vp9: use correct chroma subsampling for profile 1 inter block recon.

2 files changed, 271 insertions, 45 deletions

diff --git a/libavcodec/vp9.c b/libavcodec/vp9.c
index 2cbd1e004f..64c95fae38 100644
--- a/libavcodec/vp9.c
+++ b/libavcodec/vp9.c
@@ -2609,7 +2609,7 @@ static av_always_inline void mc_luma_scaled(VP9Context *s, vp9_scaled_mc_func sm
                                             int bw, int bh, int w, int h,
                                             const uint16_t *scale, const uint8_t *step)
 {
-#define scale_mv(n, dim) (((int64_t)n * scale[dim]) >> 14)
+#define scale_mv(n, dim) (((int64_t)(n) * scale[dim]) >> 14)
     // BUG libvpx seems to scale the two components separately. This introduces
     // rounding errors but we have to reproduce them to be exactly compatible
     // with the output from libvpx...
@@ -2653,8 +2653,8 @@ static av_always_inline void mc_chroma_scaled(VP9Context *s, vp9_scaled_mc_func
                                               const uint16_t *scale, const uint8_t *step)
 {
     // BUG https://code.google.com/p/webm/issues/detail?id=820
-    int mx = scale_mv(mv->x, 0) + (scale_mv(x * 16, 0) & ~15) + (scale_mv(x * 32, 0) & 15);
-    int my = scale_mv(mv->y, 1) + (scale_mv(y * 16, 1) & ~15) + (scale_mv(y * 32, 1) & 15);
+    int mx = scale_mv(mv->x << !s->ss_h, 0) + (scale_mv(x * 16, 0) & ~15) + (scale_mv(x * 32, 0) & 15);
+    int my = scale_mv(mv->y << !s->ss_v, 1) + (scale_mv(y * 16, 1) & ~15) + (scale_mv(y * 32, 1) & 15);
 #undef scale_mv
     int refbw_m1, refbh_m1;
     int th;
@@ -2670,7 +2670,7 @@ static av_always_inline void mc_chroma_scaled(VP9Context *s, vp9_scaled_mc_func
     // FIXME bilinear filter only needs 0/1 pixels, not 3/4
     // we use +7 because the last 7 pixels of each sbrow can be changed in
     // the longest loopfilter of the next sbrow
-    th = (y + refbh_m1 + 4 + 7) >> 5;
+    th = (y + refbh_m1 + 4 + 7) >> (6 - s->ss_v);
     ff_thread_await_progress(ref_frame, FFMAX(th, 0), 0);
     if (x < 3 || y < 3 || x + 4 >= w - refbw_m1 || y + 4 >= h - refbh_m1) {
         s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
@@ -2748,7 +2748,7 @@ static av_always_inline void mc_chroma_unscaled(VP9Context *s, vp9_mc_func (*mc)
                                                 ptrdiff_t y, ptrdiff_t x, const VP56mv *mv,
                                                 int bw, int bh, int w, int h)
 {
-    int mx = mv->x, my = mv->y, th;
+    int mx = mv->x << !s->ss_h, my = mv->y << !s->ss_v, th;
 
     y += my >> 4;
     x += mx >> 4;
@@ -2759,7 +2759,7 @@ static av_always_inline void mc_chroma_unscaled(VP9Context *s, vp9_mc_func (*mc)
     // FIXME bilinear filter only needs 0/1 pixels, not 3/4
     // we use +7 because the last 7 pixels of each sbrow can be changed in
     // the longest loopfilter of the next sbrow
-    th = (y + bh + 4 * !!my + 7) >> 5;
+    th = (y + bh + 4 * !!my + 7) >> (6 - s->ss_v);
     ff_thread_await_progress(ref_frame, FFMAX(th, 0), 0);
     if (x < !!mx * 3 || y < !!my * 3 ||
         x + !!mx * 4 > w - bw || y + !!my * 4 > h - bh) {
@@ -2833,8 +2833,8 @@ static void inter_recon(AVCodecContext *ctx)
         }
 
         // uv itxfm add
-        end_x >>= 1;
-        end_y >>= 1;
+        end_x >>= s->ss_h;
+        end_y >>= s->ss_v;
         step = 1 << (b->uvtx * 2);
         for (p = 0; p < 2; p++) {
             dst = s->dst[p + 1];
diff --git a/libavcodec/vp9_mc_template.c b/libavcodec/vp9_mc_template.c
index c6ae432e26..11b9500384 100644
--- a/libavcodec/vp9_mc_template.c
+++ b/libavcodec/vp9_mc_template.c
@@ -21,6 +21,12 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
+#define ROUNDED_DIV_MVx2(a, b) \
+    (VP56mv) { .x = ROUNDED_DIV(a.x + b.x, 2), .y = ROUNDED_DIV(a.y + b.y, 2) }
+#define ROUNDED_DIV_MVx4(a, b, c, d) \
+    (VP56mv) { .x = ROUNDED_DIV(a.x + b.x + c.x + d.x, 4), \
+               .y = ROUNDED_DIV(a.y + b.y + c.y + d.y, 4) }
+
 static void FN(inter_pred)(AVCodecContext *ctx)
 {
     static const uint8_t bwlog_tab[2][N_BS_SIZES] = {
@@ -44,6 +50,8 @@ static void FN(inter_pred)(AVCodecContext *ctx)
 
     // y inter pred
     if (b->bs > BS_8x8) {
+        VP56mv uvmv;
+
         if (b->bs == BS_8x4) {
             mc_luma_dir(s, mc[3][b->filter][0], s->dst[0], ls_y,
                         ref1->data[0], ref1->linesize[0], tref1,
@@ -52,6 +60,30 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                         s->dst[0] + 4 * ls_y, ls_y,
                         ref1->data[0], ref1->linesize[0], tref1,
                         (row << 3) + 4, col << 3, &b->mv[2][0], 8, 4, w1, h1, 0);
+            w1 = (w1 + s->ss_h) >> s->ss_h;
+            if (s->ss_v) {
+                h1 = (h1 + 1) >> 1;
+                uvmv = ROUNDED_DIV_MVx2(b->mv[0][0], b->mv[2][0]);
+                mc_chroma_dir(s, mc[3 + s->ss_h][b->filter][0],
+                              s->dst[1], s->dst[2], ls_uv,
+                              ref1->data[1], ref1->linesize[1],
+                              ref1->data[2], ref1->linesize[2], tref1,
+                              row << 2, col << (3 - s->ss_h),
+                              &uvmv, 8 >> s->ss_h, 4, w1, h1, 0);
+            } else {
+                mc_chroma_dir(s, mc[3 + s->ss_h][b->filter][0],
+                              s->dst[1], s->dst[2], ls_uv,
+                              ref1->data[1], ref1->linesize[1],
+                              ref1->data[2], ref1->linesize[2], tref1,
+                              row << 3, col << (3 - s->ss_h),
+                              &b->mv[0][0], 8 >> s->ss_h, 4, w1, h1, 0);
+                mc_chroma_dir(s, mc[3 + s->ss_h][b->filter][0],
+                              s->dst[1] + 4 * ls_uv, s->dst[2] + 4 * ls_uv, ls_uv,
+                              ref1->data[1], ref1->linesize[1],
+                              ref1->data[2], ref1->linesize[2], tref1,
+                              (row << 3) + 4, col << (3 - s->ss_h),
+                              &b->mv[2][0], 8 >> s->ss_h, 4, w1, h1, 0);
+            }
 
             if (b->comp) {
                 mc_luma_dir(s, mc[3][b->filter][1], s->dst[0], ls_y,
@@ -61,6 +93,30 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                             s->dst[0] + 4 * ls_y, ls_y,
                             ref2->data[0], ref2->linesize[0], tref2,
                             (row << 3) + 4, col << 3, &b->mv[2][1], 8, 4, w2, h2, 1);
+                w2 = (w2 + s->ss_h) >> s->ss_h;
+                if (s->ss_v) {
+                    h2 = (h2 + 1) >> 1;
+                    uvmv = ROUNDED_DIV_MVx2(b->mv[0][1], b->mv[2][1]);
+                    mc_chroma_dir(s, mc[3 + s->ss_h][b->filter][1],
+                                  s->dst[1], s->dst[2], ls_uv,
+                                  ref2->data[1], ref2->linesize[1],
+                                  ref2->data[2], ref2->linesize[2], tref2,
+                                  row << 2, col << (3 - s->ss_h),
+                                  &uvmv, 8 >> s->ss_h, 4, w2, h2, 1);
+                } else {
+                    mc_chroma_dir(s, mc[3 + s->ss_h][b->filter][1],
+                                  s->dst[1], s->dst[2], ls_uv,
+                                  ref2->data[1], ref2->linesize[1],
+                                  ref2->data[2], ref2->linesize[2], tref2,
+                                  row << 3, col << (3 - s->ss_h),
+                                  &b->mv[0][1], 8 >> s->ss_h, 4, w2, h2, 1);
+                    mc_chroma_dir(s, mc[3 + s->ss_h][b->filter][1],
+                                  s->dst[1] + 4 * ls_uv, s->dst[2] + 4 * ls_uv, ls_uv,
+                                  ref2->data[1], ref2->linesize[1],
+                                  ref2->data[2], ref2->linesize[2], tref2,
+                                  (row << 3) + 4, col << (3 - s->ss_h),
+                                  &b->mv[2][1], 8 >> s->ss_h, 4, w2, h2, 1);
+                }
             }
         } else if (b->bs == BS_4x8) {
             mc_luma_dir(s, mc[4][b->filter][0], s->dst[0], ls_y,
@@ -69,6 +125,30 @@ static void FN(inter_pred)(AVCodecContext *ctx)
             mc_luma_dir(s, mc[4][b->filter][0], s->dst[0] + 4, ls_y,
                         ref1->data[0], ref1->linesize[0], tref1,
                         row << 3, (col << 3) + 4, &b->mv[1][0], 4, 8, w1, h1, 0);
+            h1 = (h1 + s->ss_v) >> s->ss_v;
+            if (s->ss_h) {
+                w1 = (w1 + 1) >> 1;
+                uvmv = ROUNDED_DIV_MVx2(b->mv[0][0], b->mv[1][0]);
+                mc_chroma_dir(s, mc[4][b->filter][0],
+                              s->dst[1], s->dst[2], ls_uv,
+                              ref1->data[1], ref1->linesize[1],
+                              ref1->data[2], ref1->linesize[2], tref1,
+                              row << (3 - s->ss_v), col << 2,
+                              &uvmv, 4, 8 >> s->ss_v, w1, h1, 0);
+            } else {
+                mc_chroma_dir(s, mc[4][b->filter][0],
+                              s->dst[1], s->dst[2], ls_uv,
+                              ref1->data[1], ref1->linesize[1],
+                              ref1->data[2], ref1->linesize[2], tref1,
+                              row << (3 - s->ss_v), col << 3,
+                              &b->mv[0][0], 4, 8 >> s->ss_v, w1, h1, 0);
+                mc_chroma_dir(s, mc[4][b->filter][0],
+                              s->dst[1] + 4, s->dst[2] + 4, ls_uv,
+                              ref1->data[1], ref1->linesize[1],
+                              ref1->data[2], ref1->linesize[2], tref1,
+                              row << (3 - s->ss_v), (col << 3) + 4,
+                              &b->mv[1][0], 4, 8 >> s->ss_v, w1, h1, 0);
+            }
 
             if (b->comp) {
                 mc_luma_dir(s, mc[4][b->filter][1], s->dst[0], ls_y,
@@ -77,6 +157,30 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                 mc_luma_dir(s, mc[4][b->filter][1], s->dst[0] + 4, ls_y,
                             ref2->data[0], ref2->linesize[0], tref2,
                             row << 3, (col << 3) + 4, &b->mv[1][1], 4, 8, w2, h2, 1);
+                h2 = (h2 + s->ss_v) >> s->ss_v;
+                if (s->ss_h) {
+                    w2 = (w2 + 1) >> 1;
+                    uvmv = ROUNDED_DIV_MVx2(b->mv[0][1], b->mv[1][1]);
+                    mc_chroma_dir(s, mc[4][b->filter][1],
+                                  s->dst[1], s->dst[2], ls_uv,
+                                  ref2->data[1], ref2->linesize[1],
+                                  ref2->data[2], ref2->linesize[2], tref2,
+                                  row << (3 - s->ss_v), col << 2,
+                                  &uvmv, 4, 8 >> s->ss_v, w2, h2, 1);
+                } else {
+                    mc_chroma_dir(s, mc[4][b->filter][1],
+                                  s->dst[1], s->dst[2], ls_uv,
+                                  ref2->data[1], ref2->linesize[1],
+                                  ref2->data[2], ref2->linesize[2], tref2,
+                                  row << (3 - s->ss_v), col << 3,
+                                  &b->mv[0][1], 4, 8 >> s->ss_v, w2, h2, 1);
+                    mc_chroma_dir(s, mc[4][b->filter][1],
+                                  s->dst[1] + 4, s->dst[2] + 4, ls_uv,
+                                  ref2->data[1], ref2->linesize[1],
+                                  ref2->data[2], ref2->linesize[2], tref2,
+                                  row << (3 - s->ss_v), (col << 3) + 4,
+                                  &b->mv[1][1], 4, 8 >> s->ss_v, w2, h2, 1);
+                }
             }
         } else {
             av_assert2(b->bs == BS_4x4);
@@ -97,6 +201,78 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                         s->dst[0] + 4 * ls_y + 4, ls_y,
                         ref1->data[0], ref1->linesize[0], tref1,
                         (row << 3) + 4, (col << 3) + 4, &b->mv[3][0], 4, 4, w1, h1, 0);
+            if (s->ss_v) {
+                h1 = (h1 + 1) >> 1;
+                if (s->ss_h) {
+                    w1 = (w1 + 1) >> 1;
+                    uvmv = ROUNDED_DIV_MVx4(b->mv[0][0], b->mv[1][0],
+                                            b->mv[2][0], b->mv[3][0]);
+                    mc_chroma_dir(s, mc[4][b->filter][0],
+                                  s->dst[1], s->dst[2], ls_uv,
+                                  ref1->data[1], ref1->linesize[1],
+                                  ref1->data[2], ref1->linesize[2], tref1,
+                                  row << 2, col << 2,
+                                  &uvmv, 4, 4, w1, h1, 0);
+                } else {
+                    uvmv = ROUNDED_DIV_MVx2(b->mv[0][0], b->mv[2][0]);
+                    mc_chroma_dir(s, mc[4][b->filter][0],
+                                  s->dst[1], s->dst[2], ls_uv,
+                                  ref1->data[1], ref1->linesize[1],
+                                  ref1->data[2], ref1->linesize[2], tref1,
+                                  row << 2, col << 3,
+                                  &uvmv, 4, 4, w1, h1, 0);
+                    uvmv = ROUNDED_DIV_MVx2(b->mv[1][0], b->mv[3][0]);
+                    mc_chroma_dir(s, mc[4][b->filter][0],
+                                  s->dst[1] + 4, s->dst[2] + 4, ls_uv,
+                                  ref1->data[1], ref1->linesize[1],
+                                  ref1->data[2], ref1->linesize[2], tref1,
+                                  row << 2, (col << 3) + 4,
+                                  &uvmv, 4, 4, w1, h1, 0);
+                }
+            } else {
+                if (s->ss_h) {
+                    w1 = (w1 + 1) >> 1;
+                    uvmv = ROUNDED_DIV_MVx2(b->mv[0][0], b->mv[1][0]);
+                    mc_chroma_dir(s, mc[4][b->filter][0],
+                                  s->dst[1], s->dst[2], ls_uv,
+                                  ref1->data[1], ref1->linesize[1],
+                                  ref1->data[2], ref1->linesize[2], tref1,
+                                  row << 3, col << 2,
+                                  &uvmv, 4, 4, w1, h1, 0);
+                    uvmv = ROUNDED_DIV_MVx2(b->mv[2][0], b->mv[3][0]);
+                    mc_chroma_dir(s, mc[4][b->filter][0],
+                                  s->dst[1] + 4 * ls_uv, s->dst[2] + 4 * ls_uv, ls_uv,
+                                  ref1->data[1], ref1->linesize[1],
+                                  ref1->data[2], ref1->linesize[2], tref1,
+                                  (row << 3) + 4, col << 2,
+                                  &uvmv, 4, 4, w1, h1, 0);
+                } else {
+                    mc_chroma_dir(s, mc[4][b->filter][0],
+                                  s->dst[1], s->dst[2], ls_uv,
+                                  ref1->data[1], ref1->linesize[1],
+                                  ref1->data[2], ref1->linesize[2], tref1,
+                                  row << 3, col << 3,
+                                  &b->mv[0][0], 4, 4, w1, h1, 0);
+                    mc_chroma_dir(s, mc[4][b->filter][0],
+                                  s->dst[1] + 4, s->dst[2] + 4, ls_uv,
+                                  ref1->data[1], ref1->linesize[1],
+                                  ref1->data[2], ref1->linesize[2], tref1,
+                                  row << 3, (col << 3) + 4,
+                                  &b->mv[1][0], 4, 4, w1, h1, 0);
+                    mc_chroma_dir(s, mc[4][b->filter][0],
+                                  s->dst[1] + 4 * ls_uv, s->dst[2] + 4 * ls_uv, ls_uv,
+                                  ref1->data[1], ref1->linesize[1],
+                                  ref1->data[2], ref1->linesize[2], tref1,
+                                  (row << 3) + 4, col << 3,
+                                  &b->mv[2][0], 4, 4, w1, h1, 0);
+                    mc_chroma_dir(s, mc[4][b->filter][0],
+                                  s->dst[1] + 4 * ls_uv + 4, s->dst[2] + 4 * ls_uv + 4, ls_uv,
+                                  ref1->data[1], ref1->linesize[1],
+                                  ref1->data[2], ref1->linesize[2], tref1,
+                                  (row << 3) + 4, (col << 3) + 4,
+                                  &b->mv[3][0], 4, 4, w1, h1, 0);
+                }
+            }
 
             if (b->comp) {
                 mc_luma_dir(s, mc[4][b->filter][1], s->dst[0], ls_y,
@@ -113,59 +289,109 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                             s->dst[0] + 4 * ls_y + 4, ls_y,
                             ref2->data[0], ref2->linesize[0], tref2,
                             (row << 3) + 4, (col << 3) + 4, &b->mv[3][1], 4, 4, w2, h2, 1);
+                if (s->ss_v) {
+                    h2 = (h2 + 1) >> 1;
+                    if (s->ss_h) {
+                        w2 = (w2 + 1) >> 1;
+                        uvmv = ROUNDED_DIV_MVx4(b->mv[0][1], b->mv[1][1],
+                                                b->mv[2][1], b->mv[3][1]);
+                        mc_chroma_dir(s, mc[4][b->filter][1],
+                                      s->dst[1], s->dst[2], ls_uv,
+                                      ref2->data[1], ref2->linesize[1],
+                                      ref2->data[2], ref2->linesize[2], tref2,
+                                      row << 2, col << 2,
+                                      &uvmv, 4, 4, w2, h2, 1);
+                    } else {
+                        uvmv = ROUNDED_DIV_MVx2(b->mv[0][1], b->mv[2][1]);
+                        mc_chroma_dir(s, mc[4][b->filter][1],
+                                      s->dst[1], s->dst[2], ls_uv,
+                                      ref2->data[1], ref2->linesize[1],
+                                      ref2->data[2], ref2->linesize[2], tref2,
+                                      row << 2, col << 3,
+                                      &uvmv, 4, 4, w2, h2, 1);
+                        uvmv = ROUNDED_DIV_MVx2(b->mv[1][1], b->mv[3][1]);
+                        mc_chroma_dir(s, mc[4][b->filter][1],
+                                      s->dst[1] + 4, s->dst[2] + 4, ls_uv,
+                                      ref2->data[1], ref2->linesize[1],
+                                      ref2->data[2], ref2->linesize[2], tref2,
+                                      row << 2, (col << 3) + 4,
+                                      &uvmv, 4, 4, w2, h2, 1);
+                    }
+                } else {
+                    if (s->ss_h) {
+                        w2 = (w2 + 1) >> 1;
+                        uvmv = ROUNDED_DIV_MVx2(b->mv[0][1], b->mv[1][1]);
+                        mc_chroma_dir(s, mc[4][b->filter][1],
+                                      s->dst[1], s->dst[2], ls_uv,
+                                      ref2->data[1], ref2->linesize[1],
+                                      ref2->data[2], ref2->linesize[2], tref2,
+                                      row << 3, col << 2,
+                                      &uvmv, 4, 4, w2, h2, 1);
+                        uvmv = ROUNDED_DIV_MVx2(b->mv[2][1], b->mv[3][1]);
+                        mc_chroma_dir(s, mc[4][b->filter][1],
+                                      s->dst[1] + 4 * ls_uv, s->dst[2] + 4 * ls_uv, ls_uv,
+                                      ref2->data[1], ref2->linesize[1],
+                                      ref2->data[2], ref2->linesize[2], tref2,
+                                      (row << 3) + 4, col << 2,
+                                      &uvmv, 4, 4, w2, h2, 1);
+                    } else {
+                        mc_chroma_dir(s, mc[4][b->filter][1],
+                                      s->dst[1], s->dst[2], ls_uv,
+                                      ref2->data[1], ref2->linesize[1],
+                                      ref2->data[2], ref2->linesize[2], tref2,
+                                      row << 3, col << 3,
+                                      &b->mv[0][1], 4, 4, w2, h2, 1);
+                        mc_chroma_dir(s, mc[4][b->filter][1],
+                                      s->dst[1] + 4, s->dst[2] + 4, ls_uv,
+                                      ref2->data[1], ref2->linesize[1],
+                                      ref2->data[2], ref2->linesize[2], tref2,
+                                      row << 3, (col << 3) + 4,
+                                      &b->mv[1][1], 4, 4, w2, h2, 1);
+                        mc_chroma_dir(s, mc[4][b->filter][1],
+                                      s->dst[1] + 4 * ls_uv, s->dst[2] + 4 * ls_uv, ls_uv,
+                                      ref2->data[1], ref2->linesize[1],
+                                      ref2->data[2], ref2->linesize[2], tref2,
+                                      (row << 3) + 4, col << 3,
+                                      &b->mv[2][1], 4, 4, w2, h2, 1);
+                        mc_chroma_dir(s, mc[4][b->filter][1],
+                                      s->dst[1] + 4 * ls_uv + 4, s->dst[2] + 4 * ls_uv + 4, ls_uv,
+                                      ref2->data[1], ref2->linesize[1],
+                                      ref2->data[2], ref2->linesize[2], tref2,
+                                      (row << 3) + 4, (col << 3) + 4,
+                                      &b->mv[3][1], 4, 4, w2, h2, 1);
+                    }
+                }
             }
         }
     } else {
         int bwl = bwlog_tab[0][b->bs];
         int bw = bwh_tab[0][b->bs][0] * 4, bh = bwh_tab[0][b->bs][1] * 4;
+        int uvbw = bwh_tab[s->ss_h][b->bs][0] * 4, uvbh = bwh_tab[s->ss_v][b->bs][1] * 4;
 
         mc_luma_dir(s, mc[bwl][b->filter][0], s->dst[0], ls_y,
                     ref1->data[0], ref1->linesize[0], tref1,
-                    row << 3, col << 3, &b->mv[0][0],bw, bh, w1, h1, 0);
-
-        if (b->comp)
-            mc_luma_dir(s, mc[bwl][b->filter][1], s->dst[0], ls_y,
-                        ref2->data[0], ref2->linesize[0], tref2,
-                        row << 3, col << 3, &b->mv[0][1], bw, bh, w2, h2, 1);
-    }
-
-    // uv inter pred
-    {
-        int bwl = bwlog_tab[1][b->bs];
-        int bw = bwh_tab[1][b->bs][0] * 4, bh = bwh_tab[1][b->bs][1] * 4;
-        VP56mv mvuv;
-
-        w1 = (w1 + 1) >> 1;
-        h1 = (h1 + 1) >> 1;
-        if (b->comp) {
-            w2 = (w2 + 1) >> 1;
-            h2 = (h2 + 1) >> 1;
-        }
-        if (b->bs > BS_8x8) {
-            mvuv.x = ROUNDED_DIV(b->mv[0][0].x + b->mv[1][0].x + b->mv[2][0].x + b->mv[3][0].x, 4);
-            mvuv.y = ROUNDED_DIV(b->mv[0][0].y + b->mv[1][0].y + b->mv[2][0].y + b->mv[3][0].y, 4);
-        } else {
-            mvuv = b->mv[0][0];
-        }
-
-        mc_chroma_dir(s, mc[bwl][b->filter][0],
+                    row << 3, col << 3, &b->mv[0][0], bw, bh, w1, h1, 0);
+        w1 = (w1 + s->ss_h) >> s->ss_h;
+        h1 = (h1 + s->ss_v) >> s->ss_v;
+        mc_chroma_dir(s, mc[bwl + s->ss_h][b->filter][0],
                       s->dst[1], s->dst[2], ls_uv,
                       ref1->data[1], ref1->linesize[1],
                       ref1->data[2], ref1->linesize[2], tref1,
-                      row << 2, col << 2, &mvuv, bw, bh, w1, h1, 0);
+                      row << (3 - s->ss_v), col << (3 - s->ss_h),
+                      &b->mv[0][0], uvbw, uvbh, w1, h1, 0);
 
         if (b->comp) {
-            if (b->bs > BS_8x8) {
-                mvuv.x = ROUNDED_DIV(b->mv[0][1].x + b->mv[1][1].x + b->mv[2][1].x + b->mv[3][1].x, 4);
-                mvuv.y = ROUNDED_DIV(b->mv[0][1].y + b->mv[1][1].y + b->mv[2][1].y + b->mv[3][1].y, 4);
-            } else {
-                mvuv = b->mv[0][1];
-            }
-            mc_chroma_dir(s, mc[bwl][b->filter][1],
+            mc_luma_dir(s, mc[bwl][b->filter][1], s->dst[0], ls_y,
+                        ref2->data[0], ref2->linesize[0], tref2,
+                        row << 3, col << 3, &b->mv[0][1], bw, bh, w2, h2, 1);
+            w2 = (w2 + s->ss_h) >> s->ss_h;
+            h2 = (h2 + s->ss_v) >> s->ss_v;
+            mc_chroma_dir(s, mc[bwl + s->ss_h][b->filter][1],
                           s->dst[1], s->dst[2], ls_uv,
                           ref2->data[1], ref2->linesize[1],
                           ref2->data[2], ref2->linesize[2], tref2,
-                          row << 2, col << 2, &mvuv, bw, bh, w2, h2, 1);
+                          row << (3 - s->ss_v), col << (3 - s->ss_h),
+                          &b->mv[0][1], uvbw, uvbh, w2, h2, 1);
         }
     }
 }