Merge commit '848e86f74d3e6e87fa592ee8ba8c184cc5fd9a42'
authorMichael Niedermayer <michaelni@gmx.at>
Fri, 22 May 2015 18:12:51 +0000 (20:12 +0200)
committerMichael Niedermayer <michaelni@gmx.at>
Fri, 22 May 2015 18:24:41 +0000 (20:24 +0200)
* commit '848e86f74d3e6e87fa592ee8ba8c184cc5fd9a42':
  mpegvideo: Drop flags and flags2

Conflicts:
libavcodec/mpeg12dec.c
libavcodec/mpeg12enc.c
libavcodec/mpegvideo.c
libavcodec/mpegvideo_enc.c
libavcodec/mpegvideo_motion.c
libavcodec/ratecontrol.c
libavcodec/vc1_block.c
libavcodec/vc1_loopfilter.c
libavcodec/vc1_mc.c
libavcodec/vc1dec.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>
24 files changed:
1  2 
libavcodec/h261dec.c
libavcodec/h263.h
libavcodec/h263dec.c
libavcodec/ituh263enc.c
libavcodec/motion_est.c
libavcodec/mpeg12dec.c
libavcodec/mpeg12enc.c
libavcodec/mpeg4videodec.c
libavcodec/mpeg4videoenc.c
libavcodec/mpegvideo.c
libavcodec/mpegvideo.h
libavcodec/mpegvideo_enc.c
libavcodec/mpegvideo_motion.c
libavcodec/mpegvideo_xvmc.c
libavcodec/ratecontrol.c
libavcodec/rv10.c
libavcodec/snowenc.c
libavcodec/svq1enc.c
libavcodec/vc1_block.c
libavcodec/vc1_loopfilter.c
libavcodec/vc1_mc.c
libavcodec/vc1dec.c
libavcodec/wmv2.c
libavcodec/x86/mpegvideo.c

Simple merge
Simple merge
@@@ -411,12 -375,8 +411,9 @@@ int ff_h263_decode_frame(AVCodecContex
      int buf_size       = avpkt->size;
      MpegEncContext *s  = avctx->priv_data;
      int ret;
 +    int slice_ret = 0;
      AVFrame *pict = data;
  
-     s->flags  = avctx->flags;
-     s->flags2 = avctx->flags2;
      /* no supplementary picture */
      if (buf_size == 0) {
          /* special case for last picture */
Simple merge
Simple merge
@@@ -838,12 -814,16 +838,12 @@@ static int mpeg_decode_mb(MpegEncContex
              memset(s->last_mv, 0, sizeof(s->last_mv));
          }
          s->mb_intra = 1;
 -#if FF_API_XVMC
 -FF_DISABLE_DEPRECATION_WARNINGS
          // if 1, we memcpy blocks in xvmcvideo
 -        if (CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration > 1)
 +        if ((CONFIG_MPEG1_XVMC_HWACCEL || CONFIG_MPEG2_XVMC_HWACCEL) && s->pack_pblocks)
              ff_xvmc_pack_pblocks(s, -1); // inter are always full blocks
 -FF_ENABLE_DEPRECATION_WARNINGS
 -#endif /* FF_API_XVMC */
  
          if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
-             if (s->flags2 & CODEC_FLAG2_FAST) {
+             if (s->avctx->flags2 & CODEC_FLAG2_FAST) {
                  for (i = 0; i < 6; i++)
                      mpeg2_fast_decode_block_intra(s, *s->pblocks[i], i);
              } else {
                  return AVERROR_INVALIDDATA;
              }
  
 -#if FF_API_XVMC
 -FF_DISABLE_DEPRECATION_WARNINGS
              // if 1, we memcpy blocks in xvmcvideo
 -            if (CONFIG_MPEG_XVMC_DECODER && s->avctx->xvmc_acceleration > 1)
 +            if ((CONFIG_MPEG1_XVMC_HWACCEL || CONFIG_MPEG2_XVMC_HWACCEL) && s->pack_pblocks)
                  ff_xvmc_pack_pblocks(s, cbp);
 -FF_ENABLE_DEPRECATION_WARNINGS
 -#endif /* FF_API_XVMC */
  
              if (s->codec_id == AV_CODEC_ID_MPEG2VIDEO) {
-                 if (s->flags2 & CODEC_FLAG2_FAST) {
+                 if (s->avctx->flags2 & CODEC_FLAG2_FAST) {
                      for (i = 0; i < 6; i++) {
                          if (cbp & 32)
                              mpeg2_fast_decode_block_non_intra(s, *s->pblocks[i], i);
@@@ -2191,8 -2132,7 +2191,8 @@@ static int mpeg1_decode_sequence(AVCode
      s->codec_id             =
      s->avctx->codec_id      = AV_CODEC_ID_MPEG1VIDEO;
      s->out_format           = FMT_MPEG1;
-     if (s->flags & CODEC_FLAG_LOW_DELAY)
 +    s->swap_uv              = 0; // AFAIK VCR2 does not have SEQ_HEADER
+     if (s->avctx->flags & CODEC_FLAG_LOW_DELAY)
          s->low_delay = 1;
  
      if (s->avctx->debug & FF_DEBUG_PICT_INFO)
@@@ -2645,7 -2516,7 +2645,7 @@@ static int decode_chunks(AVCodecContex
                          }
                      }
                  }
-                 if (s2->pict_type == AV_PICTURE_TYPE_I || (s2->flags2 & CODEC_FLAG2_SHOW_ALL))
 -                if (s2->pict_type == AV_PICTURE_TYPE_I)
++                if (s2->pict_type == AV_PICTURE_TYPE_I || (s2->avctx->flags2 & CODEC_FLAG2_SHOW_ALL))
                      s->sync = 1;
                  if (!s2->next_picture_ptr) {
                      /* Skip P-frames if we do not have a reference frame or
@@@ -384,7 -324,7 +384,7 @@@ static void mpeg1_encode_sequence_heade
          put_bits(&s->pb, 1, 1);
          put_bits(&s->pb, 6, (uint32_t)((time_code / fps) % 60));
          put_bits(&s->pb, 6, (uint32_t)((time_code % fps)));
-         put_bits(&s->pb, 1, !!(s->flags & CODEC_FLAG_CLOSED_GOP) || s->intra_only || !s->gop_picture_number);
 -        put_bits(&s->pb, 1, !!(s->avctx->flags & CODEC_FLAG_CLOSED_GOP));
++        put_bits(&s->pb, 1, !!(s->avctx->flags & CODEC_FLAG_CLOSED_GOP) || s->intra_only || !s->gop_picture_number);
          put_bits(&s->pb, 1, 0);                     // broken link
      }
  }
Simple merge
@@@ -932,9 -923,11 +932,9 @@@ static void mpeg4_encode_gop_header(Mpe
      put_bits(&s->pb, 1, 1);
      put_bits(&s->pb, 6, seconds);
  
-     put_bits(&s->pb, 1, !!(s->flags & CODEC_FLAG_CLOSED_GOP));
+     put_bits(&s->pb, 1, !!(s->avctx->flags & CODEC_FLAG_CLOSED_GOP));
      put_bits(&s->pb, 1, 0);  // broken link == NO
  
 -    s->last_time_base = time / s->avctx->time_base.den;
 -
      ff_mpeg4_stuffing(&s->pb);
  }
  
@@@ -1115,20 -1060,6 +1115,18 @@@ void ff_mpv_decode_defaults(MpegEncCont
      ff_mpv_common_defaults(s);
  }
  
-     s->flags           = avctx->flags;
-     s->flags2          = avctx->flags2;
 +void ff_mpv_decode_init(MpegEncContext *s, AVCodecContext *avctx)
 +{
 +    s->avctx           = avctx;
 +    s->width           = avctx->coded_width;
 +    s->height          = avctx->coded_height;
 +    s->codec_id        = avctx->codec->id;
 +    s->workaround_bugs = avctx->workaround_bugs;
 +
 +    /* convert fourcc to upper case */
 +    s->codec_tag          = avpriv_toupper4(avctx->codec_tag);
 +}
 +
  static int init_er(MpegEncContext *s)
  {
      ERContext *er = &s->er;
@@@ -1330,14 -1277,13 +1328,11 @@@ av_cold int ff_mpv_common_init(MpegEncC
  
      dct_init(s);
  
-     s->flags  = s->avctx->flags;
-     s->flags2 = s->avctx->flags2;
      /* set chroma shifts */
 -    av_pix_fmt_get_chroma_sub_sample(s->avctx->pix_fmt,
 -                                     &s->chroma_x_shift,
 -                                     &s->chroma_y_shift);
 +    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);
  
      FF_ALLOCZ_OR_GOTO(s->avctx, s->picture,
                        MAX_PICTURE_COUNT * sizeof(Picture), fail);
@@@ -2527,408 -2003,6 +2522,408 @@@ void ff_print_debug_info2(AVCodecContex
      }
  }
  
-         if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
 +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;
 +
 +    // 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)) {
++        if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
 +            uint8_t *ubuf = s->edge_emu_buffer + 18 * s->linesize;
 +            uint8_t *vbuf =ubuf + 9 * s->uvlinesize;
 +            s->vdsp.emulated_edge_mc(ubuf,  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(vbuf,  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 = ubuf;
 +            ptr_cr = vbuf;
 +        }
 +    }
 +
 +    // 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))
++    if (!CONFIG_GRAY || !(s->avctx->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->avctx->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->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;
 +                }
 +            }
 +        }
 +        break;
 +    default:
 +        av_assert2(0);
 +    }
 +}
 +
  /**
   * find the lowest MB row referenced in the MVs
   */
@@@ -3078,9 -2155,9 +3073,9 @@@ void mpv_decode_mb_internal(MpegEncCont
      else if (!is_mpeg12 && (s->h263_pred || s->h263_aic))
          s->mbintra_table[mb_xy]=1;
  
-     if (   (s->flags&CODEC_FLAG_PSNR)
-         || s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor
-         || !(s->encoding && (s->intra_only || s->pict_type==AV_PICTURE_TYPE_B) && s->avctx->mb_decision != FF_MB_DECISION_RD)) { //FIXME precalc
 -    if ((s->avctx->flags & CODEC_FLAG_PSNR) ||
++    if ((s->avctx->flags & CODEC_FLAG_PSNR) || s->avctx->frame_skip_threshold || s->avctx->frame_skip_factor ||
+         !(s->encoding && (s->intra_only || s->pict_type == AV_PICTURE_TYPE_B) &&
+           s->avctx->mb_decision != FF_MB_DECISION_RD)) { // FIXME precalc
          uint8_t *dest_y, *dest_cb, *dest_cr;
          int dct_linesize, dct_offset;
          op_pixels_func (*op_pix)[4];
  skip_idct:
          if(!readable){
              s->hdsp.put_pixels_tab[0][0](s->dest[0], dest_y ,   linesize,16);
-             if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
 -            s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift);
 -            s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift);
++            if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
 +                s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[1], dest_cb, uvlinesize,16 >> s->chroma_y_shift);
 +                s->hdsp.put_pixels_tab[s->chroma_x_shift][0](s->dest[2], dest_cr, uvlinesize,16 >> s->chroma_y_shift);
 +            }
          }
      }
  }
Simple merge
@@@ -389,36 -341,12 +387,36 @@@ av_cold int ff_mpv_encode_init(AVCodecC
                           (s->mpv_flags & FF_MPV_FLAG_QP_RD)) &&
                          !s->fixed_qscale;
  
-     s->loop_filter      = !!(s->flags & CODEC_FLAG_LOOP_FILTER);
+     s->loop_filter = !!(s->avctx->flags & CODEC_FLAG_LOOP_FILTER);
  
      if (avctx->rc_max_rate && !avctx->rc_buffer_size) {
 -        av_log(avctx, AV_LOG_ERROR,
 -               "a vbv buffer size is needed, "
 -               "for encoding with a maximum bitrate\n");
 +        switch(avctx->codec_id) {
 +        case AV_CODEC_ID_MPEG1VIDEO:
 +        case AV_CODEC_ID_MPEG2VIDEO:
 +            avctx->rc_buffer_size = FFMAX(avctx->rc_max_rate, 15000000) * 112LL / 15000000 * 16384;
 +            break;
 +        case AV_CODEC_ID_MPEG4:
 +        case AV_CODEC_ID_MSMPEG4V1:
 +        case AV_CODEC_ID_MSMPEG4V2:
 +        case AV_CODEC_ID_MSMPEG4V3:
 +            if       (avctx->rc_max_rate >= 15000000) {
 +                avctx->rc_buffer_size = 320 + (avctx->rc_max_rate - 15000000LL) * (760-320) / (38400000 - 15000000);
 +            } else if(avctx->rc_max_rate >=  2000000) {
 +                avctx->rc_buffer_size =  80 + (avctx->rc_max_rate -  2000000LL) * (320- 80) / (15000000 -  2000000);
 +            } else if(avctx->rc_max_rate >=   384000) {
 +                avctx->rc_buffer_size =  40 + (avctx->rc_max_rate -   384000LL) * ( 80- 40) / ( 2000000 -   384000);
 +            } else
 +                avctx->rc_buffer_size = 40;
 +            avctx->rc_buffer_size *= 16384;
 +            break;
 +        }
 +        if (avctx->rc_buffer_size) {
 +            av_log(avctx, AV_LOG_INFO, "Automatically choosing VBV buffer size of %d kbyte\n", avctx->rc_buffer_size/8192);
 +        }
 +    }
 +
 +    if ((!avctx->rc_max_rate) != (!avctx->rc_buffer_size)) {
 +        av_log(avctx, AV_LOG_ERROR, "Either both buffer size and max rate or neither must be specified\n");
          return -1;
      }
  
          return -1;
      }
  
-     if ((s->flags & (CODEC_FLAG_INTERLACED_DCT | CODEC_FLAG_INTERLACED_ME)) &&
 +    if (s->codec_id == AV_CODEC_ID_RV10 &&
 +        (avctx->width &15 ||
 +         avctx->height&15 )) {
 +        av_log(avctx, AV_LOG_ERROR, "width and height must be a multiple of 16\n");
 +        return AVERROR(EINVAL);
 +    }
 +
 +    if (s->codec_id == AV_CODEC_ID_RV20 &&
 +        (avctx->width &3 ||
 +         avctx->height&3 )) {
 +        av_log(avctx, AV_LOG_ERROR, "width and height must be a multiple of 4\n");
 +        return AVERROR(EINVAL);
 +    }
 +
 +    if ((s->codec_id == AV_CODEC_ID_WMV1 ||
 +         s->codec_id == AV_CODEC_ID_WMV2) &&
 +         avctx->width & 1) {
 +         av_log(avctx, AV_LOG_ERROR, "width must be multiple of 2\n");
 +         return -1;
 +    }
 +
+     if ((s->avctx->flags & (CODEC_FLAG_INTERLACED_DCT | CODEC_FLAG_INTERLACED_ME)) &&
          s->codec_id != AV_CODEC_ID_MPEG4 && s->codec_id != AV_CODEC_ID_MPEG2VIDEO) {
          av_log(avctx, AV_LOG_ERROR, "interlacing not supported by codec\n");
          return -1;
@@@ -1422,7 -1258,20 +1420,7 @@@ static int select_input_picture(MpegEnc
          } else {
              int b_frames;
  
-             if (s->flags & CODEC_FLAG_PASS2) {
 -            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->avctx->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;
  
@@@ -1808,10 -1654,10 +1806,10 @@@ vbv_retry
                  goto vbv_retry;
              }
  
 -            assert(s->avctx->rc_max_rate);
 +            av_assert0(s->avctx->rc_max_rate);
          }
  
-         if (s->flags & CODEC_FLAG_PASS1)
+         if (s->avctx->flags & CODEC_FLAG_PASS1)
              ff_write_pass1_stats(s);
  
          for (i = 0; i < 4; i++) {
@@@ -320,25 -320,23 +320,25 @@@ void mpeg_motion_internal(MpegEncContex
          s->vdsp.emulated_edge_mc(s->edge_emu_buffer, ptr_y,
                                   s->linesize, s->linesize,
                                   17, 17 + field_based,
 -                                 src_x, src_y << field_based,
 +                                 src_x, src_y,
                                   s->h_edge_pos, s->v_edge_pos);
          ptr_y = s->edge_emu_buffer;
-         if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
+         if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
 -            uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
 -            s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
 +            uint8_t *ubuf = s->edge_emu_buffer + 18 * s->linesize;
 +            uint8_t *vbuf = ubuf + 9 * s->uvlinesize;
 +            uvsrc_y = (unsigned)uvsrc_y << field_based;
 +            s->vdsp.emulated_edge_mc(ubuf, ptr_cb,
                                       s->uvlinesize, s->uvlinesize,
                                       9, 9 + field_based,
 -                                     uvsrc_x, uvsrc_y << field_based,
 +                                     uvsrc_x, uvsrc_y,
                                       s->h_edge_pos >> 1, s->v_edge_pos >> 1);
 -            s->vdsp.emulated_edge_mc(uvbuf + 16, ptr_cr,
 +            s->vdsp.emulated_edge_mc(vbuf, ptr_cr,
                                       s->uvlinesize, s->uvlinesize,
                                       9, 9 + field_based,
 -                                     uvsrc_x, uvsrc_y << field_based,
 +                                     uvsrc_x, uvsrc_y,
                                       s->h_edge_pos >> 1, s->v_edge_pos >> 1);
 -            ptr_cb = uvbuf;
 -            ptr_cr = uvbuf + 16;
 +            ptr_cb = ubuf;
 +            ptr_cr = vbuf;
          }
      }
  
@@@ -546,10 -544,9 +546,10 @@@ static inline void qpel_motion(MpegEncC
                                   src_x, src_y << field_based,
                                   s->h_edge_pos, s->v_edge_pos);
          ptr_y = s->edge_emu_buffer;
-         if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
+         if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
 -            uint8_t *uvbuf = s->edge_emu_buffer + 18 * s->linesize;
 -            s->vdsp.emulated_edge_mc(uvbuf, ptr_cb,
 +            uint8_t *ubuf = s->edge_emu_buffer + 18 * s->linesize;
 +            uint8_t *vbuf = ubuf + 9 * s->uvlinesize;
 +            s->vdsp.emulated_edge_mc(ubuf, ptr_cb,
                                       s->uvlinesize, s->uvlinesize,
                                       9, 9 + field_based,
                                       uvsrc_x, uvsrc_y << field_based,
Simple merge
@@@ -158,10 -149,8 +158,10 @@@ av_cold int ff_rate_control_init(MpegEn
          rcc->last_qscale_for[i] = FF_QP2LAMBDA * 5;
      }
      rcc->buffer_index = s->avctx->rc_initial_buffer_occupancy;
 +    if (!rcc->buffer_index)
 +        rcc->buffer_index = s->avctx->rc_buffer_size * 3 / 4;
  
-     if (s->flags & CODEC_FLAG_PASS2) {
+     if (s->avctx->flags & CODEC_FLAG_PASS2) {
          int i;
          char *p;
  
@@@ -767,25 -751,19 +767,25 @@@ float ff_rate_estimate_qscale(MpegEncCo
  
      get_qminmax(&qmin, &qmax, s, pict_type);
  
 -    fps = 1 / av_q2d(s->avctx->time_base);
 +    fps = get_fps(s->avctx);
      /* update predictors */
      if (picture_number > 2 && !dry_run) {
 -        const int last_var = s->last_pict_type == AV_PICTURE_TYPE_I ? rcc->last_mb_var_sum
 -                                                                    : rcc->last_mc_mb_var_sum;
 +        const int64_t last_var =
 +            s->last_pict_type == AV_PICTURE_TYPE_I ? rcc->last_mb_var_sum
 +                                                   : rcc->last_mc_mb_var_sum;
 +        av_assert1(s->frame_bits >= s->stuffing_bits);
          update_predictor(&rcc->pred[s->last_pict_type],
                           rcc->last_qscale,
 -                         sqrt(last_var), s->frame_bits);
 +                         sqrt(last_var),
 +                         s->frame_bits - s->stuffing_bits);
      }
  
-     if (s->flags & CODEC_FLAG_PASS2) {
+     if (s->avctx->flags & CODEC_FLAG_PASS2) {
 -        assert(picture_number >= 0);
 -        assert(picture_number < rcc->num_entries);
 +        av_assert0(picture_number >= 0);
 +        if (picture_number >= rcc->num_entries) {
 +            av_log(s, AV_LOG_ERROR, "Input is longer than 2-pass log file\n");
 +            return -1;
 +        }
          rce         = &rcc->entry[picture_number];
          wanted_bits = rce->expected_bits;
      } else {
      var = pict_type == AV_PICTURE_TYPE_I ? pic->mb_var_sum : pic->mc_mb_var_sum;
  
      short_term_q = 0; /* avoid warning */
-     if (s->flags & CODEC_FLAG_PASS2) {
+     if (s->avctx->flags & CODEC_FLAG_PASS2) {
          if (pict_type != AV_PICTURE_TYPE_I)
 -            assert(pict_type == rce->new_pict_type);
 +            av_assert0(pict_type == rce->new_pict_type);
  
          q = rce->new_qscale / br_compensation;
 -        ff_dlog(s, "%f %f %f last:%d var:%d type:%d//\n", q, rce->new_qscale,
 +        ff_dlog(s, "%f %f %f last:%d var:%"PRId64" type:%d//\n", q, rce->new_qscale,
                  br_compensation, s->frame_bits, var, pict_type);
      } else {
          rce->pict_type     =
Simple merge
index a7de16a,0000000..aab7146
mode 100644,000000..100644
--- /dev/null
@@@ -1,2040 -1,0 +1,2038 @@@
-     s->m.flags   = avctx->flags;
 +/*
 + * 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 "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;
 +    int i;
 +
 +    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) {
 +        return ret;
 +    }
 +    ff_mpegvideoencdsp_init(&s->mpvencdsp, avctx);
 +
 +    ff_snow_alloc_blocks(s);
 +
 +    s->version=0;
 +
 +    s->m.avctx   = avctx;
-         s->m.flags= s->avctx->flags;
 +    s->m.bit_rate= avctx->bit_rate;
 +
 +    s->m.me.temp      =
 +    s->m.me.scratchpad= av_mallocz_array((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->mecc, s->mecc.me_cmp, s->avctx->me_cmp);
 +    ff_set_cmp(&s->mecc, s->mecc.me_sub_cmp, s->avctx->me_sub_cmp);
 +
 +    s->input_picture = av_frame_alloc();
 +    avctx->coded_frame = av_frame_alloc();
 +    if (!s->input_picture || !avctx->coded_frame)
 +        return AVERROR(ENOMEM);
 +
 +    if ((ret = ff_snow_get_buffer(s, s->input_picture)) < 0)
 +        return ret;
 +
 +    if(s->avctx->me_method == ME_ITER){
 +        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_array(size, sizeof(int16_t[2]));
 +            s->ref_scores[i]= av_mallocz_array(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;
 +
 +    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_uint8( ROUNDED_DIV(ab<<LOG2_OBMC_MAX, aa) ); //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->mecc.me_cmp[0](&s->m, src + off, dst + off, ref_stride, 16);
 +            }
 +        }
 +    }else{
 +        av_assert2(block_w==8);
 +        distortion = s->mecc.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->mecc.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) + s->intra_penalty * !!intra;
 +
 +//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{
 +                        int newx = block->mx;
 +                        int newy = block->my;
 +                        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, newx+4*(i-j), newy+(4*j), obmc_edged, &best_rd);
 +                                dia_change |= check_block_inter(s, mb_x, mb_y, newx-4*(i-j), newy-(4*j), obmc_edged, &best_rd);
 +                                dia_change |= check_block_inter(s, mb_x, mb_y, newx-(4*j), newy+4*(i-j), obmc_edged, &best_rd);
 +                                dia_change |= check_block_inter(s, mb_x, mb_y, newx+(4*j), newy-4*(i-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_DEBUG, "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,
 +                        const 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<FF_CEIL_RSHIFT(height, vshift); y++)
 +            memcpy(&s->input_picture->data[i][y * s->input_picture->linesize[i]],
 +                   &pict->data[i][y * pict->linesize[i]],
 +                   FF_CEIL_RSHIFT(width, hshift));
 +        s->mpvencdsp.draw_edges(s->input_picture->data[i], s->input_picture->linesize[i],
 +                                FF_CEIL_RSHIFT(width, hshift), FF_CEIL_RSHIFT(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
 +
 +    if (s->current_picture->data[0] && !(s->avctx->flags&CODEC_FLAG_EMU_EDGE)) {
 +        int w = s->avctx->width;
 +        int h = s->avctx->height;
 +
 +        s->mpvencdsp.draw_edges(s->current_picture->data[0],
 +                                s->current_picture->linesize[0], w   , h   ,
 +                                EDGE_WIDTH  , EDGE_WIDTH  , EDGE_TOP | EDGE_BOTTOM);
 +        if (s->current_picture->data[2]) {
 +            s->mpvencdsp.draw_edges(s->current_picture->data[1],
 +                                    s->current_picture->linesize[1], w>>s->chroma_h_shift, h>>s->chroma_v_shift,
 +                                    EDGE_WIDTH>>s->chroma_h_shift, EDGE_WIDTH>>s->chroma_v_shift, EDGE_TOP | EDGE_BOTTOM);
 +            s->mpvencdsp.draw_edges(s->current_picture->data[2],
 +                                    s->current_picture->linesize[2], w>>s->chroma_h_shift, h>>s->chroma_v_shift,
 +                                    EDGE_WIDTH>>s->chroma_h_shift, EDGE_WIDTH>>s->chroma_v_shift, EDGE_TOP | EDGE_BOTTOM);
 +        }
 +    }
 +
 +    ff_snow_frame_start(s);
 +    av_frame_unref(avctx->coded_frame);
 +    ret = av_frame_ref(avctx->coded_frame, s->current_picture);
 +    if (ret < 0)
 +        return ret;
 +
 +    s->m.current_picture_ptr= &s->m.current_picture;
 +    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 = s->last_picture[0];
 +        s->m.    new_picture.f = s->input_picture;
 +        s->m.   last_picture_ptr= &s->m.   last_picture;
 +        s->m.linesize = 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.me.dia_size = avctx->dia_size;
 +        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.mecc= s->mecc; //move
 +        s->m.qdsp= s->qdsp; //move
 +        s->m.hdsp = s->hdsp;
 +        ff_init_me(&s->m);
 +        s->hdsp = s->m.hdsp;
 +        s->mecc= s->m.mecc;
 +    }
 +
 +    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_frame_free(&avctx->coded_frame);
 +    av_freep(&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[] = {
 +    FF_MPV_COMMON_OPTS
 +    { "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 },
 +    { "intra_penalty",  "Penalty for intra blocks in block decission",      OFFSET(intra_penalty), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, INT_MAX, 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,
 +    .caps_internal  = FF_CODEC_CAP_INIT_THREADSAFE |
 +                      FF_CODEC_CAP_INIT_CLEANUP,
 +};
 +
 +
 +#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_array(width, sizeof(DWTELEM));
 +    s.temp_idwt_buffer = av_mallocz_array(width, sizeof(IDWTELEM));
 +
 +    if (!s.temp_dwt_buffer || !s.temp_idwt_buffer) {
 +        fprintf(stderr, "Failed to allocate memory\n");
 +        return 1;
 +    }
 +
 +    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
@@@ -99,7 -111,6 +99,7 @@@ static void vc1_put_signed_blocks_clamp
              s->idsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][3],
                                                s->dest[0] - v_dist * s->linesize - 8,
                                                stride_y);
-             if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
++            if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
              s->idsp.put_signed_pixels_clamped(v->block[v->topleft_blk_idx][4],
                                                s->dest[1] - 8 * s->uvlinesize - 8,
                                                s->uvlinesize);
              s->idsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][3],
                                                s->dest[0] - v_dist * s->linesize + 8,
                                                stride_y);
-             if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
++            if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
              s->idsp.put_signed_pixels_clamped(v->block[v->top_blk_idx][4],
                                                s->dest[1] - 8 * s->uvlinesize,
                                                s->uvlinesize);
@@@ -1327,7 -1416,7 +1327,7 @@@ static int vc1_decode_p_mb(VC1Context *
  
                      vc1_decode_intra_block(v, s->block[i], i, val, mquant,
                                             (i & 4) ? v->codingset2 : v->codingset);
-                     if (CONFIG_GRAY && (i>3) && (s->flags & CODEC_FLAG_GRAY))
 -                    if ((i > 3) && (s->avctx->flags & CODEC_FLAG_GRAY))
++                    if (CONFIG_GRAY && (i > 3) && (s->avctx->flags & CODEC_FLAG_GRAY))
                          continue;
                      v->vc1dsp.vc1_inv_trans_8x8(s->block[i]);
                      if (v->rangeredfrm)
                  } else if (val) {
                      pat = vc1_decode_p_block(v, s->block[i], i, mquant, ttmb, first_block,
                                               s->dest[dst_idx] + off, (i & 4) ? s->uvlinesize : s->linesize,
-                                              CONFIG_GRAY && (i & 4) && (s->flags & CODEC_FLAG_GRAY), &block_tt);
 -                                             (i & 4) && (s->avctx->flags & CODEC_FLAG_GRAY), &block_tt);
++                                             CONFIG_GRAY && (i & 4) && (s->avctx->flags & CODEC_FLAG_GRAY), &block_tt);
                      block_cbp |= pat << (i << 2);
                      if (!v->ttmbf && ttmb < 8)
                          ttmb = -1;
  
                      vc1_decode_intra_block(v, s->block[i], i, is_coded[i], mquant,
                                             (i & 4) ? v->codingset2 : v->codingset);
-                     if (CONFIG_GRAY && (i>3) && (s->flags & CODEC_FLAG_GRAY))
 -                    if ((i > 3) && (s->avctx->flags & CODEC_FLAG_GRAY))
++                    if (CONFIG_GRAY && (i > 3) && (s->avctx->flags & CODEC_FLAG_GRAY))
                          continue;
                      v->vc1dsp.vc1_inv_trans_8x8(s->block[i]);
                      if (v->rangeredfrm)
                      pat = vc1_decode_p_block(v, s->block[i], i, mquant, ttmb,
                                               first_block, s->dest[dst_idx] + off,
                                               (i & 4) ? s->uvlinesize : s->linesize,
-                                              CONFIG_GRAY && (i & 4) && (s->flags & CODEC_FLAG_GRAY),
 -                                             (i & 4) && (s->avctx->flags & CODEC_FLAG_GRAY),
++                                             CONFIG_GRAY && (i & 4) && (s->avctx->flags & CODEC_FLAG_GRAY),
                                               &block_tt);
                      block_cbp |= pat << (i << 2);
                      if (!v->ttmbf && ttmb < 8)
@@@ -1586,7 -1675,8 +1586,8 @@@ static int vc1_decode_p_mb_intfr(VC1Con
  
                  vc1_decode_intra_block(v, s->block[i], i, val, mquant,
                                         (i & 4) ? v->codingset2 : v->codingset);
-                 if (CONFIG_GRAY && (i>3) && (s->flags & CODEC_FLAG_GRAY)) continue;
 -                if ((i > 3) && (s->avctx->flags & CODEC_FLAG_GRAY))
++                if (CONFIG_GRAY && (i > 3) && (s->avctx->flags & CODEC_FLAG_GRAY))
+                     continue;
                  v->vc1dsp.vc1_inv_trans_8x8(s->block[i]);
                  if (i < 4) {
                      stride_y = s->linesize << fieldtx;
                      pat = vc1_decode_p_block(v, s->block[i], i, mquant, ttmb,
                                               first_block, s->dest[dst_idx] + off,
                                               (i & 4) ? s->uvlinesize : (s->linesize << fieldtx),
-                                              CONFIG_GRAY && (i & 4) && (s->flags & CODEC_FLAG_GRAY), &block_tt);
 -                                             (i & 4) && (s->avctx->flags & CODEC_FLAG_GRAY), &block_tt);
++                                             CONFIG_GRAY && (i & 4) && (s->avctx->flags & CODEC_FLAG_GRAY), &block_tt);
                      block_cbp |= pat << (i << 2);
                      if (!v->ttmbf && ttmb < 8)
                          ttmb = -1;
@@@ -1751,7 -1846,7 +1752,7 @@@ static int vc1_decode_p_mb_intfi(VC1Con
  
              vc1_decode_intra_block(v, s->block[i], i, val, mquant,
                                     (i & 4) ? v->codingset2 : v->codingset);
-             if (CONFIG_GRAY && (i>3) && (s->flags & CODEC_FLAG_GRAY))
 -            if ((i > 3) && (s->avctx->flags & CODEC_FLAG_GRAY))
++            if (CONFIG_GRAY && (i > 3) && (s->avctx->flags & CODEC_FLAG_GRAY))
                  continue;
              v->vc1dsp.vc1_inv_trans_8x8(s->block[i]);
              off  = (i & 4) ? 0 : ((i & 1) * 8 + (i & 2) * 4 * s->linesize);
                  pat = vc1_decode_p_block(v, s->block[i], i, mquant, ttmb,
                                           first_block, s->dest[dst_idx] + off,
                                           (i & 4) ? s->uvlinesize : s->linesize,
-                                          CONFIG_GRAY && (i & 4) && (s->flags & CODEC_FLAG_GRAY),
 -                                         (i & 4) && (s->avctx->flags & CODEC_FLAG_GRAY),
++                                         CONFIG_GRAY && (i & 4) && (s->avctx->flags & CODEC_FLAG_GRAY),
                                           &block_tt);
                  block_cbp |= pat << (i << 2);
 -                if (!v->ttmbf && ttmb < 8) ttmb = -1;
 +                if (!v->ttmbf && ttmb < 8)
 +                    ttmb = -1;
                  first_block = 0;
              }
          }
@@@ -1952,7 -2049,7 +1953,7 @@@ static void vc1_decode_b_mb(VC1Context 
  
              vc1_decode_intra_block(v, s->block[i], i, val, mquant,
                                     (i & 4) ? v->codingset2 : v->codingset);
-             if (CONFIG_GRAY && (i>3) && (s->flags & CODEC_FLAG_GRAY))
 -            if ((i > 3) && (s->avctx->flags & CODEC_FLAG_GRAY))
++            if (CONFIG_GRAY && (i > 3) && (s->avctx->flags & CODEC_FLAG_GRAY))
                  continue;
              v->vc1dsp.vc1_inv_trans_8x8(s->block[i]);
              if (v->rangeredfrm)
              vc1_decode_p_block(v, s->block[i], i, mquant, ttmb,
                                 first_block, s->dest[dst_idx] + off,
                                 (i & 4) ? s->uvlinesize : s->linesize,
-                                CONFIG_GRAY && (i & 4) && (s->flags & CODEC_FLAG_GRAY), NULL);
 -                               (i & 4) && (s->avctx->flags & CODEC_FLAG_GRAY), NULL);
++                               CONFIG_GRAY && (i & 4) && (s->avctx->flags & CODEC_FLAG_GRAY), NULL);
              if (!v->ttmbf && ttmb < 8)
                  ttmb = -1;
              first_block = 0;
@@@ -2027,7 -2124,7 +2028,7 @@@ static void vc1_decode_b_mb_intfi(VC1Co
  
              vc1_decode_intra_block(v, s->block[i], i, val, mquant,
                                     (i & 4) ? v->codingset2 : v->codingset);
-             if (CONFIG_GRAY && (i>3) && (s->flags & CODEC_FLAG_GRAY))
 -            if ((i > 3) && (s->avctx->flags & CODEC_FLAG_GRAY))
++            if (CONFIG_GRAY && (i > 3) && (s->avctx->flags & CODEC_FLAG_GRAY))
                  continue;
              v->vc1dsp.vc1_inv_trans_8x8(s->block[i]);
              if (v->rangeredfrm)
                  vc1_decode_p_block(v, s->block[i], i, mquant, ttmb,
                                     first_block, s->dest[dst_idx] + off,
                                     (i & 4) ? s->uvlinesize : s->linesize,
-                                    CONFIG_GRAY && (i & 4) && (s->flags & CODEC_FLAG_GRAY), NULL);
 -                                   (i & 4) && (s->avctx->flags & CODEC_FLAG_GRAY), NULL);
++                                   CONFIG_GRAY && (i & 4) && (s->avctx->flags & CODEC_FLAG_GRAY), NULL);
                  if (!v->ttmbf && ttmb < 8)
                      ttmb = -1;
                  first_block = 0;
@@@ -2250,7 -2342,7 +2251,7 @@@ static int vc1_decode_b_mb_intfr(VC1Con
  
              vc1_decode_intra_block(v, s->block[i], i, val, mquant,
                                     (i & 4) ? v->codingset2 : v->codingset);
-             if (CONFIG_GRAY && i > 3 && (s->flags & CODEC_FLAG_GRAY))
 -            if (i > 3 && (s->avctx->flags & CODEC_FLAG_GRAY))
++            if (CONFIG_GRAY && i > 3 && (s->avctx->flags & CODEC_FLAG_GRAY))
                  continue;
              v->vc1dsp.vc1_inv_trans_8x8(s->block[i]);
              if (i < 4) {
                      pat = vc1_decode_p_block(v, s->block[i], i, mquant, ttmb,
                                               first_block, s->dest[dst_idx] + off,
                                               (i & 4) ? s->uvlinesize : (s->linesize << fieldtx),
-                                              CONFIG_GRAY && (i & 4) && (s->flags & CODEC_FLAG_GRAY), &block_tt);
 -                                             (i & 4) && (s->avctx->flags & CODEC_FLAG_GRAY), &block_tt);
++                                             CONFIG_GRAY && (i & 4) && (s->avctx->flags & CODEC_FLAG_GRAY), &block_tt);
                      block_cbp |= pat << (i << 2);
                      if (!v->ttmbf && ttmb < 8)
                          ttmb = -1;
@@@ -2559,7 -2651,7 +2560,7 @@@ static void vc1_decode_i_blocks(VC1Cont
  
                  vc1_decode_i_block(v, s->block[k], k, val, (k < 4) ? v->codingset : v->codingset2);
  
-                 if (CONFIG_GRAY && k > 3 && (s->flags & CODEC_FLAG_GRAY))
 -                if (k > 3 && (s->avctx->flags & CODEC_FLAG_GRAY))
++                if (CONFIG_GRAY && k > 3 && (s->avctx->flags & CODEC_FLAG_GRAY))
                      continue;
                  v->vc1dsp.vc1_inv_trans_8x8(s->block[k]);
                  if (v->pq >= 9 && v->overlap) {
                  if (s->mb_x) {
                      v->vc1dsp.vc1_h_overlap(s->dest[0], s->linesize);
                      v->vc1dsp.vc1_h_overlap(s->dest[0] + 8 * s->linesize, s->linesize);
-                     if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
 -                    if (!(s->avctx->flags & CODEC_FLAG_GRAY)) {
++                    if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
                          v->vc1dsp.vc1_h_overlap(s->dest[1], s->uvlinesize);
                          v->vc1dsp.vc1_h_overlap(s->dest[2], s->uvlinesize);
                      }
                  if (!s->first_slice_line) {
                      v->vc1dsp.vc1_v_overlap(s->dest[0], s->linesize);
                      v->vc1dsp.vc1_v_overlap(s->dest[0] + 8, s->linesize);
-                     if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
 -                    if (!(s->avctx->flags & CODEC_FLAG_GRAY)) {
++                    if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
                          v->vc1dsp.vc1_v_overlap(s->dest[1], s->uvlinesize);
                          v->vc1dsp.vc1_v_overlap(s->dest[2], s->uvlinesize);
                      }
@@@ -2722,7 -2814,7 +2723,7 @@@ static void vc1_decode_i_blocks_adv(VC1
                  vc1_decode_i_block_adv(v, block[k], k, val,
                                         (k < 4) ? v->codingset : v->codingset2, mquant);
  
-                 if (CONFIG_GRAY && k > 3 && (s->flags & CODEC_FLAG_GRAY))
 -                if (k > 3 && (s->avctx->flags & CODEC_FLAG_GRAY))
++                if (CONFIG_GRAY && k > 3 && (s->avctx->flags & CODEC_FLAG_GRAY))
                      continue;
                  v->vc1dsp.vc1_inv_trans_8x8(block[k]);
              }
@@@ -40,7 -40,6 +40,7 @@@ void ff_vc1_loop_filter_iblk(VC1Contex
          if (s->mb_x)
              v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
          v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize + 8, s->linesize, pq);
-         if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
++        if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY))
          for (j = 0; j < 2; j++) {
              v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1], s->uvlinesize, pq);
              if (s->mb_x)
      if (s->mb_y == s->end_mb_y - 1) {
          if (s->mb_x) {
              v->vc1dsp.vc1_h_loop_filter16(s->dest[0], s->linesize, pq);
-             if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
++            if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
              v->vc1dsp.vc1_h_loop_filter8(s->dest[1], s->uvlinesize, pq);
              v->vc1dsp.vc1_h_loop_filter8(s->dest[2], s->uvlinesize, pq);
 +            }
          }
          v->vc1dsp.vc1_h_loop_filter16(s->dest[0] + 8, s->linesize, pq);
      }
@@@ -76,7 -73,6 +76,7 @@@ void ff_vc1_loop_filter_iblk_delayed(VC
                  if (s->mb_x >= 2)
                      v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 16, s->linesize, pq);
                  v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize - 8, s->linesize, pq);
-                 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
++                if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY))
                  for (j = 0; j < 2; j++) {
                      v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq);
                      if (s->mb_x >= 2) {
@@@ -94,7 -90,6 +94,7 @@@
                  if (s->mb_x)
                      v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize, s->linesize, pq);
                  v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 32 * s->linesize + 8, s->linesize, pq);
-                 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))
++                if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY))
                  for (j = 0; j < 2; j++) {
                      v->vc1dsp.vc1_v_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
                      if (s->mb_x >= 2) {
                  if (s->mb_x >= 2)
                      v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 16, s->linesize, pq);
                  v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize - 8, s->linesize, pq);
-                 if (s->mb_x >= 2 && (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))) {
 -                if (s->mb_x >= 2) {
++                if (s->mb_x >= 2 && (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY))) {
                      for (j = 0; j < 2; j++) {
                          v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize - 8, s->uvlinesize, pq);
                      }
                  if (s->mb_x)
                      v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize, s->linesize, pq);
                  v->vc1dsp.vc1_h_loop_filter16(s->dest[0] - 16 * s->linesize + 8, s->linesize, pq);
-                 if (s->mb_x && (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY))) {
 -                if (s->mb_x) {
++                if (s->mb_x && (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY))) {
                      for (j = 0; j < 2; j++) {
                          v->vc1dsp.vc1_h_loop_filter8(s->dest[j + 1] - 8 * s->uvlinesize, s->uvlinesize, pq);
                      }
@@@ -155,7 -150,7 +155,7 @@@ void ff_vc1_smooth_overlap_filter_iblk(
                                        v->block[v->cur_blk_idx][0]);
              v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][3],
                                        v->block[v->cur_blk_idx][2]);
-             if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
 -            if (!(s->avctx->flags & CODEC_FLAG_GRAY)) {
++            if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
                  v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][4],
                                            v->block[v->cur_blk_idx][4]);
                  v->vc1dsp.vc1_h_s_overlap(v->block[v->left_blk_idx][5],
                                            v->block[v->cur_blk_idx][0]);
                  v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][3],
                                            v->block[v->cur_blk_idx][1]);
-                 if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
 -                if (!(s->avctx->flags & CODEC_FLAG_GRAY)) {
++                if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
                      v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][4],
                                                v->block[v->cur_blk_idx][4]);
                      v->vc1dsp.vc1_v_s_overlap(v->block[v->top_blk_idx][5],
                                        v->block[v->left_blk_idx][0]);
              v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][3],
                                        v->block[v->left_blk_idx][1]);
-             if (!CONFIG_GRAY || !(s->flags & CODEC_FLAG_GRAY)) {
 -            if (!(s->avctx->flags & CODEC_FLAG_GRAY)) {
++            if (!CONFIG_GRAY || !(s->avctx->flags & CODEC_FLAG_GRAY)) {
                  v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][4],
                                            v->block[v->left_blk_idx][4]);
                  v->vc1dsp.vc1_v_s_overlap(v->block[v->topleft_blk_idx][5],
@@@ -336,9 -331,8 +336,9 @@@ void ff_vc1_apply_p_loop_filter(VC1Cont
  {
      MpegEncContext *s = &v->s;
      int i;
-     int block_count = CONFIG_GRAY && (s->flags & CODEC_FLAG_GRAY) ? 4 : 6;
++    int block_count = CONFIG_GRAY && (s->avctx->flags & CODEC_FLAG_GRAY) ? 4 : 6;
  
 -    for (i = 0; i < 6; i++) {
 +    for (i = 0; i < block_count; i++) {
          vc1_apply_p_v_loop_filter(v, i);
      }
  
@@@ -270,7 -136,7 +270,7 @@@ void ff_vc1_mc_1mv(VC1Context *v, int d
      }
  
      /* for grayscale we should not try to read from unknown area */
-     if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY) {
 -    if (s->avctx->flags & CODEC_FLAG_GRAY) {
++    if (CONFIG_GRAY && s->avctx->flags & CODEC_FLAG_GRAY) {
          srcU = s->edge_emu_buffer + 18 * s->linesize;
          srcV = s->edge_emu_buffer + 18 * s->linesize;
      }
              s->hdsp.put_no_rnd_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
      }
  
-     if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY) return;
 -    if (s->avctx->flags & CODEC_FLAG_GRAY)
++    if (CONFIG_GRAY && s->avctx->flags & CODEC_FLAG_GRAY)
+         return;
      /* Chroma MC always uses qpel bilinear */
      uvmx = (uvmx & 3) << 1;
      uvmy = (uvmy & 3) << 1;
@@@ -517,12 -515,21 +518,12 @@@ void ff_vc1_mc_4mv_chroma(VC1Context *v
  
      if (!v->field_mode && !v->s.last_picture.f->data[0])
          return;
-     if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
 -    if (s->avctx->flags & CODEC_FLAG_GRAY)
++    if (CONFIG_GRAY && s->avctx->flags & CODEC_FLAG_GRAY)
          return;
  
 -    for (k = 0; k < 4; k++) {
 -        mvx[k] = s->mv[dir][k][0];
 -        mvy[k] = s->mv[dir][k][1];
 -        intra[k] = v->mb_type[0][s->block_index[k]];
 -        if (v->field_mode)
 -            mv_f[k] = v->mv_f[dir][s->block_index[k] + v->blocks_off];
 -    }
 -
      /* calculate chroma MV vector from four luma MVs */
 -    if (!v->field_mode || (v->field_mode && !v->numref)) {
 -        valid_count = get_chroma_mv(mvx, mvy, intra, 0, &tx, &ty);
 -        chroma_ref_type = v->reffield;
 +    if (!v->field_mode || !v->numref) {
 +        int valid_count = get_chroma_mv(v, dir, &tx, &ty);
          if (!valid_count) {
              s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][0] = 0;
              s->current_picture.motion_val[1][s->block_index[0] + v->blocks_off][1] = 0;
@@@ -654,7 -686,7 +655,7 @@@ void ff_vc1_mc_4mv_chroma4(VC1Context *
      int use_ic;
      uint8_t (*lutuv)[256];
  
-     if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY)
 -    if (s->avctx->flags & CODEC_FLAG_GRAY)
++    if (CONFIG_GRAY && s->avctx->flags & CODEC_FLAG_GRAY)
          return;
  
      for (i = 0; i < 4; i++) {
@@@ -800,7 -837,7 +801,7 @@@ void ff_vc1_interp_mc(VC1Context *v
      }
  
      /* for grayscale we should not try to read from unknown area */
-     if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY) {
 -    if (s->avctx->flags & CODEC_FLAG_GRAY) {
++    if (CONFIG_GRAY && s->avctx->flags & CODEC_FLAG_GRAY) {
          srcU = s->edge_emu_buffer + 18 * s->linesize;
          srcV = s->edge_emu_buffer + 18 * s->linesize;
      }
          dxy = (my & 2) | ((mx & 2) >> 1);
  
          if (!v->rnd)
 -            s->hdsp.avg_pixels_tab[0][dxy](s->dest[0] + off, srcY, s->linesize, 16);
 +            s->hdsp.avg_pixels_tab[0][dxy](s->dest[0], srcY, s->linesize, 16);
          else
 -            s->hdsp.avg_no_rnd_pixels_tab[dxy](s->dest[0] + off, srcY, s->linesize, 16);
 +            s->hdsp.avg_no_rnd_pixels_tab[dxy](s->dest[0], srcY, s->linesize, 16);
      }
  
-     if (CONFIG_GRAY && s->flags & CODEC_FLAG_GRAY) return;
 -    if (s->avctx->flags & CODEC_FLAG_GRAY)
++    if (CONFIG_GRAY && s->avctx->flags & CODEC_FLAG_GRAY)
+         return;
      /* Chroma MC always uses qpel blilinear */
      uvmx = (uvmx & 3) << 1;
      uvmy = (uvmy & 3) << 1;
@@@ -190,7 -183,7 +190,7 @@@ static void vc1_draw_sprites(VC1Contex
      }
      alpha = av_clip_uint16(sd->coefs[1][6]);
  
-     for (plane = 0; plane < (CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY ? 1 : 3); plane++) {
 -    for (plane = 0; plane < (s->avctx->flags & CODEC_FLAG_GRAY ? 1 : 3); plane++) {
++    for (plane = 0; plane < (CONFIG_GRAY && s->avctx->flags & CODEC_FLAG_GRAY ? 1 : 3); plane++) {
          int width = v->output_width>>!!plane;
  
          for (row = 0; row < v->output_height>>!!plane; row++) {
@@@ -311,7 -301,7 +311,7 @@@ static void vc1_sprite_flush(AVCodecCon
         wrong but it looks better than doing nothing. */
  
      if (f && f->data[0])
-         for (plane = 0; plane < (CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY ? 1 : 3); plane++)
 -        for (plane = 0; plane < (s->avctx->flags & CODEC_FLAG_GRAY ? 1 : 3); plane++)
++        for (plane = 0; plane < (CONFIG_GRAY && s->avctx->flags & CODEC_FLAG_GRAY ? 1 : 3); plane++)
              for (i = 0; i < v->sprite_height>>!!plane; i++)
                  memset(f->data[plane] + i * f->linesize[plane],
                         plane ? 128 : 0, f->linesize[plane]);
@@@ -627,11 -594,6 +627,11 @@@ static int vc1_decode_frame(AVCodecCont
          int mby_start;
      } *slices = NULL, *tmp;
  
-     if(s->flags & CODEC_FLAG_LOW_DELAY)
 +    v->second_field = 0;
 +
++    if(s->avctx->flags & CODEC_FLAG_LOW_DELAY)
 +        s->low_delay = 1;
 +
      /* no supplementary picture */
      if (buf_size == 0 || (buf_size == 4 && AV_RB32(buf) == VC1_CODE_ENDOFSEQ)) {
          /* special case for last picture */
@@@ -140,16 -140,24 +140,16 @@@ void ff_mspel_motion(MpegEncContext *s
      w->wdsp.put_mspel_pixels_tab[dxy](dest_y     + 8 * linesize, ptr     + 8 * linesize, linesize);
      w->wdsp.put_mspel_pixels_tab[dxy](dest_y + 8 + 8 * linesize, ptr + 8 + 8 * linesize, linesize);
  
-     if (s->flags & CODEC_FLAG_GRAY)
+     if (s->avctx->flags & CODEC_FLAG_GRAY)
          return;
  
 -    if (s->out_format == FMT_H263) {
 -        dxy = 0;
 -        if ((motion_x & 3) != 0)
 -            dxy |= 1;
 -        if ((motion_y & 3) != 0)
 -            dxy |= 2;
 -        mx = motion_x >> 2;
 -        my = motion_y >> 2;
 -    } else {
 -        mx   = motion_x / 2;
 -        my   = motion_y / 2;
 -        dxy  = ((my & 1) << 1) | (mx & 1);
 -        mx >>= 1;
 -        my >>= 1;
 -    }
 +    dxy = 0;
 +    if ((motion_x & 3) != 0)
 +        dxy |= 1;
 +    if ((motion_y & 3) != 0)
 +        dxy |= 2;
 +    mx = motion_x >> 2;
 +    my = motion_y >> 2;
  
      src_x = s->mb_x * 8 + mx;
      src_y = s->mb_y * 8 + my;
Simple merge