Merge remote-tracking branch 'qatar/master'
authorMichael Niedermayer <michaelni@gmx.at>
Thu, 20 Dec 2012 01:51:35 +0000 (02:51 +0100)
committerMichael Niedermayer <michaelni@gmx.at>
Thu, 20 Dec 2012 01:51:35 +0000 (02:51 +0100)
* qatar/master:
  mpegvideo: increase edge_emu_buffer size for VC1
  lavc: merge latest x86inc.asm fixes with x264

Conflicts:
libavcodec/mpegvideo.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>
1  2 
libavcodec/mpegvideo.c
libavutil/x86/x86inc.asm

diff --combined libavcodec/mpegvideo.c
@@@ -5,20 -5,20 +5,20 @@@
   *
   * 4MV & hq & B-frame encoding stuff by Michael Niedermayer <michaelni@gmx.at>
   *
 - * This file is part of Libav.
 + * This file is part of FFmpeg.
   *
 - * Libav is free software; you can redistribute it and/or
 + * 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.
   *
 - * Libav is distributed in the hope that it will be useful,
 + * 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 Libav; if not, write to the Free Software
 + * License along with FFmpeg; if not, write to the Free Software
   * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
   */
  
@@@ -58,6 -58,9 +58,6 @@@ static void dct_unquantize_h263_inter_c
                                    DCTELEM *block, int n, int qscale);
  
  
 -/* enable all paranoid tests for rounding, overflows, etc... */
 -//#define PARANOID
 -
  //#define DEBUG
  
  
@@@ -135,9 -138,9 +135,9 @@@ const enum AVPixelFormat ff_hwaccel_pix
      AV_PIX_FMT_NONE
  };
  
 -const uint8_t *avpriv_mpv_find_start_code(const uint8_t *restrict p,
 +const uint8_t *avpriv_mpv_find_start_code(const uint8_t *av_restrict p,
                                            const uint8_t *end,
 -                                          uint32_t * restrict state)
 +                                          uint32_t *av_restrict state)
  {
      int i;
  
@@@ -234,12 -237,14 +234,14 @@@ static void free_frame_buffer(MpegEncCo
  
  int ff_mpv_frame_size_alloc(MpegEncContext *s, int linesize)
  {
 -    int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
 +    int alloc_size = FFALIGN(FFABS(linesize) + 64, 32);
  
      // edge emu needs blocksize + filter length - 1
      // (= 17x17 for  halfpel / 21x21 for  h264)
+     // VC1 computes luma and chroma simultaneously and needs 19X19 + 9x9
+     // at uvlinesize. It supports only YUV420 so 24x24 is enough
      // linesize * interlaced * MBsize
-     FF_ALLOCZ_OR_GOTO(s->avctx, s->edge_emu_buffer, alloc_size * 4 * 21,
 -    FF_ALLOCZ_OR_GOTO(s->avctx, s->edge_emu_buffer, alloc_size * 2 * 24,
++    FF_ALLOCZ_OR_GOTO(s->avctx, s->edge_emu_buffer, alloc_size * 4 * 24,
                        fail);
  
      FF_ALLOCZ_OR_GOTO(s->avctx, s->me.scratchpad, alloc_size * 2 * 16 * 2,
@@@ -559,32 -564,27 +561,32 @@@ int ff_mpeg_update_thread_context(AVCod
                                    const AVCodecContext *src)
  {
      int i;
 +    int err;
      MpegEncContext *s = dst->priv_data, *s1 = src->priv_data;
  
 -    if (dst == src || !s1->context_initialized)
 +    if (dst == src)
          return 0;
  
 +    av_assert0(s != s1);
 +
      // FIXME can parameters change on I-frames?
      // in that case dst may need a reinit
      if (!s->context_initialized) {
          memcpy(s, s1, sizeof(MpegEncContext));
  
          s->avctx                 = dst;
 -        s->picture_range_start  += MAX_PICTURE_COUNT;
 -        s->picture_range_end    += MAX_PICTURE_COUNT;
          s->bitstream_buffer      = NULL;
          s->bitstream_buffer_size = s->allocated_bitstream_buffer_size = 0;
  
 -        ff_MPV_common_init(s);
 +        if (s1->context_initialized){
 +            s->picture_range_start  += MAX_PICTURE_COUNT;
 +            s->picture_range_end    += MAX_PICTURE_COUNT;
 +            if((err = ff_MPV_common_init(s)) < 0)
 +                return err;
 +        }
      }
  
      if (s->height != s1->height || s->width != s1->width || s->context_reinit) {
 -        int err;
          s->context_reinit = 0;
          s->height = s1->height;
          s->width  = s1->width;
      s->picture_number       = s1->picture_number;
      s->input_picture_number = s1->input_picture_number;
  
 +    av_assert0(!s->picture || s->picture != s1->picture);
      memcpy(s->picture, s1->picture, s1->picture_count * sizeof(Picture));
      memcpy(&s->last_picture, &s1->last_picture,
             (char *) &s1->last_picture_ptr - (char *) &s1->last_picture);
      // Error/bug resilience
      s->next_p_frame_damaged = s1->next_p_frame_damaged;
      s->workaround_bugs      = s1->workaround_bugs;
 +    s->padding_bug_score    = s1->padding_bug_score;
  
      // MPEG4 timing info
      memcpy(&s->time_increment_bits, &s1->time_increment_bits,
          } else {
              av_log(s->avctx, AV_LOG_ERROR, "Context scratch buffers could not "
                     "be allocated due to unknown size.\n");
 -            return AVERROR_BUG;
          }
  
      // MPEG2/interlacing info
@@@ -746,32 -745,44 +748,32 @@@ static int init_context_frame(MpegEncCo
      c_size  = s->mb_stride * (s->mb_height + 1);
      yc_size = y_size + 2   * c_size;
  
 -    FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num + 1) * sizeof(int),
 -                      fail); // error ressilience code looks cleaner with this
 +    FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_index2xy, (s->mb_num + 1) * sizeof(int), fail); // error ressilience code looks cleaner with this
      for (y = 0; y < s->mb_height; y++)
          for (x = 0; x < s->mb_width; x++)
              s->mb_index2xy[x + y * s->mb_width] = x + y * s->mb_stride;
  
 -    s->mb_index2xy[s->mb_height * s->mb_width] =
 -        (s->mb_height - 1) * s->mb_stride + s->mb_width; // FIXME really needed?
 +    s->mb_index2xy[s->mb_height * s->mb_width] = (s->mb_height - 1) * s->mb_stride + s->mb_width; // FIXME really needed?
  
      if (s->encoding) {
          /* Allocate MV tables */
 -        FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base,
 -                          mv_table_size * 2 * sizeof(int16_t), fail);
 -        FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base,
 -                          mv_table_size * 2 * sizeof(int16_t), fail);
 -        FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base,
 -                          mv_table_size * 2 * sizeof(int16_t), fail);
 -        FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base,
 -                          mv_table_size * 2 * sizeof(int16_t), fail);
 -        FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base,
 -                          mv_table_size * 2 * sizeof(int16_t), fail);
 -        FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base,
 -                          mv_table_size * 2 * sizeof(int16_t), fail);
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->p_mv_table_base,                 mv_table_size * 2 * sizeof(int16_t), fail)
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->b_forw_mv_table_base,            mv_table_size * 2 * sizeof(int16_t), fail)
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->b_back_mv_table_base,            mv_table_size * 2 * sizeof(int16_t), fail)
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_forw_mv_table_base,      mv_table_size * 2 * sizeof(int16_t), fail)
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->b_bidir_back_mv_table_base,      mv_table_size * 2 * sizeof(int16_t), fail)
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->b_direct_mv_table_base,          mv_table_size * 2 * sizeof(int16_t), fail)
          s->p_mv_table            = s->p_mv_table_base + s->mb_stride + 1;
          s->b_forw_mv_table       = s->b_forw_mv_table_base + s->mb_stride + 1;
          s->b_back_mv_table       = s->b_back_mv_table_base + s->mb_stride + 1;
 -        s->b_bidir_forw_mv_table = s->b_bidir_forw_mv_table_base +
 -                                   s->mb_stride + 1;
 -        s->b_bidir_back_mv_table = s->b_bidir_back_mv_table_base +
 -                                   s->mb_stride + 1;
 +        s->b_bidir_forw_mv_table = s->b_bidir_forw_mv_table_base + s->mb_stride + 1;
 +        s->b_bidir_back_mv_table = s->b_bidir_back_mv_table_base + s->mb_stride + 1;
          s->b_direct_mv_table     = s->b_direct_mv_table_base + s->mb_stride + 1;
  
          /* Allocate MB type table */
 -        FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type, mb_array_size *
 -                          sizeof(uint16_t), fail); // needed for encoding
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->mb_type, mb_array_size * sizeof(uint16_t), fail) // needed for encoding
  
 -        FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size *
 -                          sizeof(int), fail);
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->lambda_table, mb_array_size * sizeof(int), fail)
  
          FF_ALLOC_OR_GOTO(s->avctx, s->cplx_tab,
                           mb_array_size * sizeof(float), fail);
      FF_ALLOCZ_OR_GOTO(s->avctx, s->error_status_table,
                        mb_array_size * sizeof(uint8_t), fail);
  
 -    if (s->codec_id == AV_CODEC_ID_MPEG4 ||
 -        (s->flags & CODEC_FLAG_INTERLACED_ME)) {
 +    if (s->codec_id == AV_CODEC_ID_MPEG4 || (s->flags & CODEC_FLAG_INTERLACED_ME)) {
          /* interlaced direct mode decoding tables */
          for (i = 0; i < 2; i++) {
              int j, k;
                      s->b_field_mv_table[i][j][k] = s->b_field_mv_table_base[i][j][k] +
                                                     s->mb_stride + 1;
                  }
 -                FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j],
 -                                  mb_array_size * 2 * sizeof(uint8_t), fail);
 -                FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j],
 -                                  mv_table_size * 2 * sizeof(int16_t), fail);
 -                s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j]
 -                                            + s->mb_stride + 1;
 +                FF_ALLOCZ_OR_GOTO(s->avctx, s->b_field_select_table [i][j], mb_array_size * 2 * sizeof(uint8_t), fail)
 +                FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_mv_table_base[i][j], mv_table_size * 2 * sizeof(int16_t), fail)
 +                s->p_field_mv_table[i][j] = s->p_field_mv_table_base[i][j] + s->mb_stride + 1;
              }
 -            FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i],
 -                              mb_array_size * 2 * sizeof(uint8_t), fail);
 +            FF_ALLOCZ_OR_GOTO(s->avctx, s->p_field_select_table[i], mb_array_size * 2 * sizeof(uint8_t), fail)
          }
      }
      if (s->out_format == FMT_H263) {
          s->coded_block = s->coded_block_base + s->b8_stride + 1;
  
          /* cbp, ac_pred, pred_dir */
 -        FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table,
 -                          mb_array_size * sizeof(uint8_t), fail);
 -        FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table,
 -                          mb_array_size * sizeof(uint8_t), fail);
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->cbp_table     , mb_array_size * sizeof(uint8_t), fail);
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->pred_dir_table, mb_array_size * sizeof(uint8_t), fail);
      }
  
      if (s->h263_pred || s->h263_plus || !s->encoding) {
          /* dc values */
          // MN: we need these for  error resilience of intra-frames
 -        FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base,
 -                          yc_size * sizeof(int16_t), fail);
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->dc_val_base, yc_size * sizeof(int16_t), fail);
          s->dc_val[0] = s->dc_val_base + s->b8_stride + 1;
          s->dc_val[1] = s->dc_val_base + y_size + s->mb_stride + 1;
          s->dc_val[2] = s->dc_val[1] + c_size;
      FF_ALLOCZ_OR_GOTO(s->avctx, s->mbskip_table, mb_array_size + 2, fail);
      // Note the + 1 is for  a quicker mpeg4 slice_end detection
  
 -    if ((s->avctx->debug & (FF_DEBUG_VIS_QP | FF_DEBUG_VIS_MB_TYPE)) ||
 -        s->avctx->debug_mv) {
 -        s->visualization_buffer[0] = av_malloc((s->mb_width * 16 +
 -                    2 * EDGE_WIDTH) * s->mb_height * 16 + 2 * EDGE_WIDTH);
 -        s->visualization_buffer[1] = av_malloc((s->mb_width * 16 +
 -                    2 * EDGE_WIDTH) * s->mb_height * 16 + 2 * EDGE_WIDTH);
 -        s->visualization_buffer[2] = av_malloc((s->mb_width * 16 +
 -                    2 * EDGE_WIDTH) * s->mb_height * 16 + 2 * EDGE_WIDTH);
 -    }
 -
      return 0;
  fail:
      return AVERROR(ENOMEM);
@@@ -884,34 -913,44 +886,34 @@@ av_cold int ff_MPV_common_init(MpegEncC
      s->flags  = s->avctx->flags;
      s->flags2 = s->avctx->flags2;
  
 -    if (s->width && s->height) {
 -        /* set chroma shifts */
 -        av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
 -                                         &s->chroma_x_shift,
 -                                         &s->chroma_y_shift);
 +    /* set chroma shifts */
 +    avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);
 +
 +    /* convert fourcc to upper case */
 +    s->codec_tag        = avpriv_toupper4(s->avctx->codec_tag);
 +    s->stream_codec_tag = avpriv_toupper4(s->avctx->stream_codec_tag);
  
 -        /* convert fourcc to upper case */
 -        s->codec_tag          = avpriv_toupper4(s->avctx->codec_tag);
 +    s->avctx->coded_frame = &s->current_picture.f;
  
 -        s->stream_codec_tag   = avpriv_toupper4(s->avctx->stream_codec_tag);
 +    if (s->encoding) {
 +        if (s->msmpeg4_version) {
 +            FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_stats,
 +                                2 * 2 * (MAX_LEVEL + 1) *
 +                                (MAX_RUN + 1) * 2 * sizeof(int), fail);
 +        }
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->avctx->stats_out, 256, fail);
  
 -        s->avctx->coded_frame = &s->current_picture.f;
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix,          64 * 32   * sizeof(int), fail)
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->q_chroma_intra_matrix,   64 * 32   * sizeof(int), fail)
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix,          64 * 32   * sizeof(int), fail)
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix16,        64 * 32 * 2 * sizeof(uint16_t), fail)
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->q_chroma_intra_matrix16, 64 * 32 * 2 * sizeof(uint16_t), fail)
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix16,        64 * 32 * 2 * sizeof(uint16_t), fail)
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->input_picture,           MAX_PICTURE_COUNT * sizeof(Picture *), fail)
 +        FF_ALLOCZ_OR_GOTO(s->avctx, s->reordered_input_picture, MAX_PICTURE_COUNT * sizeof(Picture *), fail)
  
 -        if (s->encoding) {
 -            if (s->msmpeg4_version) {
 -                FF_ALLOCZ_OR_GOTO(s->avctx, s->ac_stats,
 -                                  2 * 2 * (MAX_LEVEL + 1) *
 -                                  (MAX_RUN + 1) * 2 * sizeof(int), fail);
 -            }
 -            FF_ALLOCZ_OR_GOTO(s->avctx, s->avctx->stats_out, 256, fail);
 -
 -            FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix,
 -                              64 * 32   * sizeof(int), fail);
 -            FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix,
 -                              64 * 32   * sizeof(int), fail);
 -            FF_ALLOCZ_OR_GOTO(s->avctx, s->q_intra_matrix16,
 -                              64 * 32 * 2 * sizeof(uint16_t), fail);
 -            FF_ALLOCZ_OR_GOTO(s->avctx, s->q_inter_matrix16,
 -                              64 * 32 * 2 * sizeof(uint16_t), fail);
 -            FF_ALLOCZ_OR_GOTO(s->avctx, s->input_picture,
 -                              MAX_PICTURE_COUNT * sizeof(Picture *), fail);
 -            FF_ALLOCZ_OR_GOTO(s->avctx, s->reordered_input_picture,
 -                              MAX_PICTURE_COUNT * sizeof(Picture *), fail);
 -
 -            if (s->avctx->noise_reduction) {
 -                FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_offset,
 -                                  2 * 64 * sizeof(uint16_t), fail);
 -            }
 +        if (s->avctx->noise_reduction) {
 +            FF_ALLOCZ_OR_GOTO(s->avctx, s->dct_offset, 2 * 64 * sizeof(uint16_t), fail);
          }
      }
  
          avcodec_get_frame_defaults(&s->picture[i].f);
      }
  
 -    if (s->width && s->height) {
          if (init_context_frame(s))
              goto fail;
  
          s->parse_context.state = -1;
 -    }
  
 -    s->context_initialized = 1;
 -    s->thread_context[0]   = s;
 +        s->context_initialized = 1;
 +        s->thread_context[0]   = s;
  
 -    if (s->width && s->height) {
 +//     if (s->width && s->height) {
          if (nb_slices > 1) {
              for (i = 1; i < nb_slices; i++) {
                  s->thread_context[i] = av_malloc(sizeof(MpegEncContext));
              s->end_mb_y   = s->mb_height;
          }
          s->slice_context_count = nb_slices;
 -    }
 +//     }
  
      return 0;
   fail:
@@@ -1007,7 -1048,6 +1009,7 @@@ static int free_context_frame(MpegEncCo
      av_freep(&s->er_temp_buffer);
      av_freep(&s->mb_index2xy);
      av_freep(&s->lambda_table);
 +
      av_freep(&s->cplx_tab);
      av_freep(&s->bits_tab);
  
@@@ -1114,10 -1154,6 +1116,10 @@@ void ff_MPV_common_end(MpegEncContext *
      av_freep(&s->avctx->stats_out);
      av_freep(&s->ac_stats);
  
 +    if(s->q_chroma_intra_matrix   != s->q_intra_matrix  ) av_freep(&s->q_chroma_intra_matrix);
 +    if(s->q_chroma_intra_matrix16 != s->q_intra_matrix16) av_freep(&s->q_chroma_intra_matrix16);
 +    s->q_chroma_intra_matrix=   NULL;
 +    s->q_chroma_intra_matrix16= NULL;
      av_freep(&s->q_intra_matrix);
      av_freep(&s->q_inter_matrix);
      av_freep(&s->q_intra_matrix16);
@@@ -1282,21 -1318,7 +1284,21 @@@ static int find_unused_picture(MpegEncC
          }
      }
  
 -    return AVERROR_INVALIDDATA;
 +    av_log(s->avctx, AV_LOG_FATAL,
 +           "Internal error, picture buffer overflow\n");
 +    /* We could return -1, but the codec would crash trying to draw into a
 +     * non-existing frame anyway. This is safer than waiting for a random crash.
 +     * Also the return of this is never useful, an encoder must only allocate
 +     * as much as allowed in the specification. This has no relationship to how
 +     * much libavcodec could allocate (and MAX_PICTURE_COUNT is always large
 +     * enough for such valid streams).
 +     * Plus, a decoder has to check stream validity and remove frames if too
 +     * many reference frames are around. Waiting for "OOM" is not correct at
 +     * all. Similarly, missing reference frames have to be replaced by
 +     * interpolated/MC frames, anything else is a bug in the codec ...
 +     */
 +    abort();
 +    return -1;
  }
  
  int ff_find_unused_picture(MpegEncContext *s, int shared)
@@@ -1344,11 -1366,6 +1346,11 @@@ int ff_MPV_frame_start(MpegEncContext *
      Picture *pic;
      s->mb_skipped = 0;
  
 +    if (!ff_thread_can_start_frame(avctx)) {
 +        av_log(avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
 +        return -1;
 +    }
 +
      /* mark & release old frames */
      if (s->out_format != FMT_H264 || s->codec_id == AV_CODEC_ID_SVQ3) {
          if (s->pict_type != AV_PICTURE_TYPE_B && s->last_picture_ptr &&
                  return i;
              }
              s->last_picture_ptr = &s->picture[i];
 +            s->last_picture_ptr->f.key_frame = 0;
              if (ff_alloc_picture(s, s->last_picture_ptr, 0) < 0) {
                  s->last_picture_ptr = NULL;
                  return -1;
              }
 +
 +            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);
 +            }
 +
              ff_thread_report_progress(&s->last_picture_ptr->f, INT_MAX, 0);
              ff_thread_report_progress(&s->last_picture_ptr->f, INT_MAX, 1);
              s->last_picture_ptr->f.reference = 3;
                  return i;
              }
              s->next_picture_ptr = &s->picture[i];
 +            s->next_picture_ptr->f.key_frame = 0;
              if (ff_alloc_picture(s, s->next_picture_ptr, 0) < 0) {
                  s->next_picture_ptr = NULL;
                  return -1;
@@@ -1557,30 -1566,28 +1559,30 @@@ void ff_MPV_frame_end(MpegEncContext *s
      // just to make sure that all data is rendered.
      if (CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration) {
          ff_xvmc_field_end(s);
 -   } else if ((s->error_count || s->encoding) &&
 +   } else if((s->error_count || s->encoding || !(s->avctx->codec->capabilities&CODEC_CAP_DRAW_HORIZ_BAND)) &&
                !s->avctx->hwaccel &&
                !(s->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU) &&
                s->unrestricted_mv &&
                s->current_picture.f.reference &&
                !s->intra_only &&
 -              !(s->flags & CODEC_FLAG_EMU_EDGE)) {
 -       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->linesize,
 -                         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->uvlinesize,
 -                         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->uvlinesize,
 -                         s->h_edge_pos >> hshift, s->v_edge_pos >> vshift,
 -                         EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
 -                         EDGE_TOP | EDGE_BOTTOM);
 +              !(s->flags & CODEC_FLAG_EMU_EDGE) &&
 +              !s->avctx->lowres
 +            ) {
 +        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->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->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->h_edge_pos >> hshift, s->v_edge_pos >> vshift,
 +                          EDGE_WIDTH >> hshift, EDGE_WIDTH >> vshift,
 +                          EDGE_TOP | EDGE_BOTTOM);
      }
  
      emms_c();
@@@ -1650,11 -1657,11 +1652,11 @@@ static void draw_line(uint8_t *buf, in
          buf += sx + sy * stride;
          ex  -= sx;
          f    = ((ey - sy) << 16) / ex;
 -        for (x = 0; x = ex; x++) {
 +        for(x= 0; x <= ex; x++){
              y  = (x * f) >> 16;
              fr = (x * f) & 0xFFFF;
              buf[y * stride + x]       += (color * (0x10000 - fr)) >> 16;
 -            buf[(y + 1) * stride + x] += (color *            fr ) >> 16;
 +            if(fr) buf[(y + 1) * stride + x] += (color *            fr ) >> 16;
          }
      } else {
          if (sy > ey) {
          buf += sx + sy * stride;
          ey  -= sy;
          if (ey)
 -            f  = ((ex - sx) << 16) / ey;
 +            f = ((ex - sx) << 16) / ey;
          else
              f = 0;
 -        for (y = 0; y = ey; y++) {
 -            x  = (y * f) >> 16;
 -            fr = (y * f) & 0xFFFF;
 +        for(y= 0; y <= ey; y++){
 +            x  = (y*f) >> 16;
 +            fr = (y*f) & 0xFFFF;
              buf[y * stride + x]     += (color * (0x10000 - fr)) >> 16;
 -            buf[y * stride + x + 1] += (color *            fr ) >> 16;
 +            if(fr) buf[y * stride + x + 1] += (color *            fr ) >> 16;
          }
      }
  }
@@@ -1716,16 -1723,33 +1718,16 @@@ static void draw_arrow(uint8_t *buf, in
   */
  void ff_print_debug_info(MpegEncContext *s, AVFrame *pict)
  {
 -    if (s->avctx->hwaccel || !pict || !pict->mb_type)
 +    if (   s->avctx->hwaccel || !pict || !pict->mb_type
 +        || (s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU))
          return;
  
 +
      if (s->avctx->debug & (FF_DEBUG_SKIP | FF_DEBUG_QP | FF_DEBUG_MB_TYPE)) {
          int x,y;
  
 -        av_log(s->avctx,AV_LOG_DEBUG,"New frame, type: ");
 -        switch (pict->pict_type) {
 -        case AV_PICTURE_TYPE_I:
 -            av_log(s->avctx,AV_LOG_DEBUG,"I\n");
 -            break;
 -        case AV_PICTURE_TYPE_P:
 -            av_log(s->avctx,AV_LOG_DEBUG,"P\n");
 -            break;
 -        case AV_PICTURE_TYPE_B:
 -            av_log(s->avctx,AV_LOG_DEBUG,"B\n");
 -            break;
 -        case AV_PICTURE_TYPE_S:
 -            av_log(s->avctx,AV_LOG_DEBUG,"S\n");
 -            break;
 -        case AV_PICTURE_TYPE_SI:
 -            av_log(s->avctx,AV_LOG_DEBUG,"SI\n");
 -            break;
 -        case AV_PICTURE_TYPE_SP:
 -            av_log(s->avctx,AV_LOG_DEBUG,"SP\n");
 -            break;
 -        }
 +        av_log(s->avctx, AV_LOG_DEBUG, "New frame, type: %c\n",
 +               av_get_picture_type_char(pict->pict_type));
          for (y = 0; y < s->mb_height; y++) {
              for (x = 0; x < s->mb_width; x++) {
                  if (s->avctx->debug & FF_DEBUG_SKIP) {
                      else if (!USES_LIST(mb_type, 0))
                          av_log(s->avctx, AV_LOG_DEBUG, "<");
                      else {
 -                        assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
 +                        av_assert2(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
                          av_log(s->avctx, AV_LOG_DEBUG, "X");
                      }
  
                                     (s->codec_id == AV_CODEC_ID_H264 ? 0 : 1);
          s->low_delay = 0; // needed to see the vectors without trashing the buffers
  
 -        av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
 -                                         &h_chroma_shift, &v_chroma_shift);
 +        avcodec_get_chroma_sub_sample(s->avctx->pix_fmt, &h_chroma_shift, &v_chroma_shift);
 +
          for (i = 0; i < 3; i++) {
 -            memcpy(s->visualization_buffer[i], pict->data[i],
 -                   (i == 0) ? pict->linesize[i] * height:
 -                              pict->linesize[i] * height >> v_chroma_shift);
 +            size_t size= (i == 0) ? pict->linesize[i] * FFALIGN(height, 16):
 +                         pict->linesize[i] * FFALIGN(height, 16) >> v_chroma_shift;
 +            s->visualization_buffer[i]= av_realloc(s->visualization_buffer[i], size);
 +            memcpy(s->visualization_buffer[i], pict->data[i], size);
              pict->data[i] = s->visualization_buffer[i];
          }
          pict->type   = FF_BUFFER_TYPE_COPY;
 +        pict->opaque= NULL;
          ptr          = pict->data[0];
          block_height = 16 >> v_chroma_shift;
  
              int mb_x;
              for (mb_x = 0; mb_x < s->mb_width; mb_x++) {
                  const int mb_index = mb_x + mb_y * s->mb_stride;
 -                if ((s->avctx->debug_mv) && pict->motion_val) {
 +                if ((s->avctx->debug_mv) && pict->motion_val[0]) {
                      int type;
                      for (type = 0; type < 3; type++) {
                          int direction = 0;
                                             height, s->linesize, 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 = pict->motion_val[direction][xy][0] >> shift + sx;
 -                              int my = pict->motion_val[direction][xy][1] >> shift + sy;
 +                              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= (pict->motion_val[direction][xy][0]>>shift) + sx;
 +                              int my= (pict->motion_val[direction][xy][1]>>shift) + sy;
                                draw_arrow(ptr, sx, sy, mx, my, width, height, s->linesize, 100);
                          }
                      }
                  }
 -                if ((s->avctx->debug & FF_DEBUG_VIS_QP) && pict->motion_val) {
 +                if ((s->avctx->debug & FF_DEBUG_VIS_QP)) {
                      uint64_t c = (pict->qscale_table[mb_index] * 128 / 31) *
                                   0x0101010101010101ULL;
                      int y;
                      }
                  }
                  if ((s->avctx->debug & FF_DEBUG_VIS_MB_TYPE) &&
 -                    pict->motion_val) {
 +                    pict->motion_val[0]) {
                      int mb_type = pict->mb_type[mb_index];
                      uint64_t u,v;
                      int y;
                      } else if (!USES_LIST(mb_type, 0)) {
                          COLOR(0, 48)
                      } else {
 -                        assert(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
 +                        av_assert2(USES_LIST(mb_type, 0) && USES_LIST(mb_type, 1));
                          COLOR(300,48)
                      }
  
      }
  }
  
 +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, int 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, 2);
 +    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->dsp.emulated_edge_mc(s->edge_emu_buffer, src, 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, uvlinesize, linesize, sx, sy,
 +        uvsx, uvsy;
 +    const int lowres     = s->avctx->lowres;
 +    const int op_index   = FFMIN(lowres-1+s->chroma_x_shift, 2);
 +    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;
 +
 +    // 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) ||
 +        (unsigned) src_y > FFMAX((v_edge_pos >> field_based) - (!!sy) - h, 0)) {
 +        s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
 +                                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->dsp.emulated_edge_mc(uvbuf , ptr_cb, uvlinesize >> field_based, 9,
 +                                    9 + field_based,
 +                                    uvsrc_x, uvsrc_y << field_based,
 +                                    h_edge_pos >> 1, v_edge_pos >> 1);
 +            s->dsp.emulated_edge_mc(uvbuf + 16, ptr_cr, 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
 +    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)) {
 +        uvsx = (uvsx << 2) >> lowres;
 +        uvsy = (uvsy << 2) >> lowres;
 +        if (h >> s->chroma_y_shift) {
 +            pix_op[op_index](dest_cb, ptr_cb, uvlinesize, h >> s->chroma_y_shift, uvsx, uvsy);
 +            pix_op[op_index](dest_cr, ptr_cr, uvlinesize, h >> s->chroma_y_shift, 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, 2);
 +    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, offset, sx, sy;
 +    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 (s->flags & CODEC_FLAG_EMU_EDGE) {
 +        if ((unsigned) src_x > FFMAX(h_edge_pos - (!!sx) - block_s, 0) ||
 +            (unsigned) src_y > FFMAX(v_edge_pos - (!!sy) - block_s, 0)) {
 +            s->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, 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->dsp.emulated_edge_mc(s->edge_emu_buffer, ptr, 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;
 +
 +            }
 +            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;
 +            }
 +
 +            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->dsp.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->dsp.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;
 +                }
 +            }
 +        }
 +        break;
 +    default:
 +        av_assert2(0);
 +    }
 +}
 +
  /**
   * find the lowest MB row referenced in the MVs
   */
@@@ -2488,7 -2135,7 +2490,7 @@@ void ff_clean_intra_table_entries(MpegE
   */
  static av_always_inline
  void MPV_decode_mb_internal(MpegEncContext *s, DCTELEM block[12][64],
 -                            int is_mpeg12)
 +                            int lowres_flag, int is_mpeg12)
  {
      const int mb_xy = s->mb_y * s->mb_stride + s->mb_x;
      if(CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration){
          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 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 */
  
              if (s->mb_skipped) {
                  s->mb_skipped= 0;
 -                assert(s->pict_type!=AV_PICTURE_TYPE_I);
 +                av_assert2(s->pict_type!=AV_PICTURE_TYPE_I);
                  *mbskip_ptr = 1;
              } else if(!s->current_picture.f.reference) {
                  *mbskip_ptr = 1;
                      }
                  }
  
 -                op_qpix= s->me.qpel_put;
 -                if ((!s->no_rounding) || s->pict_type==AV_PICTURE_TYPE_B){
 -                    op_pix = s->dsp.put_pixels_tab;
 +                if(lowres_flag){
 +                    h264_chroma_mc_func *op_pix = s->dsp.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);
 +                        op_pix = s->dsp.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);
 +                    }
                  }else{
 -                    op_pix = s->dsp.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->dsp.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->dsp.put_pixels_tab;
 +                    }else{
 +                        op_pix = s->dsp.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->dsp.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);
 +                    }
                  }
              }
  
                      }else{
                          //chroma422
                          dct_linesize = uvlinesize << s->interlaced_dct;
 -                        dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize * 8;
 +                        dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize*block_size;
  
                          add_dct(s, block[4], 4, dest_cb, dct_linesize);
                          add_dct(s, block[5], 5, dest_cr, dct_linesize);
                          add_dct(s, block[6], 6, dest_cb+dct_offset, dct_linesize);
                          add_dct(s, block[7], 7, dest_cr+dct_offset, dct_linesize);
                          if(!s->chroma_x_shift){//Chroma444
 -                            add_dct(s, block[8], 8, dest_cb+8, dct_linesize);
 -                            add_dct(s, block[9], 9, dest_cr+8, dct_linesize);
 -                            add_dct(s, block[10], 10, dest_cb+8+dct_offset, dct_linesize);
 -                            add_dct(s, block[11], 11, dest_cr+8+dct_offset, dct_linesize);
 +                            add_dct(s, block[8], 8, dest_cb+block_size, dct_linesize);
 +                            add_dct(s, block[9], 9, dest_cr+block_size, dct_linesize);
 +                            add_dct(s, block[10], 10, dest_cb+block_size+dct_offset, dct_linesize);
 +                            add_dct(s, block[11], 11, dest_cr+block_size+dct_offset, dct_linesize);
                          }
                      }
                  }//fi gray
                      }else{
  
                          dct_linesize = uvlinesize << s->interlaced_dct;
 -                        dct_offset   = s->interlaced_dct ? uvlinesize : uvlinesize * 8;
 +                        dct_offset   = s->interlaced_dct? uvlinesize : uvlinesize*block_size;
  
                          s->dsp.idct_put(dest_cb,              dct_linesize, block[4]);
                          s->dsp.idct_put(dest_cr,              dct_linesize, block[5]);
                          s->dsp.idct_put(dest_cb + dct_offset, dct_linesize, block[6]);
                          s->dsp.idct_put(dest_cr + dct_offset, dct_linesize, block[7]);
                          if(!s->chroma_x_shift){//Chroma444
 -                            s->dsp.idct_put(dest_cb + 8,              dct_linesize, block[8]);
 -                            s->dsp.idct_put(dest_cr + 8,              dct_linesize, block[9]);
 -                            s->dsp.idct_put(dest_cb + 8 + dct_offset, dct_linesize, block[10]);
 -                            s->dsp.idct_put(dest_cr + 8 + dct_offset, dct_linesize, block[11]);
 +                            s->dsp.idct_put(dest_cb + block_size,              dct_linesize, block[8]);
 +                            s->dsp.idct_put(dest_cr + block_size,              dct_linesize, block[9]);
 +                            s->dsp.idct_put(dest_cb + block_size + dct_offset, dct_linesize, block[10]);
 +                            s->dsp.idct_put(dest_cr + block_size + dct_offset, dct_linesize, block[11]);
                          }
                      }
                  }//gray
@@@ -2733,12 -2368,10 +2735,12 @@@ skip_idct
  void ff_MPV_decode_mb(MpegEncContext *s, DCTELEM block[12][64]){
  #if !CONFIG_SMALL
      if(s->out_format == FMT_MPEG1) {
 -        MPV_decode_mb_internal(s, block, 1);
 +        if(s->avctx->lowres) MPV_decode_mb_internal(s, block, 1, 1);
 +        else                 MPV_decode_mb_internal(s, block, 0, 1);
      } else
  #endif
 -        MPV_decode_mb_internal(s, block, 0);
 +    if(s->avctx->lowres) MPV_decode_mb_internal(s, block, 1, 0);
 +    else                  MPV_decode_mb_internal(s, block, 0, 0);
  }
  
  /**
@@@ -2814,7 -2447,7 +2816,7 @@@ void ff_draw_horiz_band(MpegEncContext 
  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 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;
              s->dest[0] += (s->mb_y>>1) *   linesize << mb_size;
              s->dest[1] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift);
              s->dest[2] += (s->mb_y>>1) * uvlinesize << (mb_size - s->chroma_y_shift);
 -            assert((s->mb_y&1) == (s->picture_structure == PICT_BOTTOM_FIELD));
 +            av_assert1((s->mb_y&1) == (s->picture_structure == PICT_BOTTOM_FIELD));
          }
      }
  }
@@@ -2859,7 -2492,6 +2861,7 @@@ void ff_mpeg_flush(AVCodecContext *avct
      s->current_picture_ptr = s->last_picture_ptr = s->next_picture_ptr = NULL;
  
      s->mb_x= s->mb_y= 0;
 +    s->closed_gop= 0;
  
      s->parse_context.state= -1;
      s->parse_context.frame_start_found= 0;
@@@ -2879,7 -2511,10 +2881,7 @@@ static void dct_unquantize_mpeg1_intra_
  
      nCoeffs= s->block_last_index[n];
  
 -    if (n < 4)
 -        block[0] = block[0] * s->y_dc_scale;
 -    else
 -        block[0] = block[0] * s->c_dc_scale;
 +    block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
      /* XXX: only mpeg1 */
      quant_matrix = s->intra_matrix;
      for(i=1;i<=nCoeffs;i++) {
@@@ -2938,7 -2573,10 +2940,7 @@@ static void dct_unquantize_mpeg2_intra_
      if(s->alternate_scan) nCoeffs= 63;
      else nCoeffs= s->block_last_index[n];
  
 -    if (n < 4)
 -        block[0] = block[0] * s->y_dc_scale;
 -    else
 -        block[0] = block[0] * s->c_dc_scale;
 +    block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
      quant_matrix = s->intra_matrix;
      for(i=1;i<=nCoeffs;i++) {
          int j= s->intra_scantable.permutated[i];
@@@ -2966,8 -2604,10 +2968,8 @@@ static void dct_unquantize_mpeg2_intra_
      if(s->alternate_scan) nCoeffs= 63;
      else nCoeffs= s->block_last_index[n];
  
 -    if (n < 4)
 -        block[0] = block[0] * s->y_dc_scale;
 -    else
 -        block[0] = block[0] * s->c_dc_scale;
 +    block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
 +    sum += block[0];
      quant_matrix = s->intra_matrix;
      for(i=1;i<=nCoeffs;i++) {
          int j= s->intra_scantable.permutated[i];
@@@ -3029,7 -2669,10 +3031,7 @@@ static void dct_unquantize_h263_intra_c
      qmul = qscale << 1;
  
      if (!s->h263_aic) {
 -        if (n < 4)
 -            block[0] = block[0] * s->y_dc_scale;
 -        else
 -            block[0] = block[0] * s->c_dc_scale;
 +        block[0] *= n < 4 ? s->y_dc_scale : s->c_dc_scale;
          qadd = (qscale - 1) | 1;
      }else{
          qadd = 0;
diff --combined libavutil/x86/x86inc.asm
  %endif
  
  ; Name of the .rodata section.
 -; Kludge: Something on OS X fails to align .rodata even given an align attribute,
 -; so use a different read-only section.
  %macro SECTION_RODATA 0-1 16
 -    %ifidn __OUTPUT_FORMAT__,macho64
 -        SECTION .text align=%1
 -    %elifidn __OUTPUT_FORMAT__,macho
 -        SECTION .text align=%1
 -        fakegot:
 -    %elifidn __OUTPUT_FORMAT__,aout
 +    ; Kludge: Something on OS X fails to align .rodata even given an align
 +    ; attribute, so use a different read-only section. This has been fixed in
 +    ; yasm 0.8.0 and nasm 2.6.
 +    %ifdef __YASM_VERSION_ID__
 +        %if __YASM_VERSION_ID__ < 00080000h
 +            %define NEED_MACHO_RODATA_KLUDGE
 +        %endif
 +    %elifdef __NASM_VERSION_ID__
 +        %if __NASM_VERSION_ID__ < 02060000h
 +            %define NEED_MACHO_RODATA_KLUDGE
 +        %endif
 +    %endif
 +
 +    %ifidn __OUTPUT_FORMAT__,aout
          section .text
      %else
 -        SECTION .rodata align=%1
 +        %ifndef NEED_MACHO_RODATA_KLUDGE
 +            SECTION .rodata align=%1
 +        %else
 +            %ifidn __OUTPUT_FORMAT__,macho64
 +                SECTION .text align=%1
 +            %elifidn __OUTPUT_FORMAT__,macho
 +                SECTION .text align=%1
 +                fakegot:
 +            %else
 +                SECTION .rodata align=%1
 +            %endif
 +        %endif
      %endif
 +
 +    %undef NEED_MACHO_RODATA_KLUDGE
  %endmacro
  
  ; aout does not support align=
@@@ -350,7 -331,9 +350,9 @@@ DECLARE_REG_TMP_SIZE 0,1,2,3,4,5,6,7,8,
              %if stack_size < 0
                  %assign stack_size -stack_size
              %endif
-             %assign xmm_regs_used %2
+             %if mmsize != 8
+                 %assign xmm_regs_used %2
+             %endif
              %if mmsize <= 16 && HAVE_ALIGNED_STACK
                  %assign stack_size_padded stack_size + %%stack_alignment - gprsize - (stack_offset & (%%stack_alignment - 1))
                  %if xmm_regs_used > 6
                  %endif
                  SUB rsp, stack_size_padded
              %else
-                 %assign reg_num (regs_used - 1)
-                 %xdefine rstk r %+ reg_num
+                 %assign %%reg_num (regs_used - 1)
+                 %xdefine rstk r %+ %%reg_num
                  ; align stack, and save original stack location directly above
                  ; it, i.e. in [rsp+stack_size_padded], so we can restore the
                  ; stack in a single instruction (i.e. mov rsp, rstk or mov
                  %assign stack_size_padded stack_size
                  %if xmm_regs_used > 6
                      %assign stack_size_padded stack_size_padded + (xmm_regs_used - 6) * 16
+                     %if mmsize == 32 && xmm_regs_used & 1
+                         ; re-align to 32 bytes
+                         %assign stack_size_padded (stack_size_padded + 16)
+                     %endif
                  %endif
                  %if %1 < 0 ; need to store rsp on stack
                      sub  rsp, gprsize+stack_size_padded
@@@ -430,11 -417,10 +436,10 @@@ DECLARE_REG 14, R15, 12
  %macro PROLOGUE 2-5+ 0 ; #args, #regs, #xmm_regs, [stack_size,] arg_names...
      %assign num_args %1
      %assign regs_used %2
-     SETUP_STACK_POINTER %4
      ASSERT regs_used >= num_args
+     SETUP_STACK_POINTER %4
      ASSERT regs_used <= 15
      PUSH_IF_USED 7, 8, 9, 10, 11, 12, 13, 14
-     %assign xmm_regs_used 0
      ALLOC_STACK %4, %3
      %if mmsize != 8 && stack_size == 0
          WIN64_SPILL_XMM %3
@@@ -518,8 -504,8 +523,8 @@@ DECLARE_REG 14, R15, 7
  %macro PROLOGUE 2-5+ ; #args, #regs, #xmm_regs, [stack_size,] arg_names...
      %assign num_args %1
      %assign regs_used %2
-     SETUP_STACK_POINTER %4
      ASSERT regs_used >= num_args
+     SETUP_STACK_POINTER %4
      ASSERT regs_used <= 15
      PUSH_IF_USED 9, 10, 11, 12, 13, 14
      ALLOC_STACK %4
@@@ -568,12 -554,15 +573,15 @@@ DECLARE_ARG 7, 8, 9, 10, 11, 12, 13, 1
  %macro PROLOGUE 2-5+ ; #args, #regs, #xmm_regs, [stack_size,] arg_names...
      %assign num_args %1
      %assign regs_used %2
+     ASSERT regs_used >= num_args
+     %if num_args > 7
+         %assign num_args 7
+     %endif
      %if regs_used > 7
          %assign regs_used 7
      %endif
      SETUP_STACK_POINTER %4
      ASSERT regs_used <= 7
-     ASSERT regs_used >= num_args
      PUSH_IF_USED 3, 4, 5, 6
      ALLOC_STACK %4
      LOAD_IF_USED 0, 1, 2, 3, 4, 5, 6
  ; Applies any symbol mangling needed for C linkage, and sets up a define such that
  ; subsequent uses of the function name automatically refer to the mangled version.
  ; Appends cpuflags to the function name if cpuflags has been specified.
- %macro cglobal 1-2+ ; name, [PROLOGUE args]
- %if %0 == 1
-     cglobal_internal %1 %+ SUFFIX
- %else
+ %macro cglobal 1-2+ "" ; name, [PROLOGUE args]
+     ; the "" is a workaround for nasm, which fails if SUFFIX is empty
+     ; and we call cglobal_internal with just %1 %+ SUFFIX (without %2)
      cglobal_internal %1 %+ SUFFIX, %2
- %endif
  %endmacro
  %macro cglobal_internal 1-2+
      %ifndef cglobaled_%1
      %assign stack_offset 0
      %assign stack_size 0
      %assign stack_size_padded 0
-     %if %0 > 1
+     %assign xmm_regs_used 0
+     %ifnidn %2, ""
          PROLOGUE %2
      %endif
  %endmacro
@@@ -814,14 -802,6 +821,14 @@@ SECTION .note.GNU-stack noalloc noexec 
      INIT_CPUFLAGS %1
  %endmacro
  
 +; FIXME: INIT_AVX can be replaced by INIT_XMM avx
 +%macro INIT_AVX 0
 +    INIT_XMM
 +    %assign avx_enabled 1
 +    %define PALIGNR PALIGNR_SSSE3
 +    %define RESET_MM_PERMUTATION INIT_AVX
 +%endmacro
 +
  %macro INIT_YMM 0-1+
      %assign avx_enabled 1
      %define RESET_MM_PERMUTATION INIT_YMM %1