]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/h264_slice.c
h264: rename mmco_index to nb_mmco
[ffmpeg] / libavcodec / h264_slice.c
index 23824db888023f984e967f653721f234ef803ccb..634f181be5ad5849fa2f794d58c3c64d7ad203a3 100644 (file)
@@ -21,7 +21,7 @@
 
 /**
  * @file
- * H.264 / AVC / MPEG4 part10 codec.
+ * H.264 / AVC / MPEG-4 part10 codec.
  * @author Michael Niedermayer <michaelni@gmx.at>
  */
 
@@ -128,7 +128,7 @@ static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
 
     av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
     // edge emu needs blocksize + filter length - 1
-    // (= 21x21 for  h264)
+    // (= 21x21 for  H.264)
     av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
 
     av_fast_malloc(&sl->top_borders[0], &sl->top_borders_allocated[0],
@@ -282,76 +282,6 @@ static int initialize_cur_frame(H264Context *h)
     return 0;
 }
 
-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 = ff_h264_quant_div6[q];
-            int idx   = ff_h264_quant_rem6[q];
-            for (x = 0; x < 64; x++)
-                h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
-                    ((uint32_t)ff_h264_dequant8_coeff_init[idx][ff_h264_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 = ff_h264_quant_div6[q] + 2;
-            int idx   = ff_h264_quant_rem6[q];
-            for (x = 0; x < 16; x++)
-                h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
-                    ((uint32_t)ff_h264_dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
-                     h->pps.scaling_matrix4[i][x]) << shift;
-        }
-    }
-}
-
-void ff_h264_init_dequant_tables(H264Context *h)
-{
-    int i, x;
-    init_dequant4_coeff_table(h);
-    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;
-    }
-}
-
 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
 
 #define REBASE_PICTURE(pic, new_ctx, old_ctx)             \
@@ -374,30 +304,6 @@ static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
     }
 }
 
-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 ff_h264_update_thread_context(AVCodecContext *dst,
@@ -411,28 +317,40 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
     if (dst == src || !h1->context_initialized)
         return 0;
 
+    if (!h1->ps.sps)
+        return AVERROR_INVALIDDATA;
+
     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)) {
+         !h->ps.sps                                            ||
+         h->ps.sps->bit_depth_luma    != h1->ps.sps->bit_depth_luma    ||
+         h->ps.sps->chroma_format_idc != h1->ps.sps->chroma_format_idc ||
+         h->ps.sps->colorspace        != h1->ps.sps->colorspace)) {
         need_reinit = 1;
     }
 
     // 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;
+    for (i = 0; i < FF_ARRAY_ELEMS(h->ps.sps_list); i++) {
+        av_buffer_unref(&h->ps.sps_list[i]);
+        if (h1->ps.sps_list[i]) {
+            h->ps.sps_list[i] = av_buffer_ref(h1->ps.sps_list[i]);
+            if (!h->ps.sps_list[i])
+                return AVERROR(ENOMEM);
+        }
+    }
+    for (i = 0; i < FF_ARRAY_ELEMS(h->ps.pps_list); i++) {
+        av_buffer_unref(&h->ps.pps_list[i]);
+        if (h1->ps.pps_list[i]) {
+            h->ps.pps_list[i] = av_buffer_ref(h1->ps.pps_list[i]);
+            if (!h->ps.pps_list[i])
+                return AVERROR(ENOMEM);
+        }
+    }
+
+    h->ps.sps = h1->ps.sps;
 
     if (need_reinit || !inited) {
         h->width     = h1->width;
@@ -460,7 +378,6 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
     h->first_field          = h1->first_field;
     h->picture_structure    = h1->picture_structure;
     h->droppable            = h1->droppable;
-    h->low_delay            = h1->low_delay;
 
     for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
         ff_h264_unref_picture(h, &h->DPB[i]);
@@ -479,29 +396,30 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
 
     h->enable_er       = h1->enable_er;
     h->workaround_bugs = h1->workaround_bugs;
-    h->low_delay       = h1->low_delay;
     h->droppable       = h1->droppable;
 
     // extradata/NAL handling
     h->is_avc = h1->is_avc;
     h->nal_length_size = h1->nal_length_size;
 
-    // Dequantization matrices
-    // FIXME these are big - can they be only copied when PPS changes?
-    copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
+    memcpy(&h->poc,        &h1->poc,        sizeof(h->poc));
 
-    for (i = 0; i < 6; i++)
-        h->dequant4_coeff[i] = h->dequant4_buffer[0] +
-                               (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
+    h->curr_pic_num      = h1->curr_pic_num;
+    h->max_pic_num       = h1->max_pic_num;
 
-    for (i = 0; i < 6; i++)
-        h->dequant8_coeff[i] = h->dequant8_buffer[0] +
-                               (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
+    memcpy(h->short_ref,   h1->short_ref,   sizeof(h->short_ref));
+    memcpy(h->long_ref,    h1->long_ref,    sizeof(h->long_ref));
+    memcpy(h->delayed_pic, h1->delayed_pic, sizeof(h->delayed_pic));
+    memcpy(h->last_pocs,   h1->last_pocs,   sizeof(h->last_pocs));
 
-    h->dequant_coeff_pps = h1->dequant_coeff_pps;
+    h->next_output_pic   = h1->next_output_pic;
+    h->next_outputed_poc = h1->next_outputed_poc;
 
-    // POC timing
-    copy_fields(h, h1, poc_lsb, current_slice);
+    memcpy(h->mmco, h1->mmco, sizeof(h->mmco));
+    h->nb_mmco         = h1->nb_mmco;
+    h->mmco_reset      = h1->mmco_reset;
+    h->long_ref_count  = h1->long_ref_count;
+    h->short_ref_count = h1->short_ref_count;
 
     copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
     copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
@@ -512,12 +430,12 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
         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;
+        err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->nb_mmco);
+        h->poc.prev_poc_msb = h->poc.poc_msb;
+        h->poc.prev_poc_lsb = h->poc.poc_lsb;
     }
-    h->prev_frame_num_offset = h->frame_num_offset;
-    h->prev_frame_num        = h->frame_num;
+    h->poc.prev_frame_num_offset = h->poc.frame_num_offset;
+    h->poc.prev_frame_num        = h->poc.frame_num;
 
     h->recovery_frame        = h1->recovery_frame;
     h->frame_recovered       = h1->frame_recovered;
@@ -539,7 +457,7 @@ static int h264_frame_start(H264Context *h)
     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;
-    pic->frame_num               = h->frame_num;
+    pic->frame_num               = h->poc.frame_num;
     /*
      * Zero key_frame here; IDR markings per slice in frame or fields are ORed
      * in later.
@@ -549,6 +467,8 @@ static int h264_frame_start(H264Context *h)
     pic->mmco_reset  = 0;
     pic->recovered   = 0;
 
+    pic->f->pict_type = h->slice_ctx[0].slice_type;
+
     if (CONFIG_ERROR_RESILIENCE && h->enable_er)
         ff_er_frame_start(&h->slice_ctx[0].er);
 
@@ -578,6 +498,8 @@ static int h264_frame_start(H264Context *h)
 
     h->next_output_pic = NULL;
 
+    h->postpone_filter = 0;
+
     assert(h->cur_pic_ptr->long_ref == 0);
 
     return 0;
@@ -766,7 +688,7 @@ static void init_scan_tables(H264Context *h)
         h->field_scan8x8_cavlc[i]  = TRANSPOSE(field_scan8x8_cavlc[i]);
 #undef TRANSPOSE
     }
-    if (h->sps.transform_bypass) { // FIXME same ugly
+    if (h->ps.sps->transform_bypass) { // FIXME same ugly
         h->zigzag_scan_q0          = ff_zigzag_scan;
         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
@@ -793,7 +715,7 @@ static enum AVPixelFormat get_pixel_format(H264Context *h)
     enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
     const enum AVPixelFormat *choices = pix_fmts;
 
-    switch (h->sps.bit_depth_luma) {
+    switch (h->ps.sps->bit_depth_luma) {
     case 9:
         if (CHROMA444(h)) {
             if (h->avctx->colorspace == AVCOL_SPC_RGB) {
@@ -856,7 +778,7 @@ static enum AVPixelFormat get_pixel_format(H264Context *h)
         break;
     default:
         av_log(h->avctx, AV_LOG_ERROR,
-               "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
+               "Unsupported bit depth %d\n", h->ps.sps->bit_depth_luma);
         return AVERROR_INVALIDDATA;
     }
 
@@ -868,8 +790,9 @@ static enum AVPixelFormat get_pixel_format(H264Context *h)
 /* 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);
+    SPS *sps = h->ps.sps;
+    int width  = h->width  - (sps->crop_right + sps->crop_left);
+    int height = h->height - (sps->crop_top   + sps->crop_bottom);
 
     /* handle container cropping */
     if (FFALIGN(h->avctx->width,  16) == FFALIGN(width,  16) &&
@@ -885,11 +808,11 @@ static int init_dimensions(H264Context *h)
             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   =
-        h->sps.crop        = 0;
+        sps->crop_bottom =
+        sps->crop_top    =
+        sps->crop_right  =
+        sps->crop_left   =
+        sps->crop        = 0;
 
         width  = h->width;
         height = h->height;
@@ -905,21 +828,19 @@ static int init_dimensions(H264Context *h)
 
 static int h264_slice_header_init(H264Context *h)
 {
-    int nb_slices = (HAVE_THREADS &&
-                     h->avctx->active_thread_type & FF_THREAD_SLICE) ?
-                    h->avctx->thread_count : 1;
+    const SPS *sps = h->ps.sps;
     int i, ret;
 
-    ff_set_sar(h->avctx, h->sps.sar);
+    ff_set_sar(h->avctx, sps->sar);
     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)
+    if (sps->timing_info_present_flag) {
+        int64_t den = sps->time_scale;
+        if (h->sei.unregistered.x264_build < 44U)
             den *= 2;
         av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
-                  h->sps.num_units_in_tick, den, 1 << 30);
+                  sps->num_units_in_tick, den, 1 << 30);
     }
 
     ff_h264_free_tables(h);
@@ -934,36 +855,24 @@ static int h264_slice_header_init(H264Context *h)
         return ret;
     }
 
-    if (h->sps.bit_depth_luma < 8 || h->sps.bit_depth_luma > 10) {
+    if (sps->bit_depth_luma < 8 || sps->bit_depth_luma > 10) {
         av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
-               h->sps.bit_depth_luma);
+               sps->bit_depth_luma);
         return AVERROR_INVALIDDATA;
     }
 
-    h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
-    h->pixel_shift                = h->sps.bit_depth_luma > 8;
-    h->chroma_format_idc          = h->sps.chroma_format_idc;
-    h->bit_depth_luma             = h->sps.bit_depth_luma;
-
-    ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
-                    h->sps.chroma_format_idc);
-    ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
-    ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma);
-    ff_h264_pred_init(&h->hpc, h->avctx->codec_id, h->sps.bit_depth_luma,
-                      h->sps.chroma_format_idc);
-    ff_videodsp_init(&h->vdsp, h->sps.bit_depth_luma);
+    h->avctx->bits_per_raw_sample = sps->bit_depth_luma;
+    h->pixel_shift                = sps->bit_depth_luma > 8;
+    h->chroma_format_idc          = sps->chroma_format_idc;
+    h->bit_depth_luma             = sps->bit_depth_luma;
 
-    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;
+    ff_h264dsp_init(&h->h264dsp, sps->bit_depth_luma,
+                    sps->chroma_format_idc);
+    ff_h264chroma_init(&h->h264chroma, sps->bit_depth_chroma);
+    ff_h264qpel_init(&h->h264qpel, sps->bit_depth_luma);
+    ff_h264_pred_init(&h->hpc, h->avctx->codec_id, sps->bit_depth_luma,
+                      sps->chroma_format_idc);
+    ff_videodsp_init(&h->vdsp, sps->bit_depth_luma);
 
     if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
         ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
@@ -972,7 +881,7 @@ static int h264_slice_header_init(H264Context *h)
             return ret;
         }
     } else {
-        for (i = 0; i < h->slice_context_count; i++) {
+        for (i = 0; i < h->nb_slice_ctx; i++) {
             H264SliceContext *sl = &h->slice_ctx[i];
 
             sl->h264               = h;
@@ -992,32 +901,22 @@ static int h264_slice_header_init(H264Context *h)
     return 0;
 }
 
-/**
- * Decode a slice header.
- * This will (re)intialize the decoder and call h264_frame_start() as needed.
- *
- * @param h h264context
- *
- * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
- */
-int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
+static int h264_slice_header_parse(H264Context *h, H264SliceContext *sl)
 {
-    unsigned int first_mb_in_slice;
+    const SPS *sps;
+    const PPS *pps;
     unsigned int pps_id;
     int ret;
-    unsigned int slice_type, tmp, i, j;
+    unsigned int slice_type, tmp, i;
     int last_pic_structure, last_pic_droppable;
     int needs_reinit = 0;
     int field_pic_flag, bottom_field_flag;
     int frame_num, droppable, picture_structure;
     int mb_aff_frame = 0;
 
-    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(&sl->gb);
+    sl->first_mb_addr = get_ue_golomb(&sl->gb);
 
-    if (first_mb_in_slice == 0) { // FIXME better field boundary detection
+    if (sl->first_mb_addr == 0) { // FIXME better field boundary detection
         if (h->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
             ff_h264_field_end(h, sl, 1);
         }
@@ -1036,7 +935,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
     if (slice_type > 9) {
         av_log(h->avctx, AV_LOG_ERROR,
                "slice type %d too large at %d\n",
-               slice_type, first_mb_in_slice);
+               slice_type, sl->first_mb_addr);
         return AVERROR_INVALIDDATA;
     }
     if (slice_type > 4) {
@@ -1055,77 +954,59 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
         return AVERROR_INVALIDDATA;
     }
 
-    // to make a few old functions happy, it's wrong though
-    if (!h->setup_finished)
-        h->pict_type = sl->slice_type;
-
     pps_id = get_ue_golomb(&sl->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 (!h->pps_buffers[pps_id]) {
+    if (!h->ps.pps_list[pps_id]) {
         av_log(h->avctx, AV_LOG_ERROR,
                "non-existing PPS %u referenced\n",
                pps_id);
         return AVERROR_INVALIDDATA;
     }
     if (!h->setup_finished) {
-        h->pps = *h->pps_buffers[pps_id];
-    } else if (h->dequant_coeff_pps != pps_id) {
+        h->ps.pps = (const PPS*)h->ps.pps_list[pps_id]->data;
+    } else if (h->ps.pps != (const PPS*)h->ps.pps_list[pps_id]->data) {
         av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
         return AVERROR_INVALIDDATA;
     }
 
-    if (!h->sps_buffers[h->pps.sps_id]) {
+    if (!h->ps.sps_list[h->ps.pps->sps_id]) {
         av_log(h->avctx, AV_LOG_ERROR,
                "non-existing SPS %u referenced\n",
-               h->pps.sps_id);
+               h->ps.pps->sps_id);
         return AVERROR_INVALIDDATA;
     }
 
-    if (h->pps.sps_id != h->sps.sps_id ||
-        h->sps_buffers[h->pps.sps_id]->new) {
-        h->sps_buffers[h->pps.sps_id]->new = 0;
+    if (h->ps.sps != (const SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data) {
+        h->ps.sps = (SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data;
 
-        h->sps = *h->sps_buffers[h->pps.sps_id];
-
-        if (h->bit_depth_luma    != h->sps.bit_depth_luma ||
-            h->chroma_format_idc != h->sps.chroma_format_idc)
+        if (h->bit_depth_luma    != h->ps.sps->bit_depth_luma ||
+            h->chroma_format_idc != h->ps.sps->chroma_format_idc)
             needs_reinit         = 1;
-
-        if (h->flags & AV_CODEC_FLAG_LOW_DELAY ||
-            (h->sps.bitstream_restriction_flag &&
-             !h->sps.num_reorder_frames)) {
-            if (h->avctx->has_b_frames > 1 || h->delayed_pic[0])
-                av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. "
-                       "Reenabling low delay requires a codec flush.\n");
-            else
-                h->low_delay = 1;
-        }
-
-        if (h->avctx->has_b_frames < 2)
-            h->avctx->has_b_frames = !h->low_delay;
-
     }
 
+    pps = h->ps.pps;
+    sps = h->ps.sps;
+
     if (!h->setup_finished) {
-        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->profile = ff_h264_get_profile(sps);
+        h->avctx->level   = sps->level_idc;
+        h->avctx->refs    = sps->ref_frame_count;
 
-        if (h->mb_width  != h->sps.mb_width ||
-            h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag))
+        if (h->mb_width  != sps->mb_width ||
+            h->mb_height != sps->mb_height * (2 - sps->frame_mbs_only_flag))
             needs_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_width  = sps->mb_width;
+        h->mb_height = sps->mb_height * (2 - 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->chroma_y_shift = sps->chroma_format_idc <= 1; // 400 uses yuv420p
 
         h->width  = 16 * h->mb_width;
         h->height = 16 * h->mb_height;
@@ -1134,15 +1015,15 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
         if (ret < 0)
             return ret;
 
-        if (h->sps.video_signal_type_present_flag) {
-            h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
+        if (sps->video_signal_type_present_flag) {
+            h->avctx->color_range = sps->full_range ? AVCOL_RANGE_JPEG
                 : AVCOL_RANGE_MPEG;
-            if (h->sps.colour_description_present_flag) {
-                if (h->avctx->colorspace != h->sps.colorspace)
+            if (sps->colour_description_present_flag) {
+                if (h->avctx->colorspace != 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;
+                h->avctx->color_primaries = sps->color_primaries;
+                h->avctx->color_trc       = sps->color_trc;
+                h->avctx->colorspace      = sps->colorspace;
             }
         }
     }
@@ -1192,14 +1073,9 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
         }
     }
 
-    if (sl == h->slice_ctx && h->dequant_coeff_pps != pps_id) {
-        h->dequant_coeff_pps = pps_id;
-        ff_h264_init_dequant_tables(h);
-    }
-
-    frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
+    frame_num = get_bits(&sl->gb, sps->log2_max_frame_num);
     if (!h->setup_finished)
-        h->frame_num = frame_num;
+        h->poc.frame_num = frame_num;
 
     sl->mb_mbaff       = 0;
 
@@ -1207,7 +1083,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
     last_pic_droppable = h->droppable;
 
     droppable = h->nal_ref_idc == 0;
-    if (h->sps.frame_mbs_only_flag) {
+    if (sps->frame_mbs_only_flag) {
         picture_structure = PICT_FRAME;
     } else {
         field_pic_flag = get_bits1(&sl->gb);
@@ -1216,7 +1092,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
             picture_structure = PICT_TOP_FIELD + bottom_field_flag;
         } else {
             picture_structure = PICT_FRAME;
-            mb_aff_frame      = h->sps.mb_aff;
+            mb_aff_frame      = sps->mb_aff;
         }
     }
     if (!h->setup_finished) {
@@ -1242,19 +1118,19 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
     } 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 (h->poc.frame_num != h->poc.prev_frame_num) {
+            int unwrap_prev_frame_num = h->poc.prev_frame_num;
+            int max_frame_num         = 1 << sps->log2_max_frame_num;
 
-            if (unwrap_prev_frame_num > h->frame_num)
+            if (unwrap_prev_frame_num > h->poc.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 ((h->poc.frame_num - unwrap_prev_frame_num) > sps->ref_frame_count) {
+                unwrap_prev_frame_num = (h->poc.frame_num - 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;
+                h->poc.prev_frame_num = unwrap_prev_frame_num;
             }
         }
 
@@ -1277,7 +1153,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
                                               last_pic_structure == PICT_TOP_FIELD);
                 }
             } else {
-                if (h->cur_pic_ptr->frame_num != h->frame_num) {
+                if (h->cur_pic_ptr->frame_num != h->poc.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
@@ -1309,26 +1185,26 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
             }
         }
 
-        while (h->frame_num != h->prev_frame_num &&
-               h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
+        while (h->poc.frame_num != h->poc.prev_frame_num &&
+               h->poc.frame_num != (h->poc.prev_frame_num + 1) % (1 << 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);
+                   h->poc.frame_num, h->poc.prev_frame_num);
             ret = initialize_cur_frame(h);
             if (ret < 0) {
                 h->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->poc.prev_frame_num++;
+            h->poc.prev_frame_num        %= 1 << sps->log2_max_frame_num;
+            h->cur_pic_ptr->frame_num = h->poc.prev_frame_num;
             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);
+            ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->nb_mmco);
             if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
                 return ret;
             /* Error concealment: If a ref is missing, copy the previous ref
@@ -1352,7 +1228,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
                                   h->mb_height * 16);
                     h->short_ref[0]->poc = prev->poc + 2;
                 }
-                h->short_ref[0]->frame_num = h->prev_frame_num;
+                h->short_ref[0]->frame_num = h->poc.prev_frame_num;
             }
         }
 
@@ -1371,7 +1247,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
                 h->cur_pic_ptr = NULL;
                 h->first_field = FIELD_PICTURE(h);
             } else {
-                if (h->cur_pic_ptr->frame_num != h->frame_num) {
+                if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
                     /* This and the previous field had different frame_nums.
                      * Consider this field first in pair. Throw away previous
                      * one except for reference purposes. */
@@ -1397,64 +1273,53 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
         }
     }
 
-    assert(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;
-    }
-    sl->resync_mb_x = sl->mb_x =  first_mb_in_slice % h->mb_width;
-    sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
-                                 FIELD_OR_MBAFF_PICTURE(h);
-    if (h->picture_structure == PICT_BOTTOM_FIELD)
-        sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
-    assert(sl->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;
+        h->curr_pic_num = h->poc.frame_num;
+        h->max_pic_num  = 1 << 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);
+        h->curr_pic_num = 2 * h->poc.frame_num + 1;
+        h->max_pic_num  = 1 << (sps->log2_max_frame_num + 1);
     }
 
     if (h->nal_unit_type == NAL_IDR_SLICE)
         get_ue_golomb(&sl->gb); /* idr_pic_id */
 
-    if (h->sps.poc_type == 0) {
-        int poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
+    if (sps->poc_type == 0) {
+        int poc_lsb = get_bits(&sl->gb, sps->log2_max_poc_lsb);
 
         if (!h->setup_finished)
-            h->poc_lsb = poc_lsb;
+            h->poc.poc_lsb = poc_lsb;
 
-        if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
+        if (pps->pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
             int delta_poc_bottom = get_se_golomb(&sl->gb);
             if (!h->setup_finished)
-                h->delta_poc_bottom = delta_poc_bottom;
+                h->poc.delta_poc_bottom = delta_poc_bottom;
         }
     }
 
-    if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
+    if (sps->poc_type == 1 && !sps->delta_pic_order_always_zero_flag) {
         int delta_poc = get_se_golomb(&sl->gb);
 
         if (!h->setup_finished)
-            h->delta_poc[0] = delta_poc;
+            h->poc.delta_poc[0] = delta_poc;
 
-        if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
+        if (pps->pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
             delta_poc = get_se_golomb(&sl->gb);
 
             if (!h->setup_finished)
-                h->delta_poc[1] = delta_poc;
+                h->poc.delta_poc[1] = delta_poc;
         }
     }
 
-    if (!h->setup_finished)
-        ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
-
-    if (h->pps.redundant_pic_cnt_present)
+    if (pps->redundant_pic_cnt_present)
         sl->redundant_pic_count = get_ue_golomb(&sl->gb);
 
-    ret = ff_set_ref_count(h, sl);
+    if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
+        sl->direct_spatial_mv_pred = get_bits1(&sl->gb);
+
+    ret = ff_h264_parse_ref_count(&sl->list_count, sl->ref_count,
+                                  &sl->gb, pps, sl->slice_type_nos,
+                                  h->picture_structure);
     if (ret < 0)
         return ret;
 
@@ -1466,24 +1331,19 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
        }
     }
 
-    if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
-        (h->pps.weighted_bipred_idc == 1 &&
+    sl->pwt.use_weight = 0;
+    for (i = 0; i < 2; i++) {
+        sl->pwt.luma_weight_flag[i]   = 0;
+        sl->pwt.chroma_weight_flag[i] = 0;
+    }
+    if ((pps->weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
+        (pps->weighted_bipred_idc == 1 &&
          sl->slice_type_nos == AV_PICTURE_TYPE_B))
-        ff_h264_pred_weight_table(&sl->gb, &h->sps, sl->ref_count,
+        ff_h264_pred_weight_table(&sl->gb, sps, sl->ref_count,
                                   sl->slice_type_nos, &sl->pwt);
-    else if (h->pps.weighted_bipred_idc == 2 &&
-             sl->slice_type_nos == AV_PICTURE_TYPE_B) {
-        implicit_weight_table(h, sl, -1);
-    } else {
-        sl->pwt.use_weight = 0;
-        for (i = 0; i < 2; i++) {
-            sl->pwt.luma_weight_flag[i]   = 0;
-            sl->pwt.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
+    // in a field. Subsequent slices can temporarily clobber h->nb_mmco
     // 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.
@@ -1495,20 +1355,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
             return AVERROR_INVALIDDATA;
     }
 
-    if (FRAME_MBAFF(h)) {
-        ff_h264_fill_mbaff_ref_list(h, sl);
-
-        if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
-            implicit_weight_table(h, sl, 0);
-            implicit_weight_table(h, sl, 1);
-        }
-    }
-
-    if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
-        ff_h264_direct_dist_scale_factor(h, sl);
-    ff_h264_direct_ref_list_init(h, sl);
-
-    if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
+    if (sl->slice_type_nos != AV_PICTURE_TYPE_I && pps->cabac) {
         tmp = get_ue_golomb_31(&sl->gb);
         if (tmp > 2) {
             av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
@@ -1518,8 +1365,8 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
     }
 
     sl->last_qscale_diff = 0;
-    tmp = h->pps.init_qp + get_se_golomb(&sl->gb);
-    if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
+    tmp = pps->init_qp + get_se_golomb(&sl->gb);
+    if (tmp > 51 + 6 * (sps->bit_depth_luma - 8)) {
         av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
         return AVERROR_INVALIDDATA;
     }
@@ -1536,7 +1383,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
     sl->deblocking_filter     = 1;
     sl->slice_alpha_c0_offset = 0;
     sl->slice_beta_offset     = 0;
-    if (h->pps.deblocking_filter_parameters_present) {
+    if (pps->deblocking_filter_parameters_present) {
         tmp = get_ue_golomb_31(&sl->gb);
         if (tmp > 2) {
             av_log(h->avctx, AV_LOG_ERROR,
@@ -1562,6 +1409,59 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
         }
     }
 
+    return 0;
+}
+
+/**
+ * Decode a slice header.
+ * This will (re)initialize the decoder and call h264_frame_start() as needed.
+ *
+ * @param h h264context
+ *
+ * @return 0 if okay, <0 if an error occurred
+ */
+int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
+{
+    int i, j, ret = 0;
+
+    ret = h264_slice_header_parse(h, sl);
+    if (ret < 0)
+        return ret;
+
+    assert(h->mb_num == h->mb_width * h->mb_height);
+    if (sl->first_mb_addr << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
+        sl->first_mb_addr >= h->mb_num) {
+        av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
+        return AVERROR_INVALIDDATA;
+    }
+    sl->resync_mb_x = sl->mb_x =  sl->first_mb_addr % h->mb_width;
+    sl->resync_mb_y = sl->mb_y = (sl->first_mb_addr / h->mb_width) <<
+                                 FIELD_OR_MBAFF_PICTURE(h);
+    if (h->picture_structure == PICT_BOTTOM_FIELD)
+        sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
+    assert(sl->mb_y < h->mb_height);
+
+    if (!h->setup_finished)
+        ff_h264_init_poc(h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc,
+                         h->ps.sps, &h->poc, h->picture_structure, h->nal_ref_idc);
+
+    ret = ff_h264_build_ref_list(h, sl);
+    if (ret < 0)
+        return ret;
+
+    if (h->ps.pps->weighted_bipred_idc == 2 &&
+        sl->slice_type_nos == AV_PICTURE_TYPE_B) {
+        implicit_weight_table(h, sl, -1);
+        if (FRAME_MBAFF(h)) {
+            implicit_weight_table(h, sl, 0);
+            implicit_weight_table(h, sl, 1);
+        }
+    }
+
+    if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
+        ff_h264_direct_dist_scale_factor(h, sl);
+    ff_h264_direct_ref_list_init(h, sl);
+
     if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
         (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
          sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
@@ -1571,31 +1471,21 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
          h->nal_ref_idc == 0))
         sl->deblocking_filter = 0;
 
-    if (sl->deblocking_filter == 1 && h->max_contexts > 1) {
+    if (sl->deblocking_filter == 1 && h->nb_slice_ctx > 1) {
         if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) {
             /* Cheat slightly for speed:
              * Do not bother to deblock across slices. */
             sl->deblocking_filter = 2;
         } else {
-            h->max_contexts = 1;
-            if (!h->single_decode_warning) {
-                av_log(h->avctx, AV_LOG_INFO,
-                       "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
-                h->single_decode_warning = 1;
-            }
-            if (sl != h->slice_ctx) {
-                av_log(h->avctx, AV_LOG_ERROR,
-                       "Deblocking switched inside frame.\n");
-                return 1;
-            }
+            h->postpone_filter = 1;
         }
     }
     sl->qp_thresh = 15 -
                    FFMIN(sl->slice_alpha_c0_offset, sl->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);
+                          h->ps.pps->chroma_qp_index_offset[0],
+                          h->ps.pps->chroma_qp_index_offset[1]) +
+                   6 * (h->ps.sps->bit_depth_luma - 8);
 
     sl->slice_num       = ++h->current_slice;
     if (sl->slice_num >= MAX_SLICES) {
@@ -1605,7 +1495,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
 
     for (j = 0; j < 2; j++) {
         int id_list[16];
-        int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
+        int *ref2frm = h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
         for (i = 0; i < 16; i++) {
             id_list[i] = 60;
             if (j < sl->list_count && i < sl->ref_count[j] &&
@@ -1638,14 +1528,14 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
 
     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",
+               "slice:%d %s mb:%d %c%s%s frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
                sl->slice_num,
                (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
-               first_mb_in_slice,
+               sl->mb_y * h->mb_width + sl->mb_x,
                av_get_picture_type_char(sl->slice_type),
                sl->slice_type_fixed ? " fix" : "",
                h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
-               pps_id, h->frame_num,
+               h->poc.frame_num,
                h->cur_pic_ptr->field_poc[0],
                h->cur_pic_ptr->field_poc[1],
                sl->ref_count[0], sl->ref_count[1],
@@ -1693,7 +1583,7 @@ static av_always_inline void fill_filter_caches_inter(const H264Context *h,
         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] = sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
+            int (*ref2frm)[64] = h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 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]];
@@ -1708,7 +1598,7 @@ static av_always_inline void fill_filter_caches_inter(const H264Context *h,
             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] = sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
+                int (*ref2frm)[64] = h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 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]);
@@ -1741,7 +1631,7 @@ static av_always_inline void fill_filter_caches_inter(const H264Context *h,
 
     {
         int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
-        int (*ref2frm)[64] = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
+        int (*ref2frm)[64] = h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 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);
@@ -1772,9 +1662,6 @@ static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb
 
     top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
 
-    /* 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]);
@@ -1865,7 +1752,7 @@ static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb
 
     /* 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 (!CABAC(h) && h->ps.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;
@@ -1916,14 +1803,15 @@ static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x,
     const int pixel_shift    = h->pixel_shift;
     const int block_h        = 16 >> h->chroma_y_shift;
 
+    if (h->postpone_filter)
+        return;
+
     if (sl->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         = sl->mb_xy = mb_x + mb_y * h->mb_stride;
-                sl->slice_num = h->slice_table[mb_xy];
                 mb_type       = h->cur_pic.mb_type[mb_xy];
-                sl->list_count = h->list_counts[mb_xy];
 
                 if (FRAME_MBAFF(h))
                     sl->mb_mbaff               =
@@ -2040,6 +1928,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
     H264SliceContext *sl = arg;
     const H264Context *h = sl->h264;
     int lf_x_start = sl->mb_x;
+    int orig_deblock = sl->deblocking_filter;
     int ret;
 
     sl->linesize   = h->cur_pic_ptr->f->linesize[0];
@@ -2051,11 +1940,14 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
 
     sl->mb_skip_run = -1;
 
+    if (h->postpone_filter)
+        sl->deblocking_filter = 0;
+
     sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
                      avctx->codec_id != AV_CODEC_ID_H264 ||
                      (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
 
-    if (h->pps.cabac) {
+    if (h->ps.pps->cabac) {
         /* realign */
         align_get_bits(&sl->gb);
 
@@ -2100,7 +1992,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
                              sl->mb_y, ER_MB_END);
                 if (sl->mb_x >= lf_x_start)
                     loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
-                return 0;
+                goto finish;
             }
             if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
                 av_log(h->avctx, AV_LOG_ERROR,
@@ -2131,7 +2023,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
                              sl->mb_y, ER_MB_END);
                 if (sl->mb_x > lf_x_start)
                     loop_filter(h, sl, lf_x_start, sl->mb_x);
-                return 0;
+                goto finish;
             }
         }
     } else {
@@ -2185,7 +2077,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
                         er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
                                      sl->mb_x - 1, sl->mb_y, ER_MB_END);
 
-                        return 0;
+                        goto finish;
                     } else {
                         er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
                                      sl->mb_x - 1, sl->mb_y, ER_MB_END);
@@ -2205,7 +2097,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
                     if (sl->mb_x > lf_x_start)
                         loop_filter(h, sl, lf_x_start, sl->mb_x);
 
-                    return 0;
+                    goto finish;
                 } else {
                     er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
                                  sl->mb_y, ER_MB_ERROR);
@@ -2215,6 +2107,10 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
             }
         }
     }
+
+finish:
+    sl->deblocking_filter = orig_deblock;
+    return 0;
 }
 
 /**
@@ -2235,6 +2131,7 @@ int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
         int ret;
 
         h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
+        h->postpone_filter = 0;
 
         ret = decode_slice(avctx, &h->slice_ctx[0]);
         h->mb_y = h->slice_ctx[0].mb_y;
@@ -2268,6 +2165,24 @@ int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
         h->mb_y              = sl->mb_y;
         for (i = 1; i < context_count; i++)
             h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;
+
+        if (h->postpone_filter) {
+            h->postpone_filter = 0;
+
+            for (i = 0; i < context_count; i++) {
+                int y_end, x_end;
+
+                sl = &h->slice_ctx[i];
+                y_end = FFMIN(sl->mb_y + 1, h->mb_height);
+                x_end = (sl->mb_y >= h->mb_height) ? h->mb_width : sl->mb_x;
+
+                for (j = sl->resync_mb_y; j < y_end; j += 1 + FIELD_OR_MBAFF_PICTURE(h)) {
+                    sl->mb_y = j;
+                    loop_filter(h, sl, j > sl->resync_mb_y ? 0 : sl->resync_mb_x,
+                                j == y_end - 1 ? x_end : h->mb_width);
+                }
+            }
+        }
     }
 
     return 0;