h264: Add missing ff_ prefix to internally visible h264_init_dequant_tables()
[libav.git] / libavcodec / h264_slice.c
index 7ad25b9..23824db 100644 (file)
 #include "rectangle.h"
 #include "thread.h"
 
-
-static const uint8_t rem6[QP_MAX_NUM + 1] = {
-    0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
-    3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
-    0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
-};
-
-static const uint8_t div6[QP_MAX_NUM + 1] = {
-    0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3,  3,  3,
-    3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6,  6,  6,
-    7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10, 10,
-};
-
 static const uint8_t field_scan[16] = {
     0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
     0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
@@ -121,57 +108,45 @@ static const uint8_t zigzag_scan8x8_cavlc[64] = {
     5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
 };
 
-static const uint8_t dequant4_coeff_init[6][3] = {
-    { 10, 13, 16 },
-    { 11, 14, 18 },
-    { 13, 16, 20 },
-    { 14, 18, 23 },
-    { 16, 20, 25 },
-    { 18, 23, 29 },
-};
-
-static const uint8_t dequant8_coeff_init_scan[16] = {
-    0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
-};
-
-static const uint8_t dequant8_coeff_init[6][6] = {
-    { 20, 18, 32, 19, 25, 24 },
-    { 22, 19, 35, 21, 28, 26 },
-    { 26, 23, 42, 24, 33, 31 },
-    { 28, 25, 45, 26, 35, 33 },
-    { 32, 28, 51, 30, 40, 38 },
-    { 36, 32, 58, 34, 46, 43 },
-};
-
-
 static void release_unused_pictures(H264Context *h, int remove_current)
 {
     int i;
 
     /* release non reference frames */
     for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
-        if (h->DPB[i].f.buf[0] && !h->DPB[i].reference &&
+        if (h->DPB[i].f->buf[0] && !h->DPB[i].reference &&
             (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
             ff_h264_unref_picture(h, &h->DPB[i]);
         }
     }
 }
 
-static int alloc_scratch_buffers(H264Context *h, int linesize)
+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);
     }
 
@@ -209,17 +184,14 @@ static int alloc_picture(H264Context *h, H264Picture *pic)
 {
     int i, ret = 0;
 
-    av_assert0(!pic->f.data[0]);
+    av_assert0(!pic->f->data[0]);
 
-    pic->tf.f = &pic->f;
+    pic->tf.f = pic->f;
     ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
                                                    AV_GET_BUFFER_FLAG_REF : 0);
     if (ret < 0)
         goto fail;
 
-    h->linesize   = pic->f.linesize[0];
-    h->uvlinesize = pic->f.linesize[1];
-
     if (h->avctx->hwaccel) {
         const AVHWAccel *hwaccel = h->avctx->hwaccel;
         av_assert0(!pic->hwaccel_picture_private);
@@ -263,9 +235,7 @@ fail:
 
 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
 {
-    if (!pic->f.buf[0])
-        return 1;
-    if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
+    if (!pic->f->buf[0])
         return 1;
     return 0;
 }
@@ -281,14 +251,36 @@ static int find_unused_picture(H264Context *h)
     if (i == H264_MAX_PICTURE_COUNT)
         return AVERROR_INVALIDDATA;
 
-    if (h->DPB[i].needs_realloc) {
-        h->DPB[i].needs_realloc = 0;
-        ff_h264_unref_picture(h, &h->DPB[i]);
-    }
-
     return i;
 }
 
+static int initialize_cur_frame(H264Context *h)
+{
+    H264Picture *cur;
+    int ret;
+
+    release_unused_pictures(h, 1);
+    ff_h264_unref_picture(h, &h->cur_pic);
+    h->cur_pic_ptr = NULL;
+
+    ret = find_unused_picture(h);
+    if (ret < 0) {
+        av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
+        return ret;
+    }
+    cur = &h->DPB[ret];
+
+    ret = alloc_picture(h, cur);
+    if (ret < 0)
+        return ret;
+
+    ret = ff_h264_ref_picture(h, &h->cur_pic, cur);
+    if (ret < 0)
+        return ret;
+    h->cur_pic_ptr = cur;
+
+    return 0;
+}
 
 static void init_dequant8_coeff_table(H264Context *h)
 {
@@ -307,11 +299,11 @@ static void init_dequant8_coeff_table(H264Context *h)
             continue;
 
         for (q = 0; q < max_qp + 1; q++) {
-            int shift = div6[q];
-            int idx   = rem6[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)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 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;
         }
     }
@@ -333,17 +325,17 @@ static void init_dequant4_coeff_table(H264Context *h)
             continue;
 
         for (q = 0; q < max_qp + 1; q++) {
-            int shift = div6[q] + 2;
-            int idx   = rem6[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)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
+                    ((uint32_t)ff_h264_dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
                      h->pps.scaling_matrix4[i][x]) << shift;
         }
     }
 }
 
-void h264_init_dequant_tables(H264Context *h)
+void ff_h264_init_dequant_tables(H264Context *h)
 {
     int i, x;
     init_dequant4_coeff_table(h);
@@ -360,33 +352,6 @@ void h264_init_dequant_tables(H264Context *h)
     }
 }
 
-/**
- * Mimic alloc_tables(), but for every context thread.
- */
-static void clone_tables(H264Context *dst, H264SliceContext *sl,
-                         H264Context *src, int i)
-{
-    sl->intra4x4_pred_mode     = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
-    sl->mvd_table[0]           = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
-    sl->mvd_table[1]           = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
-
-    dst->non_zero_count         = src->non_zero_count;
-    dst->slice_table            = src->slice_table;
-    dst->cbp_table              = src->cbp_table;
-    dst->mb2b_xy                = src->mb2b_xy;
-    dst->mb2br_xy               = src->mb2br_xy;
-    dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
-    dst->direct_table           = src->direct_table;
-    dst->list_counts            = src->list_counts;
-    dst->DPB                    = src->DPB;
-    dst->cur_pic_ptr            = src->cur_pic_ptr;
-    dst->cur_pic                = src->cur_pic;
-    dst->bipred_scratchpad      = NULL;
-    dst->edge_emu_buffer        = NULL;
-    ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
-                      src->sps.chroma_format_idc);
-}
-
 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
 
 #define REBASE_PICTURE(pic, new_ctx, old_ctx)             \
@@ -433,14 +398,14 @@ static int copy_parameter_set(void **to, void **from, int count, int size)
     memcpy(&to->start_field, &from->start_field,                        \
            (char *)&to->end_field - (char *)&to->start_field)
 
-static int h264_slice_header_init(H264Context *h, int reinit);
+static int h264_slice_header_init(H264Context *h);
 
 int ff_h264_update_thread_context(AVCodecContext *dst,
                                   const AVCodecContext *src)
 {
     H264Context *h = dst->priv_data, *h1 = src->priv_data;
     int inited = h->context_initialized, err = 0;
-    int context_reinitialized = 0;
+    int need_reinit = 0;
     int i, ret;
 
     if (dst == src || !h1->context_initialized)
@@ -454,14 +419,22 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
          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)) {
+        need_reinit = 1;
+    }
 
-        /* set bits_per_raw_sample to the previous value. the check for changed
-         * bit depth in h264_set_parameter_from_sps() uses it and sets it to
-         * the current value */
-        h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
-
-        av_freep(&h->bipred_scratchpad);
+    // SPS/PPS
+    if ((ret = copy_parameter_set((void **)h->sps_buffers,
+                                  (void **)h1->sps_buffers,
+                                  MAX_SPS_COUNT, sizeof(SPS))) < 0)
+        return ret;
+    h->sps = h1->sps;
+    if ((ret = copy_parameter_set((void **)h->pps_buffers,
+                                  (void **)h1->pps_buffers,
+                                  MAX_PPS_COUNT, sizeof(PPS))) < 0)
+        return ret;
+    h->pps = h1->pps;
 
+    if (need_reinit || !inited) {
         h->width     = h1->width;
         h->height    = h1->height;
         h->mb_height = h1->mb_height;
@@ -470,74 +443,15 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
         h->mb_stride = h1->mb_stride;
         h->b_stride  = h1->b_stride;
 
-        if ((err = h264_slice_header_init(h, 1)) < 0) {
+        if ((err = h264_slice_header_init(h)) < 0) {
             av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
             return err;
         }
-        context_reinitialized = 1;
-
-        /* update linesize on resize. The decoder doesn't
-         * necessarily call h264_frame_start in the new thread */
-        h->linesize   = h1->linesize;
-        h->uvlinesize = h1->uvlinesize;
 
         /* copy block_offset since frame_start may not be called */
         memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
     }
 
-    if (!inited) {
-        H264SliceContext *orig_slice_ctx = h->slice_ctx;
-
-        for (i = 0; i < MAX_SPS_COUNT; i++)
-            av_freep(h->sps_buffers + i);
-
-        for (i = 0; i < MAX_PPS_COUNT; i++)
-            av_freep(h->pps_buffers + i);
-
-        memcpy(h, h1, sizeof(*h1));
-        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));
-        h->context_initialized = 0;
-
-        memset(&h->cur_pic, 0, sizeof(h->cur_pic));
-        av_frame_unref(&h->cur_pic.f);
-        h->cur_pic.tf.f = &h->cur_pic.f;
-
-        h->slice_ctx = orig_slice_ctx;
-
-        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));
-
-        h->avctx             = dst;
-        h->DPB               = NULL;
-        h->qscale_table_pool = NULL;
-        h->mb_type_pool      = NULL;
-        h->ref_index_pool    = NULL;
-        h->motion_val_pool   = NULL;
-
-        ret = ff_h264_alloc_tables(h);
-        if (ret < 0) {
-            av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
-            return ret;
-        }
-        ret = ff_h264_context_init(h);
-        if (ret < 0) {
-            av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
-            return ret;
-        }
-
-        h->rbsp_buffer      = NULL;
-        h->rbsp_buffer_size = 0;
-        h->bipred_scratchpad = NULL;
-        h->edge_emu_buffer   = NULL;
-
-        h->thread_context[0] = h;
-
-        h->context_initialized = 1;
-    }
-
     h->avctx->coded_height  = h1->avctx->coded_height;
     h->avctx->coded_width   = h1->avctx->coded_width;
     h->avctx->width         = h1->avctx->width;
@@ -550,43 +464,27 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
 
     for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
         ff_h264_unref_picture(h, &h->DPB[i]);
-        if (h1->DPB[i].f.buf[0] &&
+        if (h1->DPB[i].f->buf[0] &&
             (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
             return ret;
     }
 
     h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
     ff_h264_unref_picture(h, &h->cur_pic);
-    if (h1->cur_pic.f.buf[0]) {
+    if (h1->cur_pic.f->buf[0]) {
         ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
         if (ret < 0)
             return ret;
     }
 
+    h->enable_er       = h1->enable_er;
     h->workaround_bugs = h1->workaround_bugs;
     h->low_delay       = h1->low_delay;
     h->droppable       = h1->droppable;
 
-    /* frame_start may not be called for the next thread (if it's decoding
-     * a bottom field) so this has to be allocated here */
-    err = alloc_scratch_buffers(h, h1->linesize);
-    if (err < 0)
-        return err;
-
     // extradata/NAL handling
     h->is_avc = h1->is_avc;
-
-    // SPS/PPS
-    if ((ret = copy_parameter_set((void **)h->sps_buffers,
-                                  (void **)h1->sps_buffers,
-                                  MAX_SPS_COUNT, sizeof(SPS))) < 0)
-        return ret;
-    h->sps = h1->sps;
-    if ((ret = copy_parameter_set((void **)h->pps_buffers,
-                                  (void **)h1->pps_buffers,
-                                  MAX_PPS_COUNT, sizeof(PPS))) < 0)
-        return ret;
-    h->pps = h1->pps;
+    h->nal_length_size = h1->nal_length_size;
 
     // Dequantization matrices
     // FIXME these are big - can they be only copied when PPS changes?
@@ -603,21 +501,13 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
     h->dequant_coeff_pps = h1->dequant_coeff_pps;
 
     // POC timing
-    copy_fields(h, h1, poc_lsb, default_ref_list);
-
-    // reference lists
-    copy_fields(h, h1, short_ref, thread_context);
+    copy_fields(h, h1, poc_lsb, current_slice);
 
     copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
     copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
     copy_picture_range(h->delayed_pic, h1->delayed_pic,
                        MAX_DELAYED_PIC_COUNT + 2, h, h1);
 
-    h->last_slice_type = h1->last_slice_type;
-
-    if (context_reinitialized)
-        ff_h264_set_parameter_from_sps(h);
-
     if (!h->cur_pic_ptr)
         return 0;
 
@@ -628,7 +518,6 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
     }
     h->prev_frame_num_offset = h->frame_num_offset;
     h->prev_frame_num        = h->frame_num;
-    h->outputed_poc          = h->next_outputed_poc;
 
     h->recovery_frame        = h1->recovery_frame;
     h->frame_recovered       = h1->frame_recovered;
@@ -642,61 +531,38 @@ static int h264_frame_start(H264Context *h)
     int i, ret;
     const int pixel_shift = h->pixel_shift;
 
-    release_unused_pictures(h, 1);
-    h->cur_pic_ptr = NULL;
-
-    i = find_unused_picture(h);
-    if (i < 0) {
-        av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
-        return i;
-    }
-    pic = &h->DPB[i];
+    ret = initialize_cur_frame(h);
+    if (ret < 0)
+        return ret;
 
+    pic = h->cur_pic_ptr;
     pic->reference              = h->droppable ? 0 : h->picture_structure;
-    pic->f.coded_picture_number = h->coded_picture_number++;
+    pic->f->coded_picture_number = h->coded_picture_number++;
     pic->field_picture          = h->picture_structure != PICT_FRAME;
+    pic->frame_num               = h->frame_num;
     /*
      * Zero key_frame here; IDR markings per slice in frame or fields are ORed
      * in later.
      * See decode_nal_units().
      */
-    pic->f.key_frame = 0;
+    pic->f->key_frame = 0;
     pic->mmco_reset  = 0;
     pic->recovered   = 0;
 
-    if ((ret = alloc_picture(h, pic)) < 0)
-        return ret;
-
-    h->cur_pic_ptr = pic;
-    ff_h264_unref_picture(h, &h->cur_pic);
-    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);
-
-    assert(h->linesize && h->uvlinesize);
+    if (CONFIG_ERROR_RESILIENCE && h->enable_er)
+        ff_er_frame_start(&h->slice_ctx[0].er);
 
     for (i = 0; i < 16; i++) {
-        h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
-        h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
+        h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
+        h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
     }
     for (i = 0; i < 16; i++) {
         h->block_offset[16 + i]      =
-        h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
+        h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
         h->block_offset[48 + 16 + i] =
-        h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
+        h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
     }
 
-    /* can't be in alloc_tables because linesize isn't known there.
-     * FIXME: redo bipred weight to not require extra buffer? */
-    for (i = 0; i < h->slice_context_count; i++)
-        if (h->thread_context[i]) {
-            ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
-            if (ret < 0)
-                return ret;
-        }
-
     /* Some macroblocks can be accessed before they're available in case
      * of lost slices, MBAFF or threading. */
     memset(h->slice_table, -1,
@@ -717,7 +583,7 @@ static int h264_frame_start(H264Context *h)
     return 0;
 }
 
-static av_always_inline void backup_mb_border(H264Context *h, H264SliceContext *sl,
+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,
@@ -735,12 +601,12 @@ static av_always_inline void backup_mb_border(H264Context *h, H264SliceContext *
 
     if (!simple && FRAME_MBAFF(h)) {
         if (sl->mb_y & 1) {
-            if (!MB_MBAFF(h)) {
-                top_border = h->top_borders[0][sl->mb_x];
+            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);
-                if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
+                if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
                     if (chroma444) {
                         if (pixel_shift) {
                             AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
@@ -770,20 +636,20 @@ static av_always_inline void backup_mb_border(H264Context *h, H264SliceContext *
                     }
                 }
             }
-        } else if (MB_MBAFF(h)) {
+        } else if (MB_MBAFF(sl)) {
             top_idx = 0;
         } else
             return;
     }
 
-    top_border = h->top_borders[top_idx][sl->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);
     if (pixel_shift)
         AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
 
-    if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
+    if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
         if (chroma444) {
             if (pixel_shift) {
                 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
@@ -819,13 +685,13 @@ static av_always_inline void backup_mb_border(H264Context *h, H264SliceContext *
  * @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;
 
     for (i = 0; i < 2; i++) {
-        sl->luma_weight_flag[i]   = 0;
-        sl->chroma_weight_flag[i] = 0;
+        sl->pwt.luma_weight_flag[i]   = 0;
+        sl->pwt.chroma_weight_flag[i] = 0;
     }
 
     if (field < 0) {
@@ -836,8 +702,8 @@ static void implicit_weight_table(H264Context *h, H264SliceContext *sl, int fiel
         }
         if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
             sl->ref_list[0][0].poc + sl->ref_list[1][0].poc == 2 * cur_poc) {
-            sl->use_weight        = 0;
-            sl->use_weight_chroma = 0;
+            sl->pwt.use_weight        = 0;
+            sl->pwt.use_weight_chroma = 0;
             return;
         }
         ref_start  = 0;
@@ -850,16 +716,16 @@ static void implicit_weight_table(H264Context *h, H264SliceContext *sl, int fiel
         ref_count1 = 16 + 2 * sl->ref_count[1];
     }
 
-    sl->use_weight               = 2;
-    sl->use_weight_chroma        = 2;
-    sl->luma_log2_weight_denom   = 5;
-    sl->chroma_log2_weight_denom = 5;
+    sl->pwt.use_weight               = 2;
+    sl->pwt.use_weight_chroma        = 2;
+    sl->pwt.luma_log2_weight_denom   = 5;
+    sl->pwt.chroma_log2_weight_denom = 5;
 
     for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
         int poc0 = sl->ref_list[0][ref0].poc;
         for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
             int w = 32;
-            if (!sl->ref_list[0][ref0].long_ref && !sl->ref_list[1][ref1].long_ref) {
+            if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
                 int poc1 = sl->ref_list[1][ref1].poc;
                 int td   = av_clip_int8(poc1 - poc0);
                 if (td) {
@@ -871,10 +737,10 @@ static void implicit_weight_table(H264Context *h, H264SliceContext *sl, int fiel
                 }
             }
             if (field < 0) {
-                sl->implicit_weight[ref0][ref1][0] =
-                sl->implicit_weight[ref0][ref1][1] = w;
+                sl->pwt.implicit_weight[ref0][ref1][0] =
+                sl->pwt.implicit_weight[ref0][ref1][1] = w;
             } else {
-                sl->implicit_weight[ref0][ref1][field] = w;
+                sl->pwt.implicit_weight[ref0][ref1][field] = w;
             }
         }
     }
@@ -888,7 +754,7 @@ static void init_scan_tables(H264Context *h)
     int i;
     for (i = 0; i < 16; i++) {
 #define TRANSPOSE(x) (x >> 2) | ((x << 2) & 0xF)
-        h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
+        h->zigzag_scan[i] = TRANSPOSE(ff_zigzag_scan[i]);
         h->field_scan[i]  = TRANSPOSE(field_scan[i]);
 #undef TRANSPOSE
     }
@@ -901,7 +767,7 @@ static void init_scan_tables(H264Context *h)
 #undef TRANSPOSE
     }
     if (h->sps.transform_bypass) { // FIXME same ugly
-        h->zigzag_scan_q0          = zigzag_scan;
+        h->zigzag_scan_q0          = ff_zigzag_scan;
         h->zigzag_scan8x8_q0       = ff_zigzag_direct;
         h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
         h->field_scan_q0           = field_scan;
@@ -917,37 +783,10 @@ 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)
 {
 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
+                     CONFIG_H264_D3D11VA_HWACCEL + \
                      CONFIG_H264_VAAPI_HWACCEL + \
                      (CONFIG_H264_VDA_HWACCEL * 2) + \
                      CONFIG_H264_VDPAU_HWACCEL)
@@ -997,8 +836,11 @@ static enum AVPixelFormat get_pixel_format(H264Context *h)
 #if CONFIG_H264_DXVA2_HWACCEL
             *fmt++ = AV_PIX_FMT_DXVA2_VLD;
 #endif
+#if CONFIG_H264_D3D11VA_HWACCEL
+            *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
+#endif
 #if CONFIG_H264_VAAPI_HWACCEL
-            *fmt++ = AV_PIX_FMT_VAAPI_VLD;
+            *fmt++ = AV_PIX_FMT_VAAPI;
 #endif
 #if CONFIG_H264_VDA_HWACCEL
             *fmt++ = AV_PIX_FMT_VDA_VLD;
@@ -1028,13 +870,10 @@ 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);
-    int crop_present = h->sps.crop_left  || h->sps.crop_top ||
-                       h->sps.crop_right || h->sps.crop_bottom;
 
     /* handle container cropping */
-    if (!crop_present &&
-        FFALIGN(h->avctx->width,  16) == h->width &&
-        FFALIGN(h->avctx->height, 16) == h->height) {
+    if (FFALIGN(h->avctx->width,  16) == FFALIGN(width,  16) &&
+        FFALIGN(h->avctx->height, 16) == FFALIGN(height, 16)) {
         width  = h->avctx->width;
         height = h->avctx->height;
     }
@@ -1064,7 +903,7 @@ static int init_dimensions(H264Context *h)
     return 0;
 }
 
-static int h264_slice_header_init(H264Context *h, int reinit)
+static int h264_slice_header_init(H264Context *h)
 {
     int nb_slices = (HAVE_THREADS &&
                      h->avctx->active_thread_type & FF_THREAD_SLICE) ?
@@ -1083,8 +922,8 @@ static int h264_slice_header_init(H264Context *h, int reinit)
                   h->sps.num_units_in_tick, den, 1 << 30);
     }
 
-    if (reinit)
-        ff_h264_free_tables(h, 0);
+    ff_h264_free_tables(h);
+
     h->first_field           = 0;
     h->prev_interlaced_frame = 1;
 
@@ -1095,6 +934,25 @@ static int h264_slice_header_init(H264Context *h, int reinit)
         return ret;
     }
 
+    if (h->sps.bit_depth_luma < 8 || h->sps.bit_depth_luma > 10) {
+        av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
+               h->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);
+
     if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
         int max_slices;
         if (h->mb_height)
@@ -1108,53 +966,25 @@ 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");
             return ret;
         }
     } else {
-        for (i = 1; i < h->slice_context_count; i++) {
-            H264Context *c;
-            c                    = h->thread_context[i] = av_mallocz(sizeof(H264Context));
-            if (!c)
-                return AVERROR(ENOMEM);
-            c->avctx             = h->avctx;
-            c->vdsp              = h->vdsp;
-            c->h264dsp           = h->h264dsp;
-            c->h264qpel          = h->h264qpel;
-            c->h264chroma        = h->h264chroma;
-            c->sps               = h->sps;
-            c->pps               = h->pps;
-            c->pixel_shift       = h->pixel_shift;
-            c->width             = h->width;
-            c->height            = h->height;
-            c->linesize          = h->linesize;
-            c->uvlinesize        = h->uvlinesize;
-            c->chroma_x_shift    = h->chroma_x_shift;
-            c->chroma_y_shift    = h->chroma_y_shift;
-            c->droppable         = h->droppable;
-            c->low_delay         = h->low_delay;
-            c->mb_width          = h->mb_width;
-            c->mb_height         = h->mb_height;
-            c->mb_stride         = h->mb_stride;
-            c->mb_num            = h->mb_num;
-            c->flags             = h->flags;
-            c->workaround_bugs   = h->workaround_bugs;
-            c->pict_type         = h->pict_type;
-
-            h->slice_ctx[i].h264 = c;
-
-            init_scan_tables(c);
-            clone_tables(c, &h->slice_ctx[i], h, i);
-            c->context_initialized = 1;
-        }
+        for (i = 0; i < h->slice_context_count; i++) {
+            H264SliceContext *sl = &h->slice_ctx[i];
+
+            sl->h264               = h;
+            sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
+            sl->mvd_table[0]       = h->mvd_table[0]       + i * 8 * 2 * h->mb_stride;
+            sl->mvd_table[1]       = h->mvd_table[1]       + i * 8 * 2 * h->mb_stride;
 
-        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, sl)) < 0) {
                 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
                 return ret;
             }
+        }
     }
 
     h->context_initialized = 1;
@@ -1167,21 +997,20 @@ static int h264_slice_header_init(H264Context *h, int reinit)
  * 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;
     int ret;
     unsigned int slice_type, tmp, i, j;
-    int default_ref_list_done = 0;
     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;
@@ -1189,12 +1018,12 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
     first_mb_in_slice = get_ue_golomb(&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);
@@ -1216,11 +1045,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
     } else
         sl->slice_type_fixed = 0;
 
-    slice_type = golomb_to_pict_type[slice_type];
-    if (slice_type == AV_PICTURE_TYPE_I ||
-        (h0->current_slice != 0 && slice_type == h0->last_slice_type)) {
-        default_ref_list_done = 1;
-    }
+    slice_type         = ff_h264_golomb_to_pict_type[slice_type];
     sl->slice_type     = slice_type;
     sl->slice_type_nos = slice_type & 3;
 
@@ -1231,22 +1056,28 @@ 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;
+    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 (!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;
     }
-    h->pps = *h0->pps_buffers[pps_id];
+    if (!h->setup_finished) {
+        h->pps = *h->pps_buffers[pps_id];
+    } else if (h->dequant_coeff_pps != pps_id) {
+        av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
+        return AVERROR_INVALIDDATA;
+    }
 
-    if (!h0->sps_buffers[h->pps.sps_id]) {
+    if (!h->sps_buffers[h->pps.sps_id]) {
         av_log(h->avctx, AV_LOG_ERROR,
                "non-existing SPS %u referenced\n",
                h->pps.sps_id);
@@ -1254,65 +1085,77 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
     }
 
     if (h->pps.sps_id != h->sps.sps_id ||
-        h0->sps_buffers[h->pps.sps_id]->new) {
-        h0->sps_buffers[h->pps.sps_id]->new = 0;
+        h->sps_buffers[h->pps.sps_id]->new) {
+        h->sps_buffers[h->pps.sps_id]->new = 0;
 
-        h->sps = *h0->sps_buffers[h->pps.sps_id];
+        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) {
-            h->bit_depth_luma    = h->sps.bit_depth_luma;
-            h->chroma_format_idc = h->sps.chroma_format_idc;
+            h->chroma_format_idc != h->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 ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
-            return ret;
+
+        if (h->avctx->has_b_frames < 2)
+            h->avctx->has_b_frames = !h->low_delay;
+
     }
 
-    h->avctx->profile = ff_h264_get_profile(&h->sps);
-    h->avctx->level   = h->sps.level_idc;
-    h->avctx->refs    = h->sps.ref_frame_count;
+    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;
 
-    if (h->mb_width  != h->sps.mb_width ||
-        h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag))
-        needs_reinit = 1;
+        if (h->mb_width  != h->sps.mb_width ||
+            h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag))
+            needs_reinit = 1;
 
-    h->mb_width  = h->sps.mb_width;
-    h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
-    h->mb_num    = h->mb_width * h->mb_height;
-    h->mb_stride = h->mb_width + 1;
+        h->mb_width  = h->sps.mb_width;
+        h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
+        h->mb_num    = h->mb_width * h->mb_height;
+        h->mb_stride = h->mb_width + 1;
 
-    h->b_stride = h->mb_width * 4;
+        h->b_stride = h->mb_width * 4;
 
-    h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
+        h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
 
-    h->width  = 16 * h->mb_width;
-    h->height = 16 * h->mb_height;
+        h->width  = 16 * h->mb_width;
+        h->height = 16 * h->mb_height;
 
-    ret = init_dimensions(h);
-    if (ret < 0)
-        return ret;
+        ret = init_dimensions(h);
+        if (ret < 0)
+            return ret;
 
-    if (h->sps.video_signal_type_present_flag) {
-        h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
-                                                  : AVCOL_RANGE_MPEG;
-        if (h->sps.colour_description_present_flag) {
-            if (h->avctx->colorspace != h->sps.colorspace)
-                needs_reinit = 1;
-            h->avctx->color_primaries = h->sps.color_primaries;
-            h->avctx->color_trc       = h->sps.color_trc;
-            h->avctx->colorspace      = h->sps.colorspace;
+        if (h->sps.video_signal_type_present_flag) {
+            h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
+                : AVCOL_RANGE_MPEG;
+            if (h->sps.colour_description_present_flag) {
+                if (h->avctx->colorspace != h->sps.colorspace)
+                    needs_reinit = 1;
+                h->avctx->color_primaries = h->sps.color_primaries;
+                h->avctx->color_trc       = h->sps.color_trc;
+                h->avctx->colorspace      = h->sps.colorspace;
+            }
         }
     }
 
     if (h->context_initialized && needs_reinit) {
-        if (h != h0) {
+        h->context_initialized = 0;
+        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;
         }
 
@@ -1325,14 +1168,14 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
         av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
                "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
 
-        if ((ret = h264_slice_header_init(h, 1)) < 0) {
+        if ((ret = h264_slice_header_init(h)) < 0) {
             av_log(h->avctx, AV_LOG_ERROR,
                    "h264_slice_header_init() failed\n");
             return ret;
         }
     }
     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;
@@ -1342,52 +1185,58 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
             return ret;
         h->avctx->pix_fmt = ret;
 
-        if ((ret = h264_slice_header_init(h, 0)) < 0) {
+        if ((ret = h264_slice_header_init(h)) < 0) {
             av_log(h->avctx, AV_LOG_ERROR,
                    "h264_slice_header_init() failed\n");
             return ret;
         }
     }
 
-    if (h == h0 && h->dequant_coeff_pps != pps_id) {
+    if (sl == h->slice_ctx && h->dequant_coeff_pps != pps_id) {
         h->dequant_coeff_pps = pps_id;
-        h264_init_dequant_tables(h);
+        ff_h264_init_dequant_tables(h);
     }
 
-    h->frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
+    frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
+    if (!h->setup_finished)
+        h->frame_num = frame_num;
+
+    sl->mb_mbaff       = 0;
+
+    last_pic_structure = h->picture_structure;
+    last_pic_droppable = h->droppable;
 
-    h->mb_mbaff        = 0;
-    h->mb_aff_frame    = 0;
-    last_pic_structure = h0->picture_structure;
-    last_pic_droppable = h0->droppable;
-    h->droppable       = h->nal_ref_idc == 0;
+    droppable = h->nal_ref_idc == 0;
     if (h->sps.frame_mbs_only_flag) {
-        h->picture_structure = PICT_FRAME;
+        picture_structure = PICT_FRAME;
     } else {
         field_pic_flag = get_bits1(&sl->gb);
         if (field_pic_flag) {
             bottom_field_flag = get_bits1(&sl->gb);
-            h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
+            picture_structure = PICT_TOP_FIELD + bottom_field_flag;
         } else {
-            h->picture_structure = PICT_FRAME;
-            h->mb_aff_frame      = h->sps.mb_aff;
+            picture_structure = PICT_FRAME;
+            mb_aff_frame      = h->sps.mb_aff;
         }
     }
-    h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
+    if (!h->setup_finished) {
+        h->droppable         = droppable;
+        h->picture_structure = picture_structure;
+        h->mb_aff_frame      = mb_aff_frame;
+    }
+    sl->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
 
-    if (h0->current_slice != 0) {
-        if (last_pic_structure != h->picture_structure ||
-            last_pic_droppable != h->droppable) {
+    if (h->current_slice != 0) {
+        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);
-            h->picture_structure = last_pic_structure;
-            h->droppable         = last_pic_droppable;
             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;
         }
     } else {
@@ -1414,27 +1263,27 @@ 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 s->current_picture_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);
 
             /* figure out if we have a complementary field pair */
             if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
                 /* Previous field is unmatched. Don't display it, but let it
                  * remain for reference if marked as such. */
                 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
-                    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_droppable && 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 {
@@ -1465,9 +1314,9 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
             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);
-            ret = h264_frame_start(h);
+            ret = initialize_cur_frame(h);
             if (ret < 0) {
-                h0->first_field = 0;
+                h->first_field = 0;
                 return ret;
             }
 
@@ -1490,12 +1339,15 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
              * vectors.  Given we are concealing a lost frame, this probably
              * is not noticeable by comparison, but it should be fixed. */
             if (h->short_ref_count) {
-                if (prev) {
-                    av_image_copy(h->short_ref[0]->f.data,
-                                  h->short_ref[0]->f.linesize,
-                                  (const uint8_t **)prev->f.data,
-                                  prev->f.linesize,
-                                  h->avctx->pix_fmt,
+                if (prev &&
+                    h->short_ref[0]->f->width == prev->f->width &&
+                    h->short_ref[0]->f->height == prev->f->height &&
+                    h->short_ref[0]->f->format == prev->f->format) {
+                    av_image_copy(h->short_ref[0]->f->data,
+                                  h->short_ref[0]->f->linesize,
+                                  (const uint8_t **)prev->f->data,
+                                  prev->f->linesize,
+                                  prev->f->format,
                                   h->mb_width  * 16,
                                   h->mb_height * 16);
                     h->short_ref[0]->poc = prev->poc + 2;
@@ -1507,47 +1359,43 @@ 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->cur_pic_ptr = NULL;
-                h0->first_field = FIELD_PICTURE(h);
+                h->cur_pic_ptr = NULL;
+                h->first_field = FIELD_PICTURE(h);
             } else {
-                if (h0->cur_pic_ptr->frame_num != h->frame_num) {
+                if (h->cur_pic_ptr->frame_num != h->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. */
-                    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 {
             release_unused_pictures(h, 0);
         }
     }
-    if (h != h0 && (ret = clone_slice(h, h0)) < 0)
-        return ret;
-
-    h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
 
     assert(h->mb_num == h->mb_width * h->mb_height);
     if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
@@ -1574,20 +1422,34 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
         get_ue_golomb(&sl->gb); /* idr_pic_id */
 
     if (h->sps.poc_type == 0) {
-        h->poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
+        int 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(&sl->gb);
+        if (!h->setup_finished)
+            h->poc_lsb = poc_lsb;
+
+        if (h->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;
+        }
     }
 
     if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
-        h->delta_poc[0] = get_se_golomb(&sl->gb);
+        int delta_poc = get_se_golomb(&sl->gb);
+
+        if (!h->setup_finished)
+            h->delta_poc[0] = delta_poc;
 
-        if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
-            h->delta_poc[1] = get_se_golomb(&sl->gb);
+        if (h->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;
+        }
     }
 
-    ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->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)
         sl->redundant_pic_count = get_ue_golomb(&sl->gb);
@@ -1595,11 +1457,6 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
     ret = ff_set_ref_count(h, sl);
     if (ret < 0)
         return ret;
-    else if (ret == 1)
-        default_ref_list_done = 0;
-
-    if (!default_ref_list_done)
-        ff_h264_fill_default_ref_list(h, sl);
 
     if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
        ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
@@ -1612,15 +1469,16 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
     if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
         (h->pps.weighted_bipred_idc == 1 &&
          sl->slice_type_nos == AV_PICTURE_TYPE_B))
-        ff_pred_weight_table(h, sl);
+        ff_h264_pred_weight_table(&sl->gb, &h->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->use_weight = 0;
+        sl->pwt.use_weight = 0;
         for (i = 0; i < 2; i++) {
-            sl->luma_weight_flag[i]   = 0;
-            sl->chroma_weight_flag[i] = 0;
+            sl->pwt.luma_weight_flag[i]   = 0;
+            sl->pwt.chroma_weight_flag[i] = 0;
         }
     }
 
@@ -1630,9 +1488,9 @@ 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, &sl->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;
     }
@@ -1713,19 +1571,19 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
          h->nal_ref_idc == 0))
         sl->deblocking_filter = 0;
 
-    if (sl->deblocking_filter == 1 && h0->max_contexts > 1) {
-        if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
+    if (sl->deblocking_filter == 1 && h->max_contexts > 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 {
-            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 deblocking type 1, decoding such frames in sequential order\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 1;
@@ -1739,8 +1597,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
                           h->pps.chroma_qp_index_offset[1]) +
                    6 * (h->sps.bit_depth_luma - 8);
 
-    h0->last_slice_type = slice_type;
-    sl->slice_num       = ++h0->current_slice;
+    sl->slice_num       = ++h->current_slice;
     if (sl->slice_num >= MAX_SLICES) {
         av_log(h->avctx, AV_LOG_ERROR,
                "Too many slices, increase MAX_SLICES and recompile\n");
@@ -1752,16 +1609,16 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
         for (i = 0; i < 16; i++) {
             id_list[i] = 60;
             if (j < sl->list_count && i < sl->ref_count[j] &&
-                sl->ref_list[j][i].f.buf[0]) {
+                sl->ref_list[j][i].parent->f->buf[0]) {
                 int k;
-                AVBuffer *buf = sl->ref_list[j][i].f.buf[0]->buffer;
+                AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
                 for (k = 0; k < h->short_ref_count; k++)
-                    if (h->short_ref[k]->f.buf[0]->buffer == buf) {
+                    if (h->short_ref[k]->f->buf[0]->buffer == buf) {
                         id_list[i] = k;
                         break;
                     }
                 for (k = 0; k < h->long_ref_count; k++)
-                    if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
+                    if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
                         id_list[i] = h->short_ref_count + k;
                         break;
                     }
@@ -1795,8 +1652,8 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
                sl->qscale,
                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->pwt.use_weight,
+               sl->pwt.use_weight == 1 && sl->pwt.use_weight_chroma ? "c" : "",
                sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
     }
 
@@ -1821,7 +1678,7 @@ 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],
@@ -1836,7 +1693,7 @@ 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] = sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
+            int (*ref2frm)[64] = 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]];
@@ -1851,7 +1708,7 @@ 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] = sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
+                int (*ref2frm)[64] = 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]);
@@ -1884,7 +1741,7 @@ 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] = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2);
+        int (*ref2frm)[64] = 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);
@@ -1903,10 +1760,9 @@ 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 = sl->mb_xy;
     int top_xy, left_xy[LEFT_MBS];
@@ -1914,7 +1770,7 @@ static int fill_filter_caches(H264Context *h, H264SliceContext *sl, int mb_type)
     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. */
@@ -2051,7 +1907,7 @@ 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;
@@ -2070,32 +1926,32 @@ static void loop_filter(H264Context *h, H264SliceContext *sl, int start_x, int e
                 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);
 
                 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] +
+                dest_y  = h->cur_pic.f->data[0] +
+                          ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
+                dest_cb = h->cur_pic.f->data[1] +
                           (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
-                          mb_y * h->uvlinesize * block_h;
-                dest_cr = h->cur_pic.f.data[2] +
+                          mb_y * sl->uvlinesize * block_h;
+                dest_cr = h->cur_pic.f->data[2] +
                           (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
-                          mb_y * h->uvlinesize * block_h;
+                          mb_y * sl->uvlinesize * block_h;
                 // FIXME simplify above
 
-                if (MB_FIELD(h)) {
-                    linesize   = sl->mb_linesize   = h->linesize   * 2;
-                    uvlinesize = sl->mb_uvlinesize = h->uvlinesize * 2;
+                if (MB_FIELD(sl)) {
+                    linesize   = sl->mb_linesize   = sl->linesize   * 2;
+                    uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
                     if (mb_y & 1) { // FIXME move out of this function?
-                        dest_y  -= h->linesize   * 15;
-                        dest_cb -= h->uvlinesize * (block_h - 1);
-                        dest_cr -= h->uvlinesize * (block_h - 1);
+                        dest_y  -= sl->linesize   * 15;
+                        dest_cb -= sl->uvlinesize * (block_h - 1);
+                        dest_cr -= sl->uvlinesize * (block_h - 1);
                     }
                 } else {
-                    linesize   = sl->mb_linesize   = h->linesize;
-                    uvlinesize = sl->mb_uvlinesize = h->uvlinesize;
+                    linesize   = sl->mb_linesize   = sl->linesize;
+                    uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
                 }
                 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
                                  uvlinesize, 0);
@@ -2120,20 +1976,20 @@ static void loop_filter(H264Context *h, H264SliceContext *sl, int start_x, int e
     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 = 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 * (sl->mb_y      >> FIELD_PICTURE(h));
     int pic_height     = 16 *  h->mb_height >> FIELD_PICTURE(h);
@@ -2164,12 +2020,15 @@ static void decode_finish_row(H264Context *h, H264SliceContext *sl)
                               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;
+
+    if (!sl->h264->enable_er)
+        return;
 
     er->ref_count = sl->ref_count[0];
     ff_er_add_slice(er, startx, starty, endx, endy, status);
@@ -2179,14 +2038,22 @@ 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;
+    const H264Context *h = sl->h264;
     int lf_x_start = sl->mb_x;
+    int ret;
+
+    sl->linesize   = h->cur_pic_ptr->f->linesize[0];
+    sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
+
+    ret = alloc_scratch_buffers(sl, sl->linesize);
+    if (ret < 0)
+        return ret;
 
     sl->mb_skip_run = -1;
 
     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));
+                     (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
 
     if (h->pps.cabac) {
         /* realign */
@@ -2201,8 +2068,15 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
 
         for (;;) {
             // START_TIMER
-            int ret = ff_h264_decode_mb_cabac(h, sl);
-            int eos;
+            int ret, eos;
+
+            if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
+                av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
+                       sl->next_slice_idx);
+                return AVERROR_INVALIDDATA;
+            }
+
+            ret = ff_h264_decode_mb_cabac(h, sl);
             // STOP_TIMER("decode_mb_cabac")
 
             if (ret >= 0)
@@ -2222,7 +2096,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
 
             if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
                 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
-                er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->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);
@@ -2233,7 +2107,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
                        "error while decoding MB %d %d, bytestream %td\n",
                        sl->mb_x, sl->mb_y,
                        sl->cabac.bytestream_end - sl->cabac.bytestream);
-                er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
+                er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
                              sl->mb_y, ER_MB_ERROR);
                 return AVERROR_INVALIDDATA;
             }
@@ -2251,9 +2125,9 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
             }
 
             if (eos || sl->mb_y >= h->mb_height) {
-                tprintf(h->avctx, "slice end %d %d\n",
+                ff_tlog(h->avctx, "slice end %d %d\n",
                         get_bits_count(&sl->gb), sl->gb.size_in_bits);
-                er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->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);
@@ -2262,7 +2136,15 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
         }
     } else {
         for (;;) {
-            int ret = ff_h264_decode_mb_cavlc(h, sl);
+            int ret;
+
+            if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
+                av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
+                       sl->next_slice_idx);
+                return AVERROR_INVALIDDATA;
+            }
+
+            ret = ff_h264_decode_mb_cavlc(h, sl);
 
             if (ret >= 0)
                 ff_h264_hl_decode_mb(h, sl);
@@ -2280,7 +2162,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
             if (ret < 0) {
                 av_log(h->avctx, AV_LOG_ERROR,
                        "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
-                er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
+                er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
                              sl->mb_y, ER_MB_ERROR);
                 return ret;
             }
@@ -2296,16 +2178,16 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
                         predict_field_decoding_flag(h, sl);
                 }
                 if (sl->mb_y >= h->mb_height) {
-                    tprintf(h->avctx, "slice end %d %d\n",
+                    ff_tlog(h->avctx, "slice end %d %d\n",
                             get_bits_count(&sl->gb), sl->gb.size_in_bits);
 
                     if (get_bits_left(&sl->gb) == 0) {
-                        er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y,
+                        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, sl->resync_mb_x, sl->resync_mb_y,
+                        er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
                                      sl->mb_x - 1, sl->mb_y, ER_MB_END);
 
                         return AVERROR_INVALIDDATA;
@@ -2314,18 +2196,18 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
             }
 
             if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
-                tprintf(h->avctx, "slice end %d %d\n",
+                ff_tlog(h->avctx, "slice end %d %d\n",
                         get_bits_count(&sl->gb), sl->gb.size_in_bits);
 
                 if (get_bits_left(&sl->gb) == 0) {
-                    er_add_slice(h, sl, sl->resync_mb_x, sl->resync_mb_y,
+                    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, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
+                    er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
                                  sl->mb_y, ER_MB_ERROR);
 
                     return AVERROR_INVALIDDATA;
@@ -2344,33 +2226,48 @@ 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;
+    int i, j;
 
     if (h->avctx->hwaccel)
         return 0;
     if (context_count == 1) {
-        int ret = decode_slice(avctx, &h->slice_ctx[0]);
+        int ret;
+
+        h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
+
+        ret = decode_slice(avctx, &h->slice_ctx[0]);
         h->mb_y = h->slice_ctx[0].mb_y;
         return ret;
     } else {
-        for (i = 1; i < context_count; i++) {
-            hx                 = h->thread_context[i];
-            hx->er.error_count = 0;
+        for (i = 0; i < context_count; i++) {
+            int next_slice_idx = h->mb_width * h->mb_height;
+            int slice_idx;
+
+            sl                 = &h->slice_ctx[i];
+            sl->er.error_count = 0;
+
+            /* make sure none of those slices overlap */
+            slice_idx = sl->mb_y * h->mb_width + sl->mb_x;
+            for (j = 0; j < context_count; j++) {
+                H264SliceContext *sl2 = &h->slice_ctx[j];
+                int        slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x;
+
+                if (i == j || slice_idx2 < slice_idx)
+                    continue;
+                next_slice_idx = FFMIN(next_slice_idx, slice_idx2);
+            }
+            sl->next_slice_idx = next_slice_idx;
         }
 
         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];
         sl                   = &h->slice_ctx[context_count - 1];
         h->mb_y              = sl->mb_y;
-        h->droppable         = hx->droppable;
-        h->picture_structure = hx->picture_structure;
         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;
     }
 
     return 0;