rate distortion mb decision support

fix decoding of old %16!=0 divx
fix assertion failure in motion_est.c

Originally committed as revision 2094 to svn://svn.ffmpeg.org/ffmpeg/trunk

10 files changed, 221 insertions, 134 deletions

diff --git a/libavcodec/avcodec.h b/libavcodec/avcodec.h
index e61dd43a46..889c3147ae 100644
--- a/libavcodec/avcodec.h
+++ b/libavcodec/avcodec.h
@@ -15,8 +15,8 @@ extern "C" {
 
 #define LIBAVCODEC_VERSION_INT 0x000406
 #define LIBAVCODEC_VERSION     "0.4.6"
-#define LIBAVCODEC_BUILD       4672
-#define LIBAVCODEC_BUILD_STR   "4672"
+#define LIBAVCODEC_BUILD       4673
+#define LIBAVCODEC_BUILD_STR   "4673"
 
 #define LIBAVCODEC_IDENT	"FFmpeg" LIBAVCODEC_VERSION "b" LIBAVCODEC_BUILD_STR
 
@@ -169,7 +169,6 @@ static const int Motion_Est_QTab[] = { ME_ZERO, ME_PHODS, ME_LOG,
    Note: note not everything is supported yet 
 */
 
-#define CODEC_FLAG_HQ     0x0001  ///< brute force MB-type decission mode (slow) 
 #define CODEC_FLAG_QSCALE 0x0002  ///< use fixed qscale 
 #define CODEC_FLAG_4MV    0x0004  ///< 4 MV per MB allowed 
 #define CODEC_FLAG_QPEL   0x0010  ///< use qpel MC 
@@ -1173,7 +1172,17 @@ typedef struct AVCodecContext {
      * - decoding: set by decoder
      */
     int xvmc_acceleration;
-
+    
+    /**
+     * macroblock decision mode
+     * - encoding: set by user.
+     * - decoding: unused
+     */
+    int mb_decision;
+#define FF_MB_DECISION_SIMPLE 0        ///< uses mb_cmp
+#define FF_MB_DECISION_BITS   1        ///< chooses the one which needs the fewest bits
+#define FF_MB_DECISION_RD     2        ///< rate distoration
+    
 } AVCodecContext;
 
 
diff --git a/libavcodec/error_resilience.c b/libavcodec/error_resilience.c
index fd7eb37737..b13eea2f70 100644
--- a/libavcodec/error_resilience.c
+++ b/libavcodec/error_resilience.c
@@ -30,6 +30,14 @@
 #include "mpegvideo.h"
 #include "common.h"
 
+static void decode_mb(MpegEncContext *s){
+    s->dest[0] = s->current_picture.data[0] + (s->mb_y * 16* s->linesize  ) + s->mb_x * 16;
+    s->dest[1] = s->current_picture.data[1] + (s->mb_y * 8 * s->uvlinesize) + s->mb_x * 8;
+    s->dest[2] = s->current_picture.data[2] + (s->mb_y * 8 * s->uvlinesize) + s->mb_x * 8;
+
+    MPV_decode_mb(s, s->block);    
+}
+
 /**
  * replaces the current MB with a flat dc only version.
  */
@@ -346,7 +354,7 @@ static void guess_mv(MpegEncContext *s){
                 s->mb_y= mb_y;
                 s->mv[0][0][0]= 0;
                 s->mv[0][0][1]= 0;
-                MPV_decode_mb(s, s->block);
+                decode_mb(s);
             }
         }
         return;
@@ -480,7 +488,7 @@ int score_sum=0;
                         s->motion_val[mot_index][0]= s->mv[0][0][0]= mv_predictor[j][0];
                         s->motion_val[mot_index][1]= s->mv[0][0][1]= mv_predictor[j][1];
 
-                        MPV_decode_mb(s, s->block);
+                        decode_mb(s);
                         
                         if(mb_x>0 && fixed[mb_xy-1]){
                             int k;
@@ -513,7 +521,7 @@ score_sum+= best_score;
                     s->motion_val[mot_index][0]= s->mv[0][0][0]= mv_predictor[best_pred][0];
                     s->motion_val[mot_index][1]= s->mv[0][0][1]= mv_predictor[best_pred][1];
 
-                    MPV_decode_mb(s, s->block);
+                    decode_mb(s);
 
                     
                     if(s->mv[0][0][0] != prev_x || s->mv[0][0][1] != prev_y){
@@ -848,7 +856,7 @@ void ff_er_frame_end(MpegEncContext *s){
 
             s->mb_x= mb_x;
             s->mb_y= mb_y;
-            MPV_decode_mb(s, s->block);
+            decode_mb(s);
         }
     }
 
@@ -888,7 +896,7 @@ void ff_er_frame_end(MpegEncContext *s){
                 s->dsp.clear_blocks(s->block[0]);
                 s->mb_x= mb_x;
                 s->mb_y= mb_y;
-                MPV_decode_mb(s, s->block);
+                decode_mb(s);
             }
         }
     }else
diff --git a/libavcodec/h263.c b/libavcodec/h263.c
index b48c86648e..67321cf3f8 100644
--- a/libavcodec/h263.c
+++ b/libavcodec/h263.c
@@ -512,6 +512,53 @@ int ff_mpeg4_set_direct_mv(MpegEncContext *s, int mx, int my){
     }
 }
 
+void ff_h263_update_motion_val(MpegEncContext * s){
+    const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
+               //FIXME a lot of thet is only needed for !low_delay
+    const int wrap = s->block_wrap[0];
+    const int xy = s->block_index[0];
+    
+    s->current_picture.mbskip_table[mb_xy]= s->mb_skiped; 
+
+    if(s->mv_type != MV_TYPE_8X8){
+        int motion_x, motion_y;
+        if (s->mb_intra) {
+            motion_x = 0;
+            motion_y = 0;
+        } else if (s->mv_type == MV_TYPE_16X16) {
+            motion_x = s->mv[0][0][0];
+            motion_y = s->mv[0][0][1];
+        } else /*if (s->mv_type == MV_TYPE_FIELD)*/ {
+            int i;
+            motion_x = s->mv[0][0][0] + s->mv[0][1][0];
+            motion_y = s->mv[0][0][1] + s->mv[0][1][1];
+            motion_x = (motion_x>>1) | (motion_x&1);
+            for(i=0; i<2; i++){
+                s->field_mv_table[mb_xy][i][0]= s->mv[0][i][0];
+                s->field_mv_table[mb_xy][i][1]= s->mv[0][i][1];
+                s->field_select_table[mb_xy][i]= s->field_select[0][i];
+            }
+        }
+        
+        /* no update if 8X8 because it has been done during parsing */
+        s->motion_val[xy][0] = motion_x;
+        s->motion_val[xy][1] = motion_y;
+        s->motion_val[xy + 1][0] = motion_x;
+        s->motion_val[xy + 1][1] = motion_y;
+        s->motion_val[xy + wrap][0] = motion_x;
+        s->motion_val[xy + wrap][1] = motion_y;
+        s->motion_val[xy + 1 + wrap][0] = motion_x;
+        s->motion_val[xy + 1 + wrap][1] = motion_y;
+    }
+
+    if(s->encoding){ //FIXME encoding MUST be cleaned up
+        if (s->mv_type == MV_TYPE_8X8) 
+            s->current_picture.mb_type[mb_xy]= MB_TYPE_L0 | MB_TYPE_8x8;
+        else
+            s->current_picture.mb_type[mb_xy]= MB_TYPE_L0 | MB_TYPE_16x16;
+    }
+}
+
 #ifdef CONFIG_ENCODERS
 void mpeg4_encode_mb(MpegEncContext * s,
 		    DCTELEM block[6][64],
diff --git a/libavcodec/h263dec.c b/libavcodec/h263dec.c
index 01defcd721..332eb3b3b7 100644
--- a/libavcodec/h263dec.c
+++ b/libavcodec/h263dec.c
@@ -209,6 +209,9 @@ static int decode_slice(MpegEncContext *s){
 //printf("%d %d %06X\n", ret, get_bits_count(&s->gb), show_bits(&s->gb, 24));
             ret= s->decode_mb(s, s->block);
 
+            if (s->pict_type!=B_TYPE)
+                ff_h263_update_motion_val(s);
+
             if(ret<0){
                 const int xy= s->mb_x + s->mb_y*s->mb_stride;
                 if(ret==SLICE_END){
@@ -533,6 +536,9 @@ retry:
         if(s->divx_version && s->divx_version<500){
             s->workaround_bugs|= FF_BUG_EDGE;
         }
+        
+        if(s->avctx->codec_tag == ff_get_fourcc("DIVX") && s->divx_version==0 && s->lavc_build==0 && s->xvid_build==0 && s->vo_type==0 && s->vol_control_parameters==0 && s->low_delay)
+            s->workaround_bugs|= FF_BUG_EDGE;
 
 #if 0
         if(s->divx_version==500)
diff --git a/libavcodec/motion_est.c b/libavcodec/motion_est.c
index 47cb4141c1..505a97d83d 100644
--- a/libavcodec/motion_est.c
+++ b/libavcodec/motion_est.c
@@ -1083,7 +1083,7 @@ void ff_estimate_p_frame_motion(MpegEncContext * s,
     printf("varc=%4d avg_var=%4d (sum=%4d) vard=%4d mx=%2d my=%2d\n",
 	   varc, s->avg_mb_var, sum, vard, mx - xx, my - yy);
 #endif
-    if(s->flags&CODEC_FLAG_HQ){
+    if(s->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
         if (vard <= 64 || vard < varc)
             s->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
         else
@@ -1315,8 +1315,8 @@ static inline int check_bidir_mv(MpegEncContext * s,
         dxy = ((motion_fy & 3) << 2) | (motion_fx & 3);
         src_x = mb_x * 16 + (motion_fx >> 2);
         src_y = mb_y * 16 + (motion_fy >> 2);
-        assert(src_x >=-16 && src_x<=s->width);
-        assert(src_y >=-16 && src_y<=s->height);
+        assert(src_x >=-16 && src_x<=s->h_edge_pos);
+        assert(src_y >=-16 && src_y<=s->v_edge_pos);
 
         ptr = s->last_picture.data[0] + (src_y * s->linesize) + src_x;
         s->dsp.put_qpel_pixels_tab[0][dxy](dest_y    , ptr    , s->linesize);
@@ -1324,8 +1324,8 @@ static inline int check_bidir_mv(MpegEncContext * s,
         dxy = ((motion_by & 3) << 2) | (motion_bx & 3);
         src_x = mb_x * 16 + (motion_bx >> 2);
         src_y = mb_y * 16 + (motion_by >> 2);
-        assert(src_x >=-16 && src_x<=s->width);
-        assert(src_y >=-16 && src_y<=s->height);
+        assert(src_x >=-16 && src_x<=s->h_edge_pos);
+        assert(src_y >=-16 && src_y<=s->v_edge_pos);
     
         ptr = s->next_picture.data[0] + (src_y * s->linesize) + src_x;
         s->dsp.avg_qpel_pixels_tab[0][dxy](dest_y    , ptr    , s->linesize);
@@ -1333,8 +1333,8 @@ static inline int check_bidir_mv(MpegEncContext * s,
         dxy = ((motion_fy & 1) << 1) | (motion_fx & 1);
         src_x = mb_x * 16 + (motion_fx >> 1);
         src_y = mb_y * 16 + (motion_fy >> 1);
-        assert(src_x >=-16 && src_x<=s->width);
-        assert(src_y >=-16 && src_y<=s->height);
+        assert(src_x >=-16 && src_x<=s->h_edge_pos);
+        assert(src_y >=-16 && src_y<=s->v_edge_pos);
 
         ptr = s->last_picture.data[0] + (src_y * s->linesize) + src_x;
         s->dsp.put_pixels_tab[0][dxy](dest_y    , ptr    , s->linesize, 16);
@@ -1342,8 +1342,8 @@ static inline int check_bidir_mv(MpegEncContext * s,
         dxy = ((motion_by & 1) << 1) | (motion_bx & 1);
         src_x = mb_x * 16 + (motion_bx >> 1);
         src_y = mb_y * 16 + (motion_by >> 1);
-        assert(src_x >=-16 && src_x<=s->width);
-        assert(src_y >=-16 && src_y<=s->height);
+        assert(src_x >=-16 && src_x<=s->h_edge_pos);
+        assert(src_y >=-16 && src_y<=s->v_edge_pos);
     
         ptr = s->next_picture.data[0] + (src_y * s->linesize) + src_x;
         s->dsp.avg_pixels_tab[0][dxy](dest_y    , ptr    , s->linesize, 16);
@@ -1523,7 +1523,7 @@ void ff_estimate_b_frame_motion(MpegEncContext * s,
         s->current_picture.mc_mb_var[mb_y*s->mb_stride + mb_x] = score; //FIXME use SSE
     }
 
-    if(s->flags&CODEC_FLAG_HQ){
+    if(s->avctx->mb_decision > FF_MB_DECISION_SIMPLE){
         type= MB_TYPE_FORWARD | MB_TYPE_BACKWARD | MB_TYPE_BIDIR | MB_TYPE_DIRECT; //FIXME something smarter
         if(dmin>256*256*16) type&= ~MB_TYPE_DIRECT; //dont try direct mode if its invalid for this MB
     }
diff --git a/libavcodec/mpeg12.c b/libavcodec/mpeg12.c
index 4f007ca3ca..f15b336d2b 100644
--- a/libavcodec/mpeg12.c
+++ b/libavcodec/mpeg12.c
@@ -1924,6 +1924,7 @@ static int mpeg_decode_slice(AVCodecContext *avctx,
     s->resync_mb_x= s->mb_x;
     s->resync_mb_y= s->mb_y = start_code;
     s->mb_skip_run= 0;
+    ff_init_block_index(s);
 
     for(;;) {
 	s->dsp.clear_blocks(s->block[0]);
@@ -1959,8 +1960,12 @@ static int mpeg_decode_slice(AVCodecContext *avctx,
             s->motion_val[xy + 1 + wrap][1] = motion_y;
         }
         
-        MPV_decode_mb(s, s->block);
+        s->dest[0] += 16;
+        s->dest[1] += 8;
+        s->dest[2] += 8;
 
+        MPV_decode_mb(s, s->block);
+        
         if (++s->mb_x >= s->mb_width) {
             if(s->avctx->codec_tag == ff_get_fourcc("VCR2"))
                 exchange_uv((AVFrame*)s->current_picture_ptr);
@@ -1983,6 +1988,8 @@ static int mpeg_decode_slice(AVCodecContext *avctx,
                 }else
                     goto eos;
             }
+            
+            ff_init_block_index(s);
         }
 
         /* skip mb handling */
diff --git a/libavcodec/mpegvideo.c b/libavcodec/mpegvideo.c
index af6dd611e5..6a3b5c7185 100644
--- a/libavcodec/mpegvideo.c
+++ b/libavcodec/mpegvideo.c
@@ -50,6 +50,7 @@ static void draw_edges_c(uint8_t *buf, int wrap, int width, int height, int w);
 #ifdef CONFIG_ENCODERS
 static int dct_quantize_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
 static int dct_quantize_trellis_c(MpegEncContext *s, DCTELEM *block, int n, int qscale, int *overflow);
+static int sse_mb(MpegEncContext *s);
 #endif //CONFIG_ENCODERS
 
 #ifdef HAVE_XVMC
@@ -2425,50 +2426,6 @@ void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
     else if (s->h263_pred || s->h263_aic)
         s->mbintra_table[mb_xy]=1;
 
-    /* update motion predictor, not for B-frames as they need the motion_val from the last P/S-Frame */
-    if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE) { //FIXME move into h263.c if possible, format specific stuff shouldnt be here
-        //FIXME a lot of thet is only needed for !low_delay
-        const int wrap = s->block_wrap[0];
-        const int xy = s->block_index[0];
-        if(s->mv_type != MV_TYPE_8X8){
-            int motion_x, motion_y;
-            if (s->mb_intra) {
-                motion_x = 0;
-                motion_y = 0;
-            } else if (s->mv_type == MV_TYPE_16X16) {
-                motion_x = s->mv[0][0][0];
-                motion_y = s->mv[0][0][1];
-            } else /*if (s->mv_type == MV_TYPE_FIELD)*/ {
-                int i;
-                motion_x = s->mv[0][0][0] + s->mv[0][1][0];
-                motion_y = s->mv[0][0][1] + s->mv[0][1][1];
-                motion_x = (motion_x>>1) | (motion_x&1);
-                for(i=0; i<2; i++){
-                    s->field_mv_table[mb_xy][i][0]= s->mv[0][i][0];
-                    s->field_mv_table[mb_xy][i][1]= s->mv[0][i][1];
-                    s->field_select_table[mb_xy][i]= s->field_select[0][i];
-                }
-            }
-            
-            /* no update if 8X8 because it has been done during parsing */
-            s->motion_val[xy][0] = motion_x;
-            s->motion_val[xy][1] = motion_y;
-            s->motion_val[xy + 1][0] = motion_x;
-            s->motion_val[xy + 1][1] = motion_y;
-            s->motion_val[xy + wrap][0] = motion_x;
-            s->motion_val[xy + wrap][1] = motion_y;
-            s->motion_val[xy + 1 + wrap][0] = motion_x;
-            s->motion_val[xy + 1 + wrap][1] = motion_y;
-        }
-
-        if(s->encoding){ //FIXME encoding MUST be cleaned up
-            if (s->mv_type == MV_TYPE_8X8) 
-                s->current_picture.mb_type[mb_xy]= MB_TYPE_L0 | MB_TYPE_8x8;
-            else
-                s->current_picture.mb_type[mb_xy]= MB_TYPE_L0 | MB_TYPE_16x16;
-        }
-    }
-    
     if ((s->flags&CODEC_FLAG_PSNR) || !(s->encoding && (s->intra_only || s->pict_type==B_TYPE))) { //FIXME precalc
         uint8_t *dest_y, *dest_cb, *dest_cr;
         int dct_linesize, dct_offset;
@@ -2478,10 +2435,6 @@ void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
         const int uvlinesize= s->current_picture.linesize[1];
 
         /* avoid copy if macroblock skipped in last frame too */
-        if (s->pict_type != B_TYPE) {
-            s->current_picture.mbskip_table[mb_xy]= s->mb_skiped;
-        }
-
         /* skip only during decoding as we might trash the buffers during encoding a bit */
         if(!s->encoding){
             uint8_t *mbskip_ptr = &s->mbskip_table[mb_xy];
@@ -2506,17 +2459,6 @@ void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
             } else{
                 *mbskip_ptr = 0; /* not skipped */
             }
-        }else
-            s->mb_skiped= 0;
-
-        if(s->pict_type==B_TYPE && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME){ //FIXME precalc
-            dest_y  = s->current_picture.data[0] + mb_x * 16;
-            dest_cb = s->current_picture.data[1] + mb_x * 8;
-            dest_cr = s->current_picture.data[2] + mb_x * 8;
-        }else{
-            dest_y  = s->current_picture.data[0] + (mb_y * 16* linesize  ) + mb_x * 16;
-            dest_cb = s->current_picture.data[1] + (mb_y * 8 * uvlinesize) + mb_x * 8;
-            dest_cr = s->current_picture.data[2] + (mb_y * 8 * uvlinesize) + mb_x * 8;
         }
 
         if (s->interlaced_dct) {
@@ -2526,11 +2468,15 @@ void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
             dct_linesize = linesize;
             dct_offset = linesize * 8;
         }
+        
+        dest_y=  s->dest[0];
+        dest_cb= s->dest[1];
+        dest_cr= s->dest[2];
 
         if (!s->mb_intra) {
             /* motion handling */
             /* decoding or more than one mb_type (MC was allready done otherwise) */
-            if((!s->encoding) || (s->mb_type[mb_xy]&(s->mb_type[mb_xy]-1))){
+            if(!s->encoding){
                 if ((!s->no_rounding) || s->pict_type==B_TYPE){                
 		    op_pix = s->dsp.put_pixels_tab;
                     op_qpix= s->dsp.put_qpel_pixels_tab;
@@ -2791,6 +2737,28 @@ void ff_draw_horiz_band(MpegEncContext *s, int y, int h){
     }
 }
 
+void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
+    const int linesize= s->current_picture.linesize[0]; //not s->linesize as this woulnd be wrong for field pics
+    const int uvlinesize= s->current_picture.linesize[1];
+        
+    s->block_index[0]= s->block_wrap[0]*(s->mb_y*2 + 1) - 1 + s->mb_x*2;
+    s->block_index[1]= s->block_wrap[0]*(s->mb_y*2 + 1)     + s->mb_x*2;
+    s->block_index[2]= s->block_wrap[0]*(s->mb_y*2 + 2) - 1 + s->mb_x*2;
+    s->block_index[3]= s->block_wrap[0]*(s->mb_y*2 + 2)     + s->mb_x*2;
+    s->block_index[4]= s->block_wrap[4]*(s->mb_y + 1)                    + s->block_wrap[0]*(s->mb_height*2 + 2) + s->mb_x;
+    s->block_index[5]= s->block_wrap[4]*(s->mb_y + 1 + s->mb_height + 2) + s->block_wrap[0]*(s->mb_height*2 + 2) + s->mb_x;
+    
+    if(s->pict_type==B_TYPE && s->avctx->draw_horiz_band && s->picture_structure==PICT_FRAME){
+        s->dest[0] = s->current_picture.data[0] + s->mb_x * 16 - 16;
+        s->dest[1] = s->current_picture.data[1] + s->mb_x * 8 - 8;
+        s->dest[2] = s->current_picture.data[2] + s->mb_x * 8 - 8;
+    }else{
+        s->dest[0] = s->current_picture.data[0] + (s->mb_y * 16* linesize  ) + s->mb_x * 16 - 16;
+        s->dest[1] = s->current_picture.data[1] + (s->mb_y * 8 * uvlinesize) + s->mb_x * 8 - 8;
+        s->dest[2] = s->current_picture.data[2] + (s->mb_y * 8 * uvlinesize) + s->mb_x * 8 - 8;
+    }    
+}
+
 #ifdef CONFIG_ENCODERS
 
 static void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
@@ -2885,9 +2853,9 @@ static void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
         int wrap_y, wrap_c;
         int emu=0;
 
-        dest_y  = s->current_picture.data[0] + (mb_y * 16 * s->linesize    ) + mb_x * 16;
-        dest_cb = s->current_picture.data[1] + (mb_y * 8  * (s->uvlinesize)) + mb_x * 8;
-        dest_cr = s->current_picture.data[2] + (mb_y * 8  * (s->uvlinesize)) + mb_x * 8;
+        dest_y  = s->dest[0];
+        dest_cb = s->dest[1];
+        dest_cr = s->dest[2];
         wrap_y = s->linesize;
         wrap_c = s->uvlinesize;
         ptr_y  = s->new_picture.data[0] + (mb_y * 16 * wrap_y) + mb_x * 16;
@@ -3161,7 +3129,7 @@ static inline void copy_context_before_encode(MpegEncContext *d, MpegEncContext
     d->misc_bits= s->misc_bits;
     d->last_bits= 0;
 
-    d->mb_skiped= s->mb_skiped;
+    d->mb_skiped= 0;
     d->qscale= s->qscale;
 }
 
@@ -3206,7 +3174,8 @@ static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegE
                            PutBitContext pb[2], PutBitContext pb2[2], PutBitContext tex_pb[2],
                            int *dmin, int *next_block, int motion_x, int motion_y)
 {
-    int bits_count;
+    int score;
+    uint8_t *dest_backup[3];
     
     copy_context_before_encode(s, backup, type);
 
@@ -3216,24 +3185,44 @@ static inline void encode_mb_hq(MpegEncContext *s, MpegEncContext *backup, MpegE
         s->pb2   = pb2   [*next_block];
         s->tex_pb= tex_pb[*next_block];
     }
+    
+    if(*next_block){
+        memcpy(dest_backup, s->dest, sizeof(s->dest));
+        s->dest[0] = s->me.scratchpad;
+        s->dest[1] = s->me.scratchpad + 16;
+        s->dest[2] = s->me.scratchpad + 16 + 8;
+        assert(2*s->uvlinesize == s->linesize); //should be no prob for encoding
+        assert(s->linesize >= 64); //FIXME
+    }
 
     encode_mb(s, motion_x, motion_y);
-
-    bits_count= get_bit_count(&s->pb);
+    
+    score= get_bit_count(&s->pb);
     if(s->data_partitioning){
-        bits_count+= get_bit_count(&s->pb2);
-        bits_count+= get_bit_count(&s->tex_pb);
+        score+= get_bit_count(&s->pb2);
+        score+= get_bit_count(&s->tex_pb);
     }
+   
+    if(s->avctx->mb_decision == FF_MB_DECISION_RD){
+        MPV_decode_mb(s, s->block);
 
-    if(bits_count<*dmin){
-        *dmin= bits_count;
+        score *= s->qscale * s->qscale * 109;
+        score += sse_mb(s) << 7;
+    }
+    
+    if(*next_block){
+        memcpy(s->dest, dest_backup, sizeof(s->dest));
+    }
+
+    if(score<*dmin){
+        *dmin= score;
         *next_block^=1;
 
         copy_context_after_encode(best, s, type);
     }
 }
                 
-static inline int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
+static int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, int h, int stride){
     uint32_t *sq = squareTbl + 256;
     int acc=0;
     int x,y;
@@ -3254,6 +3243,23 @@ static inline int sse(MpegEncContext *s, uint8_t *src1, uint8_t *src2, int w, in
     return acc;
 }
 
+static int sse_mb(MpegEncContext *s){
+    int w= 16;
+    int h= 16;
+
+    if(s->mb_x*16 + 16 > s->width ) w= s->width - s->mb_x*16;
+    if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
+
+    if(w==16 && h==16)
+        return  s->dsp.sse[0](NULL, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], s->linesize)
+               +s->dsp.sse[1](NULL, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], s->uvlinesize)
+               +s->dsp.sse[1](NULL, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], s->uvlinesize);
+    else
+        return  sse(s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16, s->dest[0], w, h, s->linesize)
+               +sse(s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[1], w>>1, h>>1, s->uvlinesize)
+               +sse(s, s->new_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,s->dest[2], w>>1, h>>1, s->uvlinesize);
+}
+
 static void encode_picture(MpegEncContext *s, int picture_number)
 {
     int mb_x, mb_y, pdif = 0;
@@ -3508,29 +3514,21 @@ static void encode_picture(MpegEncContext *s, int picture_number)
     s->first_slice_line = 1;
     s->ptr_lastgob = s->pb.buf;
     for(mb_y=0; mb_y < s->mb_height; mb_y++) {
+        s->mb_x=0;
+        s->mb_y= mb_y;
+
         s->y_dc_scale= s->y_dc_scale_table[ s->qscale ];
         s->c_dc_scale= s->c_dc_scale_table[ s->qscale ];
+        ff_init_block_index(s);
         
-        s->block_index[0]= s->block_wrap[0]*(mb_y*2 + 1) - 1;
-        s->block_index[1]= s->block_wrap[0]*(mb_y*2 + 1);
-        s->block_index[2]= s->block_wrap[0]*(mb_y*2 + 2) - 1;
-        s->block_index[3]= s->block_wrap[0]*(mb_y*2 + 2);
-        s->block_index[4]= s->block_wrap[4]*(mb_y + 1)                    + s->block_wrap[0]*(s->mb_height*2 + 2);
-        s->block_index[5]= s->block_wrap[4]*(mb_y + 1 + s->mb_height + 2) + s->block_wrap[0]*(s->mb_height*2 + 2);
         for(mb_x=0; mb_x < s->mb_width; mb_x++) {
             const int xy= mb_y*s->mb_stride + mb_x;
             int mb_type= s->mb_type[xy];
 //            int d;
-            int dmin=10000000;
+            int dmin= INT_MAX;
 
             s->mb_x = mb_x;
-            s->mb_y = mb_y;
-            s->block_index[0]+=2;
-            s->block_index[1]+=2;
-            s->block_index[2]+=2;
-            s->block_index[3]+=2;
-            s->block_index[4]++;
-            s->block_index[5]++;
+            ff_update_block_index(s);
 
             /* write gob / video packet header  */
 #ifdef CONFIG_RISKY
@@ -3588,6 +3586,8 @@ static void encode_picture(MpegEncContext *s, int picture_number)
                 s->first_slice_line=0; 
             }
 
+            s->mb_skiped=0;
+
             if(mb_type & (mb_type-1)){ // more than 1 MB type possible
                 int next_block=0;
                 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
@@ -3670,9 +3670,12 @@ static void encode_picture(MpegEncContext *s, int picture_number)
                     s->mv[0][0][1] = 0;
                     encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTRA, pb, pb2, tex_pb, 
                                  &dmin, &next_block, 0, 0);
-                    /* force cleaning of ac/dc pred stuff if needed ... */
-                    if(s->h263_pred || s->h263_aic)
-                        s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
+                    if(s->h263_pred || s->h263_aic){
+                        if(best_s.mb_intra)
+                            s->mbintra_table[mb_x + mb_y*s->mb_stride]=1;
+                        else
+                            ff_clean_intra_table_entries(s); //old mode?
+                    }
                 }
                 copy_context_after_encode(s, &best_s, -1);
                 
@@ -3693,12 +3696,24 @@ static void encode_picture(MpegEncContext *s, int picture_number)
                     s->tex_pb= backup_s.tex_pb;
                 }
                 s->last_bits= get_bit_count(&s->pb);
+                
+                if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE)
+                    ff_h263_update_motion_val(s);
+        
+                if(next_block==0){
+                    s->dsp.put_pixels_tab[0][0](s->dest[0], s->me.scratchpad     , s->linesize  ,16);
+                    s->dsp.put_pixels_tab[1][0](s->dest[1], s->me.scratchpad + 16, s->uvlinesize, 8);
+                    s->dsp.put_pixels_tab[1][0](s->dest[2], s->me.scratchpad + 24, s->uvlinesize, 8);
+                }
+
+                if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
+                    MPV_decode_mb(s, s->block);
             } else {
                 int motion_x, motion_y;
                 int intra_score;
                 int inter_score= s->current_picture.mb_cmp_score[mb_x + mb_y*s->mb_stride];
                 
-              if(!(s->flags&CODEC_FLAG_HQ) && s->pict_type==P_TYPE){
+              if(s->avctx->mb_decision==FF_MB_DECISION_SIMPLE && s->pict_type==P_TYPE){ //FIXME check if the mess is usefull at all
                 /* get luma score */
                 if((s->avctx->mb_cmp&0xFF)==FF_CMP_SSE){
                     intra_score= (s->current_picture.mb_var[mb_x + mb_y*s->mb_stride]<<8) - 500; //FIXME dont scale it down so we dont have to fix it
@@ -3846,6 +3861,11 @@ static void encode_picture(MpegEncContext *s, int picture_number)
 
                 // RAL: Update last macrobloc type
                 s->last_mv_dir = s->mv_dir;
+            
+                if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE)
+                    ff_h263_update_motion_val(s);
+
+                MPV_decode_mb(s, s->block);
             }
 
             /* clean the MV table in IPS frames for direct mode in B frames */
@@ -3853,8 +3873,6 @@ static void encode_picture(MpegEncContext *s, int picture_number)
                 s->p_mv_table[xy][0]=0;
                 s->p_mv_table[xy][1]=0;
             }
-
-            MPV_decode_mb(s, s->block);
             
             if(s->flags&CODEC_FLAG_PSNR){
                 int w= 16;
@@ -3864,20 +3882,14 @@ static void encode_picture(MpegEncContext *s, int picture_number)
                 if(s->mb_y*16 + 16 > s->height) h= s->height- s->mb_y*16;
 
                 s->current_picture_ptr->error[0] += sse(
-                    s,
-                    s->new_picture    .data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
-                    s->current_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
-                    w, h, s->linesize);
+                    s, s->new_picture.data[0] + s->mb_x*16 + s->mb_y*s->linesize*16,
+                    s->dest[0], w, h, s->linesize);
                 s->current_picture_ptr->error[1] += sse(
-                    s,
-                    s->new_picture    .data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,
-                    s->current_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,
-                    w>>1, h>>1, s->uvlinesize);
+                    s, s->new_picture.data[1] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,
+                    s->dest[1], w>>1, h>>1, s->uvlinesize);
                 s->current_picture_ptr->error[2] += sse(
-                    s,
-                    s->new_picture    .data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,
-                    s->current_picture.data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,
-                    w>>1, h>>1, s->uvlinesize);
+                    s, s->new_picture    .data[2] + s->mb_x*8  + s->mb_y*s->uvlinesize*8,
+                    s->dest[2], w>>1, h>>1, s->uvlinesize);
             }
 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, get_bit_count(&s->pb));
         }
@@ -4402,7 +4414,6 @@ static void dct_unquantize_h263_c(MpegEncContext *s,
 static const AVOption mpeg4_options[] =
 {
     AVOPTION_CODEC_INT("bitrate", "desired video bitrate", bit_rate, 4, 240000000, 800000),
-    AVOPTION_CODEC_FLAG("vhq", "very high quality", flags, CODEC_FLAG_HQ, 0),
     AVOPTION_CODEC_INT("ratetol", "number of bits the bitstream is allowed to diverge from the reference"
 		       "the reference can be CBR (for CBR pass1) or VBR (for pass2)",
 		       bit_rate_tolerance, 4, 240000000, 8000),
diff --git a/libavcodec/mpegvideo.h b/libavcodec/mpegvideo.h
index 3546d85a37..a0f5d4c0b6 100644
--- a/libavcodec/mpegvideo.h
+++ b/libavcodec/mpegvideo.h
@@ -430,6 +430,7 @@ typedef struct MpegEncContext {
 
     int block_index[6]; ///< index to current MB in block based arrays with edges
     int block_wrap[6];
+    uint8_t *dest[3];
     
     int *mb_index2xy;        ///< mb_index -> mb_x + mb_y*mb_stride
 
@@ -723,14 +724,7 @@ void ff_er_add_slice(MpegEncContext *s, int startx, int starty, int endx, int en
 
 extern enum PixelFormat ff_yuv420p_list[2];
 
-static inline void ff_init_block_index(MpegEncContext *s){
-    s->block_index[0]= s->block_wrap[0]*(s->mb_y*2 + 1) - 1 + s->mb_x*2;
-    s->block_index[1]= s->block_wrap[0]*(s->mb_y*2 + 1)     + s->mb_x*2;
-    s->block_index[2]= s->block_wrap[0]*(s->mb_y*2 + 2) - 1 + s->mb_x*2;
-    s->block_index[3]= s->block_wrap[0]*(s->mb_y*2 + 2)     + s->mb_x*2;
-    s->block_index[4]= s->block_wrap[4]*(s->mb_y + 1)                    + s->block_wrap[0]*(s->mb_height*2 + 2) + s->mb_x;
-    s->block_index[5]= s->block_wrap[4]*(s->mb_y + 1 + s->mb_height + 2) + s->block_wrap[0]*(s->mb_height*2 + 2) + s->mb_x;
-}
+void ff_init_block_index(MpegEncContext *s);
 
 static inline void ff_update_block_index(MpegEncContext *s){
     s->block_index[0]+=2;
@@ -739,6 +733,9 @@ static inline void ff_update_block_index(MpegEncContext *s){
     s->block_index[3]+=2;
     s->block_index[4]++;
     s->block_index[5]++;
+    s->dest[0]+= 16;
+    s->dest[1]+= 8;
+    s->dest[2]+= 8;
 }
 
 static inline int get_bits_diff(MpegEncContext *s){
@@ -833,6 +830,7 @@ void h263_decode_init_vlc(MpegEncContext *s);
 int h263_decode_picture_header(MpegEncContext *s);
 int ff_h263_decode_gob_header(MpegEncContext *s);
 int ff_mpeg4_decode_picture_header(MpegEncContext * s, GetBitContext *gb);
+void ff_h263_update_motion_val(MpegEncContext * s);
 
 
 int intel_h263_decode_picture_header(MpegEncContext *s);
diff --git a/libavcodec/rv10.c b/libavcodec/rv10.c
index 4b351ea1ea..c19c36ec5e 100644
--- a/libavcodec/rv10.c
+++ b/libavcodec/rv10.c
@@ -466,6 +466,7 @@ static int rv10_decode_packet(AVCodecContext *avctx,
             fprintf(stderr, "ERROR at MB %d %d\n", s->mb_x, s->mb_y);
             return -1;
         }
+        ff_h263_update_motion_val(s);
         MPV_decode_mb(s, s->block);
         if (++s->mb_x == s->mb_width) {
             s->mb_x = 0;
diff --git a/libavcodec/utils.c b/libavcodec/utils.c
index eeffde2c9e..47831f2f35 100644
--- a/libavcodec/utils.c
+++ b/libavcodec/utils.c
@@ -489,7 +489,7 @@ void avcodec_string(char *buf, int buf_size, AVCodecContext *enc, int encode)
     case CODEC_TYPE_VIDEO:
         snprintf(buf, buf_size,
                  "Video: %s%s",
-                 codec_name, enc->flags & CODEC_FLAG_HQ ? " (hq)" : "");
+                 codec_name, enc->mb_decision ? " (hq)" : "");
         if (enc->codec_id == CODEC_ID_RAWVIDEO) {
             snprintf(buf + strlen(buf), buf_size - strlen(buf),
                      ", %s",