summaryrefslogtreecommitdiffstats
path: root/third_party/aom/av1/encoder/mbgraph.c
blob: 1296027dcb72f51b4a0309445788c268b6fb01ce (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
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
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
/*
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
 *
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
 */

#include <limits.h>

#include "./av1_rtcd.h"
#include "./aom_dsp_rtcd.h"

#include "aom_dsp/aom_dsp_common.h"
#include "aom_mem/aom_mem.h"
#include "aom_ports/system_state.h"
#include "av1/encoder/segmentation.h"
#include "av1/encoder/mcomp.h"
#include "av1/common/blockd.h"
#include "av1/common/reconinter.h"
#include "av1/common/reconintra.h"

static unsigned int do_16x16_motion_iteration(AV1_COMP *cpi, const MV *ref_mv,
                                              int mb_row, int mb_col) {
  MACROBLOCK *const x = &cpi->td.mb;
  MACROBLOCKD *const xd = &x->e_mbd;
  const MV_SPEED_FEATURES *const mv_sf = &cpi->sf.mv;
  const aom_variance_fn_ptr_t v_fn_ptr = cpi->fn_ptr[BLOCK_16X16];

  const MvLimits tmp_mv_limits = x->mv_limits;
  MV ref_full;
  int cost_list[5];

  // Further step/diamond searches as necessary
  int step_param = mv_sf->reduce_first_step_size;
  step_param = AOMMIN(step_param, MAX_MVSEARCH_STEPS - 2);

  av1_set_mv_search_range(&x->mv_limits, ref_mv);

  ref_full.col = ref_mv->col >> 3;
  ref_full.row = ref_mv->row >> 3;

  /*cpi->sf.search_method == HEX*/
  av1_hex_search(x, &ref_full, step_param, x->errorperbit, 0,
                 cond_cost_list(cpi, cost_list), &v_fn_ptr, 0, ref_mv);

  // Try sub-pixel MC
  // if (bestsme > error_thresh && bestsme < INT_MAX)
  {
    int distortion;
    unsigned int sse;
    cpi->find_fractional_mv_step(x, ref_mv, cpi->common.allow_high_precision_mv,
                                 x->errorperbit, &v_fn_ptr, 0,
                                 mv_sf->subpel_iters_per_step,
                                 cond_cost_list(cpi, cost_list), NULL, NULL,
                                 &distortion, &sse, NULL, 0, 0, 0);
  }

#if CONFIG_EXT_INTER
  if (has_second_ref(&xd->mi[0]->mbmi))
    xd->mi[0]->mbmi.mode = NEW_NEWMV;
  else
#endif  // CONFIG_EXT_INTER
    xd->mi[0]->mbmi.mode = NEWMV;

  xd->mi[0]->mbmi.mv[0] = x->best_mv;
#if CONFIG_EXT_INTER
  xd->mi[0]->mbmi.ref_frame[1] = NONE_FRAME;
#endif  // CONFIG_EXT_INTER

  av1_build_inter_predictors_sby(xd, mb_row, mb_col, NULL, BLOCK_16X16);

  /* restore UMV window */
  x->mv_limits = tmp_mv_limits;

  return aom_sad16x16(x->plane[0].src.buf, x->plane[0].src.stride,
                      xd->plane[0].dst.buf, xd->plane[0].dst.stride);
}

static int do_16x16_motion_search(AV1_COMP *cpi, const MV *ref_mv, int mb_row,
                                  int mb_col) {
  MACROBLOCK *const x = &cpi->td.mb;
  MACROBLOCKD *const xd = &x->e_mbd;
  unsigned int err, tmp_err;
  MV best_mv;

  // Try zero MV first
  // FIXME should really use something like near/nearest MV and/or MV prediction
  err = aom_sad16x16(x->plane[0].src.buf, x->plane[0].src.stride,
                     xd->plane[0].pre[0].buf, xd->plane[0].pre[0].stride);
  best_mv.col = best_mv.row = 0;

  // Test last reference frame using the previous best mv as the
  // starting point (best reference) for the search
  tmp_err = do_16x16_motion_iteration(cpi, ref_mv, mb_row, mb_col);
  if (tmp_err < err) {
    err = tmp_err;
    best_mv = x->best_mv.as_mv;
  }

  // If the current best reference mv is not centered on 0,0 then do a 0,0
  // based search as well.
  if (ref_mv->row != 0 || ref_mv->col != 0) {
    MV zero_ref_mv = { 0, 0 };

    tmp_err = do_16x16_motion_iteration(cpi, &zero_ref_mv, mb_row, mb_col);
    if (tmp_err < err) {
      err = tmp_err;
      best_mv = x->best_mv.as_mv;
    }
  }

  x->best_mv.as_mv = best_mv;
  return err;
}

static int do_16x16_zerozero_search(AV1_COMP *cpi, int_mv *dst_mv) {
  MACROBLOCK *const x = &cpi->td.mb;
  MACROBLOCKD *const xd = &x->e_mbd;
  unsigned int err;

  // Try zero MV first
  // FIXME should really use something like near/nearest MV and/or MV prediction
  err = aom_sad16x16(x->plane[0].src.buf, x->plane[0].src.stride,
                     xd->plane[0].pre[0].buf, xd->plane[0].pre[0].stride);

  dst_mv->as_int = 0;

  return err;
}
static int find_best_16x16_intra(AV1_COMP *cpi, PREDICTION_MODE *pbest_mode) {
  MACROBLOCK *const x = &cpi->td.mb;
  MACROBLOCKD *const xd = &x->e_mbd;
  PREDICTION_MODE best_mode = -1, mode;
  unsigned int best_err = INT_MAX;

  // calculate SATD for each intra prediction mode;
  // we're intentionally not doing 4x4, we just want a rough estimate
  for (mode = DC_PRED; mode <= TM_PRED; mode++) {
    unsigned int err;

    xd->mi[0]->mbmi.mode = mode;
    av1_predict_intra_block(xd, 16, 16, BLOCK_16X16, mode, x->plane[0].src.buf,
                            x->plane[0].src.stride, xd->plane[0].dst.buf,
                            xd->plane[0].dst.stride, 0, 0, 0);
    err = aom_sad16x16(x->plane[0].src.buf, x->plane[0].src.stride,
                       xd->plane[0].dst.buf, xd->plane[0].dst.stride);

    // find best
    if (err < best_err) {
      best_err = err;
      best_mode = mode;
    }
  }

  if (pbest_mode) *pbest_mode = best_mode;

  return best_err;
}

static void update_mbgraph_mb_stats(AV1_COMP *cpi, MBGRAPH_MB_STATS *stats,
                                    YV12_BUFFER_CONFIG *buf, int mb_y_offset,
                                    YV12_BUFFER_CONFIG *golden_ref,
                                    const MV *prev_golden_ref_mv,
                                    YV12_BUFFER_CONFIG *alt_ref, int mb_row,
                                    int mb_col) {
  MACROBLOCK *const x = &cpi->td.mb;
  MACROBLOCKD *const xd = &x->e_mbd;
  int intra_error;
  AV1_COMMON *cm = &cpi->common;

  // FIXME in practice we're completely ignoring chroma here
  x->plane[0].src.buf = buf->y_buffer + mb_y_offset;
  x->plane[0].src.stride = buf->y_stride;

  xd->plane[0].dst.buf = get_frame_new_buffer(cm)->y_buffer + mb_y_offset;
  xd->plane[0].dst.stride = get_frame_new_buffer(cm)->y_stride;

  // do intra 16x16 prediction
  intra_error = find_best_16x16_intra(cpi, &stats->ref[INTRA_FRAME].m.mode);
  if (intra_error <= 0) intra_error = 1;
  stats->ref[INTRA_FRAME].err = intra_error;

  // Golden frame MV search, if it exists and is different than last frame
  if (golden_ref) {
    int g_motion_error;
    xd->plane[0].pre[0].buf = golden_ref->y_buffer + mb_y_offset;
    xd->plane[0].pre[0].stride = golden_ref->y_stride;
    g_motion_error =
        do_16x16_motion_search(cpi, prev_golden_ref_mv, mb_row, mb_col);
    stats->ref[GOLDEN_FRAME].m.mv = x->best_mv;
    stats->ref[GOLDEN_FRAME].err = g_motion_error;
  } else {
    stats->ref[GOLDEN_FRAME].err = INT_MAX;
    stats->ref[GOLDEN_FRAME].m.mv.as_int = 0;
  }

  // Do an Alt-ref frame MV search, if it exists and is different than
  // last/golden frame.
  if (alt_ref) {
    int a_motion_error;
    xd->plane[0].pre[0].buf = alt_ref->y_buffer + mb_y_offset;
    xd->plane[0].pre[0].stride = alt_ref->y_stride;
    a_motion_error =
        do_16x16_zerozero_search(cpi, &stats->ref[ALTREF_FRAME].m.mv);

    stats->ref[ALTREF_FRAME].err = a_motion_error;
  } else {
    stats->ref[ALTREF_FRAME].err = INT_MAX;
    stats->ref[ALTREF_FRAME].m.mv.as_int = 0;
  }
}

static void update_mbgraph_frame_stats(AV1_COMP *cpi,
                                       MBGRAPH_FRAME_STATS *stats,
                                       YV12_BUFFER_CONFIG *buf,
                                       YV12_BUFFER_CONFIG *golden_ref,
                                       YV12_BUFFER_CONFIG *alt_ref) {
  MACROBLOCK *const x = &cpi->td.mb;
  MACROBLOCKD *const xd = &x->e_mbd;
  AV1_COMMON *const cm = &cpi->common;

  int mb_col, mb_row, offset = 0;
  int mb_y_offset = 0, arf_y_offset = 0, gld_y_offset = 0;
  MV gld_top_mv = { 0, 0 };
  MODE_INFO mi_local;

  av1_zero(mi_local);
  // Set up limit values for motion vectors to prevent them extending outside
  // the UMV borders.
  x->mv_limits.row_min = -BORDER_MV_PIXELS_B16;
  x->mv_limits.row_max = (cm->mb_rows - 1) * 8 + BORDER_MV_PIXELS_B16;
  xd->up_available = 0;
  xd->plane[0].dst.stride = buf->y_stride;
  xd->plane[0].pre[0].stride = buf->y_stride;
  xd->plane[1].dst.stride = buf->uv_stride;
  xd->mi[0] = &mi_local;
  mi_local.mbmi.sb_type = BLOCK_16X16;
  mi_local.mbmi.ref_frame[0] = LAST_FRAME;
  mi_local.mbmi.ref_frame[1] = NONE_FRAME;

  for (mb_row = 0; mb_row < cm->mb_rows; mb_row++) {
    MV gld_left_mv = gld_top_mv;
    int mb_y_in_offset = mb_y_offset;
    int arf_y_in_offset = arf_y_offset;
    int gld_y_in_offset = gld_y_offset;

    // Set up limit values for motion vectors to prevent them extending outside
    // the UMV borders.
    x->mv_limits.col_min = -BORDER_MV_PIXELS_B16;
    x->mv_limits.col_max = (cm->mb_cols - 1) * 8 + BORDER_MV_PIXELS_B16;
    xd->left_available = 0;

    for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) {
      MBGRAPH_MB_STATS *mb_stats = &stats->mb_stats[offset + mb_col];

      update_mbgraph_mb_stats(cpi, mb_stats, buf, mb_y_in_offset, golden_ref,
                              &gld_left_mv, alt_ref, mb_row, mb_col);
      gld_left_mv = mb_stats->ref[GOLDEN_FRAME].m.mv.as_mv;
      if (mb_col == 0) {
        gld_top_mv = gld_left_mv;
      }
      xd->left_available = 1;
      mb_y_in_offset += 16;
      gld_y_in_offset += 16;
      arf_y_in_offset += 16;
      x->mv_limits.col_min -= 16;
      x->mv_limits.col_max -= 16;
    }
    xd->up_available = 1;
    mb_y_offset += buf->y_stride * 16;
    gld_y_offset += golden_ref->y_stride * 16;
    if (alt_ref) arf_y_offset += alt_ref->y_stride * 16;
    x->mv_limits.row_min -= 16;
    x->mv_limits.row_max -= 16;
    offset += cm->mb_cols;
  }
}

// void separate_arf_mbs_byzz
static void separate_arf_mbs(AV1_COMP *cpi) {
  AV1_COMMON *const cm = &cpi->common;
  int mb_col, mb_row, offset, i;
  int mi_row, mi_col;
  int ncnt[4] = { 0 };
  int n_frames = cpi->mbgraph_n_frames;

  int *arf_not_zz;

  CHECK_MEM_ERROR(
      cm, arf_not_zz,
      aom_calloc(cm->mb_rows * cm->mb_cols * sizeof(*arf_not_zz), 1));

  // We are not interested in results beyond the alt ref itself.
  if (n_frames > cpi->rc.frames_till_gf_update_due)
    n_frames = cpi->rc.frames_till_gf_update_due;

  // defer cost to reference frames
  for (i = n_frames - 1; i >= 0; i--) {
    MBGRAPH_FRAME_STATS *frame_stats = &cpi->mbgraph_stats[i];

    for (offset = 0, mb_row = 0; mb_row < cm->mb_rows;
         offset += cm->mb_cols, mb_row++) {
      for (mb_col = 0; mb_col < cm->mb_cols; mb_col++) {
        MBGRAPH_MB_STATS *mb_stats = &frame_stats->mb_stats[offset + mb_col];

        int altref_err = mb_stats->ref[ALTREF_FRAME].err;
        int intra_err = mb_stats->ref[INTRA_FRAME].err;
        int golden_err = mb_stats->ref[GOLDEN_FRAME].err;

        // Test for altref vs intra and gf and that its mv was 0,0.
        if (altref_err > 1000 || altref_err > intra_err ||
            altref_err > golden_err) {
          arf_not_zz[offset + mb_col]++;
        }
      }
    }
  }

  // arf_not_zz is indexed by MB, but this loop is indexed by MI to avoid out
  // of bound access in segmentation_map
  for (mi_row = 0; mi_row < cm->mi_rows; mi_row++) {
    for (mi_col = 0; mi_col < cm->mi_cols; mi_col++) {
      // If any of the blocks in the sequence failed then the MB
      // goes in segment 0
      if (arf_not_zz[mi_row / 2 * cm->mb_cols + mi_col / 2]) {
        ncnt[0]++;
        cpi->segmentation_map[mi_row * cm->mi_cols + mi_col] = 0;
      } else {
        cpi->segmentation_map[mi_row * cm->mi_cols + mi_col] = 1;
        ncnt[1]++;
      }
    }
  }

  // Only bother with segmentation if over 10% of the MBs in static segment
  // if ( ncnt[1] && (ncnt[0] / ncnt[1] < 10) )
  if (1) {
    // Note % of blocks that are marked as static
    if (cm->MBs)
      cpi->static_mb_pct = (ncnt[1] * 100) / (cm->mi_rows * cm->mi_cols);

    // This error case should not be reachable as this function should
    // never be called with the common data structure uninitialized.
    else
      cpi->static_mb_pct = 0;

    av1_enable_segmentation(&cm->seg);
  } else {
    cpi->static_mb_pct = 0;
    av1_disable_segmentation(&cm->seg);
  }

  // Free localy allocated storage
  aom_free(arf_not_zz);
}

void av1_update_mbgraph_stats(AV1_COMP *cpi) {
  AV1_COMMON *const cm = &cpi->common;
  int i, n_frames = av1_lookahead_depth(cpi->lookahead);
  YV12_BUFFER_CONFIG *golden_ref = get_ref_frame_buffer(cpi, GOLDEN_FRAME);

  assert(golden_ref != NULL);

  // we need to look ahead beyond where the ARF transitions into
  // being a GF - so exit if we don't look ahead beyond that
  if (n_frames <= cpi->rc.frames_till_gf_update_due) return;

  if (n_frames > MAX_LAG_BUFFERS) n_frames = MAX_LAG_BUFFERS;

  cpi->mbgraph_n_frames = n_frames;
  for (i = 0; i < n_frames; i++) {
    MBGRAPH_FRAME_STATS *frame_stats = &cpi->mbgraph_stats[i];
    memset(frame_stats->mb_stats, 0,
           cm->mb_rows * cm->mb_cols * sizeof(*cpi->mbgraph_stats[i].mb_stats));
  }

  // do motion search to find contribution of each reference to data
  // later on in this GF group
  // FIXME really, the GF/last MC search should be done forward, and
  // the ARF MC search backwards, to get optimal results for MV caching
  for (i = 0; i < n_frames; i++) {
    MBGRAPH_FRAME_STATS *frame_stats = &cpi->mbgraph_stats[i];
    struct lookahead_entry *q_cur = av1_lookahead_peek(cpi->lookahead, i);

    assert(q_cur != NULL);

    update_mbgraph_frame_stats(cpi, frame_stats, &q_cur->img, golden_ref,
                               cpi->source);
  }

  aom_clear_system_state();

  separate_arf_mbs(cpi);
}