--- decoder.c	2004/07/10 17:49:31	1.61
+++ decoder.c	2004/08/16 22:38:06	1.67
@@ -20,7 +20,7 @@
  *  along with this program ; if not, write to the Free Software
  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  *
- * $Id: decoder.c,v 1.61 2004/07/10 17:49:31 edgomez Exp $
+ * $Id: decoder.c,v 1.67 2004/08/16 22:38:06 edgomez Exp $
  *
  ****************************************************************************/
 
@@ -48,6 +48,7 @@
 #include "image/interpolate8x8.h"
 #include "image/reduced.h"
 #include "image/font.h"
+#include "image/qpel.h"
 
 #include "bitstream/mbcoding.h"
 #include "prediction/mbprediction.h"
@@ -390,14 +391,13 @@
 				uint8_t * pY_Cur,
 				uint8_t * pU_Cur,
 				uint8_t * pV_Cur,
-				const int reduced_resolution,
+				int reduced_resolution,
 				const MACROBLOCK * pMB)
 {
-	DECLARE_ALIGNED_MATRIX(data, 6, 64, int16_t, CACHE_LINE);
+	DECLARE_ALIGNED_MATRIX(data, 1, 64, int16_t, CACHE_LINE);
 
 	int stride = dec->edged_width;
 	int next_block = stride * (reduced_resolution ? 16 : 8);
-	const int stride2 = stride/2;
 	int i;
 	const uint32_t iQuant = pMB->quant;
 	const int direction = dec->alternate_vertical_scan ? 2 : 0;
@@ -407,63 +407,98 @@
 			int direction,
 			const int quant,
 			const uint16_t *matrix);
+	typedef void (*add_residual_function_t)(
+			uint8_t *predicted_block,
+			const int16_t *residual,
+			int stride);
 
 	const get_inter_block_function_t get_inter_block = (dec->quant_type == 0)
-		? get_inter_block_h263
-		: get_inter_block_mpeg;
+		? (get_inter_block_function_t)get_inter_block_h263
+		: (get_inter_block_function_t)get_inter_block_mpeg;
+
+	const add_residual_function_t add_residual = (reduced_resolution)
+		? (add_residual_function_t)add_upsampled_8x8_16to8
+		: (add_residual_function_t)transfer_16to8add;
+
+	uint8_t *dst[6];
+	int strides[6];
 	
-	memset(&data[0], 0, 6*64*sizeof(int16_t));	/* clear */
 
-	for (i = 0; i < 6; i++) {
+	if (dec->interlacing && pMB->field_dct) {
+		next_block = stride;
+		stride *= 2;
+	}
 
-		if (cbp & (1 << (5 - i))) {	/* coded */
+	reduced_resolution = !!reduced_resolution;
+	dst[0] = pY_Cur;
+	dst[2] = pY_Cur + next_block;
+	dst[1] = dst[0] + (8<<reduced_resolution);
+	dst[3] = dst[2] + (8<<reduced_resolution);
+	dst[4] = pU_Cur;
+	dst[5] = pV_Cur;
+	strides[0] = strides[1] = strides[2] = strides[3] = stride;
+	strides[4] = stride/2;
+	strides[5] = stride/2;
+	
+	for (i = 0; i < 6; i++) {
+		/* Process only coded blocks */
+		if (cbp & (1 << (5 - i))) {
 
+			/* Clear the block */
+			memset(&data[0], 0, 64*sizeof(int16_t));
 
 			/* Decode coeffs and dequantize on the fly */
 			start_timer();
-			get_inter_block(bs, &data[i*64], direction, iQuant, get_inter_matrix(dec->mpeg_quant_matrices));
+			get_inter_block(bs, &data[0], direction, iQuant, get_inter_matrix(dec->mpeg_quant_matrices));
 			stop_coding_timer();
 
+			/* iDCT */
 			start_timer();
-			idct(&data[i * 64]);
+			idct(&data[0]);
 			stop_idct_timer();
-		}
-	}
 
-	if (dec->interlacing && pMB->field_dct) {
-		next_block = stride;
-		stride *= 2;
+			/* Add this residual to the predicted block */
+			start_timer();
+			add_residual(dst[i], &data[0], strides[i]);
+			stop_transfer_timer();
+		}
 	}
+}
 
-	start_timer();
-	if (reduced_resolution) {
-		if (cbp & 32)
-			add_upsampled_8x8_16to8(pY_Cur, &data[0 * 64], stride);
-		if (cbp & 16)
-			add_upsampled_8x8_16to8(pY_Cur + 16, &data[1 * 64], stride);
-		if (cbp & 8)
-			add_upsampled_8x8_16to8(pY_Cur + next_block, &data[2 * 64], stride);
-		if (cbp & 4)
-			add_upsampled_8x8_16to8(pY_Cur + 16 + next_block, &data[3 * 64], stride);
-		if (cbp & 2)
-			add_upsampled_8x8_16to8(pU_Cur, &data[4 * 64], stride2);
-		if (cbp & 1)
-			add_upsampled_8x8_16to8(pV_Cur, &data[5 * 64], stride2);
-	} else {
-		if (cbp & 32)
-			transfer_16to8add(pY_Cur, &data[0 * 64], stride);
-		if (cbp & 16)
-			transfer_16to8add(pY_Cur + 8, &data[1 * 64], stride);
-		if (cbp & 8)
-			transfer_16to8add(pY_Cur + next_block, &data[2 * 64], stride);
-		if (cbp & 4)
-			transfer_16to8add(pY_Cur + 8 + next_block, &data[3 * 64], stride);
-		if (cbp & 2)
-			transfer_16to8add(pU_Cur, &data[4 * 64], stride2);
-		if (cbp & 1)
-			transfer_16to8add(pV_Cur, &data[5 * 64], stride2);
-	}
-	stop_transfer_timer();
+static void __inline
+validate_vector(VECTOR * mv, unsigned int x_pos, unsigned int y_pos, const DECODER * dec)
+{
+	/* clip a vector to valid range 
+	   prevents crashes if bitstream is broken 
+	*/
+	int shift = 5 + dec->quarterpel;
+	int xborder_high = (int)(dec->mb_width - x_pos) << shift;
+	int xborder_low = (-(int)x_pos-1) << shift;
+	int yborder_high = (int)(dec->mb_height - y_pos) << shift;
+	int yborder_low = (-(int)y_pos-1) << shift;
+
+#define CHECK_MV(mv) \
+	do { \
+	if ((mv).x > xborder_high) { \
+		DPRINTF(XVID_DEBUG_MV, "mv.x > max -- %d > %d, MB %d, %d", (mv).x, xborder_high, x_pos, y_pos); \
+		(mv).x = xborder_high; \
+	} else if ((mv).x < xborder_low) { \
+		DPRINTF(XVID_DEBUG_MV, "mv.x < min -- %d < %d, MB %d, %d", (mv).x, xborder_low, x_pos, y_pos); \
+		(mv).x = xborder_low; \
+	} \
+	if ((mv).y > yborder_high) { \
+		DPRINTF(XVID_DEBUG_MV, "mv.y > max -- %d > %d, MB %d, %d", (mv).y, yborder_high, x_pos, y_pos); \
+		(mv).y = yborder_high; \
+	} else if ((mv).y < yborder_low) { \
+		DPRINTF(XVID_DEBUG_MV, "mv.y < min -- %d < %d, MB %d, %d", (mv).y, yborder_low, x_pos, y_pos); \
+		(mv).y = yborder_low; \
+	} \
+	} while (0)
+
+	CHECK_MV(mv[0]);
+	CHECK_MV(mv[1]);
+	CHECK_MV(mv[2]);
+	CHECK_MV(mv[3]);
 }
 
 /* decode an inter macroblock */
@@ -503,32 +538,7 @@
 			mv[i] = pMB->mvs[i];
 	}
 
-	for (i = 0; i < 4; i++) {
-		/* clip to valid range */
-		int border = (int)(dec->mb_width - x_pos) << (5 + dec->quarterpel);
-		if (mv[i].x > border) {
-			DPRINTF(XVID_DEBUG_MV, "mv.x > max -- %d > %d, MB %d, %d", mv[i].x, border, x_pos, y_pos);
-			mv[i].x = border;
-		} else {
-			border = (-(int)x_pos-1) << (5 + dec->quarterpel);
-			if (mv[i].x < border) {
-				DPRINTF(XVID_DEBUG_MV, "mv.x < min -- %d < %d, MB %d, %d", mv[i].x, border, x_pos, y_pos);
-				mv[i].x = border;
-			}
-		}
-
-		border = (int)(dec->mb_height - y_pos) << (5 + dec->quarterpel);
-		if (mv[i].y >  border) {
-			DPRINTF(XVID_DEBUG_MV, "mv.y > max -- %d > %d, MB %d, %d", mv[i].y, border, x_pos, y_pos);
-			mv[i].y = border;
-		} else {
-			border = (-(int)y_pos-1) << (5 + dec->quarterpel);
-			if (mv[i].y < border) {
-				DPRINTF(XVID_DEBUG_MV, "mv.y < min -- %d < %d, MB %d, %d", mv[i].y, border, x_pos, y_pos);
-				mv[i].y = border;
-			}
-		}
-	}
+	validate_vector(mv, x_pos, y_pos, dec);
 
 	start_timer();
 
@@ -1018,7 +1028,7 @@
 decoder_bf_interpolate_mbinter(DECODER * dec,
 								IMAGE forward,
 								IMAGE backward,
-								const MACROBLOCK * pMB,
+								MACROBLOCK * pMB,
 								const uint32_t x_pos,
 								const uint32_t y_pos,
 								Bitstream * bs,
@@ -1035,10 +1045,12 @@
 	pU_Cur = dec->cur.u + (y_pos << 3) * stride2 + (x_pos << 3);
 	pV_Cur = dec->cur.v + (y_pos << 3) * stride2 + (x_pos << 3);
 
+	validate_vector(pMB->mvs, x_pos, y_pos, dec);
+	validate_vector(pMB->b_mvs, x_pos, y_pos, dec);
+
 	if (!direct) {
 		uv_dx = pMB->mvs[0].x;
 		uv_dy = pMB->mvs[0].y;
-
 		b_uv_dx = pMB->b_mvs[0].x;
 		b_uv_dy = pMB->b_mvs[0].y;
 
@@ -1051,21 +1063,20 @@
 
 		uv_dx = (uv_dx >> 1) + roundtab_79[uv_dx & 0x3];
 		uv_dy = (uv_dy >> 1) + roundtab_79[uv_dy & 0x3];
-
 		b_uv_dx = (b_uv_dx >> 1) + roundtab_79[b_uv_dx & 0x3];
 		b_uv_dy = (b_uv_dy >> 1) + roundtab_79[b_uv_dy & 0x3];
 
 	} else {
-		if(dec->quarterpel) {
-			uv_dx = (pMB->mvs[0].x / 2) + (pMB->mvs[1].x / 2) + (pMB->mvs[2].x / 2) + (pMB->mvs[3].x / 2);
-			uv_dy = (pMB->mvs[0].y / 2) + (pMB->mvs[1].y / 2) + (pMB->mvs[2].y / 2) + (pMB->mvs[3].y / 2);
-			b_uv_dx = (pMB->b_mvs[0].x / 2) + (pMB->b_mvs[1].x / 2) + (pMB->b_mvs[2].x / 2) + (pMB->b_mvs[3].x / 2);
-			b_uv_dy = (pMB->b_mvs[0].y / 2) + (pMB->b_mvs[1].y / 2) + (pMB->b_mvs[2].y / 2) + (pMB->b_mvs[3].y / 2);
-		} else {
-			uv_dx = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x;
-			uv_dy = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y;
-			b_uv_dx = pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + pMB->b_mvs[3].x;
-			b_uv_dy = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y;
+		uv_dx = pMB->mvs[0].x + pMB->mvs[1].x + pMB->mvs[2].x + pMB->mvs[3].x;
+		uv_dy = pMB->mvs[0].y + pMB->mvs[1].y + pMB->mvs[2].y + pMB->mvs[3].y;
+		b_uv_dx = pMB->b_mvs[0].x + pMB->b_mvs[1].x + pMB->b_mvs[2].x + pMB->b_mvs[3].x;
+		b_uv_dy = pMB->b_mvs[0].y + pMB->b_mvs[1].y + pMB->b_mvs[2].y + pMB->b_mvs[3].y;
+
+		if (dec->quarterpel) {
+			uv_dx /= 2;
+			uv_dy /= 2;
+			b_uv_dx /= 2;
+			b_uv_dy /= 2;
 		}
 
 		uv_dx = (uv_dx >> 3) + roundtab_76[uv_dx & 0xf];
@@ -1113,68 +1124,38 @@
 
 	if(dec->quarterpel) {
 		if(!direct) {
-			interpolate16x16_quarterpel(dec->tmp.y, backward.y, dec->qtmp.y, dec->qtmp.y + 64,
- 										dec->qtmp.y + 128, 16*x_pos, 16*y_pos,
-										pMB->b_mvs[0].x, pMB->b_mvs[0].y, stride, 0);
+			interpolate16x16_add_quarterpel(dec->cur.y, backward.y, dec->qtmp.y, dec->qtmp.y + 64,
+					dec->qtmp.y + 128, 16*x_pos, 16*y_pos,
+					pMB->b_mvs[0].x, pMB->b_mvs[0].y, stride, 0);
 		} else {
-			interpolate8x8_quarterpel(dec->tmp.y, backward.y, dec->qtmp.y, dec->qtmp.y + 64,
- 										dec->qtmp.y + 128, 16*x_pos, 16*y_pos,
-										pMB->b_mvs[0].x, pMB->b_mvs[0].y, stride, 0);
-			interpolate8x8_quarterpel(dec->tmp.y, backward.y, dec->qtmp.y, dec->qtmp.y + 64,
- 										dec->qtmp.y + 128, 16*x_pos + 8, 16*y_pos,
-										pMB->b_mvs[1].x, pMB->b_mvs[1].y, stride, 0);
-			interpolate8x8_quarterpel(dec->tmp.y, backward.y, dec->qtmp.y, dec->qtmp.y + 64,
- 										dec->qtmp.y + 128, 16*x_pos, 16*y_pos + 8,
-										pMB->b_mvs[2].x, pMB->b_mvs[2].y, stride, 0);
-			interpolate8x8_quarterpel(dec->tmp.y, backward.y, dec->qtmp.y, dec->qtmp.y + 64,
- 										dec->qtmp.y + 128, 16*x_pos + 8, 16*y_pos + 8,
-										pMB->b_mvs[3].x, pMB->b_mvs[3].y, stride, 0);
+			interpolate8x8_add_quarterpel(dec->cur.y, backward.y, dec->qtmp.y, dec->qtmp.y + 64,
+					dec->qtmp.y + 128, 16*x_pos, 16*y_pos,
+					pMB->b_mvs[0].x, pMB->b_mvs[0].y, stride, 0);
+			interpolate8x8_add_quarterpel(dec->cur.y, backward.y, dec->qtmp.y, dec->qtmp.y + 64,
+					dec->qtmp.y + 128, 16*x_pos + 8, 16*y_pos,
+					pMB->b_mvs[1].x, pMB->b_mvs[1].y, stride, 0);
+			interpolate8x8_add_quarterpel(dec->cur.y, backward.y, dec->qtmp.y, dec->qtmp.y + 64,
+					dec->qtmp.y + 128, 16*x_pos, 16*y_pos + 8,
+					pMB->b_mvs[2].x, pMB->b_mvs[2].y, stride, 0);
+			interpolate8x8_add_quarterpel(dec->cur.y, backward.y, dec->qtmp.y, dec->qtmp.y + 64,
+					dec->qtmp.y + 128, 16*x_pos + 8, 16*y_pos + 8,
+					pMB->b_mvs[3].x, pMB->b_mvs[3].y, stride, 0);
 		}
 	} else {
-		interpolate8x8_switch(dec->tmp.y, backward.y, 16 * x_pos, 16 * y_pos,
-							pMB->b_mvs[0].x, pMB->b_mvs[0].y, stride, 0);
-		interpolate8x8_switch(dec->tmp.y, backward.y, 16 * x_pos + 8,
-							16 * y_pos, pMB->b_mvs[1].x, pMB->b_mvs[1].y, stride, 0);
-		interpolate8x8_switch(dec->tmp.y, backward.y, 16 * x_pos,
-							16 * y_pos + 8, pMB->b_mvs[2].x, pMB->b_mvs[2].y, stride, 0);
-		interpolate8x8_switch(dec->tmp.y, backward.y, 16 * x_pos + 8,
-							16 * y_pos + 8, pMB->b_mvs[3].x, pMB->b_mvs[3].y, stride, 0);
-	}
-
-	interpolate8x8_switch(dec->tmp.u, backward.u, 8 * x_pos, 8 * y_pos,
-						b_uv_dx, b_uv_dy, stride2, 0);
-	interpolate8x8_switch(dec->tmp.v, backward.v, 8 * x_pos, 8 * y_pos,
-						b_uv_dx, b_uv_dy, stride2, 0);
-
-	interpolate8x8_avg2(dec->cur.y + (16 * y_pos * stride) + 16 * x_pos,
-						dec->cur.y + (16 * y_pos * stride) + 16 * x_pos,
-						dec->tmp.y + (16 * y_pos * stride) + 16 * x_pos,
-						stride, 1, 8);
-
-	interpolate8x8_avg2(dec->cur.y + (16 * y_pos * stride) + 16 * x_pos + 8,
-						dec->cur.y + (16 * y_pos * stride) + 16 * x_pos + 8,
-						dec->tmp.y + (16 * y_pos * stride) + 16 * x_pos + 8,
-						stride, 1, 8);
-
-	interpolate8x8_avg2(dec->cur.y + ((16 * y_pos + 8) * stride) + 16 * x_pos,
-						dec->cur.y + ((16 * y_pos + 8) * stride) + 16 * x_pos,
-						dec->tmp.y + ((16 * y_pos + 8) * stride) + 16 * x_pos,
-						stride, 1, 8);
-
-	interpolate8x8_avg2(dec->cur.y + ((16 * y_pos + 8) * stride) + 16 * x_pos + 8,
-						dec->cur.y + ((16 * y_pos + 8) * stride) + 16 * x_pos + 8,
-						dec->tmp.y + ((16 * y_pos + 8) * stride) + 16 * x_pos + 8,
-						stride, 1, 8);
-
-	interpolate8x8_avg2(dec->cur.u + (8 * y_pos * stride2) + 8 * x_pos,
-						dec->cur.u + (8 * y_pos * stride2) + 8 * x_pos,
-						dec->tmp.u + (8 * y_pos * stride2) + 8 * x_pos,
-						stride2, 1, 8);
-
-	interpolate8x8_avg2(dec->cur.v + (8 * y_pos * stride2) + 8 * x_pos,
-						dec->cur.v + (8 * y_pos * stride2) + 8 * x_pos,
-						dec->tmp.v + (8 * y_pos * stride2) + 8 * x_pos,
-						stride2, 1, 8);
+		interpolate8x8_add_switch(dec->cur.y, backward.y, 16 * x_pos, 16 * y_pos,
+				pMB->b_mvs[0].x, pMB->b_mvs[0].y, stride, 0);
+		interpolate8x8_add_switch(dec->cur.y, backward.y, 16 * x_pos + 8,
+				16 * y_pos, pMB->b_mvs[1].x, pMB->b_mvs[1].y, stride, 0);
+		interpolate8x8_add_switch(dec->cur.y, backward.y, 16 * x_pos,
+				16 * y_pos + 8, pMB->b_mvs[2].x, pMB->b_mvs[2].y, stride, 0);
+		interpolate8x8_add_switch(dec->cur.y, backward.y, 16 * x_pos + 8,
+				16 * y_pos + 8, pMB->b_mvs[3].x, pMB->b_mvs[3].y, stride, 0);
+	}
+
+	interpolate8x8_add_switch(dec->cur.u, backward.u, 8 * x_pos, 8 * y_pos,
+			b_uv_dx, b_uv_dy, stride2, 0);
+	interpolate8x8_add_switch(dec->cur.v, backward.v, 8 * x_pos, 8 * y_pos,
+			b_uv_dx, b_uv_dy, stride2, 0);
 
 	stop_comp_timer();
 
@@ -1414,7 +1395,6 @@
 	}
 }
 
-
 int
 decoder_decode(DECODER * dec,
 				xvid_dec_frame_t * frame, xvid_dec_stats_t * stats)