vp8: move block coeff arithcoder on stack.

This prevents gcc from assuming that contents of it may have changed
between calls to vp56_range_get_prob(), thus preventing countless (and
unnecessary) movs. Decoding of sintel trailer goes from (avg+SG) 9.796
+/- 0.003 to 9.635 +/- 0.010.

1 files changed, 20 insertions, 18 deletions

diff --git a/libavcodec/vp8.c b/libavcodec/vp8.c
index d9fdd4da44..94200f68ac 100644
--- a/libavcodec/vp8.c
+++ b/libavcodec/vp8.c
@@ -707,56 +707,58 @@ void decode_mb_mode(VP8Context *s, VP8Macroblock *mb, int mb_x, int mb_y, uint8_
  * @return 0 if no coeffs were decoded
  *         otherwise, the index of the last coeff decoded plus one
  */
-static int decode_block_coeffs_internal(VP56RangeCoder *c, DCTELEM block[16],
+static int decode_block_coeffs_internal(VP56RangeCoder *r, DCTELEM block[16],
                                         uint8_t probs[16][3][NUM_DCT_TOKENS-1],
                                         int i, uint8_t *token_prob, int16_t qmul[2])
 {
+    VP56RangeCoder c = *r;
     goto skip_eob;
     do {
         int coeff;
-        if (!vp56_rac_get_prob_branchy(c, token_prob[0]))   // DCT_EOB
-            return i;
+        if (!vp56_rac_get_prob_branchy(&c, token_prob[0]))   // DCT_EOB
+            break;
 
 skip_eob:
-        if (!vp56_rac_get_prob_branchy(c, token_prob[1])) { // DCT_0
+        if (!vp56_rac_get_prob_branchy(&c, token_prob[1])) { // DCT_0
             if (++i == 16)
-                return i; // invalid input; blocks should end with EOB
+                break; // invalid input; blocks should end with EOB
             token_prob = probs[i][0];
             goto skip_eob;
         }
 
-        if (!vp56_rac_get_prob_branchy(c, token_prob[2])) { // DCT_1
+        if (!vp56_rac_get_prob_branchy(&c, token_prob[2])) { // DCT_1
             coeff = 1;
             token_prob = probs[i+1][1];
         } else {
-            if (!vp56_rac_get_prob_branchy(c, token_prob[3])) { // DCT 2,3,4
-                coeff = vp56_rac_get_prob_branchy(c, token_prob[4]);
+            if (!vp56_rac_get_prob_branchy(&c, token_prob[3])) { // DCT 2,3,4
+                coeff = vp56_rac_get_prob_branchy(&c, token_prob[4]);
                 if (coeff)
-                    coeff += vp56_rac_get_prob(c, token_prob[5]);
+                    coeff += vp56_rac_get_prob(&c, token_prob[5]);
                 coeff += 2;
             } else {
                 // DCT_CAT*
-                if (!vp56_rac_get_prob_branchy(c, token_prob[6])) {
-                    if (!vp56_rac_get_prob_branchy(c, token_prob[7])) { // DCT_CAT1
-                        coeff  = 5 + vp56_rac_get_prob(c, vp8_dct_cat1_prob[0]);
+                if (!vp56_rac_get_prob_branchy(&c, token_prob[6])) {
+                    if (!vp56_rac_get_prob_branchy(&c, token_prob[7])) { // DCT_CAT1
+                        coeff  = 5 + vp56_rac_get_prob(&c, vp8_dct_cat1_prob[0]);
                     } else {                                    // DCT_CAT2
                         coeff  = 7;
-                        coeff += vp56_rac_get_prob(c, vp8_dct_cat2_prob[0]) << 1;
-                        coeff += vp56_rac_get_prob(c, vp8_dct_cat2_prob[1]);
+                        coeff += vp56_rac_get_prob(&c, vp8_dct_cat2_prob[0]) << 1;
+                        coeff += vp56_rac_get_prob(&c, vp8_dct_cat2_prob[1]);
                     }
                 } else {    // DCT_CAT3 and up
-                    int a = vp56_rac_get_prob(c, token_prob[8]);
-                    int b = vp56_rac_get_prob(c, token_prob[9+a]);
+                    int a = vp56_rac_get_prob(&c, token_prob[8]);
+                    int b = vp56_rac_get_prob(&c, token_prob[9+a]);
                     int cat = (a<<1) + b;
                     coeff  = 3 + (8<<cat);
-                    coeff += vp8_rac_get_coeff(c, ff_vp8_dct_cat_prob[cat]);
+                    coeff += vp8_rac_get_coeff(&c, ff_vp8_dct_cat_prob[cat]);
                 }
             }
             token_prob = probs[i+1][2];
         }
-        block[zigzag_scan[i]] = (vp8_rac_get(c) ? -coeff : coeff) * qmul[!!i];
+        block[zigzag_scan[i]] = (vp8_rac_get(&c) ? -coeff : coeff) * qmul[!!i];
     } while (++i < 16);
 
+    *r = c;
     return i;
 }
 #endif