shithub: libvpx

Info  •  Files  •  Log  •  Branches

Add best_rd breakout in intra4x4 RD loop.

Encoding time of first 50 frames of bus (speed 0) @ 1500kbps goes from
1min5.4 to 1min4.0, i.e. 2.2% faster overall.

Change-Id: I8c32f2aff9a649ce7dd49d910dc5ba16b99c3bc6

--- a/vp9/encoder/vp9_rdopt.c

+++ b/vp9/encoder/vp9_rdopt.c

@@ -1178,10 +1178,11 @@

                                      ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l,

                                      int *bestrate, int *bestratey,

                                      int64_t *bestdistortion,

-                                     BLOCK_SIZE_TYPE bsize) {

+                                     BLOCK_SIZE_TYPE bsize,

+                                     int64_t rd_thresh) {

   MB_PREDICTION_MODE mode;

   MACROBLOCKD *xd = &x->e_mbd;

-  int64_t best_rd = INT64_MAX;

+  int64_t best_rd = rd_thresh;

   int rate = 0;

   int64_t distortion;

   VP9_COMMON *const cm = &cpi->common;

@@ -1262,6 +1263,8 @@

         distortion += vp9_block_error(coeff, BLOCK_OFFSET(pd->dqcoeff,

                                                           block, 16),

                                       16, &ssz) >> 2;

+        if (RDCOST(x->rdmult, x->rddiv, ratey, distortion) >= best_rd)

+          goto next;

         if (tx_type != DCT_DCT)

           vp9_short_iht4x4_add(BLOCK_OFFSET(pd->dqcoeff, block, 16),

@@ -1284,6 +1287,8 @@

       best_tx_type = tx_type;

       vpx_memcpy(a, tempa, sizeof(tempa));

       vpx_memcpy(l, templ, sizeof(templ));

+      // FIXME(rbultje) why are we storing best_dqcoeff instead of the

+      // dst buffer here?

       for (idy = 0; idy < num_4x4_blocks_high; ++idy) {

         for (idx = 0; idx < num_4x4_blocks_wide; ++idx) {

           block = ib + idy * 2 + idx;

@@ -1293,9 +1298,11 @@

         }

       }

     }

+  next:

+    {}

   }

-  if (x->skip_encode)

+  if (best_rd >= rd_thresh || x->skip_encode)

     return best_rd;

   for (idy = 0; idy < num_4x4_blocks_high; ++idy) {

@@ -1352,7 +1359,7 @@

       const int mis = xd->mode_info_stride;

       MB_PREDICTION_MODE UNINITIALIZED_IS_SAFE(best_mode);

       int UNINITIALIZED_IS_SAFE(r), UNINITIALIZED_IS_SAFE(ry);

-      int64_t UNINITIALIZED_IS_SAFE(d);

+      int64_t UNINITIALIZED_IS_SAFE(d), this_rd;

       i = idy * 2 + idx;

       if (cpi->common.frame_type == KEY_FRAME) {

@@ -1363,9 +1370,14 @@

         bmode_costs  = mb->y_mode_costs[A][L];

       }

-      total_rd += rd_pick_intra4x4block(cpi, mb, i, &best_mode, bmode_costs,

-                                        t_above + idx, t_left + idy,

-                                        &r, &ry, &d, bsize);

+      this_rd = rd_pick_intra4x4block(cpi, mb, i, &best_mode, bmode_costs,

+                                      t_above + idx, t_left + idy,

+                                      &r, &ry, &d, bsize,

+                                      best_rd - total_rd);

+      if (this_rd >= best_rd - total_rd)

+        return INT64_MAX;

+

+      total_rd += this_rd;

       cost += r;

       distortion += d;

       tot_rate_y += ry;

@@ -3511,8 +3523,9 @@

         */

       mbmi->txfm_size = TX_4X4;

-      rd_pick_intra4x4mby_modes(cpi, x, &rate, &rate_y,

-                                &distortion_y, INT64_MAX);

+      if (rd_pick_intra4x4mby_modes(cpi, x, &rate, &rate_y,

+                                    &distortion_y, best_rd) >= best_rd)

+        continue;

       rate2 += rate;

       rate2 += intra_cost_penalty;

       distortion2 += distortion_y;