1 /* 2 * Copyright (c) 2016, Alliance for Open Media. All rights reserved 3 * 4 * This source code is subject to the terms of the BSD 2 Clause License and 5 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License 6 * was not distributed with this source code in the LICENSE file, you can 7 * obtain it at www.aomedia.org/license/software. If the Alliance for Open 8 * Media Patent License 1.0 was not distributed with this source code in the 9 * PATENTS file, you can obtain it at www.aomedia.org/license/patent. 10 */ 11 12 #include <limits.h> 13 #include <math.h> 14 #include <stdio.h> 15 16 #include "config/aom_config.h" 17 #include "config/aom_dsp_rtcd.h" 18 #include "config/aom_scale_rtcd.h" 19 #include "config/av1_rtcd.h" 20 21 #include "aom_dsp/aom_dsp_common.h" 22 #include "aom_dsp/aom_filter.h" 23 #if CONFIG_DENOISE 24 #include "aom_dsp/grain_table.h" 25 #include "aom_dsp/noise_util.h" 26 #include "aom_dsp/noise_model.h" 27 #endif 28 #include "aom_dsp/psnr.h" 29 #if CONFIG_INTERNAL_STATS 30 #include "aom_dsp/ssim.h" 31 #endif 32 #include "aom_ports/aom_timer.h" 33 #include "aom_ports/mem.h" 34 #include "aom_ports/system_state.h" 35 #include "aom_scale/aom_scale.h" 36 #if CONFIG_BITSTREAM_DEBUG 37 #include "aom_util/debug_util.h" 38 #endif // CONFIG_BITSTREAM_DEBUG 39 40 #include "av1/common/alloccommon.h" 41 #include "av1/common/cdef.h" 42 #include "av1/common/filter.h" 43 #include "av1/common/idct.h" 44 #include "av1/common/reconinter.h" 45 #include "av1/common/reconintra.h" 46 #include "av1/common/resize.h" 47 #include "av1/common/tile_common.h" 48 49 #include "av1/encoder/av1_multi_thread.h" 50 #include "av1/encoder/aq_complexity.h" 51 #include "av1/encoder/aq_cyclicrefresh.h" 52 #include "av1/encoder/aq_variance.h" 53 #include "av1/encoder/bitstream.h" 54 #include "av1/encoder/context_tree.h" 55 #include "av1/encoder/encodeframe.h" 56 #include "av1/encoder/encodemv.h" 57 #include "av1/encoder/encode_strategy.h" 58 #include "av1/encoder/encoder.h" 59 #include "av1/encoder/encodetxb.h" 60 #include "av1/encoder/ethread.h" 61 #include "av1/encoder/firstpass.h" 62 #include "av1/encoder/grain_test_vectors.h" 63 #include "av1/encoder/hash_motion.h" 64 #include "av1/encoder/mbgraph.h" 65 #include "av1/encoder/pass2_strategy.h" 66 #include "av1/encoder/picklpf.h" 67 #include "av1/encoder/pickrst.h" 68 #include "av1/encoder/random.h" 69 #include "av1/encoder/ratectrl.h" 70 #include "av1/encoder/rd.h" 71 #include "av1/encoder/rdopt.h" 72 #include "av1/encoder/segmentation.h" 73 #include "av1/encoder/speed_features.h" 74 #include "av1/encoder/reconinter_enc.h" 75 #include "av1/encoder/var_based_part.h" 76 77 #define DEFAULT_EXPLICIT_ORDER_HINT_BITS 7 78 79 #if CONFIG_ENTROPY_STATS 80 FRAME_COUNTS aggregate_fc; 81 #endif // CONFIG_ENTROPY_STATS 82 83 #define AM_SEGMENT_ID_INACTIVE 7 84 #define AM_SEGMENT_ID_ACTIVE 0 85 86 // Whether to use high precision mv for altref computation. 87 #define ALTREF_HIGH_PRECISION_MV 1 88 89 // Q threshold for high precision mv. Choose a very high value for now so that 90 // HIGH_PRECISION is always chosen. 91 #define HIGH_PRECISION_MV_QTHRESH 200 92 93 // #define OUTPUT_YUV_REC 94 #ifdef OUTPUT_YUV_SKINMAP 95 FILE *yuv_skinmap_file = NULL; 96 #endif 97 #ifdef OUTPUT_YUV_REC 98 FILE *yuv_rec_file; 99 #define FILE_NAME_LEN 100 100 #endif 101 102 static INLINE void Scale2Ratio(AOM_SCALING mode, int *hr, int *hs) { 103 switch (mode) { 104 case NORMAL: 105 *hr = 1; 106 *hs = 1; 107 break; 108 case FOURFIVE: 109 *hr = 4; 110 *hs = 5; 111 break; 112 case THREEFIVE: 113 *hr = 3; 114 *hs = 5; 115 break; 116 case ONETWO: 117 *hr = 1; 118 *hs = 2; 119 break; 120 default: 121 *hr = 1; 122 *hs = 1; 123 assert(0); 124 break; 125 } 126 } 127 128 // Mark all inactive blocks as active. Other segmentation features may be set 129 // so memset cannot be used, instead only inactive blocks should be reset. 130 static void suppress_active_map(AV1_COMP *cpi) { 131 unsigned char *const seg_map = cpi->segmentation_map; 132 int i; 133 if (cpi->active_map.enabled || cpi->active_map.update) 134 for (i = 0; i < cpi->common.mi_rows * cpi->common.mi_cols; ++i) 135 if (seg_map[i] == AM_SEGMENT_ID_INACTIVE) 136 seg_map[i] = AM_SEGMENT_ID_ACTIVE; 137 } 138 139 static void apply_active_map(AV1_COMP *cpi) { 140 struct segmentation *const seg = &cpi->common.seg; 141 unsigned char *const seg_map = cpi->segmentation_map; 142 const unsigned char *const active_map = cpi->active_map.map; 143 int i; 144 145 assert(AM_SEGMENT_ID_ACTIVE == CR_SEGMENT_ID_BASE); 146 147 if (frame_is_intra_only(&cpi->common)) { 148 cpi->active_map.enabled = 0; 149 cpi->active_map.update = 1; 150 } 151 152 if (cpi->active_map.update) { 153 if (cpi->active_map.enabled) { 154 for (i = 0; i < cpi->common.mi_rows * cpi->common.mi_cols; ++i) 155 if (seg_map[i] == AM_SEGMENT_ID_ACTIVE) seg_map[i] = active_map[i]; 156 av1_enable_segmentation(seg); 157 av1_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_SKIP); 158 av1_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_H); 159 av1_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_V); 160 av1_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_U); 161 av1_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_V); 162 163 av1_set_segdata(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_H, 164 -MAX_LOOP_FILTER); 165 av1_set_segdata(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_V, 166 -MAX_LOOP_FILTER); 167 av1_set_segdata(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_U, 168 -MAX_LOOP_FILTER); 169 av1_set_segdata(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_V, 170 -MAX_LOOP_FILTER); 171 } else { 172 av1_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_SKIP); 173 av1_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_H); 174 av1_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_V); 175 av1_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_U); 176 av1_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_V); 177 if (seg->enabled) { 178 seg->update_data = 1; 179 seg->update_map = 1; 180 } 181 } 182 cpi->active_map.update = 0; 183 } 184 } 185 186 int av1_set_active_map(AV1_COMP *cpi, unsigned char *new_map_16x16, int rows, 187 int cols) { 188 if (rows == cpi->common.mb_rows && cols == cpi->common.mb_cols) { 189 unsigned char *const active_map_8x8 = cpi->active_map.map; 190 const int mi_rows = cpi->common.mi_rows; 191 const int mi_cols = cpi->common.mi_cols; 192 const int row_scale = mi_size_high[BLOCK_16X16] == 2 ? 1 : 2; 193 const int col_scale = mi_size_wide[BLOCK_16X16] == 2 ? 1 : 2; 194 cpi->active_map.update = 1; 195 if (new_map_16x16) { 196 int r, c; 197 for (r = 0; r < mi_rows; ++r) { 198 for (c = 0; c < mi_cols; ++c) { 199 active_map_8x8[r * mi_cols + c] = 200 new_map_16x16[(r >> row_scale) * cols + (c >> col_scale)] 201 ? AM_SEGMENT_ID_ACTIVE 202 : AM_SEGMENT_ID_INACTIVE; 203 } 204 } 205 cpi->active_map.enabled = 1; 206 } else { 207 cpi->active_map.enabled = 0; 208 } 209 return 0; 210 } else { 211 return -1; 212 } 213 } 214 215 int av1_get_active_map(AV1_COMP *cpi, unsigned char *new_map_16x16, int rows, 216 int cols) { 217 if (rows == cpi->common.mb_rows && cols == cpi->common.mb_cols && 218 new_map_16x16) { 219 unsigned char *const seg_map_8x8 = cpi->segmentation_map; 220 const int mi_rows = cpi->common.mi_rows; 221 const int mi_cols = cpi->common.mi_cols; 222 const int row_scale = mi_size_high[BLOCK_16X16] == 2 ? 1 : 2; 223 const int col_scale = mi_size_wide[BLOCK_16X16] == 2 ? 1 : 2; 224 225 memset(new_map_16x16, !cpi->active_map.enabled, rows * cols); 226 if (cpi->active_map.enabled) { 227 int r, c; 228 for (r = 0; r < mi_rows; ++r) { 229 for (c = 0; c < mi_cols; ++c) { 230 // Cyclic refresh segments are considered active despite not having 231 // AM_SEGMENT_ID_ACTIVE 232 new_map_16x16[(r >> row_scale) * cols + (c >> col_scale)] |= 233 seg_map_8x8[r * mi_cols + c] != AM_SEGMENT_ID_INACTIVE; 234 } 235 } 236 } 237 return 0; 238 } else { 239 return -1; 240 } 241 } 242 243 // Compute the horizontal frequency components' energy in a frame 244 // by calculuating the 16x4 Horizontal DCT. This is to be used to 245 // decide the superresolution parameters. 246 void analyze_hor_freq(const AV1_COMP *cpi, double *energy) { 247 uint64_t freq_energy[16] = { 0 }; 248 const YV12_BUFFER_CONFIG *buf = cpi->source; 249 const int bd = cpi->td.mb.e_mbd.bd; 250 const int width = buf->y_crop_width; 251 const int height = buf->y_crop_height; 252 DECLARE_ALIGNED(16, int32_t, coeff[16 * 4]); 253 int n = 0; 254 memset(freq_energy, 0, sizeof(freq_energy)); 255 if (buf->flags & YV12_FLAG_HIGHBITDEPTH) { 256 const int16_t *src16 = (const int16_t *)CONVERT_TO_SHORTPTR(buf->y_buffer); 257 for (int i = 0; i < height - 4; i += 4) { 258 for (int j = 0; j < width - 16; j += 16) { 259 av1_fwd_txfm2d_16x4(src16 + i * buf->y_stride + j, coeff, buf->y_stride, 260 H_DCT, bd); 261 for (int k = 1; k < 16; ++k) { 262 const uint64_t this_energy = 263 ((int64_t)coeff[k] * coeff[k]) + 264 ((int64_t)coeff[k + 16] * coeff[k + 16]) + 265 ((int64_t)coeff[k + 32] * coeff[k + 32]) + 266 ((int64_t)coeff[k + 48] * coeff[k + 48]); 267 freq_energy[k] += ROUND_POWER_OF_TWO(this_energy, 2 + 2 * (bd - 8)); 268 } 269 n++; 270 } 271 } 272 } else { 273 assert(bd == 8); 274 DECLARE_ALIGNED(16, int16_t, src16[16 * 4]); 275 for (int i = 0; i < height - 4; i += 4) { 276 for (int j = 0; j < width - 16; j += 16) { 277 for (int ii = 0; ii < 4; ++ii) 278 for (int jj = 0; jj < 16; ++jj) 279 src16[ii * 16 + jj] = 280 buf->y_buffer[(i + ii) * buf->y_stride + (j + jj)]; 281 av1_fwd_txfm2d_16x4(src16, coeff, 16, H_DCT, bd); 282 for (int k = 1; k < 16; ++k) { 283 const uint64_t this_energy = 284 ((int64_t)coeff[k] * coeff[k]) + 285 ((int64_t)coeff[k + 16] * coeff[k + 16]) + 286 ((int64_t)coeff[k + 32] * coeff[k + 32]) + 287 ((int64_t)coeff[k + 48] * coeff[k + 48]); 288 freq_energy[k] += ROUND_POWER_OF_TWO(this_energy, 2); 289 } 290 n++; 291 } 292 } 293 } 294 if (n) { 295 for (int k = 1; k < 16; ++k) energy[k] = (double)freq_energy[k] / n; 296 // Convert to cumulative energy 297 for (int k = 14; k > 0; --k) energy[k] += energy[k + 1]; 298 } else { 299 for (int k = 1; k < 16; ++k) energy[k] = 1e+20; 300 } 301 } 302 303 static void set_high_precision_mv(AV1_COMP *cpi, int allow_high_precision_mv, 304 int cur_frame_force_integer_mv) { 305 MACROBLOCK *const mb = &cpi->td.mb; 306 cpi->common.allow_high_precision_mv = 307 allow_high_precision_mv && cur_frame_force_integer_mv == 0; 308 const int copy_hp = 309 cpi->common.allow_high_precision_mv && cur_frame_force_integer_mv == 0; 310 int *(*src)[2] = copy_hp ? &mb->nmvcost_hp : &mb->nmvcost; 311 mb->mv_cost_stack = *src; 312 } 313 314 static BLOCK_SIZE select_sb_size(const AV1_COMP *const cpi) { 315 const AV1_COMMON *const cm = &cpi->common; 316 317 if (cpi->oxcf.superblock_size == AOM_SUPERBLOCK_SIZE_64X64) 318 return BLOCK_64X64; 319 #if CONFIG_FILEOPTIONS 320 if (cm->options && cm->options->ext_partition) 321 #endif 322 if (cpi->oxcf.superblock_size == AOM_SUPERBLOCK_SIZE_128X128) 323 return BLOCK_128X128; 324 325 assert(cpi->oxcf.superblock_size == AOM_SUPERBLOCK_SIZE_DYNAMIC); 326 327 // TODO(any): Possibly could improve this with a heuristic. 328 #if CONFIG_FILEOPTIONS 329 if (cm->options && !cm->options->ext_partition) return BLOCK_64X64; 330 #endif 331 332 // When superres / resize is on, 'cm->width / height' can change between 333 // calls, so we don't apply this heuristic there. Also, this heuristic gives 334 // compression gain for speed >= 2 only. 335 // Things break if superblock size changes per-frame which is why this 336 // heuristic is set based on configured speed rather than actual 337 // speed-features (which may change per-frame in future) 338 if (cpi->oxcf.superres_mode == SUPERRES_NONE && 339 cpi->oxcf.resize_mode == RESIZE_NONE && cpi->oxcf.speed >= 2) { 340 return (cm->width >= 480 && cm->height >= 360) ? BLOCK_128X128 341 : BLOCK_64X64; 342 } 343 344 return BLOCK_128X128; 345 } 346 347 static void setup_frame(AV1_COMP *cpi) { 348 AV1_COMMON *const cm = &cpi->common; 349 // Set up entropy context depending on frame type. The decoder mandates 350 // the use of the default context, index 0, for keyframes and inter 351 // frames where the error_resilient_mode or intra_only flag is set. For 352 // other inter-frames the encoder currently uses only two contexts; 353 // context 1 for ALTREF frames and context 0 for the others. 354 355 if (frame_is_intra_only(cm) || cm->error_resilient_mode || 356 cpi->ext_use_primary_ref_none) { 357 av1_setup_past_independence(cm); 358 } 359 360 if (cm->current_frame.frame_type == KEY_FRAME && cm->show_frame) { 361 set_sb_size(&cm->seq_params, select_sb_size(cpi)); 362 } else if (frame_is_sframe(cm)) { 363 set_sb_size(&cm->seq_params, select_sb_size(cpi)); 364 } else { 365 const RefCntBuffer *const primary_ref_buf = get_primary_ref_frame_buf(cm); 366 if (primary_ref_buf == NULL) { 367 av1_setup_past_independence(cm); 368 cm->seg.update_map = 1; 369 cm->seg.update_data = 1; 370 } else { 371 *cm->fc = primary_ref_buf->frame_context; 372 } 373 } 374 375 av1_zero(cm->cur_frame->interp_filter_selected); 376 cm->prev_frame = get_primary_ref_frame_buf(cm); 377 cpi->vaq_refresh = 0; 378 } 379 380 static void enc_setup_mi(AV1_COMMON *cm) { 381 int i; 382 int mi_rows_sb_aligned = calc_mi_size(cm->mi_rows); 383 cm->mi = cm->mip; 384 memset(cm->mip, 0, cm->mi_stride * mi_rows_sb_aligned * sizeof(*cm->mip)); 385 cm->prev_mi = cm->prev_mip; 386 // Clear top border row 387 memset(cm->prev_mip, 0, sizeof(*cm->prev_mip) * cm->mi_stride); 388 // Clear left border column 389 for (i = 0; i < mi_rows_sb_aligned; ++i) 390 memset(&cm->prev_mip[i * cm->mi_stride], 0, sizeof(*cm->prev_mip)); 391 cm->mi_grid_visible = cm->mi_grid_base; 392 cm->prev_mi_grid_visible = cm->prev_mi_grid_base; 393 394 memset(cm->mi_grid_base, 0, 395 cm->mi_stride * mi_rows_sb_aligned * sizeof(*cm->mi_grid_base)); 396 } 397 398 static int enc_alloc_mi(AV1_COMMON *cm, int mi_size) { 399 cm->mip = aom_calloc(mi_size, sizeof(*cm->mip)); 400 if (!cm->mip) return 1; 401 cm->prev_mip = aom_calloc(mi_size, sizeof(*cm->prev_mip)); 402 if (!cm->prev_mip) return 1; 403 cm->mi_alloc_size = mi_size; 404 405 cm->mi_grid_base = 406 (MB_MODE_INFO **)aom_calloc(mi_size, sizeof(MB_MODE_INFO *)); 407 if (!cm->mi_grid_base) return 1; 408 cm->prev_mi_grid_base = 409 (MB_MODE_INFO **)aom_calloc(mi_size, sizeof(MB_MODE_INFO *)); 410 if (!cm->prev_mi_grid_base) return 1; 411 412 return 0; 413 } 414 415 static void enc_free_mi(AV1_COMMON *cm) { 416 aom_free(cm->mip); 417 cm->mip = NULL; 418 aom_free(cm->prev_mip); 419 cm->prev_mip = NULL; 420 aom_free(cm->mi_grid_base); 421 cm->mi_grid_base = NULL; 422 aom_free(cm->prev_mi_grid_base); 423 cm->prev_mi_grid_base = NULL; 424 cm->mi_alloc_size = 0; 425 } 426 427 static void swap_mi_and_prev_mi(AV1_COMMON *cm) { 428 // Current mip will be the prev_mip for the next frame. 429 MB_MODE_INFO **temp_base = cm->prev_mi_grid_base; 430 MB_MODE_INFO *temp = cm->prev_mip; 431 cm->prev_mip = cm->mip; 432 cm->mip = temp; 433 434 // Update the upper left visible macroblock ptrs. 435 cm->mi = cm->mip; 436 cm->prev_mi = cm->prev_mip; 437 438 cm->prev_mi_grid_base = cm->mi_grid_base; 439 cm->mi_grid_base = temp_base; 440 cm->mi_grid_visible = cm->mi_grid_base; 441 cm->prev_mi_grid_visible = cm->prev_mi_grid_base; 442 } 443 444 void av1_initialize_enc(void) { 445 av1_rtcd(); 446 aom_dsp_rtcd(); 447 aom_scale_rtcd(); 448 av1_init_intra_predictors(); 449 av1_init_me_luts(); 450 av1_rc_init_minq_luts(); 451 av1_init_wedge_masks(); 452 } 453 454 static void dealloc_context_buffers_ext(AV1_COMP *cpi) { 455 if (cpi->mbmi_ext_base) { 456 aom_free(cpi->mbmi_ext_base); 457 cpi->mbmi_ext_base = NULL; 458 } 459 } 460 461 static void alloc_context_buffers_ext(AV1_COMP *cpi) { 462 AV1_COMMON *cm = &cpi->common; 463 int mi_size = cm->mi_cols * cm->mi_rows; 464 465 dealloc_context_buffers_ext(cpi); 466 CHECK_MEM_ERROR(cm, cpi->mbmi_ext_base, 467 aom_calloc(mi_size, sizeof(*cpi->mbmi_ext_base))); 468 } 469 470 static void reset_film_grain_chroma_params(aom_film_grain_t *pars) { 471 pars->num_cr_points = 0; 472 pars->cr_mult = 0; 473 pars->cr_luma_mult = 0; 474 memset(pars->scaling_points_cr, 0, sizeof(pars->scaling_points_cr)); 475 memset(pars->ar_coeffs_cr, 0, sizeof(pars->ar_coeffs_cr)); 476 pars->num_cb_points = 0; 477 pars->cb_mult = 0; 478 pars->cb_luma_mult = 0; 479 pars->chroma_scaling_from_luma = 0; 480 memset(pars->scaling_points_cb, 0, sizeof(pars->scaling_points_cb)); 481 memset(pars->ar_coeffs_cb, 0, sizeof(pars->ar_coeffs_cb)); 482 } 483 484 static void update_film_grain_parameters(struct AV1_COMP *cpi, 485 const AV1EncoderConfig *oxcf) { 486 AV1_COMMON *const cm = &cpi->common; 487 cpi->oxcf = *oxcf; 488 489 if (cpi->film_grain_table) { 490 aom_film_grain_table_free(cpi->film_grain_table); 491 aom_free(cpi->film_grain_table); 492 cpi->film_grain_table = NULL; 493 } 494 495 if (oxcf->film_grain_test_vector) { 496 cm->seq_params.film_grain_params_present = 1; 497 if (cm->current_frame.frame_type == KEY_FRAME) { 498 memcpy(&cm->film_grain_params, 499 film_grain_test_vectors + oxcf->film_grain_test_vector - 1, 500 sizeof(cm->film_grain_params)); 501 if (oxcf->monochrome) 502 reset_film_grain_chroma_params(&cm->film_grain_params); 503 cm->film_grain_params.bit_depth = cm->seq_params.bit_depth; 504 if (cm->seq_params.color_range == AOM_CR_FULL_RANGE) { 505 cm->film_grain_params.clip_to_restricted_range = 0; 506 } 507 } 508 } else if (oxcf->film_grain_table_filename) { 509 cm->seq_params.film_grain_params_present = 1; 510 511 cpi->film_grain_table = aom_malloc(sizeof(*cpi->film_grain_table)); 512 memset(cpi->film_grain_table, 0, sizeof(aom_film_grain_table_t)); 513 514 aom_film_grain_table_read(cpi->film_grain_table, 515 oxcf->film_grain_table_filename, &cm->error); 516 } else { 517 #if CONFIG_DENOISE 518 cm->seq_params.film_grain_params_present = (cpi->oxcf.noise_level > 0); 519 #else 520 cm->seq_params.film_grain_params_present = 0; 521 #endif 522 memset(&cm->film_grain_params, 0, sizeof(cm->film_grain_params)); 523 } 524 } 525 526 static void dealloc_compressor_data(AV1_COMP *cpi) { 527 AV1_COMMON *const cm = &cpi->common; 528 const int num_planes = av1_num_planes(cm); 529 530 dealloc_context_buffers_ext(cpi); 531 532 aom_free(cpi->tile_data); 533 cpi->tile_data = NULL; 534 535 // Delete sementation map 536 aom_free(cpi->segmentation_map); 537 cpi->segmentation_map = NULL; 538 539 av1_cyclic_refresh_free(cpi->cyclic_refresh); 540 cpi->cyclic_refresh = NULL; 541 542 aom_free(cpi->active_map.map); 543 cpi->active_map.map = NULL; 544 545 aom_free(cpi->td.mb.above_pred_buf); 546 cpi->td.mb.above_pred_buf = NULL; 547 548 aom_free(cpi->td.mb.left_pred_buf); 549 cpi->td.mb.left_pred_buf = NULL; 550 551 aom_free(cpi->td.mb.wsrc_buf); 552 cpi->td.mb.wsrc_buf = NULL; 553 554 aom_free(cpi->td.mb.inter_modes_info); 555 cpi->td.mb.inter_modes_info = NULL; 556 557 for (int i = 0; i < 2; i++) 558 for (int j = 0; j < 2; j++) { 559 aom_free(cpi->td.mb.hash_value_buffer[i][j]); 560 cpi->td.mb.hash_value_buffer[i][j] = NULL; 561 } 562 aom_free(cpi->td.mb.mask_buf); 563 cpi->td.mb.mask_buf = NULL; 564 565 aom_free(cm->tpl_mvs); 566 cm->tpl_mvs = NULL; 567 568 av1_free_ref_frame_buffers(cm->buffer_pool); 569 av1_free_txb_buf(cpi); 570 av1_free_context_buffers(cm); 571 572 aom_free_frame_buffer(&cpi->last_frame_uf); 573 av1_free_restoration_buffers(cm); 574 aom_free_frame_buffer(&cpi->trial_frame_rst); 575 aom_free_frame_buffer(&cpi->scaled_source); 576 aom_free_frame_buffer(&cpi->scaled_last_source); 577 aom_free_frame_buffer(&cpi->alt_ref_buffer); 578 av1_lookahead_destroy(cpi->lookahead); 579 580 aom_free(cpi->tile_tok[0][0]); 581 cpi->tile_tok[0][0] = 0; 582 583 aom_free(cpi->tplist[0][0]); 584 cpi->tplist[0][0] = NULL; 585 586 av1_free_pc_tree(&cpi->td, num_planes); 587 588 aom_free(cpi->td.mb.palette_buffer); 589 590 aom_free(cpi->td.mb.tmp_conv_dst); 591 for (int j = 0; j < 2; ++j) { 592 aom_free(cpi->td.mb.tmp_obmc_bufs[j]); 593 } 594 595 #if CONFIG_DENOISE 596 if (cpi->denoise_and_model) { 597 aom_denoise_and_model_free(cpi->denoise_and_model); 598 cpi->denoise_and_model = NULL; 599 } 600 #endif 601 if (cpi->film_grain_table) { 602 aom_film_grain_table_free(cpi->film_grain_table); 603 cpi->film_grain_table = NULL; 604 } 605 } 606 607 static void save_coding_context(AV1_COMP *cpi) { 608 CODING_CONTEXT *const cc = &cpi->coding_context; 609 AV1_COMMON *cm = &cpi->common; 610 611 // Stores a snapshot of key state variables which can subsequently be 612 // restored with a call to av1_restore_coding_context. These functions are 613 // intended for use in a re-code loop in av1_compress_frame where the 614 // quantizer value is adjusted between loop iterations. 615 av1_copy(cc->nmv_vec_cost, cpi->td.mb.nmv_vec_cost); 616 av1_copy(cc->nmv_costs, cpi->nmv_costs); 617 av1_copy(cc->nmv_costs_hp, cpi->nmv_costs_hp); 618 619 cc->fc = *cm->fc; 620 } 621 622 static void restore_coding_context(AV1_COMP *cpi) { 623 CODING_CONTEXT *const cc = &cpi->coding_context; 624 AV1_COMMON *cm = &cpi->common; 625 626 // Restore key state variables to the snapshot state stored in the 627 // previous call to av1_save_coding_context. 628 av1_copy(cpi->td.mb.nmv_vec_cost, cc->nmv_vec_cost); 629 av1_copy(cpi->nmv_costs, cc->nmv_costs); 630 av1_copy(cpi->nmv_costs_hp, cc->nmv_costs_hp); 631 632 *cm->fc = cc->fc; 633 } 634 635 static void configure_static_seg_features(AV1_COMP *cpi) { 636 AV1_COMMON *const cm = &cpi->common; 637 const RATE_CONTROL *const rc = &cpi->rc; 638 struct segmentation *const seg = &cm->seg; 639 640 int high_q = (int)(rc->avg_q > 48.0); 641 int qi_delta; 642 643 // Disable and clear down for KF 644 if (cm->current_frame.frame_type == KEY_FRAME) { 645 // Clear down the global segmentation map 646 memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols); 647 seg->update_map = 0; 648 seg->update_data = 0; 649 cpi->static_mb_pct = 0; 650 651 // Disable segmentation 652 av1_disable_segmentation(seg); 653 654 // Clear down the segment features. 655 av1_clearall_segfeatures(seg); 656 } else if (cpi->refresh_alt_ref_frame) { 657 // If this is an alt ref frame 658 // Clear down the global segmentation map 659 memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols); 660 seg->update_map = 0; 661 seg->update_data = 0; 662 cpi->static_mb_pct = 0; 663 664 // Disable segmentation and individual segment features by default 665 av1_disable_segmentation(seg); 666 av1_clearall_segfeatures(seg); 667 668 // Scan frames from current to arf frame. 669 // This function re-enables segmentation if appropriate. 670 av1_update_mbgraph_stats(cpi); 671 672 // If segmentation was enabled set those features needed for the 673 // arf itself. 674 if (seg->enabled) { 675 seg->update_map = 1; 676 seg->update_data = 1; 677 678 qi_delta = av1_compute_qdelta(rc, rc->avg_q, rc->avg_q * 0.875, 679 cm->seq_params.bit_depth); 680 av1_set_segdata(seg, 1, SEG_LVL_ALT_Q, qi_delta - 2); 681 av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_Y_H, -2); 682 av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_Y_V, -2); 683 av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_U, -2); 684 av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_V, -2); 685 686 av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_Y_H); 687 av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_Y_V); 688 av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_U); 689 av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_V); 690 691 av1_enable_segfeature(seg, 1, SEG_LVL_ALT_Q); 692 } 693 } else if (seg->enabled) { 694 // All other frames if segmentation has been enabled 695 696 // First normal frame in a valid gf or alt ref group 697 if (rc->frames_since_golden == 0) { 698 // Set up segment features for normal frames in an arf group 699 if (rc->source_alt_ref_active) { 700 seg->update_map = 0; 701 seg->update_data = 1; 702 703 qi_delta = av1_compute_qdelta(rc, rc->avg_q, rc->avg_q * 1.125, 704 cm->seq_params.bit_depth); 705 av1_set_segdata(seg, 1, SEG_LVL_ALT_Q, qi_delta + 2); 706 av1_enable_segfeature(seg, 1, SEG_LVL_ALT_Q); 707 708 av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_Y_H, -2); 709 av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_Y_V, -2); 710 av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_U, -2); 711 av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_V, -2); 712 713 av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_Y_H); 714 av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_Y_V); 715 av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_U); 716 av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_V); 717 718 // Segment coding disabled for compred testing 719 if (high_q || (cpi->static_mb_pct == 100)) { 720 av1_set_segdata(seg, 1, SEG_LVL_REF_FRAME, ALTREF_FRAME); 721 av1_enable_segfeature(seg, 1, SEG_LVL_REF_FRAME); 722 av1_enable_segfeature(seg, 1, SEG_LVL_SKIP); 723 } 724 } else { 725 // Disable segmentation and clear down features if alt ref 726 // is not active for this group 727 728 av1_disable_segmentation(seg); 729 730 memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols); 731 732 seg->update_map = 0; 733 seg->update_data = 0; 734 735 av1_clearall_segfeatures(seg); 736 } 737 } else if (rc->is_src_frame_alt_ref) { 738 // Special case where we are coding over the top of a previous 739 // alt ref frame. 740 // Segment coding disabled for compred testing 741 742 // Enable ref frame features for segment 0 as well 743 av1_enable_segfeature(seg, 0, SEG_LVL_REF_FRAME); 744 av1_enable_segfeature(seg, 1, SEG_LVL_REF_FRAME); 745 746 // All mbs should use ALTREF_FRAME 747 av1_clear_segdata(seg, 0, SEG_LVL_REF_FRAME); 748 av1_set_segdata(seg, 0, SEG_LVL_REF_FRAME, ALTREF_FRAME); 749 av1_clear_segdata(seg, 1, SEG_LVL_REF_FRAME); 750 av1_set_segdata(seg, 1, SEG_LVL_REF_FRAME, ALTREF_FRAME); 751 752 // Skip all MBs if high Q (0,0 mv and skip coeffs) 753 if (high_q) { 754 av1_enable_segfeature(seg, 0, SEG_LVL_SKIP); 755 av1_enable_segfeature(seg, 1, SEG_LVL_SKIP); 756 } 757 // Enable data update 758 seg->update_data = 1; 759 } else { 760 // All other frames. 761 762 // No updates.. leave things as they are. 763 seg->update_map = 0; 764 seg->update_data = 0; 765 } 766 } 767 } 768 769 static void update_reference_segmentation_map(AV1_COMP *cpi) { 770 AV1_COMMON *const cm = &cpi->common; 771 MB_MODE_INFO **mi_4x4_ptr = cm->mi_grid_visible; 772 uint8_t *cache_ptr = cm->cur_frame->seg_map; 773 int row, col; 774 775 for (row = 0; row < cm->mi_rows; row++) { 776 MB_MODE_INFO **mi_4x4 = mi_4x4_ptr; 777 uint8_t *cache = cache_ptr; 778 for (col = 0; col < cm->mi_cols; col++, mi_4x4++, cache++) 779 cache[0] = mi_4x4[0]->segment_id; 780 mi_4x4_ptr += cm->mi_stride; 781 cache_ptr += cm->mi_cols; 782 } 783 } 784 785 static void alloc_raw_frame_buffers(AV1_COMP *cpi) { 786 AV1_COMMON *cm = &cpi->common; 787 const SequenceHeader *const seq_params = &cm->seq_params; 788 const AV1EncoderConfig *oxcf = &cpi->oxcf; 789 790 if (!cpi->lookahead) { 791 int is_scale = (oxcf->resize_mode || oxcf->superres_mode); 792 cpi->lookahead = av1_lookahead_init( 793 oxcf->width, oxcf->height, seq_params->subsampling_x, 794 seq_params->subsampling_y, seq_params->use_highbitdepth, 795 oxcf->lag_in_frames, oxcf->border_in_pixels, is_scale); 796 } 797 if (!cpi->lookahead) 798 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, 799 "Failed to allocate lag buffers"); 800 801 // TODO(agrange) Check if ARF is enabled and skip allocation if not. 802 if (aom_realloc_frame_buffer( 803 &cpi->alt_ref_buffer, oxcf->width, oxcf->height, 804 seq_params->subsampling_x, seq_params->subsampling_y, 805 seq_params->use_highbitdepth, oxcf->border_in_pixels, 806 cm->byte_alignment, NULL, NULL, NULL)) 807 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, 808 "Failed to allocate altref buffer"); 809 } 810 811 static void alloc_util_frame_buffers(AV1_COMP *cpi) { 812 AV1_COMMON *const cm = &cpi->common; 813 const SequenceHeader *const seq_params = &cm->seq_params; 814 if (aom_realloc_frame_buffer( 815 &cpi->last_frame_uf, cm->width, cm->height, seq_params->subsampling_x, 816 seq_params->subsampling_y, seq_params->use_highbitdepth, 817 cpi->oxcf.border_in_pixels, cm->byte_alignment, NULL, NULL, NULL)) 818 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, 819 "Failed to allocate last frame buffer"); 820 821 if (aom_realloc_frame_buffer( 822 &cpi->trial_frame_rst, cm->superres_upscaled_width, 823 cm->superres_upscaled_height, seq_params->subsampling_x, 824 seq_params->subsampling_y, seq_params->use_highbitdepth, 825 AOM_RESTORATION_FRAME_BORDER, cm->byte_alignment, NULL, NULL, NULL)) 826 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, 827 "Failed to allocate trial restored frame buffer"); 828 829 if (aom_realloc_frame_buffer( 830 &cpi->scaled_source, cm->width, cm->height, seq_params->subsampling_x, 831 seq_params->subsampling_y, seq_params->use_highbitdepth, 832 cpi->oxcf.border_in_pixels, cm->byte_alignment, NULL, NULL, NULL)) 833 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, 834 "Failed to allocate scaled source buffer"); 835 836 if (aom_realloc_frame_buffer( 837 &cpi->scaled_last_source, cm->width, cm->height, 838 seq_params->subsampling_x, seq_params->subsampling_y, 839 seq_params->use_highbitdepth, cpi->oxcf.border_in_pixels, 840 cm->byte_alignment, NULL, NULL, NULL)) 841 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, 842 "Failed to allocate scaled last source buffer"); 843 } 844 845 static void alloc_compressor_data(AV1_COMP *cpi) { 846 AV1_COMMON *cm = &cpi->common; 847 const int num_planes = av1_num_planes(cm); 848 849 av1_alloc_context_buffers(cm, cm->width, cm->height); 850 851 int mi_rows_aligned_to_sb = 852 ALIGN_POWER_OF_TWO(cm->mi_rows, cm->seq_params.mib_size_log2); 853 int sb_rows = mi_rows_aligned_to_sb >> cm->seq_params.mib_size_log2; 854 855 av1_alloc_txb_buf(cpi); 856 857 alloc_context_buffers_ext(cpi); 858 859 aom_free(cpi->tile_tok[0][0]); 860 861 { 862 unsigned int tokens = 863 get_token_alloc(cm->mb_rows, cm->mb_cols, MAX_SB_SIZE_LOG2, num_planes); 864 CHECK_MEM_ERROR(cm, cpi->tile_tok[0][0], 865 aom_calloc(tokens, sizeof(*cpi->tile_tok[0][0]))); 866 } 867 aom_free(cpi->tplist[0][0]); 868 869 CHECK_MEM_ERROR(cm, cpi->tplist[0][0], 870 aom_calloc(sb_rows * MAX_TILE_ROWS * MAX_TILE_COLS, 871 sizeof(*cpi->tplist[0][0]))); 872 873 av1_setup_pc_tree(&cpi->common, &cpi->td); 874 } 875 876 void av1_new_framerate(AV1_COMP *cpi, double framerate) { 877 cpi->framerate = framerate < 0.1 ? 30 : framerate; 878 av1_rc_update_framerate(cpi, cpi->common.width, cpi->common.height); 879 } 880 881 static void set_tile_info(AV1_COMP *cpi) { 882 AV1_COMMON *const cm = &cpi->common; 883 int i, start_sb; 884 885 av1_get_tile_limits(cm); 886 887 // configure tile columns 888 if (cpi->oxcf.tile_width_count == 0 || cpi->oxcf.tile_height_count == 0) { 889 cm->uniform_tile_spacing_flag = 1; 890 cm->log2_tile_cols = AOMMAX(cpi->oxcf.tile_columns, cm->min_log2_tile_cols); 891 cm->log2_tile_cols = AOMMIN(cm->log2_tile_cols, cm->max_log2_tile_cols); 892 } else { 893 int mi_cols = ALIGN_POWER_OF_TWO(cm->mi_cols, cm->seq_params.mib_size_log2); 894 int sb_cols = mi_cols >> cm->seq_params.mib_size_log2; 895 int size_sb, j = 0; 896 cm->uniform_tile_spacing_flag = 0; 897 for (i = 0, start_sb = 0; start_sb < sb_cols && i < MAX_TILE_COLS; i++) { 898 cm->tile_col_start_sb[i] = start_sb; 899 size_sb = cpi->oxcf.tile_widths[j++]; 900 if (j >= cpi->oxcf.tile_width_count) j = 0; 901 start_sb += AOMMIN(size_sb, cm->max_tile_width_sb); 902 } 903 cm->tile_cols = i; 904 cm->tile_col_start_sb[i] = sb_cols; 905 } 906 av1_calculate_tile_cols(cm); 907 908 // configure tile rows 909 if (cm->uniform_tile_spacing_flag) { 910 cm->log2_tile_rows = AOMMAX(cpi->oxcf.tile_rows, cm->min_log2_tile_rows); 911 cm->log2_tile_rows = AOMMIN(cm->log2_tile_rows, cm->max_log2_tile_rows); 912 } else { 913 int mi_rows = ALIGN_POWER_OF_TWO(cm->mi_rows, cm->seq_params.mib_size_log2); 914 int sb_rows = mi_rows >> cm->seq_params.mib_size_log2; 915 int size_sb, j = 0; 916 for (i = 0, start_sb = 0; start_sb < sb_rows && i < MAX_TILE_ROWS; i++) { 917 cm->tile_row_start_sb[i] = start_sb; 918 size_sb = cpi->oxcf.tile_heights[j++]; 919 if (j >= cpi->oxcf.tile_height_count) j = 0; 920 start_sb += AOMMIN(size_sb, cm->max_tile_height_sb); 921 } 922 cm->tile_rows = i; 923 cm->tile_row_start_sb[i] = sb_rows; 924 } 925 av1_calculate_tile_rows(cm); 926 } 927 928 static void update_frame_size(AV1_COMP *cpi) { 929 AV1_COMMON *const cm = &cpi->common; 930 MACROBLOCKD *const xd = &cpi->td.mb.e_mbd; 931 932 av1_set_mb_mi(cm, cm->width, cm->height); 933 av1_init_context_buffers(cm); 934 av1_init_macroblockd(cm, xd, NULL); 935 memset(cpi->mbmi_ext_base, 0, 936 cm->mi_rows * cm->mi_cols * sizeof(*cpi->mbmi_ext_base)); 937 set_tile_info(cpi); 938 } 939 940 static void init_buffer_indices(AV1_COMP *cpi) { 941 int fb_idx; 942 for (fb_idx = 0; fb_idx < REF_FRAMES; ++fb_idx) 943 cpi->common.remapped_ref_idx[fb_idx] = fb_idx; 944 cpi->rate_index = 0; 945 cpi->rate_size = 0; 946 } 947 948 static INLINE int does_level_match(int width, int height, double fps, 949 int lvl_width, int lvl_height, 950 double lvl_fps, int lvl_dim_mult) { 951 const int64_t lvl_luma_pels = lvl_width * lvl_height; 952 const double lvl_display_sample_rate = lvl_luma_pels * lvl_fps; 953 const int64_t luma_pels = width * height; 954 const double display_sample_rate = luma_pels * fps; 955 return luma_pels <= lvl_luma_pels && 956 display_sample_rate <= lvl_display_sample_rate && 957 width <= lvl_width * lvl_dim_mult && 958 height <= lvl_height * lvl_dim_mult; 959 } 960 961 static void set_bitstream_level_tier(SequenceHeader *seq, AV1_COMMON *cm, 962 const AV1EncoderConfig *oxcf) { 963 // TODO(any): This is a placeholder function that only addresses dimensions 964 // and max display sample rates. 965 // Need to add checks for max bit rate, max decoded luma sample rate, header 966 // rate, etc. that are not covered by this function. 967 AV1_LEVEL level = SEQ_LEVEL_MAX; 968 if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 512, 969 288, 30.0, 4)) { 970 level = SEQ_LEVEL_2_0; 971 } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 972 704, 396, 30.0, 4)) { 973 level = SEQ_LEVEL_2_1; 974 } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 975 1088, 612, 30.0, 4)) { 976 level = SEQ_LEVEL_3_0; 977 } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 978 1376, 774, 30.0, 4)) { 979 level = SEQ_LEVEL_3_1; 980 } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 981 2048, 1152, 30.0, 3)) { 982 level = SEQ_LEVEL_4_0; 983 } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 984 2048, 1152, 60.0, 3)) { 985 level = SEQ_LEVEL_4_1; 986 } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 987 4096, 2176, 30.0, 2)) { 988 level = SEQ_LEVEL_5_0; 989 } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 990 4096, 2176, 60.0, 2)) { 991 level = SEQ_LEVEL_5_1; 992 } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 993 4096, 2176, 120.0, 2)) { 994 level = SEQ_LEVEL_5_2; 995 } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 996 8192, 4352, 30.0, 2)) { 997 level = SEQ_LEVEL_6_0; 998 } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 999 8192, 4352, 60.0, 2)) { 1000 } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 1001 8192, 4352, 120.0, 2)) { 1002 level = SEQ_LEVEL_6_2; 1003 } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 1004 16384, 8704, 30.0, 2)) { 1005 level = SEQ_LEVEL_7_0; 1006 } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 1007 16384, 8704, 60.0, 2)) { 1008 level = SEQ_LEVEL_7_1; 1009 } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 1010 16384, 8704, 120.0, 2)) { 1011 level = SEQ_LEVEL_7_2; 1012 } 1013 for (int i = 0; i < MAX_NUM_OPERATING_POINTS; ++i) { 1014 seq->seq_level_idx[i] = level; 1015 // Set the maximum parameters for bitrate and buffer size for this profile, 1016 // level, and tier 1017 cm->op_params[i].bitrate = max_level_bitrate( 1018 cm->seq_params.profile, seq->seq_level_idx[i], seq->tier[i]); 1019 // Level with seq_level_idx = 31 returns a high "dummy" bitrate to pass the 1020 // check 1021 if (cm->op_params[i].bitrate == 0) 1022 aom_internal_error( 1023 &cm->error, AOM_CODEC_UNSUP_BITSTREAM, 1024 "AV1 does not support this combination of profile, level, and tier."); 1025 // Buffer size in bits/s is bitrate in bits/s * 1 s 1026 cm->op_params[i].buffer_size = cm->op_params[i].bitrate; 1027 } 1028 } 1029 1030 static void init_seq_coding_tools(SequenceHeader *seq, AV1_COMMON *cm, 1031 const AV1EncoderConfig *oxcf) { 1032 seq->still_picture = (oxcf->limit == 1); 1033 seq->reduced_still_picture_hdr = seq->still_picture; 1034 seq->reduced_still_picture_hdr &= !oxcf->full_still_picture_hdr; 1035 seq->force_screen_content_tools = 2; 1036 seq->force_integer_mv = 2; 1037 seq->order_hint_info.enable_order_hint = oxcf->enable_order_hint; 1038 seq->frame_id_numbers_present_flag = 1039 !(seq->still_picture && seq->reduced_still_picture_hdr) && 1040 !oxcf->large_scale_tile && oxcf->error_resilient_mode; 1041 if (seq->still_picture && seq->reduced_still_picture_hdr) { 1042 seq->order_hint_info.enable_order_hint = 0; 1043 seq->force_screen_content_tools = 2; 1044 seq->force_integer_mv = 2; 1045 } 1046 seq->order_hint_info.order_hint_bits_minus_1 = 1047 seq->order_hint_info.enable_order_hint 1048 ? DEFAULT_EXPLICIT_ORDER_HINT_BITS - 1 1049 : -1; 1050 1051 seq->max_frame_width = 1052 oxcf->forced_max_frame_width ? oxcf->forced_max_frame_width : oxcf->width; 1053 seq->max_frame_height = oxcf->forced_max_frame_height 1054 ? oxcf->forced_max_frame_height 1055 : oxcf->height; 1056 seq->num_bits_width = 1057 (seq->max_frame_width > 1) ? get_msb(seq->max_frame_width - 1) + 1 : 1; 1058 seq->num_bits_height = 1059 (seq->max_frame_height > 1) ? get_msb(seq->max_frame_height - 1) + 1 : 1; 1060 assert(seq->num_bits_width <= 16); 1061 assert(seq->num_bits_height <= 16); 1062 1063 seq->frame_id_length = FRAME_ID_LENGTH; 1064 seq->delta_frame_id_length = DELTA_FRAME_ID_LENGTH; 1065 1066 seq->enable_dual_filter = oxcf->enable_dual_filter; 1067 seq->order_hint_info.enable_dist_wtd_comp = oxcf->enable_dist_wtd_comp; 1068 seq->order_hint_info.enable_dist_wtd_comp &= 1069 seq->order_hint_info.enable_order_hint; 1070 seq->order_hint_info.enable_ref_frame_mvs = oxcf->enable_ref_frame_mvs; 1071 seq->order_hint_info.enable_ref_frame_mvs &= 1072 seq->order_hint_info.enable_order_hint; 1073 seq->enable_superres = oxcf->enable_superres; 1074 seq->enable_cdef = oxcf->enable_cdef; 1075 seq->enable_restoration = oxcf->enable_restoration; 1076 seq->enable_warped_motion = oxcf->enable_warped_motion; 1077 seq->enable_interintra_compound = oxcf->enable_interintra_comp; 1078 seq->enable_masked_compound = oxcf->enable_masked_comp; 1079 seq->enable_intra_edge_filter = oxcf->enable_intra_edge_filter; 1080 seq->enable_filter_intra = oxcf->enable_filter_intra; 1081 1082 set_bitstream_level_tier(seq, cm, oxcf); 1083 1084 if (seq->operating_points_cnt_minus_1 == 0) { 1085 seq->operating_point_idc[0] = 0; 1086 } else { 1087 // Set operating_point_idc[] such that for the i-th operating point the 1088 // first (operating_points_cnt-i) spatial layers and the first temporal 1089 // layer are decoded Note that highest quality operating point should come 1090 // first 1091 for (int i = 0; i < seq->operating_points_cnt_minus_1 + 1; i++) 1092 seq->operating_point_idc[i] = 1093 (~(~0u << (seq->operating_points_cnt_minus_1 + 1 - i)) << 8) | 1; 1094 } 1095 } 1096 1097 static void init_config(struct AV1_COMP *cpi, AV1EncoderConfig *oxcf) { 1098 AV1_COMMON *const cm = &cpi->common; 1099 1100 cpi->oxcf = *oxcf; 1101 cpi->framerate = oxcf->init_framerate; 1102 1103 cm->seq_params.profile = oxcf->profile; 1104 cm->seq_params.bit_depth = oxcf->bit_depth; 1105 cm->seq_params.use_highbitdepth = oxcf->use_highbitdepth; 1106 cm->seq_params.color_primaries = oxcf->color_primaries; 1107 cm->seq_params.transfer_characteristics = oxcf->transfer_characteristics; 1108 cm->seq_params.matrix_coefficients = oxcf->matrix_coefficients; 1109 cm->seq_params.monochrome = oxcf->monochrome; 1110 cm->seq_params.chroma_sample_position = oxcf->chroma_sample_position; 1111 cm->seq_params.color_range = oxcf->color_range; 1112 cm->timing_info_present = oxcf->timing_info_present; 1113 cm->timing_info.num_units_in_display_tick = 1114 oxcf->timing_info.num_units_in_display_tick; 1115 cm->timing_info.time_scale = oxcf->timing_info.time_scale; 1116 cm->timing_info.equal_picture_interval = 1117 oxcf->timing_info.equal_picture_interval; 1118 cm->timing_info.num_ticks_per_picture = 1119 oxcf->timing_info.num_ticks_per_picture; 1120 1121 cm->seq_params.display_model_info_present_flag = 1122 oxcf->display_model_info_present_flag; 1123 cm->seq_params.decoder_model_info_present_flag = 1124 oxcf->decoder_model_info_present_flag; 1125 if (oxcf->decoder_model_info_present_flag) { 1126 // set the decoder model parameters in schedule mode 1127 cm->buffer_model.num_units_in_decoding_tick = 1128 oxcf->buffer_model.num_units_in_decoding_tick; 1129 cm->buffer_removal_time_present = 1; 1130 set_aom_dec_model_info(&cm->buffer_model); 1131 set_dec_model_op_parameters(&cm->op_params[0]); 1132 } else if (cm->timing_info_present && 1133 cm->timing_info.equal_picture_interval && 1134 !cm->seq_params.decoder_model_info_present_flag) { 1135 // set the decoder model parameters in resource availability mode 1136 set_resource_availability_parameters(&cm->op_params[0]); 1137 } else { 1138 cm->op_params[0].initial_display_delay = 1139 10; // Default value (not signaled) 1140 } 1141 1142 if (cm->seq_params.monochrome) { 1143 cm->seq_params.subsampling_x = 1; 1144 cm->seq_params.subsampling_y = 1; 1145 } else if (cm->seq_params.color_primaries == AOM_CICP_CP_BT_709 && 1146 cm->seq_params.transfer_characteristics == AOM_CICP_TC_SRGB && 1147 cm->seq_params.matrix_coefficients == AOM_CICP_MC_IDENTITY) { 1148 cm->seq_params.subsampling_x = 0; 1149 cm->seq_params.subsampling_y = 0; 1150 } else { 1151 if (cm->seq_params.profile == 0) { 1152 cm->seq_params.subsampling_x = 1; 1153 cm->seq_params.subsampling_y = 1; 1154 } else if (cm->seq_params.profile == 1) { 1155 cm->seq_params.subsampling_x = 0; 1156 cm->seq_params.subsampling_y = 0; 1157 } else { 1158 if (cm->seq_params.bit_depth == AOM_BITS_12) { 1159 cm->seq_params.subsampling_x = oxcf->chroma_subsampling_x; 1160 cm->seq_params.subsampling_y = oxcf->chroma_subsampling_y; 1161 } else { 1162 cm->seq_params.subsampling_x = 1; 1163 cm->seq_params.subsampling_y = 0; 1164 } 1165 } 1166 } 1167 1168 cm->width = oxcf->width; 1169 cm->height = oxcf->height; 1170 set_sb_size(&cm->seq_params, 1171 select_sb_size(cpi)); // set sb size before allocations 1172 alloc_compressor_data(cpi); 1173 1174 update_film_grain_parameters(cpi, oxcf); 1175 1176 // Single thread case: use counts in common. 1177 cpi->td.counts = &cpi->counts; 1178 1179 // change includes all joint functionality 1180 av1_change_config(cpi, oxcf); 1181 1182 cpi->static_mb_pct = 0; 1183 cpi->ref_frame_flags = 0; 1184 1185 // Reset resize pending flags 1186 cpi->resize_pending_width = 0; 1187 cpi->resize_pending_height = 0; 1188 1189 init_buffer_indices(cpi); 1190 } 1191 1192 static void set_rc_buffer_sizes(RATE_CONTROL *rc, 1193 const AV1EncoderConfig *oxcf) { 1194 const int64_t bandwidth = oxcf->target_bandwidth; 1195 const int64_t starting = oxcf->starting_buffer_level_ms; 1196 const int64_t optimal = oxcf->optimal_buffer_level_ms; 1197 const int64_t maximum = oxcf->maximum_buffer_size_ms; 1198 1199 rc->starting_buffer_level = starting * bandwidth / 1000; 1200 rc->optimal_buffer_level = 1201 (optimal == 0) ? bandwidth / 8 : optimal * bandwidth / 1000; 1202 rc->maximum_buffer_size = 1203 (maximum == 0) ? bandwidth / 8 : maximum * bandwidth / 1000; 1204 } 1205 1206 #define HIGHBD_BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX4DF, JSDAF, JSVAF) \ 1207 cpi->fn_ptr[BT].sdf = SDF; \ 1208 cpi->fn_ptr[BT].sdaf = SDAF; \ 1209 cpi->fn_ptr[BT].vf = VF; \ 1210 cpi->fn_ptr[BT].svf = SVF; \ 1211 cpi->fn_ptr[BT].svaf = SVAF; \ 1212 cpi->fn_ptr[BT].sdx4df = SDX4DF; \ 1213 cpi->fn_ptr[BT].jsdaf = JSDAF; \ 1214 cpi->fn_ptr[BT].jsvaf = JSVAF; 1215 1216 #define MAKE_BFP_SAD_WRAPPER(fnname) \ 1217 static unsigned int fnname##_bits8(const uint8_t *src_ptr, \ 1218 int source_stride, \ 1219 const uint8_t *ref_ptr, int ref_stride) { \ 1220 return fnname(src_ptr, source_stride, ref_ptr, ref_stride); \ 1221 } \ 1222 static unsigned int fnname##_bits10( \ 1223 const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ 1224 int ref_stride) { \ 1225 return fnname(src_ptr, source_stride, ref_ptr, ref_stride) >> 2; \ 1226 } \ 1227 static unsigned int fnname##_bits12( \ 1228 const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ 1229 int ref_stride) { \ 1230 return fnname(src_ptr, source_stride, ref_ptr, ref_stride) >> 4; \ 1231 } 1232 1233 #define MAKE_BFP_SADAVG_WRAPPER(fnname) \ 1234 static unsigned int fnname##_bits8( \ 1235 const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ 1236 int ref_stride, const uint8_t *second_pred) { \ 1237 return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred); \ 1238 } \ 1239 static unsigned int fnname##_bits10( \ 1240 const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ 1241 int ref_stride, const uint8_t *second_pred) { \ 1242 return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred) >> \ 1243 2; \ 1244 } \ 1245 static unsigned int fnname##_bits12( \ 1246 const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ 1247 int ref_stride, const uint8_t *second_pred) { \ 1248 return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred) >> \ 1249 4; \ 1250 } 1251 1252 #define MAKE_BFP_SAD4D_WRAPPER(fnname) \ 1253 static void fnname##_bits8(const uint8_t *src_ptr, int source_stride, \ 1254 const uint8_t *const ref_ptr[], int ref_stride, \ 1255 unsigned int *sad_array) { \ 1256 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \ 1257 } \ 1258 static void fnname##_bits10(const uint8_t *src_ptr, int source_stride, \ 1259 const uint8_t *const ref_ptr[], int ref_stride, \ 1260 unsigned int *sad_array) { \ 1261 int i; \ 1262 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \ 1263 for (i = 0; i < 4; i++) sad_array[i] >>= 2; \ 1264 } \ 1265 static void fnname##_bits12(const uint8_t *src_ptr, int source_stride, \ 1266 const uint8_t *const ref_ptr[], int ref_stride, \ 1267 unsigned int *sad_array) { \ 1268 int i; \ 1269 fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \ 1270 for (i = 0; i < 4; i++) sad_array[i] >>= 4; \ 1271 } 1272 1273 #define MAKE_BFP_JSADAVG_WRAPPER(fnname) \ 1274 static unsigned int fnname##_bits8( \ 1275 const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ 1276 int ref_stride, const uint8_t *second_pred, \ 1277 const DIST_WTD_COMP_PARAMS *jcp_param) { \ 1278 return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred, \ 1279 jcp_param); \ 1280 } \ 1281 static unsigned int fnname##_bits10( \ 1282 const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ 1283 int ref_stride, const uint8_t *second_pred, \ 1284 const DIST_WTD_COMP_PARAMS *jcp_param) { \ 1285 return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred, \ 1286 jcp_param) >> \ 1287 2; \ 1288 } \ 1289 static unsigned int fnname##_bits12( \ 1290 const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ 1291 int ref_stride, const uint8_t *second_pred, \ 1292 const DIST_WTD_COMP_PARAMS *jcp_param) { \ 1293 return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred, \ 1294 jcp_param) >> \ 1295 4; \ 1296 } 1297 1298 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad128x128) 1299 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad128x128_avg) 1300 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad128x128x4d) 1301 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad128x64) 1302 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad128x64_avg) 1303 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad128x64x4d) 1304 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad64x128) 1305 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad64x128_avg) 1306 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad64x128x4d) 1307 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad32x16) 1308 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad32x16_avg) 1309 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad32x16x4d) 1310 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad16x32) 1311 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad16x32_avg) 1312 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad16x32x4d) 1313 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad64x32) 1314 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad64x32_avg) 1315 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad64x32x4d) 1316 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad32x64) 1317 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad32x64_avg) 1318 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad32x64x4d) 1319 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad32x32) 1320 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad32x32_avg) 1321 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad32x32x4d) 1322 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad64x64) 1323 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad64x64_avg) 1324 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad64x64x4d) 1325 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad16x16) 1326 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad16x16_avg) 1327 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad16x16x4d) 1328 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad16x8) 1329 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad16x8_avg) 1330 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad16x8x4d) 1331 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad8x16) 1332 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad8x16_avg) 1333 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad8x16x4d) 1334 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad8x8) 1335 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad8x8_avg) 1336 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad8x8x4d) 1337 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad8x4) 1338 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad8x4_avg) 1339 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad8x4x4d) 1340 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad4x8) 1341 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad4x8_avg) 1342 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad4x8x4d) 1343 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad4x4) 1344 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad4x4_avg) 1345 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad4x4x4d) 1346 1347 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad4x16) 1348 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad4x16_avg) 1349 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad4x16x4d) 1350 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad16x4) 1351 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad16x4_avg) 1352 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad16x4x4d) 1353 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad8x32) 1354 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad8x32_avg) 1355 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad8x32x4d) 1356 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad32x8) 1357 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad32x8_avg) 1358 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad32x8x4d) 1359 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad16x64) 1360 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad16x64_avg) 1361 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad16x64x4d) 1362 MAKE_BFP_SAD_WRAPPER(aom_highbd_sad64x16) 1363 MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad64x16_avg) 1364 MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad64x16x4d) 1365 1366 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad128x128_avg) 1367 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad128x64_avg) 1368 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad64x128_avg) 1369 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad32x16_avg) 1370 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad16x32_avg) 1371 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad64x32_avg) 1372 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad32x64_avg) 1373 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad32x32_avg) 1374 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad64x64_avg) 1375 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad16x16_avg) 1376 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad16x8_avg) 1377 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad8x16_avg) 1378 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad8x8_avg) 1379 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad8x4_avg) 1380 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad4x8_avg) 1381 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad4x4_avg) 1382 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad4x16_avg) 1383 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad16x4_avg) 1384 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad8x32_avg) 1385 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad32x8_avg) 1386 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad16x64_avg) 1387 MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad64x16_avg) 1388 1389 #define HIGHBD_MBFP(BT, MCSDF, MCSVF) \ 1390 cpi->fn_ptr[BT].msdf = MCSDF; \ 1391 cpi->fn_ptr[BT].msvf = MCSVF; 1392 1393 #define MAKE_MBFP_COMPOUND_SAD_WRAPPER(fnname) \ 1394 static unsigned int fnname##_bits8( \ 1395 const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ 1396 int ref_stride, const uint8_t *second_pred_ptr, const uint8_t *m, \ 1397 int m_stride, int invert_mask) { \ 1398 return fnname(src_ptr, source_stride, ref_ptr, ref_stride, \ 1399 second_pred_ptr, m, m_stride, invert_mask); \ 1400 } \ 1401 static unsigned int fnname##_bits10( \ 1402 const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ 1403 int ref_stride, const uint8_t *second_pred_ptr, const uint8_t *m, \ 1404 int m_stride, int invert_mask) { \ 1405 return fnname(src_ptr, source_stride, ref_ptr, ref_stride, \ 1406 second_pred_ptr, m, m_stride, invert_mask) >> \ 1407 2; \ 1408 } \ 1409 static unsigned int fnname##_bits12( \ 1410 const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ 1411 int ref_stride, const uint8_t *second_pred_ptr, const uint8_t *m, \ 1412 int m_stride, int invert_mask) { \ 1413 return fnname(src_ptr, source_stride, ref_ptr, ref_stride, \ 1414 second_pred_ptr, m, m_stride, invert_mask) >> \ 1415 4; \ 1416 } 1417 1418 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad128x128) 1419 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad128x64) 1420 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad64x128) 1421 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad64x64) 1422 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad64x32) 1423 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad32x64) 1424 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad32x32) 1425 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad32x16) 1426 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad16x32) 1427 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad16x16) 1428 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad16x8) 1429 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad8x16) 1430 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad8x8) 1431 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad8x4) 1432 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad4x8) 1433 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad4x4) 1434 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad4x16) 1435 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad16x4) 1436 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad8x32) 1437 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad32x8) 1438 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad16x64) 1439 MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad64x16) 1440 1441 #define HIGHBD_OBFP(BT, OSDF, OVF, OSVF) \ 1442 cpi->fn_ptr[BT].osdf = OSDF; \ 1443 cpi->fn_ptr[BT].ovf = OVF; \ 1444 cpi->fn_ptr[BT].osvf = OSVF; 1445 1446 #define MAKE_OBFP_SAD_WRAPPER(fnname) \ 1447 static unsigned int fnname##_bits8(const uint8_t *ref, int ref_stride, \ 1448 const int32_t *wsrc, \ 1449 const int32_t *msk) { \ 1450 return fnname(ref, ref_stride, wsrc, msk); \ 1451 } \ 1452 static unsigned int fnname##_bits10(const uint8_t *ref, int ref_stride, \ 1453 const int32_t *wsrc, \ 1454 const int32_t *msk) { \ 1455 return fnname(ref, ref_stride, wsrc, msk) >> 2; \ 1456 } \ 1457 static unsigned int fnname##_bits12(const uint8_t *ref, int ref_stride, \ 1458 const int32_t *wsrc, \ 1459 const int32_t *msk) { \ 1460 return fnname(ref, ref_stride, wsrc, msk) >> 4; \ 1461 } 1462 1463 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad128x128) 1464 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad128x64) 1465 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad64x128) 1466 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad64x64) 1467 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad64x32) 1468 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad32x64) 1469 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad32x32) 1470 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad32x16) 1471 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad16x32) 1472 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad16x16) 1473 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad16x8) 1474 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad8x16) 1475 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad8x8) 1476 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad8x4) 1477 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad4x8) 1478 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad4x4) 1479 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad4x16) 1480 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad16x4) 1481 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad8x32) 1482 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad32x8) 1483 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad16x64) 1484 MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad64x16) 1485 1486 static void highbd_set_var_fns(AV1_COMP *const cpi) { 1487 AV1_COMMON *const cm = &cpi->common; 1488 if (cm->seq_params.use_highbitdepth) { 1489 switch (cm->seq_params.bit_depth) { 1490 case AOM_BITS_8: 1491 HIGHBD_BFP(BLOCK_64X16, aom_highbd_sad64x16_bits8, 1492 aom_highbd_sad64x16_avg_bits8, aom_highbd_8_variance64x16, 1493 aom_highbd_8_sub_pixel_variance64x16, 1494 aom_highbd_8_sub_pixel_avg_variance64x16, 1495 aom_highbd_sad64x16x4d_bits8, 1496 aom_highbd_dist_wtd_sad64x16_avg_bits8, 1497 aom_highbd_8_dist_wtd_sub_pixel_avg_variance64x16) 1498 1499 HIGHBD_BFP(BLOCK_16X64, aom_highbd_sad16x64_bits8, 1500 aom_highbd_sad16x64_avg_bits8, aom_highbd_8_variance16x64, 1501 aom_highbd_8_sub_pixel_variance16x64, 1502 aom_highbd_8_sub_pixel_avg_variance16x64, 1503 aom_highbd_sad16x64x4d_bits8, 1504 aom_highbd_dist_wtd_sad16x64_avg_bits8, 1505 aom_highbd_8_dist_wtd_sub_pixel_avg_variance16x64) 1506 1507 HIGHBD_BFP( 1508 BLOCK_32X8, aom_highbd_sad32x8_bits8, aom_highbd_sad32x8_avg_bits8, 1509 aom_highbd_8_variance32x8, aom_highbd_8_sub_pixel_variance32x8, 1510 aom_highbd_8_sub_pixel_avg_variance32x8, 1511 aom_highbd_sad32x8x4d_bits8, aom_highbd_dist_wtd_sad32x8_avg_bits8, 1512 aom_highbd_8_dist_wtd_sub_pixel_avg_variance32x8) 1513 1514 HIGHBD_BFP( 1515 BLOCK_8X32, aom_highbd_sad8x32_bits8, aom_highbd_sad8x32_avg_bits8, 1516 aom_highbd_8_variance8x32, aom_highbd_8_sub_pixel_variance8x32, 1517 aom_highbd_8_sub_pixel_avg_variance8x32, 1518 aom_highbd_sad8x32x4d_bits8, aom_highbd_dist_wtd_sad8x32_avg_bits8, 1519 aom_highbd_8_dist_wtd_sub_pixel_avg_variance8x32) 1520 1521 HIGHBD_BFP( 1522 BLOCK_16X4, aom_highbd_sad16x4_bits8, aom_highbd_sad16x4_avg_bits8, 1523 aom_highbd_8_variance16x4, aom_highbd_8_sub_pixel_variance16x4, 1524 aom_highbd_8_sub_pixel_avg_variance16x4, 1525 aom_highbd_sad16x4x4d_bits8, aom_highbd_dist_wtd_sad16x4_avg_bits8, 1526 aom_highbd_8_dist_wtd_sub_pixel_avg_variance16x4) 1527 1528 HIGHBD_BFP( 1529 BLOCK_4X16, aom_highbd_sad4x16_bits8, aom_highbd_sad4x16_avg_bits8, 1530 aom_highbd_8_variance4x16, aom_highbd_8_sub_pixel_variance4x16, 1531 aom_highbd_8_sub_pixel_avg_variance4x16, 1532 aom_highbd_sad4x16x4d_bits8, aom_highbd_dist_wtd_sad4x16_avg_bits8, 1533 aom_highbd_8_dist_wtd_sub_pixel_avg_variance4x16) 1534 1535 HIGHBD_BFP(BLOCK_32X16, aom_highbd_sad32x16_bits8, 1536 aom_highbd_sad32x16_avg_bits8, aom_highbd_8_variance32x16, 1537 aom_highbd_8_sub_pixel_variance32x16, 1538 aom_highbd_8_sub_pixel_avg_variance32x16, 1539 aom_highbd_sad32x16x4d_bits8, 1540 aom_highbd_dist_wtd_sad32x16_avg_bits8, 1541 aom_highbd_8_dist_wtd_sub_pixel_avg_variance32x16) 1542 1543 HIGHBD_BFP(BLOCK_16X32, aom_highbd_sad16x32_bits8, 1544 aom_highbd_sad16x32_avg_bits8, aom_highbd_8_variance16x32, 1545 aom_highbd_8_sub_pixel_variance16x32, 1546 aom_highbd_8_sub_pixel_avg_variance16x32, 1547 aom_highbd_sad16x32x4d_bits8, 1548 aom_highbd_dist_wtd_sad16x32_avg_bits8, 1549 aom_highbd_8_dist_wtd_sub_pixel_avg_variance16x32) 1550 1551 HIGHBD_BFP(BLOCK_64X32, aom_highbd_sad64x32_bits8, 1552 aom_highbd_sad64x32_avg_bits8, aom_highbd_8_variance64x32, 1553 aom_highbd_8_sub_pixel_variance64x32, 1554 aom_highbd_8_sub_pixel_avg_variance64x32, 1555 aom_highbd_sad64x32x4d_bits8, 1556 aom_highbd_dist_wtd_sad64x32_avg_bits8, 1557 aom_highbd_8_dist_wtd_sub_pixel_avg_variance64x32) 1558 1559 HIGHBD_BFP(BLOCK_32X64, aom_highbd_sad32x64_bits8, 1560 aom_highbd_sad32x64_avg_bits8, aom_highbd_8_variance32x64, 1561 aom_highbd_8_sub_pixel_variance32x64, 1562 aom_highbd_8_sub_pixel_avg_variance32x64, 1563 aom_highbd_sad32x64x4d_bits8, 1564 aom_highbd_dist_wtd_sad32x64_avg_bits8, 1565 aom_highbd_8_dist_wtd_sub_pixel_avg_variance32x64) 1566 1567 HIGHBD_BFP(BLOCK_32X32, aom_highbd_sad32x32_bits8, 1568 aom_highbd_sad32x32_avg_bits8, aom_highbd_8_variance32x32, 1569 aom_highbd_8_sub_pixel_variance32x32, 1570 aom_highbd_8_sub_pixel_avg_variance32x32, 1571 aom_highbd_sad32x32x4d_bits8, 1572 aom_highbd_dist_wtd_sad32x32_avg_bits8, 1573 aom_highbd_8_dist_wtd_sub_pixel_avg_variance32x32) 1574 1575 HIGHBD_BFP(BLOCK_64X64, aom_highbd_sad64x64_bits8, 1576 aom_highbd_sad64x64_avg_bits8, aom_highbd_8_variance64x64, 1577 aom_highbd_8_sub_pixel_variance64x64, 1578 aom_highbd_8_sub_pixel_avg_variance64x64, 1579 aom_highbd_sad64x64x4d_bits8, 1580 aom_highbd_dist_wtd_sad64x64_avg_bits8, 1581 aom_highbd_8_dist_wtd_sub_pixel_avg_variance64x64) 1582 1583 HIGHBD_BFP(BLOCK_16X16, aom_highbd_sad16x16_bits8, 1584 aom_highbd_sad16x16_avg_bits8, aom_highbd_8_variance16x16, 1585 aom_highbd_8_sub_pixel_variance16x16, 1586 aom_highbd_8_sub_pixel_avg_variance16x16, 1587 aom_highbd_sad16x16x4d_bits8, 1588 aom_highbd_dist_wtd_sad16x16_avg_bits8, 1589 aom_highbd_8_dist_wtd_sub_pixel_avg_variance16x16) 1590 1591 HIGHBD_BFP( 1592 BLOCK_16X8, aom_highbd_sad16x8_bits8, aom_highbd_sad16x8_avg_bits8, 1593 aom_highbd_8_variance16x8, aom_highbd_8_sub_pixel_variance16x8, 1594 aom_highbd_8_sub_pixel_avg_variance16x8, 1595 aom_highbd_sad16x8x4d_bits8, aom_highbd_dist_wtd_sad16x8_avg_bits8, 1596 aom_highbd_8_dist_wtd_sub_pixel_avg_variance16x8) 1597 1598 HIGHBD_BFP( 1599 BLOCK_8X16, aom_highbd_sad8x16_bits8, aom_highbd_sad8x16_avg_bits8, 1600 aom_highbd_8_variance8x16, aom_highbd_8_sub_pixel_variance8x16, 1601 aom_highbd_8_sub_pixel_avg_variance8x16, 1602 aom_highbd_sad8x16x4d_bits8, aom_highbd_dist_wtd_sad8x16_avg_bits8, 1603 aom_highbd_8_dist_wtd_sub_pixel_avg_variance8x16) 1604 1605 HIGHBD_BFP( 1606 BLOCK_8X8, aom_highbd_sad8x8_bits8, aom_highbd_sad8x8_avg_bits8, 1607 aom_highbd_8_variance8x8, aom_highbd_8_sub_pixel_variance8x8, 1608 aom_highbd_8_sub_pixel_avg_variance8x8, aom_highbd_sad8x8x4d_bits8, 1609 aom_highbd_dist_wtd_sad8x8_avg_bits8, 1610 aom_highbd_8_dist_wtd_sub_pixel_avg_variance8x8) 1611 1612 HIGHBD_BFP( 1613 BLOCK_8X4, aom_highbd_sad8x4_bits8, aom_highbd_sad8x4_avg_bits8, 1614 aom_highbd_8_variance8x4, aom_highbd_8_sub_pixel_variance8x4, 1615 aom_highbd_8_sub_pixel_avg_variance8x4, aom_highbd_sad8x4x4d_bits8, 1616 aom_highbd_dist_wtd_sad8x4_avg_bits8, 1617 aom_highbd_8_dist_wtd_sub_pixel_avg_variance8x4) 1618 1619 HIGHBD_BFP( 1620 BLOCK_4X8, aom_highbd_sad4x8_bits8, aom_highbd_sad4x8_avg_bits8, 1621 aom_highbd_8_variance4x8, aom_highbd_8_sub_pixel_variance4x8, 1622 aom_highbd_8_sub_pixel_avg_variance4x8, aom_highbd_sad4x8x4d_bits8, 1623 aom_highbd_dist_wtd_sad4x8_avg_bits8, 1624 aom_highbd_8_dist_wtd_sub_pixel_avg_variance4x8) 1625 1626 HIGHBD_BFP( 1627 BLOCK_4X4, aom_highbd_sad4x4_bits8, aom_highbd_sad4x4_avg_bits8, 1628 aom_highbd_8_variance4x4, aom_highbd_8_sub_pixel_variance4x4, 1629 aom_highbd_8_sub_pixel_avg_variance4x4, aom_highbd_sad4x4x4d_bits8, 1630 aom_highbd_dist_wtd_sad4x4_avg_bits8, 1631 aom_highbd_8_dist_wtd_sub_pixel_avg_variance4x4) 1632 1633 HIGHBD_BFP(BLOCK_128X128, aom_highbd_sad128x128_bits8, 1634 aom_highbd_sad128x128_avg_bits8, 1635 aom_highbd_8_variance128x128, 1636 aom_highbd_8_sub_pixel_variance128x128, 1637 aom_highbd_8_sub_pixel_avg_variance128x128, 1638 aom_highbd_sad128x128x4d_bits8, 1639 aom_highbd_dist_wtd_sad128x128_avg_bits8, 1640 aom_highbd_8_dist_wtd_sub_pixel_avg_variance128x128) 1641 1642 HIGHBD_BFP(BLOCK_128X64, aom_highbd_sad128x64_bits8, 1643 aom_highbd_sad128x64_avg_bits8, aom_highbd_8_variance128x64, 1644 aom_highbd_8_sub_pixel_variance128x64, 1645 aom_highbd_8_sub_pixel_avg_variance128x64, 1646 aom_highbd_sad128x64x4d_bits8, 1647 aom_highbd_dist_wtd_sad128x64_avg_bits8, 1648 aom_highbd_8_dist_wtd_sub_pixel_avg_variance128x64) 1649 1650 HIGHBD_BFP(BLOCK_64X128, aom_highbd_sad64x128_bits8, 1651 aom_highbd_sad64x128_avg_bits8, aom_highbd_8_variance64x128, 1652 aom_highbd_8_sub_pixel_variance64x128, 1653 aom_highbd_8_sub_pixel_avg_variance64x128, 1654 aom_highbd_sad64x128x4d_bits8, 1655 aom_highbd_dist_wtd_sad64x128_avg_bits8, 1656 aom_highbd_8_dist_wtd_sub_pixel_avg_variance64x128) 1657 1658 HIGHBD_MBFP(BLOCK_128X128, aom_highbd_masked_sad128x128_bits8, 1659 aom_highbd_8_masked_sub_pixel_variance128x128) 1660 HIGHBD_MBFP(BLOCK_128X64, aom_highbd_masked_sad128x64_bits8, 1661 aom_highbd_8_masked_sub_pixel_variance128x64) 1662 HIGHBD_MBFP(BLOCK_64X128, aom_highbd_masked_sad64x128_bits8, 1663 aom_highbd_8_masked_sub_pixel_variance64x128) 1664 HIGHBD_MBFP(BLOCK_64X64, aom_highbd_masked_sad64x64_bits8, 1665 aom_highbd_8_masked_sub_pixel_variance64x64) 1666 HIGHBD_MBFP(BLOCK_64X32, aom_highbd_masked_sad64x32_bits8, 1667 aom_highbd_8_masked_sub_pixel_variance64x32) 1668 HIGHBD_MBFP(BLOCK_32X64, aom_highbd_masked_sad32x64_bits8, 1669 aom_highbd_8_masked_sub_pixel_variance32x64) 1670 HIGHBD_MBFP(BLOCK_32X32, aom_highbd_masked_sad32x32_bits8, 1671 aom_highbd_8_masked_sub_pixel_variance32x32) 1672 HIGHBD_MBFP(BLOCK_32X16, aom_highbd_masked_sad32x16_bits8, 1673 aom_highbd_8_masked_sub_pixel_variance32x16) 1674 HIGHBD_MBFP(BLOCK_16X32, aom_highbd_masked_sad16x32_bits8, 1675 aom_highbd_8_masked_sub_pixel_variance16x32) 1676 HIGHBD_MBFP(BLOCK_16X16, aom_highbd_masked_sad16x16_bits8, 1677 aom_highbd_8_masked_sub_pixel_variance16x16) 1678 HIGHBD_MBFP(BLOCK_8X16, aom_highbd_masked_sad8x16_bits8, 1679 aom_highbd_8_masked_sub_pixel_variance8x16) 1680 HIGHBD_MBFP(BLOCK_16X8, aom_highbd_masked_sad16x8_bits8, 1681 aom_highbd_8_masked_sub_pixel_variance16x8) 1682 HIGHBD_MBFP(BLOCK_8X8, aom_highbd_masked_sad8x8_bits8, 1683 aom_highbd_8_masked_sub_pixel_variance8x8) 1684 HIGHBD_MBFP(BLOCK_4X8, aom_highbd_masked_sad4x8_bits8, 1685 aom_highbd_8_masked_sub_pixel_variance4x8) 1686 HIGHBD_MBFP(BLOCK_8X4, aom_highbd_masked_sad8x4_bits8, 1687 aom_highbd_8_masked_sub_pixel_variance8x4) 1688 HIGHBD_MBFP(BLOCK_4X4, aom_highbd_masked_sad4x4_bits8, 1689 aom_highbd_8_masked_sub_pixel_variance4x4) 1690 HIGHBD_MBFP(BLOCK_64X16, aom_highbd_masked_sad64x16_bits8, 1691 aom_highbd_8_masked_sub_pixel_variance64x16) 1692 HIGHBD_MBFP(BLOCK_16X64, aom_highbd_masked_sad16x64_bits8, 1693 aom_highbd_8_masked_sub_pixel_variance16x64) 1694 HIGHBD_MBFP(BLOCK_32X8, aom_highbd_masked_sad32x8_bits8, 1695 aom_highbd_8_masked_sub_pixel_variance32x8) 1696 HIGHBD_MBFP(BLOCK_8X32, aom_highbd_masked_sad8x32_bits8, 1697 aom_highbd_8_masked_sub_pixel_variance8x32) 1698 HIGHBD_MBFP(BLOCK_16X4, aom_highbd_masked_sad16x4_bits8, 1699 aom_highbd_8_masked_sub_pixel_variance16x4) 1700 HIGHBD_MBFP(BLOCK_4X16, aom_highbd_masked_sad4x16_bits8, 1701 aom_highbd_8_masked_sub_pixel_variance4x16) 1702 HIGHBD_OBFP(BLOCK_128X128, aom_highbd_obmc_sad128x128_bits8, 1703 aom_highbd_obmc_variance128x128, 1704 aom_highbd_obmc_sub_pixel_variance128x128) 1705 HIGHBD_OBFP(BLOCK_128X64, aom_highbd_obmc_sad128x64_bits8, 1706 aom_highbd_obmc_variance128x64, 1707 aom_highbd_obmc_sub_pixel_variance128x64) 1708 HIGHBD_OBFP(BLOCK_64X128, aom_highbd_obmc_sad64x128_bits8, 1709 aom_highbd_obmc_variance64x128, 1710 aom_highbd_obmc_sub_pixel_variance64x128) 1711 HIGHBD_OBFP(BLOCK_64X64, aom_highbd_obmc_sad64x64_bits8, 1712 aom_highbd_obmc_variance64x64, 1713 aom_highbd_obmc_sub_pixel_variance64x64) 1714 HIGHBD_OBFP(BLOCK_64X32, aom_highbd_obmc_sad64x32_bits8, 1715 aom_highbd_obmc_variance64x32, 1716 aom_highbd_obmc_sub_pixel_variance64x32) 1717 HIGHBD_OBFP(BLOCK_32X64, aom_highbd_obmc_sad32x64_bits8, 1718 aom_highbd_obmc_variance32x64, 1719 aom_highbd_obmc_sub_pixel_variance32x64) 1720 HIGHBD_OBFP(BLOCK_32X32, aom_highbd_obmc_sad32x32_bits8, 1721 aom_highbd_obmc_variance32x32, 1722 aom_highbd_obmc_sub_pixel_variance32x32) 1723 HIGHBD_OBFP(BLOCK_32X16, aom_highbd_obmc_sad32x16_bits8, 1724 aom_highbd_obmc_variance32x16, 1725 aom_highbd_obmc_sub_pixel_variance32x16) 1726 HIGHBD_OBFP(BLOCK_16X32, aom_highbd_obmc_sad16x32_bits8, 1727 aom_highbd_obmc_variance16x32, 1728 aom_highbd_obmc_sub_pixel_variance16x32) 1729 HIGHBD_OBFP(BLOCK_16X16, aom_highbd_obmc_sad16x16_bits8, 1730 aom_highbd_obmc_variance16x16, 1731 aom_highbd_obmc_sub_pixel_variance16x16) 1732 HIGHBD_OBFP(BLOCK_8X16, aom_highbd_obmc_sad8x16_bits8, 1733 aom_highbd_obmc_variance8x16, 1734 aom_highbd_obmc_sub_pixel_variance8x16) 1735 HIGHBD_OBFP(BLOCK_16X8, aom_highbd_obmc_sad16x8_bits8, 1736 aom_highbd_obmc_variance16x8, 1737 aom_highbd_obmc_sub_pixel_variance16x8) 1738 HIGHBD_OBFP(BLOCK_8X8, aom_highbd_obmc_sad8x8_bits8, 1739 aom_highbd_obmc_variance8x8, 1740 aom_highbd_obmc_sub_pixel_variance8x8) 1741 HIGHBD_OBFP(BLOCK_4X8, aom_highbd_obmc_sad4x8_bits8, 1742 aom_highbd_obmc_variance4x8, 1743 aom_highbd_obmc_sub_pixel_variance4x8) 1744 HIGHBD_OBFP(BLOCK_8X4, aom_highbd_obmc_sad8x4_bits8, 1745 aom_highbd_obmc_variance8x4, 1746 aom_highbd_obmc_sub_pixel_variance8x4) 1747 HIGHBD_OBFP(BLOCK_4X4, aom_highbd_obmc_sad4x4_bits8, 1748 aom_highbd_obmc_variance4x4, 1749 aom_highbd_obmc_sub_pixel_variance4x4) 1750 HIGHBD_OBFP(BLOCK_64X16, aom_highbd_obmc_sad64x16_bits8, 1751 aom_highbd_obmc_variance64x16, 1752 aom_highbd_obmc_sub_pixel_variance64x16) 1753 HIGHBD_OBFP(BLOCK_16X64, aom_highbd_obmc_sad16x64_bits8, 1754 aom_highbd_obmc_variance16x64, 1755 aom_highbd_obmc_sub_pixel_variance16x64) 1756 HIGHBD_OBFP(BLOCK_32X8, aom_highbd_obmc_sad32x8_bits8, 1757 aom_highbd_obmc_variance32x8, 1758 aom_highbd_obmc_sub_pixel_variance32x8) 1759 HIGHBD_OBFP(BLOCK_8X32, aom_highbd_obmc_sad8x32_bits8, 1760 aom_highbd_obmc_variance8x32, 1761 aom_highbd_obmc_sub_pixel_variance8x32) 1762 HIGHBD_OBFP(BLOCK_16X4, aom_highbd_obmc_sad16x4_bits8, 1763 aom_highbd_obmc_variance16x4, 1764 aom_highbd_obmc_sub_pixel_variance16x4) 1765 HIGHBD_OBFP(BLOCK_4X16, aom_highbd_obmc_sad4x16_bits8, 1766 aom_highbd_obmc_variance4x16, 1767 aom_highbd_obmc_sub_pixel_variance4x16) 1768 break; 1769 1770 case AOM_BITS_10: 1771 HIGHBD_BFP(BLOCK_64X16, aom_highbd_sad64x16_bits10, 1772 aom_highbd_sad64x16_avg_bits10, aom_highbd_10_variance64x16, 1773 aom_highbd_10_sub_pixel_variance64x16, 1774 aom_highbd_10_sub_pixel_avg_variance64x16, 1775 aom_highbd_sad64x16x4d_bits10, 1776 aom_highbd_dist_wtd_sad64x16_avg_bits10, 1777 aom_highbd_10_dist_wtd_sub_pixel_avg_variance64x16); 1778 1779 HIGHBD_BFP(BLOCK_16X64, aom_highbd_sad16x64_bits10, 1780 aom_highbd_sad16x64_avg_bits10, aom_highbd_10_variance16x64, 1781 aom_highbd_10_sub_pixel_variance16x64, 1782 aom_highbd_10_sub_pixel_avg_variance16x64, 1783 aom_highbd_sad16x64x4d_bits10, 1784 aom_highbd_dist_wtd_sad16x64_avg_bits10, 1785 aom_highbd_10_dist_wtd_sub_pixel_avg_variance16x64); 1786 1787 HIGHBD_BFP(BLOCK_32X8, aom_highbd_sad32x8_bits10, 1788 aom_highbd_sad32x8_avg_bits10, aom_highbd_10_variance32x8, 1789 aom_highbd_10_sub_pixel_variance32x8, 1790 aom_highbd_10_sub_pixel_avg_variance32x8, 1791 aom_highbd_sad32x8x4d_bits10, 1792 aom_highbd_dist_wtd_sad32x8_avg_bits10, 1793 aom_highbd_10_dist_wtd_sub_pixel_avg_variance32x8); 1794 1795 HIGHBD_BFP(BLOCK_8X32, aom_highbd_sad8x32_bits10, 1796 aom_highbd_sad8x32_avg_bits10, aom_highbd_10_variance8x32, 1797 aom_highbd_10_sub_pixel_variance8x32, 1798 aom_highbd_10_sub_pixel_avg_variance8x32, 1799 aom_highbd_sad8x32x4d_bits10, 1800 aom_highbd_dist_wtd_sad8x32_avg_bits10, 1801 aom_highbd_10_dist_wtd_sub_pixel_avg_variance8x32); 1802 1803 HIGHBD_BFP(BLOCK_16X4, aom_highbd_sad16x4_bits10, 1804 aom_highbd_sad16x4_avg_bits10, aom_highbd_10_variance16x4, 1805 aom_highbd_10_sub_pixel_variance16x4, 1806 aom_highbd_10_sub_pixel_avg_variance16x4, 1807 aom_highbd_sad16x4x4d_bits10, 1808 aom_highbd_dist_wtd_sad16x4_avg_bits10, 1809 aom_highbd_10_dist_wtd_sub_pixel_avg_variance16x4); 1810 1811 HIGHBD_BFP(BLOCK_4X16, aom_highbd_sad4x16_bits10, 1812 aom_highbd_sad4x16_avg_bits10, aom_highbd_10_variance4x16, 1813 aom_highbd_10_sub_pixel_variance4x16, 1814 aom_highbd_10_sub_pixel_avg_variance4x16, 1815 aom_highbd_sad4x16x4d_bits10, 1816 aom_highbd_dist_wtd_sad4x16_avg_bits10, 1817 aom_highbd_10_dist_wtd_sub_pixel_avg_variance4x16); 1818 1819 HIGHBD_BFP(BLOCK_32X16, aom_highbd_sad32x16_bits10, 1820 aom_highbd_sad32x16_avg_bits10, aom_highbd_10_variance32x16, 1821 aom_highbd_10_sub_pixel_variance32x16, 1822 aom_highbd_10_sub_pixel_avg_variance32x16, 1823 aom_highbd_sad32x16x4d_bits10, 1824 aom_highbd_dist_wtd_sad32x16_avg_bits10, 1825 aom_highbd_10_dist_wtd_sub_pixel_avg_variance32x16); 1826 1827 HIGHBD_BFP(BLOCK_16X32, aom_highbd_sad16x32_bits10, 1828 aom_highbd_sad16x32_avg_bits10, aom_highbd_10_variance16x32, 1829 aom_highbd_10_sub_pixel_variance16x32, 1830 aom_highbd_10_sub_pixel_avg_variance16x32, 1831 aom_highbd_sad16x32x4d_bits10, 1832 aom_highbd_dist_wtd_sad16x32_avg_bits10, 1833 aom_highbd_10_dist_wtd_sub_pixel_avg_variance16x32); 1834 1835 HIGHBD_BFP(BLOCK_64X32, aom_highbd_sad64x32_bits10, 1836 aom_highbd_sad64x32_avg_bits10, aom_highbd_10_variance64x32, 1837 aom_highbd_10_sub_pixel_variance64x32, 1838 aom_highbd_10_sub_pixel_avg_variance64x32, 1839 aom_highbd_sad64x32x4d_bits10, 1840 aom_highbd_dist_wtd_sad64x32_avg_bits10, 1841 aom_highbd_10_dist_wtd_sub_pixel_avg_variance64x32); 1842 1843 HIGHBD_BFP(BLOCK_32X64, aom_highbd_sad32x64_bits10, 1844 aom_highbd_sad32x64_avg_bits10, aom_highbd_10_variance32x64, 1845 aom_highbd_10_sub_pixel_variance32x64, 1846 aom_highbd_10_sub_pixel_avg_variance32x64, 1847 aom_highbd_sad32x64x4d_bits10, 1848 aom_highbd_dist_wtd_sad32x64_avg_bits10, 1849 aom_highbd_10_dist_wtd_sub_pixel_avg_variance32x64); 1850 1851 HIGHBD_BFP(BLOCK_32X32, aom_highbd_sad32x32_bits10, 1852 aom_highbd_sad32x32_avg_bits10, aom_highbd_10_variance32x32, 1853 aom_highbd_10_sub_pixel_variance32x32, 1854 aom_highbd_10_sub_pixel_avg_variance32x32, 1855 aom_highbd_sad32x32x4d_bits10, 1856 aom_highbd_dist_wtd_sad32x32_avg_bits10, 1857 aom_highbd_10_dist_wtd_sub_pixel_avg_variance32x32); 1858 1859 HIGHBD_BFP(BLOCK_64X64, aom_highbd_sad64x64_bits10, 1860 aom_highbd_sad64x64_avg_bits10, aom_highbd_10_variance64x64, 1861 aom_highbd_10_sub_pixel_variance64x64, 1862 aom_highbd_10_sub_pixel_avg_variance64x64, 1863 aom_highbd_sad64x64x4d_bits10, 1864 aom_highbd_dist_wtd_sad64x64_avg_bits10, 1865 aom_highbd_10_dist_wtd_sub_pixel_avg_variance64x64); 1866 1867 HIGHBD_BFP(BLOCK_16X16, aom_highbd_sad16x16_bits10, 1868 aom_highbd_sad16x16_avg_bits10, aom_highbd_10_variance16x16, 1869 aom_highbd_10_sub_pixel_variance16x16, 1870 aom_highbd_10_sub_pixel_avg_variance16x16, 1871 aom_highbd_sad16x16x4d_bits10, 1872 aom_highbd_dist_wtd_sad16x16_avg_bits10, 1873 aom_highbd_10_dist_wtd_sub_pixel_avg_variance16x16); 1874 1875 HIGHBD_BFP(BLOCK_16X8, aom_highbd_sad16x8_bits10, 1876 aom_highbd_sad16x8_avg_bits10, aom_highbd_10_variance16x8, 1877 aom_highbd_10_sub_pixel_variance16x8, 1878 aom_highbd_10_sub_pixel_avg_variance16x8, 1879 aom_highbd_sad16x8x4d_bits10, 1880 aom_highbd_dist_wtd_sad16x8_avg_bits10, 1881 aom_highbd_10_dist_wtd_sub_pixel_avg_variance16x8); 1882 1883 HIGHBD_BFP(BLOCK_8X16, aom_highbd_sad8x16_bits10, 1884 aom_highbd_sad8x16_avg_bits10, aom_highbd_10_variance8x16, 1885 aom_highbd_10_sub_pixel_variance8x16, 1886 aom_highbd_10_sub_pixel_avg_variance8x16, 1887 aom_highbd_sad8x16x4d_bits10, 1888 aom_highbd_dist_wtd_sad8x16_avg_bits10, 1889 aom_highbd_10_dist_wtd_sub_pixel_avg_variance8x16); 1890 1891 HIGHBD_BFP( 1892 BLOCK_8X8, aom_highbd_sad8x8_bits10, aom_highbd_sad8x8_avg_bits10, 1893 aom_highbd_10_variance8x8, aom_highbd_10_sub_pixel_variance8x8, 1894 aom_highbd_10_sub_pixel_avg_variance8x8, 1895 aom_highbd_sad8x8x4d_bits10, aom_highbd_dist_wtd_sad8x8_avg_bits10, 1896 aom_highbd_10_dist_wtd_sub_pixel_avg_variance8x8); 1897 1898 HIGHBD_BFP( 1899 BLOCK_8X4, aom_highbd_sad8x4_bits10, aom_highbd_sad8x4_avg_bits10, 1900 aom_highbd_10_variance8x4, aom_highbd_10_sub_pixel_variance8x4, 1901 aom_highbd_10_sub_pixel_avg_variance8x4, 1902 aom_highbd_sad8x4x4d_bits10, aom_highbd_dist_wtd_sad8x4_avg_bits10, 1903 aom_highbd_10_dist_wtd_sub_pixel_avg_variance8x4); 1904 1905 HIGHBD_BFP( 1906 BLOCK_4X8, aom_highbd_sad4x8_bits10, aom_highbd_sad4x8_avg_bits10, 1907 aom_highbd_10_variance4x8, aom_highbd_10_sub_pixel_variance4x8, 1908 aom_highbd_10_sub_pixel_avg_variance4x8, 1909 aom_highbd_sad4x8x4d_bits10, aom_highbd_dist_wtd_sad4x8_avg_bits10, 1910 aom_highbd_10_dist_wtd_sub_pixel_avg_variance4x8); 1911 1912 HIGHBD_BFP( 1913 BLOCK_4X4, aom_highbd_sad4x4_bits10, aom_highbd_sad4x4_avg_bits10, 1914 aom_highbd_10_variance4x4, aom_highbd_10_sub_pixel_variance4x4, 1915 aom_highbd_10_sub_pixel_avg_variance4x4, 1916 aom_highbd_sad4x4x4d_bits10, aom_highbd_dist_wtd_sad4x4_avg_bits10, 1917 aom_highbd_10_dist_wtd_sub_pixel_avg_variance4x4); 1918 1919 HIGHBD_BFP(BLOCK_128X128, aom_highbd_sad128x128_bits10, 1920 aom_highbd_sad128x128_avg_bits10, 1921 aom_highbd_10_variance128x128, 1922 aom_highbd_10_sub_pixel_variance128x128, 1923 aom_highbd_10_sub_pixel_avg_variance128x128, 1924 aom_highbd_sad128x128x4d_bits10, 1925 aom_highbd_dist_wtd_sad128x128_avg_bits10, 1926 aom_highbd_10_dist_wtd_sub_pixel_avg_variance128x128); 1927 1928 HIGHBD_BFP(BLOCK_128X64, aom_highbd_sad128x64_bits10, 1929 aom_highbd_sad128x64_avg_bits10, 1930 aom_highbd_10_variance128x64, 1931 aom_highbd_10_sub_pixel_variance128x64, 1932 aom_highbd_10_sub_pixel_avg_variance128x64, 1933 aom_highbd_sad128x64x4d_bits10, 1934 aom_highbd_dist_wtd_sad128x64_avg_bits10, 1935 aom_highbd_10_dist_wtd_sub_pixel_avg_variance128x64); 1936 1937 HIGHBD_BFP(BLOCK_64X128, aom_highbd_sad64x128_bits10, 1938 aom_highbd_sad64x128_avg_bits10, 1939 aom_highbd_10_variance64x128, 1940 aom_highbd_10_sub_pixel_variance64x128, 1941 aom_highbd_10_sub_pixel_avg_variance64x128, 1942 aom_highbd_sad64x128x4d_bits10, 1943 aom_highbd_dist_wtd_sad64x128_avg_bits10, 1944 aom_highbd_10_dist_wtd_sub_pixel_avg_variance64x128); 1945 1946 HIGHBD_MBFP(BLOCK_128X128, aom_highbd_masked_sad128x128_bits10, 1947 aom_highbd_10_masked_sub_pixel_variance128x128) 1948 HIGHBD_MBFP(BLOCK_128X64, aom_highbd_masked_sad128x64_bits10, 1949 aom_highbd_10_masked_sub_pixel_variance128x64) 1950 HIGHBD_MBFP(BLOCK_64X128, aom_highbd_masked_sad64x128_bits10, 1951 aom_highbd_10_masked_sub_pixel_variance64x128) 1952 HIGHBD_MBFP(BLOCK_64X64, aom_highbd_masked_sad64x64_bits10, 1953 aom_highbd_10_masked_sub_pixel_variance64x64) 1954 HIGHBD_MBFP(BLOCK_64X32, aom_highbd_masked_sad64x32_bits10, 1955 aom_highbd_10_masked_sub_pixel_variance64x32) 1956 HIGHBD_MBFP(BLOCK_32X64, aom_highbd_masked_sad32x64_bits10, 1957 aom_highbd_10_masked_sub_pixel_variance32x64) 1958 HIGHBD_MBFP(BLOCK_32X32, aom_highbd_masked_sad32x32_bits10, 1959 aom_highbd_10_masked_sub_pixel_variance32x32) 1960 HIGHBD_MBFP(BLOCK_32X16, aom_highbd_masked_sad32x16_bits10, 1961 aom_highbd_10_masked_sub_pixel_variance32x16) 1962 HIGHBD_MBFP(BLOCK_16X32, aom_highbd_masked_sad16x32_bits10, 1963 aom_highbd_10_masked_sub_pixel_variance16x32) 1964 HIGHBD_MBFP(BLOCK_16X16, aom_highbd_masked_sad16x16_bits10, 1965 aom_highbd_10_masked_sub_pixel_variance16x16) 1966 HIGHBD_MBFP(BLOCK_8X16, aom_highbd_masked_sad8x16_bits10, 1967 aom_highbd_10_masked_sub_pixel_variance8x16) 1968 HIGHBD_MBFP(BLOCK_16X8, aom_highbd_masked_sad16x8_bits10, 1969 aom_highbd_10_masked_sub_pixel_variance16x8) 1970 HIGHBD_MBFP(BLOCK_8X8, aom_highbd_masked_sad8x8_bits10, 1971 aom_highbd_10_masked_sub_pixel_variance8x8) 1972 HIGHBD_MBFP(BLOCK_4X8, aom_highbd_masked_sad4x8_bits10, 1973 aom_highbd_10_masked_sub_pixel_variance4x8) 1974 HIGHBD_MBFP(BLOCK_8X4, aom_highbd_masked_sad8x4_bits10, 1975 aom_highbd_10_masked_sub_pixel_variance8x4) 1976 HIGHBD_MBFP(BLOCK_4X4, aom_highbd_masked_sad4x4_bits10, 1977 aom_highbd_10_masked_sub_pixel_variance4x4) 1978 HIGHBD_MBFP(BLOCK_64X16, aom_highbd_masked_sad64x16_bits10, 1979 aom_highbd_10_masked_sub_pixel_variance64x16) 1980 HIGHBD_MBFP(BLOCK_16X64, aom_highbd_masked_sad16x64_bits10, 1981 aom_highbd_10_masked_sub_pixel_variance16x64) 1982 HIGHBD_MBFP(BLOCK_32X8, aom_highbd_masked_sad32x8_bits10, 1983 aom_highbd_10_masked_sub_pixel_variance32x8) 1984 HIGHBD_MBFP(BLOCK_8X32, aom_highbd_masked_sad8x32_bits10, 1985 aom_highbd_10_masked_sub_pixel_variance8x32) 1986 HIGHBD_MBFP(BLOCK_16X4, aom_highbd_masked_sad16x4_bits10, 1987 aom_highbd_10_masked_sub_pixel_variance16x4) 1988 HIGHBD_MBFP(BLOCK_4X16, aom_highbd_masked_sad4x16_bits10, 1989 aom_highbd_10_masked_sub_pixel_variance4x16) 1990 HIGHBD_OBFP(BLOCK_128X128, aom_highbd_obmc_sad128x128_bits10, 1991 aom_highbd_10_obmc_variance128x128, 1992 aom_highbd_10_obmc_sub_pixel_variance128x128) 1993 HIGHBD_OBFP(BLOCK_128X64, aom_highbd_obmc_sad128x64_bits10, 1994 aom_highbd_10_obmc_variance128x64, 1995 aom_highbd_10_obmc_sub_pixel_variance128x64) 1996 HIGHBD_OBFP(BLOCK_64X128, aom_highbd_obmc_sad64x128_bits10, 1997 aom_highbd_10_obmc_variance64x128, 1998 aom_highbd_10_obmc_sub_pixel_variance64x128) 1999 HIGHBD_OBFP(BLOCK_64X64, aom_highbd_obmc_sad64x64_bits10, 2000 aom_highbd_10_obmc_variance64x64, 2001 aom_highbd_10_obmc_sub_pixel_variance64x64) 2002 HIGHBD_OBFP(BLOCK_64X32, aom_highbd_obmc_sad64x32_bits10, 2003 aom_highbd_10_obmc_variance64x32, 2004 aom_highbd_10_obmc_sub_pixel_variance64x32) 2005 HIGHBD_OBFP(BLOCK_32X64, aom_highbd_obmc_sad32x64_bits10, 2006 aom_highbd_10_obmc_variance32x64, 2007 aom_highbd_10_obmc_sub_pixel_variance32x64) 2008 HIGHBD_OBFP(BLOCK_32X32, aom_highbd_obmc_sad32x32_bits10, 2009 aom_highbd_10_obmc_variance32x32, 2010 aom_highbd_10_obmc_sub_pixel_variance32x32) 2011 HIGHBD_OBFP(BLOCK_32X16, aom_highbd_obmc_sad32x16_bits10, 2012 aom_highbd_10_obmc_variance32x16, 2013 aom_highbd_10_obmc_sub_pixel_variance32x16) 2014 HIGHBD_OBFP(BLOCK_16X32, aom_highbd_obmc_sad16x32_bits10, 2015 aom_highbd_10_obmc_variance16x32, 2016 aom_highbd_10_obmc_sub_pixel_variance16x32) 2017 HIGHBD_OBFP(BLOCK_16X16, aom_highbd_obmc_sad16x16_bits10, 2018 aom_highbd_10_obmc_variance16x16, 2019 aom_highbd_10_obmc_sub_pixel_variance16x16) 2020 HIGHBD_OBFP(BLOCK_8X16, aom_highbd_obmc_sad8x16_bits10, 2021 aom_highbd_10_obmc_variance8x16, 2022 aom_highbd_10_obmc_sub_pixel_variance8x16) 2023 HIGHBD_OBFP(BLOCK_16X8, aom_highbd_obmc_sad16x8_bits10, 2024 aom_highbd_10_obmc_variance16x8, 2025 aom_highbd_10_obmc_sub_pixel_variance16x8) 2026 HIGHBD_OBFP(BLOCK_8X8, aom_highbd_obmc_sad8x8_bits10, 2027 aom_highbd_10_obmc_variance8x8, 2028 aom_highbd_10_obmc_sub_pixel_variance8x8) 2029 HIGHBD_OBFP(BLOCK_4X8, aom_highbd_obmc_sad4x8_bits10, 2030 aom_highbd_10_obmc_variance4x8, 2031 aom_highbd_10_obmc_sub_pixel_variance4x8) 2032 HIGHBD_OBFP(BLOCK_8X4, aom_highbd_obmc_sad8x4_bits10, 2033 aom_highbd_10_obmc_variance8x4, 2034 aom_highbd_10_obmc_sub_pixel_variance8x4) 2035 HIGHBD_OBFP(BLOCK_4X4, aom_highbd_obmc_sad4x4_bits10, 2036 aom_highbd_10_obmc_variance4x4, 2037 aom_highbd_10_obmc_sub_pixel_variance4x4) 2038 2039 HIGHBD_OBFP(BLOCK_64X16, aom_highbd_obmc_sad64x16_bits10, 2040 aom_highbd_10_obmc_variance64x16, 2041 aom_highbd_10_obmc_sub_pixel_variance64x16) 2042 2043 HIGHBD_OBFP(BLOCK_16X64, aom_highbd_obmc_sad16x64_bits10, 2044 aom_highbd_10_obmc_variance16x64, 2045 aom_highbd_10_obmc_sub_pixel_variance16x64) 2046 2047 HIGHBD_OBFP(BLOCK_32X8, aom_highbd_obmc_sad32x8_bits10, 2048 aom_highbd_10_obmc_variance32x8, 2049 aom_highbd_10_obmc_sub_pixel_variance32x8) 2050 2051 HIGHBD_OBFP(BLOCK_8X32, aom_highbd_obmc_sad8x32_bits10, 2052 aom_highbd_10_obmc_variance8x32, 2053 aom_highbd_10_obmc_sub_pixel_variance8x32) 2054 2055 HIGHBD_OBFP(BLOCK_16X4, aom_highbd_obmc_sad16x4_bits10, 2056 aom_highbd_10_obmc_variance16x4, 2057 aom_highbd_10_obmc_sub_pixel_variance16x4) 2058 2059 HIGHBD_OBFP(BLOCK_4X16, aom_highbd_obmc_sad4x16_bits10, 2060 aom_highbd_10_obmc_variance4x16, 2061 aom_highbd_10_obmc_sub_pixel_variance4x16) 2062 break; 2063 2064 case AOM_BITS_12: 2065 HIGHBD_BFP(BLOCK_64X16, aom_highbd_sad64x16_bits12, 2066 aom_highbd_sad64x16_avg_bits12, aom_highbd_12_variance64x16, 2067 aom_highbd_12_sub_pixel_variance64x16, 2068 aom_highbd_12_sub_pixel_avg_variance64x16, 2069 aom_highbd_sad64x16x4d_bits12, 2070 aom_highbd_dist_wtd_sad64x16_avg_bits12, 2071 aom_highbd_12_dist_wtd_sub_pixel_avg_variance64x16); 2072 2073 HIGHBD_BFP(BLOCK_16X64, aom_highbd_sad16x64_bits12, 2074 aom_highbd_sad16x64_avg_bits12, aom_highbd_12_variance16x64, 2075 aom_highbd_12_sub_pixel_variance16x64, 2076 aom_highbd_12_sub_pixel_avg_variance16x64, 2077 aom_highbd_sad16x64x4d_bits12, 2078 aom_highbd_dist_wtd_sad16x64_avg_bits12, 2079 aom_highbd_12_dist_wtd_sub_pixel_avg_variance16x64); 2080 2081 HIGHBD_BFP(BLOCK_32X8, aom_highbd_sad32x8_bits12, 2082 aom_highbd_sad32x8_avg_bits12, aom_highbd_12_variance32x8, 2083 aom_highbd_12_sub_pixel_variance32x8, 2084 aom_highbd_12_sub_pixel_avg_variance32x8, 2085 aom_highbd_sad32x8x4d_bits12, 2086 aom_highbd_dist_wtd_sad32x8_avg_bits12, 2087 aom_highbd_12_dist_wtd_sub_pixel_avg_variance32x8); 2088 2089 HIGHBD_BFP(BLOCK_8X32, aom_highbd_sad8x32_bits12, 2090 aom_highbd_sad8x32_avg_bits12, aom_highbd_12_variance8x32, 2091 aom_highbd_12_sub_pixel_variance8x32, 2092 aom_highbd_12_sub_pixel_avg_variance8x32, 2093 aom_highbd_sad8x32x4d_bits12, 2094 aom_highbd_dist_wtd_sad8x32_avg_bits12, 2095 aom_highbd_12_dist_wtd_sub_pixel_avg_variance8x32); 2096 2097 HIGHBD_BFP(BLOCK_16X4, aom_highbd_sad16x4_bits12, 2098 aom_highbd_sad16x4_avg_bits12, aom_highbd_12_variance16x4, 2099 aom_highbd_12_sub_pixel_variance16x4, 2100 aom_highbd_12_sub_pixel_avg_variance16x4, 2101 aom_highbd_sad16x4x4d_bits12, 2102 aom_highbd_dist_wtd_sad16x4_avg_bits12, 2103 aom_highbd_12_dist_wtd_sub_pixel_avg_variance16x4); 2104 2105 HIGHBD_BFP(BLOCK_4X16, aom_highbd_sad4x16_bits12, 2106 aom_highbd_sad4x16_avg_bits12, aom_highbd_12_variance4x16, 2107 aom_highbd_12_sub_pixel_variance4x16, 2108 aom_highbd_12_sub_pixel_avg_variance4x16, 2109 aom_highbd_sad4x16x4d_bits12, 2110 aom_highbd_dist_wtd_sad4x16_avg_bits12, 2111 aom_highbd_12_dist_wtd_sub_pixel_avg_variance4x16); 2112 2113 HIGHBD_BFP(BLOCK_32X16, aom_highbd_sad32x16_bits12, 2114 aom_highbd_sad32x16_avg_bits12, aom_highbd_12_variance32x16, 2115 aom_highbd_12_sub_pixel_variance32x16, 2116 aom_highbd_12_sub_pixel_avg_variance32x16, 2117 aom_highbd_sad32x16x4d_bits12, 2118 aom_highbd_dist_wtd_sad32x16_avg_bits12, 2119 aom_highbd_12_dist_wtd_sub_pixel_avg_variance32x16); 2120 2121 HIGHBD_BFP(BLOCK_16X32, aom_highbd_sad16x32_bits12, 2122 aom_highbd_sad16x32_avg_bits12, aom_highbd_12_variance16x32, 2123 aom_highbd_12_sub_pixel_variance16x32, 2124 aom_highbd_12_sub_pixel_avg_variance16x32, 2125 aom_highbd_sad16x32x4d_bits12, 2126 aom_highbd_dist_wtd_sad16x32_avg_bits12, 2127 aom_highbd_12_dist_wtd_sub_pixel_avg_variance16x32); 2128 2129 HIGHBD_BFP(BLOCK_64X32, aom_highbd_sad64x32_bits12, 2130 aom_highbd_sad64x32_avg_bits12, aom_highbd_12_variance64x32, 2131 aom_highbd_12_sub_pixel_variance64x32, 2132 aom_highbd_12_sub_pixel_avg_variance64x32, 2133 aom_highbd_sad64x32x4d_bits12, 2134 aom_highbd_dist_wtd_sad64x32_avg_bits12, 2135 aom_highbd_12_dist_wtd_sub_pixel_avg_variance64x32); 2136 2137 HIGHBD_BFP(BLOCK_32X64, aom_highbd_sad32x64_bits12, 2138 aom_highbd_sad32x64_avg_bits12, aom_highbd_12_variance32x64, 2139 aom_highbd_12_sub_pixel_variance32x64, 2140 aom_highbd_12_sub_pixel_avg_variance32x64, 2141 aom_highbd_sad32x64x4d_bits12, 2142 aom_highbd_dist_wtd_sad32x64_avg_bits12, 2143 aom_highbd_12_dist_wtd_sub_pixel_avg_variance32x64); 2144 2145 HIGHBD_BFP(BLOCK_32X32, aom_highbd_sad32x32_bits12, 2146 aom_highbd_sad32x32_avg_bits12, aom_highbd_12_variance32x32, 2147 aom_highbd_12_sub_pixel_variance32x32, 2148 aom_highbd_12_sub_pixel_avg_variance32x32, 2149 aom_highbd_sad32x32x4d_bits12, 2150 aom_highbd_dist_wtd_sad32x32_avg_bits12, 2151 aom_highbd_12_dist_wtd_sub_pixel_avg_variance32x32); 2152 2153 HIGHBD_BFP(BLOCK_64X64, aom_highbd_sad64x64_bits12, 2154 aom_highbd_sad64x64_avg_bits12, aom_highbd_12_variance64x64, 2155 aom_highbd_12_sub_pixel_variance64x64, 2156 aom_highbd_12_sub_pixel_avg_variance64x64, 2157 aom_highbd_sad64x64x4d_bits12, 2158 aom_highbd_dist_wtd_sad64x64_avg_bits12, 2159 aom_highbd_12_dist_wtd_sub_pixel_avg_variance64x64); 2160 2161 HIGHBD_BFP(BLOCK_16X16, aom_highbd_sad16x16_bits12, 2162 aom_highbd_sad16x16_avg_bits12, aom_highbd_12_variance16x16, 2163 aom_highbd_12_sub_pixel_variance16x16, 2164 aom_highbd_12_sub_pixel_avg_variance16x16, 2165 aom_highbd_sad16x16x4d_bits12, 2166 aom_highbd_dist_wtd_sad16x16_avg_bits12, 2167 aom_highbd_12_dist_wtd_sub_pixel_avg_variance16x16); 2168 2169 HIGHBD_BFP(BLOCK_16X8, aom_highbd_sad16x8_bits12, 2170 aom_highbd_sad16x8_avg_bits12, aom_highbd_12_variance16x8, 2171 aom_highbd_12_sub_pixel_variance16x8, 2172 aom_highbd_12_sub_pixel_avg_variance16x8, 2173 aom_highbd_sad16x8x4d_bits12, 2174 aom_highbd_dist_wtd_sad16x8_avg_bits12, 2175 aom_highbd_12_dist_wtd_sub_pixel_avg_variance16x8); 2176 2177 HIGHBD_BFP(BLOCK_8X16, aom_highbd_sad8x16_bits12, 2178 aom_highbd_sad8x16_avg_bits12, aom_highbd_12_variance8x16, 2179 aom_highbd_12_sub_pixel_variance8x16, 2180 aom_highbd_12_sub_pixel_avg_variance8x16, 2181 aom_highbd_sad8x16x4d_bits12, 2182 aom_highbd_dist_wtd_sad8x16_avg_bits12, 2183 aom_highbd_12_dist_wtd_sub_pixel_avg_variance8x16); 2184 2185 HIGHBD_BFP( 2186 BLOCK_8X8, aom_highbd_sad8x8_bits12, aom_highbd_sad8x8_avg_bits12, 2187 aom_highbd_12_variance8x8, aom_highbd_12_sub_pixel_variance8x8, 2188 aom_highbd_12_sub_pixel_avg_variance8x8, 2189 aom_highbd_sad8x8x4d_bits12, aom_highbd_dist_wtd_sad8x8_avg_bits12, 2190 aom_highbd_12_dist_wtd_sub_pixel_avg_variance8x8); 2191 2192 HIGHBD_BFP( 2193 BLOCK_8X4, aom_highbd_sad8x4_bits12, aom_highbd_sad8x4_avg_bits12, 2194 aom_highbd_12_variance8x4, aom_highbd_12_sub_pixel_variance8x4, 2195 aom_highbd_12_sub_pixel_avg_variance8x4, 2196 aom_highbd_sad8x4x4d_bits12, aom_highbd_dist_wtd_sad8x4_avg_bits12, 2197 aom_highbd_12_dist_wtd_sub_pixel_avg_variance8x4); 2198 2199 HIGHBD_BFP( 2200 BLOCK_4X8, aom_highbd_sad4x8_bits12, aom_highbd_sad4x8_avg_bits12, 2201 aom_highbd_12_variance4x8, aom_highbd_12_sub_pixel_variance4x8, 2202 aom_highbd_12_sub_pixel_avg_variance4x8, 2203 aom_highbd_sad4x8x4d_bits12, aom_highbd_dist_wtd_sad4x8_avg_bits12, 2204 aom_highbd_12_dist_wtd_sub_pixel_avg_variance4x8); 2205 2206 HIGHBD_BFP( 2207 BLOCK_4X4, aom_highbd_sad4x4_bits12, aom_highbd_sad4x4_avg_bits12, 2208 aom_highbd_12_variance4x4, aom_highbd_12_sub_pixel_variance4x4, 2209 aom_highbd_12_sub_pixel_avg_variance4x4, 2210 aom_highbd_sad4x4x4d_bits12, aom_highbd_dist_wtd_sad4x4_avg_bits12, 2211 aom_highbd_12_dist_wtd_sub_pixel_avg_variance4x4); 2212 2213 HIGHBD_BFP(BLOCK_128X128, aom_highbd_sad128x128_bits12, 2214 aom_highbd_sad128x128_avg_bits12, 2215 aom_highbd_12_variance128x128, 2216 aom_highbd_12_sub_pixel_variance128x128, 2217 aom_highbd_12_sub_pixel_avg_variance128x128, 2218 aom_highbd_sad128x128x4d_bits12, 2219 aom_highbd_dist_wtd_sad128x128_avg_bits12, 2220 aom_highbd_12_dist_wtd_sub_pixel_avg_variance128x128); 2221 2222 HIGHBD_BFP(BLOCK_128X64, aom_highbd_sad128x64_bits12, 2223 aom_highbd_sad128x64_avg_bits12, 2224 aom_highbd_12_variance128x64, 2225 aom_highbd_12_sub_pixel_variance128x64, 2226 aom_highbd_12_sub_pixel_avg_variance128x64, 2227 aom_highbd_sad128x64x4d_bits12, 2228 aom_highbd_dist_wtd_sad128x64_avg_bits12, 2229 aom_highbd_12_dist_wtd_sub_pixel_avg_variance128x64); 2230 2231 HIGHBD_BFP(BLOCK_64X128, aom_highbd_sad64x128_bits12, 2232 aom_highbd_sad64x128_avg_bits12, 2233 aom_highbd_12_variance64x128, 2234 aom_highbd_12_sub_pixel_variance64x128, 2235 aom_highbd_12_sub_pixel_avg_variance64x128, 2236 aom_highbd_sad64x128x4d_bits12, 2237 aom_highbd_dist_wtd_sad64x128_avg_bits12, 2238 aom_highbd_12_dist_wtd_sub_pixel_avg_variance64x128); 2239 2240 HIGHBD_MBFP(BLOCK_128X128, aom_highbd_masked_sad128x128_bits12, 2241 aom_highbd_12_masked_sub_pixel_variance128x128) 2242 HIGHBD_MBFP(BLOCK_128X64, aom_highbd_masked_sad128x64_bits12, 2243 aom_highbd_12_masked_sub_pixel_variance128x64) 2244 HIGHBD_MBFP(BLOCK_64X128, aom_highbd_masked_sad64x128_bits12, 2245 aom_highbd_12_masked_sub_pixel_variance64x128) 2246 HIGHBD_MBFP(BLOCK_64X64, aom_highbd_masked_sad64x64_bits12, 2247 aom_highbd_12_masked_sub_pixel_variance64x64) 2248 HIGHBD_MBFP(BLOCK_64X32, aom_highbd_masked_sad64x32_bits12, 2249 aom_highbd_12_masked_sub_pixel_variance64x32) 2250 HIGHBD_MBFP(BLOCK_32X64, aom_highbd_masked_sad32x64_bits12, 2251 aom_highbd_12_masked_sub_pixel_variance32x64) 2252 HIGHBD_MBFP(BLOCK_32X32, aom_highbd_masked_sad32x32_bits12, 2253 aom_highbd_12_masked_sub_pixel_variance32x32) 2254 HIGHBD_MBFP(BLOCK_32X16, aom_highbd_masked_sad32x16_bits12, 2255 aom_highbd_12_masked_sub_pixel_variance32x16) 2256 HIGHBD_MBFP(BLOCK_16X32, aom_highbd_masked_sad16x32_bits12, 2257 aom_highbd_12_masked_sub_pixel_variance16x32) 2258 HIGHBD_MBFP(BLOCK_16X16, aom_highbd_masked_sad16x16_bits12, 2259 aom_highbd_12_masked_sub_pixel_variance16x16) 2260 HIGHBD_MBFP(BLOCK_8X16, aom_highbd_masked_sad8x16_bits12, 2261 aom_highbd_12_masked_sub_pixel_variance8x16) 2262 HIGHBD_MBFP(BLOCK_16X8, aom_highbd_masked_sad16x8_bits12, 2263 aom_highbd_12_masked_sub_pixel_variance16x8) 2264 HIGHBD_MBFP(BLOCK_8X8, aom_highbd_masked_sad8x8_bits12, 2265 aom_highbd_12_masked_sub_pixel_variance8x8) 2266 HIGHBD_MBFP(BLOCK_4X8, aom_highbd_masked_sad4x8_bits12, 2267 aom_highbd_12_masked_sub_pixel_variance4x8) 2268 HIGHBD_MBFP(BLOCK_8X4, aom_highbd_masked_sad8x4_bits12, 2269 aom_highbd_12_masked_sub_pixel_variance8x4) 2270 HIGHBD_MBFP(BLOCK_4X4, aom_highbd_masked_sad4x4_bits12, 2271 aom_highbd_12_masked_sub_pixel_variance4x4) 2272 HIGHBD_MBFP(BLOCK_64X16, aom_highbd_masked_sad64x16_bits12, 2273 aom_highbd_12_masked_sub_pixel_variance64x16) 2274 HIGHBD_MBFP(BLOCK_16X64, aom_highbd_masked_sad16x64_bits12, 2275 aom_highbd_12_masked_sub_pixel_variance16x64) 2276 HIGHBD_MBFP(BLOCK_32X8, aom_highbd_masked_sad32x8_bits12, 2277 aom_highbd_12_masked_sub_pixel_variance32x8) 2278 HIGHBD_MBFP(BLOCK_8X32, aom_highbd_masked_sad8x32_bits12, 2279 aom_highbd_12_masked_sub_pixel_variance8x32) 2280 HIGHBD_MBFP(BLOCK_16X4, aom_highbd_masked_sad16x4_bits12, 2281 aom_highbd_12_masked_sub_pixel_variance16x4) 2282 HIGHBD_MBFP(BLOCK_4X16, aom_highbd_masked_sad4x16_bits12, 2283 aom_highbd_12_masked_sub_pixel_variance4x16) 2284 HIGHBD_OBFP(BLOCK_128X128, aom_highbd_obmc_sad128x128_bits12, 2285 aom_highbd_12_obmc_variance128x128, 2286 aom_highbd_12_obmc_sub_pixel_variance128x128) 2287 HIGHBD_OBFP(BLOCK_128X64, aom_highbd_obmc_sad128x64_bits12, 2288 aom_highbd_12_obmc_variance128x64, 2289 aom_highbd_12_obmc_sub_pixel_variance128x64) 2290 HIGHBD_OBFP(BLOCK_64X128, aom_highbd_obmc_sad64x128_bits12, 2291 aom_highbd_12_obmc_variance64x128, 2292 aom_highbd_12_obmc_sub_pixel_variance64x128) 2293 HIGHBD_OBFP(BLOCK_64X64, aom_highbd_obmc_sad64x64_bits12, 2294 aom_highbd_12_obmc_variance64x64, 2295 aom_highbd_12_obmc_sub_pixel_variance64x64) 2296 HIGHBD_OBFP(BLOCK_64X32, aom_highbd_obmc_sad64x32_bits12, 2297 aom_highbd_12_obmc_variance64x32, 2298 aom_highbd_12_obmc_sub_pixel_variance64x32) 2299 HIGHBD_OBFP(BLOCK_32X64, aom_highbd_obmc_sad32x64_bits12, 2300 aom_highbd_12_obmc_variance32x64, 2301 aom_highbd_12_obmc_sub_pixel_variance32x64) 2302 HIGHBD_OBFP(BLOCK_32X32, aom_highbd_obmc_sad32x32_bits12, 2303 aom_highbd_12_obmc_variance32x32, 2304 aom_highbd_12_obmc_sub_pixel_variance32x32) 2305 HIGHBD_OBFP(BLOCK_32X16, aom_highbd_obmc_sad32x16_bits12, 2306 aom_highbd_12_obmc_variance32x16, 2307 aom_highbd_12_obmc_sub_pixel_variance32x16) 2308 HIGHBD_OBFP(BLOCK_16X32, aom_highbd_obmc_sad16x32_bits12, 2309 aom_highbd_12_obmc_variance16x32, 2310 aom_highbd_12_obmc_sub_pixel_variance16x32) 2311 HIGHBD_OBFP(BLOCK_16X16, aom_highbd_obmc_sad16x16_bits12, 2312 aom_highbd_12_obmc_variance16x16, 2313 aom_highbd_12_obmc_sub_pixel_variance16x16) 2314 HIGHBD_OBFP(BLOCK_8X16, aom_highbd_obmc_sad8x16_bits12, 2315 aom_highbd_12_obmc_variance8x16, 2316 aom_highbd_12_obmc_sub_pixel_variance8x16) 2317 HIGHBD_OBFP(BLOCK_16X8, aom_highbd_obmc_sad16x8_bits12, 2318 aom_highbd_12_obmc_variance16x8, 2319 aom_highbd_12_obmc_sub_pixel_variance16x8) 2320 HIGHBD_OBFP(BLOCK_8X8, aom_highbd_obmc_sad8x8_bits12, 2321 aom_highbd_12_obmc_variance8x8, 2322 aom_highbd_12_obmc_sub_pixel_variance8x8) 2323 HIGHBD_OBFP(BLOCK_4X8, aom_highbd_obmc_sad4x8_bits12, 2324 aom_highbd_12_obmc_variance4x8, 2325 aom_highbd_12_obmc_sub_pixel_variance4x8) 2326 HIGHBD_OBFP(BLOCK_8X4, aom_highbd_obmc_sad8x4_bits12, 2327 aom_highbd_12_obmc_variance8x4, 2328 aom_highbd_12_obmc_sub_pixel_variance8x4) 2329 HIGHBD_OBFP(BLOCK_4X4, aom_highbd_obmc_sad4x4_bits12, 2330 aom_highbd_12_obmc_variance4x4, 2331 aom_highbd_12_obmc_sub_pixel_variance4x4) 2332 HIGHBD_OBFP(BLOCK_64X16, aom_highbd_obmc_sad64x16_bits12, 2333 aom_highbd_12_obmc_variance64x16, 2334 aom_highbd_12_obmc_sub_pixel_variance64x16) 2335 HIGHBD_OBFP(BLOCK_16X64, aom_highbd_obmc_sad16x64_bits12, 2336 aom_highbd_12_obmc_variance16x64, 2337 aom_highbd_12_obmc_sub_pixel_variance16x64) 2338 HIGHBD_OBFP(BLOCK_32X8, aom_highbd_obmc_sad32x8_bits12, 2339 aom_highbd_12_obmc_variance32x8, 2340 aom_highbd_12_obmc_sub_pixel_variance32x8) 2341 HIGHBD_OBFP(BLOCK_8X32, aom_highbd_obmc_sad8x32_bits12, 2342 aom_highbd_12_obmc_variance8x32, 2343 aom_highbd_12_obmc_sub_pixel_variance8x32) 2344 HIGHBD_OBFP(BLOCK_16X4, aom_highbd_obmc_sad16x4_bits12, 2345 aom_highbd_12_obmc_variance16x4, 2346 aom_highbd_12_obmc_sub_pixel_variance16x4) 2347 HIGHBD_OBFP(BLOCK_4X16, aom_highbd_obmc_sad4x16_bits12, 2348 aom_highbd_12_obmc_variance4x16, 2349 aom_highbd_12_obmc_sub_pixel_variance4x16) 2350 break; 2351 2352 default: 2353 assert(0 && 2354 "cm->seq_params.bit_depth should be AOM_BITS_8, " 2355 "AOM_BITS_10 or AOM_BITS_12"); 2356 } 2357 } 2358 } 2359 2360 static void realloc_segmentation_maps(AV1_COMP *cpi) { 2361 AV1_COMMON *const cm = &cpi->common; 2362 2363 // Create the encoder segmentation map and set all entries to 0 2364 aom_free(cpi->segmentation_map); 2365 CHECK_MEM_ERROR(cm, cpi->segmentation_map, 2366 aom_calloc(cm->mi_rows * cm->mi_cols, 1)); 2367 2368 // Create a map used for cyclic background refresh. 2369 if (cpi->cyclic_refresh) av1_cyclic_refresh_free(cpi->cyclic_refresh); 2370 CHECK_MEM_ERROR(cm, cpi->cyclic_refresh, 2371 av1_cyclic_refresh_alloc(cm->mi_rows, cm->mi_cols)); 2372 2373 // Create a map used to mark inactive areas. 2374 aom_free(cpi->active_map.map); 2375 CHECK_MEM_ERROR(cm, cpi->active_map.map, 2376 aom_calloc(cm->mi_rows * cm->mi_cols, 1)); 2377 } 2378 2379 void av1_change_config(struct AV1_COMP *cpi, const AV1EncoderConfig *oxcf) { 2380 AV1_COMMON *const cm = &cpi->common; 2381 SequenceHeader *const seq_params = &cm->seq_params; 2382 const int num_planes = av1_num_planes(cm); 2383 RATE_CONTROL *const rc = &cpi->rc; 2384 MACROBLOCK *const x = &cpi->td.mb; 2385 2386 if (seq_params->profile != oxcf->profile) seq_params->profile = oxcf->profile; 2387 seq_params->bit_depth = oxcf->bit_depth; 2388 seq_params->color_primaries = oxcf->color_primaries; 2389 seq_params->transfer_characteristics = oxcf->transfer_characteristics; 2390 seq_params->matrix_coefficients = oxcf->matrix_coefficients; 2391 seq_params->monochrome = oxcf->monochrome; 2392 seq_params->chroma_sample_position = oxcf->chroma_sample_position; 2393 seq_params->color_range = oxcf->color_range; 2394 2395 assert(IMPLIES(seq_params->profile <= PROFILE_1, 2396 seq_params->bit_depth <= AOM_BITS_10)); 2397 2398 memcpy(cpi->target_seq_level_idx, oxcf->target_seq_level_idx, 2399 sizeof(cpi->target_seq_level_idx)); 2400 cpi->keep_level_stats = 0; 2401 for (int i = 0; i < MAX_NUM_OPERATING_POINTS; ++i) { 2402 if (cpi->target_seq_level_idx[i] < SEQ_LEVELS) { 2403 cpi->keep_level_stats = 1; 2404 break; 2405 } 2406 } 2407 2408 cm->timing_info_present = oxcf->timing_info_present; 2409 cm->timing_info.num_units_in_display_tick = 2410 oxcf->timing_info.num_units_in_display_tick; 2411 cm->timing_info.time_scale = oxcf->timing_info.time_scale; 2412 cm->timing_info.equal_picture_interval = 2413 oxcf->timing_info.equal_picture_interval; 2414 cm->timing_info.num_ticks_per_picture = 2415 oxcf->timing_info.num_ticks_per_picture; 2416 2417 seq_params->display_model_info_present_flag = 2418 oxcf->display_model_info_present_flag; 2419 seq_params->decoder_model_info_present_flag = 2420 oxcf->decoder_model_info_present_flag; 2421 if (oxcf->decoder_model_info_present_flag) { 2422 // set the decoder model parameters in schedule mode 2423 cm->buffer_model.num_units_in_decoding_tick = 2424 oxcf->buffer_model.num_units_in_decoding_tick; 2425 cm->buffer_removal_time_present = 1; 2426 set_aom_dec_model_info(&cm->buffer_model); 2427 set_dec_model_op_parameters(&cm->op_params[0]); 2428 } else if (cm->timing_info_present && 2429 cm->timing_info.equal_picture_interval && 2430 !seq_params->decoder_model_info_present_flag) { 2431 // set the decoder model parameters in resource availability mode 2432 set_resource_availability_parameters(&cm->op_params[0]); 2433 } else { 2434 cm->op_params[0].initial_display_delay = 2435 10; // Default value (not signaled) 2436 } 2437 2438 update_film_grain_parameters(cpi, oxcf); 2439 2440 cpi->oxcf = *oxcf; 2441 cpi->common.options = oxcf->cfg; 2442 x->e_mbd.bd = (int)seq_params->bit_depth; 2443 x->e_mbd.global_motion = cm->global_motion; 2444 2445 if ((oxcf->pass == 0) && (oxcf->rc_mode == AOM_Q)) { 2446 rc->baseline_gf_interval = FIXED_GF_INTERVAL; 2447 } else { 2448 rc->baseline_gf_interval = (MIN_GF_INTERVAL + MAX_GF_INTERVAL) / 2; 2449 } 2450 2451 cpi->refresh_last_frame = 1; 2452 cpi->refresh_golden_frame = 0; 2453 cpi->refresh_bwd_ref_frame = 0; 2454 cpi->refresh_alt2_ref_frame = 0; 2455 2456 cm->refresh_frame_context = (oxcf->frame_parallel_decoding_mode) 2457 ? REFRESH_FRAME_CONTEXT_DISABLED 2458 : REFRESH_FRAME_CONTEXT_BACKWARD; 2459 if (oxcf->large_scale_tile) 2460 cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED; 2461 2462 if (x->palette_buffer == NULL) { 2463 CHECK_MEM_ERROR(cm, x->palette_buffer, 2464 aom_memalign(16, sizeof(*x->palette_buffer))); 2465 } 2466 2467 if (x->tmp_conv_dst == NULL) { 2468 CHECK_MEM_ERROR( 2469 cm, x->tmp_conv_dst, 2470 aom_memalign(32, MAX_SB_SIZE * MAX_SB_SIZE * sizeof(*x->tmp_conv_dst))); 2471 x->e_mbd.tmp_conv_dst = x->tmp_conv_dst; 2472 } 2473 for (int i = 0; i < 2; ++i) { 2474 if (x->tmp_obmc_bufs[i] == NULL) { 2475 CHECK_MEM_ERROR(cm, x->tmp_obmc_bufs[i], 2476 aom_memalign(32, 2 * MAX_MB_PLANE * MAX_SB_SQUARE * 2477 sizeof(*x->tmp_obmc_bufs[i]))); 2478 x->e_mbd.tmp_obmc_bufs[i] = x->tmp_obmc_bufs[i]; 2479 } 2480 } 2481 2482 av1_reset_segment_features(cm); 2483 set_high_precision_mv(cpi, 1, 0); 2484 2485 set_rc_buffer_sizes(rc, &cpi->oxcf); 2486 2487 // Under a configuration change, where maximum_buffer_size may change, 2488 // keep buffer level clipped to the maximum allowed buffer size. 2489 rc->bits_off_target = AOMMIN(rc->bits_off_target, rc->maximum_buffer_size); 2490 rc->buffer_level = AOMMIN(rc->buffer_level, rc->maximum_buffer_size); 2491 2492 // Set up frame rate and related parameters rate control values. 2493 av1_new_framerate(cpi, cpi->framerate); 2494 2495 // Set absolute upper and lower quality limits 2496 rc->worst_quality = cpi->oxcf.worst_allowed_q; 2497 rc->best_quality = cpi->oxcf.best_allowed_q; 2498 2499 cm->interp_filter = oxcf->large_scale_tile ? EIGHTTAP_REGULAR : SWITCHABLE; 2500 cm->switchable_motion_mode = 1; 2501 2502 if (cpi->oxcf.render_width > 0 && cpi->oxcf.render_height > 0) { 2503 cm->render_width = cpi->oxcf.render_width; 2504 cm->render_height = cpi->oxcf.render_height; 2505 } else { 2506 cm->render_width = cpi->oxcf.width; 2507 cm->render_height = cpi->oxcf.height; 2508 } 2509 cm->width = cpi->oxcf.width; 2510 cm->height = cpi->oxcf.height; 2511 2512 int sb_size = seq_params->sb_size; 2513 // Superblock size should not be updated after the first key frame. 2514 if (!cpi->seq_params_locked) { 2515 set_sb_size(&cm->seq_params, select_sb_size(cpi)); 2516 for (int i = 0; i < MAX_NUM_OPERATING_POINTS; ++i) 2517 seq_params->tier[i] = (oxcf->tier_mask >> i) & 1; 2518 } 2519 2520 if (cpi->initial_width || sb_size != seq_params->sb_size) { 2521 if (cm->width > cpi->initial_width || cm->height > cpi->initial_height || 2522 seq_params->sb_size != sb_size) { 2523 av1_free_context_buffers(cm); 2524 av1_free_pc_tree(&cpi->td, num_planes); 2525 alloc_compressor_data(cpi); 2526 realloc_segmentation_maps(cpi); 2527 cpi->initial_width = cpi->initial_height = 0; 2528 } 2529 } 2530 update_frame_size(cpi); 2531 2532 cpi->alt_ref_source = NULL; 2533 rc->is_src_frame_alt_ref = 0; 2534 2535 set_tile_info(cpi); 2536 2537 cpi->ext_refresh_frame_flags_pending = 0; 2538 cpi->ext_refresh_frame_context_pending = 0; 2539 2540 highbd_set_var_fns(cpi); 2541 2542 // Init sequence level coding tools 2543 // This should not be called after the first key frame. 2544 if (!cpi->seq_params_locked) { 2545 seq_params->operating_points_cnt_minus_1 = 2546 cm->number_spatial_layers > 1 ? cm->number_spatial_layers - 1 : 0; 2547 init_seq_coding_tools(&cm->seq_params, cm, oxcf); 2548 } 2549 } 2550 2551 static void init_level_info(AV1LevelInfo *level_info) { 2552 memset(level_info, 0, MAX_NUM_OPERATING_POINTS * sizeof(*level_info)); 2553 for (int i = 0; i < MAX_NUM_OPERATING_POINTS; ++i) { 2554 AV1LevelSpec *const level_spec = &level_info[i].level_spec; 2555 level_spec->level = SEQ_LEVEL_MAX; 2556 AV1LevelStats *const level_stats = &level_info[i].level_stats; 2557 level_stats->min_cropped_tile_width = INT_MAX; 2558 level_stats->min_cropped_tile_height = INT_MAX; 2559 level_stats->min_frame_width = INT_MAX; 2560 level_stats->min_frame_height = INT_MAX; 2561 level_stats->tile_width_is_valid = 1; 2562 level_stats->min_cr = 1e8; 2563 } 2564 } 2565 2566 AV1_COMP *av1_create_compressor(AV1EncoderConfig *oxcf, 2567 BufferPool *const pool) { 2568 unsigned int i; 2569 AV1_COMP *volatile const cpi = aom_memalign(32, sizeof(AV1_COMP)); 2570 AV1_COMMON *volatile const cm = cpi != NULL ? &cpi->common : NULL; 2571 2572 if (!cm) return NULL; 2573 2574 av1_zero(*cpi); 2575 2576 // The jmp_buf is valid only for the duration of the function that calls 2577 // setjmp(). Therefore, this function must reset the 'setjmp' field to 0 2578 // before it returns. 2579 if (setjmp(cm->error.jmp)) { 2580 cm->error.setjmp = 0; 2581 av1_remove_compressor(cpi); 2582 return 0; 2583 } 2584 2585 cm->error.setjmp = 1; 2586 cm->alloc_mi = enc_alloc_mi; 2587 cm->free_mi = enc_free_mi; 2588 cm->setup_mi = enc_setup_mi; 2589 2590 CHECK_MEM_ERROR(cm, cm->fc, 2591 (FRAME_CONTEXT *)aom_memalign(32, sizeof(*cm->fc))); 2592 CHECK_MEM_ERROR( 2593 cm, cm->default_frame_context, 2594 (FRAME_CONTEXT *)aom_memalign(32, sizeof(*cm->default_frame_context))); 2595 memset(cm->fc, 0, sizeof(*cm->fc)); 2596 memset(cm->default_frame_context, 0, sizeof(*cm->default_frame_context)); 2597 2598 cpi->resize_state = 0; 2599 cpi->resize_avg_qp = 0; 2600 cpi->resize_buffer_underflow = 0; 2601 2602 cpi->common.buffer_pool = pool; 2603 2604 init_config(cpi, oxcf); 2605 av1_rc_init(&cpi->oxcf, oxcf->pass, &cpi->rc); 2606 2607 cm->current_frame.frame_number = 0; 2608 cm->current_frame_id = -1; 2609 cpi->seq_params_locked = 0; 2610 cpi->partition_search_skippable_frame = 0; 2611 cpi->tile_data = NULL; 2612 cpi->last_show_frame_buf = NULL; 2613 realloc_segmentation_maps(cpi); 2614 2615 memset(cpi->nmv_costs, 0, sizeof(cpi->nmv_costs)); 2616 memset(cpi->nmv_costs_hp, 0, sizeof(cpi->nmv_costs_hp)); 2617 2618 for (i = 0; i < (sizeof(cpi->mbgraph_stats) / sizeof(cpi->mbgraph_stats[0])); 2619 i++) { 2620 CHECK_MEM_ERROR( 2621 cm, cpi->mbgraph_stats[i].mb_stats, 2622 aom_calloc(cm->MBs * sizeof(*cpi->mbgraph_stats[i].mb_stats), 1)); 2623 } 2624 2625 cpi->refresh_alt_ref_frame = 0; 2626 2627 init_level_info(cpi->level_info); 2628 2629 cpi->b_calculate_psnr = CONFIG_INTERNAL_STATS; 2630 #if CONFIG_INTERNAL_STATS 2631 cpi->b_calculate_blockiness = 1; 2632 cpi->b_calculate_consistency = 1; 2633 cpi->total_inconsistency = 0; 2634 cpi->psnr.worst = 100.0; 2635 cpi->worst_ssim = 100.0; 2636 2637 cpi->count = 0; 2638 cpi->bytes = 0; 2639 #if CONFIG_SPEED_STATS 2640 cpi->tx_search_count = 0; 2641 #endif // CONFIG_SPEED_STATS 2642 2643 if (cpi->b_calculate_psnr) { 2644 cpi->total_sq_error = 0; 2645 cpi->total_samples = 0; 2646 cpi->tot_recode_hits = 0; 2647 cpi->summed_quality = 0; 2648 cpi->summed_weights = 0; 2649 } 2650 2651 cpi->fastssim.worst = 100.0; 2652 cpi->psnrhvs.worst = 100.0; 2653 2654 if (cpi->b_calculate_blockiness) { 2655 cpi->total_blockiness = 0; 2656 cpi->worst_blockiness = 0.0; 2657 } 2658 2659 if (cpi->b_calculate_consistency) { 2660 CHECK_MEM_ERROR(cm, cpi->ssim_vars, 2661 aom_malloc(sizeof(*cpi->ssim_vars) * 4 * 2662 cpi->common.mi_rows * cpi->common.mi_cols)); 2663 cpi->worst_consistency = 100.0; 2664 } 2665 #endif 2666 #if CONFIG_ENTROPY_STATS 2667 av1_zero(aggregate_fc); 2668 #endif // CONFIG_ENTROPY_STATS 2669 2670 cpi->first_time_stamp_ever = INT64_MAX; 2671 2672 cpi->td.mb.nmvcost[0] = &cpi->nmv_costs[0][MV_MAX]; 2673 cpi->td.mb.nmvcost[1] = &cpi->nmv_costs[1][MV_MAX]; 2674 cpi->td.mb.nmvcost_hp[0] = &cpi->nmv_costs_hp[0][MV_MAX]; 2675 cpi->td.mb.nmvcost_hp[1] = &cpi->nmv_costs_hp[1][MV_MAX]; 2676 2677 #ifdef OUTPUT_YUV_SKINMAP 2678 yuv_skinmap_file = fopen("skinmap.yuv", "ab"); 2679 #endif 2680 #ifdef OUTPUT_YUV_REC 2681 yuv_rec_file = fopen("rec.yuv", "wb"); 2682 #endif 2683 2684 if (oxcf->pass == 1) { 2685 av1_init_first_pass(cpi); 2686 } else if (oxcf->pass == 2) { 2687 const size_t packet_sz = sizeof(FIRSTPASS_STATS); 2688 const int packets = (int)(oxcf->two_pass_stats_in.sz / packet_sz); 2689 2690 cpi->twopass.stats_in_start = oxcf->two_pass_stats_in.buf; 2691 cpi->twopass.stats_in = cpi->twopass.stats_in_start; 2692 cpi->twopass.stats_in_end = &cpi->twopass.stats_in[packets - 1]; 2693 2694 av1_init_second_pass(cpi); 2695 } 2696 2697 CHECK_MEM_ERROR( 2698 cm, cpi->td.mb.above_pred_buf, 2699 (uint8_t *)aom_memalign(16, MAX_MB_PLANE * MAX_SB_SQUARE * 2700 sizeof(*cpi->td.mb.above_pred_buf))); 2701 CHECK_MEM_ERROR( 2702 cm, cpi->td.mb.left_pred_buf, 2703 (uint8_t *)aom_memalign(16, MAX_MB_PLANE * MAX_SB_SQUARE * 2704 sizeof(*cpi->td.mb.left_pred_buf))); 2705 2706 CHECK_MEM_ERROR(cm, cpi->td.mb.wsrc_buf, 2707 (int32_t *)aom_memalign( 2708 16, MAX_SB_SQUARE * sizeof(*cpi->td.mb.wsrc_buf))); 2709 2710 CHECK_MEM_ERROR( 2711 cm, cpi->td.mb.inter_modes_info, 2712 (InterModesInfo *)aom_malloc(sizeof(*cpi->td.mb.inter_modes_info))); 2713 2714 for (int x = 0; x < 2; x++) 2715 for (int y = 0; y < 2; y++) 2716 CHECK_MEM_ERROR( 2717 cm, cpi->td.mb.hash_value_buffer[x][y], 2718 (uint32_t *)aom_malloc(AOM_BUFFER_SIZE_FOR_BLOCK_HASH * 2719 sizeof(*cpi->td.mb.hash_value_buffer[0][0]))); 2720 2721 cpi->td.mb.g_crc_initialized = 0; 2722 2723 CHECK_MEM_ERROR(cm, cpi->td.mb.mask_buf, 2724 (int32_t *)aom_memalign( 2725 16, MAX_SB_SQUARE * sizeof(*cpi->td.mb.mask_buf))); 2726 2727 av1_set_speed_features_framesize_independent(cpi, oxcf->speed); 2728 av1_set_speed_features_framesize_dependent(cpi, oxcf->speed); 2729 2730 for (int frame = 0; frame < MAX_LAG_BUFFERS; ++frame) { 2731 int mi_cols = ALIGN_POWER_OF_TWO(cm->mi_cols, MAX_MIB_SIZE_LOG2); 2732 int mi_rows = ALIGN_POWER_OF_TWO(cm->mi_rows, MAX_MIB_SIZE_LOG2); 2733 2734 CHECK_MEM_ERROR(cm, cpi->tpl_stats[frame].tpl_stats_ptr, 2735 aom_calloc(mi_rows * mi_cols, 2736 sizeof(*cpi->tpl_stats[frame].tpl_stats_ptr))); 2737 cpi->tpl_stats[frame].is_valid = 0; 2738 cpi->tpl_stats[frame].width = mi_cols; 2739 cpi->tpl_stats[frame].height = mi_rows; 2740 cpi->tpl_stats[frame].stride = mi_cols; 2741 cpi->tpl_stats[frame].mi_rows = cm->mi_rows; 2742 cpi->tpl_stats[frame].mi_cols = cm->mi_cols; 2743 } 2744 2745 #if CONFIG_COLLECT_PARTITION_STATS == 2 2746 av1_zero(cpi->partition_stats); 2747 #endif 2748 2749 #define BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX4DF, JSDAF, JSVAF) \ 2750 cpi->fn_ptr[BT].sdf = SDF; \ 2751 cpi->fn_ptr[BT].sdaf = SDAF; \ 2752 cpi->fn_ptr[BT].vf = VF; \ 2753 cpi->fn_ptr[BT].svf = SVF; \ 2754 cpi->fn_ptr[BT].svaf = SVAF; \ 2755 cpi->fn_ptr[BT].sdx4df = SDX4DF; \ 2756 cpi->fn_ptr[BT].jsdaf = JSDAF; \ 2757 cpi->fn_ptr[BT].jsvaf = JSVAF; 2758 2759 BFP(BLOCK_4X16, aom_sad4x16, aom_sad4x16_avg, aom_variance4x16, 2760 aom_sub_pixel_variance4x16, aom_sub_pixel_avg_variance4x16, 2761 aom_sad4x16x4d, aom_dist_wtd_sad4x16_avg, 2762 aom_dist_wtd_sub_pixel_avg_variance4x16) 2763 2764 BFP(BLOCK_16X4, aom_sad16x4, aom_sad16x4_avg, aom_variance16x4, 2765 aom_sub_pixel_variance16x4, aom_sub_pixel_avg_variance16x4, 2766 aom_sad16x4x4d, aom_dist_wtd_sad16x4_avg, 2767 aom_dist_wtd_sub_pixel_avg_variance16x4) 2768 2769 BFP(BLOCK_8X32, aom_sad8x32, aom_sad8x32_avg, aom_variance8x32, 2770 aom_sub_pixel_variance8x32, aom_sub_pixel_avg_variance8x32, 2771 aom_sad8x32x4d, aom_dist_wtd_sad8x32_avg, 2772 aom_dist_wtd_sub_pixel_avg_variance8x32) 2773 2774 BFP(BLOCK_32X8, aom_sad32x8, aom_sad32x8_avg, aom_variance32x8, 2775 aom_sub_pixel_variance32x8, aom_sub_pixel_avg_variance32x8, 2776 aom_sad32x8x4d, aom_dist_wtd_sad32x8_avg, 2777 aom_dist_wtd_sub_pixel_avg_variance32x8) 2778 2779 BFP(BLOCK_16X64, aom_sad16x64, aom_sad16x64_avg, aom_variance16x64, 2780 aom_sub_pixel_variance16x64, aom_sub_pixel_avg_variance16x64, 2781 aom_sad16x64x4d, aom_dist_wtd_sad16x64_avg, 2782 aom_dist_wtd_sub_pixel_avg_variance16x64) 2783 2784 BFP(BLOCK_64X16, aom_sad64x16, aom_sad64x16_avg, aom_variance64x16, 2785 aom_sub_pixel_variance64x16, aom_sub_pixel_avg_variance64x16, 2786 aom_sad64x16x4d, aom_dist_wtd_sad64x16_avg, 2787 aom_dist_wtd_sub_pixel_avg_variance64x16) 2788 2789 BFP(BLOCK_128X128, aom_sad128x128, aom_sad128x128_avg, aom_variance128x128, 2790 aom_sub_pixel_variance128x128, aom_sub_pixel_avg_variance128x128, 2791 aom_sad128x128x4d, aom_dist_wtd_sad128x128_avg, 2792 aom_dist_wtd_sub_pixel_avg_variance128x128) 2793 2794 BFP(BLOCK_128X64, aom_sad128x64, aom_sad128x64_avg, aom_variance128x64, 2795 aom_sub_pixel_variance128x64, aom_sub_pixel_avg_variance128x64, 2796 aom_sad128x64x4d, aom_dist_wtd_sad128x64_avg, 2797 aom_dist_wtd_sub_pixel_avg_variance128x64) 2798 2799 BFP(BLOCK_64X128, aom_sad64x128, aom_sad64x128_avg, aom_variance64x128, 2800 aom_sub_pixel_variance64x128, aom_sub_pixel_avg_variance64x128, 2801 aom_sad64x128x4d, aom_dist_wtd_sad64x128_avg, 2802 aom_dist_wtd_sub_pixel_avg_variance64x128) 2803 2804 BFP(BLOCK_32X16, aom_sad32x16, aom_sad32x16_avg, aom_variance32x16, 2805 aom_sub_pixel_variance32x16, aom_sub_pixel_avg_variance32x16, 2806 aom_sad32x16x4d, aom_dist_wtd_sad32x16_avg, 2807 aom_dist_wtd_sub_pixel_avg_variance32x16) 2808 2809 BFP(BLOCK_16X32, aom_sad16x32, aom_sad16x32_avg, aom_variance16x32, 2810 aom_sub_pixel_variance16x32, aom_sub_pixel_avg_variance16x32, 2811 aom_sad16x32x4d, aom_dist_wtd_sad16x32_avg, 2812 aom_dist_wtd_sub_pixel_avg_variance16x32) 2813 2814 BFP(BLOCK_64X32, aom_sad64x32, aom_sad64x32_avg, aom_variance64x32, 2815 aom_sub_pixel_variance64x32, aom_sub_pixel_avg_variance64x32, 2816 aom_sad64x32x4d, aom_dist_wtd_sad64x32_avg, 2817 aom_dist_wtd_sub_pixel_avg_variance64x32) 2818 2819 BFP(BLOCK_32X64, aom_sad32x64, aom_sad32x64_avg, aom_variance32x64, 2820 aom_sub_pixel_variance32x64, aom_sub_pixel_avg_variance32x64, 2821 aom_sad32x64x4d, aom_dist_wtd_sad32x64_avg, 2822 aom_dist_wtd_sub_pixel_avg_variance32x64) 2823 2824 BFP(BLOCK_32X32, aom_sad32x32, aom_sad32x32_avg, aom_variance32x32, 2825 aom_sub_pixel_variance32x32, aom_sub_pixel_avg_variance32x32, 2826 aom_sad32x32x4d, aom_dist_wtd_sad32x32_avg, 2827 aom_dist_wtd_sub_pixel_avg_variance32x32) 2828 2829 BFP(BLOCK_64X64, aom_sad64x64, aom_sad64x64_avg, aom_variance64x64, 2830 aom_sub_pixel_variance64x64, aom_sub_pixel_avg_variance64x64, 2831 aom_sad64x64x4d, aom_dist_wtd_sad64x64_avg, 2832 aom_dist_wtd_sub_pixel_avg_variance64x64) 2833 2834 BFP(BLOCK_16X16, aom_sad16x16, aom_sad16x16_avg, aom_variance16x16, 2835 aom_sub_pixel_variance16x16, aom_sub_pixel_avg_variance16x16, 2836 aom_sad16x16x4d, aom_dist_wtd_sad16x16_avg, 2837 aom_dist_wtd_sub_pixel_avg_variance16x16) 2838 2839 BFP(BLOCK_16X8, aom_sad16x8, aom_sad16x8_avg, aom_variance16x8, 2840 aom_sub_pixel_variance16x8, aom_sub_pixel_avg_variance16x8, 2841 aom_sad16x8x4d, aom_dist_wtd_sad16x8_avg, 2842 aom_dist_wtd_sub_pixel_avg_variance16x8) 2843 2844 BFP(BLOCK_8X16, aom_sad8x16, aom_sad8x16_avg, aom_variance8x16, 2845 aom_sub_pixel_variance8x16, aom_sub_pixel_avg_variance8x16, 2846 aom_sad8x16x4d, aom_dist_wtd_sad8x16_avg, 2847 aom_dist_wtd_sub_pixel_avg_variance8x16) 2848 2849 BFP(BLOCK_8X8, aom_sad8x8, aom_sad8x8_avg, aom_variance8x8, 2850 aom_sub_pixel_variance8x8, aom_sub_pixel_avg_variance8x8, aom_sad8x8x4d, 2851 aom_dist_wtd_sad8x8_avg, aom_dist_wtd_sub_pixel_avg_variance8x8) 2852 2853 BFP(BLOCK_8X4, aom_sad8x4, aom_sad8x4_avg, aom_variance8x4, 2854 aom_sub_pixel_variance8x4, aom_sub_pixel_avg_variance8x4, aom_sad8x4x4d, 2855 aom_dist_wtd_sad8x4_avg, aom_dist_wtd_sub_pixel_avg_variance8x4) 2856 2857 BFP(BLOCK_4X8, aom_sad4x8, aom_sad4x8_avg, aom_variance4x8, 2858 aom_sub_pixel_variance4x8, aom_sub_pixel_avg_variance4x8, aom_sad4x8x4d, 2859 aom_dist_wtd_sad4x8_avg, aom_dist_wtd_sub_pixel_avg_variance4x8) 2860 2861 BFP(BLOCK_4X4, aom_sad4x4, aom_sad4x4_avg, aom_variance4x4, 2862 aom_sub_pixel_variance4x4, aom_sub_pixel_avg_variance4x4, aom_sad4x4x4d, 2863 aom_dist_wtd_sad4x4_avg, aom_dist_wtd_sub_pixel_avg_variance4x4) 2864 2865 #define OBFP(BT, OSDF, OVF, OSVF) \ 2866 cpi->fn_ptr[BT].osdf = OSDF; \ 2867 cpi->fn_ptr[BT].ovf = OVF; \ 2868 cpi->fn_ptr[BT].osvf = OSVF; 2869 2870 OBFP(BLOCK_128X128, aom_obmc_sad128x128, aom_obmc_variance128x128, 2871 aom_obmc_sub_pixel_variance128x128) 2872 OBFP(BLOCK_128X64, aom_obmc_sad128x64, aom_obmc_variance128x64, 2873 aom_obmc_sub_pixel_variance128x64) 2874 OBFP(BLOCK_64X128, aom_obmc_sad64x128, aom_obmc_variance64x128, 2875 aom_obmc_sub_pixel_variance64x128) 2876 OBFP(BLOCK_64X64, aom_obmc_sad64x64, aom_obmc_variance64x64, 2877 aom_obmc_sub_pixel_variance64x64) 2878 OBFP(BLOCK_64X32, aom_obmc_sad64x32, aom_obmc_variance64x32, 2879 aom_obmc_sub_pixel_variance64x32) 2880 OBFP(BLOCK_32X64, aom_obmc_sad32x64, aom_obmc_variance32x64, 2881 aom_obmc_sub_pixel_variance32x64) 2882 OBFP(BLOCK_32X32, aom_obmc_sad32x32, aom_obmc_variance32x32, 2883 aom_obmc_sub_pixel_variance32x32) 2884 OBFP(BLOCK_32X16, aom_obmc_sad32x16, aom_obmc_variance32x16, 2885 aom_obmc_sub_pixel_variance32x16) 2886 OBFP(BLOCK_16X32, aom_obmc_sad16x32, aom_obmc_variance16x32, 2887 aom_obmc_sub_pixel_variance16x32) 2888 OBFP(BLOCK_16X16, aom_obmc_sad16x16, aom_obmc_variance16x16, 2889 aom_obmc_sub_pixel_variance16x16) 2890 OBFP(BLOCK_16X8, aom_obmc_sad16x8, aom_obmc_variance16x8, 2891 aom_obmc_sub_pixel_variance16x8) 2892 OBFP(BLOCK_8X16, aom_obmc_sad8x16, aom_obmc_variance8x16, 2893 aom_obmc_sub_pixel_variance8x16) 2894 OBFP(BLOCK_8X8, aom_obmc_sad8x8, aom_obmc_variance8x8, 2895 aom_obmc_sub_pixel_variance8x8) 2896 OBFP(BLOCK_4X8, aom_obmc_sad4x8, aom_obmc_variance4x8, 2897 aom_obmc_sub_pixel_variance4x8) 2898 OBFP(BLOCK_8X4, aom_obmc_sad8x4, aom_obmc_variance8x4, 2899 aom_obmc_sub_pixel_variance8x4) 2900 OBFP(BLOCK_4X4, aom_obmc_sad4x4, aom_obmc_variance4x4, 2901 aom_obmc_sub_pixel_variance4x4) 2902 OBFP(BLOCK_4X16, aom_obmc_sad4x16, aom_obmc_variance4x16, 2903 aom_obmc_sub_pixel_variance4x16) 2904 OBFP(BLOCK_16X4, aom_obmc_sad16x4, aom_obmc_variance16x4, 2905 aom_obmc_sub_pixel_variance16x4) 2906 OBFP(BLOCK_8X32, aom_obmc_sad8x32, aom_obmc_variance8x32, 2907 aom_obmc_sub_pixel_variance8x32) 2908 OBFP(BLOCK_32X8, aom_obmc_sad32x8, aom_obmc_variance32x8, 2909 aom_obmc_sub_pixel_variance32x8) 2910 OBFP(BLOCK_16X64, aom_obmc_sad16x64, aom_obmc_variance16x64, 2911 aom_obmc_sub_pixel_variance16x64) 2912 OBFP(BLOCK_64X16, aom_obmc_sad64x16, aom_obmc_variance64x16, 2913 aom_obmc_sub_pixel_variance64x16) 2914 2915 #define MBFP(BT, MCSDF, MCSVF) \ 2916 cpi->fn_ptr[BT].msdf = MCSDF; \ 2917 cpi->fn_ptr[BT].msvf = MCSVF; 2918 2919 MBFP(BLOCK_128X128, aom_masked_sad128x128, 2920 aom_masked_sub_pixel_variance128x128) 2921 MBFP(BLOCK_128X64, aom_masked_sad128x64, aom_masked_sub_pixel_variance128x64) 2922 MBFP(BLOCK_64X128, aom_masked_sad64x128, aom_masked_sub_pixel_variance64x128) 2923 MBFP(BLOCK_64X64, aom_masked_sad64x64, aom_masked_sub_pixel_variance64x64) 2924 MBFP(BLOCK_64X32, aom_masked_sad64x32, aom_masked_sub_pixel_variance64x32) 2925 MBFP(BLOCK_32X64, aom_masked_sad32x64, aom_masked_sub_pixel_variance32x64) 2926 MBFP(BLOCK_32X32, aom_masked_sad32x32, aom_masked_sub_pixel_variance32x32) 2927 MBFP(BLOCK_32X16, aom_masked_sad32x16, aom_masked_sub_pixel_variance32x16) 2928 MBFP(BLOCK_16X32, aom_masked_sad16x32, aom_masked_sub_pixel_variance16x32) 2929 MBFP(BLOCK_16X16, aom_masked_sad16x16, aom_masked_sub_pixel_variance16x16) 2930 MBFP(BLOCK_16X8, aom_masked_sad16x8, aom_masked_sub_pixel_variance16x8) 2931 MBFP(BLOCK_8X16, aom_masked_sad8x16, aom_masked_sub_pixel_variance8x16) 2932 MBFP(BLOCK_8X8, aom_masked_sad8x8, aom_masked_sub_pixel_variance8x8) 2933 MBFP(BLOCK_4X8, aom_masked_sad4x8, aom_masked_sub_pixel_variance4x8) 2934 MBFP(BLOCK_8X4, aom_masked_sad8x4, aom_masked_sub_pixel_variance8x4) 2935 MBFP(BLOCK_4X4, aom_masked_sad4x4, aom_masked_sub_pixel_variance4x4) 2936 2937 MBFP(BLOCK_4X16, aom_masked_sad4x16, aom_masked_sub_pixel_variance4x16) 2938 2939 MBFP(BLOCK_16X4, aom_masked_sad16x4, aom_masked_sub_pixel_variance16x4) 2940 2941 MBFP(BLOCK_8X32, aom_masked_sad8x32, aom_masked_sub_pixel_variance8x32) 2942 2943 MBFP(BLOCK_32X8, aom_masked_sad32x8, aom_masked_sub_pixel_variance32x8) 2944 2945 MBFP(BLOCK_16X64, aom_masked_sad16x64, aom_masked_sub_pixel_variance16x64) 2946 2947 MBFP(BLOCK_64X16, aom_masked_sad64x16, aom_masked_sub_pixel_variance64x16) 2948 2949 highbd_set_var_fns(cpi); 2950 2951 /* av1_init_quantizer() is first called here. Add check in 2952 * av1_frame_init_quantizer() so that av1_init_quantizer is only 2953 * called later when needed. This will avoid unnecessary calls of 2954 * av1_init_quantizer() for every frame. 2955 */ 2956 av1_init_quantizer(cpi); 2957 av1_qm_init(cm); 2958 2959 av1_loop_filter_init(cm); 2960 cm->superres_scale_denominator = SCALE_NUMERATOR; 2961 cm->superres_upscaled_width = oxcf->width; 2962 cm->superres_upscaled_height = oxcf->height; 2963 av1_loop_restoration_precal(); 2964 2965 cm->error.setjmp = 0; 2966 2967 return cpi; 2968 } 2969 2970 #if CONFIG_INTERNAL_STATS 2971 #define SNPRINT(H, T) snprintf((H) + strlen(H), sizeof(H) - strlen(H), (T)) 2972 2973 #define SNPRINT2(H, T, V) \ 2974 snprintf((H) + strlen(H), sizeof(H) - strlen(H), (T), (V)) 2975 #endif // CONFIG_INTERNAL_STATS 2976 2977 void av1_remove_compressor(AV1_COMP *cpi) { 2978 AV1_COMMON *cm; 2979 unsigned int i; 2980 int t; 2981 2982 if (!cpi) return; 2983 2984 cm = &cpi->common; 2985 const int num_planes = av1_num_planes(cm); 2986 2987 if (cm->current_frame.frame_number > 0) { 2988 #if CONFIG_ENTROPY_STATS 2989 if (cpi->oxcf.pass != 1) { 2990 fprintf(stderr, "Writing counts.stt\n"); 2991 FILE *f = fopen("counts.stt", "wb"); 2992 fwrite(&aggregate_fc, sizeof(aggregate_fc), 1, f); 2993 fclose(f); 2994 } 2995 #endif // CONFIG_ENTROPY_STATS 2996 #if CONFIG_INTERNAL_STATS 2997 aom_clear_system_state(); 2998 2999 if (cpi->oxcf.pass != 1) { 3000 char headings[512] = { 0 }; 3001 char results[512] = { 0 }; 3002 FILE *f = fopen("opsnr.stt", "a"); 3003 double time_encoded = 3004 (cpi->last_end_time_stamp_seen - cpi->first_time_stamp_ever) / 3005 10000000.000; 3006 double total_encode_time = 3007 (cpi->time_receive_data + cpi->time_compress_data) / 1000.000; 3008 const double dr = 3009 (double)cpi->bytes * (double)8 / (double)1000 / time_encoded; 3010 const double peak = (double)((1 << cpi->oxcf.input_bit_depth) - 1); 3011 const double target_rate = (double)cpi->oxcf.target_bandwidth / 1000; 3012 const double rate_err = ((100.0 * (dr - target_rate)) / target_rate); 3013 3014 if (cpi->b_calculate_psnr) { 3015 const double total_psnr = aom_sse_to_psnr( 3016 (double)cpi->total_samples, peak, (double)cpi->total_sq_error); 3017 const double total_ssim = 3018 100 * pow(cpi->summed_quality / cpi->summed_weights, 8.0); 3019 snprintf(headings, sizeof(headings), 3020 "Bitrate\tAVGPsnr\tGLBPsnr\tAVPsnrP\tGLPsnrP\t" 3021 "AOMSSIM\tVPSSIMP\tFASTSIM\tPSNRHVS\t" 3022 "WstPsnr\tWstSsim\tWstFast\tWstHVS\t" 3023 "AVPsrnY\tAPsnrCb\tAPsnrCr"); 3024 snprintf(results, sizeof(results), 3025 "%7.2f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t" 3026 "%7.3f\t%7.3f\t%7.3f\t%7.3f\t" 3027 "%7.3f\t%7.3f\t%7.3f\t%7.3f\t" 3028 "%7.3f\t%7.3f\t%7.3f", 3029 dr, cpi->psnr.stat[STAT_ALL] / cpi->count, total_psnr, 3030 cpi->psnr.stat[STAT_ALL] / cpi->count, total_psnr, total_ssim, 3031 total_ssim, cpi->fastssim.stat[STAT_ALL] / cpi->count, 3032 cpi->psnrhvs.stat[STAT_ALL] / cpi->count, cpi->psnr.worst, 3033 cpi->worst_ssim, cpi->fastssim.worst, cpi->psnrhvs.worst, 3034 cpi->psnr.stat[STAT_Y] / cpi->count, 3035 cpi->psnr.stat[STAT_U] / cpi->count, 3036 cpi->psnr.stat[STAT_V] / cpi->count); 3037 3038 if (cpi->b_calculate_blockiness) { 3039 SNPRINT(headings, "\t Block\tWstBlck"); 3040 SNPRINT2(results, "\t%7.3f", cpi->total_blockiness / cpi->count); 3041 SNPRINT2(results, "\t%7.3f", cpi->worst_blockiness); 3042 } 3043 3044 if (cpi->b_calculate_consistency) { 3045 double consistency = 3046 aom_sse_to_psnr((double)cpi->total_samples, peak, 3047 (double)cpi->total_inconsistency); 3048 3049 SNPRINT(headings, "\tConsist\tWstCons"); 3050 SNPRINT2(results, "\t%7.3f", consistency); 3051 SNPRINT2(results, "\t%7.3f", cpi->worst_consistency); 3052 } 3053 fprintf(f, "%s\t Time\tRcErr\tAbsErr\n", headings); 3054 fprintf(f, "%s\t%8.0f\t%7.2f\t%7.2f\n", results, total_encode_time, 3055 rate_err, fabs(rate_err)); 3056 } 3057 3058 fclose(f); 3059 } 3060 #endif // CONFIG_INTERNAL_STATS 3061 #if CONFIG_SPEED_STATS 3062 if (cpi->oxcf.pass != 1) { 3063 fprintf(stdout, "tx_search_count = %d\n", cpi->tx_search_count); 3064 } 3065 #endif // CONFIG_SPEED_STATS 3066 3067 #if CONFIG_COLLECT_PARTITION_STATS == 2 3068 if (cpi->oxcf.pass != 1) { 3069 av1_print_partition_stats(&cpi->partition_stats); 3070 } 3071 #endif 3072 } 3073 3074 for (int frame = 0; frame < MAX_LAG_BUFFERS; ++frame) { 3075 aom_free(cpi->tpl_stats[frame].tpl_stats_ptr); 3076 cpi->tpl_stats[frame].is_valid = 0; 3077 } 3078 3079 for (t = cpi->num_workers - 1; t >= 0; --t) { 3080 AVxWorker *const worker = &cpi->workers[t]; 3081 EncWorkerData *const thread_data = &cpi->tile_thr_data[t]; 3082 3083 // Deallocate allocated threads. 3084 aom_get_worker_interface()->end(worker); 3085 3086 // Deallocate allocated thread data. 3087 if (cpi->row_mt == 1) aom_free(thread_data->td->tctx); 3088 if (t > 0) { 3089 aom_free(thread_data->td->palette_buffer); 3090 aom_free(thread_data->td->tmp_conv_dst); 3091 for (int j = 0; j < 2; ++j) { 3092 aom_free(thread_data->td->tmp_obmc_bufs[j]); 3093 } 3094 aom_free(thread_data->td->above_pred_buf); 3095 aom_free(thread_data->td->left_pred_buf); 3096 aom_free(thread_data->td->wsrc_buf); 3097 3098 aom_free(thread_data->td->inter_modes_info); 3099 for (int x = 0; x < 2; x++) { 3100 for (int y = 0; y < 2; y++) { 3101 aom_free(thread_data->td->hash_value_buffer[x][y]); 3102 thread_data->td->hash_value_buffer[x][y] = NULL; 3103 } 3104 } 3105 aom_free(thread_data->td->mask_buf); 3106 aom_free(thread_data->td->counts); 3107 av1_free_pc_tree(thread_data->td, num_planes); 3108 aom_free(thread_data->td); 3109 } 3110 } 3111 #if CONFIG_MULTITHREAD 3112 if (cpi->row_mt == 1) { 3113 if (cpi->row_mt_mutex_ != NULL) { 3114 pthread_mutex_destroy(cpi->row_mt_mutex_); 3115 aom_free(cpi->row_mt_mutex_); 3116 } 3117 } 3118 #endif 3119 av1_row_mt_mem_dealloc(cpi); 3120 aom_free(cpi->tile_thr_data); 3121 aom_free(cpi->workers); 3122 3123 if (cpi->num_workers > 1) { 3124 av1_loop_filter_dealloc(&cpi->lf_row_sync); 3125 av1_loop_restoration_dealloc(&cpi->lr_row_sync, cpi->num_workers); 3126 } 3127 3128 dealloc_compressor_data(cpi); 3129 3130 for (i = 0; i < sizeof(cpi->mbgraph_stats) / sizeof(cpi->mbgraph_stats[0]); 3131 ++i) { 3132 aom_free(cpi->mbgraph_stats[i].mb_stats); 3133 } 3134 3135 #if CONFIG_INTERNAL_STATS 3136 aom_free(cpi->ssim_vars); 3137 cpi->ssim_vars = NULL; 3138 #endif // CONFIG_INTERNAL_STATS 3139 3140 av1_remove_common(cm); 3141 for (i = 0; i < FRAME_BUFFERS; ++i) { 3142 av1_hash_table_destroy(&cm->buffer_pool->frame_bufs[i].hash_table); 3143 } 3144 if (cpi->sf.use_hash_based_trellis) hbt_destroy(); 3145 av1_free_ref_frame_buffers(cm->buffer_pool); 3146 aom_free(cpi); 3147 3148 #ifdef OUTPUT_YUV_SKINMAP 3149 fclose(yuv_skinmap_file); 3150 #endif 3151 #ifdef OUTPUT_YUV_REC 3152 fclose(yuv_rec_file); 3153 #endif 3154 } 3155 3156 static void generate_psnr_packet(AV1_COMP *cpi) { 3157 struct aom_codec_cx_pkt pkt; 3158 int i; 3159 PSNR_STATS psnr; 3160 aom_calc_highbd_psnr(cpi->source, &cpi->common.cur_frame->buf, &psnr, 3161 cpi->td.mb.e_mbd.bd, cpi->oxcf.input_bit_depth); 3162 3163 for (i = 0; i < 4; ++i) { 3164 pkt.data.psnr.samples[i] = psnr.samples[i]; 3165 pkt.data.psnr.sse[i] = psnr.sse[i]; 3166 pkt.data.psnr.psnr[i] = psnr.psnr[i]; 3167 } 3168 pkt.kind = AOM_CODEC_PSNR_PKT; 3169 aom_codec_pkt_list_add(cpi->output_pkt_list, &pkt); 3170 } 3171 3172 int av1_use_as_reference(AV1_COMP *cpi, int ref_frame_flags) { 3173 if (ref_frame_flags > ((1 << INTER_REFS_PER_FRAME) - 1)) return -1; 3174 3175 cpi->ext_ref_frame_flags = ref_frame_flags; 3176 return 0; 3177 } 3178 3179 int av1_copy_reference_enc(AV1_COMP *cpi, int idx, YV12_BUFFER_CONFIG *sd) { 3180 AV1_COMMON *const cm = &cpi->common; 3181 const int num_planes = av1_num_planes(cm); 3182 YV12_BUFFER_CONFIG *cfg = get_ref_frame(cm, idx); 3183 if (cfg) { 3184 aom_yv12_copy_frame(cfg, sd, num_planes); 3185 return 0; 3186 } else { 3187 return -1; 3188 } 3189 } 3190 3191 int av1_set_reference_enc(AV1_COMP *cpi, int idx, YV12_BUFFER_CONFIG *sd) { 3192 AV1_COMMON *const cm = &cpi->common; 3193 const int num_planes = av1_num_planes(cm); 3194 YV12_BUFFER_CONFIG *cfg = get_ref_frame(cm, idx); 3195 if (cfg) { 3196 aom_yv12_copy_frame(sd, cfg, num_planes); 3197 return 0; 3198 } else { 3199 return -1; 3200 } 3201 } 3202 3203 int av1_update_entropy(AV1_COMP *cpi, int update) { 3204 cpi->ext_refresh_frame_context = update; 3205 cpi->ext_refresh_frame_context_pending = 1; 3206 return 0; 3207 } 3208 3209 #if defined(OUTPUT_YUV_DENOISED) || defined(OUTPUT_YUV_SKINMAP) 3210 // The denoiser buffer is allocated as a YUV 440 buffer. This function writes it 3211 // as YUV 420. We simply use the top-left pixels of the UV buffers, since we do 3212 // not denoise the UV channels at this time. If ever we implement UV channel 3213 // denoising we will have to modify this. 3214 void aom_write_yuv_frame_420(YV12_BUFFER_CONFIG *s, FILE *f) { 3215 uint8_t *src = s->y_buffer; 3216 int h = s->y_height; 3217 3218 do { 3219 fwrite(src, s->y_width, 1, f); 3220 src += s->y_stride; 3221 } while (--h); 3222 3223 src = s->u_buffer; 3224 h = s->uv_height; 3225 3226 do { 3227 fwrite(src, s->uv_width, 1, f); 3228 src += s->uv_stride; 3229 } while (--h); 3230 3231 src = s->v_buffer; 3232 h = s->uv_height; 3233 3234 do { 3235 fwrite(src, s->uv_width, 1, f); 3236 src += s->uv_stride; 3237 } while (--h); 3238 } 3239 #endif 3240 3241 #ifdef OUTPUT_YUV_REC 3242 void aom_write_one_yuv_frame(AV1_COMMON *cm, YV12_BUFFER_CONFIG *s) { 3243 uint8_t *src = s->y_buffer; 3244 int h = cm->height; 3245 if (yuv_rec_file == NULL) return; 3246 if (s->flags & YV12_FLAG_HIGHBITDEPTH) { 3247 uint16_t *src16 = CONVERT_TO_SHORTPTR(s->y_buffer); 3248 3249 do { 3250 fwrite(src16, s->y_width, 2, yuv_rec_file); 3251 src16 += s->y_stride; 3252 } while (--h); 3253 3254 src16 = CONVERT_TO_SHORTPTR(s->u_buffer); 3255 h = s->uv_height; 3256 3257 do { 3258 fwrite(src16, s->uv_width, 2, yuv_rec_file); 3259 src16 += s->uv_stride; 3260 } while (--h); 3261 3262 src16 = CONVERT_TO_SHORTPTR(s->v_buffer); 3263 h = s->uv_height; 3264 3265 do { 3266 fwrite(src16, s->uv_width, 2, yuv_rec_file); 3267 src16 += s->uv_stride; 3268 } while (--h); 3269 3270 fflush(yuv_rec_file); 3271 return; 3272 } 3273 3274 do { 3275 fwrite(src, s->y_width, 1, yuv_rec_file); 3276 src += s->y_stride; 3277 } while (--h); 3278 3279 src = s->u_buffer; 3280 h = s->uv_height; 3281 3282 do { 3283 fwrite(src, s->uv_width, 1, yuv_rec_file); 3284 src += s->uv_stride; 3285 } while (--h); 3286 3287 src = s->v_buffer; 3288 h = s->uv_height; 3289 3290 do { 3291 fwrite(src, s->uv_width, 1, yuv_rec_file); 3292 src += s->uv_stride; 3293 } while (--h); 3294 3295 fflush(yuv_rec_file); 3296 } 3297 #endif // OUTPUT_YUV_REC 3298 3299 #define GM_RECODE_LOOP_NUM4X4_FACTOR 192 3300 static int recode_loop_test_global_motion(AV1_COMP *cpi) { 3301 int i; 3302 int recode = 0; 3303 RD_COUNTS *const rdc = &cpi->td.rd_counts; 3304 AV1_COMMON *const cm = &cpi->common; 3305 for (i = LAST_FRAME; i <= ALTREF_FRAME; ++i) { 3306 if (cm->global_motion[i].wmtype != IDENTITY && 3307 rdc->global_motion_used[i] * GM_RECODE_LOOP_NUM4X4_FACTOR < 3308 cpi->gmparams_cost[i]) { 3309 cm->global_motion[i] = default_warp_params; 3310 assert(cm->global_motion[i].wmtype == IDENTITY); 3311 cpi->gmparams_cost[i] = 0; 3312 recode = 1; 3313 // TODO(sarahparker): The earlier condition for recoding here was: 3314 // "recode |= (rdc->global_motion_used[i] > 0);". Can we bring something 3315 // similar to that back to speed up global motion? 3316 } 3317 } 3318 return recode; 3319 } 3320 3321 // Function to test for conditions that indicate we should loop 3322 // back and recode a frame. 3323 static int recode_loop_test(AV1_COMP *cpi, int high_limit, int low_limit, int q, 3324 int maxq, int minq) { 3325 const RATE_CONTROL *const rc = &cpi->rc; 3326 const AV1EncoderConfig *const oxcf = &cpi->oxcf; 3327 const int frame_is_kfgfarf = frame_is_kf_gf_arf(cpi); 3328 int force_recode = 0; 3329 3330 if ((rc->projected_frame_size >= rc->max_frame_bandwidth) || 3331 (cpi->sf.recode_loop == ALLOW_RECODE) || 3332 (frame_is_kfgfarf && (cpi->sf.recode_loop == ALLOW_RECODE_KFARFGF))) { 3333 // TODO(agrange) high_limit could be greater than the scale-down threshold. 3334 if ((rc->projected_frame_size > high_limit && q < maxq) || 3335 (rc->projected_frame_size < low_limit && q > minq)) { 3336 force_recode = 1; 3337 } else if (cpi->oxcf.rc_mode == AOM_CQ) { 3338 // Deal with frame undershoot and whether or not we are 3339 // below the automatically set cq level. 3340 if (q > oxcf->cq_level && 3341 rc->projected_frame_size < ((rc->this_frame_target * 7) >> 3)) { 3342 force_recode = 1; 3343 } 3344 } 3345 } 3346 return force_recode; 3347 } 3348 3349 static void scale_references(AV1_COMP *cpi) { 3350 AV1_COMMON *cm = &cpi->common; 3351 const int num_planes = av1_num_planes(cm); 3352 MV_REFERENCE_FRAME ref_frame; 3353 const AOM_REFFRAME ref_mask[INTER_REFS_PER_FRAME] = { 3354 AOM_LAST_FLAG, AOM_LAST2_FLAG, AOM_LAST3_FLAG, AOM_GOLD_FLAG, 3355 AOM_BWD_FLAG, AOM_ALT2_FLAG, AOM_ALT_FLAG 3356 }; 3357 3358 for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) { 3359 // Need to convert from AOM_REFFRAME to index into ref_mask (subtract 1). 3360 if (cpi->ref_frame_flags & ref_mask[ref_frame - 1]) { 3361 BufferPool *const pool = cm->buffer_pool; 3362 const YV12_BUFFER_CONFIG *const ref = 3363 get_ref_frame_yv12_buf(cm, ref_frame); 3364 3365 if (ref == NULL) { 3366 cpi->scaled_ref_buf[ref_frame - 1] = NULL; 3367 continue; 3368 } 3369 3370 if (ref->y_crop_width != cm->width || ref->y_crop_height != cm->height) { 3371 // Replace the reference buffer with a copy having a thicker border, 3372 // if the reference buffer is higher resolution than the current 3373 // frame, and the border is thin. 3374 if ((ref->y_crop_width > cm->width || 3375 ref->y_crop_height > cm->height) && 3376 ref->border < AOM_BORDER_IN_PIXELS) { 3377 RefCntBuffer *ref_fb = get_ref_frame_buf(cm, ref_frame); 3378 if (aom_yv12_realloc_with_new_border( 3379 &ref_fb->buf, AOM_BORDER_IN_PIXELS, cm->byte_alignment, 3380 num_planes) != 0) { 3381 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, 3382 "Failed to allocate frame buffer"); 3383 } 3384 } 3385 int force_scaling = 0; 3386 RefCntBuffer *new_fb = cpi->scaled_ref_buf[ref_frame - 1]; 3387 if (new_fb == NULL) { 3388 const int new_fb_idx = get_free_fb(cm); 3389 if (new_fb_idx == INVALID_IDX) { 3390 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, 3391 "Unable to find free frame buffer"); 3392 } 3393 force_scaling = 1; 3394 new_fb = &pool->frame_bufs[new_fb_idx]; 3395 } 3396 3397 if (force_scaling || new_fb->buf.y_crop_width != cm->width || 3398 new_fb->buf.y_crop_height != cm->height) { 3399 if (aom_realloc_frame_buffer( 3400 &new_fb->buf, cm->width, cm->height, 3401 cm->seq_params.subsampling_x, cm->seq_params.subsampling_y, 3402 cm->seq_params.use_highbitdepth, AOM_BORDER_IN_PIXELS, 3403 cm->byte_alignment, NULL, NULL, NULL)) { 3404 if (force_scaling) { 3405 // Release the reference acquired in the get_free_fb() call above. 3406 --new_fb->ref_count; 3407 } 3408 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, 3409 "Failed to allocate frame buffer"); 3410 } 3411 av1_resize_and_extend_frame( 3412 ref, &new_fb->buf, (int)cm->seq_params.bit_depth, num_planes); 3413 cpi->scaled_ref_buf[ref_frame - 1] = new_fb; 3414 alloc_frame_mvs(cm, new_fb); 3415 } 3416 } else { 3417 RefCntBuffer *buf = get_ref_frame_buf(cm, ref_frame); 3418 buf->buf.y_crop_width = ref->y_crop_width; 3419 buf->buf.y_crop_height = ref->y_crop_height; 3420 cpi->scaled_ref_buf[ref_frame - 1] = buf; 3421 ++buf->ref_count; 3422 } 3423 } else { 3424 if (cpi->oxcf.pass != 0) cpi->scaled_ref_buf[ref_frame - 1] = NULL; 3425 } 3426 } 3427 } 3428 3429 static void release_scaled_references(AV1_COMP *cpi) { 3430 // TODO(isbs): only refresh the necessary frames, rather than all of them 3431 for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) { 3432 RefCntBuffer *const buf = cpi->scaled_ref_buf[i]; 3433 if (buf != NULL) { 3434 --buf->ref_count; 3435 cpi->scaled_ref_buf[i] = NULL; 3436 } 3437 } 3438 } 3439 3440 static void set_mv_search_params(AV1_COMP *cpi) { 3441 const AV1_COMMON *const cm = &cpi->common; 3442 const unsigned int max_mv_def = AOMMIN(cm->width, cm->height); 3443 3444 // Default based on max resolution. 3445 cpi->mv_step_param = av1_init_search_range(max_mv_def); 3446 3447 if (cpi->sf.mv.auto_mv_step_size) { 3448 if (frame_is_intra_only(cm)) { 3449 // Initialize max_mv_magnitude for use in the first INTER frame 3450 // after a key/intra-only frame. 3451 cpi->max_mv_magnitude = max_mv_def; 3452 } else { 3453 if (cm->show_frame) { 3454 // Allow mv_steps to correspond to twice the max mv magnitude found 3455 // in the previous frame, capped by the default max_mv_magnitude based 3456 // on resolution. 3457 cpi->mv_step_param = av1_init_search_range( 3458 AOMMIN(max_mv_def, 2 * cpi->max_mv_magnitude)); 3459 } 3460 cpi->max_mv_magnitude = 0; 3461 } 3462 } 3463 } 3464 3465 static void set_screen_content_options(AV1_COMP *cpi) { 3466 AV1_COMMON *cm = &cpi->common; 3467 3468 if (cm->seq_params.force_screen_content_tools != 2) { 3469 cm->allow_screen_content_tools = cm->allow_intrabc = 3470 cm->seq_params.force_screen_content_tools; 3471 return; 3472 } 3473 3474 if (cpi->oxcf.content == AOM_CONTENT_SCREEN) { 3475 cm->allow_screen_content_tools = cm->allow_intrabc = 1; 3476 return; 3477 } 3478 3479 // Estimate if the source frame is screen content, based on the portion of 3480 // blocks that have few luma colors. 3481 const uint8_t *src = cpi->source->y_buffer; 3482 assert(src != NULL); 3483 const int use_hbd = cpi->source->flags & YV12_FLAG_HIGHBITDEPTH; 3484 const int stride = cpi->source->y_stride; 3485 const int width = cpi->source->y_width; 3486 const int height = cpi->source->y_height; 3487 const int bd = cm->seq_params.bit_depth; 3488 const int blk_w = 16; 3489 const int blk_h = 16; 3490 // These threshold values are selected experimentally. 3491 const int color_thresh = 4; 3492 const unsigned int var_thresh = 0; 3493 // Counts of blocks with no more than color_thresh colors. 3494 int counts_1 = 0; 3495 // Counts of blocks with no more than color_thresh colors and variance larger 3496 // than var_thresh. 3497 int counts_2 = 0; 3498 3499 for (int r = 0; r + blk_h <= height; r += blk_h) { 3500 for (int c = 0; c + blk_w <= width; c += blk_w) { 3501 int count_buf[1 << 12]; // Maximum (1 << 12) color levels. 3502 const uint8_t *const this_src = src + r * stride + c; 3503 const int n_colors = 3504 use_hbd ? av1_count_colors_highbd(this_src, stride, blk_w, blk_h, bd, 3505 count_buf) 3506 : av1_count_colors(this_src, stride, blk_w, blk_h, count_buf); 3507 if (n_colors > 1 && n_colors <= color_thresh) { 3508 ++counts_1; 3509 struct buf_2d buf; 3510 buf.stride = stride; 3511 buf.buf = (uint8_t *)this_src; 3512 const unsigned int var = 3513 use_hbd 3514 ? av1_high_get_sby_perpixel_variance(cpi, &buf, BLOCK_16X16, bd) 3515 : av1_get_sby_perpixel_variance(cpi, &buf, BLOCK_16X16); 3516 if (var > var_thresh) ++counts_2; 3517 } 3518 } 3519 } 3520 3521 // The threshold values are selected experimentally. 3522 cm->allow_screen_content_tools = 3523 counts_1 * blk_h * blk_w * 10 > width * height; 3524 // IntraBC would force loop filters off, so we use more strict rules that also 3525 // requires that the block has high variance. 3526 cm->allow_intrabc = cm->allow_screen_content_tools && 3527 counts_2 * blk_h * blk_w * 15 > width * height; 3528 } 3529 3530 static void set_size_independent_vars(AV1_COMP *cpi) { 3531 int i; 3532 AV1_COMMON *cm = &cpi->common; 3533 for (i = LAST_FRAME; i <= ALTREF_FRAME; ++i) { 3534 cm->global_motion[i] = default_warp_params; 3535 } 3536 cpi->global_motion_search_done = 0; 3537 3538 if (frame_is_intra_only(cm)) set_screen_content_options(cpi); 3539 cpi->is_screen_content_type = (cm->allow_screen_content_tools != 0); 3540 3541 av1_set_speed_features_framesize_independent(cpi, cpi->speed); 3542 av1_set_rd_speed_thresholds(cpi); 3543 cm->interp_filter = SWITCHABLE; 3544 cm->switchable_motion_mode = 1; 3545 } 3546 3547 static void set_size_dependent_vars(AV1_COMP *cpi, int *q, int *bottom_index, 3548 int *top_index) { 3549 AV1_COMMON *const cm = &cpi->common; 3550 const AV1EncoderConfig *const oxcf = &cpi->oxcf; 3551 3552 // Setup variables that depend on the dimensions of the frame. 3553 av1_set_speed_features_framesize_dependent(cpi, cpi->speed); 3554 3555 // Decide q and q bounds. 3556 *q = av1_rc_pick_q_and_bounds(cpi, cm->width, cm->height, bottom_index, 3557 top_index); 3558 3559 if (!frame_is_intra_only(cm)) { 3560 set_high_precision_mv(cpi, (*q) < HIGH_PRECISION_MV_QTHRESH, 3561 cpi->common.cur_frame_force_integer_mv); 3562 } 3563 3564 // Configure experimental use of segmentation for enhanced coding of 3565 // static regions if indicated. 3566 // Only allowed in the second pass of a two pass encode, as it requires 3567 // lagged coding, and if the relevant speed feature flag is set. 3568 if (oxcf->pass == 2 && cpi->sf.static_segmentation) 3569 configure_static_seg_features(cpi); 3570 } 3571 3572 static void init_motion_estimation(AV1_COMP *cpi) { 3573 int y_stride = cpi->scaled_source.y_stride; 3574 int y_stride_src = (cpi->oxcf.resize_mode || cpi->oxcf.superres_mode) 3575 ? y_stride 3576 : cpi->lookahead->buf->img.y_stride; 3577 3578 if (cpi->sf.mv.search_method == NSTEP) { 3579 av1_init3smotion_compensation(&cpi->ss_cfg[SS_CFG_SRC], y_stride); 3580 av1_init3smotion_compensation(&cpi->ss_cfg[SS_CFG_LOOKAHEAD], y_stride_src); 3581 } else if (cpi->sf.mv.search_method == DIAMOND) { 3582 av1_init_dsmotion_compensation(&cpi->ss_cfg[SS_CFG_SRC], y_stride); 3583 av1_init_dsmotion_compensation(&cpi->ss_cfg[SS_CFG_LOOKAHEAD], 3584 y_stride_src); 3585 } 3586 } 3587 3588 #define COUPLED_CHROMA_FROM_LUMA_RESTORATION 0 3589 static void set_restoration_unit_size(int width, int height, int sx, int sy, 3590 RestorationInfo *rst) { 3591 (void)width; 3592 (void)height; 3593 (void)sx; 3594 (void)sy; 3595 #if COUPLED_CHROMA_FROM_LUMA_RESTORATION 3596 int s = AOMMIN(sx, sy); 3597 #else 3598 int s = 0; 3599 #endif // !COUPLED_CHROMA_FROM_LUMA_RESTORATION 3600 3601 if (width * height > 352 * 288) 3602 rst[0].restoration_unit_size = RESTORATION_UNITSIZE_MAX; 3603 else 3604 rst[0].restoration_unit_size = (RESTORATION_UNITSIZE_MAX >> 1); 3605 rst[1].restoration_unit_size = rst[0].restoration_unit_size >> s; 3606 rst[2].restoration_unit_size = rst[1].restoration_unit_size; 3607 } 3608 3609 static void init_ref_frame_bufs(AV1_COMP *cpi) { 3610 AV1_COMMON *const cm = &cpi->common; 3611 int i; 3612 BufferPool *const pool = cm->buffer_pool; 3613 cm->cur_frame = NULL; 3614 for (i = 0; i < REF_FRAMES; ++i) { 3615 cm->ref_frame_map[i] = NULL; 3616 } 3617 for (i = 0; i < FRAME_BUFFERS; ++i) { 3618 pool->frame_bufs[i].ref_count = 0; 3619 } 3620 if (cm->seq_params.force_screen_content_tools) { 3621 for (i = 0; i < FRAME_BUFFERS; ++i) { 3622 av1_hash_table_init(&pool->frame_bufs[i].hash_table, &cpi->td.mb); 3623 } 3624 } 3625 } 3626 3627 static void check_initial_width(AV1_COMP *cpi, int use_highbitdepth, 3628 int subsampling_x, int subsampling_y) { 3629 AV1_COMMON *const cm = &cpi->common; 3630 SequenceHeader *const seq_params = &cm->seq_params; 3631 3632 if (!cpi->initial_width || seq_params->use_highbitdepth != use_highbitdepth || 3633 seq_params->subsampling_x != subsampling_x || 3634 seq_params->subsampling_y != subsampling_y) { 3635 seq_params->subsampling_x = subsampling_x; 3636 seq_params->subsampling_y = subsampling_y; 3637 seq_params->use_highbitdepth = use_highbitdepth; 3638 3639 alloc_raw_frame_buffers(cpi); 3640 init_ref_frame_bufs(cpi); 3641 alloc_util_frame_buffers(cpi); 3642 3643 init_motion_estimation(cpi); // TODO(agrange) This can be removed. 3644 3645 cpi->initial_width = cm->width; 3646 cpi->initial_height = cm->height; 3647 cpi->initial_mbs = cm->MBs; 3648 } 3649 } 3650 3651 // Returns 1 if the assigned width or height was <= 0. 3652 static int set_size_literal(AV1_COMP *cpi, int width, int height) { 3653 AV1_COMMON *cm = &cpi->common; 3654 const int num_planes = av1_num_planes(cm); 3655 check_initial_width(cpi, cm->seq_params.use_highbitdepth, 3656 cm->seq_params.subsampling_x, 3657 cm->seq_params.subsampling_y); 3658 3659 if (width <= 0 || height <= 0) return 1; 3660 3661 cm->width = width; 3662 cm->height = height; 3663 3664 if (cpi->initial_width && cpi->initial_height && 3665 (cm->width > cpi->initial_width || cm->height > cpi->initial_height)) { 3666 av1_free_context_buffers(cm); 3667 av1_free_pc_tree(&cpi->td, num_planes); 3668 alloc_compressor_data(cpi); 3669 realloc_segmentation_maps(cpi); 3670 cpi->initial_width = cpi->initial_height = 0; 3671 } 3672 update_frame_size(cpi); 3673 3674 return 0; 3675 } 3676 3677 void av1_set_frame_size(AV1_COMP *cpi, int width, int height) { 3678 AV1_COMMON *const cm = &cpi->common; 3679 const SequenceHeader *const seq_params = &cm->seq_params; 3680 const int num_planes = av1_num_planes(cm); 3681 MACROBLOCKD *const xd = &cpi->td.mb.e_mbd; 3682 int ref_frame; 3683 3684 if (width != cm->width || height != cm->height) { 3685 // There has been a change in the encoded frame size 3686 set_size_literal(cpi, width, height); 3687 set_mv_search_params(cpi); 3688 // Recalculate 'all_lossless' in case super-resolution was (un)selected. 3689 cm->all_lossless = cm->coded_lossless && !av1_superres_scaled(cm); 3690 } 3691 3692 if (cpi->oxcf.pass == 2) { 3693 av1_set_target_rate(cpi, cm->width, cm->height); 3694 } 3695 3696 alloc_frame_mvs(cm, cm->cur_frame); 3697 3698 // Allocate above context buffers 3699 if (cm->num_allocated_above_context_planes < av1_num_planes(cm) || 3700 cm->num_allocated_above_context_mi_col < cm->mi_cols || 3701 cm->num_allocated_above_contexts < cm->tile_rows) { 3702 av1_free_above_context_buffers(cm, cm->num_allocated_above_contexts); 3703 if (av1_alloc_above_context_buffers(cm, cm->tile_rows)) 3704 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, 3705 "Failed to allocate context buffers"); 3706 } 3707 3708 // Reset the frame pointers to the current frame size. 3709 if (aom_realloc_frame_buffer( 3710 &cm->cur_frame->buf, cm->width, cm->height, seq_params->subsampling_x, 3711 seq_params->subsampling_y, seq_params->use_highbitdepth, 3712 cpi->oxcf.border_in_pixels, cm->byte_alignment, NULL, NULL, NULL)) 3713 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, 3714 "Failed to allocate frame buffer"); 3715 3716 const int frame_width = cm->superres_upscaled_width; 3717 const int frame_height = cm->superres_upscaled_height; 3718 set_restoration_unit_size(frame_width, frame_height, 3719 seq_params->subsampling_x, 3720 seq_params->subsampling_y, cm->rst_info); 3721 for (int i = 0; i < num_planes; ++i) 3722 cm->rst_info[i].frame_restoration_type = RESTORE_NONE; 3723 3724 av1_alloc_restoration_buffers(cm); 3725 alloc_util_frame_buffers(cpi); 3726 init_motion_estimation(cpi); 3727 3728 for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) { 3729 RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame); 3730 if (buf != NULL) { 3731 struct scale_factors *sf = get_ref_scale_factors(cm, ref_frame); 3732 av1_setup_scale_factors_for_frame(sf, buf->buf.y_crop_width, 3733 buf->buf.y_crop_height, cm->width, 3734 cm->height); 3735 if (av1_is_scaled(sf)) aom_extend_frame_borders(&buf->buf, num_planes); 3736 } 3737 } 3738 3739 av1_setup_scale_factors_for_frame(&cm->sf_identity, cm->width, cm->height, 3740 cm->width, cm->height); 3741 3742 set_ref_ptrs(cm, xd, LAST_FRAME, LAST_FRAME); 3743 } 3744 3745 static uint8_t calculate_next_resize_scale(const AV1_COMP *cpi) { 3746 // Choose an arbitrary random number 3747 static unsigned int seed = 56789; 3748 const AV1EncoderConfig *oxcf = &cpi->oxcf; 3749 if (oxcf->pass == 1) return SCALE_NUMERATOR; 3750 uint8_t new_denom = SCALE_NUMERATOR; 3751 3752 if (cpi->common.seq_params.reduced_still_picture_hdr) return SCALE_NUMERATOR; 3753 switch (oxcf->resize_mode) { 3754 case RESIZE_NONE: new_denom = SCALE_NUMERATOR; break; 3755 case RESIZE_FIXED: 3756 if (cpi->common.current_frame.frame_type == KEY_FRAME) 3757 new_denom = oxcf->resize_kf_scale_denominator; 3758 else 3759 new_denom = oxcf->resize_scale_denominator; 3760 break; 3761 case RESIZE_RANDOM: new_denom = lcg_rand16(&seed) % 9 + 8; break; 3762 default: assert(0); 3763 } 3764 return new_denom; 3765 } 3766 3767 #define ENERGY_BY_Q2_THRESH 0.01 3768 #define ENERGY_BY_AC_THRESH 0.2 3769 3770 static uint8_t get_superres_denom_from_qindex_energy(int qindex, double *energy, 3771 double threshq, 3772 double threshp) { 3773 const double q = av1_convert_qindex_to_q(qindex, AOM_BITS_8); 3774 const double tq = threshq * q * q; 3775 const double tp = threshp * energy[1]; 3776 const double thresh = AOMMIN(tq, tp); 3777 int k; 3778 for (k = 16; k > 8; --k) { 3779 if (energy[k - 1] > thresh) break; 3780 } 3781 return 3 * SCALE_NUMERATOR - k; 3782 } 3783 3784 static uint8_t get_superres_denom_for_qindex(const AV1_COMP *cpi, int qindex) { 3785 double energy[16]; 3786 analyze_hor_freq(cpi, energy); 3787 /* 3788 printf("\nenergy = ["); 3789 for (int k = 1; k < 16; ++k) printf("%f, ", energy[k]); 3790 printf("]\n"); 3791 */ 3792 return get_superres_denom_from_qindex_energy( 3793 qindex, energy, ENERGY_BY_Q2_THRESH, ENERGY_BY_AC_THRESH); 3794 } 3795 3796 static uint8_t calculate_next_superres_scale(AV1_COMP *cpi) { 3797 // Choose an arbitrary random number 3798 static unsigned int seed = 34567; 3799 const AV1EncoderConfig *oxcf = &cpi->oxcf; 3800 if (oxcf->pass == 1) return SCALE_NUMERATOR; 3801 uint8_t new_denom = SCALE_NUMERATOR; 3802 3803 // Make sure that superres mode of the frame is consistent with the 3804 // sequence-level flag. 3805 assert(IMPLIES(oxcf->superres_mode != SUPERRES_NONE, 3806 cpi->common.seq_params.enable_superres)); 3807 assert(IMPLIES(!cpi->common.seq_params.enable_superres, 3808 oxcf->superres_mode == SUPERRES_NONE)); 3809 3810 switch (oxcf->superres_mode) { 3811 case SUPERRES_NONE: new_denom = SCALE_NUMERATOR; break; 3812 case SUPERRES_FIXED: 3813 if (cpi->common.current_frame.frame_type == KEY_FRAME) 3814 new_denom = oxcf->superres_kf_scale_denominator; 3815 else 3816 new_denom = oxcf->superres_scale_denominator; 3817 break; 3818 case SUPERRES_RANDOM: new_denom = lcg_rand16(&seed) % 9 + 8; break; 3819 case SUPERRES_QTHRESH: { 3820 // Do not use superres when screen content tools are used. 3821 if (cpi->common.allow_screen_content_tools) break; 3822 if (oxcf->rc_mode == AOM_VBR || oxcf->rc_mode == AOM_CQ) 3823 av1_set_target_rate(cpi, cpi->oxcf.width, cpi->oxcf.height); 3824 int bottom_index, top_index; 3825 const int q = av1_rc_pick_q_and_bounds( 3826 cpi, cpi->oxcf.width, cpi->oxcf.height, &bottom_index, &top_index); 3827 3828 const int qthresh = (frame_is_intra_only(&cpi->common)) 3829 ? oxcf->superres_kf_qthresh 3830 : oxcf->superres_qthresh; 3831 if (q <= qthresh) { 3832 new_denom = SCALE_NUMERATOR; 3833 } else { 3834 new_denom = get_superres_denom_for_qindex(cpi, q); 3835 } 3836 break; 3837 } 3838 case SUPERRES_AUTO: { 3839 // Don't use when screen content tools are used. 3840 if (cpi->common.allow_screen_content_tools) break; 3841 // Don't use for inter frames. 3842 if (!frame_is_intra_only(&cpi->common)) break; 3843 // Don't use for keyframes that can be used as references. 3844 if (cpi->rc.frames_to_key != 1) break; 3845 3846 // Now decide the use of superres based on 'q'. 3847 int bottom_index, top_index; 3848 const int q = av1_rc_pick_q_and_bounds( 3849 cpi, cpi->oxcf.width, cpi->oxcf.height, &bottom_index, &top_index); 3850 3851 const int qthresh = 128; 3852 if (q <= qthresh) { 3853 new_denom = SCALE_NUMERATOR; 3854 } else { 3855 new_denom = get_superres_denom_for_qindex(cpi, q); 3856 } 3857 break; 3858 } 3859 default: assert(0); 3860 } 3861 return new_denom; 3862 } 3863 3864 static int dimension_is_ok(int orig_dim, int resized_dim, int denom) { 3865 return (resized_dim * SCALE_NUMERATOR >= orig_dim * denom / 2); 3866 } 3867 3868 static int dimensions_are_ok(int owidth, int oheight, size_params_type *rsz) { 3869 // Only need to check the width, as scaling is horizontal only. 3870 (void)oheight; 3871 return dimension_is_ok(owidth, rsz->resize_width, rsz->superres_denom); 3872 } 3873 3874 static int validate_size_scales(RESIZE_MODE resize_mode, 3875 SUPERRES_MODE superres_mode, int owidth, 3876 int oheight, size_params_type *rsz) { 3877 if (dimensions_are_ok(owidth, oheight, rsz)) { // Nothing to do. 3878 return 1; 3879 } 3880 3881 // Calculate current resize scale. 3882 int resize_denom = 3883 AOMMAX(DIVIDE_AND_ROUND(owidth * SCALE_NUMERATOR, rsz->resize_width), 3884 DIVIDE_AND_ROUND(oheight * SCALE_NUMERATOR, rsz->resize_height)); 3885 3886 if (resize_mode != RESIZE_RANDOM && superres_mode == SUPERRES_RANDOM) { 3887 // Alter superres scale as needed to enforce conformity. 3888 rsz->superres_denom = 3889 (2 * SCALE_NUMERATOR * SCALE_NUMERATOR) / resize_denom; 3890 if (!dimensions_are_ok(owidth, oheight, rsz)) { 3891 if (rsz->superres_denom > SCALE_NUMERATOR) --rsz->superres_denom; 3892 } 3893 } else if (resize_mode == RESIZE_RANDOM && superres_mode != SUPERRES_RANDOM) { 3894 // Alter resize scale as needed to enforce conformity. 3895 resize_denom = 3896 (2 * SCALE_NUMERATOR * SCALE_NUMERATOR) / rsz->superres_denom; 3897 rsz->resize_width = owidth; 3898 rsz->resize_height = oheight; 3899 av1_calculate_scaled_size(&rsz->resize_width, &rsz->resize_height, 3900 resize_denom); 3901 if (!dimensions_are_ok(owidth, oheight, rsz)) { 3902 if (resize_denom > SCALE_NUMERATOR) { 3903 --resize_denom; 3904 rsz->resize_width = owidth; 3905 rsz->resize_height = oheight; 3906 av1_calculate_scaled_size(&rsz->resize_width, &rsz->resize_height, 3907 resize_denom); 3908 } 3909 } 3910 } else if (resize_mode == RESIZE_RANDOM && superres_mode == SUPERRES_RANDOM) { 3911 // Alter both resize and superres scales as needed to enforce conformity. 3912 do { 3913 if (resize_denom > rsz->superres_denom) 3914 --resize_denom; 3915 else 3916 --rsz->superres_denom; 3917 rsz->resize_width = owidth; 3918 rsz->resize_height = oheight; 3919 av1_calculate_scaled_size(&rsz->resize_width, &rsz->resize_height, 3920 resize_denom); 3921 } while (!dimensions_are_ok(owidth, oheight, rsz) && 3922 (resize_denom > SCALE_NUMERATOR || 3923 rsz->superres_denom > SCALE_NUMERATOR)); 3924 } else { // We are allowed to alter neither resize scale nor superres 3925 // scale. 3926 return 0; 3927 } 3928 return dimensions_are_ok(owidth, oheight, rsz); 3929 } 3930 3931 // Calculates resize and superres params for next frame 3932 static size_params_type calculate_next_size_params(AV1_COMP *cpi) { 3933 const AV1EncoderConfig *oxcf = &cpi->oxcf; 3934 size_params_type rsz = { oxcf->width, oxcf->height, SCALE_NUMERATOR }; 3935 int resize_denom; 3936 if (oxcf->pass == 1) return rsz; 3937 if (cpi->resize_pending_width && cpi->resize_pending_height) { 3938 rsz.resize_width = cpi->resize_pending_width; 3939 rsz.resize_height = cpi->resize_pending_height; 3940 cpi->resize_pending_width = cpi->resize_pending_height = 0; 3941 } else { 3942 resize_denom = calculate_next_resize_scale(cpi); 3943 rsz.resize_width = cpi->oxcf.width; 3944 rsz.resize_height = cpi->oxcf.height; 3945 av1_calculate_scaled_size(&rsz.resize_width, &rsz.resize_height, 3946 resize_denom); 3947 } 3948 rsz.superres_denom = calculate_next_superres_scale(cpi); 3949 if (!validate_size_scales(oxcf->resize_mode, oxcf->superres_mode, oxcf->width, 3950 oxcf->height, &rsz)) 3951 assert(0 && "Invalid scale parameters"); 3952 return rsz; 3953 } 3954 3955 static void setup_frame_size_from_params(AV1_COMP *cpi, 3956 const size_params_type *rsz) { 3957 int encode_width = rsz->resize_width; 3958 int encode_height = rsz->resize_height; 3959 3960 AV1_COMMON *cm = &cpi->common; 3961 cm->superres_upscaled_width = encode_width; 3962 cm->superres_upscaled_height = encode_height; 3963 cm->superres_scale_denominator = rsz->superres_denom; 3964 av1_calculate_scaled_superres_size(&encode_width, &encode_height, 3965 rsz->superres_denom); 3966 av1_set_frame_size(cpi, encode_width, encode_height); 3967 } 3968 3969 void av1_setup_frame_size(AV1_COMP *cpi) { 3970 AV1_COMMON *cm = &cpi->common; 3971 // Reset superres params from previous frame. 3972 cm->superres_scale_denominator = SCALE_NUMERATOR; 3973 const size_params_type rsz = calculate_next_size_params(cpi); 3974 setup_frame_size_from_params(cpi, &rsz); 3975 3976 assert(is_min_tile_width_satisfied(cm)); 3977 } 3978 3979 static void superres_post_encode(AV1_COMP *cpi) { 3980 AV1_COMMON *cm = &cpi->common; 3981 const int num_planes = av1_num_planes(cm); 3982 3983 if (!av1_superres_scaled(cm)) return; 3984 3985 assert(cpi->oxcf.enable_superres); 3986 assert(!is_lossless_requested(&cpi->oxcf)); 3987 assert(!cm->all_lossless); 3988 3989 av1_superres_upscale(cm, NULL); 3990 3991 // If regular resizing is occurring the source will need to be downscaled to 3992 // match the upscaled superres resolution. Otherwise the original source is 3993 // used. 3994 if (!av1_resize_scaled(cm)) { 3995 cpi->source = cpi->unscaled_source; 3996 if (cpi->last_source != NULL) cpi->last_source = cpi->unscaled_last_source; 3997 } else { 3998 assert(cpi->unscaled_source->y_crop_width != cm->superres_upscaled_width); 3999 assert(cpi->unscaled_source->y_crop_height != cm->superres_upscaled_height); 4000 // Do downscale. cm->(width|height) has been updated by 4001 // av1_superres_upscale 4002 if (aom_realloc_frame_buffer( 4003 &cpi->scaled_source, cm->superres_upscaled_width, 4004 cm->superres_upscaled_height, cm->seq_params.subsampling_x, 4005 cm->seq_params.subsampling_y, cm->seq_params.use_highbitdepth, 4006 AOM_BORDER_IN_PIXELS, cm->byte_alignment, NULL, NULL, NULL)) 4007 aom_internal_error( 4008 &cm->error, AOM_CODEC_MEM_ERROR, 4009 "Failed to reallocate scaled source buffer for superres"); 4010 assert(cpi->scaled_source.y_crop_width == cm->superres_upscaled_width); 4011 assert(cpi->scaled_source.y_crop_height == cm->superres_upscaled_height); 4012 av1_resize_and_extend_frame(cpi->unscaled_source, &cpi->scaled_source, 4013 (int)cm->seq_params.bit_depth, num_planes); 4014 cpi->source = &cpi->scaled_source; 4015 } 4016 } 4017 4018 static void loopfilter_frame(AV1_COMP *cpi, AV1_COMMON *cm) { 4019 const int num_planes = av1_num_planes(cm); 4020 MACROBLOCKD *xd = &cpi->td.mb.e_mbd; 4021 4022 assert(IMPLIES(is_lossless_requested(&cpi->oxcf), 4023 cm->coded_lossless && cm->all_lossless)); 4024 4025 const int use_loopfilter = !cm->coded_lossless && !cm->large_scale_tile; 4026 const int use_cdef = cm->seq_params.enable_cdef && !cm->coded_lossless && 4027 !cm->large_scale_tile; 4028 const int use_restoration = cm->seq_params.enable_restoration && 4029 !cm->all_lossless && !cm->large_scale_tile; 4030 4031 struct loopfilter *lf = &cm->lf; 4032 4033 #if CONFIG_COLLECT_COMPONENT_TIMING 4034 start_timing(cpi, loop_filter_time); 4035 #endif 4036 if (use_loopfilter) { 4037 aom_clear_system_state(); 4038 av1_pick_filter_level(cpi->source, cpi, cpi->sf.lpf_pick); 4039 } else { 4040 lf->filter_level[0] = 0; 4041 lf->filter_level[1] = 0; 4042 } 4043 4044 if (lf->filter_level[0] || lf->filter_level[1]) { 4045 if (cpi->num_workers > 1) 4046 av1_loop_filter_frame_mt(&cm->cur_frame->buf, cm, xd, 0, num_planes, 0, 4047 #if LOOP_FILTER_BITMASK 4048 0, 4049 #endif 4050 cpi->workers, cpi->num_workers, 4051 &cpi->lf_row_sync); 4052 else 4053 av1_loop_filter_frame(&cm->cur_frame->buf, cm, xd, 4054 #if LOOP_FILTER_BITMASK 4055 0, 4056 #endif 4057 0, num_planes, 0); 4058 } 4059 #if CONFIG_COLLECT_COMPONENT_TIMING 4060 end_timing(cpi, loop_filter_time); 4061 #endif 4062 4063 if (use_restoration) 4064 av1_loop_restoration_save_boundary_lines(&cm->cur_frame->buf, cm, 0); 4065 4066 if (use_cdef) { 4067 #if CONFIG_COLLECT_COMPONENT_TIMING 4068 start_timing(cpi, cdef_time); 4069 #endif 4070 // Find CDEF parameters 4071 av1_cdef_search(&cm->cur_frame->buf, cpi->source, cm, xd, 4072 cpi->sf.fast_cdef_search); 4073 4074 // Apply the filter 4075 av1_cdef_frame(&cm->cur_frame->buf, cm, xd); 4076 #if CONFIG_COLLECT_COMPONENT_TIMING 4077 end_timing(cpi, cdef_time); 4078 #endif 4079 } else { 4080 cm->cdef_info.cdef_bits = 0; 4081 cm->cdef_info.cdef_strengths[0] = 0; 4082 cm->cdef_info.nb_cdef_strengths = 1; 4083 cm->cdef_info.cdef_uv_strengths[0] = 0; 4084 } 4085 4086 superres_post_encode(cpi); 4087 4088 #if CONFIG_COLLECT_COMPONENT_TIMING 4089 start_timing(cpi, loop_restoration_time); 4090 #endif 4091 if (use_restoration) { 4092 av1_loop_restoration_save_boundary_lines(&cm->cur_frame->buf, cm, 1); 4093 av1_pick_filter_restoration(cpi->source, cpi); 4094 if (cm->rst_info[0].frame_restoration_type != RESTORE_NONE || 4095 cm->rst_info[1].frame_restoration_type != RESTORE_NONE || 4096 cm->rst_info[2].frame_restoration_type != RESTORE_NONE) { 4097 if (cpi->num_workers > 1) 4098 av1_loop_restoration_filter_frame_mt(&cm->cur_frame->buf, cm, 0, 4099 cpi->workers, cpi->num_workers, 4100 &cpi->lr_row_sync, &cpi->lr_ctxt); 4101 else 4102 av1_loop_restoration_filter_frame(&cm->cur_frame->buf, cm, 0, 4103 &cpi->lr_ctxt); 4104 } 4105 } else { 4106 cm->rst_info[0].frame_restoration_type = RESTORE_NONE; 4107 cm->rst_info[1].frame_restoration_type = RESTORE_NONE; 4108 cm->rst_info[2].frame_restoration_type = RESTORE_NONE; 4109 } 4110 #if CONFIG_COLLECT_COMPONENT_TIMING 4111 end_timing(cpi, loop_restoration_time); 4112 #endif 4113 } 4114 4115 static void fix_interp_filter(InterpFilter *const interp_filter, 4116 const FRAME_COUNTS *const counts) { 4117 if (*interp_filter == SWITCHABLE) { 4118 // Check to see if only one of the filters is actually used 4119 int count[SWITCHABLE_FILTERS] = { 0 }; 4120 int num_filters_used = 0; 4121 for (int i = 0; i < SWITCHABLE_FILTERS; ++i) { 4122 for (int j = 0; j < SWITCHABLE_FILTER_CONTEXTS; ++j) 4123 count[i] += counts->switchable_interp[j][i]; 4124 num_filters_used += (count[i] > 0); 4125 } 4126 if (num_filters_used == 1) { 4127 // Only one filter is used. So set the filter at frame level 4128 for (int i = 0; i < SWITCHABLE_FILTERS; ++i) { 4129 if (count[i]) { 4130 if (i == EIGHTTAP_REGULAR) *interp_filter = i; 4131 break; 4132 } 4133 } 4134 } 4135 } 4136 } 4137 4138 static void finalize_encoded_frame(AV1_COMP *const cpi) { 4139 AV1_COMMON *const cm = &cpi->common; 4140 CurrentFrame *const current_frame = &cm->current_frame; 4141 4142 if (!cm->seq_params.reduced_still_picture_hdr && 4143 encode_show_existing_frame(cm)) { 4144 RefCntBuffer *const frame_to_show = 4145 cm->ref_frame_map[cpi->existing_fb_idx_to_show]; 4146 4147 if (frame_to_show == NULL) { 4148 aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, 4149 "Buffer does not contain a reconstructed frame"); 4150 } 4151 assert(frame_to_show->ref_count > 0); 4152 assign_frame_buffer_p(&cm->cur_frame, frame_to_show); 4153 } 4154 4155 if (!encode_show_existing_frame(cm) && 4156 cm->seq_params.film_grain_params_present && 4157 (cm->show_frame || cm->showable_frame)) { 4158 // Copy the current frame's film grain params to the its corresponding 4159 // RefCntBuffer slot. 4160 cm->cur_frame->film_grain_params = cm->film_grain_params; 4161 4162 // We must update the parameters if this is not an INTER_FRAME 4163 if (current_frame->frame_type != INTER_FRAME) 4164 cm->cur_frame->film_grain_params.update_parameters = 1; 4165 4166 // Iterate the random seed for the next frame. 4167 cm->film_grain_params.random_seed += 3381; 4168 if (cm->film_grain_params.random_seed == 0) 4169 cm->film_grain_params.random_seed = 7391; 4170 } 4171 4172 // Initialise all tiles' contexts from the global frame context 4173 for (int tile_col = 0; tile_col < cm->tile_cols; tile_col++) { 4174 for (int tile_row = 0; tile_row < cm->tile_rows; tile_row++) { 4175 const int tile_idx = tile_row * cm->tile_cols + tile_col; 4176 cpi->tile_data[tile_idx].tctx = *cm->fc; 4177 } 4178 } 4179 4180 fix_interp_filter(&cm->interp_filter, cpi->td.counts); 4181 } 4182 4183 static int get_regulated_q_overshoot(AV1_COMP *const cpi, int q_low, int q_high, 4184 int top_index, int bottom_index) { 4185 const AV1_COMMON *const cm = &cpi->common; 4186 const RATE_CONTROL *const rc = &cpi->rc; 4187 4188 av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height); 4189 4190 int q_regulated = 4191 av1_rc_regulate_q(cpi, rc->this_frame_target, bottom_index, 4192 AOMMAX(q_high, top_index), cm->width, cm->height); 4193 4194 int retries = 0; 4195 while (q_regulated < q_low && retries < 10) { 4196 av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height); 4197 q_regulated = 4198 av1_rc_regulate_q(cpi, rc->this_frame_target, bottom_index, 4199 AOMMAX(q_high, top_index), cm->width, cm->height); 4200 retries++; 4201 } 4202 return q_regulated; 4203 } 4204 4205 static int get_regulated_q_undershoot(AV1_COMP *const cpi, int q_high, 4206 int top_index, int bottom_index) { 4207 const AV1_COMMON *const cm = &cpi->common; 4208 const RATE_CONTROL *const rc = &cpi->rc; 4209 4210 av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height); 4211 int q_regulated = av1_rc_regulate_q(cpi, rc->this_frame_target, bottom_index, 4212 top_index, cm->width, cm->height); 4213 4214 int retries = 0; 4215 while (q_regulated > q_high && retries < 10) { 4216 av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height); 4217 q_regulated = av1_rc_regulate_q(cpi, rc->this_frame_target, bottom_index, 4218 top_index, cm->width, cm->height); 4219 retries++; 4220 } 4221 return q_regulated; 4222 } 4223 4224 // Called after encode_with_recode_loop() has just encoded a frame and packed 4225 // its bitstream. This function works out whether we under- or over-shot 4226 // our bitrate target and adjusts q as appropriate. Also decides whether 4227 // or not we should do another recode loop, indicated by *loop 4228 static void recode_loop_update_q(AV1_COMP *const cpi, int *const loop, 4229 int *const q, int *const q_low, 4230 int *const q_high, const int top_index, 4231 const int bottom_index, 4232 int *const undershoot_seen, 4233 int *const overshoot_seen, 4234 const int loop_at_this_size) { 4235 AV1_COMMON *const cm = &cpi->common; 4236 RATE_CONTROL *const rc = &cpi->rc; 4237 4238 int frame_over_shoot_limit = 0, frame_under_shoot_limit = 0; 4239 av1_rc_compute_frame_size_bounds(cpi, rc->this_frame_target, 4240 &frame_under_shoot_limit, 4241 &frame_over_shoot_limit); 4242 if (frame_over_shoot_limit == 0) frame_over_shoot_limit = 1; 4243 4244 if ((cm->current_frame.frame_type == KEY_FRAME) && 4245 rc->this_key_frame_forced && 4246 (rc->projected_frame_size < rc->max_frame_bandwidth)) { 4247 int last_q = *q; 4248 int64_t kf_err; 4249 4250 int64_t high_err_target = cpi->ambient_err; 4251 int64_t low_err_target = cpi->ambient_err >> 1; 4252 4253 if (cm->seq_params.use_highbitdepth) { 4254 kf_err = aom_highbd_get_y_sse(cpi->source, &cm->cur_frame->buf); 4255 } else { 4256 kf_err = aom_get_y_sse(cpi->source, &cm->cur_frame->buf); 4257 } 4258 // Prevent possible divide by zero error below for perfect KF 4259 kf_err += !kf_err; 4260 4261 // The key frame is not good enough or we can afford 4262 // to make it better without undue risk of popping. 4263 if ((kf_err > high_err_target && 4264 rc->projected_frame_size <= frame_over_shoot_limit) || 4265 (kf_err > low_err_target && 4266 rc->projected_frame_size <= frame_under_shoot_limit)) { 4267 // Lower q_high 4268 *q_high = *q > *q_low ? *q - 1 : *q_low; 4269 4270 // Adjust Q 4271 *q = (int)((*q * high_err_target) / kf_err); 4272 *q = AOMMIN(*q, (*q_high + *q_low) >> 1); 4273 } else if (kf_err < low_err_target && 4274 rc->projected_frame_size >= frame_under_shoot_limit) { 4275 // The key frame is much better than the previous frame 4276 // Raise q_low 4277 *q_low = *q < *q_high ? *q + 1 : *q_high; 4278 4279 // Adjust Q 4280 *q = (int)((*q * low_err_target) / kf_err); 4281 *q = AOMMIN(*q, (*q_high + *q_low + 1) >> 1); 4282 } 4283 4284 // Clamp Q to upper and lower limits: 4285 *q = clamp(*q, *q_low, *q_high); 4286 4287 *loop = *q != last_q; 4288 } else if (recode_loop_test(cpi, frame_over_shoot_limit, 4289 frame_under_shoot_limit, *q, 4290 AOMMAX(*q_high, top_index), bottom_index)) { 4291 // Is the projected frame size out of range and are we allowed 4292 // to attempt to recode. 4293 int last_q = *q; 4294 4295 // Frame size out of permitted range: 4296 // Update correction factor & compute new Q to try... 4297 // Frame is too large 4298 if (rc->projected_frame_size > rc->this_frame_target) { 4299 // Special case if the projected size is > the max allowed. 4300 if (rc->projected_frame_size >= rc->max_frame_bandwidth) 4301 *q_high = rc->worst_quality; 4302 4303 // Raise Qlow as to at least the current value 4304 *q_low = *q < *q_high ? *q + 1 : *q_high; 4305 4306 if (*undershoot_seen || loop_at_this_size > 2 || 4307 (loop_at_this_size == 2 && !frame_is_intra_only(cm))) { 4308 av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height); 4309 4310 *q = (*q_high + *q_low + 1) / 2; 4311 } else if (loop_at_this_size == 2 && frame_is_intra_only(cm)) { 4312 const int q_mid = (*q_high + *q_low + 1) / 2; 4313 const int q_regulated = get_regulated_q_overshoot( 4314 cpi, *q_low, *q_high, top_index, bottom_index); 4315 // Get 'q' in-between 'q_mid' and 'q_regulated' for a smooth 4316 // transition between loop_at_this_size < 2 and loop_at_this_size > 2. 4317 *q = (q_mid + q_regulated + 1) / 2; 4318 } else { 4319 *q = get_regulated_q_overshoot(cpi, *q_low, *q_high, top_index, 4320 bottom_index); 4321 } 4322 4323 *overshoot_seen = 1; 4324 } else { 4325 // Frame is too small 4326 *q_high = *q > *q_low ? *q - 1 : *q_low; 4327 4328 if (*overshoot_seen || loop_at_this_size > 2 || 4329 (loop_at_this_size == 2 && !frame_is_intra_only(cm))) { 4330 av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height); 4331 *q = (*q_high + *q_low) / 2; 4332 } else if (loop_at_this_size == 2 && frame_is_intra_only(cm)) { 4333 const int q_mid = (*q_high + *q_low) / 2; 4334 const int q_regulated = 4335 get_regulated_q_undershoot(cpi, *q_high, top_index, bottom_index); 4336 // Get 'q' in-between 'q_mid' and 'q_regulated' for a smooth 4337 // transition between loop_at_this_size < 2 and loop_at_this_size > 2. 4338 *q = (q_mid + q_regulated) / 2; 4339 4340 // Special case reset for qlow for constrained quality. 4341 // This should only trigger where there is very substantial 4342 // undershoot on a frame and the auto cq level is above 4343 // the user passsed in value. 4344 if (cpi->oxcf.rc_mode == AOM_CQ && q_regulated < *q_low) { 4345 *q_low = *q; 4346 } 4347 } else { 4348 *q = get_regulated_q_undershoot(cpi, *q_high, top_index, bottom_index); 4349 4350 // Special case reset for qlow for constrained quality. 4351 // This should only trigger where there is very substantial 4352 // undershoot on a frame and the auto cq level is above 4353 // the user passsed in value. 4354 if (cpi->oxcf.rc_mode == AOM_CQ && *q < *q_low) { 4355 *q_low = *q; 4356 } 4357 } 4358 4359 *undershoot_seen = 1; 4360 } 4361 4362 // Clamp Q to upper and lower limits: 4363 *q = clamp(*q, *q_low, *q_high); 4364 4365 *loop = (*q != last_q); 4366 } else { 4367 *loop = 0; 4368 } 4369 } 4370 4371 static int encode_with_recode_loop(AV1_COMP *cpi, size_t *size, uint8_t *dest) { 4372 AV1_COMMON *const cm = &cpi->common; 4373 RATE_CONTROL *const rc = &cpi->rc; 4374 const int allow_recode = cpi->sf.recode_loop != DISALLOW_RECODE; 4375 4376 set_size_independent_vars(cpi); 4377 4378 cpi->source->buf_8bit_valid = 0; 4379 4380 av1_setup_frame_size(cpi); 4381 4382 int top_index = 0, bottom_index = 0; 4383 int q = 0, q_low = 0, q_high = 0; 4384 set_size_dependent_vars(cpi, &q, &bottom_index, &top_index); 4385 q_low = bottom_index; 4386 q_high = top_index; 4387 4388 // Loop variables 4389 int loop_count = 0; 4390 int loop_at_this_size = 0; 4391 int loop = 0; 4392 int overshoot_seen = 0; 4393 int undershoot_seen = 0; 4394 4395 #if CONFIG_COLLECT_COMPONENT_TIMING 4396 printf("\n Encoding a frame:"); 4397 #endif 4398 do { 4399 aom_clear_system_state(); 4400 4401 // if frame was scaled calculate global_motion_search again if already 4402 // done 4403 if (loop_count > 0 && cpi->source && cpi->global_motion_search_done) { 4404 if (cpi->source->y_crop_width != cm->width || 4405 cpi->source->y_crop_height != cm->height) { 4406 cpi->global_motion_search_done = 0; 4407 } 4408 } 4409 cpi->source = 4410 av1_scale_if_required(cm, cpi->unscaled_source, &cpi->scaled_source); 4411 if (cpi->unscaled_last_source != NULL) { 4412 cpi->last_source = av1_scale_if_required(cm, cpi->unscaled_last_source, 4413 &cpi->scaled_last_source); 4414 } 4415 4416 if (!frame_is_intra_only(cm)) { 4417 if (loop_count > 0) { 4418 release_scaled_references(cpi); 4419 } 4420 scale_references(cpi); 4421 } 4422 av1_set_quantizer(cm, q); 4423 av1_init_quantizer(cpi); 4424 4425 av1_set_variance_partition_thresholds(cpi, q, 0); 4426 4427 // printf("Frame %d/%d: q = %d, frame_type = %d superres_denom = %d\n", 4428 // cm->current_frame.frame_number, cm->show_frame, q, 4429 // cm->current_frame.frame_type, cm->superres_scale_denominator); 4430 4431 if (loop_count == 0) { 4432 setup_frame(cpi); 4433 } else if (get_primary_ref_frame_buf(cm) == NULL) { 4434 // Base q-index may have changed, so we need to assign proper default coef 4435 // probs before every iteration. 4436 av1_default_coef_probs(cm); 4437 av1_setup_frame_contexts(cm); 4438 } 4439 4440 if (cpi->oxcf.aq_mode == VARIANCE_AQ) { 4441 av1_vaq_frame_setup(cpi); 4442 } else if (cpi->oxcf.aq_mode == COMPLEXITY_AQ) { 4443 av1_setup_in_frame_q_adj(cpi); 4444 } else if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ && !allow_recode) { 4445 suppress_active_map(cpi); 4446 av1_cyclic_refresh_setup(cpi); 4447 apply_active_map(cpi); 4448 } 4449 4450 if (cm->seg.enabled) { 4451 if (!cm->seg.update_data && cm->prev_frame) { 4452 segfeatures_copy(&cm->seg, &cm->prev_frame->seg); 4453 } else { 4454 calculate_segdata(&cm->seg); 4455 } 4456 } else { 4457 memset(&cm->seg, 0, sizeof(cm->seg)); 4458 } 4459 segfeatures_copy(&cm->cur_frame->seg, &cm->seg); 4460 4461 if (allow_recode) save_coding_context(cpi); 4462 #if CONFIG_COLLECT_COMPONENT_TIMING 4463 start_timing(cpi, av1_encode_frame_time); 4464 #endif 4465 // transform / motion compensation build reconstruction frame 4466 av1_encode_frame(cpi); 4467 #if CONFIG_COLLECT_COMPONENT_TIMING 4468 end_timing(cpi, av1_encode_frame_time); 4469 #endif 4470 4471 aom_clear_system_state(); 4472 4473 // Dummy pack of the bitstream using up to date stats to get an 4474 // accurate estimate of output frame size to determine if we need 4475 // to recode. 4476 if (cpi->sf.recode_loop >= ALLOW_RECODE_KFARFGF) { 4477 restore_coding_context(cpi); 4478 4479 finalize_encoded_frame(cpi); 4480 int largest_tile_id = 0; // Output from bitstream: unused here 4481 if (av1_pack_bitstream(cpi, dest, size, &largest_tile_id) != AOM_CODEC_OK) 4482 return AOM_CODEC_ERROR; 4483 4484 rc->projected_frame_size = (int)(*size) << 3; 4485 restore_coding_context(cpi); 4486 } 4487 4488 if (allow_recode && cpi->oxcf.rc_mode != AOM_Q) { 4489 // Update q and decide whether to do a recode loop 4490 recode_loop_update_q(cpi, &loop, &q, &q_low, &q_high, top_index, 4491 bottom_index, &undershoot_seen, &overshoot_seen, 4492 loop_at_this_size); 4493 } 4494 4495 // Special case for overlay frame. 4496 if (rc->is_src_frame_alt_ref && 4497 rc->projected_frame_size < rc->max_frame_bandwidth) 4498 loop = 0; 4499 4500 if (allow_recode && !cpi->sf.gm_disable_recode && 4501 recode_loop_test_global_motion(cpi)) { 4502 loop = 1; 4503 } 4504 4505 if (loop) { 4506 ++loop_count; 4507 ++loop_at_this_size; 4508 4509 #if CONFIG_INTERNAL_STATS 4510 ++cpi->tot_recode_hits; 4511 #endif 4512 } 4513 #if CONFIG_COLLECT_COMPONENT_TIMING 4514 if (loop) printf("\n Recoding:"); 4515 #endif 4516 } while (loop); 4517 4518 return AOM_CODEC_OK; 4519 } 4520 4521 #define DUMP_RECON_FRAMES 0 4522 4523 #if DUMP_RECON_FRAMES == 1 4524 // NOTE(zoeliu): For debug - Output the filtered reconstructed video. 4525 static void dump_filtered_recon_frames(AV1_COMP *cpi) { 4526 AV1_COMMON *const cm = &cpi->common; 4527 const CurrentFrame *const current_frame = &cm->current_frame; 4528 const YV12_BUFFER_CONFIG *recon_buf = &cm->cur_frame->buf; 4529 4530 if (recon_buf == NULL) { 4531 printf("Frame %d is not ready.\n", current_frame->frame_number); 4532 return; 4533 } 4534 4535 static const int flag_list[REF_FRAMES] = { 0, 4536 AOM_LAST_FLAG, 4537 AOM_LAST2_FLAG, 4538 AOM_LAST3_FLAG, 4539 AOM_GOLD_FLAG, 4540 AOM_BWD_FLAG, 4541 AOM_ALT2_FLAG, 4542 AOM_ALT_FLAG }; 4543 printf( 4544 "\n***Frame=%d (frame_offset=%d, show_frame=%d, " 4545 "show_existing_frame=%d) " 4546 "[LAST LAST2 LAST3 GOLDEN BWD ALT2 ALT]=[", 4547 current_frame->frame_number, current_frame->order_hint, cm->show_frame, 4548 cm->show_existing_frame); 4549 for (int ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) { 4550 const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame); 4551 const int ref_offset = buf != NULL ? (int)buf->order_hint : -1; 4552 printf(" %d(%c)", ref_offset, 4553 (cpi->ref_frame_flags & flag_list[ref_frame]) ? 'Y' : 'N'); 4554 } 4555 printf(" ]\n"); 4556 4557 if (!cm->show_frame) { 4558 printf("Frame %d is a no show frame, so no image dump.\n", 4559 current_frame->frame_number); 4560 return; 4561 } 4562 4563 int h; 4564 char file_name[256] = "/tmp/enc_filtered_recon.yuv"; 4565 FILE *f_recon = NULL; 4566 4567 if (current_frame->frame_number == 0) { 4568 if ((f_recon = fopen(file_name, "wb")) == NULL) { 4569 printf("Unable to open file %s to write.\n", file_name); 4570 return; 4571 } 4572 } else { 4573 if ((f_recon = fopen(file_name, "ab")) == NULL) { 4574 printf("Unable to open file %s to append.\n", file_name); 4575 return; 4576 } 4577 } 4578 printf( 4579 "\nFrame=%5d, encode_update_type[%5d]=%1d, frame_offset=%d, " 4580 "show_frame=%d, show_existing_frame=%d, source_alt_ref_active=%d, " 4581 "refresh_alt_ref_frame=%d, " 4582 "y_stride=%4d, uv_stride=%4d, cm->width=%4d, cm->height=%4d\n\n", 4583 current_frame->frame_number, cpi->twopass.gf_group.index, 4584 cpi->twopass.gf_group.update_type[cpi->twopass.gf_group.index], 4585 current_frame->order_hint, cm->show_frame, cm->show_existing_frame, 4586 cpi->rc.source_alt_ref_active, cpi->refresh_alt_ref_frame, 4587 recon_buf->y_stride, recon_buf->uv_stride, cm->width, cm->height); 4588 #if 0 4589 int ref_frame; 4590 printf("get_ref_frame_map_idx: ["); 4591 for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) 4592 printf(" %d", get_ref_frame_map_idx(cm, ref_frame)); 4593 printf(" ]\n"); 4594 #endif // 0 4595 4596 // --- Y --- 4597 for (h = 0; h < cm->height; ++h) { 4598 fwrite(&recon_buf->y_buffer[h * recon_buf->y_stride], 1, cm->width, 4599 f_recon); 4600 } 4601 // --- U --- 4602 for (h = 0; h < (cm->height >> 1); ++h) { 4603 fwrite(&recon_buf->u_buffer[h * recon_buf->uv_stride], 1, (cm->width >> 1), 4604 f_recon); 4605 } 4606 // --- V --- 4607 for (h = 0; h < (cm->height >> 1); ++h) { 4608 fwrite(&recon_buf->v_buffer[h * recon_buf->uv_stride], 1, (cm->width >> 1), 4609 f_recon); 4610 } 4611 4612 fclose(f_recon); 4613 } 4614 #endif // DUMP_RECON_FRAMES 4615 4616 static int get_interp_filter_selected(const AV1_COMMON *const cm, 4617 MV_REFERENCE_FRAME ref, 4618 InterpFilters ifilter) { 4619 const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref); 4620 if (buf == NULL) return 0; 4621 return buf->interp_filter_selected[ifilter]; 4622 } 4623 4624 static int setup_interp_filter_search_mask(AV1_COMP *cpi) { 4625 const AV1_COMMON *const cm = &cpi->common; 4626 int ref_total[REF_FRAMES] = { 0 }; 4627 4628 if (cpi->common.last_frame_type == KEY_FRAME || cpi->refresh_alt_ref_frame) 4629 return 0; 4630 4631 for (MV_REFERENCE_FRAME ref = LAST_FRAME; ref <= ALTREF_FRAME; ++ref) { 4632 for (InterpFilters ifilter = EIGHTTAP_REGULAR; ifilter <= MULTITAP_SHARP; 4633 ++ifilter) { 4634 ref_total[ref] += get_interp_filter_selected(cm, ref, ifilter); 4635 } 4636 } 4637 int ref_total_total = (ref_total[LAST2_FRAME] + ref_total[LAST3_FRAME] + 4638 ref_total[GOLDEN_FRAME] + ref_total[BWDREF_FRAME] + 4639 ref_total[ALTREF2_FRAME] + ref_total[ALTREF_FRAME]); 4640 4641 int mask = 0; 4642 for (InterpFilters ifilter = EIGHTTAP_REGULAR; ifilter <= MULTITAP_SHARP; 4643 ++ifilter) { 4644 int last_score = get_interp_filter_selected(cm, LAST_FRAME, ifilter) * 30; 4645 if (ref_total[LAST_FRAME] && last_score <= ref_total[LAST_FRAME]) { 4646 int filter_score = 4647 get_interp_filter_selected(cm, LAST2_FRAME, ifilter) * 20 + 4648 get_interp_filter_selected(cm, LAST3_FRAME, ifilter) * 20 + 4649 get_interp_filter_selected(cm, GOLDEN_FRAME, ifilter) * 20 + 4650 get_interp_filter_selected(cm, BWDREF_FRAME, ifilter) * 10 + 4651 get_interp_filter_selected(cm, ALTREF2_FRAME, ifilter) * 10 + 4652 get_interp_filter_selected(cm, ALTREF_FRAME, ifilter) * 10; 4653 if (filter_score < ref_total_total) mask |= 1 << ifilter; 4654 } 4655 } 4656 return mask; 4657 } 4658 4659 static int is_integer_mv(AV1_COMP *cpi, const YV12_BUFFER_CONFIG *cur_picture, 4660 const YV12_BUFFER_CONFIG *last_picture, 4661 hash_table *last_hash_table) { 4662 aom_clear_system_state(); 4663 // check use hash ME 4664 int k; 4665 uint32_t hash_value_1; 4666 uint32_t hash_value_2; 4667 4668 const int block_size = 8; 4669 const double threshold_current = 0.8; 4670 const double threshold_average = 0.95; 4671 const int max_history_size = 32; 4672 int T = 0; // total block 4673 int C = 0; // match with collocated block 4674 int S = 0; // smooth region but not match with collocated block 4675 int M = 0; // match with other block 4676 4677 const int pic_width = cur_picture->y_width; 4678 const int pic_height = cur_picture->y_height; 4679 for (int i = 0; i + block_size <= pic_height; i += block_size) { 4680 for (int j = 0; j + block_size <= pic_width; j += block_size) { 4681 const int x_pos = j; 4682 const int y_pos = i; 4683 int match = 1; 4684 T++; 4685 4686 // check whether collocated block match with current 4687 uint8_t *p_cur = cur_picture->y_buffer; 4688 uint8_t *p_ref = last_picture->y_buffer; 4689 int stride_cur = cur_picture->y_stride; 4690 int stride_ref = last_picture->y_stride; 4691 p_cur += (y_pos * stride_cur + x_pos); 4692 p_ref += (y_pos * stride_ref + x_pos); 4693 4694 if (cur_picture->flags & YV12_FLAG_HIGHBITDEPTH) { 4695 uint16_t *p16_cur = CONVERT_TO_SHORTPTR(p_cur); 4696 uint16_t *p16_ref = CONVERT_TO_SHORTPTR(p_ref); 4697 for (int tmpY = 0; tmpY < block_size && match; tmpY++) { 4698 for (int tmpX = 0; tmpX < block_size && match; tmpX++) { 4699 if (p16_cur[tmpX] != p16_ref[tmpX]) { 4700 match = 0; 4701 } 4702 } 4703 p16_cur += stride_cur; 4704 p16_ref += stride_ref; 4705 } 4706 } else { 4707 for (int tmpY = 0; tmpY < block_size && match; tmpY++) { 4708 for (int tmpX = 0; tmpX < block_size && match; tmpX++) { 4709 if (p_cur[tmpX] != p_ref[tmpX]) { 4710 match = 0; 4711 } 4712 } 4713 p_cur += stride_cur; 4714 p_ref += stride_ref; 4715 } 4716 } 4717 4718 if (match) { 4719 C++; 4720 continue; 4721 } 4722 4723 if (av1_hash_is_horizontal_perfect(cur_picture, block_size, x_pos, 4724 y_pos) || 4725 av1_hash_is_vertical_perfect(cur_picture, block_size, x_pos, y_pos)) { 4726 S++; 4727 continue; 4728 } 4729 4730 av1_get_block_hash_value( 4731 cur_picture->y_buffer + y_pos * stride_cur + x_pos, stride_cur, 4732 block_size, &hash_value_1, &hash_value_2, 4733 (cur_picture->flags & YV12_FLAG_HIGHBITDEPTH), &cpi->td.mb); 4734 // Hashing does not work for highbitdepth currently. 4735 // TODO(Roger): Make it work for highbitdepth. 4736 if (av1_use_hash_me(&cpi->common)) { 4737 if (av1_has_exact_match(last_hash_table, hash_value_1, hash_value_2)) { 4738 M++; 4739 } 4740 } 4741 } 4742 } 4743 4744 assert(T > 0); 4745 double csm_rate = ((double)(C + S + M)) / ((double)(T)); 4746 double m_rate = ((double)(M)) / ((double)(T)); 4747 4748 cpi->csm_rate_array[cpi->rate_index] = csm_rate; 4749 cpi->m_rate_array[cpi->rate_index] = m_rate; 4750 4751 cpi->rate_index = (cpi->rate_index + 1) % max_history_size; 4752 cpi->rate_size++; 4753 cpi->rate_size = AOMMIN(cpi->rate_size, max_history_size); 4754 4755 if (csm_rate < threshold_current) { 4756 return 0; 4757 } 4758 4759 if (C == T) { 4760 return 1; 4761 } 4762 4763 double csm_average = 0.0; 4764 double m_average = 0.0; 4765 4766 for (k = 0; k < cpi->rate_size; k++) { 4767 csm_average += cpi->csm_rate_array[k]; 4768 m_average += cpi->m_rate_array[k]; 4769 } 4770 csm_average /= cpi->rate_size; 4771 m_average /= cpi->rate_size; 4772 4773 if (csm_average < threshold_average) { 4774 return 0; 4775 } 4776 4777 if (M > (T - C - S) / 3) { 4778 return 1; 4779 } 4780 4781 if (csm_rate > 0.99 && m_rate > 0.01) { 4782 return 1; 4783 } 4784 4785 if (csm_average + m_average > 1.01) { 4786 return 1; 4787 } 4788 4789 return 0; 4790 } 4791 4792 // Refresh reference frame buffers according to refresh_frame_flags. 4793 static void refresh_reference_frames(AV1_COMP *cpi) { 4794 AV1_COMMON *const cm = &cpi->common; 4795 // All buffers are refreshed for shown keyframes and S-frames. 4796 4797 for (int ref_frame = 0; ref_frame < REF_FRAMES; ref_frame++) { 4798 if (((cm->current_frame.refresh_frame_flags >> ref_frame) & 1) == 1) { 4799 assign_frame_buffer_p(&cm->ref_frame_map[ref_frame], cm->cur_frame); 4800 } 4801 } 4802 } 4803 4804 static int encode_frame_to_data_rate(AV1_COMP *cpi, size_t *size, 4805 uint8_t *dest) { 4806 AV1_COMMON *const cm = &cpi->common; 4807 SequenceHeader *const seq_params = &cm->seq_params; 4808 CurrentFrame *const current_frame = &cm->current_frame; 4809 const AV1EncoderConfig *const oxcf = &cpi->oxcf; 4810 struct segmentation *const seg = &cm->seg; 4811 4812 #if CONFIG_COLLECT_COMPONENT_TIMING 4813 start_timing(cpi, encode_frame_to_data_rate_time); 4814 #endif 4815 4816 // frame type has been decided outside of this function call 4817 cm->cur_frame->frame_type = current_frame->frame_type; 4818 4819 cm->large_scale_tile = cpi->oxcf.large_scale_tile; 4820 cm->single_tile_decoding = cpi->oxcf.single_tile_decoding; 4821 4822 cm->allow_ref_frame_mvs &= frame_might_allow_ref_frame_mvs(cm); 4823 // cm->allow_ref_frame_mvs needs to be written into the frame header while 4824 // cm->large_scale_tile is 1, therefore, "cm->large_scale_tile=1" case is 4825 // separated from frame_might_allow_ref_frame_mvs(). 4826 cm->allow_ref_frame_mvs &= !cm->large_scale_tile; 4827 4828 cm->allow_warped_motion = 4829 cpi->oxcf.allow_warped_motion && frame_might_allow_warped_motion(cm); 4830 4831 cm->last_frame_type = current_frame->frame_type; 4832 if (cpi->oxcf.pass == 2 && cpi->sf.adaptive_interp_filter_search) 4833 cpi->sf.interp_filter_search_mask = setup_interp_filter_search_mask(cpi); 4834 4835 cpi->two_pass_partition_search = cpi->sf.two_pass_partition_search && 4836 !cpi->partition_search_skippable_frame; 4837 4838 if (encode_show_existing_frame(cm)) { 4839 restore_coding_context(cpi); 4840 4841 finalize_encoded_frame(cpi); 4842 // Build the bitstream 4843 int largest_tile_id = 0; // Output from bitstream: unused here 4844 if (av1_pack_bitstream(cpi, dest, size, &largest_tile_id) != AOM_CODEC_OK) 4845 return AOM_CODEC_ERROR; 4846 4847 if (seq_params->frame_id_numbers_present_flag && 4848 current_frame->frame_type == KEY_FRAME) { 4849 // Displaying a forward key-frame, so reset the ref buffer IDs 4850 int display_frame_id = cm->ref_frame_id[cpi->existing_fb_idx_to_show]; 4851 for (int i = 0; i < REF_FRAMES; i++) 4852 cm->ref_frame_id[i] = display_frame_id; 4853 } 4854 4855 cpi->seq_params_locked = 1; 4856 4857 #if DUMP_RECON_FRAMES == 1 4858 // NOTE(zoeliu): For debug - Output the filtered reconstructed video. 4859 dump_filtered_recon_frames(cpi); 4860 #endif // DUMP_RECON_FRAMES 4861 4862 // NOTE: Save the new show frame buffer index for --test-code=warn, i.e., 4863 // for the purpose to verify no mismatch between encoder and decoder. 4864 if (cm->show_frame) cpi->last_show_frame_buf = cm->cur_frame; 4865 4866 refresh_reference_frames(cpi); 4867 4868 // Since we allocate a spot for the OVERLAY frame in the gf group, we need 4869 // to do post-encoding update accordingly. 4870 if (cpi->rc.is_src_frame_alt_ref) { 4871 av1_set_target_rate(cpi, cm->width, cm->height); 4872 av1_rc_postencode_update(cpi, *size); 4873 } 4874 4875 ++current_frame->frame_number; 4876 4877 return AOM_CODEC_OK; 4878 } 4879 4880 // Work out whether to force_integer_mv this frame 4881 if (oxcf->pass != 1 && cpi->common.allow_screen_content_tools && 4882 !frame_is_intra_only(cm)) { 4883 if (cpi->common.seq_params.force_integer_mv == 2) { 4884 // Adaptive mode: see what previous frame encoded did 4885 if (cpi->unscaled_last_source != NULL) { 4886 cm->cur_frame_force_integer_mv = 4887 is_integer_mv(cpi, cpi->source, cpi->unscaled_last_source, 4888 cpi->previous_hash_table); 4889 } else { 4890 cpi->common.cur_frame_force_integer_mv = 0; 4891 } 4892 } else { 4893 cpi->common.cur_frame_force_integer_mv = 4894 cpi->common.seq_params.force_integer_mv; 4895 } 4896 } else { 4897 cpi->common.cur_frame_force_integer_mv = 0; 4898 } 4899 4900 // Set default state for segment based loop filter update flags. 4901 cm->lf.mode_ref_delta_update = 0; 4902 4903 // Set various flags etc to special state if it is a key frame. 4904 if (frame_is_intra_only(cm) || frame_is_sframe(cm)) { 4905 // Reset the loop filter deltas and segmentation map. 4906 av1_reset_segment_features(cm); 4907 4908 // If segmentation is enabled force a map update for key frames. 4909 if (seg->enabled) { 4910 seg->update_map = 1; 4911 seg->update_data = 1; 4912 } 4913 4914 // The alternate reference frame cannot be active for a key frame. 4915 cpi->rc.source_alt_ref_active = 0; 4916 } 4917 if (cpi->oxcf.mtu == 0) { 4918 cm->num_tg = cpi->oxcf.num_tile_groups; 4919 } else { 4920 // Use a default value for the purposes of weighting costs in probability 4921 // updates 4922 cm->num_tg = DEFAULT_MAX_NUM_TG; 4923 } 4924 4925 // For 1 pass CBR, check if we are dropping this frame. 4926 // Never drop on key frame. 4927 if (oxcf->pass == 0 && oxcf->rc_mode == AOM_CBR && 4928 current_frame->frame_type != KEY_FRAME) { 4929 if (av1_rc_drop_frame(cpi)) { 4930 av1_rc_postencode_update_drop_frame(cpi); 4931 release_scaled_references(cpi); 4932 return AOM_CODEC_OK; 4933 } 4934 } 4935 4936 aom_clear_system_state(); 4937 4938 #if CONFIG_INTERNAL_STATS 4939 memset(cpi->mode_chosen_counts, 0, 4940 MAX_MODES * sizeof(*cpi->mode_chosen_counts)); 4941 #endif 4942 4943 if (seq_params->frame_id_numbers_present_flag) { 4944 /* Non-normative definition of current_frame_id ("frame counter" with 4945 * wraparound) */ 4946 if (cm->current_frame_id == -1) { 4947 int lsb, msb; 4948 /* quasi-random initialization of current_frame_id for a key frame */ 4949 if (cpi->source->flags & YV12_FLAG_HIGHBITDEPTH) { 4950 lsb = CONVERT_TO_SHORTPTR(cpi->source->y_buffer)[0] & 0xff; 4951 msb = CONVERT_TO_SHORTPTR(cpi->source->y_buffer)[1] & 0xff; 4952 } else { 4953 lsb = cpi->source->y_buffer[0] & 0xff; 4954 msb = cpi->source->y_buffer[1] & 0xff; 4955 } 4956 cm->current_frame_id = 4957 ((msb << 8) + lsb) % (1 << seq_params->frame_id_length); 4958 4959 // S_frame is meant for stitching different streams of different 4960 // resolutions together, so current_frame_id must be the 4961 // same across different streams of the same content current_frame_id 4962 // should be the same and not random. 0x37 is a chosen number as start 4963 // point 4964 if (cpi->oxcf.sframe_enabled) cm->current_frame_id = 0x37; 4965 } else { 4966 cm->current_frame_id = 4967 (cm->current_frame_id + 1 + (1 << seq_params->frame_id_length)) % 4968 (1 << seq_params->frame_id_length); 4969 } 4970 } 4971 4972 switch (cpi->oxcf.cdf_update_mode) { 4973 case 0: // No CDF update for any frames(4~6% compression loss). 4974 cm->disable_cdf_update = 1; 4975 break; 4976 case 1: // Enable CDF update for all frames. 4977 cm->disable_cdf_update = 0; 4978 break; 4979 case 2: 4980 // Strategically determine at which frames to do CDF update. 4981 // Currently only enable CDF update for all-intra and no-show frames(1.5% 4982 // compression loss). 4983 // TODO(huisu (at) google.com): design schemes for various trade-offs between 4984 // compression quality and decoding speed. 4985 cm->disable_cdf_update = 4986 (frame_is_intra_only(cm) || !cm->show_frame) ? 0 : 1; 4987 break; 4988 } 4989 cm->timing_info_present &= !seq_params->reduced_still_picture_hdr; 4990 4991 #if CONFIG_COLLECT_COMPONENT_TIMING 4992 start_timing(cpi, encode_with_recode_loop_time); 4993 #endif 4994 if (encode_with_recode_loop(cpi, size, dest) != AOM_CODEC_OK) 4995 return AOM_CODEC_ERROR; 4996 #if CONFIG_COLLECT_COMPONENT_TIMING 4997 end_timing(cpi, encode_with_recode_loop_time); 4998 #endif 4999 5000 #ifdef OUTPUT_YUV_SKINMAP 5001 if (cpi->common.current_frame.frame_number > 1) { 5002 av1_compute_skin_map(cpi, yuv_skinmap_file); 5003 } 5004 #endif // OUTPUT_YUV_SKINMAP 5005 5006 // Special case code to reduce pulsing when key frames are forced at a 5007 // fixed interval. Note the reconstruction error if it is the frame before 5008 // the force key frame 5009 if (cpi->rc.next_key_frame_forced && cpi->rc.frames_to_key == 1) { 5010 if (seq_params->use_highbitdepth) { 5011 cpi->ambient_err = aom_highbd_get_y_sse(cpi->source, &cm->cur_frame->buf); 5012 } else { 5013 cpi->ambient_err = aom_get_y_sse(cpi->source, &cm->cur_frame->buf); 5014 } 5015 } 5016 5017 cm->cur_frame->buf.color_primaries = seq_params->color_primaries; 5018 cm->cur_frame->buf.transfer_characteristics = 5019 seq_params->transfer_characteristics; 5020 cm->cur_frame->buf.matrix_coefficients = seq_params->matrix_coefficients; 5021 cm->cur_frame->buf.monochrome = seq_params->monochrome; 5022 cm->cur_frame->buf.chroma_sample_position = 5023 seq_params->chroma_sample_position; 5024 cm->cur_frame->buf.color_range = seq_params->color_range; 5025 cm->cur_frame->buf.render_width = cm->render_width; 5026 cm->cur_frame->buf.render_height = cm->render_height; 5027 5028 // TODO(zoeliu): For non-ref frames, loop filtering may need to be turned 5029 // off. 5030 5031 // Pick the loop filter level for the frame. 5032 if (!cm->allow_intrabc) { 5033 loopfilter_frame(cpi, cm); 5034 } else { 5035 cm->lf.filter_level[0] = 0; 5036 cm->lf.filter_level[1] = 0; 5037 cm->cdef_info.cdef_bits = 0; 5038 cm->cdef_info.cdef_strengths[0] = 0; 5039 cm->cdef_info.nb_cdef_strengths = 1; 5040 cm->cdef_info.cdef_uv_strengths[0] = 0; 5041 cm->rst_info[0].frame_restoration_type = RESTORE_NONE; 5042 cm->rst_info[1].frame_restoration_type = RESTORE_NONE; 5043 cm->rst_info[2].frame_restoration_type = RESTORE_NONE; 5044 } 5045 5046 // TODO(debargha): Fix mv search range on encoder side 5047 // aom_extend_frame_inner_borders(&cm->cur_frame->buf, av1_num_planes(cm)); 5048 aom_extend_frame_borders(&cm->cur_frame->buf, av1_num_planes(cm)); 5049 5050 #ifdef OUTPUT_YUV_REC 5051 aom_write_one_yuv_frame(cm, &cm->cur_frame->buf); 5052 #endif 5053 5054 finalize_encoded_frame(cpi); 5055 // Build the bitstream 5056 int largest_tile_id = 0; // Output from pack_bitstream 5057 #if CONFIG_COLLECT_COMPONENT_TIMING 5058 start_timing(cpi, av1_pack_bitstream_final_time); 5059 #endif 5060 if (av1_pack_bitstream(cpi, dest, size, &largest_tile_id) != AOM_CODEC_OK) 5061 return AOM_CODEC_ERROR; 5062 #if CONFIG_COLLECT_COMPONENT_TIMING 5063 end_timing(cpi, av1_pack_bitstream_final_time); 5064 #endif 5065 5066 cpi->seq_params_locked = 1; 5067 5068 // Update reference frame ids for reference frames this frame will overwrite 5069 if (seq_params->frame_id_numbers_present_flag) { 5070 for (int i = 0; i < REF_FRAMES; i++) { 5071 if ((current_frame->refresh_frame_flags >> i) & 1) { 5072 cm->ref_frame_id[i] = cm->current_frame_id; 5073 } 5074 } 5075 } 5076 5077 #if DUMP_RECON_FRAMES == 1 5078 // NOTE(zoeliu): For debug - Output the filtered reconstructed video. 5079 dump_filtered_recon_frames(cpi); 5080 #endif // DUMP_RECON_FRAMES 5081 5082 if (cm->seg.enabled) { 5083 if (cm->seg.update_map) { 5084 update_reference_segmentation_map(cpi); 5085 } else if (cm->last_frame_seg_map) { 5086 memcpy(cm->cur_frame->seg_map, cm->last_frame_seg_map, 5087 cm->mi_cols * cm->mi_rows * sizeof(uint8_t)); 5088 } 5089 } 5090 5091 if (frame_is_intra_only(cm) == 0) { 5092 release_scaled_references(cpi); 5093 } 5094 5095 // NOTE: Save the new show frame buffer index for --test-code=warn, i.e., 5096 // for the purpose to verify no mismatch between encoder and decoder. 5097 if (cm->show_frame) cpi->last_show_frame_buf = cm->cur_frame; 5098 5099 refresh_reference_frames(cpi); 5100 5101 #if CONFIG_ENTROPY_STATS 5102 av1_accumulate_frame_counts(&aggregate_fc, &cpi->counts); 5103 #endif // CONFIG_ENTROPY_STATS 5104 5105 if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) { 5106 *cm->fc = cpi->tile_data[largest_tile_id].tctx; 5107 av1_reset_cdf_symbol_counters(cm->fc); 5108 } 5109 if (!cm->large_scale_tile) { 5110 cm->cur_frame->frame_context = *cm->fc; 5111 } 5112 #define EXT_TILE_DEBUG 0 5113 #if EXT_TILE_DEBUG 5114 if (cm->large_scale_tile && oxcf->pass == 2) { 5115 char fn[20] = "./fc"; 5116 fn[4] = current_frame->frame_number / 100 + '0'; 5117 fn[5] = (current_frame->frame_number % 100) / 10 + '0'; 5118 fn[6] = (current_frame->frame_number % 10) + '0'; 5119 fn[7] = '\0'; 5120 av1_print_frame_contexts(cm->fc, fn); 5121 } 5122 #endif // EXT_TILE_DEBUG 5123 #undef EXT_TILE_DEBUG 5124 5125 #if CONFIG_COLLECT_COMPONENT_TIMING 5126 end_timing(cpi, encode_frame_to_data_rate_time); 5127 5128 // Print out timing information. 5129 int i; 5130 fprintf(stderr, "\n Frame number: %d, Frame type: %s, Show Frame: %d\n", 5131 cm->current_frame.frame_number, 5132 get_frame_type_enum(cm->current_frame.frame_type), cm->show_frame); 5133 for (i = 0; i < kTimingComponents; i++) { 5134 cpi->component_time[i] += cpi->frame_component_time[i]; 5135 fprintf(stderr, " %s: %" PRId64 " us (total: %" PRId64 " us)\n", 5136 get_component_name(i), cpi->frame_component_time[i], 5137 cpi->component_time[i]); 5138 cpi->frame_component_time[i] = 0; 5139 } 5140 #endif 5141 5142 cm->last_frame_type = current_frame->frame_type; 5143 5144 av1_rc_postencode_update(cpi, *size); 5145 5146 // Store encoded frame's hash table for is_integer_mv() next time 5147 if (oxcf->pass != 1 && cpi->common.allow_screen_content_tools) { 5148 cpi->previous_hash_table = &cm->cur_frame->hash_table; 5149 } 5150 5151 // Clear the one shot update flags for segmentation map and mode/ref loop 5152 // filter deltas. 5153 cm->seg.update_map = 0; 5154 cm->seg.update_data = 0; 5155 cm->lf.mode_ref_delta_update = 0; 5156 5157 // A droppable frame might not be shown but it always 5158 // takes a space in the gf group. Therefore, even when 5159 // it is not shown, we still need update the count down. 5160 5161 if (cm->show_frame) { 5162 // TODO(zoeliu): We may only swamp mi and prev_mi for those frames that 5163 // are 5164 // being used as reference. 5165 swap_mi_and_prev_mi(cm); 5166 // Don't increment frame counters if this was an altref buffer 5167 // update not a real frame 5168 5169 ++current_frame->frame_number; 5170 } 5171 5172 return AOM_CODEC_OK; 5173 } 5174 5175 int av1_encode(AV1_COMP *const cpi, uint8_t *const dest, 5176 const EncodeFrameInput *const frame_input, 5177 const EncodeFrameParams *const frame_params, 5178 EncodeFrameResults *const frame_results) { 5179 AV1_COMMON *const cm = &cpi->common; 5180 CurrentFrame *const current_frame = &cm->current_frame; 5181 5182 cpi->unscaled_source = frame_input->source; 5183 cpi->source = frame_input->source; 5184 cpi->unscaled_last_source = frame_input->last_source; 5185 5186 current_frame->refresh_frame_flags = frame_params->refresh_frame_flags; 5187 cm->error_resilient_mode = frame_params->error_resilient_mode; 5188 cm->primary_ref_frame = frame_params->primary_ref_frame; 5189 cm->current_frame.frame_type = frame_params->frame_type; 5190 cm->show_frame = frame_params->show_frame; 5191 cpi->ref_frame_flags = frame_params->ref_frame_flags; 5192 cpi->speed = frame_params->speed; 5193 cm->show_existing_frame = frame_params->show_existing_frame; 5194 cpi->existing_fb_idx_to_show = frame_params->existing_fb_idx_to_show; 5195 5196 memcpy(cm->remapped_ref_idx, frame_params->remapped_ref_idx, 5197 REF_FRAMES * sizeof(*cm->remapped_ref_idx)); 5198 5199 cpi->refresh_last_frame = frame_params->refresh_last_frame; 5200 cpi->refresh_golden_frame = frame_params->refresh_golden_frame; 5201 cpi->refresh_bwd_ref_frame = frame_params->refresh_bwd_ref_frame; 5202 cpi->refresh_alt2_ref_frame = frame_params->refresh_alt2_ref_frame; 5203 cpi->refresh_alt_ref_frame = frame_params->refresh_alt_ref_frame; 5204 5205 if (current_frame->frame_type == KEY_FRAME && cm->show_frame) 5206 current_frame->frame_number = 0; 5207 5208 if (cm->show_existing_frame) { 5209 current_frame->order_hint = cm->cur_frame->order_hint; 5210 } else { 5211 current_frame->order_hint = 5212 current_frame->frame_number + frame_params->order_offset; 5213 current_frame->order_hint %= 5214 (1 << (cm->seq_params.order_hint_info.order_hint_bits_minus_1 + 1)); 5215 } 5216 5217 if (cpi->oxcf.pass == 1) { 5218 av1_first_pass(cpi, frame_input->ts_duration); 5219 } else if (cpi->oxcf.pass == 0 || cpi->oxcf.pass == 2) { 5220 if (encode_frame_to_data_rate(cpi, &frame_results->size, dest) != 5221 AOM_CODEC_OK) { 5222 return AOM_CODEC_ERROR; 5223 } 5224 } else { 5225 return AOM_CODEC_ERROR; 5226 } 5227 5228 return AOM_CODEC_OK; 5229 } 5230 5231 #if CONFIG_DENOISE 5232 static int apply_denoise_2d(AV1_COMP *cpi, YV12_BUFFER_CONFIG *sd, 5233 int block_size, float noise_level, 5234 int64_t time_stamp, int64_t end_time) { 5235 AV1_COMMON *const cm = &cpi->common; 5236 if (!cpi->denoise_and_model) { 5237 cpi->denoise_and_model = aom_denoise_and_model_alloc( 5238 cm->seq_params.bit_depth, block_size, noise_level); 5239 if (!cpi->denoise_and_model) { 5240 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, 5241 "Error allocating denoise and model"); 5242 return -1; 5243 } 5244 } 5245 if (!cpi->film_grain_table) { 5246 cpi->film_grain_table = aom_malloc(sizeof(*cpi->film_grain_table)); 5247 if (!cpi->film_grain_table) { 5248 aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, 5249 "Error allocating grain table"); 5250 return -1; 5251 } 5252 memset(cpi->film_grain_table, 0, sizeof(*cpi->film_grain_table)); 5253 } 5254 if (aom_denoise_and_model_run(cpi->denoise_and_model, sd, 5255 &cm->film_grain_params)) { 5256 if (cm->film_grain_params.apply_grain) { 5257 aom_film_grain_table_append(cpi->film_grain_table, time_stamp, end_time, 5258 &cm->film_grain_params); 5259 } 5260 } 5261 return 0; 5262 } 5263 #endif 5264 5265 int av1_receive_raw_frame(AV1_COMP *cpi, aom_enc_frame_flags_t frame_flags, 5266 YV12_BUFFER_CONFIG *sd, int64_t time_stamp, 5267 int64_t end_time) { 5268 AV1_COMMON *const cm = &cpi->common; 5269 const SequenceHeader *const seq_params = &cm->seq_params; 5270 int res = 0; 5271 const int subsampling_x = sd->subsampling_x; 5272 const int subsampling_y = sd->subsampling_y; 5273 const int use_highbitdepth = (sd->flags & YV12_FLAG_HIGHBITDEPTH) != 0; 5274 5275 check_initial_width(cpi, use_highbitdepth, subsampling_x, subsampling_y); 5276 5277 #if CONFIG_INTERNAL_STATS 5278 struct aom_usec_timer timer; 5279 aom_usec_timer_start(&timer); 5280 #endif 5281 #if CONFIG_DENOISE 5282 if (cpi->oxcf.noise_level > 0) 5283 if (apply_denoise_2d(cpi, sd, cpi->oxcf.noise_block_size, 5284 cpi->oxcf.noise_level, time_stamp, end_time) < 0) 5285 res = -1; 5286 #endif // CONFIG_DENOISE 5287 5288 if (av1_lookahead_push(cpi->lookahead, sd, time_stamp, end_time, 5289 use_highbitdepth, frame_flags)) 5290 res = -1; 5291 #if CONFIG_INTERNAL_STATS 5292 aom_usec_timer_mark(&timer); 5293 cpi->time_receive_data += aom_usec_timer_elapsed(&timer); 5294 #endif 5295 if ((seq_params->profile == PROFILE_0) && !seq_params->monochrome && 5296 (subsampling_x != 1 || subsampling_y != 1)) { 5297 aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM, 5298 "Non-4:2:0 color format requires profile 1 or 2"); 5299 res = -1; 5300 } 5301 if ((seq_params->profile == PROFILE_1) && 5302 !(subsampling_x == 0 && subsampling_y == 0)) { 5303 aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM, 5304 "Profile 1 requires 4:4:4 color format"); 5305 res = -1; 5306 } 5307 if ((seq_params->profile == PROFILE_2) && 5308 (seq_params->bit_depth <= AOM_BITS_10) && 5309 !(subsampling_x == 1 && subsampling_y == 0)) { 5310 aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM, 5311 "Profile 2 bit-depth < 10 requires 4:2:2 color format"); 5312 res = -1; 5313 } 5314 5315 return res; 5316 } 5317 5318 #if CONFIG_INTERNAL_STATS 5319 extern double av1_get_blockiness(const unsigned char *img1, int img1_pitch, 5320 const unsigned char *img2, int img2_pitch, 5321 int width, int height); 5322 5323 static void adjust_image_stat(double y, double u, double v, double all, 5324 ImageStat *s) { 5325 s->stat[STAT_Y] += y; 5326 s->stat[STAT_U] += u; 5327 s->stat[STAT_V] += v; 5328 s->stat[STAT_ALL] += all; 5329 s->worst = AOMMIN(s->worst, all); 5330 } 5331 5332 static void compute_internal_stats(AV1_COMP *cpi, int frame_bytes) { 5333 AV1_COMMON *const cm = &cpi->common; 5334 double samples = 0.0; 5335 uint32_t in_bit_depth = 8; 5336 uint32_t bit_depth = 8; 5337 5338 #if CONFIG_INTER_STATS_ONLY 5339 if (cm->current_frame.frame_type == KEY_FRAME) return; // skip key frame 5340 #endif 5341 cpi->bytes += frame_bytes; 5342 5343 if (cm->seq_params.use_highbitdepth) { 5344 in_bit_depth = cpi->oxcf.input_bit_depth; 5345 bit_depth = cm->seq_params.bit_depth; 5346 } 5347 if (cm->show_frame) { 5348 const YV12_BUFFER_CONFIG *orig = cpi->source; 5349 const YV12_BUFFER_CONFIG *recon = &cpi->common.cur_frame->buf; 5350 double y, u, v, frame_all; 5351 5352 cpi->count++; 5353 if (cpi->b_calculate_psnr) { 5354 PSNR_STATS psnr; 5355 double frame_ssim2 = 0.0, weight = 0.0; 5356 aom_clear_system_state(); 5357 // TODO(yaowu): unify these two versions into one. 5358 aom_calc_highbd_psnr(orig, recon, &psnr, bit_depth, in_bit_depth); 5359 5360 adjust_image_stat(psnr.psnr[1], psnr.psnr[2], psnr.psnr[3], psnr.psnr[0], 5361 &cpi->psnr); 5362 cpi->total_sq_error += psnr.sse[0]; 5363 cpi->total_samples += psnr.samples[0]; 5364 samples = psnr.samples[0]; 5365 // TODO(yaowu): unify these two versions into one. 5366 if (cm->seq_params.use_highbitdepth) 5367 frame_ssim2 = 5368 aom_highbd_calc_ssim(orig, recon, &weight, bit_depth, in_bit_depth); 5369 else 5370 frame_ssim2 = aom_calc_ssim(orig, recon, &weight); 5371 5372 cpi->worst_ssim = AOMMIN(cpi->worst_ssim, frame_ssim2); 5373 cpi->summed_quality += frame_ssim2 * weight; 5374 cpi->summed_weights += weight; 5375 5376 #if 0 5377 { 5378 FILE *f = fopen("q_used.stt", "a"); 5379 double y2 = psnr.psnr[1]; 5380 double u2 = psnr.psnr[2]; 5381 double v2 = psnr.psnr[3]; 5382 double frame_psnr2 = psnr.psnr[0]; 5383 fprintf(f, "%5d : Y%f7.3:U%f7.3:V%f7.3:F%f7.3:S%7.3f\n", 5384 cm->current_frame.frame_number, y2, u2, v2, 5385 frame_psnr2, frame_ssim2); 5386 fclose(f); 5387 } 5388 #endif 5389 } 5390 if (cpi->b_calculate_blockiness) { 5391 if (!cm->seq_params.use_highbitdepth) { 5392 const double frame_blockiness = 5393 av1_get_blockiness(orig->y_buffer, orig->y_stride, recon->y_buffer, 5394 recon->y_stride, orig->y_width, orig->y_height); 5395 cpi->worst_blockiness = AOMMAX(cpi->worst_blockiness, frame_blockiness); 5396 cpi->total_blockiness += frame_blockiness; 5397 } 5398 5399 if (cpi->b_calculate_consistency) { 5400 if (!cm->seq_params.use_highbitdepth) { 5401 const double this_inconsistency = aom_get_ssim_metrics( 5402 orig->y_buffer, orig->y_stride, recon->y_buffer, recon->y_stride, 5403 orig->y_width, orig->y_height, cpi->ssim_vars, &cpi->metrics, 1); 5404 5405 const double peak = (double)((1 << in_bit_depth) - 1); 5406 const double consistency = 5407 aom_sse_to_psnr(samples, peak, cpi->total_inconsistency); 5408 if (consistency > 0.0) 5409 cpi->worst_consistency = 5410 AOMMIN(cpi->worst_consistency, consistency); 5411 cpi->total_inconsistency += this_inconsistency; 5412 } 5413 } 5414 } 5415 5416 frame_all = 5417 aom_calc_fastssim(orig, recon, &y, &u, &v, bit_depth, in_bit_depth); 5418 adjust_image_stat(y, u, v, frame_all, &cpi->fastssim); 5419 frame_all = aom_psnrhvs(orig, recon, &y, &u, &v, bit_depth, in_bit_depth); 5420 adjust_image_stat(y, u, v, frame_all, &cpi->psnrhvs); 5421 } 5422 } 5423 #endif // CONFIG_INTERNAL_STATS 5424 int av1_get_compressed_data(AV1_COMP *cpi, unsigned int *frame_flags, 5425 size_t *size, uint8_t *dest, int64_t *time_stamp, 5426 int64_t *time_end, int flush, 5427 const aom_rational_t *timebase) { 5428 const AV1EncoderConfig *const oxcf = &cpi->oxcf; 5429 AV1_COMMON *const cm = &cpi->common; 5430 5431 #if CONFIG_BITSTREAM_DEBUG 5432 assert(cpi->oxcf.max_threads == 0 && 5433 "bitstream debug tool does not support multithreading"); 5434 bitstream_queue_record_write(); 5435 bitstream_queue_set_frame_write(cm->current_frame.frame_number * 2 + 5436 cm->show_frame); 5437 #endif 5438 5439 // Indicates whether or not to use an adaptive quantize b rather than 5440 // the traditional version 5441 cm->use_quant_b_adapt = cpi->oxcf.quant_b_adapt; 5442 5443 cm->showable_frame = 0; 5444 *size = 0; 5445 #if CONFIG_INTERNAL_STATS 5446 struct aom_usec_timer cmptimer; 5447 aom_usec_timer_start(&cmptimer); 5448 #endif 5449 set_high_precision_mv(cpi, ALTREF_HIGH_PRECISION_MV, 0); 5450 5451 // Normal defaults 5452 cm->refresh_frame_context = oxcf->frame_parallel_decoding_mode 5453 ? REFRESH_FRAME_CONTEXT_DISABLED 5454 : REFRESH_FRAME_CONTEXT_BACKWARD; 5455 if (oxcf->large_scale_tile) 5456 cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED; 5457 5458 // Initialize fields related to forward keyframes 5459 cpi->no_show_kf = 0; 5460 5461 if (assign_cur_frame_new_fb(cm) == NULL) return AOM_CODEC_ERROR; 5462 5463 const int result = av1_encode_strategy(cpi, size, dest, frame_flags, 5464 time_stamp, time_end, timebase, flush); 5465 if (result != AOM_CODEC_OK && result != -1) { 5466 return AOM_CODEC_ERROR; 5467 } else if (result == -1) { 5468 // Returning -1 indicates no frame encoded; more input is required 5469 return -1; 5470 } 5471 #if CONFIG_INTERNAL_STATS 5472 aom_usec_timer_mark(&cmptimer); 5473 cpi->time_compress_data += aom_usec_timer_elapsed(&cmptimer); 5474 #endif 5475 if (cpi->b_calculate_psnr) { 5476 if (cm->show_existing_frame || (oxcf->pass != 1 && cm->show_frame)) { 5477 generate_psnr_packet(cpi); 5478 } 5479 } 5480 if (cpi->keep_level_stats && oxcf->pass != 1) 5481 av1_update_level_info(cpi, *size, *time_stamp, *time_end); 5482 5483 #if CONFIG_INTERNAL_STATS 5484 if (oxcf->pass != 1) { 5485 compute_internal_stats(cpi, (int)(*size)); 5486 } 5487 #endif // CONFIG_INTERNAL_STATS 5488 #if CONFIG_SPEED_STATS 5489 if (cpi->oxcf.pass != 1 && !cm->show_existing_frame) { 5490 cpi->tx_search_count += cpi->td.mb.tx_search_count; 5491 cpi->td.mb.tx_search_count = 0; 5492 } 5493 #endif // CONFIG_SPEED_STATS 5494 5495 aom_clear_system_state(); 5496 5497 return 0; 5498 } 5499 5500 int av1_get_preview_raw_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *dest) { 5501 AV1_COMMON *cm = &cpi->common; 5502 if (!cm->show_frame) { 5503 return -1; 5504 } else { 5505 int ret; 5506 if (cm->cur_frame != NULL) { 5507 *dest = cm->cur_frame->buf; 5508 dest->y_width = cm->width; 5509 dest->y_height = cm->height; 5510 dest->uv_width = cm->width >> cm->seq_params.subsampling_x; 5511 dest->uv_height = cm->height >> cm->seq_params.subsampling_y; 5512 ret = 0; 5513 } else { 5514 ret = -1; 5515 } 5516 aom_clear_system_state(); 5517 return ret; 5518 } 5519 } 5520 5521 int av1_get_last_show_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *frame) { 5522 if (cpi->last_show_frame_buf == NULL) return -1; 5523 5524 *frame = cpi->last_show_frame_buf->buf; 5525 return 0; 5526 } 5527 5528 static int equal_dimensions_and_border(const YV12_BUFFER_CONFIG *a, 5529 const YV12_BUFFER_CONFIG *b) { 5530 return a->y_height == b->y_height && a->y_width == b->y_width && 5531 a->uv_height == b->uv_height && a->uv_width == b->uv_width && 5532 a->y_stride == b->y_stride && a->uv_stride == b->uv_stride && 5533 a->border == b->border && 5534 (a->flags & YV12_FLAG_HIGHBITDEPTH) == 5535 (b->flags & YV12_FLAG_HIGHBITDEPTH); 5536 } 5537 5538 aom_codec_err_t av1_copy_new_frame_enc(AV1_COMMON *cm, 5539 YV12_BUFFER_CONFIG *new_frame, 5540 YV12_BUFFER_CONFIG *sd) { 5541 const int num_planes = av1_num_planes(cm); 5542 if (!equal_dimensions_and_border(new_frame, sd)) 5543 aom_internal_error(&cm->error, AOM_CODEC_ERROR, 5544 "Incorrect buffer dimensions"); 5545 else 5546 aom_yv12_copy_frame(new_frame, sd, num_planes); 5547 5548 return cm->error.error_code; 5549 } 5550 5551 int av1_set_internal_size(AV1_COMP *cpi, AOM_SCALING horiz_mode, 5552 AOM_SCALING vert_mode) { 5553 int hr = 0, hs = 0, vr = 0, vs = 0; 5554 5555 if (horiz_mode > ONETWO || vert_mode > ONETWO) return -1; 5556 5557 Scale2Ratio(horiz_mode, &hr, &hs); 5558 Scale2Ratio(vert_mode, &vr, &vs); 5559 5560 // always go to the next whole number 5561 cpi->resize_pending_width = (hs - 1 + cpi->oxcf.width * hr) / hs; 5562 cpi->resize_pending_height = (vs - 1 + cpi->oxcf.height * vr) / vs; 5563 5564 return 0; 5565 } 5566 5567 int av1_get_quantizer(AV1_COMP *cpi) { return cpi->common.base_qindex; } 5568 5569 int av1_convert_sect5obus_to_annexb(uint8_t *buffer, size_t *frame_size) { 5570 size_t output_size = 0; 5571 size_t total_bytes_read = 0; 5572 size_t remaining_size = *frame_size; 5573 uint8_t *buff_ptr = buffer; 5574 5575 // go through each OBUs 5576 while (total_bytes_read < *frame_size) { 5577 uint8_t saved_obu_header[2]; 5578 uint64_t obu_payload_size; 5579 size_t length_of_payload_size; 5580 size_t length_of_obu_size; 5581 uint32_t obu_header_size = (buff_ptr[0] >> 2) & 0x1 ? 2 : 1; 5582 size_t obu_bytes_read = obu_header_size; // bytes read for current obu 5583 5584 // save the obu header (1 or 2 bytes) 5585 memmove(saved_obu_header, buff_ptr, obu_header_size); 5586 // clear the obu_has_size_field 5587 saved_obu_header[0] = saved_obu_header[0] & (~0x2); 5588 5589 // get the payload_size and length of payload_size 5590 if (aom_uleb_decode(buff_ptr + obu_header_size, remaining_size, 5591 &obu_payload_size, &length_of_payload_size) != 0) { 5592 return AOM_CODEC_ERROR; 5593 } 5594 obu_bytes_read += length_of_payload_size; 5595 5596 // calculate the length of size of the obu header plus payload 5597 length_of_obu_size = 5598 aom_uleb_size_in_bytes((uint64_t)(obu_header_size + obu_payload_size)); 5599 5600 // move the rest of data to new location 5601 memmove(buff_ptr + length_of_obu_size + obu_header_size, 5602 buff_ptr + obu_bytes_read, remaining_size - obu_bytes_read); 5603 obu_bytes_read += (size_t)obu_payload_size; 5604 5605 // write the new obu size 5606 const uint64_t obu_size = obu_header_size + obu_payload_size; 5607 size_t coded_obu_size; 5608 if (aom_uleb_encode(obu_size, sizeof(obu_size), buff_ptr, 5609 &coded_obu_size) != 0) { 5610 return AOM_CODEC_ERROR; 5611 } 5612 5613 // write the saved (modified) obu_header following obu size 5614 memmove(buff_ptr + length_of_obu_size, saved_obu_header, obu_header_size); 5615 5616 total_bytes_read += obu_bytes_read; 5617 remaining_size -= obu_bytes_read; 5618 buff_ptr += length_of_obu_size + obu_size; 5619 output_size += length_of_obu_size + (size_t)obu_size; 5620 } 5621 5622 *frame_size = output_size; 5623 return AOM_CODEC_OK; 5624 } 5625 5626 void av1_apply_encoding_flags(AV1_COMP *cpi, aom_enc_frame_flags_t flags) { 5627 // TODO(yunqingwang): For what references to use, external encoding flags 5628 // should be consistent with internal reference frame selection. Need to 5629 // ensure that there is not conflict between the two. In AV1 encoder, the 5630 // priority rank for 7 reference frames are: LAST, ALTREF, LAST2, LAST3, 5631 // GOLDEN, BWDREF, ALTREF2. If only one reference frame is used, it must be 5632 // LAST. 5633 cpi->ext_ref_frame_flags = AOM_REFFRAME_ALL; 5634 if (flags & 5635 (AOM_EFLAG_NO_REF_LAST | AOM_EFLAG_NO_REF_LAST2 | AOM_EFLAG_NO_REF_LAST3 | 5636 AOM_EFLAG_NO_REF_GF | AOM_EFLAG_NO_REF_ARF | AOM_EFLAG_NO_REF_BWD | 5637 AOM_EFLAG_NO_REF_ARF2)) { 5638 if (flags & AOM_EFLAG_NO_REF_LAST) { 5639 cpi->ext_ref_frame_flags = 0; 5640 } else { 5641 int ref = AOM_REFFRAME_ALL; 5642 5643 if (flags & AOM_EFLAG_NO_REF_LAST2) ref ^= AOM_LAST2_FLAG; 5644 if (flags & AOM_EFLAG_NO_REF_LAST3) ref ^= AOM_LAST3_FLAG; 5645 5646 if (flags & AOM_EFLAG_NO_REF_GF) ref ^= AOM_GOLD_FLAG; 5647 5648 if (flags & AOM_EFLAG_NO_REF_ARF) { 5649 ref ^= AOM_ALT_FLAG; 5650 ref ^= AOM_BWD_FLAG; 5651 ref ^= AOM_ALT2_FLAG; 5652 } else { 5653 if (flags & AOM_EFLAG_NO_REF_BWD) ref ^= AOM_BWD_FLAG; 5654 if (flags & AOM_EFLAG_NO_REF_ARF2) ref ^= AOM_ALT2_FLAG; 5655 } 5656 5657 av1_use_as_reference(cpi, ref); 5658 } 5659 } 5660 5661 if (flags & 5662 (AOM_EFLAG_NO_UPD_LAST | AOM_EFLAG_NO_UPD_GF | AOM_EFLAG_NO_UPD_ARF)) { 5663 int upd = AOM_REFFRAME_ALL; 5664 5665 // Refreshing LAST/LAST2/LAST3 is handled by 1 common flag. 5666 if (flags & AOM_EFLAG_NO_UPD_LAST) upd ^= AOM_LAST_FLAG; 5667 5668 if (flags & AOM_EFLAG_NO_UPD_GF) upd ^= AOM_GOLD_FLAG; 5669 5670 if (flags & AOM_EFLAG_NO_UPD_ARF) { 5671 upd ^= AOM_ALT_FLAG; 5672 upd ^= AOM_BWD_FLAG; 5673 upd ^= AOM_ALT2_FLAG; 5674 } 5675 5676 cpi->ext_refresh_last_frame = (upd & AOM_LAST_FLAG) != 0; 5677 cpi->ext_refresh_golden_frame = (upd & AOM_GOLD_FLAG) != 0; 5678 cpi->ext_refresh_alt_ref_frame = (upd & AOM_ALT_FLAG) != 0; 5679 cpi->ext_refresh_bwd_ref_frame = (upd & AOM_BWD_FLAG) != 0; 5680 cpi->ext_refresh_alt2_ref_frame = (upd & AOM_ALT2_FLAG) != 0; 5681 cpi->ext_refresh_frame_flags_pending = 1; 5682 } else { 5683 cpi->ext_refresh_frame_flags_pending = 0; 5684 } 5685 5686 cpi->ext_use_ref_frame_mvs = cpi->oxcf.allow_ref_frame_mvs & 5687 ((flags & AOM_EFLAG_NO_REF_FRAME_MVS) == 0); 5688 cpi->ext_use_error_resilient = cpi->oxcf.error_resilient_mode | 5689 ((flags & AOM_EFLAG_ERROR_RESILIENT) != 0); 5690 cpi->ext_use_s_frame = 5691 cpi->oxcf.s_frame_mode | ((flags & AOM_EFLAG_SET_S_FRAME) != 0); 5692 cpi->ext_use_primary_ref_none = (flags & AOM_EFLAG_SET_PRIMARY_REF_NONE) != 0; 5693 5694 if (flags & AOM_EFLAG_NO_UPD_ENTROPY) { 5695 av1_update_entropy(cpi, 0); 5696 } 5697 } 5698 5699 aom_fixed_buf_t *av1_get_global_headers(AV1_COMP *cpi) { 5700 if (!cpi) return NULL; 5701 5702 uint8_t header_buf[512] = { 0 }; 5703 const uint32_t sequence_header_size = 5704 write_sequence_header_obu(cpi, &header_buf[0]); 5705 assert(sequence_header_size <= sizeof(header_buf)); 5706 if (sequence_header_size == 0) return NULL; 5707 5708 const size_t obu_header_size = 1; 5709 const size_t size_field_size = aom_uleb_size_in_bytes(sequence_header_size); 5710 const size_t payload_offset = obu_header_size + size_field_size; 5711 5712 if (payload_offset + sequence_header_size > sizeof(header_buf)) return NULL; 5713 memmove(&header_buf[payload_offset], &header_buf[0], sequence_header_size); 5714 5715 if (av1_write_obu_header(cpi, OBU_SEQUENCE_HEADER, 0, &header_buf[0]) != 5716 obu_header_size) { 5717 return NULL; 5718 } 5719 5720 size_t coded_size_field_size = 0; 5721 if (aom_uleb_encode(sequence_header_size, size_field_size, 5722 &header_buf[obu_header_size], 5723 &coded_size_field_size) != 0) { 5724 return NULL; 5725 } 5726 assert(coded_size_field_size == size_field_size); 5727 5728 aom_fixed_buf_t *global_headers = 5729 (aom_fixed_buf_t *)malloc(sizeof(*global_headers)); 5730 if (!global_headers) return NULL; 5731 5732 const size_t global_header_buf_size = 5733 obu_header_size + size_field_size + sequence_header_size; 5734 5735 global_headers->buf = malloc(global_header_buf_size); 5736 if (!global_headers->buf) { 5737 free(global_headers); 5738 return NULL; 5739 } 5740 5741 memcpy(global_headers->buf, &header_buf[0], global_header_buf_size); 5742 global_headers->sz = global_header_buf_size; 5743 return global_headers; 5744 } 5745