Don't do edge emulation unless the edge pixels will be used in MC.

Do not emulate larger edges than we will actually use for this round of
MC. Decoding goes from avg+SE 29.972+/-0.023sec to 29.856+/-0.023, i.e.
0.12sec or ~0.4% faster.
(cherry picked from commit 44002d8323)

libavcodec/vp8.c

@@ -1218,21 +1218,26 @@ void vp8_mc(VP8Context *s, int luma,
             vp8_mc_func mc_func[3][3])
 {
     if (AV_RN32A(mv)) {
-        static const uint8_t idx[8] = { 0, 1, 2, 1, 2, 1, 2, 1 };
+        static const uint8_t idx[3][8] = {
-        int mx = (mv->x << luma)&7, mx_idx = idx[mx];
+            { 0, 1, 2, 1, 2, 1, 2, 1 }, // nr. of left extra pixels,
-        int my = (mv->y << luma)&7, my_idx = idx[my];
+                                        // also function pointer index
+            { 0, 3, 5, 3, 5, 3, 5, 3 }, // nr. of extra pixels required
+            { 0, 2, 3, 2, 3, 2, 3, 2 }, // nr. of right extra pixels
+        };
+        int mx = (mv->x << luma)&7, mx_idx = idx[0][mx];
+        int my = (mv->y << luma)&7, my_idx = idx[0][my];
 
         x_off += mv->x >> (3 - luma);
         y_off += mv->y >> (3 - luma);
 
         // edge emulation
         src += y_off * linesize + x_off;
-        if (x_off < 2 || x_off >= width  - block_w - 3 ||
+        if (x_off < mx_idx || x_off >= width  - block_w - idx[2][mx] ||
-            y_off < 2 || y_off >= height - block_h - 3) {
+            y_off < my_idx || y_off >= height - block_h - idx[2][my]) {
-            ff_emulated_edge_mc(s->edge_emu_buffer, src - 2 * linesize - 2, linesize,
+            ff_emulated_edge_mc(s->edge_emu_buffer, src - my_idx * linesize - mx_idx, linesize,
-                                block_w + 5, block_h + 5,
+                                block_w + idx[1][mx], block_h + idx[1][my],
-                                x_off - 2, y_off - 2, width, height);
+                                x_off - mx_idx, y_off - my_idx, width, height);
-            src = s->edge_emu_buffer + 2 + linesize * 2;
+            src = s->edge_emu_buffer + mx_idx + linesize * my_idx;
         }
         mc_func[my_idx][mx_idx](dst, linesize, src, linesize, block_h, mx, my);
     } else