Merge "Take out skip_encode speed feature in vp10"

vp10/encoder/block.h

@@ -115,7 +115,6 @@ struct macroblock {
 
   // indicate if it is in the rd search loop or encoding process
   int use_lp32x32fdct;
-  int skip_encode;
 
   // use fast quantization process
   int quant_fp;

vp10/encoder/encodeframe.c

@@ -2585,20 +2585,6 @@ static TX_MODE select_tx_mode(const VP10_COMP *cpi, MACROBLOCKD *const xd) {
     return cpi->common.tx_mode;
 }
 
-static int get_skip_encode_frame(const VP10_COMMON *cm, ThreadData *const td) {
-  unsigned int intra_count = 0, inter_count = 0;
-  int j;
-
-  for (j = 0; j < INTRA_INTER_CONTEXTS; ++j) {
-    intra_count += td->counts->intra_inter[j][0];
-    inter_count += td->counts->intra_inter[j][1];
-  }
-
-  return (intra_count << 2) < inter_count &&
-         cm->frame_type != KEY_FRAME &&
-         cm->show_frame;
-}
-
 void vp10_init_tile_data(VP10_COMP *cpi) {
   VP10_COMMON *const cm = &cpi->common;
   const int tile_cols = 1 << cm->log2_tile_cols;
@@ -2690,7 +2676,6 @@ static int input_fpmb_stats(FIRSTPASS_MB_STATS *firstpass_mb_stats,
 #endif
 
 static void encode_frame_internal(VP10_COMP *cpi) {
-  SPEED_FEATURES *const sf = &cpi->sf;
   ThreadData *const td = &cpi->td;
   MACROBLOCK *const x = &td->mb;
   VP10_COMMON *const cm = &cpi->common;
@@ -2766,9 +2751,6 @@ static void encode_frame_internal(VP10_COMP *cpi) {
     cpi->time_encode_sb_row += vpx_usec_timer_elapsed(&emr_timer);
   }
 
-  sf->skip_encode_frame = sf->skip_encode_sb ?
-      get_skip_encode_frame(cm, td) : 0;
-
 #if 0
   // Keep record of the total distortion this time around for future use
   cpi->last_frame_distortion = cpi->frame_distortion;
@@ -2962,11 +2944,6 @@ static void encode_superblock(VP10_COMP *cpi, ThreadData *td,
   x->skip_optimize = ctx->is_coded;
   ctx->is_coded = 1;
   x->use_lp32x32fdct = cpi->sf.use_lp32x32fdct;
-  x->skip_encode = (!output_enabled && cpi->sf.skip_encode_frame &&
-                    x->q_index < QIDX_SKIP_THRESH);
-
-  if (x->skip_encode)
-    return;
 
   if (!is_inter_block(mbmi)) {
     int plane;

vp10/encoder/encodemb.c

@@ -797,7 +797,7 @@ static void encode_block(int plane, int block, BLOCK_SIZE plane_bsize,
   if (p->eobs[block])
     *(args->skip) = 0;
 
-  if (x->skip_encode || p->eobs[block] == 0)
+  if (p->eobs[block] == 0)
     return;
 #if CONFIG_VP9_HIGHBITDEPTH
   if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
@@ -945,8 +945,7 @@ void vp10_encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
   src_diff = &p->src_diff[4 * (j * diff_stride + i)];
 
   mode = plane == 0 ? get_y_mode(xd->mi[0], block) : mbmi->uv_mode;
-  vp10_predict_intra_block(xd, bwl, tx_size, mode, x->skip_encode ? src : dst,
-                          x->skip_encode ? src_stride : dst_stride,
+  vp10_predict_intra_block(xd, bwl, tx_size, mode, dst, dst_stride,
                           dst, dst_stride, i, j, plane);
 
 #if CONFIG_VP9_HIGHBITDEPTH
@@ -1036,7 +1035,7 @@ void vp10_encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
                              pd->dequant, eob, scan_order->scan,
                              scan_order->iscan);
       }
-      if (!x->skip_encode && *eob)
+      if (*eob)
         vp10_inv_txfm_add_32x32(dqcoeff, dst, dst_stride, *eob, tx_type);
       break;
     case TX_16X16:
@@ -1049,7 +1048,7 @@ void vp10_encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
                        pd->dequant, eob, scan_order->scan,
                        scan_order->iscan);
       }
-      if (!x->skip_encode && *eob)
+      if (*eob)
         vp10_inv_txfm_add_16x16(dqcoeff, dst, dst_stride, *eob, tx_type);
       break;
     case TX_8X8:
@@ -1062,7 +1061,7 @@ void vp10_encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
                        pd->dequant, eob, scan_order->scan,
                        scan_order->iscan);
       }
-      if (!x->skip_encode && *eob)
+      if (*eob)
         vp10_inv_txfm_add_8x8(dqcoeff, dst, dst_stride, *eob, tx_type);
       break;
     case TX_4X4:
@@ -1076,7 +1075,7 @@ void vp10_encode_block_intra(int plane, int block, BLOCK_SIZE plane_bsize,
                        scan_order->iscan);
       }
 
-      if (!x->skip_encode && *eob) {
+      if (*eob) {
         // this is like vp10_short_idct4x4 but has a special case around eob<=1
         // which is significant (not just an optimization) for the lossless
         // case.

vp10/encoder/firstpass.c

@@ -614,7 +614,6 @@ void vp10_first_pass(VP10_COMP *cpi, const struct lookahead_entry *source) {
                      cm->mi_rows, cm->mi_cols);
 
       // Do intra 16x16 prediction.
-      x->skip_encode = 0;
       xd->mi[0]->mbmi.mode = DC_PRED;
       xd->mi[0]->mbmi.tx_size = use_dc_pred ?
          (bsize >= BLOCK_16X16 ? TX_16X16 : TX_8X8) : TX_4X4;

vp10/encoder/rdopt.c

@@ -438,19 +438,6 @@ static void dist_block(MACROBLOCK *x, int plane, int block, TX_SIZE tx_size,
                               &this_sse) >> shift;
 #endif  // CONFIG_VP9_HIGHBITDEPTH
   *out_sse = this_sse >> shift;
-
-  if (x->skip_encode && !is_inter_block(&xd->mi[0]->mbmi)) {
-    // TODO(jingning): tune the model to better capture the distortion.
-    int64_t p = (pd->dequant[1] * pd->dequant[1] *
-                    (1 << ss_txfrm_size)) >>
-#if CONFIG_VP9_HIGHBITDEPTH
-                        (shift + 2 + (bd - 8) * 2);
-#else
-                        (shift + 2);
-#endif  // CONFIG_VP9_HIGHBITDEPTH
-    *out_dist += (p >> 4);
-    *out_sse  += p;
-  }
 }
 
 static int rate_block(int plane, int block, BLOCK_SIZE plane_bsize,
@@ -903,9 +890,7 @@ static int64_t rd_pick_intra4x4block(VP10_COMP *cpi, MACROBLOCK *x,
             vp10_raster_block_offset_int16(BLOCK_8X8, block, p->src_diff);
         tran_low_t *const coeff = BLOCK_OFFSET(x->plane[0].coeff, block);
         xd->mi[0]->bmi[block].as_mode = mode;
-        vp10_predict_intra_block(xd, 1, TX_4X4, mode,
-                                x->skip_encode ? src : dst,
-                                x->skip_encode ? src_stride : dst_stride,
+        vp10_predict_intra_block(xd, 1, TX_4X4, mode, dst, dst_stride,
                                 dst, dst_stride, col + idx, row + idy, 0);
         vpx_subtract_block(4, 4, src_diff, 8, src, src_stride, dst, dst_stride);
 
@@ -961,7 +946,7 @@ static int64_t rd_pick_intra4x4block(VP10_COMP *cpi, MACROBLOCK *x,
     {}
   }
 
-  if (best_rd >= rd_thresh || x->skip_encode)
+  if (best_rd >= rd_thresh)
     return best_rd;
 
   for (idy = 0; idy < num_4x4_blocks_high * 4; ++idy)
@@ -2739,7 +2724,6 @@ void vp10_rd_pick_intra_mode_sb(VP10_COMP *cpi, MACROBLOCK *x,
   int y_skip = 0, uv_skip = 0;
   int64_t dist_y = 0, dist_uv = 0;
   TX_SIZE max_uv_tx_size;
-  x->skip_encode = 0;
   ctx->skip = 0;
   xd->mi[0]->mbmi.ref_frame[0] = INTRA_FRAME;
   xd->mi[0]->mbmi.ref_frame[1] = NONE;
@@ -2969,8 +2953,6 @@ void vp10_rd_pick_inter_mode_sb(VP10_COMP *cpi,
 
   vp10_zero(best_mbmode);
 
-  x->skip_encode = sf->skip_encode_frame && x->q_index < QIDX_SKIP_THRESH;
-
   for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; ++i)
     filter_cache[i] = INT64_MAX;
 
@@ -3563,8 +3545,6 @@ void vp10_rd_pick_inter_mode_sb_seg_skip(VP10_COMP *cpi,
   int rate2 = 0;
   const int64_t distortion2 = 0;
 
-  x->skip_encode = cpi->sf.skip_encode_frame && x->q_index < QIDX_SKIP_THRESH;
-
   estimate_ref_frame_costs(cm, xd, segment_id, ref_costs_single, ref_costs_comp,
                            &comp_mode_p);
 
@@ -3688,7 +3668,6 @@ void vp10_rd_pick_inter_mode_sub8x8(VP10_COMP *cpi,
   int internal_active_edge =
     vp10_active_edge_sb(cpi, mi_row, mi_col) && vp10_internal_image_edge(cpi);
 
-  x->skip_encode = sf->skip_encode_frame && x->q_index < QIDX_SKIP_THRESH;
   memset(x->zcoeff_blk[TX_4X4], 0, 4);
   vp10_zero(best_mbmode);
 

vp10/encoder/speed_features.c

@@ -301,7 +301,6 @@ static void set_rt_speed_feature(VP10_COMP *cpi, SPEED_FEATURES *sf,
     sf->use_square_partition_only = 1;
     sf->disable_filter_search_var_thresh = 100;
     sf->use_uv_intra_rd_estimate = 1;
-    sf->skip_encode_sb = 1;
     sf->mv.subpel_iters_per_step = 1;
     sf->adaptive_rd_thresh = 4;
     sf->mode_skip_start = 6;
@@ -384,7 +383,6 @@ static void set_rt_speed_feature(VP10_COMP *cpi, SPEED_FEATURES *sf,
     // Turn on this to use non-RD key frame coding mode.
     sf->mv.search_method = NSTEP;
     sf->mv.reduce_first_step_size = 1;
-    sf->skip_encode_sb = 0;
   }
 
   if (speed >= 7) {
@@ -481,7 +479,6 @@ void vp10_set_speed_features_framesize_independent(VP10_COMP *cpi) {
     sf->intra_uv_mode_mask[i] = INTRA_ALL;
   }
   sf->use_rd_breakout = 0;
-  sf->skip_encode_sb = 0;
   sf->use_uv_intra_rd_estimate = 0;
   sf->allow_skip_recode = 0;
   sf->lpf_pick = LPF_PICK_FROM_FULL_IMAGE;

vp10/encoder/speed_features.h

@@ -223,11 +223,6 @@ typedef struct SPEED_FEATURES {
   // mode to be evaluated. A high value means we will be faster.
   int adaptive_rd_thresh;
 
-  // Enables skipping the reconstruction step (idct, recon) in the
-  // intermediate steps assuming the last frame didn't have too many intra
-  // blocks and the q is less than a threshold.
-  int skip_encode_sb;
-  int skip_encode_frame;
   // Speed feature to allow or disallow skipping of recode at block
   // level within a frame.
   int allow_skip_recode;