Merge remote-tracking branch 'qatar/master'
authorMichael Niedermayer <michaelni@gmx.at>
Mon, 24 Mar 2014 00:24:55 +0000 (01:24 +0100)
committerMichael Niedermayer <michaelni@gmx.at>
Mon, 24 Mar 2014 00:33:02 +0000 (01:33 +0100)
* qatar/master:
  h264: Split h264 slice decoding from nal decoding

Conflicts:
libavcodec/Makefile
libavcodec/h264.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>
1  2 
libavcodec/Makefile
libavcodec/h264.c
libavcodec/h264.h
libavcodec/h264_picture.c
libavcodec/h264_slice.c

@@@ -233,9 -196,9 +233,10 @@@ OBJS-$(CONFIG_H263_ENCODER)            
                                            h263.o ituh263enc.o flvenc.o
  OBJS-$(CONFIG_H264_DECODER)            += h264.o h264_cabac.o h264_cavlc.o \
                                            h264_direct.o h264_loopfilter.o  \
-                                           h264_mb.o h264_ps.o h264_refs.o  \
-                                           h264_sei.o cabac.o
+                                           h264_mb.o h264_picture.o h264_ps.o \
+                                           h264_refs.o h264_sei.o h264_slice.o \
+                                           cabac.o
 +OBJS-$(CONFIG_H264_VDA_DECODER)        += vda_h264_dec.o
  OBJS-$(CONFIG_HEVC_DECODER)            += hevc.o hevc_mvs.o hevc_ps.o hevc_sei.o \
                                            hevc_cabac.o hevc_refs.o hevcpred.o    \
                                            hevcdsp.o hevc_filter.o cabac.o
  
  #include <assert.h>
  
- static void flush_change(H264Context *h);
  const uint16_t ff_h264_mb_sizes[4] = { 256, 384, 512, 768 };
  
- static const uint8_t rem6[QP_MAX_NUM + 1] = {
-     0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
-     3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
-     0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
-     3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
-     0, 1, 2, 3,
- };
- static const uint8_t div6[QP_MAX_NUM + 1] = {
-     0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3,  3,  3,
-     3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6,  6,  6,
-     7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
-    10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
-    14,14,14,14,
- };
- static const uint8_t field_scan[16+1] = {
-     0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
-     0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
-     2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
-     3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
- };
- static const uint8_t field_scan8x8[64+1] = {
-     0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
-     1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
-     2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
-     0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
-     2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
-     2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
-     2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
-     3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
-     3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
-     4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
-     4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
-     5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
-     5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
-     7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
-     6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
-     7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
- };
- static const uint8_t field_scan8x8_cavlc[64+1] = {
-     0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
-     2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
-     3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
-     5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
-     0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
-     1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
-     3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
-     5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
-     0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
-     1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
-     3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
-     5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
-     1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
-     1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
-     3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
-     6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
- };
- // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
- static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
-     0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
-     4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
-     3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
-     2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
-     1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
-     3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
-     2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
-     3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
-     0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
-     2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
-     1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
-     4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
-     0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
-     1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
-     0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
-     5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
- };
- static const uint8_t dequant4_coeff_init[6][3] = {
-     { 10, 13, 16 },
-     { 11, 14, 18 },
-     { 13, 16, 20 },
-     { 14, 18, 23 },
-     { 16, 20, 25 },
-     { 18, 23, 29 },
- };
- static const uint8_t dequant8_coeff_init_scan[16] = {
-     0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
- };
- static const uint8_t dequant8_coeff_init[6][6] = {
-     { 20, 18, 32, 19, 25, 24 },
-     { 22, 19, 35, 21, 28, 26 },
-     { 26, 23, 42, 24, 33, 31 },
-     { 28, 25, 45, 26, 35, 33 },
-     { 32, 28, 51, 30, 40, 38 },
-     { 36, 32, 58, 34, 46, 43 },
- };
- static const enum AVPixelFormat h264_hwaccel_pixfmt_list_420[] = {
- #if CONFIG_H264_DXVA2_HWACCEL
-     AV_PIX_FMT_DXVA2_VLD,
- #endif
- #if CONFIG_H264_VAAPI_HWACCEL
-     AV_PIX_FMT_VAAPI_VLD,
- #endif
- #if CONFIG_H264_VDA_HWACCEL
-     AV_PIX_FMT_VDA_VLD,
- #endif
- #if CONFIG_H264_VDPAU_HWACCEL
-     AV_PIX_FMT_VDPAU,
- #endif
-     AV_PIX_FMT_YUV420P,
-     AV_PIX_FMT_NONE
- };
- static const enum AVPixelFormat h264_hwaccel_pixfmt_list_jpeg_420[] = {
- #if CONFIG_H264_DXVA2_HWACCEL
-     AV_PIX_FMT_DXVA2_VLD,
- #endif
- #if CONFIG_H264_VAAPI_HWACCEL
-     AV_PIX_FMT_VAAPI_VLD,
- #endif
- #if CONFIG_H264_VDA_HWACCEL
-     AV_PIX_FMT_VDA_VLD,
- #endif
- #if CONFIG_H264_VDPAU_HWACCEL
-     AV_PIX_FMT_VDPAU,
- #endif
-     AV_PIX_FMT_YUVJ420P,
-     AV_PIX_FMT_NONE
- };
 +int avpriv_h264_has_num_reorder_frames(AVCodecContext *avctx)
 +{
 +    H264Context *h = avctx->priv_data;
 +    return h ? h->sps.num_reorder_frames : 0;
 +}
 +
  static void h264_er_decode_mb(void *opaque, int ref, int mv_dir, int mv_type,
                                int (*mv)[2][4][2],
                                int mb_x, int mb_y, int mb_intra, int mb_skipped)
@@@ -1189,50 -670,6 +709,8 @@@ av_cold int ff_h264_decode_init(AVCodec
  
      avctx->internal->allocate_progress = 1;
  
-     flush_change(h);
-     return 0;
- }
- #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
- #undef REBASE_PICTURE
- #define REBASE_PICTURE(pic, new_ctx, old_ctx)             \
-     ((pic && pic >= old_ctx->DPB &&                       \
-       pic < old_ctx->DPB + H264_MAX_PICTURE_COUNT) ?           \
-      &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
- static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
-                                H264Context *new_base,
-                                H264Context *old_base)
- {
-     int i;
-     for (i = 0; i < count; i++) {
-         assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
-                 IN_RANGE(from[i], old_base->DPB,
-                          sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
-                 !from[i]));
-         to[i] = REBASE_PICTURE(from[i], new_base, old_base);
-     }
- }
- static int copy_parameter_set(void **to, void **from, int count, int size)
- {
-     int i;
-     for (i = 0; i < count; i++) {
-         if (to[i] && !from[i]) {
-             av_freep(&to[i]);
-         } else if (from[i] && !to[i]) {
-             to[i] = av_malloc(size);
-             if (!to[i])
-                 return AVERROR(ENOMEM);
-         }
-         if (from[i])
-             memcpy(to[i], from[i], size);
-     }
++    ff_h264_flush_change(h);
 +
      return 0;
  }
  
@@@ -2042,10 -1029,11 +1043,10 @@@ static void idr(H264Context *h
  }
  
  /* forget old pics after a seek */
static void flush_change(H264Context *h)
void ff_h264_flush_change(H264Context *h)
  {
 -    int i;
 -    for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
 -        h->last_pocs[i] = INT_MIN;
 +    int i, j;
 +
      h->outputed_poc          = h->next_outputed_poc = INT_MIN;
      h->prev_interlaced_frame = 1;
      idr(h);
@@@ -2681,1327 -1281,6 +1313,8 @@@ int ff_set_ref_count(H264Context *h
      return 0;
  }
  
- static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
- {
-     switch (a) {
-     case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
-     case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
-     case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
-     default:
-         return a;
-     }
- }
- /**
-  * Decode a slice header.
-  * This will (re)intialize the decoder and call h264_frame_start() as needed.
-  *
-  * @param h h264context
-  * @param h0 h264 master context (differs from 'h' when doing sliced based
-  *           parallel decoding)
-  *
-  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
-  */
- static int decode_slice_header(H264Context *h, H264Context *h0)
- {
-     unsigned int first_mb_in_slice;
-     unsigned int pps_id;
-     int ret;
-     unsigned int slice_type, tmp, i, j;
-     int last_pic_structure, last_pic_droppable;
-     int must_reinit;
-     int needs_reinit = 0;
-     int field_pic_flag, bottom_field_flag;
-     h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
-     h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
-     first_mb_in_slice = get_ue_golomb_long(&h->gb);
-     if (first_mb_in_slice == 0) { // FIXME better field boundary detection
-         if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
-             field_end(h, 1);
-         }
-         h0->current_slice = 0;
-         if (!h0->first_field) {
-             if (h->cur_pic_ptr && !h->droppable) {
-                 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
-                                           h->picture_structure == PICT_BOTTOM_FIELD);
-             }
-             h->cur_pic_ptr = NULL;
-         }
-     }
-     slice_type = get_ue_golomb_31(&h->gb);
-     if (slice_type > 9) {
-         av_log(h->avctx, AV_LOG_ERROR,
-                "slice type %d too large at %d %d\n",
-                slice_type, h->mb_x, h->mb_y);
-         return AVERROR_INVALIDDATA;
-     }
-     if (slice_type > 4) {
-         slice_type -= 5;
-         h->slice_type_fixed = 1;
-     } else
-         h->slice_type_fixed = 0;
-     slice_type = golomb_to_pict_type[slice_type];
-     h->slice_type     = slice_type;
-     h->slice_type_nos = slice_type & 3;
-     if (h->nal_unit_type  == NAL_IDR_SLICE &&
-         h->slice_type_nos != AV_PICTURE_TYPE_I) {
-         av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
-         return AVERROR_INVALIDDATA;
-     }
-     // to make a few old functions happy, it's wrong though
-     h->pict_type = h->slice_type;
-     pps_id = get_ue_golomb(&h->gb);
-     if (pps_id >= MAX_PPS_COUNT) {
-         av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
-         return AVERROR_INVALIDDATA;
-     }
-     if (!h0->pps_buffers[pps_id]) {
-         av_log(h->avctx, AV_LOG_ERROR,
-                "non-existing PPS %u referenced\n",
-                pps_id);
-         return AVERROR_INVALIDDATA;
-     }
-     if (h0->au_pps_id >= 0 && pps_id != h0->au_pps_id) {
-         av_log(h->avctx, AV_LOG_ERROR,
-                "PPS change from %d to %d forbidden\n",
-                h0->au_pps_id, pps_id);
-         return AVERROR_INVALIDDATA;
-     }
-     h->pps = *h0->pps_buffers[pps_id];
-     if (!h0->sps_buffers[h->pps.sps_id]) {
-         av_log(h->avctx, AV_LOG_ERROR,
-                "non-existing SPS %u referenced\n",
-                h->pps.sps_id);
-         return AVERROR_INVALIDDATA;
-     }
-     if (h->pps.sps_id != h->sps.sps_id ||
-         h->pps.sps_id != h->current_sps_id ||
-         h0->sps_buffers[h->pps.sps_id]->new) {
-         h->sps = *h0->sps_buffers[h->pps.sps_id];
-         if (h->mb_width  != h->sps.mb_width ||
-             h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
-             h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
-             h->cur_chroma_format_idc != h->sps.chroma_format_idc
-         )
-             needs_reinit = 1;
-         if (h->bit_depth_luma    != h->sps.bit_depth_luma ||
-             h->chroma_format_idc != h->sps.chroma_format_idc) {
-             h->bit_depth_luma    = h->sps.bit_depth_luma;
-             h->chroma_format_idc = h->sps.chroma_format_idc;
-             needs_reinit         = 1;
-         }
-         if ((ret = h264_set_parameter_from_sps(h)) < 0)
-             return ret;
-     }
-     h->avctx->profile = ff_h264_get_profile(&h->sps);
-     h->avctx->level   = h->sps.level_idc;
-     h->avctx->refs    = h->sps.ref_frame_count;
-     must_reinit = (h->context_initialized &&
-                     (   16*h->sps.mb_width != h->avctx->coded_width
-                      || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
-                      || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
-                      || h->cur_chroma_format_idc != h->sps.chroma_format_idc
-                      || av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio)
-                      || h->mb_width  != h->sps.mb_width
-                      || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
-                     ));
-     if (non_j_pixfmt(h0->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h0, 0)))
-         must_reinit = 1;
-     h->mb_width  = h->sps.mb_width;
-     h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
-     h->mb_num    = h->mb_width * h->mb_height;
-     h->mb_stride = h->mb_width + 1;
-     h->b_stride = h->mb_width * 4;
-     h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
-     h->width  = 16 * h->mb_width;
-     h->height = 16 * h->mb_height;
-     ret = init_dimensions(h);
-     if (ret < 0)
-         return ret;
-     if (h->sps.video_signal_type_present_flag) {
-         h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
-                                                     : AVCOL_RANGE_MPEG;
-         if (h->sps.colour_description_present_flag) {
-             if (h->avctx->colorspace != h->sps.colorspace)
-                 needs_reinit = 1;
-             h->avctx->color_primaries = h->sps.color_primaries;
-             h->avctx->color_trc       = h->sps.color_trc;
-             h->avctx->colorspace      = h->sps.colorspace;
-         }
-     }
-     if (h->context_initialized &&
-         (h->width  != h->avctx->coded_width   ||
-          h->height != h->avctx->coded_height  ||
-          must_reinit ||
-          needs_reinit)) {
-         if (h != h0) {
-             av_log(h->avctx, AV_LOG_ERROR,
-                    "changing width %d -> %d / height %d -> %d on "
-                    "slice %d\n",
-                    h->width, h->avctx->coded_width,
-                    h->height, h->avctx->coded_height,
-                    h0->current_slice + 1);
-             return AVERROR_INVALIDDATA;
-         }
-         flush_change(h);
-         if ((ret = get_pixel_format(h, 1)) < 0)
-             return ret;
-         h->avctx->pix_fmt = ret;
-         av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
-                "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
-         if ((ret = h264_slice_header_init(h, 1)) < 0) {
-             av_log(h->avctx, AV_LOG_ERROR,
-                    "h264_slice_header_init() failed\n");
-             return ret;
-         }
-     }
-     if (!h->context_initialized) {
-         if (h != h0) {
-             av_log(h->avctx, AV_LOG_ERROR,
-                    "Cannot (re-)initialize context during parallel decoding.\n");
-             return AVERROR_PATCHWELCOME;
-         }
-         if ((ret = get_pixel_format(h, 1)) < 0)
-             return ret;
-         h->avctx->pix_fmt = ret;
-         if ((ret = h264_slice_header_init(h, 0)) < 0) {
-             av_log(h->avctx, AV_LOG_ERROR,
-                    "h264_slice_header_init() failed\n");
-             return ret;
-         }
-     }
-     if (h == h0 && h->dequant_coeff_pps != pps_id) {
-         h->dequant_coeff_pps = pps_id;
-         init_dequant_tables(h);
-     }
-     h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
-     h->mb_mbaff        = 0;
-     h->mb_aff_frame    = 0;
-     last_pic_structure = h0->picture_structure;
-     last_pic_droppable = h0->droppable;
-     h->droppable       = h->nal_ref_idc == 0;
-     if (h->sps.frame_mbs_only_flag) {
-         h->picture_structure = PICT_FRAME;
-     } else {
-         if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
-             av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
-             return -1;
-         }
-         field_pic_flag = get_bits1(&h->gb);
-         if (field_pic_flag) {
-             bottom_field_flag = get_bits1(&h->gb);
-             h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
-         } else {
-             h->picture_structure = PICT_FRAME;
-             h->mb_aff_frame      = h->sps.mb_aff;
-         }
-     }
-     h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
-     if (h0->current_slice != 0) {
-         if (last_pic_structure != h->picture_structure ||
-             last_pic_droppable != h->droppable) {
-             av_log(h->avctx, AV_LOG_ERROR,
-                    "Changing field mode (%d -> %d) between slices is not allowed\n",
-                    last_pic_structure, h->picture_structure);
-             h->picture_structure = last_pic_structure;
-             h->droppable         = last_pic_droppable;
-             return AVERROR_INVALIDDATA;
-         } else if (!h0->cur_pic_ptr) {
-             av_log(h->avctx, AV_LOG_ERROR,
-                    "unset cur_pic_ptr on slice %d\n",
-                    h0->current_slice + 1);
-             return AVERROR_INVALIDDATA;
-         }
-     } else {
-         /* Shorten frame num gaps so we don't have to allocate reference
-          * frames just to throw them away */
-         if (h->frame_num != h->prev_frame_num) {
-             int unwrap_prev_frame_num = h->prev_frame_num;
-             int max_frame_num         = 1 << h->sps.log2_max_frame_num;
-             if (unwrap_prev_frame_num > h->frame_num)
-                 unwrap_prev_frame_num -= max_frame_num;
-             if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
-                 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
-                 if (unwrap_prev_frame_num < 0)
-                     unwrap_prev_frame_num += max_frame_num;
-                 h->prev_frame_num = unwrap_prev_frame_num;
-             }
-         }
-         /* See if we have a decoded first field looking for a pair...
-          * Here, we're using that to see if we should mark previously
-          * decode frames as "finished".
-          * We have to do that before the "dummy" in-between frame allocation,
-          * since that can modify h->cur_pic_ptr. */
-         if (h0->first_field) {
-             assert(h0->cur_pic_ptr);
-             assert(h0->cur_pic_ptr->f.buf[0]);
-             assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
-             /* Mark old field/frame as completed */
-             if (h0->cur_pic_ptr->tf.owner == h0->avctx) {
-                 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
-                                           last_pic_structure == PICT_BOTTOM_FIELD);
-             }
-             /* figure out if we have a complementary field pair */
-             if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
-                 /* Previous field is unmatched. Don't display it, but let it
-                  * remain for reference if marked as such. */
-                 if (last_pic_structure != PICT_FRAME) {
-                     ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
-                                               last_pic_structure == PICT_TOP_FIELD);
-                 }
-             } else {
-                 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
-                     /* This and previous field were reference, but had
-                      * different frame_nums. Consider this field first in
-                      * pair. Throw away previous field except for reference
-                      * purposes. */
-                     if (last_pic_structure != PICT_FRAME) {
-                         ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
-                                                   last_pic_structure == PICT_TOP_FIELD);
-                     }
-                 } else {
-                     /* Second field in complementary pair */
-                     if (!((last_pic_structure   == PICT_TOP_FIELD &&
-                            h->picture_structure == PICT_BOTTOM_FIELD) ||
-                           (last_pic_structure   == PICT_BOTTOM_FIELD &&
-                            h->picture_structure == PICT_TOP_FIELD))) {
-                         av_log(h->avctx, AV_LOG_ERROR,
-                                "Invalid field mode combination %d/%d\n",
-                                last_pic_structure, h->picture_structure);
-                         h->picture_structure = last_pic_structure;
-                         h->droppable         = last_pic_droppable;
-                         return AVERROR_INVALIDDATA;
-                     } else if (last_pic_droppable != h->droppable) {
-                         avpriv_request_sample(h->avctx,
-                                               "Found reference and non-reference fields in the same frame, which");
-                         h->picture_structure = last_pic_structure;
-                         h->droppable         = last_pic_droppable;
-                         return AVERROR_PATCHWELCOME;
-                     }
-                 }
-             }
-         }
-         while (h->frame_num != h->prev_frame_num && !h0->first_field &&
-                h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
-             H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
-             av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
-                    h->frame_num, h->prev_frame_num);
-             if (!h->sps.gaps_in_frame_num_allowed_flag)
-                 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
-                     h->last_pocs[i] = INT_MIN;
-             ret = h264_frame_start(h);
-             if (ret < 0) {
-                 h0->first_field = 0;
-                 return ret;
-             }
-             h->prev_frame_num++;
-             h->prev_frame_num        %= 1 << h->sps.log2_max_frame_num;
-             h->cur_pic_ptr->frame_num = h->prev_frame_num;
-             h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
-             ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
-             ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
-             ret = ff_generate_sliding_window_mmcos(h, 1);
-             if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
-                 return ret;
-             ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
-             if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
-                 return ret;
-             /* Error concealment: If a ref is missing, copy the previous ref
-              * in its place.
-              * FIXME: Avoiding a memcpy would be nice, but ref handling makes
-              * many assumptions about there being no actual duplicates.
-              * FIXME: This does not copy padding for out-of-frame motion
-              * vectors.  Given we are concealing a lost frame, this probably
-              * is not noticeable by comparison, but it should be fixed. */
-             if (h->short_ref_count) {
-                 if (prev) {
-                     av_image_copy(h->short_ref[0]->f.data,
-                                   h->short_ref[0]->f.linesize,
-                                   (const uint8_t **)prev->f.data,
-                                   prev->f.linesize,
-                                   h->avctx->pix_fmt,
-                                   h->mb_width  * 16,
-                                   h->mb_height * 16);
-                     h->short_ref[0]->poc = prev->poc + 2;
-                 }
-                 h->short_ref[0]->frame_num = h->prev_frame_num;
-             }
-         }
-         /* See if we have a decoded first field looking for a pair...
-          * We're using that to see whether to continue decoding in that
-          * frame, or to allocate a new one. */
-         if (h0->first_field) {
-             assert(h0->cur_pic_ptr);
-             assert(h0->cur_pic_ptr->f.buf[0]);
-             assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
-             /* figure out if we have a complementary field pair */
-             if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
-                 /* Previous field is unmatched. Don't display it, but let it
-                  * remain for reference if marked as such. */
-                 h0->cur_pic_ptr = NULL;
-                 h0->first_field = FIELD_PICTURE(h);
-             } else {
-                 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
-                     ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
-                                               h0->picture_structure==PICT_BOTTOM_FIELD);
-                     /* This and the previous field had different frame_nums.
-                      * Consider this field first in pair. Throw away previous
-                      * one except for reference purposes. */
-                     h0->first_field = 1;
-                     h0->cur_pic_ptr = NULL;
-                 } else {
-                     /* Second field in complementary pair */
-                     h0->first_field = 0;
-                 }
-             }
-         } else {
-             /* Frame or first field in a potentially complementary pair */
-             h0->first_field = FIELD_PICTURE(h);
-         }
-         if (!FIELD_PICTURE(h) || h0->first_field) {
-             if (h264_frame_start(h) < 0) {
-                 h0->first_field = 0;
-                 return AVERROR_INVALIDDATA;
-             }
-         } else {
-             release_unused_pictures(h, 0);
-         }
-         /* Some macroblocks can be accessed before they're available in case
-         * of lost slices, MBAFF or threading. */
-         if (FIELD_PICTURE(h)) {
-             for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
-                 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
-         } else {
-             memset(h->slice_table, -1,
-                 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
-         }
-         h0->last_slice_type = -1;
-     }
-     if (h != h0 && (ret = clone_slice(h, h0)) < 0)
-         return ret;
-     /* can't be in alloc_tables because linesize isn't known there.
-      * FIXME: redo bipred weight to not require extra buffer? */
-     for (i = 0; i < h->slice_context_count; i++)
-         if (h->thread_context[i]) {
-             ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
-             if (ret < 0)
-                 return ret;
-         }
-     h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
-     av_assert1(h->mb_num == h->mb_width * h->mb_height);
-     if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
-         first_mb_in_slice >= h->mb_num) {
-         av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
-         return AVERROR_INVALIDDATA;
-     }
-     h->resync_mb_x = h->mb_x =  first_mb_in_slice % h->mb_width;
-     h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) <<
-                                FIELD_OR_MBAFF_PICTURE(h);
-     if (h->picture_structure == PICT_BOTTOM_FIELD)
-         h->resync_mb_y = h->mb_y = h->mb_y + 1;
-     av_assert1(h->mb_y < h->mb_height);
-     if (h->picture_structure == PICT_FRAME) {
-         h->curr_pic_num = h->frame_num;
-         h->max_pic_num  = 1 << h->sps.log2_max_frame_num;
-     } else {
-         h->curr_pic_num = 2 * h->frame_num + 1;
-         h->max_pic_num  = 1 << (h->sps.log2_max_frame_num + 1);
-     }
-     if (h->nal_unit_type == NAL_IDR_SLICE)
-         get_ue_golomb(&h->gb); /* idr_pic_id */
-     if (h->sps.poc_type == 0) {
-         h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
-         if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
-             h->delta_poc_bottom = get_se_golomb(&h->gb);
-     }
-     if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
-         h->delta_poc[0] = get_se_golomb(&h->gb);
-         if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
-             h->delta_poc[1] = get_se_golomb(&h->gb);
-     }
-     ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
-     if (h->pps.redundant_pic_cnt_present)
-         h->redundant_pic_count = get_ue_golomb(&h->gb);
-     ret = ff_set_ref_count(h);
-     if (ret < 0)
-         return ret;
-     if (slice_type != AV_PICTURE_TYPE_I &&
-         (h0->current_slice == 0 ||
-          slice_type != h0->last_slice_type ||
-          memcmp(h0->last_ref_count, h0->ref_count, sizeof(h0->ref_count)))) {
-         ff_h264_fill_default_ref_list(h);
-     }
-     if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
-        ret = ff_h264_decode_ref_pic_list_reordering(h);
-        if (ret < 0) {
-            h->ref_count[1] = h->ref_count[0] = 0;
-            return ret;
-        }
-     }
-     if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
-         (h->pps.weighted_bipred_idc == 1 &&
-          h->slice_type_nos == AV_PICTURE_TYPE_B))
-         ff_pred_weight_table(h);
-     else if (h->pps.weighted_bipred_idc == 2 &&
-              h->slice_type_nos == AV_PICTURE_TYPE_B) {
-         implicit_weight_table(h, -1);
-     } else {
-         h->use_weight = 0;
-         for (i = 0; i < 2; i++) {
-             h->luma_weight_flag[i]   = 0;
-             h->chroma_weight_flag[i] = 0;
-         }
-     }
-     // If frame-mt is enabled, only update mmco tables for the first slice
-     // in a field. Subsequent slices can temporarily clobber h->mmco_index
-     // or h->mmco, which will cause ref list mix-ups and decoding errors
-     // further down the line. This may break decoding if the first slice is
-     // corrupt, thus we only do this if frame-mt is enabled.
-     if (h->nal_ref_idc) {
-         ret = ff_h264_decode_ref_pic_marking(h0, &h->gb,
-                                              !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
-                                              h0->current_slice == 0);
-         if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
-             return AVERROR_INVALIDDATA;
-     }
-     if (FRAME_MBAFF(h)) {
-         ff_h264_fill_mbaff_ref_list(h);
-         if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
-             implicit_weight_table(h, 0);
-             implicit_weight_table(h, 1);
-         }
-     }
-     if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
-         ff_h264_direct_dist_scale_factor(h);
-     ff_h264_direct_ref_list_init(h);
-     if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
-         tmp = get_ue_golomb_31(&h->gb);
-         if (tmp > 2) {
-             av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
-             return AVERROR_INVALIDDATA;
-         }
-         h->cabac_init_idc = tmp;
-     }
-     h->last_qscale_diff = 0;
-     tmp = h->pps.init_qp + get_se_golomb(&h->gb);
-     if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
-         av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
-         return AVERROR_INVALIDDATA;
-     }
-     h->qscale       = tmp;
-     h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
-     h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
-     // FIXME qscale / qp ... stuff
-     if (h->slice_type == AV_PICTURE_TYPE_SP)
-         get_bits1(&h->gb); /* sp_for_switch_flag */
-     if (h->slice_type == AV_PICTURE_TYPE_SP ||
-         h->slice_type == AV_PICTURE_TYPE_SI)
-         get_se_golomb(&h->gb); /* slice_qs_delta */
-     h->deblocking_filter     = 1;
-     h->slice_alpha_c0_offset = 0;
-     h->slice_beta_offset     = 0;
-     if (h->pps.deblocking_filter_parameters_present) {
-         tmp = get_ue_golomb_31(&h->gb);
-         if (tmp > 2) {
-             av_log(h->avctx, AV_LOG_ERROR,
-                    "deblocking_filter_idc %u out of range\n", tmp);
-             return AVERROR_INVALIDDATA;
-         }
-         h->deblocking_filter = tmp;
-         if (h->deblocking_filter < 2)
-             h->deblocking_filter ^= 1;  // 1<->0
-         if (h->deblocking_filter) {
-             h->slice_alpha_c0_offset = get_se_golomb(&h->gb) * 2;
-             h->slice_beta_offset     = get_se_golomb(&h->gb) * 2;
-             if (h->slice_alpha_c0_offset >  12 ||
-                 h->slice_alpha_c0_offset < -12 ||
-                 h->slice_beta_offset >  12     ||
-                 h->slice_beta_offset < -12) {
-                 av_log(h->avctx, AV_LOG_ERROR,
-                        "deblocking filter parameters %d %d out of range\n",
-                        h->slice_alpha_c0_offset, h->slice_beta_offset);
-                 return AVERROR_INVALIDDATA;
-             }
-         }
-     }
-     if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
-         (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
-          h->slice_type_nos != AV_PICTURE_TYPE_I) ||
-         (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR  &&
-          h->slice_type_nos == AV_PICTURE_TYPE_B) ||
-         (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
-          h->nal_ref_idc == 0))
-         h->deblocking_filter = 0;
-     if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
-         if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
-             /* Cheat slightly for speed:
-              * Do not bother to deblock across slices. */
-             h->deblocking_filter = 2;
-         } else {
-             h0->max_contexts = 1;
-             if (!h0->single_decode_warning) {
-                 av_log(h->avctx, AV_LOG_INFO,
-                        "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
-                 h0->single_decode_warning = 1;
-             }
-             if (h != h0) {
-                 av_log(h->avctx, AV_LOG_ERROR,
-                        "Deblocking switched inside frame.\n");
-                 return 1;
-             }
-         }
-     }
-     h->qp_thresh = 15 -
-                    FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
-                    FFMAX3(0,
-                           h->pps.chroma_qp_index_offset[0],
-                           h->pps.chroma_qp_index_offset[1]) +
-                    6 * (h->sps.bit_depth_luma - 8);
-     h0->last_slice_type = slice_type;
-     memcpy(h0->last_ref_count, h0->ref_count, sizeof(h0->last_ref_count));
-     h->slice_num        = ++h0->current_slice;
-     if (h->slice_num)
-         h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= h->resync_mb_y;
-     if (   h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= h->resync_mb_y
-         && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= h->resync_mb_y
-         && h->slice_num >= MAX_SLICES) {
-         //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
-         av_log(h->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", h->slice_num, MAX_SLICES);
-     }
-     for (j = 0; j < 2; j++) {
-         int id_list[16];
-         int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
-         for (i = 0; i < 16; i++) {
-             id_list[i] = 60;
-             if (j < h->list_count && i < h->ref_count[j] &&
-                 h->ref_list[j][i].f.buf[0]) {
-                 int k;
-                 AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer;
-                 for (k = 0; k < h->short_ref_count; k++)
-                     if (h->short_ref[k]->f.buf[0]->buffer == buf) {
-                         id_list[i] = k;
-                         break;
-                     }
-                 for (k = 0; k < h->long_ref_count; k++)
-                     if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
-                         id_list[i] = h->short_ref_count + k;
-                         break;
-                     }
-             }
-         }
-         ref2frm[0] =
-         ref2frm[1] = -1;
-         for (i = 0; i < 16; i++)
-             ref2frm[i + 2] = 4 * id_list[i] + (h->ref_list[j][i].reference & 3);
-         ref2frm[18 + 0] =
-         ref2frm[18 + 1] = -1;
-         for (i = 16; i < 48; i++)
-             ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
-                              (h->ref_list[j][i].reference & 3);
-     }
-     if (h->ref_count[0]) h264_set_erpic(&h->er.last_pic, &h->ref_list[0][0]);
-     if (h->ref_count[1]) h264_set_erpic(&h->er.next_pic, &h->ref_list[1][0]);
-     h->er.ref_count = h->ref_count[0];
-     h0->au_pps_id = pps_id;
-     h->sps.new =
-     h0->sps_buffers[h->pps.sps_id]->new = 0;
-     h->current_sps_id = h->pps.sps_id;
-     if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
-         av_log(h->avctx, AV_LOG_DEBUG,
-                "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
-                h->slice_num,
-                (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
-                first_mb_in_slice,
-                av_get_picture_type_char(h->slice_type),
-                h->slice_type_fixed ? " fix" : "",
-                h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
-                pps_id, h->frame_num,
-                h->cur_pic_ptr->field_poc[0],
-                h->cur_pic_ptr->field_poc[1],
-                h->ref_count[0], h->ref_count[1],
-                h->qscale,
-                h->deblocking_filter,
-                h->slice_alpha_c0_offset, h->slice_beta_offset,
-                h->use_weight,
-                h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
-                h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
-     }
-     return 0;
- }
- int ff_h264_get_slice_type(const H264Context *h)
- {
-     switch (h->slice_type) {
-     case AV_PICTURE_TYPE_P:
-         return 0;
-     case AV_PICTURE_TYPE_B:
-         return 1;
-     case AV_PICTURE_TYPE_I:
-         return 2;
-     case AV_PICTURE_TYPE_SP:
-         return 3;
-     case AV_PICTURE_TYPE_SI:
-         return 4;
-     default:
-         return AVERROR_INVALIDDATA;
-     }
- }
- static av_always_inline void fill_filter_caches_inter(H264Context *h,
-                                                       int mb_type, int top_xy,
-                                                       int left_xy[LEFT_MBS],
-                                                       int top_type,
-                                                       int left_type[LEFT_MBS],
-                                                       int mb_xy, int list)
- {
-     int b_stride = h->b_stride;
-     int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
-     int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
-     if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
-         if (USES_LIST(top_type, list)) {
-             const int b_xy  = h->mb2b_xy[top_xy] + 3 * b_stride;
-             const int b8_xy = 4 * top_xy + 2;
-             int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
-             AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
-             ref_cache[0 - 1 * 8] =
-             ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
-             ref_cache[2 - 1 * 8] =
-             ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
-         } else {
-             AV_ZERO128(mv_dst - 1 * 8);
-             AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
-         }
-         if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
-             if (USES_LIST(left_type[LTOP], list)) {
-                 const int b_xy  = h->mb2b_xy[left_xy[LTOP]] + 3;
-                 const int b8_xy = 4 * left_xy[LTOP] + 1;
-                 int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
-                 AV_COPY32(mv_dst - 1 +  0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
-                 AV_COPY32(mv_dst - 1 +  8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
-                 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
-                 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
-                 ref_cache[-1 +  0] =
-                 ref_cache[-1 +  8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
-                 ref_cache[-1 + 16] =
-                 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
-             } else {
-                 AV_ZERO32(mv_dst - 1 +  0);
-                 AV_ZERO32(mv_dst - 1 +  8);
-                 AV_ZERO32(mv_dst - 1 + 16);
-                 AV_ZERO32(mv_dst - 1 + 24);
-                 ref_cache[-1 +  0] =
-                 ref_cache[-1 +  8] =
-                 ref_cache[-1 + 16] =
-                 ref_cache[-1 + 24] = LIST_NOT_USED;
-             }
-         }
-     }
-     if (!USES_LIST(mb_type, list)) {
-         fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
-         AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
-         AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
-         AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
-         AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
-         return;
-     }
-     {
-         int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
-         int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
-         uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
-         uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
-         AV_WN32A(&ref_cache[0 * 8], ref01);
-         AV_WN32A(&ref_cache[1 * 8], ref01);
-         AV_WN32A(&ref_cache[2 * 8], ref23);
-         AV_WN32A(&ref_cache[3 * 8], ref23);
-     }
-     {
-         int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
-         AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
-         AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
-         AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
-         AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
-     }
- }
- /**
-  *
-  * @return non zero if the loop filter can be skipped
-  */
- static int fill_filter_caches(H264Context *h, int mb_type)
- {
-     const int mb_xy = h->mb_xy;
-     int top_xy, left_xy[LEFT_MBS];
-     int top_type, left_type[LEFT_MBS];
-     uint8_t *nnz;
-     uint8_t *nnz_cache;
-     top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
-     /* Wow, what a mess, why didn't they simplify the interlacing & intra
-      * stuff, I can't imagine that these complex rules are worth it. */
-     left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
-     if (FRAME_MBAFF(h)) {
-         const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
-         const int curr_mb_field_flag = IS_INTERLACED(mb_type);
-         if (h->mb_y & 1) {
-             if (left_mb_field_flag != curr_mb_field_flag)
-                 left_xy[LTOP] -= h->mb_stride;
-         } else {
-             if (curr_mb_field_flag)
-                 top_xy += h->mb_stride &
-                           (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
-             if (left_mb_field_flag != curr_mb_field_flag)
-                 left_xy[LBOT] += h->mb_stride;
-         }
-     }
-     h->top_mb_xy        = top_xy;
-     h->left_mb_xy[LTOP] = left_xy[LTOP];
-     h->left_mb_xy[LBOT] = left_xy[LBOT];
-     {
-         /* For sufficiently low qp, filtering wouldn't do anything.
-          * This is a conservative estimate: could also check beta_offset
-          * and more accurate chroma_qp. */
-         int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
-         int qp        = h->cur_pic.qscale_table[mb_xy];
-         if (qp <= qp_thresh &&
-             (left_xy[LTOP] < 0 ||
-              ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
-             (top_xy < 0 ||
-              ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
-             if (!FRAME_MBAFF(h))
-                 return 1;
-             if ((left_xy[LTOP] < 0 ||
-                  ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
-                 (top_xy < h->mb_stride ||
-                  ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
-                 return 1;
-         }
-     }
-     top_type        = h->cur_pic.mb_type[top_xy];
-     left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
-     left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
-     if (h->deblocking_filter == 2) {
-         if (h->slice_table[top_xy] != h->slice_num)
-             top_type = 0;
-         if (h->slice_table[left_xy[LBOT]] != h->slice_num)
-             left_type[LTOP] = left_type[LBOT] = 0;
-     } else {
-         if (h->slice_table[top_xy] == 0xFFFF)
-             top_type = 0;
-         if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
-             left_type[LTOP] = left_type[LBOT] = 0;
-     }
-     h->top_type        = top_type;
-     h->left_type[LTOP] = left_type[LTOP];
-     h->left_type[LBOT] = left_type[LBOT];
-     if (IS_INTRA(mb_type))
-         return 0;
-     fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
-                              top_type, left_type, mb_xy, 0);
-     if (h->list_count == 2)
-         fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
-                                  top_type, left_type, mb_xy, 1);
-     nnz       = h->non_zero_count[mb_xy];
-     nnz_cache = h->non_zero_count_cache;
-     AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
-     AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
-     AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
-     AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
-     h->cbp = h->cbp_table[mb_xy];
-     if (top_type) {
-         nnz = h->non_zero_count[top_xy];
-         AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
-     }
-     if (left_type[LTOP]) {
-         nnz = h->non_zero_count[left_xy[LTOP]];
-         nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
-         nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
-         nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
-         nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
-     }
-     /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
-      * from what the loop filter needs */
-     if (!CABAC(h) && h->pps.transform_8x8_mode) {
-         if (IS_8x8DCT(top_type)) {
-             nnz_cache[4 + 8 * 0] =
-             nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
-             nnz_cache[6 + 8 * 0] =
-             nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
-         }
-         if (IS_8x8DCT(left_type[LTOP])) {
-             nnz_cache[3 + 8 * 1] =
-             nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
-         }
-         if (IS_8x8DCT(left_type[LBOT])) {
-             nnz_cache[3 + 8 * 3] =
-             nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
-         }
-         if (IS_8x8DCT(mb_type)) {
-             nnz_cache[scan8[0]] =
-             nnz_cache[scan8[1]] =
-             nnz_cache[scan8[2]] =
-             nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
-             nnz_cache[scan8[0 + 4]] =
-             nnz_cache[scan8[1 + 4]] =
-             nnz_cache[scan8[2 + 4]] =
-             nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
-             nnz_cache[scan8[0 + 8]] =
-             nnz_cache[scan8[1 + 8]] =
-             nnz_cache[scan8[2 + 8]] =
-             nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
-             nnz_cache[scan8[0 + 12]] =
-             nnz_cache[scan8[1 + 12]] =
-             nnz_cache[scan8[2 + 12]] =
-             nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
-         }
-     }
-     return 0;
- }
- static void loop_filter(H264Context *h, int start_x, int end_x)
- {
-     uint8_t *dest_y, *dest_cb, *dest_cr;
-     int linesize, uvlinesize, mb_x, mb_y;
-     const int end_mb_y       = h->mb_y + FRAME_MBAFF(h);
-     const int old_slice_type = h->slice_type;
-     const int pixel_shift    = h->pixel_shift;
-     const int block_h        = 16 >> h->chroma_y_shift;
-     if (h->deblocking_filter) {
-         for (mb_x = start_x; mb_x < end_x; mb_x++)
-             for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
-                 int mb_xy, mb_type;
-                 mb_xy         = h->mb_xy = mb_x + mb_y * h->mb_stride;
-                 h->slice_num  = h->slice_table[mb_xy];
-                 mb_type       = h->cur_pic.mb_type[mb_xy];
-                 h->list_count = h->list_counts[mb_xy];
-                 if (FRAME_MBAFF(h))
-                     h->mb_mbaff               =
-                     h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
-                 h->mb_x = mb_x;
-                 h->mb_y = mb_y;
-                 dest_y  = h->cur_pic.f.data[0] +
-                           ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
-                 dest_cb = h->cur_pic.f.data[1] +
-                           (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
-                           mb_y * h->uvlinesize * block_h;
-                 dest_cr = h->cur_pic.f.data[2] +
-                           (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
-                           mb_y * h->uvlinesize * block_h;
-                 // FIXME simplify above
-                 if (MB_FIELD(h)) {
-                     linesize   = h->mb_linesize   = h->linesize   * 2;
-                     uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2;
-                     if (mb_y & 1) { // FIXME move out of this function?
-                         dest_y  -= h->linesize   * 15;
-                         dest_cb -= h->uvlinesize * (block_h - 1);
-                         dest_cr -= h->uvlinesize * (block_h - 1);
-                     }
-                 } else {
-                     linesize   = h->mb_linesize   = h->linesize;
-                     uvlinesize = h->mb_uvlinesize = h->uvlinesize;
-                 }
-                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
-                                  uvlinesize, 0);
-                 if (fill_filter_caches(h, mb_type))
-                     continue;
-                 h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
-                 h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
-                 if (FRAME_MBAFF(h)) {
-                     ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
-                                       linesize, uvlinesize);
-                 } else {
-                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
-                                            dest_cr, linesize, uvlinesize);
-                 }
-             }
-     }
-     h->slice_type   = old_slice_type;
-     h->mb_x         = end_x;
-     h->mb_y         = end_mb_y - FRAME_MBAFF(h);
-     h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
-     h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
- }
- static void predict_field_decoding_flag(H264Context *h)
- {
-     const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
-     int mb_type     = (h->slice_table[mb_xy - 1] == h->slice_num) ?
-                       h->cur_pic.mb_type[mb_xy - 1] :
-                       (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ?
-                       h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
-     h->mb_mbaff     = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
- }
- /**
-  * Draw edges and report progress for the last MB row.
-  */
- static void decode_finish_row(H264Context *h)
- {
-     int top            = 16 * (h->mb_y      >> FIELD_PICTURE(h));
-     int pic_height     = 16 *  h->mb_height >> FIELD_PICTURE(h);
-     int height         =  16      << FRAME_MBAFF(h);
-     int deblock_border = (16 + 4) << FRAME_MBAFF(h);
-     if (h->deblocking_filter) {
-         if ((top + height) >= pic_height)
-             height += deblock_border;
-         top -= deblock_border;
-     }
-     if (top >= pic_height || (top + height) < 0)
-         return;
-     height = FFMIN(height, pic_height - top);
-     if (top < 0) {
-         height = top + height;
-         top    = 0;
-     }
-     ff_h264_draw_horiz_band(h, top, height);
-     if (h->droppable || h->er.error_occurred)
-         return;
-     ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
-                               h->picture_structure == PICT_BOTTOM_FIELD);
- }
- static void er_add_slice(H264Context *h, int startx, int starty,
-                          int endx, int endy, int status)
- {
-     if (CONFIG_ERROR_RESILIENCE) {
-         ERContext *er = &h->er;
-         ff_er_add_slice(er, startx, starty, endx, endy, status);
-     }
- }
- static int decode_slice(struct AVCodecContext *avctx, void *arg)
- {
-     H264Context *h = *(void **)arg;
-     int lf_x_start = h->mb_x;
-     h->mb_skip_run = -1;
-     av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3));
-     h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
-                     avctx->codec_id != AV_CODEC_ID_H264 ||
-                     (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
-     if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->er.error_status_table) {
-         const int start_i  = av_clip(h->resync_mb_x + h->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
-         if (start_i) {
-             int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]];
-             prev_status &= ~ VP_START;
-             if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
-                 h->er.error_occurred = 1;
-         }
-     }
-     if (h->pps.cabac) {
-         /* realign */
-         align_get_bits(&h->gb);
-         /* init cabac */
-         ff_init_cabac_decoder(&h->cabac,
-                               h->gb.buffer + get_bits_count(&h->gb) / 8,
-                               (get_bits_left(&h->gb) + 7) / 8);
-         ff_h264_init_cabac_states(h);
-         for (;;) {
-             // START_TIMER
-             int ret = ff_h264_decode_mb_cabac(h);
-             int eos;
-             // STOP_TIMER("decode_mb_cabac")
-             if (ret >= 0)
-                 ff_h264_hl_decode_mb(h);
-             // FIXME optimal? or let mb_decode decode 16x32 ?
-             if (ret >= 0 && FRAME_MBAFF(h)) {
-                 h->mb_y++;
-                 ret = ff_h264_decode_mb_cabac(h);
-                 if (ret >= 0)
-                     ff_h264_hl_decode_mb(h);
-                 h->mb_y--;
-             }
-             eos = get_cabac_terminate(&h->cabac);
-             if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
-                 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
-                 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
-                              h->mb_y, ER_MB_END);
-                 if (h->mb_x >= lf_x_start)
-                     loop_filter(h, lf_x_start, h->mb_x + 1);
-                 return 0;
-             }
-             if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
-                 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %td\n", h->cabac.bytestream_end - h->cabac.bytestream);
-             if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
-                 av_log(h->avctx, AV_LOG_ERROR,
-                        "error while decoding MB %d %d, bytestream %td\n",
-                        h->mb_x, h->mb_y,
-                        h->cabac.bytestream_end - h->cabac.bytestream);
-                 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
-                              h->mb_y, ER_MB_ERROR);
-                 return AVERROR_INVALIDDATA;
-             }
-             if (++h->mb_x >= h->mb_width) {
-                 loop_filter(h, lf_x_start, h->mb_x);
-                 h->mb_x = lf_x_start = 0;
-                 decode_finish_row(h);
-                 ++h->mb_y;
-                 if (FIELD_OR_MBAFF_PICTURE(h)) {
-                     ++h->mb_y;
-                     if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
-                         predict_field_decoding_flag(h);
-                 }
-             }
-             if (eos || h->mb_y >= h->mb_height) {
-                 tprintf(h->avctx, "slice end %d %d\n",
-                         get_bits_count(&h->gb), h->gb.size_in_bits);
-                 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
-                              h->mb_y, ER_MB_END);
-                 if (h->mb_x > lf_x_start)
-                     loop_filter(h, lf_x_start, h->mb_x);
-                 return 0;
-             }
-         }
-     } else {
-         for (;;) {
-             int ret = ff_h264_decode_mb_cavlc(h);
-             if (ret >= 0)
-                 ff_h264_hl_decode_mb(h);
-             // FIXME optimal? or let mb_decode decode 16x32 ?
-             if (ret >= 0 && FRAME_MBAFF(h)) {
-                 h->mb_y++;
-                 ret = ff_h264_decode_mb_cavlc(h);
-                 if (ret >= 0)
-                     ff_h264_hl_decode_mb(h);
-                 h->mb_y--;
-             }
-             if (ret < 0) {
-                 av_log(h->avctx, AV_LOG_ERROR,
-                        "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
-                 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
-                              h->mb_y, ER_MB_ERROR);
-                 return ret;
-             }
-             if (++h->mb_x >= h->mb_width) {
-                 loop_filter(h, lf_x_start, h->mb_x);
-                 h->mb_x = lf_x_start = 0;
-                 decode_finish_row(h);
-                 ++h->mb_y;
-                 if (FIELD_OR_MBAFF_PICTURE(h)) {
-                     ++h->mb_y;
-                     if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
-                         predict_field_decoding_flag(h);
-                 }
-                 if (h->mb_y >= h->mb_height) {
-                     tprintf(h->avctx, "slice end %d %d\n",
-                             get_bits_count(&h->gb), h->gb.size_in_bits);
-                     if (   get_bits_left(&h->gb) == 0
-                         || get_bits_left(&h->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
-                         er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
-                                      h->mb_x - 1, h->mb_y,
-                                      ER_MB_END);
-                         return 0;
-                     } else {
-                         er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
-                                      h->mb_x, h->mb_y,
-                                      ER_MB_END);
-                         return AVERROR_INVALIDDATA;
-                     }
-                 }
-             }
-             if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
-                 tprintf(h->avctx, "slice end %d %d\n",
-                         get_bits_count(&h->gb), h->gb.size_in_bits);
-                 if (get_bits_left(&h->gb) == 0) {
-                     er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
-                                  h->mb_x - 1, h->mb_y,
-                                  ER_MB_END);
-                     if (h->mb_x > lf_x_start)
-                         loop_filter(h, lf_x_start, h->mb_x);
-                     return 0;
-                 } else {
-                     er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
-                                  h->mb_y, ER_MB_ERROR);
-                     return AVERROR_INVALIDDATA;
-                 }
-             }
-         }
-     }
- }
- /**
-  * Call decode_slice() for each context.
-  *
-  * @param h h264 master context
-  * @param context_count number of contexts to execute
-  */
- static int execute_decode_slices(H264Context *h, unsigned context_count)
- {
-     AVCodecContext *const avctx = h->avctx;
-     H264Context *hx;
-     int i;
-     av_assert0(h->mb_y < h->mb_height);
-     if (h->avctx->hwaccel ||
-         h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
-         return 0;
-     if (context_count == 1) {
-         return decode_slice(avctx, &h);
-     } else {
-         av_assert0(context_count > 0);
-         for (i = 1; i < context_count; i++) {
-             hx                 = h->thread_context[i];
-             if (CONFIG_ERROR_RESILIENCE) {
-                 hx->er.error_count = 0;
-             }
-             hx->x264_build     = h->x264_build;
-         }
-         avctx->execute(avctx, decode_slice, h->thread_context,
-                        NULL, context_count, sizeof(void *));
-         /* pull back stuff from slices to master context */
-         hx                   = h->thread_context[context_count - 1];
-         h->mb_x              = hx->mb_x;
-         h->mb_y              = hx->mb_y;
-         h->droppable         = hx->droppable;
-         h->picture_structure = hx->picture_structure;
-         if (CONFIG_ERROR_RESILIENCE) {
-             for (i = 1; i < context_count; i++)
-                 h->er.error_count += h->thread_context[i]->er.error_count;
-         }
-     }
-     return 0;
- }
 +static const uint8_t start_code[] = { 0x00, 0x00, 0x01 };
 +
  static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size,
                              int parse_extradata)
  {
@@@ -4193,21 -1436,12 +1506,21 @@@ again
                  hx->inter_gb_ptr      = &hx->gb;
                  hx->data_partitioning = 0;
  
-                 if ((err = decode_slice_header(hx, h)))
+                 if ((err = ff_h264_decode_slice_header(hx, h)))
                      break;
  
 -                if (h->sei_recovery_frame_cnt >= 0 && h->recovery_frame < 0) {
 -                    h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) &
 -                                        ((1 << h->sps.log2_max_frame_num) - 1);
 +                if (h->sei_recovery_frame_cnt >= 0) {
 +                    if (h->frame_num != h->sei_recovery_frame_cnt || hx->slice_type_nos != AV_PICTURE_TYPE_I)
 +                        h->valid_recovery_point = 1;
 +
 +                    if (   h->recovery_frame < 0
 +                        || ((h->recovery_frame - h->frame_num) & ((1 << h->sps.log2_max_frame_num)-1)) > h->sei_recovery_frame_cnt) {
 +                        h->recovery_frame = (h->frame_num + h->sei_recovery_frame_cnt) &
 +                                            ((1 << h->sps.log2_max_frame_num) - 1);
 +
 +                        if (!h->valid_recovery_point)
 +                            h->recovery_frame = h->frame_num;
 +                    }
                  }
  
                  h->cur_pic_ptr->f.key_frame |=
@@@ -4515,11 -1702,10 +1828,11 @@@ not_extra
          if (avctx->flags2 & CODEC_FLAG2_CHUNKS)
              decode_postinit(h, 1);
  
-         field_end(h, 0);
+         ff_h264_field_end(h, 0);
  
 +        /* Wait for second field. */
          *got_frame = 0;
 -        if (h->next_output_pic && ((avctx->flags & CODEC_FLAG_OUTPUT_CORRUPT) ||
 +        if (h->next_output_pic && (
                                     h->next_output_pic->recovered)) {
              if (!h->next_output_pic->recovered)
                  h->next_output_pic->f.flags |= AV_FRAME_FLAG_CORRUPT;
@@@ -4558,10 -1739,9 +1871,10 @@@ static av_cold int h264_decode_end(AVCo
  {
      H264Context *h = avctx->priv_data;
  
 +    ff_h264_remove_all_refs(h);
      ff_h264_free_context(h);
  
-     unref_picture(h, &h->cur_pic);
+     ff_h264_unref_picture(h, &h->cur_pic);
  
      return 0;
  }
@@@ -4617,26 -1777,6 +1930,26 @@@ AVCodec ff_h264_decoder = 
                               CODEC_CAP_FRAME_THREADS,
      .flush                 = flush_dpb,
      .init_thread_copy      = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
-     .update_thread_context = ONLY_IF_THREADS_ENABLED(decode_update_thread_context),
+     .update_thread_context = ONLY_IF_THREADS_ENABLED(ff_h264_update_thread_context),
      .profiles              = NULL_IF_CONFIG_SMALL(profiles),
 +    .priv_class            = &h264_class,
  };
 +
 +#if CONFIG_H264_VDPAU_DECODER
 +AVCodec ff_h264_vdpau_decoder = {
 +    .name           = "h264_vdpau",
 +    .long_name      = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
 +    .type           = AVMEDIA_TYPE_VIDEO,
 +    .id             = AV_CODEC_ID_H264,
 +    .priv_data_size = sizeof(H264Context),
 +    .init           = ff_h264_decode_init,
 +    .close          = h264_decode_end,
 +    .decode         = h264_decode_frame,
 +    .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
 +    .flush          = flush_dpb,
 +    .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_VDPAU_H264,
 +                                                     AV_PIX_FMT_NONE},
 +    .profiles       = NULL_IF_CONFIG_SMALL(profiles),
 +    .priv_class     = &h264_vdpau_class,
 +};
 +#endif
@@@ -1083,4 -1035,13 +1093,15 @@@ int ff_init_poc(H264Context *h, int pic
  int ff_pred_weight_table(H264Context *h);
  int ff_set_ref_count(H264Context *h);
  
+ int ff_h264_decode_slice_header(H264Context *h, H264Context *h0);
+ int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count);
+ int ff_h264_update_thread_context(AVCodecContext *dst,
+                                   const AVCodecContext *src);
+ void ff_h264_flush_change(H264Context *h);
+ void ff_h264_free_tables(H264Context *h, int free_rbsp);
++void ff_h264_set_erpic(ERPicture *dst, H264Picture *src);
++
  #endif /* AVCODEC_H264_H */
index 0000000,d7e6ce5..36028dc
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,200 +1,210 @@@
 - * This file is part of Libav.
+ /*
+  * H.26L/H.264/AVC/JVT/14496-10/... decoder
+  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
+  *
 - * Libav is free software; you can redistribute it and/or
++ * This file is part of FFmpeg.
+  *
 - * Libav is distributed in the hope that it will be useful,
++ * 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.
+  *
 - * License along with Libav; if not, write to the Free Software
++ * 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
 -static void h264_set_erpic(ERPicture *dst, H264Picture *src)
++ * License along with FFmpeg; if not, write to the Free Software
+  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+  */
+ /**
+  * @file
+  * H.264 / AVC / MPEG4 part10 codec.
+  * @author Michael Niedermayer <michaelni@gmx.at>
+  */
+ #include "libavutil/avassert.h"
+ #include "libavutil/imgutils.h"
+ #include "libavutil/timer.h"
+ #include "internal.h"
+ #include "cabac.h"
+ #include "cabac_functions.h"
+ #include "dsputil.h"
+ #include "error_resilience.h"
+ #include "avcodec.h"
+ #include "h264.h"
+ #include "h264data.h"
+ #include "h264chroma.h"
+ #include "h264_mvpred.h"
+ #include "golomb.h"
+ #include "mathops.h"
+ #include "mpegutils.h"
+ #include "rectangle.h"
+ #include "thread.h"
++#include "vdpau_internal.h"
+ void ff_h264_unref_picture(H264Context *h, H264Picture *pic)
+ {
+     int off = offsetof(H264Picture, tf) + sizeof(pic->tf);
+     int i;
+     if (!pic->f.buf[0])
+         return;
+     ff_thread_release_buffer(h->avctx, &pic->tf);
+     av_buffer_unref(&pic->hwaccel_priv_buf);
+     av_buffer_unref(&pic->qscale_table_buf);
+     av_buffer_unref(&pic->mb_type_buf);
+     for (i = 0; i < 2; i++) {
+         av_buffer_unref(&pic->motion_val_buf[i]);
+         av_buffer_unref(&pic->ref_index_buf[i]);
+     }
+     memset((uint8_t*)pic + off, 0, sizeof(*pic) - off);
+ }
+ int ff_h264_ref_picture(H264Context *h, H264Picture *dst, H264Picture *src)
+ {
+     int ret, i;
+     av_assert0(!dst->f.buf[0]);
+     av_assert0(src->f.buf[0]);
+     src->tf.f = &src->f;
+     dst->tf.f = &dst->f;
+     ret = ff_thread_ref_frame(&dst->tf, &src->tf);
+     if (ret < 0)
+         goto fail;
+     dst->qscale_table_buf = av_buffer_ref(src->qscale_table_buf);
+     dst->mb_type_buf      = av_buffer_ref(src->mb_type_buf);
+     if (!dst->qscale_table_buf || !dst->mb_type_buf)
+         goto fail;
+     dst->qscale_table = src->qscale_table;
+     dst->mb_type      = src->mb_type;
+     for (i = 0; i < 2; i++) {
+         dst->motion_val_buf[i] = av_buffer_ref(src->motion_val_buf[i]);
+         dst->ref_index_buf[i]  = av_buffer_ref(src->ref_index_buf[i]);
+         if (!dst->motion_val_buf[i] || !dst->ref_index_buf[i])
+             goto fail;
+         dst->motion_val[i] = src->motion_val[i];
+         dst->ref_index[i]  = src->ref_index[i];
+     }
+     if (src->hwaccel_picture_private) {
+         dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf);
+         if (!dst->hwaccel_priv_buf)
+             goto fail;
+         dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data;
+     }
+     for (i = 0; i < 2; i++)
+         dst->field_poc[i] = src->field_poc[i];
+     memcpy(dst->ref_poc,   src->ref_poc,   sizeof(src->ref_poc));
+     memcpy(dst->ref_count, src->ref_count, sizeof(src->ref_count));
+     dst->poc           = src->poc;
+     dst->frame_num     = src->frame_num;
+     dst->mmco_reset    = src->mmco_reset;
+     dst->pic_id        = src->pic_id;
+     dst->long_ref      = src->long_ref;
+     dst->mbaff         = src->mbaff;
+     dst->field_picture = src->field_picture;
+     dst->needs_realloc = src->needs_realloc;
+     dst->reference     = src->reference;
++    dst->crop          = src->crop;
++    dst->crop_left     = src->crop_left;
++    dst->crop_top      = src->crop_top;
+     dst->recovered     = src->recovered;
++    dst->invalid_gap   = src->invalid_gap;
+     return 0;
+ fail:
+     ff_h264_unref_picture(h, dst);
+     return ret;
+ }
+ #if CONFIG_ERROR_RESILIENCE
 -    if (!in_setup && !h->droppable)
 -        ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
 -                                  h->picture_structure == PICT_BOTTOM_FIELD);
++void ff_h264_set_erpic(ERPicture *dst, H264Picture *src)
+ {
+     int i;
++    memset(dst, 0, sizeof(*dst));
++
+     if (!src)
+         return;
+     dst->f = &src->f;
+     dst->tf = &src->tf;
+     for (i = 0; i < 2; i++) {
+         dst->motion_val[i] = src->motion_val[i];
+         dst->ref_index[i] = src->ref_index[i];
+     }
+     dst->mb_type = src->mb_type;
+     dst->field_picture = src->field_picture;
+ }
+ #endif /* CONFIG_ERROR_RESILIENCE */
+ int ff_h264_field_end(H264Context *h, int in_setup)
+ {
+     AVCodecContext *const avctx = h->avctx;
+     int err = 0;
+     h->mb_y = 0;
 -    if (CONFIG_ERROR_RESILIENCE && !FIELD_PICTURE(h)) {
 -        h264_set_erpic(&h->er.cur_pic, h->cur_pic_ptr);
 -        h264_set_erpic(&h->er.last_pic,
 -                       h->ref_count[0] ? &h->ref_list[0][0] : NULL);
 -        h264_set_erpic(&h->er.next_pic,
 -                       h->ref_count[1] ? &h->ref_list[1][0] : NULL);
++    if (CONFIG_H264_VDPAU_DECODER &&
++        h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
++        ff_vdpau_h264_set_reference_frames(h);
+     if (in_setup || !(avctx->active_thread_type & FF_THREAD_FRAME)) {
+         if (!h->droppable) {
+             err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
+             h->prev_poc_msb = h->poc_msb;
+             h->prev_poc_lsb = h->poc_lsb;
+         }
+         h->prev_frame_num_offset = h->frame_num_offset;
+         h->prev_frame_num        = h->frame_num;
+         h->outputed_poc          = h->next_outputed_poc;
+     }
+     if (avctx->hwaccel) {
+         if (avctx->hwaccel->end_frame(avctx) < 0)
+             av_log(avctx, AV_LOG_ERROR,
+                    "hardware accelerator failed to decode picture\n");
+     }
++    if (CONFIG_H264_VDPAU_DECODER &&
++        h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
++        ff_vdpau_h264_picture_complete(h);
++
+     /*
+      * FIXME: Error handling code does not seem to support interlaced
+      * when slices span multiple rows
+      * The ff_er_add_slice calls don't work right for bottom
+      * fields; they cause massive erroneous error concealing
+      * Error marking covers both fields (top and bottom).
+      * This causes a mismatched s->error_count
+      * and a bad error table. Further, the error count goes to
+      * INT_MAX when called for bottom field, because mb_y is
+      * past end by one (callers fault) and resync_mb_y != 0
+      * causes problems for the first MB line, too.
+      */
++    if (CONFIG_ERROR_RESILIENCE && !FIELD_PICTURE(h) && h->current_slice && !h->sps.new) {
++        ff_h264_set_erpic(&h->er.cur_pic, h->cur_pic_ptr);
+         ff_er_frame_end(&h->er);
+     }
++    if (!in_setup && !h->droppable)
++        ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
++                                  h->picture_structure == PICT_BOTTOM_FIELD);
+     emms_c();
+     h->current_slice = 0;
+     return err;
+ }
index 0000000,897c8eb..7db5179
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,2384 +1,2567 @@@
 - * This file is part of Libav.
+ /*
+  * H.26L/H.264/AVC/JVT/14496-10/... decoder
+  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
+  *
 - * Libav is free software; you can redistribute it and/or
++ * This file is part of FFmpeg.
+  *
 - * Libav is distributed in the hope that it will be useful,
++ * 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.
+  *
 - * License along with Libav; if not, write to the Free Software
++ * 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
 -    0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
++ * License along with FFmpeg; if not, write to the Free Software
+  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+  */
+ /**
+  * @file
+  * H.264 / AVC / MPEG4 part10 codec.
+  * @author Michael Niedermayer <michaelni@gmx.at>
+  */
+ #include "libavutil/avassert.h"
+ #include "libavutil/imgutils.h"
+ #include "libavutil/timer.h"
+ #include "internal.h"
+ #include "cabac.h"
+ #include "cabac_functions.h"
+ #include "dsputil.h"
+ #include "error_resilience.h"
+ #include "avcodec.h"
+ #include "h264.h"
+ #include "h264data.h"
+ #include "h264chroma.h"
+ #include "h264_mvpred.h"
+ #include "golomb.h"
+ #include "mathops.h"
+ #include "mpegutils.h"
+ #include "rectangle.h"
+ #include "thread.h"
+ static const uint8_t rem6[QP_MAX_NUM + 1] = {
+     0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
+     3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
 -    7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10, 10,
++    0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
++    3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
++    0, 1, 2, 3,
+ };
+ static const uint8_t div6[QP_MAX_NUM + 1] = {
+     0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3,  3,  3,
+     3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6,  6,  6,
 -static const uint8_t field_scan[16] = {
++    7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
++   10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
++   14,14,14,14,
+ };
 -static const uint8_t field_scan8x8[64] = {
++static const uint8_t field_scan[16+1] = {
+     0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
+     0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
+     2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
+     3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
+ };
 -static const uint8_t field_scan8x8_cavlc[64] = {
++static const uint8_t field_scan8x8[64+1] = {
+     0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
+     1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
+     2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
+     0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
+     2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
+     2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
+     2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
+     3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
+     3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
+     4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
+     4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
+     5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
+     5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
+     7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
+     6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
+     7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
+ };
 -static const uint8_t zigzag_scan8x8_cavlc[64] = {
++static const uint8_t field_scan8x8_cavlc[64+1] = {
+     0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
+     2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
+     3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
+     5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
+     0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
+     1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
+     3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
+     5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
+     0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
+     1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
+     3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
+     5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
+     1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
+     1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
+     3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
+     6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
+ };
+ // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
 -    if (dst == src || !h1->context_initialized)
++static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
+     0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
+     4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
+     3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
+     2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
+     1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
+     3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
+     2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
+     3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
+     0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
+     2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
+     1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
+     4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
+     0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
+     1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
+     0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
+     5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
+ };
+ static const uint8_t dequant4_coeff_init[6][3] = {
+     { 10, 13, 16 },
+     { 11, 14, 18 },
+     { 13, 16, 20 },
+     { 14, 18, 23 },
+     { 16, 20, 25 },
+     { 18, 23, 29 },
+ };
+ static const uint8_t dequant8_coeff_init_scan[16] = {
+     0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
+ };
+ static const uint8_t dequant8_coeff_init[6][6] = {
+     { 20, 18, 32, 19, 25, 24 },
+     { 22, 19, 35, 21, 28, 26 },
+     { 26, 23, 42, 24, 33, 31 },
+     { 28, 25, 45, 26, 35, 33 },
+     { 32, 28, 51, 30, 40, 38 },
+     { 36, 32, 58, 34, 46, 43 },
+ };
+ static const enum AVPixelFormat h264_hwaccel_pixfmt_list_420[] = {
+ #if CONFIG_H264_DXVA2_HWACCEL
+     AV_PIX_FMT_DXVA2_VLD,
+ #endif
+ #if CONFIG_H264_VAAPI_HWACCEL
+     AV_PIX_FMT_VAAPI_VLD,
+ #endif
+ #if CONFIG_H264_VDA_HWACCEL
+     AV_PIX_FMT_VDA_VLD,
+ #endif
+ #if CONFIG_H264_VDPAU_HWACCEL
+     AV_PIX_FMT_VDPAU,
+ #endif
+     AV_PIX_FMT_YUV420P,
+     AV_PIX_FMT_NONE
+ };
+ static const enum AVPixelFormat h264_hwaccel_pixfmt_list_jpeg_420[] = {
+ #if CONFIG_H264_DXVA2_HWACCEL
+     AV_PIX_FMT_DXVA2_VLD,
+ #endif
+ #if CONFIG_H264_VAAPI_HWACCEL
+     AV_PIX_FMT_VAAPI_VLD,
+ #endif
+ #if CONFIG_H264_VDA_HWACCEL
+     AV_PIX_FMT_VDA_VLD,
+ #endif
+ #if CONFIG_H264_VDPAU_HWACCEL
+     AV_PIX_FMT_VDPAU,
+ #endif
+     AV_PIX_FMT_YUVJ420P,
+     AV_PIX_FMT_NONE
+ };
+ static void release_unused_pictures(H264Context *h, int remove_current)
+ {
+     int i;
+     /* release non reference frames */
+     for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
+         if (h->DPB[i].f.buf[0] && !h->DPB[i].reference &&
+             (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
+             ff_h264_unref_picture(h, &h->DPB[i]);
+         }
+     }
+ }
+ static int alloc_scratch_buffers(H264Context *h, int linesize)
+ {
+     int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
+     if (h->bipred_scratchpad)
+         return 0;
+     h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size);
+     // edge emu needs blocksize + filter length - 1
+     // (= 21x21 for  h264)
+     h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 21);
+     if (!h->bipred_scratchpad || !h->edge_emu_buffer) {
+         av_freep(&h->bipred_scratchpad);
+         av_freep(&h->edge_emu_buffer);
+         return AVERROR(ENOMEM);
+     }
+     return 0;
+ }
+ static int init_table_pools(H264Context *h)
+ {
+     const int big_mb_num    = h->mb_stride * (h->mb_height + 1) + 1;
+     const int mb_array_size = h->mb_stride * h->mb_height;
+     const int b4_stride     = h->mb_width * 4 + 1;
+     const int b4_array_size = b4_stride * h->mb_height * 4;
+     h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
+                                                av_buffer_allocz);
+     h->mb_type_pool      = av_buffer_pool_init((big_mb_num + h->mb_stride) *
+                                                sizeof(uint32_t), av_buffer_allocz);
+     h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
+                                              sizeof(int16_t), av_buffer_allocz);
+     h->ref_index_pool  = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
+     if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
+         !h->ref_index_pool) {
+         av_buffer_pool_uninit(&h->qscale_table_pool);
+         av_buffer_pool_uninit(&h->mb_type_pool);
+         av_buffer_pool_uninit(&h->motion_val_pool);
+         av_buffer_pool_uninit(&h->ref_index_pool);
+         return AVERROR(ENOMEM);
+     }
+     return 0;
+ }
+ static int alloc_picture(H264Context *h, H264Picture *pic)
+ {
+     int i, ret = 0;
+     av_assert0(!pic->f.data[0]);
+     pic->tf.f = &pic->f;
+     ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
+                                                    AV_GET_BUFFER_FLAG_REF : 0);
+     if (ret < 0)
+         goto fail;
+     h->linesize   = pic->f.linesize[0];
+     h->uvlinesize = pic->f.linesize[1];
++    pic->crop     = h->sps.crop;
++    pic->crop_top = h->sps.crop_top;
++    pic->crop_left= h->sps.crop_left;
+     if (h->avctx->hwaccel) {
+         const AVHWAccel *hwaccel = h->avctx->hwaccel;
+         av_assert0(!pic->hwaccel_picture_private);
+         if (hwaccel->priv_data_size) {
+             pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->priv_data_size);
+             if (!pic->hwaccel_priv_buf)
+                 return AVERROR(ENOMEM);
+             pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
+         }
+     }
++    if (!h->avctx->hwaccel && CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY && pic->f.data[2]) {
++        int h_chroma_shift, v_chroma_shift;
++        av_pix_fmt_get_chroma_sub_sample(pic->f.format,
++                                         &h_chroma_shift, &v_chroma_shift);
++
++        for(i=0; i<FF_CEIL_RSHIFT(h->avctx->height, v_chroma_shift); i++) {
++            memset(pic->f.data[1] + pic->f.linesize[1]*i,
++                   0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
++            memset(pic->f.data[2] + pic->f.linesize[2]*i,
++                   0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
++        }
++    }
+     if (!h->qscale_table_pool) {
+         ret = init_table_pools(h);
+         if (ret < 0)
+             goto fail;
+     }
+     pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
+     pic->mb_type_buf      = av_buffer_pool_get(h->mb_type_pool);
+     if (!pic->qscale_table_buf || !pic->mb_type_buf)
+         goto fail;
+     pic->mb_type      = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
+     pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
+     for (i = 0; i < 2; i++) {
+         pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
+         pic->ref_index_buf[i]  = av_buffer_pool_get(h->ref_index_pool);
+         if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
+             goto fail;
+         pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
+         pic->ref_index[i]  = pic->ref_index_buf[i]->data;
+     }
+     return 0;
+ fail:
+     ff_h264_unref_picture(h, pic);
+     return (ret < 0) ? ret : AVERROR(ENOMEM);
+ }
+ static inline int pic_is_unused(H264Context *h, H264Picture *pic)
+ {
+     if (!pic->f.buf[0])
+         return 1;
+     if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
+         return 1;
+     return 0;
+ }
+ static int find_unused_picture(H264Context *h)
+ {
+     int i;
+     for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
+         if (pic_is_unused(h, &h->DPB[i]))
+             break;
+     }
+     if (i == H264_MAX_PICTURE_COUNT)
+         return AVERROR_INVALIDDATA;
+     if (h->DPB[i].needs_realloc) {
+         h->DPB[i].needs_realloc = 0;
+         ff_h264_unref_picture(h, &h->DPB[i]);
+     }
+     return i;
+ }
+ static void init_dequant8_coeff_table(H264Context *h)
+ {
+     int i, j, q, x;
+     const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
+     for (i = 0; i < 6; i++) {
+         h->dequant8_coeff[i] = h->dequant8_buffer[i];
+         for (j = 0; j < i; j++)
+             if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
+                         64 * sizeof(uint8_t))) {
+                 h->dequant8_coeff[i] = h->dequant8_buffer[j];
+                 break;
+             }
+         if (j < i)
+             continue;
+         for (q = 0; q < max_qp + 1; q++) {
+             int shift = div6[q];
+             int idx   = rem6[q];
+             for (x = 0; x < 64; x++)
+                 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
+                     ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
+                      h->pps.scaling_matrix8[i][x]) << shift;
+         }
+     }
+ }
+ static void init_dequant4_coeff_table(H264Context *h)
+ {
+     int i, j, q, x;
+     const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
+     for (i = 0; i < 6; i++) {
+         h->dequant4_coeff[i] = h->dequant4_buffer[i];
+         for (j = 0; j < i; j++)
+             if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
+                         16 * sizeof(uint8_t))) {
+                 h->dequant4_coeff[i] = h->dequant4_buffer[j];
+                 break;
+             }
+         if (j < i)
+             continue;
+         for (q = 0; q < max_qp + 1; q++) {
+             int shift = div6[q] + 2;
+             int idx   = rem6[q];
+             for (x = 0; x < 16; x++)
+                 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
+                     ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
+                      h->pps.scaling_matrix4[i][x]) << shift;
+         }
+     }
+ }
+ void h264_init_dequant_tables(H264Context *h)
+ {
+     int i, x;
+     init_dequant4_coeff_table(h);
++    memset(h->dequant8_coeff, 0, sizeof(h->dequant8_coeff));
++
+     if (h->pps.transform_8x8_mode)
+         init_dequant8_coeff_table(h);
+     if (h->sps.transform_bypass) {
+         for (i = 0; i < 6; i++)
+             for (x = 0; x < 16; x++)
+                 h->dequant4_coeff[i][0][x] = 1 << 6;
+         if (h->pps.transform_8x8_mode)
+             for (i = 0; i < 6; i++)
+                 for (x = 0; x < 64; x++)
+                     h->dequant8_coeff[i][0][x] = 1 << 6;
+     }
+ }
+ /**
+  * Mimic alloc_tables(), but for every context thread.
+  */
+ static void clone_tables(H264Context *dst, H264Context *src, int i)
+ {
+     dst->intra4x4_pred_mode     = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
+     dst->non_zero_count         = src->non_zero_count;
+     dst->slice_table            = src->slice_table;
+     dst->cbp_table              = src->cbp_table;
+     dst->mb2b_xy                = src->mb2b_xy;
+     dst->mb2br_xy               = src->mb2br_xy;
+     dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
+     dst->mvd_table[0]           = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
+     dst->mvd_table[1]           = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
+     dst->direct_table           = src->direct_table;
+     dst->list_counts            = src->list_counts;
+     dst->DPB                    = src->DPB;
+     dst->cur_pic_ptr            = src->cur_pic_ptr;
+     dst->cur_pic                = src->cur_pic;
+     dst->bipred_scratchpad      = NULL;
+     dst->edge_emu_buffer        = NULL;
+     ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
+                       src->sps.chroma_format_idc);
+ }
+ #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
+ #undef REBASE_PICTURE
+ #define REBASE_PICTURE(pic, new_ctx, old_ctx)             \
+     ((pic && pic >= old_ctx->DPB &&                       \
+       pic < old_ctx->DPB + H264_MAX_PICTURE_COUNT) ?          \
+      &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
+ static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
+                                H264Context *new_base,
+                                H264Context *old_base)
+ {
+     int i;
+     for (i = 0; i < count; i++) {
+         assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
+                 IN_RANGE(from[i], old_base->DPB,
+                          sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
+                 !from[i]));
+         to[i] = REBASE_PICTURE(from[i], new_base, old_base);
+     }
+ }
+ static int copy_parameter_set(void **to, void **from, int count, int size)
+ {
+     int i;
+     for (i = 0; i < count; i++) {
+         if (to[i] && !from[i]) {
+             av_freep(&to[i]);
+         } else if (from[i] && !to[i]) {
+             to[i] = av_malloc(size);
+             if (!to[i])
+                 return AVERROR(ENOMEM);
+         }
+         if (from[i])
+             memcpy(to[i], from[i], size);
+     }
+     return 0;
+ }
+ #define copy_fields(to, from, start_field, end_field)                   \
+     memcpy(&to->start_field, &from->start_field,                        \
+            (char *)&to->end_field - (char *)&to->start_field)
+ static int h264_slice_header_init(H264Context *h, int reinit);
+ int ff_h264_update_thread_context(AVCodecContext *dst,
+                                   const AVCodecContext *src)
+ {
+     H264Context *h = dst->priv_data, *h1 = src->priv_data;
+     int inited = h->context_initialized, err = 0;
+     int context_reinitialized = 0;
+     int i, ret;
 -        /* update linesize on resize. The decoder doesn't
 -         * necessarily call h264_frame_start in the new thread */
 -        h->linesize   = h1->linesize;
 -        h->uvlinesize = h1->uvlinesize;
 -
 -        /* copy block_offset since frame_start may not be called */
 -        memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
++    if (dst == src)
+         return 0;
+     if (inited &&
+         (h->width                 != h1->width                 ||
+          h->height                != h1->height                ||
+          h->mb_width              != h1->mb_width              ||
+          h->mb_height             != h1->mb_height             ||
+          h->sps.bit_depth_luma    != h1->sps.bit_depth_luma    ||
+          h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
+          h->sps.colorspace        != h1->sps.colorspace)) {
+         /* set bits_per_raw_sample to the previous value. the check for changed
+          * bit depth in h264_set_parameter_from_sps() uses it and sets it to
+          * the current value */
+         h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
+         av_freep(&h->bipred_scratchpad);
+         h->width     = h1->width;
+         h->height    = h1->height;
+         h->mb_height = h1->mb_height;
+         h->mb_width  = h1->mb_width;
+         h->mb_num    = h1->mb_num;
+         h->mb_stride = h1->mb_stride;
+         h->b_stride  = h1->b_stride;
++        // SPS/PPS
++        if ((ret = copy_parameter_set((void **)h->sps_buffers,
++                                      (void **)h1->sps_buffers,
++                                      MAX_SPS_COUNT, sizeof(SPS))) < 0)
++            return ret;
++        h->sps = h1->sps;
++        if ((ret = copy_parameter_set((void **)h->pps_buffers,
++                                      (void **)h1->pps_buffers,
++                                      MAX_PPS_COUNT, sizeof(PPS))) < 0)
++            return ret;
++        h->pps = h1->pps;
+         if ((err = h264_slice_header_init(h, 1)) < 0) {
+             av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
+             return err;
+         }
+         context_reinitialized = 1;
 -        memcpy(h, h1, sizeof(*h1));
++#if 0
++        h264_set_parameter_from_sps(h);
++        //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted
++        h->cur_chroma_format_idc = h1->cur_chroma_format_idc;
++#endif
+     }
++    /* update linesize on resize for h264. The h264 decoder doesn't
++     * necessarily call ff_MPV_frame_start in the new thread */
++    h->linesize   = h1->linesize;
++    h->uvlinesize = h1->uvlinesize;
++
++    /* copy block_offset since frame_start may not be called */
++    memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
+     if (!inited) {
+         for (i = 0; i < MAX_SPS_COUNT; i++)
+             av_freep(h->sps_buffers + i);
+         for (i = 0; i < MAX_PPS_COUNT; i++)
+             av_freep(h->pps_buffers + i);
 -        h->context_initialized = 0;
 -
 -        memset(&h->cur_pic, 0, sizeof(h->cur_pic));
 -        av_frame_unref(&h->cur_pic.f);
 -        h->cur_pic.tf.f = &h->cur_pic.f;
++        av_freep(&h->rbsp_buffer[0]);
++        av_freep(&h->rbsp_buffer[1]);
++        memcpy(h, h1, offsetof(H264Context, intra_pcm_ptr));
++        memcpy(&h->cabac, &h1->cabac,
++               sizeof(H264Context) - offsetof(H264Context, cabac));
++        av_assert0((void*)&h->cabac == &h->mb_padding + 1);
++
+         memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
+         memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
++
+         memset(&h->er, 0, sizeof(h->er));
+         memset(&h->mb, 0, sizeof(h->mb));
+         memset(&h->mb_luma_dc, 0, sizeof(h->mb_luma_dc));
+         memset(&h->mb_padding, 0, sizeof(h->mb_padding));
 -
 -        for (i = 0; i < 2; i++) {
 -            h->rbsp_buffer[i]      = NULL;
 -            h->rbsp_buffer_size[i] = 0;
+         h->avctx             = dst;
+         h->DPB               = NULL;
+         h->qscale_table_pool = NULL;
+         h->mb_type_pool      = NULL;
+         h->ref_index_pool    = NULL;
+         h->motion_val_pool   = NULL;
++        for (i = 0; i < 2; i++) {
++            h->rbsp_buffer[i] = NULL;
++            h->rbsp_buffer_size[i] = 0;
++        }
++
++        if (h1->context_initialized) {
++        h->context_initialized = 0;
++
++        memset(&h->cur_pic, 0, sizeof(h->cur_pic));
++        av_frame_unref(&h->cur_pic.f);
++        h->cur_pic.tf.f = &h->cur_pic.f;
+         ret = ff_h264_alloc_tables(h);
+         if (ret < 0) {
+             av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
+             return ret;
+         }
+         ret = ff_h264_context_init(h);
+         if (ret < 0) {
+             av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
+             return ret;
+         }
 -
 -        h->context_initialized = 1;
+         }
++
+         h->bipred_scratchpad = NULL;
+         h->edge_emu_buffer   = NULL;
+         h->thread_context[0] = h;
 -    for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
++        h->context_initialized = h1->context_initialized;
+     }
+     h->avctx->coded_height  = h1->avctx->coded_height;
+     h->avctx->coded_width   = h1->avctx->coded_width;
+     h->avctx->width         = h1->avctx->width;
+     h->avctx->height        = h1->avctx->height;
+     h->coded_picture_number = h1->coded_picture_number;
+     h->first_field          = h1->first_field;
+     h->picture_structure    = h1->picture_structure;
+     h->qscale               = h1->qscale;
+     h->droppable            = h1->droppable;
+     h->low_delay            = h1->low_delay;
 -        if (h1->DPB[i].f.buf[0] &&
++    for (i = 0; h->DPB && i < H264_MAX_PICTURE_COUNT; i++) {
+         ff_h264_unref_picture(h, &h->DPB[i]);
 -    if ((ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic)) < 0)
++        if (h1->DPB && h1->DPB[i].f.buf[0] &&
+             (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
+             return ret;
+     }
+     h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
+     ff_h264_unref_picture(h, &h->cur_pic);
 -    /* frame_start may not be called for the next thread (if it's decoding
 -     * a bottom field) so this has to be allocated here */
 -    err = alloc_scratch_buffers(h, h1->linesize);
 -    if (err < 0)
 -        return err;
 -
++    if (h1->cur_pic.f.buf[0] && (ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic)) < 0)
+         return ret;
+     h->workaround_bugs = h1->workaround_bugs;
+     h->low_delay       = h1->low_delay;
+     h->droppable       = h1->droppable;
 -    h->last_slice_type = h1->last_slice_type;
+     // extradata/NAL handling
+     h->is_avc = h1->is_avc;
+     // SPS/PPS
+     if ((ret = copy_parameter_set((void **)h->sps_buffers,
+                                   (void **)h1->sps_buffers,
+                                   MAX_SPS_COUNT, sizeof(SPS))) < 0)
+         return ret;
+     h->sps = h1->sps;
+     if ((ret = copy_parameter_set((void **)h->pps_buffers,
+                                   (void **)h1->pps_buffers,
+                                   MAX_PPS_COUNT, sizeof(PPS))) < 0)
+         return ret;
+     h->pps = h1->pps;
+     // Dequantization matrices
+     // FIXME these are big - can they be only copied when PPS changes?
+     copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
+     for (i = 0; i < 6; i++)
+         h->dequant4_coeff[i] = h->dequant4_buffer[0] +
+                                (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
+     for (i = 0; i < 6; i++)
+         h->dequant8_coeff[i] = h->dequant8_buffer[0] +
+                                (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
+     h->dequant_coeff_pps = h1->dequant_coeff_pps;
+     // POC timing
+     copy_fields(h, h1, poc_lsb, redundant_pic_count);
+     // reference lists
+     copy_fields(h, h1, short_ref, cabac_init_idc);
+     copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
+     copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
+     copy_picture_range(h->delayed_pic, h1->delayed_pic,
+                        MAX_DELAYED_PIC_COUNT + 2, h, h1);
 -    h->frame_recovered       = h1->frame_recovered;
++    h->frame_recovered       = h1->frame_recovered;
+     if (context_reinitialized)
+         ff_h264_set_parameter_from_sps(h);
+     if (!h->cur_pic_ptr)
+         return 0;
+     if (!h->droppable) {
+         err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
+         h->prev_poc_msb = h->poc_msb;
+         h->prev_poc_lsb = h->poc_lsb;
+     }
+     h->prev_frame_num_offset = h->frame_num_offset;
+     h->prev_frame_num        = h->frame_num;
+     h->outputed_poc          = h->next_outputed_poc;
+     h->recovery_frame        = h1->recovery_frame;
 -    if (CONFIG_ERROR_RESILIENCE)
+     return err;
+ }
+ static int h264_frame_start(H264Context *h)
+ {
+     H264Picture *pic;
+     int i, ret;
+     const int pixel_shift = h->pixel_shift;
++    int c[4] = {
++        1<<(h->sps.bit_depth_luma-1),
++        1<<(h->sps.bit_depth_chroma-1),
++        1<<(h->sps.bit_depth_chroma-1),
++        -1
++    };
++
++    if (!ff_thread_can_start_frame(h->avctx)) {
++        av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
++        return -1;
++    }
+     release_unused_pictures(h, 1);
+     h->cur_pic_ptr = NULL;
+     i = find_unused_picture(h);
+     if (i < 0) {
+         av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
+         return i;
+     }
+     pic = &h->DPB[i];
+     pic->reference              = h->droppable ? 0 : h->picture_structure;
+     pic->f.coded_picture_number = h->coded_picture_number++;
+     pic->field_picture          = h->picture_structure != PICT_FRAME;
++
+     /*
+      * Zero key_frame here; IDR markings per slice in frame or fields are ORed
+      * in later.
+      * See decode_nal_units().
+      */
+     pic->f.key_frame = 0;
+     pic->mmco_reset  = 0;
+     pic->recovered   = 0;
++    pic->invalid_gap = 0;
+     if ((ret = alloc_picture(h, pic)) < 0)
+         return ret;
++    if(!h->frame_recovered && !h->avctx->hwaccel &&
++       !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
++        avpriv_color_frame(&pic->f, c);
+     h->cur_pic_ptr = pic;
+     ff_h264_unref_picture(h, &h->cur_pic);
++    if (CONFIG_ERROR_RESILIENCE) {
++        memset(&h->er.cur_pic, 0, sizeof(h->er.cur_pic));
++    }
++
+     if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
+         return ret;
 -    /* can't be in alloc_tables because linesize isn't known there.
 -     * FIXME: redo bipred weight to not require extra buffer? */
 -    for (i = 0; i < h->slice_context_count; i++)
 -        if (h->thread_context[i]) {
 -            ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
 -            if (ret < 0)
 -                return ret;
 -        }
 -
 -    /* Some macroblocks can be accessed before they're available in case
 -     * of lost slices, MBAFF or threading. */
 -    memset(h->slice_table, -1,
 -           (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
 -
 -    // s->decode = (s->flags & CODEC_FLAG_PSNR) || !s->encoding ||
 -    //             s->current_picture.f.reference /* || h->contains_intra */ || 1;
++    if (CONFIG_ERROR_RESILIENCE) {
+         ff_er_frame_start(&h->er);
++        memset(&h->er.last_pic, 0, sizeof(h->er.last_pic));
++        memset(&h->er.next_pic, 0, sizeof(h->er.next_pic));
++    }
+     assert(h->linesize && h->uvlinesize);
+     for (i = 0; i < 16; i++) {
+         h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
+         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
+     }
+     for (i = 0; i < 16; i++) {
+         h->block_offset[16 + i]      =
+         h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
+         h->block_offset[48 + 16 + i] =
+         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
+     }
 -        h->zigzag_scan_q0          = zigzag_scan;
 -        h->zigzag_scan8x8_q0       = ff_zigzag_direct;
 -        h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
 -        h->field_scan_q0           = field_scan;
 -        h->field_scan8x8_q0        = field_scan8x8;
 -        h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
++    // s->decode = (h->flags & CODEC_FLAG_PSNR) || !s->encoding ||
++    //             h->cur_pic.reference /* || h->contains_intra */ || 1;
+     /* We mark the current picture as non-reference after allocating it, so
+      * that if we break out due to an error it can be released automatically
+      * in the next ff_MPV_frame_start().
+      */
+     h->cur_pic_ptr->reference = 0;
+     h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
+     h->next_output_pic = NULL;
+     assert(h->cur_pic_ptr->long_ref == 0);
+     return 0;
+ }
+ static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
+                                               uint8_t *src_cb, uint8_t *src_cr,
+                                               int linesize, int uvlinesize,
+                                               int simple)
+ {
+     uint8_t *top_border;
+     int top_idx = 1;
+     const int pixel_shift = h->pixel_shift;
+     int chroma444 = CHROMA444(h);
+     int chroma422 = CHROMA422(h);
+     src_y  -= linesize;
+     src_cb -= uvlinesize;
+     src_cr -= uvlinesize;
+     if (!simple && FRAME_MBAFF(h)) {
+         if (h->mb_y & 1) {
+             if (!MB_MBAFF(h)) {
+                 top_border = h->top_borders[0][h->mb_x];
+                 AV_COPY128(top_border, src_y + 15 * linesize);
+                 if (pixel_shift)
+                     AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
+                 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
+                     if (chroma444) {
+                         if (pixel_shift) {
+                             AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
+                             AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
+                             AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
+                             AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
+                         } else {
+                             AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
+                             AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
+                         }
+                     } else if (chroma422) {
+                         if (pixel_shift) {
+                             AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
+                             AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
+                         } else {
+                             AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
+                             AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
+                         }
+                     } else {
+                         if (pixel_shift) {
+                             AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
+                             AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
+                         } else {
+                             AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
+                             AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
+                         }
+                     }
+                 }
+             }
+         } else if (MB_MBAFF(h)) {
+             top_idx = 0;
+         } else
+             return;
+     }
+     top_border = h->top_borders[top_idx][h->mb_x];
+     /* There are two lines saved, the line above the top macroblock
+      * of a pair, and the line above the bottom macroblock. */
+     AV_COPY128(top_border, src_y + 16 * linesize);
+     if (pixel_shift)
+         AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
+     if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
+         if (chroma444) {
+             if (pixel_shift) {
+                 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
+                 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
+                 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
+                 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
+             } else {
+                 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
+                 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
+             }
+         } else if (chroma422) {
+             if (pixel_shift) {
+                 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
+                 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
+             } else {
+                 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
+                 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
+             }
+         } else {
+             if (pixel_shift) {
+                 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
+                 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
+             } else {
+                 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
+                 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
+             }
+         }
+     }
+ }
+ /**
+  * Initialize implicit_weight table.
+  * @param field  0/1 initialize the weight for interlaced MBAFF
+  *                -1 initializes the rest
+  */
+ static void implicit_weight_table(H264Context *h, int field)
+ {
+     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
+     for (i = 0; i < 2; i++) {
+         h->luma_weight_flag[i]   = 0;
+         h->chroma_weight_flag[i] = 0;
+     }
+     if (field < 0) {
+         if (h->picture_structure == PICT_FRAME) {
+             cur_poc = h->cur_pic_ptr->poc;
+         } else {
+             cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
+         }
+         if (h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
+             h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2 * cur_poc) {
+             h->use_weight        = 0;
+             h->use_weight_chroma = 0;
+             return;
+         }
+         ref_start  = 0;
+         ref_count0 = h->ref_count[0];
+         ref_count1 = h->ref_count[1];
+     } else {
+         cur_poc    = h->cur_pic_ptr->field_poc[field];
+         ref_start  = 16;
+         ref_count0 = 16 + 2 * h->ref_count[0];
+         ref_count1 = 16 + 2 * h->ref_count[1];
+     }
+     h->use_weight               = 2;
+     h->use_weight_chroma        = 2;
+     h->luma_log2_weight_denom   = 5;
+     h->chroma_log2_weight_denom = 5;
+     for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
+         int poc0 = h->ref_list[0][ref0].poc;
+         for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
+             int w = 32;
+             if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
+                 int poc1 = h->ref_list[1][ref1].poc;
+                 int td   = av_clip(poc1 - poc0, -128, 127);
+                 if (td) {
+                     int tb = av_clip(cur_poc - poc0, -128, 127);
+                     int tx = (16384 + (FFABS(td) >> 1)) / td;
+                     int dist_scale_factor = (tb * tx + 32) >> 8;
+                     if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
+                         w = 64 - dist_scale_factor;
+                 }
+             }
+             if (field < 0) {
+                 h->implicit_weight[ref0][ref1][0] =
+                 h->implicit_weight[ref0][ref1][1] = w;
+             } else {
+                 h->implicit_weight[ref0][ref1][field] = w;
+             }
+         }
+     }
+ }
+ /**
+  * initialize scan tables
+  */
+ static void init_scan_tables(H264Context *h)
+ {
+     int i;
+     for (i = 0; i < 16; i++) {
+ #define TRANSPOSE(x) (x >> 2) | ((x << 2) & 0xF)
+         h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
+         h->field_scan[i]  = TRANSPOSE(field_scan[i]);
+ #undef TRANSPOSE
+     }
+     for (i = 0; i < 64; i++) {
+ #define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
+         h->zigzag_scan8x8[i]       = TRANSPOSE(ff_zigzag_direct[i]);
+         h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
+         h->field_scan8x8[i]        = TRANSPOSE(field_scan8x8[i]);
+         h->field_scan8x8_cavlc[i]  = TRANSPOSE(field_scan8x8_cavlc[i]);
+ #undef TRANSPOSE
+     }
+     if (h->sps.transform_bypass) { // FIXME same ugly
 -        h->zigzag_scan_q0          = h->zigzag_scan;
 -        h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
 -        h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
 -        h->field_scan_q0           = h->field_scan;
 -        h->field_scan8x8_q0        = h->field_scan8x8;
 -        h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
++        memcpy(h->zigzag_scan_q0          , zigzag_scan             , sizeof(h->zigzag_scan_q0         ));
++        memcpy(h->zigzag_scan8x8_q0       , ff_zigzag_direct        , sizeof(h->zigzag_scan8x8_q0      ));
++        memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc    , sizeof(h->zigzag_scan8x8_cavlc_q0));
++        memcpy(h->field_scan_q0           , field_scan              , sizeof(h->field_scan_q0          ));
++        memcpy(h->field_scan8x8_q0        , field_scan8x8           , sizeof(h->field_scan8x8_q0       ));
++        memcpy(h->field_scan8x8_cavlc_q0  , field_scan8x8_cavlc     , sizeof(h->field_scan8x8_cavlc_q0 ));
+     } else {
 -static enum AVPixelFormat get_pixel_format(H264Context *h)
++        memcpy(h->zigzag_scan_q0          , h->zigzag_scan          , sizeof(h->zigzag_scan_q0         ));
++        memcpy(h->zigzag_scan8x8_q0       , h->zigzag_scan8x8       , sizeof(h->zigzag_scan8x8_q0      ));
++        memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
++        memcpy(h->field_scan_q0           , h->field_scan           , sizeof(h->field_scan_q0          ));
++        memcpy(h->field_scan8x8_q0        , h->field_scan8x8        , sizeof(h->field_scan8x8_q0       ));
++        memcpy(h->field_scan8x8_cavlc_q0  , h->field_scan8x8_cavlc  , sizeof(h->field_scan8x8_cavlc_q0 ));
+     }
+ }
+ /**
+  * Replicate H264 "master" context to thread contexts.
+  */
+ static int clone_slice(H264Context *dst, H264Context *src)
+ {
+     memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
+     dst->cur_pic_ptr = src->cur_pic_ptr;
+     dst->cur_pic     = src->cur_pic;
+     dst->linesize    = src->linesize;
+     dst->uvlinesize  = src->uvlinesize;
+     dst->first_field = src->first_field;
+     dst->prev_poc_msb          = src->prev_poc_msb;
+     dst->prev_poc_lsb          = src->prev_poc_lsb;
+     dst->prev_frame_num_offset = src->prev_frame_num_offset;
+     dst->prev_frame_num        = src->prev_frame_num;
+     dst->short_ref_count       = src->short_ref_count;
+     memcpy(dst->short_ref,        src->short_ref,        sizeof(dst->short_ref));
+     memcpy(dst->long_ref,         src->long_ref,         sizeof(dst->long_ref));
+     memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
+     memcpy(dst->dequant4_coeff,   src->dequant4_coeff,   sizeof(src->dequant4_coeff));
+     memcpy(dst->dequant8_coeff,   src->dequant8_coeff,   sizeof(src->dequant8_coeff));
+     return 0;
+ }
 -    case 8:
++static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
+ {
+     switch (h->sps.bit_depth_luma) {
+     case 9:
+         if (CHROMA444(h)) {
+             if (h->avctx->colorspace == AVCOL_SPC_RGB) {
+                 return AV_PIX_FMT_GBRP9;
+             } else
+                 return AV_PIX_FMT_YUV444P9;
+         } else if (CHROMA422(h))
+             return AV_PIX_FMT_YUV422P9;
+         else
+             return AV_PIX_FMT_YUV420P9;
+         break;
+     case 10:
+         if (CHROMA444(h)) {
+             if (h->avctx->colorspace == AVCOL_SPC_RGB) {
+                 return AV_PIX_FMT_GBRP10;
+             } else
+                 return AV_PIX_FMT_YUV444P10;
+         } else if (CHROMA422(h))
+             return AV_PIX_FMT_YUV422P10;
+         else
+             return AV_PIX_FMT_YUV420P10;
+         break;
 -                return AV_PIX_FMT_GBRP;
++    case 12:
+         if (CHROMA444(h)) {
+             if (h->avctx->colorspace == AVCOL_SPC_RGB) {
 -                return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
 -                                                                 : AV_PIX_FMT_YUV444P;
++                return AV_PIX_FMT_GBRP12;
+             } else
 -            return h->avctx->get_format(h->avctx, h->avctx->codec->pix_fmts ?
++                return AV_PIX_FMT_YUV444P12;
++        } else if (CHROMA422(h))
++            return AV_PIX_FMT_YUV422P12;
++        else
++            return AV_PIX_FMT_YUV420P12;
++        break;
++    case 14:
++        if (CHROMA444(h)) {
++            if (h->avctx->colorspace == AVCOL_SPC_RGB) {
++                return AV_PIX_FMT_GBRP14;
++            } else
++                return AV_PIX_FMT_YUV444P14;
++        } else if (CHROMA422(h))
++            return AV_PIX_FMT_YUV422P14;
++        else
++            return AV_PIX_FMT_YUV420P14;
++        break;
++    case 8:
++        if (CHROMA444(h)) {
++            if (h->avctx->colorspace == AVCOL_SPC_RGB) {
++                av_log(h->avctx, AV_LOG_DEBUG, "Detected GBR colorspace.\n");
++                return AV_PIX_FMT_GBR24P;
++            } else if (h->avctx->colorspace == AVCOL_SPC_YCGCO) {
++                av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
++            }
++            return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ444P
++                                                                : AV_PIX_FMT_YUV444P;
+         } else if (CHROMA422(h)) {
+             return h->avctx->color_range == AVCOL_RANGE_JPEG ? AV_PIX_FMT_YUVJ422P
+                                                              : AV_PIX_FMT_YUV422P;
+         } else {
 -                                        h264_hwaccel_pixfmt_list_420);
++            int i;
++            const enum AVPixelFormat * fmt = h->avctx->codec->pix_fmts ?
+                                         h->avctx->codec->pix_fmts :
+                                         h->avctx->color_range == AVCOL_RANGE_JPEG ?
+                                         h264_hwaccel_pixfmt_list_jpeg_420 :
 -            c->dsp               = h->dsp;
++                                        h264_hwaccel_pixfmt_list_420;
++
++            for (i=0; fmt[i] != AV_PIX_FMT_NONE; i++)
++                if (fmt[i] == h->avctx->pix_fmt && !force_callback)
++                    return fmt[i];
++            return ff_thread_get_format(h->avctx, fmt);
+         }
+         break;
+     default:
+         av_log(h->avctx, AV_LOG_ERROR,
+                "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
+         return AVERROR_INVALIDDATA;
+     }
+ }
+ /* export coded and cropped frame dimensions to AVCodecContext */
+ static int init_dimensions(H264Context *h)
+ {
+     int width  = h->width  - (h->sps.crop_right + h->sps.crop_left);
+     int height = h->height - (h->sps.crop_top   + h->sps.crop_bottom);
++    av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
++    av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
+     /* handle container cropping */
+     if (!h->sps.crop &&
+         FFALIGN(h->avctx->width,  16) == h->width &&
+         FFALIGN(h->avctx->height, 16) == h->height) {
+         width  = h->avctx->width;
+         height = h->avctx->height;
+     }
+     if (width <= 0 || height <= 0) {
+         av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
+                width, height);
+         if (h->avctx->err_recognition & AV_EF_EXPLODE)
+             return AVERROR_INVALIDDATA;
+         av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
+         h->sps.crop_bottom = h->sps.crop_top = h->sps.crop_right = h->sps.crop_left = 0;
+         h->sps.crop        = 0;
+         width  = h->width;
+         height = h->height;
+     }
+     h->avctx->coded_width  = h->width;
+     h->avctx->coded_height = h->height;
+     h->avctx->width        = width;
+     h->avctx->height       = height;
+     return 0;
+ }
+ static int h264_slice_header_init(H264Context *h, int reinit)
+ {
+     int nb_slices = (HAVE_THREADS &&
+                      h->avctx->active_thread_type & FF_THREAD_SLICE) ?
+                     h->avctx->thread_count : 1;
+     int i, ret;
+     h->avctx->sample_aspect_ratio = h->sps.sar;
+     av_assert0(h->avctx->sample_aspect_ratio.den);
+     av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
+                                      &h->chroma_x_shift, &h->chroma_y_shift);
+     if (h->sps.timing_info_present_flag) {
+         int64_t den = h->sps.time_scale;
+         if (h->x264_build < 44U)
+             den *= 2;
+         av_reduce(&h->avctx->time_base.num, &h->avctx->time_base.den,
+                   h->sps.num_units_in_tick, den, 1 << 30);
+     }
+     h->avctx->hwaccel = ff_find_hwaccel(h->avctx);
+     if (reinit)
+         ff_h264_free_tables(h, 0);
+     h->first_field           = 0;
+     h->prev_interlaced_frame = 1;
+     init_scan_tables(h);
+     ret = ff_h264_alloc_tables(h);
+     if (ret < 0) {
+         av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
+         return ret;
+     }
+     if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
+         int max_slices;
+         if (h->mb_height)
+             max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
+         else
+             max_slices = H264_MAX_THREADS;
+         av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
+                " reducing to %d\n", nb_slices, max_slices);
+         nb_slices = max_slices;
+     }
+     h->slice_context_count = nb_slices;
+     if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
+         ret = ff_h264_context_init(h);
+         if (ret < 0) {
+             av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
+             return ret;
+         }
+     } else {
+         for (i = 1; i < h->slice_context_count; i++) {
+             H264Context *c;
+             c                    = h->thread_context[i] = av_mallocz(sizeof(H264Context));
+             if (!c)
+                 return AVERROR(ENOMEM);
+             c->avctx             = h->avctx;
 -    int default_ref_list_done = 0;
++            if (CONFIG_ERROR_RESILIENCE) {
++                c->dsp               = h->dsp;
++            }
+             c->vdsp              = h->vdsp;
+             c->h264dsp           = h->h264dsp;
+             c->h264qpel          = h->h264qpel;
+             c->h264chroma        = h->h264chroma;
+             c->sps               = h->sps;
+             c->pps               = h->pps;
+             c->pixel_shift       = h->pixel_shift;
++            c->cur_chroma_format_idc = h->cur_chroma_format_idc;
+             c->width             = h->width;
+             c->height            = h->height;
+             c->linesize          = h->linesize;
+             c->uvlinesize        = h->uvlinesize;
+             c->chroma_x_shift    = h->chroma_x_shift;
+             c->chroma_y_shift    = h->chroma_y_shift;
+             c->qscale            = h->qscale;
+             c->droppable         = h->droppable;
+             c->data_partitioning = h->data_partitioning;
+             c->low_delay         = h->low_delay;
+             c->mb_width          = h->mb_width;
+             c->mb_height         = h->mb_height;
+             c->mb_stride         = h->mb_stride;
+             c->mb_num            = h->mb_num;
+             c->flags             = h->flags;
+             c->workaround_bugs   = h->workaround_bugs;
+             c->pict_type         = h->pict_type;
+             init_scan_tables(c);
+             clone_tables(c, h, i);
+             c->context_initialized = 1;
+         }
+         for (i = 0; i < h->slice_context_count; i++)
+             if ((ret = ff_h264_context_init(h->thread_context[i])) < 0) {
+                 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
+                 return ret;
+             }
+     }
+     h->context_initialized = 1;
+     return 0;
+ }
++static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
++{
++    switch (a) {
++    case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
++    case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
++    case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
++    default:
++        return a;
++    }
++}
++
+ /**
+  * Decode a slice header.
+  * This will (re)intialize the decoder and call h264_frame_start() as needed.
+  *
+  * @param h h264context
+  * @param h0 h264 master context (differs from 'h' when doing sliced based
+  *           parallel decoding)
+  *
+  * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
+  */
+ int ff_h264_decode_slice_header(H264Context *h, H264Context *h0)
+ {
+     unsigned int first_mb_in_slice;
+     unsigned int pps_id;
+     int ret;
+     unsigned int slice_type, tmp, i, j;
 -    first_mb_in_slice = get_ue_golomb(&h->gb);
+     int last_pic_structure, last_pic_droppable;
++    int must_reinit;
+     int needs_reinit = 0;
+     int field_pic_flag, bottom_field_flag;
+     h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
+     h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
 -    if (slice_type == AV_PICTURE_TYPE_I ||
 -        (h0->current_slice != 0 && slice_type == h0->last_slice_type)) {
 -        default_ref_list_done = 1;
 -    }
++    first_mb_in_slice = get_ue_golomb_long(&h->gb);
+     if (first_mb_in_slice == 0) { // FIXME better field boundary detection
+         if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
+             ff_h264_field_end(h, 1);
+         }
+         h0->current_slice = 0;
+         if (!h0->first_field) {
+             if (h->cur_pic_ptr && !h->droppable) {
+                 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
+                                           h->picture_structure == PICT_BOTTOM_FIELD);
+             }
+             h->cur_pic_ptr = NULL;
+         }
+     }
+     slice_type = get_ue_golomb_31(&h->gb);
+     if (slice_type > 9) {
+         av_log(h->avctx, AV_LOG_ERROR,
+                "slice type %d too large at %d %d\n",
+                slice_type, h->mb_x, h->mb_y);
+         return AVERROR_INVALIDDATA;
+     }
+     if (slice_type > 4) {
+         slice_type -= 5;
+         h->slice_type_fixed = 1;
+     } else
+         h->slice_type_fixed = 0;
+     slice_type = golomb_to_pict_type[slice_type];
 -        h0->sps_buffers[h->pps.sps_id]->new = 0;
+     h->slice_type     = slice_type;
+     h->slice_type_nos = slice_type & 3;
+     if (h->nal_unit_type  == NAL_IDR_SLICE &&
+         h->slice_type_nos != AV_PICTURE_TYPE_I) {
+         av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
+         return AVERROR_INVALIDDATA;
+     }
+     // to make a few old functions happy, it's wrong though
+     h->pict_type = h->slice_type;
+     pps_id = get_ue_golomb(&h->gb);
+     if (pps_id >= MAX_PPS_COUNT) {
+         av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
+         return AVERROR_INVALIDDATA;
+     }
+     if (!h0->pps_buffers[pps_id]) {
+         av_log(h->avctx, AV_LOG_ERROR,
+                "non-existing PPS %u referenced\n",
+                pps_id);
+         return AVERROR_INVALIDDATA;
+     }
++    if (h0->au_pps_id >= 0 && pps_id != h0->au_pps_id) {
++        av_log(h->avctx, AV_LOG_ERROR,
++               "PPS change from %d to %d forbidden\n",
++               h0->au_pps_id, pps_id);
++        return AVERROR_INVALIDDATA;
++    }
+     h->pps = *h0->pps_buffers[pps_id];
+     if (!h0->sps_buffers[h->pps.sps_id]) {
+         av_log(h->avctx, AV_LOG_ERROR,
+                "non-existing SPS %u referenced\n",
+                h->pps.sps_id);
+         return AVERROR_INVALIDDATA;
+     }
+     if (h->pps.sps_id != h->sps.sps_id ||
++        h->pps.sps_id != h->current_sps_id ||
+         h0->sps_buffers[h->pps.sps_id]->new) {
 -    if (h->mb_width  != h->sps.mb_width ||
 -        h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag))
 -        needs_reinit = 1;
+         h->sps = *h0->sps_buffers[h->pps.sps_id];
++        if (h->mb_width  != h->sps.mb_width ||
++            h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
++            h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
++            h->cur_chroma_format_idc != h->sps.chroma_format_idc
++        )
++            needs_reinit = 1;
++
+         if (h->bit_depth_luma    != h->sps.bit_depth_luma ||
+             h->chroma_format_idc != h->sps.chroma_format_idc) {
+             h->bit_depth_luma    = h->sps.bit_depth_luma;
+             h->chroma_format_idc = h->sps.chroma_format_idc;
+             needs_reinit         = 1;
+         }
+         if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
+             return ret;
+     }
+     h->avctx->profile = ff_h264_get_profile(&h->sps);
+     h->avctx->level   = h->sps.level_idc;
+     h->avctx->refs    = h->sps.ref_frame_count;
 -        h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
 -                                                  : AVCOL_RANGE_MPEG;
++    must_reinit = (h->context_initialized &&
++                    (   16*h->sps.mb_width != h->avctx->coded_width
++                     || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
++                     || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
++                     || h->cur_chroma_format_idc != h->sps.chroma_format_idc
++                     || av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio)
++                     || h->mb_width  != h->sps.mb_width
++                     || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
++                    ));
++    if (non_j_pixfmt(h0->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h0, 0)))
++        must_reinit = 1;
+     h->mb_width  = h->sps.mb_width;
+     h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
+     h->mb_num    = h->mb_width * h->mb_height;
+     h->mb_stride = h->mb_width + 1;
+     h->b_stride = h->mb_width * 4;
+     h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
+     h->width  = 16 * h->mb_width;
+     h->height = 16 * h->mb_height;
+     ret = init_dimensions(h);
+     if (ret < 0)
+         return ret;
+     if (h->sps.video_signal_type_present_flag) {
 -        if ((ret = get_pixel_format(h)) < 0)
++        h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
++                                                    : AVCOL_RANGE_MPEG;
+         if (h->sps.colour_description_present_flag) {
+             if (h->avctx->colorspace != h->sps.colorspace)
+                 needs_reinit = 1;
+             h->avctx->color_primaries = h->sps.color_primaries;
+             h->avctx->color_trc       = h->sps.color_trc;
+             h->avctx->colorspace      = h->sps.colorspace;
+         }
+     }
+     if (h->context_initialized &&
+         (h->width  != h->avctx->coded_width   ||
+          h->height != h->avctx->coded_height  ||
++         must_reinit ||
+          needs_reinit)) {
+         if (h != h0) {
+             av_log(h->avctx, AV_LOG_ERROR,
+                    "changing width %d -> %d / height %d -> %d on "
+                    "slice %d\n",
+                    h->width, h->avctx->coded_width,
+                    h->height, h->avctx->coded_height,
+                    h0->current_slice + 1);
+             return AVERROR_INVALIDDATA;
+         }
+         ff_h264_flush_change(h);
 -               "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
++        if ((ret = get_pixel_format(h, 1)) < 0)
+             return ret;
+         h->avctx->pix_fmt = ret;
+         av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
 -        if ((ret = get_pixel_format(h)) < 0)
++               "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
+         if ((ret = h264_slice_header_init(h, 1)) < 0) {
+             av_log(h->avctx, AV_LOG_ERROR,
+                    "h264_slice_header_init() failed\n");
+             return ret;
+         }
+     }
+     if (!h->context_initialized) {
+         if (h != h0) {
+             av_log(h->avctx, AV_LOG_ERROR,
+                    "Cannot (re-)initialize context during parallel decoding.\n");
+             return AVERROR_PATCHWELCOME;
+         }
 -         * since that can modify s->current_picture_ptr. */
++        if ((ret = get_pixel_format(h, 1)) < 0)
+             return ret;
+         h->avctx->pix_fmt = ret;
+         if ((ret = h264_slice_header_init(h, 0)) < 0) {
+             av_log(h->avctx, AV_LOG_ERROR,
+                    "h264_slice_header_init() failed\n");
+             return ret;
+         }
+     }
+     if (h == h0 && h->dequant_coeff_pps != pps_id) {
+         h->dequant_coeff_pps = pps_id;
+         h264_init_dequant_tables(h);
+     }
+     h->frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
+     h->mb_mbaff        = 0;
+     h->mb_aff_frame    = 0;
+     last_pic_structure = h0->picture_structure;
+     last_pic_droppable = h0->droppable;
+     h->droppable       = h->nal_ref_idc == 0;
+     if (h->sps.frame_mbs_only_flag) {
+         h->picture_structure = PICT_FRAME;
+     } else {
++        if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
++            av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
++            return -1;
++        }
+         field_pic_flag = get_bits1(&h->gb);
+         if (field_pic_flag) {
+             bottom_field_flag = get_bits1(&h->gb);
+             h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
+         } else {
+             h->picture_structure = PICT_FRAME;
+             h->mb_aff_frame      = h->sps.mb_aff;
+         }
+     }
+     h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
+     if (h0->current_slice != 0) {
+         if (last_pic_structure != h->picture_structure ||
+             last_pic_droppable != h->droppable) {
+             av_log(h->avctx, AV_LOG_ERROR,
+                    "Changing field mode (%d -> %d) between slices is not allowed\n",
+                    last_pic_structure, h->picture_structure);
+             h->picture_structure = last_pic_structure;
+             h->droppable         = last_pic_droppable;
+             return AVERROR_INVALIDDATA;
+         } else if (!h0->cur_pic_ptr) {
+             av_log(h->avctx, AV_LOG_ERROR,
+                    "unset cur_pic_ptr on slice %d\n",
+                    h0->current_slice + 1);
+             return AVERROR_INVALIDDATA;
+         }
+     } else {
+         /* Shorten frame num gaps so we don't have to allocate reference
+          * frames just to throw them away */
+         if (h->frame_num != h->prev_frame_num) {
+             int unwrap_prev_frame_num = h->prev_frame_num;
+             int max_frame_num         = 1 << h->sps.log2_max_frame_num;
+             if (unwrap_prev_frame_num > h->frame_num)
+                 unwrap_prev_frame_num -= max_frame_num;
+             if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
+                 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
+                 if (unwrap_prev_frame_num < 0)
+                     unwrap_prev_frame_num += max_frame_num;
+                 h->prev_frame_num = unwrap_prev_frame_num;
+             }
+         }
+         /* See if we have a decoded first field looking for a pair...
+          * Here, we're using that to see if we should mark previously
+          * decode frames as "finished".
+          * We have to do that before the "dummy" in-between frame allocation,
 -                if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
++         * since that can modify h->cur_pic_ptr. */
+         if (h0->first_field) {
+             assert(h0->cur_pic_ptr);
+             assert(h0->cur_pic_ptr->f.buf[0]);
+             assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
++            /* Mark old field/frame as completed */
++            if (h0->cur_pic_ptr->tf.owner == h0->avctx) {
++                ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
++                                          last_pic_structure == PICT_BOTTOM_FIELD);
++            }
++
+             /* figure out if we have a complementary field pair */
+             if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
+                 /* Previous field is unmatched. Don't display it, but let it
+                  * remain for reference if marked as such. */
 -                    if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
++                if (last_pic_structure != PICT_FRAME) {
+                     ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
+                                               last_pic_structure == PICT_TOP_FIELD);
+                 }
+             } else {
+                 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
+                     /* This and previous field were reference, but had
+                      * different frame_nums. Consider this field first in
+                      * pair. Throw away previous field except for reference
+                      * purposes. */
 -        while (h->frame_num != h->prev_frame_num &&
++                    if (last_pic_structure != PICT_FRAME) {
+                         ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
+                                                   last_pic_structure == PICT_TOP_FIELD);
+                     }
+                 } else {
+                     /* Second field in complementary pair */
+                     if (!((last_pic_structure   == PICT_TOP_FIELD &&
+                            h->picture_structure == PICT_BOTTOM_FIELD) ||
+                           (last_pic_structure   == PICT_BOTTOM_FIELD &&
+                            h->picture_structure == PICT_TOP_FIELD))) {
+                         av_log(h->avctx, AV_LOG_ERROR,
+                                "Invalid field mode combination %d/%d\n",
+                                last_pic_structure, h->picture_structure);
+                         h->picture_structure = last_pic_structure;
+                         h->droppable         = last_pic_droppable;
+                         return AVERROR_INVALIDDATA;
+                     } else if (last_pic_droppable != h->droppable) {
+                         avpriv_request_sample(h->avctx,
+                                               "Found reference and non-reference fields in the same frame, which");
+                         h->picture_structure = last_pic_structure;
+                         h->droppable         = last_pic_droppable;
+                         return AVERROR_PATCHWELCOME;
+                     }
+                 }
+             }
+         }
 -    assert(h->mb_num == h->mb_width * h->mb_height);
++        while (h->frame_num != h->prev_frame_num && !h0->first_field &&
+                h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
+             H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
+             av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
+                    h->frame_num, h->prev_frame_num);
++            if (!h->sps.gaps_in_frame_num_allowed_flag)
++                for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
++                    h->last_pocs[i] = INT_MIN;
+             ret = h264_frame_start(h);
+             if (ret < 0) {
+                 h0->first_field = 0;
+                 return ret;
+             }
+             h->prev_frame_num++;
+             h->prev_frame_num        %= 1 << h->sps.log2_max_frame_num;
+             h->cur_pic_ptr->frame_num = h->prev_frame_num;
++            h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
+             ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
+             ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
+             ret = ff_generate_sliding_window_mmcos(h, 1);
+             if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
+                 return ret;
+             ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
+             if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
+                 return ret;
+             /* Error concealment: If a ref is missing, copy the previous ref
+              * in its place.
+              * FIXME: Avoiding a memcpy would be nice, but ref handling makes
+              * many assumptions about there being no actual duplicates.
+              * FIXME: This does not copy padding for out-of-frame motion
+              * vectors.  Given we are concealing a lost frame, this probably
+              * is not noticeable by comparison, but it should be fixed. */
+             if (h->short_ref_count) {
+                 if (prev) {
+                     av_image_copy(h->short_ref[0]->f.data,
+                                   h->short_ref[0]->f.linesize,
+                                   (const uint8_t **)prev->f.data,
+                                   prev->f.linesize,
+                                   h->avctx->pix_fmt,
+                                   h->mb_width  * 16,
+                                   h->mb_height * 16);
+                     h->short_ref[0]->poc = prev->poc + 2;
+                 }
+                 h->short_ref[0]->frame_num = h->prev_frame_num;
+             }
+         }
+         /* See if we have a decoded first field looking for a pair...
+          * We're using that to see whether to continue decoding in that
+          * frame, or to allocate a new one. */
+         if (h0->first_field) {
+             assert(h0->cur_pic_ptr);
+             assert(h0->cur_pic_ptr->f.buf[0]);
+             assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
+             /* figure out if we have a complementary field pair */
+             if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
+                 /* Previous field is unmatched. Don't display it, but let it
+                  * remain for reference if marked as such. */
+                 h0->cur_pic_ptr = NULL;
+                 h0->first_field = FIELD_PICTURE(h);
+             } else {
+                 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
++                    ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
++                                              h0->picture_structure==PICT_BOTTOM_FIELD);
+                     /* This and the previous field had different frame_nums.
+                      * Consider this field first in pair. Throw away previous
+                      * one except for reference purposes. */
+                     h0->first_field = 1;
+                     h0->cur_pic_ptr = NULL;
+                 } else {
+                     /* Second field in complementary pair */
+                     h0->first_field = 0;
+                 }
+             }
+         } else {
+             /* Frame or first field in a potentially complementary pair */
+             h0->first_field = FIELD_PICTURE(h);
+         }
+         if (!FIELD_PICTURE(h) || h0->first_field) {
+             if (h264_frame_start(h) < 0) {
+                 h0->first_field = 0;
+                 return AVERROR_INVALIDDATA;
+             }
+         } else {
+             release_unused_pictures(h, 0);
+         }
++        /* Some macroblocks can be accessed before they're available in case
++        * of lost slices, MBAFF or threading. */
++        if (FIELD_PICTURE(h)) {
++            for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
++                memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
++        } else {
++            memset(h->slice_table, -1,
++                (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
++        }
++        h0->last_slice_type = -1;
+     }
+     if (h != h0 && (ret = clone_slice(h, h0)) < 0)
+         return ret;
++    /* can't be in alloc_tables because linesize isn't known there.
++     * FIXME: redo bipred weight to not require extra buffer? */
++    for (i = 0; i < h->slice_context_count; i++)
++        if (h->thread_context[i]) {
++            ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
++            if (ret < 0)
++                return ret;
++        }
++
+     h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
 -    assert(h->mb_y < h->mb_height);
++    av_assert1(h->mb_num == h->mb_width * h->mb_height);
+     if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
+         first_mb_in_slice >= h->mb_num) {
+         av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
+         return AVERROR_INVALIDDATA;
+     }
+     h->resync_mb_x = h->mb_x =  first_mb_in_slice % h->mb_width;
+     h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) <<
+                                FIELD_OR_MBAFF_PICTURE(h);
+     if (h->picture_structure == PICT_BOTTOM_FIELD)
+         h->resync_mb_y = h->mb_y = h->mb_y + 1;
 -    else if (ret == 1)
 -        default_ref_list_done = 0;
++    av_assert1(h->mb_y < h->mb_height);
+     if (h->picture_structure == PICT_FRAME) {
+         h->curr_pic_num = h->frame_num;
+         h->max_pic_num  = 1 << h->sps.log2_max_frame_num;
+     } else {
+         h->curr_pic_num = 2 * h->frame_num + 1;
+         h->max_pic_num  = 1 << (h->sps.log2_max_frame_num + 1);
+     }
+     if (h->nal_unit_type == NAL_IDR_SLICE)
+         get_ue_golomb(&h->gb); /* idr_pic_id */
+     if (h->sps.poc_type == 0) {
+         h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
+         if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
+             h->delta_poc_bottom = get_se_golomb(&h->gb);
+     }
+     if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
+         h->delta_poc[0] = get_se_golomb(&h->gb);
+         if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
+             h->delta_poc[1] = get_se_golomb(&h->gb);
+     }
+     ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
+     if (h->pps.redundant_pic_cnt_present)
+         h->redundant_pic_count = get_ue_golomb(&h->gb);
+     ret = ff_set_ref_count(h);
+     if (ret < 0)
+         return ret;
 -    if (!default_ref_list_done)
 -    if (h->slice_num >= MAX_SLICES) {
 -        av_log(h->avctx, AV_LOG_ERROR,
 -               "Too many slices, increase MAX_SLICES and recompile\n");
++    if (slice_type != AV_PICTURE_TYPE_I &&
++        (h0->current_slice == 0 ||
++         slice_type != h0->last_slice_type ||
++         memcmp(h0->last_ref_count, h0->ref_count, sizeof(h0->ref_count)))) {
++
+         ff_h264_fill_default_ref_list(h);
++    }
+     if (h->slice_type_nos != AV_PICTURE_TYPE_I) {
+        ret = ff_h264_decode_ref_pic_list_reordering(h);
+        if (ret < 0) {
+            h->ref_count[1] = h->ref_count[0] = 0;
+            return ret;
+        }
+     }
+     if ((h->pps.weighted_pred && h->slice_type_nos == AV_PICTURE_TYPE_P) ||
+         (h->pps.weighted_bipred_idc == 1 &&
+          h->slice_type_nos == AV_PICTURE_TYPE_B))
+         ff_pred_weight_table(h);
+     else if (h->pps.weighted_bipred_idc == 2 &&
+              h->slice_type_nos == AV_PICTURE_TYPE_B) {
+         implicit_weight_table(h, -1);
+     } else {
+         h->use_weight = 0;
+         for (i = 0; i < 2; i++) {
+             h->luma_weight_flag[i]   = 0;
+             h->chroma_weight_flag[i] = 0;
+         }
+     }
+     // If frame-mt is enabled, only update mmco tables for the first slice
+     // in a field. Subsequent slices can temporarily clobber h->mmco_index
+     // or h->mmco, which will cause ref list mix-ups and decoding errors
+     // further down the line. This may break decoding if the first slice is
+     // corrupt, thus we only do this if frame-mt is enabled.
+     if (h->nal_ref_idc) {
+         ret = ff_h264_decode_ref_pic_marking(h0, &h->gb,
+                                              !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
+                                              h0->current_slice == 0);
+         if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
+             return AVERROR_INVALIDDATA;
+     }
+     if (FRAME_MBAFF(h)) {
+         ff_h264_fill_mbaff_ref_list(h);
+         if (h->pps.weighted_bipred_idc == 2 && h->slice_type_nos == AV_PICTURE_TYPE_B) {
+             implicit_weight_table(h, 0);
+             implicit_weight_table(h, 1);
+         }
+     }
+     if (h->slice_type_nos == AV_PICTURE_TYPE_B && !h->direct_spatial_mv_pred)
+         ff_h264_direct_dist_scale_factor(h);
+     ff_h264_direct_ref_list_init(h);
+     if (h->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
+         tmp = get_ue_golomb_31(&h->gb);
+         if (tmp > 2) {
+             av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
+             return AVERROR_INVALIDDATA;
+         }
+         h->cabac_init_idc = tmp;
+     }
+     h->last_qscale_diff = 0;
+     tmp = h->pps.init_qp + get_se_golomb(&h->gb);
+     if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
+         av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
+         return AVERROR_INVALIDDATA;
+     }
+     h->qscale       = tmp;
+     h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
+     h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
+     // FIXME qscale / qp ... stuff
+     if (h->slice_type == AV_PICTURE_TYPE_SP)
+         get_bits1(&h->gb); /* sp_for_switch_flag */
+     if (h->slice_type == AV_PICTURE_TYPE_SP ||
+         h->slice_type == AV_PICTURE_TYPE_SI)
+         get_se_golomb(&h->gb); /* slice_qs_delta */
+     h->deblocking_filter     = 1;
+     h->slice_alpha_c0_offset = 0;
+     h->slice_beta_offset     = 0;
+     if (h->pps.deblocking_filter_parameters_present) {
+         tmp = get_ue_golomb_31(&h->gb);
+         if (tmp > 2) {
+             av_log(h->avctx, AV_LOG_ERROR,
+                    "deblocking_filter_idc %u out of range\n", tmp);
+             return AVERROR_INVALIDDATA;
+         }
+         h->deblocking_filter = tmp;
+         if (h->deblocking_filter < 2)
+             h->deblocking_filter ^= 1;  // 1<->0
+         if (h->deblocking_filter) {
+             h->slice_alpha_c0_offset = get_se_golomb(&h->gb) * 2;
+             h->slice_beta_offset     = get_se_golomb(&h->gb) * 2;
+             if (h->slice_alpha_c0_offset >  12 ||
+                 h->slice_alpha_c0_offset < -12 ||
+                 h->slice_beta_offset >  12     ||
+                 h->slice_beta_offset < -12) {
+                 av_log(h->avctx, AV_LOG_ERROR,
+                        "deblocking filter parameters %d %d out of range\n",
+                        h->slice_alpha_c0_offset, h->slice_beta_offset);
+                 return AVERROR_INVALIDDATA;
+             }
+         }
+     }
+     if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
+         (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
+          h->slice_type_nos != AV_PICTURE_TYPE_I) ||
+         (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR  &&
+          h->slice_type_nos == AV_PICTURE_TYPE_B) ||
+         (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
+          h->nal_ref_idc == 0))
+         h->deblocking_filter = 0;
+     if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
+         if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
+             /* Cheat slightly for speed:
+              * Do not bother to deblock across slices. */
+             h->deblocking_filter = 2;
+         } else {
+             h0->max_contexts = 1;
+             if (!h0->single_decode_warning) {
+                 av_log(h->avctx, AV_LOG_INFO,
+                        "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
+                 h0->single_decode_warning = 1;
+             }
+             if (h != h0) {
+                 av_log(h->avctx, AV_LOG_ERROR,
+                        "Deblocking switched inside frame.\n");
+                 return 1;
+             }
+         }
+     }
+     h->qp_thresh = 15 -
+                    FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
+                    FFMAX3(0,
+                           h->pps.chroma_qp_index_offset[0],
+                           h->pps.chroma_qp_index_offset[1]) +
+                    6 * (h->sps.bit_depth_luma - 8);
+     h0->last_slice_type = slice_type;
++    memcpy(h0->last_ref_count, h0->ref_count, sizeof(h0->last_ref_count));
+     h->slice_num        = ++h0->current_slice;
 -            int (*ref2frm)[64] = h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
++
++    if (h->slice_num)
++        h0->slice_row[(h->slice_num-1)&(MAX_SLICES-1)]= h->resync_mb_y;
++    if (   h0->slice_row[h->slice_num&(MAX_SLICES-1)] + 3 >= h->resync_mb_y
++        && h0->slice_row[h->slice_num&(MAX_SLICES-1)] <= h->resync_mb_y
++        && h->slice_num >= MAX_SLICES) {
++        //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
++        av_log(h->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", h->slice_num, MAX_SLICES);
+     }
+     for (j = 0; j < 2; j++) {
+         int id_list[16];
+         int *ref2frm = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][j];
+         for (i = 0; i < 16; i++) {
+             id_list[i] = 60;
+             if (j < h->list_count && i < h->ref_count[j] &&
+                 h->ref_list[j][i].f.buf[0]) {
+                 int k;
+                 AVBuffer *buf = h->ref_list[j][i].f.buf[0]->buffer;
+                 for (k = 0; k < h->short_ref_count; k++)
+                     if (h->short_ref[k]->f.buf[0]->buffer == buf) {
+                         id_list[i] = k;
+                         break;
+                     }
+                 for (k = 0; k < h->long_ref_count; k++)
+                     if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
+                         id_list[i] = h->short_ref_count + k;
+                         break;
+                     }
+             }
+         }
+         ref2frm[0] =
+         ref2frm[1] = -1;
+         for (i = 0; i < 16; i++)
+             ref2frm[i + 2] = 4 * id_list[i] + (h->ref_list[j][i].reference & 3);
+         ref2frm[18 + 0] =
+         ref2frm[18 + 1] = -1;
+         for (i = 16; i < 48; i++)
+             ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
+                              (h->ref_list[j][i].reference & 3);
+     }
++    if (h->ref_count[0]) ff_h264_set_erpic(&h->er.last_pic, &h->ref_list[0][0]);
++    if (h->ref_count[1]) ff_h264_set_erpic(&h->er.next_pic, &h->ref_list[1][0]);
++
++    h->er.ref_count = h->ref_count[0];
++    h0->au_pps_id = pps_id;
++    h->sps.new =
++    h0->sps_buffers[h->pps.sps_id]->new = 0;
++    h->current_sps_id = h->pps.sps_id;
++
+     if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
+         av_log(h->avctx, AV_LOG_DEBUG,
+                "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
+                h->slice_num,
+                (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
+                first_mb_in_slice,
+                av_get_picture_type_char(h->slice_type),
+                h->slice_type_fixed ? " fix" : "",
+                h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
+                pps_id, h->frame_num,
+                h->cur_pic_ptr->field_poc[0],
+                h->cur_pic_ptr->field_poc[1],
+                h->ref_count[0], h->ref_count[1],
+                h->qscale,
+                h->deblocking_filter,
+                h->slice_alpha_c0_offset, h->slice_beta_offset,
+                h->use_weight,
+                h->use_weight == 1 && h->use_weight_chroma ? "c" : "",
+                h->slice_type == AV_PICTURE_TYPE_B ? (h->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
+     }
+     return 0;
+ }
+ int ff_h264_get_slice_type(const H264Context *h)
+ {
+     switch (h->slice_type) {
+     case AV_PICTURE_TYPE_P:
+         return 0;
+     case AV_PICTURE_TYPE_B:
+         return 1;
+     case AV_PICTURE_TYPE_I:
+         return 2;
+     case AV_PICTURE_TYPE_SP:
+         return 3;
+     case AV_PICTURE_TYPE_SI:
+         return 4;
+     default:
+         return AVERROR_INVALIDDATA;
+     }
+ }
+ static av_always_inline void fill_filter_caches_inter(H264Context *h,
+                                                       int mb_type, int top_xy,
+                                                       int left_xy[LEFT_MBS],
+                                                       int top_type,
+                                                       int left_type[LEFT_MBS],
+                                                       int mb_xy, int list)
+ {
+     int b_stride = h->b_stride;
+     int16_t(*mv_dst)[2] = &h->mv_cache[list][scan8[0]];
+     int8_t *ref_cache = &h->ref_cache[list][scan8[0]];
+     if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
+         if (USES_LIST(top_type, list)) {
+             const int b_xy  = h->mb2b_xy[top_xy] + 3 * b_stride;
+             const int b8_xy = 4 * top_xy + 2;
 -                int (*ref2frm)[64] = h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
++            int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
+             AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
+             ref_cache[0 - 1 * 8] =
+             ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
+             ref_cache[2 - 1 * 8] =
+             ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
+         } else {
+             AV_ZERO128(mv_dst - 1 * 8);
+             AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
+         }
+         if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
+             if (USES_LIST(left_type[LTOP], list)) {
+                 const int b_xy  = h->mb2b_xy[left_xy[LTOP]] + 3;
+                 const int b8_xy = 4 * left_xy[LTOP] + 1;
 -        int (*ref2frm)[64] = h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
++                int (*ref2frm)[64] =(void*)( h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
+                 AV_COPY32(mv_dst - 1 +  0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
+                 AV_COPY32(mv_dst - 1 +  8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
+                 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
+                 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
+                 ref_cache[-1 +  0] =
+                 ref_cache[-1 +  8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
+                 ref_cache[-1 + 16] =
+                 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
+             } else {
+                 AV_ZERO32(mv_dst - 1 +  0);
+                 AV_ZERO32(mv_dst - 1 +  8);
+                 AV_ZERO32(mv_dst - 1 + 16);
+                 AV_ZERO32(mv_dst - 1 + 24);
+                 ref_cache[-1 +  0] =
+                 ref_cache[-1 +  8] =
+                 ref_cache[-1 + 16] =
+                 ref_cache[-1 + 24] = LIST_NOT_USED;
+             }
+         }
+     }
+     if (!USES_LIST(mb_type, list)) {
+         fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
+         AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
+         AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
+         AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
+         AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
+         return;
+     }
+     {
+         int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
 -    if (h->droppable)
++        int (*ref2frm)[64] = (void*)(h->ref2frm[h->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
+         uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
+         uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
+         AV_WN32A(&ref_cache[0 * 8], ref01);
+         AV_WN32A(&ref_cache[1 * 8], ref01);
+         AV_WN32A(&ref_cache[2 * 8], ref23);
+         AV_WN32A(&ref_cache[3 * 8], ref23);
+     }
+     {
+         int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
+         AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
+         AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
+         AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
+         AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
+     }
+ }
+ /**
+  *
+  * @return non zero if the loop filter can be skipped
+  */
+ static int fill_filter_caches(H264Context *h, int mb_type)
+ {
+     const int mb_xy = h->mb_xy;
+     int top_xy, left_xy[LEFT_MBS];
+     int top_type, left_type[LEFT_MBS];
+     uint8_t *nnz;
+     uint8_t *nnz_cache;
+     top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
+     /* Wow, what a mess, why didn't they simplify the interlacing & intra
+      * stuff, I can't imagine that these complex rules are worth it. */
+     left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
+     if (FRAME_MBAFF(h)) {
+         const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
+         const int curr_mb_field_flag = IS_INTERLACED(mb_type);
+         if (h->mb_y & 1) {
+             if (left_mb_field_flag != curr_mb_field_flag)
+                 left_xy[LTOP] -= h->mb_stride;
+         } else {
+             if (curr_mb_field_flag)
+                 top_xy += h->mb_stride &
+                           (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
+             if (left_mb_field_flag != curr_mb_field_flag)
+                 left_xy[LBOT] += h->mb_stride;
+         }
+     }
+     h->top_mb_xy        = top_xy;
+     h->left_mb_xy[LTOP] = left_xy[LTOP];
+     h->left_mb_xy[LBOT] = left_xy[LBOT];
+     {
+         /* For sufficiently low qp, filtering wouldn't do anything.
+          * This is a conservative estimate: could also check beta_offset
+          * and more accurate chroma_qp. */
+         int qp_thresh = h->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
+         int qp        = h->cur_pic.qscale_table[mb_xy];
+         if (qp <= qp_thresh &&
+             (left_xy[LTOP] < 0 ||
+              ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
+             (top_xy < 0 ||
+              ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
+             if (!FRAME_MBAFF(h))
+                 return 1;
+             if ((left_xy[LTOP] < 0 ||
+                  ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
+                 (top_xy < h->mb_stride ||
+                  ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
+                 return 1;
+         }
+     }
+     top_type        = h->cur_pic.mb_type[top_xy];
+     left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
+     left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
+     if (h->deblocking_filter == 2) {
+         if (h->slice_table[top_xy] != h->slice_num)
+             top_type = 0;
+         if (h->slice_table[left_xy[LBOT]] != h->slice_num)
+             left_type[LTOP] = left_type[LBOT] = 0;
+     } else {
+         if (h->slice_table[top_xy] == 0xFFFF)
+             top_type = 0;
+         if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
+             left_type[LTOP] = left_type[LBOT] = 0;
+     }
+     h->top_type        = top_type;
+     h->left_type[LTOP] = left_type[LTOP];
+     h->left_type[LBOT] = left_type[LBOT];
+     if (IS_INTRA(mb_type))
+         return 0;
+     fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
+                              top_type, left_type, mb_xy, 0);
+     if (h->list_count == 2)
+         fill_filter_caches_inter(h, mb_type, top_xy, left_xy,
+                                  top_type, left_type, mb_xy, 1);
+     nnz       = h->non_zero_count[mb_xy];
+     nnz_cache = h->non_zero_count_cache;
+     AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
+     AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
+     AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
+     AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
+     h->cbp = h->cbp_table[mb_xy];
+     if (top_type) {
+         nnz = h->non_zero_count[top_xy];
+         AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
+     }
+     if (left_type[LTOP]) {
+         nnz = h->non_zero_count[left_xy[LTOP]];
+         nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
+         nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
+         nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
+         nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
+     }
+     /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
+      * from what the loop filter needs */
+     if (!CABAC(h) && h->pps.transform_8x8_mode) {
+         if (IS_8x8DCT(top_type)) {
+             nnz_cache[4 + 8 * 0] =
+             nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
+             nnz_cache[6 + 8 * 0] =
+             nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
+         }
+         if (IS_8x8DCT(left_type[LTOP])) {
+             nnz_cache[3 + 8 * 1] =
+             nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
+         }
+         if (IS_8x8DCT(left_type[LBOT])) {
+             nnz_cache[3 + 8 * 3] =
+             nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
+         }
+         if (IS_8x8DCT(mb_type)) {
+             nnz_cache[scan8[0]] =
+             nnz_cache[scan8[1]] =
+             nnz_cache[scan8[2]] =
+             nnz_cache[scan8[3]] = (h->cbp & 0x1000) >> 12;
+             nnz_cache[scan8[0 + 4]] =
+             nnz_cache[scan8[1 + 4]] =
+             nnz_cache[scan8[2 + 4]] =
+             nnz_cache[scan8[3 + 4]] = (h->cbp & 0x2000) >> 12;
+             nnz_cache[scan8[0 + 8]] =
+             nnz_cache[scan8[1 + 8]] =
+             nnz_cache[scan8[2 + 8]] =
+             nnz_cache[scan8[3 + 8]] = (h->cbp & 0x4000) >> 12;
+             nnz_cache[scan8[0 + 12]] =
+             nnz_cache[scan8[1 + 12]] =
+             nnz_cache[scan8[2 + 12]] =
+             nnz_cache[scan8[3 + 12]] = (h->cbp & 0x8000) >> 12;
+         }
+     }
+     return 0;
+ }
+ static void loop_filter(H264Context *h, int start_x, int end_x)
+ {
+     uint8_t *dest_y, *dest_cb, *dest_cr;
+     int linesize, uvlinesize, mb_x, mb_y;
+     const int end_mb_y       = h->mb_y + FRAME_MBAFF(h);
+     const int old_slice_type = h->slice_type;
+     const int pixel_shift    = h->pixel_shift;
+     const int block_h        = 16 >> h->chroma_y_shift;
+     if (h->deblocking_filter) {
+         for (mb_x = start_x; mb_x < end_x; mb_x++)
+             for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
+                 int mb_xy, mb_type;
+                 mb_xy         = h->mb_xy = mb_x + mb_y * h->mb_stride;
+                 h->slice_num  = h->slice_table[mb_xy];
+                 mb_type       = h->cur_pic.mb_type[mb_xy];
+                 h->list_count = h->list_counts[mb_xy];
+                 if (FRAME_MBAFF(h))
+                     h->mb_mbaff               =
+                     h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
+                 h->mb_x = mb_x;
+                 h->mb_y = mb_y;
+                 dest_y  = h->cur_pic.f.data[0] +
+                           ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
+                 dest_cb = h->cur_pic.f.data[1] +
+                           (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
+                           mb_y * h->uvlinesize * block_h;
+                 dest_cr = h->cur_pic.f.data[2] +
+                           (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
+                           mb_y * h->uvlinesize * block_h;
+                 // FIXME simplify above
+                 if (MB_FIELD(h)) {
+                     linesize   = h->mb_linesize   = h->linesize   * 2;
+                     uvlinesize = h->mb_uvlinesize = h->uvlinesize * 2;
+                     if (mb_y & 1) { // FIXME move out of this function?
+                         dest_y  -= h->linesize   * 15;
+                         dest_cb -= h->uvlinesize * (block_h - 1);
+                         dest_cr -= h->uvlinesize * (block_h - 1);
+                     }
+                 } else {
+                     linesize   = h->mb_linesize   = h->linesize;
+                     uvlinesize = h->mb_uvlinesize = h->uvlinesize;
+                 }
+                 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
+                                  uvlinesize, 0);
+                 if (fill_filter_caches(h, mb_type))
+                     continue;
+                 h->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
+                 h->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
+                 if (FRAME_MBAFF(h)) {
+                     ff_h264_filter_mb(h, mb_x, mb_y, dest_y, dest_cb, dest_cr,
+                                       linesize, uvlinesize);
+                 } else {
+                     ff_h264_filter_mb_fast(h, mb_x, mb_y, dest_y, dest_cb,
+                                            dest_cr, linesize, uvlinesize);
+                 }
+             }
+     }
+     h->slice_type   = old_slice_type;
+     h->mb_x         = end_x;
+     h->mb_y         = end_mb_y - FRAME_MBAFF(h);
+     h->chroma_qp[0] = get_chroma_qp(h, 0, h->qscale);
+     h->chroma_qp[1] = get_chroma_qp(h, 1, h->qscale);
+ }
+ static void predict_field_decoding_flag(H264Context *h)
+ {
+     const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
+     int mb_type     = (h->slice_table[mb_xy - 1] == h->slice_num) ?
+                       h->cur_pic.mb_type[mb_xy - 1] :
+                       (h->slice_table[mb_xy - h->mb_stride] == h->slice_num) ?
+                       h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
+     h->mb_mbaff     = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
+ }
+ /**
+  * Draw edges and report progress for the last MB row.
+  */
+ static void decode_finish_row(H264Context *h)
+ {
+     int top            = 16 * (h->mb_y      >> FIELD_PICTURE(h));
+     int pic_height     = 16 *  h->mb_height >> FIELD_PICTURE(h);
+     int height         =  16      << FRAME_MBAFF(h);
+     int deblock_border = (16 + 4) << FRAME_MBAFF(h);
+     if (h->deblocking_filter) {
+         if ((top + height) >= pic_height)
+             height += deblock_border;
+         top -= deblock_border;
+     }
+     if (top >= pic_height || (top + height) < 0)
+         return;
+     height = FFMIN(height, pic_height - top);
+     if (top < 0) {
+         height = top + height;
+         top    = 0;
+     }
+     ff_h264_draw_horiz_band(h, top, height);
 -#if CONFIG_ERROR_RESILIENCE
 -    ERContext *er = &h->er;
++    if (h->droppable || h->er.error_occurred)
+         return;
+     ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
+                               h->picture_structure == PICT_BOTTOM_FIELD);
+ }
+ static void er_add_slice(H264Context *h, int startx, int starty,
+                          int endx, int endy, int status)
+ {
 -    er->ref_count = h->ref_count[0];
 -    ff_er_add_slice(er, startx, starty, endx, endy, status);
 -#endif
++    if (CONFIG_ERROR_RESILIENCE) {
++        ERContext *er = &h->er;
 -            if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
++        ff_er_add_slice(er, startx, starty, endx, endy, status);
++    }
+ }
+ static int decode_slice(struct AVCodecContext *avctx, void *arg)
+ {
+     H264Context *h = *(void **)arg;
+     int lf_x_start = h->mb_x;
+     h->mb_skip_run = -1;
++    av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3));
++
+     h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
+                     avctx->codec_id != AV_CODEC_ID_H264 ||
+                     (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
++    if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->er.error_status_table) {
++        const int start_i  = av_clip(h->resync_mb_x + h->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
++        if (start_i) {
++            int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]];
++            prev_status &= ~ VP_START;
++            if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
++                h->er.error_occurred = 1;
++        }
++    }
++
+     if (h->pps.cabac) {
+         /* realign */
+         align_get_bits(&h->gb);
+         /* init cabac */
+         ff_init_cabac_decoder(&h->cabac,
+                               h->gb.buffer + get_bits_count(&h->gb) / 8,
+                               (get_bits_left(&h->gb) + 7) / 8);
+         ff_h264_init_cabac_states(h);
+         for (;;) {
+             // START_TIMER
+             int ret = ff_h264_decode_mb_cabac(h);
+             int eos;
+             // STOP_TIMER("decode_mb_cabac")
+             if (ret >= 0)
+                 ff_h264_hl_decode_mb(h);
+             // FIXME optimal? or let mb_decode decode 16x32 ?
+             if (ret >= 0 && FRAME_MBAFF(h)) {
+                 h->mb_y++;
+                 ret = ff_h264_decode_mb_cabac(h);
+                 if (ret >= 0)
+                     ff_h264_hl_decode_mb(h);
+                 h->mb_y--;
+             }
+             eos = get_cabac_terminate(&h->cabac);
+             if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
+                 h->cabac.bytestream > h->cabac.bytestream_end + 2) {
+                 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
+                              h->mb_y, ER_MB_END);
+                 if (h->mb_x >= lf_x_start)
+                     loop_filter(h, lf_x_start, h->mb_x + 1);
+                 return 0;
+             }
 -                    if (get_bits_left(&h->gb) == 0) {
++            if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
++                av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %td\n", h->cabac.bytestream_end - h->cabac.bytestream);
++            if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
+                 av_log(h->avctx, AV_LOG_ERROR,
+                        "error while decoding MB %d %d, bytestream %td\n",
+                        h->mb_x, h->mb_y,
+                        h->cabac.bytestream_end - h->cabac.bytestream);
+                 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
+                              h->mb_y, ER_MB_ERROR);
+                 return AVERROR_INVALIDDATA;
+             }
+             if (++h->mb_x >= h->mb_width) {
+                 loop_filter(h, lf_x_start, h->mb_x);
+                 h->mb_x = lf_x_start = 0;
+                 decode_finish_row(h);
+                 ++h->mb_y;
+                 if (FIELD_OR_MBAFF_PICTURE(h)) {
+                     ++h->mb_y;
+                     if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
+                         predict_field_decoding_flag(h);
+                 }
+             }
+             if (eos || h->mb_y >= h->mb_height) {
+                 tprintf(h->avctx, "slice end %d %d\n",
+                         get_bits_count(&h->gb), h->gb.size_in_bits);
+                 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
+                              h->mb_y, ER_MB_END);
+                 if (h->mb_x > lf_x_start)
+                     loop_filter(h, lf_x_start, h->mb_x);
+                 return 0;
+             }
+         }
+     } else {
+         for (;;) {
+             int ret = ff_h264_decode_mb_cavlc(h);
+             if (ret >= 0)
+                 ff_h264_hl_decode_mb(h);
+             // FIXME optimal? or let mb_decode decode 16x32 ?
+             if (ret >= 0 && FRAME_MBAFF(h)) {
+                 h->mb_y++;
+                 ret = ff_h264_decode_mb_cavlc(h);
+                 if (ret >= 0)
+                     ff_h264_hl_decode_mb(h);
+                 h->mb_y--;
+             }
+             if (ret < 0) {
+                 av_log(h->avctx, AV_LOG_ERROR,
+                        "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
+                 er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
+                              h->mb_y, ER_MB_ERROR);
+                 return ret;
+             }
+             if (++h->mb_x >= h->mb_width) {
+                 loop_filter(h, lf_x_start, h->mb_x);
+                 h->mb_x = lf_x_start = 0;
+                 decode_finish_row(h);
+                 ++h->mb_y;
+                 if (FIELD_OR_MBAFF_PICTURE(h)) {
+                     ++h->mb_y;
+                     if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
+                         predict_field_decoding_flag(h);
+                 }
+                 if (h->mb_y >= h->mb_height) {
+                     tprintf(h->avctx, "slice end %d %d\n",
+                             get_bits_count(&h->gb), h->gb.size_in_bits);
 -                                     h->mb_x - 1, h->mb_y,
++                    if (   get_bits_left(&h->gb) == 0
++                        || get_bits_left(&h->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
+                         er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
+                                      h->mb_x - 1, h->mb_y,
+                                      ER_MB_END);
+                         return 0;
+                     } else {
+                         er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
 -    if (h->mb_y >= h->mb_height) {
 -        av_log(h->avctx, AV_LOG_ERROR,
 -               "Input contains more MB rows than the frame height.\n");
 -        return AVERROR_INVALIDDATA;
 -    }
++                                     h->mb_x, h->mb_y,
+                                      ER_MB_END);
+                         return AVERROR_INVALIDDATA;
+                     }
+                 }
+             }
+             if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
+                 tprintf(h->avctx, "slice end %d %d\n",
+                         get_bits_count(&h->gb), h->gb.size_in_bits);
+                 if (get_bits_left(&h->gb) == 0) {
+                     er_add_slice(h, h->resync_mb_x, h->resync_mb_y,
+                                  h->mb_x - 1, h->mb_y,
+                                  ER_MB_END);
+                     if (h->mb_x > lf_x_start)
+                         loop_filter(h, lf_x_start, h->mb_x);
+                     return 0;
+                 } else {
+                     er_add_slice(h, h->resync_mb_x, h->resync_mb_y, h->mb_x,
+                                  h->mb_y, ER_MB_ERROR);
+                     return AVERROR_INVALIDDATA;
+                 }
+             }
+         }
+     }
+ }
+ /**
+  * Call decode_slice() for each context.
+  *
+  * @param h h264 master context
+  * @param context_count number of contexts to execute
+  */
+ int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
+ {
+     AVCodecContext *const avctx = h->avctx;
+     H264Context *hx;
+     int i;
 -    if (h->avctx->hwaccel)
++    av_assert0(h->mb_y < h->mb_height);
 -            hx->er.error_count = 0;
++    if (h->avctx->hwaccel ||
++        h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
+         return 0;
+     if (context_count == 1) {
+         return decode_slice(avctx, &h);
+     } else {
++        av_assert0(context_count > 0);
+         for (i = 1; i < context_count; i++) {
+             hx                 = h->thread_context[i];
 -        for (i = 1; i < context_count; i++)
 -            h->er.error_count += h->thread_context[i]->er.error_count;
++            if (CONFIG_ERROR_RESILIENCE) {
++                hx->er.error_count = 0;
++            }
++            hx->x264_build     = h->x264_build;
+         }
+         avctx->execute(avctx, decode_slice, h->thread_context,
+                        NULL, context_count, sizeof(void *));
+         /* pull back stuff from slices to master context */
+         hx                   = h->thread_context[context_count - 1];
+         h->mb_x              = hx->mb_x;
+         h->mb_y              = hx->mb_y;
+         h->droppable         = hx->droppable;
+         h->picture_structure = hx->picture_structure;
++        if (CONFIG_ERROR_RESILIENCE) {
++            for (i = 1; i < context_count; i++)
++                h->er.error_count += h->thread_context[i]->er.error_count;
++        }
+     }
+     return 0;
+ }