data, const int Direction)
 {
 	int32_t sad, xcf, ycf, xcb, ycb;
 	const uint8_t *ReferenceF;
@@ -671,17 +663,17 @@
 	if (sad < *(data->iMinSAD)) {
 		*(data->iMinSAD) = sad;
 		data->currentMV->x = x; data->currentMV->y = y;
-		*dir = Direction;
+		*data->dir = Direction;
 	}
 }
 
 
 static void
-CheckCandidateBits16(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
+CheckCandidateRD16(const int x, const int y, const SearchData * const data, const int Direction)
 {
 
 	int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64;
-	int32_t bits = 0;
+	int32_t rd = 0;
 	VECTOR * current;
 	const uint8_t * ptr;
 	int i, cbp = 0, t, xc, yc;
@@ -702,23 +694,23 @@
 	for(i = 0; i < 4; i++) {
 		int s = 8*((i&1) + (i>>1)*data->iEdgedWidth);
 		transfer_8to16subro(in, data->Cur + s, ptr + s, data->iEdgedWidth);
-		bits += data->temp[i] = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, i);
+		rd += data->temp[i] = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, i);
 	}
 
-	bits += t = BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
+	rd += t = BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
 
 	if (data->temp[0] + t < data->iMinSAD[1]) {
-		data->iMinSAD[1] = data->temp[0] + t; current[1].x = x; current[1].y = y; data->cbp[1] = (data->cbp[1]&~32) | cbp&32; }
+		data->iMinSAD[1] = data->temp[0] + t; current[1].x = x; current[1].y = y; data->cbp[1] = (data->cbp[1]&~32) | (cbp&32); }
 	if (data->temp[1] < data->iMinSAD[2]) {
-		data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; data->cbp[1] = (data->cbp[1]&~16) | cbp&16; }
+		data->iMinSAD[2] = data->temp[1]; current[2].x = x; current[2].y = y; data->cbp[1] = (data->cbp[1]&~16) | (cbp&16); }
 	if (data->temp[2] < data->iMinSAD[3]) {
-		data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; data->cbp[1] = (data->cbp[1]&~8) | cbp&8; }
+		data->iMinSAD[3] = data->temp[2]; current[3].x = x; current[3].y = y; data->cbp[1] = (data->cbp[1]&~8) | (cbp&8); }
 	if (data->temp[3] < data->iMinSAD[4]) {
-		data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; data->cbp[1] = (data->cbp[1]&~4) | cbp&4; }
+		data->iMinSAD[4] = data->temp[3]; current[4].x = x; current[4].y = y; data->cbp[1] = (data->cbp[1]&~4) | (cbp&4); }
 
-	bits += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len;
+	rd += BITS_MULT*xvid_cbpy_tab[15-(cbp>>2)].len;
 
-	if (bits >= data->iMinSAD[0]) return;
+	if (rd >= data->iMinSAD[0]) return;
 
 	/* chroma */
 	xc = (xc >> 1) + roundtab_79[xc & 0x3];
@@ -727,30 +719,30 @@
 	/* chroma U */
 	ptr = interpolate8x8_switch2(data->RefQ, data->RefP[4], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding);
 	transfer_8to16subro(in, data->CurU, ptr, data->iEdgedWidth/2);
-	bits += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 4);
-	if (bits >= data->iMinSAD[0]) return;
+	rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 4);
+	if (rd >= data->iMinSAD[0]) return;
 
 	/* chroma V */
 	ptr = interpolate8x8_switch2(data->RefQ, data->RefP[5], 0, 0, xc, yc, data->iEdgedWidth/2, data->rounding);
 	transfer_8to16subro(in, data->CurV, ptr, data->iEdgedWidth/2);
-	bits += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5);
+	rd += Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5);
 
-	bits += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len;
+	rd += BITS_MULT*mcbpc_inter_tab[(MODE_INTER & 7) | ((cbp & 3) << 3)].len;
 
-	if (bits < data->iMinSAD[0]) {
-		data->iMinSAD[0] = bits;
+	if (rd < data->iMinSAD[0]) {
+		data->iMinSAD[0] = rd;
 		current[0].x = x; current[0].y = y;
-		*dir = Direction;
+		*data->dir = Direction;
 		*data->cbp = cbp;
 	}
 }
 
 static void
-CheckCandidateBits8(const int x, const int y, const int Direction, int * const dir, const SearchData * const data)
+CheckCandidateRD8(const int x, const int y, const SearchData * const data, const int Direction)
 {
 
 	int16_t *in = data->dctSpace, *coeff = data->dctSpace + 64;
-	int32_t bits;
+	int32_t rd;
 	VECTOR * current;
 	const uint8_t * ptr;
 	int cbp = 0;
@@ -767,14 +759,14 @@
 	}
 
 	transfer_8to16subro(in, data->Cur, ptr, data->iEdgedWidth);
-	bits = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5);
-	bits += BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
+	rd = Block_CalcBits(coeff, in, data->dctSpace + 128, data->iQuant, data->quant_type, &cbp, 5);
+	rd += BITS_MULT*d_mv_bits(x, y, data->predMV, data->iFcode, data->qpel^data->qpel_precision, 0);
 
-	if (bits < data->iMinSAD[0]) {
+	if (rd < data->iMinSAD[0]) {
 		*data->cbp = cbp;
-		data->iMinSAD[0] = bits;
+		data->iMinSAD[0] = rd;
 		current[0].x = x; current[0].y = y;
-		*dir = Direction;
+		*data->dir = Direction;
 	}
 }
 
@@ -783,15 +775,15 @@
 /* MAINSEARCH FUNCTIONS START */
 
 static void
-AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection)
+AdvDiamondSearch(int x, int y, const SearchData * const data, int bDirection, CheckFunc * const CheckCandidate)
 {
 
 /* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
 
-	int iDirection;
+	unsigned int * const iDirection = data->dir;
 
 	for(;;) { /* forever */
-		iDirection = 0;
+		*iDirection = 0;
 		if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
 		if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
 		if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
@@ -799,9 +791,9 @@
 
 		/* now we're doing diagonal checks near our candidate */
 
-		if (iDirection) {		/* if anything found */
-			bDirection = iDirection;
-			iDirection = 0;
+		if (*iDirection) {		/* if anything found */
+			bDirection = *iDirection;
+			*iDirection = 0;
 			x = data->currentMV->x; y = data->currentMV->y;
 			if (bDirection & 3) {	/* our candidate is left or right */
 				CHECK_CANDIDATE(x, y + iDiamondSize, 8);
@@ -811,8 +803,8 @@
 				CHECK_CANDIDATE(x - iDiamondSize, y, 1);
 			}
 
-			if (iDirection) {
-				bDirection += iDirection;
+			if (*iDirection) {
+				bDirection += *iDirection;
 				x = data->currentMV->x; y = data->currentMV->y;
 			}
 		} else {				/* about to quit, eh? not so fast.... */
@@ -860,20 +852,20 @@
 				CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2 + 8);
 				break;
 			}
-			if (!iDirection) break;		/* ok, the end. really */
-			bDirection = iDirection;
+			if (!*iDirection) break;		/* ok, the end. really */
+			bDirection = *iDirection;
 			x = data->currentMV->x; y = data->currentMV->y;
 		}
 	}
 }
 
 static void
-SquareSearch(int x, int y, const SearchData * const data, int bDirection)
+SquareSearch(int x, int y, const SearchData * const data, int bDirection, CheckFunc * const CheckCandidate)
 {
-	int iDirection;
+	unsigned int * const iDirection = data->dir;
 
 	do {
-		iDirection = 0;
+		*iDirection = 0;
 		if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1+16+64);
 		if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2+32+128);
 		if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4+16+32);
@@ -883,21 +875,21 @@
 		if (bDirection & 64) CHECK_CANDIDATE(x - iDiamondSize, y + iDiamondSize, 1+8+16+64+128);
 		if (bDirection & 128) CHECK_CANDIDATE(x + iDiamondSize, y + iDiamondSize, 2+8+32+64+128);
 
-		bDirection = iDirection;
+		bDirection = *iDirection;
 		x = data->currentMV->x; y = data->currentMV->y;
-	} while (iDirection);
+	} while (*iDirection);
 }
 
 static void
-DiamondSearch(int x, int y, const SearchData * const data, int bDirection)
+DiamondSearch(int x, int y, const SearchData * const data, int bDirection, CheckFunc * const CheckCandidate)
 {
 
 /* directions: 1 - left (x-1); 2 - right (x+1), 4 - up (y-1); 8 - down (y+1) */
 
-	int iDirection;
+	unsigned int * const iDirection = data->dir;
 
 	do {
-		iDirection = 0;
+		*iDirection = 0;
 		if (bDirection & 1) CHECK_CANDIDATE(x - iDiamondSize, y, 1);
 		if (bDirection & 2) CHECK_CANDIDATE(x + iDiamondSize, y, 2);
 		if (bDirection & 4) CHECK_CANDIDATE(x, y - iDiamondSize, 4);
@@ -905,9 +897,9 @@
 
 		/* now we're doing diagonal checks near our candidate */
 
-		if (iDirection) {		/* checking if anything found */
-			bDirection = iDirection;
-			iDirection = 0;
+		if (*iDirection) {		/* checking if anything found */
+			bDirection = *iDirection;
+			*iDirection = 0;
 			x = data->currentMV->x; y = data->currentMV->y;
 			if (bDirection & 3) {	/* our candidate is left or right */
 				CHECK_CANDIDATE(x, y + iDiamondSize, 8);
@@ -916,21 +908,20 @@
 				CHECK_CANDIDATE(x + iDiamondSize, y, 2);
 				CHECK_CANDIDATE(x - iDiamondSize, y, 1);
 			}
-			bDirection += iDirection;
+			bDirection += *iDirection;
 			x = data->currentMV->x; y = data->currentMV->y;
 		}
 	}
-	while (iDirection);
+	while (*iDirection);
 }
 
 /* MAINSEARCH FUNCTIONS END */
 
 static void
-SubpelRefine(const SearchData * const data)
+SubpelRefine(const SearchData * const data, CheckFunc * const CheckCandidate)
 {
 /* Do a half-pel or q-pel refinement */
 	const VECTOR centerMV = data->qpel_precision ? *data->currentQMV : *data->currentMV;
-	int iDirection; /* only needed because macro expects it */
 
 	CHECK_CANDIDATE(centerMV.x, centerMV.y - 1, 0);
 	CHECK_CANDIDATE(centerMV.x + 1, centerMV.y - 1, 0);
@@ -970,15 +961,6 @@
 }
 
 static __inline void
-SkipMacroblockP(MACROBLOCK *pMB, const int32_t sad)
-{
-	pMB->mode = MODE_NOT_CODED;
-	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] = sad;
-}
-
-static __inline void
 ZeroMacroblockP(MACROBLOCK *pMB, const int32_t sad)
 {
 	pMB->mode = MODE_INTER;
@@ -1078,9 +1060,9 @@
 		pMB->cbp = 63;
 		pMB->sad16 = pMB->sad8[0] = pMB->sad8[1] = pMB->sad8[2] = pMB->sad8[3] = sad;
 
-	} else { /* BITS */
+	} else { /* Rate-Distortion */
 
-		int bits, intra, i, cbp, c[2] = {0, 0};
+		int min_rd, intra_rd, i, cbp, c[2] = {0, 0};
 		VECTOR backup[5], *v;
 		Data->iQuant = iQuant;
 		Data->cbp = c;
@@ -1091,34 +1073,34 @@
 			backup[i] = v[i];
 		}
 
-		bits = CountMBBitsInter(Data, pMBs, x, y, pParam, MotionFlags);
+		min_rd = findRDinter(Data, pMBs, x, y, pParam, MotionFlags);
 		cbp = *Data->cbp;
 
 		if (coding_type == S_VOP) {
-			int bits_gmc;
-			*Data->iMinSAD = bits += BITS_MULT*1; /* mcsel */
-			bits_gmc = CountMBBitsGMC(Data, vGMC, x, y);
-			if (bits_gmc < bits) {
+			int gmc_rd;
+			*Data->iMinSAD = min_rd += BITS_MULT*1; /* mcsel */
+			gmc_rd = findRDgmc(Data, vGMC, x, y);
+			if (gmc_rd < min_rd) {
 				mcsel = 1;
-				*Data->iMinSAD = bits = bits_gmc;
+				*Data->iMinSAD = min_rd = gmc_rd;
 				mode = MODE_INTER;
 				cbp = *Data->cbp;
 			}
 		}
 
 		if (inter4v) {
-			int bits_4v;
-			bits_4v = CountMBBitsInter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup);
-			if (bits_4v < bits) {
-				Data->iMinSAD[0] = bits = bits_4v;
+			int v4_rd;
+			v4_rd = findRDinter4v(Data, pMB, pMBs, x, y, pParam, MotionFlags, backup);
+			if (v4_rd < min_rd) {
+				Data->iMinSAD[0] = min_rd = v4_rd;
 				mode = MODE_INTER4V;
 				cbp = *Data->cbp;
 			}
 		}
 
-		intra = CountMBBitsIntra(Data);
-		if (intra < bits) {
-			*Data->iMinSAD = bits = intra;
+		intra_rd = findRDintra(Data);
+		if (intra_rd < min_rd) {
+			*Data->iMinSAD = min_rd = intra_rd;
 			mode = MODE_INTRA;
 		}
 
@@ -1157,7 +1139,7 @@
 	} else
 		if (mode == MODE_INTER4V) ; /* anything here? */
 	else	/* INTRA, NOT_CODED */
-		SkipMacroblockP(pMB, 0);
+		ZeroMacroblockP(pMB, 0);
 
 	pMB->mode = mode;
 }
@@ -1183,13 +1165,13 @@
 
 	uint32_t x, y;
 	uint32_t iIntra = 0;
-	int32_t quant = current->quant, sad00;
+	int32_t sad00;
 	int skip_thresh = INITIAL_SKIP_THRESH * \
 		(current->vop_flags & XVID_VOP_REDUCED ? 4:1) * \
 		(current->vop_flags & XVID_VOP_MODEDECISION_RD ? 2:1);
 
 	/* some pre-initialized thingies for SearchP */
-	int32_t temp[8];
+	int32_t temp[8]; uint32_t dir;
 	VECTOR currentMV[5];
 	VECTOR currentQMV[5];
 	int32_t iMinSAD[5];
@@ -1201,6 +1183,7 @@
 	Data.currentQMV = currentQMV;
 	Data.iMinSAD = iMinSAD;
 	Data.temp = temp;
+	Data.dir = &dir;
 	Data.iFcode = current->fcode;
 	Data.rounding = pParam->m_rounding_type;
 	Data.qpel = (current->vol_flags & XVID_VOL_QUARTERPEL ? 1:0);
@@ -1242,19 +1225,13 @@
 
 			sad00 = pMB->sad16;
 
-			if (pMB->dquant != 0) {
-				quant += DQtab[pMB->dquant];
-				if (quant > 31) quant = 31;
-				else if (quant < 1) quant = 1;
-			}
-			pMB->quant = quant;
-
 			/* initial skip decision */
 			/* no early skip for GMC (global vector = skip vector is unknown!)  */
 			if (current->coding_type != S_VOP)	{ /* 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);
+						ZeroMacroblockP(pMB, sad00);
+						pMB->mode = MODE_NOT_CODED;
 						continue;
 					}
 			}
@@ -1278,27 +1255,33 @@
 		}
 	}
 
-//	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;
 }
 
+/* check if given vector is equal to any vector checked before */
+static __inline int
+vector_repeats(const VECTOR * const pmv, const int i)
+{
+	unsigned int j;
+	for (j = 0; j < i; j++)
+		if (MVequal(pmv[i], pmv[j])) return 1; /* same vector has been checked already */
+	return 0;
+}
 
+/*	make a binary mask that prevents diamonds/squares 
+	from checking a vector which has been checked as a prediction */
 static __inline int
-make_mask(const VECTOR * const pmv, const int i)
+make_mask(const VECTOR * const pmv, const int i, const int current)
 {
-	int mask = 255, j;
+	unsigned int mask = 255, j;
 	for (j = 0; j < i; j++) {
-		if (MVequal(pmv[i], pmv[j])) return 0; /* same vector has been checked already */
-		if (pmv[i].x == pmv[j].x) {
-			if (pmv[i].y == pmv[j].y + iDiamondSize) mask &= ~4;
-			else if (pmv[i].y == pmv[j].y - iDiamondSize) mask &= ~8;
+		if (pmv[current].x == pmv[j].x) {
+			if (pmv[current].y == pmv[j].y + iDiamondSize) mask &= ~4;
+			else if (pmv[current].y == pmv[j].y - iDiamondSize) mask &= ~8;
 		} else
-			if (pmv[i].y == pmv[j].y) {
-				if (pmv[i].x == pmv[j].x + iDiamondSize) mask &= ~1;
-				else if (pmv[i].x == pmv[j].x - iDiamondSize) mask &= ~2;
+			if (pmv[current].y == pmv[j].y) {
+				if (pmv[current].x == pmv[j].x + iDiamondSize) mask &= ~1;
+				else if (pmv[current].x == pmv[j].x - iDiamondSize) mask &= ~2;
 			}
 	}
 	return mask;
@@ -1362,12 +1345,13 @@
 		MACROBLOCK * const pMB)
 {
 
-	int i, iDirection = 255, mask, threshA;
+	int i, threshA;
 	VECTOR pmv[7];
 	int inter4v = (VopFlags & XVID_VOP_INTER4V) && (pMB->dquant == 0);
+	CheckFunc * CheckCandidate;
 
-	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);
+	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
+						pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv);
 
 	get_pmvdata2(pMBs, pParam->mb_width, 0, x, y, 0, pmv, Data->temp);
 
@@ -1387,6 +1371,7 @@
 	Data->lambda16 = lambda_vec16[pMB->quant];
 	Data->lambda8 = lambda_vec8[pMB->quant];
 	Data->qpel_precision = 0;
+	*Data->dir = 0;
 
 	memset(Data->currentMV, 0, 5*sizeof(VECTOR));
 
@@ -1411,17 +1396,17 @@
 					prevMBs + x + y * pParam->mb_width, Data->rrv);
 
 	if (!Data->rrv) {
-		if (inter4v | Data->chroma) CheckCandidate = CheckCandidate16;
+		if (inter4v) CheckCandidate = CheckCandidate16;
 			else CheckCandidate = CheckCandidate16no4v; /* for extra speed */
 	} else CheckCandidate = CheckCandidate32;
 
 /* main loop. checking all predictions (but first, which is 0,0 and has been checked in MotionEstimation())*/
 
-	for (i = 1; i < 7; i++) {
-		if (!(mask = make_mask(pmv, i)) ) continue;
-		CheckCandidate(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
-		if (Data->iMinSAD[0] <= threshA) break;
-	}
+	for (i = 1; i < 7; i++)
+		if (!vector_repeats(pmv, i)) {
+			CheckCandidate(pmv[i].x, pmv[i].y, Data, i);
+			if (Data->iMinSAD[0] <= threshA) { i++; break; }
+		}
 
 	if ((Data->iMinSAD[0] <= threshA) ||
 			(MVequal(Data->currentMV[0], (prevMBs+x+y*pParam->mb_width)->mvs[0]) &&
@@ -1430,11 +1415,13 @@
 	else {
 
 		MainSearchFunc * MainSearchPtr;
+		int mask = make_mask(pmv, i, *Data->dir); // all vectors pmv[0..i-1] have been checked
+
 		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);
+		MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, mask, CheckCandidate);
 
 /* extended search, diamond starting in 0,0 and in prediction.
 	note that this search is/might be done in halfpel positions,
@@ -1450,8 +1437,8 @@
 			if (!(MVequal(startMV, backupMV))) {
 				bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
 
-				CheckCandidate(startMV.x, startMV.y, 255, &iDirection, Data);
-				MainSearchPtr(startMV.x, startMV.y, Data, 255);
+				CheckCandidate(startMV.x, startMV.y, Data, 255);
+				MainSearchPtr(startMV.x, startMV.y, Data, 255, CheckCandidate);
 				if (bSAD < Data->iMinSAD[0]) {
 					Data->currentMV[0] = backupMV;
 					Data->iMinSAD[0] = bSAD; }
@@ -1462,17 +1449,18 @@
 			if (!(MVequal(startMV, backupMV))) {
 				bSAD = Data->iMinSAD[0]; Data->iMinSAD[0] = MV_MAX_ERROR;
 
-				CheckCandidate(startMV.x, startMV.y, 255, &iDirection, Data);
-				MainSearchPtr(startMV.x, startMV.y, Data, 255);
+				CheckCandidate(startMV.x, startMV.y, Data, 255);
+				MainSearchPtr(startMV.x, startMV.y, Data, 255, CheckCandidate);
 				if (bSAD < Data->iMinSAD[0]) {
 					Data->currentMV[0] = backupMV;
-					Data->iMinSAD[0] = bSAD; }
+					Data->iMinSAD[0] = bSAD; 
+				}
 			}
 		}
 	}
 
 	if (MotionFlags & XVID_ME_HALFPELREFINE16)
-			SubpelRefine(Data);
+			SubpelRefine(Data, CheckCandidate);
 
 	for(i = 0; i < 5; i++) {
 		Data->currentQMV[i].x = 2 * Data->currentMV[i].x; /* initialize qpel vectors */
@@ -1480,11 +1468,11 @@
 	}
 
 	if (Data->qpel) {
-		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);
+		get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
+				pParam->width, pParam->height, Data->iFcode, 2, 0);
 		Data->qpel_precision = 1;
 		if (MotionFlags & XVID_ME_QUARTERPELREFINE16)
-			SubpelRefine(Data);
+			SubpelRefine(Data, CheckCandidate);
 	}
 
 	if (Data->iMinSAD[0] < (int32_t)pMB->quant * 30)
@@ -1531,6 +1519,7 @@
 		SearchData * const Data)
 {
 	int i = 0;
+	CheckFunc * CheckCandidate;
 	Data->iMinSAD = OldData->iMinSAD + 1 + block;
 	Data->currentMV = OldData->currentMV + 1 + block;
 	Data->currentQMV = OldData->currentQMV + 1 + block;
@@ -1559,8 +1548,8 @@
 		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,
-					pParam->width, pParam->height, Data->iFcode - Data->qpel, 0, Data->rrv);
+		get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3,
+					pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv);
 
 		if (!Data->rrv) CheckCandidate = CheckCandidate8;
 		else CheckCandidate = CheckCandidate16no4v;
@@ -1573,7 +1562,7 @@
 				else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND8) MainSearchPtr = AdvDiamondSearch;
 					else MainSearchPtr = DiamondSearch;
 
-			MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, 255);
+			MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, 255, CheckCandidate);
 
 			if(*(Data->iMinSAD) < temp_sad) {
 					Data->currentQMV->x = 2 * Data->currentMV->x; /* update our qpel vector */
@@ -1584,7 +1573,7 @@
 		if (MotionFlags & XVID_ME_HALFPELREFINE8) {
 			int32_t temp_sad = *(Data->iMinSAD); /* store current MinSAD */
 
-			SubpelRefine(Data); /* perform halfpel refine of current best vector */
+			SubpelRefine(Data, CheckCandidate); /* 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 */
@@ -1594,9 +1583,9 @@
 
 		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);
-				SubpelRefine(Data);
+				get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 3,
+					pParam->width, pParam->height, Data->iFcode, 2, 0);
+				SubpelRefine(Data, CheckCandidate);
 		}
 	}
 
@@ -1682,7 +1671,7 @@
 			SearchData * const Data)
 {
 
-	int i, iDirection = 255, mask;
+	int i, mask;
 	VECTOR pmv[7];
 	MainSearchFunc *MainSearchPtr;
 	*Data->iMinSAD = MV_MAX_ERROR;
@@ -1699,8 +1688,8 @@
 
 	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 - Data->qpel, 0, 0);
+	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
+				pParam->width, pParam->height, iFcode - Data->qpel, 1, 0);
 
 	pmv[0] = Data->predMV;
 	if (Data->qpel) { pmv[0].x /= 2; pmv[0].y /= 2; }
@@ -1708,29 +1697,29 @@
 	PreparePredictionsBF(pmv, x, y, pParam->mb_width, pMB, mode_current);
 
 	Data->currentMV->x = Data->currentMV->y = 0;
-	CheckCandidate = CheckCandidate16no4v;
 
 	/* main loop. checking all predictions */
-	for (i = 0; i < 7; i++) {
-		if (!(mask = make_mask(pmv, i)) ) continue;
-		CheckCandidate16no4v(pmv[i].x, pmv[i].y, mask, &iDirection, Data);
-	}
+	for (i = 0; i < 7; i++)
+		if (!vector_repeats(pmv, i) )
+			CheckCandidate16no4v(pmv[i].x, pmv[i].y, Data, i);
 
 	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);
+	mask = make_mask(pmv, 7, *Data->dir);
 
-	SubpelRefine(Data);
+	MainSearchPtr(Data->currentMV->x, Data->currentMV->y, Data, mask, CheckCandidate16no4v);
+
+	SubpelRefine(Data, CheckCandidate16no4v);
 
 	if (Data->qpel && *Data->iMinSAD < *best_sad + 300) {
 		Data->currentQMV->x = 2*Data->currentMV->x;
 		Data->currentQMV->y = 2*Data->currentMV->y;
 		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, iFcode, 1, 0);
-		SubpelRefine(Data);
+		get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
+					pParam->width, pParam->height, iFcode, 2, 0);
+		SubpelRefine(Data, CheckCandidate16no4v);
 	}
 
 	/* three bits are needed to code backward mode. four for forward */
@@ -1792,14 +1781,14 @@
 					b_Ref->u + (y*8 + b_dy/2) * stride + x*8 + b_dx/2,
 					stride);
 
-	if (sum >= 2 * MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) return; /* no skip */
+	if (sum >= MAX_CHROMA_SAD_FOR_SKIP * pMB->quant) return; /* no skip */
 
 	sum += sad8bi(pCur->v + 8*x + 8 * y * stride,
 					f_Ref->v + (y*8 + dy/2) * stride + x*8 + dx/2,
 					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) {
+	if (sum < 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];
@@ -1831,6 +1820,7 @@
 	int32_t skip_sad;
 	int k = (x + Data->iEdgedWidth*y) * 16;
 	MainSearchFunc *MainSearchPtr;
+	CheckFunc * CheckCandidate;
 
 	*Data->iMinSAD = 256*4096;
 	Data->RefP[0] = f_Ref->y + k;
@@ -1880,7 +1870,7 @@
 
 	CheckCandidate = b_mb->mode == MODE_INTER4V ? CheckCandidateDirect : CheckCandidateDirectno4v;
 
-	CheckCandidate(0, 0, 255, &k, Data);
+	CheckCandidate(0, 0, Data, 255);
 
 	/* initial (fast) skip decision */
 	if (*Data->iMinSAD < pMB->quant * INITIAL_SKIP_THRESH * (Data->chroma?3:2)) {
@@ -1906,9 +1896,9 @@
 		else if (MotionFlags & XVID_ME_ADVANCEDDIAMOND16) MainSearchPtr = AdvDiamondSearch;
 			else MainSearchPtr = DiamondSearch;
 
-	MainSearchPtr(0, 0, Data, 255);
+	MainSearchPtr(0, 0, Data, 255, CheckCandidate);
 
-	SubpelRefine(Data);
+	SubpelRefine(Data, CheckCandidate);
 
 	*best_sad = *Data->iMinSAD;
 
@@ -1967,7 +1957,7 @@
 
 {
 
-	int iDirection, i, j;
+	int i, j;
 	SearchData bData;
 
 	fData->qpel_precision = 0;
@@ -1990,13 +1980,14 @@
 	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.dir = fData->dir;
 
 	bData.bpredMV = fData->predMV = *f_predMV;
 	fData->bpredMV = bData.predMV = *b_predMV;
 	fData->currentMV[0] = fData->currentMV[2];
 
-	get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode - fData->qpel, 0, 0);
-	get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode - fData->qpel, 0, 0);
+	get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 4, pParam->width, pParam->height, fcode - fData->qpel, 1, 0);
+	get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 4, pParam->width, pParam->height, bcode - fData->qpel, 1, 0);
 
 	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_dx;
@@ -2008,44 +1999,43 @@
 	if (fData->currentMV[1].y > bData.max_dy) fData->currentMV[1].y = bData.max_dy;
 	if (fData->currentMV[1].y < bData.min_dy) fData->currentMV[1].y = bData.min_dy;
 
-	CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, 255, &iDirection, fData);
+	CheckCandidateInt(fData->currentMV[0].x, fData->currentMV[0].y, fData, 255);
 
 	/* diamond */
 	do {
-		iDirection = 255;
+		*fData->dir = 255;
 		/* forward MV moves */
 		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, fData, 0);
+		CheckCandidateInt(i, j + 1, fData, 0);
+		CheckCandidateInt(i - 1, j, fData, 0);
+		CheckCandidateInt(i, j - 1, fData, 0);
 
 		/* backward MV moves */
 		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);
-		CheckCandidateInt(i - 1, j, 0, &iDirection, &bData);
-		CheckCandidateInt(i, j - 1, 0, &iDirection, &bData);
+		CheckCandidateInt(i + 1, j, &bData, 0);
+		CheckCandidateInt(i, j + 1, &bData, 0);
+		CheckCandidateInt(i - 1, j, &bData, 0);
+		CheckCandidateInt(i, j - 1, &bData, 0);
 
-	} while (!(iDirection));
+	} while (!(*fData->dir));
 
 	/* qpel refinement */
 	if (fData->qpel) {
 		if (*fData->iMinSAD > *best_sad + 500) return;
-		CheckCandidate = CheckCandidateInt;
 		fData->qpel_precision = bData.qpel_precision = 1;
-		get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 16, pParam->width, pParam->height, fcode, 1, 0);
-		get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 16, pParam->width, pParam->height, bcode, 1, 0);
+		get_range(&fData->min_dx, &fData->max_dx, &fData->min_dy, &fData->max_dy, x, y, 4, pParam->width, pParam->height, fcode, 2, 0);
+		get_range(&bData.min_dx, &bData.max_dx, &bData.min_dy, &bData.max_dy, x, y, 4, pParam->width, pParam->height, bcode, 2, 0);
 		fData->currentQMV[2].x = fData->currentQMV[0].x = 2 * fData->currentMV[0].x;
 		fData->currentQMV[2].y = fData->currentQMV[0].y = 2 * fData->currentMV[0].y;
 		fData->currentQMV[1].x = 2 * fData->currentMV[1].x;
 		fData->currentQMV[1].y = 2 * fData->currentMV[1].y;
-		SubpelRefine(fData);
+		SubpelRefine(fData, CheckCandidateInt);
 		if (*fData->iMinSAD > *best_sad + 300) return;
 		fData->currentQMV[2] = fData->currentQMV[0];
-		SubpelRefine(&bData);
+		SubpelRefine(&bData, CheckCandidateInt);
 	}
 
 	*fData->iMinSAD += (2+3) * fData->lambda16; /* two bits are needed to code interpolate mode. */
@@ -2104,6 +2094,7 @@
 
 	SearchData Data;
 	int32_t iMinSAD;
+	uint32_t dir;
 	VECTOR currentMV[3];
 	VECTOR currentQMV[3];
 	int32_t temp[8];
@@ -2112,10 +2103,11 @@
 	Data.currentMV = currentMV; Data.currentQMV = currentQMV;
 	Data.iMinSAD = &iMinSAD;
 	Data.lambda16 = lambda_vec16[frame->quant];
-	Data.qpel = pParam->vol_flags & XVID_VOL_QUARTERPEL;
+	Data.qpel = pParam->vol_flags & XVID_VOL_QUARTERPEL ? 1 : 0;
 	Data.rounding = 0;
 	Data.chroma = frame->motion_flags & XVID_ME_CHROMA_BVOP;
 	Data.temp = temp;
+	Data.dir = &dir;
 
 	Data.RefQ = f_refV->u; /* a good place, also used in MC (for similar purpose) */
 
@@ -2222,8 +2214,7 @@
 				SearchData * const Data)
 {
 
-	int i, mask;
-	int quarterpel = (pParam->vol_flags & XVID_VOL_QUARTERPEL)? 1: 0;
+	int i;
 	VECTOR pmv[3];
 	MACROBLOCK * const pMB = &pMBs[x + y * pParam->mb_width];
 
@@ -2238,8 +2229,8 @@
 			Data->predMV = (pMB - 1)->mvs[0]; /* left instead of median */
 			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 - quarterpel, 0, 0);
+	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
+			pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, 0);
 
 	Data->Cur = pCur + (x + y * pParam->edged_width) * 16;
 	Data->RefP[0] = pRef + (x + y * pParam->edged_width) * 16;
@@ -2250,17 +2241,19 @@
 	pmv[2].y = EVEN(Data->predMV.y);
 	pmv[0].x = pmv[0].y = 0;
 
-	CheckCandidate32I(0, 0, 255, &i, Data);
+	CheckCandidate32I(0, 0, Data, 0);
 
 	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) /* diamond only if needed */
-			DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, i);
+		if (!vector_repeats(pmv, 1))
+			CheckCandidate32I(pmv[1].x, pmv[1].y, Data, 1);
+		if (!vector_repeats(pmv, 2))
+			CheckCandidate32I(pmv[2].x, pmv[2].y, Data, 2);
+
+		if (*Data->iMinSAD > 4 * MAX_SAD00_FOR_SKIP) { /* diamond only if needed */
+			unsigned int mask = make_mask(pmv, 3, *Data->dir);
+			DiamondSearch(Data->currentMV->x, Data->currentMV->y, Data, 255/*mask*/, CheckCandidate32I);
+		}
 	}
 
 	for (i = 0; i < 4; i++) {
@@ -2293,6 +2286,7 @@
 	int complexity = 0;
 
 	int32_t iMinSAD[5], temp[5];
+	uint32_t dir;
 	VECTOR currentMV[5];
 	SearchData Data;
 	Data.iEdgedWidth = pParam->edged_width;
@@ -2300,7 +2294,8 @@
 	Data.iMinSAD = iMinSAD;
 	Data.iFcode = Current->fcode;
 	Data.temp = temp;
-	CheckCandidate = CheckCandidate32I;
+	Data.dir = &dir;
+	Data.qpel = (pParam->vol_flags & XVID_VOL_QUARTERPEL)? 1: 0;
 
 	if (intraCount != 0) {
 		if (intraCount < 10) // we're right after an I frame
@@ -2336,7 +2331,7 @@
 				dev = dev16(pCurrent->y + (x + (i&1) + (y + (i>>1)) * pParam->edged_width) * 16,
 								pParam->edged_width);
 
-				complexity += dev;
+				complexity += MAX(dev, 300);
 				if (dev + IntraThresh < pMB->sad16) {
 					pMB->mode = MODE_INTRA;
 					if (++intra > ((pParam->mb_height-2)*(pParam->mb_width-2))/2) return I_VOP;
@@ -2346,7 +2341,7 @@
 					if (dev > 500 && pMB->sad16 < 1000)
 						sSAD += 1000;
 
-				sSAD += pMB->sad16;
+				sSAD += (dev < 3000) ? pMB->sad16 : pMB->sad16/2; /* blocks with big contrast differences usually have large SAD - while they look very good in b-frames */
 			}
 		}
 	}
@@ -2354,7 +2349,7 @@
 
 	sSAD /= complexity + 4*blocks;
 
-	if (intraCount > 12 && sSAD > INTRA_THRESH2 ) return I_VOP;
+	if (intraCount > 80 && sSAD > INTRA_THRESH2 ) return I_VOP;
 	if (sSAD > InterThresh ) return P_VOP;
 	emms();
 	return B_VOP;
@@ -2364,41 +2359,41 @@
 /* functions which perform BITS-based search/bitcount */
 
 static int
-CountMBBitsInter(SearchData * const Data,
-				const MACROBLOCK * const pMBs, const int x, const int y,
-				const MBParam * const pParam,
-				const uint32_t MotionFlags)
+findRDinter(SearchData * const Data,
+			const MACROBLOCK * const pMBs, const int x, const int y,
+			const MBParam * const pParam,
+			const uint32_t MotionFlags)
 {
-	int i, iDirection;
+	int i;
 	int32_t bsad[5];
 
-	CheckCandidate = CheckCandidateBits16;
-
 	if (Data->qpel) {
 		for(i = 0; i < 5; i++) {
 			Data->currentMV[i].x = Data->currentQMV[i].x/2;
 			Data->currentMV[i].y = Data->currentQMV[i].y/2;
 		}
 		Data->qpel_precision = 1;
-		CheckCandidateBits16(Data->currentQMV[0].x, Data->currentQMV[0].y, 255, &iDirection, Data);
+		CheckCandidateRD16(Data->currentQMV[0].x, Data->currentQMV[0].y, Data, 255);
 
 		if (MotionFlags & (XVID_ME_HALFPELREFINE16_RD | XVID_ME_EXTSEARCH_RD)) { /* 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);
+			get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
+						pParam->width, pParam->height, Data->iFcode - Data->qpel, 1, Data->rrv);
 			Data->qpel_precision = 0;
 			if (Data->currentQMV->x & 1 || Data->currentQMV->y & 1)
-				CheckCandidateBits16(Data->currentMV[0].x, Data->currentMV[0].y, 255, &iDirection, Data);
+				CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255);
 		}
 
 	} else { /* not qpel */
 
-		CheckCandidateBits16(Data->currentMV[0].x, Data->currentMV[0].y, 255, &iDirection, Data);
+		CheckCandidateRD16(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255);
 	}
 
-	if (MotionFlags&XVID_ME_EXTSEARCH_RD) SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, iDirection);
+	if (MotionFlags&XVID_ME_EXTSEARCH_RD) 
+		SquareSearch(Data->currentMV->x, Data->currentMV->y, Data, 255, CheckCandidateRD16);
 
-	if (MotionFlags&XVID_ME_HALFPELREFINE16_RD) SubpelRefine(Data);
+	if (MotionFlags&XVID_ME_HALFPELREFINE16_RD) 
+		SubpelRefine(Data, CheckCandidateRD16);
 
 	if (Data->qpel) {
 		if (MotionFlags&(XVID_ME_EXTSEARCH_RD | XVID_ME_HALFPELREFINE16_RD)) { /* there was halfpel-precision search */
@@ -2409,37 +2404,37 @@
 
 			/* preparing for qpel-precision search */
 			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);
+			get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
+					pParam->width, pParam->height, Data->iFcode, 2, 0);
 		}
-		if (MotionFlags&XVID_ME_QUARTERPELREFINE16_RD) SubpelRefine(Data);
+		if (MotionFlags&XVID_ME_QUARTERPELREFINE16_RD) 
+			SubpelRefine(Data, CheckCandidateRD16);
 	}
 
 	if (MotionFlags&XVID_ME_CHECKPREDICTION_RD) { /* 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);
+			CheckCandidateRD16(Data->predMV.x, Data->predMV.y, Data, 255);
 	}
 	return Data->iMinSAD[0];
 }
 
 static int
-CountMBBitsInter4v(const SearchData * const Data,
-					MACROBLOCK * const pMB, const MACROBLOCK * const pMBs,
-					const int x, const int y,
-					const MBParam * const pParam, const uint32_t MotionFlags,
-					const VECTOR * const backup)
+findRDinter4v(const SearchData * const Data,
+				MACROBLOCK * const pMB, const MACROBLOCK * const pMBs,
+				const int x, const int y,
+				const MBParam * const pParam, const uint32_t MotionFlags,
+				const VECTOR * const backup)
 {
 
-	int cbp = 0, bits = 0, t = 0, i, iDirection;
+	int cbp = 0, bits = 0, t = 0, i;
 	SearchData Data2, *Data8 = &Data2;
 	int sumx = 0, sumy = 0;
 	int16_t *in = Data->dctSpace, *coeff = Data->dctSpace + 64;
 	uint8_t * ptr;
 
 	memcpy(Data8, Data, sizeof(SearchData));
-	CheckCandidate = CheckCandidateBits8;
-
+	
 	for (i = 0; i < 4; i++) { /* for all luma blocks */
 
 		Data8->iMinSAD = Data->iMinSAD + i + 1;
@@ -2451,7 +2446,6 @@
 		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);
 		*Data8->cbp = (Data->cbp[1] & (1<<(5-i))) ? 1:0; // copy corresponding cbp bit
-//		*Data8->cbp = 1;
 
 		if(Data->qpel) {
 			Data8->predMV = get_qpmv2(pMBs, pParam->mb_width, 0, x, y, i);
@@ -2463,8 +2457,8 @@
 										Data8->predMV, Data8->iFcode, 0, 0);
 		}
 
-		get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 8,
-					pParam->width, pParam->height, Data8->iFcode, Data8->qpel, 0);
+		get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3,
+					pParam->width, pParam->height, Data8->iFcode, Data8->qpel+1, 0);
 
 		*Data8->iMinSAD += BITS_MULT*t;
 
@@ -2473,7 +2467,7 @@
 		{
 			VECTOR *v = Data8->qpel ? Data8->currentQMV : Data8->currentMV;
 			if (!MVequal (*v, backup[i+1]) )
-				CheckCandidateBits8(backup[i+1].x, backup[i+1].y, 255, &iDirection, Data8);
+				CheckCandidateRD8(backup[i+1].x, backup[i+1].y, Data8, 255);
 		}
 
 		if (Data8->qpel) {
@@ -2482,17 +2476,17 @@
 				Data8->currentMV->x = Data8->currentQMV->x/2;
 				Data8->currentMV->y = Data8->currentQMV->y/2;
 				Data8->qpel_precision = 0;
-				get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 8,
-							pParam->width, pParam->height, Data8->iFcode - 1, 0, 0);
+				get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3,
+							pParam->width, pParam->height, Data8->iFcode - 1, 1, 0);
 
 				if (Data8->currentQMV->x & 1 || Data8->currentQMV->y & 1)
-					CheckCandidateBits8(Data8->currentMV->x, Data8->currentMV->y, 255, &iDirection, Data8);
+					CheckCandidateRD8(Data8->currentMV->x, Data8->currentMV->y, Data8, 255);
 
 				if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD)
-					SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255);
+					SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255, CheckCandidateRD8);
 
 				if (MotionFlags & XVID_ME_HALFPELREFINE8_RD)
-					SubpelRefine(Data8);
+					SubpelRefine(Data8, CheckCandidateRD8);
 
 				if(s > *Data8->iMinSAD) { /* we have found a better match */
 					Data8->currentQMV->x = 2*Data8->currentMV->x;
@@ -2500,26 +2494,27 @@
 				}
 
 				Data8->qpel_precision = 1;
-				get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 8,
-							pParam->width, pParam->height, Data8->iFcode, 1, 0);
+				get_range(&Data8->min_dx, &Data8->max_dx, &Data8->min_dy, &Data8->max_dy, 2*x + (i&1), 2*y + (i>>1), 3,
+							pParam->width, pParam->height, Data8->iFcode, 2, 0);
 
 			}
-			if (MotionFlags & XVID_ME_QUARTERPELREFINE8_RD) SubpelRefine(Data8);
+			if (MotionFlags & XVID_ME_QUARTERPELREFINE8_RD) 
+				SubpelRefine(Data8, CheckCandidateRD8);
 
 		} else { /* not qpel */
 
 			if (MotionFlags & XVID_ME_EXTSEARCH8 && MotionFlags & XVID_ME_EXTSEARCH_RD) /* extsearch */
-				SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255);
+				SquareSearch(Data8->currentMV->x, Data8->currentMV->x, Data8, 255, CheckCandidateRD8);
 
 			if (MotionFlags & XVID_ME_HALFPELREFINE8_RD)
-				SubpelRefine(Data8); /* halfpel refinement */
+				SubpelRefine(Data8, CheckCandidateRD8); /* halfpel refinement */
 		}
 
 		/* checking vector equal to predicion */
 		if (i != 0 && MotionFlags & XVID_ME_CHECKPREDICTION_RD) {
 			const VECTOR * v = Data->qpel ? Data8->currentQMV : Data8->currentMV;
 			if (!MVequal(*v, Data8->predMV))
-				CheckCandidateBits8(Data8->predMV.x, Data8->predMV.y, 255, &iDirection, Data8);
+				CheckCandidateRD8(Data8->predMV.x, Data8->predMV.y, Data8, 255);
 		}
 
 		bits += *Data8->iMinSAD;
@@ -2569,7 +2564,7 @@
 }
 
 static int
-CountMBBitsIntra(const SearchData * const Data)
+findRDintra(const SearchData * const Data)
 {
 	int bits = BITS_MULT*1; /* this one is ac/dc prediction flag bit */
 	int cbp = 0, i, dc = 0;
@@ -2601,7 +2596,7 @@
 }
 
 static int
-CountMBBitsGMC(const SearchData * const Data, const IMAGE * const vGMC, const int x, const int y)
+findRDgmc(const SearchData * const Data, const IMAGE * const vGMC, const int x, const int y)
 {
 	int bits = BITS_MULT*1; /* this one is mcsel */
 	int cbp = 0, i;
@@ -2656,8 +2651,8 @@
 
 	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, 16, 0, 0);
+	get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
+				pParam->width, pParam->height, 16, 1, 0);
 
 	Data->Cur = pCur + 16*(x + y * pParam->edged_width);
 	Data->RefP[0] = pRef + 16*(x + y * pParam->edged_width);
@@ -2666,14 +2661,14 @@
 	Data->RefP[3] = pRefHV + 16*(x + y * pParam->edged_width);
 
 	Data->currentMV[0].x = Data->currentMV[0].y = 0;
-	CheckCandidate16I(0, 0, 255, &i, Data);
+	CheckCandidate16I(0, 0, Data, 255);
 
 	if ( (Data->predMV.x !=0) || (Data->predMV.y != 0) )
-		CheckCandidate16I(Data->predMV.x, Data->predMV.y, 255, &i, Data);
+		CheckCandidate16I(Data->predMV.x, Data->predMV.y, Data, 255);
 
-	AdvDiamondSearch(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255);
+	DiamondSearch(Data->currentMV[0].x, Data->currentMV[0].y, Data, 255, CheckCandidate16I);
 
-	SubpelRefine(Data);
+	SubpelRefine(Data, CheckCandidate16I);
 
 
 	/* for QPel halfpel positions are worse than in halfpel mode :( */
@@ -2681,8 +2676,8 @@
 		Data->currentQMV->x = 2*Data->currentMV->x;
 		Data->currentQMV->y = 2*Data->currentMV->y;
 		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, iFcode, 1, 0);
+		get_range(&Data->min_dx, &Data->max_dx, &Data->min_dy, &Data->max_dy, x, y, 4,
+					pParam->width, pParam->height, iFcode, 2, 0);
 		SubpelRefine(Data);
 	}
 */
@@ -2709,6 +2704,7 @@
 
 	int32_t iMinSAD[5], temp[5];
 	VECTOR currentMV[5];
+	uint32_t dir;
 	SearchData Data;
 	memset(&Data, 0, sizeof(SearchData));
 
@@ -2719,14 +2715,12 @@
 	Data.iMinSAD = &iMinSAD[0];
 	Data.iFcode = current->fcode;
 	Data.temp = temp;
-
-	CheckCandidate = CheckCandidate16I;
+	Data.dir = &dir;
 
 	if (sadInit) (*sadInit) ();
 
 	for (y = 0; y < pParam->mb_height; y ++) {
 		for (x = 0; x < pParam->mb_width; x ++) {
-
 			GMEanalyzeMB(pCurrent->y, pReference->y, pRefH->y, pRefV->y, pRefHV->y, x, y, pParam, pMBs, &Data);
 		}
 	}