]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/mpegvideo.c
h263 loop filter
[ffmpeg] / libavcodec / mpegvideo.c
index 737cd0d1c8f5bba48bcbee2b63d1768973382a1d..595ab16c2f53b838ddbb57cd304832becaf187ae 100644 (file)
  * The simplest mpeg encoder (well, it was the simplest!).
  */ 
  
-#include <ctype.h>
 #include <limits.h>
 #include "avcodec.h"
 #include "dsputil.h"
 #include "mpegvideo.h"
+#include "faandct.h"
 
 #ifdef USE_FASTMEMCPY
 #include "fastmemcpy.h"
@@ -50,12 +50,13 @@ 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
 extern int  XVMC_field_start(MpegEncContext*s, AVCodecContext *avctx);
 extern void XVMC_field_end(MpegEncContext *s);
-extern void XVMC_decode_mb(MpegEncContext *s, DCTELEM block[6][64]);
+extern void XVMC_decode_mb(MpegEncContext *s);
 #endif
 
 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w)= draw_edges_c;
@@ -87,22 +88,31 @@ static const uint8_t h263_chroma_roundtab[16] = {
     0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2,
 };
 
+static const uint8_t ff_default_chroma_qscale_table[32]={
+//  0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
+    0, 1, 2, 3, 4, 5, 6, 7, 8, 9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31
+};
+
 #ifdef CONFIG_ENCODERS
 static uint8_t (*default_mv_penalty)[MAX_MV*2+1]=NULL;
 static uint8_t default_fcode_tab[MAX_MV*2+1];
 
 enum PixelFormat ff_yuv420p_list[2]= {PIX_FMT_YUV420P, -1};
 
-static void convert_matrix(MpegEncContext *s, int (*qmat)[64], uint16_t (*qmat16)[64], uint16_t (*qmat16_bias)[64],
+static void convert_matrix(DSPContext *dsp, int (*qmat)[64], uint16_t (*qmat16)[2][64],
                            const uint16_t *quant_matrix, int bias, int qmin, int qmax)
 {
     int qscale;
 
     for(qscale=qmin; qscale<=qmax; qscale++){
         int i;
-        if (s->dsp.fdct == ff_jpeg_fdct_islow) {
+        if (dsp->fdct == ff_jpeg_fdct_islow 
+#ifdef FAAN_POSTSCALE
+            || dsp->fdct == ff_faandct
+#endif
+            ) {
             for(i=0;i<64;i++) {
-                const int j= s->dsp.idct_permutation[i];
+                const int j= dsp->idct_permutation[i];
                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
                 /* 19952         <= aanscales[i] * qscale * quant_matrix[i]           <= 249205026 */
                 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
@@ -111,9 +121,13 @@ static void convert_matrix(MpegEncContext *s, int (*qmat)[64], uint16_t (*qmat16
                 qmat[qscale][i] = (int)((uint64_t_C(1) << QMAT_SHIFT) / 
                                 (qscale * quant_matrix[j]));
             }
-        } else if (s->dsp.fdct == fdct_ifast) {
+        } else if (dsp->fdct == fdct_ifast
+#ifndef FAAN_POSTSCALE
+                   || dsp->fdct == ff_faandct
+#endif
+                   ) {
             for(i=0;i<64;i++) {
-                const int j= s->dsp.idct_permutation[i];
+                const int j= dsp->idct_permutation[i];
                 /* 16 <= qscale * quant_matrix[i] <= 7905 */
                 /* 19952         <= aanscales[i] * qscale * quant_matrix[i]           <= 249205026 */
                 /* (1<<36)/19952 >= (1<<36)/(aanscales[i] * qscale * quant_matrix[i]) >= (1<<36)/249205026 */
@@ -124,7 +138,7 @@ static void convert_matrix(MpegEncContext *s, int (*qmat)[64], uint16_t (*qmat16
             }
         } else {
             for(i=0;i<64;i++) {
-                const int j= s->dsp.idct_permutation[i];
+                const int j= dsp->idct_permutation[i];
                 /* We can safely suppose that 16 <= quant_matrix[i] <= 255
                    So 16           <= qscale * quant_matrix[i]             <= 7905
                    so (1<<19) / 16 >= (1<<19) / (qscale * quant_matrix[i]) >= (1<<19) / 7905
@@ -132,14 +146,21 @@ static void convert_matrix(MpegEncContext *s, int (*qmat)[64], uint16_t (*qmat16
                 */
                 qmat[qscale][i] = (int)((uint64_t_C(1) << QMAT_SHIFT) / (qscale * quant_matrix[j]));
 //                qmat  [qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[i]);
-                qmat16[qscale][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
+                qmat16[qscale][0][i] = (1 << QMAT_SHIFT_MMX) / (qscale * quant_matrix[j]);
 
-                if(qmat16[qscale][i]==0 || qmat16[qscale][i]==128*256) qmat16[qscale][i]=128*256-1;
-                qmat16_bias[qscale][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][i]);
+                if(qmat16[qscale][0][i]==0 || qmat16[qscale][0][i]==128*256) qmat16[qscale][0][i]=128*256-1;
+                qmat16[qscale][1][i]= ROUNDED_DIV(bias<<(16-QUANT_BIAS_SHIFT), qmat16[qscale][0][i]);
             }
         }
     }
 }
+
+static inline void update_qscale(MpegEncContext *s){
+    s->qscale= (s->lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
+    s->qscale= clip(s->qscale, s->avctx->qmin, s->avctx->qmax);
+    
+    s->lambda2= (s->lambda*s->lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
+}
 #endif //CONFIG_ENCODERS
 
 void ff_init_scantable(uint8_t *permutation, ScanTable *st, const uint8_t *src_scantable){
@@ -166,6 +187,20 @@ void ff_init_scantable(uint8_t *permutation, ScanTable *st, const uint8_t *src_s
     }
 }
 
+#ifdef CONFIG_ENCODERS
+void ff_write_quant_matrix(PutBitContext *pb, int16_t *matrix){
+    int i;
+
+    if(matrix){
+        put_bits(pb, 1, 1);
+        for(i=0;i<64;i++) {
+            put_bits(pb, 8, matrix[ ff_zigzag_direct[i] ]);
+        }
+    }else
+        put_bits(pb, 1, 0);
+}
+#endif //CONFIG_ENCODERS
+
 /* init common dct for both encoder and decoder */
 int DCT_common_init(MpegEncContext *s)
 {
@@ -218,6 +253,11 @@ int DCT_common_init(MpegEncContext *s)
     return 0;
 }
 
+static void copy_picture(Picture *dst, Picture *src){
+    *dst = *src;
+    dst->type= FF_BUFFER_TYPE_COPY;
+}
+
 /**
  * allocates a Picture
  * The pixels are allocated/set by calling get_buffer() if shared=0
@@ -239,17 +279,17 @@ static int alloc_picture(MpegEncContext *s, Picture *pic, int shared){
         r= s->avctx->get_buffer(s->avctx, (AVFrame*)pic);
         
         if(r<0 || !pic->age || !pic->type || !pic->data[0]){
-            fprintf(stderr, "get_buffer() failed (%d %d %d %p)\n", r, pic->age, pic->type, pic->data[0]);
+           av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (%d %d %d %p)\n", r, pic->age, pic->type, pic->data[0]);
             return -1;
         }
 
         if(s->linesize && (s->linesize != pic->linesize[0] || s->uvlinesize != pic->linesize[1])){
-            fprintf(stderr, "get_buffer() failed (stride changed)\n");
+            av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (stride changed)\n");
             return -1;
         }
 
         if(pic->linesize[1] != pic->linesize[2]){
-            fprintf(stderr, "get_buffer() failed (uv stride missmatch)\n");
+            av_log(s->avctx, AV_LOG_ERROR, "get_buffer() failed (uv stride missmatch)\n");
             return -1;
         }
 
@@ -262,7 +302,6 @@ static int alloc_picture(MpegEncContext *s, Picture *pic, int shared){
             CHECKED_ALLOCZ(pic->mb_var   , mb_array_size * sizeof(int16_t))
             CHECKED_ALLOCZ(pic->mc_mb_var, mb_array_size * sizeof(int16_t))
             CHECKED_ALLOCZ(pic->mb_mean  , mb_array_size * sizeof(int8_t))
-            CHECKED_ALLOCZ(pic->mb_cmp_score, mb_array_size * sizeof(int32_t))
         }
 
         CHECKED_ALLOCZ(pic->mbskip_table , mb_array_size * sizeof(uint8_t)+2) //the +2 is for the slice end check
@@ -276,6 +315,7 @@ static int alloc_picture(MpegEncContext *s, Picture *pic, int shared){
             }
         }
         pic->qstride= s->mb_stride;
+        CHECKED_ALLOCZ(pic->pan_scan , 1 * sizeof(AVPanScan))
     }
 
     //it might be nicer if the application would keep track of these but it would require a API change
@@ -302,10 +342,10 @@ static void free_picture(MpegEncContext *s, Picture *pic){
     av_freep(&pic->mb_var);
     av_freep(&pic->mc_mb_var);
     av_freep(&pic->mb_mean);
-    av_freep(&pic->mb_cmp_score);
     av_freep(&pic->mbskip_table);
     av_freep(&pic->qscale_table);
     av_freep(&pic->mb_type_base);
+    av_freep(&pic->pan_scan);
     pic->mb_type= NULL;
     for(i=0; i<2; i++){
         av_freep(&pic->motion_val[i]);
@@ -349,6 +389,10 @@ int MPV_common_init(MpegEncContext *s)
     s->block_wrap[4]=
     s->block_wrap[5]= s->mb_width + 2;
 
+    s->y_dc_scale_table=
+    s->c_dc_scale_table= ff_mpeg1_dc_scale_table;
+    s->chroma_qscale_table= ff_default_chroma_qscale_table;    
+    
     y_size = (2 * s->mb_width + 2) * (2 * s->mb_height + 2);
     c_size = (s->mb_width + 2) * (s->mb_height + 2);
     yc_size = y_size + 2 * c_size;
@@ -359,6 +403,11 @@ int MPV_common_init(MpegEncContext *s)
                         + (toupper((s->avctx->codec_tag>>16)&0xFF)<<16) 
                         + (toupper((s->avctx->codec_tag>>24)&0xFF)<<24);
 
+    s->avctx->stream_codec_tag=   toupper( s->avctx->stream_codec_tag     &0xFF)          
+                               + (toupper((s->avctx->stream_codec_tag>>8 )&0xFF)<<8 )
+                               + (toupper((s->avctx->stream_codec_tag>>16)&0xFF)<<16) 
+                               + (toupper((s->avctx->stream_codec_tag>>24)&0xFF)<<24);
+
     CHECKED_ALLOCZ(s->allocated_edge_emu_buffer, (s->width+64)*2*17*2); //(width + edge + align)*interlaced*MBsize*tolerance
     s->edge_emu_buffer= s->allocated_edge_emu_buffer + (s->width+64)*2*17;
 
@@ -407,11 +456,28 @@ int MPV_common_init(MpegEncContext *s)
 
         /* Allocate MB type table */
         CHECKED_ALLOCZ(s->mb_type  , mb_array_size * sizeof(uint8_t)) //needed for encoding
+        
+        CHECKED_ALLOCZ(s->lambda_table, mb_array_size * sizeof(int))
+        
+        CHECKED_ALLOCZ(s->q_intra_matrix, 64*32 * sizeof(int))
+        CHECKED_ALLOCZ(s->q_inter_matrix, 64*32 * sizeof(int))
+        CHECKED_ALLOCZ(s->q_intra_matrix16, 64*32*2 * sizeof(uint16_t))
+        CHECKED_ALLOCZ(s->q_inter_matrix16, 64*32*2 * sizeof(uint16_t))
+        CHECKED_ALLOCZ(s->input_picture, MAX_PICTURE_COUNT * sizeof(Picture*))
+        CHECKED_ALLOCZ(s->reordered_input_picture, MAX_PICTURE_COUNT * sizeof(Picture*))
+        
+        if(s->avctx->noise_reduction){
+            CHECKED_ALLOCZ(s->dct_error_sum, 2 * 64 * sizeof(int))
+            CHECKED_ALLOCZ(s->dct_offset, 2 * 64 * sizeof(uint16_t))
+        }
     }
+    CHECKED_ALLOCZ(s->blocks, 64*6*2 * sizeof(DCTELEM))
         
+    CHECKED_ALLOCZ(s->picture, MAX_PICTURE_COUNT * sizeof(Picture))
+
     CHECKED_ALLOCZ(s->error_status_table, mb_array_size*sizeof(uint8_t))
     
-    if (s->out_format == FMT_H263 || s->encoding) {
+    if (s->out_format == FMT_H263 || s->encoding || (s->avctx->debug&FF_DEBUG_VIS_MV)) {
         int size;
 
         /* MV prediction */
@@ -465,6 +531,10 @@ int MPV_common_init(MpegEncContext *s)
     
     s->block= s->blocks[0];
 
+    for(i=0;i<12;i++){
+        s->pblocks[i] = (short *)(&s->block[i]);
+    }
+
     s->parse_context.state= -1;
 
     s->context_initialized = 1;
@@ -522,12 +592,28 @@ void MPV_common_end(MpegEncContext *s)
     av_freep(&s->ac_stats);
     av_freep(&s->error_status_table);
     av_freep(&s->mb_index2xy);
-
-    for(i=0; i<MAX_PICTURE_COUNT; i++){
-        free_picture(s, &s->picture[i]);
+    av_freep(&s->lambda_table);
+    av_freep(&s->q_intra_matrix);
+    av_freep(&s->q_inter_matrix);
+    av_freep(&s->q_intra_matrix16);
+    av_freep(&s->q_inter_matrix16);
+    av_freep(&s->blocks);
+    av_freep(&s->input_picture);
+    av_freep(&s->reordered_input_picture);
+    av_freep(&s->dct_error_sum);
+    av_freep(&s->dct_offset);
+
+    if(s->picture){
+        for(i=0; i<MAX_PICTURE_COUNT; i++){
+            free_picture(s, &s->picture[i]);
+        }
     }
+    av_freep(&s->picture);
     avcodec_default_free_buffers(s->avctx);
     s->context_initialized = 0;
+    s->last_picture_ptr=
+    s->next_picture_ptr=
+    s->current_picture_ptr= NULL;
 }
 
 #ifdef CONFIG_ENCODERS
@@ -536,15 +622,17 @@ void MPV_common_end(MpegEncContext *s)
 int MPV_encode_init(AVCodecContext *avctx)
 {
     MpegEncContext *s = avctx->priv_data;
-    int i;
+    int i, dummy;
     int chroma_h_shift, chroma_v_shift;
 
+    avctx->pix_fmt = PIX_FMT_YUV420P; // FIXME
+
     s->bit_rate = avctx->bit_rate;
     s->bit_rate_tolerance = avctx->bit_rate_tolerance;
     s->width = avctx->width;
     s->height = avctx->height;
     if(avctx->gop_size > 600){
-        fprintf(stderr, "Warning keyframe interval too large! reducing it ...\n");
+       av_log(avctx, AV_LOG_ERROR, "Warning keyframe interval too large! reducing it ...\n");
         avctx->gop_size=600;
     }
     s->gop_size = avctx->gop_size;
@@ -583,40 +671,65 @@ int MPV_encode_init(AVCodecContext *avctx)
                         || s->avctx->dark_masking
                         || s->avctx->temporal_cplx_masking 
                         || s->avctx->spatial_cplx_masking
-                        || s->avctx->p_masking)
+                        || s->avctx->p_masking
+                        || (s->flags&CODEC_FLAG_QP_RD))
                        && !s->fixed_qscale;
     
     s->progressive_sequence= !(avctx->flags & CODEC_FLAG_INTERLACED_DCT);
+    
+    s->obmc= (s->flags & CODEC_FLAG_OBMC);
+    s->loop_filter= (s->flags & CODEC_FLAG_LOOP_FILTER);
 
-    if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4){
-        fprintf(stderr, "4MV not supporetd by codec\n");
+    if((s->flags & CODEC_FLAG_4MV) && s->codec_id != CODEC_ID_MPEG4 
+       && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
+        av_log(avctx, AV_LOG_ERROR, "4MV not supported by codec\n");
+        return -1;
+    }
+    
+    if(s->obmc && s->avctx->mb_decision != FF_MB_DECISION_SIMPLE){
+        av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with simple mb decission\n");
+        return -1;
+    }
+    
+    if(s->obmc && s->codec_id != CODEC_ID_H263 && s->codec_id != CODEC_ID_H263P){
+        av_log(avctx, AV_LOG_ERROR, "OBMC is only supported with H263(+)\n");
         return -1;
     }
     
     if(s->quarter_sample && s->codec_id != CODEC_ID_MPEG4){
-        fprintf(stderr, "qpel not supporetd by codec\n");
+        av_log(avctx, AV_LOG_ERROR, "qpel not supported by codec\n");
         return -1;
     }
 
     if(s->data_partitioning && s->codec_id != CODEC_ID_MPEG4){
-        fprintf(stderr, "data partitioning not supporetd by codec\n");
+        av_log(avctx, AV_LOG_ERROR, "data partitioning not supported by codec\n");
         return -1;
     }
     
-    if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO){
-        fprintf(stderr, "b frames not supporetd by codec\n");
+    if(s->max_b_frames && s->codec_id != CODEC_ID_MPEG4 && s->codec_id != CODEC_ID_MPEG1VIDEO && s->codec_id != CODEC_ID_MPEG2VIDEO){
+        av_log(avctx, AV_LOG_ERROR, "b frames not supported by codec\n");
         return -1;
     }
     
     if(s->mpeg_quant && s->codec_id != CODEC_ID_MPEG4){ //FIXME mpeg2 uses that too
-        fprintf(stderr, "mpeg2 style quantization not supporetd by codec\n");
+        av_log(avctx, AV_LOG_ERROR, "mpeg2 style quantization not supporetd by codec\n");
         return -1;
     }
         
+    if((s->flags & CODEC_FLAG_CBP_RD) && !(s->flags & CODEC_FLAG_TRELLIS_QUANT)){
+        av_log(avctx, AV_LOG_ERROR, "CBP RD needs trellis quant\n");
+        return -1;
+    }
+
+    if((s->flags & CODEC_FLAG_QP_RD) && s->avctx->mb_decision != FF_MB_DECISION_RD){
+        av_log(avctx, AV_LOG_ERROR, "QP RD needs mbd=2\n");
+        return -1;
+    }
+    
     if(s->codec_id==CODEC_ID_MJPEG){
         s->intra_quant_bias= 1<<(QUANT_BIAS_SHIFT-1); //(a + x/2)/x
         s->inter_quant_bias= 0;
-    }else if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO){
+    }else if(s->mpeg_quant || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO){
         s->intra_quant_bias= 3<<(QUANT_BIAS_SHIFT-3); //(a + x*3/8)/x
         s->inter_quant_bias= 0;
     }else{
@@ -631,12 +744,22 @@ int MPV_encode_init(AVCodecContext *avctx)
         
     avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_h_shift, &chroma_v_shift);
 
+    av_reduce(&s->time_increment_resolution, &dummy, s->avctx->frame_rate, s->avctx->frame_rate_base, (1<<16)-1);
+    s->time_increment_bits = av_log2(s->time_increment_resolution - 1) + 1;
+
     switch(avctx->codec->id) {
     case CODEC_ID_MPEG1VIDEO:
         s->out_format = FMT_MPEG1;
         s->low_delay= 0; //s->max_b_frames ? 0 : 1;
         avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
         break;
+    case CODEC_ID_MPEG2VIDEO:
+        s->out_format = FMT_MPEG1;
+        s->low_delay= 0; //s->max_b_frames ? 0 : 1;
+        avctx->delay= s->low_delay ? 0 : (s->max_b_frames + 1);
+        s->rtp_mode= 1; // mpeg2 must have slices
+        if(s->rtp_payload_size == 0) s->rtp_payload_size= 256*256*256;
+        break;
     case CODEC_ID_LJPEG:
     case CODEC_ID_MJPEG:
         s->out_format = FMT_MJPEG;
@@ -657,10 +780,11 @@ int MPV_encode_init(AVCodecContext *avctx)
 #ifdef CONFIG_RISKY
     case CODEC_ID_H263:
         if (h263_get_picture_format(s->width, s->height) == 7) {
-            printf("Input picture size isn't suitable for h263 codec! try h263+\n");
+            av_log(avctx, AV_LOG_INFO, "Input picture size isn't suitable for h263 codec! try h263+\n");
             return -1;
         }
         s->out_format = FMT_H263;
+       s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
         avctx->delay=0;
         s->low_delay=1;
         break;
@@ -668,11 +792,18 @@ int MPV_encode_init(AVCodecContext *avctx)
         s->out_format = FMT_H263;
         s->h263_plus = 1;
        /* Fx */
-       s->unrestricted_mv=(avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
+        s->umvplus = (avctx->flags & CODEC_FLAG_H263P_UMV) ? 1:0;
        s->h263_aic= (avctx->flags & CODEC_FLAG_H263P_AIC) ? 1:0;
+       s->modified_quant= s->h263_aic;
+       s->alt_inter_vlc= (avctx->flags & CODEC_FLAG_H263P_AIV) ? 1:0;
+       s->obmc= (avctx->flags & CODEC_FLAG_OBMC) ? 1:0;
+       s->loop_filter= (avctx->flags & CODEC_FLAG_LOOP_FILTER) ? 1:0;
+       s->unrestricted_mv= s->obmc || s->loop_filter || s->umvplus;
+        if(s->modified_quant)
+            s->chroma_qscale_table= ff_h263_chroma_qscale_table;
+
        /* /Fx */
         /* These are just to be sure */
-        s->umvplus = 1;
         avctx->delay=0;
         s->low_delay=1;
         break;
@@ -686,7 +817,6 @@ int MPV_encode_init(AVCodecContext *avctx)
         break;
     case CODEC_ID_RV10:
         s->out_format = FMT_H263;
-        s->h263_rv10 = 1;
         avctx->delay=0;
         s->low_delay=1;
         break;
@@ -767,8 +897,6 @@ int MPV_encode_init(AVCodecContext *avctx)
     }
     s->me.mv_penalty= default_mv_penalty;
     s->fcode_tab= default_fcode_tab;
-    s->y_dc_scale_table=
-    s->c_dc_scale_table= ff_mpeg1_dc_scale_table;
  
     /* dont use mv_penalty table for crap MV as it would be confused */
     //FIXME remove after fixing / removing old ME
@@ -805,18 +933,22 @@ int MPV_encode_init(AVCodecContext *avctx)
             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
         }else
 #endif
-        { /* mpeg1 */
+        { /* mpeg1/2 */
             s->intra_matrix[j] = ff_mpeg1_default_intra_matrix[i];
             s->inter_matrix[j] = ff_mpeg1_default_non_intra_matrix[i];
         }
+        if(s->avctx->intra_matrix)
+            s->intra_matrix[j] = s->avctx->intra_matrix[i];
+        if(s->avctx->inter_matrix)
+            s->inter_matrix[j] = s->avctx->inter_matrix[i];
     }
 
     /* precompute matrix */
     /* for mjpeg, we do include qscale in the matrix */
     if (s->out_format != FMT_MJPEG) {
-        convert_matrix(s, s->q_intra_matrix, s->q_intra_matrix16, s->q_intra_matrix16_bias
+        convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16
                        s->intra_matrix, s->intra_quant_bias, 1, 31);
-        convert_matrix(s, s->q_inter_matrix, s->q_inter_matrix16, s->q_inter_matrix16_bias
+        convert_matrix(&s->dsp, s->q_inter_matrix, s->q_inter_matrix16
                        s->inter_matrix, s->inter_quant_bias, 1, 31);
     }
 
@@ -846,6 +978,8 @@ int MPV_encode_end(AVCodecContext *avctx)
     MPV_common_end(s);
     if (s->out_format == FMT_MJPEG)
         mjpeg_close(s);
+
+    av_freep(&avctx->extradata);
       
     return 0;
 }
@@ -919,38 +1053,56 @@ static void draw_edges_c(uint8_t *buf, int wrap, int width, int height, int w)
     }
 }
 
-static int find_unused_picture(MpegEncContext *s, int shared){
+int ff_find_unused_picture(MpegEncContext *s, int shared){
     int i;
     
     if(shared){
         for(i=0; i<MAX_PICTURE_COUNT; i++){
-            if(s->picture[i].data[0]==NULL && s->picture[i].type==0) break;
+            if(s->picture[i].data[0]==NULL && s->picture[i].type==0) return i;
         }
     }else{
         for(i=0; i<MAX_PICTURE_COUNT; i++){
-            if(s->picture[i].data[0]==NULL && s->picture[i].type!=0) break; //FIXME
+            if(s->picture[i].data[0]==NULL && s->picture[i].type!=0) return i; //FIXME
         }
         for(i=0; i<MAX_PICTURE_COUNT; i++){
-            if(s->picture[i].data[0]==NULL) break;
+            if(s->picture[i].data[0]==NULL) return i;
         }
     }
 
-    assert(i<MAX_PICTURE_COUNT);
-    return i;
+    assert(0);
+    return -1;
 }
 
-/* generic function for encode/decode called before a frame is coded/decoded */
+static void update_noise_reduction(MpegEncContext *s){
+    int intra, i;
+
+    for(intra=0; intra<2; intra++){
+        if(s->dct_count[intra] > (1<<16)){
+            for(i=0; i<64; i++){
+                s->dct_error_sum[intra][i] >>=1;
+            }
+            s->dct_count[intra] >>= 1;
+        }
+        
+        for(i=0; i<64; i++){
+            s->dct_offset[intra][i]= (s->avctx->noise_reduction * s->dct_count[intra] + s->dct_error_sum[intra][i]/2) / (s->dct_error_sum[intra][i]+1);
+        }
+    }
+}
+
+/**
+ * generic function for encode/decode called after coding/decoding the header and before a frame is coded/decoded
+ */
 int MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx)
 {
     int i;
     AVFrame *pic;
-
     s->mb_skiped = 0;
 
     assert(s->last_picture_ptr==NULL || s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3);
 
     /* mark&release old frames */
-    if (s->pict_type != B_TYPE && s->last_picture_ptr) {
+    if (s->pict_type != B_TYPE && s->last_picture_ptr && s->last_picture_ptr->data[0]) {
         avctx->release_buffer(avctx, (AVFrame*)s->last_picture_ptr);
 
         /* release forgotten pictures */
@@ -958,7 +1110,7 @@ int MPV_frame_start(MpegEncContext *s, AVCodecContext *avctx)
         if(!s->encoding){
             for(i=0; i<MAX_PICTURE_COUNT; i++){
                 if(s->picture[i].data[0] && &s->picture[i] != s->next_picture_ptr && s->picture[i].reference){
-                    fprintf(stderr, "releasing zombie picture\n");
+                    av_log(avctx, AV_LOG_ERROR, "releasing zombie picture\n");
                     avctx->release_buffer(avctx, (AVFrame*)&s->picture[i]);                
                 }
             }
@@ -973,24 +1125,30 @@ alloc:
             }
         }
 
-        i= find_unused_picture(s, 0);
-    
-        pic= (AVFrame*)&s->picture[i];
+        if(s->current_picture_ptr && s->current_picture_ptr->data[0]==NULL)
+            pic= (AVFrame*)s->current_picture_ptr; //we allready have a unused image (maybe it was set before reading the header)
+        else{
+            i= ff_find_unused_picture(s, 0);
+            pic= (AVFrame*)&s->picture[i];
+        }
+
         pic->reference= s->pict_type != B_TYPE ? 3 : 0;
 
-        if(s->current_picture_ptr)
+        if(s->current_picture_ptr) //FIXME broken, we need a coded_picture_number in MpegEncContext
             pic->coded_picture_number= s->current_picture_ptr->coded_picture_number+1;
         
-        alloc_picture(s, (Picture*)pic, 0);
+        if( alloc_picture(s, (Picture*)pic, 0) < 0)
+            return -1;
 
-        s->current_picture_ptr= &s->picture[i];
+        s->current_picture_ptr= (Picture*)pic;
     }
 
     s->current_picture_ptr->pict_type= s->pict_type;
-    s->current_picture_ptr->quality= s->qscale;
+//    if(s->flags && CODEC_FLAG_QSCALE) 
+  //      s->current_picture_ptr->quality= s->new_picture_ptr->quality;
     s->current_picture_ptr->key_frame= s->pict_type == I_TYPE;
 
-    s->current_picture= *s->current_picture_ptr;
+    copy_picture(&s->current_picture, s->current_picture_ptr);
   
   if(s->out_format != FMT_H264 || s->codec_id == CODEC_ID_SVQ3){
     if (s->pict_type != B_TYPE) {
@@ -998,10 +1156,17 @@ alloc:
         s->next_picture_ptr= s->current_picture_ptr;
     }
     
-    if(s->last_picture_ptr) s->last_picture= *s->last_picture_ptr;
-    if(s->next_picture_ptr) s->next_picture= *s->next_picture_ptr;
-    if(s->new_picture_ptr ) s->new_picture = *s->new_picture_ptr;
+    if(s->last_picture_ptr) copy_picture(&s->last_picture, s->last_picture_ptr);
+    if(s->next_picture_ptr) copy_picture(&s->next_picture, s->next_picture_ptr);
     
+    if(s->pict_type != I_TYPE && (s->last_picture_ptr==NULL || s->last_picture_ptr->data[0]==NULL)){
+        av_log(avctx, AV_LOG_ERROR, "warning: first frame is no keyframe\n");
+        assert(s->pict_type != B_TYPE); //these should have been dropped if we dont have a reference
+        goto alloc;
+    }
+
+    assert(s->pict_type == I_TYPE || (s->last_picture_ptr && s->last_picture_ptr->data[0]));
+
     if(s->picture_structure!=PICT_FRAME){
         int i;
         for(i=0; i<4; i++){
@@ -1013,12 +1178,6 @@ alloc:
             s->next_picture.linesize[i] *=2;
         }
     }
-    
-    if(s->pict_type != I_TYPE && s->last_picture_ptr==NULL){
-        fprintf(stderr, "warning: first frame is no keyframe\n");
-        assert(s->pict_type != B_TYPE); //these should have been dropped if we dont have a reference
-        goto alloc;
-    }
   }
    
     s->hurry_up= s->avctx->hurry_up;
@@ -1026,14 +1185,19 @@ alloc:
 
     /* set dequantizer, we cant do it during init as it might change for mpeg4
        and we cant do it in the header decode as init isnt called for mpeg4 there yet */
-    if(s->out_format == FMT_H263){
-        if(s->mpeg_quant)
-            s->dct_unquantize = s->dct_unquantize_mpeg2;
-        else
-            s->dct_unquantize = s->dct_unquantize_h263;
-    }else 
+    if(s->mpeg_quant || s->codec_id == CODEC_ID_MPEG2VIDEO) 
+        s->dct_unquantize = s->dct_unquantize_mpeg2;
+    else if(s->out_format == FMT_H263)
+        s->dct_unquantize = s->dct_unquantize_h263;
+    else 
         s->dct_unquantize = s->dct_unquantize_mpeg1;
 
+    if(s->dct_error_sum){
+        assert(s->avctx->noise_reduction && s->encoding);
+
+        update_noise_reduction(s);
+    }
+        
 #ifdef HAVE_XVMC
     if(s->avctx->xvmc_acceleration)
         return XVMC_field_start(s, avctx);
@@ -1052,12 +1216,10 @@ void MPV_frame_end(MpegEncContext *s)
         XVMC_field_end(s);
     }else
 #endif
-    if(s->codec_id!=CODEC_ID_SVQ1 && s->codec_id != CODEC_ID_MPEG1VIDEO){
-        if (s->pict_type != B_TYPE && !s->intra_only && !(s->flags&CODEC_FLAG_EMU_EDGE)) {
+    if(s->unrestricted_mv && s->pict_type != B_TYPE && !s->intra_only && !(s->flags&CODEC_FLAG_EMU_EDGE)) {
             draw_edges(s->current_picture.data[0], s->linesize  , s->h_edge_pos   , s->v_edge_pos   , EDGE_WIDTH  );
             draw_edges(s->current_picture.data[1], s->uvlinesize, s->h_edge_pos>>1, s->v_edge_pos>>1, EDGE_WIDTH/2);
             draw_edges(s->current_picture.data[2], s->uvlinesize, s->h_edge_pos>>1, s->v_edge_pos>>1, EDGE_WIDTH/2);
-        }
     }
     emms_c();
     
@@ -1185,63 +1347,63 @@ void ff_print_debug_info(MpegEncContext *s, Picture *pict){
                 if(s->avctx->debug&FF_DEBUG_SKIP){
                     int count= s->mbskip_table[x + y*s->mb_stride];
                     if(count>9) count=9;
-                    printf("%1d", count);
+                    av_log(s->avctx, AV_LOG_DEBUG, "%1d", count);
                 }
                 if(s->avctx->debug&FF_DEBUG_QP){
-                    printf("%2d", pict->qscale_table[x + y*s->mb_stride]);
+                    av_log(s->avctx, AV_LOG_DEBUG, "%2d", pict->qscale_table[x + y*s->mb_stride]);
                 }
                 if(s->avctx->debug&FF_DEBUG_MB_TYPE){
                     int mb_type= pict->mb_type[x + y*s->mb_stride];
                     
                     //Type & MV direction
                     if(IS_PCM(mb_type))
-                        printf("P");
+                        av_log(s->avctx, AV_LOG_DEBUG, "P");
                     else if(IS_INTRA(mb_type) && IS_ACPRED(mb_type))
-                        printf("A");
+                        av_log(s->avctx, AV_LOG_DEBUG, "A");
                     else if(IS_INTRA4x4(mb_type))
-                        printf("i");
+                        av_log(s->avctx, AV_LOG_DEBUG, "i");
                     else if(IS_INTRA16x16(mb_type))
-                        printf("I");
+                        av_log(s->avctx, AV_LOG_DEBUG, "I");
                     else if(IS_DIRECT(mb_type) && IS_SKIP(mb_type))
-                        printf("d");
+                        av_log(s->avctx, AV_LOG_DEBUG, "d");
                     else if(IS_DIRECT(mb_type))
-                        printf("D");
+                        av_log(s->avctx, AV_LOG_DEBUG, "D");
                     else if(IS_GMC(mb_type) && IS_SKIP(mb_type))
-                        printf("g");
+                        av_log(s->avctx, AV_LOG_DEBUG, "g");
                     else if(IS_GMC(mb_type))
-                        printf("G");
+                        av_log(s->avctx, AV_LOG_DEBUG, "G");
                     else if(IS_SKIP(mb_type))
-                        printf("S");
+                        av_log(s->avctx, AV_LOG_DEBUG, "S");
                     else if(!USES_LIST(mb_type, 1))
-                        printf(">");
+                        av_log(s->avctx, AV_LOG_DEBUG, ">");
                     else if(!USES_LIST(mb_type, 0))
-                        printf("<");
+                        av_log(s->avctx, AV_LOG_DEBUG, "<");
                     else{
                         assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
-                        printf("X");
+                        av_log(s->avctx, AV_LOG_DEBUG, "X");
                     }
                     
                     //segmentation
                     if(IS_8X8(mb_type))
-                        printf("+");
+                        av_log(s->avctx, AV_LOG_DEBUG, "+");
                     else if(IS_16X8(mb_type))
-                        printf("-");
+                        av_log(s->avctx, AV_LOG_DEBUG, "-");
                     else if(IS_8X16(mb_type))
-                        printf("¦");
+                        av_log(s->avctx, AV_LOG_DEBUG, "¦");
                     else if(IS_INTRA(mb_type) || IS_16X16(mb_type))
-                        printf(" ");
+                        av_log(s->avctx, AV_LOG_DEBUG, " ");
                     else
-                        printf("?");
+                        av_log(s->avctx, AV_LOG_DEBUG, "?");
                     
                         
                     if(IS_INTERLACED(mb_type) && s->codec_id == CODEC_ID_H264)
-                        printf("=");
+                        av_log(s->avctx, AV_LOG_DEBUG, "=");
                     else
-                        printf(" ");
+                        av_log(s->avctx, AV_LOG_DEBUG, " ");
                 }
-//                printf(" ");
+//                av_log(s->avctx, AV_LOG_DEBUG, " ");
             }
-            printf("\n");
+            av_log(s->avctx, AV_LOG_DEBUG, "\n");
         }
     }
     
@@ -1265,6 +1427,16 @@ void ff_print_debug_info(MpegEncContext *s, Picture *pict){
                         int my= (s->motion_val[xy][1]>>shift) + sy;
                         draw_arrow(ptr, sx, sy, mx, my, s->width, s->height, s->linesize, 100);
                     }
+                }else if(IS_16X8(s->current_picture.mb_type[mb_index])){
+                    int i;
+                    for(i=0; i<2; i++){
+                        int sx=mb_x*16 + 8;
+                        int sy=mb_y*16 + 4 + 8*i;
+                        int xy=1 + mb_x*2 + (mb_y*2 + 1 + i)*(s->mb_width*2 + 2);
+                        int mx=(s->motion_val[xy][0]>>shift) + sx;
+                        int my=(s->motion_val[xy][1]>>shift) + sy;
+                        draw_arrow(ptr, sx, sy, mx, my, s->width, s->height, s->linesize, 100);
+                    }
                 }else{
                     int sx= mb_x*16 + 8;
                     int sy= mb_y*16 + 8;
@@ -1327,10 +1499,10 @@ static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
     if(pic_arg->linesize[1] != s->uvlinesize) direct=0;
     if(pic_arg->linesize[2] != s->uvlinesize) direct=0;
   
-//    printf("%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
+//    av_log(AV_LOG_DEBUG, "%d %d %d %d\n",pic_arg->linesize[0], pic_arg->linesize[1], s->linesize, s->uvlinesize);
     
     if(direct){
-        i= find_unused_picture(s, 1);
+        i= ff_find_unused_picture(s, 1);
 
         pic= (AVFrame*)&s->picture[i];
         pic->reference= 3;
@@ -1341,26 +1513,20 @@ static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
         }
         alloc_picture(s, (Picture*)pic, 1);
     }else{
-        i= find_unused_picture(s, 0);
+        int offset= 16;
+        i= ff_find_unused_picture(s, 0);
 
         pic= (AVFrame*)&s->picture[i];
         pic->reference= 3;
 
         alloc_picture(s, (Picture*)pic, 0);
-        for(i=0; i<4; i++){
-            /* the input will be 16 pixels to the right relative to the actual buffer start
-             * and the current_pic, so the buffer can be reused, yes its not beatifull 
-             */
-            pic->data[i]+= 16; 
-        }
 
-        if(   pic->data[0] == pic_arg->data[0] 
-           && pic->data[1] == pic_arg->data[1]
-           && pic->data[2] == pic_arg->data[2]){
+        if(   pic->data[0] + offset == pic_arg->data[0] 
+           && pic->data[1] + offset == pic_arg->data[1]
+           && pic->data[2] + offset == pic_arg->data[2]){
        // empty
         }else{
             int h_chroma_shift, v_chroma_shift;
-        
             avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
         
             for(i=0; i<3; i++){
@@ -1371,7 +1537,7 @@ static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
                 int w= s->width >>h_shift;
                 int h= s->height>>v_shift;
                 uint8_t *src= pic_arg->data[i];
-                uint8_t *dst= pic->data[i];
+                uint8_t *dst= pic->data[i] + offset;
             
                 if(src_stride==dst_stride)
                     memcpy(dst, src, src_stride*h);
@@ -1405,7 +1571,6 @@ static int load_input_picture(MpegEncContext *s, AVFrame *pic_arg){
 
 static void select_input_picture(MpegEncContext *s){
     int i;
-    const int encoding_delay= s->max_b_frames;
     int coded_pic_num=0;    
 
     if(s->reordered_input_picture[0])
@@ -1441,7 +1606,7 @@ static void select_input_picture(MpegEncContext *s){
                 }
             
                 if(b_frames > s->max_b_frames){
-                    fprintf(stderr, "warning, too many bframes in a row\n");
+                    av_log(s->avctx, AV_LOG_ERROR, "warning, too many bframes in a row\n");
                     b_frames = s->max_b_frames;
                 }
             }else if(s->b_frame_strategy==0){
@@ -1466,14 +1631,14 @@ static void select_input_picture(MpegEncContext *s){
                     s->input_picture[i]->b_frame_score=0;
                 }
             }else{
-                fprintf(stderr, "illegal b frame strategy\n");
+                av_log(s->avctx, AV_LOG_ERROR, "illegal b frame strategy\n");
                 b_frames=0;
             }
 
             emms_c();
 //static int b_count=0;
 //b_count+= b_frames;
-//printf("b_frames: %d\n", b_count);
+//av_log(s->avctx, AV_LOG_DEBUG, "b_frames: %d\n", b_count);
                         
             s->reordered_input_picture[0]= s->input_picture[b_frames];
             if(   s->picture_in_gop_number + b_frames >= s->gop_size 
@@ -1494,12 +1659,12 @@ static void select_input_picture(MpegEncContext *s){
     if(s->reordered_input_picture[0]){
         s->reordered_input_picture[0]->reference= s->reordered_input_picture[0]->pict_type!=B_TYPE ? 3 : 0;
 
-        s->new_picture= *s->reordered_input_picture[0];
+        copy_picture(&s->new_picture, s->reordered_input_picture[0]);
 
         if(s->reordered_input_picture[0]->type == FF_BUFFER_TYPE_SHARED){
             // input is a shared pix, so we cant modifiy it -> alloc a new one & ensure that the shared one is reuseable
         
-            int i= find_unused_picture(s, 0);
+            int i= ff_find_unused_picture(s, 0);
             Picture *pic= &s->picture[i];
 
             /* mark us unused / free shared pic */
@@ -1512,6 +1677,7 @@ static void select_input_picture(MpegEncContext *s){
             pic->quality   = s->reordered_input_picture[0]->quality;
             pic->coded_picture_number = s->reordered_input_picture[0]->coded_picture_number;
             pic->reference = s->reordered_input_picture[0]->reference;
+           pic->pts = s->reordered_input_picture[0]->pts;
             
             alloc_picture(s, pic, 0);
 
@@ -1524,11 +1690,10 @@ static void select_input_picture(MpegEncContext *s){
             
             s->current_picture_ptr= s->reordered_input_picture[0];
             for(i=0; i<4; i++){
-                //reverse the +16 we did before storing the input
-                s->current_picture_ptr->data[i]-=16;
+                s->new_picture.data[i]+=16;
             }
         }
-        s->current_picture= *s->current_picture_ptr;
+        copy_picture(&s->current_picture, s->current_picture_ptr);
     
         s->picture_number= s->new_picture.display_picture_number;
 //printf("dpn:%d\n", s->picture_number);
@@ -1544,7 +1709,12 @@ int MPV_encode_picture(AVCodecContext *avctx,
     AVFrame *pic_arg = data;
     int i;
 
-    init_put_bits(&s->pb, buf, buf_size, NULL, NULL);
+    if(avctx->pix_fmt != PIX_FMT_YUV420P){
+        av_log(avctx, AV_LOG_ERROR, "this codec supports only YUV420P\n");
+        return -1;
+    }
+    
+    init_put_bits(&s->pb, buf, buf_size);
 
     s->picture_in_gop_number++;
 
@@ -1556,10 +1726,6 @@ int MPV_encode_picture(AVCodecContext *avctx,
     if(s->new_picture.data[0]){
 
         s->pict_type= s->new_picture.pict_type;
-        if (s->fixed_qscale){ /* the ratecontrol needs the last qscale so we dont touch it for CBR */
-            s->qscale= (int)(s->new_picture.quality+0.5);
-            assert(s->qscale);
-        }
 //emms_c();
 //printf("qs:%f %f %d\n", s->new_picture.quality, s->current_picture.quality, s->qscale);
         MPV_frame_start(s, avctx);
@@ -1632,8 +1798,8 @@ static inline void gmc1_motion(MpegEncContext *s,
 
     dest_y+=dest_offset;
     if(s->flags&CODEC_FLAG_EMU_EDGE){
-        if(src_x<0 || src_y<0 || src_x + 17 >= s->h_edge_pos
-                              || src_y + 17 >= s->v_edge_pos){
+        if(   (unsigned)src_x >= s->h_edge_pos - 17
+           || (unsigned)src_y >= s->v_edge_pos - 17){
             ff_emulated_edge_mc(s->edge_emu_buffer, ptr, linesize, 17, 17, src_x, src_y, s->h_edge_pos, s->v_edge_pos);
             ptr= s->edge_emu_buffer;
         }
@@ -1671,8 +1837,8 @@ static inline void gmc1_motion(MpegEncContext *s,
     offset = (src_y * uvlinesize) + src_x + (src_offset>>1);
     ptr = ref_picture[1] + offset;
     if(s->flags&CODEC_FLAG_EMU_EDGE){
-        if(src_x<0 || src_y<0 || src_x + 9 >= s->h_edge_pos>>1
-                              || src_y + 9 >= s->v_edge_pos>>1){
+        if(   (unsigned)src_x >= (s->h_edge_pos>>1) - 9
+           || (unsigned)src_y >= (s->v_edge_pos>>1) - 9){
             ff_emulated_edge_mc(s->edge_emu_buffer, ptr, uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
             ptr= s->edge_emu_buffer;
             emu=1;
@@ -1824,6 +1990,42 @@ void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize, int block_w,
     }
 }
 
+static inline int hpel_motion(MpegEncContext *s, 
+                                  uint8_t *dest, uint8_t *src, 
+                                  int src_x, int src_y,
+                                  int width, int height, int stride,
+                                  int h_edge_pos, int v_edge_pos,
+                                  int w, int h, op_pixels_func *pix_op,
+                                  int motion_x, int motion_y)
+{
+    int dxy;
+    int emu=0;
+
+    dxy = ((motion_y & 1) << 1) | (motion_x & 1);
+    src_x += motion_x >> 1;
+    src_y += motion_y >> 1;
+                
+    /* WARNING: do no forget half pels */
+    src_x = clip(src_x, -16, width); //FIXME unneeded for emu?
+    if (src_x == width)
+        dxy &= ~1;
+    src_y = clip(src_y, -16, height);
+    if (src_y == height)
+        dxy &= ~2;
+    src += src_y * stride + src_x;
+
+    if(s->unrestricted_mv && (s->flags&CODEC_FLAG_EMU_EDGE)){
+        if(   (unsigned)src_x > h_edge_pos - (motion_x&1) - w
+           || (unsigned)src_y > v_edge_pos - (motion_y&1) - h){
+            ff_emulated_edge_mc(s->edge_emu_buffer, src, stride, w+1, h+1,
+                             src_x, src_y, h_edge_pos, v_edge_pos);
+            src= s->edge_emu_buffer;
+            emu=1;
+        }
+    }
+    pix_op[dxy](dest, src, stride, h);
+    return emu;
+}
 
 /* apply one mpeg motion vector to the three components */
 static inline void mpeg_motion(MpegEncContext *s,
@@ -1834,7 +2036,7 @@ static inline void mpeg_motion(MpegEncContext *s,
                                int motion_x, int motion_y, int h)
 {
     uint8_t *ptr;
-    int dxy, offset, mx, my, src_x, src_y, height, v_edge_pos, linesize, uvlinesize;
+    int dxy, offset, mx, my, src_x, src_y, height, v_edge_pos, uvlinesize;
     int emu=0;
 #if 0    
 if(s->quarter_sample)
@@ -1843,34 +2045,19 @@ if(s->quarter_sample)
     motion_y>>=1;
 }
 #endif
-    dxy = ((motion_y & 1) << 1) | (motion_x & 1);
-    src_x = s->mb_x * 16 + (motion_x >> 1);
-    src_y = s->mb_y * (16 >> field_based) + (motion_y >> 1);
-                
-    /* WARNING: do no forget half pels */
+
     height = s->height >> field_based;
     v_edge_pos = s->v_edge_pos >> field_based;
-    src_x = clip(src_x, -16, s->width);
-    if (src_x == s->width)
-        dxy &= ~1;
-    src_y = clip(src_y, -16, height);
-    if (src_y == height)
-        dxy &= ~2;
-    linesize   = s->current_picture.linesize[0] << field_based;
     uvlinesize = s->current_picture.linesize[1] << field_based;
-    ptr = ref_picture[0] + (src_y * linesize) + (src_x) + src_offset;
-    dest_y += dest_offset;
 
-    if(s->flags&CODEC_FLAG_EMU_EDGE){
-        if(src_x<0 || src_y<0 || src_x + (motion_x&1) + 16 > s->h_edge_pos
-                              || src_y + (motion_y&1) + h  > v_edge_pos){
-            ff_emulated_edge_mc(s->edge_emu_buffer, ptr - src_offset, s->linesize, 17, 17+field_based,  //FIXME linesize? and uv below
-                             src_x, src_y<<field_based, s->h_edge_pos, s->v_edge_pos);
-            ptr= s->edge_emu_buffer + src_offset;
-            emu=1;
-        }
-    }
-    pix_op[0][dxy](dest_y, ptr, linesize, h);
+    emu= hpel_motion(s, 
+                dest_y + dest_offset, ref_picture[0] + src_offset,
+                s->mb_x * 16, s->mb_y * (16 >> field_based),
+                s->width, height, s->current_picture.linesize[0] << field_based,
+                s->h_edge_pos, v_edge_pos,
+                16, h, pix_op[0],
+                motion_x, motion_y);
+
 
     if(s->flags&CODEC_FLAG_GRAY) return;
 
@@ -1915,6 +2102,87 @@ if(s->quarter_sample)
     }
     pix_op[1][dxy](dest_cr + (dest_offset >> 1), ptr, uvlinesize, h >> 1);
 }
+//FIXME move to dsputil, avg variant, 16x16 version
+static inline void put_obmc(uint8_t *dst, uint8_t *src[5], int stride){
+    int x;
+    uint8_t * const top   = src[1];
+    uint8_t * const left  = src[2];
+    uint8_t * const mid   = src[0];
+    uint8_t * const right = src[3];
+    uint8_t * const bottom= src[4];
+#define OBMC_FILTER(x, t, l, m, r, b)\
+    dst[x]= (t*top[x] + l*left[x] + m*mid[x] + r*right[x] + b*bottom[x] + 4)>>3
+#define OBMC_FILTER4(x, t, l, m, r, b)\
+    OBMC_FILTER(x         , t, l, m, r, b);\
+    OBMC_FILTER(x+1       , t, l, m, r, b);\
+    OBMC_FILTER(x  +stride, t, l, m, r, b);\
+    OBMC_FILTER(x+1+stride, t, l, m, r, b);
+    
+    x=0;
+    OBMC_FILTER (x  , 2, 2, 4, 0, 0);
+    OBMC_FILTER (x+1, 2, 1, 5, 0, 0);
+    OBMC_FILTER4(x+2, 2, 1, 5, 0, 0);
+    OBMC_FILTER4(x+4, 2, 0, 5, 1, 0);
+    OBMC_FILTER (x+6, 2, 0, 5, 1, 0);
+    OBMC_FILTER (x+7, 2, 0, 4, 2, 0);
+    x+= stride;
+    OBMC_FILTER (x  , 1, 2, 5, 0, 0);
+    OBMC_FILTER (x+1, 1, 2, 5, 0, 0);
+    OBMC_FILTER (x+6, 1, 0, 5, 2, 0);
+    OBMC_FILTER (x+7, 1, 0, 5, 2, 0);
+    x+= stride;
+    OBMC_FILTER4(x  , 1, 2, 5, 0, 0);
+    OBMC_FILTER4(x+2, 1, 1, 6, 0, 0);
+    OBMC_FILTER4(x+4, 1, 0, 6, 1, 0);
+    OBMC_FILTER4(x+6, 1, 0, 5, 2, 0);
+    x+= 2*stride;
+    OBMC_FILTER4(x  , 0, 2, 5, 0, 1);
+    OBMC_FILTER4(x+2, 0, 1, 6, 0, 1);
+    OBMC_FILTER4(x+4, 0, 0, 6, 1, 1);
+    OBMC_FILTER4(x+6, 0, 0, 5, 2, 1);
+    x+= 2*stride;
+    OBMC_FILTER (x  , 0, 2, 5, 0, 1);
+    OBMC_FILTER (x+1, 0, 2, 5, 0, 1);
+    OBMC_FILTER4(x+2, 0, 1, 5, 0, 2);
+    OBMC_FILTER4(x+4, 0, 0, 5, 1, 2);
+    OBMC_FILTER (x+6, 0, 0, 5, 2, 1);
+    OBMC_FILTER (x+7, 0, 0, 5, 2, 1);
+    x+= stride;
+    OBMC_FILTER (x  , 0, 2, 4, 0, 2);
+    OBMC_FILTER (x+1, 0, 1, 5, 0, 2);
+    OBMC_FILTER (x+6, 0, 0, 5, 1, 2);
+    OBMC_FILTER (x+7, 0, 0, 4, 2, 2);
+}
+
+/* obmc for 1 8x8 luma block */
+static inline void obmc_motion(MpegEncContext *s,
+                               uint8_t *dest, uint8_t *src,
+                               int src_x, int src_y,
+                               op_pixels_func *pix_op,
+                               int16_t mv[5][2]/* mid top left right bottom*/)
+#define MID    0
+{
+    int i;
+    uint8_t *ptr[5];
+    
+    assert(s->quarter_sample==0);
+    
+    for(i=0; i<5; i++){
+        if(i && mv[i][0]==mv[MID][0] && mv[i][1]==mv[MID][1]){
+            ptr[i]= ptr[MID];
+        }else{
+            ptr[i]= s->edge_emu_buffer + 16 + 8*(i&1) + s->linesize*8*(i>>1);
+            hpel_motion(s, ptr[i], src,
+                        src_x, src_y,
+                        s->width, s->height, s->linesize,
+                        s->h_edge_pos, s->v_edge_pos,
+                        8, 8, pix_op,
+                        mv[i][0], mv[i][1]);
+        }
+    }
+
+    put_obmc(dest, ptr, s->linesize);                
+}
 
 static inline void qpel_motion(MpegEncContext *s,
                                uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
@@ -1947,8 +2215,8 @@ static inline void qpel_motion(MpegEncContext *s,
 //printf("%d %d %d\n", src_x, src_y, dxy);
     
     if(s->flags&CODEC_FLAG_EMU_EDGE){
-        if(src_x<0 || src_y<0 || src_x + (motion_x&3) + 16 > s->h_edge_pos
-                              || src_y + (motion_y&3) + h  > v_edge_pos){
+        if(   (unsigned)src_x > s->h_edge_pos - (motion_x&3) - 16 
+           || (unsigned)src_y >    v_edge_pos - (motion_y&3) - h  ){
             ff_emulated_edge_mc(s->edge_emu_buffer, ptr - src_offset, s->linesize, 17, 17+field_based, 
                              src_x, src_y<<field_based, s->h_edge_pos, s->v_edge_pos);
             ptr= s->edge_emu_buffer + src_offset;
@@ -2023,6 +2291,55 @@ inline int ff_h263_round_chroma(int x){
     }
 }
 
+/**
+ * h263 chorma 4mv motion compensation.
+ */
+static inline void chroma_4mv_motion(MpegEncContext *s,
+                                     uint8_t *dest_cb, uint8_t *dest_cr,
+                                     uint8_t **ref_picture,
+                                     op_pixels_func *pix_op,
+                                     int mx, int my){
+    int dxy, emu=0, src_x, src_y, offset;
+    uint8_t *ptr;
+    
+    /* In case of 8X8, we construct a single chroma motion vector
+       with a special rounding */
+    mx= ff_h263_round_chroma(mx);
+    my= ff_h263_round_chroma(my);
+    
+    dxy = ((my & 1) << 1) | (mx & 1);
+    mx >>= 1;
+    my >>= 1;
+
+    src_x = s->mb_x * 8 + mx;
+    src_y = s->mb_y * 8 + my;
+    src_x = clip(src_x, -8, s->width/2);
+    if (src_x == s->width/2)
+        dxy &= ~1;
+    src_y = clip(src_y, -8, s->height/2);
+    if (src_y == s->height/2)
+        dxy &= ~2;
+    
+    offset = (src_y * (s->uvlinesize)) + src_x;
+    ptr = ref_picture[1] + offset;
+    if(s->flags&CODEC_FLAG_EMU_EDGE){
+        if(   (unsigned)src_x > (s->h_edge_pos>>1) - (dxy &1) - 8
+           || (unsigned)src_y > (s->v_edge_pos>>1) - (dxy>>1) - 8){
+            ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
+            ptr= s->edge_emu_buffer;
+            emu=1;
+        }
+    }
+    pix_op[dxy](dest_cb, ptr, s->uvlinesize, 8);
+
+    ptr = ref_picture[2] + offset;
+    if(emu){
+        ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
+        ptr= s->edge_emu_buffer;
+    }
+    pix_op[dxy](dest_cr, ptr, s->uvlinesize, 8);
+}
+
 /**
  * motion compesation of a single macroblock
  * @param s context
@@ -2040,14 +2357,74 @@ static inline void MPV_motion(MpegEncContext *s,
                               int dir, uint8_t **ref_picture, 
                               op_pixels_func (*pix_op)[4], qpel_mc_func (*qpix_op)[16])
 {
-    int dxy, offset, mx, my, src_x, src_y, motion_x, motion_y;
+    int dxy, mx, my, src_x, src_y, motion_x, motion_y;
     int mb_x, mb_y, i;
     uint8_t *ptr, *dest;
-    int emu=0;
 
     mb_x = s->mb_x;
     mb_y = s->mb_y;
 
+    if(s->obmc){
+        int16_t mv_cache[4][4][2];
+        const int xy= s->mb_x + s->mb_y*s->mb_stride;
+        const int mot_stride= s->mb_width*2 + 2;
+        const int mot_xy= 1 + mb_x*2 + (mb_y*2 + 1)*mot_stride;
+
+        assert(!s->mb_skiped);
+                
+        memcpy(mv_cache[1][1], s->motion_val[mot_xy           ], sizeof(int16_t)*4);
+        memcpy(mv_cache[2][1], s->motion_val[mot_xy+mot_stride], sizeof(int16_t)*4);
+        memcpy(mv_cache[3][1], s->motion_val[mot_xy+mot_stride], sizeof(int16_t)*4);
+
+        if(mb_y==0 || IS_INTRA(s->current_picture.mb_type[xy-s->mb_stride])){
+            memcpy(mv_cache[0][1], mv_cache[1][1], sizeof(int16_t)*4);
+        }else{
+            memcpy(mv_cache[0][1], s->motion_val[mot_xy-mot_stride], sizeof(int16_t)*4);
+        }
+
+        if(mb_x==0 || IS_INTRA(s->current_picture.mb_type[xy-1])){
+            *(int32_t*)mv_cache[1][0]= *(int32_t*)mv_cache[1][1];
+            *(int32_t*)mv_cache[2][0]= *(int32_t*)mv_cache[2][1];
+        }else{
+            *(int32_t*)mv_cache[1][0]= *(int32_t*)s->motion_val[mot_xy-1];
+            *(int32_t*)mv_cache[2][0]= *(int32_t*)s->motion_val[mot_xy-1+mot_stride];
+        }
+
+        if(mb_x+1>=s->mb_width || IS_INTRA(s->current_picture.mb_type[xy+1])){
+            *(int32_t*)mv_cache[1][3]= *(int32_t*)mv_cache[1][2];
+            *(int32_t*)mv_cache[2][3]= *(int32_t*)mv_cache[2][2];
+        }else{
+            *(int32_t*)mv_cache[1][3]= *(int32_t*)s->motion_val[mot_xy+2];
+            *(int32_t*)mv_cache[2][3]= *(int32_t*)s->motion_val[mot_xy+2+mot_stride];
+        }
+        
+        mx = 0;
+        my = 0;
+        for(i=0;i<4;i++) {
+            const int x= (i&1)+1;
+            const int y= (i>>1)+1;
+            int16_t mv[5][2]= {
+                {mv_cache[y][x  ][0], mv_cache[y][x  ][1]},
+                {mv_cache[y-1][x][0], mv_cache[y-1][x][1]},
+                {mv_cache[y][x-1][0], mv_cache[y][x-1][1]},
+                {mv_cache[y][x+1][0], mv_cache[y][x+1][1]},
+                {mv_cache[y+1][x][0], mv_cache[y+1][x][1]}};
+            //FIXME cleanup
+            obmc_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
+                        ref_picture[0],
+                        mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
+                        pix_op[1],
+                        mv);
+
+            mx += mv[0][0];
+            my += mv[0][1];
+        }
+        if(!(s->flags&CODEC_FLAG_GRAY))
+            chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
+
+        return;
+    }
+   
     switch(s->mv_type) {
     case MV_TYPE_16X16:
 #ifdef CONFIG_RISKY
@@ -2099,8 +2476,8 @@ static inline void MPV_motion(MpegEncContext *s,
                     
                 ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
                 if(s->flags&CODEC_FLAG_EMU_EDGE){
-                    if(src_x<0 || src_y<0 || src_x + (motion_x&3) + 8 > s->h_edge_pos
-                                          || src_y + (motion_y&3) + 8 > s->v_edge_pos){
+                    if(   (unsigned)src_x > s->h_edge_pos - (motion_x&3) - 8 
+                       || (unsigned)src_y > s->v_edge_pos - (motion_y&3) - 8 ){
                         ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->linesize, 9, 9, src_x, src_y, s->h_edge_pos, s->v_edge_pos);
                         ptr= s->edge_emu_buffer;
                     }
@@ -2113,73 +2490,21 @@ static inline void MPV_motion(MpegEncContext *s,
             }
         }else{
             for(i=0;i<4;i++) {
-                motion_x = s->mv[dir][i][0];
-                motion_y = s->mv[dir][i][1];
-
-                dxy = ((motion_y & 1) << 1) | (motion_x & 1);
-                src_x = mb_x * 16 + (motion_x >> 1) + (i & 1) * 8;
-                src_y = mb_y * 16 + (motion_y >> 1) + (i >>1) * 8;
-                    
-                /* WARNING: do no forget half pels */
-                src_x = clip(src_x, -16, s->width);
-                if (src_x == s->width)
-                    dxy &= ~1;
-                src_y = clip(src_y, -16, s->height);
-                if (src_y == s->height)
-                    dxy &= ~2;
-                    
-                ptr = ref_picture[0] + (src_y * s->linesize) + (src_x);
-                if(s->flags&CODEC_FLAG_EMU_EDGE){
-                    if(src_x<0 || src_y<0 || src_x + (motion_x&1) + 8 > s->h_edge_pos
-                                          || src_y + (motion_y&1) + 8 > s->v_edge_pos){
-                        ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->linesize, 9, 9, src_x, src_y, s->h_edge_pos, s->v_edge_pos);
-                        ptr= s->edge_emu_buffer;
-                    }
-                }
-                dest = dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize;
-                pix_op[1][dxy](dest, ptr, s->linesize, 8);
+                hpel_motion(s, dest_y + ((i & 1) * 8) + (i >> 1) * 8 * s->linesize,
+                            ref_picture[0],
+                            mb_x * 16 + (i & 1) * 8, mb_y * 16 + (i >>1) * 8,
+                            s->width, s->height, s->linesize,
+                            s->h_edge_pos, s->v_edge_pos,
+                            8, 8, pix_op[1],
+                            s->mv[dir][i][0], s->mv[dir][i][1]);
 
                 mx += s->mv[dir][i][0];
                 my += s->mv[dir][i][1];
             }
         }
 
-        if(s->flags&CODEC_FLAG_GRAY) break;
-        /* In case of 8X8, we construct a single chroma motion vector
-           with a special rounding */
-        mx= ff_h263_round_chroma(mx);
-        my= ff_h263_round_chroma(my);
-        dxy = ((my & 1) << 1) | (mx & 1);
-        mx >>= 1;
-        my >>= 1;
-
-        src_x = mb_x * 8 + mx;
-        src_y = mb_y * 8 + my;
-        src_x = clip(src_x, -8, s->width/2);
-        if (src_x == s->width/2)
-            dxy &= ~1;
-        src_y = clip(src_y, -8, s->height/2);
-        if (src_y == s->height/2)
-            dxy &= ~2;
-        
-        offset = (src_y * (s->uvlinesize)) + src_x;
-        ptr = ref_picture[1] + offset;
-        if(s->flags&CODEC_FLAG_EMU_EDGE){
-                if(src_x<0 || src_y<0 || src_x + (dxy &1) + 8 > s->h_edge_pos>>1
-                                      || src_y + (dxy>>1) + 8 > s->v_edge_pos>>1){
-                    ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
-                    ptr= s->edge_emu_buffer;
-                    emu=1;
-                }
-            }
-        pix_op[1][dxy](dest_cb, ptr, s->uvlinesize, 8);
-
-        ptr = ref_picture[2] + offset;
-        if(emu){
-            ff_emulated_edge_mc(s->edge_emu_buffer, ptr, s->uvlinesize, 9, 9, src_x, src_y, s->h_edge_pos>>1, s->v_edge_pos>>1);
-            ptr= s->edge_emu_buffer;
-        }
-        pix_op[1][dxy](dest_cr, ptr, s->uvlinesize, 8);
+        if(!(s->flags&CODEC_FLAG_GRAY))
+            chroma_4mv_motion(s, dest_cb, dest_cr, ref_picture, pix_op[1], mx, my);
         break;
     case MV_TYPE_FIELD:
         if (s->picture_structure == PICT_FRAME) {
@@ -2319,7 +2644,7 @@ static inline void MPV_motion(MpegEncContext *s,
         }
     }
     break;
-
+    default: assert(0);
     }
 }
 
@@ -2342,10 +2667,10 @@ static inline void add_dct(MpegEncContext *s,
 }
 
 static inline void add_dequant_dct(MpegEncContext *s, 
-                           DCTELEM *block, int i, uint8_t *dest, int line_size)
+                           DCTELEM *block, int i, uint8_t *dest, int line_size, int qscale)
 {
     if (s->block_last_index[i] >= 0) {
-        s->dct_unquantize(s, block, i, s->qscale);
+        s->dct_unquantize(s, block, i, qscale);
 
         s->dsp.idct_add (dest, line_size, block);
     }
@@ -2400,7 +2725,7 @@ void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
     const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
 #ifdef HAVE_XVMC
     if(s->avctx->xvmc_acceleration){
-        XVMC_decode_mb(s,block);
+        XVMC_decode_mb(s);//xvmc uses pblocks
         return;
     }
 #endif
@@ -2424,50 +2749,6 @@ 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;
@@ -2475,12 +2756,9 @@ void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
         qpel_mc_func (*op_qpix)[16];
         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];
+        const int readable= s->pict_type != B_TYPE || s->encoding || s->avctx->draw_horiz_band;
 
         /* 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];
@@ -2505,17 +2783,6 @@ 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) {
@@ -2525,11 +2792,19 @@ void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
             dct_linesize = linesize;
             dct_offset = linesize * 8;
         }
-
+        if(readable){
+            dest_y=  s->dest[0];
+            dest_cb= s->dest[1];
+            dest_cr= s->dest[2];
+        }else{
+            dest_y = s->edge_emu_buffer+32; //FIXME cleanup scratchpad pointers
+            dest_cb= s->edge_emu_buffer+48;
+            dest_cr= s->edge_emu_buffer+56;
+        }
         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;
@@ -2552,16 +2827,16 @@ void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
             if(s->hurry_up>1) return;
 
             /* add dct residue */
-            if(s->encoding || !(   s->mpeg2 || s->h263_msmpeg4 || s->codec_id==CODEC_ID_MPEG1VIDEO 
+            if(s->encoding || !(   s->h263_msmpeg4 || s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO
                                 || (s->codec_id==CODEC_ID_MPEG4 && !s->mpeg_quant))){
-                add_dequant_dct(s, block[0], 0, dest_y, dct_linesize);
-                add_dequant_dct(s, block[1], 1, dest_y + 8, dct_linesize);
-                add_dequant_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
-                add_dequant_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize);
+                add_dequant_dct(s, block[0], 0, dest_y, dct_linesize, s->qscale);
+                add_dequant_dct(s, block[1], 1, dest_y + 8, dct_linesize, s->qscale);
+                add_dequant_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize, s->qscale);
+                add_dequant_dct(s, block[3], 3, dest_y + dct_offset + 8, dct_linesize, s->qscale);
 
                 if(!(s->flags&CODEC_FLAG_GRAY)){
-                    add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize);
-                    add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize);
+                    add_dequant_dct(s, block[4], 4, dest_cb, uvlinesize, s->chroma_qscale);
+                    add_dequant_dct(s, block[5], 5, dest_cr, uvlinesize, s->chroma_qscale);
                 }
             } else if(s->codec_id != CODEC_ID_WMV2){
                 add_dct(s, block[0], 0, dest_y, dct_linesize);
@@ -2581,7 +2856,7 @@ void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
 #endif
         } else {
             /* dct only in intra block */
-            if(s->encoding || !(s->mpeg2 || s->codec_id==CODEC_ID_MPEG1VIDEO)){
+            if(s->encoding || !(s->codec_id==CODEC_ID_MPEG1VIDEO || s->codec_id==CODEC_ID_MPEG2VIDEO)){
                 put_dct(s, block[0], 0, dest_y, dct_linesize);
                 put_dct(s, block[1], 1, dest_y + 8, dct_linesize);
                 put_dct(s, block[2], 2, dest_y + dct_offset, dct_linesize);
@@ -2603,6 +2878,11 @@ void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64])
                 }
             }
         }
+        if(!readable){
+            s->dsp.put_pixels_tab[0][0](s->dest[0], dest_y ,   linesize,16);
+            s->dsp.put_pixels_tab[1][0](s->dest[1], dest_cb, uvlinesize, 8);
+            s->dsp.put_pixels_tab[1][0](s->dest[2], dest_cr, uvlinesize, 8);
+        }
     }
 }
 
@@ -2753,7 +3033,6 @@ static int pix_diff_vcmp16x8(uint8_t *s1, uint8_t*s2, int stride){ //FIXME move
 void ff_draw_horiz_band(MpegEncContext *s, int y, int h){
     if (s->avctx->draw_horiz_band) {
         AVFrame *src;
-        uint8_t *src_ptr[3];
         int offset[4];
         
         if(s->picture_structure != PICT_FRAME){
@@ -2790,6 +3069,28 @@ 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)
@@ -2803,24 +3104,29 @@ static void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
     for(i=0; i<6; i++) skip_dct[i]=0;
     
     if(s->adaptive_quant){
-        s->dquant= s->current_picture.qscale_table[mb_x + mb_y*s->mb_stride] - s->qscale;
+        const int last_qp= s->qscale;
+        const int mb_xy= mb_x + mb_y*s->mb_stride;
 
-        if(s->out_format==FMT_H263){
-            if     (s->dquant> 2) s->dquant= 2;
-            else if(s->dquant<-2) s->dquant=-2;
-        }
-            
-        if(s->codec_id==CODEC_ID_MPEG4){        
-            if(!s->mb_intra){
-                if(s->mv_dir&MV_DIRECT)
-                    s->dquant=0;
+        s->lambda= s->lambda_table[mb_xy];
+        update_qscale(s);
+    
+        if(!(s->flags&CODEC_FLAG_QP_RD)){
+            s->dquant= s->qscale - last_qp;
 
-                assert(s->dquant==0 || s->mv_type!=MV_TYPE_8X8);
+            if(s->out_format==FMT_H263)
+                s->dquant= clip(s->dquant, -2, 2); //FIXME RD
+            
+            if(s->codec_id==CODEC_ID_MPEG4){        
+                if(!s->mb_intra){
+                    if((s->mv_dir&MV_DIRECT) || s->mv_type==MV_TYPE_8X8)
+                        s->dquant=0;
+                }
             }
         }
-        s->qscale+= s->dquant;
+        s->qscale= last_qp + s->dquant;
+        s->chroma_qscale= s->chroma_qscale_table[ s->qscale ];
         s->y_dc_scale= s->y_dc_scale_table[ s->qscale ];
-        s->c_dc_scale= s->c_dc_scale_table[ s->qscale ];
+        s->c_dc_scale= s->c_dc_scale_table[ s->chroma_qscale ];
     }
 
     if (s->mb_intra) {
@@ -2884,9 +3190,9 @@ 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;
@@ -2981,19 +3287,6 @@ static void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
 
     }
             
-#if 0
-            {
-                float adap_parm;
-                
-                adap_parm = ((s->avg_mb_var << 1) + s->mb_var[s->mb_stride*mb_y+mb_x] + 1.0) /
-                            ((s->mb_var[s->mb_stride*mb_y+mb_x] << 1) + s->avg_mb_var + 1.0);
-            
-                printf("\ntype=%c qscale=%2d adap=%0.2f dquant=%4.2f var=%4d avgvar=%4d", 
-                        (s->mb_type[s->mb_stride*mb_y+mb_x] > 0) ? 'I' : 'P', 
-                        s->qscale, adap_parm, s->qscale*adap_parm,
-                        s->mb_var[s->mb_stride*mb_y+mb_x], s->avg_mb_var);
-            }
-#endif
     /* DCT & quantize */
     if(s->out_format==FMT_MJPEG){
         for(i=0;i<6;i++) {
@@ -3013,12 +3306,20 @@ static void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
             }else
                 s->block_last_index[i]= -1;
         }
+        
         if(s->luma_elim_threshold && !s->mb_intra)
             for(i=0; i<4; i++)
                 dct_single_coeff_elimination(s, i, s->luma_elim_threshold);
         if(s->chroma_elim_threshold && !s->mb_intra)
             for(i=4; i<6; i++)
                 dct_single_coeff_elimination(s, i, s->chroma_elim_threshold);
+
+        if(s->flags & CODEC_FLAG_CBP_RD){
+            for(i=0;i<6;i++) {
+                if(s->block_last_index[i] == -1)
+                    s->coded_score[i]= INT_MAX/256;
+            }
+        }
     }
 
     if((s->flags&CODEC_FLAG_GRAY) && s->mb_intra){
@@ -3031,6 +3332,7 @@ static void encode_mb(MpegEncContext *s, int motion_x, int motion_y)
     /* huffman encode */
     switch(s->codec_id){ //FIXME funct ptr could be slightly faster
     case CODEC_ID_MPEG1VIDEO:
+    case CODEC_ID_MPEG2VIDEO:
         mpeg1_encode_mb(s, s->block, motion_x, motion_y); break;
 #ifdef CONFIG_RISKY
     case CODEC_ID_MPEG4:
@@ -3123,7 +3425,14 @@ void ff_mpeg_flush(AVCodecContext *avctx){
                                     || s->picture[i].type == FF_BUFFER_TYPE_USER))
         avctx->release_buffer(avctx, (AVFrame*)&s->picture[i]);
     }
-    s->last_picture_ptr = s->next_picture_ptr = NULL;
+    s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL;
+    
+    s->parse_context.state= -1;
+    s->parse_context.frame_start_found= 0;
+    s->parse_context.overread= 0;
+    s->parse_context.overread_index= 0;
+    s->parse_context.index= 0;
+    s->parse_context.last_index= 0;
 }
 
 #ifdef CONFIG_ENCODERS
@@ -3160,8 +3469,9 @@ 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;
+    d->dquant= s->dquant;
 }
 
 static inline void copy_context_after_encode(MpegEncContext *d, MpegEncContext *s, int type){
@@ -3205,7 +3515,8 @@ 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);
 
@@ -3215,24 +3526,44 @@ 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);
+
+        score *= s->lambda2;
+        score += sse_mb(s) << FF_LAMBDA_SHIFT;
+    }
+    
+    if(*next_block){
+        memcpy(s->dest, dest_backup, sizeof(s->dest));
     }
 
-    if(bits_count<*dmin){
-        *dmin= bits_count;
+    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;
@@ -3253,6 +3584,23 @@ 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;
@@ -3265,9 +3613,9 @@ static void encode_picture(MpegEncContext *s, int picture_number)
     PutBitContext pb[2], pb2[2], tex_pb[2];
 
     for(i=0; i<2; i++){
-        init_put_bits(&pb    [i], bit_buf    [i], 3000, NULL, NULL);
-        init_put_bits(&pb2   [i], bit_buf2   [i], 3000, NULL, NULL);
-        init_put_bits(&tex_pb[i], bit_buf_tex[i], 3000, NULL, NULL);
+        init_put_bits(&pb    [i], bit_buf    [i], 3000);
+        init_put_bits(&pb2   [i], bit_buf2   [i], 3000);
+        init_put_bits(&tex_pb[i], bit_buf_tex[i], 3000);
     }
 
     s->picture_number = picture_number;
@@ -3279,13 +3627,13 @@ static void encode_picture(MpegEncContext *s, int picture_number)
 #ifdef CONFIG_RISKY
     /* we need to initialize some time vars before we can encode b-frames */
     // RAL: Condition added for MPEG1VIDEO
-    if (s->codec_id == CODEC_ID_MPEG1VIDEO || (s->h263_pred && !s->h263_msmpeg4))
+    if (s->codec_id == CODEC_ID_MPEG1VIDEO || s->codec_id == CODEC_ID_MPEG2VIDEO || (s->h263_pred && !s->h263_msmpeg4))
         ff_set_mpeg4_time(s, s->picture_number); 
 #endif
         
     s->scene_change_score=0;
     
-    s->qscale= (int)(s->frame_qscale + 0.5); //FIXME qscale / ... stuff for ME ratedistoration
+    s->lambda= s->current_picture_ptr->quality; //FIXME qscale / ... stuff for ME ratedistoration
     
     if(s->pict_type==I_TYPE){
         if(s->msmpeg4_version >= 3) s->no_rounding=1;
@@ -3363,7 +3711,7 @@ static void encode_picture(MpegEncContext *s, int picture_number)
     }
     emms_c();
 
-    if(s->scene_change_score > 0 && s->pict_type == P_TYPE){
+    if(s->scene_change_score > s->avctx->scenechange_threshold && s->pict_type == P_TYPE){
         s->pict_type= I_TYPE;
         memset(s->mb_type   , MB_TYPE_INTRA, sizeof(uint8_t)*s->mb_stride*s->mb_height);
 //printf("Scene change detected, encoding as I Frame %d %d\n", s->current_picture.mb_var_sum, s->current_picture.mc_mb_var_sum);
@@ -3394,10 +3742,8 @@ static void encode_picture(MpegEncContext *s, int picture_number)
         }
     }
     
-    if (s->fixed_qscale) 
-        s->frame_qscale = s->current_picture.quality;
-    else
-        s->frame_qscale = ff_rate_estimate_qscale(s);
+    if (!s->fixed_qscale) 
+        s->current_picture.quality = ff_rate_estimate_qscale(s);
 
     if(s->adaptive_quant){
 #ifdef CONFIG_RISKY
@@ -3413,9 +3759,15 @@ static void encode_picture(MpegEncContext *s, int picture_number)
         }
 #endif
 
-        s->qscale= s->current_picture.qscale_table[0];
+        s->lambda= s->lambda_table[0];
+        //FIXME broken
     }else
-        s->qscale= (int)(s->frame_qscale + 0.5);
+        s->lambda= s->current_picture.quality;
+//printf("%d %d\n", s->avctx->global_quality, s->current_picture.quality);
+    update_qscale(s);
+    
+    if(s->qscale < 3 && s->max_qcoeff<=128 && s->pict_type==I_TYPE && !(s->flags & CODEC_FLAG_QSCALE)) 
+        s->qscale= 3; //reduce cliping problems
         
     if (s->out_format == FMT_MJPEG) {
         /* for mjpeg, we do include qscale in the matrix */
@@ -3425,8 +3777,8 @@ static void encode_picture(MpegEncContext *s, int picture_number)
 
             s->intra_matrix[j] = CLAMP_TO_8BIT((ff_mpeg1_default_intra_matrix[i] * s->qscale) >> 3);
         }
-        convert_matrix(s, s->q_intra_matrix, s->q_intra_matrix16, 
-                       s->q_intra_matrix16_bias, s->intra_matrix, s->intra_quant_bias, 8, 8);
+        convert_matrix(&s->dsp, s->q_intra_matrix, s->q_intra_matrix16, 
+                       s->intra_matrix, s->intra_quant_bias, 8, 8);
     }
     
     //FIXME var duplication
@@ -3449,7 +3801,7 @@ static void encode_picture(MpegEncContext *s, int picture_number)
             msmpeg4_encode_picture_header(s, picture_number);
         else if (s->h263_pred)
             mpeg4_encode_picture_header(s, picture_number);
-        else if (s->h263_rv10) 
+        else if (s->codec_id == CODEC_ID_RV10) 
             rv10_encode_picture_header(s, picture_number);
         else if (s->codec_id == CODEC_ID_FLV1)
             ff_flv_encode_picture_header(s, picture_number);
@@ -3460,6 +3812,8 @@ static void encode_picture(MpegEncContext *s, int picture_number)
     case FMT_MPEG1:
         mpeg1_encode_picture_header(s, picture_number);
         break;
+    case FMT_H264:
+        break;
     }
     bits= get_bit_count(&s->pb);
     s->header_bits= bits - s->last_bits;
@@ -3507,41 +3861,33 @@ 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->chroma_qscale= s->chroma_qscale_table[ s->qscale ];
         s->y_dc_scale= s->y_dc_scale_table[ s->qscale ];
-        s->c_dc_scale= s->c_dc_scale_table[ s->qscale ];
+        s->c_dc_scale= s->c_dc_scale_table[ s->chroma_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
-            if(s->rtp_mode){
+            if(s->rtp_mode && mb_y + mb_x>0){
                 int current_packet_size, is_gob_start;
                 
                 current_packet_size= pbBufPtr(&s->pb) - s->ptr_lastgob;
                 is_gob_start=0;
                 
                 if(s->codec_id==CODEC_ID_MPEG4){
-                    if(current_packet_size >= s->rtp_payload_size
-                       && s->mb_y + s->mb_x>0){
+                    if(current_packet_size >= s->rtp_payload_size){
 
                         if(s->partitioned_frame){
                             ff_mpeg4_merge_partitions(s);
@@ -3559,14 +3905,21 @@ static void encode_picture(MpegEncContext *s, int picture_number)
                     }
                 }else if(s->codec_id==CODEC_ID_MPEG1VIDEO){
                     if(   current_packet_size >= s->rtp_payload_size 
-                       && s->mb_y + s->mb_x>0 && s->mb_skip_run==0){
+                       && s->mb_skip_run==0){
+                        ff_mpeg1_encode_slice_header(s);
+                        ff_mpeg1_clean_buffers(s);
+                        is_gob_start=1;
+                    }
+                }else if(s->codec_id==CODEC_ID_MPEG2VIDEO){
+                    if(   (   current_packet_size >= s->rtp_payload_size || mb_x==0)
+                       && s->mb_skip_run==0){
                         ff_mpeg1_encode_slice_header(s);
                         ff_mpeg1_clean_buffers(s);
                         is_gob_start=1;
                     }
                 }else{
                     if(current_packet_size >= s->rtp_payload_size
-                       && s->mb_x==0 && s->mb_y>0 && s->mb_y%s->gob_index==0){
+                       && s->mb_x==0 && s->mb_y%s->gob_index==0){
                        
                         h263_encode_gob_header(s, mb_y);                       
                         is_gob_start=1;
@@ -3587,7 +3940,10 @@ static void encode_picture(MpegEncContext *s, int picture_number)
                 s->first_slice_line=0; 
             }
 
-            if(mb_type & (mb_type-1)){ // more than 1 MB type possible
+            s->mb_skiped=0;
+            s->dquant=0; //only for QP_RD
+
+            if(mb_type & (mb_type-1) || (s->flags & CODEC_FLAG_QP_RD)){ // more than 1 MB type possible
                 int next_block=0;
                 int pb_bits_count, pb2_bits_count, tex_pb_bits_count;
 
@@ -3609,6 +3965,15 @@ static void encode_picture(MpegEncContext *s, int picture_number)
                     encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTER, pb, pb2, tex_pb, 
                                  &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
                 }
+                if(mb_type&MB_TYPE_SKIPED){
+                    s->mv_dir = MV_DIR_FORWARD;
+                    s->mv_type = MV_TYPE_16X16;
+                    s->mb_intra= 0;
+                    s->mv[0][0][0] = 0;
+                    s->mv[0][0][1] = 0;
+                    encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_SKIPED, pb, pb2, tex_pb, 
+                                 &dmin, &next_block, s->mv[0][0][0], s->mv[0][0][1]);
+                }
                 if(mb_type&MB_TYPE_INTER4V){                 
                     s->mv_dir = MV_DIR_FORWARD;
                     s->mv_type = MV_TYPE_8X8;
@@ -3669,10 +4034,67 @@ 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?
+                    }
                 }
+
+                if(s->flags & CODEC_FLAG_QP_RD){
+                    if(best_s.mv_type==MV_TYPE_16X16 && !(best_s.mv_dir&MV_DIRECT)){
+                        const int last_qp= backup_s.qscale;
+                        int dquant, dir, qp, dc[6];
+                        DCTELEM ac[6][16];
+                        const int mvdir= (best_s.mv_dir&MV_DIR_BACKWARD) ? 1 : 0;
+                        
+                        assert(backup_s.dquant == 0);
+
+                        //FIXME intra
+                        s->mv_dir= best_s.mv_dir;
+                        s->mv_type = MV_TYPE_16X16;
+                        s->mb_intra= best_s.mb_intra;
+                        s->mv[0][0][0] = best_s.mv[0][0][0];
+                        s->mv[0][0][1] = best_s.mv[0][0][1];
+                        s->mv[1][0][0] = best_s.mv[1][0][0];
+                        s->mv[1][0][1] = best_s.mv[1][0][1];
+                        
+                        dir= s->pict_type == B_TYPE ? 2 : 1;
+                        if(last_qp + dir > s->avctx->qmax) dir= -dir;
+                        for(dquant= dir; dquant<=2 && dquant>=-2; dquant += dir){
+                            qp= last_qp + dquant;
+                            if(qp < s->avctx->qmin || qp > s->avctx->qmax)
+                                break;
+                            backup_s.dquant= dquant;
+                            if(s->mb_intra){
+                                for(i=0; i<6; i++){
+                                    dc[i]= s->dc_val[0][ s->block_index[i] ];
+                                    memcpy(ac[i], s->ac_val[0][s->block_index[i]], sizeof(DCTELEM)*16);
+                                }
+                            }
+
+                            encode_mb_hq(s, &backup_s, &best_s, MB_TYPE_INTER /* wrong but unused */, pb, pb2, tex_pb, 
+                                         &dmin, &next_block, s->mv[mvdir][0][0], s->mv[mvdir][0][1]);
+                            if(best_s.qscale != qp){
+                                if(s->mb_intra){
+                                    for(i=0; i<6; i++){
+                                        s->dc_val[0][ s->block_index[i] ]= dc[i];
+                                        memcpy(s->ac_val[0][s->block_index[i]], ac[i], sizeof(DCTELEM)*16);
+                                    }
+                                }
+                                if(dir > 0 && dquant==dir){
+                                    dquant= 0;
+                                    dir= -dir;
+                                }else
+                                    break;
+                            }
+                        }
+                        qp= best_s.qscale;
+                        s->current_picture.qscale_table[xy]= qp;
+                    }
+                }
+
                 copy_context_after_encode(s, &best_s, -1);
                 
                 pb_bits_count= get_bit_count(&s->pb);
@@ -3692,92 +4114,22 @@ static void encode_picture(MpegEncContext *s, int picture_number)
                     s->tex_pb= backup_s.tex_pb;
                 }
                 s->last_bits= get_bit_count(&s->pb);
-            } 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){
-                /* 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
-                }else{
-                    uint8_t *dest_y;
-
-                    int mean= s->current_picture.mb_mean[mb_x + mb_y*s->mb_stride]; //FIXME
-                    mean*= 0x01010101;
-                    
-                    dest_y  = s->new_picture.data[0] + (mb_y * 16 * s->linesize    ) + mb_x * 16;
-                
-                    for(i=0; i<16; i++){
-                        *(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 0]) = mean;
-                        *(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 4]) = mean;
-                        *(uint32_t*)(&s->me.scratchpad[i*s->linesize+ 8]) = mean;
-                        *(uint32_t*)(&s->me.scratchpad[i*s->linesize+12]) = mean;
-                    }
-
-                    s->mb_intra=1;
-                    intra_score= s->dsp.mb_cmp[0](s, s->me.scratchpad, dest_y, s->linesize);
-                                        
-/*                    printf("intra:%7d inter:%7d var:%7d mc_var.%7d\n", intra_score>>8, inter_score>>8, 
-                        s->current_picture.mb_var[mb_x + mb_y*s->mb_stride],
-                        s->current_picture.mc_mb_var[mb_x + mb_y*s->mb_stride]);*/
-                }
-                
-                /* get chroma score */
-                if(s->avctx->mb_cmp&FF_CMP_CHROMA){
-                    int i;
-                    
-                    s->mb_intra=1;
-                    for(i=1; i<3; i++){
-                        uint8_t *dest_c;
-                        int mean;
-                        
-                        if(s->out_format == FMT_H263){
-                            mean= (s->dc_val[i][mb_x + (mb_y+1)*(s->mb_width+2)] + 4)>>3; //FIXME not exact but simple ;)
-                        }else{
-                            mean= (s->last_dc[i] + 4)>>3;
-                        }
-                        dest_c = s->new_picture.data[i] + (mb_y * 8  * (s->uvlinesize)) + mb_x * 8;
-                        
-                        mean*= 0x01010101;
-                        for(i=0; i<8; i++){
-                            *(uint32_t*)(&s->me.scratchpad[i*s->uvlinesize+ 0]) = mean;
-                            *(uint32_t*)(&s->me.scratchpad[i*s->uvlinesize+ 4]) = mean;
-                        }
-                        
-                        intra_score+= s->dsp.mb_cmp[1](s, s->me.scratchpad, dest_c, s->uvlinesize);
-                    }                
-                }
-
-                /* bias */
-                switch(s->avctx->mb_cmp&0xFF){
-                default:
-                case FF_CMP_SAD:
-                    intra_score+= 32*s->qscale;
-                    break;
-                case FF_CMP_SSE:
-                    intra_score+= 24*s->qscale*s->qscale;
-                    break;
-                case FF_CMP_SATD:
-                    intra_score+= 96*s->qscale;
-                    break;
-                case FF_CMP_DCT:
-                    intra_score+= 48*s->qscale;
-                    break;
-                case FF_CMP_BIT:
-                    intra_score+= 16;
-                    break;
-                case FF_CMP_PSNR:
-                case FF_CMP_RD:
-                    intra_score+= (s->qscale*s->qscale*109*8 + 64)>>7;
-                    break;
+               
+#ifdef CONFIG_RISKY
+                if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE)
+                    ff_h263_update_motion_val(s);
+#endif
+        
+                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(intra_score < inter_score)
-                    mb_type= MB_TYPE_INTRA;
-              }  
-                
+                if(s->avctx->mb_decision == FF_MB_DECISION_BITS)
+                    MPV_decode_mb(s, s->block);
+            } else {
+                int motion_x, motion_y;
                 s->mv_type=MV_TYPE_16X16;
                 // only one MB-Type possible
                 
@@ -3838,13 +4190,20 @@ static void encode_picture(MpegEncContext *s, int picture_number)
                     break;
                 default:
                     motion_x=motion_y=0; //gcc warning fix
-                    printf("illegal MB type\n");
+                    av_log(s->avctx, AV_LOG_ERROR, "illegal MB type\n");
                 }
 
                 encode_mb(s, motion_x, motion_y);
 
                 // RAL: Update last macrobloc type
                 s->last_mv_dir = s->mv_dir;
+            
+#ifdef CONFIG_RISKY
+                if (s->out_format == FMT_H263 && s->pict_type!=B_TYPE)
+                    ff_h263_update_motion_val(s);
+#endif
+               
+                MPV_decode_mb(s, s->block);
             }
 
             /* clean the MV table in IPS frames for direct mode in B frames */
@@ -3852,8 +4211,6 @@ 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;
@@ -3863,21 +4220,17 @@ 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);
             }
+            if(s->loop_filter)
+                ff_h263_loop_filter(s);
 //printf("MB %d %d bits\n", s->mb_x+s->mb_y*s->mb_stride, get_bit_count(&s->pb));
         }
     }
@@ -3909,6 +4262,30 @@ static void encode_picture(MpegEncContext *s, int picture_number)
     }
 }
 
+void ff_denoise_dct(MpegEncContext *s, DCTELEM *block){
+    const int intra= s->mb_intra;
+    int i;
+
+    s->dct_count[intra]++;
+
+    for(i=0; i<64; i++){
+        int level= block[i];
+
+        if(level){
+            if(level>0){
+                s->dct_error_sum[intra][i] += level;
+                level -= s->dct_offset[intra][i];
+                if(level<0) level=0;
+            }else{
+                s->dct_error_sum[intra][i] -= level;
+                level += s->dct_offset[intra][i];
+                if(level>0) level=0;
+            }
+            block[i]= level;
+        }
+    }
+}
+
 static int dct_quantize_trellis_c(MpegEncContext *s, 
                         DCTELEM *block, int n,
                         int qscale, int *overflow){
@@ -3924,17 +4301,23 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
     int last_level=0;
     int last_score= 0;
     int last_i= 0;
+    int not_coded_score= 0;
     int coeff[3][64];
     int coeff_count[64];
-    int lambda, qmul, qadd, start_i, last_non_zero, i;
+    int qmul, qadd, start_i, last_non_zero, i, dc;
     const int esc_length= s->ac_esc_length;
     uint8_t * length;
     uint8_t * last_length;
     int score_limit=0;
     int left_limit= 0;
+    const int lambda= s->lambda2 >> (FF_LAMBDA_SHIFT - 6);
+    const int patch_table= s->out_format == FMT_MPEG1 && !s->mb_intra;
         
     s->dsp.fdct (block);
-
+    
+    if(s->dct_error_sum)
+        ff_denoise_dct(s, block);
+    
     qmul= qscale*16;
     qadd= ((qscale-1)|1)*8;
 
@@ -3957,7 +4340,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
         start_i = 1;
         last_non_zero = 0;
         qmat = s->q_intra_matrix[qscale];
-        if(s->mpeg_quant || s->codec_id== CODEC_ID_MPEG1VIDEO)
+        if(s->mpeg_quant || s->out_format == FMT_MPEG1)
             bias= 1<<(QMAT_SHIFT-1);
         length     = s->intra_ac_vlc_length;
         last_length= s->intra_ac_vlc_last_length;
@@ -3993,6 +4376,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
 //                coeff[2][k]= -level+2;
             }
             coeff_count[k]= FFMIN(level, 2);
+            assert(coeff_count[k]);
             max |=level;
             last_non_zero = i;
         }else{
@@ -4008,9 +4392,13 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
         return last_non_zero;
     }
 
-    lambda= (qscale*qscale*64*105 + 64)>>7; //FIXME finetune
-        
     score_tab[0]= 0;
+    
+    if(patch_table){
+//        length[UNI_AC_ENC_INDEX(0, 63)]=
+//        length[UNI_AC_ENC_INDEX(0, 65)]= 2;
+    }
+
     for(i=0; i<=last_non_zero - start_i; i++){
         int level_index, run, j;
         const int dct_coeff= block[ scantable[i + start_i] ];
@@ -4018,6 +4406,7 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
         int best_score=256*256*256*120;
 
         last_score += zero_distoration;
+        not_coded_score += zero_distoration;
         for(level_index=0; level_index < coeff_count[i]; level_index++){
             int distoration;
             int level= coeff[level_index][i];
@@ -4116,10 +4505,13 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
         
         //Note: there is a vlc code in mpeg4 which is 1 bit shorter then another one with a shorter run and the same level
         while(score_tab[ left_limit ] > score_limit + lambda) left_limit++;
+    
+        if(patch_table){
+//            length[UNI_AC_ENC_INDEX(0, 63)]=
+//            length[UNI_AC_ENC_INDEX(0, 65)]= 3;
+        }
     }
 
-        //FIXME add some cbp penalty
-
     if(s->out_format != FMT_H263){
         last_score= 256*256*256*120;
         for(i= left_limit; i<=last_non_zero - start_i + 1; i++){
@@ -4134,13 +4526,60 @@ static int dct_quantize_trellis_c(MpegEncContext *s,
             }
         }
     }
+
+    s->coded_score[n] = last_score - not_coded_score;
     
+    dc= block[0];
     last_non_zero= last_i - 1 + start_i;
     memset(block + start_i, 0, (64-start_i)*sizeof(DCTELEM));
     
     if(last_non_zero < start_i)
         return last_non_zero;
-    
+
+    if(last_non_zero == 0 && start_i == 0){
+        int best_level= 0;
+        int best_score= dc * dc;
+        
+        for(i=0; i<coeff_count[0]; i++){
+            int level= coeff[i][0];
+            int unquant_coeff, score, distoration;
+
+            if(s->out_format == FMT_H263){
+                if(level>0){
+                    unquant_coeff= (level*qmul + qadd)>>3;
+                }else{
+                    unquant_coeff= (level*qmul - qadd)>>3;
+                }
+            }else{ //MPEG1
+                    if (level < 0) {
+                        unquant_coeff = ((((-level) << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
+                        unquant_coeff = -((unquant_coeff - 1) | 1);
+                    } else {
+                        unquant_coeff = (((  level  << 1) + 1) * qscale * ((int) s->inter_matrix[0])) >> 4;
+                        unquant_coeff =   (unquant_coeff - 1) | 1;
+                    }
+            }
+            unquant_coeff = (unquant_coeff + 4) >> 3;
+            unquant_coeff<<= 3 + 3;
+
+            distoration= (unquant_coeff - dc) * (unquant_coeff - dc);
+            level+=64;
+            if((level&(~127)) == 0)
+                score= distoration + last_length[UNI_AC_ENC_INDEX(0, level)]*lambda;
+            else
+                score= distoration + esc_length*lambda;
+
+            if(score < best_score){
+                best_score= score;
+                best_level= level - 64;
+            }
+        }
+        block[0]= best_level;
+        s->coded_score[n] = best_score - dc*dc;
+        if(best_level == 0) return -1;
+        else                return last_non_zero;
+    }
+
     i= last_i;
     assert(last_level);
 //FIXME use permutated scantable
@@ -4170,6 +4609,9 @@ static int dct_quantize_c(MpegEncContext *s,
 
     s->dsp.fdct (block);
 
+    if(s->dct_error_sum)
+        ff_denoise_dct(s, block);
+
     if (s->mb_intra) {
         if (!s->h263_aic) {
             if (n < 4)
@@ -4401,7 +4843,6 @@ 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),
@@ -4487,6 +4928,16 @@ AVCodec mpeg1video_encoder = {
 
 #ifdef CONFIG_RISKY
 
+AVCodec mpeg2video_encoder = {
+    "mpeg2video",
+    CODEC_TYPE_VIDEO,
+    CODEC_ID_MPEG2VIDEO,
+    sizeof(MpegEncContext),
+    MPV_encode_init,
+    MPV_encode_picture,
+    MPV_encode_end,
+};
+
 AVCodec h263_encoder = {
     "h263",
     CODEC_TYPE_VIDEO,
@@ -4595,4 +5046,3 @@ AVCodec mjpeg_encoder = {
 };
 
 #endif //CONFIG_ENCODERS
-