vp8: make mv_min/max thread-local if using partition threading.

Fixes tsan warnings like this in fate-vp8-test-vector-007:

WARNING: ThreadSanitizer: data race (pid=65909)
  Write of size 4 at 0x7d8c0000e088 by thread T1:
    #0 vp8_decode_mb_row_sliced vp8.c:2519 (ffmpeg:x86_64+0x100995ede)
[..]
  Previous write of size 4 at 0x7d8c0000e088 by thread T2:
    #0 vp8_decode_mb_row_sliced vp8.c:2519 (ffmpeg:x86_64+0x100995ede)

(cherry picked from commit fed92adbb3fc6cbf735e3df9a2f7d0a2917fcfbd)
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

2 files changed, 40 insertions, 32 deletions

diff --git a/libavcodec/vp8.c b/libavcodec/vp8.c
index 9bc1d95220..fe7aa23491 100644
--- a/libavcodec/vp8.c
+++ b/libavcodec/vp8.c
@@ -772,7 +772,7 @@ static int vp8_decode_frame_header(VP8Context *s, const uint8_t *buf, int buf_si
 }
 
 static av_always_inline
-void clamp_mv(VP8Context *s, VP56mv *dst, const VP56mv *src)
+void clamp_mv(VP8mvbounds *s, VP56mv *dst, const VP56mv *src)
 {
     dst->x = av_clip(src->x, av_clip(s->mv_min.x, INT16_MIN, INT16_MAX),
                              av_clip(s->mv_max.x, INT16_MIN, INT16_MAX));
@@ -1031,7 +1031,7 @@ void vp7_decode_mvs(VP8Context *s, VP8Macroblock *mb,
 }
 
 static av_always_inline
-void vp8_decode_mvs(VP8Context *s, VP8Macroblock *mb,
+void vp8_decode_mvs(VP8Context *s, VP8mvbounds *mv_bounds, VP8Macroblock *mb,
                     int mb_x, int mb_y, int layout)
 {
     VP8Macroblock *mb_edge[3] = { 0      /* top */,
@@ -1102,7 +1102,7 @@ void vp8_decode_mvs(VP8Context *s, VP8Macroblock *mb,
         if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_NEAREST]][1])) {
             if (vp56_rac_get_prob_branchy(c, vp8_mode_contexts[cnt[CNT_NEAR]][2])) {
                 /* Choose the best mv out of 0,0 and the nearest mv */
-                clamp_mv(s, &mb->mv, &near_mv[CNT_ZERO + (cnt[CNT_NEAREST] >= cnt[CNT_ZERO])]);
+                clamp_mv(mv_bounds, &mb->mv, &near_mv[CNT_ZERO + (cnt[CNT_NEAREST] >= cnt[CNT_ZERO])]);
                 cnt[CNT_SPLITMV] = ((mb_edge[VP8_EDGE_LEFT]->mode    == VP8_MVMODE_SPLIT) +
                                     (mb_edge[VP8_EDGE_TOP]->mode     == VP8_MVMODE_SPLIT)) * 2 +
                                     (mb_edge[VP8_EDGE_TOPLEFT]->mode == VP8_MVMODE_SPLIT);
@@ -1116,11 +1116,11 @@ void vp8_decode_mvs(VP8Context *s, VP8Macroblock *mb,
                     mb->bmv[0] = mb->mv;
                 }
             } else {
-                clamp_mv(s, &mb->mv, &near_mv[CNT_NEAR]);
+                clamp_mv(mv_bounds, &mb->mv, &near_mv[CNT_NEAR]);
                 mb->bmv[0] = mb->mv;
             }
         } else {
-            clamp_mv(s, &mb->mv, &near_mv[CNT_NEAREST]);
+            clamp_mv(mv_bounds, &mb->mv, &near_mv[CNT_NEAREST]);
             mb->bmv[0] = mb->mv;
         }
     } else {
@@ -1166,7 +1166,8 @@ void decode_intra4x4_modes(VP8Context *s, VP56RangeCoder *c, VP8Macroblock *mb,
 }
 
 static av_always_inline
-void decode_mb_mode(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y,
+void decode_mb_mode(VP8Context *s, VP8mvbounds *mv_bounds,
+                    VP8Macroblock *mb, int mb_x, int mb_y,
                     uint8_t *segment, uint8_t *ref, int layout, int is_vp7)
 {
     VP56RangeCoder *c = &s->c;
@@ -1230,7 +1231,7 @@ void decode_mb_mode(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y,
         if (is_vp7)
             vp7_decode_mvs(s, mb, mb_x, mb_y, layout);
         else
-            vp8_decode_mvs(s, mb, mb_x, mb_y, layout);
+            vp8_decode_mvs(s, mv_bounds, mb, mb_x, mb_y, layout);
     } else {
         // intra MB, 16.1
         mb->mode = vp8_rac_get_tree(c, vp8_pred16x16_tree_inter, s->prob->pred16x16);
@@ -2205,8 +2206,8 @@ void vp78_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *curframe,
     VP8Context *s = avctx->priv_data;
     int mb_x, mb_y;
 
-    s->mv_min.y = -MARGIN;
-    s->mv_max.y = ((s->mb_height - 1) << 6) + MARGIN;
+    s->mv_bounds.mv_min.y = -MARGIN;
+    s->mv_bounds.mv_max.y = ((s->mb_height - 1) << 6) + MARGIN;
     for (mb_y = 0; mb_y < s->mb_height; mb_y++) {
         VP8Macroblock *mb = s->macroblocks_base +
                             ((s->mb_width + 1) * (mb_y + 1) + 1);
@@ -2214,20 +2215,20 @@ void vp78_decode_mv_mb_modes(AVCodecContext *avctx, VP8Frame *curframe,
 
         AV_WN32A(s->intra4x4_pred_mode_left, DC_PRED * 0x01010101);
 
-        s->mv_min.x = -MARGIN;
-        s->mv_max.x = ((s->mb_width - 1) << 6) + MARGIN;
+        s->mv_bounds.mv_min.x = -MARGIN;
+        s->mv_bounds.mv_max.x = ((s->mb_width - 1) << 6) + MARGIN;
         for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) {
             if (mb_y == 0)
                 AV_WN32A((mb - s->mb_width - 1)->intra4x4_pred_mode_top,
                          DC_PRED * 0x01010101);
-            decode_mb_mode(s, mb, mb_x, mb_y, curframe->seg_map->data + mb_xy,
+            decode_mb_mode(s, &s->mv_bounds, mb, mb_x, mb_y, curframe->seg_map->data + mb_xy,
                            prev_frame && prev_frame->seg_map ?
                            prev_frame->seg_map->data + mb_xy : NULL, 1, is_vp7);
-            s->mv_min.x -= 64;
-            s->mv_max.x -= 64;
+            s->mv_bounds.mv_min.x -= 64;
+            s->mv_bounds.mv_max.x -= 64;
         }
-        s->mv_min.y -= 64;
-        s->mv_max.y -= 64;
+        s->mv_bounds.mv_min.y -= 64;
+        s->mv_bounds.mv_max.y -= 64;
     }
 }
 
@@ -2325,8 +2326,8 @@ static av_always_inline int decode_mb_row_no_filter(AVCodecContext *avctx, void
     if (!is_vp7 || mb_y == 0)
         memset(td->left_nnz, 0, sizeof(td->left_nnz));
 
-    s->mv_min.x = -MARGIN;
-    s->mv_max.x = ((s->mb_width - 1) << 6) + MARGIN;
+    td->mv_bounds.mv_min.x = -MARGIN;
+    td->mv_bounds.mv_max.x = ((s->mb_width - 1) << 6) + MARGIN;
 
     for (mb_x = 0; mb_x < s->mb_width; mb_x++, mb_xy++, mb++) {
         if (c->end <= c->buffer && c->bits >= 0)
@@ -2350,7 +2351,7 @@ static av_always_inline int decode_mb_row_no_filter(AVCodecContext *avctx, void
                          dst[2] - dst[1], 2);
 
         if (!s->mb_layout)
-            decode_mb_mode(s, mb, mb_x, mb_y, curframe->seg_map->data + mb_xy,
+            decode_mb_mode(s, &td->mv_bounds, mb, mb_x, mb_y, curframe->seg_map->data + mb_xy,
                            prev_frame && prev_frame->seg_map ?
                            prev_frame->seg_map->data + mb_xy : NULL, 0, is_vp7);
 
@@ -2397,8 +2398,8 @@ static av_always_inline int decode_mb_row_no_filter(AVCodecContext *avctx, void
         dst[0]      += 16;
         dst[1]      += 8;
         dst[2]      += 8;
-        s->mv_min.x -= 64;
-        s->mv_max.x -= 64;
+        td->mv_bounds.mv_min.x -= 64;
+        td->mv_bounds.mv_max.x -= 64;
 
         if (mb_x == s->mb_width + 1) {
             update_pos(td, mb_y, s->mb_width + 3);
@@ -2504,6 +2505,8 @@ int vp78_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata, int jobnr,
     int ret;
 
     td->thread_nr = threadnr;
+    td->mv_bounds.mv_min.y   = -MARGIN - 64 * threadnr;
+    td->mv_bounds.mv_max.y   = ((s->mb_height - 1) << 6) + MARGIN - 64 * threadnr;
     for (mb_y = jobnr; mb_y < s->mb_height; mb_y += num_jobs) {
         atomic_store(&td->thread_mb_pos, mb_y << 16);
         ret = s->decode_mb_row_no_filter(avctx, tdata, jobnr, threadnr);
@@ -2515,8 +2518,8 @@ int vp78_decode_mb_row_sliced(AVCodecContext *avctx, void *tdata, int jobnr,
             s->filter_mb_row(avctx, tdata, jobnr, threadnr);
         update_pos(td, mb_y, INT_MAX & 0xFFFF);
 
-        s->mv_min.y -= 64;
-        s->mv_max.y -= 64;
+        td->mv_bounds.mv_min.y -= 64 * num_jobs;
+        td->mv_bounds.mv_max.y -= 64 * num_jobs;
 
         if (avctx->active_thread_type == FF_THREAD_FRAME)
             ff_thread_report_progress(&curframe->tf, mb_y, 0);
@@ -2662,8 +2665,8 @@ int vp78_decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
     s->num_jobs   = num_jobs;
     s->curframe   = curframe;
     s->prev_frame = prev_frame;
-    s->mv_min.y   = -MARGIN;
-    s->mv_max.y   = ((s->mb_height - 1) << 6) + MARGIN;
+    s->mv_bounds.mv_min.y   = -MARGIN;
+    s->mv_bounds.mv_max.y   = ((s->mb_height - 1) << 6) + MARGIN;
     for (i = 0; i < MAX_THREADS; i++) {
         VP8ThreadData *td = &s->thread_data[i];
         atomic_init(&td->thread_mb_pos, 0);
diff --git a/libavcodec/vp8.h b/libavcodec/vp8.h
index d7e7680276..8263997e3f 100644
--- a/libavcodec/vp8.h
+++ b/libavcodec/vp8.h
@@ -93,6 +93,16 @@ typedef struct VP8Macroblock {
     VP56mv bmv[16];
 } VP8Macroblock;
 
+typedef struct VP8intmv {
+    int x;
+    int y;
+} VP8intmv;
+
+typedef struct VP8mvbounds {
+    VP8intmv mv_min;
+    VP8intmv mv_max;
+} VP8mvbounds;
+
 typedef struct VP8ThreadData {
     DECLARE_ALIGNED(16, int16_t, block)[6][4][16];
     DECLARE_ALIGNED(16, int16_t, block_dc)[16];
@@ -122,6 +132,7 @@ typedef struct VP8ThreadData {
 #define EDGE_EMU_LINESIZE 32
     DECLARE_ALIGNED(16, uint8_t, edge_emu_buffer)[21 * EDGE_EMU_LINESIZE];
     VP8FilterStrength *filter_strength;
+    VP8mvbounds mv_bounds;
 } VP8ThreadData;
 
 typedef struct VP8Frame {
@@ -129,11 +140,6 @@ typedef struct VP8Frame {
     AVBufferRef *seg_map;
 } VP8Frame;
 
-typedef struct VP8intmv {
-    int x;
-    int y;
-} VP8intmv;
-
 #define MAX_THREADS 8
 typedef struct VP8Context {
     VP8ThreadData *thread_data;
@@ -152,8 +158,7 @@ typedef struct VP8Context {
     uint8_t deblock_filter;
     uint8_t mbskip_enabled;
     uint8_t profile;
-    VP8intmv mv_min;
-    VP8intmv mv_max;
+    VP8mvbounds mv_bounds;
 
     int8_t sign_bias[4]; ///< one state [0, 1] per ref frame type
     int ref_count[3];