Merge remote-tracking branch 'qatar/master'
authorMichael Niedermayer <michaelni@gmx.at>
Wed, 9 Apr 2014 12:36:47 +0000 (14:36 +0200)
committerMichael Niedermayer <michaelni@gmx.at>
Wed, 9 Apr 2014 13:18:46 +0000 (15:18 +0200)
* qatar/master:
  mpegvideo: operate with pointers to AVFrames instead of whole structs

Conflicts:
libavcodec/h261dec.c
libavcodec/h263dec.c
libavcodec/intrax8.c
libavcodec/mpeg12enc.c
libavcodec/mpegvideo.c
libavcodec/mpegvideo.h
libavcodec/mpegvideo_enc.c
libavcodec/mpegvideo_motion.c
libavcodec/mpegvideo_xvmc.c
libavcodec/msmpeg4.c
libavcodec/ratecontrol.c
libavcodec/vaapi.c
libavcodec/vc1dec.c
libavcodec/vdpau_vc1.c

See: fc567ac49e17151f00f31b59030cd10f952612ef
Merged-by: Michael Niedermayer <michaelni@gmx.at>
35 files changed:
1  2 
libavcodec/dxva2_mpeg2.c
libavcodec/dxva2_vc1.c
libavcodec/h261dec.c
libavcodec/h263dec.c
libavcodec/intrax8.c
libavcodec/motion_est.c
libavcodec/mpeg12dec.c
libavcodec/mpeg12enc.c
libavcodec/mpeg4videoenc.c
libavcodec/mpegvideo.c
libavcodec/mpegvideo.h
libavcodec/mpegvideo_enc.c
libavcodec/mpegvideo_motion.c
libavcodec/mpegvideo_xvmc.c
libavcodec/msmpeg4.c
libavcodec/mss2.c
libavcodec/pthread_frame.c
libavcodec/ratecontrol.c
libavcodec/rv10.c
libavcodec/rv30.c
libavcodec/rv34.c
libavcodec/rv40.c
libavcodec/snowenc.c
libavcodec/svq1enc.c
libavcodec/utils.c
libavcodec/vaapi_mpeg.c
libavcodec/vaapi_mpeg2.c
libavcodec/vaapi_mpeg4.c
libavcodec/vaapi_vc1.c
libavcodec/vc1.c
libavcodec/vc1dec.c
libavcodec/vdpau.c
libavcodec/vdpau_mpeg12.c
libavcodec/vdpau_mpeg4.c
libavcodec/vdpau_vc1.c

Simple merge
Simple merge
@@@ -641,12 -632,12 +641,12 @@@ retry
      }
      ff_MPV_frame_end(s);
  
-     av_assert0(s->current_picture.f.pict_type == s->current_picture_ptr->f.pict_type);
-     av_assert0(s->current_picture.f.pict_type == s->pict_type);
 -    assert(s->current_picture.f->pict_type == s->current_picture_ptr->f->pict_type);
 -    assert(s->current_picture.f->pict_type == s->pict_type);
++    av_assert0(s->current_picture.f->pict_type == s->current_picture_ptr->f->pict_type);
++    av_assert0(s->current_picture.f->pict_type == s->pict_type);
  
-     if ((ret = av_frame_ref(pict, &s->current_picture_ptr->f)) < 0)
+     if ((ret = av_frame_ref(pict, s->current_picture_ptr->f)) < 0)
          return ret;
 -    ff_print_debug_info(s, s->current_picture_ptr);
 +    ff_print_debug_info(s, s->current_picture_ptr, pict);
  
      *got_frame = 1;
  
@@@ -640,40 -611,23 +640,40 @@@ frame_end
      if (!s->divx_packed && avctx->hwaccel)
          ff_thread_finish_setup(avctx);
  
-     av_assert1(s->current_picture.f.pict_type == s->current_picture_ptr->f.pict_type);
-     av_assert1(s->current_picture.f.pict_type == s->pict_type);
 -    assert(s->current_picture.f->pict_type ==
 -           s->current_picture_ptr->f->pict_type);
 -    assert(s->current_picture.f->pict_type == s->pict_type);
++    av_assert1(s->current_picture.f->pict_type == s->current_picture_ptr->f->pict_type);
++    av_assert1(s->current_picture.f->pict_type == s->pict_type);
      if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
-         if ((ret = av_frame_ref(pict, &s->current_picture_ptr->f)) < 0)
+         if ((ret = av_frame_ref(pict, s->current_picture_ptr->f)) < 0)
              return ret;
 -        ff_print_debug_info(s, s->current_picture_ptr);
 +        ff_print_debug_info(s, s->current_picture_ptr, pict);
 +        ff_mpv_export_qp_table(s, pict, s->current_picture_ptr, FF_QSCALE_TYPE_MPEG1);
      } else if (s->last_picture_ptr != NULL) {
-         if ((ret = av_frame_ref(pict, &s->last_picture_ptr->f)) < 0)
+         if ((ret = av_frame_ref(pict, s->last_picture_ptr->f)) < 0)
              return ret;
 -        ff_print_debug_info(s, s->last_picture_ptr);
 +        ff_print_debug_info(s, s->last_picture_ptr, pict);
 +        ff_mpv_export_qp_table(s, pict, s->last_picture_ptr, FF_QSCALE_TYPE_MPEG1);
      }
  
 -    if (s->last_picture_ptr || s->low_delay)
 +    if (s->last_picture_ptr || s->low_delay) {
 +        if (   pict->format == AV_PIX_FMT_YUV420P
 +            && (s->codec_tag == AV_RL32("GEOV") || s->codec_tag == AV_RL32("GEOX"))) {
 +            int x, y, p;
 +            av_frame_make_writable(pict);
 +            for (p=0; p<3; p++) {
 +                int w = FF_CEIL_RSHIFT(pict-> width, !!p);
 +                int h = FF_CEIL_RSHIFT(pict->height, !!p);
 +                int linesize = pict->linesize[p];
 +                for (y=0; y<(h>>1); y++)
 +                    for (x=0; x<w; x++)
 +                        FFSWAP(int,
 +                               pict->data[p][x + y*linesize],
 +                               pict->data[p][x + (h-1-y)*linesize]);
 +            }
 +        }
          *got_frame = 1;
 +    }
  
 -    if (ret && (avctx->err_recognition & AV_EF_EXPLODE))
 +    if (slice_ret < 0 && (avctx->err_recognition & AV_EF_EXPLODE))
          return ret;
      else
          return get_consumed_bytes(s, buf_size);
@@@ -640,11 -643,11 +640,11 @@@ static int x8_decode_intra_mb(IntraX8Co
      }else{
          w->dsp.spatial_compensation[w->orient]( s->edge_emu_buffer,
                                              s->dest[chroma],
-                                             s->current_picture.f.linesize[!!chroma] );
+                                             s->current_picture.f->linesize[!!chroma] );
      }
      if(!zeros_only)
 -        s->dsp.idct_add ( s->dest[chroma],
 +        w->wdsp.idct_add (s->dest[chroma],
-                           s->current_picture.f.linesize[!!chroma],
+                           s->current_picture.f->linesize[!!chroma],
                            s->block[0] );
  
  block_placed:
@@@ -881,11 -863,11 +881,11 @@@ void ff_estimate_p_frame_motion(MpegEnc
      int mb_type=0;
      Picture * const pic= &s->current_picture;
  
-     init_ref(c, s->new_picture.f.data, s->last_picture.f.data, NULL, 16*mb_x, 16*mb_y, 0);
+     init_ref(c, s->new_picture.f->data, s->last_picture.f->data, NULL, 16*mb_x, 16*mb_y, 0);
  
 -    assert(s->quarter_sample==0 || s->quarter_sample==1);
 -    assert(s->linesize == c->stride);
 -    assert(s->uvlinesize == c->uvstride);
 +    av_assert0(s->quarter_sample==0 || s->quarter_sample==1);
 +    av_assert0(s->linesize == c->stride);
 +    av_assert0(s->uvlinesize == c->uvstride);
  
      c->penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
      c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
@@@ -1080,9 -1062,9 +1080,9 @@@ int ff_pre_estimate_p_frame_motion(Mpeg
      int P[10][2];
      const int shift= 1+s->quarter_sample;
      const int xy= mb_x + mb_y*s->mb_stride;
-     init_ref(c, s->new_picture.f.data, s->last_picture.f.data, NULL, 16*mb_x, 16*mb_y, 0);
+     init_ref(c, s->new_picture.f->data, s->last_picture.f->data, NULL, 16*mb_x, 16*mb_y, 0);
  
 -    assert(s->quarter_sample==0 || s->quarter_sample==1);
 +    av_assert0(s->quarter_sample==0 || s->quarter_sample==1);
  
      c->pre_penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_pre_cmp);
      c->current_mv_penalty= c->mv_penalty[s->f_code] + MAX_MV;
@@@ -1566,14 -1517,9 +1566,14 @@@ static void mpeg_decode_picture_coding_
                  s->pict_type = AV_PICTURE_TYPE_P;
          } else
              s->pict_type = AV_PICTURE_TYPE_B;
-         s->current_picture.f.pict_type = s->pict_type;
-         s->current_picture.f.key_frame = s->pict_type == AV_PICTURE_TYPE_I;
+         s->current_picture.f->pict_type = s->pict_type;
+         s->current_picture.f->key_frame = s->pict_type == AV_PICTURE_TYPE_I;
      }
 +    s->mpeg_f_code[0][0] += !s->mpeg_f_code[0][0];
 +    s->mpeg_f_code[0][1] += !s->mpeg_f_code[0][1];
 +    s->mpeg_f_code[1][0] += !s->mpeg_f_code[1][0];
 +    s->mpeg_f_code[1][1] += !s->mpeg_f_code[1][1];
 +
      s->intra_dc_precision         = get_bits(&s->gb, 2);
      s->picture_structure          = get_bits(&s->gb, 2);
      s->top_field_first            = get_bits1(&s->gb);
@@@ -2028,22 -2004,20 +2028,22 @@@ static int slice_end(AVCodecContext *av
          ff_MPV_frame_end(s);
  
          if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
-             int ret = av_frame_ref(pict, &s->current_picture_ptr->f);
+             int ret = av_frame_ref(pict, s->current_picture_ptr->f);
              if (ret < 0)
                  return ret;
 -            ff_print_debug_info(s, s->current_picture_ptr);
 +            ff_print_debug_info(s, s->current_picture_ptr, pict);
 +            ff_mpv_export_qp_table(s, pict, s->current_picture_ptr, FF_QSCALE_TYPE_MPEG2);
          } else {
              if (avctx->active_thread_type & FF_THREAD_FRAME)
                  s->picture_number++;
              /* latency of 1 frame for I- and P-frames */
              /* XXX: use another variable than picture_number */
              if (s->last_picture_ptr != NULL) {
-                 int ret = av_frame_ref(pict, &s->last_picture_ptr->f);
+                 int ret = av_frame_ref(pict, s->last_picture_ptr->f);
                  if (ret < 0)
                      return ret;
 -                ff_print_debug_info(s, s->last_picture_ptr);
 +                ff_print_debug_info(s, s->last_picture_ptr, pict);
 +                ff_mpv_export_qp_table(s, pict, s->last_picture_ptr, FF_QSCALE_TYPE_MPEG2);
              }
          }
  
@@@ -336,15 -293,19 +336,15 @@@ static void mpeg1_encode_sequence_heade
          /* time code: we must convert from the real frame rate to a
           * fake MPEG frame rate in case of low frame rate */
          fps       = (framerate.num + framerate.den / 2) / framerate.den;
-         time_code = s->current_picture_ptr->f.coded_picture_number +
+         time_code = s->current_picture_ptr->f->coded_picture_number +
                      s->avctx->timecode_frame_start;
  
-         s->gop_picture_number = s->current_picture_ptr->f.coded_picture_number;
+         s->gop_picture_number = s->current_picture_ptr->f->coded_picture_number;
 -        if (s->drop_frame_timecode) {
 -            /* only works for NTSC 29.97 */
 -            int d = time_code / 17982;
 -            int m = time_code % 17982;
 -            /* not needed since -2,-1 / 1798 in C returns 0 */
 -            // if (m < 2)
 -            //     m += 2;
 -            time_code += 18 * d + 2 * ((m - 2) / 1798);
 -        }
 +
 +        av_assert0(s->drop_frame_timecode == !!(s->tc.flags & AV_TIMECODE_FLAG_DROPFRAME));
 +        if (s->drop_frame_timecode)
 +            time_code = av_timecode_adjust_ntsc_framenum2(time_code, fps);
 +
          put_bits(&s->pb, 5, (uint32_t)((time_code / (fps * 3600)) % 24));
          put_bits(&s->pb, 6, (uint32_t)((time_code / (fps *   60)) % 60));
          put_bits(&s->pb, 1, 1);
@@@ -669,9 -669,13 +669,9 @@@ void ff_mpeg4_encode_mb(MpegEncContext 
  
                      x = s->mb_x * 16;
                      y = s->mb_y * 16;
 -                    if (x + 16 > s->width)
 -                        x = s->width - 16;
 -                    if (y + 16 > s->height)
 -                        y = s->height - 16;
  
                      offset = x + y * s->linesize;
-                     p_pic  = s->new_picture.f.data[0] + offset;
+                     p_pic  = s->new_picture.f->data[0] + offset;
  
                      s->mb_skipped = 1;
                      for (i = 0; i < s->max_b_frames; i++) {
                          int diff;
                          Picture *pic = s->reordered_input_picture[i + 1];
  
-                         if (!pic || pic->f.pict_type != AV_PICTURE_TYPE_B)
+                         if (!pic || pic->f->pict_type != AV_PICTURE_TYPE_B)
                              break;
  
-                         b_pic = pic->f.data[0] + offset;
+                         b_pic = pic->f->data[0] + offset;
                          if (!pic->shared)
                              b_pic += INPLACE_OFFSET;
 -                        diff = s->dsp.sad[0](NULL, p_pic, b_pic, s->linesize, 16);
 +
 +                        if (x + 16 > s->width || y + 16 > s->height) {
 +                            int x1, y1;
 +                            int xe = FFMIN(16, s->width - x);
 +                            int ye = FFMIN(16, s->height - y);
 +                            diff = 0;
 +                            for (y1 = 0; y1 < ye; y1++) {
 +                                for (x1 = 0; x1 < xe; x1++) {
 +                                    diff += FFABS(p_pic[x1 + y1 * s->linesize] - b_pic[x1 + y1 * s->linesize]);
 +                                }
 +                            }
 +                            diff = diff * 256 / (xe * ye);
 +                        } else {
 +                            diff = s->dsp.sad[0](NULL, p_pic, b_pic, s->linesize, 16);
 +                        }
                          if (diff > s->qscale * 70) {  // FIXME check that 70 is optimal
                              s->mb_skipped = 0;
                              break;
@@@ -916,16 -906,17 +916,16 @@@ static void mpeg4_encode_gop_header(Mpe
      put_bits(&s->pb, 16, 0);
      put_bits(&s->pb, 16, GOP_STARTCODE);
  
-     time = s->current_picture_ptr->f.pts;
+     time = s->current_picture_ptr->f->pts;
      if (s->reordered_input_picture[1])
-         time = FFMIN(time, s->reordered_input_picture[1]->f.pts);
+         time = FFMIN(time, s->reordered_input_picture[1]->f->pts);
      time = time * s->avctx->time_base.num;
 +    s->last_time_base = FFUDIV(time, s->avctx->time_base.den);
  
 -    seconds  = time / s->avctx->time_base.den;
 -    minutes  = seconds / 60;
 -    seconds %= 60;
 -    hours    = minutes / 60;
 -    minutes %= 60;
 -    hours   %= 24;
 +    seconds = FFUDIV(time, s->avctx->time_base.den);
 +    minutes = FFUDIV(seconds, 60); seconds = FFUMOD(seconds, 60);
 +    hours   = FFUDIV(minutes, 60); minutes = FFUMOD(minutes, 60);
 +    hours   = FFUMOD(hours  , 24);
  
      put_bits(&s->pb, 5, hours);
      put_bits(&s->pb, 6, minutes);
@@@ -331,12 -342,11 +331,12 @@@ static void mpeg_er_decode_mb(void *opa
  
      s->dsp.clear_blocks(s->block[0]);
  
-     s->dest[0] = s->current_picture.f.data[0] + (s->mb_y *  16                       * s->linesize)   + s->mb_x *  16;
-     s->dest[1] = s->current_picture.f.data[1] + (s->mb_y * (16 >> s->chroma_y_shift) * s->uvlinesize) + s->mb_x * (16 >> s->chroma_x_shift);
-     s->dest[2] = s->current_picture.f.data[2] + (s->mb_y * (16 >> s->chroma_y_shift) * s->uvlinesize) + s->mb_x * (16 >> s->chroma_x_shift);
+     s->dest[0] = s->current_picture.f->data[0] + (s->mb_y *  16                       * s->linesize)   + s->mb_x *  16;
+     s->dest[1] = s->current_picture.f->data[1] + (s->mb_y * (16 >> s->chroma_y_shift) * s->uvlinesize) + s->mb_x * (16 >> s->chroma_x_shift);
+     s->dest[2] = s->current_picture.f->data[2] + (s->mb_y * (16 >> s->chroma_y_shift) * s->uvlinesize) + s->mb_x * (16 >> s->chroma_x_shift);
  
 -    assert(ref == 0);
 +    if (ref)
 +        av_log(s->avctx, AV_LOG_DEBUG, "Interlaced error concealment is not fully implemented\n");
      ff_MPV_decode_mb(s, s->block);
  }
  
@@@ -583,16 -584,11 +583,16 @@@ int ff_alloc_picture(MpegEncContext *s
  {
      int i, ret;
  
 +    if (pic->qscale_table_buf)
 +        if (   pic->alloc_mb_width  != s->mb_width
 +            || pic->alloc_mb_height != s->mb_height)
 +            ff_free_picture_tables(pic);
 +
      if (shared) {
-         av_assert0(pic->f.data[0]);
 -        assert(pic->f->data[0]);
++        av_assert0(pic->f->data[0]);
          pic->shared = 1;
      } else {
-         av_assert0(!pic->f.buf[0]);
 -        assert(!pic->f->buf[0]);
++        av_assert0(!pic->f->buf[0]);
  
          if (alloc_frame_buffer(s, pic) < 0)
              return -1;
@@@ -919,11 -900,9 +919,11 @@@ int ff_mpeg_update_thread_context(AVCod
      s->coded_picture_number = s1->coded_picture_number;
      s->picture_number       = s1->picture_number;
  
 +    av_assert0(!s->picture || s->picture != s1->picture);
 +    if(s->picture)
      for (i = 0; i < MAX_PICTURE_COUNT; i++) {
          ff_mpeg_unref_picture(s, &s->picture[i]);
-         if (s1->picture[i].f.buf[0] &&
+         if (s1->picture[i].f->buf[0] &&
              (ret = ff_mpeg_ref_picture(s, &s->picture[i], &s1->picture[i])) < 0)
              return ret;
      }
@@@ -1262,10 -1259,21 +1264,20 @@@ av_cold int ff_MPV_common_init(MpegEncC
      memset(&s->next_picture, 0, sizeof(s->next_picture));
      memset(&s->last_picture, 0, sizeof(s->last_picture));
      memset(&s->current_picture, 0, sizeof(s->current_picture));
-     av_frame_unref(&s->next_picture.f);
-     av_frame_unref(&s->last_picture.f);
-     av_frame_unref(&s->current_picture.f);
+     memset(&s->new_picture, 0, sizeof(s->new_picture));
+     s->next_picture.f = av_frame_alloc();
+     if (!s->next_picture.f)
+         goto fail;
+     s->last_picture.f = av_frame_alloc();
+     if (!s->last_picture.f)
+         goto fail;
+     s->current_picture.f = av_frame_alloc();
+     if (!s->current_picture.f)
+         goto fail;
+     s->new_picture.f = av_frame_alloc();
+     if (!s->new_picture.f)
+         goto fail;
  
 -    if (s->width && s->height) {
          if (init_context_frame(s))
              goto fail;
  
@@@ -1582,9 -1596,7 +1601,9 @@@ static void release_unused_pictures(Mpe
  
  static inline int pic_is_unused(MpegEncContext *s, Picture *pic)
  {
-     if (pic->f.buf[0] == NULL)
 +    if (pic == s->last_picture_ptr)
 +        return 0;
+     if (pic->f->buf[0] == NULL)
          return 1;
      if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
          return 1;
@@@ -1597,7 -1609,7 +1616,7 @@@ static int find_unused_picture(MpegEncC
  
      if (shared) {
          for (i = 0; i < MAX_PICTURE_COUNT; i++) {
-             if (s->picture[i].f.buf[0] == NULL && &s->picture[i] != s->last_picture_ptr)
 -            if (s->picture[i].f->buf[0] == NULL)
++            if (s->picture[i].f->buf[0] == NULL && &s->picture[i] != s->last_picture_ptr)
                  return i;
          }
      } else {
@@@ -1743,10 -1736,7 +1762,10 @@@ int ff_MPV_frame_start(MpegEncContext *
          int h_chroma_shift, v_chroma_shift;
          av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
                                           &h_chroma_shift, &v_chroma_shift);
-         if (s->pict_type == AV_PICTURE_TYPE_B && s->next_picture_ptr && s->next_picture_ptr->f.buf[0])
 -        if (s->pict_type != AV_PICTURE_TYPE_I)
++        if (s->pict_type == AV_PICTURE_TYPE_B && s->next_picture_ptr && s->next_picture_ptr->f->buf[0])
 +            av_log(avctx, AV_LOG_DEBUG,
 +                   "allocating dummy last picture for B frame\n");
 +        else if (s->pict_type != AV_PICTURE_TYPE_I)
              av_log(avctx, AV_LOG_ERROR,
                     "warning: first frame is no keyframe\n");
          else if (s->picture_structure != PICT_FRAME)
          s->last_picture_ptr = &s->picture[i];
  
          s->last_picture_ptr->reference   = 3;
-         s->last_picture_ptr->f.key_frame = 0;
-         s->last_picture_ptr->f.pict_type = AV_PICTURE_TYPE_P;
 -        s->last_picture_ptr->f->pict_type = AV_PICTURE_TYPE_I;
++        s->last_picture_ptr->f->key_frame = 0;
++        s->last_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
  
          if (ff_alloc_picture(s, s->last_picture_ptr, 0) < 0) {
              s->last_picture_ptr = NULL;
              return -1;
          }
  
 -        memset(s->last_picture_ptr->f->data[0], 0,
 -               avctx->height * s->last_picture_ptr->f->linesize[0]);
 -        memset(s->last_picture_ptr->f->data[1], 0x80,
 -               (avctx->height >> v_chroma_shift) *
 -               s->last_picture_ptr->f->linesize[1]);
 -        memset(s->last_picture_ptr->f->data[2], 0x80,
 -               (avctx->height >> v_chroma_shift) *
 -               s->last_picture_ptr->f->linesize[2]);
 +        if (!avctx->hwaccel) {
 +            for(i=0; i<avctx->height; i++)
-                 memset(s->last_picture_ptr->f.data[0] + s->last_picture_ptr->f.linesize[0]*i,
++                memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i,
 +                       0x80, avctx->width);
 +            for(i=0; i<FF_CEIL_RSHIFT(avctx->height, v_chroma_shift); i++) {
-                 memset(s->last_picture_ptr->f.data[1] + s->last_picture_ptr->f.linesize[1]*i,
++                memset(s->last_picture_ptr->f->data[1] + s->last_picture_ptr->f->linesize[1]*i,
 +                       0x80, FF_CEIL_RSHIFT(avctx->width, h_chroma_shift));
-                 memset(s->last_picture_ptr->f.data[2] + s->last_picture_ptr->f.linesize[2]*i,
++                memset(s->last_picture_ptr->f->data[2] + s->last_picture_ptr->f->linesize[2]*i,
 +                       0x80, FF_CEIL_RSHIFT(avctx->width, h_chroma_shift));
 +            }
 +
 +            if(s->codec_id == AV_CODEC_ID_FLV1 || s->codec_id == AV_CODEC_ID_H263){
 +                for(i=0; i<avctx->height; i++)
-                 memset(s->last_picture_ptr->f.data[0] + s->last_picture_ptr->f.linesize[0]*i, 16, avctx->width);
++                memset(s->last_picture_ptr->f->data[0] + s->last_picture_ptr->f->linesize[0]*i, 16, avctx->width);
 +            }
 +        }
  
          ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 0);
          ff_thread_report_progress(&s->last_picture_ptr->tf, INT_MAX, 1);
          s->next_picture_ptr = &s->picture[i];
  
          s->next_picture_ptr->reference   = 3;
-         s->next_picture_ptr->f.key_frame = 0;
-         s->next_picture_ptr->f.pict_type = AV_PICTURE_TYPE_P;
 -        s->next_picture_ptr->f->pict_type = AV_PICTURE_TYPE_I;
++        s->next_picture_ptr->f->key_frame = 0;
++        s->next_picture_ptr->f->pict_type = AV_PICTURE_TYPE_P;
  
          if (ff_alloc_picture(s, s->next_picture_ptr, 0) < 0) {
              s->next_picture_ptr = NULL;
          ff_thread_report_progress(&s->next_picture_ptr->tf, INT_MAX, 1);
      }
  
-     memset(s->last_picture.f.data, 0, sizeof(s->last_picture.f.data));
-     memset(s->next_picture.f.data, 0, sizeof(s->next_picture.f.data));
 +#if 0 // BUFREF-FIXME
++    memset(s->last_picture.f->data, 0, sizeof(s->last_picture.f->data));
++    memset(s->next_picture.f->data, 0, sizeof(s->next_picture.f->data));
 +#endif
      if (s->last_picture_ptr) {
          ff_mpeg_unref_picture(s, &s->last_picture);
-         if (s->last_picture_ptr->f.buf[0] &&
+         if (s->last_picture_ptr->f->buf[0] &&
              (ret = ff_mpeg_ref_picture(s, &s->last_picture,
                                         s->last_picture_ptr)) < 0)
              return ret;
              return ret;
      }
  
 -    if (s->pict_type != AV_PICTURE_TYPE_I &&
 -        !(s->last_picture_ptr && s->last_picture_ptr->f->buf[0])) {
 -        av_log(s, AV_LOG_ERROR,
 -               "Non-reference picture received and no reference available\n");
 -        return AVERROR_INVALIDDATA;
 -    }
 +    av_assert0(s->pict_type == AV_PICTURE_TYPE_I || (s->last_picture_ptr &&
-                                                  s->last_picture_ptr->f.buf[0]));
++                                                 s->last_picture_ptr->f->buf[0]));
  
      if (s->picture_structure!= PICT_FRAME) {
          int i;
@@@ -2038,620 -1963,13 +2057,620 @@@ void ff_print_debug_info2(AVCodecContex
  
  
                      if (IS_INTERLACED(mb_type))
 -                        av_log(s->avctx, AV_LOG_DEBUG, "=");
 +                        av_log(avctx, AV_LOG_DEBUG, "=");
                      else
 -                        av_log(s->avctx, AV_LOG_DEBUG, " ");
 +                        av_log(avctx, AV_LOG_DEBUG, " ");
 +                }
 +            }
 +            av_log(avctx, AV_LOG_DEBUG, "\n");
 +        }
 +    }
 +
 +    if ((avctx->debug & (FF_DEBUG_VIS_QP | FF_DEBUG_VIS_MB_TYPE)) ||
 +        (avctx->debug_mv)) {
 +        const int shift = 1 + quarter_sample;
 +        int mb_y;
 +        uint8_t *ptr;
 +        int i;
 +        int h_chroma_shift, v_chroma_shift, block_height;
 +        const int width          = avctx->width;
 +        const int height         = avctx->height;
 +        const int mv_sample_log2 = avctx->codec_id == AV_CODEC_ID_H264 || avctx->codec_id == AV_CODEC_ID_SVQ3 ? 2 : 1;
 +        const int mv_stride      = (mb_width << mv_sample_log2) +
 +                                   (avctx->codec->id == AV_CODEC_ID_H264 ? 0 : 1);
 +
 +        *low_delay = 0; // needed to see the vectors without trashing the buffers
 +
 +        avcodec_get_chroma_sub_sample(avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
 +
 +        av_frame_make_writable(pict);
 +
 +        pict->opaque = NULL;
 +        ptr          = pict->data[0];
 +        block_height = 16 >> v_chroma_shift;
 +
 +        for (mb_y = 0; mb_y < mb_height; mb_y++) {
 +            int mb_x;
 +            for (mb_x = 0; mb_x < mb_width; mb_x++) {
 +                const int mb_index = mb_x + mb_y * mb_stride;
 +                if ((avctx->debug_mv) && motion_val[0]) {
 +                    int type;
 +                    for (type = 0; type < 3; type++) {
 +                        int direction = 0;
 +                        switch (type) {
 +                        case 0:
 +                            if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_P_FOR)) ||
 +                                (pict->pict_type!= AV_PICTURE_TYPE_P))
 +                                continue;
 +                            direction = 0;
 +                            break;
 +                        case 1:
 +                            if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_B_FOR)) ||
 +                                (pict->pict_type!= AV_PICTURE_TYPE_B))
 +                                continue;
 +                            direction = 0;
 +                            break;
 +                        case 2:
 +                            if ((!(avctx->debug_mv & FF_DEBUG_VIS_MV_B_BACK)) ||
 +                                (pict->pict_type!= AV_PICTURE_TYPE_B))
 +                                continue;
 +                            direction = 1;
 +                            break;
 +                        }
 +                        if (!USES_LIST(mbtype_table[mb_index], direction))
 +                            continue;
 +
 +                        if (IS_8X8(mbtype_table[mb_index])) {
 +                            int i;
 +                            for (i = 0; i < 4; i++) {
 +                                int sx = mb_x * 16 + 4 + 8 * (i & 1);
 +                                int sy = mb_y * 16 + 4 + 8 * (i >> 1);
 +                                int xy = (mb_x * 2 + (i & 1) +
 +                                          (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1);
 +                                int mx = (motion_val[direction][xy][0] >> shift) + sx;
 +                                int my = (motion_val[direction][xy][1] >> shift) + sy;
 +                                draw_arrow(ptr, sx, sy, mx, my, width,
 +                                           height, pict->linesize[0], 100);
 +                            }
 +                        } else if (IS_16X8(mbtype_table[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 = (mb_x * 2 + (mb_y * 2 + i) * mv_stride) << (mv_sample_log2 - 1);
 +                                int mx = (motion_val[direction][xy][0] >> shift);
 +                                int my = (motion_val[direction][xy][1] >> shift);
 +
 +                                if (IS_INTERLACED(mbtype_table[mb_index]))
 +                                    my *= 2;
 +
 +                            draw_arrow(ptr, sx, sy, mx + sx, my + sy, width,
 +                                       height, pict->linesize[0], 100);
 +                            }
 +                        } else if (IS_8X16(mbtype_table[mb_index])) {
 +                            int i;
 +                            for (i = 0; i < 2; i++) {
 +                                int sx = mb_x * 16 + 4 + 8 * i;
 +                                int sy = mb_y * 16 + 8;
 +                                int xy = (mb_x * 2 + i + mb_y * 2 * mv_stride) << (mv_sample_log2 - 1);
 +                                int mx = motion_val[direction][xy][0] >> shift;
 +                                int my = motion_val[direction][xy][1] >> shift;
 +
 +                                if (IS_INTERLACED(mbtype_table[mb_index]))
 +                                    my *= 2;
 +
 +                                draw_arrow(ptr, sx, sy, mx + sx, my + sy, width,
 +                                           height, pict->linesize[0], 100);
 +                            }
 +                        } else {
 +                              int sx= mb_x * 16 + 8;
 +                              int sy= mb_y * 16 + 8;
 +                              int xy= (mb_x + mb_y * mv_stride) << mv_sample_log2;
 +                              int mx= (motion_val[direction][xy][0]>>shift) + sx;
 +                              int my= (motion_val[direction][xy][1]>>shift) + sy;
 +                              draw_arrow(ptr, sx, sy, mx, my, width, height, pict->linesize[0], 100);
 +                        }
 +                    }
 +                }
 +                if ((avctx->debug & FF_DEBUG_VIS_QP)) {
 +                    uint64_t c = (qscale_table[mb_index] * 128 / 31) *
 +                                 0x0101010101010101ULL;
 +                    int y;
 +                    for (y = 0; y < block_height; y++) {
 +                        *(uint64_t *)(pict->data[1] + 8 * mb_x +
 +                                      (block_height * mb_y + y) *
 +                                      pict->linesize[1]) = c;
 +                        *(uint64_t *)(pict->data[2] + 8 * mb_x +
 +                                      (block_height * mb_y + y) *
 +                                      pict->linesize[2]) = c;
 +                    }
 +                }
 +                if ((avctx->debug & FF_DEBUG_VIS_MB_TYPE) &&
 +                    motion_val[0]) {
 +                    int mb_type = mbtype_table[mb_index];
 +                    uint64_t u,v;
 +                    int y;
 +#define COLOR(theta, r) \
 +    u = (int)(128 + r * cos(theta * 3.141592 / 180)); \
 +    v = (int)(128 + r * sin(theta * 3.141592 / 180));
 +
 +
 +                    u = v = 128;
 +                    if (IS_PCM(mb_type)) {
 +                        COLOR(120, 48)
 +                    } else if ((IS_INTRA(mb_type) && IS_ACPRED(mb_type)) ||
 +                               IS_INTRA16x16(mb_type)) {
 +                        COLOR(30, 48)
 +                    } else if (IS_INTRA4x4(mb_type)) {
 +                        COLOR(90, 48)
 +                    } else if (IS_DIRECT(mb_type) && IS_SKIP(mb_type)) {
 +                        // COLOR(120, 48)
 +                    } else if (IS_DIRECT(mb_type)) {
 +                        COLOR(150, 48)
 +                    } else if (IS_GMC(mb_type) && IS_SKIP(mb_type)) {
 +                        COLOR(170, 48)
 +                    } else if (IS_GMC(mb_type)) {
 +                        COLOR(190, 48)
 +                    } else if (IS_SKIP(mb_type)) {
 +                        // COLOR(180, 48)
 +                    } else if (!USES_LIST(mb_type, 1)) {
 +                        COLOR(240, 48)
 +                    } else if (!USES_LIST(mb_type, 0)) {
 +                        COLOR(0, 48)
 +                    } else {
 +                        av_assert2(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
 +                        COLOR(300,48)
 +                    }
 +
 +                    u *= 0x0101010101010101ULL;
 +                    v *= 0x0101010101010101ULL;
 +                    for (y = 0; y < block_height; y++) {
 +                        *(uint64_t *)(pict->data[1] + 8 * mb_x +
 +                                      (block_height * mb_y + y) * pict->linesize[1]) = u;
 +                        *(uint64_t *)(pict->data[2] + 8 * mb_x +
 +                                      (block_height * mb_y + y) * pict->linesize[2]) = v;
 +                    }
 +
 +                    // segmentation
 +                    if (IS_8X8(mb_type) || IS_16X8(mb_type)) {
 +                        *(uint64_t *)(pict->data[0] + 16 * mb_x + 0 +
 +                                      (16 * mb_y + 8) * pict->linesize[0]) ^= 0x8080808080808080ULL;
 +                        *(uint64_t *)(pict->data[0] + 16 * mb_x + 8 +
 +                                      (16 * mb_y + 8) * pict->linesize[0]) ^= 0x8080808080808080ULL;
 +                    }
 +                    if (IS_8X8(mb_type) || IS_8X16(mb_type)) {
 +                        for (y = 0; y < 16; y++)
 +                            pict->data[0][16 * mb_x + 8 + (16 * mb_y + y) *
 +                                          pict->linesize[0]] ^= 0x80;
 +                    }
 +                    if (IS_8X8(mb_type) && mv_sample_log2 >= 2) {
 +                        int dm = 1 << (mv_sample_log2 - 2);
 +                        for (i = 0; i < 4; i++) {
 +                            int sx = mb_x * 16 + 8 * (i & 1);
 +                            int sy = mb_y * 16 + 8 * (i >> 1);
 +                            int xy = (mb_x * 2 + (i & 1) +
 +                                     (mb_y * 2 + (i >> 1)) * mv_stride) << (mv_sample_log2 - 1);
 +                            // FIXME bidir
 +                            int32_t *mv = (int32_t *) &motion_val[0][xy];
 +                            if (mv[0] != mv[dm] ||
 +                                mv[dm * mv_stride] != mv[dm * (mv_stride + 1)])
 +                                for (y = 0; y < 8; y++)
 +                                    pict->data[0][sx + 4 + (sy + y) * pict->linesize[0]] ^= 0x80;
 +                            if (mv[0] != mv[dm * mv_stride] || mv[dm] != mv[dm * (mv_stride + 1)])
 +                                *(uint64_t *)(pict->data[0] + sx + (sy + 4) *
 +                                              pict->linesize[0]) ^= 0x8080808080808080ULL;
 +                        }
 +                    }
 +
 +                    if (IS_INTERLACED(mb_type) &&
 +                        avctx->codec->id == AV_CODEC_ID_H264) {
 +                        // hmm
 +                    }
 +                }
 +                mbskip_table[mb_index] = 0;
 +            }
 +        }
 +    }
 +}
 +
 +void ff_print_debug_info(MpegEncContext *s, Picture *p, AVFrame *pict)
 +{
 +    ff_print_debug_info2(s->avctx, pict, s->mbskip_table, p->mb_type,
 +                         p->qscale_table, p->motion_val, &s->low_delay,
 +                         s->mb_width, s->mb_height, s->mb_stride, s->quarter_sample);
 +}
 +
 +int ff_mpv_export_qp_table(MpegEncContext *s, AVFrame *f, Picture *p, int qp_type)
 +{
 +    AVBufferRef *ref = av_buffer_ref(p->qscale_table_buf);
 +    int offset = 2*s->mb_stride + 1;
 +    if(!ref)
 +        return AVERROR(ENOMEM);
 +    av_assert0(ref->size >= offset + s->mb_stride * ((f->height+15)/16));
 +    ref->size -= offset;
 +    ref->data += offset;
 +    return av_frame_set_qp_table(f, ref, s->mb_stride, qp_type);
 +}
 +
 +static inline int hpel_motion_lowres(MpegEncContext *s,
 +                                     uint8_t *dest, uint8_t *src,
 +                                     int field_based, int field_select,
 +                                     int src_x, int src_y,
 +                                     int width, int height, ptrdiff_t stride,
 +                                     int h_edge_pos, int v_edge_pos,
 +                                     int w, int h, h264_chroma_mc_func *pix_op,
 +                                     int motion_x, int motion_y)
 +{
 +    const int lowres   = s->avctx->lowres;
 +    const int op_index = FFMIN(lowres, 3);
 +    const int s_mask   = (2 << lowres) - 1;
 +    int emu = 0;
 +    int sx, sy;
 +
 +    if (s->quarter_sample) {
 +        motion_x /= 2;
 +        motion_y /= 2;
 +    }
 +
 +    sx = motion_x & s_mask;
 +    sy = motion_y & s_mask;
 +    src_x += motion_x >> lowres + 1;
 +    src_y += motion_y >> lowres + 1;
 +
 +    src   += src_y * stride + src_x;
 +
 +    if ((unsigned)src_x > FFMAX( h_edge_pos - (!!sx) - w,                 0) ||
 +        (unsigned)src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
 +        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src,
 +                                 s->linesize, s->linesize,
 +                                 w + 1, (h + 1) << field_based,
 +                                 src_x, src_y   << field_based,
 +                                 h_edge_pos, v_edge_pos);
 +        src = s->edge_emu_buffer;
 +        emu = 1;
 +    }
 +
 +    sx = (sx << 2) >> lowres;
 +    sy = (sy << 2) >> lowres;
 +    if (field_select)
 +        src += s->linesize;
 +    pix_op[op_index](dest, src, stride, h, sx, sy);
 +    return emu;
 +}
 +
 +/* apply one mpeg motion vector to the three components */
 +static av_always_inline void mpeg_motion_lowres(MpegEncContext *s,
 +                                                uint8_t *dest_y,
 +                                                uint8_t *dest_cb,
 +                                                uint8_t *dest_cr,
 +                                                int field_based,
 +                                                int bottom_field,
 +                                                int field_select,
 +                                                uint8_t **ref_picture,
 +                                                h264_chroma_mc_func *pix_op,
 +                                                int motion_x, int motion_y,
 +                                                int h, int mb_y)
 +{
 +    uint8_t *ptr_y, *ptr_cb, *ptr_cr;
 +    int mx, my, src_x, src_y, uvsrc_x, uvsrc_y, sx, sy, uvsx, uvsy;
 +    ptrdiff_t uvlinesize, linesize;
 +    const int lowres     = s->avctx->lowres;
 +    const int op_index   = FFMIN(lowres-1+s->chroma_x_shift, 3);
 +    const int block_s    = 8>>lowres;
 +    const int s_mask     = (2 << lowres) - 1;
 +    const int h_edge_pos = s->h_edge_pos >> lowres;
 +    const int v_edge_pos = s->v_edge_pos >> lowres;
-     linesize   = s->current_picture.f.linesize[0] << field_based;
-     uvlinesize = s->current_picture.f.linesize[1] << field_based;
++    linesize   = s->current_picture.f->linesize[0] << field_based;
++    uvlinesize = s->current_picture.f->linesize[1] << field_based;
 +
 +    // FIXME obviously not perfect but qpel will not work in lowres anyway
 +    if (s->quarter_sample) {
 +        motion_x /= 2;
 +        motion_y /= 2;
 +    }
 +
 +    if(field_based){
 +        motion_y += (bottom_field - field_select)*((1 << lowres)-1);
 +    }
 +
 +    sx = motion_x & s_mask;
 +    sy = motion_y & s_mask;
 +    src_x = s->mb_x * 2 * block_s + (motion_x >> lowres + 1);
 +    src_y = (mb_y * 2 * block_s >> field_based) + (motion_y >> lowres + 1);
 +
 +    if (s->out_format == FMT_H263) {
 +        uvsx    = ((motion_x >> 1) & s_mask) | (sx & 1);
 +        uvsy    = ((motion_y >> 1) & s_mask) | (sy & 1);
 +        uvsrc_x = src_x >> 1;
 +        uvsrc_y = src_y >> 1;
 +    } else if (s->out_format == FMT_H261) {
 +        // even chroma mv's are full pel in H261
 +        mx      = motion_x / 4;
 +        my      = motion_y / 4;
 +        uvsx    = (2 * mx) & s_mask;
 +        uvsy    = (2 * my) & s_mask;
 +        uvsrc_x = s->mb_x * block_s + (mx >> lowres);
 +        uvsrc_y =    mb_y * block_s + (my >> lowres);
 +    } else {
 +        if(s->chroma_y_shift){
 +            mx      = motion_x / 2;
 +            my      = motion_y / 2;
 +            uvsx    = mx & s_mask;
 +            uvsy    = my & s_mask;
 +            uvsrc_x = s->mb_x * block_s                 + (mx >> lowres + 1);
 +            uvsrc_y =   (mb_y * block_s >> field_based) + (my >> lowres + 1);
 +        } else {
 +            if(s->chroma_x_shift){
 +            //Chroma422
 +                mx = motion_x / 2;
 +                uvsx = mx & s_mask;
 +                uvsy = motion_y & s_mask;
 +                uvsrc_y = src_y;
 +                uvsrc_x = s->mb_x*block_s               + (mx >> (lowres+1));
 +            } else {
 +            //Chroma444
 +                uvsx = motion_x & s_mask;
 +                uvsy = motion_y & s_mask;
 +                uvsrc_x = src_x;
 +                uvsrc_y = src_y;
 +            }
 +        }
 +    }
 +
 +    ptr_y  = ref_picture[0] + src_y   * linesize   + src_x;
 +    ptr_cb = ref_picture[1] + uvsrc_y * uvlinesize + uvsrc_x;
 +    ptr_cr = ref_picture[2] + uvsrc_y * uvlinesize + uvsrc_x;
 +
 +    if ((unsigned) src_x > FFMAX( h_edge_pos - (!!sx) - 2 * block_s,       0) || uvsrc_y<0 ||
 +        (unsigned) src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
 +        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
 +                                 linesize >> field_based, linesize >> field_based,
 +                                 17, 17 + field_based,
 +                                src_x, src_y << field_based, h_edge_pos,
 +                                v_edge_pos);
 +        ptr_y = s->edge_emu_buffer;
 +        if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
 +            uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
 +            s->vdsp.emulated_edge_mc(uvbuf,  ptr_cb,
 +                                     uvlinesize >> field_based, uvlinesize >> field_based,
 +                                     9, 9 + field_based,
 +                                    uvsrc_x, uvsrc_y << field_based,
 +                                    h_edge_pos >> 1, v_edge_pos >> 1);
 +            s->vdsp.emulated_edge_mc(uvbuf + 16,  ptr_cr,
 +                                     uvlinesize >> field_based,uvlinesize >> field_based,
 +                                     9, 9 + field_based,
 +                                    uvsrc_x, uvsrc_y << field_based,
 +                                    h_edge_pos >> 1, v_edge_pos >> 1);
 +            ptr_cb = uvbuf;
 +            ptr_cr = uvbuf + 16;
 +        }
 +    }
 +
-     // FIXME use this for field pix too instead of the obnoxious hack which changes picture.f.data
++    // FIXME use this for field pix too instead of the obnoxious hack which changes picture.f->data
 +    if (bottom_field) {
 +        dest_y  += s->linesize;
 +        dest_cb += s->uvlinesize;
 +        dest_cr += s->uvlinesize;
 +    }
 +
 +    if (field_select) {
 +        ptr_y   += s->linesize;
 +        ptr_cb  += s->uvlinesize;
 +        ptr_cr  += s->uvlinesize;
 +    }
 +
 +    sx = (sx << 2) >> lowres;
 +    sy = (sy << 2) >> lowres;
 +    pix_op[lowres - 1](dest_y, ptr_y, linesize, h, sx, sy);
 +
 +    if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
 +        int hc = s->chroma_y_shift ? (h+1-bottom_field)>>1 : h;
 +        uvsx = (uvsx << 2) >> lowres;
 +        uvsy = (uvsy << 2) >> lowres;
 +        if (hc) {
 +            pix_op[op_index](dest_cb, ptr_cb, uvlinesize, hc, uvsx, uvsy);
 +            pix_op[op_index](dest_cr, ptr_cr, uvlinesize, hc, uvsx, uvsy);
 +        }
 +    }
 +    // FIXME h261 lowres loop filter
 +}
 +
 +static inline void chroma_4mv_motion_lowres(MpegEncContext *s,
 +                                            uint8_t *dest_cb, uint8_t *dest_cr,
 +                                            uint8_t **ref_picture,
 +                                            h264_chroma_mc_func * pix_op,
 +                                            int mx, int my)
 +{
 +    const int lowres     = s->avctx->lowres;
 +    const int op_index   = FFMIN(lowres, 3);
 +    const int block_s    = 8 >> lowres;
 +    const int s_mask     = (2 << lowres) - 1;
 +    const int h_edge_pos = s->h_edge_pos >> lowres + 1;
 +    const int v_edge_pos = s->v_edge_pos >> lowres + 1;
 +    int emu = 0, src_x, src_y, sx, sy;
 +    ptrdiff_t offset;
 +    uint8_t *ptr;
 +
 +    if (s->quarter_sample) {
 +        mx /= 2;
 +        my /= 2;
 +    }
 +
 +    /* 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);
 +
 +    sx = mx & s_mask;
 +    sy = my & s_mask;
 +    src_x = s->mb_x * block_s + (mx >> lowres + 1);
 +    src_y = s->mb_y * block_s + (my >> lowres + 1);
 +
 +    offset = src_y * s->uvlinesize + src_x;
 +    ptr = ref_picture[1] + offset;
 +    if ((unsigned) src_x > FFMAX(h_edge_pos - (!!sx) - block_s, 0) ||
 +        (unsigned) src_y > FFMAX(v_edge_pos - (!!sy) - block_s, 0)) {
 +        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
 +                                 s->uvlinesize, s->uvlinesize,
 +                                 9, 9,
 +                                 src_x, src_y, h_edge_pos, v_edge_pos);
 +        ptr = s->edge_emu_buffer;
 +        emu = 1;
 +    }
 +    sx = (sx << 2) >> lowres;
 +    sy = (sy << 2) >> lowres;
 +    pix_op[op_index](dest_cb, ptr, s->uvlinesize, block_s, sx, sy);
 +
 +    ptr = ref_picture[2] + offset;
 +    if (emu) {
 +        s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr,
 +                                 s->uvlinesize, s->uvlinesize,
 +                                 9, 9,
 +                                 src_x, src_y, h_edge_pos, v_edge_pos);
 +        ptr = s->edge_emu_buffer;
 +    }
 +    pix_op[op_index](dest_cr, ptr, s->uvlinesize, block_s, sx, sy);
 +}
 +
 +/**
 + * motion compensation of a single macroblock
 + * @param s context
 + * @param dest_y luma destination pointer
 + * @param dest_cb chroma cb/u destination pointer
 + * @param dest_cr chroma cr/v destination pointer
 + * @param dir direction (0->forward, 1->backward)
 + * @param ref_picture array[3] of pointers to the 3 planes of the reference picture
 + * @param pix_op halfpel motion compensation function (average or put normally)
 + * the motion vectors are taken from s->mv and the MV type from s->mv_type
 + */
 +static inline void MPV_motion_lowres(MpegEncContext *s,
 +                                     uint8_t *dest_y, uint8_t *dest_cb,
 +                                     uint8_t *dest_cr,
 +                                     int dir, uint8_t **ref_picture,
 +                                     h264_chroma_mc_func *pix_op)
 +{
 +    int mx, my;
 +    int mb_x, mb_y, i;
 +    const int lowres  = s->avctx->lowres;
 +    const int block_s = 8 >>lowres;
 +
 +    mb_x = s->mb_x;
 +    mb_y = s->mb_y;
 +
 +    switch (s->mv_type) {
 +    case MV_TYPE_16X16:
 +        mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
 +                           0, 0, 0,
 +                           ref_picture, pix_op,
 +                           s->mv[dir][0][0], s->mv[dir][0][1],
 +                           2 * block_s, mb_y);
 +        break;
 +    case MV_TYPE_8X8:
 +        mx = 0;
 +        my = 0;
 +        for (i = 0; i < 4; i++) {
 +            hpel_motion_lowres(s, dest_y + ((i & 1) + (i >> 1) *
 +                               s->linesize) * block_s,
 +                               ref_picture[0], 0, 0,
 +                               (2 * mb_x + (i & 1)) * block_s,
 +                               (2 * mb_y + (i >> 1)) * block_s,
 +                               s->width, s->height, s->linesize,
 +                               s->h_edge_pos >> lowres, s->v_edge_pos >> lowres,
 +                               block_s, block_s, pix_op,
 +                               s->mv[dir][i][0], s->mv[dir][i][1]);
 +
 +            mx += s->mv[dir][i][0];
 +            my += s->mv[dir][i][1];
 +        }
 +
 +        if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
 +            chroma_4mv_motion_lowres(s, dest_cb, dest_cr, ref_picture,
 +                                     pix_op, mx, my);
 +        break;
 +    case MV_TYPE_FIELD:
 +        if (s->picture_structure == PICT_FRAME) {
 +            /* top field */
 +            mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
 +                               1, 0, s->field_select[dir][0],
 +                               ref_picture, pix_op,
 +                               s->mv[dir][0][0], s->mv[dir][0][1],
 +                               block_s, mb_y);
 +            /* bottom field */
 +            mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
 +                               1, 1, s->field_select[dir][1],
 +                               ref_picture, pix_op,
 +                               s->mv[dir][1][0], s->mv[dir][1][1],
 +                               block_s, mb_y);
 +        } else {
 +            if (s->picture_structure != s->field_select[dir][0] + 1 &&
 +                s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
-                 ref_picture = s->current_picture_ptr->f.data;
++                ref_picture = s->current_picture_ptr->f->data;
 +
 +            }
 +            mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
 +                               0, 0, s->field_select[dir][0],
 +                               ref_picture, pix_op,
 +                               s->mv[dir][0][0],
 +                               s->mv[dir][0][1], 2 * block_s, mb_y >> 1);
 +            }
 +        break;
 +    case MV_TYPE_16X8:
 +        for (i = 0; i < 2; i++) {
 +            uint8_t **ref2picture;
 +
 +            if (s->picture_structure == s->field_select[dir][i] + 1 ||
 +                s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
 +                ref2picture = ref_picture;
 +            } else {
-                 ref2picture = s->current_picture_ptr->f.data;
++                ref2picture = s->current_picture_ptr->f->data;
 +            }
 +
 +            mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
 +                               0, 0, s->field_select[dir][i],
 +                               ref2picture, pix_op,
 +                               s->mv[dir][i][0], s->mv[dir][i][1] +
 +                               2 * block_s * i, block_s, mb_y >> 1);
 +
 +            dest_y  +=  2 * block_s *  s->linesize;
 +            dest_cb += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
 +            dest_cr += (2 * block_s >> s->chroma_y_shift) * s->uvlinesize;
 +        }
 +        break;
 +    case MV_TYPE_DMV:
 +        if (s->picture_structure == PICT_FRAME) {
 +            for (i = 0; i < 2; i++) {
 +                int j;
 +                for (j = 0; j < 2; j++) {
 +                    mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
 +                                       1, j, j ^ i,
 +                                       ref_picture, pix_op,
 +                                       s->mv[dir][2 * i + j][0],
 +                                       s->mv[dir][2 * i + j][1],
 +                                       block_s, mb_y);
 +                }
 +                pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
 +            }
 +        } else {
 +            for (i = 0; i < 2; i++) {
 +                mpeg_motion_lowres(s, dest_y, dest_cb, dest_cr,
 +                                   0, 0, s->picture_structure != i + 1,
 +                                   ref_picture, pix_op,
 +                                   s->mv[dir][2 * i][0],s->mv[dir][2 * i][1],
 +                                   2 * block_s, mb_y >> 1);
 +
 +                // after put we make avg of the same block
 +                pix_op = s->h264chroma.avg_h264_chroma_pixels_tab;
 +
 +                // opposite parity is always in the same
 +                // frame if this is second field
 +                if (!s->first_field) {
-                     ref_picture = s->current_picture_ptr->f.data;
++                    ref_picture = s->current_picture_ptr->f->data;
                  }
              }
 -            av_log(s->avctx, AV_LOG_DEBUG, "\n");
          }
 +        break;
 +    default:
 +        av_assert2(0);
      }
  }
  
@@@ -2810,10 -2129,10 +2829,10 @@@ void MPV_decode_mb_internal(MpegEncCont
          int dct_linesize, dct_offset;
          op_pixels_func (*op_pix)[4];
          qpel_mc_func (*op_qpix)[16];
-         const int linesize   = s->current_picture.f.linesize[0]; //not s->linesize as this would be wrong for field pics
-         const int uvlinesize = s->current_picture.f.linesize[1];
+         const int linesize   = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
+         const int uvlinesize = s->current_picture.f->linesize[1];
 -        const int readable= s->pict_type != AV_PICTURE_TYPE_B || s->encoding || s->avctx->draw_horiz_band;
 -        const int block_size = 8;
 +        const int readable= s->pict_type != AV_PICTURE_TYPE_B || s->encoding || s->avctx->draw_horiz_band || lowres_flag;
 +        const int block_size= lowres_flag ? 8>>s->avctx->lowres : 8;
  
          /* avoid copy if macroblock skipped in last frame too */
          /* skip only during decoding as we might trash the buffers during encoding a bit */
                      }
                  }
  
 -                op_qpix= s->me.qpel_put;
 -                if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
 -                    op_pix = s->hdsp.put_pixels_tab;
 +                if(lowres_flag){
 +                    h264_chroma_mc_func *op_pix = s->h264chroma.put_h264_chroma_pixels_tab;
 +
 +                    if (s->mv_dir & MV_DIR_FORWARD) {
-                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f.data, op_pix);
++                        MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix);
 +                        op_pix = s->h264chroma.avg_h264_chroma_pixels_tab;
 +                    }
 +                    if (s->mv_dir & MV_DIR_BACKWARD) {
-                         MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f.data, op_pix);
++                        MPV_motion_lowres(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix);
 +                    }
                  }else{
 -                    op_pix = s->hdsp.put_no_rnd_pixels_tab;
 -                }
 -                if (s->mv_dir & MV_DIR_FORWARD) {
 -                    ff_MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix, op_qpix);
 -                    op_pix = s->hdsp.avg_pixels_tab;
 -                    op_qpix= s->me.qpel_avg;
 -                }
 -                if (s->mv_dir & MV_DIR_BACKWARD) {
 -                    ff_MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix, op_qpix);
 +                    op_qpix = s->me.qpel_put;
 +                    if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
 +                        op_pix = s->hdsp.put_pixels_tab;
 +                    }else{
 +                        op_pix = s->hdsp.put_no_rnd_pixels_tab;
 +                    }
 +                    if (s->mv_dir & MV_DIR_FORWARD) {
-                         ff_MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f.data, op_pix, op_qpix);
++                        ff_MPV_motion(s, dest_y, dest_cb, dest_cr, 0, s->last_picture.f->data, op_pix, op_qpix);
 +                        op_pix = s->hdsp.avg_pixels_tab;
 +                        op_qpix= s->me.qpel_avg;
 +                    }
 +                    if (s->mv_dir & MV_DIR_BACKWARD) {
-                         ff_MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f.data, op_pix, op_qpix);
++                        ff_MPV_motion(s, dest_y, dest_cb, dest_cr, 1, s->next_picture.f->data, op_pix, op_qpix);
 +                    }
                  }
              }
  
@@@ -3022,15 -2327,15 +3041,15 @@@ void ff_MPV_decode_mb(MpegEncContext *s
  
  void ff_mpeg_draw_horiz_band(MpegEncContext *s, int y, int h)
  {
-     ff_draw_horiz_band(s->avctx, &s->current_picture_ptr->f,
-                        &s->last_picture_ptr->f, y, h, s->picture_structure,
 -    ff_draw_horiz_band(s->avctx, s->current_picture.f,
 -                       s->last_picture.f, y, h, s->picture_structure,
++    ff_draw_horiz_band(s->avctx, s->current_picture_ptr->f,
++                       s->last_picture_ptr ? s->last_picture_ptr->f : NULL, y, h, s->picture_structure,
                         s->first_field, s->low_delay);
  }
  
  void ff_init_block_index(MpegEncContext *s){ //FIXME maybe rename
-     const int linesize   = s->current_picture.f.linesize[0]; //not s->linesize as this would be wrong for field pics
-     const int uvlinesize = s->current_picture.f.linesize[1];
+     const int linesize   = s->current_picture.f->linesize[0]; //not s->linesize as this would be wrong for field pics
+     const int uvlinesize = s->current_picture.f->linesize[1];
 -    const int mb_size= 4;
 +    const int mb_size= 4 - s->avctx->lowres;
  
      s->block_index[0]= s->b8_stride*(s->mb_y*2    ) - 2 + s->mb_x*2;
      s->block_index[1]= s->b8_stride*(s->mb_y*2    ) - 1 + s->mb_x*2;
@@@ -92,8 -91,7 +92,8 @@@ struct MpegEncContext
   * Picture.
   */
  typedef struct Picture{
-     struct AVFrame f;
+     struct AVFrame *f;
 +    uint8_t avframe_padding[1024]; // hack to allow linking to a avutil with larger AVFrame
      ThreadFrame tf;
  
      AVBufferRef *qscale_table_buf;
@@@ -803,7 -684,10 +803,7 @@@ av_cold int ff_MPV_encode_init(AVCodecC
      if (ff_MPV_common_init(s) < 0)
          return -1;
  
-     s->avctx->coded_frame = &s->current_picture.f;
 -    if (ARCH_X86)
 -        ff_MPV_encode_init_x86(s);
 -
+     s->avctx->coded_frame = s->current_picture.f;
  
      if (s->msmpeg4_version) {
          FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_stats,
@@@ -1091,12 -972,8 +1091,12 @@@ static int load_input_picture(MpegEncCo
                      int w = s->width  >> h_shift;
                      int h = s->height >> v_shift;
                      uint8_t *src = pic_arg->data[i];
-                     uint8_t *dst = pic->f.data[i];
+                     uint8_t *dst = pic->f->data[i];
  
 +                    if (s->codec_id == AV_CODEC_ID_AMV && !(s->avctx->flags & CODEC_FLAG_EMU_EDGE)) {
 +                        h = ((s->height + 15)/16*16) >> v_shift;
 +                    }
 +
                      if (!s->avctx->rc_buffer_size)
                          dst += INPLACE_OFFSET;
  
@@@ -1150,16 -1018,16 +1150,16 @@@ static int skip_check(MpegEncContext *s
          for (y = 0; y < s->mb_height * bw; y++) {
              for (x = 0; x < s->mb_width * bw; x++) {
                  int off = p->shared ? 0 : 16;
-                 uint8_t *dptr = p->f.data[plane] + 8 * (x + y * stride) + off;
-                 uint8_t *rptr = ref->f.data[plane] + 8 * (x + y * stride);
+                 uint8_t *dptr = p->f->data[plane] + 8 * (x + y * stride) + off;
+                 uint8_t *rptr = ref->f->data[plane] + 8 * (x + y * stride);
                  int v   = s->dsp.frame_skip_cmp[1](s, dptr, rptr, stride, 8);
  
 -                switch (s->avctx->frame_skip_exp) {
 +                switch (FFABS(s->avctx->frame_skip_exp)) {
                  case 0: score    =  FFMAX(score, v);          break;
                  case 1: score   += FFABS(v);                  break;
 -                case 2: score   += v * v;                     break;
 -                case 3: score64 += FFABS(v * v * (int64_t)v); break;
 -                case 4: score64 += v * v * (int64_t)(v * v);  break;
 +                case 2: score64 += v * (int64_t)v;                       break;
 +                case 3: score64 += FFABS(v * (int64_t)v * v);            break;
 +                case 4: score64 += (v * (int64_t)v) * (v * (int64_t)v);  break;
                  }
              }
          }
@@@ -1313,19 -1177,6 +1313,19 @@@ static int select_input_picture(MpegEnc
  
      /* set next picture type & ordering */
      if (s->reordered_input_picture[0] == NULL && s->input_picture[0]) {
-                 av_frame_unref(&s->input_picture[0]->f);
 +        if (s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor) {
 +            if (s->picture_in_gop_number < s->gop_size &&
 +                s->next_picture_ptr &&
 +                skip_check(s, s->input_picture[0], s->next_picture_ptr)) {
 +                // FIXME check that te gop check above is +-1 correct
++                av_frame_unref(s->input_picture[0]->f);
 +
 +                ff_vbv_update(s, 0);
 +
 +                goto no_output_pic;
 +            }
 +        }
 +
          if (/*s->picture_in_gop_number >= s->gop_size ||*/
              s->next_picture_ptr == NULL || s->intra_only) {
              s->reordered_input_picture[0] = s->input_picture[0];
          } else {
              int b_frames;
  
 -            if (s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor) {
 -                if (s->picture_in_gop_number < s->gop_size &&
 -                    skip_check(s, s->input_picture[0], s->next_picture_ptr)) {
 -                    // FIXME check that te gop check above is +-1 correct
 -                    av_frame_unref(s->input_picture[0]->f);
 -
 -                    emms_c();
 -                    ff_vbv_update(s, 0);
 -
 -                    goto no_output_pic;
 -                }
 -            }
 -
              if (s->flags & CODEC_FLAG_PASS2) {
                  for (i = 0; i < s->max_b_frames + 1; i++) {
-                     int pict_num = s->input_picture[0]->f.display_picture_number + i;
+                     int pict_num = s->input_picture[0]->f->display_picture_number + i;
  
                      if (pict_num >= s->rc_context.num_entries)
                          break;
@@@ -1489,15 -1355,15 +1489,15 @@@ static void frame_end(MpegEncContext *s
          const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(s->avctx->pix_fmt);
          int hshift = desc->log2_chroma_w;
          int vshift = desc->log2_chroma_h;
-         s->dsp.draw_edges(s->current_picture.f.data[0], s->current_picture.f.linesize[0],
 -        s->dsp.draw_edges(s->current_picture.f->data[0], s->linesize,
++        s->dsp.draw_edges(s->current_picture.f->data[0], s->current_picture.f->linesize[0],
                            s->h_edge_pos, s->v_edge_pos,
                            EDGE_WIDTH, EDGE_WIDTH,
                            EDGE_TOP | EDGE_BOTTOM);
-         s->dsp.draw_edges(s->current_picture.f.data[1], s->current_picture.f.linesize[1],
 -        s->dsp.draw_edges(s->current_picture.f->data[1], s->uvlinesize,
++        s->dsp.draw_edges(s->current_picture.f->data[1], s->current_picture.f->linesize[1],
                            s->h_edge_pos >> hshift, s->v_edge_pos >> vshift,
                            EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
                            EDGE_TOP | EDGE_BOTTOM);
-         s->dsp.draw_edges(s->current_picture.f.data[2], s->current_picture.f.linesize[2],
 -        s->dsp.draw_edges(s->current_picture.f->data[2], s->uvlinesize,
++        s->dsp.draw_edges(s->current_picture.f->data[2], s->current_picture.f->linesize[2],
                            s->h_edge_pos >> hshift, s->v_edge_pos >> vshift,
                            EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
                            EDGE_TOP | EDGE_BOTTOM);
      if (s->pict_type!= AV_PICTURE_TYPE_B)
          s->last_non_b_pict_type = s->pict_type;
  
-     s->avctx->coded_frame = &s->current_picture_ptr->f;
 -    if (s->encoding) {
 -        /* release non-reference frames */
 -        for (i = 0; i < MAX_PICTURE_COUNT; i++) {
 -            if (!s->picture[i].reference)
 -                ff_mpeg_unref_picture(s, &s->picture[i]);
 -        }
 -    }
 -
+     s->avctx->coded_frame = s->current_picture_ptr->f;
  
  }
  
@@@ -1624,8 -1498,9 +1624,8 @@@ int ff_MPV_encode_picture(AVCodecContex
      }
  
      /* output? */
-     if (s->new_picture.f.data[0]) {
+     if (s->new_picture.f->data[0]) {
 -        if (!pkt->data &&
 -            (ret = ff_alloc_packet(pkt, s->mb_width*s->mb_height*MAX_MB_BYTES)) < 0)
 +        if ((ret = ff_alloc_packet2(avctx, pkt, s->mb_width*s->mb_height*(MAX_MB_BYTES+100)+10000)) < 0)
              return ret;
          if (s->mb_info) {
              s->mb_info_ptr = av_packet_new_side_data(pkt,
@@@ -1789,16 -1663,16 +1789,16 @@@ vbv_retry
          s->total_bits     += s->frame_bits;
          avctx->frame_bits  = s->frame_bits;
  
-         pkt->pts = s->current_picture.f.pts;
+         pkt->pts = s->current_picture.f->pts;
 -        if (!s->low_delay) {
 +        if (!s->low_delay && s->pict_type != AV_PICTURE_TYPE_B) {
-             if (!s->current_picture.f.coded_picture_number)
+             if (!s->current_picture.f->coded_picture_number)
                  pkt->dts = pkt->pts - s->dts_delta;
              else
                  pkt->dts = s->reordered_pts;
 -            s->reordered_pts = s->input_picture[0]->f->pts;
 +            s->reordered_pts = pkt->pts;
          } else
              pkt->dts = pkt->pts;
-         if (s->current_picture.f.key_frame)
+         if (s->current_picture.f->key_frame)
              pkt->flags |= AV_PKT_FLAG_KEY;
          if (s->mb_info)
              av_packet_shrink_side_data(pkt, AV_PKT_DATA_H263_MB_INFO, s->mb_info_size);
@@@ -1985,17 -1850,15 +1985,17 @@@ static av_always_inline void encode_mb_
  
      wrap_y = s->linesize;
      wrap_c = s->uvlinesize;
-     ptr_y  = s->new_picture.f.data[0] +
+     ptr_y  = s->new_picture.f->data[0] +
               (mb_y * 16 * wrap_y)              + mb_x * 16;
-     ptr_cb = s->new_picture.f.data[1] +
+     ptr_cb = s->new_picture.f->data[1] +
 -             (mb_y * mb_block_height * wrap_c) + mb_x * 8;
 +             (mb_y * mb_block_height * wrap_c) + mb_x * mb_block_width;
-     ptr_cr = s->new_picture.f.data[2] +
+     ptr_cr = s->new_picture.f->data[2] +
 -             (mb_y * mb_block_height * wrap_c) + mb_x * 8;
 +             (mb_y * mb_block_height * wrap_c) + mb_x * mb_block_width;
  
 -    if (mb_x * 16 + 16 > s->width || mb_y * 16 + 16 > s->height) {
 +    if((mb_x * 16 + 16 > s->width || mb_y * 16 + 16 > s->height) && s->codec_id != AV_CODEC_ID_AMV){
          uint8_t *ebuf = s->edge_emu_buffer + 32;
 +        int cw = (s->width  + s->chroma_x_shift) >> s->chroma_x_shift;
 +        int ch = (s->height + s->chroma_y_shift) >> s->chroma_y_shift;
          s->vdsp.emulated_edge_mc(ebuf, ptr_y,
                                   wrap_y, wrap_y,
                                   16, 16, mb_x * 16, mb_y * 16,
@@@ -2656,13 -2495,8 +2656,13 @@@ static int encode_thread(AVCodecContex
          /* note: quant matrix value (8) is implied here */
          s->last_dc[i] = 128 << s->intra_dc_precision;
  
-         s->current_picture.f.error[i] = 0;
+         s->current_picture.f->error[i] = 0;
      }
 +    if(s->codec_id==AV_CODEC_ID_AMV){
 +        s->last_dc[0] = 128*8/13;
 +        s->last_dc[1] = 128*8/14;
 +        s->last_dc[2] = 128*8/14;
 +    }
      s->mb_skip_run = 0;
      memset(s->last_mv, 0, sizeof(s->last_mv));
  
@@@ -3557,15 -3345,14 +3557,15 @@@ static int encode_picture(MpegEncContex
      }
  
      //FIXME var duplication
-     s->current_picture_ptr->f.key_frame =
-     s->current_picture.f.key_frame = s->pict_type == AV_PICTURE_TYPE_I; //FIXME pic_ptr
-     s->current_picture_ptr->f.pict_type =
-     s->current_picture.f.pict_type = s->pict_type;
+     s->current_picture_ptr->f->key_frame =
+     s->current_picture.f->key_frame = s->pict_type == AV_PICTURE_TYPE_I; //FIXME pic_ptr
+     s->current_picture_ptr->f->pict_type =
+     s->current_picture.f->pict_type = s->pict_type;
  
-     if (s->current_picture.f.key_frame)
+     if (s->current_picture.f->key_frame)
          s->picture_in_gop_number=0;
  
 +    s->mb_x = s->mb_y = 0;
      s->last_bits= put_bits_count(&s->pb);
      switch(s->out_format) {
      case FMT_MJPEG:
@@@ -897,9 -898,9 +897,9 @@@ static av_always_inline void MPV_motion
                                    s->mv[dir][1][0], s->mv[dir][1][1], 8, mb_y);
              }
          } else {
 -            if (s->picture_structure != s->field_select[dir][0] + 1 &&
 -                s->pict_type != AV_PICTURE_TYPE_B && !s->first_field) {
 +            if (   s->picture_structure != s->field_select[dir][0] + 1 && s->pict_type != AV_PICTURE_TYPE_B && !s->first_field
 +                || !ref_picture[0]) {
-                 ref_picture = s->current_picture_ptr->f.data;
+                 ref_picture = s->current_picture_ptr->f->data;
              }
  
              mpeg_motion(s, dest_y, dest_cb, dest_cr,
          for (i = 0; i < 2; i++) {
              uint8_t **ref2picture;
  
 -            if (s->picture_structure == s->field_select[dir][i] + 1
 -                || s->pict_type == AV_PICTURE_TYPE_B || s->first_field) {
 +            if ((s->picture_structure == s->field_select[dir][i] + 1
 +                || s->pict_type == AV_PICTURE_TYPE_B || s->first_field) && ref_picture[0]) {
                  ref2picture = ref_picture;
              } else {
-                 ref2picture = s->current_picture_ptr->f.data;
+                 ref2picture = s->current_picture_ptr->f->data;
              }
  
              mpeg_motion(s, dest_y, dest_cb, dest_cr,
                  pix_op = s->hdsp.avg_pixels_tab;
              }
          } else {
-                 ref_picture = s->current_picture_ptr->f.data;
 +            if (!ref_picture[0]) {
++                ref_picture = s->current_picture_ptr->f->data;
 +            }
              for (i = 0; i < 2; i++) {
                  mpeg_motion(s, dest_y, dest_cb, dest_cr,
                              s->picture_structure != i + 1,
@@@ -84,10 -74,9 +84,10 @@@ void ff_xvmc_pack_pblocks(MpegEncContex
   * This function should be called for every new field and/or frame.
   * It should be safe to call the function a few times for the same field.
   */
 -int ff_xvmc_field_start(MpegEncContext *s, AVCodecContext *avctx)
 +static int ff_xvmc_field_start(AVCodecContext *avctx, const uint8_t *buf, uint32_t buf_size)
  {
-     struct xvmc_pix_fmt *last, *next, *render = (struct xvmc_pix_fmt*)s->current_picture.f.data[2];
 +    struct MpegEncContext *s = avctx->priv_data;
+     struct xvmc_pix_fmt *last, *next, *render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2];
      const int mb_block_count = 4 + (1 << s->chroma_format);
  
      assert(avctx);
@@@ -153,10 -142,9 +153,10 @@@ return -1
   * some leftover blocks, for example from error_resilience(), may remain.
   * It should be safe to call the function a few times for the same field.
   */
 -void ff_xvmc_field_end(MpegEncContext *s)
 +static int ff_xvmc_field_end(AVCodecContext *avctx)
  {
-     struct xvmc_pix_fmt *render = (struct xvmc_pix_fmt*)s->current_picture.f.data[2];
 +    struct MpegEncContext *s = avctx->priv_data;
+     struct xvmc_pix_fmt *render = (struct xvmc_pix_fmt*)s->current_picture.f->data[2];
      assert(render);
  
      if (render->filled_mv_blocks_num > 0)
@@@ -273,18 -275,17 +273,18 @@@ int ff_msmpeg4_pred_dc(MpegEncContext *
                      *dir_ptr = 0;
                  }
              }else{
 +                int bs = 8 >> s->avctx->lowres;
                  if(n<4){
                      wrap= s->linesize;
-                     dest= s->current_picture.f.data[0] + (((n >> 1) + 2*s->mb_y) * bs*  wrap ) + ((n & 1) + 2*s->mb_x) * bs;
 -                    dest= s->current_picture.f->data[0] + (((n >> 1) + 2*s->mb_y) * 8*  wrap ) + ((n & 1) + 2*s->mb_x) * 8;
++                    dest= s->current_picture.f->data[0] + (((n >> 1) + 2*s->mb_y) * bs*  wrap ) + ((n & 1) + 2*s->mb_x) * bs;
                  }else{
                      wrap= s->uvlinesize;
-                     dest= s->current_picture.f.data[n - 3] + (s->mb_y * bs * wrap) + s->mb_x * bs;
 -                    dest= s->current_picture.f->data[n - 3] + (s->mb_y * 8 * wrap) + s->mb_x * 8;
++                    dest= s->current_picture.f->data[n - 3] + (s->mb_y * bs * wrap) + s->mb_x * bs;
                  }
                  if(s->mb_x==0) a= (1024 + (scale>>1))/scale;
 -                else           a= get_dc(dest-8, wrap, scale*8);
 +                else           a= get_dc(dest-bs, wrap, scale*8>>(2*s->avctx->lowres), bs);
                  if(s->mb_y==0) c= (1024 + (scale>>1))/scale;
 -                else           c= get_dc(dest-8*wrap, wrap, scale*8);
 +                else           c= get_dc(dest-bs*wrap, wrap, scale*8>>(2*s->avctx->lowres), bs);
  
                  if (s->h263_aic_dir==0) {
                      pred= a;
Simple merge
Simple merge
@@@ -47,11 -47,11 +47,11 @@@ void ff_write_pass1_stats(MpegEncContex
  {
      snprintf(s->avctx->stats_out, 256,
               "in:%d out:%d type:%d q:%d itex:%d ptex:%d mv:%d misc:%d "
 -             "fcode:%d bcode:%d mc-var:%d var:%d icount:%d skipcount:%d hbits:%d;\n",
 +             "fcode:%d bcode:%d mc-var:%"PRId64" var:%"PRId64" icount:%d skipcount:%d hbits:%d;\n",
-              s->current_picture_ptr->f.display_picture_number,
-              s->current_picture_ptr->f.coded_picture_number,
+              s->current_picture_ptr->f->display_picture_number,
+              s->current_picture_ptr->f->coded_picture_number,
               s->pict_type,
-              s->current_picture.f.quality,
+              s->current_picture.f->quality,
               s->i_tex_bits,
               s->p_tex_bits,
               s->mv_bits,
@@@ -762,15 -739,13 +762,15 @@@ static int rv10_decode_frame(AVCodecCon
          ff_MPV_frame_end(s);
  
          if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
-             if ((ret = av_frame_ref(pict, &s->current_picture_ptr->f)) < 0)
+             if ((ret = av_frame_ref(pict, s->current_picture_ptr->f)) < 0)
                  return ret;
 -            ff_print_debug_info(s, s->current_picture_ptr);
 +            ff_print_debug_info(s, s->current_picture_ptr, pict);
 +            ff_mpv_export_qp_table(s, pict, s->current_picture_ptr, FF_QSCALE_TYPE_MPEG1);
          } else if (s->last_picture_ptr != NULL) {
-             if ((ret = av_frame_ref(pict, &s->last_picture_ptr->f)) < 0)
+             if ((ret = av_frame_ref(pict, s->last_picture_ptr->f)) < 0)
                  return ret;
 -            ff_print_debug_info(s, s->last_picture_ptr);
 +            ff_print_debug_info(s, s->last_picture_ptr, pict);
 +            ff_mpv_export_qp_table(s, pict,s->last_picture_ptr, FF_QSCALE_TYPE_MPEG1);
          }
  
          if (s->last_picture_ptr || s->low_delay) {
Simple merge
@@@ -1591,16 -1583,14 +1591,16 @@@ static int finish_frame(AVCodecContext 
          ff_thread_report_progress(&s->current_picture_ptr->tf, INT_MAX, 0);
  
      if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
-         if ((ret = av_frame_ref(pict, &s->current_picture_ptr->f)) < 0)
+         if ((ret = av_frame_ref(pict, s->current_picture_ptr->f)) < 0)
              return ret;
 -        ff_print_debug_info(s, s->current_picture_ptr);
 +        ff_print_debug_info(s, s->current_picture_ptr, pict);
 +        ff_mpv_export_qp_table(s, pict, s->current_picture_ptr, FF_QSCALE_TYPE_MPEG1);
          got_picture = 1;
      } else if (s->last_picture_ptr != NULL) {
-         if ((ret = av_frame_ref(pict, &s->last_picture_ptr->f)) < 0)
+         if ((ret = av_frame_ref(pict, s->last_picture_ptr->f)) < 0)
              return ret;
 -        ff_print_debug_info(s, s->last_picture_ptr);
 +        ff_print_debug_info(s, s->last_picture_ptr, pict);
 +        ff_mpv_export_qp_table(s, pict, s->last_picture_ptr, FF_QSCALE_TYPE_MPEG1);
          got_picture = 1;
      }
  
Simple merge
index ace2648,0000000..fd79e69
mode 100644,000000..100644
--- /dev/null
@@@ -1,2009 -1,0 +1,2011 @@@
-     s->m.last_picture.f.pts = s->m.current_picture.f.pts;
-     s->m.current_picture.f.pts = pict->pts;
 +/*
 + * Copyright (C) 2004 Michael Niedermayer <michaelni@gmx.at>
 + *
 + * This file is part of FFmpeg.
 + *
 + * FFmpeg is free software; you can redistribute it and/or
 + * modify it under the terms of the GNU Lesser General Public
 + * License as published by the Free Software Foundation; either
 + * version 2.1 of the License, or (at your option) any later version.
 + *
 + * FFmpeg is distributed in the hope that it will be useful,
 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 + * Lesser General Public License for more details.
 + *
 + * You should have received a copy of the GNU Lesser General Public
 + * License along with FFmpeg; if not, write to the Free Software
 + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 + */
 +
 +#include "libavutil/intmath.h"
 +#include "libavutil/log.h"
 +#include "libavutil/opt.h"
 +#include "avcodec.h"
 +#include "dsputil.h"
 +#include "internal.h"
 +#include "snow_dwt.h"
 +#include "snow.h"
 +
 +#include "rangecoder.h"
 +#include "mathops.h"
 +
 +#include "mpegvideo.h"
 +#include "h263.h"
 +
 +static av_cold int encode_init(AVCodecContext *avctx)
 +{
 +    SnowContext *s = avctx->priv_data;
 +    int plane_index, ret;
 +
 +    if(avctx->prediction_method == DWT_97
 +       && (avctx->flags & CODEC_FLAG_QSCALE)
 +       && avctx->global_quality == 0){
 +        av_log(avctx, AV_LOG_ERROR, "The 9/7 wavelet is incompatible with lossless mode.\n");
 +        return -1;
 +    }
 +
 +    s->spatial_decomposition_type= avctx->prediction_method; //FIXME add decorrelator type r transform_type
 +
 +    s->mv_scale       = (avctx->flags & CODEC_FLAG_QPEL) ? 2 : 4;
 +    s->block_max_depth= (avctx->flags & CODEC_FLAG_4MV ) ? 1 : 0;
 +
 +    for(plane_index=0; plane_index<3; plane_index++){
 +        s->plane[plane_index].diag_mc= 1;
 +        s->plane[plane_index].htaps= 6;
 +        s->plane[plane_index].hcoeff[0]=  40;
 +        s->plane[plane_index].hcoeff[1]= -10;
 +        s->plane[plane_index].hcoeff[2]=   2;
 +        s->plane[plane_index].fast_mc= 1;
 +    }
 +
 +    if ((ret = ff_snow_common_init(avctx)) < 0) {
 +        ff_snow_common_end(avctx->priv_data);
 +        return ret;
 +    }
 +    ff_snow_alloc_blocks(s);
 +
 +    s->version=0;
 +
 +    s->m.avctx   = avctx;
 +    s->m.flags   = avctx->flags;
 +    s->m.bit_rate= avctx->bit_rate;
 +
 +    s->m.me.temp      =
 +    s->m.me.scratchpad= av_mallocz((avctx->width+64)*2*16*2*sizeof(uint8_t));
 +    s->m.me.map       = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
 +    s->m.me.score_map = av_mallocz(ME_MAP_SIZE*sizeof(uint32_t));
 +    s->m.obmc_scratchpad= av_mallocz(MB_SIZE*MB_SIZE*12*sizeof(uint32_t));
 +    if (!s->m.me.scratchpad || !s->m.me.map || !s->m.me.score_map || !s->m.obmc_scratchpad)
 +        return AVERROR(ENOMEM);
 +
 +    ff_h263_encode_init(&s->m); //mv_penalty
 +
 +    s->max_ref_frames = FFMAX(FFMIN(avctx->refs, MAX_REF_FRAMES), 1);
 +
 +    if(avctx->flags&CODEC_FLAG_PASS1){
 +        if(!avctx->stats_out)
 +            avctx->stats_out = av_mallocz(256);
 +
 +        if (!avctx->stats_out)
 +            return AVERROR(ENOMEM);
 +    }
 +    if((avctx->flags&CODEC_FLAG_PASS2) || !(avctx->flags&CODEC_FLAG_QSCALE)){
 +        if(ff_rate_control_init(&s->m) < 0)
 +            return -1;
 +    }
 +    s->pass1_rc= !(avctx->flags & (CODEC_FLAG_QSCALE|CODEC_FLAG_PASS2));
 +
 +    switch(avctx->pix_fmt){
 +    case AV_PIX_FMT_YUV444P:
 +//    case AV_PIX_FMT_YUV422P:
 +    case AV_PIX_FMT_YUV420P:
 +//    case AV_PIX_FMT_YUV411P:
 +    case AV_PIX_FMT_YUV410P:
 +        s->nb_planes = 3;
 +        s->colorspace_type= 0;
 +        break;
 +    case AV_PIX_FMT_GRAY8:
 +        s->nb_planes = 1;
 +        s->colorspace_type = 1;
 +        break;
 +/*    case AV_PIX_FMT_RGB32:
 +        s->colorspace= 1;
 +        break;*/
 +    default:
 +        av_log(avctx, AV_LOG_ERROR, "pixel format not supported\n");
 +        return -1;
 +    }
 +    avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_h_shift, &s->chroma_v_shift);
 +
 +    ff_set_cmp(&s->dsp, s->dsp.me_cmp, s->avctx->me_cmp);
 +    ff_set_cmp(&s->dsp, s->dsp.me_sub_cmp, s->avctx->me_sub_cmp);
 +
 +    s->input_picture = av_frame_alloc();
 +    if (!s->input_picture)
 +        return AVERROR(ENOMEM);
 +    if ((ret = ff_get_buffer(s->avctx, s->input_picture, AV_GET_BUFFER_FLAG_REF)) < 0)
 +        return ret;
 +
 +    if(s->avctx->me_method == ME_ITER){
 +        int i;
 +        int size= s->b_width * s->b_height << 2*s->block_max_depth;
 +        for(i=0; i<s->max_ref_frames; i++){
 +            s->ref_mvs[i]= av_mallocz(size*sizeof(int16_t[2]));
 +            s->ref_scores[i]= av_mallocz(size*sizeof(uint32_t));
 +            if (!s->ref_mvs[i] || !s->ref_scores[i])
 +                return AVERROR(ENOMEM);
 +        }
 +    }
 +
 +    return 0;
 +}
 +
 +//near copy & paste from dsputil, FIXME
 +static int pix_sum(uint8_t * pix, int line_size, int w, int h)
 +{
 +    int s, i, j;
 +
 +    s = 0;
 +    for (i = 0; i < h; i++) {
 +        for (j = 0; j < w; j++) {
 +            s += pix[0];
 +            pix ++;
 +        }
 +        pix += line_size - w;
 +    }
 +    return s;
 +}
 +
 +//near copy & paste from dsputil, FIXME
 +static int pix_norm1(uint8_t * pix, int line_size, int w)
 +{
 +    int s, i, j;
 +    uint32_t *sq = ff_square_tab + 256;
 +
 +    s = 0;
 +    for (i = 0; i < w; i++) {
 +        for (j = 0; j < w; j ++) {
 +            s += sq[pix[0]];
 +            pix ++;
 +        }
 +        pix += line_size - w;
 +    }
 +    return s;
 +}
 +
 +static inline int get_penalty_factor(int lambda, int lambda2, int type){
 +    switch(type&0xFF){
 +    default:
 +    case FF_CMP_SAD:
 +        return lambda>>FF_LAMBDA_SHIFT;
 +    case FF_CMP_DCT:
 +        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
 +    case FF_CMP_W53:
 +        return (4*lambda)>>(FF_LAMBDA_SHIFT);
 +    case FF_CMP_W97:
 +        return (2*lambda)>>(FF_LAMBDA_SHIFT);
 +    case FF_CMP_SATD:
 +    case FF_CMP_DCT264:
 +        return (2*lambda)>>FF_LAMBDA_SHIFT;
 +    case FF_CMP_RD:
 +    case FF_CMP_PSNR:
 +    case FF_CMP_SSE:
 +    case FF_CMP_NSSE:
 +        return lambda2>>FF_LAMBDA_SHIFT;
 +    case FF_CMP_BIT:
 +        return 1;
 +    }
 +}
 +
 +//FIXME copy&paste
 +#define P_LEFT P[1]
 +#define P_TOP P[2]
 +#define P_TOPRIGHT P[3]
 +#define P_MEDIAN P[4]
 +#define P_MV1 P[9]
 +#define FLAG_QPEL   1 //must be 1
 +
 +static int encode_q_branch(SnowContext *s, int level, int x, int y){
 +    uint8_t p_buffer[1024];
 +    uint8_t i_buffer[1024];
 +    uint8_t p_state[sizeof(s->block_state)];
 +    uint8_t i_state[sizeof(s->block_state)];
 +    RangeCoder pc, ic;
 +    uint8_t *pbbak= s->c.bytestream;
 +    uint8_t *pbbak_start= s->c.bytestream_start;
 +    int score, score2, iscore, i_len, p_len, block_s, sum, base_bits;
 +    const int w= s->b_width  << s->block_max_depth;
 +    const int h= s->b_height << s->block_max_depth;
 +    const int rem_depth= s->block_max_depth - level;
 +    const int index= (x + y*w) << rem_depth;
 +    const int block_w= 1<<(LOG2_MB_SIZE - level);
 +    int trx= (x+1)<<rem_depth;
 +    int try= (y+1)<<rem_depth;
 +    const BlockNode *left  = x ? &s->block[index-1] : &null_block;
 +    const BlockNode *top   = y ? &s->block[index-w] : &null_block;
 +    const BlockNode *right = trx<w ? &s->block[index+1] : &null_block;
 +    const BlockNode *bottom= try<h ? &s->block[index+w] : &null_block;
 +    const BlockNode *tl    = y && x ? &s->block[index-w-1] : left;
 +    const BlockNode *tr    = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
 +    int pl = left->color[0];
 +    int pcb= left->color[1];
 +    int pcr= left->color[2];
 +    int pmx, pmy;
 +    int mx=0, my=0;
 +    int l,cr,cb;
 +    const int stride= s->current_picture->linesize[0];
 +    const int uvstride= s->current_picture->linesize[1];
 +    uint8_t *current_data[3]= { s->input_picture->data[0] + (x + y*  stride)*block_w,
 +                                s->input_picture->data[1] + ((x*block_w)>>s->chroma_h_shift) + ((y*uvstride*block_w)>>s->chroma_v_shift),
 +                                s->input_picture->data[2] + ((x*block_w)>>s->chroma_h_shift) + ((y*uvstride*block_w)>>s->chroma_v_shift)};
 +    int P[10][2];
 +    int16_t last_mv[3][2];
 +    int qpel= !!(s->avctx->flags & CODEC_FLAG_QPEL); //unused
 +    const int shift= 1+qpel;
 +    MotionEstContext *c= &s->m.me;
 +    int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
 +    int mx_context= av_log2(2*FFABS(left->mx - top->mx));
 +    int my_context= av_log2(2*FFABS(left->my - top->my));
 +    int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
 +    int ref, best_ref, ref_score, ref_mx, ref_my;
 +
 +    av_assert0(sizeof(s->block_state) >= 256);
 +    if(s->keyframe){
 +        set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA);
 +        return 0;
 +    }
 +
 +//    clip predictors / edge ?
 +
 +    P_LEFT[0]= left->mx;
 +    P_LEFT[1]= left->my;
 +    P_TOP [0]= top->mx;
 +    P_TOP [1]= top->my;
 +    P_TOPRIGHT[0]= tr->mx;
 +    P_TOPRIGHT[1]= tr->my;
 +
 +    last_mv[0][0]= s->block[index].mx;
 +    last_mv[0][1]= s->block[index].my;
 +    last_mv[1][0]= right->mx;
 +    last_mv[1][1]= right->my;
 +    last_mv[2][0]= bottom->mx;
 +    last_mv[2][1]= bottom->my;
 +
 +    s->m.mb_stride=2;
 +    s->m.mb_x=
 +    s->m.mb_y= 0;
 +    c->skip= 0;
 +
 +    av_assert1(c->  stride ==   stride);
 +    av_assert1(c->uvstride == uvstride);
 +
 +    c->penalty_factor    = get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_cmp);
 +    c->sub_penalty_factor= get_penalty_factor(s->lambda, s->lambda2, c->avctx->me_sub_cmp);
 +    c->mb_penalty_factor = get_penalty_factor(s->lambda, s->lambda2, c->avctx->mb_cmp);
 +    c->current_mv_penalty= c->mv_penalty[s->m.f_code=1] + MAX_MV;
 +
 +    c->xmin = - x*block_w - 16+3;
 +    c->ymin = - y*block_w - 16+3;
 +    c->xmax = - (x+1)*block_w + (w<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-3;
 +    c->ymax = - (y+1)*block_w + (h<<(LOG2_MB_SIZE - s->block_max_depth)) + 16-3;
 +
 +    if(P_LEFT[0]     > (c->xmax<<shift)) P_LEFT[0]    = (c->xmax<<shift);
 +    if(P_LEFT[1]     > (c->ymax<<shift)) P_LEFT[1]    = (c->ymax<<shift);
 +    if(P_TOP[0]      > (c->xmax<<shift)) P_TOP[0]     = (c->xmax<<shift);
 +    if(P_TOP[1]      > (c->ymax<<shift)) P_TOP[1]     = (c->ymax<<shift);
 +    if(P_TOPRIGHT[0] < (c->xmin<<shift)) P_TOPRIGHT[0]= (c->xmin<<shift);
 +    if(P_TOPRIGHT[0] > (c->xmax<<shift)) P_TOPRIGHT[0]= (c->xmax<<shift); //due to pmx no clip
 +    if(P_TOPRIGHT[1] > (c->ymax<<shift)) P_TOPRIGHT[1]= (c->ymax<<shift);
 +
 +    P_MEDIAN[0]= mid_pred(P_LEFT[0], P_TOP[0], P_TOPRIGHT[0]);
 +    P_MEDIAN[1]= mid_pred(P_LEFT[1], P_TOP[1], P_TOPRIGHT[1]);
 +
 +    if (!y) {
 +        c->pred_x= P_LEFT[0];
 +        c->pred_y= P_LEFT[1];
 +    } else {
 +        c->pred_x = P_MEDIAN[0];
 +        c->pred_y = P_MEDIAN[1];
 +    }
 +
 +    score= INT_MAX;
 +    best_ref= 0;
 +    for(ref=0; ref<s->ref_frames; ref++){
 +        init_ref(c, current_data, s->last_picture[ref]->data, NULL, block_w*x, block_w*y, 0);
 +
 +        ref_score= ff_epzs_motion_search(&s->m, &ref_mx, &ref_my, P, 0, /*ref_index*/ 0, last_mv,
 +                                         (1<<16)>>shift, level-LOG2_MB_SIZE+4, block_w);
 +
 +        av_assert2(ref_mx >= c->xmin);
 +        av_assert2(ref_mx <= c->xmax);
 +        av_assert2(ref_my >= c->ymin);
 +        av_assert2(ref_my <= c->ymax);
 +
 +        ref_score= c->sub_motion_search(&s->m, &ref_mx, &ref_my, ref_score, 0, 0, level-LOG2_MB_SIZE+4, block_w);
 +        ref_score= ff_get_mb_score(&s->m, ref_mx, ref_my, 0, 0, level-LOG2_MB_SIZE+4, block_w, 0);
 +        ref_score+= 2*av_log2(2*ref)*c->penalty_factor;
 +        if(s->ref_mvs[ref]){
 +            s->ref_mvs[ref][index][0]= ref_mx;
 +            s->ref_mvs[ref][index][1]= ref_my;
 +            s->ref_scores[ref][index]= ref_score;
 +        }
 +        if(score > ref_score){
 +            score= ref_score;
 +            best_ref= ref;
 +            mx= ref_mx;
 +            my= ref_my;
 +        }
 +    }
 +    //FIXME if mb_cmp != SSE then intra cannot be compared currently and mb_penalty vs. lambda2
 +
 +  //  subpel search
 +    base_bits= get_rac_count(&s->c) - 8*(s->c.bytestream - s->c.bytestream_start);
 +    pc= s->c;
 +    pc.bytestream_start=
 +    pc.bytestream= p_buffer; //FIXME end/start? and at the other stoo
 +    memcpy(p_state, s->block_state, sizeof(s->block_state));
 +
 +    if(level!=s->block_max_depth)
 +        put_rac(&pc, &p_state[4 + s_context], 1);
 +    put_rac(&pc, &p_state[1 + left->type + top->type], 0);
 +    if(s->ref_frames > 1)
 +        put_symbol(&pc, &p_state[128 + 1024 + 32*ref_context], best_ref, 0);
 +    pred_mv(s, &pmx, &pmy, best_ref, left, top, tr);
 +    put_symbol(&pc, &p_state[128 + 32*(mx_context + 16*!!best_ref)], mx - pmx, 1);
 +    put_symbol(&pc, &p_state[128 + 32*(my_context + 16*!!best_ref)], my - pmy, 1);
 +    p_len= pc.bytestream - pc.bytestream_start;
 +    score += (s->lambda2*(get_rac_count(&pc)-base_bits))>>FF_LAMBDA_SHIFT;
 +
 +    block_s= block_w*block_w;
 +    sum = pix_sum(current_data[0], stride, block_w, block_w);
 +    l= (sum + block_s/2)/block_s;
 +    iscore = pix_norm1(current_data[0], stride, block_w) - 2*l*sum + l*l*block_s;
 +
 +    if (s->nb_planes > 2) {
 +        block_s= block_w*block_w>>(s->chroma_h_shift + s->chroma_v_shift);
 +        sum = pix_sum(current_data[1], uvstride, block_w>>s->chroma_h_shift, block_w>>s->chroma_v_shift);
 +        cb= (sum + block_s/2)/block_s;
 +    //    iscore += pix_norm1(&current_mb[1][0], uvstride, block_w>>1) - 2*cb*sum + cb*cb*block_s;
 +        sum = pix_sum(current_data[2], uvstride, block_w>>s->chroma_h_shift, block_w>>s->chroma_v_shift);
 +        cr= (sum + block_s/2)/block_s;
 +    //    iscore += pix_norm1(&current_mb[2][0], uvstride, block_w>>1) - 2*cr*sum + cr*cr*block_s;
 +    }else
 +        cb = cr = 0;
 +
 +    ic= s->c;
 +    ic.bytestream_start=
 +    ic.bytestream= i_buffer; //FIXME end/start? and at the other stoo
 +    memcpy(i_state, s->block_state, sizeof(s->block_state));
 +    if(level!=s->block_max_depth)
 +        put_rac(&ic, &i_state[4 + s_context], 1);
 +    put_rac(&ic, &i_state[1 + left->type + top->type], 1);
 +    put_symbol(&ic, &i_state[32],  l-pl , 1);
 +    if (s->nb_planes > 2) {
 +        put_symbol(&ic, &i_state[64], cb-pcb, 1);
 +        put_symbol(&ic, &i_state[96], cr-pcr, 1);
 +    }
 +    i_len= ic.bytestream - ic.bytestream_start;
 +    iscore += (s->lambda2*(get_rac_count(&ic)-base_bits))>>FF_LAMBDA_SHIFT;
 +
 +//    assert(score==256*256*256*64-1);
 +    av_assert1(iscore < 255*255*256 + s->lambda2*10);
 +    av_assert1(iscore >= 0);
 +    av_assert1(l>=0 && l<=255);
 +    av_assert1(pl>=0 && pl<=255);
 +
 +    if(level==0){
 +        int varc= iscore >> 8;
 +        int vard= score >> 8;
 +        if (vard <= 64 || vard < varc)
 +            c->scene_change_score+= ff_sqrt(vard) - ff_sqrt(varc);
 +        else
 +            c->scene_change_score+= s->m.qscale;
 +    }
 +
 +    if(level!=s->block_max_depth){
 +        put_rac(&s->c, &s->block_state[4 + s_context], 0);
 +        score2 = encode_q_branch(s, level+1, 2*x+0, 2*y+0);
 +        score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+0);
 +        score2+= encode_q_branch(s, level+1, 2*x+0, 2*y+1);
 +        score2+= encode_q_branch(s, level+1, 2*x+1, 2*y+1);
 +        score2+= s->lambda2>>FF_LAMBDA_SHIFT; //FIXME exact split overhead
 +
 +        if(score2 < score && score2 < iscore)
 +            return score2;
 +    }
 +
 +    if(iscore < score){
 +        pred_mv(s, &pmx, &pmy, 0, left, top, tr);
 +        memcpy(pbbak, i_buffer, i_len);
 +        s->c= ic;
 +        s->c.bytestream_start= pbbak_start;
 +        s->c.bytestream= pbbak + i_len;
 +        set_blocks(s, level, x, y, l, cb, cr, pmx, pmy, 0, BLOCK_INTRA);
 +        memcpy(s->block_state, i_state, sizeof(s->block_state));
 +        return iscore;
 +    }else{
 +        memcpy(pbbak, p_buffer, p_len);
 +        s->c= pc;
 +        s->c.bytestream_start= pbbak_start;
 +        s->c.bytestream= pbbak + p_len;
 +        set_blocks(s, level, x, y, pl, pcb, pcr, mx, my, best_ref, 0);
 +        memcpy(s->block_state, p_state, sizeof(s->block_state));
 +        return score;
 +    }
 +}
 +
 +static void encode_q_branch2(SnowContext *s, int level, int x, int y){
 +    const int w= s->b_width  << s->block_max_depth;
 +    const int rem_depth= s->block_max_depth - level;
 +    const int index= (x + y*w) << rem_depth;
 +    int trx= (x+1)<<rem_depth;
 +    BlockNode *b= &s->block[index];
 +    const BlockNode *left  = x ? &s->block[index-1] : &null_block;
 +    const BlockNode *top   = y ? &s->block[index-w] : &null_block;
 +    const BlockNode *tl    = y && x ? &s->block[index-w-1] : left;
 +    const BlockNode *tr    = y && trx<w && ((x&1)==0 || level==0) ? &s->block[index-w+(1<<rem_depth)] : tl; //FIXME use lt
 +    int pl = left->color[0];
 +    int pcb= left->color[1];
 +    int pcr= left->color[2];
 +    int pmx, pmy;
 +    int ref_context= av_log2(2*left->ref) + av_log2(2*top->ref);
 +    int mx_context= av_log2(2*FFABS(left->mx - top->mx)) + 16*!!b->ref;
 +    int my_context= av_log2(2*FFABS(left->my - top->my)) + 16*!!b->ref;
 +    int s_context= 2*left->level + 2*top->level + tl->level + tr->level;
 +
 +    if(s->keyframe){
 +        set_blocks(s, level, x, y, pl, pcb, pcr, 0, 0, 0, BLOCK_INTRA);
 +        return;
 +    }
 +
 +    if(level!=s->block_max_depth){
 +        if(same_block(b,b+1) && same_block(b,b+w) && same_block(b,b+w+1)){
 +            put_rac(&s->c, &s->block_state[4 + s_context], 1);
 +        }else{
 +            put_rac(&s->c, &s->block_state[4 + s_context], 0);
 +            encode_q_branch2(s, level+1, 2*x+0, 2*y+0);
 +            encode_q_branch2(s, level+1, 2*x+1, 2*y+0);
 +            encode_q_branch2(s, level+1, 2*x+0, 2*y+1);
 +            encode_q_branch2(s, level+1, 2*x+1, 2*y+1);
 +            return;
 +        }
 +    }
 +    if(b->type & BLOCK_INTRA){
 +        pred_mv(s, &pmx, &pmy, 0, left, top, tr);
 +        put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 1);
 +        put_symbol(&s->c, &s->block_state[32], b->color[0]-pl , 1);
 +        if (s->nb_planes > 2) {
 +            put_symbol(&s->c, &s->block_state[64], b->color[1]-pcb, 1);
 +            put_symbol(&s->c, &s->block_state[96], b->color[2]-pcr, 1);
 +        }
 +        set_blocks(s, level, x, y, b->color[0], b->color[1], b->color[2], pmx, pmy, 0, BLOCK_INTRA);
 +    }else{
 +        pred_mv(s, &pmx, &pmy, b->ref, left, top, tr);
 +        put_rac(&s->c, &s->block_state[1 + (left->type&1) + (top->type&1)], 0);
 +        if(s->ref_frames > 1)
 +            put_symbol(&s->c, &s->block_state[128 + 1024 + 32*ref_context], b->ref, 0);
 +        put_symbol(&s->c, &s->block_state[128 + 32*mx_context], b->mx - pmx, 1);
 +        put_symbol(&s->c, &s->block_state[128 + 32*my_context], b->my - pmy, 1);
 +        set_blocks(s, level, x, y, pl, pcb, pcr, b->mx, b->my, b->ref, 0);
 +    }
 +}
 +
 +static int get_dc(SnowContext *s, int mb_x, int mb_y, int plane_index){
 +    int i, x2, y2;
 +    Plane *p= &s->plane[plane_index];
 +    const int block_size = MB_SIZE >> s->block_max_depth;
 +    const int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
 +    const int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
 +    const uint8_t *obmc  = plane_index ? ff_obmc_tab[s->block_max_depth+s->chroma_h_shift] : ff_obmc_tab[s->block_max_depth];
 +    const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
 +    const int ref_stride= s->current_picture->linesize[plane_index];
 +    uint8_t *src= s-> input_picture->data[plane_index];
 +    IDWTELEM *dst= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4; //FIXME change to unsigned
 +    const int b_stride = s->b_width << s->block_max_depth;
 +    const int w= p->width;
 +    const int h= p->height;
 +    int index= mb_x + mb_y*b_stride;
 +    BlockNode *b= &s->block[index];
 +    BlockNode backup= *b;
 +    int ab=0;
 +    int aa=0;
 +
 +    av_assert2(s->chroma_h_shift == s->chroma_v_shift); //obmc stuff above
 +
 +    b->type|= BLOCK_INTRA;
 +    b->color[plane_index]= 0;
 +    memset(dst, 0, obmc_stride*obmc_stride*sizeof(IDWTELEM));
 +
 +    for(i=0; i<4; i++){
 +        int mb_x2= mb_x + (i &1) - 1;
 +        int mb_y2= mb_y + (i>>1) - 1;
 +        int x= block_w*mb_x2 + block_w/2;
 +        int y= block_h*mb_y2 + block_h/2;
 +
 +        add_yblock(s, 0, NULL, dst + (i&1)*block_w + (i>>1)*obmc_stride*block_h, NULL, obmc,
 +                    x, y, block_w, block_h, w, h, obmc_stride, ref_stride, obmc_stride, mb_x2, mb_y2, 0, 0, plane_index);
 +
 +        for(y2= FFMAX(y, 0); y2<FFMIN(h, y+block_h); y2++){
 +            for(x2= FFMAX(x, 0); x2<FFMIN(w, x+block_w); x2++){
 +                int index= x2-(block_w*mb_x - block_w/2) + (y2-(block_h*mb_y - block_h/2))*obmc_stride;
 +                int obmc_v= obmc[index];
 +                int d;
 +                if(y<0) obmc_v += obmc[index + block_h*obmc_stride];
 +                if(x<0) obmc_v += obmc[index + block_w];
 +                if(y+block_h>h) obmc_v += obmc[index - block_h*obmc_stride];
 +                if(x+block_w>w) obmc_v += obmc[index - block_w];
 +                //FIXME precalculate this or simplify it somehow else
 +
 +                d = -dst[index] + (1<<(FRAC_BITS-1));
 +                dst[index] = d;
 +                ab += (src[x2 + y2*ref_stride] - (d>>FRAC_BITS)) * obmc_v;
 +                aa += obmc_v * obmc_v; //FIXME precalculate this
 +            }
 +        }
 +    }
 +    *b= backup;
 +
 +    return av_clip( ROUNDED_DIV(ab<<LOG2_OBMC_MAX, aa), 0, 255); //FIXME we should not need clipping
 +}
 +
 +static inline int get_block_bits(SnowContext *s, int x, int y, int w){
 +    const int b_stride = s->b_width << s->block_max_depth;
 +    const int b_height = s->b_height<< s->block_max_depth;
 +    int index= x + y*b_stride;
 +    const BlockNode *b     = &s->block[index];
 +    const BlockNode *left  = x ? &s->block[index-1] : &null_block;
 +    const BlockNode *top   = y ? &s->block[index-b_stride] : &null_block;
 +    const BlockNode *tl    = y && x ? &s->block[index-b_stride-1] : left;
 +    const BlockNode *tr    = y && x+w<b_stride ? &s->block[index-b_stride+w] : tl;
 +    int dmx, dmy;
 +//  int mx_context= av_log2(2*FFABS(left->mx - top->mx));
 +//  int my_context= av_log2(2*FFABS(left->my - top->my));
 +
 +    if(x<0 || x>=b_stride || y>=b_height)
 +        return 0;
 +/*
 +1            0      0
 +01X          1-2    1
 +001XX        3-6    2-3
 +0001XXX      7-14   4-7
 +00001XXXX   15-30   8-15
 +*/
 +//FIXME try accurate rate
 +//FIXME intra and inter predictors if surrounding blocks are not the same type
 +    if(b->type & BLOCK_INTRA){
 +        return 3+2*( av_log2(2*FFABS(left->color[0] - b->color[0]))
 +                   + av_log2(2*FFABS(left->color[1] - b->color[1]))
 +                   + av_log2(2*FFABS(left->color[2] - b->color[2])));
 +    }else{
 +        pred_mv(s, &dmx, &dmy, b->ref, left, top, tr);
 +        dmx-= b->mx;
 +        dmy-= b->my;
 +        return 2*(1 + av_log2(2*FFABS(dmx)) //FIXME kill the 2* can be merged in lambda
 +                    + av_log2(2*FFABS(dmy))
 +                    + av_log2(2*b->ref));
 +    }
 +}
 +
 +static int get_block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index, uint8_t (*obmc_edged)[MB_SIZE * 2]){
 +    Plane *p= &s->plane[plane_index];
 +    const int block_size = MB_SIZE >> s->block_max_depth;
 +    const int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
 +    const int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
 +    const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
 +    const int ref_stride= s->current_picture->linesize[plane_index];
 +    uint8_t *dst= s->current_picture->data[plane_index];
 +    uint8_t *src= s->  input_picture->data[plane_index];
 +    IDWTELEM *pred= (IDWTELEM*)s->m.obmc_scratchpad + plane_index*block_size*block_size*4;
 +    uint8_t *cur = s->scratchbuf;
 +    uint8_t *tmp = s->emu_edge_buffer;
 +    const int b_stride = s->b_width << s->block_max_depth;
 +    const int b_height = s->b_height<< s->block_max_depth;
 +    const int w= p->width;
 +    const int h= p->height;
 +    int distortion;
 +    int rate= 0;
 +    const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp);
 +    int sx= block_w*mb_x - block_w/2;
 +    int sy= block_h*mb_y - block_h/2;
 +    int x0= FFMAX(0,-sx);
 +    int y0= FFMAX(0,-sy);
 +    int x1= FFMIN(block_w*2, w-sx);
 +    int y1= FFMIN(block_h*2, h-sy);
 +    int i,x,y;
 +
 +    av_assert2(s->chroma_h_shift == s->chroma_v_shift); //obmc and square assumtions below chckinhg only block_w
 +
 +    ff_snow_pred_block(s, cur, tmp, ref_stride, sx, sy, block_w*2, block_h*2, &s->block[mb_x + mb_y*b_stride], plane_index, w, h);
 +
 +    for(y=y0; y<y1; y++){
 +        const uint8_t *obmc1= obmc_edged[y];
 +        const IDWTELEM *pred1 = pred + y*obmc_stride;
 +        uint8_t *cur1 = cur + y*ref_stride;
 +        uint8_t *dst1 = dst + sx + (sy+y)*ref_stride;
 +        for(x=x0; x<x1; x++){
 +#if FRAC_BITS >= LOG2_OBMC_MAX
 +            int v = (cur1[x] * obmc1[x]) << (FRAC_BITS - LOG2_OBMC_MAX);
 +#else
 +            int v = (cur1[x] * obmc1[x] + (1<<(LOG2_OBMC_MAX - FRAC_BITS-1))) >> (LOG2_OBMC_MAX - FRAC_BITS);
 +#endif
 +            v = (v + pred1[x]) >> FRAC_BITS;
 +            if(v&(~255)) v= ~(v>>31);
 +            dst1[x] = v;
 +        }
 +    }
 +
 +    /* copy the regions where obmc[] = (uint8_t)256 */
 +    if(LOG2_OBMC_MAX == 8
 +        && (mb_x == 0 || mb_x == b_stride-1)
 +        && (mb_y == 0 || mb_y == b_height-1)){
 +        if(mb_x == 0)
 +            x1 = block_w;
 +        else
 +            x0 = block_w;
 +        if(mb_y == 0)
 +            y1 = block_h;
 +        else
 +            y0 = block_h;
 +        for(y=y0; y<y1; y++)
 +            memcpy(dst + sx+x0 + (sy+y)*ref_stride, cur + x0 + y*ref_stride, x1-x0);
 +    }
 +
 +    if(block_w==16){
 +        /* FIXME rearrange dsputil to fit 32x32 cmp functions */
 +        /* FIXME check alignment of the cmp wavelet vs the encoding wavelet */
 +        /* FIXME cmps overlap but do not cover the wavelet's whole support.
 +         * So improving the score of one block is not strictly guaranteed
 +         * to improve the score of the whole frame, thus iterative motion
 +         * estimation does not always converge. */
 +        if(s->avctx->me_cmp == FF_CMP_W97)
 +            distortion = ff_w97_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
 +        else if(s->avctx->me_cmp == FF_CMP_W53)
 +            distortion = ff_w53_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
 +        else{
 +            distortion = 0;
 +            for(i=0; i<4; i++){
 +                int off = sx+16*(i&1) + (sy+16*(i>>1))*ref_stride;
 +                distortion += s->dsp.me_cmp[0](&s->m, src + off, dst + off, ref_stride, 16);
 +            }
 +        }
 +    }else{
 +        av_assert2(block_w==8);
 +        distortion = s->dsp.me_cmp[0](&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, block_w*2);
 +    }
 +
 +    if(plane_index==0){
 +        for(i=0; i<4; i++){
 +/* ..RRr
 + * .RXx.
 + * rxx..
 + */
 +            rate += get_block_bits(s, mb_x + (i&1) - (i>>1), mb_y + (i>>1), 1);
 +        }
 +        if(mb_x == b_stride-2)
 +            rate += get_block_bits(s, mb_x + 1, mb_y + 1, 1);
 +    }
 +    return distortion + rate*penalty_factor;
 +}
 +
 +static int get_4block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index){
 +    int i, y2;
 +    Plane *p= &s->plane[plane_index];
 +    const int block_size = MB_SIZE >> s->block_max_depth;
 +    const int block_w    = plane_index ? block_size>>s->chroma_h_shift : block_size;
 +    const int block_h    = plane_index ? block_size>>s->chroma_v_shift : block_size;
 +    const uint8_t *obmc  = plane_index ? ff_obmc_tab[s->block_max_depth+s->chroma_h_shift] : ff_obmc_tab[s->block_max_depth];
 +    const int obmc_stride= plane_index ? (2*block_size)>>s->chroma_h_shift : 2*block_size;
 +    const int ref_stride= s->current_picture->linesize[plane_index];
 +    uint8_t *dst= s->current_picture->data[plane_index];
 +    uint8_t *src= s-> input_picture->data[plane_index];
 +    //FIXME zero_dst is const but add_yblock changes dst if add is 0 (this is never the case for dst=zero_dst
 +    // const has only been removed from zero_dst to suppress a warning
 +    static IDWTELEM zero_dst[4096]; //FIXME
 +    const int b_stride = s->b_width << s->block_max_depth;
 +    const int w= p->width;
 +    const int h= p->height;
 +    int distortion= 0;
 +    int rate= 0;
 +    const int penalty_factor= get_penalty_factor(s->lambda, s->lambda2, s->avctx->me_cmp);
 +
 +    av_assert2(s->chroma_h_shift == s->chroma_v_shift); //obmc and square assumtions below
 +
 +    for(i=0; i<9; i++){
 +        int mb_x2= mb_x + (i%3) - 1;
 +        int mb_y2= mb_y + (i/3) - 1;
 +        int x= block_w*mb_x2 + block_w/2;
 +        int y= block_h*mb_y2 + block_h/2;
 +
 +        add_yblock(s, 0, NULL, zero_dst, dst, obmc,
 +                   x, y, block_w, block_h, w, h, /*dst_stride*/0, ref_stride, obmc_stride, mb_x2, mb_y2, 1, 1, plane_index);
 +
 +        //FIXME find a cleaner/simpler way to skip the outside stuff
 +        for(y2= y; y2<0; y2++)
 +            memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w);
 +        for(y2= h; y2<y+block_h; y2++)
 +            memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, block_w);
 +        if(x<0){
 +            for(y2= y; y2<y+block_h; y2++)
 +                memcpy(dst + x + y2*ref_stride, src + x + y2*ref_stride, -x);
 +        }
 +        if(x+block_w > w){
 +            for(y2= y; y2<y+block_h; y2++)
 +                memcpy(dst + w + y2*ref_stride, src + w + y2*ref_stride, x+block_w - w);
 +        }
 +
 +        av_assert1(block_w== 8 || block_w==16);
 +        distortion += s->dsp.me_cmp[block_w==8](&s->m, src + x + y*ref_stride, dst + x + y*ref_stride, ref_stride, block_h);
 +    }
 +
 +    if(plane_index==0){
 +        BlockNode *b= &s->block[mb_x+mb_y*b_stride];
 +        int merged= same_block(b,b+1) && same_block(b,b+b_stride) && same_block(b,b+b_stride+1);
 +
 +/* ..RRRr
 + * .RXXx.
 + * .RXXx.
 + * rxxx.
 + */
 +        if(merged)
 +            rate = get_block_bits(s, mb_x, mb_y, 2);
 +        for(i=merged?4:0; i<9; i++){
 +            static const int dxy[9][2] = {{0,0},{1,0},{0,1},{1,1},{2,0},{2,1},{-1,2},{0,2},{1,2}};
 +            rate += get_block_bits(s, mb_x + dxy[i][0], mb_y + dxy[i][1], 1);
 +        }
 +    }
 +    return distortion + rate*penalty_factor;
 +}
 +
 +static int encode_subband_c0run(SnowContext *s, SubBand *b, const IDWTELEM *src, const IDWTELEM *parent, int stride, int orientation){
 +    const int w= b->width;
 +    const int h= b->height;
 +    int x, y;
 +
 +    if(1){
 +        int run=0;
 +        int *runs = s->run_buffer;
 +        int run_index=0;
 +        int max_index;
 +
 +        for(y=0; y<h; y++){
 +            for(x=0; x<w; x++){
 +                int v, p=0;
 +                int /*ll=0, */l=0, lt=0, t=0, rt=0;
 +                v= src[x + y*stride];
 +
 +                if(y){
 +                    t= src[x + (y-1)*stride];
 +                    if(x){
 +                        lt= src[x - 1 + (y-1)*stride];
 +                    }
 +                    if(x + 1 < w){
 +                        rt= src[x + 1 + (y-1)*stride];
 +                    }
 +                }
 +                if(x){
 +                    l= src[x - 1 + y*stride];
 +                    /*if(x > 1){
 +                        if(orientation==1) ll= src[y + (x-2)*stride];
 +                        else               ll= src[x - 2 + y*stride];
 +                    }*/
 +                }
 +                if(parent){
 +                    int px= x>>1;
 +                    int py= y>>1;
 +                    if(px<b->parent->width && py<b->parent->height)
 +                        p= parent[px + py*2*stride];
 +                }
 +                if(!(/*ll|*/l|lt|t|rt|p)){
 +                    if(v){
 +                        runs[run_index++]= run;
 +                        run=0;
 +                    }else{
 +                        run++;
 +                    }
 +                }
 +            }
 +        }
 +        max_index= run_index;
 +        runs[run_index++]= run;
 +        run_index=0;
 +        run= runs[run_index++];
 +
 +        put_symbol2(&s->c, b->state[30], max_index, 0);
 +        if(run_index <= max_index)
 +            put_symbol2(&s->c, b->state[1], run, 3);
 +
 +        for(y=0; y<h; y++){
 +            if(s->c.bytestream_end - s->c.bytestream < w*40){
 +                av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
 +                return -1;
 +            }
 +            for(x=0; x<w; x++){
 +                int v, p=0;
 +                int /*ll=0, */l=0, lt=0, t=0, rt=0;
 +                v= src[x + y*stride];
 +
 +                if(y){
 +                    t= src[x + (y-1)*stride];
 +                    if(x){
 +                        lt= src[x - 1 + (y-1)*stride];
 +                    }
 +                    if(x + 1 < w){
 +                        rt= src[x + 1 + (y-1)*stride];
 +                    }
 +                }
 +                if(x){
 +                    l= src[x - 1 + y*stride];
 +                    /*if(x > 1){
 +                        if(orientation==1) ll= src[y + (x-2)*stride];
 +                        else               ll= src[x - 2 + y*stride];
 +                    }*/
 +                }
 +                if(parent){
 +                    int px= x>>1;
 +                    int py= y>>1;
 +                    if(px<b->parent->width && py<b->parent->height)
 +                        p= parent[px + py*2*stride];
 +                }
 +                if(/*ll|*/l|lt|t|rt|p){
 +                    int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p));
 +
 +                    put_rac(&s->c, &b->state[0][context], !!v);
 +                }else{
 +                    if(!run){
 +                        run= runs[run_index++];
 +
 +                        if(run_index <= max_index)
 +                            put_symbol2(&s->c, b->state[1], run, 3);
 +                        av_assert2(v);
 +                    }else{
 +                        run--;
 +                        av_assert2(!v);
 +                    }
 +                }
 +                if(v){
 +                    int context= av_log2(/*FFABS(ll) + */3*FFABS(l) + FFABS(lt) + 2*FFABS(t) + FFABS(rt) + FFABS(p));
 +                    int l2= 2*FFABS(l) + (l<0);
 +                    int t2= 2*FFABS(t) + (t<0);
 +
 +                    put_symbol2(&s->c, b->state[context + 2], FFABS(v)-1, context-4);
 +                    put_rac(&s->c, &b->state[0][16 + 1 + 3 + ff_quant3bA[l2&0xFF] + 3*ff_quant3bA[t2&0xFF]], v<0);
 +                }
 +            }
 +        }
 +    }
 +    return 0;
 +}
 +
 +static int encode_subband(SnowContext *s, SubBand *b, const IDWTELEM *src, const IDWTELEM *parent, int stride, int orientation){
 +//    encode_subband_qtree(s, b, src, parent, stride, orientation);
 +//    encode_subband_z0run(s, b, src, parent, stride, orientation);
 +    return encode_subband_c0run(s, b, src, parent, stride, orientation);
 +//    encode_subband_dzr(s, b, src, parent, stride, orientation);
 +}
 +
 +static av_always_inline int check_block(SnowContext *s, int mb_x, int mb_y, int p[3], int intra, uint8_t (*obmc_edged)[MB_SIZE * 2], int *best_rd){
 +    const int b_stride= s->b_width << s->block_max_depth;
 +    BlockNode *block= &s->block[mb_x + mb_y * b_stride];
 +    BlockNode backup= *block;
 +    unsigned value;
 +    int rd, index;
 +
 +    av_assert2(mb_x>=0 && mb_y>=0);
 +    av_assert2(mb_x<b_stride);
 +
 +    if(intra){
 +        block->color[0] = p[0];
 +        block->color[1] = p[1];
 +        block->color[2] = p[2];
 +        block->type |= BLOCK_INTRA;
 +    }else{
 +        index= (p[0] + 31*p[1]) & (ME_CACHE_SIZE-1);
 +        value= s->me_cache_generation + (p[0]>>10) + (p[1]<<6) + (block->ref<<12);
 +        if(s->me_cache[index] == value)
 +            return 0;
 +        s->me_cache[index]= value;
 +
 +        block->mx= p[0];
 +        block->my= p[1];
 +        block->type &= ~BLOCK_INTRA;
 +    }
 +
 +    rd= get_block_rd(s, mb_x, mb_y, 0, obmc_edged);
 +
 +//FIXME chroma
 +    if(rd < *best_rd){
 +        *best_rd= rd;
 +        return 1;
 +    }else{
 +        *block= backup;
 +        return 0;
 +    }
 +}
 +
 +/* special case for int[2] args we discard afterwards,
 + * fixes compilation problem with gcc 2.95 */
 +static av_always_inline int check_block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, uint8_t (*obmc_edged)[MB_SIZE * 2], int *best_rd){
 +    int p[2] = {p0, p1};
 +    return check_block(s, mb_x, mb_y, p, 0, obmc_edged, best_rd);
 +}
 +
 +static av_always_inline int check_4block_inter(SnowContext *s, int mb_x, int mb_y, int p0, int p1, int ref, int *best_rd){
 +    const int b_stride= s->b_width << s->block_max_depth;
 +    BlockNode *block= &s->block[mb_x + mb_y * b_stride];
 +    BlockNode backup[4];
 +    unsigned value;
 +    int rd, index;
 +
 +    /* We don't initialize backup[] during variable declaration, because
 +     * that fails to compile on MSVC: "cannot convert from 'BlockNode' to
 +     * 'int16_t'". */
 +    backup[0] = block[0];
 +    backup[1] = block[1];
 +    backup[2] = block[b_stride];
 +    backup[3] = block[b_stride + 1];
 +
 +    av_assert2(mb_x>=0 && mb_y>=0);
 +    av_assert2(mb_x<b_stride);
 +    av_assert2(((mb_x|mb_y)&1) == 0);
 +
 +    index= (p0 + 31*p1) & (ME_CACHE_SIZE-1);
 +    value= s->me_cache_generation + (p0>>10) + (p1<<6) + (block->ref<<12);
 +    if(s->me_cache[index] == value)
 +        return 0;
 +    s->me_cache[index]= value;
 +
 +    block->mx= p0;
 +    block->my= p1;
 +    block->ref= ref;
 +    block->type &= ~BLOCK_INTRA;
 +    block[1]= block[b_stride]= block[b_stride+1]= *block;
 +
 +    rd= get_4block_rd(s, mb_x, mb_y, 0);
 +
 +//FIXME chroma
 +    if(rd < *best_rd){
 +        *best_rd= rd;
 +        return 1;
 +    }else{
 +        block[0]= backup[0];
 +        block[1]= backup[1];
 +        block[b_stride]= backup[2];
 +        block[b_stride+1]= backup[3];
 +        return 0;
 +    }
 +}
 +
 +static void iterative_me(SnowContext *s){
 +    int pass, mb_x, mb_y;
 +    const int b_width = s->b_width  << s->block_max_depth;
 +    const int b_height= s->b_height << s->block_max_depth;
 +    const int b_stride= b_width;
 +    int color[3];
 +
 +    {
 +        RangeCoder r = s->c;
 +        uint8_t state[sizeof(s->block_state)];
 +        memcpy(state, s->block_state, sizeof(s->block_state));
 +        for(mb_y= 0; mb_y<s->b_height; mb_y++)
 +            for(mb_x= 0; mb_x<s->b_width; mb_x++)
 +                encode_q_branch(s, 0, mb_x, mb_y);
 +        s->c = r;
 +        memcpy(s->block_state, state, sizeof(s->block_state));
 +    }
 +
 +    for(pass=0; pass<25; pass++){
 +        int change= 0;
 +
 +        for(mb_y= 0; mb_y<b_height; mb_y++){
 +            for(mb_x= 0; mb_x<b_width; mb_x++){
 +                int dia_change, i, j, ref;
 +                int best_rd= INT_MAX, ref_rd;
 +                BlockNode backup, ref_b;
 +                const int index= mb_x + mb_y * b_stride;
 +                BlockNode *block= &s->block[index];
 +                BlockNode *tb =                   mb_y            ? &s->block[index-b_stride  ] : NULL;
 +                BlockNode *lb = mb_x                              ? &s->block[index         -1] : NULL;
 +                BlockNode *rb = mb_x+1<b_width                    ? &s->block[index         +1] : NULL;
 +                BlockNode *bb =                   mb_y+1<b_height ? &s->block[index+b_stride  ] : NULL;
 +                BlockNode *tlb= mb_x           && mb_y            ? &s->block[index-b_stride-1] : NULL;
 +                BlockNode *trb= mb_x+1<b_width && mb_y            ? &s->block[index-b_stride+1] : NULL;
 +                BlockNode *blb= mb_x           && mb_y+1<b_height ? &s->block[index+b_stride-1] : NULL;
 +                BlockNode *brb= mb_x+1<b_width && mb_y+1<b_height ? &s->block[index+b_stride+1] : NULL;
 +                const int b_w= (MB_SIZE >> s->block_max_depth);
 +                uint8_t obmc_edged[MB_SIZE * 2][MB_SIZE * 2];
 +
 +                if(pass && (block->type & BLOCK_OPT))
 +                    continue;
 +                block->type |= BLOCK_OPT;
 +
 +                backup= *block;
 +
 +                if(!s->me_cache_generation)
 +                    memset(s->me_cache, 0, sizeof(s->me_cache));
 +                s->me_cache_generation += 1<<22;
 +
 +                //FIXME precalculate
 +                {
 +                    int x, y;
 +                    for (y = 0; y < b_w * 2; y++)
 +                        memcpy(obmc_edged[y], ff_obmc_tab[s->block_max_depth] + y * b_w * 2, b_w * 2);
 +                    if(mb_x==0)
 +                        for(y=0; y<b_w*2; y++)
 +                            memset(obmc_edged[y], obmc_edged[y][0] + obmc_edged[y][b_w-1], b_w);
 +                    if(mb_x==b_stride-1)
 +                        for(y=0; y<b_w*2; y++)
 +                            memset(obmc_edged[y]+b_w, obmc_edged[y][b_w] + obmc_edged[y][b_w*2-1], b_w);
 +                    if(mb_y==0){
 +                        for(x=0; x<b_w*2; x++)
 +                            obmc_edged[0][x] += obmc_edged[b_w-1][x];
 +                        for(y=1; y<b_w; y++)
 +                            memcpy(obmc_edged[y], obmc_edged[0], b_w*2);
 +                    }
 +                    if(mb_y==b_height-1){
 +                        for(x=0; x<b_w*2; x++)
 +                            obmc_edged[b_w*2-1][x] += obmc_edged[b_w][x];
 +                        for(y=b_w; y<b_w*2-1; y++)
 +                            memcpy(obmc_edged[y], obmc_edged[b_w*2-1], b_w*2);
 +                    }
 +                }
 +
 +                //skip stuff outside the picture
 +                if(mb_x==0 || mb_y==0 || mb_x==b_width-1 || mb_y==b_height-1){
 +                    uint8_t *src= s->  input_picture->data[0];
 +                    uint8_t *dst= s->current_picture->data[0];
 +                    const int stride= s->current_picture->linesize[0];
 +                    const int block_w= MB_SIZE >> s->block_max_depth;
 +                    const int block_h= MB_SIZE >> s->block_max_depth;
 +                    const int sx= block_w*mb_x - block_w/2;
 +                    const int sy= block_h*mb_y - block_h/2;
 +                    const int w= s->plane[0].width;
 +                    const int h= s->plane[0].height;
 +                    int y;
 +
 +                    for(y=sy; y<0; y++)
 +                        memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2);
 +                    for(y=h; y<sy+block_h*2; y++)
 +                        memcpy(dst + sx + y*stride, src + sx + y*stride, block_w*2);
 +                    if(sx<0){
 +                        for(y=sy; y<sy+block_h*2; y++)
 +                            memcpy(dst + sx + y*stride, src + sx + y*stride, -sx);
 +                    }
 +                    if(sx+block_w*2 > w){
 +                        for(y=sy; y<sy+block_h*2; y++)
 +                            memcpy(dst + w + y*stride, src + w + y*stride, sx+block_w*2 - w);
 +                    }
 +                }
 +
 +                // intra(black) = neighbors' contribution to the current block
 +                for(i=0; i < s->nb_planes; i++)
 +                    color[i]= get_dc(s, mb_x, mb_y, i);
 +
 +                // get previous score (cannot be cached due to OBMC)
 +                if(pass > 0 && (block->type&BLOCK_INTRA)){
 +                    int color0[3]= {block->color[0], block->color[1], block->color[2]};
 +                    check_block(s, mb_x, mb_y, color0, 1, obmc_edged, &best_rd);
 +                }else
 +                    check_block_inter(s, mb_x, mb_y, block->mx, block->my, obmc_edged, &best_rd);
 +
 +                ref_b= *block;
 +                ref_rd= best_rd;
 +                for(ref=0; ref < s->ref_frames; ref++){
 +                    int16_t (*mvr)[2]= &s->ref_mvs[ref][index];
 +                    if(s->ref_scores[ref][index] > s->ref_scores[ref_b.ref][index]*3/2) //FIXME tune threshold
 +                        continue;
 +                    block->ref= ref;
 +                    best_rd= INT_MAX;
 +
 +                    check_block_inter(s, mb_x, mb_y, mvr[0][0], mvr[0][1], obmc_edged, &best_rd);
 +                    check_block_inter(s, mb_x, mb_y, 0, 0, obmc_edged, &best_rd);
 +                    if(tb)
 +                        check_block_inter(s, mb_x, mb_y, mvr[-b_stride][0], mvr[-b_stride][1], obmc_edged, &best_rd);
 +                    if(lb)
 +                        check_block_inter(s, mb_x, mb_y, mvr[-1][0], mvr[-1][1], obmc_edged, &best_rd);
 +                    if(rb)
 +                        check_block_inter(s, mb_x, mb_y, mvr[1][0], mvr[1][1], obmc_edged, &best_rd);
 +                    if(bb)
 +                        check_block_inter(s, mb_x, mb_y, mvr[b_stride][0], mvr[b_stride][1], obmc_edged, &best_rd);
 +
 +                    /* fullpel ME */
 +                    //FIXME avoid subpel interpolation / round to nearest integer
 +                    do{
 +                        dia_change=0;
 +                        for(i=0; i<FFMAX(s->avctx->dia_size, 1); i++){
 +                            for(j=0; j<i; j++){
 +                                dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my+(4*j), obmc_edged, &best_rd);
 +                                dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my-(4*j), obmc_edged, &best_rd);
 +                                dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+4*(i-j), block->my-(4*j), obmc_edged, &best_rd);
 +                                dia_change |= check_block_inter(s, mb_x, mb_y, block->mx-4*(i-j), block->my+(4*j), obmc_edged, &best_rd);
 +                            }
 +                        }
 +                    }while(dia_change);
 +                    /* subpel ME */
 +                    do{
 +                        static const int square[8][2]= {{+1, 0},{-1, 0},{ 0,+1},{ 0,-1},{+1,+1},{-1,-1},{+1,-1},{-1,+1},};
 +                        dia_change=0;
 +                        for(i=0; i<8; i++)
 +                            dia_change |= check_block_inter(s, mb_x, mb_y, block->mx+square[i][0], block->my+square[i][1], obmc_edged, &best_rd);
 +                    }while(dia_change);
 +                    //FIXME or try the standard 2 pass qpel or similar
 +
 +                    mvr[0][0]= block->mx;
 +                    mvr[0][1]= block->my;
 +                    if(ref_rd > best_rd){
 +                        ref_rd= best_rd;
 +                        ref_b= *block;
 +                    }
 +                }
 +                best_rd= ref_rd;
 +                *block= ref_b;
 +                check_block(s, mb_x, mb_y, color, 1, obmc_edged, &best_rd);
 +                //FIXME RD style color selection
 +                if(!same_block(block, &backup)){
 +                    if(tb ) tb ->type &= ~BLOCK_OPT;
 +                    if(lb ) lb ->type &= ~BLOCK_OPT;
 +                    if(rb ) rb ->type &= ~BLOCK_OPT;
 +                    if(bb ) bb ->type &= ~BLOCK_OPT;
 +                    if(tlb) tlb->type &= ~BLOCK_OPT;
 +                    if(trb) trb->type &= ~BLOCK_OPT;
 +                    if(blb) blb->type &= ~BLOCK_OPT;
 +                    if(brb) brb->type &= ~BLOCK_OPT;
 +                    change ++;
 +                }
 +            }
 +        }
 +        av_log(s->avctx, AV_LOG_ERROR, "pass:%d changed:%d\n", pass, change);
 +        if(!change)
 +            break;
 +    }
 +
 +    if(s->block_max_depth == 1){
 +        int change= 0;
 +        for(mb_y= 0; mb_y<b_height; mb_y+=2){
 +            for(mb_x= 0; mb_x<b_width; mb_x+=2){
 +                int i;
 +                int best_rd, init_rd;
 +                const int index= mb_x + mb_y * b_stride;
 +                BlockNode *b[4];
 +
 +                b[0]= &s->block[index];
 +                b[1]= b[0]+1;
 +                b[2]= b[0]+b_stride;
 +                b[3]= b[2]+1;
 +                if(same_block(b[0], b[1]) &&
 +                   same_block(b[0], b[2]) &&
 +                   same_block(b[0], b[3]))
 +                    continue;
 +
 +                if(!s->me_cache_generation)
 +                    memset(s->me_cache, 0, sizeof(s->me_cache));
 +                s->me_cache_generation += 1<<22;
 +
 +                init_rd= best_rd= get_4block_rd(s, mb_x, mb_y, 0);
 +
 +                //FIXME more multiref search?
 +                check_4block_inter(s, mb_x, mb_y,
 +                                   (b[0]->mx + b[1]->mx + b[2]->mx + b[3]->mx + 2) >> 2,
 +                                   (b[0]->my + b[1]->my + b[2]->my + b[3]->my + 2) >> 2, 0, &best_rd);
 +
 +                for(i=0; i<4; i++)
 +                    if(!(b[i]->type&BLOCK_INTRA))
 +                        check_4block_inter(s, mb_x, mb_y, b[i]->mx, b[i]->my, b[i]->ref, &best_rd);
 +
 +                if(init_rd != best_rd)
 +                    change++;
 +            }
 +        }
 +        av_log(s->avctx, AV_LOG_ERROR, "pass:4mv changed:%d\n", change*4);
 +    }
 +}
 +
 +static void encode_blocks(SnowContext *s, int search){
 +    int x, y;
 +    int w= s->b_width;
 +    int h= s->b_height;
 +
 +    if(s->avctx->me_method == ME_ITER && !s->keyframe && search)
 +        iterative_me(s);
 +
 +    for(y=0; y<h; y++){
 +        if(s->c.bytestream_end - s->c.bytestream < w*MB_SIZE*MB_SIZE*3){ //FIXME nicer limit
 +            av_log(s->avctx, AV_LOG_ERROR, "encoded frame too large\n");
 +            return;
 +        }
 +        for(x=0; x<w; x++){
 +            if(s->avctx->me_method == ME_ITER || !search)
 +                encode_q_branch2(s, 0, x, y);
 +            else
 +                encode_q_branch (s, 0, x, y);
 +        }
 +    }
 +}
 +
 +static void quantize(SnowContext *s, SubBand *b, IDWTELEM *dst, DWTELEM *src, int stride, int bias){
 +    const int w= b->width;
 +    const int h= b->height;
 +    const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
 +    const int qmul= ff_qexp[qlog&(QROOT-1)]<<((qlog>>QSHIFT) + ENCODER_EXTRA_BITS);
 +    int x,y, thres1, thres2;
 +
 +    if(s->qlog == LOSSLESS_QLOG){
 +        for(y=0; y<h; y++)
 +            for(x=0; x<w; x++)
 +                dst[x + y*stride]= src[x + y*stride];
 +        return;
 +    }
 +
 +    bias= bias ? 0 : (3*qmul)>>3;
 +    thres1= ((qmul - bias)>>QEXPSHIFT) - 1;
 +    thres2= 2*thres1;
 +
 +    if(!bias){
 +        for(y=0; y<h; y++){
 +            for(x=0; x<w; x++){
 +                int i= src[x + y*stride];
 +
 +                if((unsigned)(i+thres1) > thres2){
 +                    if(i>=0){
 +                        i<<= QEXPSHIFT;
 +                        i/= qmul; //FIXME optimize
 +                        dst[x + y*stride]=  i;
 +                    }else{
 +                        i= -i;
 +                        i<<= QEXPSHIFT;
 +                        i/= qmul; //FIXME optimize
 +                        dst[x + y*stride]= -i;
 +                    }
 +                }else
 +                    dst[x + y*stride]= 0;
 +            }
 +        }
 +    }else{
 +        for(y=0; y<h; y++){
 +            for(x=0; x<w; x++){
 +                int i= src[x + y*stride];
 +
 +                if((unsigned)(i+thres1) > thres2){
 +                    if(i>=0){
 +                        i<<= QEXPSHIFT;
 +                        i= (i + bias) / qmul; //FIXME optimize
 +                        dst[x + y*stride]=  i;
 +                    }else{
 +                        i= -i;
 +                        i<<= QEXPSHIFT;
 +                        i= (i + bias) / qmul; //FIXME optimize
 +                        dst[x + y*stride]= -i;
 +                    }
 +                }else
 +                    dst[x + y*stride]= 0;
 +            }
 +        }
 +    }
 +}
 +
 +static void dequantize(SnowContext *s, SubBand *b, IDWTELEM *src, int stride){
 +    const int w= b->width;
 +    const int h= b->height;
 +    const int qlog= av_clip(s->qlog + b->qlog, 0, QROOT*16);
 +    const int qmul= ff_qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
 +    const int qadd= (s->qbias*qmul)>>QBIAS_SHIFT;
 +    int x,y;
 +
 +    if(s->qlog == LOSSLESS_QLOG) return;
 +
 +    for(y=0; y<h; y++){
 +        for(x=0; x<w; x++){
 +            int i= src[x + y*stride];
 +            if(i<0){
 +                src[x + y*stride]= -((-i*qmul + qadd)>>(QEXPSHIFT)); //FIXME try different bias
 +            }else if(i>0){
 +                src[x + y*stride]=  (( i*qmul + qadd)>>(QEXPSHIFT));
 +            }
 +        }
 +    }
 +}
 +
 +static void decorrelate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){
 +    const int w= b->width;
 +    const int h= b->height;
 +    int x,y;
 +
 +    for(y=h-1; y>=0; y--){
 +        for(x=w-1; x>=0; x--){
 +            int i= x + y*stride;
 +
 +            if(x){
 +                if(use_median){
 +                    if(y && x+1<w) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
 +                    else  src[i] -= src[i - 1];
 +                }else{
 +                    if(y) src[i] -= mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]);
 +                    else  src[i] -= src[i - 1];
 +                }
 +            }else{
 +                if(y) src[i] -= src[i - stride];
 +            }
 +        }
 +    }
 +}
 +
 +static void correlate(SnowContext *s, SubBand *b, IDWTELEM *src, int stride, int inverse, int use_median){
 +    const int w= b->width;
 +    const int h= b->height;
 +    int x,y;
 +
 +    for(y=0; y<h; y++){
 +        for(x=0; x<w; x++){
 +            int i= x + y*stride;
 +
 +            if(x){
 +                if(use_median){
 +                    if(y && x+1<w) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - stride + 1]);
 +                    else  src[i] += src[i - 1];
 +                }else{
 +                    if(y) src[i] += mid_pred(src[i - 1], src[i - stride], src[i - 1] + src[i - stride] - src[i - 1 - stride]);
 +                    else  src[i] += src[i - 1];
 +                }
 +            }else{
 +                if(y) src[i] += src[i - stride];
 +            }
 +        }
 +    }
 +}
 +
 +static void encode_qlogs(SnowContext *s){
 +    int plane_index, level, orientation;
 +
 +    for(plane_index=0; plane_index<FFMIN(s->nb_planes, 2); plane_index++){
 +        for(level=0; level<s->spatial_decomposition_count; level++){
 +            for(orientation=level ? 1:0; orientation<4; orientation++){
 +                if(orientation==2) continue;
 +                put_symbol(&s->c, s->header_state, s->plane[plane_index].band[level][orientation].qlog, 1);
 +            }
 +        }
 +    }
 +}
 +
 +static void encode_header(SnowContext *s){
 +    int plane_index, i;
 +    uint8_t kstate[32];
 +
 +    memset(kstate, MID_STATE, sizeof(kstate));
 +
 +    put_rac(&s->c, kstate, s->keyframe);
 +    if(s->keyframe || s->always_reset){
 +        ff_snow_reset_contexts(s);
 +        s->last_spatial_decomposition_type=
 +        s->last_qlog=
 +        s->last_qbias=
 +        s->last_mv_scale=
 +        s->last_block_max_depth= 0;
 +        for(plane_index=0; plane_index<2; plane_index++){
 +            Plane *p= &s->plane[plane_index];
 +            p->last_htaps=0;
 +            p->last_diag_mc=0;
 +            memset(p->last_hcoeff, 0, sizeof(p->last_hcoeff));
 +        }
 +    }
 +    if(s->keyframe){
 +        put_symbol(&s->c, s->header_state, s->version, 0);
 +        put_rac(&s->c, s->header_state, s->always_reset);
 +        put_symbol(&s->c, s->header_state, s->temporal_decomposition_type, 0);
 +        put_symbol(&s->c, s->header_state, s->temporal_decomposition_count, 0);
 +        put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0);
 +        put_symbol(&s->c, s->header_state, s->colorspace_type, 0);
 +        if (s->nb_planes > 2) {
 +            put_symbol(&s->c, s->header_state, s->chroma_h_shift, 0);
 +            put_symbol(&s->c, s->header_state, s->chroma_v_shift, 0);
 +        }
 +        put_rac(&s->c, s->header_state, s->spatial_scalability);
 +//        put_rac(&s->c, s->header_state, s->rate_scalability);
 +        put_symbol(&s->c, s->header_state, s->max_ref_frames-1, 0);
 +
 +        encode_qlogs(s);
 +    }
 +
 +    if(!s->keyframe){
 +        int update_mc=0;
 +        for(plane_index=0; plane_index<FFMIN(s->nb_planes, 2); plane_index++){
 +            Plane *p= &s->plane[plane_index];
 +            update_mc |= p->last_htaps   != p->htaps;
 +            update_mc |= p->last_diag_mc != p->diag_mc;
 +            update_mc |= !!memcmp(p->last_hcoeff, p->hcoeff, sizeof(p->hcoeff));
 +        }
 +        put_rac(&s->c, s->header_state, update_mc);
 +        if(update_mc){
 +            for(plane_index=0; plane_index<FFMIN(s->nb_planes, 2); plane_index++){
 +                Plane *p= &s->plane[plane_index];
 +                put_rac(&s->c, s->header_state, p->diag_mc);
 +                put_symbol(&s->c, s->header_state, p->htaps/2-1, 0);
 +                for(i= p->htaps/2; i; i--)
 +                    put_symbol(&s->c, s->header_state, FFABS(p->hcoeff[i]), 0);
 +            }
 +        }
 +        if(s->last_spatial_decomposition_count != s->spatial_decomposition_count){
 +            put_rac(&s->c, s->header_state, 1);
 +            put_symbol(&s->c, s->header_state, s->spatial_decomposition_count, 0);
 +            encode_qlogs(s);
 +        }else
 +            put_rac(&s->c, s->header_state, 0);
 +    }
 +
 +    put_symbol(&s->c, s->header_state, s->spatial_decomposition_type - s->last_spatial_decomposition_type, 1);
 +    put_symbol(&s->c, s->header_state, s->qlog            - s->last_qlog    , 1);
 +    put_symbol(&s->c, s->header_state, s->mv_scale        - s->last_mv_scale, 1);
 +    put_symbol(&s->c, s->header_state, s->qbias           - s->last_qbias   , 1);
 +    put_symbol(&s->c, s->header_state, s->block_max_depth - s->last_block_max_depth, 1);
 +
 +}
 +
 +static void update_last_header_values(SnowContext *s){
 +    int plane_index;
 +
 +    if(!s->keyframe){
 +        for(plane_index=0; plane_index<2; plane_index++){
 +            Plane *p= &s->plane[plane_index];
 +            p->last_diag_mc= p->diag_mc;
 +            p->last_htaps  = p->htaps;
 +            memcpy(p->last_hcoeff, p->hcoeff, sizeof(p->hcoeff));
 +        }
 +    }
 +
 +    s->last_spatial_decomposition_type  = s->spatial_decomposition_type;
 +    s->last_qlog                        = s->qlog;
 +    s->last_qbias                       = s->qbias;
 +    s->last_mv_scale                    = s->mv_scale;
 +    s->last_block_max_depth             = s->block_max_depth;
 +    s->last_spatial_decomposition_count = s->spatial_decomposition_count;
 +}
 +
 +static int qscale2qlog(int qscale){
 +    return rint(QROOT*log2(qscale / (float)FF_QP2LAMBDA))
 +           + 61*QROOT/8; ///< 64 > 60
 +}
 +
 +static int ratecontrol_1pass(SnowContext *s, AVFrame *pict)
 +{
 +    /* Estimate the frame's complexity as a sum of weighted dwt coefficients.
 +     * FIXME we know exact mv bits at this point,
 +     * but ratecontrol isn't set up to include them. */
 +    uint32_t coef_sum= 0;
 +    int level, orientation, delta_qlog;
 +
 +    for(level=0; level<s->spatial_decomposition_count; level++){
 +        for(orientation=level ? 1 : 0; orientation<4; orientation++){
 +            SubBand *b= &s->plane[0].band[level][orientation];
 +            IDWTELEM *buf= b->ibuf;
 +            const int w= b->width;
 +            const int h= b->height;
 +            const int stride= b->stride;
 +            const int qlog= av_clip(2*QROOT + b->qlog, 0, QROOT*16);
 +            const int qmul= ff_qexp[qlog&(QROOT-1)]<<(qlog>>QSHIFT);
 +            const int qdiv= (1<<16)/qmul;
 +            int x, y;
 +            //FIXME this is ugly
 +            for(y=0; y<h; y++)
 +                for(x=0; x<w; x++)
 +                    buf[x+y*stride]= b->buf[x+y*stride];
 +            if(orientation==0)
 +                decorrelate(s, b, buf, stride, 1, 0);
 +            for(y=0; y<h; y++)
 +                for(x=0; x<w; x++)
 +                    coef_sum+= abs(buf[x+y*stride]) * qdiv >> 16;
 +        }
 +    }
 +
 +    /* ugly, ratecontrol just takes a sqrt again */
 +    av_assert0(coef_sum < INT_MAX);
 +    coef_sum = (uint64_t)coef_sum * coef_sum >> 16;
 +
 +    if(pict->pict_type == AV_PICTURE_TYPE_I){
 +        s->m.current_picture.mb_var_sum= coef_sum;
 +        s->m.current_picture.mc_mb_var_sum= 0;
 +    }else{
 +        s->m.current_picture.mc_mb_var_sum= coef_sum;
 +        s->m.current_picture.mb_var_sum= 0;
 +    }
 +
 +    pict->quality= ff_rate_estimate_qscale(&s->m, 1);
 +    if (pict->quality < 0)
 +        return INT_MIN;
 +    s->lambda= pict->quality * 3/2;
 +    delta_qlog= qscale2qlog(pict->quality) - s->qlog;
 +    s->qlog+= delta_qlog;
 +    return delta_qlog;
 +}
 +
 +static void calculate_visual_weight(SnowContext *s, Plane *p){
 +    int width = p->width;
 +    int height= p->height;
 +    int level, orientation, x, y;
 +
 +    for(level=0; level<s->spatial_decomposition_count; level++){
 +        for(orientation=level ? 1 : 0; orientation<4; orientation++){
 +            SubBand *b= &p->band[level][orientation];
 +            IDWTELEM *ibuf= b->ibuf;
 +            int64_t error=0;
 +
 +            memset(s->spatial_idwt_buffer, 0, sizeof(*s->spatial_idwt_buffer)*width*height);
 +            ibuf[b->width/2 + b->height/2*b->stride]= 256*16;
 +            ff_spatial_idwt(s->spatial_idwt_buffer, s->temp_idwt_buffer, width, height, width, s->spatial_decomposition_type, s->spatial_decomposition_count);
 +            for(y=0; y<height; y++){
 +                for(x=0; x<width; x++){
 +                    int64_t d= s->spatial_idwt_buffer[x + y*width]*16;
 +                    error += d*d;
 +                }
 +            }
 +
 +            b->qlog= (int)(log(352256.0/sqrt(error)) / log(pow(2.0, 1.0/QROOT))+0.5);
 +        }
 +    }
 +}
 +
 +static int encode_frame(AVCodecContext *avctx, AVPacket *pkt,
 +                        AVFrame *pict, int *got_packet)
 +{
 +    SnowContext *s = avctx->priv_data;
 +    RangeCoder * const c= &s->c;
 +    AVFrame *pic = pict;
 +    const int width= s->avctx->width;
 +    const int height= s->avctx->height;
 +    int level, orientation, plane_index, i, y, ret;
 +    uint8_t rc_header_bak[sizeof(s->header_state)];
 +    uint8_t rc_block_bak[sizeof(s->block_state)];
 +
 +    if ((ret = ff_alloc_packet2(avctx, pkt, s->b_width*s->b_height*MB_SIZE*MB_SIZE*3 + FF_MIN_BUFFER_SIZE)) < 0)
 +        return ret;
 +
 +    ff_init_range_encoder(c, pkt->data, pkt->size);
 +    ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
 +
 +    for(i=0; i < s->nb_planes; i++){
 +        int hshift= i ? s->chroma_h_shift : 0;
 +        int vshift= i ? s->chroma_v_shift : 0;
 +        for(y=0; y<(height>>vshift); y++)
 +            memcpy(&s->input_picture->data[i][y * s->input_picture->linesize[i]],
 +                   &pict->data[i][y * pict->linesize[i]],
 +                   width>>hshift);
 +        s->dsp.draw_edges(s->input_picture->data[i], s->input_picture->linesize[i],
 +                            width >> hshift, height >> vshift,
 +                            EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
 +                            EDGE_TOP | EDGE_BOTTOM);
 +
 +    }
 +    emms_c();
 +    s->new_picture = pict;
 +
 +    s->m.picture_number= avctx->frame_number;
 +    if(avctx->flags&CODEC_FLAG_PASS2){
 +        s->m.pict_type = pic->pict_type = s->m.rc_context.entry[avctx->frame_number].new_pict_type;
 +        s->keyframe = pic->pict_type == AV_PICTURE_TYPE_I;
 +        if(!(avctx->flags&CODEC_FLAG_QSCALE)) {
 +            pic->quality = ff_rate_estimate_qscale(&s->m, 0);
 +            if (pic->quality < 0)
 +                return -1;
 +        }
 +    }else{
 +        s->keyframe= avctx->gop_size==0 || avctx->frame_number % avctx->gop_size == 0;
 +        s->m.pict_type = pic->pict_type = s->keyframe ? AV_PICTURE_TYPE_I : AV_PICTURE_TYPE_P;
 +    }
 +
 +    if(s->pass1_rc && avctx->frame_number == 0)
 +        pic->quality = 2*FF_QP2LAMBDA;
 +    if (pic->quality) {
 +        s->qlog   = qscale2qlog(pic->quality);
 +        s->lambda = pic->quality * 3/2;
 +    }
 +    if (s->qlog < 0 || (!pic->quality && (avctx->flags & CODEC_FLAG_QSCALE))) {
 +        s->qlog= LOSSLESS_QLOG;
 +        s->lambda = 0;
 +    }//else keep previous frame's qlog until after motion estimation
 +
 +    ff_snow_frame_start(s);
 +    avctx->coded_frame= s->current_picture;
 +
 +    s->m.current_picture_ptr= &s->m.current_picture;
-         s->m.current_picture.f.data[0] = s->current_picture->data[0];
-         s->m.   last_picture.f.data[0] = s->last_picture[0]->data[0];
-         s->m.    new_picture.f.data[0] = s->  input_picture->data[0];
++    s->m.current_picture.f = s->current_picture;
++    s->m.current_picture.f->pts = pict->pts;
 +    if(pic->pict_type == AV_PICTURE_TYPE_P){
 +        int block_width = (width +15)>>4;
 +        int block_height= (height+15)>>4;
 +        int stride= s->current_picture->linesize[0];
 +
 +        av_assert0(s->current_picture->data[0]);
 +        av_assert0(s->last_picture[0]->data[0]);
 +
 +        s->m.avctx= s->avctx;
-         s->m.   last_picture.f.linesize[0] =
-         s->m.    new_picture.f.linesize[0] =
-         s->m.current_picture.f.linesize[0] = stride;
++        s->m.   last_picture.f = s->last_picture[0];
++        s->m.    new_picture.f = s->input_picture;
++        s->m.current_picture.f->data[0] = s->current_picture->data[0];
++        s->m.   last_picture.f->data[0] = s->last_picture[0]->data[0];
++        s->m.    new_picture.f->data[0] = s->  input_picture->data[0];
 +        s->m.   last_picture_ptr= &s->m.   last_picture;
 +        s->m.linesize=
-     s->m.current_picture.f.display_picture_number =
-     s->m.current_picture.f.coded_picture_number   = avctx->frame_number;
-     s->m.current_picture.f.quality                = pic->quality;
++        s->m.   last_picture.f->linesize[0] =
++        s->m.    new_picture.f->linesize[0] =
++        s->m.current_picture.f->linesize[0] = stride;
 +        s->m.uvlinesize= s->current_picture->linesize[1];
 +        s->m.width = width;
 +        s->m.height= height;
 +        s->m.mb_width = block_width;
 +        s->m.mb_height= block_height;
 +        s->m.mb_stride=   s->m.mb_width+1;
 +        s->m.b8_stride= 2*s->m.mb_width+1;
 +        s->m.f_code=1;
 +        s->m.pict_type = pic->pict_type;
 +        s->m.me_method= s->avctx->me_method;
 +        s->m.me.scene_change_score=0;
 +        s->m.flags= s->avctx->flags;
 +        s->m.quarter_sample= (s->avctx->flags & CODEC_FLAG_QPEL)!=0;
 +        s->m.out_format= FMT_H263;
 +        s->m.unrestricted_mv= 1;
 +
 +        s->m.lambda = s->lambda;
 +        s->m.qscale= (s->m.lambda*139 + FF_LAMBDA_SCALE*64) >> (FF_LAMBDA_SHIFT + 7);
 +        s->lambda2= s->m.lambda2= (s->m.lambda*s->m.lambda + FF_LAMBDA_SCALE/2) >> FF_LAMBDA_SHIFT;
 +
 +        s->m.dsp= s->dsp; //move
 +        s->m.hdsp = s->hdsp;
 +        ff_init_me(&s->m);
 +        s->hdsp = s->m.hdsp;
 +        s->dsp= s->m.dsp;
 +    }
 +
 +    if(s->pass1_rc){
 +        memcpy(rc_header_bak, s->header_state, sizeof(s->header_state));
 +        memcpy(rc_block_bak, s->block_state, sizeof(s->block_state));
 +    }
 +
 +redo_frame:
 +
 +    s->spatial_decomposition_count= 5;
 +
 +    while(   !(width >>(s->chroma_h_shift + s->spatial_decomposition_count))
 +          || !(height>>(s->chroma_v_shift + s->spatial_decomposition_count)))
 +        s->spatial_decomposition_count--;
 +
 +    if (s->spatial_decomposition_count <= 0) {
 +        av_log(avctx, AV_LOG_ERROR, "Resolution too low\n");
 +        return AVERROR(EINVAL);
 +    }
 +
 +    s->m.pict_type = pic->pict_type;
 +    s->qbias = pic->pict_type == AV_PICTURE_TYPE_P ? 2 : 0;
 +
 +    ff_snow_common_init_after_header(avctx);
 +
 +    if(s->last_spatial_decomposition_count != s->spatial_decomposition_count){
 +        for(plane_index=0; plane_index < s->nb_planes; plane_index++){
 +            calculate_visual_weight(s, &s->plane[plane_index]);
 +        }
 +    }
 +
 +    encode_header(s);
 +    s->m.misc_bits = 8*(s->c.bytestream - s->c.bytestream_start);
 +    encode_blocks(s, 1);
 +    s->m.mv_bits = 8*(s->c.bytestream - s->c.bytestream_start) - s->m.misc_bits;
 +
 +    for(plane_index=0; plane_index < s->nb_planes; plane_index++){
 +        Plane *p= &s->plane[plane_index];
 +        int w= p->width;
 +        int h= p->height;
 +        int x, y;
 +//        int bits= put_bits_count(&s->c.pb);
 +
 +        if (!s->memc_only) {
 +            //FIXME optimize
 +            if(pict->data[plane_index]) //FIXME gray hack
 +                for(y=0; y<h; y++){
 +                    for(x=0; x<w; x++){
 +                        s->spatial_idwt_buffer[y*w + x]= pict->data[plane_index][y*pict->linesize[plane_index] + x]<<FRAC_BITS;
 +                    }
 +                }
 +            predict_plane(s, s->spatial_idwt_buffer, plane_index, 0);
 +
 +            if(   plane_index==0
 +               && pic->pict_type == AV_PICTURE_TYPE_P
 +               && !(avctx->flags&CODEC_FLAG_PASS2)
 +               && s->m.me.scene_change_score > s->avctx->scenechange_threshold){
 +                ff_init_range_encoder(c, pkt->data, pkt->size);
 +                ff_build_rac_states(c, 0.05*(1LL<<32), 256-8);
 +                pic->pict_type= AV_PICTURE_TYPE_I;
 +                s->keyframe=1;
 +                s->current_picture->key_frame=1;
 +                goto redo_frame;
 +            }
 +
 +            if(s->qlog == LOSSLESS_QLOG){
 +                for(y=0; y<h; y++){
 +                    for(x=0; x<w; x++){
 +                        s->spatial_dwt_buffer[y*w + x]= (s->spatial_idwt_buffer[y*w + x] + (1<<(FRAC_BITS-1))-1)>>FRAC_BITS;
 +                    }
 +                }
 +            }else{
 +                for(y=0; y<h; y++){
 +                    for(x=0; x<w; x++){
 +                        s->spatial_dwt_buffer[y*w + x]=s->spatial_idwt_buffer[y*w + x]<<ENCODER_EXTRA_BITS;
 +                    }
 +                }
 +            }
 +
 +            ff_spatial_dwt(s->spatial_dwt_buffer, s->temp_dwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count);
 +
 +            if(s->pass1_rc && plane_index==0){
 +                int delta_qlog = ratecontrol_1pass(s, pic);
 +                if (delta_qlog <= INT_MIN)
 +                    return -1;
 +                if(delta_qlog){
 +                    //reordering qlog in the bitstream would eliminate this reset
 +                    ff_init_range_encoder(c, pkt->data, pkt->size);
 +                    memcpy(s->header_state, rc_header_bak, sizeof(s->header_state));
 +                    memcpy(s->block_state, rc_block_bak, sizeof(s->block_state));
 +                    encode_header(s);
 +                    encode_blocks(s, 0);
 +                }
 +            }
 +
 +            for(level=0; level<s->spatial_decomposition_count; level++){
 +                for(orientation=level ? 1 : 0; orientation<4; orientation++){
 +                    SubBand *b= &p->band[level][orientation];
 +
 +                    quantize(s, b, b->ibuf, b->buf, b->stride, s->qbias);
 +                    if(orientation==0)
 +                        decorrelate(s, b, b->ibuf, b->stride, pic->pict_type == AV_PICTURE_TYPE_P, 0);
 +                    if (!s->no_bitstream)
 +                    encode_subband(s, b, b->ibuf, b->parent ? b->parent->ibuf : NULL, b->stride, orientation);
 +                    av_assert0(b->parent==NULL || b->parent->stride == b->stride*2);
 +                    if(orientation==0)
 +                        correlate(s, b, b->ibuf, b->stride, 1, 0);
 +                }
 +            }
 +
 +            for(level=0; level<s->spatial_decomposition_count; level++){
 +                for(orientation=level ? 1 : 0; orientation<4; orientation++){
 +                    SubBand *b= &p->band[level][orientation];
 +
 +                    dequantize(s, b, b->ibuf, b->stride);
 +                }
 +            }
 +
 +            ff_spatial_idwt(s->spatial_idwt_buffer, s->temp_idwt_buffer, w, h, w, s->spatial_decomposition_type, s->spatial_decomposition_count);
 +            if(s->qlog == LOSSLESS_QLOG){
 +                for(y=0; y<h; y++){
 +                    for(x=0; x<w; x++){
 +                        s->spatial_idwt_buffer[y*w + x]<<=FRAC_BITS;
 +                    }
 +                }
 +            }
 +            predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
 +        }else{
 +            //ME/MC only
 +            if(pic->pict_type == AV_PICTURE_TYPE_I){
 +                for(y=0; y<h; y++){
 +                    for(x=0; x<w; x++){
 +                        s->current_picture->data[plane_index][y*s->current_picture->linesize[plane_index] + x]=
 +                            pict->data[plane_index][y*pict->linesize[plane_index] + x];
 +                    }
 +                }
 +            }else{
 +                memset(s->spatial_idwt_buffer, 0, sizeof(IDWTELEM)*w*h);
 +                predict_plane(s, s->spatial_idwt_buffer, plane_index, 1);
 +            }
 +        }
 +        if(s->avctx->flags&CODEC_FLAG_PSNR){
 +            int64_t error= 0;
 +
 +            if(pict->data[plane_index]) //FIXME gray hack
 +                for(y=0; y<h; y++){
 +                    for(x=0; x<w; x++){
 +                        int d= s->current_picture->data[plane_index][y*s->current_picture->linesize[plane_index] + x] - pict->data[plane_index][y*pict->linesize[plane_index] + x];
 +                        error += d*d;
 +                    }
 +                }
 +            s->avctx->error[plane_index] += error;
 +            s->current_picture->error[plane_index] = error;
 +        }
 +
 +    }
 +
 +    update_last_header_values(s);
 +
 +    ff_snow_release_buffer(avctx);
 +
 +    s->current_picture->coded_picture_number = avctx->frame_number;
 +    s->current_picture->pict_type = pict->pict_type;
 +    s->current_picture->quality = pict->quality;
 +    s->m.frame_bits = 8*(s->c.bytestream - s->c.bytestream_start);
 +    s->m.p_tex_bits = s->m.frame_bits - s->m.misc_bits - s->m.mv_bits;
++    s->m.current_picture.f->display_picture_number =
++    s->m.current_picture.f->coded_picture_number   = avctx->frame_number;
++    s->m.current_picture.f->quality                = pic->quality;
 +    s->m.total_bits += 8*(s->c.bytestream - s->c.bytestream_start);
 +    if(s->pass1_rc)
 +        if (ff_rate_estimate_qscale(&s->m, 0) < 0)
 +            return -1;
 +    if(avctx->flags&CODEC_FLAG_PASS1)
 +        ff_write_pass1_stats(&s->m);
 +    s->m.last_pict_type = s->m.pict_type;
 +    avctx->frame_bits = s->m.frame_bits;
 +    avctx->mv_bits = s->m.mv_bits;
 +    avctx->misc_bits = s->m.misc_bits;
 +    avctx->p_tex_bits = s->m.p_tex_bits;
 +
 +    emms_c();
 +
 +    pkt->size = ff_rac_terminate(c);
 +    if (avctx->coded_frame->key_frame)
 +        pkt->flags |= AV_PKT_FLAG_KEY;
 +    *got_packet = 1;
 +
 +    return 0;
 +}
 +
 +static av_cold int encode_end(AVCodecContext *avctx)
 +{
 +    SnowContext *s = avctx->priv_data;
 +
 +    ff_snow_common_end(s);
 +    ff_rate_control_uninit(&s->m);
 +    av_frame_free(&s->input_picture);
 +    av_free(avctx->stats_out);
 +
 +    return 0;
 +}
 +
 +#define OFFSET(x) offsetof(SnowContext, x)
 +#define VE AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
 +static const AVOption options[] = {
 +    { "memc_only",      "Only do ME/MC (I frames -> ref, P frame -> ME+MC).",   OFFSET(memc_only), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
 +    { "no_bitstream",   "Skip final bitstream writeout.",                    OFFSET(no_bitstream), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, VE },
 +    { NULL },
 +};
 +
 +static const AVClass snowenc_class = {
 +    .class_name = "snow encoder",
 +    .item_name  = av_default_item_name,
 +    .option     = options,
 +    .version    = LIBAVUTIL_VERSION_INT,
 +};
 +
 +AVCodec ff_snow_encoder = {
 +    .name           = "snow",
 +    .long_name      = NULL_IF_CONFIG_SMALL("Snow"),
 +    .type           = AVMEDIA_TYPE_VIDEO,
 +    .id             = AV_CODEC_ID_SNOW,
 +    .priv_data_size = sizeof(SnowContext),
 +    .init           = encode_init,
 +    .encode2        = encode_frame,
 +    .close          = encode_end,
 +    .pix_fmts       = (const enum AVPixelFormat[]){
 +        AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV444P,
 +        AV_PIX_FMT_GRAY8,
 +        AV_PIX_FMT_NONE
 +    },
 +    .priv_class     = &snowenc_class,
 +};
 +
 +
 +#ifdef TEST
 +#undef malloc
 +#undef free
 +#undef printf
 +
 +#include "libavutil/lfg.h"
 +#include "libavutil/mathematics.h"
 +
 +int main(void){
 +#define width  256
 +#define height 256
 +    int buffer[2][width*height];
 +    SnowContext s;
 +    int i;
 +    AVLFG prng;
 +    s.spatial_decomposition_count=6;
 +    s.spatial_decomposition_type=1;
 +
 +    s.temp_dwt_buffer  = av_mallocz(width * sizeof(DWTELEM));
 +    s.temp_idwt_buffer = av_mallocz(width * sizeof(IDWTELEM));
 +
 +    av_lfg_init(&prng, 1);
 +
 +    printf("testing 5/3 DWT\n");
 +    for(i=0; i<width*height; i++)
 +        buffer[0][i] = buffer[1][i] = av_lfg_get(&prng) % 54321 - 12345;
 +
 +    ff_spatial_dwt(buffer[0], s.temp_dwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
 +    ff_spatial_idwt((IDWTELEM*)buffer[0], s.temp_idwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
 +
 +    for(i=0; i<width*height; i++)
 +        if(buffer[0][i]!= buffer[1][i]) printf("fsck: %6d %12d %7d\n",i, buffer[0][i], buffer[1][i]);
 +
 +    printf("testing 9/7 DWT\n");
 +    s.spatial_decomposition_type=0;
 +    for(i=0; i<width*height; i++)
 +        buffer[0][i] = buffer[1][i] = av_lfg_get(&prng) % 54321 - 12345;
 +
 +    ff_spatial_dwt(buffer[0], s.temp_dwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
 +    ff_spatial_idwt((IDWTELEM*)buffer[0], s.temp_idwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
 +
 +    for(i=0; i<width*height; i++)
 +        if(FFABS(buffer[0][i] - buffer[1][i])>20) printf("fsck: %6d %12d %7d\n",i, buffer[0][i], buffer[1][i]);
 +
 +    {
 +    int level, orientation, x, y;
 +    int64_t errors[8][4];
 +    int64_t g=0;
 +
 +        memset(errors, 0, sizeof(errors));
 +        s.spatial_decomposition_count=3;
 +        s.spatial_decomposition_type=0;
 +        for(level=0; level<s.spatial_decomposition_count; level++){
 +            for(orientation=level ? 1 : 0; orientation<4; orientation++){
 +                int w= width  >> (s.spatial_decomposition_count-level);
 +                int h= height >> (s.spatial_decomposition_count-level);
 +                int stride= width  << (s.spatial_decomposition_count-level);
 +                DWTELEM *buf= buffer[0];
 +                int64_t error=0;
 +
 +                if(orientation&1) buf+=w;
 +                if(orientation>1) buf+=stride>>1;
 +
 +                memset(buffer[0], 0, sizeof(int)*width*height);
 +                buf[w/2 + h/2*stride]= 256*256;
 +                ff_spatial_idwt((IDWTELEM*)buffer[0], s.temp_idwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
 +                for(y=0; y<height; y++){
 +                    for(x=0; x<width; x++){
 +                        int64_t d= buffer[0][x + y*width];
 +                        error += d*d;
 +                        if(FFABS(width/2-x)<9 && FFABS(height/2-y)<9 && level==2) printf("%8"PRId64" ", d);
 +                    }
 +                    if(FFABS(height/2-y)<9 && level==2) printf("\n");
 +                }
 +                error= (int)(sqrt(error)+0.5);
 +                errors[level][orientation]= error;
 +                if(g) g=av_gcd(g, error);
 +                else g= error;
 +            }
 +        }
 +        printf("static int const visual_weight[][4]={\n");
 +        for(level=0; level<s.spatial_decomposition_count; level++){
 +            printf("  {");
 +            for(orientation=0; orientation<4; orientation++){
 +                printf("%8"PRId64",", errors[level][orientation]/g);
 +            }
 +            printf("},\n");
 +        }
 +        printf("};\n");
 +        {
 +            int level=2;
 +            int w= width  >> (s.spatial_decomposition_count-level);
 +            //int h= height >> (s.spatial_decomposition_count-level);
 +            int stride= width  << (s.spatial_decomposition_count-level);
 +            DWTELEM *buf= buffer[0];
 +            int64_t error=0;
 +
 +            buf+=w;
 +            buf+=stride>>1;
 +
 +            memset(buffer[0], 0, sizeof(int)*width*height);
 +            for(y=0; y<height; y++){
 +                for(x=0; x<width; x++){
 +                    int tab[4]={0,2,3,1};
 +                    buffer[0][x+width*y]= 256*256*tab[(x&1) + 2*(y&1)];
 +                }
 +            }
 +            ff_spatial_dwt(buffer[0], s.temp_dwt_buffer, width, height, width, s.spatial_decomposition_type, s.spatial_decomposition_count);
 +            for(y=0; y<height; y++){
 +                for(x=0; x<width; x++){
 +                    int64_t d= buffer[0][x + y*width];
 +                    error += d*d;
 +                    if(FFABS(width/2-x)<9 && FFABS(height/2-y)<9) printf("%8"PRId64" ", d);
 +                }
 +                if(FFABS(height/2-y)<9) printf("\n");
 +            }
 +        }
 +
 +    }
 +    return 0;
 +}
 +#endif /* TEST */
Simple merge
Simple merge
index 0b9958d,0000000..63879e5
mode 100644,000000..100644
--- /dev/null
@@@ -1,48 -1,0 +1,48 @@@
-                                   ff_vaapi_get_surface_id(&s->current_picture_ptr->f));
 +/*
 + * Video Acceleration API (video decoding)
 + * HW decode acceleration for MPEG-2, MPEG-4, H.264 and VC-1
 + *
 + * Copyright (C) 2013 Anton Khirnov
 + *
 + * This file is part of FFmpeg.
 + *
 + * FFmpeg is free software; you can redistribute it and/or
 + * modify it under the terms of the GNU Lesser General Public
 + * License as published by the Free Software Foundation; either
 + * version 2.1 of the License, or (at your option) any later version.
 + *
 + * FFmpeg is distributed in the hope that it will be useful,
 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 + * Lesser General Public License for more details.
 + *
 + * You should have received a copy of the GNU Lesser General Public
 + * License along with FFmpeg; if not, write to the Free Software
 + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 + */
 +
 +#include "avcodec.h"
 +#include "vaapi_internal.h"
 +
 +int ff_vaapi_mpeg_end_frame(AVCodecContext *avctx)
 +{
 +    struct vaapi_context * const vactx = avctx->hwaccel_context;
 +    MpegEncContext *s = avctx->priv_data;
 +    int ret;
 +
 +    ret = ff_vaapi_commit_slices(vactx);
 +    if (ret < 0)
 +        goto finish;
 +
 +    ret = ff_vaapi_render_picture(vactx,
++                                  ff_vaapi_get_surface_id(s->current_picture_ptr->f));
 +    if (ret < 0)
 +        goto finish;
 +
 +    ff_mpeg_draw_horiz_band(s, 0, s->avctx->height);
 +
 +finish:
 +    ff_vaapi_common_end_frame(avctx);
 +    return ret;
 +}
 +
Simple merge
Simple merge
Simple merge
Simple merge
@@@ -373,16 -385,16 +373,16 @@@ static void vc1_mc_1mv(VC1Context *v, i
      }
      if (!dir) {
          if (v->field_mode && (v->cur_field_type != v->ref_field_type[dir]) && v->second_field) {
-             srcY = s->current_picture.f.data[0];
-             srcU = s->current_picture.f.data[1];
-             srcV = s->current_picture.f.data[2];
+             srcY = s->current_picture.f->data[0];
+             srcU = s->current_picture.f->data[1];
+             srcV = s->current_picture.f->data[2];
              luty  = v->curr_luty;
              lutuv = v->curr_lutuv;
 -            use_ic = v->curr_use_ic;
 +            use_ic = *v->curr_use_ic;
          } else {
-             srcY = s->last_picture.f.data[0];
-             srcU = s->last_picture.f.data[1];
-             srcV = s->last_picture.f.data[2];
+             srcY = s->last_picture.f->data[0];
+             srcU = s->last_picture.f->data[1];
+             srcV = s->last_picture.f->data[2];
              luty  = v->last_luty;
              lutuv = v->last_lutuv;
              use_ic = v->last_use_ic;
@@@ -570,11 -580,11 +570,11 @@@ static void vc1_mc_4mv_luma(VC1Context 
  
      if (!dir) {
          if (v->field_mode && (v->cur_field_type != v->ref_field_type[dir]) && v->second_field) {
-             srcY = s->current_picture.f.data[0];
+             srcY = s->current_picture.f->data[0];
              luty = v->curr_luty;
 -            use_ic = v->curr_use_ic;
 +            use_ic = *v->curr_use_ic;
          } else {
-             srcY = s->last_picture.f.data[0];
+             srcY = s->last_picture.f->data[0];
              luty = v->last_luty;
              use_ic = v->last_use_ic;
          }
@@@ -871,13 -879,13 +871,13 @@@ static void vc1_mc_4mv_chroma(VC1Contex
  
      if (!dir) {
          if (v->field_mode && (v->cur_field_type != chroma_ref_type) && v->second_field) {
-             srcU = s->current_picture.f.data[1];
-             srcV = s->current_picture.f.data[2];
+             srcU = s->current_picture.f->data[1];
+             srcV = s->current_picture.f->data[2];
              lutuv = v->curr_lutuv;
 -            use_ic = v->curr_use_ic;
 +            use_ic = *v->curr_use_ic;
          } else {
-             srcU = s->last_picture.f.data[1];
-             srcV = s->last_picture.f.data[2];
+             srcU = s->last_picture.f->data[1];
+             srcV = s->last_picture.f->data[2];
              lutuv = v->last_lutuv;
              use_ic = v->last_use_ic;
          }
@@@ -1004,13 -1012,13 +1004,13 @@@ static void vc1_mc_4mv_chroma4(VC1Conte
          uvsrc_x = av_clip(uvsrc_x, -8, s->avctx->coded_width  >> 1);
          uvsrc_y = av_clip(uvsrc_y, -8, s->avctx->coded_height >> 1);
          if (i < 2 ? dir : dir2) {
-             srcU = s->next_picture.f.data[1];
-             srcV = s->next_picture.f.data[2];
 -            srcU = s->next_picture.f->data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
 -            srcV = s->next_picture.f->data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
++            srcU = s->next_picture.f->data[1];
++            srcV = s->next_picture.f->data[2];
              lutuv  = v->next_lutuv;
              use_ic = v->next_use_ic;
          } else {
-             srcU = s->last_picture.f.data[1];
-             srcV = s->last_picture.f.data[2];
 -            srcU = s->last_picture.f->data[1] + uvsrc_y * s->uvlinesize + uvsrc_x;
 -            srcV = s->last_picture.f->data[2] + uvsrc_y * s->uvlinesize + uvsrc_x;
++            srcU = s->last_picture.f->data[1];
++            srcV = s->last_picture.f->data[2];
              lutuv  = v->last_lutuv;
              use_ic = v->last_use_ic;
          }
@@@ -5459,13 -5445,9 +5459,13 @@@ static int vc1_decode_sprites(VC1Contex
      AVCodecContext *avctx = s->avctx;
      SpriteData sd;
  
 -    vc1_parse_sprites(v, gb, &sd);
 +    memset(&sd, 0, sizeof(sd));
 +
 +    ret = vc1_parse_sprites(v, gb, &sd);
 +    if (ret < 0)
 +        return ret;
  
-     if (!s->current_picture.f.data[0]) {
+     if (!s->current_picture.f->data[0]) {
          av_log(avctx, AV_LOG_ERROR, "Got no sprites\n");
          return -1;
      }
@@@ -5994,14 -5935,9 +5994,14 @@@ static int vc1_decode_frame(AVCodecCont
          goto err;
      }
  
 +    if ((s->mb_height >> v->field_mode) == 0) {
 +        av_log(v->s.avctx, AV_LOG_ERROR, "image too short\n");
 +        goto err;
 +    }
 +
      // for skipping the frame
-     s->current_picture.f.pict_type = s->pict_type;
-     s->current_picture.f.key_frame = s->pict_type == AV_PICTURE_TYPE_I;
+     s->current_picture.f->pict_type = s->pict_type;
+     s->current_picture.f->key_frame = s->pict_type == AV_PICTURE_TYPE_I;
  
      /* skip B-frames if we don't have reference frames */
      if (s->last_picture_ptr == NULL && (s->pict_type == AV_PICTURE_TYPE_B || s->droppable)) {
          goto err;
      }
  
-     v->s.current_picture_ptr->f.interlaced_frame = (v->fcm != PROGRESSIVE);
-     v->s.current_picture_ptr->f.top_field_first  = v->tff;
 +    v->s.current_picture_ptr->field_picture = v->field_mode;
++    v->s.current_picture_ptr->f->interlaced_frame = (v->fcm != PROGRESSIVE);
++    v->s.current_picture_ptr->f->top_field_first  = v->tff;
 +
      // process pulldown flags
-     s->current_picture_ptr->f.repeat_pict = 0;
+     s->current_picture_ptr->f->repeat_pict = 0;
      // Pulldown flags are only valid when 'broadcast' has been set.
      // So ticks_per_frame will be 2
      if (v->rff) {
      s->me.qpel_put = s->dsp.put_qpel_pixels_tab;
      s->me.qpel_avg = s->dsp.avg_qpel_pixels_tab;
  
 -    if (avctx->hwaccel) {
 -        if (avctx->hwaccel->start_frame(avctx, buf, buf_size) < 0)
 -            goto err;
 -        if (avctx->hwaccel->decode_slice(avctx, buf_start, (buf + buf_size) - buf_start) < 0)
 -            goto err;
 -        if (avctx->hwaccel->end_frame(avctx) < 0)
 -            goto err;
 +    if ((CONFIG_VC1_VDPAU_DECODER)
 +        &&s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU) {
 +        if (v->field_mode && buf_start_second_field) {
 +            ff_vdpau_vc1_decode_picture(s, buf_start, buf_start_second_field - buf_start);
 +            ff_vdpau_vc1_decode_picture(s, buf_start_second_field, (buf + buf_size) - buf_start_second_field);
 +        } else {
 +            ff_vdpau_vc1_decode_picture(s, buf_start, (buf + buf_size) - buf_start);
 +        }
 +    } else if (avctx->hwaccel) {
 +        if (v->field_mode && buf_start_second_field) {
 +            // decode first field
 +            s->picture_structure = PICT_BOTTOM_FIELD - v->tff;
 +            if (avctx->hwaccel->start_frame(avctx, buf_start, buf_start_second_field - buf_start) < 0)
 +                goto err;
 +            if (avctx->hwaccel->decode_slice(avctx, buf_start, buf_start_second_field - buf_start) < 0)
 +                goto err;
 +            if (avctx->hwaccel->end_frame(avctx) < 0)
 +                goto err;
 +
 +            // decode second field
 +            s->gb = slices[n_slices1 + 1].gb;
 +            s->picture_structure = PICT_TOP_FIELD + v->tff;
 +            v->second_field = 1;
 +            v->pic_header_flag = 0;
 +            if (ff_vc1_parse_frame_header_adv(v, &s->gb) < 0) {
 +                av_log(avctx, AV_LOG_ERROR, "parsing header for second field failed");
 +                goto err;
 +            }
-             v->s.current_picture_ptr->f.pict_type = v->s.pict_type;
++            v->s.current_picture_ptr->f->pict_type = v->s.pict_type;
 +
 +            if (avctx->hwaccel->start_frame(avctx, buf_start_second_field, (buf + buf_size) - buf_start_second_field) < 0)
 +                goto err;
 +            if (avctx->hwaccel->decode_slice(avctx, buf_start_second_field, (buf + buf_size) - buf_start_second_field) < 0)
 +                goto err;
 +            if (avctx->hwaccel->end_frame(avctx) < 0)
 +                goto err;
 +        } else {
 +            s->picture_structure = PICT_FRAME;
 +            if (avctx->hwaccel->start_frame(avctx, buf_start, (buf + buf_size) - buf_start) < 0)
 +                goto err;
 +            if (avctx->hwaccel->decode_slice(avctx, buf_start, (buf + buf_size) - buf_start) < 0)
 +                goto err;
 +            if (avctx->hwaccel->end_frame(avctx) < 0)
 +                goto err;
 +        }
      } else {
          int header_ret = 0;
  
@@@ -6205,14 -6086,14 +6205,14 @@@ image
          *got_frame = 1;
      } else {
          if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay) {
-             if ((ret = av_frame_ref(pict, &s->current_picture_ptr->f)) < 0)
+             if ((ret = av_frame_ref(pict, s->current_picture_ptr->f)) < 0)
                  goto err;
 -            ff_print_debug_info(s, s->current_picture_ptr);
 +            ff_print_debug_info(s, s->current_picture_ptr, pict);
              *got_frame = 1;
          } else if (s->last_picture_ptr != NULL) {
-             if ((ret = av_frame_ref(pict, &s->last_picture_ptr->f)) < 0)
+             if ((ret = av_frame_ref(pict, s->last_picture_ptr->f)) < 0)
                  goto err;
 -            ff_print_debug_info(s, s->last_picture_ptr);
 +            ff_print_debug_info(s, s->last_picture_ptr, pict);
              *got_frame = 1;
          }
      }
@@@ -65,21 -57,8 +65,21 @@@ int ff_vdpau_mpeg_end_frame(AVCodecCont
      MpegEncContext *s = avctx->priv_data;
      Picture *pic = s->current_picture_ptr;
      struct vdpau_picture_context *pic_ctx = pic->hwaccel_picture_private;
-     VdpVideoSurface surf = ff_vdpau_get_surface_id(&pic->f);
+     VdpVideoSurface surf = ff_vdpau_get_surface_id(pic->f);
  
-         res = hwctx->render2(avctx, &pic->f, (void *)&pic_ctx->info,
 +#if FF_API_BUFS_VDPAU
 +FF_DISABLE_DEPRECATION_WARNINGS
 +    hwctx->info = pic_ctx->info;
 +    hwctx->bitstream_buffers = pic_ctx->bitstream_buffers;
 +    hwctx->bitstream_buffers_used = pic_ctx->bitstream_buffers_used;
 +    hwctx->bitstream_buffers_allocated = pic_ctx->bitstream_buffers_allocated;
 +FF_ENABLE_DEPRECATION_WARNINGS
 +#endif
 +
 +    if (!hwctx->render) {
++        res = hwctx->render2(avctx, pic->f, (void *)&pic_ctx->info,
 +                             pic_ctx->bitstream_buffers_used, pic_ctx->bitstream_buffers);
 +    } else
      hwctx->render(hwctx->decoder, surf, (void *)&pic_ctx->info,
                    pic_ctx->bitstream_buffers_used, pic_ctx->bitstream_buffers);
  
@@@ -117,343 -88,6 +117,343 @@@ int ff_vdpau_add_buffer(struct vdpau_pi
      return 0;
  }
  
-     render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
 +/* Obsolete non-hwaccel VDPAU support below... */
 +
 +void ff_vdpau_h264_set_reference_frames(H264Context *h)
 +{
 +    struct vdpau_render_state *render, *render_ref;
 +    VdpReferenceFrameH264 *rf, *rf2;
 +    H264Picture *pic;
 +    int i, list, pic_frame_idx;
 +
 +    render = (struct vdpau_render_state *)h->cur_pic_ptr->f.data[0];
 +    assert(render);
 +
 +    rf = &render->info.h264.referenceFrames[0];
 +#define H264_RF_COUNT FF_ARRAY_ELEMS(render->info.h264.referenceFrames)
 +
 +    for (list = 0; list < 2; ++list) {
 +        H264Picture **lp = list ? h->long_ref : h->short_ref;
 +        int ls = list ? 16 : h->short_ref_count;
 +
 +        for (i = 0; i < ls; ++i) {
 +            pic = lp[i];
 +            if (!pic || !pic->reference)
 +                continue;
 +            pic_frame_idx = pic->long_ref ? pic->pic_id : pic->frame_num;
 +
 +            render_ref = (struct vdpau_render_state *)pic->f.data[0];
 +            assert(render_ref);
 +
 +            rf2 = &render->info.h264.referenceFrames[0];
 +            while (rf2 != rf) {
 +                if (
 +                    (rf2->surface == render_ref->surface)
 +                    && (rf2->is_long_term == pic->long_ref)
 +                    && (rf2->frame_idx == pic_frame_idx)
 +                )
 +                    break;
 +                ++rf2;
 +            }
 +            if (rf2 != rf) {
 +                rf2->top_is_reference    |= (pic->reference & PICT_TOP_FIELD)    ? VDP_TRUE : VDP_FALSE;
 +                rf2->bottom_is_reference |= (pic->reference & PICT_BOTTOM_FIELD) ? VDP_TRUE : VDP_FALSE;
 +                continue;
 +            }
 +
 +            if (rf >= &render->info.h264.referenceFrames[H264_RF_COUNT])
 +                continue;
 +
 +            rf->surface             = render_ref->surface;
 +            rf->is_long_term        = pic->long_ref;
 +            rf->top_is_reference    = (pic->reference & PICT_TOP_FIELD)    ? VDP_TRUE : VDP_FALSE;
 +            rf->bottom_is_reference = (pic->reference & PICT_BOTTOM_FIELD) ? VDP_TRUE : VDP_FALSE;
 +            rf->field_order_cnt[0]  = pic->field_poc[0];
 +            rf->field_order_cnt[1]  = pic->field_poc[1];
 +            rf->frame_idx           = pic_frame_idx;
 +
 +            ++rf;
 +        }
 +    }
 +
 +    for (; rf < &render->info.h264.referenceFrames[H264_RF_COUNT]; ++rf) {
 +        rf->surface             = VDP_INVALID_HANDLE;
 +        rf->is_long_term        = 0;
 +        rf->top_is_reference    = 0;
 +        rf->bottom_is_reference = 0;
 +        rf->field_order_cnt[0]  = 0;
 +        rf->field_order_cnt[1]  = 0;
 +        rf->frame_idx           = 0;
 +    }
 +}
 +
 +void ff_vdpau_add_data_chunk(uint8_t *data, const uint8_t *buf, int buf_size)
 +{
 +    struct vdpau_render_state *render = (struct vdpau_render_state*)data;
 +    assert(render);
 +
 +    render->bitstream_buffers= av_fast_realloc(
 +        render->bitstream_buffers,
 +        &render->bitstream_buffers_allocated,
 +        sizeof(*render->bitstream_buffers)*(render->bitstream_buffers_used + 1)
 +    );
 +
 +    render->bitstream_buffers[render->bitstream_buffers_used].struct_version  = VDP_BITSTREAM_BUFFER_VERSION;
 +    render->bitstream_buffers[render->bitstream_buffers_used].bitstream       = buf;
 +    render->bitstream_buffers[render->bitstream_buffers_used].bitstream_bytes = buf_size;
 +    render->bitstream_buffers_used++;
 +}
 +
 +#if CONFIG_H264_VDPAU_DECODER
 +void ff_vdpau_h264_picture_start(H264Context *h)
 +{
 +    struct vdpau_render_state *render;
 +    int i;
 +
 +    render = (struct vdpau_render_state *)h->cur_pic_ptr->f.data[0];
 +    assert(render);
 +
 +    for (i = 0; i < 2; ++i) {
 +        int foc = h->cur_pic_ptr->field_poc[i];
 +        if (foc == INT_MAX)
 +            foc = 0;
 +        render->info.h264.field_order_cnt[i] = foc;
 +    }
 +
 +    render->info.h264.frame_num = h->frame_num;
 +}
 +
 +void ff_vdpau_h264_picture_complete(H264Context *h)
 +{
 +    struct vdpau_render_state *render;
 +
 +    render = (struct vdpau_render_state *)h->cur_pic_ptr->f.data[0];
 +    assert(render);
 +
 +    render->info.h264.slice_count = h->slice_num;
 +    if (render->info.h264.slice_count < 1)
 +        return;
 +
 +    render->info.h264.is_reference                           = (h->cur_pic_ptr->reference & 3) ? VDP_TRUE : VDP_FALSE;
 +    render->info.h264.field_pic_flag                         = h->picture_structure != PICT_FRAME;
 +    render->info.h264.bottom_field_flag                      = h->picture_structure == PICT_BOTTOM_FIELD;
 +    render->info.h264.num_ref_frames                         = h->sps.ref_frame_count;
 +    render->info.h264.mb_adaptive_frame_field_flag           = h->sps.mb_aff && !render->info.h264.field_pic_flag;
 +    render->info.h264.constrained_intra_pred_flag            = h->pps.constrained_intra_pred;
 +    render->info.h264.weighted_pred_flag                     = h->pps.weighted_pred;
 +    render->info.h264.weighted_bipred_idc                    = h->pps.weighted_bipred_idc;
 +    render->info.h264.frame_mbs_only_flag                    = h->sps.frame_mbs_only_flag;
 +    render->info.h264.transform_8x8_mode_flag                = h->pps.transform_8x8_mode;
 +    render->info.h264.chroma_qp_index_offset                 = h->pps.chroma_qp_index_offset[0];
 +    render->info.h264.second_chroma_qp_index_offset          = h->pps.chroma_qp_index_offset[1];
 +    render->info.h264.pic_init_qp_minus26                    = h->pps.init_qp - 26;
 +    render->info.h264.num_ref_idx_l0_active_minus1           = h->pps.ref_count[0] - 1;
 +    render->info.h264.num_ref_idx_l1_active_minus1           = h->pps.ref_count[1] - 1;
 +    render->info.h264.log2_max_frame_num_minus4              = h->sps.log2_max_frame_num - 4;
 +    render->info.h264.pic_order_cnt_type                     = h->sps.poc_type;
 +    render->info.h264.log2_max_pic_order_cnt_lsb_minus4      = h->sps.poc_type ? 0 : h->sps.log2_max_poc_lsb - 4;
 +    render->info.h264.delta_pic_order_always_zero_flag       = h->sps.delta_pic_order_always_zero_flag;
 +    render->info.h264.direct_8x8_inference_flag              = h->sps.direct_8x8_inference_flag;
 +    render->info.h264.entropy_coding_mode_flag               = h->pps.cabac;
 +    render->info.h264.pic_order_present_flag                 = h->pps.pic_order_present;
 +    render->info.h264.deblocking_filter_control_present_flag = h->pps.deblocking_filter_parameters_present;
 +    render->info.h264.redundant_pic_cnt_present_flag         = h->pps.redundant_pic_cnt_present;
 +    memcpy(render->info.h264.scaling_lists_4x4, h->pps.scaling_matrix4, sizeof(render->info.h264.scaling_lists_4x4));
 +    memcpy(render->info.h264.scaling_lists_8x8[0], h->pps.scaling_matrix8[0], sizeof(render->info.h264.scaling_lists_8x8[0]));
 +    memcpy(render->info.h264.scaling_lists_8x8[1], h->pps.scaling_matrix8[3], sizeof(render->info.h264.scaling_lists_8x8[0]));
 +
 +    ff_h264_draw_horiz_band(h, 0, h->avctx->height);
 +    render->bitstream_buffers_used = 0;
 +}
 +#endif /* CONFIG_H264_VDPAU_DECODER */
 +
 +#if CONFIG_MPEG_VDPAU_DECODER || CONFIG_MPEG1_VDPAU_DECODER
 +void ff_vdpau_mpeg_picture_complete(MpegEncContext *s, const uint8_t *buf,
 +                                    int buf_size, int slice_count)
 +{
 +    struct vdpau_render_state *render, *last, *next;
 +    int i;
 +
 +    if (!s->current_picture_ptr) return;
 +
-         next = (struct vdpau_render_state *)s->next_picture.f.data[0];
++    render = (struct vdpau_render_state *)s->current_picture_ptr->f->data[0];
 +    assert(render);
 +
 +    /* fill VdpPictureInfoMPEG1Or2 struct */
 +    render->info.mpeg.picture_structure          = s->picture_structure;
 +    render->info.mpeg.picture_coding_type        = s->pict_type;
 +    render->info.mpeg.intra_dc_precision         = s->intra_dc_precision;
 +    render->info.mpeg.frame_pred_frame_dct       = s->frame_pred_frame_dct;
 +    render->info.mpeg.concealment_motion_vectors = s->concealment_motion_vectors;
 +    render->info.mpeg.intra_vlc_format           = s->intra_vlc_format;
 +    render->info.mpeg.alternate_scan             = s->alternate_scan;
 +    render->info.mpeg.q_scale_type               = s->q_scale_type;
 +    render->info.mpeg.top_field_first            = s->top_field_first;
 +    render->info.mpeg.full_pel_forward_vector    = s->full_pel[0]; // MPEG-1 only.  Set 0 for MPEG-2
 +    render->info.mpeg.full_pel_backward_vector   = s->full_pel[1]; // MPEG-1 only.  Set 0 for MPEG-2
 +    render->info.mpeg.f_code[0][0]               = s->mpeg_f_code[0][0]; // For MPEG-1 fill both horiz. & vert.
 +    render->info.mpeg.f_code[0][1]               = s->mpeg_f_code[0][1];
 +    render->info.mpeg.f_code[1][0]               = s->mpeg_f_code[1][0];
 +    render->info.mpeg.f_code[1][1]               = s->mpeg_f_code[1][1];
 +    for (i = 0; i < 64; ++i) {
 +        render->info.mpeg.intra_quantizer_matrix[i]     = s->intra_matrix[i];
 +        render->info.mpeg.non_intra_quantizer_matrix[i] = s->inter_matrix[i];
 +    }
 +
 +    render->info.mpeg.forward_reference          = VDP_INVALID_HANDLE;
 +    render->info.mpeg.backward_reference         = VDP_INVALID_HANDLE;
 +
 +    switch(s->pict_type){
 +    case  AV_PICTURE_TYPE_B:
-         last = (struct vdpau_render_state *)s->last_picture.f.data[0];
++        next = (struct vdpau_render_state *)s->next_picture.f->data[0];
 +        assert(next);
 +        render->info.mpeg.backward_reference     = next->surface;
 +        // no return here, going to set forward prediction
 +    case  AV_PICTURE_TYPE_P:
-     ff_vdpau_add_data_chunk(s->current_picture_ptr->f.data[0], buf, buf_size);
++        last = (struct vdpau_render_state *)s->last_picture.f->data[0];
 +        if (!last) // FIXME: Does this test make sense?
 +            last = render; // predict second field from the first
 +        render->info.mpeg.forward_reference      = last->surface;
 +    }
 +
-     render = (struct vdpau_render_state *)s->current_picture.f.data[0];
++    ff_vdpau_add_data_chunk(s->current_picture_ptr->f->data[0], buf, buf_size);
 +
 +    render->info.mpeg.slice_count                = slice_count;
 +
 +    if (slice_count)
 +        ff_mpeg_draw_horiz_band(s, 0, s->avctx->height);
 +    render->bitstream_buffers_used               = 0;
 +}
 +#endif /* CONFIG_MPEG_VDPAU_DECODER || CONFIG_MPEG1_VDPAU_DECODER */
 +
 +#if CONFIG_VC1_VDPAU_DECODER
 +void ff_vdpau_vc1_decode_picture(MpegEncContext *s, const uint8_t *buf,
 +                                 int buf_size)
 +{
 +    VC1Context *v = s->avctx->priv_data;
 +    struct vdpau_render_state *render, *last, *next;
 +
-         next = (struct vdpau_render_state *)s->next_picture.f.data[0];
++    render = (struct vdpau_render_state *)s->current_picture.f->data[0];
 +    assert(render);
 +
 +    /*  fill LvPictureInfoVC1 struct */
 +    render->info.vc1.frame_coding_mode  = v->fcm ? v->fcm + 1 : 0;
 +    render->info.vc1.postprocflag       = v->postprocflag;
 +    render->info.vc1.pulldown           = v->broadcast;
 +    render->info.vc1.interlace          = v->interlace;
 +    render->info.vc1.tfcntrflag         = v->tfcntrflag;
 +    render->info.vc1.finterpflag        = v->finterpflag;
 +    render->info.vc1.psf                = v->psf;
 +    render->info.vc1.dquant             = v->dquant;
 +    render->info.vc1.panscan_flag       = v->panscanflag;
 +    render->info.vc1.refdist_flag       = v->refdist_flag;
 +    render->info.vc1.quantizer          = v->quantizer_mode;
 +    render->info.vc1.extended_mv        = v->extended_mv;
 +    render->info.vc1.extended_dmv       = v->extended_dmv;
 +    render->info.vc1.overlap            = v->overlap;
 +    render->info.vc1.vstransform        = v->vstransform;
 +    render->info.vc1.loopfilter         = v->s.loop_filter;
 +    render->info.vc1.fastuvmc           = v->fastuvmc;
 +    render->info.vc1.range_mapy_flag    = v->range_mapy_flag;
 +    render->info.vc1.range_mapy         = v->range_mapy;
 +    render->info.vc1.range_mapuv_flag   = v->range_mapuv_flag;
 +    render->info.vc1.range_mapuv        = v->range_mapuv;
 +    /* Specific to simple/main profile only */
 +    render->info.vc1.multires           = v->multires;
 +    render->info.vc1.syncmarker         = v->resync_marker;
 +    render->info.vc1.rangered           = v->rangered | (v->rangeredfrm << 1);
 +    render->info.vc1.maxbframes         = v->s.max_b_frames;
 +
 +    render->info.vc1.deblockEnable      = v->postprocflag & 1;
 +    render->info.vc1.pquant             = v->pq;
 +
 +    render->info.vc1.forward_reference  = VDP_INVALID_HANDLE;
 +    render->info.vc1.backward_reference = VDP_INVALID_HANDLE;
 +
 +    if (v->bi_type)
 +        render->info.vc1.picture_type = 4;
 +    else
 +        render->info.vc1.picture_type = s->pict_type - 1 + s->pict_type / 3;
 +
 +    switch(s->pict_type){
 +    case  AV_PICTURE_TYPE_B:
-         last = (struct vdpau_render_state *)s->last_picture.f.data[0];
++        next = (struct vdpau_render_state *)s->next_picture.f->data[0];
 +        assert(next);
 +        render->info.vc1.backward_reference = next->surface;
 +        // no break here, going to set forward prediction
 +    case  AV_PICTURE_TYPE_P:
-     ff_vdpau_add_data_chunk(s->current_picture_ptr->f.data[0], buf, buf_size);
++        last = (struct vdpau_render_state *)s->last_picture.f->data[0];
 +        if (!last) // FIXME: Does this test make sense?
 +            last = render; // predict second field from the first
 +        render->info.vc1.forward_reference = last->surface;
 +    }
 +
-     render = (struct vdpau_render_state *)s->current_picture_ptr->f.data[0];
++    ff_vdpau_add_data_chunk(s->current_picture_ptr->f->data[0], buf, buf_size);
 +
 +    render->info.vc1.slice_count          = 1;
 +
 +    ff_mpeg_draw_horiz_band(s, 0, s->avctx->height);
 +    render->bitstream_buffers_used        = 0;
 +}
 +#endif /* (CONFIG_VC1_VDPAU_DECODER */
 +
 +#if CONFIG_MPEG4_VDPAU_DECODER
 +void ff_vdpau_mpeg4_decode_picture(Mpeg4DecContext *ctx, const uint8_t *buf,
 +                                   int buf_size)
 +{
 +    MpegEncContext *s = &ctx->m;
 +    struct vdpau_render_state *render, *last, *next;
 +    int i;
 +
 +    if (!s->current_picture_ptr) return;
 +
-         next = (struct vdpau_render_state *)s->next_picture.f.data[0];
++    render = (struct vdpau_render_state *)s->current_picture_ptr->f->data[0];
 +    assert(render);
 +
 +    /* fill VdpPictureInfoMPEG4Part2 struct */
 +    render->info.mpeg4.trd[0]                            = s->pp_time;
 +    render->info.mpeg4.trb[0]                            = s->pb_time;
 +    render->info.mpeg4.trd[1]                            = s->pp_field_time >> 1;
 +    render->info.mpeg4.trb[1]                            = s->pb_field_time >> 1;
 +    render->info.mpeg4.vop_time_increment_resolution     = s->avctx->time_base.den;
 +    render->info.mpeg4.vop_coding_type                   = 0;
 +    render->info.mpeg4.vop_fcode_forward                 = s->f_code;
 +    render->info.mpeg4.vop_fcode_backward                = s->b_code;
 +    render->info.mpeg4.resync_marker_disable             = !ctx->resync_marker;
 +    render->info.mpeg4.interlaced                        = !s->progressive_sequence;
 +    render->info.mpeg4.quant_type                        = s->mpeg_quant;
 +    render->info.mpeg4.quarter_sample                    = s->quarter_sample;
 +    render->info.mpeg4.short_video_header                = s->avctx->codec->id == AV_CODEC_ID_H263;
 +    render->info.mpeg4.rounding_control                  = s->no_rounding;
 +    render->info.mpeg4.alternate_vertical_scan_flag      = s->alternate_scan;
 +    render->info.mpeg4.top_field_first                   = s->top_field_first;
 +    for (i = 0; i < 64; ++i) {
 +        render->info.mpeg4.intra_quantizer_matrix[i]     = s->intra_matrix[i];
 +        render->info.mpeg4.non_intra_quantizer_matrix[i] = s->inter_matrix[i];
 +    }
 +    render->info.mpeg4.forward_reference                 = VDP_INVALID_HANDLE;
 +    render->info.mpeg4.backward_reference                = VDP_INVALID_HANDLE;
 +
 +    switch (s->pict_type) {
 +    case AV_PICTURE_TYPE_B:
-         last = (struct vdpau_render_state *)s->last_picture.f.data[0];
++        next = (struct vdpau_render_state *)s->next_picture.f->data[0];
 +        assert(next);
 +        render->info.mpeg4.backward_reference     = next->surface;
 +        render->info.mpeg4.vop_coding_type        = 2;
 +        // no break here, going to set forward prediction
 +    case AV_PICTURE_TYPE_P:
-     ff_vdpau_add_data_chunk(s->current_picture_ptr->f.data[0], buf, buf_size);
++        last = (struct vdpau_render_state *)s->last_picture.f->data[0];
 +        assert(last);
 +        render->info.mpeg4.forward_reference      = last->surface;
 +    }
 +
++    ff_vdpau_add_data_chunk(s->current_picture_ptr->f->data[0], buf, buf_size);
 +
 +    ff_mpeg_draw_horiz_band(s, 0, s->avctx->height);
 +    render->bitstream_buffers_used = 0;
 +}
 +#endif /* CONFIG_MPEG4_VDPAU_DECODER */
 +
  int av_vdpau_get_profile(AVCodecContext *avctx, VdpDecoderProfile *profile)
  {
  #define PROFILE(prof)       \
Simple merge
Simple merge
@@@ -44,18 -44,14 +44,18 @@@ static int vdpau_vc1_start_frame(AVCode
  
      switch (s->pict_type) {
      case AV_PICTURE_TYPE_B:
-         ref = ff_vdpau_get_surface_id(&s->next_picture.f);
 +        if (s->next_picture_ptr) {
+         ref = ff_vdpau_get_surface_id(s->next_picture.f);
          assert(ref != VDP_INVALID_HANDLE);
          info->backward_reference = ref;
 +        }
          /* fall-through */
      case AV_PICTURE_TYPE_P:
-         ref = ff_vdpau_get_surface_id(&s->last_picture.f);
 +        if (s->last_picture_ptr) {
+         ref = ff_vdpau_get_surface_id(s->last_picture.f);
          assert(ref != VDP_INVALID_HANDLE);
          info->forward_reference  = ref;
 +        }
      }
  
      info->slice_count       = 0;