[FFmpeg-cvslog] Merge commit 'b77fffa127663028169c5ed543956af4b9496c29'

Clément Bœsch git at videolan.org
Mon Jun 20 15:34:56 CEST 2016


ffmpeg | branch: master | Clément Bœsch <clement at stupeflix.com> | Mon Jun 20 13:00:17 2016 +0200| [405398989093f5ef19cb63eab6a9205a0c40ce4f] | committer: Clément Bœsch

Merge commit 'b77fffa127663028169c5ed543956af4b9496c29'

* commit 'b77fffa127663028169c5ed543956af4b9496c29':
  h264: make slice threading work with deblocking_filter=1

again label, and SLICE_SINGLETHREAD error handling are preserved as that
SLICE_SINGLETHREAD can be raised on a remaining case.

slice_context_count is also kept since it's still in use.

Merged-by: Clément Bœsch <clement at stupeflix.com>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=405398989093f5ef19cb63eab6a9205a0c40ce4f
---

 libavcodec/h264.h       |    7 ++++-
 libavcodec/h264_slice.c |   65 ++++++++++++++++++++++++++++++-----------------
 2 files changed, 48 insertions(+), 24 deletions(-)

diff --git a/libavcodec/h264.h b/libavcodec/h264.h
index 5e9827d..61d94cd 100644
--- a/libavcodec/h264.h
+++ b/libavcodec/h264.h
@@ -407,7 +407,6 @@ typedef struct H264SliceContext {
     H264Ref ref_list[2][48];        /**< 0..15: frame refs, 16..47: mbaff field refs.
                                          *   Reordered version of default_ref_list
                                          *   according to picture reordering in slice header */
-    int ref2frm[MAX_SLICES][2][64];     ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
 
     const uint8_t *intra_pcm_ptr;
     int16_t *dc_val_base;
@@ -494,6 +493,11 @@ typedef struct H264Context {
     int context_initialized;
     int flags;
     int workaround_bugs;
+    /* Set when slice threading is used and at least one slice uses deblocking
+     * mode 1 (i.e. across slice boundaries). Then we disable the loop filter
+     * during normal MB decoding and execute it serially at the end.
+     */
+    int postpone_filter;
 
     int8_t(*intra4x4_pred_mode);
     H264PredContext hpc;
@@ -685,6 +689,7 @@ typedef struct H264Context {
     AVBufferPool *mb_type_pool;
     AVBufferPool *motion_val_pool;
     AVBufferPool *ref_index_pool;
+    int ref2frm[MAX_SLICES][2][64];     ///< reference to frame number lists, used in the loop filter, the first 2 are for -2,-1
 } H264Context;
 
 extern const uint16_t ff_h264_mb_sizes[4];
diff --git a/libavcodec/h264_slice.c b/libavcodec/h264_slice.c
index 0e13fd9..8d8e939 100644
--- a/libavcodec/h264_slice.c
+++ b/libavcodec/h264_slice.c
@@ -551,6 +551,8 @@ static int h264_frame_start(H264Context *h)
 
     h->next_output_pic = NULL;
 
+    h->postpone_filter = 0;
+
     assert(h->cur_pic_ptr->long_ref == 0);
 
     return 0;
@@ -1040,7 +1042,7 @@ static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
  *
  * @param h h264context
  *
- * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
+ * @return 0 if okay, <0 if an error occurred
  */
 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
 {
@@ -1742,20 +1744,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
              * Do not bother to deblock across slices. */
             sl->deblocking_filter = 2;
         } else {
-            h->max_contexts = 1;
-            if (!h->single_decode_warning) {
-                av_log(h->avctx, AV_LOG_INFO,
-                       "Cannot parallelize 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");
-                h->single_decode_warning = 1;
-            }
-            if (sl != h->slice_ctx) {
-                av_log(h->avctx, AV_LOG_ERROR,
-                       "Deblocking switched inside frame.\n");
-                return SLICE_SINGLETHREAD;
-            }
+            h->postpone_filter = 1;
         }
     }
     sl->qp_thresh = 15 -
@@ -1778,7 +1767,7 @@ int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
 
     for (j = 0; j < 2; j++) {
         int id_list[16];
-        int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
+        int *ref2frm = h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
         for (i = 0; i < 16; i++) {
             id_list[i] = 60;
             if (j < sl->list_count && i < sl->ref_count[j] &&
@@ -1869,7 +1858,7 @@ static av_always_inline void fill_filter_caches_inter(const H264Context *h,
         if (USES_LIST(top_type, list)) {
             const int b_xy  = h->mb2b_xy[top_xy] + 3 * b_stride;
             const int b8_xy = 4 * top_xy + 2;
-            int *ref2frm = sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
+            int *ref2frm = h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][list] + (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[h->cur_pic.ref_index[list][b8_xy + 0]];
@@ -1884,7 +1873,7 @@ static av_always_inline void fill_filter_caches_inter(const H264Context *h,
             if (USES_LIST(left_type[LTOP], list)) {
                 const int b_xy  = h->mb2b_xy[left_xy[LTOP]] + 3;
                 const int b8_xy = 4 * left_xy[LTOP] + 1;
-                int *ref2frm = sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
+                int *ref2frm = h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][list] + (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]);
@@ -1917,7 +1906,7 @@ static av_always_inline void fill_filter_caches_inter(const H264Context *h,
 
     {
         int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
-        int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
+        int *ref2frm = h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
         uint32_t ref01 = (pack16to32(ref2frm[ref[0]], ref2frm[ref[1]]) & 0x00FF00FF) * 0x0101;
         uint32_t ref23 = (pack16to32(ref2frm[ref[2]], ref2frm[ref[3]]) & 0x00FF00FF) * 0x0101;
         AV_WN32A(&ref_cache[0 * 8], ref01);
@@ -2092,6 +2081,9 @@ static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x,
     const int pixel_shift    = h->pixel_shift;
     const int block_h        = 16 >> h->chroma_y_shift;
 
+    if (h->postpone_filter)
+        return;
+
     if (sl->deblocking_filter) {
         for (mb_x = start_x; mb_x < end_x; mb_x++)
             for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
@@ -2215,6 +2207,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
     H264SliceContext *sl = arg;
     const H264Context *h = sl->h264;
     int lf_x_start = sl->mb_x;
+    int orig_deblock = sl->deblocking_filter;
     int ret;
 
     sl->linesize   = h->cur_pic_ptr->f->linesize[0];
@@ -2228,6 +2221,9 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
 
     av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * sl->linesize * ((scan8[15] - scan8[0]) >> 3));
 
+    if (h->postpone_filter)
+        sl->deblocking_filter = 0;
+
     sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
                      avctx->codec_id != AV_CODEC_ID_H264 ||
                      (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
@@ -2290,7 +2286,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
                              sl->mb_y, ER_MB_END);
                 if (sl->mb_x >= lf_x_start)
                     loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
-                return 0;
+                goto finish;
             }
             if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
                 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
@@ -2323,7 +2319,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
                              sl->mb_y, ER_MB_END);
                 if (sl->mb_x > lf_x_start)
                     loop_filter(h, sl, lf_x_start, sl->mb_x);
-                return 0;
+                goto finish;
             }
         }
     } else {
@@ -2380,7 +2376,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
                         er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
                                      sl->mb_x - 1, sl->mb_y, ER_MB_END);
 
-                        return 0;
+                        goto finish;
                     } else {
                         er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
                                      sl->mb_x, sl->mb_y, ER_MB_END);
@@ -2400,7 +2396,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
                     if (sl->mb_x > lf_x_start)
                         loop_filter(h, sl, lf_x_start, sl->mb_x);
 
-                    return 0;
+                    goto finish;
                 } else {
                     er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
                                  sl->mb_y, ER_MB_ERROR);
@@ -2410,6 +2406,10 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg)
             }
         }
     }
+
+finish:
+    sl->deblocking_filter = orig_deblock;
+    return 0;
 }
 
 /**
@@ -2438,6 +2438,7 @@ int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
         int ret;
 
         h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
+        h->postpone_filter = 0;
 
         ret = decode_slice(avctx, &h->slice_ctx[0]);
         h->mb_y = h->slice_ctx[0].mb_y;
@@ -2476,6 +2477,24 @@ int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
             for (i = 1; i < context_count; i++)
                 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;
         }
+
+        if (h->postpone_filter) {
+            h->postpone_filter = 0;
+
+            for (i = 0; i < context_count; i++) {
+                int y_end, x_end;
+
+                sl = &h->slice_ctx[i];
+                y_end = FFMIN(sl->mb_y + 1, h->mb_height);
+                x_end = (sl->mb_y >= h->mb_height) ? h->mb_width : sl->mb_x;
+
+                for (j = sl->resync_mb_y; j < y_end; j += 1 + FIELD_OR_MBAFF_PICTURE(h)) {
+                    sl->mb_y = j;
+                    loop_filter(h, sl, j > sl->resync_mb_y ? 0 : sl->resync_mb_x,
+                                j == y_end - 1 ? x_end : h->mb_width);
+                }
+            }
+        }
     }
 
     return 0;


======================================================================

diff --cc libavcodec/h264_slice.c
index 0e13fd9,240feb9..8d8e939
--- a/libavcodec/h264_slice.c
+++ b/libavcodec/h264_slice.c
@@@ -1869,12 -1589,12 +1858,12 @@@ static av_always_inline void fill_filte
          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 = sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
 -            int (*ref2frm)[64] = h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
++            int *ref2frm = h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][list] + (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]];
 +            ref_cache[1 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 0]];
              ref_cache[2 - 1 * 8] =
 -            ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
 +            ref_cache[3 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 1]];
          } else {
              AV_ZERO128(mv_dst - 1 * 8);
              AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
@@@ -1884,7 -1604,7 +1873,7 @@@
              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 = sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
 -                int (*ref2frm)[64] = h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
++                int *ref2frm = h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][list] + (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]);
@@@ -1917,9 -1637,9 +1906,9 @@@
  
      {
          int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
-         int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
 -        int (*ref2frm)[64] = h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
 -        uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
 -        uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
++        int *ref2frm = h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
 +        uint32_t ref01 = (pack16to32(ref2frm[ref[0]], ref2frm[ref[1]]) & 0x00FF00FF) * 0x0101;
 +        uint32_t ref23 = (pack16to32(ref2frm[ref[2]], ref2frm[ref[3]]) & 0x00FF00FF) * 0x0101;
          AV_WN32A(&ref_cache[0 * 8], ref01);
          AV_WN32A(&ref_cache[1 * 8], ref01);
          AV_WN32A(&ref_cache[2 * 8], ref23);
@@@ -2226,8 -1951,9 +2219,11 @@@ static int decode_slice(struct AVCodecC
  
      sl->mb_skip_run = -1;
  
 +    av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * sl->linesize * ((scan8[15] - scan8[0]) >> 3));
 +
+     if (h->postpone_filter)
+         sl->deblocking_filter = 0;
+ 
      sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
                       avctx->codec_id != AV_CODEC_ID_H264 ||
                       (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
@@@ -2290,13 -2003,11 +2286,13 @@@
                               sl->mb_y, ER_MB_END);
                  if (sl->mb_x >= lf_x_start)
                      loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
-                 return 0;
+                 goto finish;
              }
 -            if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
 +            if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
 +                av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
 +            if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
                  av_log(h->avctx, AV_LOG_ERROR,
 -                       "error while decoding MB %d %d, bytestream %td\n",
 +                       "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
                         sl->mb_x, sl->mb_y,
                         sl->cabac.bytestream_end - sl->cabac.bytestream);
                  er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
@@@ -2380,10 -2088,10 +2376,10 @@@
                          er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
                                       sl->mb_x - 1, sl->mb_y, ER_MB_END);
  
-                         return 0;
+                         goto finish;
                      } else {
                          er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
 -                                     sl->mb_x - 1, sl->mb_y, ER_MB_END);
 +                                     sl->mb_x, sl->mb_y, ER_MB_END);
  
                          return AVERROR_INVALIDDATA;
                      }
@@@ -2472,10 -2174,26 +2473,28 @@@ int ff_h264_execute_decode_slices(H264C
          /* pull back stuff from slices to master context */
          sl                   = &h->slice_ctx[context_count - 1];
          h->mb_y              = sl->mb_y;
 -        for (i = 1; i < context_count; i++)
 -            h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;
 +        if (CONFIG_ERROR_RESILIENCE) {
 +            for (i = 1; i < context_count; i++)
 +                h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;
 +        }
+ 
+         if (h->postpone_filter) {
+             h->postpone_filter = 0;
+ 
+             for (i = 0; i < context_count; i++) {
+                 int y_end, x_end;
+ 
+                 sl = &h->slice_ctx[i];
+                 y_end = FFMIN(sl->mb_y + 1, h->mb_height);
+                 x_end = (sl->mb_y >= h->mb_height) ? h->mb_width : sl->mb_x;
+ 
+                 for (j = sl->resync_mb_y; j < y_end; j += 1 + FIELD_OR_MBAFF_PICTURE(h)) {
+                     sl->mb_y = j;
+                     loop_filter(h, sl, j > sl->resync_mb_y ? 0 : sl->resync_mb_x,
+                                 j == y_end - 1 ? x_end : h->mb_width);
+                 }
+             }
+         }
      }
  
      return 0;



More information about the ffmpeg-cvslog mailing list