aom/av1/encoder/encoder.h

846 строки
26 KiB
C
Исходник Обычный вид История

/*
* 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.
*/
#ifndef AV1_ENCODER_ENCODER_H_
#define AV1_ENCODER_ENCODER_H_
#include <stdio.h>
#include "./aom_config.h"
#include "aom/aomcx.h"
#include "av1/common/alloccommon.h"
#include "av1/common/entropymode.h"
#include "av1/common/thread_common.h"
#include "av1/common/onyxc_int.h"
#include "av1/encoder/aq_cyclicrefresh.h"
#if CONFIG_ANS
#include "aom_dsp/buf_ans.h"
#endif
#include "av1/encoder/context_tree.h"
#include "av1/encoder/encodemb.h"
#include "av1/encoder/firstpass.h"
#include "av1/encoder/lookahead.h"
#include "av1/encoder/mbgraph.h"
#include "av1/encoder/mcomp.h"
#include "av1/encoder/quantize.h"
#include "av1/encoder/ratectrl.h"
#include "av1/encoder/rd.h"
#include "av1/encoder/speed_features.h"
#include "av1/encoder/tokenize.h"
#include "av1/encoder/variance_tree.h"
#if CONFIG_INTERNAL_STATS
#include "aom_dsp/ssim.h"
#endif
#include "aom_dsp/variance.h"
#include "aom/internal/aom_codec_internal.h"
#include "aom_util/aom_thread.h"
#ifdef __cplusplus
extern "C" {
#endif
typedef struct {
int nmvjointcost[MV_JOINTS];
int nmvcosts[2][MV_VALS];
int nmvcosts_hp[2][MV_VALS];
#if CONFIG_REF_MV
int nmv_vec_cost[NMV_CONTEXTS][MV_JOINTS];
int nmv_costs[NMV_CONTEXTS][2][MV_VALS];
int nmv_costs_hp[NMV_CONTEXTS][2][MV_VALS];
#endif
// 0 = Intra, Last, GF, ARF
signed char last_ref_lf_deltas[TOTAL_REFS_PER_FRAME];
// 0 = ZERO_MV, MV
signed char last_mode_lf_deltas[MAX_MODE_LF_DELTAS];
FRAME_CONTEXT fc;
} CODING_CONTEXT;
typedef enum {
// regular inter frame
REGULAR_FRAME = 0,
// alternate reference frame
ARF_FRAME = 1,
// overlay frame
OVERLAY_FRAME = 2,
// golden frame
GLD_FRAME = 3,
#if CONFIG_EXT_REFS
// backward reference frame
BRF_FRAME = 4,
// extra alternate reference frame
EXT_ARF_FRAME = 5
#endif
} FRAME_CONTEXT_INDEX;
typedef enum {
NORMAL = 0,
FOURFIVE = 1,
THREEFIVE = 2,
ONETWO = 3
} AOM_SCALING;
typedef enum {
// Good Quality Fast Encoding. The encoder balances quality with the amount of
// time it takes to encode the output. Speed setting controls how fast.
GOOD,
// The encoder places priority on the quality of the output over encoding
// speed. The output is compressed at the highest possible quality. This
// option takes the longest amount of time to encode. Speed setting ignored.
BEST,
// Realtime/Live Encoding. This mode is optimized for realtime encoding (for
// example, capturing a television signal or feed from a live camera). Speed
// setting controls how fast.
REALTIME
} MODE;
typedef enum {
FRAMEFLAGS_KEY = 1 << 0,
FRAMEFLAGS_GOLDEN = 1 << 1,
#if CONFIG_EXT_REFS
FRAMEFLAGS_BWDREF = 1 << 2,
FRAMEFLAGS_ALTREF = 1 << 3,
#else
FRAMEFLAGS_ALTREF = 1 << 2,
#endif // CONFIG_EXT_REFS
} FRAMETYPE_FLAGS;
typedef enum {
NO_AQ = 0,
VARIANCE_AQ = 1,
COMPLEXITY_AQ = 2,
CYCLIC_REFRESH_AQ = 3,
#if CONFIG_DELTA_Q
DELTA_AQ = 4,
#endif
AQ_MODE_COUNT // This should always be the last member of the enum
} AQ_MODE;
typedef enum {
RESIZE_NONE = 0, // No frame resizing allowed.
RESIZE_FIXED = 1, // All frames are coded at the specified dimension.
RESIZE_DYNAMIC = 2 // Coded size of each frame is determined by the codec.
} RESIZE_TYPE;
typedef struct AV1EncoderConfig {
BITSTREAM_PROFILE profile;
aom_bit_depth_t bit_depth; // Codec bit-depth.
int width; // width of data passed to the compressor
int height; // height of data passed to the compressor
unsigned int input_bit_depth; // Input bit depth.
double init_framerate; // set to passed in framerate
int64_t target_bandwidth; // bandwidth to be used in bits per second
int noise_sensitivity; // pre processing blur: recommendation 0
int sharpness; // sharpening output: recommendation 0:
int speed;
// maximum allowed bitrate for any intra frame in % of bitrate target.
unsigned int rc_max_intra_bitrate_pct;
// maximum allowed bitrate for any inter frame in % of bitrate target.
unsigned int rc_max_inter_bitrate_pct;
// percent of rate boost for golden frame in CBR mode.
unsigned int gf_cbr_boost_pct;
MODE mode;
int pass;
// Key Framing Operations
int auto_key; // autodetect cut scenes and set the keyframes
int key_freq; // maximum distance to key frame.
int lag_in_frames; // how many frames lag before we start encoding
// ----------------------------------------------------------------
// DATARATE CONTROL OPTIONS
// vbr, cbr, constrained quality or constant quality
enum aom_rc_mode rc_mode;
// buffer targeting aggressiveness
int under_shoot_pct;
int over_shoot_pct;
// buffering parameters
int64_t starting_buffer_level_ms;
int64_t optimal_buffer_level_ms;
int64_t maximum_buffer_size_ms;
// Frame drop threshold.
int drop_frames_water_mark;
// controlling quality
int fixed_q;
int worst_allowed_q;
int best_allowed_q;
int cq_level;
AQ_MODE aq_mode; // Adaptive Quantization mode
#if CONFIG_AOM_QM
int using_qm;
int qm_minlevel;
int qm_maxlevel;
#endif
#if CONFIG_TILE_GROUPS
unsigned int num_tile_groups;
unsigned int mtu;
#endif
// Internal frame size scaling.
RESIZE_TYPE resize_mode;
int scaled_frame_width;
int scaled_frame_height;
// Enable feature to reduce the frame quantization every x frames.
int frame_periodic_boost;
// two pass datarate control
int two_pass_vbrbias; // two pass datarate control tweaks
int two_pass_vbrmin_section;
int two_pass_vbrmax_section;
// END DATARATE CONTROL OPTIONS
// ----------------------------------------------------------------
int enable_auto_arf;
#if CONFIG_EXT_REFS
int enable_auto_brf; // (b)ackward (r)ef (f)rame
#endif // CONFIG_EXT_REFS
/* Bitfield defining the error resiliency features to enable.
* Can provide decodable frames after losses in previous
* frames and decodable partitions after losses in the same frame.
*/
unsigned int error_resilient_mode;
/* Bitfield defining the parallel decoding mode where the
* decoding in successive frames may be conducted in parallel
* just by decoding the frame headers.
*/
unsigned int frame_parallel_decoding_mode;
int arnr_max_frames;
int arnr_strength;
int min_gf_interval;
int max_gf_interval;
int tile_columns;
int tile_rows;
#if CONFIG_DEBLOCKING_ACROSS_TILES
int loop_filter_across_tiles_enabled;
#endif // CONFIG_DEBLOCKING_ACROSS_TILES
int max_threads;
aom_fixed_buf_t two_pass_stats_in;
struct aom_codec_pkt_list *output_pkt_list;
#if CONFIG_FP_MB_STATS
aom_fixed_buf_t firstpass_mb_stats_in;
#endif
aom_tune_metric tuning;
aom_tune_content content;
#if CONFIG_AOM_HIGHBITDEPTH
int use_highbitdepth;
#endif
aom_color_space_t color_space;
int color_range;
int render_width;
int render_height;
Make superblock size variable at the frame level. The uncompressed frame header contains a bit to signal whether the frame is encoded using 64x64 or 128x128 superblocks. This can vary between any 2 frames. vpxenc gained the --sb-size={64,128,dynamic} option, which allows the configuration of the superblock size used (default is dynamic). 64/128 will force the encoder to always use the specified superblock size. Dynamic would enable the encoder to choose the sb size for each frame, but this is not implemented yet (dynamic does the same as 128 for now). Constraints on tile sizes depend on the superblock size, the following is a summary of the current bitstream syntax and semantics: If both --enable-ext-tile is OFF and --enable-ext-partition is OFF: The tile coding in this case is the same as VP9. In particular, tiles have a minimum width of 256 pixels and a maximum width of 4096 pixels. The tile width must be multiples of 64 pixels (except for the rightmost tile column). There can be a maximum of 64 tile columns and 4 tile rows. If --enable-ext-tile is OFF and --enable-ext-partition is ON: Same constraints as above, except that tile width must be multiples of 128 pixels (except for the rightmost tile column). There is no change in the bitstream syntax used for coding the tile configuration if --enable-ext-tile is OFF. If --enable-ext-tile is ON and --enable-ext-partition is ON: This is the new large scale tile coding configuration. The minimum/maximum tile width and height are 64/4096 pixels. Tile width and height must be multiples of 64 pixels. The uncompressed header contains two 6 bit fields that hold the tile width/heigh in units of 64 pixels. The maximum number of tile rows/columns is only limited by the maximum frame size of 65536x65536 pixels that can be coded in the bitstream. This yields a maximum of 1024x1024 tile rows and columns (of 64x64 tiles in a 65536x65536 frame). If both --enable-ext-tile is ON and --enable-ext-partition is ON: Same applies as above, except that in the bitstream the 2 fields containing the tile width/height are in units of the superblock size, and the superblock size itself is also coded in the bitstream. If the uncompressed header signals the use of 64x64 superblocks, then the tile width/height fields are 6 bits wide and are in units of 64 pixels. If the uncompressed header signals the use of 128x128 superblocks, then the tile width/height fields are 5 bits wide and are in units of 128 pixels. The above is a summary of the bitstream. The user interface to vpxenc (and the equivalent encoder API) behaves a follows: If --enable-ext-tile is OFF: No change in the user interface. --tile-columns and --tile-rows specify the base 2 logarithm of the desired number of tile columns and tile rows. The actual number of tile rows and tile columns, and the particular tile width and tile height are computed by the codec ensuring all of the above constraints are respected. If --enable-ext-tile is ON, but --enable-ext-partition is OFF: No change in the user interface. --tile-columns and --tile-rows specify the WIDTH and HEIGHT of the tiles in unit of 64 pixels. The valid values are in the range [1, 64] (which corresponds to [64, 4096] pixels in increments of 64. If both --enable-ext-tile is ON and --enable-ext-partition is ON: If --sb-size=64 (default): The user interface is the same as in the previous point. --tile-columns and --tile-rows specify tile WIDTH and HEIGHT, in units of 64 pixels, in the range [1, 64] (which corresponds to [64, 4096] pixels in increments of 64). If --sb-size=128 or --sb-size=dynamic: --tile-columns and --tile-rows specify tile WIDTH and HEIGHT, in units of 128 pixels in the range [1, 32] (which corresponds to [128, 4096] pixels in increments of 128). Change-Id: Idc9beee1ad12ff1634e83671985d14c680f9179a
2016-03-24 16:56:05 +03:00
#if CONFIG_EXT_PARTITION
aom_superblock_size_t superblock_size;
Make superblock size variable at the frame level. The uncompressed frame header contains a bit to signal whether the frame is encoded using 64x64 or 128x128 superblocks. This can vary between any 2 frames. vpxenc gained the --sb-size={64,128,dynamic} option, which allows the configuration of the superblock size used (default is dynamic). 64/128 will force the encoder to always use the specified superblock size. Dynamic would enable the encoder to choose the sb size for each frame, but this is not implemented yet (dynamic does the same as 128 for now). Constraints on tile sizes depend on the superblock size, the following is a summary of the current bitstream syntax and semantics: If both --enable-ext-tile is OFF and --enable-ext-partition is OFF: The tile coding in this case is the same as VP9. In particular, tiles have a minimum width of 256 pixels and a maximum width of 4096 pixels. The tile width must be multiples of 64 pixels (except for the rightmost tile column). There can be a maximum of 64 tile columns and 4 tile rows. If --enable-ext-tile is OFF and --enable-ext-partition is ON: Same constraints as above, except that tile width must be multiples of 128 pixels (except for the rightmost tile column). There is no change in the bitstream syntax used for coding the tile configuration if --enable-ext-tile is OFF. If --enable-ext-tile is ON and --enable-ext-partition is ON: This is the new large scale tile coding configuration. The minimum/maximum tile width and height are 64/4096 pixels. Tile width and height must be multiples of 64 pixels. The uncompressed header contains two 6 bit fields that hold the tile width/heigh in units of 64 pixels. The maximum number of tile rows/columns is only limited by the maximum frame size of 65536x65536 pixels that can be coded in the bitstream. This yields a maximum of 1024x1024 tile rows and columns (of 64x64 tiles in a 65536x65536 frame). If both --enable-ext-tile is ON and --enable-ext-partition is ON: Same applies as above, except that in the bitstream the 2 fields containing the tile width/height are in units of the superblock size, and the superblock size itself is also coded in the bitstream. If the uncompressed header signals the use of 64x64 superblocks, then the tile width/height fields are 6 bits wide and are in units of 64 pixels. If the uncompressed header signals the use of 128x128 superblocks, then the tile width/height fields are 5 bits wide and are in units of 128 pixels. The above is a summary of the bitstream. The user interface to vpxenc (and the equivalent encoder API) behaves a follows: If --enable-ext-tile is OFF: No change in the user interface. --tile-columns and --tile-rows specify the base 2 logarithm of the desired number of tile columns and tile rows. The actual number of tile rows and tile columns, and the particular tile width and tile height are computed by the codec ensuring all of the above constraints are respected. If --enable-ext-tile is ON, but --enable-ext-partition is OFF: No change in the user interface. --tile-columns and --tile-rows specify the WIDTH and HEIGHT of the tiles in unit of 64 pixels. The valid values are in the range [1, 64] (which corresponds to [64, 4096] pixels in increments of 64. If both --enable-ext-tile is ON and --enable-ext-partition is ON: If --sb-size=64 (default): The user interface is the same as in the previous point. --tile-columns and --tile-rows specify tile WIDTH and HEIGHT, in units of 64 pixels, in the range [1, 64] (which corresponds to [64, 4096] pixels in increments of 64). If --sb-size=128 or --sb-size=dynamic: --tile-columns and --tile-rows specify tile WIDTH and HEIGHT, in units of 128 pixels in the range [1, 32] (which corresponds to [128, 4096] pixels in increments of 128). Change-Id: Idc9beee1ad12ff1634e83671985d14c680f9179a
2016-03-24 16:56:05 +03:00
#endif // CONFIG_EXT_PARTITION
} AV1EncoderConfig;
static INLINE int is_lossless_requested(const AV1EncoderConfig *cfg) {
return cfg->best_allowed_q == 0 && cfg->worst_allowed_q == 0;
}
// TODO(jingning) All spatially adaptive variables should go to TileDataEnc.
typedef struct TileDataEnc {
TileInfo tile_info;
int thresh_freq_fact[BLOCK_SIZES][MAX_MODES];
int mode_map[BLOCK_SIZES][MAX_MODES];
int m_search_count;
int ex_search_count;
New experiment: Perceptual Vector Quantization from Daala PVQ replaces the scalar quantizer and coefficient coding with a new design originally developed in Daala. It currently depends on the Daala entropy coder although it could be adapted to work with another entropy coder if needed: ./configure --enable-experimental --enable-daala_ec --enable-pvq The version of PVQ in this commit is adapted from the following revision of Daala: https://github.com/xiph/daala/commit/fb51c1ade6a31b668a0157d89de8f0a4493162a8 More information about PVQ: - https://people.xiph.org/~jm/daala/pvq_demo/ - https://jmvalin.ca/papers/spie_pvq.pdf The following files are copied as-is from Daala with minimal adaptations, therefore we disable clang-format on those files to make it easier to synchronize the AV1 and Daala codebases in the future: av1/common/generic_code.c av1/common/generic_code.h av1/common/laplace_tables.c av1/common/partition.c av1/common/partition.h av1/common/pvq.c av1/common/pvq.h av1/common/state.c av1/common/state.h av1/common/zigzag.h av1/common/zigzag16.c av1/common/zigzag32.c av1/common/zigzag4.c av1/common/zigzag64.c av1/common/zigzag8.c av1/decoder/decint.h av1/decoder/generic_decoder.c av1/decoder/laplace_decoder.c av1/decoder/pvq_decoder.c av1/decoder/pvq_decoder.h av1/encoder/daala_compat_enc.c av1/encoder/encint.h av1/encoder/generic_encoder.c av1/encoder/laplace_encoder.c av1/encoder/pvq_encoder.c av1/encoder/pvq_encoder.h Known issues: - Lossless mode is not supported, '--lossless=1' will give the same result as '--end-usage=q --cq-level=1'. - High bit depth is not supported by PVQ. Change-Id: I1ae0d6517b87f4c1ccea944b2e12dc906979f25e
2016-11-05 02:36:56 +03:00
#if CONFIG_PVQ
PVQ_QUEUE pvq_q;
#endif
} TileDataEnc;
typedef struct RD_COUNTS {
av1_coeff_count coef_counts[TX_SIZES][PLANE_TYPES];
int64_t comp_pred_diff[REFERENCE_MODES];
} RD_COUNTS;
typedef struct ThreadData {
MACROBLOCK mb;
RD_COUNTS rd_counts;
FRAME_COUNTS *counts;
PICK_MODE_CONTEXT *leaf_tree;
PC_TREE *pc_tree;
PC_TREE *pc_root[MAX_MIB_SIZE_LOG2 - MIN_MIB_SIZE_LOG2 + 1];
VAR_TREE *var_tree;
VAR_TREE *var_root[MAX_MIB_SIZE_LOG2 - MIN_MIB_SIZE_LOG2 + 1];
} ThreadData;
struct EncWorkerData;
typedef struct ActiveMap {
int enabled;
int update;
unsigned char *map;
} ActiveMap;
#define NUM_STAT_TYPES 4 // types of stats: Y, U, V and ALL
typedef struct IMAGE_STAT {
double stat[NUM_STAT_TYPES];
double worst;
} ImageStat;
#undef NUM_STAT_TYPES
typedef struct {
int ref_count;
YV12_BUFFER_CONFIG buf;
} EncRefCntBuffer;
#if CONFIG_ENTROPY
typedef struct SUBFRAME_STATS {
av1_coeff_probs_model coef_probs_buf[COEF_PROBS_BUFS][TX_SIZES][PLANE_TYPES];
av1_coeff_count coef_counts_buf[COEF_PROBS_BUFS][TX_SIZES][PLANE_TYPES];
unsigned int eob_counts_buf[COEF_PROBS_BUFS][TX_SIZES][PLANE_TYPES][REF_TYPES]
[COEF_BANDS][COEFF_CONTEXTS];
av1_coeff_probs_model enc_starting_coef_probs[TX_SIZES][PLANE_TYPES];
} SUBFRAME_STATS;
#endif // CONFIG_ENTROPY
typedef struct TileBufferEnc {
uint8_t *data;
size_t size;
} TileBufferEnc;
typedef struct AV1_COMP {
QUANTS quants;
ThreadData td;
MB_MODE_INFO_EXT *mbmi_ext_base;
DECLARE_ALIGNED(16, int16_t, y_dequant[QINDEX_RANGE][8]); // 8: SIMD width
DECLARE_ALIGNED(16, int16_t, uv_dequant[QINDEX_RANGE][8]); // 8: SIMD width
#if CONFIG_NEW_QUANT
DECLARE_ALIGNED(16, dequant_val_type_nuq,
y_dequant_val_nuq[QUANT_PROFILES][QINDEX_RANGE][COEF_BANDS]);
DECLARE_ALIGNED(16, dequant_val_type_nuq,
uv_dequant_val_nuq[QUANT_PROFILES][QINDEX_RANGE][COEF_BANDS]);
#endif // CONFIG_NEW_QUANT
AV1_COMMON common;
AV1EncoderConfig oxcf;
struct lookahead_ctx *lookahead;
struct lookahead_entry *alt_ref_source;
YV12_BUFFER_CONFIG *Source;
YV12_BUFFER_CONFIG *Last_Source; // NULL for first frame and alt_ref frames
YV12_BUFFER_CONFIG *un_scaled_source;
YV12_BUFFER_CONFIG scaled_source;
YV12_BUFFER_CONFIG *unscaled_last_source;
YV12_BUFFER_CONFIG scaled_last_source;
// Up-sampled reference buffers
// NOTE(zoeliu): It is needed to allocate sufficient space to the up-sampled
// reference buffers, which should include the up-sampled version of all the
// possibly stored references plus the currently coded frame itself.
EncRefCntBuffer upsampled_ref_bufs[REF_FRAMES + 1];
int upsampled_ref_idx[REF_FRAMES + 1];
// For a still frame, this flag is set to 1 to skip partition search.
int partition_search_skippable_frame;
int scaled_ref_idx[TOTAL_REFS_PER_FRAME];
#if CONFIG_EXT_REFS
int lst_fb_idxes[LAST_REF_FRAMES];
#else
int lst_fb_idx;
#endif // CONFIG_EXT_REFS
int gld_fb_idx;
#if CONFIG_EXT_REFS
int bwd_fb_idx; // BWD_REF_FRAME
#endif // CONFIG_EXT_REFS
int alt_fb_idx;
int last_show_frame_buf_idx; // last show frame buffer index
int refresh_last_frame;
int refresh_golden_frame;
#if CONFIG_EXT_REFS
int refresh_bwd_ref_frame;
#endif // CONFIG_EXT_REFS
int refresh_alt_ref_frame;
int ext_refresh_frame_flags_pending;
int ext_refresh_last_frame;
int ext_refresh_golden_frame;
int ext_refresh_alt_ref_frame;
int ext_refresh_frame_context_pending;
int ext_refresh_frame_context;
YV12_BUFFER_CONFIG last_frame_uf;
#if CONFIG_LOOP_RESTORATION
YV12_BUFFER_CONFIG last_frame_db;
YV12_BUFFER_CONFIG trial_frame_rst;
uint8_t *extra_rstbuf; // Extra buffers used in restoration search
RestorationInfo rst_search[MAX_MB_PLANE]; // Used for encoder side search
#endif // CONFIG_LOOP_RESTORATION
// Ambient reconstruction err target for force key frames
int64_t ambient_err;
RD_OPT rd;
CODING_CONTEXT coding_context;
#if CONFIG_REF_MV
int nmv_costs[NMV_CONTEXTS][2][MV_VALS];
int nmv_costs_hp[NMV_CONTEXTS][2][MV_VALS];
#endif
int nmvcosts[2][MV_VALS];
int nmvcosts_hp[2][MV_VALS];
int nmvsadcosts[2][MV_VALS];
int nmvsadcosts_hp[2][MV_VALS];
int64_t last_time_stamp_seen;
int64_t last_end_time_stamp_seen;
int64_t first_time_stamp_ever;
RATE_CONTROL rc;
double framerate;
// NOTE(zoeliu): Any inter frame allows maximum of REF_FRAMES inter
// references; Plus the currently coded frame itself, it is needed to allocate
// sufficient space to the size of the maximum possible number of frames.
int interp_filter_selected[REF_FRAMES + 1][SWITCHABLE];
struct aom_codec_pkt_list *output_pkt_list;
MBGRAPH_FRAME_STATS mbgraph_stats[MAX_LAG_BUFFERS];
int mbgraph_n_frames; // number of frames filled in the above
int static_mb_pct; // % forced skip mbs by segmentation
int ref_frame_flags;
SPEED_FEATURES sf;
unsigned int max_mv_magnitude;
int mv_step_param;
int allow_comp_inter_inter;
Make superblock size variable at the frame level. The uncompressed frame header contains a bit to signal whether the frame is encoded using 64x64 or 128x128 superblocks. This can vary between any 2 frames. vpxenc gained the --sb-size={64,128,dynamic} option, which allows the configuration of the superblock size used (default is dynamic). 64/128 will force the encoder to always use the specified superblock size. Dynamic would enable the encoder to choose the sb size for each frame, but this is not implemented yet (dynamic does the same as 128 for now). Constraints on tile sizes depend on the superblock size, the following is a summary of the current bitstream syntax and semantics: If both --enable-ext-tile is OFF and --enable-ext-partition is OFF: The tile coding in this case is the same as VP9. In particular, tiles have a minimum width of 256 pixels and a maximum width of 4096 pixels. The tile width must be multiples of 64 pixels (except for the rightmost tile column). There can be a maximum of 64 tile columns and 4 tile rows. If --enable-ext-tile is OFF and --enable-ext-partition is ON: Same constraints as above, except that tile width must be multiples of 128 pixels (except for the rightmost tile column). There is no change in the bitstream syntax used for coding the tile configuration if --enable-ext-tile is OFF. If --enable-ext-tile is ON and --enable-ext-partition is ON: This is the new large scale tile coding configuration. The minimum/maximum tile width and height are 64/4096 pixels. Tile width and height must be multiples of 64 pixels. The uncompressed header contains two 6 bit fields that hold the tile width/heigh in units of 64 pixels. The maximum number of tile rows/columns is only limited by the maximum frame size of 65536x65536 pixels that can be coded in the bitstream. This yields a maximum of 1024x1024 tile rows and columns (of 64x64 tiles in a 65536x65536 frame). If both --enable-ext-tile is ON and --enable-ext-partition is ON: Same applies as above, except that in the bitstream the 2 fields containing the tile width/height are in units of the superblock size, and the superblock size itself is also coded in the bitstream. If the uncompressed header signals the use of 64x64 superblocks, then the tile width/height fields are 6 bits wide and are in units of 64 pixels. If the uncompressed header signals the use of 128x128 superblocks, then the tile width/height fields are 5 bits wide and are in units of 128 pixels. The above is a summary of the bitstream. The user interface to vpxenc (and the equivalent encoder API) behaves a follows: If --enable-ext-tile is OFF: No change in the user interface. --tile-columns and --tile-rows specify the base 2 logarithm of the desired number of tile columns and tile rows. The actual number of tile rows and tile columns, and the particular tile width and tile height are computed by the codec ensuring all of the above constraints are respected. If --enable-ext-tile is ON, but --enable-ext-partition is OFF: No change in the user interface. --tile-columns and --tile-rows specify the WIDTH and HEIGHT of the tiles in unit of 64 pixels. The valid values are in the range [1, 64] (which corresponds to [64, 4096] pixels in increments of 64. If both --enable-ext-tile is ON and --enable-ext-partition is ON: If --sb-size=64 (default): The user interface is the same as in the previous point. --tile-columns and --tile-rows specify tile WIDTH and HEIGHT, in units of 64 pixels, in the range [1, 64] (which corresponds to [64, 4096] pixels in increments of 64). If --sb-size=128 or --sb-size=dynamic: --tile-columns and --tile-rows specify tile WIDTH and HEIGHT, in units of 128 pixels in the range [1, 32] (which corresponds to [128, 4096] pixels in increments of 128). Change-Id: Idc9beee1ad12ff1634e83671985d14c680f9179a
2016-03-24 16:56:05 +03:00
uint8_t *segmentation_map;
CYCLIC_REFRESH *cyclic_refresh;
ActiveMap active_map;
fractional_mv_step_fp *find_fractional_mv_step;
av1_full_search_fn_t full_search_sad; // It is currently unused.
av1_diamond_search_fn_t diamond_search_sad;
aom_variance_fn_ptr_t fn_ptr[BLOCK_SIZES];
uint64_t time_receive_data;
uint64_t time_compress_data;
uint64_t time_pick_lpf;
uint64_t time_encode_sb_row;
#if CONFIG_FP_MB_STATS
int use_fp_mb_stats;
#endif
TWO_PASS twopass;
YV12_BUFFER_CONFIG alt_ref_buffer;
#if CONFIG_INTERNAL_STATS
unsigned int mode_chosen_counts[MAX_MODES];
int count;
uint64_t total_sq_error;
uint64_t total_samples;
ImageStat psnr;
double total_blockiness;
double worst_blockiness;
int bytes;
double summed_quality;
double summed_weights;
unsigned int tot_recode_hits;
double worst_ssim;
ImageStat fastssim;
ImageStat psnrhvs;
int b_calculate_blockiness;
int b_calculate_consistency;
double total_inconsistency;
double worst_consistency;
Ssimv *ssim_vars;
Metrics metrics;
#endif
int b_calculate_psnr;
int droppable;
int initial_width;
int initial_height;
int initial_mbs; // Number of MBs in the full-size frame; to be used to
// normalize the firstpass stats. This will differ from the
// number of MBs in the current frame when the frame is
// scaled.
// Store frame variance info in SOURCE_VAR_BASED_PARTITION search type.
DIFF *source_diff_var;
// The threshold used in SOURCE_VAR_BASED_PARTITION search type.
unsigned int source_var_thresh;
int frames_till_next_var_check;
int frame_flags;
search_site_config ss_cfg;
int mbmode_cost[BLOCK_SIZE_GROUPS][INTRA_MODES];
#if CONFIG_REF_MV
int newmv_mode_cost[NEWMV_MODE_CONTEXTS][2];
int zeromv_mode_cost[ZEROMV_MODE_CONTEXTS][2];
int refmv_mode_cost[REFMV_MODE_CONTEXTS][2];
int drl_mode_cost0[DRL_MODE_CONTEXTS][2];
#if CONFIG_EXT_INTER
int new2mv_mode_cost[2];
#endif // CONFIG_EXT_INTER
#endif
unsigned int inter_mode_cost[INTER_MODE_CONTEXTS][INTER_MODES];
#if CONFIG_EXT_INTER
unsigned int inter_compound_mode_cost[INTER_MODE_CONTEXTS]
[INTER_COMPOUND_MODES];
unsigned int interintra_mode_cost[BLOCK_SIZE_GROUPS][INTERINTRA_MODES];
#endif // CONFIG_EXT_INTER
#if CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
int motion_mode_cost[BLOCK_SIZES][MOTION_MODES];
#if CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
int motion_mode_cost1[BLOCK_SIZES][2];
#endif // CONFIG_MOTION_VAR && CONFIG_WARPED_MOTION
#endif // CONFIG_MOTION_VAR || CONFIG_WARPED_MOTION
int intra_uv_mode_cost[INTRA_MODES][INTRA_MODES];
int y_mode_costs[INTRA_MODES][INTRA_MODES][INTRA_MODES];
int switchable_interp_costs[SWITCHABLE_FILTER_CONTEXTS][SWITCHABLE_FILTERS];
#if CONFIG_EXT_PARTITION_TYPES
int partition_cost[PARTITION_CONTEXTS][EXT_PARTITION_TYPES];
#else
int partition_cost[PARTITION_CONTEXTS][PARTITION_TYPES];
#endif
#if CONFIG_PALETTE
int palette_y_size_cost[PALETTE_BLOCK_SIZES][PALETTE_SIZES];
int palette_uv_size_cost[PALETTE_BLOCK_SIZES][PALETTE_SIZES];
int palette_y_color_cost[PALETTE_MAX_SIZE - 1][PALETTE_COLOR_CONTEXTS]
[PALETTE_COLORS];
int palette_uv_color_cost[PALETTE_MAX_SIZE - 1][PALETTE_COLOR_CONTEXTS]
[PALETTE_COLORS];
#endif // CONFIG_PALETTE
int tx_size_cost[TX_SIZES - 1][TX_SIZE_CONTEXTS][TX_SIZES];
#if CONFIG_EXT_TX
int inter_tx_type_costs[EXT_TX_SETS_INTER][EXT_TX_SIZES][TX_TYPES];
int intra_tx_type_costs[EXT_TX_SETS_INTRA][EXT_TX_SIZES][INTRA_MODES]
[TX_TYPES];
#else
int intra_tx_type_costs[EXT_TX_SIZES][TX_TYPES][TX_TYPES];
int inter_tx_type_costs[EXT_TX_SIZES][TX_TYPES];
#endif // CONFIG_EXT_TX
#if CONFIG_EXT_INTRA
#if CONFIG_INTRA_INTERP
int intra_filter_cost[INTRA_FILTERS + 1][INTRA_FILTERS];
#endif // CONFIG_INTRA_INTERP
#endif // CONFIG_EXT_INTRA
#if CONFIG_LOOP_RESTORATION
int switchable_restore_cost[RESTORE_SWITCHABLE_TYPES];
#endif // CONFIG_LOOP_RESTORATION
#if CONFIG_GLOBAL_MOTION
int gmtype_cost[TRANS_TYPES];
#endif // CONFIG_GLOBAL_MOTION
int multi_arf_allowed;
int multi_arf_enabled;
int multi_arf_last_grp_enabled;
TileDataEnc *tile_data;
int allocated_tiles; // Keep track of memory allocated for tiles.
TOKENEXTRA *tile_tok[MAX_TILE_ROWS][MAX_TILE_COLS];
unsigned int tok_count[MAX_TILE_ROWS][MAX_TILE_COLS];
TileBufferEnc tile_buffers[MAX_TILE_ROWS][MAX_TILE_COLS];
int resize_pending;
int resize_state;
int resize_scale_num;
int resize_scale_den;
int resize_avg_qp;
int resize_buffer_underflow;
int resize_count;
// VAR_BASED_PARTITION thresholds
// 0 - threshold_128x128;
// 1 - threshold_64x64;
// 2 - threshold_32x32;
// 3 - threshold_16x16;
// 4 - threshold_8x8;
int64_t vbp_thresholds[5];
int64_t vbp_threshold_minmax;
int64_t vbp_threshold_sad;
BLOCK_SIZE vbp_bsize_min;
// VARIANCE_AQ segment map refresh
int vaq_refresh;
// Multi-threading
int num_workers;
AVxWorker *workers;
struct EncWorkerData *tile_thr_data;
AV1LfSync lf_row_sync;
#if CONFIG_ENTROPY
SUBFRAME_STATS subframe_stats;
// TODO(yaowu): minimize the size of count buffers
SUBFRAME_STATS wholeframe_stats;
av1_coeff_stats branch_ct_buf[COEF_PROBS_BUFS][TX_SIZES][PLANE_TYPES];
#endif // CONFIG_ENTROPY
#if CONFIG_ANS
struct BufAnsCoder buf_ans;
#endif
#if CONFIG_EXT_REFS
int refresh_frame_mask;
int existing_fb_idx_to_show;
int is_arf_filter_off[MAX_EXT_ARFS + 1];
int num_extra_arfs;
int arf_map[MAX_EXT_ARFS + 1];
#endif // CONFIG_EXT_REFS
#if CONFIG_GLOBAL_MOTION
// Stores number of prediction blocks using global motion and the
// number of 4x4 blocks using it per reference frame.
int global_motion_used[TOTAL_REFS_PER_FRAME][2];
int global_motion_search_done;
#endif
#if CONFIG_REFERENCE_BUFFER
SequenceHeader seq_params;
#endif
} AV1_COMP;
void av1_initialize_enc(void);
struct AV1_COMP *av1_create_compressor(AV1EncoderConfig *oxcf,
BufferPool *const pool);
void av1_remove_compressor(AV1_COMP *cpi);
void av1_change_config(AV1_COMP *cpi, const AV1EncoderConfig *oxcf);
// receive a frames worth of data. caller can assume that a copy of this
// frame is made and not just a copy of the pointer..
int av1_receive_raw_frame(AV1_COMP *cpi, unsigned int frame_flags,
YV12_BUFFER_CONFIG *sd, int64_t time_stamp,
int64_t end_time_stamp);
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);
int av1_get_preview_raw_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *dest);
int av1_get_last_show_frame(AV1_COMP *cpi, YV12_BUFFER_CONFIG *frame);
int av1_use_as_reference(AV1_COMP *cpi, int ref_frame_flags);
void av1_update_reference(AV1_COMP *cpi, int ref_frame_flags);
int av1_copy_reference_enc(AV1_COMP *cpi, AOM_REFFRAME ref_frame_flag,
YV12_BUFFER_CONFIG *sd);
int av1_set_reference_enc(AV1_COMP *cpi, AOM_REFFRAME ref_frame_flag,
YV12_BUFFER_CONFIG *sd);
int av1_update_entropy(AV1_COMP *cpi, int update);
int av1_set_active_map(AV1_COMP *cpi, unsigned char *map, int rows, int cols);
int av1_get_active_map(AV1_COMP *cpi, unsigned char *map, int rows, int cols);
int av1_set_internal_size(AV1_COMP *cpi, AOM_SCALING horiz_mode,
AOM_SCALING vert_mode);
int av1_set_size_literal(AV1_COMP *cpi, unsigned int width,
unsigned int height);
int av1_get_quantizer(struct AV1_COMP *cpi);
void av1_full_to_model_counts(av1_coeff_count_model *model_count,
av1_coeff_count *full_count);
static INLINE int frame_is_kf_gf_arf(const AV1_COMP *cpi) {
return frame_is_intra_only(&cpi->common) || cpi->refresh_alt_ref_frame ||
(cpi->refresh_golden_frame && !cpi->rc.is_src_frame_alt_ref);
}
static INLINE int get_ref_frame_map_idx(const AV1_COMP *cpi,
MV_REFERENCE_FRAME ref_frame) {
#if CONFIG_EXT_REFS
if (ref_frame >= LAST_FRAME && ref_frame <= LAST3_FRAME)
return cpi->lst_fb_idxes[ref_frame - 1];
#else
if (ref_frame == LAST_FRAME) return cpi->lst_fb_idx;
#endif // CONFIG_EXT_REFS
else if (ref_frame == GOLDEN_FRAME)
return cpi->gld_fb_idx;
#if CONFIG_EXT_REFS
else if (ref_frame == BWDREF_FRAME)
return cpi->bwd_fb_idx;
#endif // CONFIG_EXT_REFS
else
return cpi->alt_fb_idx;
}
static INLINE int get_ref_frame_buf_idx(const AV1_COMP *cpi,
MV_REFERENCE_FRAME ref_frame) {
const AV1_COMMON *const cm = &cpi->common;
const int map_idx = get_ref_frame_map_idx(cpi, ref_frame);
return (map_idx != INVALID_IDX) ? cm->ref_frame_map[map_idx] : INVALID_IDX;
}
static INLINE YV12_BUFFER_CONFIG *get_ref_frame_buffer(
const AV1_COMP *cpi, MV_REFERENCE_FRAME ref_frame) {
const AV1_COMMON *const cm = &cpi->common;
const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
return buf_idx != INVALID_IDX ? &cm->buffer_pool->frame_bufs[buf_idx].buf
: NULL;
}
static INLINE const YV12_BUFFER_CONFIG *get_upsampled_ref(
const AV1_COMP *cpi, const MV_REFERENCE_FRAME ref_frame) {
// Use up-sampled reference frames.
const int buf_idx =
cpi->upsampled_ref_idx[get_ref_frame_map_idx(cpi, ref_frame)];
return &cpi->upsampled_ref_bufs[buf_idx].buf;
}
#if CONFIG_EXT_REFS
static INLINE int enc_is_ref_frame_buf(AV1_COMP *cpi, RefCntBuffer *frame_buf) {
MV_REFERENCE_FRAME ref_frame;
AV1_COMMON *const cm = &cpi->common;
for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) {
const int buf_idx = get_ref_frame_buf_idx(cpi, ref_frame);
if (buf_idx == INVALID_IDX) continue;
if (frame_buf == &cm->buffer_pool->frame_bufs[buf_idx]) break;
}
return (ref_frame <= ALTREF_FRAME);
}
#endif // CONFIG_EXT_REFS
static INLINE unsigned int get_token_alloc(int mb_rows, int mb_cols) {
// We assume 3 planes all at full resolution. We assume up to 1 token per
// pixel, and then allow a head room of 1 EOSB token per 4x4 block per plane,
// plus EOSB_TOKEN per plane.
return mb_rows * mb_cols * (16 * 16 + 17) * 3;
}
// Get the allocated token size for a tile. It does the same calculation as in
// the frame token allocation.
static INLINE unsigned int allocated_tokens(TileInfo tile) {
int tile_mb_rows = (tile.mi_row_end - tile.mi_row_start + 1) >> 1;
int tile_mb_cols = (tile.mi_col_end - tile.mi_col_start + 1) >> 1;
return get_token_alloc(tile_mb_rows, tile_mb_cols);
}
void av1_alloc_compressor_data(AV1_COMP *cpi);
void av1_scale_references(AV1_COMP *cpi);
void av1_update_reference_frames(AV1_COMP *cpi);
void av1_set_high_precision_mv(AV1_COMP *cpi, int allow_high_precision_mv);
YV12_BUFFER_CONFIG *av1_scale_if_required_fast(AV1_COMMON *cm,
YV12_BUFFER_CONFIG *unscaled,
YV12_BUFFER_CONFIG *scaled);
YV12_BUFFER_CONFIG *av1_scale_if_required(AV1_COMMON *cm,
YV12_BUFFER_CONFIG *unscaled,
YV12_BUFFER_CONFIG *scaled);
void av1_apply_encoding_flags(AV1_COMP *cpi, aom_enc_frame_flags_t flags);
static INLINE int is_altref_enabled(const AV1_COMP *const cpi) {
return cpi->oxcf.mode != REALTIME && cpi->oxcf.lag_in_frames > 0 &&
cpi->oxcf.enable_auto_arf;
}
// TODO(zoeliu): To set up cpi->oxcf.enable_auto_brf
#if 0 && CONFIG_EXT_REFS
static INLINE int is_bwdref_enabled(const AV1_COMP *const cpi) {
// NOTE(zoeliu): The enabling of bi-predictive frames depends on the use of
// alt_ref, and now will be off when the alt_ref interval is
// not sufficiently large.
return is_altref_enabled(cpi) && cpi->oxcf.enable_auto_brf;
}
#endif // CONFIG_EXT_REFS
static INLINE void set_ref_ptrs(const AV1_COMMON *cm, MACROBLOCKD *xd,
MV_REFERENCE_FRAME ref0,
MV_REFERENCE_FRAME ref1) {
xd->block_refs[0] =
&cm->frame_refs[ref0 >= LAST_FRAME ? ref0 - LAST_FRAME : 0];
xd->block_refs[1] =
&cm->frame_refs[ref1 >= LAST_FRAME ? ref1 - LAST_FRAME : 0];
}
static INLINE int get_chessboard_index(const int frame_index) {
return frame_index & 0x1;
}
static INLINE int *cond_cost_list(const struct AV1_COMP *cpi, int *cost_list) {
return cpi->sf.mv.subpel_search_method != SUBPEL_TREE ? cost_list : NULL;
}
void av1_new_framerate(AV1_COMP *cpi, double framerate);
#define LAYER_IDS_TO_IDX(sl, tl, num_tl) ((sl) * (num_tl) + (tl))
// Update up-sampled reference frame index.
static INLINE void uref_cnt_fb(EncRefCntBuffer *ubufs, int *uidx,
int new_uidx) {
const int ref_index = *uidx;
if (ref_index >= 0 && ubufs[ref_index].ref_count > 0)
ubufs[ref_index].ref_count--;
*uidx = new_uidx;
ubufs[new_uidx].ref_count++;
}
#ifdef __cplusplus
} // extern "C"
#endif
#endif // AV1_ENCODER_ENCODER_H_