/* * Copyright (c) 2016, Alliance for Open Media. All rights reserved * * This source code is subject to the terms of the BSD 2 Clause License and * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License * was not distributed with this source code in the LICENSE file, you can * obtain it at www.aomedia.org/license/software. If the Alliance for Open * Media Patent License 1.0 was not distributed with this source code in the * PATENTS file, you can obtain it at www.aomedia.org/license/patent. */ #include #include #include #include "config/aom_config.h" #include "config/aom_dsp_rtcd.h" #include "config/aom_scale_rtcd.h" #include "config/av1_rtcd.h" #include "aom_dsp/aom_dsp_common.h" #include "aom_dsp/aom_filter.h" #if CONFIG_DENOISE #include "aom_dsp/grain_table.h" #include "aom_dsp/noise_util.h" #include "aom_dsp/noise_model.h" #endif #include "aom_dsp/psnr.h" #if CONFIG_INTERNAL_STATS #include "aom_dsp/ssim.h" #endif #include "aom_ports/aom_timer.h" #include "aom_ports/mem.h" #include "aom_ports/system_state.h" #include "aom_scale/aom_scale.h" #if CONFIG_BITSTREAM_DEBUG #include "aom_util/debug_util.h" #endif // CONFIG_BITSTREAM_DEBUG #include "av1/common/alloccommon.h" #include "av1/common/cdef.h" #include "av1/common/filter.h" #include "av1/common/idct.h" #include "av1/common/reconinter.h" #include "av1/common/reconintra.h" #include "av1/common/resize.h" #include "av1/common/tile_common.h" #include "av1/encoder/av1_multi_thread.h" #include "av1/encoder/aq_complexity.h" #include "av1/encoder/aq_cyclicrefresh.h" #include "av1/encoder/aq_variance.h" #include "av1/encoder/bitstream.h" #include "av1/encoder/context_tree.h" #include "av1/encoder/encodeframe.h" #include "av1/encoder/encodemv.h" #include "av1/encoder/encode_strategy.h" #include "av1/encoder/encoder.h" #include "av1/encoder/encodetxb.h" #include "av1/encoder/ethread.h" #include "av1/encoder/firstpass.h" #include "av1/encoder/grain_test_vectors.h" #include "av1/encoder/hash_motion.h" #include "av1/encoder/mbgraph.h" #include "av1/encoder/picklpf.h" #include "av1/encoder/pickrst.h" #include "av1/encoder/random.h" #include "av1/encoder/ratectrl.h" #include "av1/encoder/rd.h" #include "av1/encoder/rdopt.h" #include "av1/encoder/segmentation.h" #include "av1/encoder/speed_features.h" #include "av1/encoder/reconinter_enc.h" #define DEFAULT_EXPLICIT_ORDER_HINT_BITS 7 #if CONFIG_ENTROPY_STATS FRAME_COUNTS aggregate_fc; #endif // CONFIG_ENTROPY_STATS #define AM_SEGMENT_ID_INACTIVE 7 #define AM_SEGMENT_ID_ACTIVE 0 // Whether to use high precision mv for altref computation. #define ALTREF_HIGH_PRECISION_MV 1 // Q threshold for high precision mv. Choose a very high value for now so that // HIGH_PRECISION is always chosen. #define HIGH_PRECISION_MV_QTHRESH 200 // #define OUTPUT_YUV_REC #ifdef OUTPUT_YUV_SKINMAP FILE *yuv_skinmap_file = NULL; #endif #ifdef OUTPUT_YUV_REC FILE *yuv_rec_file; #define FILE_NAME_LEN 100 #endif // Estimate if the source frame is screen content, based on the portion of // blocks that have no more than 4 (experimentally selected) luma colors. static int is_screen_content(const uint8_t *src, int use_hbd, int bd, int stride, int width, int height) { assert(src != NULL); int counts = 0; const int blk_w = 16; const int blk_h = 16; const int limit = 4; for (int r = 0; r + blk_h <= height; r += blk_h) { for (int c = 0; c + blk_w <= width; c += blk_w) { int count_buf[1 << 12]; // Maximum (1 << 12) color levels. const int n_colors = use_hbd ? av1_count_colors_highbd(src + r * stride + c, stride, blk_w, blk_h, bd, count_buf) : av1_count_colors(src + r * stride + c, stride, blk_w, blk_h, count_buf); if (n_colors > 1 && n_colors <= limit) counts++; } } // The threshold is 10%. return counts * blk_h * blk_w * 10 > width * height; } static INLINE void Scale2Ratio(AOM_SCALING mode, int *hr, int *hs) { switch (mode) { case NORMAL: *hr = 1; *hs = 1; break; case FOURFIVE: *hr = 4; *hs = 5; break; case THREEFIVE: *hr = 3; *hs = 5; break; case ONETWO: *hr = 1; *hs = 2; break; default: *hr = 1; *hs = 1; assert(0); break; } } // Mark all inactive blocks as active. Other segmentation features may be set // so memset cannot be used, instead only inactive blocks should be reset. static void suppress_active_map(AV1_COMP *cpi) { unsigned char *const seg_map = cpi->segmentation_map; int i; if (cpi->active_map.enabled || cpi->active_map.update) for (i = 0; i < cpi->common.mi_rows * cpi->common.mi_cols; ++i) if (seg_map[i] == AM_SEGMENT_ID_INACTIVE) seg_map[i] = AM_SEGMENT_ID_ACTIVE; } static void apply_active_map(AV1_COMP *cpi) { struct segmentation *const seg = &cpi->common.seg; unsigned char *const seg_map = cpi->segmentation_map; const unsigned char *const active_map = cpi->active_map.map; int i; assert(AM_SEGMENT_ID_ACTIVE == CR_SEGMENT_ID_BASE); if (frame_is_intra_only(&cpi->common)) { cpi->active_map.enabled = 0; cpi->active_map.update = 1; } if (cpi->active_map.update) { if (cpi->active_map.enabled) { for (i = 0; i < cpi->common.mi_rows * cpi->common.mi_cols; ++i) if (seg_map[i] == AM_SEGMENT_ID_ACTIVE) seg_map[i] = active_map[i]; av1_enable_segmentation(seg); av1_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_SKIP); av1_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_H); av1_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_V); av1_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_U); av1_enable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_V); av1_set_segdata(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_H, -MAX_LOOP_FILTER); av1_set_segdata(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_V, -MAX_LOOP_FILTER); av1_set_segdata(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_U, -MAX_LOOP_FILTER); av1_set_segdata(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_V, -MAX_LOOP_FILTER); } else { av1_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_SKIP); av1_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_H); av1_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_Y_V); av1_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_U); av1_disable_segfeature(seg, AM_SEGMENT_ID_INACTIVE, SEG_LVL_ALT_LF_V); if (seg->enabled) { seg->update_data = 1; seg->update_map = 1; } } cpi->active_map.update = 0; } } int av1_set_active_map(AV1_COMP *cpi, unsigned char *new_map_16x16, int rows, int cols) { if (rows == cpi->common.mb_rows && cols == cpi->common.mb_cols) { unsigned char *const active_map_8x8 = cpi->active_map.map; const int mi_rows = cpi->common.mi_rows; const int mi_cols = cpi->common.mi_cols; const int row_scale = mi_size_high[BLOCK_16X16] == 2 ? 1 : 2; const int col_scale = mi_size_wide[BLOCK_16X16] == 2 ? 1 : 2; cpi->active_map.update = 1; if (new_map_16x16) { int r, c; for (r = 0; r < mi_rows; ++r) { for (c = 0; c < mi_cols; ++c) { active_map_8x8[r * mi_cols + c] = new_map_16x16[(r >> row_scale) * cols + (c >> col_scale)] ? AM_SEGMENT_ID_ACTIVE : AM_SEGMENT_ID_INACTIVE; } } cpi->active_map.enabled = 1; } else { cpi->active_map.enabled = 0; } return 0; } else { return -1; } } int av1_get_active_map(AV1_COMP *cpi, unsigned char *new_map_16x16, int rows, int cols) { if (rows == cpi->common.mb_rows && cols == cpi->common.mb_cols && new_map_16x16) { unsigned char *const seg_map_8x8 = cpi->segmentation_map; const int mi_rows = cpi->common.mi_rows; const int mi_cols = cpi->common.mi_cols; const int row_scale = mi_size_high[BLOCK_16X16] == 2 ? 1 : 2; const int col_scale = mi_size_wide[BLOCK_16X16] == 2 ? 1 : 2; memset(new_map_16x16, !cpi->active_map.enabled, rows * cols); if (cpi->active_map.enabled) { int r, c; for (r = 0; r < mi_rows; ++r) { for (c = 0; c < mi_cols; ++c) { // Cyclic refresh segments are considered active despite not having // AM_SEGMENT_ID_ACTIVE new_map_16x16[(r >> row_scale) * cols + (c >> col_scale)] |= seg_map_8x8[r * mi_cols + c] != AM_SEGMENT_ID_INACTIVE; } } } return 0; } else { return -1; } } // Compute the horizontal frequency components' energy in a frame // by calculuating the 16x4 Horizontal DCT. This is to be used to // decide the superresolution parameters. void analyze_hor_freq(const AV1_COMP *cpi, double *energy) { uint64_t freq_energy[16] = { 0 }; const YV12_BUFFER_CONFIG *buf = cpi->source; const int bd = cpi->td.mb.e_mbd.bd; const int width = buf->y_crop_width; const int height = buf->y_crop_height; DECLARE_ALIGNED(16, int32_t, coeff[16 * 4]); int n = 0; memset(freq_energy, 0, sizeof(freq_energy)); if (buf->flags & YV12_FLAG_HIGHBITDEPTH) { const int16_t *src16 = (const int16_t *)CONVERT_TO_SHORTPTR(buf->y_buffer); for (int i = 0; i < height - 4; i += 4) { for (int j = 0; j < width - 16; j += 16) { av1_fwd_txfm2d_16x4(src16 + i * buf->y_stride + j, coeff, buf->y_stride, H_DCT, bd); for (int k = 1; k < 16; ++k) { const uint64_t this_energy = ((int64_t)coeff[k] * coeff[k]) + ((int64_t)coeff[k + 16] * coeff[k + 16]) + ((int64_t)coeff[k + 32] * coeff[k + 32]) + ((int64_t)coeff[k + 48] * coeff[k + 48]); freq_energy[k] += ROUND_POWER_OF_TWO(this_energy, 2 + 2 * (bd - 8)); } n++; } } } else { assert(bd == 8); DECLARE_ALIGNED(16, int16_t, src16[16 * 4]); for (int i = 0; i < height - 4; i += 4) { for (int j = 0; j < width - 16; j += 16) { for (int ii = 0; ii < 4; ++ii) for (int jj = 0; jj < 16; ++jj) src16[ii * 16 + jj] = buf->y_buffer[(i + ii) * buf->y_stride + (j + jj)]; av1_fwd_txfm2d_16x4(src16, coeff, 16, H_DCT, bd); for (int k = 1; k < 16; ++k) { const uint64_t this_energy = ((int64_t)coeff[k] * coeff[k]) + ((int64_t)coeff[k + 16] * coeff[k + 16]) + ((int64_t)coeff[k + 32] * coeff[k + 32]) + ((int64_t)coeff[k + 48] * coeff[k + 48]); freq_energy[k] += ROUND_POWER_OF_TWO(this_energy, 2); } n++; } } } if (n) { for (int k = 1; k < 16; ++k) energy[k] = (double)freq_energy[k] / n; // Convert to cumulative energy for (int k = 14; k > 0; --k) energy[k] += energy[k + 1]; } else { for (int k = 1; k < 16; ++k) energy[k] = 1e+20; } } static void set_high_precision_mv(AV1_COMP *cpi, int allow_high_precision_mv, int cur_frame_force_integer_mv) { MACROBLOCK *const mb = &cpi->td.mb; cpi->common.allow_high_precision_mv = allow_high_precision_mv && cur_frame_force_integer_mv == 0; const int copy_hp = cpi->common.allow_high_precision_mv && cur_frame_force_integer_mv == 0; int *(*src)[2] = copy_hp ? &mb->nmvcost_hp : &mb->nmvcost; mb->mv_cost_stack = *src; } static BLOCK_SIZE select_sb_size(const AV1_COMP *const cpi) { const AV1_COMMON *const cm = &cpi->common; if (cpi->oxcf.superblock_size == AOM_SUPERBLOCK_SIZE_64X64) return BLOCK_64X64; #if CONFIG_FILEOPTIONS if (cm->options && cm->options->ext_partition) #endif if (cpi->oxcf.superblock_size == AOM_SUPERBLOCK_SIZE_128X128) return BLOCK_128X128; assert(cpi->oxcf.superblock_size == AOM_SUPERBLOCK_SIZE_DYNAMIC); // TODO(any): Possibly could improve this with a heuristic. #if CONFIG_FILEOPTIONS if (cm->options && !cm->options->ext_partition) return BLOCK_64X64; #endif // When superres / resize is on, 'cm->width / height' can change between // calls, so we don't apply this heuristic there. Also, this heuristic gives // compression gain for speed >= 2 only. // Things break if superblock size changes per-frame which is why this // heuristic is set based on configured speed rather than actual // speed-features (which may change per-frame in future) if (cpi->oxcf.superres_mode == SUPERRES_NONE && cpi->oxcf.resize_mode == RESIZE_NONE && cpi->oxcf.speed >= 2) { return (cm->width >= 480 && cm->height >= 360) ? BLOCK_128X128 : BLOCK_64X64; } return BLOCK_128X128; } static void setup_frame(AV1_COMP *cpi) { AV1_COMMON *const cm = &cpi->common; // Set up entropy context depending on frame type. The decoder mandates // the use of the default context, index 0, for keyframes and inter // frames where the error_resilient_mode or intra_only flag is set. For // other inter-frames the encoder currently uses only two contexts; // context 1 for ALTREF frames and context 0 for the others. if (frame_is_intra_only(cm) || cm->error_resilient_mode || cpi->ext_use_primary_ref_none) { av1_setup_past_independence(cm); } if (cm->current_frame.frame_type == KEY_FRAME && cm->show_frame) { cpi->refresh_golden_frame = 1; cpi->refresh_alt_ref_frame = 1; av1_zero(cpi->interp_filter_selected); set_sb_size(&cm->seq_params, select_sb_size(cpi)); } else if (frame_is_sframe(cm)) { cpi->refresh_golden_frame = 1; cpi->refresh_alt_ref_frame = 1; av1_zero(cpi->interp_filter_selected); set_sb_size(&cm->seq_params, select_sb_size(cpi)); } else { const RefCntBuffer *const primary_ref_buf = get_primary_ref_frame_buf(cm); if (primary_ref_buf == NULL) { av1_setup_past_independence(cm); cm->seg.update_map = 1; cm->seg.update_data = 1; } else { *cm->fc = primary_ref_buf->frame_context; } av1_zero(cpi->interp_filter_selected[0]); } cm->prev_frame = get_primary_ref_frame_buf(cm); cpi->vaq_refresh = 0; } static void enc_setup_mi(AV1_COMMON *cm) { int i; int mi_rows_sb_aligned = calc_mi_size(cm->mi_rows); cm->mi = cm->mip; memset(cm->mip, 0, cm->mi_stride * mi_rows_sb_aligned * sizeof(*cm->mip)); cm->prev_mi = cm->prev_mip; // Clear top border row memset(cm->prev_mip, 0, sizeof(*cm->prev_mip) * cm->mi_stride); // Clear left border column for (i = 0; i < mi_rows_sb_aligned; ++i) memset(&cm->prev_mip[i * cm->mi_stride], 0, sizeof(*cm->prev_mip)); cm->mi_grid_visible = cm->mi_grid_base; cm->prev_mi_grid_visible = cm->prev_mi_grid_base; memset(cm->mi_grid_base, 0, cm->mi_stride * mi_rows_sb_aligned * sizeof(*cm->mi_grid_base)); } static int enc_alloc_mi(AV1_COMMON *cm, int mi_size) { cm->mip = aom_calloc(mi_size, sizeof(*cm->mip)); if (!cm->mip) return 1; cm->prev_mip = aom_calloc(mi_size, sizeof(*cm->prev_mip)); if (!cm->prev_mip) return 1; cm->mi_alloc_size = mi_size; cm->mi_grid_base = (MB_MODE_INFO **)aom_calloc(mi_size, sizeof(MB_MODE_INFO *)); if (!cm->mi_grid_base) return 1; cm->prev_mi_grid_base = (MB_MODE_INFO **)aom_calloc(mi_size, sizeof(MB_MODE_INFO *)); if (!cm->prev_mi_grid_base) return 1; return 0; } static void enc_free_mi(AV1_COMMON *cm) { aom_free(cm->mip); cm->mip = NULL; aom_free(cm->prev_mip); cm->prev_mip = NULL; aom_free(cm->mi_grid_base); cm->mi_grid_base = NULL; aom_free(cm->prev_mi_grid_base); cm->prev_mi_grid_base = NULL; cm->mi_alloc_size = 0; } static void swap_mi_and_prev_mi(AV1_COMMON *cm) { // Current mip will be the prev_mip for the next frame. MB_MODE_INFO **temp_base = cm->prev_mi_grid_base; MB_MODE_INFO *temp = cm->prev_mip; cm->prev_mip = cm->mip; cm->mip = temp; // Update the upper left visible macroblock ptrs. cm->mi = cm->mip; cm->prev_mi = cm->prev_mip; cm->prev_mi_grid_base = cm->mi_grid_base; cm->mi_grid_base = temp_base; cm->mi_grid_visible = cm->mi_grid_base; cm->prev_mi_grid_visible = cm->prev_mi_grid_base; } void av1_initialize_enc(void) { av1_rtcd(); aom_dsp_rtcd(); aom_scale_rtcd(); av1_init_intra_predictors(); av1_init_me_luts(); av1_rc_init_minq_luts(); av1_init_wedge_masks(); } static void dealloc_context_buffers_ext(AV1_COMP *cpi) { if (cpi->mbmi_ext_base) { aom_free(cpi->mbmi_ext_base); cpi->mbmi_ext_base = NULL; } } static void alloc_context_buffers_ext(AV1_COMP *cpi) { AV1_COMMON *cm = &cpi->common; int mi_size = cm->mi_cols * cm->mi_rows; dealloc_context_buffers_ext(cpi); CHECK_MEM_ERROR(cm, cpi->mbmi_ext_base, aom_calloc(mi_size, sizeof(*cpi->mbmi_ext_base))); } static void reset_film_grain_chroma_params(aom_film_grain_t *pars) { pars->num_cr_points = 0; pars->cr_mult = 0; pars->cr_luma_mult = 0; memset(pars->scaling_points_cr, 0, sizeof(pars->scaling_points_cr)); memset(pars->ar_coeffs_cr, 0, sizeof(pars->ar_coeffs_cr)); pars->num_cb_points = 0; pars->cb_mult = 0; pars->cb_luma_mult = 0; pars->chroma_scaling_from_luma = 0; memset(pars->scaling_points_cb, 0, sizeof(pars->scaling_points_cb)); memset(pars->ar_coeffs_cb, 0, sizeof(pars->ar_coeffs_cb)); } static void update_film_grain_parameters(struct AV1_COMP *cpi, const AV1EncoderConfig *oxcf) { AV1_COMMON *const cm = &cpi->common; cpi->oxcf = *oxcf; if (cpi->film_grain_table) { aom_film_grain_table_free(cpi->film_grain_table); aom_free(cpi->film_grain_table); cpi->film_grain_table = NULL; } if (oxcf->film_grain_test_vector) { cm->seq_params.film_grain_params_present = 1; if (cm->current_frame.frame_type == KEY_FRAME) { memcpy(&cm->film_grain_params, film_grain_test_vectors + oxcf->film_grain_test_vector - 1, sizeof(cm->film_grain_params)); if (oxcf->monochrome) reset_film_grain_chroma_params(&cm->film_grain_params); cm->film_grain_params.bit_depth = cm->seq_params.bit_depth; if (cm->seq_params.color_range == AOM_CR_FULL_RANGE) { cm->film_grain_params.clip_to_restricted_range = 0; } } } else if (oxcf->film_grain_table_filename) { cpi->film_grain_table = aom_malloc(sizeof(*cpi->film_grain_table)); memset(cpi->film_grain_table, 0, sizeof(aom_film_grain_table_t)); aom_film_grain_table_read(cpi->film_grain_table, oxcf->film_grain_table_filename, &cm->error); } else { cm->seq_params.film_grain_params_present = 0; memset(&cm->film_grain_params, 0, sizeof(cm->film_grain_params)); } } static void dealloc_compressor_data(AV1_COMP *cpi) { AV1_COMMON *const cm = &cpi->common; const int num_planes = av1_num_planes(cm); dealloc_context_buffers_ext(cpi); aom_free(cpi->tile_data); cpi->tile_data = NULL; // Delete sementation map aom_free(cpi->segmentation_map); cpi->segmentation_map = NULL; av1_cyclic_refresh_free(cpi->cyclic_refresh); cpi->cyclic_refresh = NULL; aom_free(cpi->active_map.map); cpi->active_map.map = NULL; aom_free(cpi->td.mb.above_pred_buf); cpi->td.mb.above_pred_buf = NULL; aom_free(cpi->td.mb.left_pred_buf); cpi->td.mb.left_pred_buf = NULL; aom_free(cpi->td.mb.wsrc_buf); cpi->td.mb.wsrc_buf = NULL; aom_free(cpi->td.mb.inter_modes_info); cpi->td.mb.inter_modes_info = NULL; for (int i = 0; i < 2; i++) for (int j = 0; j < 2; j++) { aom_free(cpi->td.mb.hash_value_buffer[i][j]); cpi->td.mb.hash_value_buffer[i][j] = NULL; } aom_free(cpi->td.mb.mask_buf); cpi->td.mb.mask_buf = NULL; aom_free(cm->tpl_mvs); cm->tpl_mvs = NULL; av1_free_ref_frame_buffers(cm->buffer_pool); av1_free_txb_buf(cpi); av1_free_context_buffers(cm); aom_free_frame_buffer(&cpi->last_frame_uf); av1_free_restoration_buffers(cm); aom_free_frame_buffer(&cpi->trial_frame_rst); aom_free_frame_buffer(&cpi->scaled_source); aom_free_frame_buffer(&cpi->scaled_last_source); aom_free_frame_buffer(&cpi->alt_ref_buffer); av1_lookahead_destroy(cpi->lookahead); aom_free(cpi->tile_tok[0][0]); cpi->tile_tok[0][0] = 0; aom_free(cpi->tplist[0][0]); cpi->tplist[0][0] = NULL; av1_free_pc_tree(&cpi->td, num_planes); aom_free(cpi->td.mb.palette_buffer); aom_free(cpi->td.mb.tmp_conv_dst); for (int j = 0; j < 2; ++j) { aom_free(cpi->td.mb.tmp_obmc_bufs[j]); } #if CONFIG_DENOISE if (cpi->denoise_and_model) { aom_denoise_and_model_free(cpi->denoise_and_model); cpi->denoise_and_model = NULL; } #endif if (cpi->film_grain_table) { aom_film_grain_table_free(cpi->film_grain_table); cpi->film_grain_table = NULL; } } static void save_coding_context(AV1_COMP *cpi) { CODING_CONTEXT *const cc = &cpi->coding_context; AV1_COMMON *cm = &cpi->common; // Stores a snapshot of key state variables which can subsequently be // restored with a call to av1_restore_coding_context. These functions are // intended for use in a re-code loop in av1_compress_frame where the // quantizer value is adjusted between loop iterations. av1_copy(cc->nmv_vec_cost, cpi->td.mb.nmv_vec_cost); av1_copy(cc->nmv_costs, cpi->nmv_costs); av1_copy(cc->nmv_costs_hp, cpi->nmv_costs_hp); cc->fc = *cm->fc; } static void restore_coding_context(AV1_COMP *cpi) { CODING_CONTEXT *const cc = &cpi->coding_context; AV1_COMMON *cm = &cpi->common; // Restore key state variables to the snapshot state stored in the // previous call to av1_save_coding_context. av1_copy(cpi->td.mb.nmv_vec_cost, cc->nmv_vec_cost); av1_copy(cpi->nmv_costs, cc->nmv_costs); av1_copy(cpi->nmv_costs_hp, cc->nmv_costs_hp); *cm->fc = cc->fc; } static void configure_static_seg_features(AV1_COMP *cpi) { AV1_COMMON *const cm = &cpi->common; const RATE_CONTROL *const rc = &cpi->rc; struct segmentation *const seg = &cm->seg; int high_q = (int)(rc->avg_q > 48.0); int qi_delta; // Disable and clear down for KF if (cm->current_frame.frame_type == KEY_FRAME) { // Clear down the global segmentation map memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols); seg->update_map = 0; seg->update_data = 0; cpi->static_mb_pct = 0; // Disable segmentation av1_disable_segmentation(seg); // Clear down the segment features. av1_clearall_segfeatures(seg); } else if (cpi->refresh_alt_ref_frame) { // If this is an alt ref frame // Clear down the global segmentation map memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols); seg->update_map = 0; seg->update_data = 0; cpi->static_mb_pct = 0; // Disable segmentation and individual segment features by default av1_disable_segmentation(seg); av1_clearall_segfeatures(seg); // Scan frames from current to arf frame. // This function re-enables segmentation if appropriate. av1_update_mbgraph_stats(cpi); // If segmentation was enabled set those features needed for the // arf itself. if (seg->enabled) { seg->update_map = 1; seg->update_data = 1; qi_delta = av1_compute_qdelta(rc, rc->avg_q, rc->avg_q * 0.875, cm->seq_params.bit_depth); av1_set_segdata(seg, 1, SEG_LVL_ALT_Q, qi_delta - 2); av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_Y_H, -2); av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_Y_V, -2); av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_U, -2); av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_V, -2); av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_Y_H); av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_Y_V); av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_U); av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_V); av1_enable_segfeature(seg, 1, SEG_LVL_ALT_Q); } } else if (seg->enabled) { // All other frames if segmentation has been enabled // First normal frame in a valid gf or alt ref group if (rc->frames_since_golden == 0) { // Set up segment features for normal frames in an arf group if (rc->source_alt_ref_active) { seg->update_map = 0; seg->update_data = 1; qi_delta = av1_compute_qdelta(rc, rc->avg_q, rc->avg_q * 1.125, cm->seq_params.bit_depth); av1_set_segdata(seg, 1, SEG_LVL_ALT_Q, qi_delta + 2); av1_enable_segfeature(seg, 1, SEG_LVL_ALT_Q); av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_Y_H, -2); av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_Y_V, -2); av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_U, -2); av1_set_segdata(seg, 1, SEG_LVL_ALT_LF_V, -2); av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_Y_H); av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_Y_V); av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_U); av1_enable_segfeature(seg, 1, SEG_LVL_ALT_LF_V); // Segment coding disabled for compred testing if (high_q || (cpi->static_mb_pct == 100)) { av1_set_segdata(seg, 1, SEG_LVL_REF_FRAME, ALTREF_FRAME); av1_enable_segfeature(seg, 1, SEG_LVL_REF_FRAME); av1_enable_segfeature(seg, 1, SEG_LVL_SKIP); } } else { // Disable segmentation and clear down features if alt ref // is not active for this group av1_disable_segmentation(seg); memset(cpi->segmentation_map, 0, cm->mi_rows * cm->mi_cols); seg->update_map = 0; seg->update_data = 0; av1_clearall_segfeatures(seg); } } else if (rc->is_src_frame_alt_ref) { // Special case where we are coding over the top of a previous // alt ref frame. // Segment coding disabled for compred testing // Enable ref frame features for segment 0 as well av1_enable_segfeature(seg, 0, SEG_LVL_REF_FRAME); av1_enable_segfeature(seg, 1, SEG_LVL_REF_FRAME); // All mbs should use ALTREF_FRAME av1_clear_segdata(seg, 0, SEG_LVL_REF_FRAME); av1_set_segdata(seg, 0, SEG_LVL_REF_FRAME, ALTREF_FRAME); av1_clear_segdata(seg, 1, SEG_LVL_REF_FRAME); av1_set_segdata(seg, 1, SEG_LVL_REF_FRAME, ALTREF_FRAME); // Skip all MBs if high Q (0,0 mv and skip coeffs) if (high_q) { av1_enable_segfeature(seg, 0, SEG_LVL_SKIP); av1_enable_segfeature(seg, 1, SEG_LVL_SKIP); } // Enable data update seg->update_data = 1; } else { // All other frames. // No updates.. leave things as they are. seg->update_map = 0; seg->update_data = 0; } } } static void update_reference_segmentation_map(AV1_COMP *cpi) { AV1_COMMON *const cm = &cpi->common; MB_MODE_INFO **mi_4x4_ptr = cm->mi_grid_visible; uint8_t *cache_ptr = cm->cur_frame->seg_map; int row, col; for (row = 0; row < cm->mi_rows; row++) { MB_MODE_INFO **mi_4x4 = mi_4x4_ptr; uint8_t *cache = cache_ptr; for (col = 0; col < cm->mi_cols; col++, mi_4x4++, cache++) cache[0] = mi_4x4[0]->segment_id; mi_4x4_ptr += cm->mi_stride; cache_ptr += cm->mi_cols; } } static void alloc_raw_frame_buffers(AV1_COMP *cpi) { AV1_COMMON *cm = &cpi->common; const SequenceHeader *const seq_params = &cm->seq_params; const AV1EncoderConfig *oxcf = &cpi->oxcf; if (!cpi->lookahead) cpi->lookahead = av1_lookahead_init( oxcf->width, oxcf->height, seq_params->subsampling_x, seq_params->subsampling_y, seq_params->use_highbitdepth, oxcf->lag_in_frames, oxcf->border_in_pixels); if (!cpi->lookahead) aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, "Failed to allocate lag buffers"); // TODO(agrange) Check if ARF is enabled and skip allocation if not. if (aom_realloc_frame_buffer( &cpi->alt_ref_buffer, oxcf->width, oxcf->height, seq_params->subsampling_x, seq_params->subsampling_y, seq_params->use_highbitdepth, oxcf->border_in_pixels, cm->byte_alignment, NULL, NULL, NULL)) aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, "Failed to allocate altref buffer"); } static void alloc_util_frame_buffers(AV1_COMP *cpi) { AV1_COMMON *const cm = &cpi->common; const SequenceHeader *const seq_params = &cm->seq_params; if (aom_realloc_frame_buffer( &cpi->last_frame_uf, cm->width, cm->height, seq_params->subsampling_x, seq_params->subsampling_y, seq_params->use_highbitdepth, cpi->oxcf.border_in_pixels, cm->byte_alignment, NULL, NULL, NULL)) aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, "Failed to allocate last frame buffer"); if (aom_realloc_frame_buffer( &cpi->trial_frame_rst, cm->superres_upscaled_width, cm->superres_upscaled_height, seq_params->subsampling_x, seq_params->subsampling_y, seq_params->use_highbitdepth, AOM_RESTORATION_FRAME_BORDER, cm->byte_alignment, NULL, NULL, NULL)) aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, "Failed to allocate trial restored frame buffer"); if (aom_realloc_frame_buffer( &cpi->scaled_source, cm->width, cm->height, seq_params->subsampling_x, seq_params->subsampling_y, seq_params->use_highbitdepth, cpi->oxcf.border_in_pixels, cm->byte_alignment, NULL, NULL, NULL)) aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, "Failed to allocate scaled source buffer"); if (aom_realloc_frame_buffer( &cpi->scaled_last_source, cm->width, cm->height, seq_params->subsampling_x, seq_params->subsampling_y, seq_params->use_highbitdepth, cpi->oxcf.border_in_pixels, cm->byte_alignment, NULL, NULL, NULL)) aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, "Failed to allocate scaled last source buffer"); } static void alloc_compressor_data(AV1_COMP *cpi) { AV1_COMMON *cm = &cpi->common; const int num_planes = av1_num_planes(cm); av1_alloc_context_buffers(cm, cm->width, cm->height); int mi_rows_aligned_to_sb = ALIGN_POWER_OF_TWO(cm->mi_rows, cm->seq_params.mib_size_log2); int sb_rows = mi_rows_aligned_to_sb >> cm->seq_params.mib_size_log2; av1_alloc_txb_buf(cpi); alloc_context_buffers_ext(cpi); aom_free(cpi->tile_tok[0][0]); { unsigned int tokens = get_token_alloc(cm->mb_rows, cm->mb_cols, MAX_SB_SIZE_LOG2, num_planes); CHECK_MEM_ERROR(cm, cpi->tile_tok[0][0], aom_calloc(tokens, sizeof(*cpi->tile_tok[0][0]))); } aom_free(cpi->tplist[0][0]); CHECK_MEM_ERROR(cm, cpi->tplist[0][0], aom_calloc(sb_rows * MAX_TILE_ROWS * MAX_TILE_COLS, sizeof(*cpi->tplist[0][0]))); av1_setup_pc_tree(&cpi->common, &cpi->td); } void av1_new_framerate(AV1_COMP *cpi, double framerate) { cpi->framerate = framerate < 0.1 ? 30 : framerate; av1_rc_update_framerate(cpi, cpi->common.width, cpi->common.height); } static void set_tile_info(AV1_COMP *cpi) { AV1_COMMON *const cm = &cpi->common; int i, start_sb; av1_get_tile_limits(cm); // configure tile columns if (cpi->oxcf.tile_width_count == 0 || cpi->oxcf.tile_height_count == 0) { cm->uniform_tile_spacing_flag = 1; cm->log2_tile_cols = AOMMAX(cpi->oxcf.tile_columns, cm->min_log2_tile_cols); cm->log2_tile_cols = AOMMIN(cm->log2_tile_cols, cm->max_log2_tile_cols); } else { int mi_cols = ALIGN_POWER_OF_TWO(cm->mi_cols, cm->seq_params.mib_size_log2); int sb_cols = mi_cols >> cm->seq_params.mib_size_log2; int size_sb, j = 0; cm->uniform_tile_spacing_flag = 0; for (i = 0, start_sb = 0; start_sb < sb_cols && i < MAX_TILE_COLS; i++) { cm->tile_col_start_sb[i] = start_sb; size_sb = cpi->oxcf.tile_widths[j++]; if (j >= cpi->oxcf.tile_width_count) j = 0; start_sb += AOMMIN(size_sb, cm->max_tile_width_sb); } cm->tile_cols = i; cm->tile_col_start_sb[i] = sb_cols; } av1_calculate_tile_cols(cm); // configure tile rows if (cm->uniform_tile_spacing_flag) { cm->log2_tile_rows = AOMMAX(cpi->oxcf.tile_rows, cm->min_log2_tile_rows); cm->log2_tile_rows = AOMMIN(cm->log2_tile_rows, cm->max_log2_tile_rows); } else { int mi_rows = ALIGN_POWER_OF_TWO(cm->mi_rows, cm->seq_params.mib_size_log2); int sb_rows = mi_rows >> cm->seq_params.mib_size_log2; int size_sb, j = 0; for (i = 0, start_sb = 0; start_sb < sb_rows && i < MAX_TILE_ROWS; i++) { cm->tile_row_start_sb[i] = start_sb; size_sb = cpi->oxcf.tile_heights[j++]; if (j >= cpi->oxcf.tile_height_count) j = 0; start_sb += AOMMIN(size_sb, cm->max_tile_height_sb); } cm->tile_rows = i; cm->tile_row_start_sb[i] = sb_rows; } av1_calculate_tile_rows(cm); } static void update_frame_size(AV1_COMP *cpi) { AV1_COMMON *const cm = &cpi->common; MACROBLOCKD *const xd = &cpi->td.mb.e_mbd; av1_set_mb_mi(cm, cm->width, cm->height); av1_init_context_buffers(cm); av1_init_macroblockd(cm, xd, NULL); memset(cpi->mbmi_ext_base, 0, cm->mi_rows * cm->mi_cols * sizeof(*cpi->mbmi_ext_base)); set_tile_info(cpi); } static void init_buffer_indices(AV1_COMP *cpi) { int fb_idx; for (fb_idx = 0; fb_idx < REF_FRAMES; ++fb_idx) cpi->common.remapped_ref_idx[fb_idx] = fb_idx; cpi->rate_index = 0; cpi->rate_size = 0; } static INLINE int does_level_match(int width, int height, double fps, int lvl_width, int lvl_height, double lvl_fps, int lvl_dim_mult) { const int64_t lvl_luma_pels = lvl_width * lvl_height; const double lvl_display_sample_rate = lvl_luma_pels * lvl_fps; const int64_t luma_pels = width * height; const double display_sample_rate = luma_pels * fps; return luma_pels <= lvl_luma_pels && display_sample_rate <= lvl_display_sample_rate && width <= lvl_width * lvl_dim_mult && height <= lvl_height * lvl_dim_mult; } static void set_bitstream_level_tier(SequenceHeader *seq, AV1_COMMON *cm, const AV1EncoderConfig *oxcf) { // TODO(any): This is a placeholder function that only addresses dimensions // and max display sample rates. // Need to add checks for max bit rate, max decoded luma sample rate, header // rate, etc. that are not covered by this function. (void)oxcf; BitstreamLevel bl = { 9, 3 }; if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 512, 288, 30.0, 4)) { bl.major = 2; bl.minor = 0; } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 704, 396, 30.0, 4)) { bl.major = 2; bl.minor = 1; } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 1088, 612, 30.0, 4)) { bl.major = 3; bl.minor = 0; } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 1376, 774, 30.0, 4)) { bl.major = 3; bl.minor = 1; } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 2048, 1152, 30.0, 3)) { bl.major = 4; bl.minor = 0; } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 2048, 1152, 60.0, 3)) { bl.major = 4; bl.minor = 1; } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 4096, 2176, 30.0, 2)) { bl.major = 5; bl.minor = 0; } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 4096, 2176, 60.0, 2)) { bl.major = 5; bl.minor = 1; } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 4096, 2176, 120.0, 2)) { bl.major = 5; bl.minor = 2; } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 8192, 4352, 30.0, 2)) { bl.major = 6; bl.minor = 0; } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 8192, 4352, 60.0, 2)) { bl.major = 6; bl.minor = 1; } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 8192, 4352, 120.0, 2)) { bl.major = 6; bl.minor = 2; } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 16384, 8704, 30.0, 2)) { bl.major = 7; bl.minor = 0; } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 16384, 8704, 60.0, 2)) { bl.major = 7; bl.minor = 1; } else if (does_level_match(oxcf->width, oxcf->height, oxcf->init_framerate, 16384, 8704, 120.0, 2)) { bl.major = 7; bl.minor = 2; } for (int i = 0; i < MAX_NUM_OPERATING_POINTS; ++i) { seq->level[i] = bl; seq->tier[i] = 0; // setting main tier by default // Set the maximum parameters for bitrate and buffer size for this profile, // level, and tier cm->op_params[i].bitrate = max_level_bitrate( cm->seq_params.profile, major_minor_to_seq_level_idx(seq->level[i]), seq->tier[i]); // Level with seq_level_idx = 31 returns a high "dummy" bitrate to pass the // check if (cm->op_params[i].bitrate == 0) aom_internal_error( &cm->error, AOM_CODEC_UNSUP_BITSTREAM, "AV1 does not support this combination of profile, level, and tier."); // Buffer size in bits/s is bitrate in bits/s * 1 s cm->op_params[i].buffer_size = cm->op_params[i].bitrate; } } static void init_seq_coding_tools(SequenceHeader *seq, AV1_COMMON *cm, const AV1EncoderConfig *oxcf) { seq->still_picture = (oxcf->limit == 1); seq->reduced_still_picture_hdr = seq->still_picture; seq->reduced_still_picture_hdr &= !oxcf->full_still_picture_hdr; seq->force_screen_content_tools = 2; seq->force_integer_mv = 2; seq->order_hint_info.enable_order_hint = oxcf->enable_order_hint; seq->frame_id_numbers_present_flag = !(seq->still_picture && seq->reduced_still_picture_hdr) && !oxcf->large_scale_tile && oxcf->error_resilient_mode; if (seq->still_picture && seq->reduced_still_picture_hdr) { seq->order_hint_info.enable_order_hint = 0; seq->force_screen_content_tools = 2; seq->force_integer_mv = 2; } seq->order_hint_info.order_hint_bits_minus_1 = seq->order_hint_info.enable_order_hint ? DEFAULT_EXPLICIT_ORDER_HINT_BITS - 1 : -1; seq->max_frame_width = oxcf->forced_max_frame_width ? oxcf->forced_max_frame_width : oxcf->width; seq->max_frame_height = oxcf->forced_max_frame_height ? oxcf->forced_max_frame_height : oxcf->height; seq->num_bits_width = (seq->max_frame_width > 1) ? get_msb(seq->max_frame_width - 1) + 1 : 1; seq->num_bits_height = (seq->max_frame_height > 1) ? get_msb(seq->max_frame_height - 1) + 1 : 1; assert(seq->num_bits_width <= 16); assert(seq->num_bits_height <= 16); seq->frame_id_length = FRAME_ID_LENGTH; seq->delta_frame_id_length = DELTA_FRAME_ID_LENGTH; seq->enable_dual_filter = oxcf->enable_dual_filter; seq->order_hint_info.enable_dist_wtd_comp = oxcf->enable_dist_wtd_comp; seq->order_hint_info.enable_dist_wtd_comp &= seq->order_hint_info.enable_order_hint; seq->order_hint_info.enable_ref_frame_mvs = oxcf->enable_ref_frame_mvs; seq->order_hint_info.enable_ref_frame_mvs &= seq->order_hint_info.enable_order_hint; seq->enable_superres = oxcf->enable_superres; seq->enable_cdef = oxcf->enable_cdef; seq->enable_restoration = oxcf->enable_restoration; seq->enable_warped_motion = oxcf->enable_warped_motion; seq->enable_interintra_compound = oxcf->enable_interintra_comp; seq->enable_masked_compound = oxcf->enable_masked_comp; seq->enable_intra_edge_filter = oxcf->enable_intra_edge_filter; seq->enable_filter_intra = oxcf->enable_filter_intra; set_bitstream_level_tier(seq, cm, oxcf); if (seq->operating_points_cnt_minus_1 == 0) { seq->operating_point_idc[0] = 0; } else { // Set operating_point_idc[] such that for the i-th operating point the // first (operating_points_cnt-i) spatial layers and the first temporal // layer are decoded Note that highest quality operating point should come // first for (int i = 0; i < seq->operating_points_cnt_minus_1 + 1; i++) seq->operating_point_idc[i] = (~(~0u << (seq->operating_points_cnt_minus_1 + 1 - i)) << 8) | 1; } } static void init_config(struct AV1_COMP *cpi, AV1EncoderConfig *oxcf) { AV1_COMMON *const cm = &cpi->common; cpi->oxcf = *oxcf; cpi->framerate = oxcf->init_framerate; cm->seq_params.profile = oxcf->profile; cm->seq_params.bit_depth = oxcf->bit_depth; cm->seq_params.use_highbitdepth = oxcf->use_highbitdepth; cm->seq_params.color_primaries = oxcf->color_primaries; cm->seq_params.transfer_characteristics = oxcf->transfer_characteristics; cm->seq_params.matrix_coefficients = oxcf->matrix_coefficients; cm->seq_params.monochrome = oxcf->monochrome; cm->seq_params.chroma_sample_position = oxcf->chroma_sample_position; cm->seq_params.color_range = oxcf->color_range; cm->timing_info_present = oxcf->timing_info_present; cm->timing_info.num_units_in_display_tick = oxcf->timing_info.num_units_in_display_tick; cm->timing_info.time_scale = oxcf->timing_info.time_scale; cm->timing_info.equal_picture_interval = oxcf->timing_info.equal_picture_interval; cm->timing_info.num_ticks_per_picture = oxcf->timing_info.num_ticks_per_picture; cm->seq_params.display_model_info_present_flag = oxcf->display_model_info_present_flag; cm->seq_params.decoder_model_info_present_flag = oxcf->decoder_model_info_present_flag; if (oxcf->decoder_model_info_present_flag) { // set the decoder model parameters in schedule mode cm->buffer_model.num_units_in_decoding_tick = oxcf->buffer_model.num_units_in_decoding_tick; cm->buffer_removal_time_present = 1; set_aom_dec_model_info(&cm->buffer_model); set_dec_model_op_parameters(&cm->op_params[0]); } else if (cm->timing_info_present && cm->timing_info.equal_picture_interval && !cm->seq_params.decoder_model_info_present_flag) { // set the decoder model parameters in resource availability mode set_resource_availability_parameters(&cm->op_params[0]); } else { cm->op_params[0].initial_display_delay = 10; // Default value (not signaled) } if (cm->seq_params.monochrome) { cm->seq_params.subsampling_x = 1; cm->seq_params.subsampling_y = 1; } else if (cm->seq_params.color_primaries == AOM_CICP_CP_BT_709 && cm->seq_params.transfer_characteristics == AOM_CICP_TC_SRGB && cm->seq_params.matrix_coefficients == AOM_CICP_MC_IDENTITY) { cm->seq_params.subsampling_x = 0; cm->seq_params.subsampling_y = 0; } else { if (cm->seq_params.profile == 0) { cm->seq_params.subsampling_x = 1; cm->seq_params.subsampling_y = 1; } else if (cm->seq_params.profile == 1) { cm->seq_params.subsampling_x = 0; cm->seq_params.subsampling_y = 0; } else { if (cm->seq_params.bit_depth == AOM_BITS_12) { cm->seq_params.subsampling_x = oxcf->chroma_subsampling_x; cm->seq_params.subsampling_y = oxcf->chroma_subsampling_y; } else { cm->seq_params.subsampling_x = 1; cm->seq_params.subsampling_y = 0; } } } cm->width = oxcf->width; cm->height = oxcf->height; set_sb_size(&cm->seq_params, select_sb_size(cpi)); // set sb size before allocations alloc_compressor_data(cpi); update_film_grain_parameters(cpi, oxcf); // Single thread case: use counts in common. cpi->td.counts = &cpi->counts; // change includes all joint functionality av1_change_config(cpi, oxcf); cpi->static_mb_pct = 0; cpi->ref_frame_flags = 0; // Reset resize pending flags cpi->resize_pending_width = 0; cpi->resize_pending_height = 0; init_buffer_indices(cpi); } static void set_rc_buffer_sizes(RATE_CONTROL *rc, const AV1EncoderConfig *oxcf) { const int64_t bandwidth = oxcf->target_bandwidth; const int64_t starting = oxcf->starting_buffer_level_ms; const int64_t optimal = oxcf->optimal_buffer_level_ms; const int64_t maximum = oxcf->maximum_buffer_size_ms; rc->starting_buffer_level = starting * bandwidth / 1000; rc->optimal_buffer_level = (optimal == 0) ? bandwidth / 8 : optimal * bandwidth / 1000; rc->maximum_buffer_size = (maximum == 0) ? bandwidth / 8 : maximum * bandwidth / 1000; } #define HIGHBD_BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX4DF, JSDAF, JSVAF) \ cpi->fn_ptr[BT].sdf = SDF; \ cpi->fn_ptr[BT].sdaf = SDAF; \ cpi->fn_ptr[BT].vf = VF; \ cpi->fn_ptr[BT].svf = SVF; \ cpi->fn_ptr[BT].svaf = SVAF; \ cpi->fn_ptr[BT].sdx4df = SDX4DF; \ cpi->fn_ptr[BT].jsdaf = JSDAF; \ cpi->fn_ptr[BT].jsvaf = JSVAF; #define MAKE_BFP_SAD_WRAPPER(fnname) \ static unsigned int fnname##_bits8(const uint8_t *src_ptr, \ int source_stride, \ const uint8_t *ref_ptr, int ref_stride) { \ return fnname(src_ptr, source_stride, ref_ptr, ref_stride); \ } \ static unsigned int fnname##_bits10( \ const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ int ref_stride) { \ return fnname(src_ptr, source_stride, ref_ptr, ref_stride) >> 2; \ } \ static unsigned int fnname##_bits12( \ const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ int ref_stride) { \ return fnname(src_ptr, source_stride, ref_ptr, ref_stride) >> 4; \ } #define MAKE_BFP_SADAVG_WRAPPER(fnname) \ static unsigned int fnname##_bits8( \ const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ int ref_stride, const uint8_t *second_pred) { \ return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred); \ } \ static unsigned int fnname##_bits10( \ const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ int ref_stride, const uint8_t *second_pred) { \ return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred) >> \ 2; \ } \ static unsigned int fnname##_bits12( \ const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ int ref_stride, const uint8_t *second_pred) { \ return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred) >> \ 4; \ } #define MAKE_BFP_SAD4D_WRAPPER(fnname) \ static void fnname##_bits8(const uint8_t *src_ptr, int source_stride, \ const uint8_t *const ref_ptr[], int ref_stride, \ unsigned int *sad_array) { \ fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \ } \ static void fnname##_bits10(const uint8_t *src_ptr, int source_stride, \ const uint8_t *const ref_ptr[], int ref_stride, \ unsigned int *sad_array) { \ int i; \ fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \ for (i = 0; i < 4; i++) sad_array[i] >>= 2; \ } \ static void fnname##_bits12(const uint8_t *src_ptr, int source_stride, \ const uint8_t *const ref_ptr[], int ref_stride, \ unsigned int *sad_array) { \ int i; \ fnname(src_ptr, source_stride, ref_ptr, ref_stride, sad_array); \ for (i = 0; i < 4; i++) sad_array[i] >>= 4; \ } #define MAKE_BFP_JSADAVG_WRAPPER(fnname) \ static unsigned int fnname##_bits8( \ const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ int ref_stride, const uint8_t *second_pred, \ const DIST_WTD_COMP_PARAMS *jcp_param) { \ return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred, \ jcp_param); \ } \ static unsigned int fnname##_bits10( \ const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ int ref_stride, const uint8_t *second_pred, \ const DIST_WTD_COMP_PARAMS *jcp_param) { \ return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred, \ jcp_param) >> \ 2; \ } \ static unsigned int fnname##_bits12( \ const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ int ref_stride, const uint8_t *second_pred, \ const DIST_WTD_COMP_PARAMS *jcp_param) { \ return fnname(src_ptr, source_stride, ref_ptr, ref_stride, second_pred, \ jcp_param) >> \ 4; \ } MAKE_BFP_SAD_WRAPPER(aom_highbd_sad128x128) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad128x128_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad128x128x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad128x64) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad128x64_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad128x64x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad64x128) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad64x128_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad64x128x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad32x16) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad32x16_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad32x16x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad16x32) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad16x32_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad16x32x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad64x32) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad64x32_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad64x32x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad32x64) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad32x64_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad32x64x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad32x32) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad32x32_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad32x32x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad64x64) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad64x64_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad64x64x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad16x16) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad16x16_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad16x16x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad16x8) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad16x8_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad16x8x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad8x16) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad8x16_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad8x16x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad8x8) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad8x8_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad8x8x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad8x4) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad8x4_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad8x4x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad4x8) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad4x8_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad4x8x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad4x4) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad4x4_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad4x4x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad4x16) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad4x16_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad4x16x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad16x4) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad16x4_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad16x4x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad8x32) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad8x32_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad8x32x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad32x8) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad32x8_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad32x8x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad16x64) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad16x64_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad16x64x4d) MAKE_BFP_SAD_WRAPPER(aom_highbd_sad64x16) MAKE_BFP_SADAVG_WRAPPER(aom_highbd_sad64x16_avg) MAKE_BFP_SAD4D_WRAPPER(aom_highbd_sad64x16x4d) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad128x128_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad128x64_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad64x128_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad32x16_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad16x32_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad64x32_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad32x64_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad32x32_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad64x64_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad16x16_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad16x8_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad8x16_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad8x8_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad8x4_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad4x8_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad4x4_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad4x16_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad16x4_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad8x32_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad32x8_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad16x64_avg) MAKE_BFP_JSADAVG_WRAPPER(aom_highbd_dist_wtd_sad64x16_avg) #define HIGHBD_MBFP(BT, MCSDF, MCSVF) \ cpi->fn_ptr[BT].msdf = MCSDF; \ cpi->fn_ptr[BT].msvf = MCSVF; #define MAKE_MBFP_COMPOUND_SAD_WRAPPER(fnname) \ static unsigned int fnname##_bits8( \ const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ int ref_stride, const uint8_t *second_pred_ptr, const uint8_t *m, \ int m_stride, int invert_mask) { \ return fnname(src_ptr, source_stride, ref_ptr, ref_stride, \ second_pred_ptr, m, m_stride, invert_mask); \ } \ static unsigned int fnname##_bits10( \ const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ int ref_stride, const uint8_t *second_pred_ptr, const uint8_t *m, \ int m_stride, int invert_mask) { \ return fnname(src_ptr, source_stride, ref_ptr, ref_stride, \ second_pred_ptr, m, m_stride, invert_mask) >> \ 2; \ } \ static unsigned int fnname##_bits12( \ const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, \ int ref_stride, const uint8_t *second_pred_ptr, const uint8_t *m, \ int m_stride, int invert_mask) { \ return fnname(src_ptr, source_stride, ref_ptr, ref_stride, \ second_pred_ptr, m, m_stride, invert_mask) >> \ 4; \ } MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad128x128) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad128x64) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad64x128) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad64x64) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad64x32) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad32x64) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad32x32) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad32x16) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad16x32) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad16x16) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad16x8) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad8x16) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad8x8) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad8x4) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad4x8) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad4x4) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad4x16) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad16x4) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad8x32) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad32x8) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad16x64) MAKE_MBFP_COMPOUND_SAD_WRAPPER(aom_highbd_masked_sad64x16) #define HIGHBD_OBFP(BT, OSDF, OVF, OSVF) \ cpi->fn_ptr[BT].osdf = OSDF; \ cpi->fn_ptr[BT].ovf = OVF; \ cpi->fn_ptr[BT].osvf = OSVF; #define MAKE_OBFP_SAD_WRAPPER(fnname) \ static unsigned int fnname##_bits8(const uint8_t *ref, int ref_stride, \ const int32_t *wsrc, \ const int32_t *msk) { \ return fnname(ref, ref_stride, wsrc, msk); \ } \ static unsigned int fnname##_bits10(const uint8_t *ref, int ref_stride, \ const int32_t *wsrc, \ const int32_t *msk) { \ return fnname(ref, ref_stride, wsrc, msk) >> 2; \ } \ static unsigned int fnname##_bits12(const uint8_t *ref, int ref_stride, \ const int32_t *wsrc, \ const int32_t *msk) { \ return fnname(ref, ref_stride, wsrc, msk) >> 4; \ } MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad128x128) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad128x64) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad64x128) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad64x64) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad64x32) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad32x64) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad32x32) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad32x16) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad16x32) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad16x16) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad16x8) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad8x16) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad8x8) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad8x4) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad4x8) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad4x4) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad4x16) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad16x4) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad8x32) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad32x8) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad16x64) MAKE_OBFP_SAD_WRAPPER(aom_highbd_obmc_sad64x16) static void highbd_set_var_fns(AV1_COMP *const cpi) { AV1_COMMON *const cm = &cpi->common; if (cm->seq_params.use_highbitdepth) { switch (cm->seq_params.bit_depth) { case AOM_BITS_8: HIGHBD_BFP(BLOCK_64X16, aom_highbd_sad64x16_bits8, aom_highbd_sad64x16_avg_bits8, aom_highbd_8_variance64x16, aom_highbd_8_sub_pixel_variance64x16, aom_highbd_8_sub_pixel_avg_variance64x16, aom_highbd_sad64x16x4d_bits8, aom_highbd_dist_wtd_sad64x16_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance64x16) HIGHBD_BFP(BLOCK_16X64, aom_highbd_sad16x64_bits8, aom_highbd_sad16x64_avg_bits8, aom_highbd_8_variance16x64, aom_highbd_8_sub_pixel_variance16x64, aom_highbd_8_sub_pixel_avg_variance16x64, aom_highbd_sad16x64x4d_bits8, aom_highbd_dist_wtd_sad16x64_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance16x64) HIGHBD_BFP( BLOCK_32X8, aom_highbd_sad32x8_bits8, aom_highbd_sad32x8_avg_bits8, aom_highbd_8_variance32x8, aom_highbd_8_sub_pixel_variance32x8, aom_highbd_8_sub_pixel_avg_variance32x8, aom_highbd_sad32x8x4d_bits8, aom_highbd_dist_wtd_sad32x8_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance32x8) HIGHBD_BFP( BLOCK_8X32, aom_highbd_sad8x32_bits8, aom_highbd_sad8x32_avg_bits8, aom_highbd_8_variance8x32, aom_highbd_8_sub_pixel_variance8x32, aom_highbd_8_sub_pixel_avg_variance8x32, aom_highbd_sad8x32x4d_bits8, aom_highbd_dist_wtd_sad8x32_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance8x32) HIGHBD_BFP( BLOCK_16X4, aom_highbd_sad16x4_bits8, aom_highbd_sad16x4_avg_bits8, aom_highbd_8_variance16x4, aom_highbd_8_sub_pixel_variance16x4, aom_highbd_8_sub_pixel_avg_variance16x4, aom_highbd_sad16x4x4d_bits8, aom_highbd_dist_wtd_sad16x4_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance16x4) HIGHBD_BFP( BLOCK_4X16, aom_highbd_sad4x16_bits8, aom_highbd_sad4x16_avg_bits8, aom_highbd_8_variance4x16, aom_highbd_8_sub_pixel_variance4x16, aom_highbd_8_sub_pixel_avg_variance4x16, aom_highbd_sad4x16x4d_bits8, aom_highbd_dist_wtd_sad4x16_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance4x16) HIGHBD_BFP(BLOCK_32X16, aom_highbd_sad32x16_bits8, aom_highbd_sad32x16_avg_bits8, aom_highbd_8_variance32x16, aom_highbd_8_sub_pixel_variance32x16, aom_highbd_8_sub_pixel_avg_variance32x16, aom_highbd_sad32x16x4d_bits8, aom_highbd_dist_wtd_sad32x16_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance32x16) HIGHBD_BFP(BLOCK_16X32, aom_highbd_sad16x32_bits8, aom_highbd_sad16x32_avg_bits8, aom_highbd_8_variance16x32, aom_highbd_8_sub_pixel_variance16x32, aom_highbd_8_sub_pixel_avg_variance16x32, aom_highbd_sad16x32x4d_bits8, aom_highbd_dist_wtd_sad16x32_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance16x32) HIGHBD_BFP(BLOCK_64X32, aom_highbd_sad64x32_bits8, aom_highbd_sad64x32_avg_bits8, aom_highbd_8_variance64x32, aom_highbd_8_sub_pixel_variance64x32, aom_highbd_8_sub_pixel_avg_variance64x32, aom_highbd_sad64x32x4d_bits8, aom_highbd_dist_wtd_sad64x32_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance64x32) HIGHBD_BFP(BLOCK_32X64, aom_highbd_sad32x64_bits8, aom_highbd_sad32x64_avg_bits8, aom_highbd_8_variance32x64, aom_highbd_8_sub_pixel_variance32x64, aom_highbd_8_sub_pixel_avg_variance32x64, aom_highbd_sad32x64x4d_bits8, aom_highbd_dist_wtd_sad32x64_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance32x64) HIGHBD_BFP(BLOCK_32X32, aom_highbd_sad32x32_bits8, aom_highbd_sad32x32_avg_bits8, aom_highbd_8_variance32x32, aom_highbd_8_sub_pixel_variance32x32, aom_highbd_8_sub_pixel_avg_variance32x32, aom_highbd_sad32x32x4d_bits8, aom_highbd_dist_wtd_sad32x32_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance32x32) HIGHBD_BFP(BLOCK_64X64, aom_highbd_sad64x64_bits8, aom_highbd_sad64x64_avg_bits8, aom_highbd_8_variance64x64, aom_highbd_8_sub_pixel_variance64x64, aom_highbd_8_sub_pixel_avg_variance64x64, aom_highbd_sad64x64x4d_bits8, aom_highbd_dist_wtd_sad64x64_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance64x64) HIGHBD_BFP(BLOCK_16X16, aom_highbd_sad16x16_bits8, aom_highbd_sad16x16_avg_bits8, aom_highbd_8_variance16x16, aom_highbd_8_sub_pixel_variance16x16, aom_highbd_8_sub_pixel_avg_variance16x16, aom_highbd_sad16x16x4d_bits8, aom_highbd_dist_wtd_sad16x16_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance16x16) HIGHBD_BFP( BLOCK_16X8, aom_highbd_sad16x8_bits8, aom_highbd_sad16x8_avg_bits8, aom_highbd_8_variance16x8, aom_highbd_8_sub_pixel_variance16x8, aom_highbd_8_sub_pixel_avg_variance16x8, aom_highbd_sad16x8x4d_bits8, aom_highbd_dist_wtd_sad16x8_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance16x8) HIGHBD_BFP( BLOCK_8X16, aom_highbd_sad8x16_bits8, aom_highbd_sad8x16_avg_bits8, aom_highbd_8_variance8x16, aom_highbd_8_sub_pixel_variance8x16, aom_highbd_8_sub_pixel_avg_variance8x16, aom_highbd_sad8x16x4d_bits8, aom_highbd_dist_wtd_sad8x16_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance8x16) HIGHBD_BFP( BLOCK_8X8, aom_highbd_sad8x8_bits8, aom_highbd_sad8x8_avg_bits8, aom_highbd_8_variance8x8, aom_highbd_8_sub_pixel_variance8x8, aom_highbd_8_sub_pixel_avg_variance8x8, aom_highbd_sad8x8x4d_bits8, aom_highbd_dist_wtd_sad8x8_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance8x8) HIGHBD_BFP( BLOCK_8X4, aom_highbd_sad8x4_bits8, aom_highbd_sad8x4_avg_bits8, aom_highbd_8_variance8x4, aom_highbd_8_sub_pixel_variance8x4, aom_highbd_8_sub_pixel_avg_variance8x4, aom_highbd_sad8x4x4d_bits8, aom_highbd_dist_wtd_sad8x4_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance8x4) HIGHBD_BFP( BLOCK_4X8, aom_highbd_sad4x8_bits8, aom_highbd_sad4x8_avg_bits8, aom_highbd_8_variance4x8, aom_highbd_8_sub_pixel_variance4x8, aom_highbd_8_sub_pixel_avg_variance4x8, aom_highbd_sad4x8x4d_bits8, aom_highbd_dist_wtd_sad4x8_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance4x8) HIGHBD_BFP( BLOCK_4X4, aom_highbd_sad4x4_bits8, aom_highbd_sad4x4_avg_bits8, aom_highbd_8_variance4x4, aom_highbd_8_sub_pixel_variance4x4, aom_highbd_8_sub_pixel_avg_variance4x4, aom_highbd_sad4x4x4d_bits8, aom_highbd_dist_wtd_sad4x4_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance4x4) HIGHBD_BFP(BLOCK_128X128, aom_highbd_sad128x128_bits8, aom_highbd_sad128x128_avg_bits8, aom_highbd_8_variance128x128, aom_highbd_8_sub_pixel_variance128x128, aom_highbd_8_sub_pixel_avg_variance128x128, aom_highbd_sad128x128x4d_bits8, aom_highbd_dist_wtd_sad128x128_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance128x128) HIGHBD_BFP(BLOCK_128X64, aom_highbd_sad128x64_bits8, aom_highbd_sad128x64_avg_bits8, aom_highbd_8_variance128x64, aom_highbd_8_sub_pixel_variance128x64, aom_highbd_8_sub_pixel_avg_variance128x64, aom_highbd_sad128x64x4d_bits8, aom_highbd_dist_wtd_sad128x64_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance128x64) HIGHBD_BFP(BLOCK_64X128, aom_highbd_sad64x128_bits8, aom_highbd_sad64x128_avg_bits8, aom_highbd_8_variance64x128, aom_highbd_8_sub_pixel_variance64x128, aom_highbd_8_sub_pixel_avg_variance64x128, aom_highbd_sad64x128x4d_bits8, aom_highbd_dist_wtd_sad64x128_avg_bits8, aom_highbd_8_dist_wtd_sub_pixel_avg_variance64x128) HIGHBD_MBFP(BLOCK_128X128, aom_highbd_masked_sad128x128_bits8, aom_highbd_8_masked_sub_pixel_variance128x128) HIGHBD_MBFP(BLOCK_128X64, aom_highbd_masked_sad128x64_bits8, aom_highbd_8_masked_sub_pixel_variance128x64) HIGHBD_MBFP(BLOCK_64X128, aom_highbd_masked_sad64x128_bits8, aom_highbd_8_masked_sub_pixel_variance64x128) HIGHBD_MBFP(BLOCK_64X64, aom_highbd_masked_sad64x64_bits8, aom_highbd_8_masked_sub_pixel_variance64x64) HIGHBD_MBFP(BLOCK_64X32, aom_highbd_masked_sad64x32_bits8, aom_highbd_8_masked_sub_pixel_variance64x32) HIGHBD_MBFP(BLOCK_32X64, aom_highbd_masked_sad32x64_bits8, aom_highbd_8_masked_sub_pixel_variance32x64) HIGHBD_MBFP(BLOCK_32X32, aom_highbd_masked_sad32x32_bits8, aom_highbd_8_masked_sub_pixel_variance32x32) HIGHBD_MBFP(BLOCK_32X16, aom_highbd_masked_sad32x16_bits8, aom_highbd_8_masked_sub_pixel_variance32x16) HIGHBD_MBFP(BLOCK_16X32, aom_highbd_masked_sad16x32_bits8, aom_highbd_8_masked_sub_pixel_variance16x32) HIGHBD_MBFP(BLOCK_16X16, aom_highbd_masked_sad16x16_bits8, aom_highbd_8_masked_sub_pixel_variance16x16) HIGHBD_MBFP(BLOCK_8X16, aom_highbd_masked_sad8x16_bits8, aom_highbd_8_masked_sub_pixel_variance8x16) HIGHBD_MBFP(BLOCK_16X8, aom_highbd_masked_sad16x8_bits8, aom_highbd_8_masked_sub_pixel_variance16x8) HIGHBD_MBFP(BLOCK_8X8, aom_highbd_masked_sad8x8_bits8, aom_highbd_8_masked_sub_pixel_variance8x8) HIGHBD_MBFP(BLOCK_4X8, aom_highbd_masked_sad4x8_bits8, aom_highbd_8_masked_sub_pixel_variance4x8) HIGHBD_MBFP(BLOCK_8X4, aom_highbd_masked_sad8x4_bits8, aom_highbd_8_masked_sub_pixel_variance8x4) HIGHBD_MBFP(BLOCK_4X4, aom_highbd_masked_sad4x4_bits8, aom_highbd_8_masked_sub_pixel_variance4x4) HIGHBD_MBFP(BLOCK_64X16, aom_highbd_masked_sad64x16_bits8, aom_highbd_8_masked_sub_pixel_variance64x16) HIGHBD_MBFP(BLOCK_16X64, aom_highbd_masked_sad16x64_bits8, aom_highbd_8_masked_sub_pixel_variance16x64) HIGHBD_MBFP(BLOCK_32X8, aom_highbd_masked_sad32x8_bits8, aom_highbd_8_masked_sub_pixel_variance32x8) HIGHBD_MBFP(BLOCK_8X32, aom_highbd_masked_sad8x32_bits8, aom_highbd_8_masked_sub_pixel_variance8x32) HIGHBD_MBFP(BLOCK_16X4, aom_highbd_masked_sad16x4_bits8, aom_highbd_8_masked_sub_pixel_variance16x4) HIGHBD_MBFP(BLOCK_4X16, aom_highbd_masked_sad4x16_bits8, aom_highbd_8_masked_sub_pixel_variance4x16) HIGHBD_OBFP(BLOCK_128X128, aom_highbd_obmc_sad128x128_bits8, aom_highbd_obmc_variance128x128, aom_highbd_obmc_sub_pixel_variance128x128) HIGHBD_OBFP(BLOCK_128X64, aom_highbd_obmc_sad128x64_bits8, aom_highbd_obmc_variance128x64, aom_highbd_obmc_sub_pixel_variance128x64) HIGHBD_OBFP(BLOCK_64X128, aom_highbd_obmc_sad64x128_bits8, aom_highbd_obmc_variance64x128, aom_highbd_obmc_sub_pixel_variance64x128) HIGHBD_OBFP(BLOCK_64X64, aom_highbd_obmc_sad64x64_bits8, aom_highbd_obmc_variance64x64, aom_highbd_obmc_sub_pixel_variance64x64) HIGHBD_OBFP(BLOCK_64X32, aom_highbd_obmc_sad64x32_bits8, aom_highbd_obmc_variance64x32, aom_highbd_obmc_sub_pixel_variance64x32) HIGHBD_OBFP(BLOCK_32X64, aom_highbd_obmc_sad32x64_bits8, aom_highbd_obmc_variance32x64, aom_highbd_obmc_sub_pixel_variance32x64) HIGHBD_OBFP(BLOCK_32X32, aom_highbd_obmc_sad32x32_bits8, aom_highbd_obmc_variance32x32, aom_highbd_obmc_sub_pixel_variance32x32) HIGHBD_OBFP(BLOCK_32X16, aom_highbd_obmc_sad32x16_bits8, aom_highbd_obmc_variance32x16, aom_highbd_obmc_sub_pixel_variance32x16) HIGHBD_OBFP(BLOCK_16X32, aom_highbd_obmc_sad16x32_bits8, aom_highbd_obmc_variance16x32, aom_highbd_obmc_sub_pixel_variance16x32) HIGHBD_OBFP(BLOCK_16X16, aom_highbd_obmc_sad16x16_bits8, aom_highbd_obmc_variance16x16, aom_highbd_obmc_sub_pixel_variance16x16) HIGHBD_OBFP(BLOCK_8X16, aom_highbd_obmc_sad8x16_bits8, aom_highbd_obmc_variance8x16, aom_highbd_obmc_sub_pixel_variance8x16) HIGHBD_OBFP(BLOCK_16X8, aom_highbd_obmc_sad16x8_bits8, aom_highbd_obmc_variance16x8, aom_highbd_obmc_sub_pixel_variance16x8) HIGHBD_OBFP(BLOCK_8X8, aom_highbd_obmc_sad8x8_bits8, aom_highbd_obmc_variance8x8, aom_highbd_obmc_sub_pixel_variance8x8) HIGHBD_OBFP(BLOCK_4X8, aom_highbd_obmc_sad4x8_bits8, aom_highbd_obmc_variance4x8, aom_highbd_obmc_sub_pixel_variance4x8) HIGHBD_OBFP(BLOCK_8X4, aom_highbd_obmc_sad8x4_bits8, aom_highbd_obmc_variance8x4, aom_highbd_obmc_sub_pixel_variance8x4) HIGHBD_OBFP(BLOCK_4X4, aom_highbd_obmc_sad4x4_bits8, aom_highbd_obmc_variance4x4, aom_highbd_obmc_sub_pixel_variance4x4) HIGHBD_OBFP(BLOCK_64X16, aom_highbd_obmc_sad64x16_bits8, aom_highbd_obmc_variance64x16, aom_highbd_obmc_sub_pixel_variance64x16) HIGHBD_OBFP(BLOCK_16X64, aom_highbd_obmc_sad16x64_bits8, aom_highbd_obmc_variance16x64, aom_highbd_obmc_sub_pixel_variance16x64) HIGHBD_OBFP(BLOCK_32X8, aom_highbd_obmc_sad32x8_bits8, aom_highbd_obmc_variance32x8, aom_highbd_obmc_sub_pixel_variance32x8) HIGHBD_OBFP(BLOCK_8X32, aom_highbd_obmc_sad8x32_bits8, aom_highbd_obmc_variance8x32, aom_highbd_obmc_sub_pixel_variance8x32) HIGHBD_OBFP(BLOCK_16X4, aom_highbd_obmc_sad16x4_bits8, aom_highbd_obmc_variance16x4, aom_highbd_obmc_sub_pixel_variance16x4) HIGHBD_OBFP(BLOCK_4X16, aom_highbd_obmc_sad4x16_bits8, aom_highbd_obmc_variance4x16, aom_highbd_obmc_sub_pixel_variance4x16) break; case AOM_BITS_10: HIGHBD_BFP(BLOCK_64X16, aom_highbd_sad64x16_bits10, aom_highbd_sad64x16_avg_bits10, aom_highbd_10_variance64x16, aom_highbd_10_sub_pixel_variance64x16, aom_highbd_10_sub_pixel_avg_variance64x16, aom_highbd_sad64x16x4d_bits10, aom_highbd_dist_wtd_sad64x16_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance64x16); HIGHBD_BFP(BLOCK_16X64, aom_highbd_sad16x64_bits10, aom_highbd_sad16x64_avg_bits10, aom_highbd_10_variance16x64, aom_highbd_10_sub_pixel_variance16x64, aom_highbd_10_sub_pixel_avg_variance16x64, aom_highbd_sad16x64x4d_bits10, aom_highbd_dist_wtd_sad16x64_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance16x64); HIGHBD_BFP(BLOCK_32X8, aom_highbd_sad32x8_bits10, aom_highbd_sad32x8_avg_bits10, aom_highbd_10_variance32x8, aom_highbd_10_sub_pixel_variance32x8, aom_highbd_10_sub_pixel_avg_variance32x8, aom_highbd_sad32x8x4d_bits10, aom_highbd_dist_wtd_sad32x8_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance32x8); HIGHBD_BFP(BLOCK_8X32, aom_highbd_sad8x32_bits10, aom_highbd_sad8x32_avg_bits10, aom_highbd_10_variance8x32, aom_highbd_10_sub_pixel_variance8x32, aom_highbd_10_sub_pixel_avg_variance8x32, aom_highbd_sad8x32x4d_bits10, aom_highbd_dist_wtd_sad8x32_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance8x32); HIGHBD_BFP(BLOCK_16X4, aom_highbd_sad16x4_bits10, aom_highbd_sad16x4_avg_bits10, aom_highbd_10_variance16x4, aom_highbd_10_sub_pixel_variance16x4, aom_highbd_10_sub_pixel_avg_variance16x4, aom_highbd_sad16x4x4d_bits10, aom_highbd_dist_wtd_sad16x4_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance16x4); HIGHBD_BFP(BLOCK_4X16, aom_highbd_sad4x16_bits10, aom_highbd_sad4x16_avg_bits10, aom_highbd_10_variance4x16, aom_highbd_10_sub_pixel_variance4x16, aom_highbd_10_sub_pixel_avg_variance4x16, aom_highbd_sad4x16x4d_bits10, aom_highbd_dist_wtd_sad4x16_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance4x16); HIGHBD_BFP(BLOCK_32X16, aom_highbd_sad32x16_bits10, aom_highbd_sad32x16_avg_bits10, aom_highbd_10_variance32x16, aom_highbd_10_sub_pixel_variance32x16, aom_highbd_10_sub_pixel_avg_variance32x16, aom_highbd_sad32x16x4d_bits10, aom_highbd_dist_wtd_sad32x16_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance32x16); HIGHBD_BFP(BLOCK_16X32, aom_highbd_sad16x32_bits10, aom_highbd_sad16x32_avg_bits10, aom_highbd_10_variance16x32, aom_highbd_10_sub_pixel_variance16x32, aom_highbd_10_sub_pixel_avg_variance16x32, aom_highbd_sad16x32x4d_bits10, aom_highbd_dist_wtd_sad16x32_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance16x32); HIGHBD_BFP(BLOCK_64X32, aom_highbd_sad64x32_bits10, aom_highbd_sad64x32_avg_bits10, aom_highbd_10_variance64x32, aom_highbd_10_sub_pixel_variance64x32, aom_highbd_10_sub_pixel_avg_variance64x32, aom_highbd_sad64x32x4d_bits10, aom_highbd_dist_wtd_sad64x32_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance64x32); HIGHBD_BFP(BLOCK_32X64, aom_highbd_sad32x64_bits10, aom_highbd_sad32x64_avg_bits10, aom_highbd_10_variance32x64, aom_highbd_10_sub_pixel_variance32x64, aom_highbd_10_sub_pixel_avg_variance32x64, aom_highbd_sad32x64x4d_bits10, aom_highbd_dist_wtd_sad32x64_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance32x64); HIGHBD_BFP(BLOCK_32X32, aom_highbd_sad32x32_bits10, aom_highbd_sad32x32_avg_bits10, aom_highbd_10_variance32x32, aom_highbd_10_sub_pixel_variance32x32, aom_highbd_10_sub_pixel_avg_variance32x32, aom_highbd_sad32x32x4d_bits10, aom_highbd_dist_wtd_sad32x32_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance32x32); HIGHBD_BFP(BLOCK_64X64, aom_highbd_sad64x64_bits10, aom_highbd_sad64x64_avg_bits10, aom_highbd_10_variance64x64, aom_highbd_10_sub_pixel_variance64x64, aom_highbd_10_sub_pixel_avg_variance64x64, aom_highbd_sad64x64x4d_bits10, aom_highbd_dist_wtd_sad64x64_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance64x64); HIGHBD_BFP(BLOCK_16X16, aom_highbd_sad16x16_bits10, aom_highbd_sad16x16_avg_bits10, aom_highbd_10_variance16x16, aom_highbd_10_sub_pixel_variance16x16, aom_highbd_10_sub_pixel_avg_variance16x16, aom_highbd_sad16x16x4d_bits10, aom_highbd_dist_wtd_sad16x16_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance16x16); HIGHBD_BFP(BLOCK_16X8, aom_highbd_sad16x8_bits10, aom_highbd_sad16x8_avg_bits10, aom_highbd_10_variance16x8, aom_highbd_10_sub_pixel_variance16x8, aom_highbd_10_sub_pixel_avg_variance16x8, aom_highbd_sad16x8x4d_bits10, aom_highbd_dist_wtd_sad16x8_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance16x8); HIGHBD_BFP(BLOCK_8X16, aom_highbd_sad8x16_bits10, aom_highbd_sad8x16_avg_bits10, aom_highbd_10_variance8x16, aom_highbd_10_sub_pixel_variance8x16, aom_highbd_10_sub_pixel_avg_variance8x16, aom_highbd_sad8x16x4d_bits10, aom_highbd_dist_wtd_sad8x16_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance8x16); HIGHBD_BFP( BLOCK_8X8, aom_highbd_sad8x8_bits10, aom_highbd_sad8x8_avg_bits10, aom_highbd_10_variance8x8, aom_highbd_10_sub_pixel_variance8x8, aom_highbd_10_sub_pixel_avg_variance8x8, aom_highbd_sad8x8x4d_bits10, aom_highbd_dist_wtd_sad8x8_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance8x8); HIGHBD_BFP( BLOCK_8X4, aom_highbd_sad8x4_bits10, aom_highbd_sad8x4_avg_bits10, aom_highbd_10_variance8x4, aom_highbd_10_sub_pixel_variance8x4, aom_highbd_10_sub_pixel_avg_variance8x4, aom_highbd_sad8x4x4d_bits10, aom_highbd_dist_wtd_sad8x4_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance8x4); HIGHBD_BFP( BLOCK_4X8, aom_highbd_sad4x8_bits10, aom_highbd_sad4x8_avg_bits10, aom_highbd_10_variance4x8, aom_highbd_10_sub_pixel_variance4x8, aom_highbd_10_sub_pixel_avg_variance4x8, aom_highbd_sad4x8x4d_bits10, aom_highbd_dist_wtd_sad4x8_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance4x8); HIGHBD_BFP( BLOCK_4X4, aom_highbd_sad4x4_bits10, aom_highbd_sad4x4_avg_bits10, aom_highbd_10_variance4x4, aom_highbd_10_sub_pixel_variance4x4, aom_highbd_10_sub_pixel_avg_variance4x4, aom_highbd_sad4x4x4d_bits10, aom_highbd_dist_wtd_sad4x4_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance4x4); HIGHBD_BFP(BLOCK_128X128, aom_highbd_sad128x128_bits10, aom_highbd_sad128x128_avg_bits10, aom_highbd_10_variance128x128, aom_highbd_10_sub_pixel_variance128x128, aom_highbd_10_sub_pixel_avg_variance128x128, aom_highbd_sad128x128x4d_bits10, aom_highbd_dist_wtd_sad128x128_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance128x128); HIGHBD_BFP(BLOCK_128X64, aom_highbd_sad128x64_bits10, aom_highbd_sad128x64_avg_bits10, aom_highbd_10_variance128x64, aom_highbd_10_sub_pixel_variance128x64, aom_highbd_10_sub_pixel_avg_variance128x64, aom_highbd_sad128x64x4d_bits10, aom_highbd_dist_wtd_sad128x64_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance128x64); HIGHBD_BFP(BLOCK_64X128, aom_highbd_sad64x128_bits10, aom_highbd_sad64x128_avg_bits10, aom_highbd_10_variance64x128, aom_highbd_10_sub_pixel_variance64x128, aom_highbd_10_sub_pixel_avg_variance64x128, aom_highbd_sad64x128x4d_bits10, aom_highbd_dist_wtd_sad64x128_avg_bits10, aom_highbd_10_dist_wtd_sub_pixel_avg_variance64x128); HIGHBD_MBFP(BLOCK_128X128, aom_highbd_masked_sad128x128_bits10, aom_highbd_10_masked_sub_pixel_variance128x128) HIGHBD_MBFP(BLOCK_128X64, aom_highbd_masked_sad128x64_bits10, aom_highbd_10_masked_sub_pixel_variance128x64) HIGHBD_MBFP(BLOCK_64X128, aom_highbd_masked_sad64x128_bits10, aom_highbd_10_masked_sub_pixel_variance64x128) HIGHBD_MBFP(BLOCK_64X64, aom_highbd_masked_sad64x64_bits10, aom_highbd_10_masked_sub_pixel_variance64x64) HIGHBD_MBFP(BLOCK_64X32, aom_highbd_masked_sad64x32_bits10, aom_highbd_10_masked_sub_pixel_variance64x32) HIGHBD_MBFP(BLOCK_32X64, aom_highbd_masked_sad32x64_bits10, aom_highbd_10_masked_sub_pixel_variance32x64) HIGHBD_MBFP(BLOCK_32X32, aom_highbd_masked_sad32x32_bits10, aom_highbd_10_masked_sub_pixel_variance32x32) HIGHBD_MBFP(BLOCK_32X16, aom_highbd_masked_sad32x16_bits10, aom_highbd_10_masked_sub_pixel_variance32x16) HIGHBD_MBFP(BLOCK_16X32, aom_highbd_masked_sad16x32_bits10, aom_highbd_10_masked_sub_pixel_variance16x32) HIGHBD_MBFP(BLOCK_16X16, aom_highbd_masked_sad16x16_bits10, aom_highbd_10_masked_sub_pixel_variance16x16) HIGHBD_MBFP(BLOCK_8X16, aom_highbd_masked_sad8x16_bits10, aom_highbd_10_masked_sub_pixel_variance8x16) HIGHBD_MBFP(BLOCK_16X8, aom_highbd_masked_sad16x8_bits10, aom_highbd_10_masked_sub_pixel_variance16x8) HIGHBD_MBFP(BLOCK_8X8, aom_highbd_masked_sad8x8_bits10, aom_highbd_10_masked_sub_pixel_variance8x8) HIGHBD_MBFP(BLOCK_4X8, aom_highbd_masked_sad4x8_bits10, aom_highbd_10_masked_sub_pixel_variance4x8) HIGHBD_MBFP(BLOCK_8X4, aom_highbd_masked_sad8x4_bits10, aom_highbd_10_masked_sub_pixel_variance8x4) HIGHBD_MBFP(BLOCK_4X4, aom_highbd_masked_sad4x4_bits10, aom_highbd_10_masked_sub_pixel_variance4x4) HIGHBD_MBFP(BLOCK_64X16, aom_highbd_masked_sad64x16_bits10, aom_highbd_10_masked_sub_pixel_variance64x16) HIGHBD_MBFP(BLOCK_16X64, aom_highbd_masked_sad16x64_bits10, aom_highbd_10_masked_sub_pixel_variance16x64) HIGHBD_MBFP(BLOCK_32X8, aom_highbd_masked_sad32x8_bits10, aom_highbd_10_masked_sub_pixel_variance32x8) HIGHBD_MBFP(BLOCK_8X32, aom_highbd_masked_sad8x32_bits10, aom_highbd_10_masked_sub_pixel_variance8x32) HIGHBD_MBFP(BLOCK_16X4, aom_highbd_masked_sad16x4_bits10, aom_highbd_10_masked_sub_pixel_variance16x4) HIGHBD_MBFP(BLOCK_4X16, aom_highbd_masked_sad4x16_bits10, aom_highbd_10_masked_sub_pixel_variance4x16) HIGHBD_OBFP(BLOCK_128X128, aom_highbd_obmc_sad128x128_bits10, aom_highbd_10_obmc_variance128x128, aom_highbd_10_obmc_sub_pixel_variance128x128) HIGHBD_OBFP(BLOCK_128X64, aom_highbd_obmc_sad128x64_bits10, aom_highbd_10_obmc_variance128x64, aom_highbd_10_obmc_sub_pixel_variance128x64) HIGHBD_OBFP(BLOCK_64X128, aom_highbd_obmc_sad64x128_bits10, aom_highbd_10_obmc_variance64x128, aom_highbd_10_obmc_sub_pixel_variance64x128) HIGHBD_OBFP(BLOCK_64X64, aom_highbd_obmc_sad64x64_bits10, aom_highbd_10_obmc_variance64x64, aom_highbd_10_obmc_sub_pixel_variance64x64) HIGHBD_OBFP(BLOCK_64X32, aom_highbd_obmc_sad64x32_bits10, aom_highbd_10_obmc_variance64x32, aom_highbd_10_obmc_sub_pixel_variance64x32) HIGHBD_OBFP(BLOCK_32X64, aom_highbd_obmc_sad32x64_bits10, aom_highbd_10_obmc_variance32x64, aom_highbd_10_obmc_sub_pixel_variance32x64) HIGHBD_OBFP(BLOCK_32X32, aom_highbd_obmc_sad32x32_bits10, aom_highbd_10_obmc_variance32x32, aom_highbd_10_obmc_sub_pixel_variance32x32) HIGHBD_OBFP(BLOCK_32X16, aom_highbd_obmc_sad32x16_bits10, aom_highbd_10_obmc_variance32x16, aom_highbd_10_obmc_sub_pixel_variance32x16) HIGHBD_OBFP(BLOCK_16X32, aom_highbd_obmc_sad16x32_bits10, aom_highbd_10_obmc_variance16x32, aom_highbd_10_obmc_sub_pixel_variance16x32) HIGHBD_OBFP(BLOCK_16X16, aom_highbd_obmc_sad16x16_bits10, aom_highbd_10_obmc_variance16x16, aom_highbd_10_obmc_sub_pixel_variance16x16) HIGHBD_OBFP(BLOCK_8X16, aom_highbd_obmc_sad8x16_bits10, aom_highbd_10_obmc_variance8x16, aom_highbd_10_obmc_sub_pixel_variance8x16) HIGHBD_OBFP(BLOCK_16X8, aom_highbd_obmc_sad16x8_bits10, aom_highbd_10_obmc_variance16x8, aom_highbd_10_obmc_sub_pixel_variance16x8) HIGHBD_OBFP(BLOCK_8X8, aom_highbd_obmc_sad8x8_bits10, aom_highbd_10_obmc_variance8x8, aom_highbd_10_obmc_sub_pixel_variance8x8) HIGHBD_OBFP(BLOCK_4X8, aom_highbd_obmc_sad4x8_bits10, aom_highbd_10_obmc_variance4x8, aom_highbd_10_obmc_sub_pixel_variance4x8) HIGHBD_OBFP(BLOCK_8X4, aom_highbd_obmc_sad8x4_bits10, aom_highbd_10_obmc_variance8x4, aom_highbd_10_obmc_sub_pixel_variance8x4) HIGHBD_OBFP(BLOCK_4X4, aom_highbd_obmc_sad4x4_bits10, aom_highbd_10_obmc_variance4x4, aom_highbd_10_obmc_sub_pixel_variance4x4) HIGHBD_OBFP(BLOCK_64X16, aom_highbd_obmc_sad64x16_bits10, aom_highbd_10_obmc_variance64x16, aom_highbd_10_obmc_sub_pixel_variance64x16) HIGHBD_OBFP(BLOCK_16X64, aom_highbd_obmc_sad16x64_bits10, aom_highbd_10_obmc_variance16x64, aom_highbd_10_obmc_sub_pixel_variance16x64) HIGHBD_OBFP(BLOCK_32X8, aom_highbd_obmc_sad32x8_bits10, aom_highbd_10_obmc_variance32x8, aom_highbd_10_obmc_sub_pixel_variance32x8) HIGHBD_OBFP(BLOCK_8X32, aom_highbd_obmc_sad8x32_bits10, aom_highbd_10_obmc_variance8x32, aom_highbd_10_obmc_sub_pixel_variance8x32) HIGHBD_OBFP(BLOCK_16X4, aom_highbd_obmc_sad16x4_bits10, aom_highbd_10_obmc_variance16x4, aom_highbd_10_obmc_sub_pixel_variance16x4) HIGHBD_OBFP(BLOCK_4X16, aom_highbd_obmc_sad4x16_bits10, aom_highbd_10_obmc_variance4x16, aom_highbd_10_obmc_sub_pixel_variance4x16) break; case AOM_BITS_12: HIGHBD_BFP(BLOCK_64X16, aom_highbd_sad64x16_bits12, aom_highbd_sad64x16_avg_bits12, aom_highbd_12_variance64x16, aom_highbd_12_sub_pixel_variance64x16, aom_highbd_12_sub_pixel_avg_variance64x16, aom_highbd_sad64x16x4d_bits12, aom_highbd_dist_wtd_sad64x16_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance64x16); HIGHBD_BFP(BLOCK_16X64, aom_highbd_sad16x64_bits12, aom_highbd_sad16x64_avg_bits12, aom_highbd_12_variance16x64, aom_highbd_12_sub_pixel_variance16x64, aom_highbd_12_sub_pixel_avg_variance16x64, aom_highbd_sad16x64x4d_bits12, aom_highbd_dist_wtd_sad16x64_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance16x64); HIGHBD_BFP(BLOCK_32X8, aom_highbd_sad32x8_bits12, aom_highbd_sad32x8_avg_bits12, aom_highbd_12_variance32x8, aom_highbd_12_sub_pixel_variance32x8, aom_highbd_12_sub_pixel_avg_variance32x8, aom_highbd_sad32x8x4d_bits12, aom_highbd_dist_wtd_sad32x8_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance32x8); HIGHBD_BFP(BLOCK_8X32, aom_highbd_sad8x32_bits12, aom_highbd_sad8x32_avg_bits12, aom_highbd_12_variance8x32, aom_highbd_12_sub_pixel_variance8x32, aom_highbd_12_sub_pixel_avg_variance8x32, aom_highbd_sad8x32x4d_bits12, aom_highbd_dist_wtd_sad8x32_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance8x32); HIGHBD_BFP(BLOCK_16X4, aom_highbd_sad16x4_bits12, aom_highbd_sad16x4_avg_bits12, aom_highbd_12_variance16x4, aom_highbd_12_sub_pixel_variance16x4, aom_highbd_12_sub_pixel_avg_variance16x4, aom_highbd_sad16x4x4d_bits12, aom_highbd_dist_wtd_sad16x4_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance16x4); HIGHBD_BFP(BLOCK_4X16, aom_highbd_sad4x16_bits12, aom_highbd_sad4x16_avg_bits12, aom_highbd_12_variance4x16, aom_highbd_12_sub_pixel_variance4x16, aom_highbd_12_sub_pixel_avg_variance4x16, aom_highbd_sad4x16x4d_bits12, aom_highbd_dist_wtd_sad4x16_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance4x16); HIGHBD_BFP(BLOCK_32X16, aom_highbd_sad32x16_bits12, aom_highbd_sad32x16_avg_bits12, aom_highbd_12_variance32x16, aom_highbd_12_sub_pixel_variance32x16, aom_highbd_12_sub_pixel_avg_variance32x16, aom_highbd_sad32x16x4d_bits12, aom_highbd_dist_wtd_sad32x16_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance32x16); HIGHBD_BFP(BLOCK_16X32, aom_highbd_sad16x32_bits12, aom_highbd_sad16x32_avg_bits12, aom_highbd_12_variance16x32, aom_highbd_12_sub_pixel_variance16x32, aom_highbd_12_sub_pixel_avg_variance16x32, aom_highbd_sad16x32x4d_bits12, aom_highbd_dist_wtd_sad16x32_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance16x32); HIGHBD_BFP(BLOCK_64X32, aom_highbd_sad64x32_bits12, aom_highbd_sad64x32_avg_bits12, aom_highbd_12_variance64x32, aom_highbd_12_sub_pixel_variance64x32, aom_highbd_12_sub_pixel_avg_variance64x32, aom_highbd_sad64x32x4d_bits12, aom_highbd_dist_wtd_sad64x32_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance64x32); HIGHBD_BFP(BLOCK_32X64, aom_highbd_sad32x64_bits12, aom_highbd_sad32x64_avg_bits12, aom_highbd_12_variance32x64, aom_highbd_12_sub_pixel_variance32x64, aom_highbd_12_sub_pixel_avg_variance32x64, aom_highbd_sad32x64x4d_bits12, aom_highbd_dist_wtd_sad32x64_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance32x64); HIGHBD_BFP(BLOCK_32X32, aom_highbd_sad32x32_bits12, aom_highbd_sad32x32_avg_bits12, aom_highbd_12_variance32x32, aom_highbd_12_sub_pixel_variance32x32, aom_highbd_12_sub_pixel_avg_variance32x32, aom_highbd_sad32x32x4d_bits12, aom_highbd_dist_wtd_sad32x32_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance32x32); HIGHBD_BFP(BLOCK_64X64, aom_highbd_sad64x64_bits12, aom_highbd_sad64x64_avg_bits12, aom_highbd_12_variance64x64, aom_highbd_12_sub_pixel_variance64x64, aom_highbd_12_sub_pixel_avg_variance64x64, aom_highbd_sad64x64x4d_bits12, aom_highbd_dist_wtd_sad64x64_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance64x64); HIGHBD_BFP(BLOCK_16X16, aom_highbd_sad16x16_bits12, aom_highbd_sad16x16_avg_bits12, aom_highbd_12_variance16x16, aom_highbd_12_sub_pixel_variance16x16, aom_highbd_12_sub_pixel_avg_variance16x16, aom_highbd_sad16x16x4d_bits12, aom_highbd_dist_wtd_sad16x16_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance16x16); HIGHBD_BFP(BLOCK_16X8, aom_highbd_sad16x8_bits12, aom_highbd_sad16x8_avg_bits12, aom_highbd_12_variance16x8, aom_highbd_12_sub_pixel_variance16x8, aom_highbd_12_sub_pixel_avg_variance16x8, aom_highbd_sad16x8x4d_bits12, aom_highbd_dist_wtd_sad16x8_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance16x8); HIGHBD_BFP(BLOCK_8X16, aom_highbd_sad8x16_bits12, aom_highbd_sad8x16_avg_bits12, aom_highbd_12_variance8x16, aom_highbd_12_sub_pixel_variance8x16, aom_highbd_12_sub_pixel_avg_variance8x16, aom_highbd_sad8x16x4d_bits12, aom_highbd_dist_wtd_sad8x16_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance8x16); HIGHBD_BFP( BLOCK_8X8, aom_highbd_sad8x8_bits12, aom_highbd_sad8x8_avg_bits12, aom_highbd_12_variance8x8, aom_highbd_12_sub_pixel_variance8x8, aom_highbd_12_sub_pixel_avg_variance8x8, aom_highbd_sad8x8x4d_bits12, aom_highbd_dist_wtd_sad8x8_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance8x8); HIGHBD_BFP( BLOCK_8X4, aom_highbd_sad8x4_bits12, aom_highbd_sad8x4_avg_bits12, aom_highbd_12_variance8x4, aom_highbd_12_sub_pixel_variance8x4, aom_highbd_12_sub_pixel_avg_variance8x4, aom_highbd_sad8x4x4d_bits12, aom_highbd_dist_wtd_sad8x4_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance8x4); HIGHBD_BFP( BLOCK_4X8, aom_highbd_sad4x8_bits12, aom_highbd_sad4x8_avg_bits12, aom_highbd_12_variance4x8, aom_highbd_12_sub_pixel_variance4x8, aom_highbd_12_sub_pixel_avg_variance4x8, aom_highbd_sad4x8x4d_bits12, aom_highbd_dist_wtd_sad4x8_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance4x8); HIGHBD_BFP( BLOCK_4X4, aom_highbd_sad4x4_bits12, aom_highbd_sad4x4_avg_bits12, aom_highbd_12_variance4x4, aom_highbd_12_sub_pixel_variance4x4, aom_highbd_12_sub_pixel_avg_variance4x4, aom_highbd_sad4x4x4d_bits12, aom_highbd_dist_wtd_sad4x4_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance4x4); HIGHBD_BFP(BLOCK_128X128, aom_highbd_sad128x128_bits12, aom_highbd_sad128x128_avg_bits12, aom_highbd_12_variance128x128, aom_highbd_12_sub_pixel_variance128x128, aom_highbd_12_sub_pixel_avg_variance128x128, aom_highbd_sad128x128x4d_bits12, aom_highbd_dist_wtd_sad128x128_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance128x128); HIGHBD_BFP(BLOCK_128X64, aom_highbd_sad128x64_bits12, aom_highbd_sad128x64_avg_bits12, aom_highbd_12_variance128x64, aom_highbd_12_sub_pixel_variance128x64, aom_highbd_12_sub_pixel_avg_variance128x64, aom_highbd_sad128x64x4d_bits12, aom_highbd_dist_wtd_sad128x64_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance128x64); HIGHBD_BFP(BLOCK_64X128, aom_highbd_sad64x128_bits12, aom_highbd_sad64x128_avg_bits12, aom_highbd_12_variance64x128, aom_highbd_12_sub_pixel_variance64x128, aom_highbd_12_sub_pixel_avg_variance64x128, aom_highbd_sad64x128x4d_bits12, aom_highbd_dist_wtd_sad64x128_avg_bits12, aom_highbd_12_dist_wtd_sub_pixel_avg_variance64x128); HIGHBD_MBFP(BLOCK_128X128, aom_highbd_masked_sad128x128_bits12, aom_highbd_12_masked_sub_pixel_variance128x128) HIGHBD_MBFP(BLOCK_128X64, aom_highbd_masked_sad128x64_bits12, aom_highbd_12_masked_sub_pixel_variance128x64) HIGHBD_MBFP(BLOCK_64X128, aom_highbd_masked_sad64x128_bits12, aom_highbd_12_masked_sub_pixel_variance64x128) HIGHBD_MBFP(BLOCK_64X64, aom_highbd_masked_sad64x64_bits12, aom_highbd_12_masked_sub_pixel_variance64x64) HIGHBD_MBFP(BLOCK_64X32, aom_highbd_masked_sad64x32_bits12, aom_highbd_12_masked_sub_pixel_variance64x32) HIGHBD_MBFP(BLOCK_32X64, aom_highbd_masked_sad32x64_bits12, aom_highbd_12_masked_sub_pixel_variance32x64) HIGHBD_MBFP(BLOCK_32X32, aom_highbd_masked_sad32x32_bits12, aom_highbd_12_masked_sub_pixel_variance32x32) HIGHBD_MBFP(BLOCK_32X16, aom_highbd_masked_sad32x16_bits12, aom_highbd_12_masked_sub_pixel_variance32x16) HIGHBD_MBFP(BLOCK_16X32, aom_highbd_masked_sad16x32_bits12, aom_highbd_12_masked_sub_pixel_variance16x32) HIGHBD_MBFP(BLOCK_16X16, aom_highbd_masked_sad16x16_bits12, aom_highbd_12_masked_sub_pixel_variance16x16) HIGHBD_MBFP(BLOCK_8X16, aom_highbd_masked_sad8x16_bits12, aom_highbd_12_masked_sub_pixel_variance8x16) HIGHBD_MBFP(BLOCK_16X8, aom_highbd_masked_sad16x8_bits12, aom_highbd_12_masked_sub_pixel_variance16x8) HIGHBD_MBFP(BLOCK_8X8, aom_highbd_masked_sad8x8_bits12, aom_highbd_12_masked_sub_pixel_variance8x8) HIGHBD_MBFP(BLOCK_4X8, aom_highbd_masked_sad4x8_bits12, aom_highbd_12_masked_sub_pixel_variance4x8) HIGHBD_MBFP(BLOCK_8X4, aom_highbd_masked_sad8x4_bits12, aom_highbd_12_masked_sub_pixel_variance8x4) HIGHBD_MBFP(BLOCK_4X4, aom_highbd_masked_sad4x4_bits12, aom_highbd_12_masked_sub_pixel_variance4x4) HIGHBD_MBFP(BLOCK_64X16, aom_highbd_masked_sad64x16_bits12, aom_highbd_12_masked_sub_pixel_variance64x16) HIGHBD_MBFP(BLOCK_16X64, aom_highbd_masked_sad16x64_bits12, aom_highbd_12_masked_sub_pixel_variance16x64) HIGHBD_MBFP(BLOCK_32X8, aom_highbd_masked_sad32x8_bits12, aom_highbd_12_masked_sub_pixel_variance32x8) HIGHBD_MBFP(BLOCK_8X32, aom_highbd_masked_sad8x32_bits12, aom_highbd_12_masked_sub_pixel_variance8x32) HIGHBD_MBFP(BLOCK_16X4, aom_highbd_masked_sad16x4_bits12, aom_highbd_12_masked_sub_pixel_variance16x4) HIGHBD_MBFP(BLOCK_4X16, aom_highbd_masked_sad4x16_bits12, aom_highbd_12_masked_sub_pixel_variance4x16) HIGHBD_OBFP(BLOCK_128X128, aom_highbd_obmc_sad128x128_bits12, aom_highbd_12_obmc_variance128x128, aom_highbd_12_obmc_sub_pixel_variance128x128) HIGHBD_OBFP(BLOCK_128X64, aom_highbd_obmc_sad128x64_bits12, aom_highbd_12_obmc_variance128x64, aom_highbd_12_obmc_sub_pixel_variance128x64) HIGHBD_OBFP(BLOCK_64X128, aom_highbd_obmc_sad64x128_bits12, aom_highbd_12_obmc_variance64x128, aom_highbd_12_obmc_sub_pixel_variance64x128) HIGHBD_OBFP(BLOCK_64X64, aom_highbd_obmc_sad64x64_bits12, aom_highbd_12_obmc_variance64x64, aom_highbd_12_obmc_sub_pixel_variance64x64) HIGHBD_OBFP(BLOCK_64X32, aom_highbd_obmc_sad64x32_bits12, aom_highbd_12_obmc_variance64x32, aom_highbd_12_obmc_sub_pixel_variance64x32) HIGHBD_OBFP(BLOCK_32X64, aom_highbd_obmc_sad32x64_bits12, aom_highbd_12_obmc_variance32x64, aom_highbd_12_obmc_sub_pixel_variance32x64) HIGHBD_OBFP(BLOCK_32X32, aom_highbd_obmc_sad32x32_bits12, aom_highbd_12_obmc_variance32x32, aom_highbd_12_obmc_sub_pixel_variance32x32) HIGHBD_OBFP(BLOCK_32X16, aom_highbd_obmc_sad32x16_bits12, aom_highbd_12_obmc_variance32x16, aom_highbd_12_obmc_sub_pixel_variance32x16) HIGHBD_OBFP(BLOCK_16X32, aom_highbd_obmc_sad16x32_bits12, aom_highbd_12_obmc_variance16x32, aom_highbd_12_obmc_sub_pixel_variance16x32) HIGHBD_OBFP(BLOCK_16X16, aom_highbd_obmc_sad16x16_bits12, aom_highbd_12_obmc_variance16x16, aom_highbd_12_obmc_sub_pixel_variance16x16) HIGHBD_OBFP(BLOCK_8X16, aom_highbd_obmc_sad8x16_bits12, aom_highbd_12_obmc_variance8x16, aom_highbd_12_obmc_sub_pixel_variance8x16) HIGHBD_OBFP(BLOCK_16X8, aom_highbd_obmc_sad16x8_bits12, aom_highbd_12_obmc_variance16x8, aom_highbd_12_obmc_sub_pixel_variance16x8) HIGHBD_OBFP(BLOCK_8X8, aom_highbd_obmc_sad8x8_bits12, aom_highbd_12_obmc_variance8x8, aom_highbd_12_obmc_sub_pixel_variance8x8) HIGHBD_OBFP(BLOCK_4X8, aom_highbd_obmc_sad4x8_bits12, aom_highbd_12_obmc_variance4x8, aom_highbd_12_obmc_sub_pixel_variance4x8) HIGHBD_OBFP(BLOCK_8X4, aom_highbd_obmc_sad8x4_bits12, aom_highbd_12_obmc_variance8x4, aom_highbd_12_obmc_sub_pixel_variance8x4) HIGHBD_OBFP(BLOCK_4X4, aom_highbd_obmc_sad4x4_bits12, aom_highbd_12_obmc_variance4x4, aom_highbd_12_obmc_sub_pixel_variance4x4) HIGHBD_OBFP(BLOCK_64X16, aom_highbd_obmc_sad64x16_bits12, aom_highbd_12_obmc_variance64x16, aom_highbd_12_obmc_sub_pixel_variance64x16) HIGHBD_OBFP(BLOCK_16X64, aom_highbd_obmc_sad16x64_bits12, aom_highbd_12_obmc_variance16x64, aom_highbd_12_obmc_sub_pixel_variance16x64) HIGHBD_OBFP(BLOCK_32X8, aom_highbd_obmc_sad32x8_bits12, aom_highbd_12_obmc_variance32x8, aom_highbd_12_obmc_sub_pixel_variance32x8) HIGHBD_OBFP(BLOCK_8X32, aom_highbd_obmc_sad8x32_bits12, aom_highbd_12_obmc_variance8x32, aom_highbd_12_obmc_sub_pixel_variance8x32) HIGHBD_OBFP(BLOCK_16X4, aom_highbd_obmc_sad16x4_bits12, aom_highbd_12_obmc_variance16x4, aom_highbd_12_obmc_sub_pixel_variance16x4) HIGHBD_OBFP(BLOCK_4X16, aom_highbd_obmc_sad4x16_bits12, aom_highbd_12_obmc_variance4x16, aom_highbd_12_obmc_sub_pixel_variance4x16) break; default: assert(0 && "cm->seq_params.bit_depth should be AOM_BITS_8, " "AOM_BITS_10 or AOM_BITS_12"); } } } static void realloc_segmentation_maps(AV1_COMP *cpi) { AV1_COMMON *const cm = &cpi->common; // Create the encoder segmentation map and set all entries to 0 aom_free(cpi->segmentation_map); CHECK_MEM_ERROR(cm, cpi->segmentation_map, aom_calloc(cm->mi_rows * cm->mi_cols, 1)); // Create a map used for cyclic background refresh. if (cpi->cyclic_refresh) av1_cyclic_refresh_free(cpi->cyclic_refresh); CHECK_MEM_ERROR(cm, cpi->cyclic_refresh, av1_cyclic_refresh_alloc(cm->mi_rows, cm->mi_cols)); // Create a map used to mark inactive areas. aom_free(cpi->active_map.map); CHECK_MEM_ERROR(cm, cpi->active_map.map, aom_calloc(cm->mi_rows * cm->mi_cols, 1)); } void av1_change_config(struct AV1_COMP *cpi, const AV1EncoderConfig *oxcf) { AV1_COMMON *const cm = &cpi->common; SequenceHeader *const seq_params = &cm->seq_params; const int num_planes = av1_num_planes(cm); RATE_CONTROL *const rc = &cpi->rc; MACROBLOCK *const x = &cpi->td.mb; if (seq_params->profile != oxcf->profile) seq_params->profile = oxcf->profile; seq_params->bit_depth = oxcf->bit_depth; seq_params->color_primaries = oxcf->color_primaries; seq_params->transfer_characteristics = oxcf->transfer_characteristics; seq_params->matrix_coefficients = oxcf->matrix_coefficients; seq_params->monochrome = oxcf->monochrome; seq_params->chroma_sample_position = oxcf->chroma_sample_position; seq_params->color_range = oxcf->color_range; assert(IMPLIES(seq_params->profile <= PROFILE_1, seq_params->bit_depth <= AOM_BITS_10)); cm->timing_info_present = oxcf->timing_info_present; cm->timing_info.num_units_in_display_tick = oxcf->timing_info.num_units_in_display_tick; cm->timing_info.time_scale = oxcf->timing_info.time_scale; cm->timing_info.equal_picture_interval = oxcf->timing_info.equal_picture_interval; cm->timing_info.num_ticks_per_picture = oxcf->timing_info.num_ticks_per_picture; seq_params->display_model_info_present_flag = oxcf->display_model_info_present_flag; seq_params->decoder_model_info_present_flag = oxcf->decoder_model_info_present_flag; if (oxcf->decoder_model_info_present_flag) { // set the decoder model parameters in schedule mode cm->buffer_model.num_units_in_decoding_tick = oxcf->buffer_model.num_units_in_decoding_tick; cm->buffer_removal_time_present = 1; set_aom_dec_model_info(&cm->buffer_model); set_dec_model_op_parameters(&cm->op_params[0]); } else if (cm->timing_info_present && cm->timing_info.equal_picture_interval && !seq_params->decoder_model_info_present_flag) { // set the decoder model parameters in resource availability mode set_resource_availability_parameters(&cm->op_params[0]); } else { cm->op_params[0].initial_display_delay = 10; // Default value (not signaled) } update_film_grain_parameters(cpi, oxcf); cpi->oxcf = *oxcf; cpi->common.options = oxcf->cfg; x->e_mbd.bd = (int)seq_params->bit_depth; x->e_mbd.global_motion = cm->global_motion; if ((oxcf->pass == 0) && (oxcf->rc_mode == AOM_Q)) { rc->baseline_gf_interval = FIXED_GF_INTERVAL; } else { rc->baseline_gf_interval = (MIN_GF_INTERVAL + MAX_GF_INTERVAL) / 2; } cpi->refresh_last_frame = 1; cpi->refresh_golden_frame = 0; cpi->refresh_bwd_ref_frame = 0; cpi->refresh_alt2_ref_frame = 0; cm->refresh_frame_context = (oxcf->frame_parallel_decoding_mode) ? REFRESH_FRAME_CONTEXT_DISABLED : REFRESH_FRAME_CONTEXT_BACKWARD; if (oxcf->large_scale_tile) cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED; if (x->palette_buffer == NULL) { CHECK_MEM_ERROR(cm, x->palette_buffer, aom_memalign(16, sizeof(*x->palette_buffer))); } if (x->tmp_conv_dst == NULL) { CHECK_MEM_ERROR( cm, x->tmp_conv_dst, aom_memalign(32, MAX_SB_SIZE * MAX_SB_SIZE * sizeof(*x->tmp_conv_dst))); x->e_mbd.tmp_conv_dst = x->tmp_conv_dst; } for (int i = 0; i < 2; ++i) { if (x->tmp_obmc_bufs[i] == NULL) { CHECK_MEM_ERROR(cm, x->tmp_obmc_bufs[i], aom_memalign(32, 2 * MAX_MB_PLANE * MAX_SB_SQUARE * sizeof(*x->tmp_obmc_bufs[i]))); x->e_mbd.tmp_obmc_bufs[i] = x->tmp_obmc_bufs[i]; } } av1_reset_segment_features(cm); set_high_precision_mv(cpi, 1, 0); set_rc_buffer_sizes(rc, &cpi->oxcf); // Under a configuration change, where maximum_buffer_size may change, // keep buffer level clipped to the maximum allowed buffer size. rc->bits_off_target = AOMMIN(rc->bits_off_target, rc->maximum_buffer_size); rc->buffer_level = AOMMIN(rc->buffer_level, rc->maximum_buffer_size); // Set up frame rate and related parameters rate control values. av1_new_framerate(cpi, cpi->framerate); // Set absolute upper and lower quality limits rc->worst_quality = cpi->oxcf.worst_allowed_q; rc->best_quality = cpi->oxcf.best_allowed_q; cm->interp_filter = oxcf->large_scale_tile ? EIGHTTAP_REGULAR : SWITCHABLE; cm->switchable_motion_mode = 1; if (cpi->oxcf.render_width > 0 && cpi->oxcf.render_height > 0) { cm->render_width = cpi->oxcf.render_width; cm->render_height = cpi->oxcf.render_height; } else { cm->render_width = cpi->oxcf.width; cm->render_height = cpi->oxcf.height; } cm->width = cpi->oxcf.width; cm->height = cpi->oxcf.height; int sb_size = seq_params->sb_size; // Superblock size should not be updated after the first key frame. if (!cpi->seq_params_locked) { set_sb_size(&cm->seq_params, select_sb_size(cpi)); } if (cpi->initial_width || sb_size != seq_params->sb_size) { if (cm->width > cpi->initial_width || cm->height > cpi->initial_height || seq_params->sb_size != sb_size) { av1_free_context_buffers(cm); av1_free_pc_tree(&cpi->td, num_planes); alloc_compressor_data(cpi); realloc_segmentation_maps(cpi); cpi->initial_width = cpi->initial_height = 0; } } update_frame_size(cpi); cpi->alt_ref_source = NULL; rc->is_src_frame_alt_ref = 0; rc->is_bwd_ref_frame = 0; rc->is_last_bipred_frame = 0; rc->is_bipred_frame = 0; set_tile_info(cpi); cpi->ext_refresh_frame_flags_pending = 0; cpi->ext_refresh_frame_context_pending = 0; highbd_set_var_fns(cpi); // Init sequence level coding tools // This should not be called after the first key frame. if (!cpi->seq_params_locked) { seq_params->operating_points_cnt_minus_1 = cm->number_spatial_layers > 1 ? cm->number_spatial_layers - 1 : 0; init_seq_coding_tools(&cm->seq_params, cm, oxcf); } } AV1_COMP *av1_create_compressor(AV1EncoderConfig *oxcf, BufferPool *const pool) { unsigned int i; AV1_COMP *volatile const cpi = aom_memalign(32, sizeof(AV1_COMP)); AV1_COMMON *volatile const cm = cpi != NULL ? &cpi->common : NULL; if (!cm) return NULL; av1_zero(*cpi); // The jmp_buf is valid only for the duration of the function that calls // setjmp(). Therefore, this function must reset the 'setjmp' field to 0 // before it returns. if (setjmp(cm->error.jmp)) { cm->error.setjmp = 0; av1_remove_compressor(cpi); return 0; } cm->error.setjmp = 1; cm->alloc_mi = enc_alloc_mi; cm->free_mi = enc_free_mi; cm->setup_mi = enc_setup_mi; CHECK_MEM_ERROR(cm, cm->fc, (FRAME_CONTEXT *)aom_memalign(32, sizeof(*cm->fc))); CHECK_MEM_ERROR( cm, cm->default_frame_context, (FRAME_CONTEXT *)aom_memalign(32, sizeof(*cm->default_frame_context))); memset(cm->fc, 0, sizeof(*cm->fc)); memset(cm->default_frame_context, 0, sizeof(*cm->default_frame_context)); cpi->resize_state = 0; cpi->resize_avg_qp = 0; cpi->resize_buffer_underflow = 0; cpi->common.buffer_pool = pool; init_config(cpi, oxcf); av1_rc_init(&cpi->oxcf, oxcf->pass, &cpi->rc); cm->current_frame.frame_number = 0; cm->current_frame_id = -1; cpi->seq_params_locked = 0; cpi->partition_search_skippable_frame = 0; cpi->tile_data = NULL; cpi->last_show_frame_buf = NULL; realloc_segmentation_maps(cpi); memset(cpi->nmv_costs, 0, sizeof(cpi->nmv_costs)); memset(cpi->nmv_costs_hp, 0, sizeof(cpi->nmv_costs_hp)); for (i = 0; i < (sizeof(cpi->mbgraph_stats) / sizeof(cpi->mbgraph_stats[0])); i++) { CHECK_MEM_ERROR( cm, cpi->mbgraph_stats[i].mb_stats, aom_calloc(cm->MBs * sizeof(*cpi->mbgraph_stats[i].mb_stats), 1)); } #if CONFIG_FP_MB_STATS cpi->use_fp_mb_stats = 0; if (cpi->use_fp_mb_stats) { // a place holder used to store the first pass mb stats in the first pass CHECK_MEM_ERROR(cm, cpi->twopass.frame_mb_stats_buf, aom_calloc(cm->MBs * sizeof(uint8_t), 1)); } else { cpi->twopass.frame_mb_stats_buf = NULL; } #endif cpi->refresh_alt_ref_frame = 0; cpi->b_calculate_psnr = CONFIG_INTERNAL_STATS; #if CONFIG_INTERNAL_STATS cpi->b_calculate_blockiness = 1; cpi->b_calculate_consistency = 1; cpi->total_inconsistency = 0; cpi->psnr.worst = 100.0; cpi->worst_ssim = 100.0; cpi->count = 0; cpi->bytes = 0; #if CONFIG_SPEED_STATS cpi->tx_search_count = 0; #endif // CONFIG_SPEED_STATS if (cpi->b_calculate_psnr) { cpi->total_sq_error = 0; cpi->total_samples = 0; cpi->tot_recode_hits = 0; cpi->summed_quality = 0; cpi->summed_weights = 0; } cpi->fastssim.worst = 100.0; cpi->psnrhvs.worst = 100.0; if (cpi->b_calculate_blockiness) { cpi->total_blockiness = 0; cpi->worst_blockiness = 0.0; } if (cpi->b_calculate_consistency) { CHECK_MEM_ERROR(cm, cpi->ssim_vars, aom_malloc(sizeof(*cpi->ssim_vars) * 4 * cpi->common.mi_rows * cpi->common.mi_cols)); cpi->worst_consistency = 100.0; } #endif #if CONFIG_ENTROPY_STATS av1_zero(aggregate_fc); #endif // CONFIG_ENTROPY_STATS cpi->first_time_stamp_ever = INT64_MAX; cpi->td.mb.nmvcost[0] = &cpi->nmv_costs[0][MV_MAX]; cpi->td.mb.nmvcost[1] = &cpi->nmv_costs[1][MV_MAX]; cpi->td.mb.nmvcost_hp[0] = &cpi->nmv_costs_hp[0][MV_MAX]; cpi->td.mb.nmvcost_hp[1] = &cpi->nmv_costs_hp[1][MV_MAX]; #ifdef OUTPUT_YUV_SKINMAP yuv_skinmap_file = fopen("skinmap.yuv", "ab"); #endif #ifdef OUTPUT_YUV_REC yuv_rec_file = fopen("rec.yuv", "wb"); #endif if (oxcf->pass == 1) { av1_init_first_pass(cpi); } else if (oxcf->pass == 2) { const size_t packet_sz = sizeof(FIRSTPASS_STATS); const int packets = (int)(oxcf->two_pass_stats_in.sz / packet_sz); #if CONFIG_FP_MB_STATS if (cpi->use_fp_mb_stats) { const size_t psz = cpi->common.MBs * sizeof(uint8_t); const int ps = (int)(oxcf->firstpass_mb_stats_in.sz / psz); cpi->twopass.firstpass_mb_stats.mb_stats_start = oxcf->firstpass_mb_stats_in.buf; cpi->twopass.firstpass_mb_stats.mb_stats_end = cpi->twopass.firstpass_mb_stats.mb_stats_start + (ps - 1) * cpi->common.MBs * sizeof(uint8_t); } #endif cpi->twopass.stats_in_start = oxcf->two_pass_stats_in.buf; cpi->twopass.stats_in = cpi->twopass.stats_in_start; cpi->twopass.stats_in_end = &cpi->twopass.stats_in[packets - 1]; av1_init_second_pass(cpi); } CHECK_MEM_ERROR( cm, cpi->td.mb.above_pred_buf, (uint8_t *)aom_memalign(16, MAX_MB_PLANE * MAX_SB_SQUARE * sizeof(*cpi->td.mb.above_pred_buf))); CHECK_MEM_ERROR( cm, cpi->td.mb.left_pred_buf, (uint8_t *)aom_memalign(16, MAX_MB_PLANE * MAX_SB_SQUARE * sizeof(*cpi->td.mb.left_pred_buf))); CHECK_MEM_ERROR(cm, cpi->td.mb.wsrc_buf, (int32_t *)aom_memalign( 16, MAX_SB_SQUARE * sizeof(*cpi->td.mb.wsrc_buf))); CHECK_MEM_ERROR( cm, cpi->td.mb.inter_modes_info, (InterModesInfo *)aom_malloc(sizeof(*cpi->td.mb.inter_modes_info))); for (int x = 0; x < 2; x++) for (int y = 0; y < 2; y++) CHECK_MEM_ERROR( cm, cpi->td.mb.hash_value_buffer[x][y], (uint32_t *)aom_malloc(AOM_BUFFER_SIZE_FOR_BLOCK_HASH * sizeof(*cpi->td.mb.hash_value_buffer[0][0]))); cpi->td.mb.g_crc_initialized = 0; CHECK_MEM_ERROR(cm, cpi->td.mb.mask_buf, (int32_t *)aom_memalign( 16, MAX_SB_SQUARE * sizeof(*cpi->td.mb.mask_buf))); av1_set_speed_features_framesize_independent(cpi, oxcf->speed); av1_set_speed_features_framesize_dependent(cpi, oxcf->speed); for (int frame = 0; frame < MAX_LAG_BUFFERS; ++frame) { int mi_cols = ALIGN_POWER_OF_TWO(cm->mi_cols, MAX_MIB_SIZE_LOG2); int mi_rows = ALIGN_POWER_OF_TWO(cm->mi_rows, MAX_MIB_SIZE_LOG2); CHECK_MEM_ERROR(cm, cpi->tpl_stats[frame].tpl_stats_ptr, aom_calloc(mi_rows * mi_cols, sizeof(*cpi->tpl_stats[frame].tpl_stats_ptr))); cpi->tpl_stats[frame].is_valid = 0; cpi->tpl_stats[frame].width = mi_cols; cpi->tpl_stats[frame].height = mi_rows; cpi->tpl_stats[frame].stride = mi_cols; cpi->tpl_stats[frame].mi_rows = cm->mi_rows; cpi->tpl_stats[frame].mi_cols = cm->mi_cols; } #if CONFIG_COLLECT_PARTITION_STATS av1_zero(cpi->partition_stats); #endif #define BFP(BT, SDF, SDAF, VF, SVF, SVAF, SDX4DF, JSDAF, JSVAF) \ cpi->fn_ptr[BT].sdf = SDF; \ cpi->fn_ptr[BT].sdaf = SDAF; \ cpi->fn_ptr[BT].vf = VF; \ cpi->fn_ptr[BT].svf = SVF; \ cpi->fn_ptr[BT].svaf = SVAF; \ cpi->fn_ptr[BT].sdx4df = SDX4DF; \ cpi->fn_ptr[BT].jsdaf = JSDAF; \ cpi->fn_ptr[BT].jsvaf = JSVAF; BFP(BLOCK_4X16, aom_sad4x16, aom_sad4x16_avg, aom_variance4x16, aom_sub_pixel_variance4x16, aom_sub_pixel_avg_variance4x16, aom_sad4x16x4d, aom_dist_wtd_sad4x16_avg, aom_dist_wtd_sub_pixel_avg_variance4x16) BFP(BLOCK_16X4, aom_sad16x4, aom_sad16x4_avg, aom_variance16x4, aom_sub_pixel_variance16x4, aom_sub_pixel_avg_variance16x4, aom_sad16x4x4d, aom_dist_wtd_sad16x4_avg, aom_dist_wtd_sub_pixel_avg_variance16x4) BFP(BLOCK_8X32, aom_sad8x32, aom_sad8x32_avg, aom_variance8x32, aom_sub_pixel_variance8x32, aom_sub_pixel_avg_variance8x32, aom_sad8x32x4d, aom_dist_wtd_sad8x32_avg, aom_dist_wtd_sub_pixel_avg_variance8x32) BFP(BLOCK_32X8, aom_sad32x8, aom_sad32x8_avg, aom_variance32x8, aom_sub_pixel_variance32x8, aom_sub_pixel_avg_variance32x8, aom_sad32x8x4d, aom_dist_wtd_sad32x8_avg, aom_dist_wtd_sub_pixel_avg_variance32x8) BFP(BLOCK_16X64, aom_sad16x64, aom_sad16x64_avg, aom_variance16x64, aom_sub_pixel_variance16x64, aom_sub_pixel_avg_variance16x64, aom_sad16x64x4d, aom_dist_wtd_sad16x64_avg, aom_dist_wtd_sub_pixel_avg_variance16x64) BFP(BLOCK_64X16, aom_sad64x16, aom_sad64x16_avg, aom_variance64x16, aom_sub_pixel_variance64x16, aom_sub_pixel_avg_variance64x16, aom_sad64x16x4d, aom_dist_wtd_sad64x16_avg, aom_dist_wtd_sub_pixel_avg_variance64x16) BFP(BLOCK_128X128, aom_sad128x128, aom_sad128x128_avg, aom_variance128x128, aom_sub_pixel_variance128x128, aom_sub_pixel_avg_variance128x128, aom_sad128x128x4d, aom_dist_wtd_sad128x128_avg, aom_dist_wtd_sub_pixel_avg_variance128x128) BFP(BLOCK_128X64, aom_sad128x64, aom_sad128x64_avg, aom_variance128x64, aom_sub_pixel_variance128x64, aom_sub_pixel_avg_variance128x64, aom_sad128x64x4d, aom_dist_wtd_sad128x64_avg, aom_dist_wtd_sub_pixel_avg_variance128x64) BFP(BLOCK_64X128, aom_sad64x128, aom_sad64x128_avg, aom_variance64x128, aom_sub_pixel_variance64x128, aom_sub_pixel_avg_variance64x128, aom_sad64x128x4d, aom_dist_wtd_sad64x128_avg, aom_dist_wtd_sub_pixel_avg_variance64x128) BFP(BLOCK_32X16, aom_sad32x16, aom_sad32x16_avg, aom_variance32x16, aom_sub_pixel_variance32x16, aom_sub_pixel_avg_variance32x16, aom_sad32x16x4d, aom_dist_wtd_sad32x16_avg, aom_dist_wtd_sub_pixel_avg_variance32x16) BFP(BLOCK_16X32, aom_sad16x32, aom_sad16x32_avg, aom_variance16x32, aom_sub_pixel_variance16x32, aom_sub_pixel_avg_variance16x32, aom_sad16x32x4d, aom_dist_wtd_sad16x32_avg, aom_dist_wtd_sub_pixel_avg_variance16x32) BFP(BLOCK_64X32, aom_sad64x32, aom_sad64x32_avg, aom_variance64x32, aom_sub_pixel_variance64x32, aom_sub_pixel_avg_variance64x32, aom_sad64x32x4d, aom_dist_wtd_sad64x32_avg, aom_dist_wtd_sub_pixel_avg_variance64x32) BFP(BLOCK_32X64, aom_sad32x64, aom_sad32x64_avg, aom_variance32x64, aom_sub_pixel_variance32x64, aom_sub_pixel_avg_variance32x64, aom_sad32x64x4d, aom_dist_wtd_sad32x64_avg, aom_dist_wtd_sub_pixel_avg_variance32x64) BFP(BLOCK_32X32, aom_sad32x32, aom_sad32x32_avg, aom_variance32x32, aom_sub_pixel_variance32x32, aom_sub_pixel_avg_variance32x32, aom_sad32x32x4d, aom_dist_wtd_sad32x32_avg, aom_dist_wtd_sub_pixel_avg_variance32x32) BFP(BLOCK_64X64, aom_sad64x64, aom_sad64x64_avg, aom_variance64x64, aom_sub_pixel_variance64x64, aom_sub_pixel_avg_variance64x64, aom_sad64x64x4d, aom_dist_wtd_sad64x64_avg, aom_dist_wtd_sub_pixel_avg_variance64x64) BFP(BLOCK_16X16, aom_sad16x16, aom_sad16x16_avg, aom_variance16x16, aom_sub_pixel_variance16x16, aom_sub_pixel_avg_variance16x16, aom_sad16x16x4d, aom_dist_wtd_sad16x16_avg, aom_dist_wtd_sub_pixel_avg_variance16x16) BFP(BLOCK_16X8, aom_sad16x8, aom_sad16x8_avg, aom_variance16x8, aom_sub_pixel_variance16x8, aom_sub_pixel_avg_variance16x8, aom_sad16x8x4d, aom_dist_wtd_sad16x8_avg, aom_dist_wtd_sub_pixel_avg_variance16x8) BFP(BLOCK_8X16, aom_sad8x16, aom_sad8x16_avg, aom_variance8x16, aom_sub_pixel_variance8x16, aom_sub_pixel_avg_variance8x16, aom_sad8x16x4d, aom_dist_wtd_sad8x16_avg, aom_dist_wtd_sub_pixel_avg_variance8x16) BFP(BLOCK_8X8, aom_sad8x8, aom_sad8x8_avg, aom_variance8x8, aom_sub_pixel_variance8x8, aom_sub_pixel_avg_variance8x8, aom_sad8x8x4d, aom_dist_wtd_sad8x8_avg, aom_dist_wtd_sub_pixel_avg_variance8x8) BFP(BLOCK_8X4, aom_sad8x4, aom_sad8x4_avg, aom_variance8x4, aom_sub_pixel_variance8x4, aom_sub_pixel_avg_variance8x4, aom_sad8x4x4d, aom_dist_wtd_sad8x4_avg, aom_dist_wtd_sub_pixel_avg_variance8x4) BFP(BLOCK_4X8, aom_sad4x8, aom_sad4x8_avg, aom_variance4x8, aom_sub_pixel_variance4x8, aom_sub_pixel_avg_variance4x8, aom_sad4x8x4d, aom_dist_wtd_sad4x8_avg, aom_dist_wtd_sub_pixel_avg_variance4x8) BFP(BLOCK_4X4, aom_sad4x4, aom_sad4x4_avg, aom_variance4x4, aom_sub_pixel_variance4x4, aom_sub_pixel_avg_variance4x4, aom_sad4x4x4d, aom_dist_wtd_sad4x4_avg, aom_dist_wtd_sub_pixel_avg_variance4x4) #define OBFP(BT, OSDF, OVF, OSVF) \ cpi->fn_ptr[BT].osdf = OSDF; \ cpi->fn_ptr[BT].ovf = OVF; \ cpi->fn_ptr[BT].osvf = OSVF; OBFP(BLOCK_128X128, aom_obmc_sad128x128, aom_obmc_variance128x128, aom_obmc_sub_pixel_variance128x128) OBFP(BLOCK_128X64, aom_obmc_sad128x64, aom_obmc_variance128x64, aom_obmc_sub_pixel_variance128x64) OBFP(BLOCK_64X128, aom_obmc_sad64x128, aom_obmc_variance64x128, aom_obmc_sub_pixel_variance64x128) OBFP(BLOCK_64X64, aom_obmc_sad64x64, aom_obmc_variance64x64, aom_obmc_sub_pixel_variance64x64) OBFP(BLOCK_64X32, aom_obmc_sad64x32, aom_obmc_variance64x32, aom_obmc_sub_pixel_variance64x32) OBFP(BLOCK_32X64, aom_obmc_sad32x64, aom_obmc_variance32x64, aom_obmc_sub_pixel_variance32x64) OBFP(BLOCK_32X32, aom_obmc_sad32x32, aom_obmc_variance32x32, aom_obmc_sub_pixel_variance32x32) OBFP(BLOCK_32X16, aom_obmc_sad32x16, aom_obmc_variance32x16, aom_obmc_sub_pixel_variance32x16) OBFP(BLOCK_16X32, aom_obmc_sad16x32, aom_obmc_variance16x32, aom_obmc_sub_pixel_variance16x32) OBFP(BLOCK_16X16, aom_obmc_sad16x16, aom_obmc_variance16x16, aom_obmc_sub_pixel_variance16x16) OBFP(BLOCK_16X8, aom_obmc_sad16x8, aom_obmc_variance16x8, aom_obmc_sub_pixel_variance16x8) OBFP(BLOCK_8X16, aom_obmc_sad8x16, aom_obmc_variance8x16, aom_obmc_sub_pixel_variance8x16) OBFP(BLOCK_8X8, aom_obmc_sad8x8, aom_obmc_variance8x8, aom_obmc_sub_pixel_variance8x8) OBFP(BLOCK_4X8, aom_obmc_sad4x8, aom_obmc_variance4x8, aom_obmc_sub_pixel_variance4x8) OBFP(BLOCK_8X4, aom_obmc_sad8x4, aom_obmc_variance8x4, aom_obmc_sub_pixel_variance8x4) OBFP(BLOCK_4X4, aom_obmc_sad4x4, aom_obmc_variance4x4, aom_obmc_sub_pixel_variance4x4) OBFP(BLOCK_4X16, aom_obmc_sad4x16, aom_obmc_variance4x16, aom_obmc_sub_pixel_variance4x16) OBFP(BLOCK_16X4, aom_obmc_sad16x4, aom_obmc_variance16x4, aom_obmc_sub_pixel_variance16x4) OBFP(BLOCK_8X32, aom_obmc_sad8x32, aom_obmc_variance8x32, aom_obmc_sub_pixel_variance8x32) OBFP(BLOCK_32X8, aom_obmc_sad32x8, aom_obmc_variance32x8, aom_obmc_sub_pixel_variance32x8) OBFP(BLOCK_16X64, aom_obmc_sad16x64, aom_obmc_variance16x64, aom_obmc_sub_pixel_variance16x64) OBFP(BLOCK_64X16, aom_obmc_sad64x16, aom_obmc_variance64x16, aom_obmc_sub_pixel_variance64x16) #define MBFP(BT, MCSDF, MCSVF) \ cpi->fn_ptr[BT].msdf = MCSDF; \ cpi->fn_ptr[BT].msvf = MCSVF; MBFP(BLOCK_128X128, aom_masked_sad128x128, aom_masked_sub_pixel_variance128x128) MBFP(BLOCK_128X64, aom_masked_sad128x64, aom_masked_sub_pixel_variance128x64) MBFP(BLOCK_64X128, aom_masked_sad64x128, aom_masked_sub_pixel_variance64x128) MBFP(BLOCK_64X64, aom_masked_sad64x64, aom_masked_sub_pixel_variance64x64) MBFP(BLOCK_64X32, aom_masked_sad64x32, aom_masked_sub_pixel_variance64x32) MBFP(BLOCK_32X64, aom_masked_sad32x64, aom_masked_sub_pixel_variance32x64) MBFP(BLOCK_32X32, aom_masked_sad32x32, aom_masked_sub_pixel_variance32x32) MBFP(BLOCK_32X16, aom_masked_sad32x16, aom_masked_sub_pixel_variance32x16) MBFP(BLOCK_16X32, aom_masked_sad16x32, aom_masked_sub_pixel_variance16x32) MBFP(BLOCK_16X16, aom_masked_sad16x16, aom_masked_sub_pixel_variance16x16) MBFP(BLOCK_16X8, aom_masked_sad16x8, aom_masked_sub_pixel_variance16x8) MBFP(BLOCK_8X16, aom_masked_sad8x16, aom_masked_sub_pixel_variance8x16) MBFP(BLOCK_8X8, aom_masked_sad8x8, aom_masked_sub_pixel_variance8x8) MBFP(BLOCK_4X8, aom_masked_sad4x8, aom_masked_sub_pixel_variance4x8) MBFP(BLOCK_8X4, aom_masked_sad8x4, aom_masked_sub_pixel_variance8x4) MBFP(BLOCK_4X4, aom_masked_sad4x4, aom_masked_sub_pixel_variance4x4) MBFP(BLOCK_4X16, aom_masked_sad4x16, aom_masked_sub_pixel_variance4x16) MBFP(BLOCK_16X4, aom_masked_sad16x4, aom_masked_sub_pixel_variance16x4) MBFP(BLOCK_8X32, aom_masked_sad8x32, aom_masked_sub_pixel_variance8x32) MBFP(BLOCK_32X8, aom_masked_sad32x8, aom_masked_sub_pixel_variance32x8) MBFP(BLOCK_16X64, aom_masked_sad16x64, aom_masked_sub_pixel_variance16x64) MBFP(BLOCK_64X16, aom_masked_sad64x16, aom_masked_sub_pixel_variance64x16) highbd_set_var_fns(cpi); /* av1_init_quantizer() is first called here. Add check in * av1_frame_init_quantizer() so that av1_init_quantizer is only * called later when needed. This will avoid unnecessary calls of * av1_init_quantizer() for every frame. */ av1_init_quantizer(cpi); av1_qm_init(cm); av1_loop_filter_init(cm); cm->superres_scale_denominator = SCALE_NUMERATOR; cm->superres_upscaled_width = oxcf->width; cm->superres_upscaled_height = oxcf->height; av1_loop_restoration_precal(); cm->error.setjmp = 0; return cpi; } #if CONFIG_INTERNAL_STATS #define SNPRINT(H, T) snprintf((H) + strlen(H), sizeof(H) - strlen(H), (T)) #define SNPRINT2(H, T, V) \ snprintf((H) + strlen(H), sizeof(H) - strlen(H), (T), (V)) #endif // CONFIG_INTERNAL_STATS void av1_remove_compressor(AV1_COMP *cpi) { AV1_COMMON *cm; unsigned int i; int t; if (!cpi) return; cm = &cpi->common; const int num_planes = av1_num_planes(cm); if (cm->current_frame.frame_number > 0) { #if CONFIG_ENTROPY_STATS if (cpi->oxcf.pass != 1) { fprintf(stderr, "Writing counts.stt\n"); FILE *f = fopen("counts.stt", "wb"); fwrite(&aggregate_fc, sizeof(aggregate_fc), 1, f); fclose(f); } #endif // CONFIG_ENTROPY_STATS #if CONFIG_INTERNAL_STATS aom_clear_system_state(); if (cpi->oxcf.pass != 1) { char headings[512] = { 0 }; char results[512] = { 0 }; FILE *f = fopen("opsnr.stt", "a"); double time_encoded = (cpi->last_end_time_stamp_seen - cpi->first_time_stamp_ever) / 10000000.000; double total_encode_time = (cpi->time_receive_data + cpi->time_compress_data) / 1000.000; const double dr = (double)cpi->bytes * (double)8 / (double)1000 / time_encoded; const double peak = (double)((1 << cpi->oxcf.input_bit_depth) - 1); const double target_rate = (double)cpi->oxcf.target_bandwidth / 1000; const double rate_err = ((100.0 * (dr - target_rate)) / target_rate); if (cpi->b_calculate_psnr) { const double total_psnr = aom_sse_to_psnr( (double)cpi->total_samples, peak, (double)cpi->total_sq_error); const double total_ssim = 100 * pow(cpi->summed_quality / cpi->summed_weights, 8.0); snprintf(headings, sizeof(headings), "Bitrate\tAVGPsnr\tGLBPsnr\tAVPsnrP\tGLPsnrP\t" "AOMSSIM\tVPSSIMP\tFASTSIM\tPSNRHVS\t" "WstPsnr\tWstSsim\tWstFast\tWstHVS\t" "AVPsrnY\tAPsnrCb\tAPsnrCr"); snprintf(results, sizeof(results), "%7.2f\t%7.3f\t%7.3f\t%7.3f\t%7.3f\t" "%7.3f\t%7.3f\t%7.3f\t%7.3f\t" "%7.3f\t%7.3f\t%7.3f\t%7.3f\t" "%7.3f\t%7.3f\t%7.3f", dr, cpi->psnr.stat[STAT_ALL] / cpi->count, total_psnr, cpi->psnr.stat[STAT_ALL] / cpi->count, total_psnr, total_ssim, total_ssim, cpi->fastssim.stat[STAT_ALL] / cpi->count, cpi->psnrhvs.stat[STAT_ALL] / cpi->count, cpi->psnr.worst, cpi->worst_ssim, cpi->fastssim.worst, cpi->psnrhvs.worst, cpi->psnr.stat[STAT_Y] / cpi->count, cpi->psnr.stat[STAT_U] / cpi->count, cpi->psnr.stat[STAT_V] / cpi->count); if (cpi->b_calculate_blockiness) { SNPRINT(headings, "\t Block\tWstBlck"); SNPRINT2(results, "\t%7.3f", cpi->total_blockiness / cpi->count); SNPRINT2(results, "\t%7.3f", cpi->worst_blockiness); } if (cpi->b_calculate_consistency) { double consistency = aom_sse_to_psnr((double)cpi->total_samples, peak, (double)cpi->total_inconsistency); SNPRINT(headings, "\tConsist\tWstCons"); SNPRINT2(results, "\t%7.3f", consistency); SNPRINT2(results, "\t%7.3f", cpi->worst_consistency); } fprintf(f, "%s\t Time\tRcErr\tAbsErr\n", headings); fprintf(f, "%s\t%8.0f\t%7.2f\t%7.2f\n", results, total_encode_time, rate_err, fabs(rate_err)); } fclose(f); } #endif // CONFIG_INTERNAL_STATS #if CONFIG_SPEED_STATS if (cpi->oxcf.pass != 1) { fprintf(stdout, "tx_search_count = %d\n", cpi->tx_search_count); } #endif // CONFIG_SPEED_STATS #if CONFIG_COLLECT_PARTITION_STATS if (cpi->oxcf.pass != 1) { av1_print_partition_stats(&cpi->partition_stats); } #endif } for (int frame = 0; frame < MAX_LAG_BUFFERS; ++frame) { aom_free(cpi->tpl_stats[frame].tpl_stats_ptr); cpi->tpl_stats[frame].is_valid = 0; } for (t = cpi->num_workers - 1; t >= 0; --t) { AVxWorker *const worker = &cpi->workers[t]; EncWorkerData *const thread_data = &cpi->tile_thr_data[t]; // Deallocate allocated threads. aom_get_worker_interface()->end(worker); // Deallocate allocated thread data. if (cpi->row_mt == 1) aom_free(thread_data->td->tctx); if (t > 0) { aom_free(thread_data->td->palette_buffer); aom_free(thread_data->td->tmp_conv_dst); for (int j = 0; j < 2; ++j) { aom_free(thread_data->td->tmp_obmc_bufs[j]); } aom_free(thread_data->td->above_pred_buf); aom_free(thread_data->td->left_pred_buf); aom_free(thread_data->td->wsrc_buf); aom_free(thread_data->td->inter_modes_info); for (int x = 0; x < 2; x++) { for (int y = 0; y < 2; y++) { aom_free(thread_data->td->hash_value_buffer[x][y]); thread_data->td->hash_value_buffer[x][y] = NULL; } } aom_free(thread_data->td->mask_buf); aom_free(thread_data->td->counts); av1_free_pc_tree(thread_data->td, num_planes); aom_free(thread_data->td); } } #if CONFIG_MULTITHREAD if (cpi->row_mt == 1) { if (cpi->row_mt_mutex_ != NULL) { pthread_mutex_destroy(cpi->row_mt_mutex_); aom_free(cpi->row_mt_mutex_); } } #endif av1_row_mt_mem_dealloc(cpi); aom_free(cpi->tile_thr_data); aom_free(cpi->workers); if (cpi->num_workers > 1) { av1_loop_filter_dealloc(&cpi->lf_row_sync); av1_loop_restoration_dealloc(&cpi->lr_row_sync, cpi->num_workers); } dealloc_compressor_data(cpi); for (i = 0; i < sizeof(cpi->mbgraph_stats) / sizeof(cpi->mbgraph_stats[0]); ++i) { aom_free(cpi->mbgraph_stats[i].mb_stats); } #if CONFIG_FP_MB_STATS if (cpi->use_fp_mb_stats) { aom_free(cpi->twopass.frame_mb_stats_buf); cpi->twopass.frame_mb_stats_buf = NULL; } #endif #if CONFIG_INTERNAL_STATS aom_free(cpi->ssim_vars); cpi->ssim_vars = NULL; #endif // CONFIG_INTERNAL_STATS av1_remove_common(cm); for (i = 0; i < FRAME_BUFFERS; ++i) { av1_hash_table_destroy(&cm->buffer_pool->frame_bufs[i].hash_table); } if (cpi->sf.use_hash_based_trellis) hbt_destroy(); av1_free_ref_frame_buffers(cm->buffer_pool); aom_free(cpi); #ifdef OUTPUT_YUV_SKINMAP fclose(yuv_skinmap_file); #endif #ifdef OUTPUT_YUV_REC fclose(yuv_rec_file); #endif } static void generate_psnr_packet(AV1_COMP *cpi) { struct aom_codec_cx_pkt pkt; int i; PSNR_STATS psnr; aom_calc_highbd_psnr(cpi->source, &cpi->common.cur_frame->buf, &psnr, cpi->td.mb.e_mbd.bd, cpi->oxcf.input_bit_depth); for (i = 0; i < 4; ++i) { pkt.data.psnr.samples[i] = psnr.samples[i]; pkt.data.psnr.sse[i] = psnr.sse[i]; pkt.data.psnr.psnr[i] = psnr.psnr[i]; } pkt.kind = AOM_CODEC_PSNR_PKT; aom_codec_pkt_list_add(cpi->output_pkt_list, &pkt); } int av1_use_as_reference(AV1_COMP *cpi, int ref_frame_flags) { if (ref_frame_flags > ((1 << INTER_REFS_PER_FRAME) - 1)) return -1; cpi->ext_ref_frame_flags = ref_frame_flags; return 0; } int av1_copy_reference_enc(AV1_COMP *cpi, int idx, YV12_BUFFER_CONFIG *sd) { AV1_COMMON *const cm = &cpi->common; const int num_planes = av1_num_planes(cm); YV12_BUFFER_CONFIG *cfg = get_ref_frame(cm, idx); if (cfg) { aom_yv12_copy_frame(cfg, sd, num_planes); return 0; } else { return -1; } } int av1_set_reference_enc(AV1_COMP *cpi, int idx, YV12_BUFFER_CONFIG *sd) { AV1_COMMON *const cm = &cpi->common; const int num_planes = av1_num_planes(cm); YV12_BUFFER_CONFIG *cfg = get_ref_frame(cm, idx); if (cfg) { aom_yv12_copy_frame(sd, cfg, num_planes); return 0; } else { return -1; } } int av1_update_entropy(AV1_COMP *cpi, int update) { cpi->ext_refresh_frame_context = update; cpi->ext_refresh_frame_context_pending = 1; return 0; } #if defined(OUTPUT_YUV_DENOISED) || defined(OUTPUT_YUV_SKINMAP) // The denoiser buffer is allocated as a YUV 440 buffer. This function writes it // as YUV 420. We simply use the top-left pixels of the UV buffers, since we do // not denoise the UV channels at this time. If ever we implement UV channel // denoising we will have to modify this. void aom_write_yuv_frame_420(YV12_BUFFER_CONFIG *s, FILE *f) { uint8_t *src = s->y_buffer; int h = s->y_height; do { fwrite(src, s->y_width, 1, f); src += s->y_stride; } while (--h); src = s->u_buffer; h = s->uv_height; do { fwrite(src, s->uv_width, 1, f); src += s->uv_stride; } while (--h); src = s->v_buffer; h = s->uv_height; do { fwrite(src, s->uv_width, 1, f); src += s->uv_stride; } while (--h); } #endif #ifdef OUTPUT_YUV_REC void aom_write_one_yuv_frame(AV1_COMMON *cm, YV12_BUFFER_CONFIG *s) { uint8_t *src = s->y_buffer; int h = cm->height; if (yuv_rec_file == NULL) return; if (s->flags & YV12_FLAG_HIGHBITDEPTH) { uint16_t *src16 = CONVERT_TO_SHORTPTR(s->y_buffer); do { fwrite(src16, s->y_width, 2, yuv_rec_file); src16 += s->y_stride; } while (--h); src16 = CONVERT_TO_SHORTPTR(s->u_buffer); h = s->uv_height; do { fwrite(src16, s->uv_width, 2, yuv_rec_file); src16 += s->uv_stride; } while (--h); src16 = CONVERT_TO_SHORTPTR(s->v_buffer); h = s->uv_height; do { fwrite(src16, s->uv_width, 2, yuv_rec_file); src16 += s->uv_stride; } while (--h); fflush(yuv_rec_file); return; } do { fwrite(src, s->y_width, 1, yuv_rec_file); src += s->y_stride; } while (--h); src = s->u_buffer; h = s->uv_height; do { fwrite(src, s->uv_width, 1, yuv_rec_file); src += s->uv_stride; } while (--h); src = s->v_buffer; h = s->uv_height; do { fwrite(src, s->uv_width, 1, yuv_rec_file); src += s->uv_stride; } while (--h); fflush(yuv_rec_file); } #endif // OUTPUT_YUV_REC #define GM_RECODE_LOOP_NUM4X4_FACTOR 192 static int recode_loop_test_global_motion(AV1_COMP *cpi) { int i; int recode = 0; RD_COUNTS *const rdc = &cpi->td.rd_counts; AV1_COMMON *const cm = &cpi->common; for (i = LAST_FRAME; i <= ALTREF_FRAME; ++i) { if (cm->global_motion[i].wmtype != IDENTITY && rdc->global_motion_used[i] * GM_RECODE_LOOP_NUM4X4_FACTOR < cpi->gmparams_cost[i]) { cm->global_motion[i] = default_warp_params; assert(cm->global_motion[i].wmtype == IDENTITY); cpi->gmparams_cost[i] = 0; recode = 1; // TODO(sarahparker): The earlier condition for recoding here was: // "recode |= (rdc->global_motion_used[i] > 0);". Can we bring something // similar to that back to speed up global motion? } } return recode; } // Function to test for conditions that indicate we should loop // back and recode a frame. static int recode_loop_test(AV1_COMP *cpi, int high_limit, int low_limit, int q, int maxq, int minq) { const RATE_CONTROL *const rc = &cpi->rc; const AV1EncoderConfig *const oxcf = &cpi->oxcf; const int frame_is_kfgfarf = frame_is_kf_gf_arf(cpi); int force_recode = 0; if ((rc->projected_frame_size >= rc->max_frame_bandwidth) || (cpi->sf.recode_loop == ALLOW_RECODE) || (frame_is_kfgfarf && (cpi->sf.recode_loop == ALLOW_RECODE_KFARFGF))) { // TODO(agrange) high_limit could be greater than the scale-down threshold. if ((rc->projected_frame_size > high_limit && q < maxq) || (rc->projected_frame_size < low_limit && q > minq)) { force_recode = 1; } else if (cpi->oxcf.rc_mode == AOM_CQ) { // Deal with frame undershoot and whether or not we are // below the automatically set cq level. if (q > oxcf->cq_level && rc->projected_frame_size < ((rc->this_frame_target * 7) >> 3)) { force_recode = 1; } } } return force_recode; } #define DUMP_REF_FRAME_IMAGES 0 #if DUMP_REF_FRAME_IMAGES == 1 static int dump_one_image(AV1_COMMON *cm, const YV12_BUFFER_CONFIG *const ref_buf, char *file_name) { int h; FILE *f_ref = NULL; if (ref_buf == NULL) { printf("Frame data buffer is NULL.\n"); return AOM_CODEC_MEM_ERROR; } if ((f_ref = fopen(file_name, "wb")) == NULL) { printf("Unable to open file %s to write.\n", file_name); return AOM_CODEC_MEM_ERROR; } // --- Y --- for (h = 0; h < cm->height; ++h) { fwrite(&ref_buf->y_buffer[h * ref_buf->y_stride], 1, cm->width, f_ref); } // --- U --- for (h = 0; h < (cm->height >> 1); ++h) { fwrite(&ref_buf->u_buffer[h * ref_buf->uv_stride], 1, (cm->width >> 1), f_ref); } // --- V --- for (h = 0; h < (cm->height >> 1); ++h) { fwrite(&ref_buf->v_buffer[h * ref_buf->uv_stride], 1, (cm->width >> 1), f_ref); } fclose(f_ref); return AOM_CODEC_OK; } static void dump_ref_frame_images(AV1_COMP *cpi) { AV1_COMMON *const cm = &cpi->common; MV_REFERENCE_FRAME ref_frame; for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) { char file_name[256] = ""; snprintf(file_name, sizeof(file_name), "/tmp/enc_F%d_ref_%d.yuv", cm->current_frame.frame_number, ref_frame); dump_one_image(cm, get_ref_frame_yv12_buf(cpi, ref_frame), file_name); } } #endif // DUMP_REF_FRAME_IMAGES == 1 // This function is used to shift the virtual indices of last reference frames // as follows: // LAST_FRAME -> LAST2_FRAME -> LAST3_FRAME // when the LAST_FRAME is updated. static INLINE void shift_last_ref_frames(AV1_COMP *cpi) { // TODO(isbs): shift the scaled indices as well for (int ref_frame = LAST3_FRAME; ref_frame > LAST_FRAME; --ref_frame) { const int ref_idx = ref_frame - LAST_FRAME; cpi->common.remapped_ref_idx[ref_idx] = cpi->common.remapped_ref_idx[ref_idx - 1]; if (!cpi->rc.is_src_frame_alt_ref) { memcpy(cpi->interp_filter_selected[ref_frame], cpi->interp_filter_selected[ref_frame - 1], sizeof(cpi->interp_filter_selected[ref_frame - 1])); } } } // This function is used to shift the virtual indices of bwd reference // frames as follows: // BWD_REF -> ALT2_REF -> EXT_REF // to clear a space to store the closest bwdref static INLINE void rshift_bwd_ref_frames(AV1_COMP *cpi) { // TODO(isbs): shift the scaled indices as well static const int ordered_bwd[3] = { BWDREF_FRAME, ALTREF2_FRAME, EXTREF_FRAME }; for (int i = 2; i > 0; --i) { // [0] is allocated to the current coded frame, i.e. bwdref memcpy(cpi->interp_filter_selected[ordered_bwd[i]], cpi->interp_filter_selected[ordered_bwd[i - 1]], sizeof(cpi->interp_filter_selected[ordered_bwd[i - 1]])); cpi->common.remapped_ref_idx[ordered_bwd[i] - LAST_FRAME] = cpi->common.remapped_ref_idx[ordered_bwd[i - 1] - LAST_FRAME]; } } // This function is used to shift the virtual indices of bwd reference // frames as follows: // BWD_REF <- ALT2_REF <- EXT_REF // to update the bwd reference frame for coding the next frame. static INLINE void lshift_bwd_ref_frames(AV1_COMP *cpi) { // TODO(isbs): shift the scaled indices as well static const int ordered_bwd[3] = { BWDREF_FRAME, ALTREF2_FRAME, EXTREF_FRAME }; for (int i = 0; i < 2; ++i) { // [0] is allocated to the current coded frame, i.e. bwdref memcpy(cpi->interp_filter_selected[ordered_bwd[i]], cpi->interp_filter_selected[ordered_bwd[i + 1]], sizeof(cpi->interp_filter_selected[ordered_bwd[i + 1]])); cpi->common.remapped_ref_idx[ordered_bwd[i] - LAST_FRAME] = cpi->common.remapped_ref_idx[ordered_bwd[i + 1] - LAST_FRAME]; } } static void update_reference_frames(AV1_COMP *cpi) { AV1_COMMON *const cm = &cpi->common; // NOTE: Save the new show frame buffer index for --test-code=warn, i.e., // for the purpose to verify no mismatch between encoder and decoder. if (cm->show_frame) cpi->last_show_frame_buf = cm->cur_frame; // In the case of show_existing frame, we will not send fresh flag // to decoder. Any change in the reference frame buffer can be done by // switching the virtual indices. if (cm->show_existing_frame) { // If we are not indicating to the decoder that this frame is // a show_existing_frame, which occurs in error_resilient mode, // we still want to refresh the LAST_FRAME when the current frame // was the source of an ext_arf. cpi->refresh_last_frame = !encode_show_existing_frame(cm) && cpi->rc.is_src_frame_ext_arf; cpi->refresh_golden_frame = 0; cpi->refresh_bwd_ref_frame = 0; cpi->refresh_alt2_ref_frame = 0; cpi->refresh_alt_ref_frame = 0; cpi->rc.is_bwd_ref_frame = 0; cpi->rc.is_last_bipred_frame = 0; cpi->rc.is_bipred_frame = 0; } // At this point the new frame has been encoded. // If any buffer copy / swapping is signaled it should be done here. // Only update all of the reference buffers if a KEY_FRAME is also a // show_frame. This ensures a fwd keyframe does not update all of the buffers if ((cm->current_frame.frame_type == KEY_FRAME && cm->show_frame) || frame_is_sframe(cm)) { for (int ref_frame = 0; ref_frame < REF_FRAMES; ++ref_frame) { assign_frame_buffer_p(&cm->ref_frame_map[cm->remapped_ref_idx[ref_frame]], cm->cur_frame); } return; } if (av1_preserve_existing_gf(cpi)) { // We have decided to preserve the previously existing golden frame as our // new ARF frame. However, in the short term in function // av1_bitstream.c::get_refresh_mask() we left it in the GF slot and, if // we're updating the GF with the current decoded frame, we save it to the // ARF slot instead. // We now have to update the ARF with the current frame and swap gld_fb_idx // and alt_fb_idx so that, overall, we've stored the old GF in the new ARF // slot and, if we're updating the GF, the current frame becomes the new GF. int tmp; // ARF in general is a better reference than overlay. We shouldkeep ARF as // reference instead of replacing it with overlay. if (!cpi->preserve_arf_as_gld) { assign_frame_buffer_p( &cm->ref_frame_map[get_ref_frame_map_idx(cm, ALTREF_FRAME)], cm->cur_frame); } tmp = get_ref_frame_map_idx(cm, ALTREF_FRAME); cm->remapped_ref_idx[ALTREF_FRAME - LAST_FRAME] = get_ref_frame_map_idx(cm, GOLDEN_FRAME); cm->remapped_ref_idx[GOLDEN_FRAME - LAST_FRAME] = tmp; // TODO(zoeliu): Do we need to copy cpi->interp_filter_selected[0] over to // cpi->interp_filter_selected[GOLDEN_FRAME]? } else if (cpi->rc.is_src_frame_ext_arf && encode_show_existing_frame(cm)) { #if CONFIG_DEBUG const GF_GROUP *const gf_group = &cpi->twopass.gf_group; assert(gf_group->update_type[gf_group->index] == INTNL_OVERLAY_UPDATE); #endif // CONFIG_DEBUG const int bwdref_to_show = (cpi->new_bwdref_update_rule == 1) ? BWDREF_FRAME : ALTREF2_FRAME; // Deal with the special case for showing existing internal ALTREF_FRAME // Refresh the LAST_FRAME with the ALTREF_FRAME and retire the LAST3_FRAME // by updating the virtual indices. const int last3_remapped_idx = get_ref_frame_map_idx(cm, LAST3_FRAME); shift_last_ref_frames(cpi); cm->remapped_ref_idx[LAST_FRAME - LAST_FRAME] = get_ref_frame_map_idx(cm, bwdref_to_show); memcpy(cpi->interp_filter_selected[LAST_FRAME], cpi->interp_filter_selected[bwdref_to_show], sizeof(cpi->interp_filter_selected[bwdref_to_show])); if (cpi->new_bwdref_update_rule == 1) { lshift_bwd_ref_frames(cpi); // pass outdated forward reference frame (previous LAST3) to the // spared space cm->remapped_ref_idx[EXTREF_FRAME - LAST_FRAME] = last3_remapped_idx; } else { cm->remapped_ref_idx[bwdref_to_show - LAST_FRAME] = last3_remapped_idx; } } else { /* For non key/golden frames */ // === ALTREF_FRAME === if (cpi->refresh_alt_ref_frame) { int arf_idx = get_ref_frame_map_idx(cm, ALTREF_FRAME); assign_frame_buffer_p(&cm->ref_frame_map[arf_idx], cm->cur_frame); memcpy(cpi->interp_filter_selected[ALTREF_FRAME], cpi->interp_filter_selected[0], sizeof(cpi->interp_filter_selected[0])); } // === GOLDEN_FRAME === if (cpi->refresh_golden_frame) { assign_frame_buffer_p( &cm->ref_frame_map[get_ref_frame_map_idx(cm, GOLDEN_FRAME)], cm->cur_frame); memcpy(cpi->interp_filter_selected[GOLDEN_FRAME], cpi->interp_filter_selected[0], sizeof(cpi->interp_filter_selected[0])); } // === BWDREF_FRAME === if (cpi->refresh_bwd_ref_frame) { if (cpi->new_bwdref_update_rule) { // We shift the backward reference frame as follows: // BWDREF -> ALTREF2 -> EXTREF // and assign the newly coded frame to BWDREF so that it always // keeps the nearest future frame const int tmp = get_ref_frame_map_idx(cm, EXTREF_FRAME); assign_frame_buffer_p(&cm->ref_frame_map[tmp], cm->cur_frame); rshift_bwd_ref_frames(cpi); cm->remapped_ref_idx[BWDREF_FRAME - LAST_FRAME] = tmp; } else { assign_frame_buffer_p( &cm->ref_frame_map[get_ref_frame_map_idx(cm, BWDREF_FRAME)], cm->cur_frame); } memcpy(cpi->interp_filter_selected[BWDREF_FRAME], cpi->interp_filter_selected[0], sizeof(cpi->interp_filter_selected[0])); } // === ALTREF2_FRAME === if (cpi->refresh_alt2_ref_frame) { assign_frame_buffer_p( &cm->ref_frame_map[get_ref_frame_map_idx(cm, ALTREF2_FRAME)], cm->cur_frame); memcpy(cpi->interp_filter_selected[ALTREF2_FRAME], cpi->interp_filter_selected[0], sizeof(cpi->interp_filter_selected[0])); } } if (cpi->refresh_last_frame) { // NOTE(zoeliu): We have two layers of mapping (1) from the per-frame // reference to the reference frame buffer virtual index; and then (2) from // the virtual index to the reference frame buffer (RefCntBuffer): // // LAST_FRAME, ..., EXTREF_FRAME // | | // v v // remapped_ref_idx[LAST_FRAME - 1], ..., remapped_ref_idx[EXTREF_FRAME - 1] // | | // v v // ref_frame_map[], ..., ref_frame_map[] // // When refresh_last_frame is set, it is intended to retire LAST3_FRAME, // have the other 2 LAST reference frames shifted as follows: // LAST_FRAME -> LAST2_FRAME -> LAST3_FRAME // , and then have LAST_FRAME refreshed by the newly coded frame. // // To fulfill it, the decoder will be notified to execute following 2 steps: // // (a) To change ref_frame_map[] and have the virtual index of LAST3_FRAME // to point to the newly coded frame, i.e. // ref_frame_map[lst_fb_idexes[2]] => cur_frame; // // (b) To change the 1st layer mapping to have LAST_FRAME mapped to the // original virtual index of LAST3_FRAME and have the other mappings // shifted as follows: // LAST_FRAME, LAST2_FRAME, LAST3_FRAME // | | | // v v v // remapped_ref_idx[2], remapped_ref_idx[0], remapped_ref_idx[1] assign_frame_buffer_p( &cm->ref_frame_map[get_ref_frame_map_idx(cm, LAST3_FRAME)], cm->cur_frame); int last3_remapped_idx = get_ref_frame_map_idx(cm, LAST3_FRAME); shift_last_ref_frames(cpi); cm->remapped_ref_idx[LAST_FRAME - LAST_FRAME] = last3_remapped_idx; assert(!encode_show_existing_frame(cm)); memcpy(cpi->interp_filter_selected[LAST_FRAME], cpi->interp_filter_selected[0], sizeof(cpi->interp_filter_selected[0])); // If the new structure is used, we will always have overlay frames coupled // with bwdref frames. Therefore, we won't have to perform this update // in advance (we do this update when the overlay frame shows up). if (cpi->new_bwdref_update_rule == 0 && cpi->rc.is_last_bipred_frame) { // Refresh the LAST_FRAME with the BWDREF_FRAME and retire the // LAST3_FRAME by updating the virtual indices. // // NOTE: The source frame for BWDREF does not have a holding position as // the OVERLAY frame for ALTREF's. Hence, to resolve the reference // virtual index reshuffling for BWDREF, the encoder always // specifies a LAST_BIPRED right before BWDREF and completes the // reshuffling job accordingly. last3_remapped_idx = get_ref_frame_map_idx(cm, LAST3_FRAME); shift_last_ref_frames(cpi); cm->remapped_ref_idx[LAST_FRAME - LAST_FRAME] = get_ref_frame_map_idx(cm, BWDREF_FRAME); cm->remapped_ref_idx[BWDREF_FRAME - LAST_FRAME] = last3_remapped_idx; memcpy(cpi->interp_filter_selected[LAST_FRAME], cpi->interp_filter_selected[BWDREF_FRAME], sizeof(cpi->interp_filter_selected[BWDREF_FRAME])); } } #if DUMP_REF_FRAME_IMAGES == 1 // Dump out all reference frame images. dump_ref_frame_images(cpi); #endif // DUMP_REF_FRAME_IMAGES } static void scale_references(AV1_COMP *cpi) { AV1_COMMON *cm = &cpi->common; const int num_planes = av1_num_planes(cm); MV_REFERENCE_FRAME ref_frame; const AOM_REFFRAME ref_mask[INTER_REFS_PER_FRAME] = { AOM_LAST_FLAG, AOM_LAST2_FLAG, AOM_LAST3_FLAG, AOM_GOLD_FLAG, AOM_BWD_FLAG, AOM_ALT2_FLAG, AOM_ALT_FLAG }; for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) { // Need to convert from AOM_REFFRAME to index into ref_mask (subtract 1). if (cpi->ref_frame_flags & ref_mask[ref_frame - 1]) { BufferPool *const pool = cm->buffer_pool; const YV12_BUFFER_CONFIG *const ref = get_ref_frame_yv12_buf(cm, ref_frame); if (ref == NULL) { cpi->scaled_ref_buf[ref_frame - 1] = NULL; continue; } if (ref->y_crop_width != cm->width || ref->y_crop_height != cm->height) { int force_scaling = 0; RefCntBuffer *new_fb = cpi->scaled_ref_buf[ref_frame - 1]; if (new_fb == NULL) { const int new_fb_idx = get_free_fb(cm); if (new_fb_idx == INVALID_IDX) { aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, "Unable to find free frame buffer"); } force_scaling = 1; new_fb = &pool->frame_bufs[new_fb_idx]; } if (force_scaling || new_fb->buf.y_crop_width != cm->width || new_fb->buf.y_crop_height != cm->height) { if (aom_realloc_frame_buffer( &new_fb->buf, cm->width, cm->height, cm->seq_params.subsampling_x, cm->seq_params.subsampling_y, cm->seq_params.use_highbitdepth, cpi->oxcf.border_in_pixels, cm->byte_alignment, NULL, NULL, NULL)) { if (force_scaling) { // Release the reference acquired in the get_free_fb() call above. --new_fb->ref_count; } aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, "Failed to allocate frame buffer"); } av1_resize_and_extend_frame( ref, &new_fb->buf, (int)cm->seq_params.bit_depth, num_planes); cpi->scaled_ref_buf[ref_frame - 1] = new_fb; alloc_frame_mvs(cm, new_fb); } } else { RefCntBuffer *buf = get_ref_frame_buf(cm, ref_frame); buf->buf.y_crop_width = ref->y_crop_width; buf->buf.y_crop_height = ref->y_crop_height; cpi->scaled_ref_buf[ref_frame - 1] = buf; ++buf->ref_count; } } else { if (cpi->oxcf.pass != 0) cpi->scaled_ref_buf[ref_frame - 1] = NULL; } } } static void release_scaled_references(AV1_COMP *cpi) { // TODO(isbs): only refresh the necessary frames, rather than all of them for (int i = 0; i < INTER_REFS_PER_FRAME; ++i) { RefCntBuffer *const buf = cpi->scaled_ref_buf[i]; if (buf != NULL) { --buf->ref_count; cpi->scaled_ref_buf[i] = NULL; } } } static void set_mv_search_params(AV1_COMP *cpi) { const AV1_COMMON *const cm = &cpi->common; const unsigned int max_mv_def = AOMMIN(cm->width, cm->height); // Default based on max resolution. cpi->mv_step_param = av1_init_search_range(max_mv_def); if (cpi->sf.mv.auto_mv_step_size) { if (frame_is_intra_only(cm)) { // Initialize max_mv_magnitude for use in the first INTER frame // after a key/intra-only frame. cpi->max_mv_magnitude = max_mv_def; } else { if (cm->show_frame) { // Allow mv_steps to correspond to twice the max mv magnitude found // in the previous frame, capped by the default max_mv_magnitude based // on resolution. cpi->mv_step_param = av1_init_search_range( AOMMIN(max_mv_def, 2 * cpi->max_mv_magnitude)); } cpi->max_mv_magnitude = 0; } } } static void set_size_independent_vars(AV1_COMP *cpi) { int i; AV1_COMMON *cm = &cpi->common; for (i = LAST_FRAME; i <= ALTREF_FRAME; ++i) { cm->global_motion[i] = default_warp_params; } cpi->global_motion_search_done = 0; av1_set_speed_features_framesize_independent(cpi, cpi->speed); av1_set_rd_speed_thresholds(cpi); cm->interp_filter = SWITCHABLE; cm->switchable_motion_mode = 1; if (frame_is_intra_only(cm)) { if (cm->seq_params.force_screen_content_tools == 2) { cm->allow_screen_content_tools = cpi->oxcf.content == AOM_CONTENT_SCREEN || is_screen_content(cpi->source->y_buffer, cpi->source->flags & YV12_FLAG_HIGHBITDEPTH, cm->seq_params.bit_depth, cpi->source->y_stride, cpi->source->y_width, cpi->source->y_height); } else { cm->allow_screen_content_tools = cm->seq_params.force_screen_content_tools; } } cpi->is_screen_content_type = (cm->allow_screen_content_tools != 0); } static void set_size_dependent_vars(AV1_COMP *cpi, int *q, int *bottom_index, int *top_index) { AV1_COMMON *const cm = &cpi->common; const AV1EncoderConfig *const oxcf = &cpi->oxcf; // Setup variables that depend on the dimensions of the frame. av1_set_speed_features_framesize_dependent(cpi, cpi->speed); // Decide q and q bounds. *q = av1_rc_pick_q_and_bounds(cpi, cm->width, cm->height, bottom_index, top_index); if (!frame_is_intra_only(cm)) { set_high_precision_mv(cpi, (*q) < HIGH_PRECISION_MV_QTHRESH, cpi->common.cur_frame_force_integer_mv); } // Configure experimental use of segmentation for enhanced coding of // static regions if indicated. // Only allowed in the second pass of a two pass encode, as it requires // lagged coding, and if the relevant speed feature flag is set. if (oxcf->pass == 2 && cpi->sf.static_segmentation) configure_static_seg_features(cpi); } static void init_motion_estimation(AV1_COMP *cpi) { int y_stride = cpi->scaled_source.y_stride; if (cpi->sf.mv.search_method == NSTEP) { av1_init3smotion_compensation(&cpi->ss_cfg, y_stride); } else if (cpi->sf.mv.search_method == DIAMOND) { av1_init_dsmotion_compensation(&cpi->ss_cfg, y_stride); } } #define COUPLED_CHROMA_FROM_LUMA_RESTORATION 0 static void set_restoration_unit_size(int width, int height, int sx, int sy, RestorationInfo *rst) { (void)width; (void)height; (void)sx; (void)sy; #if COUPLED_CHROMA_FROM_LUMA_RESTORATION int s = AOMMIN(sx, sy); #else int s = 0; #endif // !COUPLED_CHROMA_FROM_LUMA_RESTORATION if (width * height > 352 * 288) rst[0].restoration_unit_size = RESTORATION_UNITSIZE_MAX; else rst[0].restoration_unit_size = (RESTORATION_UNITSIZE_MAX >> 1); rst[1].restoration_unit_size = rst[0].restoration_unit_size >> s; rst[2].restoration_unit_size = rst[1].restoration_unit_size; } static void init_ref_frame_bufs(AV1_COMP *cpi) { AV1_COMMON *const cm = &cpi->common; int i; BufferPool *const pool = cm->buffer_pool; cm->cur_frame = NULL; for (i = 0; i < REF_FRAMES; ++i) { cm->ref_frame_map[i] = NULL; } for (i = 0; i < FRAME_BUFFERS; ++i) { pool->frame_bufs[i].ref_count = 0; } if (cm->seq_params.force_screen_content_tools) { for (i = 0; i < FRAME_BUFFERS; ++i) { av1_hash_table_init(&pool->frame_bufs[i].hash_table, &cpi->td.mb); } } } static void check_initial_width(AV1_COMP *cpi, int use_highbitdepth, int subsampling_x, int subsampling_y) { AV1_COMMON *const cm = &cpi->common; SequenceHeader *const seq_params = &cm->seq_params; if (!cpi->initial_width || seq_params->use_highbitdepth != use_highbitdepth || seq_params->subsampling_x != subsampling_x || seq_params->subsampling_y != subsampling_y) { seq_params->subsampling_x = subsampling_x; seq_params->subsampling_y = subsampling_y; seq_params->use_highbitdepth = use_highbitdepth; alloc_raw_frame_buffers(cpi); init_ref_frame_bufs(cpi); alloc_util_frame_buffers(cpi); init_motion_estimation(cpi); // TODO(agrange) This can be removed. cpi->initial_width = cm->width; cpi->initial_height = cm->height; cpi->initial_mbs = cm->MBs; } } // Returns 1 if the assigned width or height was <= 0. static int set_size_literal(AV1_COMP *cpi, int width, int height) { AV1_COMMON *cm = &cpi->common; const int num_planes = av1_num_planes(cm); check_initial_width(cpi, cm->seq_params.use_highbitdepth, cm->seq_params.subsampling_x, cm->seq_params.subsampling_y); if (width <= 0 || height <= 0) return 1; cm->width = width; cm->height = height; if (cpi->initial_width && cpi->initial_height && (cm->width > cpi->initial_width || cm->height > cpi->initial_height)) { av1_free_context_buffers(cm); av1_free_pc_tree(&cpi->td, num_planes); alloc_compressor_data(cpi); realloc_segmentation_maps(cpi); cpi->initial_width = cpi->initial_height = 0; } update_frame_size(cpi); return 0; } void av1_set_frame_size(AV1_COMP *cpi, int width, int height) { AV1_COMMON *const cm = &cpi->common; const SequenceHeader *const seq_params = &cm->seq_params; const int num_planes = av1_num_planes(cm); MACROBLOCKD *const xd = &cpi->td.mb.e_mbd; int ref_frame; if (width != cm->width || height != cm->height) { // There has been a change in the encoded frame size set_size_literal(cpi, width, height); set_mv_search_params(cpi); // Recalculate 'all_lossless' in case super-resolution was (un)selected. cm->all_lossless = cm->coded_lossless && !av1_superres_scaled(cm); } if (cpi->oxcf.pass == 2) { av1_set_target_rate(cpi, cm->width, cm->height); } alloc_frame_mvs(cm, cm->cur_frame); // Allocate above context buffers if (cm->num_allocated_above_context_planes < av1_num_planes(cm) || cm->num_allocated_above_context_mi_col < cm->mi_cols || cm->num_allocated_above_contexts < cm->tile_rows) { av1_free_above_context_buffers(cm, cm->num_allocated_above_contexts); if (av1_alloc_above_context_buffers(cm, cm->tile_rows)) aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, "Failed to allocate context buffers"); } // Reset the frame pointers to the current frame size. if (aom_realloc_frame_buffer( &cm->cur_frame->buf, cm->width, cm->height, seq_params->subsampling_x, seq_params->subsampling_y, seq_params->use_highbitdepth, cpi->oxcf.border_in_pixels, cm->byte_alignment, NULL, NULL, NULL)) aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, "Failed to allocate frame buffer"); const int frame_width = cm->superres_upscaled_width; const int frame_height = cm->superres_upscaled_height; set_restoration_unit_size(frame_width, frame_height, seq_params->subsampling_x, seq_params->subsampling_y, cm->rst_info); for (int i = 0; i < num_planes; ++i) cm->rst_info[i].frame_restoration_type = RESTORE_NONE; av1_alloc_restoration_buffers(cm); alloc_util_frame_buffers(cpi); init_motion_estimation(cpi); for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) { RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame); if (buf != NULL) { struct scale_factors *sf = get_ref_scale_factors(cm, ref_frame); av1_setup_scale_factors_for_frame(sf, buf->buf.y_crop_width, buf->buf.y_crop_height, cm->width, cm->height); if (av1_is_scaled(sf)) aom_extend_frame_borders(&buf->buf, num_planes); } } av1_setup_scale_factors_for_frame(&cm->sf_identity, cm->width, cm->height, cm->width, cm->height); set_ref_ptrs(cm, xd, LAST_FRAME, LAST_FRAME); } static uint8_t calculate_next_resize_scale(const AV1_COMP *cpi) { // Choose an arbitrary random number static unsigned int seed = 56789; const AV1EncoderConfig *oxcf = &cpi->oxcf; if (oxcf->pass == 1) return SCALE_NUMERATOR; uint8_t new_denom = SCALE_NUMERATOR; if (cpi->common.seq_params.reduced_still_picture_hdr) return SCALE_NUMERATOR; switch (oxcf->resize_mode) { case RESIZE_NONE: new_denom = SCALE_NUMERATOR; break; case RESIZE_FIXED: if (cpi->common.current_frame.frame_type == KEY_FRAME) new_denom = oxcf->resize_kf_scale_denominator; else new_denom = oxcf->resize_scale_denominator; break; case RESIZE_RANDOM: new_denom = lcg_rand16(&seed) % 9 + 8; break; default: assert(0); } return new_denom; } #define ENERGY_BY_Q2_THRESH 0.01 #define ENERGY_BY_AC_THRESH 0.2 static uint8_t get_superres_denom_from_qindex_energy(int qindex, double *energy, double threshq, double threshp) { const double q = av1_convert_qindex_to_q(qindex, AOM_BITS_8); const double tq = threshq * q * q; const double tp = threshp * energy[1]; const double thresh = AOMMIN(tq, tp); int k; for (k = 16; k > 8; --k) { if (energy[k - 1] > thresh) break; } return 3 * SCALE_NUMERATOR - k; } static uint8_t get_superres_denom_for_qindex(const AV1_COMP *cpi, int qindex) { double energy[16]; analyze_hor_freq(cpi, energy); /* printf("\nenergy = ["); for (int k = 1; k < 16; ++k) printf("%f, ", energy[k]); printf("]\n"); */ return get_superres_denom_from_qindex_energy( qindex, energy, ENERGY_BY_Q2_THRESH, ENERGY_BY_AC_THRESH); } static uint8_t calculate_next_superres_scale(AV1_COMP *cpi) { // Choose an arbitrary random number static unsigned int seed = 34567; const AV1EncoderConfig *oxcf = &cpi->oxcf; if (oxcf->pass == 1) return SCALE_NUMERATOR; uint8_t new_denom = SCALE_NUMERATOR; // Make sure that superres mode of the frame is consistent with the // sequence-level flag. assert(IMPLIES(oxcf->superres_mode != SUPERRES_NONE, cpi->common.seq_params.enable_superres)); assert(IMPLIES(!cpi->common.seq_params.enable_superres, oxcf->superres_mode == SUPERRES_NONE)); switch (oxcf->superres_mode) { case SUPERRES_NONE: new_denom = SCALE_NUMERATOR; break; case SUPERRES_FIXED: if (cpi->common.current_frame.frame_type == KEY_FRAME) new_denom = oxcf->superres_kf_scale_denominator; else new_denom = oxcf->superres_scale_denominator; break; case SUPERRES_RANDOM: new_denom = lcg_rand16(&seed) % 9 + 8; break; case SUPERRES_QTHRESH: { // Do not use superres when screen content tools are used. if (cpi->common.allow_screen_content_tools) break; if (oxcf->rc_mode == AOM_VBR || oxcf->rc_mode == AOM_CQ) av1_set_target_rate(cpi, cpi->oxcf.width, cpi->oxcf.height); int bottom_index, top_index; const int q = av1_rc_pick_q_and_bounds( cpi, cpi->oxcf.width, cpi->oxcf.height, &bottom_index, &top_index); const int qthresh = (frame_is_intra_only(&cpi->common)) ? oxcf->superres_kf_qthresh : oxcf->superres_qthresh; if (q <= qthresh) { new_denom = SCALE_NUMERATOR; } else { new_denom = get_superres_denom_for_qindex(cpi, q); } break; } default: assert(0); } return new_denom; } static int dimension_is_ok(int orig_dim, int resized_dim, int denom) { return (resized_dim * SCALE_NUMERATOR >= orig_dim * denom / 2); } static int dimensions_are_ok(int owidth, int oheight, size_params_type *rsz) { // Only need to check the width, as scaling is horizontal only. (void)oheight; return dimension_is_ok(owidth, rsz->resize_width, rsz->superres_denom); } static int validate_size_scales(RESIZE_MODE resize_mode, SUPERRES_MODE superres_mode, int owidth, int oheight, size_params_type *rsz) { if (dimensions_are_ok(owidth, oheight, rsz)) { // Nothing to do. return 1; } // Calculate current resize scale. int resize_denom = AOMMAX(DIVIDE_AND_ROUND(owidth * SCALE_NUMERATOR, rsz->resize_width), DIVIDE_AND_ROUND(oheight * SCALE_NUMERATOR, rsz->resize_height)); if (resize_mode != RESIZE_RANDOM && superres_mode == SUPERRES_RANDOM) { // Alter superres scale as needed to enforce conformity. rsz->superres_denom = (2 * SCALE_NUMERATOR * SCALE_NUMERATOR) / resize_denom; if (!dimensions_are_ok(owidth, oheight, rsz)) { if (rsz->superres_denom > SCALE_NUMERATOR) --rsz->superres_denom; } } else if (resize_mode == RESIZE_RANDOM && superres_mode != SUPERRES_RANDOM) { // Alter resize scale as needed to enforce conformity. resize_denom = (2 * SCALE_NUMERATOR * SCALE_NUMERATOR) / rsz->superres_denom; rsz->resize_width = owidth; rsz->resize_height = oheight; av1_calculate_scaled_size(&rsz->resize_width, &rsz->resize_height, resize_denom); if (!dimensions_are_ok(owidth, oheight, rsz)) { if (resize_denom > SCALE_NUMERATOR) { --resize_denom; rsz->resize_width = owidth; rsz->resize_height = oheight; av1_calculate_scaled_size(&rsz->resize_width, &rsz->resize_height, resize_denom); } } } else if (resize_mode == RESIZE_RANDOM && superres_mode == SUPERRES_RANDOM) { // Alter both resize and superres scales as needed to enforce conformity. do { if (resize_denom > rsz->superres_denom) --resize_denom; else --rsz->superres_denom; rsz->resize_width = owidth; rsz->resize_height = oheight; av1_calculate_scaled_size(&rsz->resize_width, &rsz->resize_height, resize_denom); } while (!dimensions_are_ok(owidth, oheight, rsz) && (resize_denom > SCALE_NUMERATOR || rsz->superres_denom > SCALE_NUMERATOR)); } else { // We are allowed to alter neither resize scale nor superres // scale. return 0; } return dimensions_are_ok(owidth, oheight, rsz); } // Calculates resize and superres params for next frame static size_params_type calculate_next_size_params(AV1_COMP *cpi) { const AV1EncoderConfig *oxcf = &cpi->oxcf; size_params_type rsz = { oxcf->width, oxcf->height, SCALE_NUMERATOR }; int resize_denom; if (oxcf->pass == 1) return rsz; if (cpi->resize_pending_width && cpi->resize_pending_height) { rsz.resize_width = cpi->resize_pending_width; rsz.resize_height = cpi->resize_pending_height; cpi->resize_pending_width = cpi->resize_pending_height = 0; } else { resize_denom = calculate_next_resize_scale(cpi); rsz.resize_width = cpi->oxcf.width; rsz.resize_height = cpi->oxcf.height; av1_calculate_scaled_size(&rsz.resize_width, &rsz.resize_height, resize_denom); } rsz.superres_denom = calculate_next_superres_scale(cpi); if (!validate_size_scales(oxcf->resize_mode, oxcf->superres_mode, oxcf->width, oxcf->height, &rsz)) assert(0 && "Invalid scale parameters"); return rsz; } static void setup_frame_size_from_params(AV1_COMP *cpi, const size_params_type *rsz) { int encode_width = rsz->resize_width; int encode_height = rsz->resize_height; AV1_COMMON *cm = &cpi->common; cm->superres_upscaled_width = encode_width; cm->superres_upscaled_height = encode_height; cm->superres_scale_denominator = rsz->superres_denom; av1_calculate_scaled_superres_size(&encode_width, &encode_height, rsz->superres_denom); av1_set_frame_size(cpi, encode_width, encode_height); } void av1_setup_frame_size(AV1_COMP *cpi) { // Reset superres params from previous frame. cpi->common.superres_scale_denominator = SCALE_NUMERATOR; const size_params_type rsz = calculate_next_size_params(cpi); setup_frame_size_from_params(cpi, &rsz); } static void superres_post_encode(AV1_COMP *cpi) { AV1_COMMON *cm = &cpi->common; const int num_planes = av1_num_planes(cm); if (!av1_superres_scaled(cm)) return; assert(cpi->oxcf.enable_superres); assert(!is_lossless_requested(&cpi->oxcf)); assert(!cm->all_lossless); av1_superres_upscale(cm, NULL); // If regular resizing is occurring the source will need to be downscaled to // match the upscaled superres resolution. Otherwise the original source is // used. if (!av1_resize_scaled(cm)) { cpi->source = cpi->unscaled_source; if (cpi->last_source != NULL) cpi->last_source = cpi->unscaled_last_source; } else { assert(cpi->unscaled_source->y_crop_width != cm->superres_upscaled_width); assert(cpi->unscaled_source->y_crop_height != cm->superres_upscaled_height); // Do downscale. cm->(width|height) has been updated by // av1_superres_upscale if (aom_realloc_frame_buffer( &cpi->scaled_source, cm->superres_upscaled_width, cm->superres_upscaled_height, cm->seq_params.subsampling_x, cm->seq_params.subsampling_y, cm->seq_params.use_highbitdepth, AOM_BORDER_IN_PIXELS, cm->byte_alignment, NULL, NULL, NULL)) aom_internal_error( &cm->error, AOM_CODEC_MEM_ERROR, "Failed to reallocate scaled source buffer for superres"); assert(cpi->scaled_source.y_crop_width == cm->superres_upscaled_width); assert(cpi->scaled_source.y_crop_height == cm->superres_upscaled_height); av1_resize_and_extend_frame(cpi->unscaled_source, &cpi->scaled_source, (int)cm->seq_params.bit_depth, num_planes); cpi->source = &cpi->scaled_source; } } static void loopfilter_frame(AV1_COMP *cpi, AV1_COMMON *cm) { const int num_planes = av1_num_planes(cm); MACROBLOCKD *xd = &cpi->td.mb.e_mbd; assert(IMPLIES(is_lossless_requested(&cpi->oxcf), cm->coded_lossless && cm->all_lossless)); const int no_loopfilter = cm->coded_lossless || cm->large_scale_tile; const int no_cdef = !cm->seq_params.enable_cdef || cm->coded_lossless || cm->large_scale_tile; const int no_restoration = !cm->seq_params.enable_restoration || cm->all_lossless || cm->large_scale_tile; struct loopfilter *lf = &cm->lf; if (no_loopfilter) { lf->filter_level[0] = 0; lf->filter_level[1] = 0; } else { struct aom_usec_timer timer; aom_clear_system_state(); aom_usec_timer_start(&timer); av1_pick_filter_level(cpi->source, cpi, cpi->sf.lpf_pick); aom_usec_timer_mark(&timer); cpi->time_pick_lpf += aom_usec_timer_elapsed(&timer); } if (lf->filter_level[0] || lf->filter_level[1]) { if (cpi->num_workers > 1) av1_loop_filter_frame_mt(&cm->cur_frame->buf, cm, xd, 0, num_planes, 0, #if LOOP_FILTER_BITMASK 0, #endif cpi->workers, cpi->num_workers, &cpi->lf_row_sync); else av1_loop_filter_frame(&cm->cur_frame->buf, cm, xd, #if LOOP_FILTER_BITMASK 0, #endif 0, num_planes, 0); } if (!no_restoration) av1_loop_restoration_save_boundary_lines(&cm->cur_frame->buf, cm, 0); if (no_cdef) { cm->cdef_info.cdef_bits = 0; cm->cdef_info.cdef_strengths[0] = 0; cm->cdef_info.nb_cdef_strengths = 1; cm->cdef_info.cdef_uv_strengths[0] = 0; } else { // Find CDEF parameters av1_cdef_search(&cm->cur_frame->buf, cpi->source, cm, xd, cpi->sf.fast_cdef_search); // Apply the filter av1_cdef_frame(&cm->cur_frame->buf, cm, xd); } superres_post_encode(cpi); if (no_restoration) { cm->rst_info[0].frame_restoration_type = RESTORE_NONE; cm->rst_info[1].frame_restoration_type = RESTORE_NONE; cm->rst_info[2].frame_restoration_type = RESTORE_NONE; } else { av1_loop_restoration_save_boundary_lines(&cm->cur_frame->buf, cm, 1); av1_pick_filter_restoration(cpi->source, cpi); if (cm->rst_info[0].frame_restoration_type != RESTORE_NONE || cm->rst_info[1].frame_restoration_type != RESTORE_NONE || cm->rst_info[2].frame_restoration_type != RESTORE_NONE) { if (cpi->num_workers > 1) av1_loop_restoration_filter_frame_mt(&cm->cur_frame->buf, cm, 0, cpi->workers, cpi->num_workers, &cpi->lr_row_sync, &cpi->lr_ctxt); else av1_loop_restoration_filter_frame(&cm->cur_frame->buf, cm, 0, &cpi->lr_ctxt); } } } static int get_refresh_frame_flags(const AV1_COMP *const cpi) { const AV1_COMMON *const cm = &cpi->common; // Switch frames and shown key-frames overwrite all reference slots if ((cm->current_frame.frame_type == KEY_FRAME && cm->show_frame) || frame_is_sframe(cm)) return 0xFF; int refresh_mask = 0; // NOTE(zoeliu): When LAST_FRAME is to get refreshed, the decoder will be // notified to get LAST3_FRAME refreshed and then the virtual indexes for all // the 3 LAST reference frames will be updated accordingly, i.e.: // (1) The original virtual index for LAST3_FRAME will become the new virtual // index for LAST_FRAME; and // (2) The original virtual indexes for LAST_FRAME and LAST2_FRAME will be // shifted and become the new virtual indexes for LAST2_FRAME and // LAST3_FRAME. refresh_mask |= (cpi->refresh_last_frame << get_ref_frame_map_idx(cm, LAST3_FRAME)); const int bwd_ref_frame = (cpi->new_bwdref_update_rule == 1) ? EXTREF_FRAME : BWDREF_FRAME; refresh_mask |= (cpi->refresh_bwd_ref_frame << get_ref_frame_map_idx(cm, bwd_ref_frame)); refresh_mask |= (cpi->refresh_alt2_ref_frame << get_ref_frame_map_idx(cm, ALTREF2_FRAME)); if (av1_preserve_existing_gf(cpi)) { // We have decided to preserve the previously existing golden frame as our // new ARF frame. However, in the short term we leave it in the GF slot and, // if we're updating the GF with the current decoded frame, we save it // instead to the ARF slot. // Later, in the function av1_encoder.c:av1_update_reference_frames() we // will swap gld_fb_idx and alt_fb_idx to achieve our objective. We do it // there so that it can be done outside of the recode loop. // Note: This is highly specific to the use of ARF as a forward reference, // and this needs to be generalized as other uses are implemented // (like RTC/temporal scalability). if (!cpi->preserve_arf_as_gld) { refresh_mask |= (cpi->refresh_golden_frame << get_ref_frame_map_idx(cm, ALTREF_FRAME)); } } else { refresh_mask |= (cpi->refresh_golden_frame << get_ref_frame_map_idx(cm, GOLDEN_FRAME)); refresh_mask |= (cpi->refresh_alt_ref_frame << get_ref_frame_map_idx(cm, ALTREF_FRAME)); } return refresh_mask; } static void fix_interp_filter(InterpFilter *const interp_filter, const FRAME_COUNTS *const counts) { if (*interp_filter == SWITCHABLE) { // Check to see if only one of the filters is actually used int count[SWITCHABLE_FILTERS] = { 0 }; int num_filters_used = 0; for (int i = 0; i < SWITCHABLE_FILTERS; ++i) { for (int j = 0; j < SWITCHABLE_FILTER_CONTEXTS; ++j) count[i] += counts->switchable_interp[j][i]; num_filters_used += (count[i] > 0); } if (num_filters_used == 1) { // Only one filter is used. So set the filter at frame level for (int i = 0; i < SWITCHABLE_FILTERS; ++i) { if (count[i]) { if (i == EIGHTTAP_REGULAR) *interp_filter = i; break; } } } } } static void finalize_encoded_frame(AV1_COMP *const cpi) { AV1_COMMON *const cm = &cpi->common; CurrentFrame *const current_frame = &cm->current_frame; // This bitfield indicates which reference frame slots will be overwritten by // the current frame current_frame->refresh_frame_flags = get_refresh_frame_flags(cpi); if (!cm->seq_params.reduced_still_picture_hdr && encode_show_existing_frame(cm)) { RefCntBuffer *const frame_to_show = cm->ref_frame_map[cpi->existing_fb_idx_to_show]; if (frame_to_show == NULL) { aom_internal_error(&cm->error, AOM_CODEC_UNSUP_BITSTREAM, "Buffer does not contain a reconstructed frame"); } assert(frame_to_show->ref_count > 0); assign_frame_buffer_p(&cm->cur_frame, frame_to_show); } if (!encode_show_existing_frame(cm) && cm->seq_params.film_grain_params_present && (cm->show_frame || cm->showable_frame)) { // Copy the current frame's film grain params to the its corresponding // RefCntBuffer slot. cm->cur_frame->film_grain_params = cm->film_grain_params; // We must update the parameters if this is not an INTER_FRAME if (current_frame->frame_type != INTER_FRAME) cm->cur_frame->film_grain_params.update_parameters = 1; // Iterate the random seed for the next frame. cm->film_grain_params.random_seed += 3381; if (cm->film_grain_params.random_seed == 0) cm->film_grain_params.random_seed = 7391; } // Initialise all tiles' contexts from the global frame context for (int tile_col = 0; tile_col < cm->tile_cols; tile_col++) { for (int tile_row = 0; tile_row < cm->tile_rows; tile_row++) { const int tile_idx = tile_row * cm->tile_cols + tile_col; cpi->tile_data[tile_idx].tctx = *cm->fc; } } fix_interp_filter(&cm->interp_filter, cpi->td.counts); } // Called after encode_with_recode_loop() has just encoded a frame and packed // its bitstream. This function works out whether we under- or over-shot // our bitrate target and adjusts q as appropriate. Also decides whether // or not we should do another recode loop, indicated by *loop static void recode_loop_update_q(AV1_COMP *const cpi, int *const loop, int *const q, int *const q_low, int *const q_high, const int top_index, const int bottom_index, int *const undershoot_seen, int *const overshoot_seen, const int loop_at_this_size) { AV1_COMMON *const cm = &cpi->common; RATE_CONTROL *const rc = &cpi->rc; int frame_over_shoot_limit = 0, frame_under_shoot_limit = 0; av1_rc_compute_frame_size_bounds(cpi, rc->this_frame_target, &frame_under_shoot_limit, &frame_over_shoot_limit); if (frame_over_shoot_limit == 0) frame_over_shoot_limit = 1; if ((cm->current_frame.frame_type == KEY_FRAME) && rc->this_key_frame_forced && (rc->projected_frame_size < rc->max_frame_bandwidth)) { int last_q = *q; int64_t kf_err; int64_t high_err_target = cpi->ambient_err; int64_t low_err_target = cpi->ambient_err >> 1; if (cm->seq_params.use_highbitdepth) { kf_err = aom_highbd_get_y_sse(cpi->source, &cm->cur_frame->buf); } else { kf_err = aom_get_y_sse(cpi->source, &cm->cur_frame->buf); } // Prevent possible divide by zero error below for perfect KF kf_err += !kf_err; // The key frame is not good enough or we can afford // to make it better without undue risk of popping. if ((kf_err > high_err_target && rc->projected_frame_size <= frame_over_shoot_limit) || (kf_err > low_err_target && rc->projected_frame_size <= frame_under_shoot_limit)) { // Lower q_high *q_high = *q > *q_low ? *q - 1 : *q_low; // Adjust Q *q = (int)((*q * high_err_target) / kf_err); *q = AOMMIN(*q, (*q_high + *q_low) >> 1); } else if (kf_err < low_err_target && rc->projected_frame_size >= frame_under_shoot_limit) { // The key frame is much better than the previous frame // Raise q_low *q_low = *q < *q_high ? *q + 1 : *q_high; // Adjust Q *q = (int)((*q * low_err_target) / kf_err); *q = AOMMIN(*q, (*q_high + *q_low + 1) >> 1); } // Clamp Q to upper and lower limits: *q = clamp(*q, *q_low, *q_high); *loop = *q != last_q; } else if (recode_loop_test(cpi, frame_over_shoot_limit, frame_under_shoot_limit, *q, AOMMAX(*q_high, top_index), bottom_index)) { // Is the projected frame size out of range and are we allowed // to attempt to recode. int last_q = *q; int retries = 0; // Frame size out of permitted range: // Update correction factor & compute new Q to try... // Frame is too large if (rc->projected_frame_size > rc->this_frame_target) { // Special case if the projected size is > the max allowed. if (rc->projected_frame_size >= rc->max_frame_bandwidth) *q_high = rc->worst_quality; // Raise Qlow as to at least the current value *q_low = *q < *q_high ? *q + 1 : *q_high; if (*undershoot_seen || loop_at_this_size > 1) { // Update rate_correction_factor unless av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height); *q = (*q_high + *q_low + 1) / 2; } else { // Update rate_correction_factor unless av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height); *q = av1_rc_regulate_q(cpi, rc->this_frame_target, bottom_index, AOMMAX(*q_high, top_index), cm->width, cm->height); while (*q < *q_low && retries < 10) { av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height); *q = av1_rc_regulate_q(cpi, rc->this_frame_target, bottom_index, AOMMAX(*q_high, top_index), cm->width, cm->height); retries++; } } *overshoot_seen = 1; } else { // Frame is too small *q_high = *q > *q_low ? *q - 1 : *q_low; if (*overshoot_seen || loop_at_this_size > 1) { av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height); *q = (*q_high + *q_low) / 2; } else { av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height); *q = av1_rc_regulate_q(cpi, rc->this_frame_target, bottom_index, top_index, cm->width, cm->height); // Special case reset for qlow for constrained quality. // This should only trigger where there is very substantial // undershoot on a frame and the auto cq level is above // the user passsed in value. if (cpi->oxcf.rc_mode == AOM_CQ && *q < *q_low) { *q_low = *q; } while (*q > *q_high && retries < 10) { av1_rc_update_rate_correction_factors(cpi, cm->width, cm->height); *q = av1_rc_regulate_q(cpi, rc->this_frame_target, bottom_index, top_index, cm->width, cm->height); retries++; } } *undershoot_seen = 1; } // Clamp Q to upper and lower limits: *q = clamp(*q, *q_low, *q_high); *loop = (*q != last_q); } else { *loop = 0; } } static int encode_with_recode_loop(AV1_COMP *cpi, size_t *size, uint8_t *dest) { AV1_COMMON *const cm = &cpi->common; RATE_CONTROL *const rc = &cpi->rc; const int allow_recode = cpi->sf.recode_loop != DISALLOW_RECODE; set_size_independent_vars(cpi); cpi->source->buf_8bit_valid = 0; av1_setup_frame_size(cpi); int top_index = 0, bottom_index = 0; int q = 0, q_low = 0, q_high = 0; set_size_dependent_vars(cpi, &q, &bottom_index, &top_index); q_low = bottom_index; q_high = top_index; // Loop variables int loop_count = 0; int loop_at_this_size = 0; int loop = 0; int overshoot_seen = 0; int undershoot_seen = 0; do { aom_clear_system_state(); // if frame was scaled calculate global_motion_search again if already // done if (loop_count > 0 && cpi->source && cpi->global_motion_search_done) { if (cpi->source->y_crop_width != cm->width || cpi->source->y_crop_height != cm->height) { cpi->global_motion_search_done = 0; } } cpi->source = av1_scale_if_required(cm, cpi->unscaled_source, &cpi->scaled_source); if (cpi->unscaled_last_source != NULL) { cpi->last_source = av1_scale_if_required(cm, cpi->unscaled_last_source, &cpi->scaled_last_source); } if (!frame_is_intra_only(cm)) { if (loop_count > 0) { release_scaled_references(cpi); } scale_references(cpi); } av1_set_quantizer(cm, q); av1_init_quantizer(cpi); // printf("Frame %d/%d: q = %d, frame_type = %d superres_denom = %d\n", // cm->current_frame.frame_number, cm->show_frame, q, // cm->current_frame.frame_type, cm->superres_scale_denominator); if (loop_count == 0) { setup_frame(cpi); } else if (get_primary_ref_frame_buf(cm) == NULL) { // Base q-index may have changed, so we need to assign proper default coef // probs before every iteration. av1_default_coef_probs(cm); av1_setup_frame_contexts(cm); } if (cpi->oxcf.aq_mode == VARIANCE_AQ) { av1_vaq_frame_setup(cpi); } else if (cpi->oxcf.aq_mode == COMPLEXITY_AQ) { av1_setup_in_frame_q_adj(cpi); } else if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ && !allow_recode) { suppress_active_map(cpi); av1_cyclic_refresh_setup(cpi); apply_active_map(cpi); } if (cm->seg.enabled) { if (!cm->seg.update_data && cm->prev_frame) { segfeatures_copy(&cm->seg, &cm->prev_frame->seg); } else { calculate_segdata(&cm->seg); } } else { memset(&cm->seg, 0, sizeof(cm->seg)); } segfeatures_copy(&cm->cur_frame->seg, &cm->seg); if (allow_recode) save_coding_context(cpi); // transform / motion compensation build reconstruction frame av1_encode_frame(cpi); // Update some stats from cyclic refresh, and check if we should not update // golden reference, for 1 pass CBR. if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ && cm->current_frame.frame_type != KEY_FRAME && (cpi->oxcf.pass == 0 && cpi->oxcf.rc_mode == AOM_CBR)) { av1_cyclic_refresh_check_golden_update(cpi); } aom_clear_system_state(); // Dummy pack of the bitstream using up to date stats to get an // accurate estimate of output frame size to determine if we need // to recode. if (cpi->sf.recode_loop >= ALLOW_RECODE_KFARFGF) { restore_coding_context(cpi); finalize_encoded_frame(cpi); int largest_tile_id = 0; // Output from bitstream: unused here if (av1_pack_bitstream(cpi, dest, size, &largest_tile_id) != AOM_CODEC_OK) return AOM_CODEC_ERROR; rc->projected_frame_size = (int)(*size) << 3; restore_coding_context(cpi); } if (allow_recode && cpi->oxcf.rc_mode != AOM_Q) { // Update q and decide whether to do a recode loop recode_loop_update_q(cpi, &loop, &q, &q_low, &q_high, top_index, bottom_index, &undershoot_seen, &overshoot_seen, loop_at_this_size); } // Special case for overlay frame. if (rc->is_src_frame_alt_ref && rc->projected_frame_size < rc->max_frame_bandwidth) loop = 0; if (allow_recode && !cpi->sf.gm_disable_recode && recode_loop_test_global_motion(cpi)) { loop = 1; } if (loop) { ++loop_count; ++loop_at_this_size; #if CONFIG_INTERNAL_STATS ++cpi->tot_recode_hits; #endif } } while (loop); return AOM_CODEC_OK; } #define DUMP_RECON_FRAMES 0 #if DUMP_RECON_FRAMES == 1 // NOTE(zoeliu): For debug - Output the filtered reconstructed video. static void dump_filtered_recon_frames(AV1_COMP *cpi) { AV1_COMMON *const cm = &cpi->common; const CurrentFrame *const current_frame = &cm->current_frame; const YV12_BUFFER_CONFIG *recon_buf = &cm->cur_frame->buf; if (recon_buf == NULL) { printf("Frame %d is not ready.\n", current_frame->frame_number); return; } static const int flag_list[REF_FRAMES] = { 0, AOM_LAST_FLAG, AOM_LAST2_FLAG, AOM_LAST3_FLAG, AOM_GOLD_FLAG, AOM_BWD_FLAG, AOM_ALT2_FLAG, AOM_ALT_FLAG }; printf( "\n***Frame=%d (frame_offset=%d, show_frame=%d, " "show_existing_frame=%d) " "[LAST LAST2 LAST3 GOLDEN BWD ALT2 ALT]=[", current_frame->frame_number, current_frame->order_hint, cm->show_frame, cm->show_existing_frame); for (int ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) { const RefCntBuffer *const buf = get_ref_frame_buf(cm, ref_frame); const int ref_offset = buf != NULL ? (int)buf->order_hint : -1; printf(" %d(%c-%d-%4.2f)", ref_offset, (cpi->ref_frame_flags & flag_list[ref_frame]) ? 'Y' : 'N', buf ? (int)buf->frame_rf_level : -1, buf ? rate_factor_deltas[buf->frame_rf_level] : -1); } printf(" ]\n"); if (!cm->show_frame) { printf("Frame %d is a no show frame, so no image dump.\n", current_frame->frame_number); return; } int h; char file_name[256] = "/tmp/enc_filtered_recon.yuv"; FILE *f_recon = NULL; if (current_frame->frame_number == 0) { if ((f_recon = fopen(file_name, "wb")) == NULL) { printf("Unable to open file %s to write.\n", file_name); return; } } else { if ((f_recon = fopen(file_name, "ab")) == NULL) { printf("Unable to open file %s to append.\n", file_name); return; } } printf( "\nFrame=%5d, encode_update_type[%5d]=%1d, frame_offset=%d, " "show_frame=%d, show_existing_frame=%d, source_alt_ref_active=%d, " "refresh_alt_ref_frame=%d, rf_level=%d, " "y_stride=%4d, uv_stride=%4d, cm->width=%4d, cm->height=%4d\n\n", current_frame->frame_number, cpi->twopass.gf_group.index, cpi->twopass.gf_group.update_type[cpi->twopass.gf_group.index], current_frame->order_hint, cm->show_frame, cm->show_existing_frame, cpi->rc.source_alt_ref_active, cpi->refresh_alt_ref_frame, cpi->twopass.gf_group.rf_level[cpi->twopass.gf_group.index], recon_buf->y_stride, recon_buf->uv_stride, cm->width, cm->height); #if 0 int ref_frame; printf("get_ref_frame_map_idx: ["); for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) printf(" %d", get_ref_frame_map_idx(cm, ref_frame)); printf(" ]\n"); #endif // 0 // --- Y --- for (h = 0; h < cm->height; ++h) { fwrite(&recon_buf->y_buffer[h * recon_buf->y_stride], 1, cm->width, f_recon); } // --- U --- for (h = 0; h < (cm->height >> 1); ++h) { fwrite(&recon_buf->u_buffer[h * recon_buf->uv_stride], 1, (cm->width >> 1), f_recon); } // --- V --- for (h = 0; h < (cm->height >> 1); ++h) { fwrite(&recon_buf->v_buffer[h * recon_buf->uv_stride], 1, (cm->width >> 1), f_recon); } fclose(f_recon); } #endif // DUMP_RECON_FRAMES static int setup_interp_filter_search_mask(AV1_COMP *cpi) { InterpFilters ifilter; int ref_total[REF_FRAMES] = { 0 }; MV_REFERENCE_FRAME ref; int mask = 0; int arf_idx = ALTREF_FRAME; if (cpi->common.last_frame_type == KEY_FRAME || cpi->refresh_alt_ref_frame) return mask; for (ref = LAST_FRAME; ref <= ALTREF_FRAME; ++ref) for (ifilter = EIGHTTAP_REGULAR; ifilter <= MULTITAP_SHARP; ++ifilter) ref_total[ref] += cpi->interp_filter_selected[ref][ifilter]; for (ifilter = EIGHTTAP_REGULAR; ifilter <= MULTITAP_SHARP; ++ifilter) { if ((ref_total[LAST_FRAME] && cpi->interp_filter_selected[LAST_FRAME][ifilter] * 30 <= ref_total[LAST_FRAME]) && (((cpi->interp_filter_selected[LAST2_FRAME][ifilter] * 20) + (cpi->interp_filter_selected[LAST3_FRAME][ifilter] * 20) + (cpi->interp_filter_selected[GOLDEN_FRAME][ifilter] * 20) + (cpi->interp_filter_selected[BWDREF_FRAME][ifilter] * 10) + (cpi->interp_filter_selected[ALTREF2_FRAME][ifilter] * 10) + (cpi->interp_filter_selected[arf_idx][ifilter] * 10)) < (ref_total[LAST2_FRAME] + ref_total[LAST3_FRAME] + ref_total[GOLDEN_FRAME] + ref_total[BWDREF_FRAME] + ref_total[ALTREF2_FRAME] + ref_total[ALTREF_FRAME]))) mask |= 1 << ifilter; } return mask; } static int is_integer_mv(AV1_COMP *cpi, const YV12_BUFFER_CONFIG *cur_picture, const YV12_BUFFER_CONFIG *last_picture, hash_table *last_hash_table) { aom_clear_system_state(); // check use hash ME int k; uint32_t hash_value_1; uint32_t hash_value_2; const int block_size = 8; const double threshold_current = 0.8; const double threshold_average = 0.95; const int max_history_size = 32; int T = 0; // total block int C = 0; // match with collocated block int S = 0; // smooth region but not match with collocated block int M = 0; // match with other block const int pic_width = cur_picture->y_width; const int pic_height = cur_picture->y_height; for (int i = 0; i + block_size <= pic_height; i += block_size) { for (int j = 0; j + block_size <= pic_width; j += block_size) { const int x_pos = j; const int y_pos = i; int match = 1; T++; // check whether collocated block match with current uint8_t *p_cur = cur_picture->y_buffer; uint8_t *p_ref = last_picture->y_buffer; int stride_cur = cur_picture->y_stride; int stride_ref = last_picture->y_stride; p_cur += (y_pos * stride_cur + x_pos); p_ref += (y_pos * stride_ref + x_pos); if (cur_picture->flags & YV12_FLAG_HIGHBITDEPTH) { uint16_t *p16_cur = CONVERT_TO_SHORTPTR(p_cur); uint16_t *p16_ref = CONVERT_TO_SHORTPTR(p_ref); for (int tmpY = 0; tmpY < block_size && match; tmpY++) { for (int tmpX = 0; tmpX < block_size && match; tmpX++) { if (p16_cur[tmpX] != p16_ref[tmpX]) { match = 0; } } p16_cur += stride_cur; p16_ref += stride_ref; } } else { for (int tmpY = 0; tmpY < block_size && match; tmpY++) { for (int tmpX = 0; tmpX < block_size && match; tmpX++) { if (p_cur[tmpX] != p_ref[tmpX]) { match = 0; } } p_cur += stride_cur; p_ref += stride_ref; } } if (match) { C++; continue; } if (av1_hash_is_horizontal_perfect(cur_picture, block_size, x_pos, y_pos) || av1_hash_is_vertical_perfect(cur_picture, block_size, x_pos, y_pos)) { S++; continue; } av1_get_block_hash_value( cur_picture->y_buffer + y_pos * stride_cur + x_pos, stride_cur, block_size, &hash_value_1, &hash_value_2, (cur_picture->flags & YV12_FLAG_HIGHBITDEPTH), &cpi->td.mb); // Hashing does not work for highbitdepth currently. // TODO(Roger): Make it work for highbitdepth. if (av1_use_hash_me(&cpi->common)) { if (av1_has_exact_match(last_hash_table, hash_value_1, hash_value_2)) { M++; } } } } assert(T > 0); double csm_rate = ((double)(C + S + M)) / ((double)(T)); double m_rate = ((double)(M)) / ((double)(T)); cpi->csm_rate_array[cpi->rate_index] = csm_rate; cpi->m_rate_array[cpi->rate_index] = m_rate; cpi->rate_index = (cpi->rate_index + 1) % max_history_size; cpi->rate_size++; cpi->rate_size = AOMMIN(cpi->rate_size, max_history_size); if (csm_rate < threshold_current) { return 0; } if (C == T) { return 1; } double csm_average = 0.0; double m_average = 0.0; for (k = 0; k < cpi->rate_size; k++) { csm_average += cpi->csm_rate_array[k]; m_average += cpi->m_rate_array[k]; } csm_average /= cpi->rate_size; m_average /= cpi->rate_size; if (csm_average < threshold_average) { return 0; } if (M > (T - C - S) / 3) { return 1; } if (csm_rate > 0.99 && m_rate > 0.01) { return 1; } if (csm_average + m_average > 1.01) { return 1; } return 0; } static int encode_frame_to_data_rate(AV1_COMP *cpi, size_t *size, uint8_t *dest, unsigned int *frame_flags) { AV1_COMMON *const cm = &cpi->common; SequenceHeader *const seq_params = &cm->seq_params; CurrentFrame *const current_frame = &cm->current_frame; const AV1EncoderConfig *const oxcf = &cpi->oxcf; struct segmentation *const seg = &cm->seg; // frame type has been decided outside of this function call cm->cur_frame->frame_type = current_frame->frame_type; cm->large_scale_tile = cpi->oxcf.large_scale_tile; cm->single_tile_decoding = cpi->oxcf.single_tile_decoding; cm->allow_ref_frame_mvs &= frame_might_allow_ref_frame_mvs(cm); // cm->allow_ref_frame_mvs needs to be written into the frame header while // cm->large_scale_tile is 1, therefore, "cm->large_scale_tile=1" case is // separated from frame_might_allow_ref_frame_mvs(). cm->allow_ref_frame_mvs &= !cm->large_scale_tile; cm->allow_warped_motion = cpi->oxcf.allow_warped_motion && frame_might_allow_warped_motion(cm); cm->last_frame_type = current_frame->frame_type; if (cpi->oxcf.pass == 2 && cpi->sf.adaptive_interp_filter_search) cpi->sf.interp_filter_search_mask = setup_interp_filter_search_mask(cpi); if (encode_show_existing_frame(cm)) { // NOTE(zoeliu): In BIDIR_PRED, the existing frame to show is the current // BWDREF_FRAME in the reference frame buffer. cpi->frame_flags = *frame_flags; restore_coding_context(cpi); finalize_encoded_frame(cpi); // Build the bitstream int largest_tile_id = 0; // Output from bitstream: unused here if (av1_pack_bitstream(cpi, dest, size, &largest_tile_id) != AOM_CODEC_OK) return AOM_CODEC_ERROR; if (seq_params->frame_id_numbers_present_flag && current_frame->frame_type == KEY_FRAME) { // Displaying a forward key-frame, so reset the ref buffer IDs int display_frame_id = cm->ref_frame_id[cpi->existing_fb_idx_to_show]; for (int i = 0; i < REF_FRAMES; i++) cm->ref_frame_id[i] = display_frame_id; } cpi->seq_params_locked = 1; #if DUMP_RECON_FRAMES == 1 // NOTE(zoeliu): For debug - Output the filtered reconstructed video. dump_filtered_recon_frames(cpi); #endif // DUMP_RECON_FRAMES // Update the LAST_FRAME in the reference frame buffer. // NOTE: // (1) For BWDREF_FRAME as the show_existing_frame, the reference frame // update has been done previously when handling the LAST_BIPRED_FRAME // right before BWDREF_FRAME (in the display order); // (2) For INTNL_OVERLAY as the show_existing_frame, the reference frame // update will be done when the following is called, which will // exchange // the virtual indexes between LAST_FRAME and ALTREF2_FRAME, so that // LAST3 will get retired, LAST2 becomes LAST3, LAST becomes LAST2, // and // ALTREF2_FRAME will serve as the new LAST_FRAME. update_reference_frames(cpi); // Update frame flags cpi->frame_flags &= ~FRAMEFLAGS_GOLDEN; cpi->frame_flags &= ~FRAMEFLAGS_BWDREF; cpi->frame_flags &= ~FRAMEFLAGS_ALTREF; *frame_flags = cpi->frame_flags & ~FRAMEFLAGS_KEY; // Since we allocate a spot for the OVERLAY frame in the gf group, we need // to do post-encoding update accordingly. if (cpi->rc.is_src_frame_alt_ref) { av1_set_target_rate(cpi, cm->width, cm->height); av1_rc_postencode_update(cpi, *size); } ++current_frame->frame_number; return AOM_CODEC_OK; } // Work out whether to force_integer_mv this frame if (oxcf->pass != 1 && cpi->common.allow_screen_content_tools && !frame_is_intra_only(cm)) { if (cpi->common.seq_params.force_integer_mv == 2) { // Adaptive mode: see what previous frame encoded did if (cpi->unscaled_last_source != NULL) { cm->cur_frame_force_integer_mv = is_integer_mv(cpi, cpi->source, cpi->unscaled_last_source, cpi->previous_hash_table); } else { cpi->common.cur_frame_force_integer_mv = 0; } } else { cpi->common.cur_frame_force_integer_mv = cpi->common.seq_params.force_integer_mv; } } else { cpi->common.cur_frame_force_integer_mv = 0; } // Set default state for segment based loop filter update flags. cm->lf.mode_ref_delta_update = 0; // Set various flags etc to special state if it is a key frame. if (frame_is_intra_only(cm) || frame_is_sframe(cm)) { // Reset the loop filter deltas and segmentation map. av1_reset_segment_features(cm); // If segmentation is enabled force a map update for key frames. if (seg->enabled) { seg->update_map = 1; seg->update_data = 1; } // The alternate reference frame cannot be active for a key frame. cpi->rc.source_alt_ref_active = 0; } if (cpi->oxcf.mtu == 0) { cm->num_tg = cpi->oxcf.num_tile_groups; } else { // Use a default value for the purposes of weighting costs in probability // updates cm->num_tg = DEFAULT_MAX_NUM_TG; } // For 1 pass CBR, check if we are dropping this frame. // Never drop on key frame. if (oxcf->pass == 0 && oxcf->rc_mode == AOM_CBR && current_frame->frame_type != KEY_FRAME) { if (av1_rc_drop_frame(cpi)) { av1_rc_postencode_update_drop_frame(cpi); release_scaled_references(cpi); return AOM_CODEC_OK; } } aom_clear_system_state(); #if CONFIG_INTERNAL_STATS memset(cpi->mode_chosen_counts, 0, MAX_MODES * sizeof(*cpi->mode_chosen_counts)); #endif if (seq_params->frame_id_numbers_present_flag) { /* Non-normative definition of current_frame_id ("frame counter" with * wraparound) */ if (cm->current_frame_id == -1) { int lsb, msb; /* quasi-random initialization of current_frame_id for a key frame */ if (cpi->source->flags & YV12_FLAG_HIGHBITDEPTH) { lsb = CONVERT_TO_SHORTPTR(cpi->source->y_buffer)[0] & 0xff; msb = CONVERT_TO_SHORTPTR(cpi->source->y_buffer)[1] & 0xff; } else { lsb = cpi->source->y_buffer[0] & 0xff; msb = cpi->source->y_buffer[1] & 0xff; } cm->current_frame_id = ((msb << 8) + lsb) % (1 << seq_params->frame_id_length); // S_frame is meant for stitching different streams of different // resolutions together, so current_frame_id must be the // same across different streams of the same content current_frame_id // should be the same and not random. 0x37 is a chosen number as start // point if (cpi->oxcf.sframe_enabled) cm->current_frame_id = 0x37; } else { cm->current_frame_id = (cm->current_frame_id + 1 + (1 << seq_params->frame_id_length)) % (1 << seq_params->frame_id_length); } } switch (cpi->oxcf.cdf_update_mode) { case 0: // No CDF update for any frames(4~6% compression loss). cm->disable_cdf_update = 1; break; case 1: // Enable CDF update for all frames. cm->disable_cdf_update = 0; break; case 2: // Strategically determine at which frames to do CDF update. // Currently only enable CDF update for all-intra and no-show frames(1.5% // compression loss). // TODO(huisu@google.com): design schemes for various trade-offs between // compression quality and decoding speed. cm->disable_cdf_update = (frame_is_intra_only(cm) || !cm->show_frame) ? 0 : 1; break; } cm->timing_info_present &= !seq_params->reduced_still_picture_hdr; if (encode_with_recode_loop(cpi, size, dest) != AOM_CODEC_OK) return AOM_CODEC_ERROR; #ifdef OUTPUT_YUV_SKINMAP if (cpi->common.current_frame.frame_number > 1) { av1_compute_skin_map(cpi, yuv_skinmap_file); } #endif // OUTPUT_YUV_SKINMAP // Special case code to reduce pulsing when key frames are forced at a // fixed interval. Note the reconstruction error if it is the frame before // the force key frame if (cpi->rc.next_key_frame_forced && cpi->rc.frames_to_key == 1) { if (seq_params->use_highbitdepth) { cpi->ambient_err = aom_highbd_get_y_sse(cpi->source, &cm->cur_frame->buf); } else { cpi->ambient_err = aom_get_y_sse(cpi->source, &cm->cur_frame->buf); } } // If the encoder forced a KEY_FRAME decision or if frame is an S_FRAME if ((current_frame->frame_type == KEY_FRAME && cm->show_frame) || frame_is_sframe(cm)) { cpi->refresh_last_frame = 1; } cm->cur_frame->buf.color_primaries = seq_params->color_primaries; cm->cur_frame->buf.transfer_characteristics = seq_params->transfer_characteristics; cm->cur_frame->buf.matrix_coefficients = seq_params->matrix_coefficients; cm->cur_frame->buf.monochrome = seq_params->monochrome; cm->cur_frame->buf.chroma_sample_position = seq_params->chroma_sample_position; cm->cur_frame->buf.color_range = seq_params->color_range; cm->cur_frame->buf.render_width = cm->render_width; cm->cur_frame->buf.render_height = cm->render_height; // TODO(zoeliu): For non-ref frames, loop filtering may need to be turned // off. // Pick the loop filter level for the frame. if (!cm->allow_intrabc) { loopfilter_frame(cpi, cm); } else { cm->lf.filter_level[0] = 0; cm->lf.filter_level[1] = 0; cm->cdef_info.cdef_bits = 0; cm->cdef_info.cdef_strengths[0] = 0; cm->cdef_info.nb_cdef_strengths = 1; cm->cdef_info.cdef_uv_strengths[0] = 0; cm->rst_info[0].frame_restoration_type = RESTORE_NONE; cm->rst_info[1].frame_restoration_type = RESTORE_NONE; cm->rst_info[2].frame_restoration_type = RESTORE_NONE; } // TODO(debargha): Fix mv search range on encoder side // aom_extend_frame_inner_borders(&cm->cur_frame->buf, av1_num_planes(cm)); aom_extend_frame_borders(&cm->cur_frame->buf, av1_num_planes(cm)); #ifdef OUTPUT_YUV_REC aom_write_one_yuv_frame(cm, &cm->cur_frame->buf); #endif finalize_encoded_frame(cpi); // Build the bitstream int largest_tile_id = 0; // Output from pack_bitstream if (av1_pack_bitstream(cpi, dest, size, &largest_tile_id) != AOM_CODEC_OK) return AOM_CODEC_ERROR; cpi->seq_params_locked = 1; // Update reference frame ids for reference frames this frame will overwrite if (seq_params->frame_id_numbers_present_flag) { for (int i = 0; i < REF_FRAMES; i++) { if ((current_frame->refresh_frame_flags >> i) & 1) { cm->ref_frame_id[i] = cm->current_frame_id; } } } #if DUMP_RECON_FRAMES == 1 // NOTE(zoeliu): For debug - Output the filtered reconstructed video. dump_filtered_recon_frames(cpi); #endif // DUMP_RECON_FRAMES if (cm->seg.enabled) { if (cm->seg.update_map) { update_reference_segmentation_map(cpi); } else if (cm->last_frame_seg_map) { memcpy(cm->cur_frame->seg_map, cm->last_frame_seg_map, cm->mi_cols * cm->mi_rows * sizeof(uint8_t)); } } if (frame_is_intra_only(cm) == 0) { release_scaled_references(cpi); } update_reference_frames(cpi); #if CONFIG_ENTROPY_STATS av1_accumulate_frame_counts(&aggregate_fc, &cpi->counts); #endif // CONFIG_ENTROPY_STATS if (cm->refresh_frame_context == REFRESH_FRAME_CONTEXT_BACKWARD) { *cm->fc = cpi->tile_data[largest_tile_id].tctx; av1_reset_cdf_symbol_counters(cm->fc); } if (!cm->large_scale_tile) { cm->cur_frame->frame_context = *cm->fc; } #define EXT_TILE_DEBUG 0 #if EXT_TILE_DEBUG if (cm->large_scale_tile && oxcf->pass == 2) { char fn[20] = "./fc"; fn[4] = current_frame->frame_number / 100 + '0'; fn[5] = (current_frame->frame_number % 100) / 10 + '0'; fn[6] = (current_frame->frame_number % 10) + '0'; fn[7] = '\0'; av1_print_frame_contexts(cm->fc, fn); } #endif // EXT_TILE_DEBUG #undef EXT_TILE_DEBUG if (cpi->refresh_golden_frame == 1) cpi->frame_flags |= FRAMEFLAGS_GOLDEN; else cpi->frame_flags &= ~FRAMEFLAGS_GOLDEN; if (cpi->refresh_alt_ref_frame == 1) cpi->frame_flags |= FRAMEFLAGS_ALTREF; else cpi->frame_flags &= ~FRAMEFLAGS_ALTREF; if (cpi->refresh_bwd_ref_frame == 1) cpi->frame_flags |= FRAMEFLAGS_BWDREF; else cpi->frame_flags &= ~FRAMEFLAGS_BWDREF; cm->last_frame_type = current_frame->frame_type; av1_rc_postencode_update(cpi, *size); if (current_frame->frame_type == KEY_FRAME) { // Tell the caller that the frame was coded as a key frame *frame_flags = cpi->frame_flags | FRAMEFLAGS_KEY; } else { *frame_flags = cpi->frame_flags & ~FRAMEFLAGS_KEY; } // Store encoded frame's hash table for is_integer_mv() next time if (oxcf->pass != 1 && cpi->common.allow_screen_content_tools) { cpi->previous_hash_table = &cm->cur_frame->hash_table; } // Clear the one shot update flags for segmentation map and mode/ref loop // filter deltas. cm->seg.update_map = 0; cm->seg.update_data = 0; cm->lf.mode_ref_delta_update = 0; // A droppable frame might not be shown but it always // takes a space in the gf group. Therefore, even when // it is not shown, we still need update the count down. if (cm->show_frame) { // TODO(zoeliu): We may only swamp mi and prev_mi for those frames that // are // being used as reference. swap_mi_and_prev_mi(cm); // Don't increment frame counters if this was an altref buffer // update not a real frame ++current_frame->frame_number; } return AOM_CODEC_OK; } int av1_encode(AV1_COMP *const cpi, uint8_t *const dest, const EncodeFrameInput *const frame_input, const EncodeFrameParams *const frame_params, EncodeFrameResults *const frame_results) { AV1_COMMON *const cm = &cpi->common; CurrentFrame *const current_frame = &cm->current_frame; cpi->unscaled_source = frame_input->source; cpi->source = frame_input->source; cpi->unscaled_last_source = frame_input->last_source; cm->error_resilient_mode = frame_params->error_resilient_mode; cm->primary_ref_frame = frame_params->primary_ref_frame; cm->current_frame.frame_type = frame_params->frame_type; cm->show_frame = frame_params->show_frame; cpi->ref_frame_flags = frame_params->ref_frame_flags; cpi->speed = frame_params->speed; if (current_frame->frame_type == KEY_FRAME && cm->show_frame) current_frame->frame_number = 0; if (cm->show_existing_frame) { current_frame->order_hint = cm->cur_frame->order_hint; } else { current_frame->order_hint = current_frame->frame_number + frame_params->order_offset; current_frame->order_hint %= (1 << (cm->seq_params.order_hint_info.order_hint_bits_minus_1 + 1)); } if (cpi->oxcf.pass == 1) { av1_first_pass(cpi, frame_input->ts_duration); } else if (cpi->oxcf.pass == 0 || cpi->oxcf.pass == 2) { if (encode_frame_to_data_rate(cpi, &frame_results->size, dest, frame_params->frame_flags) != AOM_CODEC_OK) { return AOM_CODEC_ERROR; } } else { return AOM_CODEC_ERROR; } return AOM_CODEC_OK; } #if CONFIG_DENOISE static int apply_denoise_2d(AV1_COMP *cpi, YV12_BUFFER_CONFIG *sd, int block_size, float noise_level, int64_t time_stamp, int64_t end_time) { AV1_COMMON *const cm = &cpi->common; if (!cpi->denoise_and_model) { cpi->denoise_and_model = aom_denoise_and_model_alloc( cm->seq_params.bit_depth, block_size, noise_level); if (!cpi->denoise_and_model) { aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, "Error allocating denoise and model"); return -1; } } if (!cpi->film_grain_table) { cpi->film_grain_table = aom_malloc(sizeof(*cpi->film_grain_table)); if (!cpi->film_grain_table) { aom_internal_error(&cm->error, AOM_CODEC_MEM_ERROR, "Error allocating grain table"); return -1; } memset(cpi->film_grain_table, 0, sizeof(*cpi->film_grain_table)); } if (aom_denoise_and_model_run(cpi->denoise_and_model, sd, &cm->film_grain_params)) { if (cm->film_grain_params.apply_grain) { aom_film_grain_table_append(cpi->film_grain_table, time_stamp, end_time, &cm->film_grain_params); } } return 0; } #endif int av1_receive_raw_frame(AV1_COMP *cpi, aom_enc_frame_flags_t frame_flags, YV12_BUFFER_CONFIG *sd, int64_t time_stamp, int64_t end_time) { AV1_COMMON *const cm = &cpi->common; const SequenceHeader *const seq_params = &cm->seq_params; struct aom_usec_timer timer; int res = 0; const int subsampling_x = sd->subsampling_x; const int subsampling_y = sd->subsampling_y; const int use_highbitdepth = (sd->flags & YV12_FLAG_HIGHBITDEPTH) != 0; check_initial_width(cpi, use_highbitdepth, subsampling_x, subsampling_y); aom_usec_timer_start(&timer); #if CONFIG_DENOISE if (cpi->oxcf.noise_level > 0) if (apply_denoise_2d(cpi, sd, cpi->oxcf.noise_block_size, cpi->oxcf.noise_level, time_stamp, end_time) < 0) res = -1; #endif // CONFIG_DENOISE if (av1_lookahead_push(cpi->lookahead, sd, time_stamp, end_time, use_highbitdepth, frame_flags)) res = -1; aom_usec_timer_mark(&timer); cpi->time_receive_data += aom_usec_timer_elapsed(&timer); if ((seq_params->profile == PROFILE_0) && !seq_params->monochrome && (subsampling_x != 1 || subsampling_y != 1)) { aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM, "Non-4:2:0 color format requires profile 1 or 2"); res = -1; } if ((seq_params->profile == PROFILE_1) && !(subsampling_x == 0 && subsampling_y == 0)) { aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM, "Profile 1 requires 4:4:4 color format"); res = -1; } if ((seq_params->profile == PROFILE_2) && (seq_params->bit_depth <= AOM_BITS_10) && !(subsampling_x == 1 && subsampling_y == 0)) { aom_internal_error(&cm->error, AOM_CODEC_INVALID_PARAM, "Profile 2 bit-depth < 10 requires 4:2:2 color format"); res = -1; } return res; } #if CONFIG_INTERNAL_STATS extern double av1_get_blockiness(const unsigned char *img1, int img1_pitch, const unsigned char *img2, int img2_pitch, int width, int height); static void adjust_image_stat(double y, double u, double v, double all, ImageStat *s) { s->stat[STAT_Y] += y; s->stat[STAT_U] += u; s->stat[STAT_V] += v; s->stat[STAT_ALL] += all; s->worst = AOMMIN(s->worst, all); } static void compute_internal_stats(AV1_COMP *cpi, int frame_bytes) { AV1_COMMON *const cm = &cpi->common; double samples = 0.0; uint32_t in_bit_depth = 8; uint32_t bit_depth = 8; #if CONFIG_INTER_STATS_ONLY if (cm->current_frame.frame_type == KEY_FRAME) return; // skip key frame #endif cpi->bytes += frame_bytes; if (cm->seq_params.use_highbitdepth) { in_bit_depth = cpi->oxcf.input_bit_depth; bit_depth = cm->seq_params.bit_depth; } if (cm->show_frame) { const YV12_BUFFER_CONFIG *orig = cpi->source; const YV12_BUFFER_CONFIG *recon = &cpi->common.cur_frame->buf; double y, u, v, frame_all; cpi->count++; if (cpi->b_calculate_psnr) { PSNR_STATS psnr; double frame_ssim2 = 0.0, weight = 0.0; aom_clear_system_state(); // TODO(yaowu): unify these two versions into one. aom_calc_highbd_psnr(orig, recon, &psnr, bit_depth, in_bit_depth); adjust_image_stat(psnr.psnr[1], psnr.psnr[2], psnr.psnr[3], psnr.psnr[0], &cpi->psnr); cpi->total_sq_error += psnr.sse[0]; cpi->total_samples += psnr.samples[0]; samples = psnr.samples[0]; // TODO(yaowu): unify these two versions into one. if (cm->seq_params.use_highbitdepth) frame_ssim2 = aom_highbd_calc_ssim(orig, recon, &weight, bit_depth, in_bit_depth); else frame_ssim2 = aom_calc_ssim(orig, recon, &weight); cpi->worst_ssim = AOMMIN(cpi->worst_ssim, frame_ssim2); cpi->summed_quality += frame_ssim2 * weight; cpi->summed_weights += weight; #if 0 { FILE *f = fopen("q_used.stt", "a"); double y2 = psnr.psnr[1]; double u2 = psnr.psnr[2]; double v2 = psnr.psnr[3]; double frame_psnr2 = psnr.psnr[0]; fprintf(f, "%5d : Y%f7.3:U%f7.3:V%f7.3:F%f7.3:S%7.3f\n", cm->current_frame.frame_number, y2, u2, v2, frame_psnr2, frame_ssim2); fclose(f); } #endif } if (cpi->b_calculate_blockiness) { if (!cm->seq_params.use_highbitdepth) { const double frame_blockiness = av1_get_blockiness(orig->y_buffer, orig->y_stride, recon->y_buffer, recon->y_stride, orig->y_width, orig->y_height); cpi->worst_blockiness = AOMMAX(cpi->worst_blockiness, frame_blockiness); cpi->total_blockiness += frame_blockiness; } if (cpi->b_calculate_consistency) { if (!cm->seq_params.use_highbitdepth) { const double this_inconsistency = aom_get_ssim_metrics( orig->y_buffer, orig->y_stride, recon->y_buffer, recon->y_stride, orig->y_width, orig->y_height, cpi->ssim_vars, &cpi->metrics, 1); const double peak = (double)((1 << in_bit_depth) - 1); const double consistency = aom_sse_to_psnr(samples, peak, cpi->total_inconsistency); if (consistency > 0.0) cpi->worst_consistency = AOMMIN(cpi->worst_consistency, consistency); cpi->total_inconsistency += this_inconsistency; } } } frame_all = aom_calc_fastssim(orig, recon, &y, &u, &v, bit_depth, in_bit_depth); adjust_image_stat(y, u, v, frame_all, &cpi->fastssim); frame_all = aom_psnrhvs(orig, recon, &y, &u, &v, bit_depth, in_bit_depth); adjust_image_stat(y, u, v, frame_all, &cpi->psnrhvs); } } #endif // CONFIG_INTERNAL_STATS // Don't allow a show_existing_frame to coincide with an error resilient or // S-Frame. An exception can be made in the case of a keyframe, since it does // not depend on any previous frames. static int allow_show_existing(const AV1_COMP *const cpi) { if (cpi->common.current_frame.frame_number == 0) return 0; const struct lookahead_entry *lookahead_src = av1_lookahead_peek(cpi->lookahead, 0); if (lookahead_src == NULL) return 1; const int is_error_resilient = cpi->oxcf.error_resilient_mode || (lookahead_src->flags & AOM_EFLAG_ERROR_RESILIENT); const int is_s_frame = cpi->oxcf.s_frame_mode || (lookahead_src->flags & AOM_EFLAG_SET_S_FRAME); const int is_key_frame = (cpi->rc.frames_to_key == 0) || (cpi->frame_flags & FRAMEFLAGS_KEY); return !(is_error_resilient || is_s_frame) || is_key_frame; } int av1_get_compressed_data(AV1_COMP *cpi, unsigned int *frame_flags, size_t *size, uint8_t *dest, int64_t *time_stamp, int64_t *time_end, int flush, const aom_rational_t *timebase) { const AV1EncoderConfig *const oxcf = &cpi->oxcf; AV1_COMMON *const cm = &cpi->common; struct aom_usec_timer cmptimer; #if CONFIG_BITSTREAM_DEBUG assert(cpi->oxcf.max_threads == 0 && "bitstream debug tool does not support multithreading"); bitstream_queue_record_write(); bitstream_queue_set_frame_write(current_frame->frame_number * 2 + cm->show_frame); #endif // Indicates whether or not to use an adaptive quantize b rather than // the traditional version cm->use_quant_b_adapt = cpi->oxcf.quant_b_adapt; cm->showable_frame = 0; *size = 0; aom_usec_timer_start(&cmptimer); set_high_precision_mv(cpi, ALTREF_HIGH_PRECISION_MV, 0); // Normal defaults cm->refresh_frame_context = oxcf->frame_parallel_decoding_mode ? REFRESH_FRAME_CONTEXT_DISABLED : REFRESH_FRAME_CONTEXT_BACKWARD; if (oxcf->large_scale_tile) cm->refresh_frame_context = REFRESH_FRAME_CONTEXT_DISABLED; // default reference buffers update config av1_configure_buffer_updates(cpi, LF_UPDATE); // Initialize fields related to forward keyframes cpi->no_show_kf = 0; cm->show_existing_frame &= allow_show_existing(cpi); if (assign_cur_frame_new_fb(cm) == NULL) return AOM_CODEC_ERROR; const int result = av1_encode_strategy(cpi, size, dest, frame_flags, time_stamp, time_end, timebase, flush); if (result != AOM_CODEC_OK && result != -1) { return AOM_CODEC_ERROR; } else if (result == -1) { // Returning -1 indicates no frame encoded; more input is required return -1; } aom_usec_timer_mark(&cmptimer); cpi->time_compress_data += aom_usec_timer_elapsed(&cmptimer); if (cpi->b_calculate_psnr) { if (cm->show_existing_frame || (oxcf->pass != 1 && cm->show_frame)) { generate_psnr_packet(cpi); } } #if CONFIG_INTERNAL_STATS if (oxcf->pass != 1) { compute_internal_stats(cpi, (int)(*size)); } #endif // CONFIG_INTERNAL_STATS #if CONFIG_SPEED_STATS if (cpi->oxcf.pass != 1 && !cm->show_existing_frame) { cpi->tx_search_count += cpi->td.mb.tx_search_count; cpi->td.mb.tx_search_count = 0; } #endif // CONFIG_SPEED_STATS aom_clear_system_state(); return 0; } int av1_get_preview_raw_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *dest) { AV1_COMMON *cm = &cpi->common; if (!cm->show_frame) { return -1; } else { int ret; if (cm->cur_frame != NULL) { *dest = cm->cur_frame->buf; dest->y_width = cm->width; dest->y_height = cm->height; dest->uv_width = cm->width >> cm->seq_params.subsampling_x; dest->uv_height = cm->height >> cm->seq_params.subsampling_y; ret = 0; } else { ret = -1; } aom_clear_system_state(); return ret; } } int av1_get_last_show_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *frame) { if (cpi->last_show_frame_buf == NULL) return -1; *frame = cpi->last_show_frame_buf->buf; return 0; } static int equal_dimensions_and_border(const YV12_BUFFER_CONFIG *a, const YV12_BUFFER_CONFIG *b) { return a->y_height == b->y_height && a->y_width == b->y_width && a->uv_height == b->uv_height && a->uv_width == b->uv_width && a->y_stride == b->y_stride && a->uv_stride == b->uv_stride && a->border == b->border && (a->flags & YV12_FLAG_HIGHBITDEPTH) == (b->flags & YV12_FLAG_HIGHBITDEPTH); } aom_codec_err_t av1_copy_new_frame_enc(AV1_COMMON *cm, YV12_BUFFER_CONFIG *new_frame, YV12_BUFFER_CONFIG *sd) { const int num_planes = av1_num_planes(cm); if (!equal_dimensions_and_border(new_frame, sd)) aom_internal_error(&cm->error, AOM_CODEC_ERROR, "Incorrect buffer dimensions"); else aom_yv12_copy_frame(new_frame, sd, num_planes); return cm->error.error_code; } int av1_set_internal_size(AV1_COMP *cpi, AOM_SCALING horiz_mode, AOM_SCALING vert_mode) { int hr = 0, hs = 0, vr = 0, vs = 0; if (horiz_mode > ONETWO || vert_mode > ONETWO) return -1; Scale2Ratio(horiz_mode, &hr, &hs); Scale2Ratio(vert_mode, &vr, &vs); // always go to the next whole number cpi->resize_pending_width = (hs - 1 + cpi->oxcf.width * hr) / hs; cpi->resize_pending_height = (vs - 1 + cpi->oxcf.height * vr) / vs; return 0; } int av1_get_quantizer(AV1_COMP *cpi) { return cpi->common.base_qindex; } int av1_convert_sect5obus_to_annexb(uint8_t *buffer, size_t *frame_size) { size_t output_size = 0; size_t total_bytes_read = 0; size_t remaining_size = *frame_size; uint8_t *buff_ptr = buffer; // go through each OBUs while (total_bytes_read < *frame_size) { uint8_t saved_obu_header[2]; uint64_t obu_payload_size; size_t length_of_payload_size; size_t length_of_obu_size; uint32_t obu_header_size = (buff_ptr[0] >> 2) & 0x1 ? 2 : 1; size_t obu_bytes_read = obu_header_size; // bytes read for current obu // save the obu header (1 or 2 bytes) memmove(saved_obu_header, buff_ptr, obu_header_size); // clear the obu_has_size_field saved_obu_header[0] = saved_obu_header[0] & (~0x2); // get the payload_size and length of payload_size if (aom_uleb_decode(buff_ptr + obu_header_size, remaining_size, &obu_payload_size, &length_of_payload_size) != 0) { return AOM_CODEC_ERROR; } obu_bytes_read += length_of_payload_size; // calculate the length of size of the obu header plus payload length_of_obu_size = aom_uleb_size_in_bytes((uint64_t)(obu_header_size + obu_payload_size)); // move the rest of data to new location memmove(buff_ptr + length_of_obu_size + obu_header_size, buff_ptr + obu_bytes_read, remaining_size - obu_bytes_read); obu_bytes_read += (size_t)obu_payload_size; // write the new obu size const uint64_t obu_size = obu_header_size + obu_payload_size; size_t coded_obu_size; if (aom_uleb_encode(obu_size, sizeof(obu_size), buff_ptr, &coded_obu_size) != 0) { return AOM_CODEC_ERROR; } // write the saved (modified) obu_header following obu size memmove(buff_ptr + length_of_obu_size, saved_obu_header, obu_header_size); total_bytes_read += obu_bytes_read; remaining_size -= obu_bytes_read; buff_ptr += length_of_obu_size + obu_size; output_size += length_of_obu_size + (size_t)obu_size; } *frame_size = output_size; return AOM_CODEC_OK; } void av1_apply_encoding_flags(AV1_COMP *cpi, aom_enc_frame_flags_t flags) { // TODO(yunqingwang): For what references to use, external encoding flags // should be consistent with internal reference frame selection. Need to // ensure that there is not conflict between the two. In AV1 encoder, the // priority rank for 7 reference frames are: LAST, ALTREF, LAST2, LAST3, // GOLDEN, BWDREF, ALTREF2. If only one reference frame is used, it must be // LAST. cpi->ext_ref_frame_flags = AOM_REFFRAME_ALL; if (flags & (AOM_EFLAG_NO_REF_LAST | AOM_EFLAG_NO_REF_LAST2 | AOM_EFLAG_NO_REF_LAST3 | AOM_EFLAG_NO_REF_GF | AOM_EFLAG_NO_REF_ARF | AOM_EFLAG_NO_REF_BWD | AOM_EFLAG_NO_REF_ARF2)) { if (flags & AOM_EFLAG_NO_REF_LAST) { cpi->ext_ref_frame_flags = 0; } else { int ref = AOM_REFFRAME_ALL; if (flags & AOM_EFLAG_NO_REF_LAST2) ref ^= AOM_LAST2_FLAG; if (flags & AOM_EFLAG_NO_REF_LAST3) ref ^= AOM_LAST3_FLAG; if (flags & AOM_EFLAG_NO_REF_GF) ref ^= AOM_GOLD_FLAG; if (flags & AOM_EFLAG_NO_REF_ARF) { ref ^= AOM_ALT_FLAG; ref ^= AOM_BWD_FLAG; ref ^= AOM_ALT2_FLAG; } else { if (flags & AOM_EFLAG_NO_REF_BWD) ref ^= AOM_BWD_FLAG; if (flags & AOM_EFLAG_NO_REF_ARF2) ref ^= AOM_ALT2_FLAG; } av1_use_as_reference(cpi, ref); } } if (flags & (AOM_EFLAG_NO_UPD_LAST | AOM_EFLAG_NO_UPD_GF | AOM_EFLAG_NO_UPD_ARF)) { int upd = AOM_REFFRAME_ALL; // Refreshing LAST/LAST2/LAST3 is handled by 1 common flag. if (flags & AOM_EFLAG_NO_UPD_LAST) upd ^= AOM_LAST_FLAG; if (flags & AOM_EFLAG_NO_UPD_GF) upd ^= AOM_GOLD_FLAG; if (flags & AOM_EFLAG_NO_UPD_ARF) { upd ^= AOM_ALT_FLAG; upd ^= AOM_BWD_FLAG; upd ^= AOM_ALT2_FLAG; } cpi->ext_refresh_last_frame = (upd & AOM_LAST_FLAG) != 0; cpi->ext_refresh_golden_frame = (upd & AOM_GOLD_FLAG) != 0; cpi->ext_refresh_alt_ref_frame = (upd & AOM_ALT_FLAG) != 0; cpi->ext_refresh_bwd_ref_frame = (upd & AOM_BWD_FLAG) != 0; cpi->ext_refresh_alt2_ref_frame = (upd & AOM_ALT2_FLAG) != 0; cpi->ext_refresh_frame_flags_pending = 1; } cpi->ext_use_ref_frame_mvs = cpi->oxcf.allow_ref_frame_mvs & ((flags & AOM_EFLAG_NO_REF_FRAME_MVS) == 0); cpi->ext_use_error_resilient = cpi->oxcf.error_resilient_mode | ((flags & AOM_EFLAG_ERROR_RESILIENT) != 0); cpi->ext_use_s_frame = cpi->oxcf.s_frame_mode | ((flags & AOM_EFLAG_SET_S_FRAME) != 0); cpi->ext_use_primary_ref_none = (flags & AOM_EFLAG_SET_PRIMARY_REF_NONE) != 0; if (flags & AOM_EFLAG_NO_UPD_ENTROPY) { av1_update_entropy(cpi, 0); } } aom_fixed_buf_t *av1_get_global_headers(AV1_COMP *cpi) { if (!cpi) return NULL; uint8_t header_buf[512] = { 0 }; const uint32_t sequence_header_size = write_sequence_header_obu(cpi, &header_buf[0]); assert(sequence_header_size <= sizeof(header_buf)); if (sequence_header_size == 0) return NULL; const size_t obu_header_size = 1; const size_t size_field_size = aom_uleb_size_in_bytes(sequence_header_size); const size_t payload_offset = obu_header_size + size_field_size; if (payload_offset + sequence_header_size > sizeof(header_buf)) return NULL; memmove(&header_buf[payload_offset], &header_buf[0], sequence_header_size); if (write_obu_header(OBU_SEQUENCE_HEADER, 0, &header_buf[0]) != obu_header_size) { return NULL; } size_t coded_size_field_size = 0; if (aom_uleb_encode(sequence_header_size, size_field_size, &header_buf[obu_header_size], &coded_size_field_size) != 0) { return NULL; } assert(coded_size_field_size == size_field_size); aom_fixed_buf_t *global_headers = (aom_fixed_buf_t *)malloc(sizeof(*global_headers)); if (!global_headers) return NULL; const size_t global_header_buf_size = obu_header_size + size_field_size + sequence_header_size; global_headers->buf = malloc(global_header_buf_size); if (!global_headers->buf) { free(global_headers); return NULL; } memcpy(global_headers->buf, &header_buf[0], global_header_buf_size); global_headers->sz = global_header_buf_size; return global_headers; }