shithub: libvpx

Info  •  Files  •  Log  •  Branches

Remove the token state array from greedy optimize_b.

Reduces memory usage, and speeds up encoding for some difficult clips.
No impact on output or metrics.

Ported from aomedia patch:
https://aomedia-review.googlesource.com/c/14501

Change-Id: I26ec69af8336f9e80da486a1cfbfc89a3596954d

--- a/vp9/encoder/vp9_encodemb.c

+++ b/vp9/encoder/vp9_encodemb.c

@@ -57,12 +57,6 @@

 #if USE_GREEDY_OPTIMIZE_B

-typedef struct {

-  int16_t token;

-  tran_low_t qc;

-  tran_low_t dqc;

-} vp9_token_state;

-

 // 'num' can be negative, but 'shift' must be non-negative.

 #define RIGHT_SHIFT_POSSIBLY_NEGATIVE(num, shift) \

   ((num) >= 0) ? (num) >> (shift) : -((-(num)) >> (shift))

@@ -73,7 +67,6 @@

   struct macroblock_plane *const p = &mb->plane[plane];

   struct macroblockd_plane *const pd = &xd->plane[plane];

   const int ref = is_inter_block(xd->mi[0]);

-  vp9_token_state tokens[1025][2];

   uint8_t token_cache[1024];

   const tran_low_t *const coeff = BLOCK_OFFSET(p->coeff, block);

   tran_low_t *const qcoeff = BLOCK_OFFSET(p->qcoeff, block);

@@ -110,22 +103,16 @@

   int64_t accu_error = ((int64_t)1) << 50;

   int64_t best_block_rd_cost = INT64_MAX;

   int x_prev = 1;

+  tran_low_t before_best_eob_qc = 0;

+  tran_low_t before_best_eob_dqc = 0;

+

   assert((!plane_type && !plane) || (plane_type && plane));

   assert(eob <= default_eob);

   for (i = 0; i < eob; i++) {

     const int rc = scan[i];

-    int x = qcoeff[rc];

-    t0 = vp9_get_token(x);

-    tokens[i][0].qc = x;

-    tokens[i][0].token = t0;

-    tokens[i][0].dqc = dqcoeff[rc];

-    token_cache[rc] = vp9_pt_energy_class[t0];

+    token_cache[rc] = vp9_pt_energy_class[vp9_get_token(qcoeff[rc])];

   }

-  tokens[eob][0].token = EOB_TOKEN;

-  tokens[eob][0].qc = 0;

-  tokens[eob][0].dqc = 0;

-  tokens[eob][1] = tokens[eob][0];

   final_eob = 0;

   // Initial RD cost.

@@ -144,8 +131,8 @@

     const int token_tree_sel_cur = (x_prev == 0);

     token_costs_cur = token_costs + band_cur;

     if (x == 0) {  // No need to search

-      rate0 =

-          (*token_costs_cur)[token_tree_sel_cur][ctx_cur][tokens[i][0].token];

+      const int token = vp9_get_token(x);

+      rate0 = (*token_costs_cur)[token_tree_sel_cur][ctx_cur][token];

       accu_rate += rate0;

       x_prev = 0;

       // Note: accu_error does not change.

@@ -205,6 +192,7 @@

         int rdcost_better_for_x1, eob_rdcost_better_for_x1;

         int dqc0, dqc1;

         int64_t best_eob_cost_cur;

+        int use_x1;

         // Calculate RD Cost effect on the next coeff for the two candidates.

         int64_t next_bits0 = 0;

@@ -214,6 +202,8 @@

         if (i < default_eob - 1) {

           int ctx_next, token_tree_sel_next;

           const int band_next = band_translate[i + 1];

+          const int token_next =

+              (i + 1 != eob) ? vp9_get_token(qcoeff[scan[i + 1]]) : EOB_TOKEN;

           unsigned int(

               *const token_costs_next)[2][COEFF_CONTEXTS][ENTROPY_TOKENS] =

               token_costs + band_next;

@@ -220,15 +210,15 @@

           token_cache[rc] = vp9_pt_energy_class[t0];

           ctx_next = get_coef_context(nb, token_cache, i + 1);

           token_tree_sel_next = (x == 0);

-          next_bits0 = (*token_costs_next)[token_tree_sel_next][ctx_next]

-                                          [tokens[i + 1][0].token];

+          next_bits0 =

+              (*token_costs_next)[token_tree_sel_next][ctx_next][token_next];

           next_eob_bits0 =

               (*token_costs_next)[token_tree_sel_next][ctx_next][EOB_TOKEN];

           token_cache[rc] = vp9_pt_energy_class[t1];

           ctx_next = get_coef_context(nb, token_cache, i + 1);

           token_tree_sel_next = (x1 == 0);

-          next_bits1 = (*token_costs_next)[token_tree_sel_next][ctx_next]

-                                          [tokens[i + 1][0].token];

+          next_bits1 =

+              (*token_costs_next)[token_tree_sel_next][ctx_next][token_next];

           if (x1 != 0) {

             next_eob_bits1 =

                 (*token_costs_next)[token_tree_sel_next][ctx_next][EOB_TOKEN];

@@ -279,21 +269,20 @@

         assert(accu_error >= 0);

         x_prev = qcoeff[rc];  // Update based on selected quantized value.

-        best_eob_cost_cur = eob_cost0;

-        tokens[i][1].token = t0;

-        tokens[i][1].qc = x;

-        tokens[i][1].dqc = dqc0;

-        if ((x1 != 0) && eob_rdcost_better_for_x1) {

-          best_eob_cost_cur = eob_cost1;

-          tokens[i][1].token = t1;

-          tokens[i][1].qc = x1;

-          tokens[i][1].dqc = dqc1;

-        }

+        use_x1 = (x1 != 0) && eob_rdcost_better_for_x1;

+        best_eob_cost_cur = use_x1 ? eob_cost1 : eob_cost0;

         // Determine whether to move the eob position to i+1

         if (best_eob_cost_cur < best_block_rd_cost) {

           best_block_rd_cost = best_eob_cost_cur;

           final_eob = i + 1;

+          if (use_x1) {

+            before_best_eob_qc = x1;

+            before_best_eob_dqc = dqc1;

+          } else {

+            before_best_eob_qc = x;

+            before_best_eob_dqc = dqc0;

+          }

         }

       }

     }

@@ -301,11 +290,11 @@

   assert(final_eob <= eob);

   if (final_eob > 0) {

     int rc;

-    assert(tokens[final_eob - 1][1].qc != 0);

+    assert(before_best_eob_qc != 0);

     i = final_eob - 1;

     rc = scan[i];

-    qcoeff[rc] = tokens[i][1].qc;

-    dqcoeff[rc] = tokens[i][1].dqc;

+    qcoeff[rc] = before_best_eob_qc;

+    dqcoeff[rc] = before_best_eob_dqc;

   }

   for (i = final_eob; i < eob; i++) {

     int rc = scan[i];