Merge commit 'bf03a878a76dea29b36f368759e9f66102b39a5f'
[ffmpeg.git] / libavcodec / h264_slice.c
index 7c51796..a4aa671 100644 (file)
@@ -2,20 +2,20 @@
  * H.26L/H.264/AVC/JVT/14496-10/... decoder
  * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
  *
- * This file is part of Libav.
+ * This file is part of FFmpeg.
  *
- * Libav is free software; you can redistribute it and/or
+ * FFmpeg is free software; you can redistribute it and/or
  * modify it under the terms of the GNU Lesser General Public
  * License as published by the Free Software Foundation; either
  * version 2.1 of the License, or (at your option) any later version.
  *
- * Libav is distributed in the hope that it will be useful,
+ * FFmpeg is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  * Lesser General Public License for more details.
  *
  * You should have received a copy of the GNU Lesser General Public
- * License along with Libav; if not, write to the Free Software
+ * License along with FFmpeg; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
 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,
+    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,
+    7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
+   10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
+   14,14,14,14,
 };
 
-static const uint8_t field_scan[16] = {
+static const uint8_t field_scan[16+1] = {
     0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
     0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
     2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
     3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
 };
 
-static const uint8_t field_scan8x8[64] = {
+static const uint8_t field_scan8x8[64+1] = {
     0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
     1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
     2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
@@ -82,7 +86,7 @@ static const uint8_t field_scan8x8[64] = {
     7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
 };
 
-static const uint8_t field_scan8x8_cavlc[64] = {
+static const uint8_t field_scan8x8_cavlc[64+1] = {
     0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
     2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
     3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
@@ -102,7 +106,7 @@ static const uint8_t field_scan8x8_cavlc[64] = {
 };
 
 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
-static const uint8_t zigzag_scan8x8_cavlc[64] = {
+static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
     0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
     4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
     3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
@@ -219,6 +223,9 @@ static int alloc_picture(H264Context *h, H264Picture *pic)
 
     h->linesize   = pic->f.linesize[0];
     h->uvlinesize = pic->f.linesize[1];
+    pic->crop     = h->sps.crop;
+    pic->crop_top = h->sps.crop_top;
+    pic->crop_left= h->sps.crop_left;
 
     if (h->avctx->hwaccel) {
         const AVHWAccel *hwaccel = h->avctx->hwaccel;
@@ -230,6 +237,18 @@ static int alloc_picture(H264Context *h, H264Picture *pic)
             pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
         }
     }
+    if (!h->avctx->hwaccel && CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY && pic->f.data[2]) {
+        int h_chroma_shift, v_chroma_shift;
+        av_pix_fmt_get_chroma_sub_sample(pic->f.format,
+                                         &h_chroma_shift, &v_chroma_shift);
+
+        for(i=0; i<FF_CEIL_RSHIFT(h->avctx->height, v_chroma_shift); i++) {
+            memset(pic->f.data[1] + pic->f.linesize[1]*i,
+                   0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
+            memset(pic->f.data[2] + pic->f.linesize[2]*i,
+                   0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
+        }
+    }
 
     if (!h->qscale_table_pool) {
         ret = init_table_pools(h);
@@ -343,10 +362,12 @@ static void init_dequant4_coeff_table(H264Context *h)
     }
 }
 
-void h264_init_dequant_tables(H264Context *h)
+void ff_h264_init_dequant_tables(H264Context *h)
 {
     int i, x;
     init_dequant4_coeff_table(h);
+    memset(h->dequant8_coeff, 0, sizeof(h->dequant8_coeff));
+
     if (h->pps.transform_8x8_mode)
         init_dequant8_coeff_table(h);
     if (h->sps.transform_bypass) {
@@ -390,9 +411,9 @@ static void clone_tables(H264Context *dst, H264SliceContext *sl,
 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
 
 #define REBASE_PICTURE(pic, new_ctx, old_ctx)             \
-    ((pic && pic >= old_ctx->DPB &&                       \
-      pic < old_ctx->DPB + H264_MAX_PICTURE_COUNT) ?          \
-     &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
+    (((pic) && (pic) >= (old_ctx)->DPB &&                       \
+      (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ?          \
+     &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
 
 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
                                H264Context *new_base,
@@ -430,8 +451,8 @@ static int copy_parameter_set(void **to, void **from, int count, int size)
 }
 
 #define copy_fields(to, from, start_field, end_field)                   \
-    memcpy(&to->start_field, &from->start_field,                        \
-           (char *)&to->end_field - (char *)&to->start_field)
+    memcpy(&(to)->start_field, &(from)->start_field,                        \
+           (char *)&(to)->end_field - (char *)&(to)->start_field)
 
 static int h264_slice_header_init(H264Context *h, int reinit);
 
@@ -443,7 +464,7 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
     int context_reinitialized = 0;
     int i, ret;
 
-    if (dst == src || !h1->context_initialized)
+    if (dst == src)
         return 0;
 
     if (inited &&
@@ -469,21 +490,37 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
         h->mb_num    = h1->mb_num;
         h->mb_stride = h1->mb_stride;
         h->b_stride  = h1->b_stride;
+        // SPS/PPS
+        if ((ret = copy_parameter_set((void **)h->sps_buffers,
+                                      (void **)h1->sps_buffers,
+                                      MAX_SPS_COUNT, sizeof(SPS))) < 0)
+            return ret;
+        h->sps = h1->sps;
+        if ((ret = copy_parameter_set((void **)h->pps_buffers,
+                                      (void **)h1->pps_buffers,
+                                      MAX_PPS_COUNT, sizeof(PPS))) < 0)
+            return ret;
+        h->pps = h1->pps;
 
         if ((err = h264_slice_header_init(h, 1)) < 0) {
-            av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
+            av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed\n");
             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 0
+        h264_set_parameter_from_sps(h);
+        //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted
+        h->cur_chroma_format_idc = h1->cur_chroma_format_idc;
+#endif
     }
+    /* update linesize on resize for h264. The h264 decoder doesn't
+     * necessarily call ff_mpv_frame_start in the new thread */
+    h->linesize   = h1->linesize;
+    h->uvlinesize = h1->uvlinesize;
+
+    /* copy block_offset since frame_start may not be called */
+    memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
 
     if (!inited) {
         H264SliceContext *orig_slice_ctx = h->slice_ctx;
@@ -494,15 +531,20 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
         for (i = 0; i < MAX_PPS_COUNT; i++)
             av_freep(h->pps_buffers + i);
 
-        memcpy(h, h1, sizeof(*h1));
+        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, offsetof(H264Context, cabac));
+        memcpy(&h->cabac, &h1->cabac,
+               sizeof(H264Context) - offsetof(H264Context, cabac));
+
+
         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->er, 0, sizeof(h->er));
         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;
+        memset(&h->last_pic_for_ec, 0, sizeof(h->last_pic_for_ec));
 
         h->slice_ctx = orig_slice_ctx;
 
@@ -516,6 +558,28 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
         h->mb_type_pool      = NULL;
         h->ref_index_pool    = NULL;
         h->motion_val_pool   = NULL;
+        h->intra4x4_pred_mode= NULL;
+        h->non_zero_count    = NULL;
+        h->slice_table_base  = NULL;
+        h->slice_table       = NULL;
+        h->cbp_table         = NULL;
+        h->chroma_pred_mode_table = NULL;
+        memset(h->mvd_table, 0, sizeof(h->mvd_table));
+        h->direct_table      = NULL;
+        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;
+
+        memset(&h->cur_pic, 0, sizeof(h->cur_pic));
+        av_frame_unref(&h->cur_pic.f);
+        h->cur_pic.tf.f = &h->cur_pic.f;
 
         ret = ff_h264_alloc_tables(h);
         if (ret < 0) {
@@ -527,17 +591,13 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
             av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
             return ret;
         }
-
-        for (i = 0; i < 2; i++) {
-            h->rbsp_buffer[i]      = NULL;
-            h->rbsp_buffer_size[i] = 0;
         }
+
         h->bipred_scratchpad = NULL;
         h->edge_emu_buffer   = NULL;
 
         h->thread_context[0] = h;
-
-        h->context_initialized = 1;
+        h->context_initialized = h1->context_initialized;
     }
 
     h->avctx->coded_height  = h1->avctx->coded_height;
@@ -550,9 +610,9 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
     h->droppable            = h1->droppable;
     h->low_delay            = h1->low_delay;
 
-    for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
+    for (i = 0; h->DPB && i < H264_MAX_PICTURE_COUNT; i++) {
         ff_h264_unref_picture(h, &h->DPB[i]);
-        if (h1->DPB[i].f.buf[0] &&
+        if (h1->DPB && h1->DPB[i].f.buf[0] &&
             (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
             return ret;
     }
@@ -569,12 +629,6 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
     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;
 
@@ -615,7 +669,7 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
     copy_picture_range(h->delayed_pic, h1->delayed_pic,
                        MAX_DELAYED_PIC_COUNT + 2, h, h1);
 
-    h->last_slice_type = h1->last_slice_type;
+    h->frame_recovered       = h1->frame_recovered;
 
     if (context_reinitialized)
         ff_h264_set_parameter_from_sps(h);
@@ -633,7 +687,6 @@ int ff_h264_update_thread_context(AVCodecContext *dst,
     h->outputed_poc          = h->next_outputed_poc;
 
     h->recovery_frame        = h1->recovery_frame;
-    h->frame_recovered       = h1->frame_recovered;
 
     return err;
 }
@@ -643,6 +696,17 @@ static int h264_frame_start(H264Context *h)
     H264Picture *pic;
     int i, ret;
     const int pixel_shift = h->pixel_shift;
+    int c[4] = {
+        1<<(h->sps.bit_depth_luma-1),
+        1<<(h->sps.bit_depth_chroma-1),
+        1<<(h->sps.bit_depth_chroma-1),
+        -1
+    };
+
+    if (!ff_thread_can_start_frame(h->avctx)) {
+        av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
+        return -1;
+    }
 
     release_unused_pictures(h, 1);
     h->cur_pic_ptr = NULL;
@@ -657,6 +721,7 @@ static int h264_frame_start(H264Context *h)
     pic->reference              = h->droppable ? 0 : h->picture_structure;
     pic->f.coded_picture_number = h->coded_picture_number++;
     pic->field_picture          = h->picture_structure != PICT_FRAME;
+
     /*
      * Zero key_frame here; IDR markings per slice in frame or fields are ORed
      * in later.
@@ -665,17 +730,29 @@ static int h264_frame_start(H264Context *h)
     pic->f.key_frame = 0;
     pic->mmco_reset  = 0;
     pic->recovered   = 0;
+    pic->invalid_gap = 0;
+    pic->sei_recovery_frame_cnt = h->sei_recovery_frame_cnt;
 
     if ((ret = alloc_picture(h, pic)) < 0)
         return ret;
+    if(!h->frame_recovered && !h->avctx->hwaccel &&
+       !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
+        avpriv_color_frame(&pic->f, c);
 
     h->cur_pic_ptr = pic;
     ff_h264_unref_picture(h, &h->cur_pic);
+    if (CONFIG_ERROR_RESILIENCE) {
+        ff_h264_set_erpic(&h->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)
+    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);
+    }
 
     assert(h->linesize && h->uvlinesize);
 
@@ -690,20 +767,6 @@ static int h264_frame_start(H264Context *h)
         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((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,
-           (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
-
     /* We mark the current picture as non-reference after allocating it, so
      * that if we break out due to an error it can be released automatically
      * in the next ff_mpv_frame_start().
@@ -888,13 +951,13 @@ static void init_scan_tables(H264Context *h)
 {
     int i;
     for (i = 0; i < 16; i++) {
-#define TRANSPOSE(x) (x >> 2) | ((x << 2) & 0xF)
+#define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
         h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
         h->field_scan[i]  = TRANSPOSE(field_scan[i]);
 #undef TRANSPOSE
     }
     for (i = 0; i < 64; i++) {
-#define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
+#define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
         h->zigzag_scan8x8[i]       = TRANSPOSE(ff_zigzag_direct[i]);
         h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
         h->field_scan8x8[i]        = TRANSPOSE(field_scan8x8[i]);
@@ -902,19 +965,19 @@ 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_scan8x8_q0       = ff_zigzag_direct;
-        h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
-        h->field_scan_q0           = field_scan;
-        h->field_scan8x8_q0        = field_scan8x8;
-        h->field_scan8x8_cavlc_q0  = field_scan8x8_cavlc;
+        memcpy(h->zigzag_scan_q0          , zigzag_scan             , sizeof(h->zigzag_scan_q0         ));
+        memcpy(h->zigzag_scan8x8_q0       , ff_zigzag_direct        , sizeof(h->zigzag_scan8x8_q0      ));
+        memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc    , sizeof(h->zigzag_scan8x8_cavlc_q0));
+        memcpy(h->field_scan_q0           , field_scan              , sizeof(h->field_scan_q0          ));
+        memcpy(h->field_scan8x8_q0        , field_scan8x8           , sizeof(h->field_scan8x8_q0       ));
+        memcpy(h->field_scan8x8_cavlc_q0  , field_scan8x8_cavlc     , sizeof(h->field_scan8x8_cavlc_q0 ));
     } else {
-        h->zigzag_scan_q0          = h->zigzag_scan;
-        h->zigzag_scan8x8_q0       = h->zigzag_scan8x8;
-        h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
-        h->field_scan_q0           = h->field_scan;
-        h->field_scan8x8_q0        = h->field_scan8x8;
-        h->field_scan8x8_cavlc_q0  = h->field_scan8x8_cavlc;
+        memcpy(h->zigzag_scan_q0          , h->zigzag_scan          , sizeof(h->zigzag_scan_q0         ));
+        memcpy(h->zigzag_scan8x8_q0       , h->zigzag_scan8x8       , sizeof(h->zigzag_scan8x8_q0      ));
+        memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
+        memcpy(h->field_scan_q0           , h->field_scan           , sizeof(h->field_scan_q0          ));
+        memcpy(h->field_scan8x8_q0        , h->field_scan8x8        , sizeof(h->field_scan8x8_q0       ));
+        memcpy(h->field_scan8x8_cavlc_q0  , h->field_scan8x8_cavlc  , sizeof(h->field_scan8x8_cavlc_q0 ));
     }
 }
 
@@ -946,7 +1009,7 @@ static int clone_slice(H264Context *dst, H264Context *src)
     return 0;
 }
 
-static enum AVPixelFormat get_pixel_format(H264Context *h)
+static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
 {
 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
                      CONFIG_H264_VAAPI_HWACCEL + \
@@ -954,6 +1017,7 @@ static enum AVPixelFormat get_pixel_format(H264Context *h)
                      CONFIG_H264_VDPAU_HWACCEL)
     enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
     const enum AVPixelFormat *choices = pix_fmts;
+    int i;
 
     switch (h->sps.bit_depth_luma) {
     case 9:
@@ -978,11 +1042,35 @@ static enum AVPixelFormat get_pixel_format(H264Context *h)
         else
             *fmt++ = AV_PIX_FMT_YUV420P10;
         break;
+    case 12:
+        if (CHROMA444(h)) {
+            if (h->avctx->colorspace == AVCOL_SPC_RGB) {
+                *fmt++ = AV_PIX_FMT_GBRP12;
+            } else
+                *fmt++ = AV_PIX_FMT_YUV444P12;
+        } else if (CHROMA422(h))
+            *fmt++ = AV_PIX_FMT_YUV422P12;
+        else
+            *fmt++ = AV_PIX_FMT_YUV420P12;
+        break;
+    case 14:
+        if (CHROMA444(h)) {
+            if (h->avctx->colorspace == AVCOL_SPC_RGB) {
+                *fmt++ = AV_PIX_FMT_GBRP14;
+            } else
+                *fmt++ = AV_PIX_FMT_YUV444P14;
+        } else if (CHROMA422(h))
+            *fmt++ = AV_PIX_FMT_YUV422P14;
+        else
+            *fmt++ = AV_PIX_FMT_YUV420P14;
+        break;
     case 8:
 #if CONFIG_H264_VDPAU_HWACCEL
         *fmt++ = AV_PIX_FMT_VDPAU;
 #endif
         if (CHROMA444(h)) {
+            if (h->avctx->colorspace == AVCOL_SPC_YCGCO)
+                av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
             if (h->avctx->colorspace == AVCOL_SPC_RGB)
                 *fmt++ = AV_PIX_FMT_GBRP;
             else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
@@ -1021,7 +1109,10 @@ static enum AVPixelFormat get_pixel_format(H264Context *h)
 
     *fmt = AV_PIX_FMT_NONE;
 
-    return ff_get_format(h->avctx, choices);
+    for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
+        if (choices[i] == h->avctx->pix_fmt && !force_callback)
+            return choices[i];
+    return ff_thread_get_format(h->avctx, choices);
 }
 
 /* export coded and cropped frame dimensions to AVCodecContext */
@@ -1031,6 +1122,8 @@ static int init_dimensions(H264Context *h)
     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;
+    av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
+    av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
 
     /* handle container cropping */
     if (!crop_present &&
@@ -1081,7 +1174,7 @@ static int h264_slice_header_init(H264Context *h, int reinit)
         if (h->x264_build < 44U)
             den *= 2;
         av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
-                  h->sps.num_units_in_tick, den, 1 << 30);
+                  h->sps.num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
     }
 
     if (reinit)
@@ -1093,7 +1186,7 @@ static int h264_slice_header_init(H264Context *h, int reinit)
     ret = ff_h264_alloc_tables(h);
     if (ret < 0) {
         av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
-        return ret;
+        goto fail;
     }
 
     if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
@@ -1112,14 +1205,16 @@ static int h264_slice_header_init(H264Context *h, int reinit)
         ret = ff_h264_context_init(h);
         if (ret < 0) {
             av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
-            return ret;
+            goto fail;
         }
     } 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);
+            if (!c) {
+                ret = AVERROR(ENOMEM);
+                goto fail;
+            }
             c->avctx             = h->avctx;
             c->vdsp              = h->vdsp;
             c->h264dsp           = h->h264dsp;
@@ -1128,6 +1223,7 @@ static int h264_slice_header_init(H264Context *h, int reinit)
             c->sps               = h->sps;
             c->pps               = h->pps;
             c->pixel_shift       = h->pixel_shift;
+            c->cur_chroma_format_idc = h->cur_chroma_format_idc;
             c->width             = h->width;
             c->height            = h->height;
             c->linesize          = h->linesize;
@@ -1154,13 +1250,28 @@ 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) {
                 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
-                return ret;
+                goto fail;
             }
     }
 
     h->context_initialized = 1;
 
     return 0;
+fail:
+    ff_h264_free_tables(h, 0);
+    h->context_initialized = 0;
+    return ret;
+}
+
+static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
+{
+    switch (a) {
+    case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
+    case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
+    case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
+    default:
+        return a;
+    }
 }
 
 /**
@@ -1179,15 +1290,18 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
     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 must_reinit;
     int needs_reinit = 0;
     int field_pic_flag, bottom_field_flag;
+    int first_slice = h == h0 && !h0->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(&h->gb);
+    first_mb_in_slice = get_ue_golomb_long(&h->gb);
 
     if (first_mb_in_slice == 0) { // FIXME better field boundary detection
         if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
@@ -1218,10 +1332,6 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
         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;
-    }
     sl->slice_type     = slice_type;
     sl->slice_type_nos = slice_type & 3;
 
@@ -1231,6 +1341,15 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
         return AVERROR_INVALIDDATA;
     }
 
+    if (
+        (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
+        (h->avctx->skip_frame >= AVDISCARD_BIDIR  && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
+        (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
+        (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE) ||
+         h->avctx->skip_frame >= AVDISCARD_ALL) {
+         return SLICE_SKIPED;
+     }
+
     // to make a few old functions happy, it's wrong though
     h->pict_type = sl->slice_type;
 
@@ -1245,21 +1364,43 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
                pps_id);
         return AVERROR_INVALIDDATA;
     }
-    h->pps = *h0->pps_buffers[pps_id];
+    if (h0->au_pps_id >= 0 && pps_id != h0->au_pps_id) {
+        av_log(h->avctx, AV_LOG_ERROR,
+               "PPS change from %d to %d forbidden\n",
+               h0->au_pps_id, pps_id);
+        return AVERROR_INVALIDDATA;
+    }
 
-    if (!h0->sps_buffers[h->pps.sps_id]) {
+    pps = h0->pps_buffers[pps_id];
+
+    if (!h0->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];
+
+    if (pps->sps_id != h->sps.sps_id ||
+        pps->sps_id != h->current_sps_id ||
+        h0->sps_buffers[pps->sps_id]->new) {
 
-    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;
+        if (!first_slice) {
+            av_log(h->avctx, AV_LOG_ERROR,
+               "SPS changed in the middle of the frame\n");
+            return AVERROR_INVALIDDATA;
+        }
 
         h->sps = *h0->sps_buffers[h->pps.sps_id];
 
+        if (h->mb_width  != h->sps.mb_width ||
+            h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
+            h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
+            h->cur_chroma_format_idc != h->sps.chroma_format_idc
+        )
+            needs_reinit = 1;
+
         if (h->bit_depth_luma    != h->sps.bit_depth_luma ||
             h->chroma_format_idc != h->sps.chroma_format_idc) {
             h->bit_depth_luma    = h->sps.bit_depth_luma;
@@ -1274,9 +1415,19 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
     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;
+    must_reinit = (h->context_initialized &&
+                    (   16*h->sps.mb_width != h->avctx->coded_width
+                     || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
+                     || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
+                     || h->cur_chroma_format_idc != h->sps.chroma_format_idc
+                     || h->mb_width  != h->sps.mb_width
+                     || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
+                    ));
+    if (non_j_pixfmt(h0->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h0, 0)))
+        must_reinit = 1;
+
+    if (first_slice && av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio))
+        must_reinit = 1;
 
     h->mb_width  = h->sps.mb_width;
     h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
@@ -1295,8 +1446,8 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
         return ret;
 
     if (h->sps.video_signal_type_present_flag) {
-        h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
-                                                  : AVCOL_RANGE_MPEG;
+        h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
+                                                    : AVCOL_RANGE_MPEG;
         if (h->sps.colour_description_present_flag) {
             if (h->avctx->colorspace != h->sps.colorspace)
                 needs_reinit = 1;
@@ -1306,7 +1457,8 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
         }
     }
 
-    if (h->context_initialized && needs_reinit) {
+    if (h->context_initialized &&
+        (must_reinit || needs_reinit)) {
         if (h != h0) {
             av_log(h->avctx, AV_LOG_ERROR,
                    "changing width %d -> %d / height %d -> %d on "
@@ -1317,14 +1469,16 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
             return AVERROR_INVALIDDATA;
         }
 
+        av_assert1(first_slice);
+
         ff_h264_flush_change(h);
 
-        if ((ret = get_pixel_format(h)) < 0)
+        if ((ret = get_pixel_format(h, 1)) < 0)
             return ret;
         h->avctx->pix_fmt = ret;
 
         av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
-               "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
+               "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
 
         if ((ret = h264_slice_header_init(h, 1)) < 0) {
             av_log(h->avctx, AV_LOG_ERROR,
@@ -1339,7 +1493,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
             return AVERROR_PATCHWELCOME;
         }
 
-        if ((ret = get_pixel_format(h)) < 0)
+        if ((ret = get_pixel_format(h, 1)) < 0)
             return ret;
         h->avctx->pix_fmt = ret;
 
@@ -1350,40 +1504,48 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
         }
     }
 
-    if (h == h0 && h->dequant_coeff_pps != pps_id) {
+    if (first_slice && 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(&h->gb, h->sps.log2_max_frame_num);
+    frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
+    if (!first_slice) {
+        if (h0->frame_num != frame_num) {
+            av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
+                   h0->frame_num, frame_num);
+            return AVERROR_INVALIDDATA;
+        }
+    }
 
     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 {
+        if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
+            av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
+            return -1;
+        }
         field_pic_flag = get_bits1(&h->gb);
+
         if (field_pic_flag) {
             bottom_field_flag = get_bits1(&h->gb);
-            h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
+            picture_structure = PICT_TOP_FIELD + bottom_field_flag;
         } else {
-            h->picture_structure = PICT_FRAME;
+            picture_structure = PICT_FRAME;
             h->mb_aff_frame      = h->sps.mb_aff;
         }
     }
-    h->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
-
-    if (h0->current_slice != 0) {
-        if (last_pic_structure != h->picture_structure ||
-            last_pic_droppable != h->droppable) {
+    if (h0->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);
-            h->picture_structure = last_pic_structure;
-            h->droppable         = last_pic_droppable;
             return AVERROR_INVALIDDATA;
         } else if (!h0->cur_pic_ptr) {
             av_log(h->avctx, AV_LOG_ERROR,
@@ -1391,7 +1553,14 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
                    h0->current_slice + 1);
             return AVERROR_INVALIDDATA;
         }
-    } else {
+    }
+
+    h->picture_structure = picture_structure;
+    h->droppable         = droppable;
+    h->frame_num         = frame_num;
+    h->mb_field_decoding_flag = picture_structure != PICT_FRAME;
+
+    if (h0->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) {
@@ -1414,17 +1583,23 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
          * Here, we're using that to see if we should mark previously
          * decode frames as "finished".
          * We have to do that before the "dummy" in-between frame allocation,
-         * since that can modify s->current_picture_ptr. */
+         * since that can modify h->cur_pic_ptr. */
         if (h0->first_field) {
             assert(h0->cur_pic_ptr);
             assert(h0->cur_pic_ptr->f.buf[0]);
             assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
 
+            /* Mark old field/frame as completed */
+            if (h0->cur_pic_ptr->tf.owner == h0->avctx) {
+                ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
+                                          last_pic_structure == PICT_BOTTOM_FIELD);
+            }
+
             /* figure out if we have a complementary field pair */
             if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
                 /* Previous field is unmatched. Don't display it, but let it
                  * remain for reference if marked as such. */
-                if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
+                if (last_pic_structure != PICT_FRAME) {
                     ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
                                               last_pic_structure == PICT_TOP_FIELD);
                 }
@@ -1434,7 +1609,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
                      * 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) {
+                    if (last_pic_structure != PICT_FRAME) {
                         ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
                                                   last_pic_structure == PICT_TOP_FIELD);
                     }
@@ -1461,11 +1636,14 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
             }
         }
 
-        while (h->frame_num != h->prev_frame_num &&
+        while (h->frame_num != h->prev_frame_num && !h0->first_field &&
                h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
             H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
             av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
                    h->frame_num, h->prev_frame_num);
+            if (!h->sps.gaps_in_frame_num_allowed_flag)
+                for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
+                    h->last_pocs[i] = INT_MIN;
             ret = h264_frame_start(h);
             if (ret < 0) {
                 h0->first_field = 0;
@@ -1475,6 +1653,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
             h->prev_frame_num++;
             h->prev_frame_num        %= 1 << h->sps.log2_max_frame_num;
             h->cur_pic_ptr->frame_num = h->prev_frame_num;
+            h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
             ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
             ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
             ret = ff_generate_sliding_window_mmcos(h, 1);
@@ -1517,10 +1696,14 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
             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);
             } 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);
                     /* This and the previous field had different frame_nums.
                      * Consider this field first in pair. Throw away previous
                      * one except for reference purposes. */
@@ -1544,13 +1727,32 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
         } else {
             release_unused_pictures(h, 0);
         }
+        /* Some macroblocks can be accessed before they're available in case
+        * of lost slices, MBAFF or threading. */
+        if (FIELD_PICTURE(h)) {
+            for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
+                memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
+        } else {
+            memset(h->slice_table, -1,
+                (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
+        }
+        h0->last_slice_type = -1;
     }
     if (h != h0 && (ret = clone_slice(h, h0)) < 0)
         return ret;
 
+    /* can't be in alloc_tables because linesize isn't known there.
+     * FIXME: redo bipred weight to not require extra buffer? */
+    for (i = 0; i < h->slice_context_count; i++)
+        if (h->thread_context[i]) {
+            ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
+            if (ret < 0)
+                return ret;
+        }
+
     h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
 
-    assert(h->mb_num == h->mb_width * h->mb_height);
+    av_assert1(h->mb_num == h->mb_width * h->mb_height);
     if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
         first_mb_in_slice >= h->mb_num) {
         av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
@@ -1561,7 +1763,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
                                FIELD_OR_MBAFF_PICTURE(h);
     if (h->picture_structure == PICT_BOTTOM_FIELD)
         h->resync_mb_y = h->mb_y = h->mb_y + 1;
-    assert(h->mb_y < h->mb_height);
+    av_assert1(h->mb_y < h->mb_height);
 
     if (h->picture_structure == PICT_FRAME) {
         h->curr_pic_num = h->frame_num;
@@ -1596,11 +1798,14 @@ 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)
+    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)))) {
+
         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);
@@ -1707,6 +1912,8 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
 
     if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
         (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
+         h->nal_unit_type != NAL_IDR_SLICE) ||
+        (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
          sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
         (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR  &&
          sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
@@ -1723,13 +1930,16 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
             h0->max_contexts = 1;
             if (!h0->single_decode_warning) {
                 av_log(h->avctx, AV_LOG_INFO,
-                       "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
+                       "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;
             }
             if (h != h0) {
                 av_log(h->avctx, AV_LOG_ERROR,
                        "Deblocking switched inside frame.\n");
-                return 1;
+                return SLICE_SINGLETHREAD;
             }
         }
     }
@@ -1741,10 +1951,16 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Contex
                    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;
-    if (sl->slice_num >= MAX_SLICES) {
-        av_log(h->avctx, AV_LOG_ERROR,
-               "Too many slices, increase MAX_SLICES and recompile\n");
+
+    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
+        && 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);
     }
 
     for (j = 0; j < 2; j++) {
@@ -1780,6 +1996,11 @@ 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->sps.new =
+    h0->sps_buffers[h->pps.sps_id]->new = 0;
+    h->current_sps_id = h->pps.sps_id;
+
     if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
         av_log(h->avctx, AV_LOG_DEBUG,
                "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
@@ -1837,7 +2058,7 @@ 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] = (void*)(sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
             AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
             ref_cache[0 - 1 * 8] =
             ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
@@ -1852,7 +2073,7 @@ 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] =(void*)( sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
                 AV_COPY32(mv_dst - 1 +  0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
                 AV_COPY32(mv_dst - 1 +  8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
                 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
@@ -1885,7 +2106,7 @@ 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] = (void*)(sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
         uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
         uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
         AV_WN32A(&ref_cache[0 * 8], ref01);
@@ -2158,7 +2379,7 @@ static void decode_finish_row(H264Context *h, H264SliceContext *sl)
 
     ff_h264_draw_horiz_band(h, sl, top, height);
 
-    if (h->droppable)
+    if (h->droppable || h->er.error_occurred)
         return;
 
     ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
@@ -2169,12 +2390,11 @@ static void er_add_slice(H264Context *h, H264SliceContext *sl,
                          int startx, int starty,
                          int endx, int endy, int status)
 {
-#if CONFIG_ERROR_RESILIENCE
-    ERContext *er = &h->er;
+    if (CONFIG_ERROR_RESILIENCE) {
+        ERContext *er = &h->er;
 
-    er->ref_count = sl->ref_count[0];
-    ff_er_add_slice(er, startx, starty, endx, endy, status);
-#endif
+        ff_er_add_slice(er, startx, starty, endx, endy, status);
+    }
 }
 
 static int decode_slice(struct AVCodecContext *avctx, void *arg)
@@ -2185,10 +2405,22 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
 
     h->mb_skip_run = -1;
 
+    av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3));
+
     h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
                     avctx->codec_id != AV_CODEC_ID_H264 ||
                     (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
 
+    if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->er.error_status_table) {
+        const int start_i  = av_clip(h->resync_mb_x + h->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
+        if (start_i) {
+            int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]];
+            prev_status &= ~ VP_START;
+            if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
+                h->er.error_occurred = 1;
+        }
+    }
+
     if (h->pps.cabac) {
         /* realign */
         align_get_bits(&h->gb);
@@ -2229,9 +2461,11 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
                     loop_filter(h, sl, lf_x_start, h->mb_x + 1);
                 return 0;
             }
-            if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
+            if (h->cabac.bytestream > h->cabac.bytestream_end + 2 )
+                av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", h->cabac.bytestream_end - h->cabac.bytestream);
+            if (ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 4) {
                 av_log(h->avctx, AV_LOG_ERROR,
-                       "error while decoding MB %d %d, bytestream %td\n",
+                       "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
                        h->mb_x, h->mb_y,
                        h->cabac.bytestream_end - h->cabac.bytestream);
                 er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, h->mb_x,
@@ -2300,14 +2534,15 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
                     tprintf(h->avctx, "slice end %d %d\n",
                             get_bits_count(&h->gb), h->gb.size_in_bits);
 
-                    if (get_bits_left(&h->gb) == 0) {
+                    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);
 
                         return 0;
                     } else {
                         er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y,
-                                     h->mb_x - 1, h->mb_y, ER_MB_END);
+                                     h->mb_x, h->mb_y, ER_MB_END);
 
                         return AVERROR_INVALIDDATA;
                     }
@@ -2348,14 +2583,21 @@ int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
     H264Context *hx;
     int i;
 
-    if (h->avctx->hwaccel)
+    av_assert0(h->mb_y < h->mb_height);
+
+    if (h->avctx->hwaccel ||
+        h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
         return 0;
     if (context_count == 1) {
         return decode_slice(avctx, &h->slice_ctx[0]);
     } else {
+        av_assert0(context_count > 0);
         for (i = 1; i < context_count; i++) {
             hx                 = h->thread_context[i];
-            hx->er.error_count = 0;
+            if (CONFIG_ERROR_RESILIENCE) {
+                hx->er.error_count = 0;
+            }
+            hx->x264_build     = h->x264_build;
         }
 
         avctx->execute(avctx, decode_slice, h->slice_ctx,
@@ -2367,8 +2609,10 @@ int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
         h->mb_y              = hx->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;
+        if (CONFIG_ERROR_RESILIENCE) {
+            for (i = 1; i < context_count; i++)
+                h->er.error_count += h->thread_context[i]->er.error_count;
+        }
     }
 
     return 0;