libvpx: Pull from upstream

source/libvpx/vp9/encoder/vp9_rd.c

 /*
  *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS.  All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
@@ skipping 75 matching lines @@
             vp9_cost_tokens((int *)c[t][i][j][k][0][l], probs,
                             vp9_coef_tree);
             vp9_cost_tokens_skip((int *)c[t][i][j][k][1][l], probs,
                                  vp9_coef_tree);
             assert(c[t][i][j][k][0][l][EOB_TOKEN] ==
                    c[t][i][j][k][1][l][EOB_TOKEN]);
           }
 }
 
 // Values are now correlated to quantizer.
-static int sad_per_bit16lut[QINDEX_RANGE];
-static int sad_per_bit4lut[QINDEX_RANGE];
+static int sad_per_bit16lut_8[QINDEX_RANGE];
+static int sad_per_bit4lut_8[QINDEX_RANGE];
 
-void vp9_init_me_luts() {
+#if CONFIG_VP9_HIGHBITDEPTH
+static int sad_per_bit16lut_10[QINDEX_RANGE];
+static int sad_per_bit4lut_10[QINDEX_RANGE];
+static int sad_per_bit16lut_12[QINDEX_RANGE];
+static int sad_per_bit4lut_12[QINDEX_RANGE];
+#endif
+
+static void init_me_luts_bd(int *bit16lut, int *bit4lut, int range,
+                            vpx_bit_depth_t bit_depth) {
   int i;
-
   // Initialize the sad lut tables using a formulaic calculation for now.
   // This is to make it easier to resolve the impact of experimental changes
   // to the quantizer tables.
-  for (i = 0; i < QINDEX_RANGE; ++i) {
-    const double q = vp9_convert_qindex_to_q(i);
-    sad_per_bit16lut[i] = (int)(0.0418 * q + 2.4107);
-    sad_per_bit4lut[i] = (int)(0.063 * q + 2.742);
+  for (i = 0; i < range; i++) {
+    const double q = vp9_convert_qindex_to_q(i, bit_depth);
+    bit16lut[i] = (int)(0.0418 * q + 2.4107);
+    bit4lut[i] = (int)(0.063 * q + 2.742);
   }
 }
 
+void vp9_init_me_luts() {
+  init_me_luts_bd(sad_per_bit16lut_8, sad_per_bit4lut_8, QINDEX_RANGE,
+                  VPX_BITS_8);
+#if CONFIG_VP9_HIGHBITDEPTH
+  init_me_luts_bd(sad_per_bit16lut_10, sad_per_bit4lut_10, QINDEX_RANGE,
+                  VPX_BITS_10);
+  init_me_luts_bd(sad_per_bit16lut_12, sad_per_bit4lut_12, QINDEX_RANGE,
+                  VPX_BITS_12);
+#endif
+}
+
 static const int rd_boost_factor[16] = {
   64, 32, 32, 32, 24, 16, 12, 12,
   8, 8, 4, 4, 2, 2, 1, 0
 };
 static const int rd_frame_type_factor[FRAME_UPDATE_TYPES] = {
-128, 144, 128, 128, 144
+  128, 144, 128, 128, 144
 };
 
 int vp9_compute_rd_mult(const VP9_COMP *cpi, int qindex) {
-  const int q = vp9_dc_quant(qindex, 0);
+  const int q = vp9_dc_quant(qindex, 0, cpi->common.bit_depth);
+#if CONFIG_VP9_HIGHBITDEPTH
+  int rdmult = 0;
+  switch (cpi->common.bit_depth) {
+    case VPX_BITS_8:
+      rdmult = 88 * q * q / 24;
+      break;
+    case VPX_BITS_10:
+      rdmult = ROUND_POWER_OF_TWO(88 * q * q / 24, 4);
+      break;
+    case VPX_BITS_12:
+      rdmult = ROUND_POWER_OF_TWO(88 * q * q / 24, 8);
+      break;
+    default:
+      assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
+      return -1;
+  }
+#else
   int rdmult = 88 * q * q / 24;
-
+#endif
   if (cpi->oxcf.pass == 2 && (cpi->common.frame_type != KEY_FRAME)) {
     const GF_GROUP *const gf_group = &cpi->twopass.gf_group;
     const FRAME_UPDATE_TYPE frame_type = gf_group->update_type[gf_group->index];
     const int boost_index = MIN(15, (cpi->rc.gfu_boost / 100));
 
     rdmult = (rdmult * rd_frame_type_factor[frame_type]) >> 7;
     rdmult += ((rdmult * rd_boost_factor[boost_index]) >> 7);
   }
   return rdmult;
 }
 
-static int compute_rd_thresh_factor(int qindex) {
+static int compute_rd_thresh_factor(int qindex, vpx_bit_depth_t bit_depth) {
+  double q;
+#if CONFIG_VP9_HIGHBITDEPTH
+  switch (bit_depth) {
+    case VPX_BITS_8:
+      q = vp9_dc_quant(qindex, 0, VPX_BITS_8) / 4.0;
+      break;
+    case VPX_BITS_10:
+      q = vp9_dc_quant(qindex, 0, VPX_BITS_10) / 16.0;
+      break;
+    case VPX_BITS_12:
+      q = vp9_dc_quant(qindex, 0, VPX_BITS_12) / 64.0;
+      break;
+    default:
+      assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
+      return -1;
+  }
+#else
+  (void) bit_depth;
+  q = vp9_dc_quant(qindex, 0, VPX_BITS_8) / 4.0;
+#endif
   // TODO(debargha): Adjust the function below.
-  const int q = (int)(pow(vp9_dc_quant(qindex, 0) / 4.0, RD_THRESH_POW) * 5.12);
-  return MAX(q, 8);
+  return MAX((int)(pow(q, RD_THRESH_POW) * 5.12), 8);
 }
 
 void vp9_initialize_me_consts(VP9_COMP *cpi, int qindex) {
-  cpi->mb.sadperbit16 = sad_per_bit16lut[qindex];
-  cpi->mb.sadperbit4 = sad_per_bit4lut[qindex];
+#if CONFIG_VP9_HIGHBITDEPTH
+  switch (cpi->common.bit_depth) {
+    case VPX_BITS_8:
+      cpi->mb.sadperbit16 = sad_per_bit16lut_8[qindex];
+      cpi->mb.sadperbit4 = sad_per_bit4lut_8[qindex];
+      break;
+    case VPX_BITS_10:
+      cpi->mb.sadperbit16 = sad_per_bit16lut_10[qindex];
+      cpi->mb.sadperbit4 = sad_per_bit4lut_10[qindex];
+      break;
+    case VPX_BITS_12:
+      cpi->mb.sadperbit16 = sad_per_bit16lut_12[qindex];
+      cpi->mb.sadperbit4 = sad_per_bit4lut_12[qindex];
+      break;
+    default:
+      assert(0 && "bit_depth should be VPX_BITS_8, VPX_BITS_10 or VPX_BITS_12");
+  }
+#else
+  cpi->mb.sadperbit16 = sad_per_bit16lut_8[qindex];
+  cpi->mb.sadperbit4 = sad_per_bit4lut_8[qindex];
+#endif
 }
 
 static void set_block_thresholds(const VP9_COMMON *cm, RD_OPT *rd) {
   int i, bsize, segment_id;
 
   for (segment_id = 0; segment_id < MAX_SEGMENTS; ++segment_id) {
     const int qindex =
         clamp(vp9_get_qindex(&cm->seg, segment_id, cm->base_qindex) +
-                  cm->y_dc_delta_q,
-              0, MAXQ);
-    const int q = compute_rd_thresh_factor(qindex);
+              cm->y_dc_delta_q, 0, MAXQ);
+    const int q = compute_rd_thresh_factor(qindex, cm->bit_depth);
 
     for (bsize = 0; bsize < BLOCK_SIZES; ++bsize) {
       // Threshold here seems unnecessarily harsh but fine given actual
       // range of values used for cpi->sf.thresh_mult[].
       const int t = q * rd_thresh_block_size_factor[bsize];
       const int thresh_max = INT_MAX / t;
 
       if (bsize >= BLOCK_8X8) {
         for (i = 0; i < MAX_MODES; ++i)
           rd->threshes[segment_id][bsize][i] =
@@ skipping 191 matching lines @@
     default:
       assert(0 && "Invalid transform size.");
       break;
   }
 }
 
 void vp9_mv_pred(VP9_COMP *cpi, MACROBLOCK *x,
                  uint8_t *ref_y_buffer, int ref_y_stride,
                  int ref_frame, BLOCK_SIZE block_size) {
   MACROBLOCKD *xd = &x->e_mbd;
-  MB_MODE_INFO *mbmi = &xd->mi[0]->mbmi;
+  MB_MODE_INFO *mbmi = &xd->mi[0].src_mi->mbmi;
   int i;
   int zero_seen = 0;
   int best_index = 0;
   int best_sad = INT_MAX;
   int this_sad = INT_MAX;
   int max_mv = 0;
   uint8_t *src_y_ptr = x->plane[0].src.buf;
   uint8_t *ref_y_ptr;
   const int num_mv_refs = MAX_MV_REF_CANDIDATES +
                     (cpi->sf.adaptive_motion_search &&
@@ skipping 58 matching lines @@
 const YV12_BUFFER_CONFIG *vp9_get_scaled_ref_frame(const VP9_COMP *cpi,
                                                    int ref_frame) {
   const VP9_COMMON *const cm = &cpi->common;
   const int ref_idx = cm->ref_frame_map[get_ref_frame_idx(cpi, ref_frame)];
   const int scaled_idx = cpi->scaled_ref_idx[ref_frame - 1];
   return (scaled_idx != ref_idx) ? &cm->frame_bufs[scaled_idx].buf : NULL;
 }
 
 int vp9_get_switchable_rate(const VP9_COMP *cpi) {
   const MACROBLOCKD *const xd = &cpi->mb.e_mbd;
-  const MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
+  const MB_MODE_INFO *const mbmi = &xd->mi[0].src_mi->mbmi;
   const int ctx = vp9_get_pred_context_switchable_interp(xd);
   return SWITCHABLE_INTERP_RATE_FACTOR *
              cpi->switchable_interp_costs[ctx][mbmi->interp_filter];
 }
 
 void vp9_set_rd_speed_thresholds(VP9_COMP *cpi) {
   int i;
   RD_OPT *const rd = &cpi->rd;
   SPEED_FEATURES *const sf = &cpi->sf;
 
   // Set baseline threshold values.
   for (i = 0; i < MAX_MODES; ++i)
     rd->thresh_mult[i] = cpi->oxcf.mode == BEST ? -500 : 0;
 
-  rd->thresh_mult[THR_NEARESTMV] = 0;
-  rd->thresh_mult[THR_NEARESTG] = 0;
-  rd->thresh_mult[THR_NEARESTA] = 0;
+  if (sf->adaptive_rd_thresh) {
+    rd->thresh_mult[THR_NEARESTMV] = 300;
+    rd->thresh_mult[THR_NEARESTG] = 300;
+    rd->thresh_mult[THR_NEARESTA] = 300;
+  } else {
+    rd->thresh_mult[THR_NEARESTMV] = 0;
+    rd->thresh_mult[THR_NEARESTG] = 0;
+    rd->thresh_mult[THR_NEARESTA] = 0;
+  }
 
   rd->thresh_mult[THR_DC] += 1000;
 
   rd->thresh_mult[THR_NEWMV] += 1000;
   rd->thresh_mult[THR_NEWA] += 1000;
   rd->thresh_mult[THR_NEWG] += 1000;
 
   // Adjust threshold only in real time mode, which only uses last
   // reference frame.
   rd->thresh_mult[THR_NEWMV] += sf->elevate_newmv_thresh;
@@ skipping 18 matching lines @@
   rd->thresh_mult[THR_COMP_ZEROGA] += 2500;
 
   rd->thresh_mult[THR_H_PRED] += 2000;
   rd->thresh_mult[THR_V_PRED] += 2000;
   rd->thresh_mult[THR_D45_PRED ] += 2500;
   rd->thresh_mult[THR_D135_PRED] += 2500;
   rd->thresh_mult[THR_D117_PRED] += 2500;
   rd->thresh_mult[THR_D153_PRED] += 2500;
   rd->thresh_mult[THR_D207_PRED] += 2500;
   rd->thresh_mult[THR_D63_PRED] += 2500;
-
-  // Disable frame modes if flags not set.
-  if (!(cpi->ref_frame_flags & VP9_LAST_FLAG)) {
-    rd->thresh_mult[THR_NEWMV    ] = INT_MAX;
-    rd->thresh_mult[THR_NEARESTMV] = INT_MAX;
-    rd->thresh_mult[THR_ZEROMV   ] = INT_MAX;
-    rd->thresh_mult[THR_NEARMV   ] = INT_MAX;
-  }
-  if (!(cpi->ref_frame_flags & VP9_GOLD_FLAG)) {
-    rd->thresh_mult[THR_NEARESTG ] = INT_MAX;
-    rd->thresh_mult[THR_ZEROG    ] = INT_MAX;
-    rd->thresh_mult[THR_NEARG    ] = INT_MAX;
-    rd->thresh_mult[THR_NEWG     ] = INT_MAX;
-  }
-  if (!(cpi->ref_frame_flags & VP9_ALT_FLAG)) {
-    rd->thresh_mult[THR_NEARESTA ] = INT_MAX;
-    rd->thresh_mult[THR_ZEROA    ] = INT_MAX;
-    rd->thresh_mult[THR_NEARA    ] = INT_MAX;
-    rd->thresh_mult[THR_NEWA     ] = INT_MAX;
-  }
-
-  if ((cpi->ref_frame_flags & (VP9_LAST_FLAG | VP9_ALT_FLAG)) !=
-      (VP9_LAST_FLAG | VP9_ALT_FLAG)) {
-    rd->thresh_mult[THR_COMP_ZEROLA   ] = INT_MAX;
-    rd->thresh_mult[THR_COMP_NEARESTLA] = INT_MAX;
-    rd->thresh_mult[THR_COMP_NEARLA   ] = INT_MAX;
-    rd->thresh_mult[THR_COMP_NEWLA    ] = INT_MAX;
-  }
-  if ((cpi->ref_frame_flags & (VP9_GOLD_FLAG | VP9_ALT_FLAG)) !=
-      (VP9_GOLD_FLAG | VP9_ALT_FLAG)) {
-    rd->thresh_mult[THR_COMP_ZEROGA   ] = INT_MAX;
-    rd->thresh_mult[THR_COMP_NEARESTGA] = INT_MAX;
-    rd->thresh_mult[THR_COMP_NEARGA   ] = INT_MAX;
-    rd->thresh_mult[THR_COMP_NEWGA    ] = INT_MAX;
-  }
 }
 
 void vp9_set_rd_speed_thresholds_sub8x8(VP9_COMP *cpi) {
   const SPEED_FEATURES *const sf = &cpi->sf;
   RD_OPT *const rd = &cpi->rd;
   int i;
 
   for (i = 0; i < MAX_REFS; ++i)
     rd->thresh_mult_sub8x8[i] = cpi->oxcf.mode == BEST ? -500 : 0;
 
   rd->thresh_mult_sub8x8[THR_LAST] += 2500;
   rd->thresh_mult_sub8x8[THR_GOLD] += 2500;
   rd->thresh_mult_sub8x8[THR_ALTR] += 2500;
   rd->thresh_mult_sub8x8[THR_INTRA] += 2500;
   rd->thresh_mult_sub8x8[THR_COMP_LA] += 4500;
   rd->thresh_mult_sub8x8[THR_COMP_GA] += 4500;
 
   // Check for masked out split cases.
   for (i = 0; i < MAX_REFS; ++i)
     if (sf->disable_split_mask & (1 << i))
       rd->thresh_mult_sub8x8[i] = INT_MAX;
-
-  // Disable mode test if frame flag is not set.
-  if (!(cpi->ref_frame_flags & VP9_LAST_FLAG))
-    rd->thresh_mult_sub8x8[THR_LAST] = INT_MAX;
-  if (!(cpi->ref_frame_flags & VP9_GOLD_FLAG))
-    rd->thresh_mult_sub8x8[THR_GOLD] = INT_MAX;
-  if (!(cpi->ref_frame_flags & VP9_ALT_FLAG))
-    rd->thresh_mult_sub8x8[THR_ALTR] = INT_MAX;
-  if ((cpi->ref_frame_flags & (VP9_LAST_FLAG | VP9_ALT_FLAG)) !=
-      (VP9_LAST_FLAG | VP9_ALT_FLAG))
-    rd->thresh_mult_sub8x8[THR_COMP_LA] = INT_MAX;
-  if ((cpi->ref_frame_flags & (VP9_GOLD_FLAG | VP9_ALT_FLAG)) !=
-      (VP9_GOLD_FLAG | VP9_ALT_FLAG))
-    rd->thresh_mult_sub8x8[THR_COMP_GA] = INT_MAX;
 }