--- motion_comp.c	2002/03/08 02:44:50	1.1
+++ motion_comp.c	2003/01/03 12:43:38	1.11.2.18
@@ -1,147 +1,438 @@
+// 30.10.2002	corrected qpel chroma rounding
+// 04.10.2002	added qpel support to MBMotionCompensation
+// 01.05.2002   updated MBMotionCompensationBVOP
+// 14.04.2002   bframe compensation
+
 #include "../encoder.h"
 #include "../utils/mbfunctions.h"
 #include "../image/interpolate8x8.h"
+#include "../image/reduced.h"
 #include "../utils/timer.h"
+#include "motion.h"
+
+static __inline void
+compensate16x16_interpolate(int16_t * const dct_codes,
+							uint8_t * const cur,
+							const uint8_t * const ref,
+							const uint8_t * const refh,
+							const uint8_t * const refv,
+							const uint8_t * const refhv,
+							uint8_t * const tmp,
+							uint32_t x,
+							uint32_t y,
+							const int32_t dx,
+							const int32_t dy,
+							const uint32_t stride,
+							const int quarterpel,
+							const int reduced_resolution,
+							const uint32_t rounding)
+{
+	const uint8_t * ptr;
+
+	if (!reduced_resolution) {
+
+		if(quarterpel) {
+			if (dx&3 | dy&3) {
+				interpolate16x16_quarterpel(tmp - y * stride - x,
+											(uint8_t *) ref, tmp + 32,
+											tmp + 64, tmp + 96, x, y, dx, dy, stride, rounding);
+				ptr = tmp;
+			} else ptr =  ref + (y + dy/4)*stride + x + dx/4; // fullpixel position
+
+		} else ptr = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride);
+
+		transfer_8to16sub(dct_codes, cur + y * stride + x,
+							  ptr, stride);
+		transfer_8to16sub(dct_codes+64, cur + y * stride + x + 8,
+							  ptr + 8, stride);
+		transfer_8to16sub(dct_codes+128, cur + y * stride + x + 8*stride,
+							  ptr + 8*stride, stride);
+		transfer_8to16sub(dct_codes+192, cur + y * stride + x + 8*stride+8,
+							  ptr + 8*stride + 8, stride);
+
+	} else { //reduced_resolution
+	
+		x *= 2; y *= 2;
+
+		ptr = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride);
+		
+		filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride);
+		filter_diff_18x18_to_8x8(dct_codes, ptr, stride);
+
+		filter_18x18_to_8x8(dct_codes+64, cur+y*stride + x + 16, stride);
+		filter_diff_18x18_to_8x8(dct_codes+64, ptr + 16, stride);
 
-#define ABS(X) (((X)>0)?(X):-(X))
-#define SIGN(X) (((X)>0)?1:-1)
+		filter_18x18_to_8x8(dct_codes+128, cur+(y+16)*stride + x, stride);
+		filter_diff_18x18_to_8x8(dct_codes+128, ptr + 16*stride, stride);
 
-static __inline void compensate8x8_halfpel(
-				int16_t * const dct_codes,
-				uint8_t * const cur,
-				const uint8_t * const ref,
-				const uint8_t * const refh,
-				const uint8_t * const refv,
-				const uint8_t * const refhv,
-				const uint32_t x, const uint32_t y,
-				const int32_t dx,  const int dy,
-				const uint32_t stride)
+		filter_18x18_to_8x8(dct_codes+192, cur+(y+16)*stride + x + 16, stride);
+		filter_diff_18x18_to_8x8(dct_codes+192, ptr + 16*stride + 16, stride);
+
+		transfer32x32_copy(cur + y*stride + x, ptr, stride);
+	}
+}
+
+static __inline void
+compensate8x8_interpolate(	int16_t * const dct_codes,
+							uint8_t * const cur,
+							const uint8_t * const ref,
+							const uint8_t * const refh,
+							const uint8_t * const refv,
+							const uint8_t * const refhv,
+							uint8_t * const tmp,
+							uint32_t x,
+							uint32_t y,
+							const int32_t dx,
+							const int32_t dy,
+							const uint32_t stride,
+							const uint32_t quarterpel,
+							const int reduced_resolution,
+							const uint32_t rounding)
 {
-	int32_t ddx,ddy;
+	const uint8_t * ptr;
 
-	switch ( ((dx&1)<<1) + (dy&1) )   // ((dx%2)?2:0)+((dy%2)?1:0)
-    {
-    case 0 :
-		ddx = dx/2;
-		ddy = dy/2;
-		transfer_8to16sub(dct_codes, cur + y*stride + x, 
-				ref + (y+ddy)*stride + x+ddx, stride);
-		break;
+	if (!reduced_resolution) {
 
-    case 1 :
-		ddx = dx/2;
-		ddy = (dy-1)/2;
-		transfer_8to16sub(dct_codes, cur + y*stride + x, 
-				refv + (y+ddy)*stride + x+ddx, stride);
-		break;
+		if(quarterpel) {
+			if (dx&3 | dy&3) {
+				interpolate8x8_quarterpel(tmp - y*stride - x,
+										(uint8_t *) ref, tmp + 32,
+										tmp + 64, tmp + 96, x, y, dx, dy, stride, rounding);
+				ptr = tmp;
+			} else ptr = ref + (y + dy/4)*stride + x + dx/4; // fullpixel position
+		} else ptr = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride);
 
-    case 2 :
-		ddx = (dx-1)/2;
-		ddy = dy/2;
-		transfer_8to16sub(dct_codes, cur + y*stride + x, 
-				refh + (y+ddy)*stride + x+ddx, stride);
-		break;
+			transfer_8to16sub(dct_codes, cur + y * stride + x, ptr, stride);
 
-	default :	// case 3:
-		ddx = (dx-1)/2;
-		ddy = (dy-1)/2;
-		transfer_8to16sub(dct_codes, cur + y*stride + x, 
-				refhv + (y+ddy)*stride + x+ddx, stride);
-		break;
-    }
+	} else { //reduced_resolution
+
+		x *= 2; y *= 2;
+
+		ptr = get_ref(ref, refh, refv, refhv, x, y, 1, dx, dy, stride);
+
+		filter_18x18_to_8x8(dct_codes, cur+y*stride + x, stride);
+		filter_diff_18x18_to_8x8(dct_codes, ptr, stride);
+		
+		transfer16x16_copy(cur + y*stride + x, ptr, stride);
+	}
 }
 
+/* XXX: slow, inelegant... */
+static void
+interpolate18x18_switch(uint8_t * const cur,
+						const uint8_t * const refn,
+						const uint32_t x,
+						const uint32_t y,
+						const int32_t dx,
+						const int dy,
+						const uint32_t stride,
+						const uint32_t rounding)
+{
+	interpolate8x8_switch(cur, refn, x-1, y-1, dx, dy, stride, rounding);
+	interpolate8x8_switch(cur, refn, x+7, y-1, dx, dy, stride, rounding);
+	interpolate8x8_switch(cur, refn, x+9, y-1, dx, dy, stride, rounding);
+
+	interpolate8x8_switch(cur, refn, x-1, y+7, dx, dy, stride, rounding);
+	interpolate8x8_switch(cur, refn, x+7, y+7, dx, dy, stride, rounding);
+	interpolate8x8_switch(cur, refn, x+9, y+7, dx, dy, stride, rounding);
+
+	interpolate8x8_switch(cur, refn, x-1, y+9, dx, dy, stride, rounding);
+	interpolate8x8_switch(cur, refn, x+7, y+9, dx, dy, stride, rounding);
+	interpolate8x8_switch(cur, refn, x+9, y+9, dx, dy, stride, rounding);
+}
 
+static void
+CompensateChroma(	int dx, int dy,
+					const int i, const int j,
+					IMAGE * const Cur,
+					const IMAGE * const Ref,
+					uint8_t * const temp,
+					int16_t * const coeff,
+					const uint32_t stride,
+					const int rounding,
+					const int rrv)
+{ /* uv-block-based compensation */
+
+	if (!rrv) {
+		transfer_8to16sub(coeff, Cur->u + 8 * j * stride + 8 * i,
+				 			interpolate8x8_switch2(temp, Ref->u, 8 * i, 8 * j,
+													dx, dy, stride, rounding),
+							stride);
+
+		transfer_8to16sub(coeff + 64, Cur->v + 8 * j * stride + 8 * i,
+ 				 			interpolate8x8_switch2(temp, Ref->v, 8 * i, 8 * j,
+													dx, dy, stride, rounding),
+							stride);
+	} else {
+		uint8_t * current, * reference;
+
+		current = Cur->u + 16*j*stride + 16*i;
+		reference = temp - 16*j*stride - 16*i;
+		interpolate18x18_switch(reference, Ref->u, 16*i, 16*j, dx, dy, stride, rounding);
+		filter_18x18_to_8x8(coeff, current, stride);
+		filter_diff_18x18_to_8x8(coeff, temp, stride);
+		transfer16x16_copy(current, temp, stride);
+
+		current = Cur->v + 16*j*stride + 16*i;
+		interpolate18x18_switch(reference, Ref->v, 16*i, 16*j, dx, dy, stride, rounding);
+		filter_18x18_to_8x8(coeff + 64, current, stride);
+		filter_diff_18x18_to_8x8(coeff + 64, temp, stride);
+		transfer16x16_copy(current, temp, stride);
+	}
+}
+
+void
+MBMotionCompensation(MACROBLOCK * const mb,
+					const uint32_t i,
+					const uint32_t j,
+					const IMAGE * const ref,
+					const IMAGE * const refh,
+					const IMAGE * const refv,
+					const IMAGE * const refhv,
+					IMAGE * const cur,
+					int16_t * dct_codes,
+					const uint32_t width,
+					const uint32_t height,
+					const uint32_t edged_width,
+					const int quarterpel,
+					const int reduced_resolution,
+					const uint32_t rounding)
+{
+	int32_t dx = (quarterpel ? mb->qmvs[0].x : mb->mvs[0].x);
+	int32_t dy = (quarterpel ? mb->qmvs[0].y : mb->mvs[0].y);
+	uint8_t * const tmp = refv->u;
+
+	if ( mb->mode == MODE_NOT_CODED && dx==0 && dy==0 && !reduced_resolution) {	/* quick copy */
+		transfer16x16_copy(cur->y + 16 * (i + j * edged_width),
+						   ref->y + 16 * (i + j * edged_width),
+						   edged_width);
+	
+		transfer8x8_copy(cur->u + 8 * (i + j * edged_width/2),
+							ref->u + 8 * (i + j * edged_width/2),
+							edged_width / 2);
+		transfer8x8_copy(cur->v + 8 * (i + j * edged_width/2),
+							ref->v + 8 * (i + j * edged_width/2),
+							edged_width / 2);
+		return;
+	}
+
+	if ((mb->mode == MODE_NOT_CODED || mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q) /*&& !quarterpel*/) {
+
+	/* quick MODE_NOT_CODED for GMC with MV!=(0,0) is still needed */
+
+		if (reduced_resolution) {
+			dx = RRV_MV_SCALEUP(dx);
+			dy = RRV_MV_SCALEUP(dy);
+		}
+
+		compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, ref->y, refh->y,
+							refv->y, refhv->y, tmp, 16 * i, 16 * j, dx, dy,
+							edged_width, quarterpel, reduced_resolution, rounding);
+		
+		dx /= 1 + quarterpel;
+		dy /= 1 + quarterpel;
+		dx = (dx >> 1) + roundtab_79[dx & 0x3];
+		dy = (dy >> 1) + roundtab_79[dy & 0x3];
+
+	} else {					// mode == MODE_INTER4V
+		int k, sumx = 0, sumy = 0;
+		const VECTOR * const mvs = (quarterpel ? mb->qmvs : mb->mvs);
+
+		for (k = 0; k < 4; k++) {
+			dx = mvs[k].x;
+			dy = mvs[k].y;
+			sumx += dx / (1 + quarterpel);
+			sumy += dy / (1 + quarterpel);
+
+			if (reduced_resolution){
+				dx = RRV_MV_SCALEUP(dx);
+				dy = RRV_MV_SCALEUP(dy);
+			}
+
+			compensate8x8_interpolate(&dct_codes[k * 64], cur->y, ref->y, refh->y,
+									refv->y, refhv->y, tmp, 16 * i + 8*(k&1), 16 * j + 8*(k>>1), dx,
+									dy, edged_width, quarterpel, reduced_resolution, rounding);
+		}
+		dx = (sumx >> 3) + roundtab_76[sumx & 0xf];
+		dy = (sumy >> 3) + roundtab_76[sumy & 0xf];
+	}
 
-void MBMotionCompensation(
-			MACROBLOCK * const mb,
-		    const uint32_t i,
-			const uint32_t j,
-		    const IMAGE * const ref,
-			const IMAGE * const refh,
-		    const IMAGE * const refv,
-			const IMAGE * const refhv,
-		    IMAGE * const cur,
-		    int16_t dct_codes[][64],
-			const uint32_t width, 
-			const uint32_t height,
-			const uint32_t edged_width,
-			const uint32_t rounding)
+	CompensateChroma(dx, dy, i, j, cur, ref, tmp,
+					&dct_codes[4 * 64], edged_width / 2, rounding, reduced_resolution);
+
+}
+
+
+void
+MBMotionCompensationBVOP(MBParam * pParam,
+						MACROBLOCK * const mb,
+						const uint32_t i,
+						const uint32_t j,
+						IMAGE * const cur,
+						const IMAGE * const f_ref,
+						const IMAGE * const f_refh,
+						const IMAGE * const f_refv,
+						const IMAGE * const f_refhv,
+						const IMAGE * const b_ref,
+						const IMAGE * const b_refh,
+						const IMAGE * const b_refv,
+						const IMAGE * const b_refhv,
+						int16_t * dct_codes)
 {
-	static const uint32_t roundtab[16] =
-		{ 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2 };
+	const uint32_t edged_width = pParam->edged_width;
+	int32_t dx, dy, b_dx, b_dy, sumx, sumy, b_sumx, b_sumy;
+	int k;
+	const int quarterpel = pParam->m_quarterpel;
+	const uint8_t * ptr1, * ptr2;
+	uint8_t * const tmp = f_refv->u;
+	const VECTOR * const fmvs = (quarterpel ? mb->qmvs : mb->mvs);
+	const VECTOR * const bmvs = (quarterpel ? mb->b_qmvs : mb->b_mvs);
+
+	switch (mb->mode) {
+	case MODE_FORWARD:
+		dx = fmvs->x; dy = fmvs->y;
+
+		compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, f_ref->y, f_refh->y,
+							f_refv->y, f_refhv->y, tmp, 16 * i, 16 * j, dx,
+							dy, edged_width, quarterpel, 0, 0);
+
+		dx /= 1 + quarterpel;
+		dy /= 1 + quarterpel;
+		CompensateChroma(	(dx >> 1) + roundtab_79[dx & 0x3],
+							(dy >> 1) + roundtab_79[dy & 0x3],
+							i, j, cur, f_ref, tmp,
+							&dct_codes[4 * 64], edged_width / 2, 0, 0);
+
+		return;
+
+	case MODE_BACKWARD:
+		b_dx = bmvs->x; b_dy = bmvs->y;
+
+		compensate16x16_interpolate(&dct_codes[0 * 64], cur->y, b_ref->y, b_refh->y,
+							b_refv->y, b_refhv->y, tmp, 16 * i, 16 * j, b_dx,
+							b_dy, edged_width, quarterpel, 0, 0);
+
+		b_dx /= 1 + quarterpel;
+		b_dy /= 1 + quarterpel;
+		CompensateChroma(	(b_dx >> 1) + roundtab_79[b_dx & 0x3],
+							(b_dy >> 1) + roundtab_79[b_dy & 0x3],
+							i, j, cur, b_ref, tmp,
+							&dct_codes[4 * 64], edged_width / 2, 0, 0);
+
+		return;
+
+	case MODE_INTERPOLATE: /* _could_ use DIRECT, but would be overkill (no 4MV there) */
+	case MODE_DIRECT_NO4V:
+		dx = fmvs->x; dy = fmvs->y;
+		b_dx = bmvs->x; b_dy = bmvs->y;
+
+		if (quarterpel) {
+			
+			if (dx&3 | dy&3) {
+				interpolate16x16_quarterpel(tmp - i * 16 - j * 16 * edged_width,
+					(uint8_t *) f_ref->y, tmp + 32,
+					tmp + 64, tmp + 96, 16*i, 16*j, dx, dy, edged_width, 0);
+				ptr1 = tmp;
+			} else ptr1 = f_ref->y + (16*j + dy/4)*edged_width + 16*i + dx/4; // fullpixel position
+
+			if (b_dx&3 | b_dy&3) {
+				interpolate16x16_quarterpel(tmp - i * 16 - j * 16 * edged_width + 16,
+					(uint8_t *) b_ref->y, tmp + 32,
+					tmp + 64, tmp + 96, 16*i, 16*j, b_dx, b_dy, edged_width, 0);
+				ptr2 = tmp + 16;
+			} else ptr2 = b_ref->y + (16*j + b_dy/4)*edged_width + 16*i + b_dx/4; // fullpixel position
+
+			b_dx /= 2;
+			b_dy /= 2;
+			dx /= 2;
+			dy /= 2;
+
+		} else {
+			ptr1 = get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y,
+							i, j, 16, dx, dy, edged_width);
+
+			ptr2 = get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y,
+							i, j, 16, b_dx, b_dy, edged_width);
+		}
+		for (k = 0; k < 4; k++)
+				transfer_8to16sub2(&dct_codes[k * 64],
+									cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width,
+									ptr1 + (k&1)*8 + (k>>1)*8*edged_width,
+									ptr2 + (k&1)*8 + (k>>1)*8*edged_width, edged_width);
+
 
+		dx = (dx >> 1) + roundtab_79[dx & 0x3];
+		dy = (dy >> 1) + roundtab_79[dy & 0x3];
 
-	if (mb->mode == MODE_INTER || mb->mode == MODE_INTER_Q)
-	{
-		int32_t dx = mb->mvs[0].x;
-		int32_t dy = mb->mvs[0].y;
-
-		compensate8x8_halfpel(dct_codes[0], cur->y, ref->y, refh->y, refv->y, refhv->y,
-								16*i,     16*j,     dx, dy, edged_width);
-		compensate8x8_halfpel(dct_codes[1], cur->y, ref->y, refh->y, refv->y, refhv->y,
-								16*i + 8, 16*j,     dx, dy, edged_width);
-		compensate8x8_halfpel(dct_codes[2], cur->y, ref->y, refh->y, refv->y, refhv->y,
-								16*i,     16*j + 8, dx, dy, edged_width);
-		compensate8x8_halfpel(dct_codes[3], cur->y, ref->y, refh->y, refv->y, refhv->y,
-								16*i + 8, 16*j + 8, dx, dy, edged_width);
-
-		dx = (dx & 3) ? (dx >> 1) | 1 : dx / 2;
-		dy = (dy & 3) ? (dy >> 1) | 1 : dy / 2;
-
-		/* uv-image-based compensation
-		compensate8x8_halfpel(dct_codes[4], cur->u, ref->u, refh->u, refv->u, refhv->u,
-								8*i, 8*j, dx, dy, edged_width/2);
-		compensate8x8_halfpel(dct_codes[5], cur->v, ref->v, refh->v, refv->v, refhv->v,
-								8*i, 8*j, dx, dy, edged_width/2);		*/
-
-		/* uv-block-based compensation */
-		interpolate8x8_switch(refv->u, ref->u, 8*i, 8*j, dx, dy, edged_width/2, rounding);
-		transfer_8to16sub(dct_codes[4], 
-				cur->u + 8*j*edged_width/2 + 8*i, 
-				refv->u + 8*j*edged_width/2 + 8*i, edged_width/2);
-
-		interpolate8x8_switch(refv->v, ref->v, 8*i, 8*j, dx, dy, edged_width/2, rounding);
-		transfer_8to16sub(dct_codes[5], 
-				cur->v + 8*j*edged_width/2 + 8*i, 
-				refv->v + 8*j*edged_width/2 + 8*i, edged_width/2);
-
-	}
-	else	// mode == MODE_INTER4V
-	{
-		int32_t sum, dx, dy;
-
-		compensate8x8_halfpel(dct_codes[0], cur->y, ref->y, refh->y, refv->y, refhv->y,
-								16*i,     16*j,     mb->mvs[0].x, mb->mvs[0].y, edged_width);
-		compensate8x8_halfpel(dct_codes[1], cur->y, ref->y, refh->y, refv->y, refhv->y,
-								16*i + 8, 16*j,     mb->mvs[1].x, mb->mvs[1].y, edged_width);
-		compensate8x8_halfpel(dct_codes[2], cur->y, ref->y, refh->y, refv->y, refhv->y,
-								16*i,     16*j + 8, mb->mvs[2].x, mb->mvs[2].y, edged_width);
-		compensate8x8_halfpel(dct_codes[3], cur->y, ref->y, refh->y, refv->y, refhv->y,
-								16*i + 8, 16*j + 8, mb->mvs[3].x, mb->mvs[3].y, edged_width);
-
-		sum = mb->mvs[0].x + mb->mvs[1].x + mb->mvs[2].x + mb->mvs[3].x;
-		dx = (sum ? SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0);
-
-		sum = mb->mvs[0].y + mb->mvs[1].y + mb->mvs[2].y + mb->mvs[3].y;
-		dy = (sum ? SIGN(sum) * (roundtab[ABS(sum) % 16] + (ABS(sum) / 16) * 2) : 0);
-
-		/* uv-image-based compensation
-		compensate8x8_halfpel(dct_codes[4], cur->u, ref->u, refh->u, refv->u, refhv->u,
-								8*i, 8*j, dx, dy, edged_width/2);
-		compensate8x8_halfpel(dct_codes[5], cur->v, ref->v, refh->v, refv->v, refhv->v,
-								8*i, 8*j, dx, dy, edged_width/2);		*/
-
-		/* uv-block-based compensation */
-		interpolate8x8_switch(refv->u, ref->u, 8*i, 8*j, dx, dy, edged_width/2, rounding);
-		transfer_8to16sub(dct_codes[4], 
-				cur->u + 8*j*edged_width/2 + 8*i, 
-				refv->u + 8*j*edged_width/2 + 8*i, edged_width/2);
-
-		interpolate8x8_switch(refv->v, ref->v, 8*i, 8*j, dx, dy, edged_width/2, rounding);
-		transfer_8to16sub(dct_codes[5], 
-				cur->v + 8*j*edged_width/2 + 8*i, 
-				refv->v + 8*j*edged_width/2 + 8*i, edged_width/2);
+		b_dx = (b_dx >> 1) + roundtab_79[b_dx & 0x3];
+		b_dy = (b_dy >> 1) + roundtab_79[b_dy & 0x3];
 
+		break;
+	
+	default: // MODE_DIRECT
+		sumx = sumy = b_sumx = b_sumy = 0;
+
+		for (k = 0; k < 4; k++) {
+			
+			dx = fmvs[k].x; dy = fmvs[k].y;
+			b_dx = bmvs[k].x; b_dy = bmvs[k].y;
+
+			if (quarterpel) {
+				sumx += dx/2; sumy += dy/2;
+				b_sumx += b_dx/2; b_sumy += b_dy/2;
+
+				if (dx&3 | dy&3) {
+					interpolate8x8_quarterpel(tmp - (i * 16+(k&1)*8) - (j * 16+((k>>1)*8)) * edged_width,
+						(uint8_t *) f_ref->y, 
+						tmp + 32, tmp + 64, tmp + 96, 
+						16*i + (k&1)*8, 16*j + (k>>1)*8, dx, dy, edged_width, 0);
+					ptr1 = tmp;
+				} else ptr1 = f_ref->y + (16*j + (k>>1)*8 + dy/4)*edged_width + 16*i + (k&1)*8 + dx/4;
+
+				if (b_dx&3 | b_dy&3) {
+					interpolate8x8_quarterpel(tmp - (i * 16+(k&1)*8) - (j * 16+((k>>1)*8)) * edged_width + 16,
+						(uint8_t *) b_ref->y,
+						tmp + 16, tmp + 32, tmp + 48, 
+						16*i + (k&1)*8, 16*j + (k>>1)*8, b_dx, b_dy, edged_width, 0);
+					ptr2 = tmp + 16;
+				} else ptr2 = b_ref->y + (16*j + (k>>1)*8 + b_dy/4)*edged_width + 16*i + (k&1)*8 + b_dx/4;
+			} else {
+				sumx += dx; sumy += dy;
+				b_sumx += b_dx; b_sumy += b_dy;
+
+				ptr1 = get_ref(f_ref->y, f_refh->y, f_refv->y, f_refhv->y, 
+								2*i + (k&1), 2*j + (k>>1), 8, dx, dy, edged_width);
+				ptr2 = get_ref(b_ref->y, b_refh->y, b_refv->y, b_refhv->y, 
+								2*i + (k&1), 2*j + (k>>1), 8, b_dx, b_dy,  edged_width);
+			}
+			transfer_8to16sub2(&dct_codes[k * 64],
+								cur->y + (i * 16+(k&1)*8) + (j * 16+((k>>1)*8)) * edged_width,
+								ptr1, ptr2,	edged_width);
+						
+		}
+
+		dx = (sumx >> 3) + roundtab_76[sumx & 0xf];
+		dy = (sumy >> 3) + roundtab_76[sumy & 0xf];
+		b_dx = (b_sumx >> 3) + roundtab_76[b_sumx & 0xf];
+		b_dy = (b_sumy >> 3) + roundtab_76[b_sumy & 0xf];
+
+		break;
 	}
+
+	// uv block-based chroma interpolation for direct and interpolate modes
+	transfer_8to16sub2(&dct_codes[4 * 64],
+						cur->u + (j * 8) * edged_width / 2 + (i * 8),
+						interpolate8x8_switch2(tmp, b_ref->u, 8 * i, 8 * j,
+												b_dx, b_dy, edged_width / 2, 0),
+						interpolate8x8_switch2(tmp + 8, f_ref->u, 8 * i, 8 * j,
+												dx, dy, edged_width / 2, 0),
+						edged_width / 2);
+
+	transfer_8to16sub2(&dct_codes[5 * 64],
+						cur->v + (j * 8) * edged_width / 2 + (i * 8),
+						interpolate8x8_switch2(tmp, b_ref->v, 8 * i, 8 * j,
+												b_dx, b_dy, edged_width / 2, 0),
+						interpolate8x8_switch2(tmp + 8, f_ref->v, 8 * i, 8 * j,
+												dx, dy, edged_width / 2, 0),
+						edged_width / 2);
 }