]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/h264_slice.c
Merge commit '51d8725a6e4c7f989d6f0f38b59cb54a2d72846c'
[ffmpeg] / libavcodec / h264_slice.c
index 11ed1f1554d59f4b732e3212e1e3d47f2416661d..eef28aedc00b921354d6546f497b5aa86d8981d9 100644 (file)
@@ -161,21 +161,32 @@ static void release_unused_pictures(H264Context *h, int remove_current)
     }
 }
 
-static int alloc_scratch_buffers(H264Context *h, int linesize)
+static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
 {
+    const H264Context *h = sl->h264;
     int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
 
-    if (h->bipred_scratchpad)
-        return 0;
-
-    h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size);
+    av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 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);
+    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],
+                   h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
+    av_fast_malloc(&sl->top_borders[1], &sl->top_borders_allocated[1],
+                   h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
+
+    if (!sl->bipred_scratchpad || !sl->edge_emu_buffer ||
+        !sl->top_borders[0]    || !sl->top_borders[1]) {
+        av_freep(&sl->bipred_scratchpad);
+        av_freep(&sl->edge_emu_buffer);
+        av_freep(&sl->top_borders[0]);
+        av_freep(&sl->top_borders[1]);
+
+        sl->bipred_scratchpad_allocated = 0;
+        sl->edge_emu_buffer_allocated   = 0;
+        sl->top_borders_allocated[0]    = 0;
+        sl->top_borders_allocated[1]    = 0;
         return AVERROR(ENOMEM);
     }
 
@@ -402,8 +413,6 @@ static void clone_tables(H264Context *dst, H264SliceContext *sl,
     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);
 }
@@ -481,8 +490,6 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
          * 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;
@@ -531,20 +538,18 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
         for (i = 0; i < MAX_PPS_COUNT; i++)
             av_freep(h->pps_buffers + i);
 
-        av_freep(&h->rbsp_buffer[0]);
-        av_freep(&h->rbsp_buffer[1]);
         ff_h264_unref_picture(h, &h->last_pic_for_ec);
         memcpy(h, h1, sizeof(H264Context));
 
         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->cur_pic, 0, sizeof(h->cur_pic));
         memset(&h->last_pic_for_ec, 0, sizeof(h->last_pic_for_ec));
 
         h->slice_ctx = orig_slice_ctx;
 
+        memset(&h->slice_ctx[0].er,         0, sizeof(h->slice_ctx[0].er));
         memset(&h->slice_ctx[0].mb,         0, sizeof(h->slice_ctx[0].mb));
         memset(&h->slice_ctx[0].mb_luma_dc, 0, sizeof(h->slice_ctx[0].mb_luma_dc));
         memset(&h->slice_ctx[0].mb_padding, 0, sizeof(h->slice_ctx[0].mb_padding));
@@ -566,10 +571,6 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
         h->list_counts       = NULL;
         h->mb2b_xy           = NULL;
         h->mb2br_xy          = 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;
@@ -583,16 +584,13 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
             av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
             return ret;
         }
-        ret = ff_h264_context_init(h);
+        ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
         if (ret < 0) {
             av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
             return ret;
         }
         }
 
-        h->bipred_scratchpad = NULL;
-        h->edge_emu_buffer   = NULL;
-
         h->thread_context[0] = h;
         h->context_initialized = h1->context_initialized;
     }
@@ -656,10 +654,10 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
     h->dequant_coeff_pps = h1->dequant_coeff_pps;
 
     // POC timing
-    copy_fields(h, h1, poc_lsb, redundant_pic_count);
+    copy_fields(h, h1, poc_lsb, default_ref_list);
 
     // reference lists
-    copy_fields(h, h1, short_ref, cabac_init_idc);
+    copy_fields(h, h1, short_ref, thread_context);
 
     copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
     copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
@@ -739,16 +737,16 @@ static int h264_frame_start(H264Context *h)
     h->cur_pic_ptr = pic;
     ff_h264_unref_picture(h, &h->cur_pic);
     if (CONFIG_ERROR_RESILIENCE) {
-        ff_h264_set_erpic(&h->er.cur_pic, NULL);
+        ff_h264_set_erpic(&h->slice_ctx[0].er.cur_pic, NULL);
     }
 
     if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
         return ret;
 
     if (CONFIG_ERROR_RESILIENCE) {
-        ff_er_frame_start(&h->er);
-        ff_h264_set_erpic(&h->er.last_pic, NULL);
-        ff_h264_set_erpic(&h->er.next_pic, NULL);
+        ff_er_frame_start(&h->slice_ctx[0].er);
+        ff_h264_set_erpic(&h->slice_ctx[0].er.last_pic, NULL);
+        ff_h264_set_erpic(&h->slice_ctx[0].er.next_pic, NULL);
     }
 
     assert(h->linesize && h->uvlinesize);
@@ -779,7 +777,8 @@ static int h264_frame_start(H264Context *h)
     return 0;
 }
 
-static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
+static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
+                                              uint8_t *src_y,
                                               uint8_t *src_cb, uint8_t *src_cr,
                                               int linesize, int uvlinesize,
                                               int simple)
@@ -795,9 +794,9 @@ static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
     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];
+        if (sl->mb_y & 1) {
+            if (!MB_MBAFF(sl)) {
+                top_border = sl->top_borders[0][sl->mb_x];
                 AV_COPY128(top_border, src_y + 15 * linesize);
                 if (pixel_shift)
                     AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
@@ -831,13 +830,13 @@ static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
                     }
                 }
             }
-        } else if (MB_MBAFF(h)) {
+        } else if (MB_MBAFF(sl)) {
             top_idx = 0;
         } else
             return;
     }
 
-    top_border = h->top_borders[top_idx][h->mb_x];
+    top_border = sl->top_borders[top_idx][sl->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);
@@ -880,7 +879,7 @@ static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
  * @param field  0/1 initialize the weight for interlaced MBAFF
  *                -1 initializes the rest
  */
-static void implicit_weight_table(H264Context *h, H264SliceContext *sl, int field)
+static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
 {
     int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
 
@@ -978,34 +977,6 @@ static void init_scan_tables(H264Context *h)
     }
 }
 
-/**
- * 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;
-}
-
 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
 {
 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
@@ -1199,7 +1170,7 @@ static int h264_slice_header_init(H264Context *h, int reinit)
     h->slice_context_count = nb_slices;
 
     if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
-        ret = ff_h264_context_init(h);
+        ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
         if (ret < 0) {
             av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
             goto fail;
@@ -1245,7 +1216,7 @@ static int h264_slice_header_init(H264Context *h, int reinit)
         }
 
         for (i = 0; i < h->slice_context_count; i++)
-            if ((ret = ff_h264_context_init(h->thread_context[i])) < 0) {
+            if ((ret = ff_h264_slice_context_init(h, &h->slice_ctx[i])) < 0) {
                 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
                 goto fail;
             }
@@ -1276,12 +1247,10 @@ static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
  * 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, H264SliceContext *sl, H264Context *h0)
+int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
 {
     unsigned int first_mb_in_slice;
     unsigned int pps_id;
@@ -1291,22 +1260,22 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
     int must_reinit;
     int needs_reinit = 0;
     int field_pic_flag, bottom_field_flag;
-    int first_slice = h == h0 && !h0->current_slice;
+    int first_slice = sl == h->slice_ctx && !h->current_slice;
     int frame_num, picture_structure, droppable;
     PPS *pps;
 
     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);
+    first_mb_in_slice = get_ue_golomb_long(&sl->gb);
 
     if (first_mb_in_slice == 0) { // FIXME better field boundary detection
-        if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
+        if (h->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
             ff_h264_field_end(h, sl, 1);
         }
 
-        h0->current_slice = 0;
-        if (!h0->first_field) {
+        h->current_slice = 0;
+        if (!h->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);
@@ -1315,11 +1284,11 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
         }
     }
 
-    slice_type = get_ue_golomb_31(&h->gb);
+    slice_type = get_ue_golomb_31(&sl->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);
+               "slice type %d too large at %d\n",
+               slice_type, first_mb_in_slice);
         return AVERROR_INVALIDDATA;
     }
     if (slice_type > 4) {
@@ -1329,6 +1298,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
         sl->slice_type_fixed = 0;
 
     slice_type = golomb_to_pict_type[slice_type];
+
     sl->slice_type     = slice_type;
     sl->slice_type_nos = slice_type & 3;
 
@@ -1350,38 +1320,38 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
     // to make a few old functions happy, it's wrong though
     h->pict_type = sl->slice_type;
 
-    pps_id = get_ue_golomb(&h->gb);
+    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 (!h0->pps_buffers[pps_id]) {
+    if (!h->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) {
+    if (h->au_pps_id >= 0 && pps_id != h->au_pps_id) {
         av_log(h->avctx, AV_LOG_ERROR,
                "PPS change from %d to %d forbidden\n",
-               h0->au_pps_id, pps_id);
+               h->au_pps_id, pps_id);
         return AVERROR_INVALIDDATA;
     }
 
-    pps = h0->pps_buffers[pps_id];
+    pps = h->pps_buffers[pps_id];
 
-    if (!h0->sps_buffers[pps->sps_id]) {
+    if (!h->sps_buffers[pps->sps_id]) {
         av_log(h->avctx, AV_LOG_ERROR,
                "non-existing SPS %u referenced\n",
                h->pps.sps_id);
         return AVERROR_INVALIDDATA;
     }
     if (first_slice)
-        h->pps = *h0->pps_buffers[pps_id];
+        h->pps = *h->pps_buffers[pps_id];
 
     if (pps->sps_id != h->sps.sps_id ||
         pps->sps_id != h->current_sps_id ||
-        h0->sps_buffers[pps->sps_id]->new) {
+        h->sps_buffers[pps->sps_id]->new) {
 
         if (!first_slice) {
             av_log(h->avctx, AV_LOG_ERROR,
@@ -1389,7 +1359,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
             return AVERROR_INVALIDDATA;
         }
 
-        h->sps = *h0->sps_buffers[h->pps.sps_id];
+        h->sps = *h->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) ||
@@ -1420,7 +1390,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
                      || 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)))
+    if (non_j_pixfmt(h->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h, 0)))
         must_reinit = 1;
 
     if (first_slice && av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio))
@@ -1456,13 +1426,13 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
 
     if (h->context_initialized &&
         (must_reinit || needs_reinit)) {
-        if (h != h0) {
+        if (sl != h->slice_ctx) {
             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);
+                   h->current_slice + 1);
             return AVERROR_INVALIDDATA;
         }
 
@@ -1484,7 +1454,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
         }
     }
     if (!h->context_initialized) {
-        if (h != h0) {
+        if (sl != h->slice_ctx) {
             av_log(h->avctx, AV_LOG_ERROR,
                    "Cannot (re-)initialize context during parallel decoding.\n");
             return AVERROR_PATCHWELCOME;
@@ -1506,19 +1476,19 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
         ff_h264_init_dequant_tables(h);
     }
 
-    frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
+    frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
     if (!first_slice) {
-        if (h0->frame_num != frame_num) {
+        if (h->frame_num != frame_num) {
             av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
-                   h0->frame_num, frame_num);
+                   h->frame_num, frame_num);
             return AVERROR_INVALIDDATA;
         }
     }
 
-    h->mb_mbaff        = 0;
+    sl->mb_mbaff       = 0;
     h->mb_aff_frame    = 0;
-    last_pic_structure = h0->picture_structure;
-    last_pic_droppable = h0->droppable;
+    last_pic_structure = h->picture_structure;
+    last_pic_droppable = h->droppable;
     droppable          = h->nal_ref_idc == 0;
     if (h->sps.frame_mbs_only_flag) {
         picture_structure = PICT_FRAME;
@@ -1527,27 +1497,27 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
             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);
+        field_pic_flag = get_bits1(&sl->gb);
 
         if (field_pic_flag) {
-            bottom_field_flag = get_bits1(&h->gb);
+            bottom_field_flag = get_bits1(&sl->gb);
             picture_structure = PICT_TOP_FIELD + bottom_field_flag;
         } else {
             picture_structure = PICT_FRAME;
             h->mb_aff_frame      = h->sps.mb_aff;
         }
     }
-    if (h0->current_slice) {
+    if (h->current_slice) {
         if (last_pic_structure != picture_structure ||
             last_pic_droppable != 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);
             return AVERROR_INVALIDDATA;
-        } else if (!h0->cur_pic_ptr) {
+        } else if (!h->cur_pic_ptr) {
             av_log(h->avctx, AV_LOG_ERROR,
                    "unset cur_pic_ptr on slice %d\n",
-                   h0->current_slice + 1);
+                   h->current_slice + 1);
             return AVERROR_INVALIDDATA;
         }
     }
@@ -1555,9 +1525,9 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
     h->picture_structure = picture_structure;
     h->droppable         = droppable;
     h->frame_num         = frame_num;
-    h->mb_field_decoding_flag = picture_structure != PICT_FRAME;
+    sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
 
-    if (h0->current_slice == 0) {
+    if (h->current_slice == 0) {
         /* 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) {
@@ -1581,14 +1551,14 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
          * 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);
+        if (h->first_field) {
+            assert(h->cur_pic_ptr);
+            assert(h->cur_pic_ptr->f.buf[0]);
+            assert(h->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,
+            if (h->cur_pic_ptr->tf.owner == h->avctx) {
+                ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
                                           last_pic_structure == PICT_BOTTOM_FIELD);
             }
 
@@ -1597,17 +1567,17 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
                 /* 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,
+                    ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
                                               last_pic_structure == PICT_TOP_FIELD);
                 }
             } else {
-                if (h0->cur_pic_ptr->frame_num != h->frame_num) {
+                if (h->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,
+                        ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
                                                   last_pic_structure == PICT_TOP_FIELD);
                     }
                 } else {
@@ -1633,7 +1603,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
             }
         }
 
-        while (h->frame_num != h->prev_frame_num && !h0->first_field &&
+        while (h->frame_num != h->prev_frame_num && !h->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",
@@ -1643,7 +1613,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
                     h->last_pocs[i] = INT_MIN;
             ret = h264_frame_start(h);
             if (ret < 0) {
-                h0->first_field = 0;
+                h->first_field = 0;
                 return ret;
             }
 
@@ -1684,41 +1654,41 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
         /* 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);
+        if (h->first_field) {
+            assert(h->cur_pic_ptr);
+            assert(h->cur_pic_ptr->f.buf[0]);
+            assert(h->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->missing_fields ++;
-                h0->cur_pic_ptr = NULL;
-                h0->first_field = FIELD_PICTURE(h);
+                h->missing_fields ++;
+                h->cur_pic_ptr = NULL;
+                h->first_field = FIELD_PICTURE(h);
             } else {
-                h0->missing_fields = 0;
-                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);
+                h->missing_fields = 0;
+                if (h->cur_pic_ptr->frame_num != h->frame_num) {
+                    ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
+                                              h->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;
+                    h->first_field = 1;
+                    h->cur_pic_ptr = NULL;
                 } else {
                     /* Second field in complementary pair */
-                    h0->first_field = 0;
+                    h->first_field = 0;
                 }
             }
         } else {
             /* Frame or first field in a potentially complementary pair */
-            h0->first_field = FIELD_PICTURE(h);
+            h->first_field = FIELD_PICTURE(h);
         }
 
-        if (!FIELD_PICTURE(h) || h0->first_field) {
+        if (!FIELD_PICTURE(h) || h->first_field) {
             if (h264_frame_start(h) < 0) {
-                h0->first_field = 0;
+                h->first_field = 0;
                 return AVERROR_INVALIDDATA;
             }
         } else {
@@ -1733,19 +1703,9 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
             memset(h->slice_table, -1,
                 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
         }
-        h0->last_slice_type = -1;
+        h->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
 
@@ -1755,12 +1715,12 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
         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);
+    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)
-        h->resync_mb_y = h->mb_y = h->mb_y + 1;
-    av_assert1(h->mb_y < h->mb_height);
+        sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
+    av_assert1(sl->mb_y < h->mb_height);
 
     if (h->picture_structure == PICT_FRAME) {
         h->curr_pic_num = h->frame_num;
@@ -1771,35 +1731,35 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
     }
 
     if (h->nal_unit_type == NAL_IDR_SLICE)
-        get_ue_golomb(&h->gb); /* idr_pic_id */
+        get_ue_golomb(&sl->gb); /* idr_pic_id */
 
     if (h->sps.poc_type == 0) {
-        h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
+        h->poc_lsb = get_bits(&sl->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);
+            h->delta_poc_bottom = get_se_golomb(&sl->gb);
     }
 
     if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
-        h->delta_poc[0] = get_se_golomb(&h->gb);
+        h->delta_poc[0] = get_se_golomb(&sl->gb);
 
         if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
-            h->delta_poc[1] = get_se_golomb(&h->gb);
+            h->delta_poc[1] = get_se_golomb(&sl->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);
+        sl->redundant_pic_count = get_ue_golomb(&sl->gb);
 
     ret = ff_set_ref_count(h, sl);
     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, sl->ref_count, sizeof(sl->ref_count)))) {
+        (h->current_slice == 0 ||
+         slice_type != h->last_slice_type ||
+         memcmp(h->last_ref_count, sl->ref_count, sizeof(sl->ref_count)))) {
 
         ff_h264_fill_default_ref_list(h, sl);
     }
@@ -1833,9 +1793,9 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
     // 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,
+        ret = ff_h264_decode_ref_pic_marking(h, &sl->gb,
                                              !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
-                                             h0->current_slice == 0);
+                                             h->current_slice == 0);
         if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
             return AVERROR_INVALIDDATA;
     }
@@ -1854,16 +1814,16 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
     ff_h264_direct_ref_list_init(h, sl);
 
     if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
-        tmp = get_ue_golomb_31(&h->gb);
+        tmp = get_ue_golomb_31(&sl->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;
+        sl->cabac_init_idc = tmp;
     }
 
     sl->last_qscale_diff = 0;
-    tmp = h->pps.init_qp + get_se_golomb(&h->gb);
+    tmp = h->pps.init_qp + get_se_golomb(&sl->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;
@@ -1873,35 +1833,35 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
     sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
     // FIXME qscale / qp ... stuff
     if (sl->slice_type == AV_PICTURE_TYPE_SP)
-        get_bits1(&h->gb); /* sp_for_switch_flag */
+        get_bits1(&sl->gb); /* sp_for_switch_flag */
     if (sl->slice_type == AV_PICTURE_TYPE_SP ||
         sl->slice_type == AV_PICTURE_TYPE_SI)
-        get_se_golomb(&h->gb); /* slice_qs_delta */
+        get_se_golomb(&sl->gb); /* slice_qs_delta */
 
-    h->deblocking_filter     = 1;
-    h->slice_alpha_c0_offset = 0;
-    h->slice_beta_offset     = 0;
+    sl->deblocking_filter     = 1;
+    sl->slice_alpha_c0_offset = 0;
+    sl->slice_beta_offset     = 0;
     if (h->pps.deblocking_filter_parameters_present) {
-        tmp = get_ue_golomb_31(&h->gb);
+        tmp = get_ue_golomb_31(&sl->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) {
+        sl->deblocking_filter = tmp;
+        if (sl->deblocking_filter < 2)
+            sl->deblocking_filter ^= 1;  // 1<->0
+
+        if (sl->deblocking_filter) {
+            sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
+            sl->slice_beta_offset     = get_se_golomb(&sl->gb) * 2;
+            if (sl->slice_alpha_c0_offset >  12 ||
+                sl->slice_alpha_c0_offset < -12 ||
+                sl->slice_beta_offset >  12     ||
+                sl->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);
+                       sl->slice_alpha_c0_offset, sl->slice_beta_offset);
                 return AVERROR_INVALIDDATA;
             }
         }
@@ -1916,24 +1876,24 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
          sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
         (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
          h->nal_ref_idc == 0))
-        h->deblocking_filter = 0;
+        sl->deblocking_filter = 0;
 
-    if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
+    if (sl->deblocking_filter == 1 && h->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;
+            sl->deblocking_filter = 2;
         } else {
-            h0->max_contexts = 1;
-            if (!h0->single_decode_warning) {
+            h->max_contexts = 1;
+            if (!h->single_decode_warning) {
                 av_log(h->avctx, AV_LOG_INFO,
                        "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
                        "To parallelize slice decoding you need video encoded with disable_deblocking_filter_idc set to 2 (deblock only edges that do not cross slices).\n"
                        "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
                        "but will generate non-standard-compliant output.\n");
-                h0->single_decode_warning = 1;
+                h->single_decode_warning = 1;
             }
-            if (h != h0) {
+            if (sl != h->slice_ctx) {
                 av_log(h->avctx, AV_LOG_ERROR,
                        "Deblocking switched inside frame.\n");
                 return SLICE_SINGLETHREAD;
@@ -1941,20 +1901,20 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
         }
     }
     sl->qp_thresh = 15 -
-                   FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
+                   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);
 
-    h0->last_slice_type = slice_type;
-    memcpy(h0->last_ref_count, sl->ref_count, sizeof(h0->last_ref_count));
-    sl->slice_num       = ++h0->current_slice;
+    h->last_slice_type = slice_type;
+    memcpy(h->last_ref_count, sl->ref_count, sizeof(h->last_ref_count));
+    sl->slice_num       = ++h->current_slice;
 
     if (sl->slice_num)
-        h0->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= h->resync_mb_y;
-    if (   h0->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= h->resync_mb_y
-        && h0->slice_row[sl->slice_num&(MAX_SLICES-1)] <= h->resync_mb_y
+        h->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
+    if (   h->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
+        && h->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
         && sl->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", sl->slice_num, MAX_SLICES);
@@ -1993,9 +1953,9 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
                              (sl->ref_list[j][i].reference & 3);
     }
 
-    h0->au_pps_id = pps_id;
+    h->au_pps_id = pps_id;
     h->sps.new =
-    h0->sps_buffers[h->pps.sps_id]->new = 0;
+    h->sps_buffers[h->pps.sps_id]->new = 0;
     h->current_sps_id = h->pps.sps_id;
 
     if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
@@ -2012,8 +1972,8 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
                h->cur_pic_ptr->field_poc[1],
                sl->ref_count[0], sl->ref_count[1],
                sl->qscale,
-               h->deblocking_filter,
-               h->slice_alpha_c0_offset, h->slice_beta_offset,
+               sl->deblocking_filter,
+               sl->slice_alpha_c0_offset, sl->slice_beta_offset,
                sl->use_weight,
                sl->use_weight == 1 && sl->use_weight_chroma ? "c" : "",
                sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
@@ -2040,7 +2000,7 @@ int ff_h264_get_slice_type(const H264SliceContext *sl)
     }
 }
 
-static av_always_inline void fill_filter_caches_inter(H264Context *h,
+static av_always_inline void fill_filter_caches_inter(const H264Context *h,
                                                       H264SliceContext *sl,
                                                       int mb_type, int top_xy,
                                                       int left_xy[LEFT_MBS],
@@ -2055,7 +2015,7 @@ static av_always_inline void fill_filter_caches_inter(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] = (void*)(sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
+            int (*ref2frm)[64] = (void*)(sl->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]];
@@ -2070,7 +2030,7 @@ static av_always_inline void fill_filter_caches_inter(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] =(void*)( sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
+                int (*ref2frm)[64] =(void*)( sl->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]);
@@ -2103,7 +2063,7 @@ static av_always_inline void fill_filter_caches_inter(H264Context *h,
 
     {
         int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
-        int (*ref2frm)[64] = (void*)(sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
+        int (*ref2frm)[64] = (void*)(sl->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);
@@ -2113,7 +2073,7 @@ static av_always_inline void fill_filter_caches_inter(H264Context *h,
     }
 
     {
-        int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
+        int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->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);
@@ -2125,15 +2085,15 @@ static av_always_inline void fill_filter_caches_inter(H264Context *h,
  *
  * @return non zero if the loop filter can be skipped
  */
-static int fill_filter_caches(H264Context *h, H264SliceContext *sl, int mb_type)
+static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
 {
-    const int mb_xy = h->mb_xy;
+    const int mb_xy = sl->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));
+    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. */
@@ -2142,7 +2102,7 @@ static int fill_filter_caches(H264Context *h, H264SliceContext *sl, int mb_type)
     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 (sl->mb_y & 1) {
             if (left_mb_field_flag != curr_mb_field_flag)
                 left_xy[LTOP] -= h->mb_stride;
         } else {
@@ -2181,7 +2141,7 @@ static int fill_filter_caches(H264Context *h, H264SliceContext *sl, int mb_type)
     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 (sl->deblocking_filter == 2) {
         if (h->slice_table[top_xy] != sl->slice_num)
             top_type = 0;
         if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
@@ -2270,30 +2230,30 @@ static int fill_filter_caches(H264Context *h, H264SliceContext *sl, int mb_type)
     return 0;
 }
 
-static void loop_filter(H264Context *h, H264SliceContext *sl, int start_x, int end_x)
+static void loop_filter(const H264Context *h, H264SliceContext *sl, 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 end_mb_y       = sl->mb_y + FRAME_MBAFF(h);
     const int old_slice_type = sl->slice_type;
     const int pixel_shift    = h->pixel_shift;
     const int block_h        = 16 >> h->chroma_y_shift;
 
-    if (h->deblocking_filter) {
+    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         = h->mb_xy = mb_x + mb_y * h->mb_stride;
+                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))
-                    h->mb_mbaff               =
-                    h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
+                    sl->mb_mbaff               =
+                    sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
 
-                h->mb_x = mb_x;
-                h->mb_y = mb_y;
+                sl->mb_x = mb_x;
+                sl->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] +
@@ -2304,7 +2264,7 @@ static void loop_filter(H264Context *h, H264SliceContext *sl, int start_x, int e
                           mb_y * h->uvlinesize * block_h;
                 // FIXME simplify above
 
-                if (MB_FIELD(h)) {
+                if (MB_FIELD(sl)) {
                     linesize   = sl->mb_linesize   = h->linesize   * 2;
                     uvlinesize = sl->mb_uvlinesize = h->uvlinesize * 2;
                     if (mb_y & 1) { // FIXME move out of this function?
@@ -2316,7 +2276,7 @@ static void loop_filter(H264Context *h, H264SliceContext *sl, int start_x, int e
                     linesize   = sl->mb_linesize   = h->linesize;
                     uvlinesize = sl->mb_uvlinesize = h->uvlinesize;
                 }
-                backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
+                backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
                                  uvlinesize, 0);
                 if (fill_filter_caches(h, sl, mb_type))
                     continue;
@@ -2333,33 +2293,33 @@ static void loop_filter(H264Context *h, H264SliceContext *sl, int start_x, int e
             }
     }
     sl->slice_type  = old_slice_type;
-    h->mb_x         = end_x;
-    h->mb_y         = end_mb_y - FRAME_MBAFF(h);
+    sl->mb_x         = end_x;
+    sl->mb_y         = end_mb_y - FRAME_MBAFF(h);
     sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
     sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
 }
 
-static void predict_field_decoding_flag(H264Context *h, H264SliceContext *sl)
+static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
 {
-    const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
+    const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
     int mb_type     = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
                       h->cur_pic.mb_type[mb_xy - 1] :
                       (h->slice_table[mb_xy - h->mb_stride] == sl->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;
+    sl->mb_mbaff    = sl->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, H264SliceContext *sl)
+static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
 {
-    int top            = 16 * (h->mb_y      >> FIELD_PICTURE(h));
+    int top            = 16 * (sl->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 (sl->deblocking_filter) {
         if ((top + height) >= pic_height)
             height += deblock_border;
         top -= deblock_border;
@@ -2376,19 +2336,19 @@ static void decode_finish_row(H264Context *h, H264SliceContext *sl)
 
     ff_h264_draw_horiz_band(h, sl, top, height);
 
-    if (h->droppable || h->er.error_occurred)
+    if (h->droppable || sl->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, H264SliceContext *sl,
+static void er_add_slice(H264SliceContext *sl,
                          int startx, int starty,
                          int endx, int endy, int status)
 {
     if (CONFIG_ERROR_RESILIENCE) {
-        ERContext *er = &h->er;
+        ERContext *er = &sl->er;
 
         ff_er_add_slice(er, startx, starty, endx, endy, status);
     }
@@ -2397,35 +2357,40 @@ static void er_add_slice(H264Context *h, H264SliceContext *sl,
 static int decode_slice(struct AVCodecContext *avctx, void *arg)
 {
     H264SliceContext *sl = arg;
-    H264Context       *h = sl->h264;
-    int lf_x_start = h->mb_x;
+    const H264Context *h = avctx->priv_data;
+    int lf_x_start = sl->mb_x;
+    int ret;
+
+    ret = alloc_scratch_buffers(sl, h->linesize);
+    if (ret < 0)
+        return ret;
 
-    h->mb_skip_run = -1;
+    sl->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));
+    sl->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 (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && sl->er.error_status_table) {
+        const int start_i  = av_clip(sl->resync_mb_x + sl->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]];
+            int prev_status = sl->er.error_status_table[sl->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;
+                sl->er.error_occurred = 1;
         }
     }
 
     if (h->pps.cabac) {
         /* realign */
-        align_get_bits(&h->gb);
+        align_get_bits(&sl->gb);
 
         /* init cabac */
         ff_init_cabac_decoder(&sl->cabac,
-                              h->gb.buffer + get_bits_count(&h->gb) / 8,
-                              (get_bits_left(&h->gb) + 7) / 8);
+                              sl->gb.buffer + get_bits_count(&sl->gb) / 8,
+                              (get_bits_left(&sl->gb) + 7) / 8);
 
         ff_h264_init_cabac_states(h, sl);
 
@@ -2440,22 +2405,22 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
 
             // FIXME optimal? or let mb_decode decode 16x32 ?
             if (ret >= 0 && FRAME_MBAFF(h)) {
-                h->mb_y++;
+                sl->mb_y++;
 
                 ret = ff_h264_decode_mb_cabac(h, sl);
 
                 if (ret >= 0)
                     ff_h264_hl_decode_mb(h, sl);
-                h->mb_y--;
+                sl->mb_y--;
             }
             eos = get_cabac_terminate(&sl->cabac);
 
             if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
                 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
-                er_add_slice(h, sl, 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, sl, lf_x_start, h->mb_x + 1);
+                er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
+                             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;
             }
             if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
@@ -2463,32 +2428,32 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
             if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
                 av_log(h->avctx, AV_LOG_ERROR,
                        "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
-                       h->mb_x, h->mb_y,
+                       sl->mb_x, sl->mb_y,
                        sl->cabac.bytestream_end - sl->cabac.bytestream);
-                er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, h->mb_x,
-                             h->mb_y, ER_MB_ERROR);
+                er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
+                             sl->mb_y, ER_MB_ERROR);
                 return AVERROR_INVALIDDATA;
             }
 
-            if (++h->mb_x >= h->mb_width) {
-                loop_filter(h, sl, lf_x_start, h->mb_x);
-                h->mb_x = lf_x_start = 0;
+            if (++sl->mb_x >= h->mb_width) {
+                loop_filter(h, sl, lf_x_start, sl->mb_x);
+                sl->mb_x = lf_x_start = 0;
                 decode_finish_row(h, sl);
-                ++h->mb_y;
+                ++sl->mb_y;
                 if (FIELD_OR_MBAFF_PICTURE(h)) {
-                    ++h->mb_y;
-                    if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
+                    ++sl->mb_y;
+                    if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
                         predict_field_decoding_flag(h, sl);
                 }
             }
 
-            if (eos || h->mb_y >= h->mb_height) {
+            if (eos || sl->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, sl, 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, sl, lf_x_start, h->mb_x);
+                        get_bits_count(&sl->gb), sl->gb.size_in_bits);
+                er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
+                             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;
             }
         }
@@ -2501,65 +2466,65 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
 
             // FIXME optimal? or let mb_decode decode 16x32 ?
             if (ret >= 0 && FRAME_MBAFF(h)) {
-                h->mb_y++;
+                sl->mb_y++;
                 ret = ff_h264_decode_mb_cavlc(h, sl);
 
                 if (ret >= 0)
                     ff_h264_hl_decode_mb(h, sl);
-                h->mb_y--;
+                sl->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, sl, h->resync_mb_x, h->resync_mb_y, h->mb_x,
-                             h->mb_y, ER_MB_ERROR);
+                       "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
+                er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
+                             sl->mb_y, ER_MB_ERROR);
                 return ret;
             }
 
-            if (++h->mb_x >= h->mb_width) {
-                loop_filter(h, sl, lf_x_start, h->mb_x);
-                h->mb_x = lf_x_start = 0;
+            if (++sl->mb_x >= h->mb_width) {
+                loop_filter(h, sl, lf_x_start, sl->mb_x);
+                sl->mb_x = lf_x_start = 0;
                 decode_finish_row(h, sl);
-                ++h->mb_y;
+                ++sl->mb_y;
                 if (FIELD_OR_MBAFF_PICTURE(h)) {
-                    ++h->mb_y;
-                    if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
+                    ++sl->mb_y;
+                    if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
                         predict_field_decoding_flag(h, sl);
                 }
-                if (h->mb_y >= h->mb_height) {
+                if (sl->mb_y >= h->mb_height) {
                     tprintf(h->avctx, "slice end %d %d\n",
-                            get_bits_count(&h->gb), h->gb.size_in_bits);
+                            get_bits_count(&sl->gb), sl->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, sl, h->resync_mb_x, h->resync_mb_y,
-                                     h->mb_x - 1, h->mb_y, ER_MB_END);
+                    if (   get_bits_left(&sl->gb) == 0
+                        || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
+                        er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
+                                     sl->mb_x - 1, sl->mb_y, ER_MB_END);
 
                         return 0;
                     } else {
-                        er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y,
-                                     h->mb_x, h->mb_y, ER_MB_END);
+                        er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
+                                     sl->mb_x, sl->mb_y, ER_MB_END);
 
                         return AVERROR_INVALIDDATA;
                     }
                 }
             }
 
-            if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
+            if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
                 tprintf(h->avctx, "slice end %d %d\n",
-                        get_bits_count(&h->gb), h->gb.size_in_bits);
+                        get_bits_count(&sl->gb), sl->gb.size_in_bits);
 
-                if (get_bits_left(&h->gb) == 0) {
-                    er_add_slice(h, sl, 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, sl, lf_x_start, h->mb_x);
+                if (get_bits_left(&sl->gb) == 0) {
+                    er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
+                                 sl->mb_x - 1, 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;
                 } else {
-                    er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, h->mb_x,
-                                 h->mb_y, ER_MB_ERROR);
+                    er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
+                                 sl->mb_y, ER_MB_ERROR);
 
                     return AVERROR_INVALIDDATA;
                 }
@@ -2577,38 +2542,36 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
 {
     AVCodecContext *const avctx = h->avctx;
-    H264Context *hx;
+    H264SliceContext *sl;
     int i;
 
-    av_assert0(h->mb_y < h->mb_height);
+    av_assert0(context_count && h->slice_ctx[context_count - 1].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->slice_ctx[0]);
+        int ret = decode_slice(avctx, &h->slice_ctx[0]);
+        h->mb_y = h->slice_ctx[0].mb_y;
+        return ret;
     } else {
         av_assert0(context_count > 0);
         for (i = 1; i < context_count; i++) {
-            hx                 = h->thread_context[i];
+            sl                 = &h->slice_ctx[i];
             if (CONFIG_ERROR_RESILIENCE) {
-                hx->er.error_count = 0;
+                sl->er.error_count = 0;
             }
-            hx->x264_build     = h->x264_build;
         }
 
         avctx->execute(avctx, decode_slice, h->slice_ctx,
                        NULL, context_count, sizeof(h->slice_ctx[0]));
 
         /* 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;
+        sl                   = &h->slice_ctx[context_count - 1];
+        h->mb_y              = sl->mb_y;
         if (CONFIG_ERROR_RESILIENCE) {
             for (i = 1; i < context_count; i++)
-                h->er.error_count += h->thread_context[i]->er.error_count;
+                h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;
         }
     }