vp9_encodeframe.c 123 KB
Newer Older
John Koleszar's avatar
John Koleszar committed
1
/*
2
 *  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
John Koleszar's avatar
John Koleszar committed
3
 *
4
 *  Use of this source code is governed by a BSD-style license
5 6
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
7
 *  in the file PATENTS.  All contributing project authors may
8
 *  be found in the AUTHORS file in the root of the source tree.
John Koleszar's avatar
John Koleszar committed
9 10
 */

Dmitry Kovalev's avatar
Dmitry Kovalev committed
11 12 13 14
#include <limits.h>
#include <math.h>
#include <stdio.h>

Jim Bankoski's avatar
Jim Bankoski committed
15
#include "./vp9_rtcd.h"
Dmitry Kovalev's avatar
Dmitry Kovalev committed
16 17 18 19
#include "./vpx_config.h"

#include "vpx_ports/vpx_timer.h"

20
#include "vp9/common/vp9_common.h"
Yaowu Xu's avatar
Yaowu Xu committed
21
#include "vp9/common/vp9_entropy.h"
22
#include "vp9/common/vp9_entropymode.h"
23
#include "vp9/common/vp9_idct.h"
Dmitry Kovalev's avatar
Dmitry Kovalev committed
24 25 26
#include "vp9/common/vp9_mvref_common.h"
#include "vp9/common/vp9_pred_common.h"
#include "vp9/common/vp9_quant_common.h"
27
#include "vp9/common/vp9_reconintra.h"
Dmitry Kovalev's avatar
Dmitry Kovalev committed
28
#include "vp9/common/vp9_reconinter.h"
29
#include "vp9/common/vp9_seg_common.h"
30
#include "vp9/common/vp9_systemdependent.h"
31
#include "vp9/common/vp9_tile_common.h"
32

33
#include "vp9/encoder/vp9_aq_complexity.h"
Marco Paniconi's avatar
Marco Paniconi committed
34 35
#include "vp9/encoder/vp9_aq_cyclicrefresh.h"
#include "vp9/encoder/vp9_aq_variance.h"
Dmitry Kovalev's avatar
Dmitry Kovalev committed
36 37 38
#include "vp9/encoder/vp9_encodeframe.h"
#include "vp9/encoder/vp9_encodemb.h"
#include "vp9/encoder/vp9_encodemv.h"
39
#include "vp9/encoder/vp9_extend.h"
40
#include "vp9/encoder/vp9_pickmode.h"
Dmitry Kovalev's avatar
Dmitry Kovalev committed
41 42
#include "vp9/encoder/vp9_rdopt.h"
#include "vp9/encoder/vp9_segmentation.h"
43
#include "vp9/encoder/vp9_tokenize.h"
44

45 46 47 48 49 50
#define GF_ZEROMV_ZBIN_BOOST 0
#define LF_ZEROMV_ZBIN_BOOST 0
#define MV_ZBIN_BOOST        0
#define SPLIT_MV_ZBIN_BOOST  0
#define INTRA_ZBIN_BOOST     0

Jim Bankoski's avatar
Jim Bankoski committed
51
static void encode_superblock(VP9_COMP *cpi, TOKENEXTRA **t, int output_enabled,
52 53
                              int mi_row, int mi_col, BLOCK_SIZE bsize,
                              PICK_MODE_CONTEXT *ctx);
54

55
// Motion vector component magnitude threshold for defining fast motion.
56
#define FAST_MOTION_MV_THRESH 24
57

58 59 60 61
// This is used as a reference when computing the source variance for the
//  purposes of activity masking.
// Eventually this should be replaced by custom no-reference routines,
//  which will be faster.
62 63 64 65 66 67 68 69 70 71 72
static const uint8_t VP9_VAR_OFFS[64] = {
  128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128,
  128, 128, 128, 128, 128, 128, 128, 128
};

73 74 75 76 77 78
typedef struct {
  unsigned int sse;
  int sum;
  unsigned int var;
} diff;

79
static unsigned int get_sby_perpixel_variance(VP9_COMP *cpi,
80
                                              const struct buf_2d *ref,
81
                                              BLOCK_SIZE bs) {
82 83 84
  unsigned int sse;
  const unsigned int var = cpi->fn_ptr[bs].vf(ref->buf, ref->stride,
                                              VP9_VAR_OFFS, 0, &sse);
85
  return ROUND_POWER_OF_TWO(var, num_pels_log2_lookup[bs]);
86 87
}

88
static unsigned int get_sby_perpixel_diff_variance(VP9_COMP *cpi,
89 90
                                                   const struct buf_2d *ref,
                                                   int mi_row, int mi_col,
91
                                                   BLOCK_SIZE bs) {
92 93 94 95 96 97
  const YV12_BUFFER_CONFIG *last = get_ref_frame_buffer(cpi, LAST_FRAME);
  const uint8_t* last_y = &last->y_buffer[mi_row * MI_SIZE * last->y_stride +
                                              mi_col * MI_SIZE];
  unsigned int sse;
  const unsigned int var = cpi->fn_ptr[bs].vf(ref->buf, ref->stride,
                                              last_y, last->y_stride, &sse);
98 99 100 101 102 103
  return ROUND_POWER_OF_TWO(var, num_pels_log2_lookup[bs]);
}

static BLOCK_SIZE get_rd_var_based_fixed_partition(VP9_COMP *cpi,
                                                   int mi_row,
                                                   int mi_col) {
104
  unsigned int var = get_sby_perpixel_diff_variance(cpi, &cpi->mb.plane[0].src,
105 106 107
                                                    mi_row, mi_col,
                                                    BLOCK_64X64);
  if (var < 8)
108
    return BLOCK_64X64;
109
  else if (var < 128)
110
    return BLOCK_32X32;
111 112 113 114
  else if (var < 2048)
    return BLOCK_16X16;
  else
    return BLOCK_8X8;
115 116
}

117 118 119
static BLOCK_SIZE get_nonrd_var_based_fixed_partition(VP9_COMP *cpi,
                                                      int mi_row,
                                                      int mi_col) {
120
  unsigned int var = get_sby_perpixel_diff_variance(cpi, &cpi->mb.plane[0].src,
121 122
                                                    mi_row, mi_col,
                                                    BLOCK_64X64);
123
  if (var < 4)
124
    return BLOCK_64X64;
125
  else if (var < 10)
126 127
    return BLOCK_32X32;
  else
128
    return BLOCK_16X16;
129 130
}

131 132
// Lighter version of set_offsets that only sets the mode info
// pointers.
133
static INLINE void set_modeinfo_offsets(VP9_COMMON *const cm,
134 135 136
                                        MACROBLOCKD *const xd,
                                        int mi_row,
                                        int mi_col) {
137 138 139
  const int idx_str = xd->mi_stride * mi_row + mi_col;
  xd->mi = cm->mi_grid_visible + idx_str;
  xd->mi[0] = cm->mi + idx_str;
140 141
}

Alex Converse's avatar
Alex Converse committed
142
static int is_block_in_mb_map(const VP9_COMP *cpi, int mi_row, int mi_col,
143
                              BLOCK_SIZE bsize) {
Alex Converse's avatar
Alex Converse committed
144
  const VP9_COMMON *const cm = &cpi->common;
145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167
  const int mb_rows = cm->mb_rows;
  const int mb_cols = cm->mb_cols;
  const int mb_row = mi_row >> 1;
  const int mb_col = mi_col >> 1;
  const int mb_width = num_8x8_blocks_wide_lookup[bsize] >> 1;
  const int mb_height = num_8x8_blocks_high_lookup[bsize] >> 1;
  int r, c;
  if (bsize <= BLOCK_16X16) {
    return cpi->active_map[mb_row * mb_cols + mb_col];
  }
  for (r = 0; r < mb_height; ++r) {
    for (c = 0; c < mb_width; ++c) {
      int row = mb_row + r;
      int col = mb_col + c;
      if (row >= mb_rows || col >= mb_cols)
        continue;
      if (cpi->active_map[row * mb_cols + col])
        return 1;
    }
  }
  return 0;
}

Alex Converse's avatar
Alex Converse committed
168 169 170 171 172 173 174 175 176 177
static int check_active_map(const VP9_COMP *cpi, const MACROBLOCK *x,
                            int mi_row, int mi_col,
                            BLOCK_SIZE bsize) {
  if (cpi->active_map_enabled && !x->e_mbd.lossless) {
    return is_block_in_mb_map(cpi, mi_row, mi_col, bsize);
  } else {
    return 1;
  }
}

178 179 180 181 182 183 184 185 186 187
static void set_offsets(VP9_COMP *cpi, const TileInfo *const tile,
                        int mi_row, int mi_col, BLOCK_SIZE bsize) {
  MACROBLOCK *const x = &cpi->mb;
  VP9_COMMON *const cm = &cpi->common;
  MACROBLOCKD *const xd = &x->e_mbd;
  MB_MODE_INFO *mbmi;
  const int mi_width = num_8x8_blocks_wide_lookup[bsize];
  const int mi_height = num_8x8_blocks_high_lookup[bsize];
  const struct segmentation *const seg = &cm->seg;

188
  set_skip_context(xd, mi_row, mi_col);
189 190

  // Activity map pointer
Alex Converse's avatar
Alex Converse committed
191
  x->in_active_map = check_active_map(cpi, x, mi_row, mi_col, bsize);
192 193 194

  set_modeinfo_offsets(cm, xd, mi_row, mi_col);

195
  mbmi = &xd->mi[0]->mbmi;
196 197

  // Set up destination pointers.
198
  vp9_setup_dst_planes(xd->plane, get_frame_new_buffer(cm), mi_row, mi_col);
199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215

  // Set up limit values for MV components.
  // Mv beyond the range do not produce new/different prediction block.
  x->mv_row_min = -(((mi_row + mi_height) * MI_SIZE) + VP9_INTERP_EXTEND);
  x->mv_col_min = -(((mi_col + mi_width) * MI_SIZE) + VP9_INTERP_EXTEND);
  x->mv_row_max = (cm->mi_rows - mi_row) * MI_SIZE + VP9_INTERP_EXTEND;
  x->mv_col_max = (cm->mi_cols - mi_col) * MI_SIZE + VP9_INTERP_EXTEND;

  // Set up distance of MB to edge of frame in 1/8th pel units.
  assert(!(mi_col & (mi_width - 1)) && !(mi_row & (mi_height - 1)));
  set_mi_row_col(xd, tile, mi_row, mi_height, mi_col, mi_width,
                 cm->mi_rows, cm->mi_cols);

  // Set up source buffers.
  vp9_setup_src_planes(x, cpi->Source, mi_row, mi_col);

  // R/D setup.
Dmitry Kovalev's avatar
Dmitry Kovalev committed
216 217
  x->rddiv = cpi->rd.RDDIV;
  x->rdmult = cpi->rd.RDMULT;
218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234

  // Setup segment ID.
  if (seg->enabled) {
    if (cpi->oxcf.aq_mode != VARIANCE_AQ) {
      const uint8_t *const map = seg->update_map ? cpi->segmentation_map
                                                 : cm->last_frame_seg_map;
      mbmi->segment_id = vp9_get_segment_id(cm, map, bsize, mi_row, mi_col);
    }
    vp9_init_plane_quantizers(cpi, x);

    x->encode_breakout = cpi->segment_encode_breakout[mbmi->segment_id];
  } else {
    mbmi->segment_id = 0;
    x->encode_breakout = cpi->encode_breakout;
  }
}

235
static void duplicate_mode_info_in_sb(VP9_COMMON * const cm,
236 237 238 239 240 241 242 243 244 245
                                     MACROBLOCKD *const xd,
                                     int mi_row,
                                     int mi_col,
                                     BLOCK_SIZE bsize) {
  const int block_width = num_8x8_blocks_wide_lookup[bsize];
  const int block_height = num_8x8_blocks_high_lookup[bsize];
  int i, j;
  for (j = 0; j < block_height; ++j)
    for (i = 0; i < block_width; ++i) {
      if (mi_row + j < cm->mi_rows && mi_col + i < cm->mi_cols)
246
        xd->mi[j * xd->mi_stride + i] = xd->mi[0];
247 248 249 250 251 252 253 254 255
    }
}

static void set_block_size(VP9_COMP * const cpi,
                           int mi_row, int mi_col,
                           BLOCK_SIZE bsize) {
  if (cpi->common.mi_cols > mi_col && cpi->common.mi_rows > mi_row) {
    MACROBLOCKD *const xd = &cpi->mb.e_mbd;
    set_modeinfo_offsets(&cpi->common, xd, mi_row, mi_col);
256
    xd->mi[0]->mbmi.sb_type = bsize;
257
    duplicate_mode_info_in_sb(&cpi->common, xd, mi_row, mi_col, bsize);
258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374
  }
}

typedef struct {
  int64_t sum_square_error;
  int64_t sum_error;
  int count;
  int variance;
} var;

typedef struct {
  var none;
  var horz[2];
  var vert[2];
} partition_variance;

typedef struct {
  partition_variance part_variances;
  var split[4];
} v8x8;

typedef struct {
  partition_variance part_variances;
  v8x8 split[4];
} v16x16;

typedef struct {
  partition_variance part_variances;
  v16x16 split[4];
} v32x32;

typedef struct {
  partition_variance part_variances;
  v32x32 split[4];
} v64x64;

typedef struct {
  partition_variance *part_variances;
  var *split[4];
} variance_node;

typedef enum {
  V16X16,
  V32X32,
  V64X64,
} TREE_LEVEL;

static void tree_to_node(void *data, BLOCK_SIZE bsize, variance_node *node) {
  int i;
  switch (bsize) {
    case BLOCK_64X64: {
      v64x64 *vt = (v64x64 *) data;
      node->part_variances = &vt->part_variances;
      for (i = 0; i < 4; i++)
        node->split[i] = &vt->split[i].part_variances.none;
      break;
    }
    case BLOCK_32X32: {
      v32x32 *vt = (v32x32 *) data;
      node->part_variances = &vt->part_variances;
      for (i = 0; i < 4; i++)
        node->split[i] = &vt->split[i].part_variances.none;
      break;
    }
    case BLOCK_16X16: {
      v16x16 *vt = (v16x16 *) data;
      node->part_variances = &vt->part_variances;
      for (i = 0; i < 4; i++)
        node->split[i] = &vt->split[i].part_variances.none;
      break;
    }
    case BLOCK_8X8: {
      v8x8 *vt = (v8x8 *) data;
      node->part_variances = &vt->part_variances;
      for (i = 0; i < 4; i++)
        node->split[i] = &vt->split[i];
      break;
    }
    default: {
      assert(0);
    }
  }
}

// Set variance values given sum square error, sum error, count.
static void fill_variance(int64_t s2, int64_t s, int c, var *v) {
  v->sum_square_error = s2;
  v->sum_error = s;
  v->count = c;
  if (c > 0)
    v->variance = (int)(256 *
                        (v->sum_square_error - v->sum_error * v->sum_error /
                         v->count) / v->count);
  else
    v->variance = 0;
}

void sum_2_variances(const var *a, const var *b, var *r) {
  fill_variance(a->sum_square_error + b->sum_square_error,
                a->sum_error + b->sum_error, a->count + b->count, r);
}

static void fill_variance_tree(void *data, BLOCK_SIZE bsize) {
  variance_node node;
  tree_to_node(data, bsize, &node);
  sum_2_variances(node.split[0], node.split[1], &node.part_variances->horz[0]);
  sum_2_variances(node.split[2], node.split[3], &node.part_variances->horz[1]);
  sum_2_variances(node.split[0], node.split[2], &node.part_variances->vert[0]);
  sum_2_variances(node.split[1], node.split[3], &node.part_variances->vert[1]);
  sum_2_variances(&node.part_variances->vert[0], &node.part_variances->vert[1],
                  &node.part_variances->none);
}

static int set_vt_partitioning(VP9_COMP *cpi,
                               void *data,
                               BLOCK_SIZE bsize,
                               int mi_row,
Yaowu Xu's avatar
Yaowu Xu committed
375
                               int mi_col) {
376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391
  VP9_COMMON * const cm = &cpi->common;
  variance_node vt;
  const int block_width = num_8x8_blocks_wide_lookup[bsize];
  const int block_height = num_8x8_blocks_high_lookup[bsize];
  // TODO(debargha): Choose this more intelligently.
  const int64_t threshold_multiplier = 25;
  int64_t threshold = threshold_multiplier * cpi->common.base_qindex;
  assert(block_height == block_width);

  tree_to_node(data, bsize, &vt);

  // Split none is available only if we have more than half a block size
  // in width and height inside the visible image.
  if (mi_col + block_width / 2 < cm->mi_cols &&
      mi_row + block_height / 2 < cm->mi_rows &&
      vt.part_variances->none.variance < threshold) {
Yaowu Xu's avatar
Yaowu Xu committed
392
    set_block_size(cpi, mi_row, mi_col, bsize);
393 394 395 396 397 398 399 400
    return 1;
  }

  // Vertical split is available on all but the bottom border.
  if (mi_row + block_height / 2 < cm->mi_rows &&
      vt.part_variances->vert[0].variance < threshold &&
      vt.part_variances->vert[1].variance < threshold) {
    BLOCK_SIZE subsize = get_subsize(bsize, PARTITION_VERT);
Yaowu Xu's avatar
Yaowu Xu committed
401 402
    set_block_size(cpi, mi_row, mi_col, subsize);
    set_block_size(cpi, mi_row, mi_col + block_width / 2, subsize);
403 404 405 406 407 408 409 410
    return 1;
  }

  // Horizontal split is available on all but the right border.
  if (mi_col + block_width / 2 < cm->mi_cols &&
      vt.part_variances->horz[0].variance < threshold &&
      vt.part_variances->horz[1].variance < threshold) {
    BLOCK_SIZE subsize = get_subsize(bsize, PARTITION_HORZ);
Yaowu Xu's avatar
Yaowu Xu committed
411 412
    set_block_size(cpi, mi_row, mi_col, subsize);
    set_block_size(cpi, mi_row + block_height / 2, mi_col, subsize);
413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450
    return 1;
  }
  return 0;
}

// TODO(debargha): Fix this function and make it work as expected.
static void choose_partitioning(VP9_COMP *cpi,
                                const TileInfo *const tile,
                                int mi_row, int mi_col) {
  VP9_COMMON * const cm = &cpi->common;
  MACROBLOCK *x = &cpi->mb;
  MACROBLOCKD *xd = &cpi->mb.e_mbd;

  int i, j, k;
  v64x64 vt;
  uint8_t *s;
  const uint8_t *d;
  int sp;
  int dp;
  int pixels_wide = 64, pixels_high = 64;
  int_mv nearest_mv, near_mv;
  const YV12_BUFFER_CONFIG *yv12 = get_ref_frame_buffer(cpi, LAST_FRAME);
  const struct scale_factors *const sf = &cm->frame_refs[LAST_FRAME - 1].sf;

  vp9_zero(vt);
  set_offsets(cpi, tile, mi_row, mi_col, BLOCK_64X64);

  if (xd->mb_to_right_edge < 0)
    pixels_wide += (xd->mb_to_right_edge >> 3);
  if (xd->mb_to_bottom_edge < 0)
    pixels_high += (xd->mb_to_bottom_edge >> 3);

  s = x->plane[0].src.buf;
  sp = x->plane[0].src.stride;

  if (cm->frame_type != KEY_FRAME) {
    vp9_setup_pre_planes(xd, 0, yv12, mi_row, mi_col, sf);

451 452
    xd->mi[0]->mbmi.ref_frame[0] = LAST_FRAME;
    xd->mi[0]->mbmi.sb_type = BLOCK_64X64;
453
    vp9_find_best_ref_mvs(xd, cm->allow_high_precision_mv,
454
                          xd->mi[0]->mbmi.ref_mvs[LAST_FRAME],
455 456
                          &nearest_mv, &near_mv);

457
    xd->mi[0]->mbmi.mv[0] = nearest_mv;
458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480
    vp9_build_inter_predictors_sby(xd, mi_row, mi_col, BLOCK_64X64);

    d = xd->plane[0].dst.buf;
    dp = xd->plane[0].dst.stride;
  } else {
    d = VP9_VAR_OFFS;
    dp = 0;
  }

  // Fill in the entire tree of 8x8 variances for splits.
  for (i = 0; i < 4; i++) {
    const int x32_idx = ((i & 1) << 5);
    const int y32_idx = ((i >> 1) << 5);
    for (j = 0; j < 4; j++) {
      const int x16_idx = x32_idx + ((j & 1) << 4);
      const int y16_idx = y32_idx + ((j >> 1) << 4);
      v16x16 *vst = &vt.split[i].split[j];
      for (k = 0; k < 4; k++) {
        int x_idx = x16_idx + ((k & 1) << 3);
        int y_idx = y16_idx + ((k >> 1) << 3);
        unsigned int sse = 0;
        int sum = 0;
        if (x_idx < pixels_wide && y_idx < pixels_high)
481 482
          vp9_get_sse_sum_8x8(s + y_idx * sp + x_idx, sp,
                              d + y_idx * dp + x_idx, dp, &sse, &sum);
483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498
        fill_variance(sse, sum, 64, &vst->split[k].part_variances.none);
      }
    }
  }
  // Fill the rest of the variance tree by summing split partition values.
  for (i = 0; i < 4; i++) {
    for (j = 0; j < 4; j++) {
      fill_variance_tree(&vt.split[i].split[j], BLOCK_16X16);
    }
    fill_variance_tree(&vt.split[i], BLOCK_32X32);
  }
  fill_variance_tree(&vt, BLOCK_64X64);

  // Now go through the entire structure,  splitting every block size until
  // we get to one that's got a variance lower than our threshold,  or we
  // hit 8x8.
Yaowu Xu's avatar
Yaowu Xu committed
499 500
  if (!set_vt_partitioning(cpi, &vt, BLOCK_64X64,
                           mi_row, mi_col)) {
501 502 503
    for (i = 0; i < 4; ++i) {
      const int x32_idx = ((i & 1) << 2);
      const int y32_idx = ((i >> 1) << 2);
Yaowu Xu's avatar
Yaowu Xu committed
504 505
      if (!set_vt_partitioning(cpi, &vt.split[i], BLOCK_32X32,
                               (mi_row + y32_idx), (mi_col + x32_idx))) {
506 507 508 509 510 511 512 513 514
        for (j = 0; j < 4; ++j) {
          const int x16_idx = ((j & 1) << 1);
          const int y16_idx = ((j >> 1) << 1);
          // NOTE: This is a temporary hack to disable 8x8 partitions,
          // since it works really bad - possibly due to a bug
#define DISABLE_8X8_VAR_BASED_PARTITION
#ifdef DISABLE_8X8_VAR_BASED_PARTITION
          if (mi_row + y32_idx + y16_idx + 1 < cm->mi_rows &&
              mi_row + x32_idx + x16_idx + 1 < cm->mi_cols) {
Yaowu Xu's avatar
Yaowu Xu committed
515
            set_block_size(cpi,
516 517 518 519 520 521 522
                           (mi_row + y32_idx + y16_idx),
                           (mi_col + x32_idx + x16_idx),
                           BLOCK_16X16);
          } else {
            for (k = 0; k < 4; ++k) {
              const int x8_idx = (k & 1);
              const int y8_idx = (k >> 1);
Yaowu Xu's avatar
Yaowu Xu committed
523
              set_block_size(cpi,
524 525 526 527 528 529 530 531 532 533 534 535 536
                             (mi_row + y32_idx + y16_idx + y8_idx),
                             (mi_col + x32_idx + x16_idx + x8_idx),
                             BLOCK_8X8);
            }
          }
#else
          if (!set_vt_partitioning(cpi, &vt.split[i].split[j], tile,
                                   BLOCK_16X16,
                                   (mi_row + y32_idx + y16_idx),
                                   (mi_col + x32_idx + x16_idx), 2)) {
            for (k = 0; k < 4; ++k) {
              const int x8_idx = (k & 1);
              const int y8_idx = (k >> 1);
Yaowu Xu's avatar
Yaowu Xu committed
537
              set_block_size(cpi,
538 539 540 541 542 543 544 545 546 547 548 549
                             (mi_row + y32_idx + y16_idx + y8_idx),
                             (mi_col + x32_idx + x16_idx + x8_idx),
                             BLOCK_8X8);
            }
          }
#endif
        }
      }
    }
  }
}

550
// Original activity measure from Tim T's code.
551
static unsigned int tt_activity_measure(MACROBLOCK *x) {
John Koleszar's avatar
John Koleszar committed
552
  unsigned int sse;
553 554 555 556 557 558 559 560 561
  // TODO: This could also be done over smaller areas (8x8), but that would
  // require extensive changes elsewhere, as lambda is assumed to be fixed
  // over an entire MB in most of the code.
  // Another option is to compute four 8x8 variances, and pick a single
  // lambda using a non-linear combination (e.g., the smallest, or second
  // smallest, etc.).
  const unsigned int act = vp9_variance16x16(x->plane[0].src.buf,
                                             x->plane[0].src.stride,
                                             VP9_VAR_OFFS, 0, &sse) << 4;
562
  // If the region is flat, lower the activity some more.
563
  return act < (8 << 12) ? MIN(act, 5 << 12) : act;
564 565
}

Jim Bankoski's avatar
Jim Bankoski committed
566
static void update_state(VP9_COMP *cpi, PICK_MODE_CONTEXT *ctx,
567 568
                         int mi_row, int mi_col, BLOCK_SIZE bsize,
                         int output_enabled) {
Ronald S. Bultje's avatar
Ronald S. Bultje committed
569
  int i, x_idx, y;
570
  VP9_COMMON *const cm = &cpi->common;
Dmitry Kovalev's avatar
Dmitry Kovalev committed
571
  RD_OPT *const rd_opt = &cpi->rd;
572 573
  MACROBLOCK *const x = &cpi->mb;
  MACROBLOCKD *const xd = &x->e_mbd;
574 575
  struct macroblock_plane *const p = x->plane;
  struct macroblockd_plane *const pd = xd->plane;
John Koleszar's avatar
John Koleszar committed
576
  MODE_INFO *mi = &ctx->mic;
577 578
  MB_MODE_INFO *const mbmi = &xd->mi[0]->mbmi;
  MODE_INFO *mi_addr = xd->mi[0];
579
  const struct segmentation *const seg = &cm->seg;
580

581
  const int mis = cm->mi_stride;
Jim Bankoski's avatar
Jim Bankoski committed
582
  const int mi_width = num_8x8_blocks_wide_lookup[bsize];
Jim Bankoski's avatar
Jim Bankoski committed
583
  const int mi_height = num_8x8_blocks_high_lookup[bsize];
584
  int max_plane;
Adrian Grange's avatar
Adrian Grange committed
585

586
  assert(mi->mbmi.sb_type == bsize);
587

588 589
  *mi_addr = *mi;

Paul Wilkins's avatar
Paul Wilkins committed
590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605
  // If segmentation in use
  if (seg->enabled && output_enabled) {
    // For in frame complexity AQ copy the segment id from the segment map.
    if (cpi->oxcf.aq_mode == COMPLEXITY_AQ) {
      const uint8_t *const map = seg->update_map ? cpi->segmentation_map
                                                 : cm->last_frame_seg_map;
      mi_addr->mbmi.segment_id =
        vp9_get_segment_id(cm, map, bsize, mi_row, mi_col);
    }
    // Else for cyclic refresh mode update the segment map, set the segment id
    // and then update the quantizer.
    else if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) {
      vp9_cyclic_refresh_update_segment(cpi, &xd->mi[0]->mbmi,
                                        mi_row, mi_col, bsize, 1);
      vp9_init_plane_quantizers(cpi, x);
    }
606
  }
607

608 609
  max_plane = is_inter_block(mbmi) ? MAX_MB_PLANE : 1;
  for (i = 0; i < max_plane; ++i) {
610
    p[i].coeff = ctx->coeff_pbuf[i][1];
611
    p[i].qcoeff = ctx->qcoeff_pbuf[i][1];
612
    pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][1];
613
    p[i].eobs = ctx->eobs_pbuf[i][1];
614 615
  }

616 617
  for (i = max_plane; i < MAX_MB_PLANE; ++i) {
    p[i].coeff = ctx->coeff_pbuf[i][2];
618
    p[i].qcoeff = ctx->qcoeff_pbuf[i][2];
619
    pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][2];
620
    p[i].eobs = ctx->eobs_pbuf[i][2];
621 622
  }

John Koleszar's avatar
John Koleszar committed
623 624
  // Restore the coding context of the MB to that that was in place
  // when the mode was picked for it
625 626
  for (y = 0; y < mi_height; y++)
    for (x_idx = 0; x_idx < mi_width; x_idx++)
James Zern's avatar
James Zern committed
627
      if ((xd->mb_to_right_edge >> (3 + MI_SIZE_LOG2)) + mi_width > x_idx
628
        && (xd->mb_to_bottom_edge >> (3 + MI_SIZE_LOG2)) + mi_height > y) {
629
        xd->mi[x_idx + y * mis] = mi_addr;
630
      }
631

632
  if (cpi->oxcf.aq_mode)
Dmitry Kovalev's avatar
Dmitry Kovalev committed
633
    vp9_init_plane_quantizers(cpi, x);
634

635 636
  // FIXME(rbultje) I'm pretty sure this should go to the end of this block
  // (i.e. after the output_enabled)
637 638
  if (bsize < BLOCK_32X32) {
    if (bsize < BLOCK_16X16)
639 640
      ctx->tx_rd_diff[ALLOW_16X16] = ctx->tx_rd_diff[ALLOW_8X8];
    ctx->tx_rd_diff[ALLOW_32X32] = ctx->tx_rd_diff[ALLOW_16X16];
Ronald S. Bultje's avatar
Ronald S. Bultje committed
641
  }
Adrian Grange's avatar
Adrian Grange committed
642

643
  if (is_inter_block(mbmi) && mbmi->sb_type < BLOCK_8X8) {
644 645
    mbmi->mv[0].as_int = mi->bmi[3].as_mv[0].as_int;
    mbmi->mv[1].as_int = mi->bmi[3].as_mv[1].as_int;
John Koleszar's avatar
John Koleszar committed
646 647
  }

648
  x->skip = ctx->skip;
649
  vpx_memcpy(x->zcoeff_blk[mbmi->tx_size], ctx->zcoeff_blk,
650
             sizeof(uint8_t) * ctx->num_4x4_blk);
651

Ronald S. Bultje's avatar
Ronald S. Bultje committed
652 653 654
  if (!output_enabled)
    return;

655
  if (!vp9_segfeature_active(&cm->seg, mbmi->segment_id, SEG_LVL_SKIP)) {
656
    for (i = 0; i < TX_MODES; i++)
Dmitry Kovalev's avatar
Dmitry Kovalev committed
657
      rd_opt->tx_select_diff[i] += ctx->tx_rd_diff[i];
658 659
  }

660
#if CONFIG_INTERNAL_STATS
661
  if (frame_is_intra_only(cm)) {
John Koleszar's avatar
John Koleszar committed
662
    static const int kf_mode_index[] = {
663 664 665 666
      THR_DC        /*DC_PRED*/,
      THR_V_PRED    /*V_PRED*/,
      THR_H_PRED    /*H_PRED*/,
      THR_D45_PRED  /*D45_PRED*/,
John Koleszar's avatar
John Koleszar committed
667 668 669
      THR_D135_PRED /*D135_PRED*/,
      THR_D117_PRED /*D117_PRED*/,
      THR_D153_PRED /*D153_PRED*/,
Dmitry Kovalev's avatar
Dmitry Kovalev committed
670
      THR_D207_PRED /*D207_PRED*/,
671 672
      THR_D63_PRED  /*D63_PRED*/,
      THR_TM        /*TM_PRED*/,
John Koleszar's avatar
John Koleszar committed
673
    };
674
    ++cpi->mode_chosen_counts[kf_mode_index[mbmi->mode]];
John Koleszar's avatar
John Koleszar committed
675 676
  } else {
    // Note how often each mode chosen as best
677 678 679 680
    ++cpi->mode_chosen_counts[ctx->best_mode_index];
  }
#endif
  if (!frame_is_intra_only(cm)) {
681
    if (is_inter_block(mbmi)) {
682
      vp9_update_mv_count(cm, xd);
683 684 685 686 687

      if (cm->interp_filter == SWITCHABLE) {
        const int ctx = vp9_get_pred_context_switchable_interp(xd);
        ++cm->counts.switchable_interp[ctx][mbmi->interp_filter];
      }
688
    }
Adrian Grange's avatar
Adrian Grange committed
689

Dmitry Kovalev's avatar
Dmitry Kovalev committed
690 691 692
    rd_opt->comp_pred_diff[SINGLE_REFERENCE] += ctx->single_pred_diff;
    rd_opt->comp_pred_diff[COMPOUND_REFERENCE] += ctx->comp_pred_diff;
    rd_opt->comp_pred_diff[REFERENCE_MODE_SELECT] += ctx->hybrid_pred_diff;
693

694
    for (i = 0; i < SWITCHABLE_FILTER_CONTEXTS; ++i)
Dmitry Kovalev's avatar
Dmitry Kovalev committed
695
      rd_opt->filter_diff[i] += ctx->best_filter_diff[i];
John Koleszar's avatar
John Koleszar committed
696
  }
Adrian Grange's avatar
Adrian Grange committed
697 698
}

Jim Bankoski's avatar
Jim Bankoski committed
699
void vp9_setup_src_planes(MACROBLOCK *x, const YV12_BUFFER_CONFIG *src,
700 701 702 703 704
                          int mi_row, int mi_col) {
  uint8_t *const buffers[4] = {src->y_buffer, src->u_buffer, src->v_buffer,
                               src->alpha_buffer};
  const int strides[4] = {src->y_stride, src->uv_stride, src->uv_stride,
                          src->alpha_stride};
705 706
  int i;

707 708 709
  // Set current frame pointer.
  x->e_mbd.cur_buf = src;

710 711
  for (i = 0; i < MAX_MB_PLANE; i++)
    setup_pred_plane(&x->plane[i].src, buffers[i], strides[i], mi_row, mi_col,
Jim Bankoski's avatar
Jim Bankoski committed
712
                     NULL, x->e_mbd.plane[i].subsampling_x,
713
                     x->e_mbd.plane[i].subsampling_y);
John Koleszar's avatar
John Koleszar committed
714 715
}

716 717 718 719
static void rd_pick_sb_modes(VP9_COMP *cpi, const TileInfo *const tile,
                             int mi_row, int mi_col,
                             int *totalrate, int64_t *totaldist,
                             BLOCK_SIZE bsize, PICK_MODE_CONTEXT *ctx,
720
                             int64_t best_rd, int block) {
721 722 723
  VP9_COMMON *const cm = &cpi->common;
  MACROBLOCK *const x = &cpi->mb;
  MACROBLOCKD *const xd = &x->e_mbd;
724
  MB_MODE_INFO *mbmi;
725 726
  struct macroblock_plane *const p = x->plane;
  struct macroblockd_plane *const pd = xd->plane;
727
  const AQ_MODE aq_mode = cpi->oxcf.aq_mode;
728
  int i, orig_rdmult;
729 730
  double rdmult_ratio;

731
  vp9_clear_system_state();
732
  rdmult_ratio = 1.0;  // avoid uninitialized warnings
Ronald S. Bultje's avatar
Ronald S. Bultje committed
733

734
  // Use the lower precision, but faster, 32x32 fdct for mode selection.
735
  x->use_lp32x32fdct = 1;
736

737 738 739
  // TODO(JBB): Most other places in the code instead of calling the function
  // and then checking if its not the first 8x8 we put the check in the
  // calling function.  Do that here.
740
  if (bsize < BLOCK_8X8) {
741 742
    // When ab_index = 0 all sub-blocks are handled, so for ab_index != 0
    // there is nothing to be done.
743
    if (block != 0) {
744 745
      *totalrate = 0;
      *totaldist = 0;
746
      return;
747
    }
748
  }
749

James Zern's avatar
James Zern committed
750
  set_offsets(cpi, tile, mi_row, mi_col, bsize);
751
  mbmi = &xd->mi[0]->mbmi;
752
  mbmi->sb_type = bsize;
753

754
  for (i = 0; i < MAX_MB_PLANE; ++i) {
755
    p[i].coeff = ctx->coeff_pbuf[i][0];
756
    p[i].qcoeff = ctx->qcoeff_pbuf[i][0];
757
    pd[i].dqcoeff = ctx->dqcoeff_pbuf[i][0];
758
    p[i].eobs = ctx->eobs_pbuf[i][0];
759
  }
760
  ctx->is_coded = 0;
761
  x->skip_recode = 0;
762

763
  // Set to zero to make sure we do not use the previous encoded frame stats
764
  mbmi->skip = 0;
765

766
  x->source_variance = get_sby_perpixel_variance(cpi, &x->plane[0].src, bsize);
767

768 769 770
  // Save rdmult before it might be changed, so it can be restored later.
  orig_rdmult = x->rdmult;

771
  if (aq_mode == VARIANCE_AQ) {
772 773
    const int energy = bsize <= BLOCK_16X16 ? x->mb_energy
                                            : vp9_block_energy(cpi, x, bsize);
Paul Wilkins's avatar
Paul Wilkins committed
774 775 776
    if (cm->frame_type == KEY_FRAME ||
        cpi->refresh_alt_ref_frame ||
        (cpi->refresh_golden_frame && !cpi->rc.is_src_frame_alt_ref)) {
777
      mbmi->segment_id = vp9_vaq_segment_id(energy);
Paul Wilkins's avatar
Paul Wilkins committed
778 779 780
    } else {
      const uint8_t *const map = cm->seg.update_map ? cpi->segmentation_map
                                                    : cm->last_frame_seg_map;
781
      mbmi->segment_id = vp9_get_segment_id(cm, map, bsize, mi_row, mi_col);
Paul Wilkins's avatar
Paul Wilkins committed
782 783
    }

784
    rdmult_ratio = vp9_vaq_rdmult_ratio(energy);
Dmitry Kovalev's avatar
Dmitry Kovalev committed
785
    vp9_init_plane_quantizers(cpi, x);
786
    vp9_clear_system_state();
787
    x->rdmult = (int)round(x->rdmult * rdmult_ratio);
788
  } else if (aq_mode == COMPLEXITY_AQ) {
789 790
    const int mi_offset = mi_row * cm->mi_cols + mi_col;
    unsigned char complexity = cpi->complexity_map[mi_offset];
Paul Wilkins's avatar
Paul Wilkins committed
791 792
    const int is_edge = (mi_row <= 1) || (mi_row >= (cm->mi_rows - 2)) ||
                        (mi_col <= 1) || (mi_col >= (cm->mi_cols - 2));
793 794
    if (!is_edge && (complexity > 128))
      x->rdmult += ((x->rdmult * (complexity - 128)) / 256);
795
  } else if (aq_mode == CYCLIC_REFRESH_AQ) {
796
    const uint8_t *const map = cm->seg.update_map ? cpi->segmentation_map
797
                                                  : cm->last_frame_seg_map;
798 799
    // If segment 1, use rdmult for that segment.
    if (vp9_get_segment_id(cm, map, bsize, mi_row, mi_col))
800
      x->rdmult = vp9_cyclic_refresh_get_rdmult(cpi->cyclic_refresh);
801
  }
802

803 804
  // Find best coding mode & reconstruct the MB so it is available
  // as a predictor for MBs that follow in the SB
805
  if (frame_is_intra_only(cm)) {
806 807
    vp9_rd_pick_intra_mode_sb(cpi, x, totalrate, totaldist, bsize, ctx,
                              best_rd);
808 809
  } else {
    if (bsize >= BLOCK_8X8)
James Zern's avatar
James Zern committed
810 811
      vp9_rd_pick_inter_mode_sb(cpi, x, tile, mi_row, mi_col,
                                totalrate, totaldist, bsize, ctx, best_rd);
812
    else
James Zern's avatar
James Zern committed
813
      vp9_rd_pick_inter_mode_sub8x8(cpi, x, tile, mi_row, mi_col, totalrate,
814 815
                                    totaldist, bsize, ctx, best_rd);
  }
816

817 818 819 820 821
  x->rdmult = orig_rdmult;

  if (aq_mode == VARIANCE_AQ && *totalrate != INT_MAX) {
    vp9_clear_system_state();
    *totalrate = (int)round(*totalrate * rdmult_ratio);
Paul Wilkins's avatar
Paul Wilkins committed
822
  }
Ronald S. Bultje's avatar
Ronald S. Bultje committed
823
}
Adrian Grange's avatar
Adrian Grange committed
824

825 826
static void update_stats(VP9_COMP *cpi) {
  VP9_COMMON *const cm = &cpi->common;
827 828
  const MACROBLOCK *const x = &cpi->mb;
  const MACROBLOCKD *const xd = &x->e_mbd;
829
  const MODE_INFO *const mi = xd->mi[0];
830
  const MB_MODE_INFO *const mbmi = &mi->mbmi;
Adrian Grange's avatar
Adrian Grange committed
831

832
  if (!frame_is_intra_only(cm)) {
833
    const int seg_ref_active = vp9_segfeature_active(&cm->seg, mbmi->segment_id,
834
                                                     SEG_LVL_REF_FRAME);
835 836 837 838 839 840 841 842 843 844 845
    if (!seg_ref_active) {
      FRAME_COUNTS *const counts = &cm->counts;
      const int inter_block = is_inter_block(mbmi);

      counts->intra_inter[vp9_get_intra_inter_context(xd)][inter_block]++;

      // If the segment reference feature is enabled we have only a single
      // reference frame allowed for the segment so exclude it from
      // the reference frame counts used to work out probabilities.
      if (inter_block) {
        const MV_REFERENCE_FRAME ref0 = mbmi->ref_frame[0];
846

847 848 849 850 851 852 853 854 855 856 857 858 859 860
        if (cm->reference_mode == REFERENCE_MODE_SELECT)
          counts->comp_inter[vp9_get_reference_mode_context(cm, xd)]
                            [has_second_ref(mbmi)]++;

        if (has_second_ref(mbmi)) {
          counts->comp_ref[vp9_get_pred_context_comp_ref_p(cm, xd)]
                          [ref0 == GOLDEN_FRAME]++;
        } else {
          counts->single_ref[vp9_get_pred_context_single_ref_p1(xd)][0]
                            [ref0 != LAST_FRAME]++;
          if (ref0 != LAST_FRAME)
            counts->single_ref[vp9_get_pred_context_single_ref_p2(xd)][1]
                              [ref0 != GOLDEN_FRAME]++;
        }
Ronald S. Bultje's avatar
Ronald S. Bultje committed
861
      }
Ronald S. Bultje's avatar
Ronald S. Bultje committed
862
    }
Ronald S. Bultje's avatar
Ronald S. Bultje committed
863 864
  }
}
John Koleszar's avatar
John Koleszar committed
865

866 867 868
static void restore_context(VP9_COMP *cpi, int mi_row, int mi_col,
                            ENTROPY_CONTEXT a[16 * MAX_MB_PLANE],
                            ENTROPY_CONTEXT l[16 * MAX_MB_PLANE],
Jim Bankoski's avatar
Jim Bankoski committed
869
                            PARTITION_CONTEXT sa[8], PARTITION_CONTEXT sl[8],
870
                            BLOCK_SIZE bsize) {
Jim Bankoski's avatar
Jim Bankoski committed
871 872
  MACROBLOCK *const x = &cpi->mb;
  MACROBLOCKD *const xd = &x->e_mbd;
873
  int p;
Jim Bankoski's avatar
Jim Bankoski committed
874 875
  const int num_4x4_blocks_wide = num_4x4_blocks_wide_lookup[bsize];
  const int num_4x4_blocks_high = num_4x4_blocks_high_lookup[bsize];
Jim Bankoski's avatar
Jim Bankoski committed
876 877
  int mi_width = num_8x8_blocks_wide_lookup[bsize];
  int mi_height = num_8x8_blocks_high_lookup[bsize];
878
  for (p = 0; p < MAX_MB_PLANE; p++) {
Jim Bankoski's avatar
Jim Bankoski committed
879
    vpx_memcpy(
880
        xd->above_context[p] + ((mi_col * 2) >> xd->plane[p].subsampling_x),
Jim Bankoski's avatar
Jim Bankoski committed
881 882 883
        a + num_4x4_blocks_wide * p,
        (sizeof(ENTROPY_CONTEXT) * num_4x4_blocks_wide) >>
        xd->plane[p].subsampling_x);
Jim Bankoski's avatar
Jim Bankoski committed
884
    vpx_memcpy(
885
        xd->left_context[p]
Jim Bankoski's avatar
Jim Bankoski committed
886 887 888 889 890
            + ((mi_row & MI_MASK) * 2 >> xd->plane[p].subsampling_y),
        l + num_4x4_blocks_high * p,
        (sizeof(ENTROPY_CONTEXT) * num_4x4_blocks_high) >>
        xd->plane[p].subsampling_y);
  }
891 892 893 894
  vpx_memcpy(xd->above_seg_context + mi_col, sa,
             sizeof(*xd->above_seg_context) * mi_width);
  vpx_memcpy(xd->left_seg_context + (mi_row & MI_MASK), sl,
             sizeof(xd->left_seg_context[0]) * mi_height);
895
}
Jim Bankoski's avatar
Jim Bankoski committed
896
static void save_context(VP9_COMP *cpi, int mi_row, int mi_col,
Jim Bankoski's avatar
Jim Bankoski committed
897 898 899
                         ENTROPY_CONTEXT a[16 * MAX_MB_PLANE],
                         ENTROPY_CONTEXT l[16 * MAX_MB_PLANE],
                         PARTITION_CONTEXT sa[8], PARTITION_CONTEXT sl[8],
900
                         BLOCK_SIZE bsize) {
Jim Bankoski's avatar
Jim Bankoski committed
901 902
  const MACROBLOCK *const x = &cpi->mb;
  const MACROBLOCKD *const xd = &x->e_mbd;
Jim Bankoski's avatar
Jim Bankoski committed
903
  int p;
Jim Bankoski's avatar
Jim Bankoski committed
904 905
  const int num_4x4_blocks_wide = num_4x4_blocks_wide_lookup[bsize];
  const int num_4x4_blocks_high = num_4x4_blocks_high_lookup[bsize];
Jim Bankoski's avatar
Jim Bankoski committed
906 907
  int mi_width = num_8x8_blocks_wide_lookup[bsize];
  int mi_height = num_8x8_blocks_high_lookup[bsize];
Jim Bankoski's avatar
Jim Bankoski committed
908 909 910

  // buffer the above/left context information of the block in search.
  for (p = 0; p < MAX_MB_PLANE; ++p) {
Jim Bankoski's avatar
Jim Bankoski committed
911
    vpx_memcpy(
Jim Bankoski's avatar
Jim Bankoski committed
912
        a + num_4x4_blocks_wide * p,
913
        xd->above_context[p] + (mi_col * 2 >> xd->plane[p].subsampling_x),
Jim Bankoski's avatar
Jim Bankoski committed
914 915
        (sizeof(ENTROPY_CONTEXT) * num_4x4_blocks_wide) >>
        xd->plane[p].subsampling_x);
Jim Bankoski's avatar
Jim Bankoski committed
916
    vpx_memcpy(
Jim Bankoski's avatar
Jim Bankoski committed
917
        l + num_4x4_blocks_high * p,
918
        xd->left_context[p]
Jim Bankoski's avatar
Jim Bankoski committed
919