h264: switch to h2645_parse for NAL parsing
[ffmpeg.git] / libavcodec / error_resilience.c
index efd1dd5..2c7836f 100644 (file)
  *
  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
  *
- * This library is free software; you can redistribute it and/or
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
  * License as published by the Free Software Foundation; either
- * version 2 of the License, or (at your option) any later version.
+ * version 2.1 of the License, or (at your option) any later version.
  *
- * This library is distributed in the hope that it will be useful,
+ * Libav is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  * Lesser General Public License for more details.
  *
  * You should have received a copy of the GNU Lesser General Public
- * License along with this library; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
+
 /**
- * @file error_resilience.c
+ * @file
  * Error resilience / concealment.
  */
 
 #include <limits.h>
+
+#include "libavutil/internal.h"
 #include "avcodec.h"
-#include "dsputil.h"
+#include "error_resilience.h"
+#include "me_cmp.h"
+#include "mpegutils.h"
 #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;
+#include "rectangle.h"
+#include "thread.h"
+#include "version.h"
 
-    MPV_decode_mb(s, s->block);    
+/**
+ * @param stride the number of MVs to get to the next row
+ * @param mv_step the number of MVs per row or column in a macroblock
+ */
+static void set_mv_strides(ERContext *s, int *mv_step, int *stride)
+{
+    if (s->avctx->codec_id == AV_CODEC_ID_H264) {
+        assert(s->quarter_sample);
+        *mv_step = 4;
+        *stride  = s->mb_width * 4;
+    } else {
+        *mv_step = 2;
+        *stride  = s->b8_stride;
+    }
 }
 
 /**
- * replaces the current MB with a flat dc only version.
+ * Replace the current MB with a flat dc-only version.
  */
-static void put_dc(MpegEncContext *s, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr, int mb_x, int mb_y)
+static void put_dc(ERContext *s, uint8_t *dest_y, uint8_t *dest_cb,
+                   uint8_t *dest_cr, int mb_x, int mb_y)
 {
+    int *linesize = s->cur_pic.f->linesize;
     int dc, dcu, dcv, y, i;
-    for(i=0; i<4; i++){
-        dc= s->dc_val[0][mb_x*2 + (i&1) + (mb_y*2 + (i>>1))*s->b8_stride];
-        if(dc<0) dc=0;
-        else if(dc>2040) dc=2040;
-        for(y=0; y<8; y++){
+    for (i = 0; i < 4; i++) {
+        dc = s->dc_val[0][mb_x * 2 + (i &  1) + (mb_y * 2 + (i >> 1)) * s->b8_stride];
+        if (dc < 0)
+            dc = 0;
+        else if (dc > 2040)
+            dc = 2040;
+        for (y = 0; y < 8; y++) {
             int x;
-            for(x=0; x<8; x++){
-                dest_y[x + (i&1)*8 + (y + (i>>1)*8)*s->linesize]= dc/8;
-            }
+            for (x = 0; x < 8; x++)
+                dest_y[x + (i &  1) * 8 + (y + (i >> 1) * 8) * linesize[0]] = dc / 8;
         }
     }
-    dcu = s->dc_val[1][mb_x + mb_y*s->mb_stride];
-    dcv = s->dc_val[2][mb_x + mb_y*s->mb_stride];
-    if     (dcu<0   ) dcu=0;
-    else if(dcu>2040) dcu=2040;
-    if     (dcv<0   ) dcv=0;
-    else if(dcv>2040) dcv=2040;
-    for(y=0; y<8; y++){
+    dcu = s->dc_val[1][mb_x + mb_y * s->mb_stride];
+    dcv = s->dc_val[2][mb_x + mb_y * s->mb_stride];
+    if (dcu < 0)
+        dcu = 0;
+    else if (dcu > 2040)
+        dcu = 2040;
+    if (dcv < 0)
+        dcv = 0;
+    else if (dcv > 2040)
+        dcv = 2040;
+    for (y = 0; y < 8; y++) {
         int x;
-        for(x=0; x<8; x++){
-            dest_cb[x + y*(s->uvlinesize)]= dcu/8;
-            dest_cr[x + y*(s->uvlinesize)]= dcv/8;
+        for (x = 0; x < 8; x++) {
+            dest_cb[x + y * linesize[1]] = dcu / 8;
+            dest_cr[x + y * linesize[2]] = dcv / 8;
         }
     }
 }
 
-static void filter181(int16_t *data, int width, int height, int stride){
-    int x,y;
+static void filter181(int16_t *data, int width, int height, int stride)
+{
+    int x, y;
 
     /* horizontal filter */
-    for(y=1; y<height-1; y++){
-        int prev_dc= data[0 + y*stride];
+    for (y = 1; y < height - 1; y++) {
+        int prev_dc = data[0 + y * stride];
 
-        for(x=1; x<width-1; x++){
+        for (x = 1; x < width - 1; x++) {
             int dc;
-            
-            dc= - prev_dc 
-                + data[x     + y*stride]*8
-                - data[x + 1 + y*stride];
-            dc= (dc*10923 + 32768)>>16;
-            prev_dc= data[x + y*stride];
-            data[x + y*stride]= dc;
+            dc = -prev_dc +
+                 data[x     + y * stride] * 8 -
+                 data[x + 1 + y * stride];
+            dc = (dc * 10923 + 32768) >> 16;
+            prev_dc = data[x + y * stride];
+            data[x + y * stride] = dc;
         }
     }
-    
+
     /* vertical filter */
-    for(x=1; x<width-1; x++){
-        int prev_dc= data[x];
+    for (x = 1; x < width - 1; x++) {
+        int prev_dc = data[x];
 
-        for(y=1; y<height-1; y++){
+        for (y = 1; y < height - 1; y++) {
             int dc;
-            
-            dc= - prev_dc 
-                + data[x +  y   *stride]*8
-                - data[x + (y+1)*stride];
-            dc= (dc*10923 + 32768)>>16;
-            prev_dc= data[x + y*stride];
-            data[x + y*stride]= dc;
+
+            dc = -prev_dc +
+                 data[x +  y      * stride] * 8 -
+                 data[x + (y + 1) * stride];
+            dc = (dc * 10923 + 32768) >> 16;
+            prev_dc = data[x + y * stride];
+            data[x + y * stride] = dc;
         }
     }
 }
 
 /**
- * guess the dc of blocks which dont have a undamaged dc
- * @param w    width in 8 pixel blocks
- * @param h    height in 8 pixel blocks
+ * guess the dc of blocks which do not have an undamaged dc
+ * @param w     width in 8 pixel blocks
+ * @param h     height in 8 pixel blocks
  */
-static void guess_dc(MpegEncContext *s, int16_t *dc, int w, int h, int stride, int is_luma){
+static void guess_dc(ERContext *s, int16_t *dc, int w,
+                     int h, int stride, int is_luma)
+{
     int b_x, b_y;
 
-    for(b_y=0; b_y<h; b_y++){
-        for(b_x=0; b_x<w; b_x++){
-            int color[4]={1024,1024,1024,1024};
-            int distance[4]={9999,9999,9999,9999};
+    for (b_y = 0; b_y < h; b_y++) {
+        for (b_x = 0; b_x < w; b_x++) {
+            int color[4]    = { 1024, 1024, 1024, 1024 };
+            int distance[4] = { 9999, 9999, 9999, 9999 };
             int mb_index, error, j;
             int64_t guess, weight_sum;
-            
-            mb_index= (b_x>>is_luma) + (b_y>>is_luma)*s->mb_stride;
-            
-            error= s->error_status_table[mb_index];
-            
-            if(IS_INTER(s->current_picture.mb_type[mb_index])) continue; //inter
-            if(!(error&DC_ERROR)) continue;           //dc-ok
-            
+            mb_index = (b_x >> is_luma) + (b_y >> is_luma) * s->mb_stride;
+            error    = s->error_status_table[mb_index];
+
+            if (IS_INTER(s->cur_pic.mb_type[mb_index]))
+                continue; // inter
+            if (!(error & ER_DC_ERROR))
+                continue; // dc-ok
+
             /* right block */
-            for(j=b_x+1; j<w; j++){
-                int mb_index_j= (j>>is_luma) + (b_y>>is_luma)*s->mb_stride;
-                int error_j= s->error_status_table[mb_index_j];
-                int intra_j= IS_INTRA(s->current_picture.mb_type[mb_index_j]);
-                if(intra_j==0 || !(error_j&DC_ERROR)){
-                    color[0]= dc[j + b_y*stride];
-                    distance[0]= j-b_x;
+            for (j = b_x + 1; j < w; j++) {
+                int mb_index_j = (j >> is_luma) + (b_y >> is_luma) * s->mb_stride;
+                int error_j    = s->error_status_table[mb_index_j];
+                int intra_j    = IS_INTRA(s->cur_pic.mb_type[mb_index_j]);
+                if (intra_j == 0 || !(error_j & ER_DC_ERROR)) {
+                    color[0]    = dc[j + b_y * stride];
+                    distance[0] = j - b_x;
                     break;
                 }
             }
-            
+
             /* left block */
-            for(j=b_x-1; j>=0; j--){
-                int mb_index_j= (j>>is_luma) + (b_y>>is_luma)*s->mb_stride;
-                int error_j= s->error_status_table[mb_index_j];
-                int intra_j= IS_INTRA(s->current_picture.mb_type[mb_index_j]);
-                if(intra_j==0 || !(error_j&DC_ERROR)){
-                    color[1]= dc[j + b_y*stride];
-                    distance[1]= b_x-j;
+            for (j = b_x - 1; j >= 0; j--) {
+                int mb_index_j = (j >> is_luma) + (b_y >> is_luma) * s->mb_stride;
+                int error_j    = s->error_status_table[mb_index_j];
+                int intra_j    = IS_INTRA(s->cur_pic.mb_type[mb_index_j]);
+                if (intra_j == 0 || !(error_j & ER_DC_ERROR)) {
+                    color[1]    = dc[j + b_y * stride];
+                    distance[1] = b_x - j;
                     break;
                 }
             }
 
             /* bottom block */
-            for(j=b_y+1; j<h; j++){
-                int mb_index_j= (b_x>>is_luma) + (j>>is_luma)*s->mb_stride;
-                int error_j= s->error_status_table[mb_index_j];
-                int intra_j= IS_INTRA(s->current_picture.mb_type[mb_index_j]);
-                if(intra_j==0 || !(error_j&DC_ERROR)){
-                    color[2]= dc[b_x + j*stride];
-                    distance[2]= j-b_y;
+            for (j = b_y + 1; j < h; j++) {
+                int mb_index_j = (b_x >> is_luma) + (j >> is_luma) * s->mb_stride;
+                int error_j    = s->error_status_table[mb_index_j];
+                int intra_j    = IS_INTRA(s->cur_pic.mb_type[mb_index_j]);
+
+                if (intra_j == 0 || !(error_j & ER_DC_ERROR)) {
+                    color[2]    = dc[b_x + j * stride];
+                    distance[2] = j - b_y;
                     break;
                 }
             }
 
             /* top block */
-            for(j=b_y-1; j>=0; j--){
-                int mb_index_j= (b_x>>is_luma) + (j>>is_luma)*s->mb_stride;
-                int error_j= s->error_status_table[mb_index_j];
-                int intra_j= IS_INTRA(s->current_picture.mb_type[mb_index_j]);
-                if(intra_j==0 || !(error_j&DC_ERROR)){
-                    color[3]= dc[b_x + j*stride];
-                    distance[3]= b_y-j;
+            for (j = b_y - 1; j >= 0; j--) {
+                int mb_index_j = (b_x >> is_luma) + (j >> is_luma) * s->mb_stride;
+                int error_j    = s->error_status_table[mb_index_j];
+                int intra_j    = IS_INTRA(s->cur_pic.mb_type[mb_index_j]);
+                if (intra_j == 0 || !(error_j & ER_DC_ERROR)) {
+                    color[3]    = dc[b_x + j * stride];
+                    distance[3] = b_y - j;
                     break;
                 }
             }
-            
-            weight_sum=0;
-            guess=0;
-            for(j=0; j<4; j++){
-                int64_t weight= 256*256*256*16/distance[j];
-                guess+= weight*(int64_t)color[j];
-                weight_sum+= weight;
-            }
-            guess= (guess + weight_sum/2) / weight_sum;
 
-            dc[b_x + b_y*stride]= guess;
+            weight_sum = 0;
+            guess      = 0;
+            for (j = 0; j < 4; j++) {
+                int64_t weight  = 256 * 256 * 256 * 16 / distance[j];
+                guess          += weight * (int64_t) color[j];
+                weight_sum     += weight;
+            }
+            guess = (guess + weight_sum / 2) / weight_sum;
+            dc[b_x + b_y * stride] = guess;
         }
     }
 }
 
 /**
  * simple horizontal deblocking filter used for error resilience
- * @param w    width in 8 pixel blocks
- * @param h    height in 8 pixel blocks
+ * @param w     width in 8 pixel blocks
+ * @param h     height in 8 pixel blocks
  */
-static void h_block_filter(MpegEncContext *s, uint8_t *dst, int w, int h, int stride, int is_luma){
-    int b_x, b_y;
-    uint8_t *cm = cropTbl + MAX_NEG_CROP;
-
-    for(b_y=0; b_y<h; b_y++){
-        for(b_x=0; b_x<w-1; b_x++){
+static void h_block_filter(ERContext *s, uint8_t *dst, int w,
+                           int h, int stride, int is_luma)
+{
+    int b_x, b_y, mvx_stride, mvy_stride;
+    const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
+    set_mv_strides(s, &mvx_stride, &mvy_stride);
+    mvx_stride >>= is_luma;
+    mvy_stride *= mvx_stride;
+
+    for (b_y = 0; b_y < h; b_y++) {
+        for (b_x = 0; b_x < w - 1; b_x++) {
             int y;
-            int left_status = s->error_status_table[( b_x   >>is_luma) + (b_y>>is_luma)*s->mb_stride];
-            int right_status= s->error_status_table[((b_x+1)>>is_luma) + (b_y>>is_luma)*s->mb_stride];
-            int left_intra=   IS_INTRA(s->current_picture.mb_type      [( b_x   >>is_luma) + (b_y>>is_luma)*s->mb_stride]);
-            int right_intra=  IS_INTRA(s->current_picture.mb_type      [((b_x+1)>>is_luma) + (b_y>>is_luma)*s->mb_stride]);
-            int left_damage =  left_status&(DC_ERROR|AC_ERROR|MV_ERROR);
-            int right_damage= right_status&(DC_ERROR|AC_ERROR|MV_ERROR);
-            int offset= b_x*8 + b_y*stride*8;
-            int16_t *left_mv=  s->current_picture.motion_val[0][s->b8_stride*(b_y<<(1-is_luma)) + ( b_x   <<(1-is_luma))];
-            int16_t *right_mv= s->current_picture.motion_val[0][s->b8_stride*(b_y<<(1-is_luma)) + ((b_x+1)<<(1-is_luma))];
-            
-            if(!(left_damage||right_damage)) continue; // both undamaged
-            
-            if(   (!left_intra) && (!right_intra) 
-               && ABS(left_mv[0]-right_mv[0]) + ABS(left_mv[1]+right_mv[1]) < 2) continue;
-            
-            for(y=0; y<8; y++){
-                int a,b,c,d;
-                
-                a= dst[offset + 7 + y*stride] - dst[offset + 6 + y*stride];
-                b= dst[offset + 8 + y*stride] - dst[offset + 7 + y*stride];
-                c= dst[offset + 9 + y*stride] - dst[offset + 8 + y*stride];
-                
-                d= ABS(b) - ((ABS(a) + ABS(c) + 1)>>1);
-                d= FFMAX(d, 0);
-                if(b<0) d= -d;
-                
-                if(d==0) continue;
-
-                if(!(left_damage && right_damage))
-                    d= d*16/9;
-                
-                if(left_damage){
-                    dst[offset + 7 + y*stride] = cm[dst[offset + 7 + y*stride] + ((d*7)>>4)];
-                    dst[offset + 6 + y*stride] = cm[dst[offset + 6 + y*stride] + ((d*5)>>4)];
-                    dst[offset + 5 + y*stride] = cm[dst[offset + 5 + y*stride] + ((d*3)>>4)];
-                    dst[offset + 4 + y*stride] = cm[dst[offset + 4 + y*stride] + ((d*1)>>4)];
+            int left_status  = s->error_status_table[( b_x      >> is_luma) + (b_y >> is_luma) * s->mb_stride];
+            int right_status = s->error_status_table[((b_x + 1) >> is_luma) + (b_y >> is_luma) * s->mb_stride];
+            int left_intra   = IS_INTRA(s->cur_pic.mb_type[( b_x      >> is_luma) + (b_y >> is_luma) * s->mb_stride]);
+            int right_intra  = IS_INTRA(s->cur_pic.mb_type[((b_x + 1) >> is_luma) + (b_y >> is_luma) * s->mb_stride]);
+            int left_damage  = left_status & ER_MB_ERROR;
+            int right_damage = right_status & ER_MB_ERROR;
+            int offset       = b_x * 8 + b_y * stride * 8;
+            int16_t *left_mv  = s->cur_pic.motion_val[0][mvy_stride * b_y + mvx_stride *  b_x];
+            int16_t *right_mv = s->cur_pic.motion_val[0][mvy_stride * b_y + mvx_stride * (b_x + 1)];
+            if (!(left_damage || right_damage))
+                continue; // both undamaged
+            if ((!left_intra) && (!right_intra) &&
+                FFABS(left_mv[0] - right_mv[0]) +
+                FFABS(left_mv[1] + right_mv[1]) < 2)
+                continue;
+
+            for (y = 0; y < 8; y++) {
+                int a, b, c, d;
+
+                a = dst[offset + 7 + y * stride] - dst[offset + 6 + y * stride];
+                b = dst[offset + 8 + y * stride] - dst[offset + 7 + y * stride];
+                c = dst[offset + 9 + y * stride] - dst[offset + 8 + y * stride];
+
+                d = FFABS(b) - ((FFABS(a) + FFABS(c) + 1) >> 1);
+                d = FFMAX(d, 0);
+                if (b < 0)
+                    d = -d;
+
+                if (d == 0)
+                    continue;
+
+                if (!(left_damage && right_damage))
+                    d = d * 16 / 9;
+
+                if (left_damage) {
+                    dst[offset + 7 + y * stride] = cm[dst[offset + 7 + y * stride] + ((d * 7) >> 4)];
+                    dst[offset + 6 + y * stride] = cm[dst[offset + 6 + y * stride] + ((d * 5) >> 4)];
+                    dst[offset + 5 + y * stride] = cm[dst[offset + 5 + y * stride] + ((d * 3) >> 4)];
+                    dst[offset + 4 + y * stride] = cm[dst[offset + 4 + y * stride] + ((d * 1) >> 4)];
                 }
-                if(right_damage){
-                    dst[offset + 8 + y*stride] = cm[dst[offset + 8 + y*stride] - ((d*7)>>4)];
-                    dst[offset + 9 + y*stride] = cm[dst[offset + 9 + y*stride] - ((d*5)>>4)];
-                    dst[offset + 10+ y*stride] = cm[dst[offset +10 + y*stride] - ((d*3)>>4)];
-                    dst[offset + 11+ y*stride] = cm[dst[offset +11 + y*stride] - ((d*1)>>4)];
+                if (right_damage) {
+                    dst[offset + 8 + y * stride] = cm[dst[offset +  8 + y * stride] - ((d * 7) >> 4)];
+                    dst[offset + 9 + y * stride] = cm[dst[offset +  9 + y * stride] - ((d * 5) >> 4)];
+                    dst[offset + 10+ y * stride] = cm[dst[offset + 10 + y * stride] - ((d * 3) >> 4)];
+                    dst[offset + 11+ y * stride] = cm[dst[offset + 11 + y * stride] - ((d * 1) >> 4)];
                 }
             }
         }
@@ -252,777 +284,936 @@ static void h_block_filter(MpegEncContext *s, uint8_t *dst, int w, int h, int st
 
 /**
  * simple vertical deblocking filter used for error resilience
- * @param w    width in 8 pixel blocks
- * @param h    height in 8 pixel blocks
+ * @param w     width in 8 pixel blocks
+ * @param h     height in 8 pixel blocks
  */
-static void v_block_filter(MpegEncContext *s, uint8_t *dst, int w, int h, int stride, int is_luma){
-    int b_x, b_y;
-    uint8_t *cm = cropTbl + MAX_NEG_CROP;
-
-    for(b_y=0; b_y<h-1; b_y++){
-        for(b_x=0; b_x<w; b_x++){
+static void v_block_filter(ERContext *s, uint8_t *dst, int w, int h,
+                           int stride, int is_luma)
+{
+    int b_x, b_y, mvx_stride, mvy_stride;
+    const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
+    set_mv_strides(s, &mvx_stride, &mvy_stride);
+    mvx_stride >>= is_luma;
+    mvy_stride *= mvx_stride;
+
+    for (b_y = 0; b_y < h - 1; b_y++) {
+        for (b_x = 0; b_x < w; b_x++) {
             int x;
-            int top_status   = s->error_status_table[(b_x>>is_luma) + ( b_y   >>is_luma)*s->mb_stride];
-            int bottom_status= s->error_status_table[(b_x>>is_luma) + ((b_y+1)>>is_luma)*s->mb_stride];
-            int top_intra=     IS_INTRA(s->current_picture.mb_type      [(b_x>>is_luma) + ( b_y   >>is_luma)*s->mb_stride]);
-            int bottom_intra=  IS_INTRA(s->current_picture.mb_type      [(b_x>>is_luma) + ((b_y+1)>>is_luma)*s->mb_stride]);
-            int top_damage =      top_status&(DC_ERROR|AC_ERROR|MV_ERROR);
-            int bottom_damage= bottom_status&(DC_ERROR|AC_ERROR|MV_ERROR);
-            int offset= b_x*8 + b_y*stride*8;
-            int16_t *top_mv=    s->current_picture.motion_val[0][s->b8_stride*( b_y   <<(1-is_luma)) + (b_x<<(1-is_luma))];
-            int16_t *bottom_mv= s->current_picture.motion_val[0][s->b8_stride*((b_y+1)<<(1-is_luma)) + (b_x<<(1-is_luma))];
-            
-            if(!(top_damage||bottom_damage)) continue; // both undamaged
-            
-            if(   (!top_intra) && (!bottom_intra) 
-               && ABS(top_mv[0]-bottom_mv[0]) + ABS(top_mv[1]+bottom_mv[1]) < 2) continue;
-            
-            for(x=0; x<8; x++){
-                int a,b,c,d;
-                
-                a= dst[offset + x + 7*stride] - dst[offset + x + 6*stride];
-                b= dst[offset + x + 8*stride] - dst[offset + x + 7*stride];
-                c= dst[offset + x + 9*stride] - dst[offset + x + 8*stride];
-                
-                d= ABS(b) - ((ABS(a) + ABS(c)+1)>>1);
-                d= FFMAX(d, 0);
-                if(b<0) d= -d;
-                
-                if(d==0) continue;
-
-                if(!(top_damage && bottom_damage))
-                    d= d*16/9;
-                
-                if(top_damage){
-                    dst[offset + x +  7*stride] = cm[dst[offset + x +  7*stride] + ((d*7)>>4)];
-                    dst[offset + x +  6*stride] = cm[dst[offset + x +  6*stride] + ((d*5)>>4)];
-                    dst[offset + x +  5*stride] = cm[dst[offset + x +  5*stride] + ((d*3)>>4)];
-                    dst[offset + x +  4*stride] = cm[dst[offset + x +  4*stride] + ((d*1)>>4)];
+            int top_status    = s->error_status_table[(b_x >> is_luma) +  (b_y      >> is_luma) * s->mb_stride];
+            int bottom_status = s->error_status_table[(b_x >> is_luma) + ((b_y + 1) >> is_luma) * s->mb_stride];
+            int top_intra     = IS_INTRA(s->cur_pic.mb_type[(b_x >> is_luma) + ( b_y      >> is_luma) * s->mb_stride]);
+            int bottom_intra  = IS_INTRA(s->cur_pic.mb_type[(b_x >> is_luma) + ((b_y + 1) >> is_luma) * s->mb_stride]);
+            int top_damage    = top_status & ER_MB_ERROR;
+            int bottom_damage = bottom_status & ER_MB_ERROR;
+            int offset        = b_x * 8 + b_y * stride * 8;
+
+            int16_t *top_mv    = s->cur_pic.motion_val[0][mvy_stride *  b_y      + mvx_stride * b_x];
+            int16_t *bottom_mv = s->cur_pic.motion_val[0][mvy_stride * (b_y + 1) + mvx_stride * b_x];
+
+            if (!(top_damage || bottom_damage))
+                continue; // both undamaged
+
+            if ((!top_intra) && (!bottom_intra) &&
+                FFABS(top_mv[0] - bottom_mv[0]) +
+                FFABS(top_mv[1] + bottom_mv[1]) < 2)
+                continue;
+
+            for (x = 0; x < 8; x++) {
+                int a, b, c, d;
+
+                a = dst[offset + x + 7 * stride] - dst[offset + x + 6 * stride];
+                b = dst[offset + x + 8 * stride] - dst[offset + x + 7 * stride];
+                c = dst[offset + x + 9 * stride] - dst[offset + x + 8 * stride];
+
+                d = FFABS(b) - ((FFABS(a) + FFABS(c) + 1) >> 1);
+                d = FFMAX(d, 0);
+                if (b < 0)
+                    d = -d;
+
+                if (d == 0)
+                    continue;
+
+                if (!(top_damage && bottom_damage))
+                    d = d * 16 / 9;
+
+                if (top_damage) {
+                    dst[offset + x +  7 * stride] = cm[dst[offset + x +  7 * stride] + ((d * 7) >> 4)];
+                    dst[offset + x +  6 * stride] = cm[dst[offset + x +  6 * stride] + ((d * 5) >> 4)];
+                    dst[offset + x +  5 * stride] = cm[dst[offset + x +  5 * stride] + ((d * 3) >> 4)];
+                    dst[offset + x +  4 * stride] = cm[dst[offset + x +  4 * stride] + ((d * 1) >> 4)];
                 }
-                if(bottom_damage){
-                    dst[offset + x +  8*stride] = cm[dst[offset + x +  8*stride] - ((d*7)>>4)];
-                    dst[offset + x +  9*stride] = cm[dst[offset + x +  9*stride] - ((d*5)>>4)];
-                    dst[offset + x + 10*stride] = cm[dst[offset + x + 10*stride] - ((d*3)>>4)];
-                    dst[offset + x + 11*stride] = cm[dst[offset + x + 11*stride] - ((d*1)>>4)];
+                if (bottom_damage) {
+                    dst[offset + x +  8 * stride] = cm[dst[offset + x +  8 * stride] - ((d * 7) >> 4)];
+                    dst[offset + x +  9 * stride] = cm[dst[offset + x +  9 * stride] - ((d * 5) >> 4)];
+                    dst[offset + x + 10 * stride] = cm[dst[offset + x + 10 * stride] - ((d * 3) >> 4)];
+                    dst[offset + x + 11 * stride] = cm[dst[offset + x + 11 * stride] - ((d * 1) >> 4)];
                 }
             }
         }
     }
 }
 
-static void guess_mv(MpegEncContext *s){
-    uint8_t fixed[s->mb_stride * s->mb_height];
+static void guess_mv(ERContext *s)
+{
+    uint8_t *fixed = s->er_temp_buffer;
 #define MV_FROZEN    3
 #define MV_CHANGED   2
 #define MV_UNCHANGED 1
     const int mb_stride = s->mb_stride;
-    const int mb_width = s->mb_width;
-    const int mb_height= s->mb_height;
+    const int mb_width  = s->mb_width;
+    const int mb_height = s->mb_height;
     int i, depth, num_avail;
-    int mb_x, mb_y;
-   
-    num_avail=0;
-    for(i=0; i<s->mb_num; i++){
-        const int mb_xy= s->mb_index2xy[ i ];
-        int f=0;
-        int error= s->error_status_table[mb_xy];
-
-        if(IS_INTRA(s->current_picture.mb_type[mb_xy])) f=MV_FROZEN; //intra //FIXME check
-        if(!(error&MV_ERROR)) f=MV_FROZEN;           //inter with undamaged MV
-        
-        fixed[mb_xy]= f;
-        if(f==MV_FROZEN)
+    int mb_x, mb_y, mot_step, mot_stride;
+
+    set_mv_strides(s, &mot_step, &mot_stride);
+
+    num_avail = 0;
+    for (i = 0; i < s->mb_num; i++) {
+        const int mb_xy = s->mb_index2xy[i];
+        int f = 0;
+        int error = s->error_status_table[mb_xy];
+
+        if (IS_INTRA(s->cur_pic.mb_type[mb_xy]))
+            f = MV_FROZEN; // intra // FIXME check
+        if (!(error & ER_MV_ERROR))
+            f = MV_FROZEN; // inter with undamaged MV
+
+        fixed[mb_xy] = f;
+        if (f == MV_FROZEN)
             num_avail++;
     }
-    
-    if((!(s->avctx->error_concealment&FF_EC_GUESS_MVS)) || num_avail <= mb_width/2){
-        for(mb_y=0; mb_y<s->mb_height; mb_y++){
-            for(mb_x=0; mb_x<s->mb_width; mb_x++){
-                const int mb_xy= mb_x + mb_y*s->mb_stride;
-                
-                if(IS_INTRA(s->current_picture.mb_type[mb_xy]))  continue;
-                if(!(s->error_status_table[mb_xy]&MV_ERROR)) continue;
-
-                s->mv_dir = MV_DIR_FORWARD;
-                s->mb_intra=0;
-                s->mv_type = MV_TYPE_16X16;
-                s->mb_skiped=0;
-
-               s->dsp.clear_blocks(s->block[0]);
-
-                s->mb_x= mb_x;
-                s->mb_y= mb_y;
-                s->mv[0][0][0]= 0;
-                s->mv[0][0][1]= 0;
-                decode_mb(s);
+
+    if ((!(s->avctx->error_concealment&FF_EC_GUESS_MVS)) ||
+        num_avail <= mb_width / 2) {
+        for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
+            for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
+                const int mb_xy = mb_x + mb_y * s->mb_stride;
+                int mv_dir = (s->last_pic.f && s->last_pic.f->data[0]) ? MV_DIR_FORWARD : MV_DIR_BACKWARD;
+
+                if (IS_INTRA(s->cur_pic.mb_type[mb_xy]))
+                    continue;
+                if (!(s->error_status_table[mb_xy] & ER_MV_ERROR))
+                    continue;
+
+                s->mv[0][0][0] = 0;
+                s->mv[0][0][1] = 0;
+                s->decode_mb(s->opaque, 0, mv_dir, MV_TYPE_16X16, &s->mv,
+                             mb_x, mb_y, 0, 0);
             }
         }
         return;
     }
-    
-    for(depth=0;; depth++){
+
+    for (depth = 0; ; depth++) {
         int changed, pass, none_left;
 
-        none_left=1;
-        changed=1;
-        for(pass=0; (changed || pass<2) && pass<10; pass++){
+        none_left = 1;
+        changed   = 1;
+        for (pass = 0; (changed || pass < 2) && pass < 10; pass++) {
             int mb_x, mb_y;
-int score_sum=0;
-            changed=0;
-            for(mb_y=0; mb_y<s->mb_height; mb_y++){
-                for(mb_x=0; mb_x<s->mb_width; mb_x++){
-                    const int mb_xy= mb_x + mb_y*s->mb_stride;
-                    int mv_predictor[8][2]={{0}};
-                    int pred_count=0;
+            int score_sum = 0;
+
+            changed = 0;
+            for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
+                for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
+                    const int mb_xy        = mb_x + mb_y * s->mb_stride;
+                    int mv_predictor[8][2] = { { 0 } };
+                    int ref[8]             = { 0 };
+                    int pred_count         = 0;
                     int j;
-                    int best_score=256*256*256*64;
-                    int best_pred=0;
-                    const int mot_stride= s->b8_stride;
-                    const int mot_index= mb_x*2 + mb_y*2*mot_stride;
-                    int prev_x= s->current_picture.motion_val[0][mot_index][0];
-                    int prev_y= s->current_picture.motion_val[0][mot_index][1];
-
-                    if((mb_x^mb_y^pass)&1) continue;
-                    
-                    if(fixed[mb_xy]==MV_FROZEN) continue;
-                    assert(!IS_INTRA(s->current_picture.mb_type[mb_xy]));
-                    assert(s->last_picture_ptr && s->last_picture_ptr->data[0]);
-                    
-                    j=0;
-                    if(mb_x>0           && fixed[mb_xy-1        ]==MV_FROZEN) j=1;
-                    if(mb_x+1<mb_width  && fixed[mb_xy+1        ]==MV_FROZEN) j=1;
-                    if(mb_y>0           && fixed[mb_xy-mb_stride]==MV_FROZEN) j=1;
-                    if(mb_y+1<mb_height && fixed[mb_xy+mb_stride]==MV_FROZEN) j=1;
-                    if(j==0) continue;
-
-                    j=0;
-                    if(mb_x>0           && fixed[mb_xy-1        ]==MV_CHANGED) j=1;
-                    if(mb_x+1<mb_width  && fixed[mb_xy+1        ]==MV_CHANGED) j=1;
-                    if(mb_y>0           && fixed[mb_xy-mb_stride]==MV_CHANGED) j=1;
-                    if(mb_y+1<mb_height && fixed[mb_xy+mb_stride]==MV_CHANGED) j=1;
-                    if(j==0 && pass>1) continue;
-                    
-                    none_left=0;
-                    
-                    if(mb_x>0 && fixed[mb_xy-1]){
-                        mv_predictor[pred_count][0]= s->current_picture.motion_val[0][mot_index - 2][0];
-                        mv_predictor[pred_count][1]= s->current_picture.motion_val[0][mot_index - 2][1];
+                    int best_score         = 256 * 256 * 256 * 64;
+                    int best_pred          = 0;
+                    const int mot_index    = (mb_x + mb_y * mot_stride) * mot_step;
+                    int prev_x = 0, prev_y = 0, prev_ref = 0;
+
+                    if ((mb_x ^ mb_y ^ pass) & 1)
+                        continue;
+
+                    if (fixed[mb_xy] == MV_FROZEN)
+                        continue;
+
+                    j = 0;
+                    if (mb_x > 0             && fixed[mb_xy - 1]         == MV_FROZEN)
+                        j = 1;
+                    if (mb_x + 1 < mb_width  && fixed[mb_xy + 1]         == MV_FROZEN)
+                        j = 1;
+                    if (mb_y > 0             && fixed[mb_xy - mb_stride] == MV_FROZEN)
+                        j = 1;
+                    if (mb_y + 1 < mb_height && fixed[mb_xy + mb_stride] == MV_FROZEN)
+                        j = 1;
+                    if (j == 0)
+                        continue;
+
+                    j = 0;
+                    if (mb_x > 0             && fixed[mb_xy - 1        ] == MV_CHANGED)
+                        j = 1;
+                    if (mb_x + 1 < mb_width  && fixed[mb_xy + 1        ] == MV_CHANGED)
+                        j = 1;
+                    if (mb_y > 0             && fixed[mb_xy - mb_stride] == MV_CHANGED)
+                        j = 1;
+                    if (mb_y + 1 < mb_height && fixed[mb_xy + mb_stride] == MV_CHANGED)
+                        j = 1;
+                    if (j == 0 && pass > 1)
+                        continue;
+
+                    none_left = 0;
+
+                    if (mb_x > 0 && fixed[mb_xy - 1]) {
+                        mv_predictor[pred_count][0] =
+                            s->cur_pic.motion_val[0][mot_index - mot_step][0];
+                        mv_predictor[pred_count][1] =
+                            s->cur_pic.motion_val[0][mot_index - mot_step][1];
+                        ref[pred_count] =
+                            s->cur_pic.ref_index[0][4 * (mb_xy - 1)];
                         pred_count++;
                     }
-                    if(mb_x+1<mb_width && fixed[mb_xy+1]){
-                        mv_predictor[pred_count][0]= s->current_picture.motion_val[0][mot_index + 2][0];
-                        mv_predictor[pred_count][1]= s->current_picture.motion_val[0][mot_index + 2][1];
+                    if (mb_x + 1 < mb_width && fixed[mb_xy + 1]) {
+                        mv_predictor[pred_count][0] =
+                            s->cur_pic.motion_val[0][mot_index + mot_step][0];
+                        mv_predictor[pred_count][1] =
+                            s->cur_pic.motion_val[0][mot_index + mot_step][1];
+                        ref[pred_count] =
+                            s->cur_pic.ref_index[0][4 * (mb_xy + 1)];
                         pred_count++;
                     }
-                    if(mb_y>0 && fixed[mb_xy-mb_stride]){
-                        mv_predictor[pred_count][0]= s->current_picture.motion_val[0][mot_index - mot_stride*2][0];
-                        mv_predictor[pred_count][1]= s->current_picture.motion_val[0][mot_index - mot_stride*2][1];
+                    if (mb_y > 0 && fixed[mb_xy - mb_stride]) {
+                        mv_predictor[pred_count][0] =
+                            s->cur_pic.motion_val[0][mot_index - mot_stride * mot_step][0];
+                        mv_predictor[pred_count][1] =
+                            s->cur_pic.motion_val[0][mot_index - mot_stride * mot_step][1];
+                        ref[pred_count] =
+                            s->cur_pic.ref_index[0][4 * (mb_xy - s->mb_stride)];
                         pred_count++;
                     }
-                    if(mb_y+1<mb_height && fixed[mb_xy+mb_stride]){
-                        mv_predictor[pred_count][0]= s->current_picture.motion_val[0][mot_index + mot_stride*2][0];
-                        mv_predictor[pred_count][1]= s->current_picture.motion_val[0][mot_index + mot_stride*2][1];
+                    if (mb_y + 1<mb_height && fixed[mb_xy + mb_stride]) {
+                        mv_predictor[pred_count][0] =
+                            s->cur_pic.motion_val[0][mot_index + mot_stride * mot_step][0];
+                        mv_predictor[pred_count][1] =
+                            s->cur_pic.motion_val[0][mot_index + mot_stride * mot_step][1];
+                        ref[pred_count] =
+                            s->cur_pic.ref_index[0][4 * (mb_xy + s->mb_stride)];
                         pred_count++;
                     }
-                    if(pred_count==0) continue;
-                    
-                    if(pred_count>1){
-                        int sum_x=0, sum_y=0;
-                        int max_x, max_y, min_x, min_y;
-
-                        for(j=0; j<pred_count; j++){
-                            sum_x+= mv_predictor[j][0];
-                            sum_y+= mv_predictor[j][1];
+                    if (pred_count == 0)
+                        continue;
+
+                    if (pred_count > 1) {
+                        int sum_x = 0, sum_y = 0, sum_r = 0;
+                        int max_x, max_y, min_x, min_y, max_r, min_r;
+
+                        for (j = 0; j < pred_count; j++) {
+                            sum_x += mv_predictor[j][0];
+                            sum_y += mv_predictor[j][1];
+                            sum_r += ref[j];
+                            if (j && ref[j] != ref[j - 1])
+                                goto skip_mean_and_median;
                         }
-                    
+
                         /* mean */
-                        mv_predictor[pred_count][0] = sum_x/j;
-                        mv_predictor[pred_count][1] = sum_y/j;
-                    
+                        mv_predictor[pred_count][0] = sum_x / j;
+                        mv_predictor[pred_count][1] = sum_y / j;
+                                 ref[pred_count]    = sum_r / j;
+
                         /* median */
-                        if(pred_count>=3){
-                            min_y= min_x= 99999;
-                            max_y= max_x=-99999;
-                        }else{
-                            min_x=min_y=max_x=max_y=0;
+                        if (pred_count >= 3) {
+                            min_y = min_x = min_r =  99999;
+                            max_y = max_x = max_r = -99999;
+                        } else {
+                            min_x = min_y = max_x = max_y = min_r = max_r = 0;
                         }
-                        for(j=0; j<pred_count; j++){
-                            max_x= FFMAX(max_x, mv_predictor[j][0]);
-                            max_y= FFMAX(max_y, mv_predictor[j][1]);
-                            min_x= FFMIN(min_x, mv_predictor[j][0]);
-                            min_y= FFMIN(min_y, mv_predictor[j][1]);
+                        for (j = 0; j < pred_count; j++) {
+                            max_x = FFMAX(max_x, mv_predictor[j][0]);
+                            max_y = FFMAX(max_y, mv_predictor[j][1]);
+                            max_r = FFMAX(max_r, ref[j]);
+                            min_x = FFMIN(min_x, mv_predictor[j][0]);
+                            min_y = FFMIN(min_y, mv_predictor[j][1]);
+                            min_r = FFMIN(min_r, ref[j]);
                         }
-                        mv_predictor[pred_count+1][0] = sum_x - max_x - min_x;
-                        mv_predictor[pred_count+1][1] = sum_y - max_y - min_y;
-                        
-                        if(pred_count==4){
-                            mv_predictor[pred_count+1][0] /= 2;
-                            mv_predictor[pred_count+1][1] /= 2;
+                        mv_predictor[pred_count + 1][0] = sum_x - max_x - min_x;
+                        mv_predictor[pred_count + 1][1] = sum_y - max_y - min_y;
+                                 ref[pred_count + 1]    = sum_r - max_r - min_r;
+
+                        if (pred_count == 4) {
+                            mv_predictor[pred_count + 1][0] /= 2;
+                            mv_predictor[pred_count + 1][1] /= 2;
+                                     ref[pred_count + 1]    /= 2;
                         }
-                        pred_count+=2;
+                        pred_count += 2;
                     }
-                    
+
+skip_mean_and_median:
                     /* zero MV */
                     pred_count++;
 
+                    if (!fixed[mb_xy]) {
+                        if (s->avctx->codec_id == AV_CODEC_ID_H264) {
+                            // FIXME
+                        } else {
+                            ff_thread_await_progress(s->last_pic.tf,
+                                                     mb_y, 0);
+                        }
+                        if (!s->last_pic.motion_val[0] ||
+                            !s->last_pic.ref_index[0])
+                            goto skip_last_mv;
+                        prev_x   = s->last_pic.motion_val[0][mot_index][0];
+                        prev_y   = s->last_pic.motion_val[0][mot_index][1];
+                        prev_ref = s->last_pic.ref_index[0][4 * mb_xy];
+                    } else {
+                        prev_x   = s->cur_pic.motion_val[0][mot_index][0];
+                        prev_y   = s->cur_pic.motion_val[0][mot_index][1];
+                        prev_ref = s->cur_pic.ref_index[0][4 * mb_xy];
+                    }
+
                     /* last MV */
-                    mv_predictor[pred_count][0]= s->current_picture.motion_val[0][mot_index][0];
-                    mv_predictor[pred_count][1]= s->current_picture.motion_val[0][mot_index][1];
-                    pred_count++;                    
-                    
-                    s->mv_dir = MV_DIR_FORWARD;
-                    s->mb_intra=0;
-                    s->mv_type = MV_TYPE_16X16;
-                    s->mb_skiped=0;
-
-                   s->dsp.clear_blocks(s->block[0]);
-
-                    s->mb_x= mb_x;
-                    s->mb_y= mb_y;
-
-                    for(j=0; j<pred_count; j++){
-                        int score=0;
-                        uint8_t *src= s->current_picture.data[0] + mb_x*16 + mb_y*16*s->linesize;
-
-                        s->current_picture.motion_val[0][mot_index][0]= s->mv[0][0][0]= mv_predictor[j][0];
-                        s->current_picture.motion_val[0][mot_index][1]= s->mv[0][0][1]= mv_predictor[j][1];
-
-                        decode_mb(s);
-                        
-                        if(mb_x>0 && fixed[mb_xy-1]){
+                    mv_predictor[pred_count][0] = prev_x;
+                    mv_predictor[pred_count][1] = prev_y;
+                             ref[pred_count]    = prev_ref;
+                    pred_count++;
+
+skip_last_mv:
+
+                    for (j = 0; j < pred_count; j++) {
+                        int *linesize = s->cur_pic.f->linesize;
+                        int score = 0;
+                        uint8_t *src = s->cur_pic.f->data[0] +
+                                       mb_x * 16 + mb_y * 16 * linesize[0];
+
+                        s->cur_pic.motion_val[0][mot_index][0] =
+                            s->mv[0][0][0] = mv_predictor[j][0];
+                        s->cur_pic.motion_val[0][mot_index][1] =
+                            s->mv[0][0][1] = mv_predictor[j][1];
+
+                        // predictor intra or otherwise not available
+                        if (ref[j] < 0)
+                            continue;
+
+                        s->decode_mb(s->opaque, ref[j], MV_DIR_FORWARD,
+                                     MV_TYPE_16X16, &s->mv, mb_x, mb_y, 0, 0);
+
+                        if (mb_x > 0 && fixed[mb_xy - 1]) {
                             int k;
-                            for(k=0; k<16; k++)
-                                score += ABS(src[k*s->linesize-1 ]-src[k*s->linesize   ]);
+                            for (k = 0; k < 16; k++)
+                                score += FFABS(src[k * linesize[0] - 1] -
+                                               src[k * linesize[0]]);
                         }
-                        if(mb_x+1<mb_width && fixed[mb_xy+1]){
+                        if (mb_x + 1 < mb_width && fixed[mb_xy + 1]) {
                             int k;
-                            for(k=0; k<16; k++)
-                                score += ABS(src[k*s->linesize+15]-src[k*s->linesize+16]);
+                            for (k = 0; k < 16; k++)
+                                score += FFABS(src[k * linesize[0] + 15] -
+                                               src[k * linesize[0] + 16]);
                         }
-                        if(mb_y>0 && fixed[mb_xy-mb_stride]){
+                        if (mb_y > 0 && fixed[mb_xy - mb_stride]) {
                             int k;
-                            for(k=0; k<16; k++)
-                                score += ABS(src[k-s->linesize   ]-src[k               ]);
+                            for (k = 0; k < 16; k++)
+                                score += FFABS(src[k - linesize[0]] - src[k]);
                         }
-                        if(mb_y+1<mb_height && fixed[mb_xy+mb_stride]){
+                        if (mb_y + 1 < mb_height && fixed[mb_xy + mb_stride]) {
                             int k;
-                            for(k=0; k<16; k++)
-                                score += ABS(src[k+s->linesize*15]-src[k+s->linesize*16]);
+                            for (k = 0; k < 16; k++)
+                                score += FFABS(src[k + linesize[0] * 15] -
+                                               src[k + linesize[0] * 16]);
                         }
-                        
-                        if(score <= best_score){ // <= will favor the last MV
-                            best_score= score;
-                            best_pred= j;
+
+                        if (score <= best_score) { // <= will favor the last MV
+                            best_score = score;
+                            best_pred  = j;
                         }
                     }
-score_sum+= best_score;
-//FIXME no need to set s->current_picture.motion_val[0][mot_index][0] explicit
-                    s->current_picture.motion_val[0][mot_index][0]= s->mv[0][0][0]= mv_predictor[best_pred][0];
-                    s->current_picture.motion_val[0][mot_index][1]= s->mv[0][0][1]= mv_predictor[best_pred][1];
+                    score_sum += best_score;
+                    s->mv[0][0][0] = mv_predictor[best_pred][0];
+                    s->mv[0][0][1] = mv_predictor[best_pred][1];
+
+                    for (i = 0; i < mot_step; i++)
+                        for (j = 0; j < mot_step; j++) {
+                            s->cur_pic.motion_val[0][mot_index + i + j * mot_stride][0] = s->mv[0][0][0];
+                            s->cur_pic.motion_val[0][mot_index + i + j * mot_stride][1] = s->mv[0][0][1];
+                        }
+
+                    s->decode_mb(s->opaque, ref[best_pred], MV_DIR_FORWARD,
+                                 MV_TYPE_16X16, &s->mv, mb_x, mb_y, 0, 0);
 
-                    decode_mb(s);
 
-                    
-                    if(s->mv[0][0][0] != prev_x || s->mv[0][0][1] != prev_y){
-                        fixed[mb_xy]=MV_CHANGED;
+                    if (s->mv[0][0][0] != prev_x || s->mv[0][0][1] != prev_y) {
+                        fixed[mb_xy] = MV_CHANGED;
                         changed++;
-                    }else
-                        fixed[mb_xy]=MV_UNCHANGED;
+                    } else
+                        fixed[mb_xy] = MV_UNCHANGED;
                 }
             }
-
-//            printf(".%d/%d", changed, score_sum); fflush(stdout);
         }
-        
-        if(none_left) 
+
+        if (none_left)
             return;
-            
-        for(i=0; i<s->mb_num; i++){
-            int mb_xy= s->mb_index2xy[i];
-            if(fixed[mb_xy])
-                fixed[mb_xy]=MV_FROZEN;
+
+        for (i = 0; i < s->mb_num; i++) {
+            int mb_xy = s->mb_index2xy[i];
+            if (fixed[mb_xy])
+                fixed[mb_xy] = MV_FROZEN;
         }
-//        printf(":"); fflush(stdout);
     }
 }
-    
-static int is_intra_more_likely(MpegEncContext *s){
+
+static int is_intra_more_likely(ERContext *s)
+{
     int is_intra_likely, i, j, undamaged_count, skip_amount, mb_x, mb_y;
-    
-    if(s->last_picture_ptr==NULL) return 1; //no previous frame available -> use spatial prediction
-
-    undamaged_count=0;
-    for(i=0; i<s->mb_num; i++){
-        const int mb_xy= s->mb_index2xy[i];
-        const int error= s->error_status_table[mb_xy];
-        if(!((error&DC_ERROR) && (error&MV_ERROR)))
+
+    if (!s->last_pic.f || !s->last_pic.f->data[0])
+        return 1; // no previous frame available -> use spatial prediction
+
+    undamaged_count = 0;
+    for (i = 0; i < s->mb_num; i++) {
+        const int mb_xy = s->mb_index2xy[i];
+        const int error = s->error_status_table[mb_xy];
+        if (!((error & ER_DC_ERROR) && (error & ER_MV_ERROR)))
             undamaged_count++;
     }
-    
-    if(undamaged_count < 5) return 0; //allmost all MBs damaged -> use temporal prediction
-    
-    skip_amount= FFMAX(undamaged_count/50, 1); //check only upto 50 MBs 
-    is_intra_likely=0;
-
-    j=0;
-    for(mb_y= 0; mb_y<s->mb_height-1; mb_y++){
-        for(mb_x= 0; mb_x<s->mb_width; mb_x++){
+
+    if (s->avctx->codec_id == AV_CODEC_ID_H264 && s->ref_count <= 0)
+        return 1;
+
+    if (undamaged_count < 5)
+        return 0; // almost all MBs damaged -> use temporal prediction
+
+#if FF_API_XVMC
+FF_DISABLE_DEPRECATION_WARNINGS
+    // prevent dsp.sad() check, that requires access to the image
+    if (CONFIG_MPEG_XVMC_DECODER    &&
+        s->avctx->xvmc_acceleration &&
+        s->cur_pic.f->pict_type == AV_PICTURE_TYPE_I)
+        return 1;
+FF_ENABLE_DEPRECATION_WARNINGS
+#endif /* FF_API_XVMC */
+
+    skip_amount     = FFMAX(undamaged_count / 50, 1); // check only up to 50 MBs
+    is_intra_likely = 0;
+
+    j = 0;
+    for (mb_y = 0; mb_y < s->mb_height - 1; mb_y++) {
+        for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
             int error;
-            const int mb_xy= mb_x + mb_y*s->mb_stride;
-
-            error= s->error_status_table[mb_xy];
-            if((error&DC_ERROR) && (error&MV_ERROR))
-                continue; //skip damaged
-        
-            j++;    
-            if((j%skip_amount) != 0) continue; //skip a few to speed things up
-    
-            if(s->pict_type==I_TYPE){
-                uint8_t *mb_ptr     = s->current_picture.data[0] + mb_x*16 + mb_y*16*s->linesize;
-                uint8_t *last_mb_ptr= s->last_picture.data   [0] + mb_x*16 + mb_y*16*s->linesize;
-    
-               is_intra_likely += s->dsp.sad[0](NULL, last_mb_ptr, mb_ptr                    , s->linesize, 16);
-                is_intra_likely -= s->dsp.sad[0](NULL, last_mb_ptr, last_mb_ptr+s->linesize*16, s->linesize, 16);
-            }else{
-                if(IS_INTRA(s->current_picture.mb_type[mb_xy]))
+            const int mb_xy = mb_x + mb_y * s->mb_stride;
+
+            error = s->error_status_table[mb_xy];
+            if ((error & ER_DC_ERROR) && (error & ER_MV_ERROR))
+                continue; // skip damaged
+
+            j++;
+            // skip a few to speed things up
+            if ((j % skip_amount) != 0)
+                continue;
+
+            if (s->cur_pic.f->pict_type == AV_PICTURE_TYPE_I) {
+                int *linesize = s->cur_pic.f->linesize;
+                uint8_t *mb_ptr      = s->cur_pic.f->data[0] +
+                                       mb_x * 16 + mb_y * 16 * linesize[0];
+                uint8_t *last_mb_ptr = s->last_pic.f->data[0] +
+                                       mb_x * 16 + mb_y * 16 * linesize[0];
+
+                if (s->avctx->codec_id == AV_CODEC_ID_H264) {
+                    // FIXME
+                } else {
+                    ff_thread_await_progress(s->last_pic.tf, mb_y, 0);
+                }
+                is_intra_likely += s->mecc.sad[0](NULL, last_mb_ptr, mb_ptr,
+                                                  linesize[0], 16);
+                is_intra_likely -= s->mecc.sad[0](NULL, last_mb_ptr,
+                                                  last_mb_ptr + linesize[0] * 16,
+                                                  linesize[0], 16);
+            } else {
+                if (IS_INTRA(s->cur_pic.mb_type[mb_xy]))
                    is_intra_likely++;
                 else
                    is_intra_likely--;
             }
         }
     }
-//printf("is_intra_likely: %d type:%d\n", is_intra_likely, s->pict_type);
-    return is_intra_likely > 0;    
+    return is_intra_likely > 0;
 }
 
-void ff_er_frame_start(MpegEncContext *s){
-    if(!s->error_resilience) return;
+void ff_er_frame_start(ERContext *s)
+{
+    if (!s->avctx->error_concealment)
+        return;
+
+    if (!s->mecc_inited) {
+        ff_me_cmp_init(&s->mecc, s->avctx);
+        s->mecc_inited = 1;
+    }
 
-    memset(s->error_status_table, MV_ERROR|AC_ERROR|DC_ERROR|VP_START|AC_END|DC_END|MV_END, s->mb_stride*s->mb_height*sizeof(uint8_t));
-    s->error_count= 3*s->mb_num;
+    memset(s->error_status_table, ER_MB_ERROR | VP_START | ER_MB_END,
+           s->mb_stride * s->mb_height * sizeof(uint8_t));
+    s->error_count    = 3 * s->mb_num;
+    s->error_occurred = 0;
 }
 
 /**
- * adds a slice.
- * @param endx x component of the last macroblock, can be -1 for the last of the previous line
- * @param status the status at the end (MV_END, AC_ERROR, ...), it is assumed that no earlier end or
- *               error of the same type occured
+ * Add a slice.
+ * @param endx   x component of the last macroblock, can be -1
+ *               for the last of the previous line
+ * @param status the status at the end (ER_MV_END, ER_AC_ERROR, ...), it is
+ *               assumed that no earlier end or error of the same type occurred
  */
-void ff_er_add_slice(MpegEncContext *s, int startx, int starty, int endx, int endy, int status){
-    const int start_i= clip(startx + starty * s->mb_width    , 0, s->mb_num-1);
-    const int end_i  = clip(endx   + endy   * s->mb_width    , 0, s->mb_num);
-    const int start_xy= s->mb_index2xy[start_i];
-    const int end_xy  = s->mb_index2xy[end_i];
-    int mask= -1;
-    
-    if(!s->error_resilience) return;
+void ff_er_add_slice(ERContext *s, int startx, int starty,
+                     int endx, int endy, int status)
+{
+    const int start_i  = av_clip(startx + starty * s->mb_width, 0, s->mb_num - 1);
+    const int end_i    = av_clip(endx   + endy   * s->mb_width, 0, s->mb_num);
+    const int start_xy = s->mb_index2xy[start_i];
+    const int end_xy   = s->mb_index2xy[end_i];
+    int mask           = -1;
+
+    if (s->avctx->hwaccel)
+        return;
+
+    if (start_i > end_i || start_xy > end_xy) {
+        av_log(s->avctx, AV_LOG_ERROR,
+               "internal error, slice end before start\n");
+        return;
+    }
+
+    if (!s->avctx->error_concealment)
+        return;
 
     mask &= ~VP_START;
-    if(status & (AC_ERROR|AC_END)){
-        mask &= ~(AC_ERROR|AC_END);
+    if (status & (ER_AC_ERROR | ER_AC_END)) {
+        mask           &= ~(ER_AC_ERROR | ER_AC_END);
         s->error_count -= end_i - start_i + 1;
     }
-    if(status & (DC_ERROR|DC_END)){
-        mask &= ~(DC_ERROR|DC_END);
+    if (status & (ER_DC_ERROR | ER_DC_END)) {
+        mask           &= ~(ER_DC_ERROR | ER_DC_END);
         s->error_count -= end_i - start_i + 1;
     }
-    if(status & (MV_ERROR|MV_END)){
-        mask &= ~(MV_ERROR|MV_END);
+    if (status & (ER_MV_ERROR | ER_MV_END)) {
+        mask           &= ~(ER_MV_ERROR | ER_MV_END);
         s->error_count -= end_i - start_i + 1;
     }
 
-    if(status & (AC_ERROR|DC_ERROR|MV_ERROR)) s->error_count= INT_MAX;
+    if (status & ER_MB_ERROR) {
+        s->error_occurred = 1;
+        s->error_count    = INT_MAX;
+    }
 
-    if(mask == ~0x7F){
-        memset(&s->error_status_table[start_xy], 0, (end_xy - start_xy) * sizeof(uint8_t));
-    }else{
+    if (mask == ~0x7F) {
+        memset(&s->error_status_table[start_xy], 0,
+               (end_xy - start_xy) * sizeof(uint8_t));
+    } else {
         int i;
-        for(i=start_xy; i<end_xy; i++){
-            s->error_status_table[ i ] &= mask;
-        }
+        for (i = start_xy; i < end_xy; i++)
+            s->error_status_table[i] &= mask;
     }
 
-    if(end_i == s->mb_num) 
-        s->error_count= INT_MAX;
-    else{
+    if (end_i == s->mb_num)
+        s->error_count = INT_MAX;
+    else {
         s->error_status_table[end_xy] &= mask;
         s->error_status_table[end_xy] |= status;
     }
+
     s->error_status_table[start_xy] |= VP_START;
 
-    if(start_xy > 0 && s->avctx->thread_count <= 1 && s->avctx->skip_top*s->mb_width < start_i){
-        int prev_status= s->error_status_table[ s->mb_index2xy[start_i - 1] ];
-        
+    if (start_xy > 0 && s->avctx->thread_count <= 1 &&
+        s->avctx->skip_top * s->mb_width < start_i) {
+        int prev_status = s->error_status_table[s->mb_index2xy[start_i - 1]];
+
         prev_status &= ~ VP_START;
-        if(prev_status != (MV_END|DC_END|AC_END)) s->error_count= INT_MAX;
+        if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
+            s->error_count = INT_MAX;
     }
 }
 
-void ff_er_frame_end(MpegEncContext *s){
+void ff_er_frame_end(ERContext *s)
+{
+    int *linesize = s->cur_pic.f->linesize;
     int i, mb_x, mb_y, error, error_type, dc_error, mv_error, ac_error;
     int distance;
-    int threshold_part[4]= {100,100,100};
-    int threshold= 50;
+    int threshold_part[4] = { 100, 100, 100 };
+    int threshold = 50;
     int is_intra_likely;
-    int size = s->b8_stride * 2 * s->mb_height;
-    Picture *pic= s->current_picture_ptr;
-    
-    if(!s->error_resilience || s->error_count==0 || 
-       s->error_count==3*s->mb_width*(s->avctx->skip_top + s->avctx->skip_bottom)) return;
-
-    if(s->current_picture.motion_val[0] == NULL){
-        av_log(s->avctx, AV_LOG_ERROR, "Warning MVs not available\n");
-            
-        for(i=0; i<2; i++){
-            pic->ref_index[i]= av_mallocz(size * sizeof(uint8_t));
-            pic->motion_val_base[i]= av_mallocz((size+2) * 2 * sizeof(uint16_t));
-            pic->motion_val[i]= pic->motion_val_base[i]+2;
-        }
-        pic->motion_subsample_log2= 3;
-        s->current_picture= *s->current_picture_ptr;
-    }
-    
-    for(i=0; i<2; i++){
-        if(pic->ref_index[i])
-            memset(pic->ref_index[i], 0, size * sizeof(uint8_t));
+
+    /* We do not support ER of field pictures yet,
+     * though it should not crash if enabled. */
+    if (!s->avctx->error_concealment || s->error_count == 0            ||
+        s->avctx->hwaccel                                              ||
+        !s->cur_pic.f                                                  ||
+        s->cur_pic.field_picture                                       ||
+        s->error_count == 3 * s->mb_width *
+                          (s->avctx->skip_top + s->avctx->skip_bottom)) {
+        return;
+    };
+
+    if (!s->cur_pic.motion_val[0] || !s->cur_pic.ref_index[0]) {
+        av_log(s->avctx, AV_LOG_ERROR, "MVs not available, ER not possible.\n");
+        return;
     }
 
-    if(s->avctx->debug&FF_DEBUG_ER){
-        for(mb_y=0; mb_y<s->mb_height; mb_y++){
-            for(mb_x=0; mb_x<s->mb_width; mb_x++){
-                int status= s->error_status_table[mb_x + mb_y*s->mb_stride];
-            
-                av_log(s->avctx, AV_LOG_DEBUG, "%2X ", status); 
+    if (s->avctx->debug & FF_DEBUG_ER) {
+        for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
+            for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
+                int status = s->error_status_table[mb_x + mb_y * s->mb_stride];
+
+                av_log(s->avctx, AV_LOG_DEBUG, "%2X ", status);
             }
             av_log(s->avctx, AV_LOG_DEBUG, "\n");
         }
     }
-    
-#if 1
+
     /* handle overlapping slices */
-    for(error_type=1; error_type<=3; error_type++){
-        int end_ok=0;
-
-        for(i=s->mb_num-1; i>=0; i--){
-            const int mb_xy= s->mb_index2xy[i];
-            int error= s->error_status_table[mb_xy];
-        
-            if(error&(1<<error_type))
-                end_ok=1;
-            if(error&(8<<error_type))
-                end_ok=1;
-
-            if(!end_ok)
-                s->error_status_table[mb_xy]|= 1<<error_type;
-
-            if(error&VP_START)
-                end_ok=0;
+    for (error_type = 1; error_type <= 3; error_type++) {
+        int end_ok = 0;
+
+        for (i = s->mb_num - 1; i >= 0; i--) {
+            const int mb_xy = s->mb_index2xy[i];
+            int error       = s->error_status_table[mb_xy];
+
+            if (error & (1 << error_type))
+                end_ok = 1;
+            if (error & (8 << error_type))
+                end_ok = 1;
+
+            if (!end_ok)
+                s->error_status_table[mb_xy] |= 1 << error_type;
+
+            if (error & VP_START)
+                end_ok = 0;
         }
     }
-#endif
-#if 1
+
     /* handle slices with partitions of different length */
-    if(s->partitioned_frame){
-        int end_ok=0;
-
-        for(i=s->mb_num-1; i>=0; i--){
-            const int mb_xy= s->mb_index2xy[i];
-            int error= s->error_status_table[mb_xy];
-        
-            if(error&AC_END)
-                end_ok=0;
-            if((error&MV_END) || (error&DC_END) || (error&AC_ERROR))
-                end_ok=1;
-
-            if(!end_ok)
-                s->error_status_table[mb_xy]|= AC_ERROR;
-
-            if(error&VP_START)
-                end_ok=0;
+    if (s->partitioned_frame) {
+        int end_ok = 0;
+
+        for (i = s->mb_num - 1; i >= 0; i--) {
+            const int mb_xy = s->mb_index2xy[i];
+            int error       = s->error_status_table[mb_xy];
+
+            if (error & ER_AC_END)
+                end_ok = 0;
+            if ((error & ER_MV_END) ||
+                (error & ER_DC_END) ||
+                (error & ER_AC_ERROR))
+                end_ok = 1;
+
+            if (!end_ok)
+                s->error_status_table[mb_xy]|= ER_AC_ERROR;
+
+            if (error & VP_START)
+                end_ok = 0;
         }
     }
-#endif
+
     /* handle missing slices */
-    if(s->error_resilience>=4){
-        int end_ok=1;
-                
-        for(i=s->mb_num-2; i>=s->mb_width+100; i--){ //FIXME +100 hack
-            const int mb_xy= s->mb_index2xy[i];
-            int error1= s->error_status_table[mb_xy  ];
-            int error2= s->error_status_table[s->mb_index2xy[i+1]];
-        
-            if(error1&VP_START)
-                end_ok=1;
-             
-            if(   error2==(VP_START|DC_ERROR|AC_ERROR|MV_ERROR|AC_END|DC_END|MV_END)
-               && error1!=(VP_START|DC_ERROR|AC_ERROR|MV_ERROR|AC_END|DC_END|MV_END) 
-               && ((error1&AC_END) || (error1&DC_END) || (error1&MV_END))){ //end & uninited
-                end_ok=0;
+    if (s->avctx->err_recognition & AV_EF_EXPLODE) {
+        int end_ok = 1;
+
+        // FIXME + 100 hack
+        for (i = s->mb_num - 2; i >= s->mb_width + 100; i--) {
+            const int mb_xy = s->mb_index2xy[i];
+            int error1 = s->error_status_table[mb_xy];
+            int error2 = s->error_status_table[s->mb_index2xy[i + 1]];
+
+            if (error1 & VP_START)
+                end_ok = 1;
+
+            if (error2 == (VP_START | ER_MB_ERROR | ER_MB_END) &&
+                error1 != (VP_START | ER_MB_ERROR | ER_MB_END) &&
+                ((error1 & ER_AC_END) || (error1 & ER_DC_END) ||
+                (error1 & ER_MV_END))) {
+                // end & uninit
+                end_ok = 0;
             }
-        
-            if(!end_ok)
-                s->error_status_table[mb_xy]|= DC_ERROR|AC_ERROR|MV_ERROR;
+
+            if (!end_ok)
+                s->error_status_table[mb_xy] |= ER_MB_ERROR;
         }
     }
-    
-#if 1
+
     /* backward mark errors */
-    distance=9999999;
-    for(error_type=1; error_type<=3; error_type++){
-        for(i=s->mb_num-1; i>=0; i--){
-            const int mb_xy= s->mb_index2xy[i];
-            int error= s->error_status_table[mb_xy];
-            
-            if(!s->mbskip_table[mb_xy]) //FIXME partition specific
-                distance++;            
-            if(error&(1<<error_type))
-                distance= 0;
-
-            if(s->partitioned_frame){
-                if(distance < threshold_part[error_type-1])
-                    s->error_status_table[mb_xy]|= 1<<error_type;
-            }else{
-                if(distance < threshold)
-                    s->error_status_table[mb_xy]|= 1<<error_type;
+    distance = 9999999;
+    for (error_type = 1; error_type <= 3; error_type++) {
+        for (i = s->mb_num - 1; i >= 0; i--) {
+            const int mb_xy = s->mb_index2xy[i];
+            int       error = s->error_status_table[mb_xy];
+
+            if (s->mbskip_table && !s->mbskip_table[mb_xy]) // FIXME partition specific
+                distance++;
+            if (error & (1 << error_type))
+                distance = 0;
+
+            if (s->partitioned_frame) {
+                if (distance < threshold_part[error_type - 1])
+                    s->error_status_table[mb_xy] |= 1 << error_type;
+            } else {
+                if (distance < threshold)
+                    s->error_status_table[mb_xy] |= 1 << error_type;
             }
 
-            if(error&VP_START)
-                distance= 9999999;
+            if (error & VP_START)
+                distance = 9999999;
         }
     }
-#endif
 
     /* forward mark errors */
-    error=0;
-    for(i=0; i<s->mb_num; i++){
-        const int mb_xy= s->mb_index2xy[i];
-        int old_error= s->error_status_table[mb_xy];
-        
-        if(old_error&VP_START)
-            error= old_error& (DC_ERROR|AC_ERROR|MV_ERROR);
-        else{
-            error|= old_error& (DC_ERROR|AC_ERROR|MV_ERROR);
-            s->error_status_table[mb_xy]|= error;
+    error = 0;
+    for (i = 0; i < s->mb_num; i++) {
+        const int mb_xy = s->mb_index2xy[i];
+        int old_error   = s->error_status_table[mb_xy];
+
+        if (old_error & VP_START) {
+            error = old_error & ER_MB_ERROR;
+        } else {
+            error |= old_error & ER_MB_ERROR;
+            s->error_status_table[mb_xy] |= error;
         }
     }
-#if 1
+
     /* handle not partitioned case */
-    if(!s->partitioned_frame){
-        for(i=0; i<s->mb_num; i++){
-            const int mb_xy= s->mb_index2xy[i];
-            error= s->error_status_table[mb_xy];
-            if(error&(AC_ERROR|DC_ERROR|MV_ERROR))
-                error|= AC_ERROR|DC_ERROR|MV_ERROR;
-            s->error_status_table[mb_xy]= error;
+    if (!s->partitioned_frame) {
+        for (i = 0; i < s->mb_num; i++) {
+            const int mb_xy = s->mb_index2xy[i];
+            error = s->error_status_table[mb_xy];
+            if (error & ER_MB_ERROR)
+                error |= ER_MB_ERROR;
+            s->error_status_table[mb_xy] = error;
         }
     }
-#endif
-
-    dc_error= ac_error= mv_error=0;
-    for(i=0; i<s->mb_num; i++){
-        const int mb_xy= s->mb_index2xy[i];
-        error= s->error_status_table[mb_xy];
-        if(error&DC_ERROR) dc_error ++;
-        if(error&AC_ERROR) ac_error ++;
-        if(error&MV_ERROR) mv_error ++;
+
+    dc_error = ac_error = mv_error = 0;
+    for (i = 0; i < s->mb_num; i++) {
+        const int mb_xy = s->mb_index2xy[i];
+        error = s->error_status_table[mb_xy];
+        if (error & ER_DC_ERROR)
+            dc_error++;
+        if (error & ER_AC_ERROR)
+            ac_error++;
+        if (error & ER_MV_ERROR)
+            mv_error++;
     }
-    av_log(s->avctx, AV_LOG_INFO, "concealing %d DC, %d AC, %d MV errors\n", dc_error, ac_error, mv_error);
+    av_log(s->avctx, AV_LOG_INFO, "concealing %d DC, %d AC, %d MV errors\n",
+           dc_error, ac_error, mv_error);
 
-    is_intra_likely= is_intra_more_likely(s);
+    is_intra_likely = is_intra_more_likely(s);
 
     /* set unknown mb-type to most likely */
-    for(i=0; i<s->mb_num; i++){
-        const int mb_xy= s->mb_index2xy[i];
-        error= s->error_status_table[mb_xy];
-        if(!((error&DC_ERROR) && (error&MV_ERROR)))
+    for (i = 0; i < s->mb_num; i++) {
+        const int mb_xy = s->mb_index2xy[i];
+        error = s->error_status_table[mb_xy];
+        if (!((error & ER_DC_ERROR) && (error & ER_MV_ERROR)))
             continue;
 
-        if(is_intra_likely)
-            s->current_picture.mb_type[mb_xy]= MB_TYPE_INTRA4x4;
+        if (is_intra_likely)
+            s->cur_pic.mb_type[mb_xy] = MB_TYPE_INTRA4x4;
         else
-            s->current_picture.mb_type[mb_xy]= MB_TYPE_16x16 | MB_TYPE_L0;
+            s->cur_pic.mb_type[mb_xy] = MB_TYPE_16x16 | MB_TYPE_L0;
     }
-    
+
+    // change inter to intra blocks if no reference frames are available
+    if (!(s->last_pic.f && s->last_pic.f->data[0]) &&
+        !(s->next_pic.f && s->next_pic.f->data[0]))
+        for (i = 0; i < s->mb_num; i++) {
+            const int mb_xy = s->mb_index2xy[i];
+            if (!IS_INTRA(s->cur_pic.mb_type[mb_xy]))
+                s->cur_pic.mb_type[mb_xy] = MB_TYPE_INTRA4x4;
+        }
+
     /* handle inter blocks with damaged AC */
-    for(mb_y=0; mb_y<s->mb_height; mb_y++){
-        for(mb_x=0; mb_x<s->mb_width; mb_x++){
-            const int mb_xy= mb_x + mb_y * s->mb_stride;
-            const int mb_type= s->current_picture.mb_type[mb_xy];
-            error= s->error_status_table[mb_xy];
-
-            if(IS_INTRA(mb_type)) continue; //intra
-            if(error&MV_ERROR) continue;              //inter with damaged MV
-            if(!(error&AC_ERROR)) continue;           //undamaged inter
-            
-            s->mv_dir = MV_DIR_FORWARD;
-            s->mb_intra=0;
-            s->mb_skiped=0;
-            if(IS_8X8(mb_type)){
-                int mb_index= mb_x*2 + mb_y*2*s->b8_stride;
+    for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
+        for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
+            const int mb_xy   = mb_x + mb_y * s->mb_stride;
+            const int mb_type = s->cur_pic.mb_type[mb_xy];
+            const int dir     = !(s->last_pic.f && s->last_pic.f->data[0]);
+            const int mv_dir  = dir ? MV_DIR_BACKWARD : MV_DIR_FORWARD;
+            int mv_type;
+
+            error = s->error_status_table[mb_xy];
+
+            if (IS_INTRA(mb_type))
+                continue; // intra
+            if (error & ER_MV_ERROR)
+                continue; // inter with damaged MV
+            if (!(error & ER_AC_ERROR))
+                continue; // undamaged inter
+
+            if (IS_8X8(mb_type)) {
+                int mb_index = mb_x * 2 + mb_y * 2 * s->b8_stride;
                 int j;
-                s->mv_type = MV_TYPE_8X8;
-                for(j=0; j<4; j++){
-                    s->mv[0][j][0] = s->current_picture.motion_val[0][ mb_index + (j&1) + (j>>1)*s->b8_stride ][0];
-                    s->mv[0][j][1] = s->current_picture.motion_val[0][ mb_index + (j&1) + (j>>1)*s->b8_stride ][1];
+                mv_type = MV_TYPE_8X8;
+                for (j = 0; j < 4; j++) {
+                    s->mv[0][j][0] = s->cur_pic.motion_val[dir][mb_index + (j & 1) + (j >> 1) * s->b8_stride][0];
+                    s->mv[0][j][1] = s->cur_pic.motion_val[dir][mb_index + (j & 1) + (j >> 1) * s->b8_stride][1];
                 }
-            }else{
-                s->mv_type = MV_TYPE_16X16;
-                s->mv[0][0][0] = s->current_picture.motion_val[0][ mb_x*2 + mb_y*2*s->b8_stride ][0];
-                s->mv[0][0][1] = s->current_picture.motion_val[0][ mb_x*2 + mb_y*2*s->b8_stride ][1];
+            } else {
+                mv_type     = MV_TYPE_16X16;
+                s->mv[0][0][0] = s->cur_pic.motion_val[dir][mb_x * 2 + mb_y * 2 * s->b8_stride][0];
+                s->mv[0][0][1] = s->cur_pic.motion_val[dir][mb_x * 2 + mb_y * 2 * s->b8_stride][1];
             }
-        
-           s->dsp.clear_blocks(s->block[0]);
 
-            s->mb_x= mb_x;
-            s->mb_y= mb_y;
-            decode_mb(s);
+            s->decode_mb(s->opaque, 0 /* FIXME h264 partitioned slices need this set */,
+                         mv_dir, mv_type, &s->mv, mb_x, mb_y, 0, 0);
         }
     }
 
     /* guess MVs */
-    if(s->pict_type==B_TYPE){
-        for(mb_y=0; mb_y<s->mb_height; mb_y++){
-            for(mb_x=0; mb_x<s->mb_width; mb_x++){
-                int xy= mb_x*2 + mb_y*2*s->b8_stride;
-                const int mb_xy= mb_x + mb_y * s->mb_stride;
-                const int mb_type= s->current_picture.mb_type[mb_xy];
-                error= s->error_status_table[mb_xy];
-
-                if(IS_INTRA(mb_type)) continue;
-                if(!(error&MV_ERROR)) continue;           //inter with undamaged MV
-                if(!(error&AC_ERROR)) continue;           //undamaged inter
-            
-                s->mv_dir = MV_DIR_FORWARD|MV_DIR_BACKWARD;
-                s->mb_intra=0;
-                s->mv_type = MV_TYPE_16X16;
-                s->mb_skiped=0;
-                
-                if(s->pp_time){
-                    int time_pp= s->pp_time;
-                    int time_pb= s->pb_time;
-            
-                    s->mv[0][0][0] = s->next_picture.motion_val[0][xy][0]*time_pb/time_pp;
-                    s->mv[0][0][1] = s->next_picture.motion_val[0][xy][1]*time_pb/time_pp;
-                    s->mv[1][0][0] = s->next_picture.motion_val[0][xy][0]*(time_pb - time_pp)/time_pp;
-                    s->mv[1][0][1] = s->next_picture.motion_val[0][xy][1]*(time_pb - time_pp)/time_pp;
-                }else{
-                    s->mv[0][0][0]= 0;
-                    s->mv[0][0][1]= 0;
-                    s->mv[1][0][0]= 0;
-                    s->mv[1][0][1]= 0;
+    if (s->cur_pic.f->pict_type == AV_PICTURE_TYPE_B) {
+        for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
+            for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
+                int       xy      = mb_x * 2 + mb_y * 2 * s->b8_stride;
+                const int mb_xy   = mb_x + mb_y * s->mb_stride;
+                const int mb_type = s->cur_pic.mb_type[mb_xy];
+                int mv_dir = MV_DIR_FORWARD | MV_DIR_BACKWARD;
+
+                error = s->error_status_table[mb_xy];
+
+                if (IS_INTRA(mb_type))
+                    continue;
+                if (!(error & ER_MV_ERROR))
+                    continue; // inter with undamaged MV
+                if (!(error & ER_AC_ERROR))
+                    continue; // undamaged inter
+
+                if (!(s->last_pic.f && s->last_pic.f->data[0]))
+                    mv_dir &= ~MV_DIR_FORWARD;
+                if (!(s->next_pic.f && s->next_pic.f->data[0]))
+                    mv_dir &= ~MV_DIR_BACKWARD;
+
+                if (s->pp_time) {
+                    int time_pp = s->pp_time;
+                    int time_pb = s->pb_time;
+
+                    ff_thread_await_progress(s->next_pic.tf, mb_y, 0);
+
+                    s->mv[0][0][0] = s->next_pic.motion_val[0][xy][0] *  time_pb            / time_pp;
+                    s->mv[0][0][1] = s->next_pic.motion_val[0][xy][1] *  time_pb            / time_pp;
+                    s->mv[1][0][0] = s->next_pic.motion_val[0][xy][0] * (time_pb - time_pp) / time_pp;
+                    s->mv[1][0][1] = s->next_pic.motion_val[0][xy][1] * (time_pb - time_pp) / time_pp;
+                } else {
+                    s->mv[0][0][0] = 0;
+                    s->mv[0][0][1] = 0;
+                    s->mv[1][0][0] = 0;
+                    s->mv[1][0][1] = 0;
                 }
 
-                s->dsp.clear_blocks(s->block[0]);
-                s->mb_x= mb_x;
-                s->mb_y= mb_y;
-                decode_mb(s);
+                s->decode_mb(s->opaque, 0, mv_dir, MV_TYPE_16X16, &s->mv,
+                             mb_x, mb_y, 0, 0);
             }
         }
-    }else
+    } else
         guess_mv(s);
 
-#ifdef HAVE_XVMC
+#if FF_API_XVMC
+FF_DISABLE_DEPRECATION_WARNINGS
     /* the filters below are not XvMC compatible, skip them */
-    if(s->avctx->xvmc_acceleration) goto ec_clean;
-#endif
+    if (CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration)
+        goto ec_clean;
+FF_ENABLE_DEPRECATION_WARNINGS
+#endif /* FF_API_XVMC */
     /* fill DC for inter blocks */
-    for(mb_y=0; mb_y<s->mb_height; mb_y++){
-        for(mb_x=0; mb_x<s->mb_width; mb_x++){
+    for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
+        for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
             int dc, dcu, dcv, y, n;
             int16_t *dc_ptr;
             uint8_t *dest_y, *dest_cb, *dest_cr;
-            const int mb_xy= mb_x + mb_y * s->mb_stride;
-            const int mb_type= s->current_picture.mb_type[mb_xy];
-           
-            error= s->error_status_table[mb_xy];
-
-            if(IS_INTRA(mb_type) && s->partitioned_frame) continue;
-//            if(error&MV_ERROR) continue; //inter data damaged FIXME is this good?
-            
-            dest_y = s->current_picture.data[0] + mb_x*16 + mb_y*16*s->linesize;
-            dest_cb= s->current_picture.data[1] + mb_x*8  + mb_y*8 *s->uvlinesize;
-            dest_cr= s->current_picture.data[2] + mb_x*8  + mb_y*8 *s->uvlinesize;
-           
-            dc_ptr= &s->dc_val[0][mb_x*2 + mb_y*2*s->b8_stride];
-            for(n=0; n<4; n++){
-                dc=0;
-                for(y=0; y<8; y++){
+            const int mb_xy   = mb_x + mb_y * s->mb_stride;
+            const int mb_type = s->cur_pic.mb_type[mb_xy];
+
+            error = s->error_status_table[mb_xy];
+
+            if (IS_INTRA(mb_type) && s->partitioned_frame)
+                continue;
+            // if (error & ER_MV_ERROR)
+            //     continue; // inter data damaged FIXME is this good?
+
+            dest_y  = s->cur_pic.f->data[0] + mb_x * 16 + mb_y * 16 * linesize[0];
+            dest_cb = s->cur_pic.f->data[1] + mb_x *  8 + mb_y *  8 * linesize[1];
+            dest_cr = s->cur_pic.f->data[2] + mb_x *  8 + mb_y *  8 * linesize[2];
+
+            dc_ptr = &s->dc_val[0][mb_x * 2 + mb_y * 2 * s->b8_stride];
+            for (n = 0; n < 4; n++) {
+                dc = 0;
+                for (y = 0; y < 8; y++) {
                     int x;
-                    for(x=0; x<8; x++){
-                       dc+= dest_y[x + (n&1)*8 + (y + (n>>1)*8)*s->linesize];
-                    }
+                    for (x = 0; x < 8; x++)
+                       dc += dest_y[x + (n & 1) * 8 +
+                             (y + (n >> 1) * 8) * linesize[0]];
                 }
-                dc_ptr[(n&1) + (n>>1)*s->b8_stride]= (dc+4)>>3;
+                dc_ptr[(n & 1) + (n >> 1) * s->b8_stride] = (dc + 4) >> 3;
             }
 
-            dcu=dcv=0;
-            for(y=0; y<8; y++){
+            dcu = dcv = 0;
+            for (y = 0; y < 8; y++) {
                 int x;
-                for(x=0; x<8; x++){
-                    dcu+=dest_cb[x + y*(s->uvlinesize)];
-                    dcv+=dest_cr[x + y*(s->uvlinesize)];
+                for (x = 0; x < 8; x++) {
+                    dcu += dest_cb[x + y * linesize[1]];
+                    dcv += dest_cr[x + y * linesize[2]];
                 }
             }
-            s->dc_val[1][mb_x + mb_y*s->mb_stride]= (dcu+4)>>3;
-            s->dc_val[2][mb_x + mb_y*s->mb_stride]= (dcv+4)>>3;   
+            s->dc_val[1][mb_x + mb_y * s->mb_stride] = (dcu + 4) >> 3;
+            s->dc_val[2][mb_x + mb_y * s->mb_stride] = (dcv + 4) >> 3;
         }
     }
-#if 1
+
     /* guess DC for damaged blocks */
-    guess_dc(s, s->dc_val[0], s->mb_width*2, s->mb_height*2, s->b8_stride, 1);
-    guess_dc(s, s->dc_val[1], s->mb_width  , s->mb_height  , s->mb_stride, 0);
-    guess_dc(s, s->dc_val[2], s->mb_width  , s->mb_height  , s->mb_stride, 0);
-#endif   
+    guess_dc(s, s->dc_val[0], s->mb_width * 2, s->mb_height * 2, s->b8_stride, 1);
+    guess_dc(s, s->dc_val[1], s->mb_width, s->mb_height, s->mb_stride, 0);
+    guess_dc(s, s->dc_val[2], s->mb_width, s->mb_height, s->mb_stride, 0);
+
     /* filter luma DC */
-    filter181(s->dc_val[0], s->mb_width*2, s->mb_height*2, s->b8_stride);
-    
-#if 1
+    filter181(s->dc_val[0], s->mb_width * 2, s->mb_height * 2, s->b8_stride);
+
     /* render DC only intra */
-    for(mb_y=0; mb_y<s->mb_height; mb_y++){
-        for(mb_x=0; mb_x<s->mb_width; mb_x++){
+    for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
+        for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
             uint8_t *dest_y, *dest_cb, *dest_cr;
-            const int mb_xy= mb_x + mb_y * s->mb_stride;
-            const int mb_type= s->current_picture.mb_type[mb_xy];
-
-            error= s->error_status_table[mb_xy];
-
-            if(IS_INTER(mb_type)) continue;
-            if(!(error&AC_ERROR)) continue;              //undamaged
-            
-            dest_y = s->current_picture.data[0] + mb_x*16 + mb_y*16*s->linesize;
-            dest_cb= s->current_picture.data[1] + mb_x*8  + mb_y*8 *s->uvlinesize;
-            dest_cr= s->current_picture.data[2] + mb_x*8  + mb_y*8 *s->uvlinesize;
-            
+            const int mb_xy   = mb_x + mb_y * s->mb_stride;
+            const int mb_type = s->cur_pic.mb_type[mb_xy];
+
+            error = s->error_status_table[mb_xy];
+
+            if (IS_INTER(mb_type))
+                continue;
+            if (!(error & ER_AC_ERROR))
+                continue; // undamaged
+
+            dest_y  = s->cur_pic.f->data[0] + mb_x * 16 + mb_y * 16 * linesize[0];
+            dest_cb = s->cur_pic.f->data[1] + mb_x *  8 + mb_y *  8 * linesize[1];
+            dest_cr = s->cur_pic.f->data[2] + mb_x *  8 + mb_y *  8 * linesize[2];
+
             put_dc(s, dest_y, dest_cb, dest_cr, mb_x, mb_y);
         }
     }
-#endif
-    
-    if(s->avctx->error_concealment&FF_EC_DEBLOCK){
+
+    if (s->avctx->error_concealment & FF_EC_DEBLOCK) {
         /* filter horizontal block boundaries */
-        h_block_filter(s, s->current_picture.data[0], s->mb_width*2, s->mb_height*2, s->linesize  , 1);
-        h_block_filter(s, s->current_picture.data[1], s->mb_width  , s->mb_height  , s->uvlinesize, 0);
-        h_block_filter(s, s->current_picture.data[2], s->mb_width  , s->mb_height  , s->uvlinesize, 0);
+        h_block_filter(s, s->cur_pic.f->data[0], s->mb_width * 2,
+                       s->mb_height * 2, linesize[0], 1);
+        h_block_filter(s, s->cur_pic.f->data[1], s->mb_width,
+                       s->mb_height, linesize[1], 0);
+        h_block_filter(s, s->cur_pic.f->data[2], s->mb_width,
+                       s->mb_height, linesize[2], 0);
 
         /* filter vertical block boundaries */
-        v_block_filter(s, s->current_picture.data[0], s->mb_width*2, s->mb_height*2, s->linesize  , 1);
-        v_block_filter(s, s->current_picture.data[1], s->mb_width  , s->mb_height  , s->uvlinesize, 0);
-        v_block_filter(s, s->current_picture.data[2], s->mb_width  , s->mb_height  , s->uvlinesize, 0);
+        v_block_filter(s, s->cur_pic.f->data[0], s->mb_width * 2,
+                       s->mb_height * 2, linesize[0], 1);
+        v_block_filter(s, s->cur_pic.f->data[1], s->mb_width,
+                       s->mb_height, linesize[1], 0);
+        v_block_filter(s, s->cur_pic.f->data[2], s->mb_width,
+                       s->mb_height, linesize[2], 0);
     }
 
-#ifdef HAVE_XVMC
 ec_clean:
-#endif
     /* clean a few tables */
-    for(i=0; i<s->mb_num; i++){
-        const int mb_xy= s->mb_index2xy[i];
-        int error= s->error_status_table[mb_xy];
-        
-        if(s->pict_type!=B_TYPE && (error&(DC_ERROR|MV_ERROR|AC_ERROR))){
-            s->mbskip_table[mb_xy]=0;
+    for (i = 0; i < s->mb_num; i++) {
+        const int mb_xy = s->mb_index2xy[i];
+        int       error = s->error_status_table[mb_xy];
+
+        if (s->mbskip_table && s->cur_pic.f->pict_type != AV_PICTURE_TYPE_B &&
+            (error & (ER_DC_ERROR | ER_MV_ERROR | ER_AC_ERROR))) {
+            s->mbskip_table[mb_xy] = 0;
         }
-        s->mbintra_table[mb_xy]=1;
-    }    
+        if (s->mbintra_table)
+            s->mbintra_table[mb_xy] = 1;
+    }
+
+    memset(&s->cur_pic, 0, sizeof(ERPicture));
+    memset(&s->last_pic, 0, sizeof(ERPicture));
+    memset(&s->next_pic, 0, sizeof(ERPicture));
 }