vp9: add inter-frame profile 2/3 suport.

3 files changed, 167 insertions, 87 deletions

diff --git a/libavcodec/vp9.c b/libavcodec/vp9.c
index ba62a295d0..db3f541c5e 100644
--- a/libavcodec/vp9.c
+++ b/libavcodec/vp9.c
@@ -2733,7 +2733,7 @@ static av_always_inline void mc_luma_scaled(VP9Context *s, vp9_scaled_mc_func sm
                                             const uint8_t *ref, ptrdiff_t ref_stride,
                                             ThreadFrame *ref_frame,
                                             ptrdiff_t y, ptrdiff_t x, const VP56mv *mv,
-                                            int bw, int bh, int w, int h,
+                                            int bw, int bh, int w, int h, int bytesperpixel,
                                             const uint16_t *scale, const uint8_t *step)
 {
 #define scale_mv(n, dim) (((int64_t)(n) * scale[dim]) >> 14)
@@ -2747,7 +2747,7 @@ static av_always_inline void mc_luma_scaled(VP9Context *s, vp9_scaled_mc_func sm
 
     y = my >> 4;
     x = mx >> 4;
-    ref += y * ref_stride + x;
+    ref += y * ref_stride + x * bytesperpixel;
     mx &= 15;
     my &= 15;
     refbw_m1 = ((bw - 1) * step[0] + mx) >> 4;
@@ -2759,12 +2759,12 @@ static av_always_inline void mc_luma_scaled(VP9Context *s, vp9_scaled_mc_func sm
     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,
-                                 ref - 3 * ref_stride - 3,
-                                 144, ref_stride,
+                                 ref - 3 * ref_stride - 3 * bytesperpixel,
+                                 288, ref_stride,
                                  refbw_m1 + 8, refbh_m1 + 8,
                                  x - 3, y - 3, w, h);
-        ref = s->edge_emu_buffer + 3 * 144 + 3;
-        ref_stride = 144;
+        ref = s->edge_emu_buffer + 3 * 288 + 3 * bytesperpixel;
+        ref_stride = 288;
     }
     smc(dst, dst_stride, ref, ref_stride, bh, mx, my, step[0], step[1]);
 }
@@ -2776,7 +2776,7 @@ static av_always_inline void mc_chroma_scaled(VP9Context *s, vp9_scaled_mc_func
                                               const uint8_t *ref_v, ptrdiff_t src_stride_v,
                                               ThreadFrame *ref_frame,
                                               ptrdiff_t y, ptrdiff_t x, const VP56mv *mv,
-                                              int bw, int bh, int w, int h,
+                                              int bw, int bh, int w, int h, int bytesperpixel,
                                               const uint16_t *scale, const uint8_t *step)
 {
     // BUG https://code.google.com/p/webm/issues/detail?id=820
@@ -2788,8 +2788,8 @@ static av_always_inline void mc_chroma_scaled(VP9Context *s, vp9_scaled_mc_func
 
     y = my >> 4;
     x = mx >> 4;
-    ref_u += y * src_stride_u + x;
-    ref_v += y * src_stride_v + x;
+    ref_u += y * src_stride_u + x * bytesperpixel;
+    ref_v += y * src_stride_v + x * bytesperpixel;
     mx &= 15;
     my &= 15;
     refbw_m1 = ((bw - 1) * step[0] + mx) >> 4;
@@ -2801,51 +2801,60 @@ static av_always_inline void mc_chroma_scaled(VP9Context *s, vp9_scaled_mc_func
     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,
-                                 ref_u - 3 * src_stride_u - 3,
-                                 144, src_stride_u,
+                                 ref_u - 3 * src_stride_u - 3 * bytesperpixel,
+                                 288, src_stride_u,
                                  refbw_m1 + 8, refbh_m1 + 8,
                                  x - 3, y - 3, w, h);
-        ref_u = s->edge_emu_buffer + 3 * 144 + 3;
-        smc(dst_u, dst_stride, ref_u, 144, bh, mx, my, step[0], step[1]);
+        ref_u = s->edge_emu_buffer + 3 * 288 + 3 * bytesperpixel;
+        smc(dst_u, dst_stride, ref_u, 288, bh, mx, my, step[0], step[1]);
 
         s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
-                                 ref_v - 3 * src_stride_v - 3,
-                                 144, src_stride_v,
+                                 ref_v - 3 * src_stride_v - 3 * bytesperpixel,
+                                 288, src_stride_v,
                                  refbw_m1 + 8, refbh_m1 + 8,
                                  x - 3, y - 3, w, h);
-        ref_v = s->edge_emu_buffer + 3 * 144 + 3;
-        smc(dst_v, dst_stride, ref_v, 144, bh, mx, my, step[0], step[1]);
+        ref_v = s->edge_emu_buffer + 3 * 288 + 3 * bytesperpixel;
+        smc(dst_v, dst_stride, ref_v, 288, bh, mx, my, step[0], step[1]);
     } else {
         smc(dst_u, dst_stride, ref_u, src_stride_u, bh, mx, my, step[0], step[1]);
         smc(dst_v, dst_stride, ref_v, src_stride_v, bh, mx, my, step[0], step[1]);
     }
 }
 
-#define FN(x) x##_scaled
 #define mc_luma_dir(s, mc, dst, dst_ls, src, src_ls, tref, row, col, mv, bw, bh, w, h, i) \
     mc_luma_scaled(s, s->dsp.s##mc, dst, dst_ls, src, src_ls, tref, row, col, \
-                   mv, bw, bh, w, h, s->mvscale[b->ref[i]], s->mvstep[b->ref[i]])
+                   mv, bw, bh, w, h, bytesperpixel, \
+                   s->mvscale[b->ref[i]], s->mvstep[b->ref[i]])
 #define mc_chroma_dir(s, mc, dstu, dstv, dst_ls, srcu, srcu_ls, srcv, srcv_ls, tref, \
                       row, col, mv, bw, bh, w, h, i) \
     mc_chroma_scaled(s, s->dsp.s##mc, dstu, dstv, dst_ls, srcu, srcu_ls, srcv, srcv_ls, tref, \
-                     row, col, mv, bw, bh, w, h, s->mvscale[b->ref[i]], s->mvstep[b->ref[i]])
+                     row, col, mv, bw, bh, w, h, bytesperpixel, \
+                     s->mvscale[b->ref[i]], s->mvstep[b->ref[i]])
+#define FN(x) x##_scaled_8bpp
+#define BYTES_PER_PIXEL 1
+#include "vp9_mc_template.c"
+#undef FN
+#undef BYTES_PER_PIXEL
+#define FN(x) x##_scaled_16bpp
+#define BYTES_PER_PIXEL 2
 #include "vp9_mc_template.c"
 #undef mc_luma_dir
 #undef mc_chroma_dir
 #undef FN
+#undef BYTES_PER_PIXEL
 
 static av_always_inline void mc_luma_unscaled(VP9Context *s, vp9_mc_func (*mc)[2],
                                               uint8_t *dst, ptrdiff_t dst_stride,
                                               const uint8_t *ref, ptrdiff_t ref_stride,
                                               ThreadFrame *ref_frame,
                                               ptrdiff_t y, ptrdiff_t x, const VP56mv *mv,
-                                              int bw, int bh, int w, int h)
+                                              int bw, int bh, int w, int h, int bytesperpixel)
 {
     int mx = mv->x, my = mv->y, th;
 
     y += my >> 3;
     x += mx >> 3;
-    ref += y * ref_stride + x;
+    ref += y * ref_stride + x * bytesperpixel;
     mx &= 7;
     my &= 7;
     // FIXME bilinear filter only needs 0/1 pixels, not 3/4
@@ -2856,12 +2865,12 @@ static av_always_inline void mc_luma_unscaled(VP9Context *s, vp9_mc_func (*mc)[2
     if (x < !!mx * 3 || y < !!my * 3 ||
         x + !!mx * 4 > w - bw || y + !!my * 4 > h - bh) {
         s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
-                                 ref - !!my * 3 * ref_stride - !!mx * 3,
-                                 80, ref_stride,
+                                 ref - !!my * 3 * ref_stride - !!mx * 3 * bytesperpixel,
+                                 160, ref_stride,
                                  bw + !!mx * 7, bh + !!my * 7,
                                  x - !!mx * 3, y - !!my * 3, w, h);
-        ref = s->edge_emu_buffer + !!my * 3 * 80 + !!mx * 3;
-        ref_stride = 80;
+        ref = s->edge_emu_buffer + !!my * 3 * 160 + !!mx * 3 * bytesperpixel;
+        ref_stride = 160;
     }
     mc[!!mx][!!my](dst, dst_stride, ref, ref_stride, bh, mx << 1, my << 1);
 }
@@ -2873,14 +2882,14 @@ static av_always_inline void mc_chroma_unscaled(VP9Context *s, vp9_mc_func (*mc)
                                                 const uint8_t *ref_v, ptrdiff_t src_stride_v,
                                                 ThreadFrame *ref_frame,
                                                 ptrdiff_t y, ptrdiff_t x, const VP56mv *mv,
-                                                int bw, int bh, int w, int h)
+                                                int bw, int bh, int w, int h, int bytesperpixel)
 {
     int mx = mv->x << !s->ss_h, my = mv->y << !s->ss_v, th;
 
     y += my >> 4;
     x += mx >> 4;
-    ref_u += y * src_stride_u + x;
-    ref_v += y * src_stride_v + x;
+    ref_u += y * src_stride_u + x * bytesperpixel;
+    ref_v += y * src_stride_v + x * bytesperpixel;
     mx &= 15;
     my &= 15;
     // FIXME bilinear filter only needs 0/1 pixels, not 3/4
@@ -2891,49 +2900,64 @@ static av_always_inline void mc_chroma_unscaled(VP9Context *s, vp9_mc_func (*mc)
     if (x < !!mx * 3 || y < !!my * 3 ||
         x + !!mx * 4 > w - bw || y + !!my * 4 > h - bh) {
         s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
-                                 ref_u - !!my * 3 * src_stride_u - !!mx * 3,
-                                 80, src_stride_u,
+                                 ref_u - !!my * 3 * src_stride_u - !!mx * 3 * bytesperpixel,
+                                 160, src_stride_u,
                                  bw + !!mx * 7, bh + !!my * 7,
                                  x - !!mx * 3, y - !!my * 3, w, h);
-        ref_u = s->edge_emu_buffer + !!my * 3 * 80 + !!mx * 3;
-        mc[!!mx][!!my](dst_u, dst_stride, ref_u, 80, bh, mx, my);
+        ref_u = s->edge_emu_buffer + !!my * 3 * 160 + !!mx * 3 * bytesperpixel;
+        mc[!!mx][!!my](dst_u, dst_stride, ref_u, 160, bh, mx, my);
 
         s->vdsp.emulated_edge_mc(s->edge_emu_buffer,
-                                 ref_v - !!my * 3 * src_stride_v - !!mx * 3,
-                                 80, src_stride_v,
+                                 ref_v - !!my * 3 * src_stride_v - !!mx * 3 * bytesperpixel,
+                                 160, src_stride_v,
                                  bw + !!mx * 7, bh + !!my * 7,
                                  x - !!mx * 3, y - !!my * 3, w, h);
-        ref_v = s->edge_emu_buffer + !!my * 3 * 80 + !!mx * 3;
-        mc[!!mx][!!my](dst_v, dst_stride, ref_v, 80, bh, mx, my);
+        ref_v = s->edge_emu_buffer + !!my * 3 * 160 + !!mx * 3 * bytesperpixel;
+        mc[!!mx][!!my](dst_v, dst_stride, ref_v, 160, bh, mx, my);
     } else {
         mc[!!mx][!!my](dst_u, dst_stride, ref_u, src_stride_u, bh, mx, my);
         mc[!!mx][!!my](dst_v, dst_stride, ref_v, src_stride_v, bh, mx, my);
     }
 }
 
-#define FN(x) x
 #define mc_luma_dir(s, mc, dst, dst_ls, src, src_ls, tref, row, col, mv, bw, bh, w, h, i) \
     mc_luma_unscaled(s, s->dsp.mc, dst, dst_ls, src, src_ls, tref, row, col, \
-                     mv, bw, bh, w, h)
+                     mv, bw, bh, w, h, bytesperpixel)
 #define mc_chroma_dir(s, mc, dstu, dstv, dst_ls, srcu, srcu_ls, srcv, srcv_ls, tref, \
                       row, col, mv, bw, bh, w, h, i) \
     mc_chroma_unscaled(s, s->dsp.mc, dstu, dstv, dst_ls, srcu, srcu_ls, srcv, srcv_ls, tref, \
-                       row, col, mv, bw, bh, w, h)
+                       row, col, mv, bw, bh, w, h, bytesperpixel)
+#define FN(x) x##_8bpp
+#define BYTES_PER_PIXEL 1
+#include "vp9_mc_template.c"
+#undef FN
+#undef BYTES_PER_PIXEL
+#define FN(x) x##_16bpp
+#define BYTES_PER_PIXEL 2
 #include "vp9_mc_template.c"
 #undef mc_luma_dir_dir
 #undef mc_chroma_dir_dir
 #undef FN
+#undef BYTES_PER_PIXEL
 
-static void inter_recon(AVCodecContext *ctx)
+static av_always_inline void inter_recon(AVCodecContext *ctx, int bytesperpixel)
 {
     VP9Context *s = ctx->priv_data;
     VP9Block *b = s->b;
     int row = s->row, col = s->col;
 
     if (s->mvscale[b->ref[0]][0] || (b->comp && s->mvscale[b->ref[1]][0])) {
-        inter_pred_scaled(ctx);
+        if (bytesperpixel == 1) {
+            inter_pred_scaled_8bpp(ctx);
+        } else {
+            inter_pred_scaled_16bpp(ctx);
+        }
     } else {
-        inter_pred(ctx);
+        if (bytesperpixel == 1) {
+            inter_pred_8bpp(ctx);
+        } else {
+            inter_pred_16bpp(ctx);
+        }
     }
     if (!b->skip) {
         /* mostly copied intra_recon() */
@@ -2949,12 +2973,13 @@ static void inter_recon(AVCodecContext *ctx)
         // y itxfm add
         for (n = 0, y = 0; y < end_y; y += step1d) {
             uint8_t *ptr = dst;
-            for (x = 0; x < end_x; x += step1d, ptr += 4 * step1d, n += step) {
+            for (x = 0; x < end_x; x += step1d,
+                 ptr += 4 * step1d * bytesperpixel, n += step) {
                 int eob = b->tx > TX_8X8 ? AV_RN16A(&s->eob[n]) : s->eob[n];
 
                 if (eob)
                     s->dsp.itxfm_add[tx][DCT_DCT](ptr, s->y_stride,
-                                                  s->block + 16 * n, eob);
+                                                  s->block + 16 * n * bytesperpixel, eob);
             }
             dst += 4 * s->y_stride * step1d;
         }
@@ -2967,12 +2992,13 @@ static void inter_recon(AVCodecContext *ctx)
             dst = s->dst[p + 1];
             for (n = 0, y = 0; y < end_y; y += uvstep1d) {
                 uint8_t *ptr = dst;
-                for (x = 0; x < end_x; x += uvstep1d, ptr += 4 * uvstep1d, n += step) {
+                for (x = 0; x < end_x; x += uvstep1d,
+                     ptr += 4 * uvstep1d * bytesperpixel, n += step) {
                     int eob = b->uvtx > TX_8X8 ? AV_RN16A(&s->uveob[p][n]) : s->uveob[p][n];
 
                     if (eob)
                         s->dsp.itxfm_add[uvtx][DCT_DCT](ptr, s->uv_stride,
-                                                        s->uvblock[p] + 16 * n, eob);
+                                                        s->uvblock[p] + 16 * n * bytesperpixel, eob);
                 }
                 dst += 4 * uvstep1d * s->uv_stride;
             }
@@ -2980,6 +3006,16 @@ static void inter_recon(AVCodecContext *ctx)
     }
 }
 
+static void inter_recon_8bpp(AVCodecContext *ctx)
+{
+    inter_recon(ctx, 1);
+}
+
+static void inter_recon_16bpp(AVCodecContext *ctx)
+{
+    inter_recon(ctx, 2);
+}
+
 static av_always_inline void mask_edges(uint8_t (*mask)[8][4], int ss_h, int ss_v,
                                         int row_and_7, int col_and_7,
                                         int w, int h, int col_end, int row_end,
@@ -3211,7 +3247,11 @@ static void decode_b(AVCodecContext *ctx, int row, int col,
             intra_recon_8bpp(ctx, yoff, uvoff);
         }
     } else {
-        inter_recon(ctx);
+        if (s->bpp > 8) {
+            inter_recon_16bpp(ctx);
+        } else {
+            inter_recon_8bpp(ctx);
+        }
     }
     if (emu[0]) {
         int w = FFMIN(s->cols - col, w4) * 8, h = FFMIN(s->rows - row, h4) * 8, n, o = 0;
diff --git a/libavcodec/vp9_mc_template.c b/libavcodec/vp9_mc_template.c
index f94438fa06..250e0a691c 100644
--- a/libavcodec/vp9_mc_template.c
+++ b/libavcodec/vp9_mc_template.c
@@ -40,6 +40,7 @@ static void FN(inter_pred)(AVCodecContext *ctx)
     AVFrame *ref1 = tref1->f, *ref2;
     int w1 = ref1->width, h1 = ref1->height, w2, h2;
     ptrdiff_t ls_y = s->y_stride, ls_uv = s->uv_stride;
+    int bytesperpixel = BYTES_PER_PIXEL;
 
     if (b->comp) {
         tref2 = &s->refs[s->refidx[b->ref[1]]];
@@ -138,7 +139,7 @@ static void FN(inter_pred)(AVCodecContext *ctx)
             mc_luma_dir(s, mc[4][b->filter][0], s->dst[0], ls_y,
                         ref1->data[0], ref1->linesize[0], tref1,
                         row << 3, col << 3, &b->mv[0][0], 4, 8, w1, h1, 0);
-            mc_luma_dir(s, mc[4][b->filter][0], s->dst[0] + 4, ls_y,
+            mc_luma_dir(s, mc[4][b->filter][0], s->dst[0] + 4 * bytesperpixel, 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;
@@ -159,7 +160,8 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                               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,
+                              s->dst[1] + 4 * bytesperpixel,
+                              s->dst[2] + 4 * bytesperpixel, ls_uv,
                               ref1->data[1], ref1->linesize[1],
                               ref1->data[2], ref1->linesize[2], tref1,
                               row << (3 - s->ss_v), (col << 3) + 4,
@@ -170,7 +172,7 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                 mc_luma_dir(s, mc[4][b->filter][1], s->dst[0], ls_y,
                             ref2->data[0], ref2->linesize[0], tref2,
                             row << 3, col << 3, &b->mv[0][1], 4, 8, w2, h2, 1);
-                mc_luma_dir(s, mc[4][b->filter][1], s->dst[0] + 4, ls_y,
+                mc_luma_dir(s, mc[4][b->filter][1], s->dst[0] + 4 * bytesperpixel, 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;
@@ -191,7 +193,8 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                                   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,
+                                  s->dst[1] + 4 * bytesperpixel,
+                                  s->dst[2] + 4 * bytesperpixel, ls_uv,
                                   ref2->data[1], ref2->linesize[1],
                                   ref2->data[2], ref2->linesize[2], tref2,
                                   row << (3 - s->ss_v), (col << 3) + 4,
@@ -206,7 +209,7 @@ static void FN(inter_pred)(AVCodecContext *ctx)
             mc_luma_dir(s, mc[4][b->filter][0], s->dst[0], ls_y,
                         ref1->data[0], ref1->linesize[0], tref1,
                         row << 3, col << 3, &b->mv[0][0], 4, 4, w1, h1, 0);
-            mc_luma_dir(s, mc[4][b->filter][0], s->dst[0] + 4, ls_y,
+            mc_luma_dir(s, mc[4][b->filter][0], s->dst[0] + 4 * bytesperpixel, ls_y,
                         ref1->data[0], ref1->linesize[0], tref1,
                         row << 3, (col << 3) + 4, &b->mv[1][0], 4, 4, w1, h1, 0);
             mc_luma_dir(s, mc[4][b->filter][0],
@@ -214,7 +217,7 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                         ref1->data[0], ref1->linesize[0], tref1,
                         (row << 3) + 4, col << 3, &b->mv[2][0], 4, 4, w1, h1, 0);
             mc_luma_dir(s, mc[4][b->filter][0],
-                        s->dst[0] + 4 * ls_y + 4, ls_y,
+                        s->dst[0] + 4 * ls_y + 4 * bytesperpixel, 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) {
@@ -239,7 +242,8 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                                   &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,
+                                  s->dst[1] + 4 * bytesperpixel,
+                                  s->dst[2] + 4 * bytesperpixel, ls_uv,
                                   ref1->data[1], ref1->linesize[1],
                                   ref1->data[2], ref1->linesize[2], tref1,
                                   row << 2, (col << 3) + 4,
@@ -273,7 +277,8 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                                   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,
+                                  s->dst[1] + 4 * bytesperpixel,
+                                  s->dst[2] + 4 * bytesperpixel, ls_uv,
                                   ref1->data[1], ref1->linesize[1],
                                   ref1->data[2], ref1->linesize[2], tref1,
                                   row << 3, (col << 3) + 4,
@@ -285,7 +290,8 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                                   (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,
+                                  s->dst[1] + 4 * ls_uv + 4 * bytesperpixel,
+                                  s->dst[2] + 4 * ls_uv + 4 * bytesperpixel, ls_uv,
                                   ref1->data[1], ref1->linesize[1],
                                   ref1->data[2], ref1->linesize[2], tref1,
                                   (row << 3) + 4, (col << 3) + 4,
@@ -297,7 +303,7 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                 mc_luma_dir(s, mc[4][b->filter][1], s->dst[0], ls_y,
                             ref2->data[0], ref2->linesize[0], tref2,
                             row << 3, col << 3, &b->mv[0][1], 4, 4, w2, h2, 1);
-                mc_luma_dir(s, mc[4][b->filter][1], s->dst[0] + 4, ls_y,
+                mc_luma_dir(s, mc[4][b->filter][1], s->dst[0] + 4 * bytesperpixel, ls_y,
                             ref2->data[0], ref2->linesize[0], tref2,
                             row << 3, (col << 3) + 4, &b->mv[1][1], 4, 4, w2, h2, 1);
                 mc_luma_dir(s, mc[4][b->filter][1],
@@ -305,7 +311,7 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                             ref2->data[0], ref2->linesize[0], tref2,
                             (row << 3) + 4, col << 3, &b->mv[2][1], 4, 4, w2, h2, 1);
                 mc_luma_dir(s, mc[4][b->filter][1],
-                            s->dst[0] + 4 * ls_y + 4, ls_y,
+                            s->dst[0] + 4 * ls_y + 4 * bytesperpixel, 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) {
@@ -330,7 +336,8 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                                       &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,
+                                      s->dst[1] + 4 * bytesperpixel,
+                                      s->dst[2] + 4 * bytesperpixel, ls_uv,
                                       ref2->data[1], ref2->linesize[1],
                                       ref2->data[2], ref2->linesize[2], tref2,
                                       row << 2, (col << 3) + 4,
@@ -364,7 +371,8 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                                       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,
+                                      s->dst[1] + 4 * bytesperpixel,
+                                      s->dst[2] + 4 * bytesperpixel, ls_uv,
                                       ref2->data[1], ref2->linesize[1],
                                       ref2->data[2], ref2->linesize[2], tref2,
                                       row << 3, (col << 3) + 4,
@@ -376,7 +384,8 @@ static void FN(inter_pred)(AVCodecContext *ctx)
                                       (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,
+                                      s->dst[1] + 4 * ls_uv + 4 * bytesperpixel,
+                                      s->dst[2] + 4 * ls_uv + 4 * bytesperpixel, ls_uv,
                                       ref2->data[1], ref2->linesize[1],
                                       ref2->data[2], ref2->linesize[2], tref2,
                                       (row << 3) + 4, (col << 3) + 4,
diff --git a/libavcodec/vp9dsp_template.c b/libavcodec/vp9dsp_template.c
index a8d056b4ba..18b14e8490 100644
--- a/libavcodec/vp9dsp_template.c
+++ b/libavcodec/vp9dsp_template.c
@@ -1911,22 +1911,27 @@ static av_always_inline void copy_c(uint8_t *dst, ptrdiff_t dst_stride,
                                     int w, int h)
 {
     do {
-        memcpy(dst, src, w);
+        memcpy(dst, src, w * sizeof(pixel));
 
         dst += dst_stride;
         src += src_stride;
     } while (--h);
 }
 
-static av_always_inline void avg_c(uint8_t *dst, ptrdiff_t dst_stride,
-                                   const uint8_t *src, ptrdiff_t src_stride,
+static av_always_inline void avg_c(uint8_t *_dst, ptrdiff_t dst_stride,
+                                   const uint8_t *_src, ptrdiff_t src_stride,
                                    int w, int h)
 {
+    pixel *dst = (pixel *) _dst;
+    const pixel *src = (const pixel *) _src;
+
+    dst_stride /= sizeof(pixel);
+    src_stride /= sizeof(pixel);
     do {
         int x;
 
         for (x = 0; x < w; x += 4)
-            AV_WN32A(&dst[x], rnd_avg32(AV_RN32A(&dst[x]), AV_RN32(&src[x])));
+            AV_WN4PA(&dst[x], rnd_avg_pixel4(AV_RN4PA(&dst[x]), AV_RN4P(&src[x])));
 
         dst += dst_stride;
         src += src_stride;
@@ -2010,7 +2015,7 @@ static const int16_t vp9_subpel_filters[3][16][8] = {
 };
 
 #define FILTER_8TAP(src, x, F, stride) \
-    av_clip_uint8((F[0] * src[x + -3 * stride] + \
+    av_clip_pixel((F[0] * src[x + -3 * stride] + \
                    F[1] * src[x + -2 * stride] + \
                    F[2] * src[x + -1 * stride] + \
                    F[3] * src[x + +0 * stride] + \
@@ -2019,11 +2024,16 @@ static const int16_t vp9_subpel_filters[3][16][8] = {
                    F[6] * src[x + +3 * stride] + \
                    F[7] * src[x + +4 * stride] + 64) >> 7)
 
-static av_always_inline void do_8tap_1d_c(uint8_t *dst, ptrdiff_t dst_stride,
-                                          const uint8_t *src, ptrdiff_t src_stride,
+static av_always_inline void do_8tap_1d_c(uint8_t *_dst, ptrdiff_t dst_stride,
+                                          const uint8_t *_src, ptrdiff_t src_stride,
                                           int w, int h, ptrdiff_t ds,
                                           const int16_t *filter, int avg)
 {
+    pixel *dst = (pixel *) _dst;
+    const pixel *src = (const pixel *) _src;
+
+    dst_stride /= sizeof(pixel);
+    src_stride /= sizeof(pixel);
     do {
         int x;
 
@@ -2047,21 +2057,25 @@ static av_noinline void opn##_8tap_1d_##dir##_c(uint8_t *dst, ptrdiff_t dst_stri
     do_8tap_1d_c(dst, dst_stride, src, src_stride, w, h, ds, filter, opa); \
 }
 
-filter_8tap_1d_fn(put, 0, v, src_stride)
+filter_8tap_1d_fn(put, 0, v, src_stride / sizeof(pixel))
 filter_8tap_1d_fn(put, 0, h, 1)
-filter_8tap_1d_fn(avg, 1, v, src_stride)
+filter_8tap_1d_fn(avg, 1, v, src_stride / sizeof(pixel))
 filter_8tap_1d_fn(avg, 1, h, 1)
 
 #undef filter_8tap_1d_fn
 
-static av_always_inline void do_8tap_2d_c(uint8_t *dst, ptrdiff_t dst_stride,
-                                          const uint8_t *src, ptrdiff_t src_stride,
+static av_always_inline void do_8tap_2d_c(uint8_t *_dst, ptrdiff_t dst_stride,
+                                          const uint8_t *_src, ptrdiff_t src_stride,
                                           int w, int h, const int16_t *filterx,
                                           const int16_t *filtery, int avg)
 {
     int tmp_h = h + 7;
-    uint8_t tmp[64 * 71], *tmp_ptr = tmp;
+    pixel tmp[64 * 71], *tmp_ptr = tmp;
+    pixel *dst = (pixel *) _dst;
+    const pixel *src = (const pixel *) _src;
 
+    dst_stride /= sizeof(pixel);
+    src_stride /= sizeof(pixel);
     src -= src_stride * 3;
     do {
         int x;
@@ -2125,10 +2139,15 @@ static void avg##_8tap_##type##_##sz##hv_c(uint8_t *dst, ptrdiff_t dst_stride, \
 #define FILTER_BILIN(src, x, mxy, stride) \
     (src[x] + ((mxy * (src[x + stride] - src[x]) + 8) >> 4))
 
-static av_always_inline void do_bilin_1d_c(uint8_t *dst, ptrdiff_t dst_stride,
-                                           const uint8_t *src, ptrdiff_t src_stride,
+static av_always_inline void do_bilin_1d_c(uint8_t *_dst, ptrdiff_t dst_stride,
+                                           const uint8_t *_src, ptrdiff_t src_stride,
                                            int w, int h, ptrdiff_t ds, int mxy, int avg)
 {
+    pixel *dst = (pixel *) _dst;
+    const pixel *src = (const pixel *) _src;
+
+    dst_stride /= sizeof(pixel);
+    src_stride /= sizeof(pixel);
     do {
         int x;
 
@@ -2152,20 +2171,24 @@ static av_noinline void opn##_bilin_1d_##dir##_c(uint8_t *dst, ptrdiff_t dst_str
     do_bilin_1d_c(dst, dst_stride, src, src_stride, w, h, ds, mxy, opa); \
 }
 
-bilin_1d_fn(put, 0, v, src_stride)
+bilin_1d_fn(put, 0, v, src_stride / sizeof(pixel))
 bilin_1d_fn(put, 0, h, 1)
-bilin_1d_fn(avg, 1, v, src_stride)
+bilin_1d_fn(avg, 1, v, src_stride / sizeof(pixel))
 bilin_1d_fn(avg, 1, h, 1)
 
 #undef bilin_1d_fn
 
-static av_always_inline void do_bilin_2d_c(uint8_t *dst, ptrdiff_t dst_stride,
-                                           const uint8_t *src, ptrdiff_t src_stride,
+static av_always_inline void do_bilin_2d_c(uint8_t *_dst, ptrdiff_t dst_stride,
+                                           const uint8_t *_src, ptrdiff_t src_stride,
                                            int w, int h, int mx, int my, int avg)
 {
-    uint8_t tmp[64 * 65], *tmp_ptr = tmp;
+    pixel tmp[64 * 65], *tmp_ptr = tmp;
     int tmp_h = h + 1;
+    pixel *dst = (pixel *) _dst;
+    const pixel *src = (const pixel *) _src;
 
+    dst_stride /= sizeof(pixel);
+    src_stride /= sizeof(pixel);
     do {
         int x;
 
@@ -2299,15 +2322,19 @@ static av_cold void vp9dsp_mc_init(VP9DSPContext *dsp)
 #undef init_subpel3
 }
 
-static av_always_inline void do_scaled_8tap_c(uint8_t *dst, ptrdiff_t dst_stride,
-                                              const uint8_t *src, ptrdiff_t src_stride,
+static av_always_inline void do_scaled_8tap_c(uint8_t *_dst, ptrdiff_t dst_stride,
+                                              const uint8_t *_src, ptrdiff_t src_stride,
                                               int w, int h, int mx, int my,
                                               int dx, int dy, int avg,
                                               const int16_t (*filters)[8])
 {
     int tmp_h = (((h - 1) * dy + my) >> 4) + 8;
-    uint8_t tmp[64 * 135], *tmp_ptr = tmp;
+    pixel tmp[64 * 135], *tmp_ptr = tmp;
+    pixel *dst = (pixel *) _dst;
+    const pixel *src = (const pixel *) _src;
 
+    dst_stride /= sizeof(pixel);
+    src_stride /= sizeof(pixel);
     src -= src_stride * 3;
     do {
         int x;
@@ -2369,14 +2396,18 @@ static void avg##_scaled_##type##_##sz##_c(uint8_t *dst, ptrdiff_t dst_stride, \
                         vp9_subpel_filters[type_idx]); \
 }
 
-static av_always_inline void do_scaled_bilin_c(uint8_t *dst, ptrdiff_t dst_stride,
-                                               const uint8_t *src, ptrdiff_t src_stride,
+static av_always_inline void do_scaled_bilin_c(uint8_t *_dst, ptrdiff_t dst_stride,
+                                               const uint8_t *_src, ptrdiff_t src_stride,
                                                int w, int h, int mx, int my,
                                                int dx, int dy, int avg)
 {
-    uint8_t tmp[64 * 129], *tmp_ptr = tmp;
+    pixel tmp[64 * 129], *tmp_ptr = tmp;
     int tmp_h = (((h - 1) * dy + my) >> 4) + 2;
+    pixel *dst = (pixel *) _dst;
+    const pixel *src = (const pixel *) _src;
 
+    dst_stride /= sizeof(pixel);
+    src_stride /= sizeof(pixel);
     do {
         int x;
         int imx = mx, ioff = 0;